text
stringlengths 4
2.78M
| meta
dict |
|---|---|
---
abstract: 'A long-standing problem in astrophysics is to measure the mass associated with galaxies. Gravitational lensing provides one of the cleanest ways to make this measurement. To date, the most powerful lensing probes of galactic mass have been multiply imaged QSO’s (strong lensing of a background point source) and galaxy-galaxy lensing (weak deformation of many background galaxies). Here we point out that the mass associated with galaxies also lenses the Cosmic Microwave Background (CMB) and this effect is potentially detectable in small scale experiments. The signal is small (roughly a few tenths of $\mu$K) but has a characteristic shape and extends out well beyond the visible region of the galaxy.'
author:
- 'Scott Dodelson$^{1,2}$ and Glenn D. Starkman$^{3}$'
title: 'Galaxy-CMB Lensing'
---
Introduction.
=============
Flat rotation curves at large distance from galactic centers imply that the mass associated with a galaxy extends far beyond the region that is visible. Indeed, it now appears that the vast majority of the mass of a galaxy is in an unseen component, dark matter. How much dark matter is there in a galaxy and how far out does the distribution go? How does the matter associated with a single galaxy compare with the overdensity around that galaxy due to large scale structure? How is the dark matter distributed inside the galaxy? Is there substructure, as expected in simulations of Cold Dark Matter (CDM) models [@moore; @andrei; @bullock], or does the absence of many satellites for our Galaxy imply that the CDM models need to be modified[@spergel; @yoshida; @colin; @burkert]?
These questions increasingly interest not only astrophysicists studying galaxies and their properties, but also cosmologists and particle physicists, for these phenomena may depend on the power spectrum of the matter in the universe. This in turn depends on such fundamental quantities as neutrino masses and the details of inflation. If the substructure dilemma remains, the properties of dark matter, such as its scattering cross-section or annihilation rate, may be responsible. These properties may be important clues in identifying the elementary particles which constitute the dark matter and the fundamental theory governing their interactions.
The traditional method of studying the mass distribution in galaxies – measuring rotation velocities – has recently been supplemented with a relatively new technique: studying the deflection of light rays as they pass by or through the galaxy. There are a number of ways in which gravitational lensing can probe the mass distribution of an object like a galaxy, but two have emerged recently as particularly promising. First is the phenomenon of multiply imaged QSOs. Light from a background point source (the QSO) gets lensed so that multiple images are seen. The separations between these images, and more importantly their magnifications, are sensitive to the local mass distribution [@mao; @metcalfe; @dalal; @chiba; @keeton]. Second, images of background galaxies are distorted by a foreground lensing galaxy, and the amplitude of this distortion as a function of projected distance from the center of the lens galaxy contains important information about the mass associated with that galaxy [@tyson; @brainerd; @fischer; @smith; @wkl; @mckay; @hoekstra; @klein]. Due to the presence of large scale structure, the lensing mass extends far beyond the visible region of the galaxy. If the diameter of the visible region is of order $10$ kpc, the region within a sphere almost a thousand times larger will be overdense. Understanding this overdensity is an important step along the way to understanding galaxy formation and the correlation between mass and light in the universe [@berwei; @guzik; @cooray; @weidav; @bwb; @jain].
In this paper, we explore the possibility that the most distant sources of all, the Cosmic Microwave Background (CMB) anisotropies generated at redshift $z\simeq 1100$, can serve as the sources which are lensed by foreground galaxies. The problem is very similar to the lensing of the CMB by foreground galaxy clusters [@sz], except that the amplitude of the signal is considerably smaller here. The observed temperature $T$ as a function of 2D position $\vec\theta$ on the sky is T() &=& T(’) = T(-) [\
]{}&& T() - T() where $\tilde T$ is the background CMB field originating from the unlensed position $\vec\theta'=\vec\theta -\vec{\delta\theta}$, and $\delta\theta$ is the deflection due to the lens. As illustrated in [@sz], the CMB anisotropy field on small scales is almost purely a dipole (or gradient; on these small scales the two terms are interchangeable); aligning the $y-$ axis with the dipole leads to a simple expression for the observed temperature: T() ( \_y - \_y) .[\[eq:tbasic\]]{}Here $\partial T/\partial\theta_y$ is the temperature gradient, which is of course zero on average. Its rms fluctuation though is equal to $0.22\mu K$ arcsec$^{-1}$ in the standard $\Lambda$CDM cosmology with $\Omega_{\rm matter}=0.3$, $\Omega_{\rm baryon} = 0.04$, $H_0=70$ km sec$^{-1}$Mpc$^{-1}$ and $\Omega_\Lambda=0.7$ (the rest of this paper assumes this background cosmology). The deviations from a pure dipole arise from the deflection angle $\vec{\delta\theta}$. We will see that the deflection angle induced by the mass associated with a galaxy is typically of order an arcsecond, so the expected signal is small, tenths of $\mu$K per galaxy. Eq. [\[\[eq:tbasic\]\]]{} breaks down on scales larger than ten arcminutes when the background CMB no longer behaves as a pure gradient due to the power on the scales of the acoustic peaks [@mc]. Even on smaller scales, the “noise” due to the quadrupole can be significant as we will see in §[III]{}.
§[II]{} introduces the basic formula relating the deflection angle to the mass distribution and presents the deflection angle for several simple mass distributions. §[III]{} considers the lensing due to the large scale overdensities surrounding a galaxy. These extend out to several Mpc (or tens of arcminutes for a galaxy at redshift $0.1$). Given the resolution and sensitivity of current CMB experiments, this large scale lensing is likely to be the easiest target for future experiments. Just as in the galaxy-galaxy lensing measurements, the signal from a single foreground galaxy is quite small, so the statistics of many galaxies must be used to beat down the noise. §[IV]{} discusses the foregrounds that might contaminate this signal. We let our fancy take flight in §[V]{} where we consider the signal due to a single galaxy and show that in principle the lensed CMB can detect the sub-structure that simulations predict must exist in galactic halos. Observing this signal would require experiments with significantly better resolution and sensitivity than those currently planned, but if the small scale problems of CDM persist, we should keep this technique in mind as a way of definitively resolving the “small scale crisis.”
One final comment about previous work. Several years ago, Peiris and Spergel[@peiris] analyzed many aspects of the CMB-galaxy cross-correlations. Their analysis, focused on the WMAP experiment[@wmap], and therefore is for scales larger than we consider here.
Deflection Angle and Mass Models
================================
The deflection angle of a photon due to a galaxy at redshift $z_L$, at comoving distance from us, ${\chi_L}$, is[@mcex] = -8G [([\_S]{}-[\_L]{}) [\_S]{}]{} .[\[eq:dtheta\]]{}Here we take the photon to be emitted from a comoving distance ${\chi_S}=1.4\times 10^4$ Mpc away from us, corresponding to the surface of last scattering. Its unperturbed path has at all times an angular distance $\theta$ from the axis connecting us to the center of the galaxy. The integral is over the two dimensional mass density $\Sigma$ of the galaxy, which is assumed to be azimuthally symmetric.
A singular isothermal sphere has a density profile proportional to $r^{-2}$, so that the surface density is proportional to $R^{-1}$, where $r\, (R)$ is the 3D (2D) radius from the center of the galaxy. For this profile then the term in square brackets in [Eq. \[\[eq:dtheta\]\]]{} is independent of $\theta$ and the amplitude is fixed by the correlation length $r_0$ defined via (r) = \_m (r\_0/r)\^2 .Here $\rho_m$ is the average matter density in the universe, so that $8\pi G \rho_m = 0.9 H_0^2$ with $c/H_0 = 4300$ Mpc. The surface density for this distribution is $\Sigma(R)=\pi \rho_m r_o^2/R$, so the amplitude of the deflection angle due to a singular isothermal sphere is $0.9 H_0^2 \pi r_0^2 = 1'' (r_0/5.6\, {\rm Mpc})^2$ for ${\chi_S}\gg {\chi_L}$. Simulations [@guzik] suggest that the distribution of matter around galaxies, which is formally measured by the galaxy-mass cross-correlation function may indeed be represented by this distribution with $r_0\simeq 7$ Mpc. The distortion, at least on scales larger than about $100$ kpc, is therefore expected to be constant with an amplitude of a little more than an arcsecond. Figure [\[fig:rhoint\]]{} shows this deflection angle; it falls off on large scales since we do not include contributions from the density beyond $r_0$. Also shown is a more conservative choice of $\rho=\rho_m (r_0/r)^{-1.8}.$ Even for this choice, the expected deflection angle is of order an arcsecond.
It might seem at first that detecting such a small deflection will be hopeless given that the rms deflection of a photon from the CMB is as much as 3 arcminutes. However, this is due almost entirely to lensing by large scale (low $\ell$) fluctuations. Large patches of the CMB sky will thus be nearly uniformly deflected by up to several arcminutes, however, by $\ell=10^4$, the rms deflection is down to about half an arc-second. Thus, CMB lensing by galaxies will take place against a relatively smoothly-lensed background, which is still well described by a local dipole.
= 0.7
The above estimates are for a statistical sample of galaxies, accounting for the fact that the universe is clustered, so we expect halos of dark matter to reside near each other. These estimates definitely break down on scales smaller than $\sim 100$ kpc. On these smaller scales, a more appropriate distribution is given by the Navarro, Frenk, and White (NFW) [@nfw] profile: (xr/[r\_s]{}) = [[\_s]{}x(1+x)\^2]{} .The NFW profile can be described by two parameters: the scale radius ${r_s}$ and the density parameter ${\rho_s}$. More commonly, these are traded in for the concentration $c$, which is the ratio of the virial radius to the scale radius, and ${v_{\rm max}}$, the maximum rotational velocity due to this mass distribution. The virial radius is defined as that within which the average density is equal to $\Delta\rho_m$ with $\Delta=337$[@bryan; @bul337] for the $\Lambda$CDM model in which we are working. The maximum rotational velocity can be determined analytically in terms of ${r_s}$ and ${\rho_s}$; it is ${v_{\rm max}}= 0.46(4\pi G{\rho_s}{r_s}^2)^{1/2}$. Figure [\[fig:rhoint\]]{} shows the deflection due to an NFW profile with ${v_{\rm max}}=200$ km sec$^{-1}$ and concentration $c=8$. Analytically, |\_[NFW]{} = -0.54” ( [ [v\_[max]{}]{}200 [km sec]{}\^[-1]{}]{} )\^2 [([\_S]{}-[\_L]{})[\_S]{}]{} f() , [\[eq:nfwlens\]]{} where $f(\lambda)$ is a smooth function which reaches its peak of one at $\lambda=1.3$: f() $\lambda$ is defined as .
Large Scale Galaxy-CMB Lensing
==============================
Recently, the Sloan Digital Sky Survey (SDSS) detected the signal of galaxy-galaxy lensing out to a Mpc [@mckay]. It is useful to recap their technique to place galaxy-CMB lensing in context. They select lenses as those galaxies bright enough for spectra to have been taken (Petrosian magnitude $r^*<17.6$) and background galaxies fainter than $18$th magnitude. For the background galaxies, photometric redshifts are used. About $30,000$ lens galaxies are chosen from SDSS commissioning data and these are probed by about $3\times 10^6$ background galaxies. Thus each lens has about a hundred background galaxies behind it. The error in the shear produced by a single foreground galaxy is $0.4/\sqrt{N_{bg}}$ where the $0.4$ comes from a combination of instrumental noise and the intrinsic ellipticities of the background galaxies. Thus, if $N_{bg}=100$, the noise in the shear is $0.04$. This is almost a factor of ten larger than the expected signal. So the SDSS survey cannot detect galaxy-galaxy lensing due to a single galaxy. Instead they must average the signal over many foreground galaxies (each with a signal to noise of order $0.1$). SDSS nevertheless provides a useful measure of the galaxy-mass correlation function because there are many ($3\times 10^4$) foreground galaxies over which to average.
How does the SDSS signal to noise compare to that obtainable with the CMB? We have determined (Eqs. [\[\[eq:tbasic\]\]]{} and [\[\[eq:dtheta\]\]]{}) that the expected signal in a CMB experiment is T() - [T\_y]{} =[T\_y]{} b() where $\phi$ is the angle between the angular position $\vec\theta$ and the $y$-axis along which the background gradient is oriented and $b(\theta)$ is roughly constant with an amplitude of order one arcsecond. We also need to consider the deviation of the background anisotropy pattern $\tilde T$ from a pure gradient. On scales larger than $5-10'$, the background field is more complicated. For simplicity we will assume that the background dipole can be removed but the quadrupole cannot, and it serves as a source of noise. That is, we consider the next term in the expansion of $\tilde T$: T() + [12]{} \_i\_j [\^2T\_i \_j]{} .This second term on the right also has mean of zero, but variance equal to \_Q\^2 = [332]{} \^4 \_l [l\^5 C\_l2]{} = (1.5 K)\^2 ( [1’]{} )\^4 where the second equality holds in the standard cosmology we are considering. If the quadrupole cannot be removed, then this source of noise adds in quadrature with that due to the atmosphere and/or instrument.
We can now compute the signal to noise expected for lensing of the CMB due to a single galaxy. Consider a CMB experiment which maps the sky into $N_p$ pixels each of area $\Delta\Omega$ with instrumental/atmospheric noise per pixel $\sigma_n$. Then, ( [SN]{} )\^2 &=& ( [T\_y]{} )\^2 \_[i]{}\^[N\_p]{} [\^2\_i \_n\^2 + \_Q\^2(\_i)]{} b\^2(\_i)&& ( [T\_y]{} b)\^2 = \^2 .[\[eq:sn\]]{}So, for a CMB experiment to achieve resolution comparable to SDSS (signal to noise per galaxy of $0.1$), beams of order an arcminute with noise per pixel of order $10\mu$K are needed. This is within the range expected of upcoming experiments. A note of caution: the $\simeq$ sign in [Eq. \[\[eq:sn\]\]]{} is a warning that this signal to noise estimate assumes the pixel size is significantly smaller than the scale at which the background dipole approximation breaks down; i.e. $\Delta\Omega \ll (10')^2$.
One might wonder whether, given the functional dependance of $\delta\vec\theta$ on the lens distance/redshift ([Eq. \[\[eq:nfwlens\]\]]{} and Ref. [@song] for example), it might not be more advantageous to look to higher redshift objects rather than SDSS galaxies as lenses For example, lyman-alpha/star-forming regions at $z\simeq 3$ might be expected to give significantly higher a $S/N$. Studying the mass distribution in such objects would indeed be of significant interest; however, since these high redshift objects will subtend a significantly smaller solid angle than medium-to-low redshift galaxies, it will have to wait until a dramatic improvement in angular resolution before their CMB-lensing signal can be properly studied. We therefore focus our attention on SDSS-type galaxies as lenses.
Since the noise from the quadrupole dominates over instrumental noise when $\theta> 1' (\sigma_n/1.5\mu K)^{1/2}$, and a fair fraction of the signal comes from these larger scales, reducing instrumental noise is not as a important as going to higher resolution and/or covering more sky. That is, signal to noise estimates typically scale as $(\Delta\Omega \sigma_n^2)^{-1/2}$ when the dominant noise is instrumental. Here, though, the scaling is $(\Delta\Omega \sigma_n)^{-1/2}$. For fixed resolution, then, the final signal to noise scales as $N_{rm gal}^{1/2} \sigma_n^{-1/2}
\propto N_p^{1/2} \sigma_n^{-1/2}$. The number of pixels covered if the total time is fixed is inversely proportional to the time spent on each individual pixel, while $\sigma_n$ is inversely proportional to the square root of time per pixel. Therefore, the final signal to noise scales increases as less time is spent on each pixel: an experiment intent on measuring CMB-galaxy lensing should strive for large sky coverage at the expense of sensitivity.
For CMB experiments with larger beams, the approximation that the background source is a gradient breaks down. Information is still contained in the cross-correlation of the CMB and galaxy surveys [@peiris]. As we have seen the signal is of order a tenth of a $\mu$K, while the noise per pixel due to primordial CMB fluctuations is naively of order $50\mu$K. Without a sophisticated algorithm to extract the signal, we would then need of order $500^2$ galaxies to beat down the noise. Going to larger redshifts to pick up these galaxies would not necessarily help since the galaxy-dark matter cross-correlation functions falls off at high redshifts, so the signal is significantly smaller. Also, a galaxy projects to a smaller angular size at high redshift, necessitating even higher resolution. Although we do not pursue it here, two possible approaches are: (i) to assume the form of the lensing profile and fit for its amplitude [@sz; @peiris] or (ii) to extend the likelihood technique developed by Hirata and Seljak [@hirata] to this cross-correlation case.
Noise
=====
Before we assert that the lensing signal from galaxies is observable, we must identify the other potential sources of noise. Extensive discussions of CMB noise sources on small angular scales are to be found in [@tegmark; @toffolatti; @teho], and are reviewed in [@sz] in the context of cluster lensing, where the issues are similar.
The expected sources of noise (other than detector noise and intrinsic noise from the CMB itself, both considered in §[III]{}) are: Galactic synchrotron, free-free and dust emission, thermal Sunyaev-Zeldovich (SZ) from galaxy clusters and filaments, kinetic SZ (including Ostriker-Vishniac) resulting from bulk gas motions, and unresolved point sources. We briefly review each of these in turn.
### Galactic emissions
Galactic synchrotron and free-free emission both decline with increasing $\ell$. Fluctuations due to free-free emission are already below $0.1\mu$K by $\ell=10^4$ above $30$ GHz [@teho]. Above $100$ GHz synchrotron emission fluctuations are below $0.1\mu$K by $\ell=10^3$; at 30GHz they are predicted to be $\sim2\mu$K at $\ell\simeq 1000$, and declining as approximately $\ell^{1/3}$ [@tegmark]. This puts them below $1\mu$K below one arcminute. Temperature fluctuations due to Galactic dust emissions are expected to be below $1\mu$K at frequencies below 217 GHz for $\ell > 3000$ if the dust emission is vibrational. The situation was less clear if the dust emission is rotational [@teho], but the recent WMAP[@wmapfor] constraints suggest that rotational dust will not be a problem.
### Thermal SZ
Of particular interest for galaxies being observed in the foreground of an SZ cluster, are fluctuations in the cluster SZ. Clusters will themselves have substructure – component galaxies, fluctuations in electron temperature and density across the cluster. To the extent that these are resolvable, whether in the CMB data itself, in cluster radio maps, or in X-ray maps (especially with future X-ray interferometers), they could be subtracted and one could select as target galaxies those that do not have significant cluster structure in their backgrounds.
The unmodeled thermal SZ foreground has been calculated by a number of groups[@bond; @zhang; @hernquist]. The power is proportional to $\sigma_8^7$, so there remains significant uncertainty in the amplitude. If $\sigma_8$ is set to its WMAP value of $0.8$[@wmapspe], then the expected rms amplitude of the thermal SZ on the scales of interest is of order $5\mu$K. Of course, the signal (or in this case, the noise) goes away at $217$GHz so it is always possible in principle to avoid this source of contamination. (The relatvistic correction which spoils the null can be expected to be severely depresssed below the $5\mu$K level.)
### kinetic SZ and Ostriker-Vishniac
The kinetic SZ effect is the Doppler shift imprinted on CMB photons when they scatter off electrons in a moving gas. In the linear regime for the matter fluctuations this is called the Ostriker-Vishniac (OV) effect [@vishniac]. The kinetic effect has the same spectral shape as the primordial CMB (and the lensing signal we are after) so it is more difficult to eliminate than the thermal effect. The amplitude of the effect though is of order $2 \mu$K[@hu; @sil1; @sil2; @gnedin; @hernquist; @val; @ma; @zhang; @zh2], so it is not a major stumbling block.
### Point Sources
Toffolatti [*et. al.*]{} [@toffolatti] considered the contribution of unresolved point sources to fluctuations in the anisotropy of the CMB on 1-100 arcminute scales, at the 8 Planck wavebands from 30GHz to 857GHz. These sources include, in the low frequency channels, radio loud AGNs, flat-spectrum radio-galaxies, quasars, and high-z BL-Lacs. Dust emission from distant dust-rich young galaxies is the dominant source at higher frequencies. Assuming that point sources with flux below 1mJy remain unsubtracted, they find (figures 5 and 6) that in all channels $\langle(\Delta T/T)^2\rangle^{1/2} > 10^{-6}$ at 1 arcminute (best channel – 143 GHz), and in many cases is much bigger. It is also increasing with decreasing $\theta$ as approximately $\theta^{-1}$, suggesting that in the best channels (between about 44 and 150 GHz) we can expect point sources to contribute about 5-10 microK of noise at 6 arcsecond , and maybe 10-15 microK of noise at 1 arcsecond.
To make much improvement in these noise figures you would need to be able to accurately subtract very low flux point sources – about 0.1 mJy to achieve a a factor of 3, 0.01mJy to get down an order of magnitude. It is also true that point sources will not exhibit the coherent extended structure expected from lensing and may possibly be filtered out. The biggest confusion is likely to arise if we attempt to use galaxy lensing to measure the proto-galactic substructures. On this scale the foreground dust emission is dominant. However, it also has a rather different energy spectrum than the SZ-distorted CMB, which may aid in its subtraction.
### Other Sources of Confusion
There are several other effects that produce signals similar to that considered here. Moving clusters produce nonlinear corrections to the Integrated Sachs Wolfe Effect on small scales, and the pattern produced has the same structure as that due to lensing. The signal is quite small though[@aghanim; @coo2] ($\sim 0.3\mu$K) and on larger scales than interest us. Coherent electron rotational velocities in clusters lead to a dipolar signature[@cooche; @chluba] and these are on relatively small scales. But the amplitude is also very small and the induced signal is not aligned with the cosmic dipole.
Single Galaxy Lensing
=====================
The lensing signal due to the halo surrounding a single galaxy is given by [Eq. \[\[eq:nfwlens\]\]]{}. Since the signal is very small, we cannot hope to measure it with the current, or even currently planned, CMB experiments. Here we simply motivate future experiments by illustrating that the potential pay-off is large: lensing of the CMB in principle allows us to differentiate between an NFW profile and an isothermal profile, thereby testing one result of numerical simulations. More ambitiously, very high resolution measurements could detect the sub-structure predicted by current theories of structure formation.
Detecting the shape of the dark matter profile around a single galaxy with CMB lensing requires resolution and sensitivity far beyond current capabilities. Figure [\[fig:med\]]{} shows the different signals induced by an NFW profile and an isothermal profile. The galaxy is situated at $z=0.1$ (the lower the redshift, the larger it appears, and therefore the less difficult is the issue of resolution). A number of groups [@bul337] have found that typical values of the concentration parameter of galaxies at redshift one are $5-10$ in CDM models. The galaxy doing the lensing in Figure [\[fig:med\]]{} has concentration parameter equal to eight and a maximum circular velocity of $200$ km sec$^{-1}$. This corresponds to a scale radius ${r_s}=44$ kpc and a virial mass of $2\times 10^{12} h^{-1} M_\odot$.
= 0.7
Even Figure [\[fig:med\]]{}, with its optimistic noise and resolution parameters, paints too pretty a picture. For, we have neglected all foregrounds, save for the local quadrupole of the CMB. Noise due to the quadrupole kicks in at surprisingly small scales, because of the exquisite sensitivity required to discriminate these models.
The surface density of a smooth NFW profile with these parameters is depicted in Figure [\[fig:sig\]]{}. Unhindered by experimental complications, we are led to ask whether a smooth NFW profile could be distinguished from the clumpy profiles found in simulations.
To construct a clumpy halo, we use the distribution of sub-halos measured by [@moore]. They find that the cumulative number of halos with circular velocity greater than ${v_{\rm max}}$ is roughly equal to $0.03 ({v_{\rm max}}/v_{\rm parent})^{-3}$. This means that the distribution of halos dn/d[v\_[max]{}]{} .[\[eq:dndv\]]{} Here the lower cut-off arises because we don’t know what the simulations predict on such small scales. Note that with this distribution, if we take ${v_{\rm cutoff}}= 0.05v_{\rm parent}$, there are $240$ sub-halos in the galaxy.
To generate a distribution of velocities of sub-halos, first normalize [Eq. \[\[eq:dndv\]\]]{} to unity: $dn^{\rm norm}/d{v_{\rm max}}= (1/N) dn/d{v_{\rm max}}$ and then given a random number $\eta$ between zero and one, define the velocity of one sub-halo $v$ via = \_[[v\_[cutoff]{}]{}]{}\^v d[v\_[max]{}]{}[dn\^[norm]{}d[v\_[max]{}]{}]{}.In this case we can do the integral analytically, so [v\_[max]{}]{}= [ [v\_[cutoff]{}]{}(1 - )\^[1/3]{}]{} .
We can use the same technique to generate the spatial distribution of halos. Suppose that the number of halos is distributed according to an NFW profile. We want $\Sigma(R)$ from the sub-structure to follow an NFW profile. The total mass within a radius $R$ is $2\pi\int_0^R dR'\, R'\,\Sigma_{\rm NFW}(R')$. So the fraction of the total mass which is within $R$ is (R) = [ \_0\^[R/[r\_s]{}]{} dxx\_[NFW]{}(x) \_0\^[R\_[max]{}/[r\_s]{}]{} dxx\_[NFW]{}(x) ]{} .Here we have cut off the distribution at $R_{\rm max}= 10{r_s}$; note that $\eta$ lies between zero and one, so a random number between zero and one can be associated with a given value of $R$. The angular coordinate of the sub-halo is also assigned randomly. This distribution is shown in the right panel of Figure [\[fig:sig\]]{}.
Figure [\[fig:txty\]]{} shows the imprint on the CMB resulting from both the smooth and the clumpy galaxy. Note the characteristic signal first pointed out in Ref. [@sz]: a hot and cold lobe on either side of the galactic center. (Here we have subtracted off the dipole.) Traces of the substructure are still evident in the right panel of Figure [\[fig:txty\]]{}. However, these are much less obvious than those in the surface density plot of Figure [\[fig:sig\]]{}. The similarity of the two panels in Figure [\[fig:txty\]]{} is reflection of the fact that the galaxy is lensing a large “sheet,” the background dipole, as opposed to the point-like QSO’s with which we are more familiar [@pkeet].
A realistic CMB experiment will have a finite width beam. Figure [\[fig:gauss\]]{} shows the results of smoothing the signal with a beam of full width half maximum of $5''$. Even after such smoothing, there remains a noticeable difference between the clumpy and smooth galaxies.
Conclusion
==========
Mass surrounding a galaxy lenses the background cosmic photons arriving from the surface of last scattering at $z\simeq 1100$. This lensing can be used to probe the dark matter distribution around galaxies. On the relevant scales – of order an arcminute – the background is nearly a dipole, and the amplitude of the signal is of order $1''\times \partial T/\partial\theta \sim 0.2\mu$K. This signal can be used in two fundamentally different ways. On the one hand, one can average over many foreground galaxies so as to obtain a measure of the galaxy-mass correlation function. We have shown that given enough sky coverage a CMB experiment with an arcminute beam size and noise of $10\mu$K per pixel would be able to make a measurement of this correlation function that is competitive with SDSS. Instruments like the Penn Bolometric Array – a $3$mm camera for the $100$m Green Bank telescope – hold some promise for beginning a more detailed exploration of galaxy mass distributions. It is worth noting that, unlike measurements of lensing-induced shear or amplification, which are sensitive to the second derivative of the lens gravitational potential, $\Phi$, this proposed measurement of the deflection angle is sensitive to $\nabla\Phi$. In principle, the inversion to obtain $\Phi$, is far easier from $\nabla\Phi$ than from $\nabla\nabla\Phi$."
In the distant future, we might be able to go further and probe the dark matter halo of an individual galaxy. For noise levels reached today, the signal to noise from a single galaxy is of order $1/100$. Either noise levels will have to come down appreciably, or a different source needs to be used. This latter option is intriguing. One particularly appealing possibility is to use the gradient caused by the thermal Sunyaev-Zel’dovich (SZ) effect (the scattering of CMB photons by electrons in an ionized plasma) from a background galaxy cluster. This results from CMB photons scattering off hot electrons in the cluster’s ionized IGM. A typical expected temperature gradient is of order $500\mu$K arcmin$^{-1}$, almost two orders of magnitude above that from the primordial CMB. If a galaxy were positioned along the line of sight to a cluster, where the SZ effect would look nearly like a gradient on sub-arcminute scales, then we would expect the galaxy lensing signal to be $\sim 5-10 \mu$K. This is then comparable to current noise levels in CMB experiments, so might be the best hope of measuring this fascinating signal in the near future.
The work of SD is supported by the DOE, by NASA grant NAG5-10842, and by NSF Grant PHY-0079251. GDS is supported by a DOE grant to astrophysics theory group at CWRU. GDS thanks the astrophysics group at Fermilab where this work began and the Kavli Institute for Theoretical Physics where much of this work was completed. The Kavli Institute is supported in part by National Science Foundation grant PHY99-07949. We thank Asantha Cooray and Andrey Kravstov for very helpful comments on an earlier draft and Lam Hui and Charles Keeton for their insight and advice.
[99]{}
B. Moore et al., [ [*Astrophys. J.*]{} [**524**]{}, L19 (1999) ]{}
A. Klypin, A. V. Kravtsov, O. Valenzuela, and F. Prada, [ [*Astrophys. J.*]{} [**522**]{}, 82 (1999) ]{}
J. S. Bullock, A. V. Kravtsov, and D. H. Weinberg, [ [*Astrophys. J.*]{} [**539**]{}, 517 (2000) ]{}
D. N. Spergel and P. J. Steinhardt, [ [*Phys. Rev. Letters*]{} [**84**]{}, 3760 (2000) ]{}
N. Yoshida, V. Springel, S. D. M. White, and G. Tormen, [ [*Astrophys. J.*]{} [**544**]{}, L87 (2000) ]{}
P. Colin, V. Avila-Reese, O. Valenzuela, and C. Firmiani, [ [*Astrophys. J.*]{} [**581**]{}, 777 (2002) ]{}
E. D’Onghia and A. Burkert, [ [*Astrophys. J.*]{} [**586**]{}, 12 (2003) ]{}
S. Mao and P. Schneider, [[*Monthly Notices of Royal Astronomical Society*]{} [**295**]{}, 587 (1998)]{}
R. B. Metcalfe and P. Madau, [ [*Astrophys. J.*]{} [**563**]{}, 9 (2001) ]{}
N. Dalal and C. S. Kochanek, [ [*Astrophys. J.*]{} [**572**]{}, 25 (2002) ]{}
M. Chiba, [ [*Astrophys. J.*]{} [**565**]{}, 17 (2002) ]{}
C. Keeton, astro-ph/0111595
J. A. Tyson, F. Valdes, J. F. Jarvis, and A. P. Mills, [ [*Astrophys. J.*]{} [**281**]{}, L59 (1984) ]{}
T. G. Brainerd, R. D. Blanford, and I. Smail, [ [*Astrophys. J.*]{} [**466**]{}, 623 (1996) ]{}
P. Fischer et al., [[*Astronomical J.*]{} [**120**]{}, 1198 (2000)]{}
D. Smith, G. Bernstein, P. Fischer, and M. Jarvis, [ [*Astrophys. J.*]{} [**551**]{}, 643 (2000) ]{}
G. Wilson, N. Kaiser, G. A. Luppino, and L. L. Cowie, [ [*Astrophys. J.*]{} [**555**]{}, 572 (2001) ]{}
T. A. McKay et al., [[astro-ph/]{}[0108013]{}]{}
Y.-S. Song, A. Cooray, L. Knox, and M. Zaldarriaga, [[astro-ph/]{}[0209001]{}]{}
H. Hoekstra et al., [[astro-ph/]{}[0206103]{}]{}
M. Kleinheinrich et al., [[astro-ph/]{}[0304208]{}]{}
A. A. Berlind and D. H. Weinberg, [ [*Astrophys. J.*]{} [**575**]{}, 587 (2002) ]{}
J. Guzik and U. Seljak, [[*Monthly Notices of Royal Astronomical Society*]{} [**321**]{}, 439 (2001)]{}
A. Cooray, [[astro-ph/]{}[0206068]{}]{}
D. H. Weinberg, R. Dav’e, N. Katz, and L. Hernquist, [[astro-ph/]{}[0212356]{}]{}
A. A. Berlind et al., [[astro-ph/]{}[0212357]{}]{}
B. Jain, R. Scranton, and R K. Sheth, [[astro-ph/]{}[0304203]{}]{}
U. Seljak and M. Zaldarriaga, [ [*Astrophys. J.*]{} [**538**]{}, 57 (2000) ]{}
S. Dodelson, [*Modern Cosmology*]{} (Academic Press, Amsterdam, 2003)
H. V. Peiris and D. N. Spergel, [ [*Astrophys. J.*]{} [**540**]{}, 605 (2000) ]{}
C. L. Bennett et al., [[astro-ph/]{}[0302207]{}]{}
See, e.g., J. A. Peacock, [*Cosmological Physics*]{} (Cambridge University Press, Cambridge, 1999) or Ref. [@mc], Chapter 10.
J. F. Navarro, C. S. Frenk, and S. D. M. White, [ [*Astrophys. J.*]{} [**462**]{}, 563 (1996) ]{}
G. Bryan and M. Norman, [ [*Astrophys. J.*]{} [**495**]{}, 80 (1998) ]{}
J. Bullock et al., [[*Monthly Notices of Royal Astronomical Society*]{} [**321**]{}, 559 (2001)]{}
C. Hirata and U. Seljak, [ [*Phys. Rev. D*]{} [**67**]{}, 043001 (2003) ]{}
M. Tegmark and G. Efstathiou, [[*Monthly Notices of Royal Astronomical Society*]{} [**281**]{}, 1297 (1996)]{}
M. Toffolatti et al., [[*Monthly Notices of Royal Astronomical Society*]{} [**297**]{}, 117 (1998)]{}
M. Tegmark, D. J. Eisenstein, W. Hu, and A. de Oliveira-Costa, [ [*Astrophys. J.*]{} [**530**]{}, 133 (2000) ]{}
C. L. Bennett et al., [[astro-ph/]{}[0302208]{}]{}
P. J. Zhang, U.-L. Pen, and B. Wang, [ [*Astrophys. J.*]{} [**577**]{}, 555 (2002) ]{}
J. R. Bond, M. I. Ruetalo, J. W. Wadsley, and M. D. Gladders, [[astro-ph/]{}[0112499]{}]{}
V. Springel, M. White, and L. Hernquist, [ [*Astrophys. J.*]{} [**549**]{}, 681 (2001) ]{}; M. White, L. Hernquist, and V. Springel, [ [*Astrophys. J.*]{} [**579**]{}, 16 (2002) ]{} D. N. Spergel et al., [[astro-ph/]{}[0302209]{}]{}
J. P. Ostriker and E. Vishniac, [ [*Astrophys. J.*]{} [**306**]{}, 51 (1986) ]{}; E. Vishniac, [ [*Astrophys. J.*]{} [**322**]{}, 597 (1987) ]{}
W. Hu, [ [*Astrophys. J.*]{} [**529**]{}, 12 (2000) ]{}
A. C. da Silva, et al. [[*Monthly Notices of Royal Astronomical Society*]{} [**326**]{}, 155 (2001)]{}
A. C. da Silva, et al., [ [*Astrophys. J.*]{} [**561**]{}, L15 (2001) ]{}
N. Y. Gnedin and A. Jaffe, [ [*Astrophys. J.*]{} [**551**]{}, 3 (2001) ]{}
P. Valageas, A. Balbi, and J. Silk, [[*Astronom and Astrophys.*]{} [**367**]{}, 1 (2001)]{}
C.-P. Ma and J. N. Fry, [ [*Phys. Rev. Letters*]{} [**88**]{}, 211301 (2002) ]{}
P. Zhang, U-L. Pen, and H. Trac, [[astro-ph/]{}[0304534]{}]{}
N. Aghanim, S. Prunet, O. Forni, and F. R. Bouchet, [[*Astronom and Astrophys.*]{} [**334**]{}, 409 (1998)]{}
A. Cooray, [ [*Phys. Rev. D*]{} [**65**]{}, 083518 (2002) ]{}
A. Cooray and X. Chen, [ [*Astrophys. J.*]{} [**573**]{}, 43 (2002) ]{}
J. Chluba and K. Mannheim, [[astro-ph/]{}[0208392]{}]{}
C. Keeton, private communication
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'A.S.Holevo\'
title: Sending quantum information with Gaussian states
---
Information characteristics of a quantum channel
================================================
During the last couple of years an impressive progress has been achieved in the theory of transmission of classical information through quantum communication channels (see [@Hol] for a comprehensive survey). The problem of sending quantum information is much less understood; we refer in particular to the papers [@Lin], [@Sch], [@Cer], initiating the study of this problem, where the reader can find further references. In this paper we make a contribution to this study by considering rather concrete situation: sending Gaussian (quasifree) states through linear Bosonic channels.
Consider quantum system in a Hilbert space ${\cal H}$, with a fixed density operator $\rho .$ A[* channel* ]{}is a transformation of quantum states as presented by density operators, given by the relation
$$T[\rho ]=\sum_{j}A_{j}\rho A_{j}^{*},$$ where $A_{j}$ are bounded operators in ${\cal H}$ satisfying $\sum_{j}A_{j}^{*}A_{j}=I.$ Let us denote $H(\rho )=-{\rm Tr}\rho \log \,\rho
$ the von Neumann entropy of a density operator $\rho .$ We call $\rho $ the input state, and $T[\rho ]$ the output state of the channel. There are three important entropy quantities related to the couple $(\rho ,T):$
1\) The entropy of the input state $H(\rho );$
2\) The entropy of the output state $H(T[\rho ]);$
3\) The entropy exchange $H(\rho ,T).$
While the definition and the meaning of the first two entropies is obvious, the third quantity is somewhat more sophisticated. To define it one introduces the[* reference system*]{}, described by the Hilbert space ${\cal H}_{R},$ isomorphic to the Hilbert space ${\cal H}_{Q}=$ ${\cal H}$ of the initial system. Then according to [@Lin], [@Sch]$,$there exists [*purification*]{} of the state $\rho ,$ i.e. a unit vector $|\psi \rangle
\in {\cal H}_{Q}\otimes {\cal H}_{R}$ such that
$$\rho ={\rm Tr}_{R}|\psi \rangle \langle \psi |.$$ The [*entropy exchange*]{} is then defined as
$$H(\rho ,T)=H((T\otimes Id)[|\psi \rangle \langle \psi |]),$$ that is as the entropy of the output state of the dilated channel $(T\otimes
Id)$ applied to the input which is purification of the input state $\rho .$ One can then show that $H(\rho ,T)$ is equal to the entropy increase in the channel environment $E$ provided the channel is represented by a unitary interaction with the environment system being initially in a pure state [@Lin], [@Sch]$.$
From these three entropies one can construct three other quantities, bearing some analogy with the classical mutual information. In general, if $\rho
_{12}$ is a density operator in a tensor product Hilbert space ${\cal H}_{1}\otimes {\cal H}_{2},$ and $\rho _{1},\rho _{2}$ are the partial states of $\rho _{12},$resp., in ${\cal H}_{1},{\cal H}_{2},$then one can introduce the quantity $$C_{12}=H(\rho _{1})+H(\rho _{2})-H(\rho _{12}).$$ This quantity is nonnegative by subadditivity of quantum entropy, and to certain extent reflects information correlation between the two systems [@Lin], although its operational meaning is still not completely clarified. Then from the above three entropies one can construct three quantum information quantities [@Lin], [@Cer]:
1\) [*Quantum mutual information*]{} $$I=H(\rho )+H(T[\rho ])-H(\rho ,T),$$ reflecting quantum information transfer from the reference system $R$ to the output of the initial system $Q^{\prime }.$ The important component of it is the [*coherent information*]{} $H(T[\rho ])-H(\rho ,T)$, supremum of which with respect to input states $\rho $ was conjectured as the quantum capacity of the channel $T$ [@Sch].
2\) [*Loss* ]{} $$L=H(\rho )+H(\rho ,T)-H(T[\rho ]),$$ which reflects quantum information transfer from the reference system $R$ to the output of the environment $E^{\prime }.$
3\) [*Noise*]{} $$N=H(T[\rho ])+H(\rho ,T)-H(\rho ),$$ which reflects quantum information transfer from the output of the environment $E^{\prime }$ to the output of the initial system $Q^{\prime }.$ In [@Cer] quantum Wenn’s diagrams were introduced to visualize the relations between the entropy and the information quantities. However, in contrast to classical case, some areas in these diagrams representing conditional entropies may have negative measure. We use another graphic representation via the [*information triangle*]{}. In this representation the entropies $H(\rho ),H(T[\rho ]),H(\rho ,T)$ are associated with the sides of the triangle, and the information quantities $I,L,N$ are attached to its vertices. The deficiency of this picture is that the representation of the information quantities is only qualitative: roughly, the bigger is the quantity - the bigger is distance from the corresponding vertex to the opposite side of the triangle, and vice versa.
Although the entropy and information quantities described above were studied in some detail from the general point of view, they are far from being completely understood, and concrete examples in which they can be explicitly evaluated are certainly welcome. In quantum statistics there is one large class of states for which many explicit calculations are possible – the so called quasifree states of canonical commutation relations, in many respect analogous to the classical Gaussian probability distributions. They are the states of the maximal entropy among all states with fixed second moments, for example, mean energy for a quadratic Hamiltonian. The aim of the present paper is the study the behavior of the information triangle for Gaussian input state and the most common attenuation/amplification channel.
Quantum Gaussian states
=======================
In this Section we repeat some results of [@Hol75], [@Hol82], [@Sohma] and give a new variant of the expression for the entropy of a general quantum Gaussian state. Let $q_{j},p_{j}$ be the canonical observables satisfying the Heisenberg CCR $$\lbrack q_{j},p_{k}]=i\delta _{jk}\hbar
I,\;\;[q_{j},q_{k}]=0,\;\;[p_{j},p_{k}]=0.$$ Let us introduce the column vector $$R=[q_{1},\dots ,q_{s};p_{1},\dots ,p_{s}]^{T}.$$ We also introduce real column $2s$-vector $z=[x_{1},\dots ,x_{s};y_{1},\dots
,y_{s}]^{T}$, and the unitary operators in ${\cal H}$ $$V(z)=\exp \,i\sum_{j=1}^{s}(x_{j}q_{j}+y_{j}p_{j})=\exp \,i\,R^{T}z.$$ The operators $V(z)$ satisfy the Weyl-Segal CCR $$V(z)V(z^{\prime })=\exp [i/2\Delta (z,z^{\prime })]V(z+z^{\prime }),
\label{weyl}$$ where $$\Delta (z,z^{\prime })=\hbar \sum_{j=1}^{s}(x_{j}^{\prime
}y_{j}-x_{j}y_{j}^{\prime })$$ is the canonical symplectic form. The Weyl -Segal CCR is the rigorous counterpart of the Heisenberg CCR, involving only bounded operators. We denote by
$$\Delta =\left[
\begin{array}{ccc}
\begin{array}{cccc}
0 & & & \\
& 0 & & \\
& & \ddots & \\
& & & 0
\end{array}
&
\begin{array}{c}
| \\
| \\
| \\
|
\end{array}
&
\begin{array}{cccc}
\hbar & & & 0 \\
& \hbar & & \\
& & \ddots & \\
0 & & & \hbar
\end{array}
\\
\begin{array}{cccc}
- & - & - & -
\end{array}
& - &
\begin{array}{cccc}
- & - & - & -
\end{array}
\\
\begin{array}{cccc}
-\hbar & & & 0 \\
& -\hbar & & \\
& & \ddots & \\
0 & & & -\hbar
\end{array}
&
\begin{array}{c}
| \\
| \\
| \\
|
\end{array}
&
\begin{array}{cccc}
0 & & & \\
& 0 & & \\
& & \ddots & \\
& & & 0
\end{array}
\end{array}
\right] \label{delta}$$
the $(2s)\times (2s)$-skew-symmetric [*commutation matrix*]{} of components of the vector $R$. Most of the results below are valid for the case where the commutation matrix is arbitrary skew-symmetric matrix, not necessarily of the canonical form (\[delta\]).
[** **]{}The density operator $\rho $ is called [*Gaussian*]{}, if its quantum characteristic function has the form $${\rm Tr}\rho V(z)=\exp (i\,m^{T}z-\frac{1}{2}z^{T}\alpha z),$$ where $m$ is column ($2s$)-vector and $\alpha $ is real symmetric $(2s)\times (2s)$-matrix.
One can show that $$m={\rm Tr}\rho R\;;\;\alpha -\frac{i}{2}\Delta ={\rm Tr}R\rho R^{T}$$ (cf[*.* ]{}[@Hol75], [@Hol82]). The [*mean*]{} $m$ can be arbitrary vector; in what follows we will be interested in the case $m=0.$ The necessary and sufficient condition on the [*correlation matrix*]{} $\alpha $ is the matrix uncertainty relation $$\alpha -\frac{i}{2}\Delta \geq 0. \label{n-s condition}$$ This condition is equivalent to its transpose $\alpha +\frac{i}{2}\Delta
\geq 0,$and to the following matrix generalization of the Heisenberg uncertainty relation $$\Delta ^{-1}\alpha \Delta ^{-1}+\frac{1}{4}\alpha ^{-1}\geq 0,
\label{heisenberg}$$ which is obtained by combining together (\[n-s condition\]) and its transpose. The state $\rho $ is pure if and only if the equality holds in this equation, or $$(\Delta ^{-1}\alpha )^{2}=-\ \frac{1}{4}I. \label{pure}$$
Let us introduce the function
$$g(x)=(x+1)\log (x+1)-x\log x,\qquad x>0.$$
We shall also use the matrix function abs$(\cdot )$, which is defined as follows: for a diagonalizable matrix $M=T{\rm diag}(m_{j})T^{-1},$ we put ${\rm abs}M=T{\rm diag}(|m_{j}|)T^{-1}.$ In [@Sohma] it was shown that the entropy of the Gaussian state is equal to $$H(\rho )=\frac{1}{2}{\rm Sp}G(-(\Delta ^{-1}\alpha )^{2}),$$ where $$G(a^{2})=(a+\frac{1}{2})\log (a+\frac{1}{2})-(a-\frac{1}{2})\log (a-\frac{1}{2}),$$ and Sp denotes trace of a matrix, as distinct from trace of operator. The matrix $\ \Delta ^{-1}\alpha $ has purely imaginary eigenvalues $\pm ia_{j}$ and is diagonalizable. Since $G(a^{2})=g(|a|-\frac{1}{2}),$ we obtain another expression
$$H(\rho )=\frac{1}{2}{\rm Sp}g({\rm abs}(\Delta ^{-1}\alpha )-\frac{I}{2}),
\label{abs}$$
which will be used in the sequel.
Purification of Gaussian states
===============================
Let us denote ${\cal H}_{Q}={\cal H}_{1}$ the Hilbert space of irreducible representation $z\rightarrow V_{1}(z)$ of the CCR (\[weyl\]), ${\cal H}_{R}={\cal H}_{2}$ the Hilbert space of irreducible representation $z\rightarrow V_{2}(z)$ of the CCR
$$V_{2}(z)V_{2}(z^{\prime })=\exp [-i/2\Delta (z,z^{\prime
})]V_{2}(z+z^{\prime }).$$ For example, $V_{2}(z)=\exp
\,i\sum_{j=1}^{s}(x_{j}p_{j}^{(2)}+y_{j}q_{j}^{(2)}),$ where $q_{j}^{(2)},p_{j}^{(2)}$ satisfy the Heisenberg CCR in ${\cal H}_{2}.$ In ${\cal H}_{1}\otimes {\cal H}_{2}$ the operators $V(z_{1},z_{2})=V_{1}(z_{1})\otimes V_{2}(z_{2})$ satisfy the CCR $$V(z_{1},z_{2})V(z_{1}^{\prime },z_{2}^{\prime })=\exp [i/2\Delta
(z_{1},z_{2};z_{1}^{\prime },z_{2}^{\prime })]V(z_{1}+z^{\prime
}{}_{1},z_{2}+z^{\prime }{}_{2}),$$ where $$\Delta (z_{1},z_{2};z_{1}^{\prime },z_{2}^{\prime })=\Delta
(z_{1},z_{1}^{\prime })-\Delta (z_{2},z_{2}^{\prime }).$$
Following [@Hol72] we introduce Gaussian state $\rho _{12}$ in ${\cal H}_{1}\otimes {\cal H}_{2}$ with the correlation matrix $$\alpha _{12}=\left[
\begin{array}{ll}
\alpha & \Delta \sqrt{-(\Delta ^{-1}\alpha )^{2}-I/4} \\
-\Delta \sqrt{-(\Delta ^{-1}\alpha )^{2}-I/4} & \alpha
\end{array}
\right] ,$$ that is $$\begin{aligned}
&&{\rm Tr}\rho _{12}V(z_{1},z_{2}) \\
&=&\exp [-\frac{1}{2}(z_{1}^{T}\alpha z_{1}+z_{2}^{T}\alpha
z_{2}+z_{1}^{T}\Delta \sqrt{-(\Delta ^{-1}\alpha )^{2}-I/4}z_{2}-z_{2}^{T}\Delta \sqrt{-(\Delta ^{-1}\alpha )^{2}-I/4}z_{1})].\end{aligned}$$ Obviously, $\rho _{1}={\rm Tr}_{2}\rho _{12}.$
Let us show that $\rho _{12}$ is pure. With $$\Delta _{12}=\left[
\begin{array}{ll}
\Delta & 0 \\
0 & -\Delta
\end{array}
\right] ,$$ we have $$\Delta _{12}^{-1}\alpha _{12}=\left[
\begin{array}{ll}
\Delta ^{-1}\alpha & \sqrt{-(\Delta ^{-1}\alpha )^{2}-I/4} \\
\sqrt{-(\Delta ^{-1}\alpha )^{2}-I/4} & -\Delta ^{-1}\alpha
\end{array}
\right] ,$$ and it is easy to check that $(\Delta _{12}^{-1}\alpha _{12})^{2}=-I/4.$ By the criterium (\[pure\]) $\rho _{12}$ is pure.
We shall be interested in the particular subclass of Gaussian states most familiar in quantum optics, which we call gauge-invariant. These are the states having the P-representation $$\rho =\pi ^{-s}|\det N|^{-1}\int \exp (-\zeta ^{\dagger }N^{-1}\zeta )|\zeta
\rangle \langle \zeta |d^{2s}\zeta \label{p-repr}$$ (see e.g. [@Hel], Sec. V, 5. II). Here $\zeta \in {\bf C}^{s}$, $|\zeta
\rangle $are the coherent vectors in ${\cal H}$, $a|\zeta \rangle =\zeta
|\zeta \rangle $, $N$ is positive Hermitian matrix such that $$N={\rm Tr}a\,\rho \,a^{\dagger }$$ (we use here vector notations, where $a=[a_{1},\dots ,a_{s}]^{T}$ is a column vector and $a^{\dagger }=[a_{1}^{\dagger },\dots ,a_{s}^{\dagger }]$ is a row vector) and $a_{j}=\frac{1}{2\hbar }(q_{j}+ip_{j}).$As shown in [@Sohma], the correlation matrix of such states is $$\alpha =\hbar \left[
\begin{array}{ll}
{\rm Re}N+I/2 & -{\rm Im}N \\
{\rm Im}N & {\rm Re}N+I/2
\end{array}
\right] ,$$
The real $2s\times 2s-$ matrices of such form can be rewritten as complex $s\times s-$ matrices, by using the correspondence $$\left[
\begin{array}{ll}
A & -B \\
B & A
\end{array}
\right] \leftrightarrow A+iB,$$ which is in fact algebraic isomorphism, provided $A^{T}=A,B^{T}=-B.$ Apparently, $$\frac{1}{2}{\rm Sp}\left[
\begin{array}{ll}
A & -B \\
B & A
\end{array}
\right] ={\rm Sp} (A+iB).$$
By using this correspondence, we have $$\alpha \leftrightarrow \hbar (N+I/2),\qquad \Delta \leftrightarrow -i\hbar
I,$$ and $$\Delta ^{-1}\alpha \leftrightarrow i(N+I/2).$$ In particular, the formula (\[abs\]) becomes $$H(\rho )={\rm Sp}g(N),$$ which is well known (see e.g. [@Sohma]) and confirms (\[abs\]).
For future use we also need the correspondence $$\Delta _{12}^{-1}\alpha _{12}\leftrightarrow \left[
\begin{array}{ll}
i(N+I/2) & \sqrt{N^{2}+N} \\
\sqrt{N^{2}+N} & -i(N+I/2)
\end{array}
\right] . \label{d-1a}$$ For the case of one degree of freedom we shall be interested in the following Section, $N$ is just nonnegative number and $\rho $ is [*elementary*]{} Gaussian state with the characteristic function $$\exp \left[ -\frac{\hbar }{2}\left( N+\frac{1}{2}\right) |z|^{2}\right] ,
\label{one mode charc. fct. }$$ where we put $|z|^{2}=(x^{2}+y^{2}).$
Attenuation/amplification channel
=================================
Let us consider CCR with one degree of freedom described by one mode annihilation operator $a=\frac{1}{2\hbar }(q+ip),$ and let $a_{0}$ be another mode in the Hilbert space ${\cal H}_{0}={\cal H}_{E}$ describing environment. Let the environment be initially in the vacuum state, which is described by the characteristic function (\[one mode charc. fct. \]) with $N=0$ i.e. $\exp [-\frac{\hbar }{4}|z|^{2}],$
The linear attenuator with coefficient $k<1$ is described by the transformation $$a\rightarrow ka+\sqrt{1-k^{2}}a_{0}$$ in the Heisenberg picture. Similarly, the linear amplifier with coefficient $k>1$ is described by the transformation $$a\rightarrow ka+\sqrt{k^{2}-1}a_{0}.$$ It follows that the corresponding transformations $T_{k}[\rho ]$ of states in the Schroedinger picture have, correspondingly, the characteristic functions $${\rm Tr}T_{k}[\rho ]V(z)={\rm Tr}\rho V(kz)\exp [-\frac{\hbar }{4}(1-k^{2})|z|^{2}],\qquad k<1, \label{atten}$$ $${\rm Tr}T_{k}[\rho ]V(z)={\rm Tr}\rho V(kz)\exp [-\frac{\hbar }{4}(k^{2}-1)|z|^{2}],\qquad k>1, \label{amplif}$$ see [@Hol72a]. Let the input state $\rho $ of the system have the characteristic function (\[one mode charc. fct. \]), i.e. $\exp [-\frac{\hbar }{4}(N+\frac{1}{2})|z|^{2}].$ The entropy of $\rho $ is $$H(\rho )=g(N).$$
From (\[atten\]), (\[amplif\]) we find that the output state $T_{k}[\rho
]$ is again elementary Gaussian with $N$ replaced by
$$\widetilde{N}=k^{2}N,\qquad k<1;\qquad \widetilde{N}=k^{2}N+(k^{2}-1),\qquad
k>1.$$ Thus $$H(T_{k}[\rho ])=g(k^{2}N),\qquad k<1;\qquad H(T_{k}[\rho
])=g(k^{2}N+(k^{2}-1)),\qquad k>1.$$
Now we calculate the entropy exchange $H(\rho ,T_{k}).$ The (pure) input state $\rho _{12}$ of the extended system ${\cal H}_{1}\otimes {\cal H}_{2}$ is characterized by the $2\times 2-$matrix (\[d-1a\]). The action of the extended channel $(T\otimes Id)$ transforms this matrix into $$\Delta _{12}^{-1}\widetilde{\alpha }_{12}\leftrightarrow \left[
\begin{array}{ll}
i(\widetilde{N}+\frac{1}{2}) & k\sqrt{N^{2}+N} \\
k\sqrt{N^{2}+N} & -i(N+\frac{1}{2})
\end{array}
\right] .$$ From formula (\[abs\]) we deduce $H(\rho ,T_{k})=g(|\lambda _{1}|-\frac{1}{2})+g(|\lambda _{2}|-\frac{1}{2}),$ where $\lambda _{1},\lambda _{2}$ are the eigenvalues of the complex matrix in the right-hand side. The eigenvalues are: $\lambda _{1}=\frac{i}{2},$$$\lambda _{2}=-i[(1-k^{2})N+\frac{1}{2}],\qquad k<1;\qquad i[(k^{2}-1)(N+1)+\frac{1}{2}],\qquad k>1.$$ Therefore we obtain $$H(\rho ,T_{k})=g((1-k^{2})N)\qquad k<1;\qquad g((k^{2}-1)(N+1)),\qquad k>1.$$
The behavior of the entropies $H(T_{k}[\rho ]),H(\rho ,T_{k})$ as functions of $k$ is clear from Fig.1 . In particular, for all $N$ the coherent information $H(T_{k}[\rho ])-H(\rho ,T_{k})$ turns out to be positive for $k>1/\sqrt{2}$ and negative otherwise. It tends to $-H(\rho )$ for $k\rightarrow 0,$ is equal to $H(\rho )$ for $k=1,$and quickly tends to zero as $k\rightarrow \infty $ (Fig.2; on both plots $N=1$). The behavior of the information triangle shows that loss dominates for $k\rightarrow 0,$ mutual information for $k\div 1,$ while the noise – as $k\rightarrow \infty .$ This agrees with what one should expect on physical grounds from quantities presenting quantum mutual information, loss and noise and gives further support for their use in quantum information theory. However, negativity of the coherent information for $k<1/\sqrt{2}$ looks somewhat mysterious and waits for a physical explanation.
Acknowledgments. This work was initiated when the author was visiting Physical Department of the University of Milan under the contract with Italian Ministry of Foreign Affairs provided by A. Volta Center for Scientific Culture. The author is grateful to Prof. L. Lanz for his hospitality and stimulating discussions. Part of this work was completed during the 1998 Elsag-Bailey – I.S.I. Foundation research meeting on quantum computation.
[99]{} C. Adami and N. J. Cerf,“Capacity of noisy quantum channels”, , vol. [**A56**]{}, pp. 3470-3485, 1972. H. Barnum, M. A. Nielsen, B. Schumacher, “Information transmission through noisy quantum channels”,LANL Report no. quant-ph/9702049, Feb. 1997.
C. W. Helstrom, , chapter 5, Academic press, 1976. A. S. Holevo,“Generalized free states of the C$^{*}$-algebra of the CCR. ”, , vol. [**6**]{}, no.1, pp. 3-20, 1971. A. S. Holevo,“Towards the mathematical theory of quantum communication channels”, , vol. [**8**]{}, no.1, pp. 63-71, 1972. A. S. Holevo, “Some statistical problems for quantum Gaussian states”, , vol. [**IT-21**]{}, no.5, pp. 533-543, 1975. A. S. Holevo, , chapter 5, North-Holland, 1982. A. S. Holevo, “Coding theorems for Quantum Channels”, , No.4, 1998. A. S. Holevo, M. Sohma and O. Hirota, “The capacity of quantum Gaussian channels ”, Preprint 1998.
G. Lindblad, “Quantum entropy and quantum measurements”, , vol. [**378**]{}, Quantum Aspects of Optical Communication, Ed. by C. Benjaballah, O. Hirota, S. Reynaud, pp.71-80, 1991.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The purpose of this short article is to build on the work of Ghirardi, Marinatto and Weber (Ghirardi, Marinatto & Weber 2002; Ghirardi & Marinatto 2003, 2004, 2005) and Ladyman, Linnebo and Bigaj (2013), in supporting a redefinition of entanglement for “indistinguishable” systems, particularly fermions. According to the proposal, non-separability of the joint state is insufficient for entanglement. The redefinition is justified by its physical significance, as enshrined in three biconditionals whose analogues hold of “distinguishable” systems.'
author:
- |
Adam Caulton\
[email protected]
title: 'Physical Entanglement in Permutation-Invariant Quantum Mechanics'
---
Introduction
============
In this article, I wish to give support for a redefinition of ‘entanglement’ for “indistinguishable” systems; i.e. systems for which permutation invariance is imposed. This redefinition was first proposed by Ghirardi, Marinatto and Weber (Ghirardi, Marinatto & Weber 2002; Ghirardi & Marinatto 2003, 2004, 2005), and has recently been endorsed by Ladyman, Linnebo and Bigaj (2013). My contribution here will be to prove that the proposed redefinition enjoys a physical significance that is not shared by the standard concept, according to which a joint state of a quantum assembly is entangled iff it is non-separable; i.e. inexpressible as a product state.
The physical significance of the concept, which I here call *GMW-entanglement*, is enshrined in three biconditionals the analogues of which hold for the standard concept of entanglement for “distinguishable” systems; i.e. systems for which permutation invariance is not imposed. These three biconditionals are:
1. The joint state of any two-system assembly is entangled iff it violates a Bell inequality.
2. The joint state of any assembly is not entangled iff the constituent systems’ states are pure.
3. The joint state of any assembly is not entangled iff the constituent systems’ states determine the joint state.
Each of the three biconditionals can be construed in two way: (i) as about the standard notion of entanglement, as applied to “distinguishable” quantum systems (i.e. for which permutation-invariance is not imposed); and (ii) as about GMW-entanglement, as applied to “indistinguishable” quantum systems. The biconditionals under (i) are well-known (the first is a Theorem due to Gisin 1991); under (ii) they are not known.
Proving the first biconditional under construal (ii) is the main work of this paper. It will be crucial to this proof that a couple of other concepts be understood rather differently, in a permutation-invariant setting, than in the usual setting. In particular, we need to revise our understanding of what counts as a *local* operation and how to extract the states of constituent systems from the joint state. The revision of both of these concepts is necessary for the following reason: in the “distinguishable” case, these concepts make essential appeal to the factor Hilbert spaces that make up the assembly’s joint Hilbert space; and our best understanding of permutation invariance is one in which factor Hilbert space indices—or, equivalently, the order in which they stand in the tensor product—have no physical meaning. The second and third biconditionals will drop out of a proper revision of these concepts.
In Section \[PIinQM\], I briefly review the topic of permutation invariance in quantum mechanics, and argue that its best interpretation is one that treats the invariance as reflecting by a representational redundancy in the standard quantum formalism. It is the fact of this redundancy which motivates the revisions in the concepts of local operation, constituents’ states and entanglement. In Section \[WhatIsE\], Gisin’s Theorem and GMW-entanglement are both reviewed, and some confusions cleared up. Section \[BellIn\] contains the proposed redefinitions of local operations, constituents’ states and entanglement, and proofs of the three biconditionals.
Permutation invariance, symmetric operators and the wedge product {#PIinQM}
=================================================================
Permutation-invariant quantum mechanics is standard quantum mechanics with the additional condition of permutation invariance. We begin with the single-system Hilbert space $\mathcal{H}$ equipped with an algebra of quantities. From this we define the $N$-fold tensor product $\otimes^N \mathcal{H}$, the *prima facie* state space for $N$ “indistinguishable” systems (their indistinguishability is expressed by the fact that any two factor Hilbert spaces are unitarily equivalent). The joint Hilbert space $\otimes^N \mathcal{H}$ carries a natural unitary representation $U:S_N\to\mathcal{U}(\otimes^N \mathcal{H})$ of the group $S_N$ of permutations on $N$ symbols. For example, the permutation $(ij)$, which swaps systems $i$ and $j$, is represented by the unitary operator $U(ij)$ defined on basis states (having chosen an orthonormal basis $\{|\phi_k\rangle\}$ on $\mathcal{H}$) by $$\begin{aligned}
&& U(ij)|\phi_{k_1}\rangle\otimes\ldots\otimes|\phi_{k_i}\rangle\otimes\ldots\otimes|\phi_{k_j}\rangle\otimes\ldots\otimes|\phi_{k_N}\rangle
\nonumber\\
&& \qquad\qquad\qquad\qquad\qquad =
|\phi_{k_1}\rangle\otimes\ldots\otimes|\phi_{k_j}\rangle\otimes\ldots\otimes|\phi_{k_i}\rangle\otimes\ldots\otimes|\phi_{k_N}\rangle\end{aligned}$$ and then extended by linearity.
Permutation invariance, otherwise known as the *Indistinguishability Postulate* (Messiah & Greenberg 1964, French & Krause 2006), is the condition on any operator $Q\in\mathcal{B}(\otimes^N \mathcal{H})$, which is to represent a legitimate physical quantity, that it be *symmetric*;[^1] i.e. for all permutations $\pi\in S_N$ and all states $|\psi\rangle\in\otimes^N \mathcal{H}$, $$\langle\psi|U^\dag(\pi)QU(\pi)|\psi\rangle = \langle\psi|Q|\psi\rangle$$ The representation $U$ is reducible, and decomposes into various irreducible representations, each irreducible representation corresponding to a different *symmetry type*; namely bosonic states, fermionic states and (if $N\geqslant 3$) a variety of paraparticle states (see e.g. Tung 1985, Ch. 5). If we consider only the information provided by the symmetric operators, we treat permutation invariance as a superselection rule, and each superselection sector corresponds to one of these symmetry types.
What does it mean to “impose” permutation invariance? Isn’t it rather that permutation invariance holds of some operators and not others? I propose to impose permutation invariance means to lay it down as a necessary condition on any operator’s receiving a physical interpretation. This justifies, and is justified by, treating the factor Hilbert space labels—i.e. the order in which single-system operators and states lie in the tensor product—as nothing but an artefact of the mathematical formalism of quantum mechanics.
What is the justification for interpreting the factor Hilbert space labels in this way? The ultimate justification is of course that it leads to an empirically adequate theory. It is an empirical fact that elementary particles exhibit statistics consistent with their being either bosons or fermions. But this fact is logically weaker than the claim that factor Hilbert space labels represent nothing. It *could* be that factor Hilbert space labels represent (or name), for example, the constituent systems, and that the joint state of any assembly of elementary particles remains in the fermionic or bosonic sector under all actual physical evolutions due only to the fact that the corresponding Hamiltonian happens to be permutation-invariant. Indeed, this interpretative gloss is either explicitly propounded or implicitly assumed by most authors in the literature (e.g. French & Redhead 1988; Butterfield 1993; Huggett 1999, 2003; French & Krause 2006; Muller & Saunders 2008; Muller & Seevnick 2009; Caulton 2013). However, it may be argued that the physical emptiness of the factor Hilbert space labels offers the *best explanation* of the empirical fact that permutation invariance seems always to hold true. This suggestion is in line with a more general interpretative stance in physics: that any exact symmetry is a symptom of representational redundancy in the corresponding theory’s formalism.
The focus of this paper is fermionic states and their compositional structure. Picking some orthonormal basis $\{|\phi_i\rangle\}$ in $\mathcal{H}$, these states are spanned by states of the form $$\frac{1}{\sqrt{N!}}\sum_{\pi\in S_N}(-1)^{\deg\pi}|\phi_{i_{\pi(1)}}\rangle\otimes|\phi_{i_{\pi(2)}}\rangle\otimes\ldots \otimes|\phi_{i_{\pi(N)}}\rangle$$ and carry the alternating irreducible representation of $S_N$; i.e. any permutation $\pi$ is represented by multiplication by $(-1)^{\deg\pi}$, where $\deg\pi$ is the *degree* of the permutation $\pi$ (i.e. the number of pairwise swaps into which $\pi$ may be decomposed).
Following Ladyman, Linnebo and Bigaj (2013), we may use the mathematical apparatus of *Grassmann* or *exterior algebras* to represent fermionic states. The exterior algebra $\Lambda(V)$ over the vector space $V$ (over the field of complex numbers $\mathbb{C}$) is obtained by quotienting the tensor algebra $T(V) := \bigoplus_{k=0}^{\infty} T^k(V) = \mathbb{C}\oplus V \oplus (V\otimes V) \oplus (V\otimes V\otimes V) \oplus \ldots$ with the equivalence relation $\sim$ defined so that $\alpha\sim\beta$ iff $\alpha$ and $\beta$ have the same anti-symmetrization;[^2] i.e. $$\Lambda(V) := T(V)/\sim\ .$$ For example, $[x\otimes y] = [-y\otimes x]$ and $[x \otimes x] = [\mathbf{0}]$. We may set $V=\mathcal{H}$, then there is a natural isomorphism $\iota$ from the elements of $\Lambda(\mathcal{H})$ onto the vectors of the fermionic Fock space $\mathcal{F_-(H)} := \bigoplus_{N=0}^{\dim\mathcal{H}} \mathcal{A}(\otimes^N\mathcal{H})$. $\iota$ simply takes any $\sim$-equivalence class of degree-$r$ vectors of $T^r(\mathcal{H})$ to the anti-symmetric degree-$r$ vector in $\mathcal{A}(\otimes^r\mathcal{H})$ that is their common anti-symmetrization. Therefore we may pick out any $N$-fermion state in $\mathcal{A}(\otimes^N\mathcal{H})$ by specifying its pre-image under $\iota$ in $\Lambda^N(\mathcal{H})$ (i.e. the subalgebra of $\Lambda(\mathcal{H})$ containing only degree-$N$ vectors).
Elements of $\Lambda(V)$ are called *decomposable* iff they are equivalence classes $[x_{i_1}\otimes x_{i_2}\otimes \ldots\otimes x_{i_r}]$ containing product vectors. (Not all elements are decomposable.) To anticipate, the decomposable elements of $\Lambda^N(\mathcal{H})$ correspond to states of $\mathcal{A}(\otimes^N\mathcal{H})$ that are non-GMW-entangled.
The product on the exterior algebra is the *exterior* or *wedge product* $\wedge$, defined by its action on decomposable elements as follows: $$\label{wedgeprod}
[x_{i_1}\otimes x_{i_2}\otimes \ldots\otimes x_{i_r}] \wedge [x_{i_{r+1}}\otimes x_{i_{r+2}}\otimes \ldots\otimes x_{i_{r+s}}] = [x_{i_1}\otimes x_{i_2}\otimes \ldots\otimes x_{i_{r+s}}] \ ,$$ where $\{x_1,x_2, \ldots x_{\dim V}\}$ is an orthonormal basis for $V$ and each $i_k\in \{1,2,\ldots, \dim V\}$. We then extend the definition of $\wedge$ to non-decomposable elements by bilinearity. (Note that if there is a pair ${i_j} = i_k$ for $j\neq k$, then the righthand side of (\[wedgeprod\]) is $[\mathbf{0}]$.) For any $\alpha\in\Lambda^r(V)$ and any $\beta\in\Lambda^s(V)$, $
\alpha \wedge \beta = (-1)^{rs}\beta\wedge\alpha \in \Lambda^{r+s}(V)
$.
In the following, I will, like Ladyman, Linnebo and Bigaj (2013), make use of a harmless abuse of notation by referring to anti-symmetric states by their corresponding wedge product. In particular, given an orthonormal basis $\{|\phi_i\rangle\}$ on $\mathcal{H}$, $$|\phi_{i_1}\rangle\wedge|\phi_{i_2}\rangle\wedge\ldots \wedge|\phi_{i_N}\rangle$$ will be used as a shorthand for $$\frac{1}{\sqrt{N!}}\sum_{\pi\in S_N}(-1)^{\deg\pi}|\phi_{i_{\pi(1)}}\rangle\otimes|\phi_{i_{\pi(2)}}\rangle\otimes\ldots \otimes|\phi_{i_{\pi(N)}}\rangle\ .$$ It is common to represent an $r$-dimensional subspace of $V$ by a wedge product of $r$ degree-1 vectors (i.e. vectors in $V$). Correspondingly, joint states of $r$ fermions which correspond to decomposable degree-$r$ vectors—i.e. joint states which are non-GMW-entangled—may be aptly (that is: completely and non-redundantly) represented by $r$-dimensional subspaces of $\mathcal{H}$. I return to this point at the end of Section \[BellIn\].
What is entanglement? {#WhatIsE}
=====================
Entanglement is standardly defined formally as the non-separability of the assembly’s state; i.e. a state is entangled iff it cannot be written as a product state (see e.g. Nielsen & Chuang 2010, 96). The physical significance of this definition is underpinned by a biconditional, one half of which is Gisin’s Theorem, which applies to assemblies of two (“distinguishable”) subsystems:
Let $|\psi\rangle \in\mathcal{H}_1\otimes\mathcal{H}_2$. If $|\psi\rangle$ is entangled (i.e. $|\psi\rangle$ is not a product state), then $|\psi\rangle$ violates a Bell inequality. That is, there is some state $|\chi\rangle \in \mathfrak{h}_1\otimes\mathfrak{h}_2$, where $\mathfrak{h}_1\leqslant \mathcal{H}_1, \mathfrak{h}_2\leqslant \mathcal{H}_2$ and $\dim\mathfrak{h}_1 = \dim\mathfrak{h}_2$, accessible from $|\psi\rangle$ by a local operation, and a triplet of $2\times 2$ matrices $\bm\sigma^{(1)} = (\sigma^{(1)}_x, \sigma^{(1)}_y, \sigma^{(1)}_z)$ on $\mathcal{H}_1$ and a triplet of $2\times 2$ matrices $\bm\sigma^{(2)} = (\sigma^{(2)}_x, \sigma^{(2)}_y, \sigma^{(2)}_z),$ on $\mathcal{H}_2$, each satisfying $$[\sigma^{(i)}_a,\sigma^{(i)}_b] = 2i\epsilon_{abc}\sigma^{(i)}_c \ ,\quad
\{\sigma^{(i)}_a,\sigma^{(i)}_b\} = 2\delta_{ab}\ ,$$ and four 3-vectors $\mathbf{a}, \mathbf{a}', \mathbf{b}, \mathbf{b}'$ such that $$\mathcal{I}:=|E(\mathbf{a},\mathbf{b}) - E(\mathbf{a},\mathbf{b}')| + |E(\mathbf{a}',\mathbf{b}) + E(\mathbf{a}',\mathbf{b}')| > 2\ ,$$ where$$E(\mathbf{a},\mathbf{b}) := \langle\chi|\mathbf{a}.\bm\sigma^{(1)}\otimes\mathbf{b}.\bm\sigma^{(2)}|\chi\rangle\ ,$$ etc.
*Proof* See Gisin (1991). $\Box$
So for the joint Hilbert space $\mathcal{H}_1\otimes\mathcal{H}_2$ to contain any entangled states, we must have $\dim\mathcal{H}_1, \dim\mathcal{H}_2\geqslant 2$. The other half of the biconditional is the “easy half”:
Let $|\psi\rangle \in\mathcal{H}_1\otimes\mathcal{H}_2$. If $|\psi\rangle$ is not entangled (i.e. $|\psi\rangle$ is expressible as a product state), then $|\psi\rangle$ satisfies any Bell inequality; that is, for any state $|\chi\rangle$ accessible from $|\psi\rangle$ by a local operation, and any triplet of $2\times 2$ matrices $\bm\sigma^{(1)} = (\sigma^{(1)}_x, \sigma^{(1)}_y, \sigma^{(1)}_z)$ on $\mathcal{H}_1$ and any triplet of $2\times 2$ matrices $\bm\sigma^{(2)} = (\sigma^{(2)}_x, \sigma^{(2)}_y, \sigma^{(2)}_z),$ on $\mathcal{H}_2$, each satisfying $$[\sigma^{(i)}_a,\sigma^{(i)}_b] = 2i\epsilon_{abc}\sigma^{(i)}_c \ ,\quad
\{\sigma^{(i)}_a,\sigma^{(i)}_b\} = 2\delta_{ab}\ ,$$ and any four 3-vectors $\mathbf{a}, \mathbf{a}', \mathbf{b}, \mathbf{b}'$, then $$\mathcal{I}:=|E(\mathbf{a},\mathbf{b}) - E(\mathbf{a},\mathbf{b}')| + |E(\mathbf{a}',\mathbf{b}) + E(\mathbf{a}',\mathbf{b}')| \leqslant 2\ .$$
*Proof.* Since $|\psi\rangle$ is non-entangled, then it has the form $$|\psi\rangle = |\phi\rangle\otimes|\theta\rangle$$ for some $|\phi\rangle$ and $|\theta\rangle$. Any state accessible from $|\psi\rangle$ by a local operation also has this form, so we proceed with $|\chi\rangle=|\psi\rangle$. Any $E(\mathbf{a}, \mathbf{b})$ then takes the form $$\begin{aligned}
E(\mathbf{a}, \mathbf{b}) &:=& \langle\psi|\mathbf{a}.\bm\sigma^{(1)}\otimes\mathbf{b}.\bm\sigma^{(2)}|\psi\rangle \\
&=& \langle\phi|\mathbf{a}.\bm\sigma^{(1)}|\phi\rangle\langle\theta|\mathbf{b}.\bm\sigma^{(2)}|\theta\rangle \\
&=:& \alpha\beta\end{aligned}$$ where $\alpha:=\langle\phi|\mathbf{a}.\bm\sigma^{(1)}|\phi\rangle$ and $\beta:=\langle\theta|\mathbf{b}.\bm\sigma^{(2)}|\theta\rangle$. If we similarly define $\alpha', \beta'$, then $$\begin{aligned}
\mathcal{I} &=&|\alpha(\beta - \beta')| + |\alpha'(\beta+\beta')|\ ,\end{aligned}$$ and since $|\alpha|, |\alpha'|,|\beta|, |\beta'|\leqslant 1$, there is no set of values for which $\mathcal{I}$ exceeds 2. $\Box$
Let $|\psi\rangle \in\mathcal{H}_1\otimes\mathcal{H}_2$. $|\psi\rangle$ is entangled iff it violates a Bell inequality.
This biconditional gives entanglement physical meaning, since the Bell inequalities represent physically realisable results—at least *so long as* we assume that any bounded self-adjoint operator represents a physical quantity.
However, when we turn to permutation-invariant quantum mechanics, the significance of this biconditional should be doubted. Permutation invariance puts restrictions on the available algebra of quantities for the joint system, and some of those prohibited quantities are involved in the definition of the correlation functions $E(\mathbf{a}, \mathbf{b})$. In a permutation-invariant setting, $\mathcal{H}_1 = \mathcal{H}_2$ and the only symmetric correlation functions are of the form $$E(\mathbf{a}, \mathbf{a}) := \langle\psi|\mathbf{a}.\bm\sigma\otimes\mathbf{a}.\bm\sigma|\psi\rangle\ .$$ Yet the Bell inequality requires us to independently vary the quantities on each system. Therefore, under permutation invariance the usual Bell inequality cannot even be constructed.
Two responses are available to us, only one of which is normally taken. The common response is to refrain from the interpretative strategy endorsed in Section \[PIinQM\], and to lift the restriction on the joint algebra placed by permutation invariance. Permutation-invariance is then construed as nothing more than a “dynamical inaccessibility”: the prohibited quantitiess still have physical meaning; it is just that dynamical evolution under them is unavailable to the joint system. Any proponent of this response may still want to say that the biconditional linking entanglement to the violation of a Bell inequality can be taken seriously, and that therefore non-separability is the right definition of entanglement.
However, as I argued in Section \[PIinQM\], we ought to take a stronger reading of permutation invariance. Under this reading, any element in the mathematical formalism that is not invariant under arbitrary permutation should not be given a physical interpretation. In that case, the non-symmetric quantities used in the definition of the correlation functions simply cannot be given any physical meaning. In that case, we must renounce the idea that non-separability provides a physically adequate definition of entanglement.
These doubts have been expressed by Ghirardi, Marinatto and Weber in a series of papers (Ghirardi, Marinatto & Weber 2002; Ghirardi & Marinatto 2003, 2004, 2005). They propose an alternative definition of entanglement, which have called *GMW-entanglement*. Although not their explicit definition, it turns out to be equivalent to following:
A joint state is *GMW-entangled* iff it is not the anti-symmetrization of a product state.
So, for example, the spin-singlet state $|\!\uparrow\rangle\wedge|\!\downarrow\rangle = \frac{1}{\sqrt{2}}(|\!\uparrow\rangle\otimes|\!\downarrow\rangle - |\!\downarrow\rangle\otimes|\!\uparrow\rangle)$ counts as non-GMW-entangled.
It may come as a surprise that a state which we have all learned to think of as *maximally* entangled—indeed, the state most commonly used to illustrate the violation of a Bell inequality—should come out as *non*-entangled on *any* reasonable definition. But there need be no confusion here. The singlet state is indeed entangled, *so long as* we have access to the full algebra of bounded operators. If we do not, as in the case of permutation invariance, then that attribution needs to be revised.
But aren’t *electrons*, which are fermions, and therefore subject to permutation invariance, involved in physical violations of the Bell inequality? And don’t those violations arise in particular when the electrons are prepared in the singlet state? The answer to both these questions is Yes, but we need to be careful about all of the electrons’ degrees of freedom. As Ghirardi, Marinatto & Weber (2003, 3) and Ladyman, Linnebo & Bigaj (2013, 216) point out, the full state in the standard EPRB experiment is $$\label{RealEPR}
\frac{1}{2}\left(|L\rangle_1\otimes|R\rangle_2 + |R\rangle_1\otimes|L\rangle_2\right)\otimes\left(|\!\uparrow\rangle_1\otimes|\!\downarrow\rangle_2 - |\!\downarrow\rangle_1\otimes|\!\uparrow\rangle_2\right)\ ,$$ where $|L\rangle$ and $|R\rangle$ represent spatial wavefunctions concentrated at the left-hand and right-hand sides of the lab respectively. Written using the wedge product, this state is $$\label{RealEPR2}
\frac{1}{\sqrt{2}}\left(|L,\uparrow\rangle\wedge|R,\downarrow\rangle - |L,\downarrow\rangle\wedge|R,\uparrow\rangle \right)\ ,$$ which is manifestly *not* expressible as the anti-symmetrization of a product state. So it counts as GMW-entangled.
But we still lack some way of making sense of Bell inequality violation under permutation invariance—one that agrees with the prevailing belief that state (\[RealEPR\]) violates a Bell inequality. It is the purpose of the next Section to do that.
Bell inequalities, local operations and constituent states under permutation invariance {#BellIn}
=======================================================================================
In order to define a Bell inequality in a permutation-invariant setting, we need some way of picking out the subsystems that is permutation-invariant—in particular, we may not appeal to the factor Hilbert space labels. Our solution, inspired by Ghirardi, Marinatto & Weber (2002) and Dieks & Lubberdink (2011), is to appeal to the *states* of the subsystems. This may be seen as the quantum analogue of Russell’s (1905) strategy of picking out objects with a property that is uniquely satisfied.
I illustrate the strategy for the case $N=2$; its generalisation to $N>2$ will be obvious. The quantum analogue of a 1-place formula is a projector that acts on the single-system Hilbert space $\mathcal{H}$. So to pick out two subsystems we select two projectors $E_1, E_2$ on $\mathcal{H}$ such that $E_1\perp E_2$, i.e. $E_1E_2 = E_2E_1 = 0$; I call this condition *orthogonality*. The orthogonality of the projectors is crucial, since it is necessary and sufficient to ensure that, for any joint state, the two projectors do not select the same subsystem.
However, there is still the danger that one of the projectors, $E_1$ say, will pick out both subsystems. Since the subsystems are fermions, we can rely on Pauli exclusion to protect us from this if we impose $\dim E_1 = \dim E_2 = 1$. However, this condition is far too strong, since it will select subsystems only in the corresponding pure states, and we want to allow the subsystems to occupy mixed states. (In fact GMW-entangled states are precisely those for which we can ascribe the subsystems mixed states; see Caulton ms.)
It is sufficient to demand that the joint state $|\psi\rangle$ be an eigenstate of the projector $$\label{IndProj}
E_1\otimes E_2 + E_2\otimes E_1$$ with eigenvalue 1; I call this condition *exhaustion*. Note that this projector is permutation-invariant. I propose that we interpret it as picking out those joint states in which one subsystem is in a state in ran$(E_1)$ and the other is in a state in ran$(E_2)$. But it must be emphasized that (\[IndProj\]) should *not* be interpreted as the quantum disjunction, ‘Subsystem 1 is in a state in ran$(E_1)$ and subsystem 2 is in a state in ran$(E_2)$ QOR subsystem 1 is in a state in ran$(E_2)$ and subsystem 2 is in a state in ran$(E_1)$.’ The individual disjuncts of this proposition are not permutation-invariant and so have no physical interpretation. Rather, we must interpret (\[IndProj\]) *primitively* as the proposition ‘Exactly one of the subsystems is in a state in ran$(E_1)$ and exactly one of the subsystems is in a state in ran$(E_2)$.’ Interpreting the projector primitively in this way (i.e. not as a disjunction) is supported by the following fact: if $\dim E_1 = \dim E_2 = 1$, then (\[IndProj\]) projects onto a single ray in the fermionic Hilbert space, and so could not be a non-trivial disjunction of other propositions.
Therefore our two conditions on what we might call *individuating projectors* $E_1$ and $E_2$ are that they be: (i) orthogonal; and (ii) exhaustive. A pair of individuating projectors can *always* be found for any given 2-fermion state. (For a proof, see Caulton ms.). The same is not true for bosonic or paraparticle states.
Once we have these individuating projectors, we can define operators associated with the corresponding subsystems. The proposal is simple: any operator $A$ on $\mathcal{H}$ is associated with the subsystem individuated by $E_i$ iff $E_iAE_i = A$. (Note that if we had demanded that $\dim E_1 = \dim E_2 = 1$, then the algebra of operators associated with each subsystem would be Abelian.) We can now define permutation-invariant operators on the joint system which act separately on each subsystem; i.e. they are the permutation-invariant analogues of $A\otimes \mathds{1}$ and $\mathds{1}\otimes B$. They have the general form $$\label{JointObs}
E_1AE_1\otimes E_2BE_2 + E_2BE_2\otimes E_1AE_1\ , \quad \mbox{where}\ A,B\in\mathcal{B(H)}\ .$$
All this leads to the following proposal for what is for a fermionic joint state $|\psi\rangle$ to violate a permutation-invariant Bell inequality:
Let $|\psi\rangle \in \mathcal{A(H\otimes H)}$. $|\psi\rangle$ violates a permutation-invariant Bell inequality iff there is some state $|\chi\rangle$, accessible from $|\psi\rangle$ by a local operation, and two projectors $E_1, E_2$ on $\mathcal{H}$, such that $E_1\perp E_2$ and $$\left(E_1\otimes E_2 + E_2\otimes E_1\right)|\chi\rangle = |\chi\rangle\ ,$$ and two triplets of $2\times 2$ matrices $\bm\sigma^{(1)} = (\sigma^{(1)}_x, \sigma^{(1)}_y, \sigma^{(1)}_z), \bm\sigma^{(2)} = (\sigma^{(2)}_x, \sigma^{(2)}_y, \sigma^{(2)}_z)$, satisfying $$[\sigma^{(i)}_a,\sigma^{(i)}_b] = 2i\epsilon_{abc}\sigma^{(i)}_c \ ,\quad
\{\sigma^{(i)}_a,\sigma^{(i)}_b\} = 2\delta_{ab}\ , \quad
E_i\sigma^{(i)}_aE_i = \sigma^{(i)}_a\ ,$$ and four 3-vectors $\mathbf{a}, \mathbf{a}', \mathbf{b}, \mathbf{b}'$ such that $$\mathcal{I}_{PI} := |F(\mathbf{a},\mathbf{b}) -F(\mathbf{a},\mathbf{b}')| + |F(\mathbf{a}', \mathbf{b}) + F(\mathbf{a}',\mathbf{b}')| > 2\ ,$$ where$$F(\mathbf{a},\mathbf{b}) := \langle\chi|\left(\mathbf{a}.\bm\sigma^{(1)}\otimes\mathbf{b}.\bm\sigma^{(2)} + \mathbf{b}.\bm\sigma^{(2)}\otimes\mathbf{a}.\bm\sigma^{(1)}\right)|\chi\rangle\ ,$$ etc.
It is important to notice that the formal explication of a “local” operation, used in the definition above, must also change under permutation-invariance. The guiding physical idea is the same for both: just as, for “distinguishable” systems, a local operation is one that acts on each subsystem—i.e. each factor Hilbert space—independently (and so has product form), so too under permutation-invariance a local operation should act on each subsystem—as individuated by $E_1$ and $E_2$—independently. So under permutation invariance a local operation is one whose form is given in (\[JointObs\]).
We are now ready to prove the biconditional linking GMW-entanglement to the violation of a permutation-invariant Bell inequality. Each direction of the biconditional will be proved separately.
Let $|\psi\rangle \in\mathcal{A}(\mathcal{H}\otimes\mathcal{H})$. If $|\psi\rangle$ is not GMW-entangled, (i.e. $|\psi\rangle$ is the anti-symmetrization of a product state) then $|\psi\rangle$ satisfies any Bell inequality for symmetric quantities. That is, for any state $|\chi\rangle$ accessible from $|\psi\rangle$ by a local operation, and any two projectors $E_1, E_2$ on $\mathcal{H}$ such that
1. $E_1\perp E_2$; and
2. $
\left(E_1\otimes E_2 + E_2\otimes E_1\right)|\chi\rangle = |\chi\rangle\ ;
$
there is no pair of triplets of $2\times 2$ matrices $\bm\sigma^{(1)} = (\sigma^{(1)}_x, \sigma^{(1)}_y, \sigma^{(1)}_z), \bm\sigma^{(2)} = (\sigma^{(2)}_x, \sigma^{(2)}_y, \sigma^{(2)}_z)$ satisfying $$[\sigma^{(i)}_a,\sigma^{(i)}_b] = 2i\epsilon_{abc}\sigma^{(i)}_c \ ,\quad
\{\sigma^{(i)}_a,\sigma^{(i)}_b\} = 2\delta_{ab}\ , \quad
E_i\sigma^{(i)}_aE_i = \sigma^{(i)}_a$$ for which, for some choice of four 3-vectors $\mathbf{a}, \mathbf{a}', \mathbf{b}, \mathbf{b}'$, $$\mathcal{I}_{PI} := |F(\mathbf{a},\mathbf{b}) -F(\mathbf{a},\mathbf{b}')| + |F(\mathbf{a}', \mathbf{b}) + F(\mathbf{a}',\mathbf{b}')| > 2\ ,$$ where$$F(\mathbf{a},\mathbf{b}) := \langle\chi|\left(\mathbf{a}.\bm\sigma^{(1)}\otimes\mathbf{b}.\bm\sigma^{(2)} + \mathbf{b}.\bm\sigma^{(2)}\otimes\mathbf{a}.\bm\sigma^{(1)}\right)|\chi\rangle\ ,$$ etc.
*Proof.* If $\dim\mathcal{H}<4$, then no pair of triplets of $2\times 2$ matrices satisfying the above conditions can be found. So we assume $\dim\mathcal{H}\geqslant 4$. Any two projectors $E_1$ and $E_2$ satisfying the above conditions must satisfy $E_1|\phi_1\rangle = |\phi_1\rangle, E_2|\phi_2\rangle = |\phi_2\rangle, E_1|\phi_2\rangle = E_2|\phi_1\rangle = 0$, where $|\psi\rangle$ can be written $$|\psi\rangle = |\phi_{1}\rangle\wedge|\phi_{2}\rangle \ .$$ Any state accessible from $|\psi\rangle$ by a local operation (in the permutation-invariant sense) also has this form, so we proceed with $|\chi\rangle=|\psi\rangle$. Since $E_1\perp E_2$ and $E_i\sigma^{(i)}_aE_i = \sigma^{(i)}_a$, $\sigma^{(i)}|\phi_j\rangle = 0$ if $i\neq j$ ($i,j\in\{1,2\}$). Therefore $$\begin{aligned}
F(\mathbf{a}, \mathbf{b}) &=&
\langle\phi_1|\mathbf{a}.\bm\sigma^{(1)}|\phi_1\rangle\langle\phi_2|\mathbf{b}.\bm\sigma^{(2)}|\phi_2\rangle \\
&=:& \alpha\beta\end{aligned}$$ where $\alpha:=\langle\phi_1|\mathbf{a}.\bm\sigma^{(1)}|\phi_1\rangle$ and $\beta:=\langle\phi_2|\mathbf{b}.\bm\sigma^{(2)}|\phi_2\rangle$. If we similarly define $\alpha', \beta'$, then $$\begin{aligned}
\mathcal{I}_{PI} &=&|\alpha(\beta - \beta')| + |\alpha'(\beta+\beta')|\ ,\end{aligned}$$ and since $|\alpha|, |\alpha'|,|\beta|, |\beta'|\leqslant 1$, there is no set of values for which $\mathcal{I}_{PI}$ exceeds 2. $\Box$
An important example of a non-GMW-entangled state is $$|L,\uparrow\rangle\wedge|R,\downarrow\rangle := \frac{1}{\sqrt{2}}(|L,\uparrow\rangle\otimes|R,\downarrow\rangle +|R,\downarrow\rangle\otimes |L,\uparrow\rangle)\ .$$ No permutation-invariant Bell inequality is violated for this state.
For the second half of the biconditional, we will need a lemma (also used by Schliemann *et al* 2001 and Ghirardi & Marinatto 2004), which is the fermionic analogue of the Schmidt bi-orthogonal decomposition theorem; I merely report it here.
\[antisymmatrix\] For any antisymmetric $d\times d$ complex matrix $A$ (i.e. $A\in\mathcal{M}(d,\mathbb{C})$ and $A^T=-A$), there exists a unitary transformation $U$ such that $A = UZU^T$, where $Z$ is a block-diagonal matrix of the form $$Z = \mbox{diag}[Z_1, \ldots Z_r, Z_0], \quad \mbox{where} \
Z_i = \left(\begin{array}{cc} 0 & c_i \\ -c_i & 0\end{array}\right) \ \mbox{and}\ c_i \in \mathbb{C}$$ and $Z_0$ is the $(d-2r)\times(d-2r)$ zero matrix.
*Proof.* See Mehta (1989).
Let $|\psi\rangle \in\mathcal{A}(\mathcal{H}\otimes\mathcal{H})$. If $|\psi\rangle$ is GMW-entangled (i.e. $|\psi\rangle$ is not the anti-symmetrization of a product state), then $|\psi\rangle$ violates a Bell inequality for symmetric quantities. That is, there is some state $|\chi\rangle$, accessible from $|\psi\rangle$ by a local operation, and two projectors $E_1, E_2$ on $\mathcal{H}$, such that $E_1\perp E_2$ and $$\left(E_1\otimes E_2 + E_2\otimes E_1\right)|\chi\rangle = |\chi\rangle\ ,$$ and two triplets of $2\times 2$ matrices $\bm\sigma^{(1)} = (\sigma^{(1)}_x, \sigma^{(1)}_y, \sigma^{(1)}_z), \bm\sigma^{(2)} = (\sigma^{(2)}_x, \sigma^{(2)}_y, \sigma^{(2)}_z)$, satisfying $$[\sigma^{(i)}_a,\sigma^{(i)}_b] = 2i\epsilon_{abc}\sigma^{(i)}_c \ ,\quad
\{\sigma^{(i)}_a,\sigma^{(i)}_b\} = 2\delta_{ab}\ , \quad
E_i\sigma^{(i)}_aE_i = \sigma^{(i)}_a\ ,$$ and four 3-vectors $\mathbf{a}, \mathbf{a}', \mathbf{b}, \mathbf{b}'$ such that $$\mathcal{I}_{PI}:=|F(\mathbf{a},\mathbf{b}) -F(\mathbf{a},\mathbf{b}')| + |F(\mathbf{a}', \mathbf{b}) + F(\mathbf{a}',\mathbf{b}')| > 2\ ,$$ where$$F(\mathbf{a},\mathbf{b}) := \langle\chi|\left(\mathbf{a}.\bm\sigma^{(1)}\otimes\mathbf{b}.\bm\sigma^{(2)} + \mathbf{b}.\bm\sigma^{(2)}\otimes\mathbf{a}.\bm\sigma^{(1)}\right)|\chi\rangle\ ,$$ etc.
*Proof.* $|\psi\rangle$ has the general form $$|\psi\rangle = \sum_{ij} a_{ij}|\theta_i\rangle\otimes|\theta_j\rangle$$ where $a_{ij} = -a_{ji}$. We can represent $|\psi\rangle$ as a complex $d\times d$ anti-symmetric matrix $A$. Any unitary transformation $U$ on $\mathcal{H}$ corresponds to the transformation $A\mapsto UAU^T$. So, given Lemma \[antisymmatrix\], we can find a basis $\{|\phi_i\rangle\}$ such that $$|\psi\rangle = \sum_{i=1}^{\llcorner \frac{d}{2}\lrcorner}c_i|\phi_{2i-1}\rangle\wedge|\phi_{2i}\rangle \ .$$ If $|\psi\rangle$ is GMW-entangled, then we can order the basis vectors so that $c_1,c_2\neq 0$. Now define $$|\chi\rangle := \frac{c_1|\phi_1\rangle\wedge|\phi_2\rangle + c_2|\phi_3\rangle\wedge|\phi_4\rangle}{\sqrt{|c_1|^2 + |c_2|^2}}\ .$$ $|\chi\rangle$ may be obtained from $|\psi\rangle$ by a local, selective operation. The idea now is to treat the state $|\chi\rangle$ analogously to the entangled state $$c_1|\phi_1\rangle\otimes|\phi_2\rangle + c_2|\phi_3\rangle\otimes|\phi_4\rangle\ ,$$ which is subject to Gisin’s Theorem. Define $$E_1 := |\phi_1\rangle\langle\phi_1| + |\phi_3\rangle\langle\phi_3|\ ,
\qquad
E_2 := |\phi_2\rangle\langle\phi_2| + |\phi_4\rangle\langle\phi_4|\ .$$ Then it may be checked that $\left(E_1\otimes E_2 + E_2\otimes E_1\right)|\chi\rangle = |\chi\rangle$. The proof now follows analogously to Gisin (1991). We define Pauli-like matrices for the factor spaces. Let $$\begin{aligned}
\sigma^{(1)}_x &:=& |\phi_1\rangle\langle\phi_3| + |\phi_3\rangle\langle\phi_1|\\
\sigma^{(1)}_y &:=& -i\left(|\phi_1\rangle\langle\phi_3| - |\phi_3\rangle\langle\phi_1|\right)\\
\sigma^{(1)}_z &:=& |\phi_1\rangle\langle\phi_1| - |\phi_3\rangle\langle\phi_3|\end{aligned}$$ and $$\begin{aligned}
\sigma^{(2)}_x &:=& |\phi_2\rangle\langle\phi_4| + |\phi_4\rangle\langle\phi_2|\\
\sigma^{(2)}_y &:=& -i\left(|\phi_2\rangle\langle\phi_4| - |\phi_4\rangle\langle\phi_2|\right)\\
\sigma^{(2)}_z &:=& |\phi_2\rangle\langle\phi_2| - |\phi_4\rangle\langle\phi_4|\end{aligned}$$ It may be checked that these operators satisfy the conditions above. Some calculation yields $$\begin{aligned}
F(\mathbf{a}, \mathbf{b}) &:=& \langle\chi|\left(\mathbf{a.\bm\sigma}^{(1)}\otimes \mathbf{b.\bm\sigma}^{(2)}+\mathbf{b.\bm\sigma}^{(2)}\otimes \mathbf{a.\bm\sigma}^{(1)} \right)|\chi\rangle\nonumber\\
&=& a_zb_z + \frac{2\Re \mbox{e}(c_1c_2^*)}{|c_1|^2 + |c_2|^2}(a_xb_x - a_yb_y) + \frac{2\Im \mbox{m}(c_1c_2^*)}{|c_1|^2 + |c_2|^2}(a_xb_y + a_yb_x) \\
&=:& a_zb_z + \xi\cos\gamma(a_xb_x - a_yb_y) + \xi\sin\gamma(a_xb_y + a_yb_x)\end{aligned}$$ where $\xi := \frac{2|c_1c_2|}{|c_1|^2 + |c_2|^2}$ and $\gamma := \arg(c_1c_2^*)$. Note that $0<\xi\leqslant 1$. We now choose $a_x = \sin\alpha, a_y = 0, a_z = \cos\alpha; b_x = \sin\beta\cos\gamma, b_y = \sin\beta\sin\gamma, b_z = \cos\beta$ to obtain $$F(\mathbf{a}, \mathbf{b}) = \cos\alpha\cos\beta + \xi\sin\alpha\sin\beta$$ Making similar choices for $\mathbf{a}', \mathbf{b}'$, and selecting $\alpha =0, \alpha' =\frac{\pi}{2}$, we obtain $$\left|F(\mathbf{a}, \mathbf{b}) - F(\mathbf{a}, \mathbf{b}')\right| + |F(\mathbf{a}', \mathbf{b}) + F(\mathbf{a}', \mathbf{b}')| = \left|\cos\beta - \cos\beta'\right| + \xi\left|\sin\beta + \sin\beta'\right|$$ We may choose $\cos\beta = -\cos\beta' =: \eta,\ \sin\beta = \sin\beta' = \sqrt{1-\eta^2}$, for which $$\left|F(\mathbf{a}, \mathbf{b}) - F(\mathbf{a}, \mathbf{b}')\right| + |F(\mathbf{a}', \mathbf{b}) + F(\mathbf{a}', \mathbf{b}')| = 2(\eta + \xi\sqrt{1-\eta^2}).$$ This quantity is maximal for $\eta = \frac{1}{\sqrt{1+4\xi^2}}$, for which it takes the value $\frac{2(1+2\xi^2)}{\sqrt{1+4\xi^2}}$, which is strictly greater than 2 for all $\xi>0$; i.e. for all non-zero $c_1$ and $c_2$. $\Box$
Let $|\psi\rangle \in\mathcal{A}(\mathcal{H}\otimes\mathcal{H})$. $|\psi\rangle$ is GMW-entangled iff it violates a permutation-invariant Bell inequality.
As mentioned above, an important example of a GMW-entangled state is the EPRB state of two electrons: $$\label{indist}
\frac{1}{\sqrt{2}}\left(|L,\uparrow\rangle\wedge|R,\downarrow\rangle - |L,\downarrow\rangle\wedge|R,\uparrow\rangle \right)\ .$$ If we use the individuating projectors $|L\rangle\langle L|\otimes\mathds{1}_{spin}$ and $|R\rangle\langle R|\otimes\mathds{1}_{spin}$, and the subsystems do not change their location, then this state is physically equivalent (because unitarily equivalent) to the state $$\label{dist}
\frac{1}{\sqrt{2}}\left(|\!\uparrow\rangle_L\otimes|\!\downarrow\rangle_R - |\!\downarrow\rangle_L\otimes|\!\uparrow\rangle_R \right)\ \in \mathbb{C}^2\otimes\mathbb{C}^2 ,$$ in which the subsystems are indexed by their locations $L$ and $R$, and permutation invariance is *not* imposed.[^3] (This is essentially pointed out by Huggett & Imbo 2009.) In particular, state (\[dist\]) violates the standard Bell inequality.
Two more biconditionals characterise entanglement for “distinguishable” systems, both of which can be extended to GMW-entanglement under permutation invariance. The first biconditional is that any joint state $|\psi\rangle$ is not entangled iff the constituent systems occupy pure states. In a permutation-invariant setting, we may say that constituent systems occupy pure states just in case individuating projectors $E_1, E_2$ may be found that satisfy our two conditions above (orthogonality and exhaustion) *and* $\dim E_1 = \dim E_2 = 1$. This conditional is obviously equivalent to $|\psi\rangle$’s being the anti-symmetrization of a product state, i.e. $|\psi\rangle$’s being non-GMW-entangled.
The second biconditional is that $|\psi\rangle$ is not entangled iff the constituents’ states determine the joint state. (Or in metaphysicians’ jargon: $|\psi\rangle$ is not entangled iff the joint state supervenes on the constituents’ states.) This biconditional is linked to the first by the following two facts: (i) the joint state is always pure; and (ii) pure states are maximally informative (and so *a fortiori* more informative than mixed states). Since, by the first biconditional, the constituent states are pure iff the joint state is not entangled, the constituents’ states carry enough information to collectively determine the joint state iff the joint state is not entangled. This reasoning carries over for GMW-entanglement in the permutation-invariant setting, so long the symmetry type of the constituents is given. In particular, for fermions: any collection of $n$ mutually orthgonal single-system pure states serves to determine a unique, non-GMW-entangled joint state: namely, their anti-symmetric combination, or wedge product.
It is important to note that the above suggestions for redefinition entanglement, constituents’ states and local operations all rely on our two conditions, (i) orthogonality, $E_1\perp E_2$, and (ii) exhaustion, $(E_1\otimes E_2 + E_2\otimes E_1)|\psi\rangle = |\psi\rangle$, holding; and that (i) and (ii) are impossible to satisfy for certain bosonic states. (In particular, product states with identical factors: $|\phi\rangle\otimes|\phi\rangle$.) It is not yet known how to make sense of physical entanglement for such states.
I conclude with a curious feature of fermionic joint states. It almost goes without saying that in the case of “distinguishable” systems *any* given joint state determines the constituents’ states; one merely has to perform the appropriate partial trace on the joint system’s density operator $|\psi\rangle\langle\psi|$. One might therefore expect the same to be true under permutation invariance. However, the partial trace has no physical meaning under permutation-invariance, since it requires selecting a preferred factor Hilbert space, a selection that, according to our favoured interpretation of permutation invariance, has no physical significance.
We must therefore revise how we extract constituents’ states in a permutation-invariant setting. The most natural alternative—for non-GMW-entangled states at least—is to look to the degree-1 vectors into which the joint state is decomposable. However, as is familiar from the use of the wedge product in differential geometry, any degree-$n$ decomposable wedge product (any non-GMW-entangled joint state of fermions) can be decomposed in several—indeed continuum-many—ways, as the anti-symmetric combination of a family of $n$ orthogonal degree-1 vectors ($n$ orthogonal constituents’ pure states). Therefore there is a sense in which a non-GMW-entangled fermionic joint state *fails* to determine the states of the constituents; a phenomenon that is, so to speak, the opposite of entanglement.
However, any two families of $n$ degree-1 vectors, into which a non-GMW-entangled fermionic state may be decomposed, have the same linear span. So, as mentioned in Section \[PIinQM\], any non-GMW-entangled state of $N$ constituents may be associated with an $N$-dimensional subspace of the single-system Hilbert space $\mathcal{H}$. We may say that any non-GMW-entangled fermionic joint state fails to determine its constituents’ states in a way exactly analogous to the failure of a multi-dimensional vector space to determine the 1-dimensional rays whose span it is. This phenomenon is unique to fermions, and suggests one more revision to our standard concepts: the sense in which an assembly is composed from its constituents. A full discussion of that is a matter for another paper.
References
==========
=-12pt
Bell, J. S. (1964), ‘On the Einstein-Podolsky-Rosen paradox’, *Physics* **1**, pp. 195-200.
Bell, J. S. (1976), ‘Einstein-Podolsky-Rosen experiments’, *Proceedings of the Symposium on Frontier Problems in High Energy Physics*, Pisa, pp. 33-45.
Butterfield, J. N. (1993), ‘Interpretation and identity in quantum theory’, *Studies in the History and Philosophy of Science* **24**, pp. 443-76.
Caulton, A. (2013). ‘Discerning “indistinguishable” quantum systems’, *Philosophy of Science* **80**, pp. 49-72.
Caulton, A. (ms.). ‘Individuation, entanglement and composition in permutation-invariant quantum mechanics’.
Dieks, D. and Lubberdink, A. (2011), ‘How Classical Particles Emerge from the Quantum World’, *Foundations of Physics* **41**, pp. 1051-1064.
Eckert, K., Schliemann, J., Bruß, D. and Lewenstein, M. (2002), ‘Quantum Correlations in Systems of Indistinguishable Particles’, *Annals of Physics* **299**, pp. 88-127.
French, S. and Krause, D. (2006), *Identity in Physics: A Historical, Philosophical and Formal Analysis.* Oxford: Oxford University Press.
French, S. and Redhead, M. (1988), ‘Quantum physics and the identity of indiscernibles’, *British Journal for the Philosophy of Science* **39**, pp. 233-46.
Ghirardi, G. and Marinatto, L. (2003), ‘Entanglement and Properties’, *Fortschritte der Physik* **51**, pp. 379-387.
Ghirardi, G. and Marinatto, L. (2004), ‘General Criterion for the Entanglement of Two Indistinguishable Particles’, *Physical Review A* **70**, 012109-1-10.
Ghirardi, G. and Marinatto, L. (2005), ‘Identical Particles and Entanglement’, *Optics and Spectroscopy* **99**, pp. 386-390.
Ghirardi, G., Marinatto, L. and Weber, T. (2002), ‘Entanglement and Properties of Composite Quantum Systems: A Conceptual and Mathematical Analysis’, *Journal of Statistical Physics* **108**, pp. 49-122.
Gisin, N. (1991), ‘Bell’s Inequality Holds for all Non-Product States’, *Physics Letters A* **154**, pp. 201-202.
Greiner, W. and Müller, B. (1994), *Quantum Mechanics: Symmetries*, 2nd revised edition. Berlin: Springer.
Huggett, N. (1999), ‘On the significance of permutation symmetry’, *British Journal for the Philosophy of Science* **50**, pp. 325-47.
Huggett, N. (2003), ‘Quarticles and the Identity of Indiscernibles’, in K. Brading and E. Castellani (eds.), *Symmetries in Physics: New Reflections*, Cambridge: Cambridge University Press, pp. 239-249.
Huggett, N. and Imbo, T. (2009), ‘Indistinguishability’, in D. Greenberger, K. Hentschel and F. Weinert (eds.), *Compendium of Quantum Physics*, Berlin: Springer, pp. 311-317.
Ladyman, J., Linnebo, Ø., and Bigaj, T. (2013), ‘Entanglement and non-factorizability’, *Studies in History and Philosophy of Modern Physics* **44**, pp. 215-221.
Mac Lane, S. and Birkoff, G. (1991), *Algebra*, Third Edition. Providence, RI: AMS Chelsea.
Mehta, M. L. (1989), *Matrix Theory: Selected Topics and Useful Results*. Delhi: Hindustan Publishing Corporation.
Messiah, A. M. L. and Greenberg, O. W. (1964), ‘Symmetrization Postulate and Its Experimental Foundation’, *Physical Review* **136**, pp. B248-B267.
Muller, F. A. and Saunders, S. (2008), ‘Discerning Fermions’, *British Journal for the Philosophy of Science*, **59**, pp. 499-548.
Muller, F. A. and Seevinck, M. (2009), ‘Discerning Elementary Particles’, *Philosophy of Science*, **76**, pp. 179-200.
Nielsen, M. A. and Chuang, I. L. (2010), *Quantum Computation and Quantum Information*, 10th anniversary edition. Cambridge: Cambridge University Press.
Russell, B. (1905), ‘On Denoting’, *Mind* **14**, pp. 479-493.
Schliemann, J., Cirac, J. I., Kuś, M., Lewenstein, M. & Loss, D. (2001), ‘Quantum correlations in two-fermion systems.’ *Physical Review A* **64**: 022303.
Tung, W.-K. (1985), *Group Theory in Physics.* River Edge: World Scientific.
[^1]: This use of ‘symmetric’ is not to be confused with the condition that $\langle\psi|Q\phi\rangle = \langle Q\psi|\phi\rangle$ for all $|\psi\rangle, |\phi\rangle\in\mbox{dom}(Q)$.
[^2]: Equivalently, $\Lambda(V)$ is the quotient algebra $T(V)/D(V^2)$ of $T(V)$ by the two-sided ideal $D(V^2)$ generated by all 2-vectors of the form $x\otimes x$. See e.g. Mac Lane & Birkoff (1991, §XVI.6).
[^3]: To be more precise: the Hilbert space $\mathfrak{H}$ spanned by the four fermionic states $\{|L,\uparrow\rangle\wedge|R,\uparrow\rangle, |L,\uparrow\rangle\wedge|R,\downarrow\rangle, |L,\downarrow\rangle\wedge|R,\uparrow\rangle, |L,\downarrow\rangle\wedge|R,\downarrow\rangle\}$, and its associated algebra of bounded operators, is unitarily equivalent to the Hilbert space spanned by the four “distinguishable”-system states $\{|\!\uparrow\rangle_L\otimes|\!\uparrow\rangle_R, |\!\uparrow\rangle_L\otimes|\!\downarrow\rangle_R, |\!\downarrow\rangle_L\otimes|\!\uparrow\rangle_R, |\!\downarrow\rangle_L\otimes|\!\downarrow\rangle_R\}$, and its associated algebra of bounded operators. The relevant unitary is the restriction of $\sqrt{2}|L\rangle_1\langle L|\otimes\mathds{1}^{(1)}_{spin}\otimes |R\rangle_2\langle R|\otimes\mathds{1}^{(2)}_{spin}$ to $\mathfrak{H}$, which sends (\[indist\]) to (\[dist\]). This unitary equivalence is discussed in a more general setting in Caulton (ms.).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We report measurements of the charmless semileptonic decays $B^0 \to \pi^- / \rho^- \ell^{+} \nu$ and $B^+ \to \pi^0 / \rho^0 \ell^{+} \nu$, based on a sample of $2.75 \times 10^8$ $B \bar{B}$ events collected at the $\Upsilon(4S)$ resonance with the Belle detector at the KEKB $e^+e^-$ asymmetric collider. In this analysis, the accompanying $B$ meson is reconstructed in the semileptonic mode $B \rightarrow D^{(*)} \ell \nu$, enabling detection of the signal modes with high purity. We measure the branching fractions ${\mathcal B}(B^0 \to \pi^- \ell^+ \nu) =
(1.38\pm 0.19\pm 0.14\pm 0.03) \times 10^{-4}$, ${\mathcal B}(B^0 \to \rho^- \ell^+ \nu) =
(2.17\pm 0.54\pm 0.31\pm 0.08) \times 10^{-4}$, ${\mathcal B}(B^+ \to \pi^0 \ell^+ \nu) =
(0.77\pm 0.14\pm 0.08\pm 0.00) \times 10^{-4}$ and ${\mathcal B}(B^+ \to \rho^0 \ell^+ \nu) =
(1.33\pm 0.23\pm 0.17\pm 0.05) \times 10^{-4}$, where the errors are statistical, experimental systematic, and systematic due to form-factor uncertainties, respectively. For each mode we also present the partial branching fractions in three $q^2$ intervals: $q^2 < 8$, $8 \leq q^2 < 16$, and $q^2 \geq 16$GeV$^2/c^2$. From our partial branching fractions for $B \to \pi \ell \nu$ and recent results for the form factor from unquenched Lattice QCD calculations, we obtain values of the CKM matrix element $|V_{ub}|$.
address:
- 'Budker Institute of Nuclear Physics, Novosibirsk, Russia'
- 'Chiba University, Chiba, Japan'
- 'University of Cincinnati, Cincinnati, OH, USA'
- 'University of Hawaii, Honolulu, HI, USA'
- 'High Energy Accelerator Research Organization (KEK), Tsukuba, Japan'
- 'Institute for High Energy Physics, Protvino, Russia'
- 'Institute of High Energy Physics, Vienna, Austria'
- 'Institute for Theoretical and Experimental Physics, Moscow, Russia'
- 'J. Stefan Institute, Ljubljana, Slovenia'
- 'Kanagawa University, Yokohama, Japan'
- 'Korea University, Seoul, South Korea'
- 'Kyungpook National University, Taegu, South Korea'
- 'Swiss Federal Institute of Technology of Lausanne, EPFL, Lausanne, Switzerland'
- 'University of Ljubljana, Ljubljana, Slovenia'
- 'University of Maribor, Maribor, Slovenia'
- 'University of Melbourne, Victoria, Australia'
- 'Nagoya University, Nagoya, Japan'
- 'Nara Women’s University, Nara, Japan'
- 'National Central University, Chung-li, Taiwan'
- 'National United University, Miao Li, Taiwan'
- 'Department of Physics, National Taiwan University, Taipei, Taiwan'
- 'H. Niewodniczanski Institute of Nuclear Physics, Krakow, Poland'
- 'Nippon Dental University, Niigata, Japan'
- 'Niigata University, Niigata, Japan'
- 'Nova Gorica Polytechnic, Nova Gorica, Slovenia'
- 'Osaka City University, Osaka, Japan'
- 'Osaka University, Osaka, Japan'
- 'Panjab University, Chandigarh, India'
- 'Peking University, Beijing, PR China'
- 'Princeton University, Princeton, NJ, USA'
- 'Saga University, Saga, Japan'
- 'University of Science and Technology of China, Hefei, PR China'
- 'Seoul National University, Seoul, South Korea'
- 'Shinshu University, Nagano, Japan'
- 'Sungkyunkwan University, Suwon, South Korea'
- 'University of Sydney, Sydney, NSW, Australia'
- 'Tata Institute of Fundamental Research, Bombay, India'
- 'Toho University, Funabashi, Japan'
- 'Tohoku Gakuin University, Tagajo, Japan'
- 'Tohoku University, Sendai, Japan'
- 'Department of Physics, University of Tokyo, Tokyo, Japan'
- 'Tokyo Institute of Technology, Tokyo, Japan'
- 'Tokyo Metropolitan University, Tokyo, Japan'
- 'Tokyo University of Agriculture and Technology, Tokyo, Japan'
- 'University of Tsukuba, Tsukuba, Japan'
- 'Virginia Polytechnic Institute and State University, Blacksburg, VA, USA'
- 'Yonsei University, Seoul, South Korea'
author:
- 'T. Hokuue'
- 'K. Abe'
- 'K. Abe'
- 'I. Adachi'
- 'H. Aihara'
- 'Y. Asano'
- 'T. Aushev'
- 'A. M. Bakich'
- 'V. Balagura'
- 'E. Barberio'
- 'M. Barbero'
- 'A. Bay'
- 'I. Bedny'
- 'K. Belous'
- 'U. Bitenc'
- 'I. Bizjak'
- 'S. Blyth'
- 'A. Bondar'
- 'A. Bozek'
- 'M. Bračko'
- 'T. E. Browder'
- 'P. Chang'
- 'A. Chen'
- 'W. T. Chen'
- 'Y. Choi'
- 'A. Chuvikov'
- 'S. Cole'
- 'J. Dalseno'
- 'M. Danilov'
- 'M. Dash'
- 'A. Drutskoy'
- 'S. Eidelman'
- 'N. Gabyshev'
- 'A. Garmash'
- 'T. Gershon'
- 'G. Gokhroo'
- 'A. Gorišek'
- 'H. Ha'
- 'J. Haba'
- 'K. Hara'
- 'T. Hara'
- 'K. Hayasaka'
- 'H. Hayashii'
- 'M. Hazumi'
- 'L. Hinz'
- 'Y. Hoshi'
- 'S. Hou'
- 'W.-S. Hou'
- 'T. Iijima'
- 'K. Ikado'
- 'A. Imoto'
- 'K. Inami'
- 'A. Ishikawa'
- 'R. Itoh'
- 'M. Iwasaki'
- 'Y. Iwasaki'
- 'H. Kakuno'
- 'J. H. Kang'
- 'P. Kapusta'
- 'S. U. Kataoka'
- 'H. Kawai'
- 'T. Kawasaki'
- 'H. R. Khan'
- 'H. J. Kim'
- 'H. O. Kim'
- 'K. Kinoshita'
- 'Y. Kozakai'
- 'P. Križan'
- 'P. Krokovny'
- 'R. Kulasiri'
- 'R. Kumar'
- 'C. C. Kuo'
- 'A. Kuzmin'
- 'Y.-J. Kwon'
- 'J. Lee'
- 'T. Lesiak'
- 'J. Li'
- 'A. Limosani'
- 'S.-W. Lin'
- 'G. Majumder'
- 'F. Mandl'
- 'T. Matsumoto'
- 'A. Matyja'
- 'S. McOnie'
- 'W. Mitaroff'
- 'H. Miyake'
- 'H. Miyata'
- 'Y. Miyazaki'
- 'D. Mohapatra'
- 'I. Nakamura'
- 'E. Nakano'
- 'Z. Natkaniec'
- 'S. Nishida'
- 'O. Nitoh'
- 'T. Nozaki'
- 'S. Ogawa'
- 'T. Ohshima'
- 'T. Okabe'
- 'S. Okuno'
- 'S. L. Olsen'
- 'Y. Onuki'
- 'P. Pakhlov'
- 'C. W. Park'
- 'H. Park'
- 'L. S. Peak'
- 'R. Pestotnik'
- 'L. E. Piilonen'
- 'Y. Sakai'
- 'N. Sato'
- 'N. Satoyama'
- 'K. Sayeed'
- 'T. Schietinger'
- 'O. Schneider'
- 'C. Schwanda'
- 'A. J. Schwartz'
- 'K. Senyo'
- 'M. E. Sevior'
- 'M. Shapkin'
- 'H. Shibuya'
- 'B. Shwartz'
- 'A. Somov'
- 'R. Stamen'
- 'S. Stanič'
- 'M. Starič'
- 'H. Stoeck'
- 'A. Sugiyama'
- 'S. Suzuki'
- 'S. Y. Suzuki'
- 'O. Tajima'
- 'F. Takasaki'
- 'K. Tamai'
- 'M. Tanaka'
- 'G. N. Taylor'
- 'Y. Teramoto'
- 'X. C. Tian'
- 'T. Tsukamoto'
- 'K. Ueno'
- 'T. Uglov'
- 'Y. Unno'
- 'S. Uno'
- 'P. Urquijo'
- 'Y. Usov'
- 'G. Varner'
- 'K. E. Varvell'
- 'S. Villa'
- 'C. C. Wang'
- 'C. H. Wang'
- 'M.-Z. Wang'
- 'Y. Watanabe'
- 'E. Won'
- 'A. Yamaguchi'
- 'Y. Yamashita'
- 'M. Yamauchi'
- 'J. Ying'
- 'L. M. Zhang'
- 'Z. P. Zhang'
- 'D. Zürcher'
title: |
\
Measurements of branching fractions and $q^2$ distributions for $B \to \pi \ell \nu$ and $B \to \rho \ell \nu$ Decays with $B \to D^{(*)} \ell \nu$ Decay Tagging
---
2.cm
-1.5cm Belle Prerpint 2006-10\
KEK Preprint 2006-4
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
, and
,
Semileptonic ,B decay ,exclusive 12.15.Hh, 12.38.Gc, 13.25.Hw
Introduction {#sec:Introduction}
============
Exclusive $B \to X_u \ell \nu$ decays proceed dominantly via a $b \to u W^-$ tree process and can be used to determine $|V_{ub}|$, one of the smallest and least known elements of the Cabibbo-Kobayashi-Maskawa matrix [@KM]. However, the need to translate the observed rate to a $|V_{ub}|$ value using model-dependent decay form-factors (FF) has resulted in large theoretical uncertainties. The recent release of FF results for $B \to \pi \ell \nu$ calculated by unquenched Lattice QCD (LQCD) [@FNAL04; @HPQCD04] makes possible the first model-independent determination of $|V_{ub}|$. Since LQCD results are available only in the high $q^2$ region ($\geq 16$GeV$^2/c^2$), a clean measurement of the partial $B \to \pi \ell \nu$ branching fraction in the same high $q^2$ region is needed.
There have been several measurements in the past by CLEO, BaBar and Belle for the $B \to \pi \ell \nu$, $\rho \ell \nu$, $\eta \ell \nu$ and $\omega \ell \nu$ modes [@CLEO1996; @CLEO2000; @CLEO2003; @BABAR2003; @BELLE2004; @BABAR2005]. The analyses in these measurements utilize the method, originally developed by CLEO, where the $B$ decays are reconstructed by inferring the undetected neutrino mass from missing energy and momentum (“$\nu$-reconstruction method”) [@CLEO1996]. In the $B$-factory era, we will improve the statistical precision by simply applying the $\nu$-reconstruction method using a large amount of data. However, the poor signal-to-noise ratio will limit the systematic uncertainty of the measurement.
In this paper we present measurements of $B^0 \to \pi^- / \rho^- \ell^+ \nu$ and $B^+ \to \pi^0 / \rho^0 \ell^+ \nu$ decays using $B \to D^{(*)} \ell \nu$ decay tagging. We reconstruct the entire decay chain from the $\Upsilon(4S)$, $\Upsilon(4S) \to B_{\rm sig}B_{\rm tag}$, $B_{\rm sig} \to \pi / \rho \ell \nu$ and $B_{\rm tag} \to D^{(*)} \ell \bar{\nu}$ with several $D^{(*)}$ sub-modes. The back-to-back correlation of the two $B$ mesons in the $\Upsilon(4S)$ rest frame allows us to constrain the kinematics of the double semileptonic decay. The signal is reconstructed in four modes, $B^0 \to \pi^- / \rho^- \ell^+ \nu$ and $B^+ \to \pi^0 / \rho^0 \ell^+ \nu$. Yields and branching fractions are extracted from a simultaneous fit of the $B^0$ and $B^+$ samples in three intervals of $q^2$, accounting for cross-feed between modes as well as other backgrounds. We have applied this method to $B \to \pi / \rho \ell \nu$ decays for the first time, and have succeeded in reconstructing these decays with significantly improved signal-to-noise ratios compared to the $\nu$-reconstruction method. Inclusion of charge conjugate decays is implied throughout this paper.
Data Set and Experiment {#sec:data_exp}
=======================
The analysis is based on data recorded with the Belle detector at the KEKB collider operating at the center-of-mass (c.m.) energy for the $\Upsilon(4S)$ resonance [@KEKB]. The $\Upsilon(4S)$ dataset that is used corresponds to an integrated luminosity of 253 fb$^{-1}$ and contains $2.75 \times 10^8$ $B \bar{B}$ events.
The Belle detector is a large-solid-angle magnetic spectrometer that consists of a silicon vertex detector (SVD), a 50-layer central drift chamber (CDC), an array of aerogel threshold Čerenkov counters (ACC), a barrel-like arrangement of time-of-flight scintillation counters (TOF), and an electromagnetic calorimeter comprised of CsI(Tl) crystals (ECL) located inside a super-conducting solenoid coil that provides a 1.5 T magnetic field. An iron flux-return located outside of the coil is instrumented to detect $K_L^0$ mesons and to identify muons (KLM). The detector is described in detail elsewhere [@BELLE]. Two inner detector configurations were used. A 2.0 cm beam pipe and a 3-layer silicon vertex detector was used for the first sample of $152 \times 10^6$ $B\bar{B}$ pairs, while a 1.5 cm beam pipe, a 4-layer silicon detector, and a small-cell inner drift chamber were used to record the remaining 123 million $B\bar{B}$ pairs [@Ushiroda].
A detailed Monte Carlo (MC) simulation, which fully describes the detector geometry and response and is based on GEANT [@GEANT], is applied to estimate the signal detection efficiency and to study the background. To examine the FF dependence, MC samples for the $B \to \pi / \rho \ell \nu$ signal decays are generated with different form-factor models: a quark model (ISGW II [@ISGW2]), light cone sum rules (LCSR for $B \to \pi \ell \nu$ [@Ball04_pi] and $B \to \rho \ell \nu$ [@Ball04_rho]) and quenched lattice QCD (UKQCD [@UKQCD]). We also use unquenched lattice QCD (FNAL [@FNAL04] and HPQCD [@HPQCD04]) for $B \to \pi \ell \nu$ and a relativistic quark model (Melikhov [@Melikhov]) for $B \to \rho \ell \nu$. To model the cross-feed from other $B \to X_u \ell \nu$ decays, MC samples are generated with the ISGW II model for the resonant components ($B \to \pi \ell \nu$ and $B \to \rho \ell \nu$ components are excluded in this sample) and the DeFazio-Neubert model [@Fazio-Neubert] for the non-resonant component. To model the $B\bar{B}$ and continuum backgrounds, large generic $B\bar{B}$ and $q\bar{q}$ Monte Carlo (based on Evtgen [@Evtgen]) samples are used.
Event Reconstruction and Selection
==================================
Charged particle tracks are reconstructed from hits in the SVD and the CDC. They are required to satisfy track quality cuts based on their impact parameters relative to the measured interaction point (IP) of the two beams. Charged kaons are identified by combining information on ionization loss ($dE/dx$) in the CDC, Čherenkov light yields in the ACC and time-of-flight measured by the TOF system. For the nominal requirement, the kaon identification efficiency is approximately $88\%$ and the rate for misidentification of pions as kaons is about $8\%$. Hadron tracks that are not identified as kaons are treated as pions. Tracks satisfying the lepton identification criteria, as described below, are removed from consideration.
Neutral pions are reconstructed using $\gamma$ pairs with an invariant mass between 117 and 150MeV/$c^2$. Each $\gamma$ is required to have a minimum energy of $50$ MeV. $K_S^0$ mesons are reconstructed using pairs of tracks that are consistent with having a common vertex and that have an invariant mass within $\pm 12$MeV/$c^2$ of the known $K_S^0$ mass.
Electron identification is based on a combination of $dE/dx$ in the CDC, the response of the ACC, shower shape in the ECL and the ratio of energy deposit in the ECL to the momentum measured by the tracking system. Muons are identified by their signals in the KLM resistive plate counters, which are interleaved with the iron of the solenoid return yoke. The lepton identification efficiencies are estimated to be about 90% for both electrons and muons in the momentum region above 1.2GeV/$c$, where leptons from prompt $B$ decays dominate. The hadron misidentification rate is measured using reconstructed $K_S^0 \to \pi^+ \pi^-$ and found to be less than 0.2% for electrons and 1.5% for muons in the same momentum region.
For the reconstruction of $B_{\rm tag} \to D^{(*)} \ell \bar{\nu}$, the lepton candidate is required to have the correct sign charge with respect to the $D$ meson flavor and a laboratory momentum ($p_{\ell}^{lab}$) greater than 1.0GeV/$c$. The $D$ meson candidates are reconstructed by using seven decay modes of $D^+$: $D^+ \to K^- \pi^+ \pi^+$, $K^- \pi^+ \pi^+ \pi^0$, $K_S^0 \pi^+$, $K_S^0 \pi^+ \pi^0$, $K_S^0 \pi^+ \pi^+ \pi^-$, $K^+ K_S^0$, $K^+ K^- \pi^+$; and ten decay modes of $D^0$: $D^0 \to K^- \pi^+$, $K^- \pi^+ \pi^0$, $K^- \pi^+ \pi^+ \pi^-$, $K_S^0 \pi^0$, $K_S^0 \pi^+ \pi^-$, $K_S^0 \pi^+ \pi^- \pi^0$, $K^- \pi^+ \pi^+ \pi^- \pi^0$, $K^+ K^-$, $K_S^0 K^+ K^- $, $K_S^0 K^- \pi^+$. The candidates are required to have an invariant mass $m_D$ within $\pm 2.5\sigma$ ($\sigma$ is a standard deviation) of the nominal $D$ mass, where the mass resolution $\sigma$ is dependent on the decay mode. $D^{*}$ mesons are reconstructed in the modes $D^{*+} \to D^0 \pi^+$, $D^+ \pi^0$ and $D^{*0} \to D^0 \pi^0$ by combining a $D$ meson candidate and a charged or neutral pion. Each $D^*$ candidate is required to have a mass difference $\Delta m = m_{\bar{D}\pi} - m_{\bar{D}}$ within $\pm 2.5 \sigma$ of the nominal values.
For the reconstruction of $B_{\rm sig} \to X_u \ell \nu$, the lepton candidate is required to have the right sign charge with respect to the $X_u$ system and $p_{\ell}^{lab}$ greater than 0.8 GeV/$c$. The $X_u$ system may consist of one pion or two pions ( $N_{\pi^+}=1$ or $N_{\pi^+}=N_{\pi^0}=1$ for a $\bar{B^0}$ tag and $N_{\pi^0}=1$ or $N_{\pi^+}=N_{\pi^-}=1$ for a $B^-$ tag). The event is required to have no additional charged tracks or $\pi^0$ candidates. We also require that the residual energy from neutral clusters be less than 0.15 GeV ($E_{\rm neut} < 0.15$GeV). The two leptons on the tag and the signal sides are required to have opposite charge. The loss of signal due to $B^0 - \bar{B^0}$ mixing is estimated by MC simulation.
We then impose a constraint based on the kinematics of the double semileptonic decay in the $\Upsilon(4S)$ rest frame. In the semileptonic decay on each side, $B_{1(2)} \to Y_{1(2)} \nu$ ($Y_1 = D^{(*)} \ell$ and $Y_2 = X_u \ell$), the angle between the $B_{1(2)}$ meson and the detected $Y_{1(2)}$ system $\theta_{B_{1(2)}}$ is calculated from the relation, ${P_{\nu}^{*}}^2 = (P_{B}^{*} - P_{Y}^{*})^2 = 0$ ($P^*$: 4-momentum vector) and the known $p_B^*$ (the absolute momentum of the mother $B$ meson). This means that the $B_{1(2)}$ direction is constrained on the surface of a cone defined with the angle $\theta_{B_{1(2)}}^*$ around the direction of the $Y_{1(2)}$ system, as shown graphically in Fig. \[fig:double\_cone\]. The back-to-back relation of the two $B$ meson directions then implies that the real $B$ direction is on the intersection of the two cones when one of the $B$ systems is spatially inverted. Denoting $\theta_{12}^*$ the angle between the $p_{Y1}^*$ and $-p_{Y2}^*$, the $B$ directional vector $\vec{n}_B = (x_B, y_B, z_B)$ is given by, $z_B = \mbox{cos}\theta_{B_1}^*$, $y_B = (\mbox{cos}\theta_{B_2}^* - \mbox{cos}\theta_{B_2}^*\mbox{cos}\theta_{12}^*)
/ \mbox{sin}\theta_{12}^*$, and $$\begin{aligned}
{x_B}^2 = 1-\frac{1}{\mbox{sin}^2\theta_{12}^*}
(\mbox{cos}^2\theta_{B_1}^* + \mbox{cos}^2\theta_{B_2}^*
- 2 \mbox{cos}\theta_{B_1}^*\mbox{cos}\theta_{B_2}^*\mbox{cos}\theta_{12}^*)
\label{eq:x_B}\end{aligned}$$ with the coordinate definition in Fig. \[fig:double\_cone\], where the $p_{Y1}^*$ and $p_{Y2}^*$ are aligned along the $z$-axis and in the $y-z$ plane, respectively. If the hypothesis of the double semileptonic decay is correct and all the decay products are detected except for the two neutrinos, $x_B^2$ must range from 0 to 1. Events passing a rather loose cut $x_B^2 > -2.0$ are used for signal extraction at a later stage of the analysis.
Since the direction of the $B$ meson is not uniquely determined, we calculate, $q^2$ as $q^{2} = (E^{*}_{\rm beam} - E^{*}_{X_u})^2 - p^{*}_{X_u}{^2}$, using the beam energy ($E^*_{\rm beam}$), energy ($E^*_{X_u}$) and momentum ($p^*_{X_u}$) of the $X_u$ system and neglecting the momentum of the $B$ meson in the c.m. system. The signal Monte Carlo simulation finds that the $q^2$ resolution depends on the reconstructed $q^2$ and is in the range 0.32-0.95GeV$^2/c^2$.
According to Monte Carlo simulation, the largest backgrounds originate from $B \to X_c \ell \nu$ and non-signal $B \to X_u \ell \nu$ decays, where some particles escape detection. There are sizable contributions from cross talk between the $\bar{B^0}$ and $B^+$ tags. The contribution from $q\bar{q}$ processes is found to be negligible.
For events selected as described above, the signal MC simulation indicates that the total detection efficiency ($\epsilon_{\rm total}$), averaged over electron and muon channels, is $1.98\times 10^{-3}$ for $B^0 \to \pi^- \ell^+ \nu$ and $0.76\times 10^{-3}$ for $B^0 \to \rho^- \ell^+ \nu$, $1.49\times 10^{-3}$ for $B^+ \to \pi^0 \ell^+ \nu$ and $1.78\times 10^{-3}$ for $B^+ \to \rho^0 \ell^+ \nu$ assuming the LCSR FF model. Here, $\epsilon_{\rm total}$ is defined with respect to the number of $B \bar{B}$ pairs, where one $B$ decays into the signal mode, and includes the loss of signal due to $B^0 - \bar{B^0}$ mixing. Because of the loose lepton momentum cut ($p_{\ell}^{lab}$ $>0.8$GeV/$c$), the variation of efficiency with different FF models is relatively small. Table \[tbl:Matrix\_sig\] gives the matrix $\epsilon(q^2_{\rm rec.}, q^2_{\rm true})$, the efficiency for a signal event generated with true $q^2$ in the bin $q^2_{\rm true}$ to be reconstructed in the bin $q^2_{\rm rec.}$.
= 1pt
------------------ ------- ---------------------- ----------- ------- ---------------------- ----------- ------- --------------------- ----------- ------- ---------------------- -----------
$q^2_{\rm true}$ $ \pi^- \ell^+ \nu $ $\rho^- \ell^+ \nu $ $\pi^0 \ell^+ \nu $ $\rho^0 \ell^+ \nu $
(GeV$^2/c^2$) $< 8$ $ 8 - 16$ $\geq 16$ $< 8$ $ 8 - 16$ $\geq 16$ $< 8$ $ 8 - 16$ $\geq 16$ $< 8$ $ 8 - 16$ $\geq 16$
$< 8$ 1.71 0.21 0.00 0.59 0.03 0.00 1.27 0.07 0.00 1.50 0.08 0.00
$8 - 16$ 0.05 1.82 0.24 0.07 0.65 0.05 0.10 1.43 0.06 0.10 1.71 0.13
$\geq 16$ 0.00 0.03 1.89 0.02 0.13 0.81 0.01 0.09 1.45 0.01 0.08 1.82
------------------ ------- ---------------------- ----------- ------- ---------------------- ----------- ------- --------------------- ----------- ------- ---------------------- -----------
: Detection efficiency matrix $\epsilon(q^2_{\rm rec.}, q^2_{\rm true})$ based on the LCSR model in units of $10^{-3}$.[]{data-label="tbl:Matrix_sig"}
To check the validity of the method, we apply the procedure described above to reconstruct $B_{\rm sig}^0 \to D^{*-} \ell^{+} \nu$ followed by $D^{*-} \to \bar{D^0} \pi^{-}, \bar{D^0} \to K^{+} \pi^{-}$ for a $\bar{B}^0$ tag and $B_{\rm sig}^+ \to \bar{D}^{*0} \ell^{+} \nu$ followed by $\bar{D}^{*0} \to \bar{D^0} \pi^{0}, \bar{D^0} \to K^{+} \pi^{-}$ for a $B^-$ tag, with the same requirement on the tagging side. Figures \[fig:dstlnu\]-a) and c) show the $M_{K\pi\pi}$ distributions that are obtained in data and expected from MC. As a result, we obtained $224.7 \pm 15.4$ ($295.9 \pm 17.6$) $\bar{B}^0 ~(B^-)$ tagged events. These values are in good agreement with expected values $224.5 \pm 9.5$ ($288.6 \pm 11.7$) calculated from the branching fractions ${\mathcal B}(B^{0} \to D^{*-} \ell^{+} \nu)$, ${\mathcal B}(D^{*-(0)} \to \bar{D^0} \pi^{-(0)})$ and ${\mathcal B}(\bar{D^0} \to K^{+} \pi^{-})$ in [@PDG2005] and efficiencies obtained from MC. Here, we use ${\mathcal B}(B^+ \to \bar{D}^{*0} \ell^{+} \nu)$ calculated from ${\mathcal B}(B^{0} \to D^{*-} \ell^{+} \nu)$ and the liftime ratio [@PDG2005]; ${\mathcal B}(B^+ \to \bar{D}^{*0} \ell^{+} \nu) = {\mathcal B}(B^{0} \to D^{*-} \ell^{+} \nu) \times (\tau_{B^+} / \tau_{B^0})$. The ratio of the reconstructed to expected value, $R=1.00 \pm 0.08 \pm 0.05 ~(1.03 \pm 0.07 \pm 0.05)$ where the first error is statistical error and the second is due to the uncertainty of the branching fractions from [@PDG2005], is consistent with unity. Figures \[fig:dstlnu\]-b) and d) show a comparison of the reconstructed $x_B^2$ distribution in the above data samples with MC simulation. Data and MC are in good agreement.
-- --
-- --
Extraction of Branching Fractions {#sec:SignalExtraction}
=================================
The $B^0 \to \pi^- / \rho^- \ell^+ \nu$ and $B^+ \to \pi^0 / \rho^0 \ell^+ \nu$ signals are extracted using binned maximum likelihood fits to the two-dimensional $(x_B^2, M_X)$ distribution, where $M_X$ is the nominal pion mass for $B \to \pi \ell^+ \nu$ candidates and the invariant mass of two pions for $B \to \rho \ell^+ \nu$ candidates. The fit includes seven components: the four signal modes and the other $B^0 \to X_u^- \ell^+ \nu$ and $B^+ \to X_u^0 \ell^+ \nu$ backgrounds, the background from $B \bar{B}$ events containing no $B \to X_u\ell\nu$. The PDF (probability density function) for each fit component is determined from MC simulation. The $\pi / \rho$ signal events exhibit characteristic behavior in both the $x_B^2$ and $M_X$ distributions; other $B \to X_u \ell^+ \nu$ events exhibit a weak peaking structure in $x_B^2$ but a broad distribution in $M_X$; the $B \bar{B}$ background has a relatively flat distribution in $x_B^2$ and a broad structure in $M_X$. The PDFs in $(x_B^2, M_X)$ for each of the seven fit components are obtained from MC for both $\bar{B}^0$ and $B^-$ tag candidates. We then fit the two $(x_B^2, M_X)$ distributions for both $\bar{B}^0$ and $B^-$ tags simultaneously; The fitting is constrained so that the sum of the deduced branching fractions for $B \to \pi \ell^+ \nu$, $B \to \rho \ell^+ \nu$ and $B \to$ other $X_u \ell^+ \nu$ is equal to the total inclusive branching fraction ${\mathcal B}(B \to X_u \ell \nu) = (0.25 \pm 0.06)$% [@BELLE_Kakuno]. Figure \[fig:fit\_allq2\] shows the projections on $M_X$ and $x_B^2$ of the fitting result for data in the entire $q^2$ region. The extracted yields for the signal components are $N(B^0 \to \pi^- \ell^+ \nu) = 155.8 \pm 20.0$, $N(B^0 \to \rho^- \ell^+ \nu) = 92.9 \pm 19.4$, $N(B^+ \to \pi^0 \ell^+ \nu) = 69.0 \pm 11.4$ and $N(B^+ \to \rho^0 \ell^+ \nu) = 135.4 \pm 24.8$, with the LCSR model used for the four signal PDFs.
-- -- --
-- -- --
Figure \[fig:fit\_subq2\] shows projections of the data, separated into three $q^2$ bins, $q^2 < 8$ GeV$^2/c^2$, $8 \leq q^2 < 16$ GeV$^2/c^2$ and $q^2 \geq 16$ GeV$^2/c^2$. Here the normalizations of the other $B \to X_u \ell \nu$ and the $B \bar{B}$ background components are fixed to those obtained in the above fitting for the entire $q^2$ region. The extracted numbers of events for low/ medium/ high $q^2$ bins are, $N(B^0 \to \pi^- \ell^+ \nu) = 64.8 \pm 11.9 ~/~ 63.2 \pm 12.4 ~/~ 40.6 \pm 11.3$, $N(B^0 \to \rho^- \ell^+ \nu) = 22.1 \pm 8.0 ~/~ 53.2 \pm 13.5 ~/~ 30.9 \pm 16.0 $, $N(B^+ \to \pi^0 \ell^+ \nu) = 18.1 \pm 5.1 ~/~ 34.5 \pm 8.3 ~/~ 18.6 \pm 6.5 $ and $N(B^+ \to \rho^0 \ell^+ \nu) = 47.2 \pm 11.2 ~/~ 68.3 \pm 16.5 ~/~ 32.5 \pm 12.3 $. Table \[tbl:FFdep\_q2\_Xulnu\] summarizes the extracted branching fractions. The branching fractions are calculated for each signal FF-model, where we take the average for cross-feed FF-models. The results are unfolded using the efficiency matrix $\epsilon(q^2_{\rm rec.}, q^2_{\rm true})$ for the three $q^2$ intervals prepared for each signal FF-model. We calculate the total branching fraction by taking the sum of the partial branching fractions in the three $q^2$ intervals.
-- -- --
-- -- --
= 3pt
Mode Model ${\mathcal B}_{\rm total}$ ${\mathcal B}_{<8}$ ${\mathcal B}_{8-16}$ ${\mathcal B}_{\geq 16}$ $\chi^2/n$ $Prob.$
--------------------- ---------- ---------------------------- --------------------- ----------------------- -------------------------- ------------ ---------
$\pi^- \ell^+ \nu$ LCSR $1.40 \pm 0.19 $ $0.52 \pm 0.11 $ $0.51 \pm 0.11 $ $0.36 \pm 0.10 $ 0.4/2 0.81
ISGW II $1.36 \pm 0.19 $ $0.52 \pm 0.11 $ $0.48 \pm 0.11 $ $0.36 \pm 0.10 $ 3.6/2 0.17
UKQCD $1.39 \pm 0.19 $ $0.53 \pm 0.11 $ $0.50 \pm 0.11 $ $0.36 \pm 0.10 $ 0.2/2 0.89
FNAL $1.39 \pm 0.19 $ $0.52 \pm 0.11 $ $0.51 \pm 0.11 $ $0.36 \pm 0.10 $ 0.3/2 0.86
HPQCD $1.39 \pm 0.19 $ $0.52 \pm 0.11 $ $0.50 \pm 0.11 $ $0.36 \pm 0.10 $ 0.5/2 0.79
Average $1.38 \pm 0.19 $ $0.52 \pm 0.11 $ $0.50 \pm 0.11 $ $0.36 \pm 0.10 $ – –
$\rho^- \ell^+ \nu$ LCSR $2.21 \pm 0.54 $ $0.48 \pm 0.19 $ $1.18 \pm 0.33 $ $0.55 \pm 0.39 $ 2.4/2 0.43
ISGW II $2.09 \pm 0.54 $ $0.44 \pm 0.17 $ $1.14 \pm 0.32 $ $0.51 \pm 0.39 $ 0.5/2 0.78
UKQCD $2.19 \pm 0.54 $ $0.48 \pm 0.19 $ $1.20 \pm 0.33 $ $0.51 \pm 0.38 $ 1.7/2 0.43
Melikhov $2.20 \pm 0.54 $ $0.48 \pm 0.19 $ $1.22 \pm 0.34 $ $0.50 \pm 0.37 $ 1.8/2 0.41
Average $2.17 \pm 0.54 $ $0.47 \pm 0.19 $ $1.19 \pm 0.33 $ $0.52 \pm 0.38 $ – –
$\pi^0 \ell^+ \nu$ LCSR $0.77 \pm 0.14 $ $0.19 \pm 0.06 $ $0.39 \pm 0.10 $ $0.20 \pm 0.08 $ 2.9/2 0.24
ISGW II $0.77 \pm 0.14 $ $0.19 \pm 0.06 $ $0.39 \pm 0.10 $ $0.20 \pm 0.08 $ 7.8/2 0.02
UKQCD $0.77 \pm 0.14 $ $0.19 \pm 0.06 $ $0.38 \pm 0.10 $ $0.20 \pm 0.08 $ 2.5/2 0.28
FNAL $0.77 \pm 0.14 $ $0.19 \pm 0.06 $ $0.39 \pm 0.10 $ $0.20 \pm 0.08 $ 2.8/2 0.25
HPQCD $0.77 \pm 0.14 $ $0.19 \pm 0.06 $ $0.39 \pm 0.10 $ $0.20 \pm 0.08 $ 4.8/2 0.09
Average $0.77 \pm 0.14 $ $0.19 \pm 0.06 $ $0.39 \pm 0.10 $ $0.20 \pm 0.08 $ – –
$\rho^0 \ell^+ \nu$ LCSR $1.36 \pm 0.23 $ $0.43 \pm 0.11 $ $0.63 \pm 0.16 $ $0.29 \pm 0.12 $ 0.9/2 0.64
ISGW II $1.29 \pm 0.22 $ $0.45 \pm 0.11 $ $0.57 \pm 0.15 $ $0.27 \pm 0.12 $ 1.4/2 0.50
UKQCD $1.34 \pm 0.23 $ $0.45 \pm 0.11 $ $0.62 \pm 0.16 $ $0.28 \pm 0.12 $ 0.2/2 0.91
Melikhov $1.35 \pm 0.23 $ $0.44 \pm 0.11 $ $0.63 \pm 0.17 $ $0.28 \pm 0.12 $ 0.2/2 0.92
Average $1.33 \pm 0.23 $ $0.44 \pm 0.11 $ $0.61 \pm 0.16 $ $0.28 \pm 0.12 $ – –
: Extracted branching fractions for each signal mode with different FF models in units of 10$^{-4}$: the total branching fraction and the partial branching fractions in three $q^2$ intervals. $\chi^2/n$ and the associated probability for this $\chi^2$ indicate the quality of the fit for the FF shape to the observed $q^2$ distribution.[]{data-label="tbl:FFdep_q2_Xulnu"}
Systematic Errors {#sec:SystematicErrors}
=================
Tables \[tbl:systematic\_pi\] and [\[tbl:systematic\_pi0\]]{} summarize the experimental systematic errors on the branching fractions. The experimental systematic errors can be categorized as originating from uncertainties in the signal reconstruction efficiency, the background estimation, and the normalization. The total experimental systematic error is the quadratic sum of all individual ones. We also consider the systematic error due to the dependence on the FF model.
The effect from the uncertainty on the signal reconstruction efficiency is evaluated based on the efficiency calibration with the $B_{\rm sig} \to D^{*} \ell^+ \nu$ sample, discussed above. The error is taken to be that on the ratio of observed to expected number of the calibration signals (9.3% for $B^0 \to \pi^-/\rho^- \ell^+ \nu$, 9.2% for $B \to \pi^0/\rho^0 \ell^+ \nu$). This gives the largest contribution to the systematic error. Note that this error is dominated by the statistics of the calibration signals, as explained above. Therefore, accumulation of additional integrated luminosity in the future will help to reduce this uncertainty. We also include residual errors for the reconstruction of the signal side: 1% and 2% for the detection of each charged and neutral pion, respectively, and 2% for the charged pion selection and 2.1% for the lepton selection.
The systematic error due to the uncertainty on the inclusive branching fraction ${\mathcal B}(B \to X_u \ell \nu)$, which is used to constrain $B \to X_u \ell^+ \nu$ background, is estimated by varying this parameter by its $\pm 1 \sigma$ error. The uncertainty in the $B \bar{B}$ background shape after our pion multiplicity selection requirements ($N_{\pi^+}=1$ or $N_{\pi^+}=N_{\pi^0}=1$ for a $\bar{B^0}$ tag and $N_{\pi^0}=1$ or $N_{\pi^+}=N_{\pi^-}=1$ for a $B^-$ tag) is studied in the simulation by randomly removing charged tracks and $\pi^0$ according to the error in detection efficiency (1% for a charged track, 2% for $\pi^0$), and also by reassigning identified charged kaons as pions according to the uncertainty in the kaon identification efficiency (2%). The resultant changes in the extracted branching fractions are assigned as systematic errors. We find a significant uncertainty in the high $q^2$ region ($q^2 > 16$ GeV$^2/c^2$) for $B \to \rho \ell^+ \nu$ due to the poor signal-to-noise ratio. We also vary the fraction of $B \to D^{**} \ell \nu$ decays in the $B\bar{B}$ background MC by the error quoted in [@PDG2005] to test the $B \to X_c \ell \nu$ model dependence in the $B\bar{B}$ background shape. To assess the uncertainty due to the production rate of $K_L^0$, we vary the production rate in the MC simulation by the uncertainty in the inclusive branching fraction for $B \to K^0 ~X$ quoted in [@PDG2005].
For the normalization, we consider the uncertainty in the number of $B^0 \bar{B^0}$ and $B^+ B^-$ pairs: the ratio of $B^+B^-$ to $B^0 \bar{B^0}$ pairs ($f_+/f_0$), $f_+/f_0 = 1.029\pm 0.035$ [@HFAG05] , the mixing parameter ($\chi_d$), $\chi_d = 0.186\pm 0.004$ [@PDG2005], and the measured number of $B \bar{B}$ pairs ($N_{B\bar{B}}$, 1.1%).
= 2pt
----------------------------------- --------- --------- ---------- -------- ------ --------- -------- ---------- -------- ------
Source $q^2<8$ $ 8-16$ $\geq16$ $< 16$ all $q^2<8$ $8-16$ $\geq16$ $< 16$ all
Tracking efficiency 1 1 1 1 1 1 1 1 1 1
$\pi^{0}$ reconstruction – – – – – 2 2 2 2 2
Lepton identification 2.1 2.1 2.1 2.1 2.1 2.1 2.1 2.1 2.1 2.1
Kaon identification 2 2 2 2 2 2 2 2 2 2
$D^* \ell \nu$ calibration 9.3 9.3 9.3 9.3 9.3 9.3 9.3 9.3 9.3 9.3
$Br(X_u \ell \nu)$ in the fitting 0.8 2.4 1.8 1.6 1.4 4.8 3.9 23.8 1.9 7.1
$B\bar{B}$ background shape 1.1 2.2 2.8 1.2 1.3 3.8 2.9 17.0 3.0 6.1
$Br(D^{**} \ell \nu)$ 1.0 1.5 0.2 1.2 0.9 0.5 0.3 2.5 0.3 0.8
$K_L^0$ production rate 0.1 0.3 0.4 0.2 0.3 1.0 0.7 2.9 0.8 1.3
$N_{B\overline{B}}$ 1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1
$\it{f}_+ / \it{f}_0$ 1.7 1.7 1.7 1.7 1.7 1.7 1.7 1.7 1.7 1.7
$\chi_d$ 2.2 2.2 2.2 2.2 2.2 2.2 2.2 2.2 2.2 2.2
exp. total 10.4 10.9 10.8 10.5 10.5 12.1 11.5 31.3 11.1 14.1
FF for signal 0.7 3.8 0.9 2.2 1.8 6.1 3.5 6.8 4.3 3.6
FF for cross-feed 1.8 2.1 1.5 1.9 1.4 0.5 0.7 2.4 0.6 1.0
FF total 1.9 4.3 1.7 2.9 2.3 6.1 3.6 7.2 4.3 3.7
----------------------------------- --------- --------- ---------- -------- ------ --------- -------- ---------- -------- ------
: Summary of systematic errors (%) for ${\mathcal B}(B^0 \to \pi^- / \rho^- \ell^+ \nu)$.[]{data-label="tbl:systematic_pi"}
= 2pt
----------------------------------- --------- --------- ---------- -------- ------ --------- -------- ---------- -------- ------
Source $q^2<8$ $ 8-16$ $\geq16$ $< 16$ all $q^2<8$ $8-16$ $\geq16$ $< 16$ all
Tracking efficiency – – – – – 2 2 2 2 2
$\pi^{0}$ reconstruction 2 2 2 2 2 – – – – –
Lepton identification 2.1 2.1 2.1 2.1 2.1 2.1 2.1 2.1 2.1 2.1
Kaon identification – – – – – 4 4 4 4 4
$D^* \ell \nu$ calibration 9.2 9.2 9.2 9.2 9.2 9.2 9.2 9.2 9.2 9.2
$Br(X_u \ell \nu)$ in the fitting 0.2 3.1 3.0 2.1 1.2 2.0 3.7 20.0 3.0 6.6
$B\bar{B}$ background shape 1.9 5.5 2.7 4.3 3.7 5.3 4.3 16.3 1.5 2.8
$Br(D^{**} \ell \nu)$ 1.3 0.8 0.8 0.9 0.9 0.2 1.6 3.0 0.9 1.4
$K_L^0$ production rate 0.3 1.1 0.6 0.8 0.8 0.3 0.2 1.9 0.1 0.4
$N_{B\overline{B}}$ 1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1 1.1
$\it{f}_+ / \it{f}_0$ 1.7 1.7 1.7 1.7 1.7 1.7 1.7 1.7 1.7 1.7
exp. total 10.1 11.8 10.7 11.0 10.7 12.1 12.2 28.1 11.2 12.9
FF for signal 1.2 0.5 1.3 0.3 0.2 2.1 7.1 3.9 3.7 3.5
FF for cross-feed 0.7 0.8 0.6 0.8 0.6 3.3 1.1 1.0 1.5 1.2
FF total 1.4 0.9 1.4 0.8 0.6 3.9 7.2 4.0 4.0 3.7
----------------------------------- --------- --------- ---------- -------- ------ --------- -------- ---------- -------- ------
: Summary of systematic errors (%) for ${\mathcal B}(B^+ \to \pi^0 / \rho^0 \ell^+ \nu)$.[]{data-label="tbl:systematic_pi0"}
The dependence of the extracted branching fractions on the FF model has been studied by repeating the above fitting procedure with various FF models for the signal mode and also for the cross-feed mode ($B \to \pi \ell \nu \leftrightarrow B \to \rho \ell \nu$). We consider the models listed in Table \[tbl:FFdep\_q2\_Xulnu\]. For the extracted ${\mathcal B}(B \to \pi^- \ell^+ \nu ~(\pi^0 \ell^+ \nu))$, the standard deviation among the models is $<1.7 ~(0.9)$% for $B \to \pi \ell^+ \nu$ and $<1.9 ~(0.5)$% for $B \to \rho \ell^+ \nu$. For ${\mathcal B}(B \to \rho^- \ell^+ \nu ~(\rho^0 \ell^+ \nu))$, the standard deviation is $<2.9 ~(3.6)$% for $B \to \rho \ell^+ \nu$ and $<1.0 ~(1.3)$% for $B \to \pi \ell^+ \nu$. The total error due to FF model dependence is the quadratic sum of the maximum variations with the signal and cross-feed FF models.
Results
=======
Table \[tbl:summary\_br\] summarizes our measurements of the total and partial branching fractions for the four signal modes. Each branching fraction is obtained by taking the simple average of the values obtained from the FF models shown in Table \[tbl:FFdep\_q2\_Xulnu\]. The errors shown in the table are statistical, experimental systematic, and model dependence due to form-factor uncertainties. The obtained branching fractions for $B^0 \to \pi^-/\rho^- \ell^+ \nu$ are consistent with the existing measurements by CLEO [@CLEO2003] and BaBar [@BABAR2005]. The overall uncertainty on our result for $B^0 \to \pi^- \ell^+ \nu$ (17%) is comparable to those on CLEO and BaBar results based on $\nu$-reconstruction. Our results for $B^0 \to \rho^- \ell^+ \nu$ have the smallest uncertainty.
Modes $q^2$ region (GeV$^2/c^2)$ Branching fraction ($\times 10^{-4}$)
----------------------------- ---------------------------- ---------------------------------------
$B^0 \to \pi^- \ell^+ \nu$ Total $1.38 \pm 0.19 \pm 0.14 \pm 0.03$
$ > 16$ $0.36 \pm 0.10 \pm 0.04 \pm 0.01$
$ < 16$ $1.02 \pm 0.16 \pm 0.11 \pm 0.03$
$B^+ \to \pi^0 \ell^+ \nu$ Total $0.77 \pm 0.14 \pm 0.08 \pm 0.00 $
$ > 16$ $0.20 \pm 0.08 \pm 0.02 \pm 0.00 $
$ < 16$ $0.57 \pm 0.12 \pm 0.06 \pm 0.00 $
$B^0 \to \rho^- \ell^+ \nu$ Total $2.17 \pm 0.54 \pm 0.31 \pm 0.08$
$B^+ \to \rho^0 \ell^+ \nu$ Total $1.33 \pm 0.23 \pm 0.17 \pm 0.05$
: Summary of the obtained branching fractions. The errors are statistical, experimental systematic, and systematic due to form-factor uncertainties.[]{data-label="tbl:summary_br"}
Figure \[fig:q2dist\] presents the measured $q^2$ distributions for each signal mode, overlaid with the best fits of FF shapes to the data. To be self-consistent, the shape of a particular FF model is fit to the $q^2$ distribution extracted with the same FF model. The quality of the fit in terms of $\chi^2$ and the probability of $\chi^2$, shown in Table \[tbl:FFdep\_q2\_Xulnu\], may provide one way to discriminate among the models. From our results, the ISGW II model is disfavored for $B \to \pi \ell^+ \nu$.
-- --
-- --
In this work, the $B^0 \to \pi^- \ell^+ \nu / B^+ \to \pi^0 \ell^+ \nu$ and $B^0 \to \rho^- \ell^+ \nu / B^+ \to \rho^0 \ell^+ \nu$ signals are extracted separately, which allows us to test the isospin relations. From the obtained branching fractions and the $B$ meson lifetimes in [@PDG2005], the ratios of decay rates are found to be, $$\begin{aligned}
\frac{\Gamma(B^0 \to \pi^- \ell^+ \nu)}{\Gamma(B^+ \to \pi^0 \ell^+ \nu)}
= (1.92 \pm 0.43 \pm 0.28),\\
\frac{\Gamma(B^0 \to \rho^- \ell^+ \nu)}{\Gamma(B^+ \to \rho^0 \ell^+ \nu)}
= (1.74 \pm 0.53 \pm 0.33),\end{aligned}$$ where the first and second errors are statistical and systematic errors, respectively. Both ratios are found to be consistent with the isospin relations; $\Gamma(B^0 \to \pi^-(\rho^-) \ell^+ \nu) =
2\Gamma(B^+ \to \pi^0(\rho^0) \ell^+ \nu)$.
The obtained branching fractions in Table \[tbl:summary\_br\] can be used to extract $|V_{ub}|$ using the relation, $$|V_{ub}| = \sqrt{ \frac{{\mathcal B}(B \to \pi \ell^+ \nu) }
{{\tilde \Gamma}_{thy}~\tau_{B} } },$$ where ${\tilde \Gamma}_{thy}$ is the form-factor normalization, predicted from theories. In Table \[tbl:summary\_br\], we list the partial branching fractions for $B \to \pi \ell^+ \nu$ decays in the $q^2$ region above 16 GeV$^2/c^2$, where the LQCD calculations are most reliable. The table provides also the results in the region below 16 GeV$^2/c^2$, so that one can deduce $|V_{ub}|$ based on other approaches such as LCSR calculations [@Ball04_pi; @Ball04_rho].
In this paper we calculate $|V_{ub}|$ based on the $B \to \pi \ell^+ \nu$ data in the high $q^2$ region and the form factor predicted by recent unquenched LQCD calculations. Their predictions (${\tilde \Gamma}_{thy}$) for the $q^2 \geq 16$GeV$^2/c^2$ region are ${\tilde \Gamma}_{thy}(B^0 \to \pi^- \ell^+ \nu) = 1.83 \pm 0.50$ ps$^{-1}$ (FNAL) [@FNAL04] and ${\tilde \Gamma}_{thy}(B^0 \to \pi^- \ell^+ \nu) = 1.46 \pm 0.35$ ps$^{-1}$ (HPQCD) [@HPQCD04]. We use $\tau_{B^0} = 1.532\pm 0.009$ ps and $\tau_{B^+} = 1.638\pm 0.011$ ps [@PDG2005], and we use isospin symmetry to relate ${\tilde \Gamma}_{thy}$ for $B^0 \to \pi^-$ and $B^+ \to \pi^0$ transitions. The results for $B^0 \to \pi^- \ell^+ \nu$ and $B^+ \to \pi^0 \ell^+ \nu$ are then averaged, weighted by their respective statistical errors.
Theory ${\tilde \Gamma}_{thy}$(ps$^{-1}$) Mode $|V_{ub}| (\times 10^{-3})$
-------- ------------------------------------ --------------------------------------- -------------------------------------------
FNAL $1.83 \pm 0.50$ $\pi^- \ell^+ \nu$ $3.59 \pm 0.51 \pm 0.20 ^{+0.62}_{-0.41}$
$\pi^0 \ell^+ \nu$ $3.63 \pm 0.70 \pm 0.20 ^{+0.63}_{-0.41}$
$\pi^- \ell^+ \nu + \pi^0 \ell^+ \nu$ $3.60 \pm 0.41 \pm 0.20 ^{+0.62}_{-0.41}$
HPQCD $1.46 \pm 0.35$ $\pi^- \ell^+ \nu$ $4.02 \pm 0.57 \pm 0.22 ^{+0.59}_{-0.41}$
$\pi^0 \ell^+ \nu$ $4.06 \pm 0.78 \pm 0.22 ^{+0.60}_{-0.41}$
$\pi^- \ell^+ \nu + \pi^0 \ell^+ \nu$ $4.03 \pm 0.46 \pm 0.22 ^{+0.59}_{-0.41}$
: Summary of $|V_{ub}|$ obtained from the $B \to \pi \ell^+ \nu$ data in the $q^2 \ge 16$GeV$^2/c^2$ region. The first and second errors are experimental statistical and systematic errors, respectively. The third error stems from the error on ${\tilde \Gamma}_{thy}$ quoted by the LQCD authors.[]{data-label="tbl:summary_vub"}
Table \[tbl:summary\_vub\] summarizes the results, where the first and second errors are the experimental statistical and systematic errors, respectively. The third error is based on the error on ${\tilde \Gamma}_{thy}$ quoted by the LQCD authors. These theoretical errors are asymmetric because we assign them by taking the variation in $|V_{ub}|$ when ${\tilde \Gamma}_{thy}$ is varied by the quoted errors. The values are in agreement with those from inclusive $B\to X_u\ell\nu$ decays [@HFAG_LP05].
To summarize, we have measured the branching fractions of the decays $B\to \pi\ell\nu$ and $B\to\rho\ell\nu$ in $2.75\times 10^8$ $B\bar
B$ events using a method which tags one $B$ in the mode $B\to
D^{(*)}\ell\nu$. Our results are consistent with previous measurements, and their precision is comparable to that of results from other experiments. The ratios of results for neutral and charged $B$ meson modes are found to be consistent with isospin. The partial rates are measured in three bins of $q^2$ and compared with distributions predicted by several theories. From the rate in the region $q^2 \ge 16$ GeV$^2/c^2$ and recent results from LQCD calculations, we extract $|V_{ub}|$: $$\begin{aligned}
|V_{ub}|_{(q^2 \geq 16 ~{\rm GeV}^2/c^2)}^{\pi^- \ell^+ \nu + \pi^0 \ell^+ \nu } & = &
(3.60 \pm 0.41 \pm 0.20 ^{+0.62}_{-0.41}) \times 10^{-3} (\mbox{FNAL LQCD}), \\
|V_{ub}|_{(q^2 \geq 16 ~{\rm GeV}^2/c^2)}^{\pi^- \ell^+ \nu + \pi^0 \ell^+ \nu } & = &
(4.03 \pm 0.46 \pm 0.22 ^{+0.59}_{-0.41}) \times 10^{-3} (\mbox{HPQCD LQCD}).\end{aligned}$$
The experimental precision on these values is 13%, currently dominated by the statistical error of 11%. By accumulating more integrated luminosity, a measurement with errors below 10% is feasible. With improvements to unquenched LQCD calculations, the present method may provide a precise determination of $|V_{ub}|$.
We thank the KEKB group for the excellent operation of the accelerator, the KEK cryogenics group for the efficient operation of the solenoid, and the KEK computer group and the National Institute of Informatics for valuable computing and Super-SINET network support. We acknowledge support from the Ministry of Education, Culture, Sports, Science, and Technology of Japan and the Japan Society for the Promotion of Science; the Australian Research Council and the Australian Department of Education, Science and Training; the National Science Foundation of China and the Knowledge Innovation Program of Chinese Academy of Sciencies under contract No. 10575109 and IHEP-U-503; the Department of Science and Technology of India; the BK21 program of the Ministry of Education of Korea, and the CHEP SRC program and Basic Research program (grant No. R01-2005-000-10089-0) of the Korea Science and Engineering Foundation; the Polish State Committee for Scientific Research under contract No. 2P03B 01324; the Ministry of Science and Technology of the Russian Federation; the Slovenian Research Agency; the Swiss National Science Foundation; the National Science Council and the Ministry of Education of Taiwan; and the U.S.Department of Energy.
[99]{}
N. Cabibbo, Phys. Rev. Lett. 10, 531 (1963); M. Kobayashi and T. Maskawa, Prog. Theor. Phys. [**49**]{}, 652 (1973).
M. Okamoto [*et al.*]{}, Nucl.Phys.Proc.Suppl.140, 461 (2005). E. Gulez [*et al.*]{} (HPQCD Collaboration), hep-lat/0601021 (2006).
J. P. Alexander [*et al.*]{} (CLEO Collaboration), Phys. Rev. Lett. [**77**]{}, 5000 (1996).
J. P. Alexander [*et al.*]{} (CLEO Collaboration), Phys. Rev. [**D61**]{}, 052001 (2000).
S. B. Athar [*et al.*]{} (CLEO Collaboration), Phys. Rev. [**D 68**]{}, 072003 (2003).
B. Aubert [*et al.*]{} (BaBar Collaboration), Phys. Rev. Lett. [**90**]{}, 181801 (2003).
C. Schwanda [*et al.*]{} (Belle Collaboration), Phys. Rev. Lett. [**93**]{}, 131803 (2004).
B. Aubert [*et al.*]{} (BaBar Collaboration), Phys. Rev. [**D 72**]{}, 051102 (2005).
S. Kurokawa and E. Kikutani, Nucl. Instr. and Meth. A [**499**]{}, 1 (2003), and other papers included in this volume.
A. Abashian [*et al.*]{} (Belle Collaboration), Nucl. Instr. and Meth. A [**479**]{}, 117 (2002).
Y. Ushiroda (Belle SVD2 Group), Nucl. Instr. and Meth.A [**511**]{} 6 (2003).
R. Brun [*et al.*]{}, GEANT3.21, CERN Report DD/EE/84-1 (1984).
D. Scora and N. Isgur, Phys. Rev. D [**52**]{}, 2783 (1995).
P. Ball and R. Zwicky, Phys. Rev. D [**71**]{}, 014015 (2005).
P. Ball and R. Zwicky, Phys. Rev. D [**71**]{}, 014029 (2005).
L. Del Debbio, J. M. Flynn, L. Lellouch, and J. Nieves (UKQCD Collaboration), Phys. Lett. B [**416**]{}, 392 (1998).
D. Melikhov and B. Stech, Phys. Rev. D [**62**]{}, 014006 (2000).
F. De Fazio and M. Neubert, JHEP [**9906**]{}, 017 (1999).
Evtgen event generator, http://hep.ucsb.edu/people/lange/EvtGen/.
S. Eidelman et al., Phys. Lett. B 592, 1 (2004) and 2005 partial update for the 2006 edition available on the PDG WWW pages (URL: http://pdg.lbl.gov/).
H. Kakuno [*et al.*]{} (Belle Collaboration), Phys. Rev. Lett [**92**]{}, 071802 (2004); the inclusive branching fraction used in the fitting is based on the partial branching fraction $\Delta {\mathcal B}(B \to X_u \ell \nu;
M_X<1.7$GeV/$c^2, q^2>8$ GeV$^2/c^2$) and a calculation of $f_u$ shown in hep-ex/0408115.
Heavy Flavor Averaging Group, hep-ex/0505100
Heavy Flavor Averaging Group (HFAG), E. Barberio [*et al.*]{}, hep-ex/0603003 and online update at http://www.slac.stanford.edu/xorg/hfag
| {
"pile_set_name": "ArXiv"
} |
**On the spectrum of discrete Schrödinger equation**
**with one-dimensional perturbation[^1]**
**V. V. Borzov$^{\,a}$ E. V. Damaskinsky$^{\,b}$**
[$^a$ Department of Mathematics, St.Petersburg State University of Telecommunications,\
Prospekt Bolshevikov 22-1, St.Petersburg, 193232, Russia; [email protected]]{}
[$^b$ Mathematical Department, VI(IT), Russia, 191123, Zacharievskaya 22, St.Petersburg, Russia; [email protected]]{}
**Abstract**
> We consider the spectrum of the discrete Schrödinger equation with one-dimensional perturbation. We obtain the explicit form of scattering matrix and find the exact condition of absence of singular part of the spectrum. We calculated also the eigenvalue that appears if this condition is not true. In the last part of our paper we give few remarks on the case of two-dimensional perturbations.
Introduction
============
The problem of a change in the spectrum of self-adjoint operator with one-dimensional perturbation in the well-known Friedrichs - Faddeev model [@1],[@2] on a finite segment $[a,b]$ is well studied. Description of this model and references see in the monograph [@3]. Let $v(\lambda,\tau)=\beta u(\lambda)\overline{u(\tau)}$ is the kernel of the integral operator of the perturbation. The function $v(\lambda,\tau)$ satisfies the Hölder condition with index $\alpha_0,$ and $u(a)=u(b)=0.$ It is well-known that in the case of one-dimensional perturbation the absolutely continuous part of the spectrum not changed, and the singular part of the continuous spectrum for arbitrary function $u(\lambda)$ does not only when $\alpha_0>\frac{1}{2}$. It is also known that if the coupling constant $ \beta $ is small, then the discrete spectrum of the perturbed operator from outside the segment $[a,b]$ is missing. However, there is some positive critical value $ \beta_0 $ such that if $\vert \beta\vert> \beta_0 $, then there is one simple eigenvalue $\lambda$ from outside the segment $[a,b]$. In the general case, even for one-dimensional perturbations, provided that $\alpha_0\leq\frac{1}{2}$ the singular spectrum can be quite complicated (see theorem 6.7.10 in [@3]).
In this paper we consider a special case of one-dimensional perturbations in Friedrichs-Faddeev model on the finite segment $[-2,2]$. Namely the case when the kernel of the integral operator of the perturbation takes the form $u(\lambda)=U_{k}(\frac{\lambda}{2})f(\lambda),$ ($k\in Z_{+}$) where $$f(\lambda)=\frac{1}{\pi}\sqrt{1-\frac{\lambda^2}{4}}.$$ and $U_{n}$ are the Chebyshev polynomials of the 2-nd kind. These polynomials satisfy the following recurrence relations $$\label{triv03}
\lambda U_{n}(\ts\frac{\lambda}{2})=U_{n+1}(\frac{\ts\lambda}{2})+ U_{n-1}(\ts\frac{\lambda}{2}),\quad n\geq 1,
\qquad U_{0}(\ts\frac{\lambda}{2})=1,\qquad U_{1}(\ts\frac{\lambda}{2})=\lambda.$$ In our case the function $u(\lambda)$ equals to zero at the ends of the segment and satisfies the Hölder condition with index $\alpha_0=\frac{1}{2}.$
Taking into account possible applications of our results it is convenient to consider the model as a discrete Schrödinger operator with a local one-dimensional perturbation. This model is found in the study of some problems of atomic physics (see [@7]), and also in the investigation of non-equivalent representations algebra of generalized Chebyshev oscillator (see [@8], [@8a]).
Let us turn to a precise formulation of the problem. We give the definition of the considered Schrödinger operator following § 4.1 of the monograph [@4]. Take as a starting point, the Jacobi matrix $\mathcal{J}_{k}^{(\beta)}=\{\mathnormal{j}_{i,n}^{(\beta,k)}\}_{i,n=0}^{\infty},$ where $$\begin{gathered}
\mathnormal{j}_{i,n}^{(\beta,k)}=
\left\{\begin{aligned}1,&\quad\text{if}\quad |i-n|=1 ,\\
\beta\delta_{nk},&\quad\text{if}\quad i=n ,\\
0,&\quad \text{if}\quad |i-n|>1,\end{aligned}\right.\end{gathered}$$ $\delta_{nk}$ is the Kronecker delta and $\beta$ is a real number.
Let $\mathcal{H}_{k}=\mathrm{L}_{2}(\mathcal{R};\mu_{k})$ be a separable Hilbert space where $\mu_{k}$ is a Borel probability measure, such that $$\nu_{n}=\int_{-\infty}^{+\infty}\lambda^{n} \mu_{k}(d\lambda)< \infty,\quad n\geq 0,$$ and $\nu_{0}=1.$ We consider $\mathcal{J}_{k}^{(\beta)}$ as the matrix of linear operator $H_{k}^{(\beta)}$ in $\mathcal{H}_{k}$ defined on the orthonormal basis $\{\phi_{n}^{(k)}(\lambda)\}_{n=0}^{\infty}$ of the space by relations $$\begin{gathered}
\label{triv1}
H_{k}^{(\beta)}\phi_{n}^{(k)}\!=\!\phi_{n+1}^{(k)}\!+\!\phi_{n-1}^{(k)}\!+\!\beta\delta_{nk}\phi_{n}^{(k)},
\quad n=1,2,..,\\
H_{1}^{(\beta)}\phi_{0}^{(k)}=\beta\delta_{0k}\phi_{0}^{(k)}+ \phi_{1}^{(k)}.
\label{triv1`}\end{gathered}$$ The symmetric operator $H_{k}^{(\beta)}$ is defined on the set of finite linear combinations of basis vectors. The set is dense in the space $\mathcal{H}_{k}.$ It is known [@4] that the deficiency indices of the operator $H_{k}^{(\beta)}$ are equal to (0,0). Hence its closure is a self-adjoint operator with a simple spectrum in $\mathcal{H}_{k}$. We denote this operator by the same symbol $H_{k}^{(\beta)}$. The operator $H_{k}^{(\beta)}$ defined by (\[triv1\]), (\[triv1‘\]) is considered discrete Schrödinger operator with a local one-dimensional perturbation.
Further, denote by $\mathcal{J}_0$ the Jacobi matrix $\mathcal{J}_0=\mathcal{J}_{k}^{(0)}$ and by $H_{0}$ the corresponding self-adjoint operator $H_{0}=H_{k}^{(0)}$.
In the following we will assume that $\mathcal{H}_{k}$ is a space of the spectral representation, i.e. the operator $H_{k}^{(\beta)}$ in the space $\mathcal{H}_{k}$ is the operator of multiplication by the variable $\lambda$.
It is known [@9] that the operator $H_{0}$ in spectral representation defined on the Hilbert space $\mathcal{H}_{0}=\mathrm{L}_{2}(\mathcal{R};\mu_{0})$ with $$\label{4}
d\mu_{0}(\lambda)=\frac{1}{\pi}
\left\{\begin{aligned}\sqrt{1-\frac{\lambda^2}{4}}d\lambda,&\quad\text{if}\quad |\lambda|\leq 2,\\
0,\qquad&\quad \text{if}\quad |\lambda|>2,\end{aligned}\right.$$ The Chebyshev polynomials of the 2-nd kind $\{U_{n}(\frac{\lambda}{2})\}_{n=0}^{\infty}$ form an orthonormal basis in the space $\mathcal{H}_{0}$. “Perturbed” operator $H_{k}^{(\beta)}$ defined on the Hilbert space $\mathcal{H}_{k},$ where the polynomials $\{\phi_{n}^{(k)}(\lambda))\}_{n=0}^{\infty}$ form an orthonormal basis. These polynomials satisfy the following recurrence relations $$\label{triv001}
\lambda\phi_{n}^{(k)}(\lambda)=\phi_{n+1}^{(k)}(\lambda)+\beta\delta_{nk}\phi_{n}^{(k)}(\lambda)+
\phi_{n-1}^{(k)}(\lambda),\quad n\geq 1,$$ $$\label{triv002}
\phi_{0}^{(k)}(\lambda)=1,\quad\phi_{1}^{(k)}(\lambda)=(\lambda-\beta\delta_{0k}).$$
[**Our aim is to study the singular part of the spectrum of self-adjoint “perturbed” Hamiltonian $H_{k}^{(\beta)}$, ($\beta\in \mathbb{R}$) for arbitrary $k\in {Z_{+}}.$**]{} This operator occurs after adding one-dimensional perturbation to the “free” Hamiltonian $H_{0}$ (corresponding to the case $\beta=0$). We use a new representation of the resolvent $ R_{0}(z)=(H_{0}-z)^{-1}$ of self-adjoint unperturbed operator $H_{0}$ in $\mathcal{H}_{0}$ . By (.,.) we denote the scalar product in the Hilbert space $\mathcal{H}_{0}$ . Computing the roots of the denominator of the resolvent $$D(z)= 1 + \beta(R_{0}(z)U_k,U_k)$$ on the real axis, we obtain the following results:
1\. Exact condition of the absence of singular spectrum on the continuous spectrum (on the segment $[-2,2]$).
2\. The critical value $\beta_{0}$ of the coupling constant $\beta$ equals to $\frac{1}{k+1}.$
3\. The boundary points $\pm 2$ of the continuous spectrum are resonances.
4\. The explicit dependence $\lambda=\lambda(\beta)$ of the coupling constant $\beta$ the eigenvalue $\lambda$ which is outside the segment $[-2,+2]$.
5\. The explicit form of the scattering matrix.
Besides, we build the orthogonality measure $\mu_{k}$ of the polynomials $\phi_{n}^{(k)}(t).$
The orthogonality measure $d\mu_{k}$
====================================
We consider Jacobi matrix $\mathcal{J}_{k}^{(\beta)}$ and the matrix $$B_k=\{\mathnormal{j}_{i,m}^{(k)}\}_{i,m=1}^{\infty},\quad \mathnormal{j}_{i,m}^{(k)}=\beta \delta_{i,k}\delta_{m,k}$$ for $\beta\in {\mathbb{R}}.$ Denote by $V_k$ the one-dimensional self-adjoint operator corresponding to the matrix $B_k$ in a Hilbert space $\mathcal{H}_{0}.$ We consider the “perturbed” operator $H_{k}^{(\beta)},$ corresponding to the Jacobi matrix $\mathcal{J}_{k}^{(\beta)}$ in $\mathcal{H}_{0},$ as a sum of $H_{0}$ and $V_k$: $$H_{k}^{(\beta)}=H_{0}+V_k,\quad V_k=\beta (.,U_{k})U_{k},\quad\quad
\Vert U_{k}\Vert=1,\quad \beta=\overline{\beta}.$$ We will compute the density $\mu_k^{\prime}(\lambda)$ of a measure $\mu_{k}$. Let $E_{0}(\lambda)\!=\!E_{0}(-\infty,\lambda)$ and $E_{k}^{(\beta)}(\lambda)\!=\!E_{k}^{(\beta)}(-\infty,\lambda)$ — generating functions of the spectral measures $E_0$ and $E_{k}^{(\beta)}$ of self-adjoint operators $H_0$ and $H_{k}^{(\beta)}.$ Let’s denote by $\it{L_k},\it{L}_{k}^{(\beta)}$ — linear spans of the sets $\{E_{0}(X)U_k\}$ and $\{E_{k}^{(\beta)}(X)U_k\}$, respectively. Then restrictions $\tilde{H}_{0},\,\, \tilde{H}_{k}^{(\beta)}\!,$ of the operators $H_{0}$ and $H_{k}^{(\beta)}$ to the subspace $\it{L_0} =\it{L}_{k}^{(\beta)}$ of Hilbert space $\mathcal{H}_{0}$ have a simple spectrum. Therefore, they are multiplication operators on $\lambda$ in $\mathrm{L}_{2}(\textbf{R};\varrho_{0}^{(k)})$ and $\mathrm{L}_{2}(\textbf{R};\varrho_{k}),$ where the measure $\varrho_{0}^{(k)}$ (we denote $\varrho_{0}^{(0)}$ by $\varrho_{0}$) and $\varrho_{k}$ are defined by equalities $$\varrho_{0}^{(k)}(.)\!=\! (E_{0}(.)U_k,U_k),\quad\!
\varrho_{k}(.)\!=\!( E_{k}^{(\beta)}(.)U_k,U_k).$$ We introduce also the measure $\varrho_{k,0}$ $$\begin{aligned}
\varrho_{k,0}(.)= ( E_{k}^{(\beta)}(.)U_0,U_0).\end{aligned}$$ It is known that $$\varrho_{0}(.)=\mu_0(.),\quad \varrho_{k,0}(.)=\mu_{k}(.).$$ Then we have $$\begin{aligned}
\label{triv104}
{\varrho_{0}^{(k)}}^{\prime}(\lambda)={\vert U_{k}(\ts\frac{\lambda}{2})\vert}^2\mu_{0}^{\prime}(\lambda), \\[4pt]
{\varrho_{k}}^{\prime}(\lambda)={\vert U_{k}(\ts\frac{\lambda}{2})\vert}^2\mu_{k}^{\prime}(\lambda),
\label{triv105}\end{aligned}$$ where the measure $\mu_0$ defined by . We consider the resolvent $R_k(z)=(H_{k}^{(\beta)}-z)^{-1}$ of self-adjoint operator $H_{k}^{(\beta)}$ in $\mathcal{H}_{0}$, and denote by $D_k(z)$ the following expression $$D_k(z)=1+\beta (R_0(z)U_{k},U_{k}).$$ From ([@3]) we know that $D_k(\lambda+i 0)\neq 0$ for almost all $\lambda$ and $$\label{triv77`}
{\rho_0^{(k)}}^{\prime}(\lambda)={\vert D_k(\lambda+i0)\vert}^2 \rho_k^{\prime}(\lambda).$$ Then we get from (\[triv104\]),(\[triv105\]) , (\[triv77‘\]) for almost all $\lambda $ $$\label{triv10}
\mu_k^{\prime}(\lambda)=\frac{\mu_0^{\prime}(\lambda)}{{\vert D_k(\lambda+i 0)\vert}^2}.$$ From (\[triv77‘\]) and (\[triv10\]) it follows that absolutely continuous parts of the measures $\varrho_{0,a}^{(k)}$ and $\varrho_{k,a}$ are equivalent, i.e. $\sigma_{a}(H_{k}^{(\beta)})=\sigma_{0,a}=[-2,2].$
By the Privalov theorem [@5], using the well-known expression for the resolvent of the operator $H_{0}$ via the Cauchy-Stieltjes integral with respect to the spectral measure, we have $$\begin{gathered}
(R_0(\lambda\pm i 0)U_k,U_k)=\pm\pi i
{\rho_0^{(k)}}^{\prime}(\lambda)+I^{(k)}(\lambda)\\
I^{(k)}(\lambda)= v.p.\int_{-\infty}^{+\infty}(t-\lambda)^{-1}
{\rho_0^{(k)}}^{\prime}(t)\,dt,\end{gathered}$$ We get from (\[triv104\]) that $$\label{triv505}
I^{(k)}(\lambda)= v.p.\int_{-2}^{+2}(t-\lambda)^{-1}
\left(U_{k}(\ts\frac{t}{2})\right)^2\mu_{0}^{\prime}(t)\,dt,$$
Using recurrence relations , for the Chebyshev polynomials one can calculate the integral $I^{(k)}(\lambda)$[^2] $$\begin{gathered}
I^{(k)}(\lambda)=
\left\{\begin{aligned}
b_{k}(\lambda)U_{k}(\ts{\frac{\lambda}{2}}), & \quad\text{if}\quad |\lambda|\leq 2,\\
(-1)^{k}a^{k+1}(\lambda)U_{k}(\ts{\frac{\lambda}{2}}),& \quad\text{if}\quad |\lambda|>2,\end{aligned}\right.
\label{triv30}\end{gathered}$$ where $$b_{k}(\lambda)=-\frac{\lambda}{2}\,U_{k}(\ts\frac{\lambda}{2})+U_{k-1}(\ts\frac{\lambda}{2}),$$ and $$\label{triv05}
a(\lambda)=-\frac{2}{\lambda}\,\,\frac{1}{1+\sqrt{1-\frac{4}{\lambda^2}}}\,.$$ We have for $|\lambda|\leq 2 $ $$D_k(\lambda+i 0)=1+\beta\left(\pm\pi i\mu_{0}^{\prime}(\lambda)U_{k}^{2}(\ts\frac{\lambda}{2})+
\widetilde{L}_k(\lambda)\right),$$ $$\widetilde{L}_k(\lambda)=-\frac{\lambda}{2}\,U_{k}^{2}(\ts\frac{\lambda}{2})+
U_{k}(\ts\frac{\lambda}{2})\,U_{k-1}(\ts\frac{\lambda}{2}).
\label{triv130}$$ In the case when $|\lambda|>2$ we have $$D_k(\lambda+i 0)\!=\!1+\beta(-1)^{k}a^{k+1}(\lambda)U_{k}(\ts\frac{\lambda}{2}).$$ Using the identity[^3] $$U_{k}^{2}\big(\ts{\frac{\lambda}{2}}\big)+U_{k-1}^{2}\big(\frac{\lambda}{2}\big)-
\lambda U_{k-1}\big(\frac{\lambda}{2}\big)U_{k}\big(\frac{\lambda}{2}\big)=1,
\label{triv405}$$ we get $$\label{triv45}
{\vert D_k(\lambda+i0)\vert}^2\!=\!
\left\{\!\begin{aligned}
1\!+\!\beta(\beta\! -\! \lambda)A_{k}\!+\!
2\beta B_k,\quad &|\lambda|\leq 2,\\
(1+(-1)^{k}\beta a^{k+1}(\lambda)A_k,\quad & |\lambda|>2.
\end{aligned}\right.$$ where $$A_k=U_{k}^{2}(\ts\frac{\lambda}{2}), \quad B_k=U_{k}(\frac{\lambda}{2})U_{k-1}(\ts\frac{\lambda}{2}).$$ Finally, from \[4\], (\[triv10\]) and (\[triv45\]) it follows that $$\mu_{k}^{\prime}(\lambda)=
\frac{1}{\pi}
\left\{\begin{aligned}
\frac{\sqrt{1-\frac{\lambda^2}{4}}}
{1+\beta(\beta - \lambda)A_{k}+2\beta B_k},&\quad\text{if}\quad |\lambda|\leq 2,\\
0,\qquad\qquad\qquad &\quad\text{if}\quad |\lambda|>2.
\end{aligned}\right.$$
Scattering Matrix for ($H_{0},H_{k}^{(\beta)}$)
===============================================
Now we turn to the scattering matrix. According to theorem 6.7.3 ([@3]) the scattering matrix $S(\lambda)$ for a pair of self-adjoint operators $H_{0}$ and $H_{k}^{(\beta)}$ is calculated for almost all $\lambda\in\widehat{\sigma}_0$ by the following formula $$\label{triv14`}
S^{(k)}(\lambda)=I(\lambda)- 2\pi\beta i D_{k}^{-1}(\lambda+i 0)
\langle .,\widetilde{U}_{k} (\frac{\lambda}{2})\rangle_{h(\lambda)}
\widetilde{U}_{k} (\frac{\lambda}{2}),$$ Here $\widehat{\sigma}_0=[-2,2]$ is the core of the spectrum of the operator $H_{0}$ (the minimal Borel support of the spectral measures $E_0$). By $\langle ., .\rangle_{h(\lambda)}$ we denote the scalar product in “infinitesimal subspace” $h(\lambda)$ of a direct integral which corresponds to the absolutely continuous part of the operator $H_{0}$. Namely, we denote by $H_{0}^{(a)}$ the restriction of the operator $H_{0}$ on the absolutely continuous subspace $\mathcal{H}_{0}^{(a)}$ of the operator $H_{0}$ and consider the decomposition of the subspace $\mathcal{H}_{0}^{(a)}$ into a direct integral (see [@3] for details) $$\mathcal{H}_{0}^{(a)}\longrightarrow \int_{\widehat{\sigma}_0}^{}\oplus h(\lambda)\,d\lambda.
\label{triv15}$$
The element $\widetilde{U}_{k}(.)$ in the formula (\[triv14‘\]) is the representative element of $U_k$ in the decomposition (\[triv15\]). In our case $$\widetilde{U}_{k}(\ts\frac{\lambda}{2})=\ds\frac{1}{\sqrt{\pi}}\,
\sqrt[4]{1-\ds\frac{t^2}{4}}\,U_{k}(\ts\frac{\lambda}{2}) \quad\text{if}\quad
|\lambda|\leq 2.
\label{triv16`}$$ Substituting (\[triv16‘\]) in (\[triv14‘\]), we get for almost all $\lambda\in [-2,2]$ $$S^{(k)}= I-\widetilde{I}, \quad |\lambda|\leq 2,$$ where $$\widetilde{I}=
\frac{2\beta\sqrt{\pi}\,i\,\sqrt[4]{1-\frac{{\lambda}^2}{4}}\,U_{k}(\frac{\lambda}{2})}
{1+\beta(i\,\sqrt{1-\frac{\lambda^2}{4}}\,
U_{k}^{2}(\frac{\lambda}{2})+\widetilde{L}_k(\lambda))}\langle .,\widetilde{U}_{k}\rangle_{h(\lambda)}$$ and $\widetilde{L}_k$ is defined by (\[triv130\]).
Point spectrum of the operator $H_{k}^{(\beta)}$
================================================
According to 6.7.6 [@3], if the point $ \lambda_0 $ belongs to the point spectrum $\sigma(H_{k}^{(\beta)})$ on the segment $[-2,2]$, then it is a solution of the equation $${\vert D_k(\lambda+i0)\vert}^2 = 0.
\label{triv49}$$
According to (\[triv45\]), we rewrite this equation in the form $$1+\beta(\beta - \lambda)U_{k}^{2}(\ts\frac{\lambda}{2})+
2\beta U_{k}(\ts\frac{\lambda}{2})U_{k-1}(\ts\frac{\lambda}{2})=0.$$ This is a quadratic equation with respect to the variable $\beta$: $$U_{k}^{2}(\ts\frac{\lambda}{2}) \beta^{2}-U_{k} (\ts\frac{\lambda}{2})
\left(\lambda U_{k}(\ts\frac{\lambda}{2})-2U_{k-1}(\ts\frac{\lambda}{2})\right)\beta +1=0.
\label{triv59}$$ For any $\lambda$ satisfying the inequality $U_{k}(\frac{\lambda}{2})\neq 0$ it is easy to obtain the solution of the quadratic equation $$\beta_{\pm}(\lambda)=\frac{\lambda U_{k}(\frac{\lambda}{2})-
2U_{k-1}(\frac{\lambda}{2})}{2U_{k}(\frac{\lambda}{2})}
\pm \sqrt{\frac{\lambda^2}{4}-1}.$$ Since $\beta\in \mathbb{R}$, from (\[triv59\]) it is clear that if $
\vert \lambda \vert < 2 $ solutions of equation (\[triv49\]) does not exist. Hence, we proved the absence of the point spectrum of the operator $H_{k}^{(\beta)}$ in the interval (-2,2).
Consider the boundary points $\pm 2$. Those numbers are solutions of the equation (\[triv59\]). More precisely, since $$\label{triv008}
U_{k}(1)=k+1, \quad U_{k}(-1)=(-1)^{k}(k+1),$$ then $$\beta_{\pm}(2)=\frac{1}{k+1}, \quad \beta_{\pm}(-2) =-\frac{1}{k+1}.$$ Now we will show that points $\pm 2$ are not eigenvalues of the operator $H_{k}^{(\pm \frac{1}{k+1})}.$
Using recurrence relations (\[triv001\]), (\[triv002\]) it is easy to show that the polynomials $\phi_{n}^{(k)}$ are calculated from the following formulas[^4] $$\begin{gathered}
\phi_{0}^{(k)}=U_{0},\, \phi_{1}^{(k)}=U_{1}, ...,\phi_{k}^{(k)}=U_{k},\quad\nonumber\\[3pt]
\phi_{k+s}^{(k)}=U_{k+s}-\beta U_{k-1+s}-\beta U_{k-3+s}-
\ldots -\beta U_{k+1-s},\qquad 1\leq s\leq k+1,\label{t25}\\[3pt]
\phi_{n}^{(k)}=U_{n}-\beta U_{n-1}-\beta U_{n-3}-
\ldots -\beta U_{n-(2k+1)},\qquad n\geq 2k+2.\nonumber\end{gathered}$$
From and it follows that $$\begin{gathered}
\sum_{n=0}^{\infty}{\vert \phi_{n}^{(k)}(\pm 2)\vert}^2\geq
\sum_{n=2k+2}^{\infty}{\vert \phi_{n}^{(k)}(\pm 2)\vert}^2 =
\infty,\end{gathered}$$ since for $n\geq 2k+2$ we have $$\left|\phi_{n}^{(k)}(\pm 2)\right|^2=(k+1)^2.$$ Thus the boundary points $\pm 2$ of the continuous spectrum are not eigenvalues of the operator $H_{k}^{(\pm 1/{k+1})}.$
Now we have to calculate the eigenvalues $\lambda(\beta)$ of the operator $H_{k}^{(\beta)}$ for $\vert\beta\vert > \frac{1}{k+1},$ i.e. outside the segment $[-2,+2]$. For this we need to find the solution of equation (\[triv49\]) outside of segment $[-2,+2]$. Taking into account (\[triv45\]), the equation (\[triv49\]) can be written outside of segment $[-2,+2]$ in the following way $$\label{triv109}
1+(-1)^{k}\beta a^{k+1}(\lambda)U_{k}(\ts\frac{\lambda}{2})=0.$$ As can be seen from (\[triv05\]), the function $a^{k+1}(\lambda)\neq 0$ for $|\lambda|>2$. Then we can find a solution of the equation (\[triv109\]) $$\beta=(-1)^{k+1}\,\frac{-\lambda\left(1+\sqrt{1-\frac{4}{\lambda^2}}\,\right)^{k+1}}{2}
\,\,\frac{(-\lambda/2)^k}{U_{k}(\lambda/2)}.$$ It is easy to see that $$\beta(2)=\frac{1}{k+1}, \quad \beta(-2) =-\frac{1}{k+1},$$ and the function $ \vert\beta(\lambda)\vert $ is monotonically increasing from $\frac{1}{k+1}$ to $\infty$ if $|\lambda|$ increases monotonically from $2$ to $\infty$. Therefore the function $\beta(\lambda)$ has a unique inverse function $\lambda=\lambda(\beta)$ and $\vert\beta(\lambda)\vert >\frac{1}{k+1}$ as $|\lambda|>2.$ The function $\lambda=\lambda(\beta)$ is the sought eigenvalue.
As shown above when $\vert\beta\vert\leq \frac{1}{k+1}$, the operator $H_{k}^{(\beta)}$ has no point spectrum on the segment $[-2,+2]$. Besides, for any $\beta$ such that $\vert\beta\vert > \frac{1}{k+1}$ there is some eigenvalue $ \lambda(\beta)$ of the operator $H_{k}^{(\beta)}$ on the interval $(2,+\infty)$ (if $\beta > \frac{1}{k+1}$) or on the interval $(-\infty,-2)$ (if $\beta < - \frac{1}{k+1}$).
This means that the boundary points $\pm 2$ of the continuous spectrum are resonances of operators $H_{k}^{(\pm\frac{1}{k+1})},$ respectively.
Two-dimensional perturbation: Case $\vec{k}=(1,2)$
==================================================
Now we consider a two-dimensional perturbation of the operator $H_{0},$ which is determined by the Jacobi matrix $$B_{1,2}=\{\mathnormal{j}_{i,m}^{(1,2)}\}_{i,m=1}^{\infty},\,\,
\mathnormal{j}_{i,m}^{(1,2)}\!=\beta_{1}\delta_{i 1}\delta_{m 1}+
\beta_{2}\delta_{i 2}\delta_{m 2}$$ for $\beta_1,\beta_{2}\in {\textbf{R}}.$ Denote by $V_{1,2}$ two-dimensional self-adjoint operator, corresponding to the matrix $B_{1,2}$ in the Hilbert space $\mathcal{H}_{0}.$ In addition, we consider the Jacobi matrix $$\mathcal{J}_{1,2}^{(\beta_1,\beta_2)}=\{\mathnormal{j}_{i,m}^{(\beta_1,\beta_2)}\}_{i,m=1}^{\infty},$$ where $$\begin{gathered}
\mathnormal{j}_{i,m}^{(\beta_1,\beta_2)}\!=\!
\left\{\begin{aligned}1,&\quad\text{if}\quad |i-m|=1 ,\\
\beta_{1}\delta_{i1}\delta_{m1}+\beta_{2}\delta_{i2}\delta_{m2},
&\quad\text{if}\quad |i-m|\neq1 .\end{aligned}\right.\end{gathered}$$
Denote by $\Phi_{(1,2)}=\{\phi_{n}^{(1,2)}(t)\}_{n=0}^{\infty}$ the set of Jacobi polynomials related with the matrix $\mathcal{J}_{1,2}^{(\beta_1,\beta_2)}.$ The polynomials $ \Phi_{(1,2)}$ are defined the following recurrence relations $$\begin{gathered}
t\phi_{n}^{(1,2)}(t)=\phi_{n+1}^{(1,2)}(t)+a_{n}\phi_{n}^{(1,2)}(t)+b_{n-1}
\phi_{n-1}^{(1,2)}(t),\\
\phi_{0}^{(1,2)}=1,\,n\geq 0,\\
b_n=1-\delta_{n,0},\, a_0=\beta_1,\, a_1=\beta_2,\, a_n=0,\, n\geq 2.\end{gathered}$$ The Jacobi matrix $\mathcal{J}_{1,2}^{(\beta_1,\beta_2)}$ corresponds to the operator $H_{1,2}^{(\beta_1,\beta_2)}$ $$\begin{gathered}
H_{1,2}^{(\beta_1,\beta_2)}=H_{0}+V_{1,2},\\
V_{1,2}=\beta_1 (.,U_0)U_0+\beta_2 (.,U_1)U_1.\end{gathered}$$ The polynomials $\Phi_{(1,2)}$ form an orthonormal in $\mathcal{H}_{12}=\mathrm{L}_{2}(\textbf{R};d\mu_{1,2}(t)).$ Just as it was done previously, we can calculate the measure $\mu_{1,2}$ $$\begin{gathered}
\mu_{1,2}^{\prime}(t)=
\left\{\begin{aligned}\frac{\frac{1}{\pi}\sqrt{1-\frac{t^2}{4}}}{Q},&\quad |t|\leq 2,\\
0,\qquad &\quad |t|>2,\end{aligned}\right.\end{gathered}$$ $$Q=\beta_1^{2}+(1-\beta_{1}\beta_2)^2+
t\left(2\beta_{2}-\beta_1(1+\beta_{1}\beta_2+2\beta_2^2)\right)+
t^{2}\beta_{2}(\beta_{2}+2\beta_1) -t^{3}\beta_2.$$
[**Remarks** ]{}
1\. Note that if $\beta_2=0$, we have $\mu_{1,2}^{\prime}(t)=\mu_{1}^{\prime}(t)$ and if $\beta_1=0$, we have $\mu_{1,2}^{\prime}(t)=\mu_{2}^{\prime}(t).$
2\. In contrast to the case of one-dimensional perturbations, for two-dimensional case we can’t say that the orthogonality measure $\mu_{1,2}$ for polynomials $\Phi_{(1,2)}(t)=\{\phi_{n}^{(1,2)}(t))\}_{n=0}^{\infty}$ in the Hilbert space $\mathcal{H}_{12}=\mathrm{L}_{2}(\textbf{R}; d\mu_{1,2}(t))$ coincides with absolutely continuous part of the measure $\mu_{1,2}$ for some real $\beta_1\neq 0,\beta_2\neq 0$.
3\. However, when $\vert\beta_1\vert\leq 1,\quad\beta_2=0$ or $\vert\beta_2\vert\leq\frac{1}{2},\quad\beta_1=0$, and $\beta_1,\beta_2\in {\textbf{R}},$ the orthogonality measure $\mu_{1,2}$ of the polynomials $\Phi_{(1,2)}(t)=\{\phi_{n}^{(1,2)}(t))\}_{n=0}^{\infty}$ in $\mathcal{H}_{k}$ coincides with absolutely continuous part of the measures $\mu_{1,2}$.
Acknowledgement {#acknowledgement .unnumbered}
===============
EVD grateful to RFBR for financial support under the grant 15-01-03148.
[9]{}
K.O. Friedrichs, [*Perturbation of Spectra in Hilbert Space*]{}, Amer. Math. Soc., Providence, RI, 1965.
L.D. Faddeev, [*On Friedrichs Model in Theory of Perturbation of continious Spectrum*]{}, Proceedings Math. Inst. of Steklov, 73 ,1964.
D.R. Yafaev, [*Mathematical scattering theory: General theory*]{}, vol. 105 Transl. Math. Monographs - Amer. Math. Soc., Providence, RI, 1992.
N.I. Akhiezer, [*The Classical Moment Problem and Some Related Question in Analysis*]{}, English translation, Oliver and Boyed, Edinburgh, 1965.
I.I. Privalov, [*Boundary properties of analytic functions*]{}(in Russian),- Gos. Izdat. Tehn.-Teor. Lit., Moscow - Leningrad, 1950, German transl. I.I. Priwalow, [*Randeigenschaften analytischer Funktionen*]{}, 1956 - VEB Deutscher Verlag, Berlin., 1965.
Yu.N. Demkov, V. N. Ostrovskii, [*Zero-range potentials and their applications in atomic physics*]{}, Plenum Press 1988 (english translation of russian 1975 original).
V.V. Borzov, E.V. Damaskinsky, [*Composite model for generalized Chebyshev oscillator*]{}, Proceedings of the “DAYS ON DIFFRACTION?2010”.
V.V. Borzov, E.V. Damaskinsky, [*Invariance of the generalized oscillator under linear transformation of the related system of orthogonal polynomials*]{}, Theor. Math. Phys. (in press).
T.S.Chihara, [*An Introduction to Orthogonal Polynomials*]{}, Gordon and Breach, New York, 1978.
I.S. Gradshteyn, I.M. Ryzhik, [*Table of Integrals, Series, and Products*]{}, Seventh Edition, Academic Press (Translated from Russian by Scripta Technica, Inc.) 2007.
Appendix 1. The proof of equation when $\vert\lambda\vert >2$ {#appendix-1.-the-proof-of-equation-when-vertlambdavert-2 .unnumbered}
=============================================================
Here we prove the relation $$\label{n01}
I^{(k)}(\lambda)=\int_{-2}^{+2}\frac{1}{(t-\lambda)}
\left(U_{k}(\ts\frac{t}{2})\right)^2\mu_{0}^{\prime}(t)\,dt
=(-1)^{k}a^{k+1}(\lambda)U_{k}(\ts{\frac{\lambda}{2}}), \quad\text{if}\quad |\lambda|>2.$$ For this aim we on the first step proved the additional relation $$\label{n02}
I^{(k)}(\lambda)=\int_{-2}^{+2}(\frac{1}{(t-\lambda)}
\left(U_{k}(\ts\frac{t}{2})\right)^2\mu_{0}^{\prime}(t)\,dt
=a\,\,\frac{1-a^{2(k+1)}}{1-a^2}, \quad\text{if}\quad |\lambda|>2,$$ where $a=a(\lambda)$ defined by as $$\label{n03}
a(\lambda)=-\frac{2}{\lambda}\,\,\frac{1}{1+\sqrt{1-\frac{4}{\lambda^2}}}\,,$$ from which it follows that $$\label{n04}
\lambda=-\frac{a^{2}+1}{a}.$$
On the second step we demonstrate that right hand sides of the relations and are equal, that proves the formula .
We will prove formula by induction. First we will check the validity of for $k=0$ and $k=1$. After the change of variable $t=2\cos x$ in the integral $I^{(0)}(\lambda)$ we obtain $$I^{(0)}(\lambda)=\int_{-2}^{+2}\frac{1}{(t-\lambda)} \mu_{0}^{\prime}(t)\,dt=
\frac{2}{\pi}\int_{0}^{\pi}\frac{\sin^{2}x}{-\lambda+2\cos x}\,dx.$$ According to eq. (3.644.4) in [@10] at $p=-\lambda, q=2$ and the formula , one can show that the last integral is equal to $$\label{n05}
I^{(0)}(\lambda)=a(\lambda).$$
Next from the recurrence relations we obtain $$\label{n06}
U_{k}^2={\lambda}^{2}U_{k-1}^2-2\lambda U_{k-1}U_{k-2}+U_{k-2}^2,\quad k\geq 1.$$ Substituting in the right hand side of equality instead of $U_{k}^2$ we have ($k\geq 2$) $$I^{(k)}(\lambda)=\int_{-2}^{+2}\frac{t^{2}}{(t-\lambda)}U_{k-1}^2\, \mu_{0}^{\prime}(t)\,dt-
2\int_{-2}^{+2}\frac{t}{(t-\lambda)}U_{k-1}U_{k-2}\, \mu_{0}^{\prime}(t)\,dt+I^{(k-2)}(\lambda).$$ Using (\[triv405\]) we replace the product $-tU_{k-1}U_{k-2}$ in this formula by the expression $$1-U_{k-1}^2-U_{k-2}^2,$$ and obtain $$I^{(k)}=\int_{-2}^{+2}\frac{t^{2}}{(t-\lambda)}U_{k-1}^2 \,\mu_{0}^{\prime}(t)\,dt+
2a(\lambda)-2I^{(k-1)}-I^{(k-2)},\quad k\geq 2.$$ From orthonormality of Chebyshev polynomials with respect to $d\mu_{0}$ and that the function $tU_{k-1}^{2}(t)\mu_{0}^{\prime}(t)$ is odd it follows that the first integral in the right hand side of this equality is equal to $$\begin{gathered}
\int_{-2}^{+2}tU_{k-1}^2 \,\mu_{0}^{\prime}(t)\,dt+\lambda\int_{-2}^{+2}U_{k-1}^2 \,\mu_{0}^{\prime}(t)\,dt+
\int_{-2}^{+2}\frac{\lambda^{2}}{(t-\lambda)}U_{k-1}^2 \,\mu_{0}^{\prime}(t)\,dt=\nonumber\\
0+\lambda+\lambda^{2}I^{(k-1)}=\lambda + \lambda^{2}I^{(k-1)}.\end{gathered}$$ Then we have $$\label{n07}
I^{(k)}=\lambda+2a+(\lambda^2 -2)I^{(k-1)}-I^{(k-2)},\quad k\geq 2.$$ From we get $$\lambda+2a=\frac{a^{2}-1}{a},\quad \lambda^2 -2=\frac{a^{4}+1}{a^2}.$$ Using the relations we can rewrite in the form $$\label{n08}
I^{(k)}=\frac{a^{2}-1}{a}+\frac{a^{4}+1}{a^2}I^{(k-1)}-I^{(k-2)},\quad k\geq 2.$$ Note that when $ k=1 $ the equality gives $$\label{n09}
I^{(1)}=\lambda+\lambda^{2}I^{(0)}=\lambda+\lambda^{2}a.$$
From relations and follows the validity of the formula (and also ) for $k=0$ and $k=1$. Now let us prove that by induction the formula is true for any $n\geq 2$. We suppose that is true for $k=n-1$ and $ k=n-2$ and prove the validity of this formula for $k=n$. To do this, we substitute into the right side of equality with $k=n$ expression of $I^{(n-1)}$ and $I^{(n-2)}$ obtained from at $k=n-1 $ and $ k=n-2$, respectively, and check that it coincides with the right hand side of for $k=n$.
We have $$\begin{gathered}
I^{(n)}=\frac{a^{2}-1}{a}+\frac{a^{4}+1}{a^2}(a+a^3+...+a^{2n-1})-
(a+a^3+...+a^{2n-3})=\nn\\
-\frac{1}{a}+(a^3+...+a^{2n+1})+
(\frac{1}{a}+a+...+a^{2n-3})-(a^3+a^5+...+a^{2n-3})=\nn\\
(a+a^3+...+a^{2n+1})=a\frac{1-a^{2(k+1)}}{1-a^2},\qquad n\geq 2,\end{gathered}$$ i.e. formula are valid for $ k=n.$
Now let us prove the validity of formula for $ n\geq 2$. Let us suppose that is true for $k=n-1$ and $ k=n-2.$ Therefore, the right-hand sides of equations and coincide when $k=n-1$ and $k=n-2,$ respectively. We will show that then the right part of equal to the right side when $k\geq 2$.
Using the formula , we can rewrite recurrence relations in the following form $$aU_n=-((a^{2}+1)U_{n-1}+aU_{n-2}),\quad n\geq 2.$$ This allows us to rewrite as follows $$\begin{gathered}
I^{(n)}=(-1)^{n}a^{n+1}U_{n}=(-1)^{n}a^{n}(-(a^{2}+1)U_{n-1}-aU_{n-2})=\\
(-1)^{n-1}a^{n}U_{n-1}(a^{2}+1)-(-1)^{n-2}a^{n-1}U_{n-2}a^{2}=\\
(a^{2}+1)I^{(n-1)}-a^{2}I^{(n-2)}.\end{gathered}$$ Substituting the expressions $I^{(n-1)}$ and $I^{(n-2)}$ using the formula , we obtain $$\begin{gathered}
I^{(n)}=(a^{2}+1)(a+a^3+...+a^{2n-1})-a^{2}(a+a^3+...+
a^{2n-3})=\\
(a^3+...+a^{2n+1})+(a+a^3+...+a^{2n-1})-
(a^3+a^5+...+a^{2n-1})=\\
(a+a^3+...+a^{2n+1})=a\frac{1-a^{2(k+1)}}{1-a^2},\quad n\geq 2.\end{gathered}$$ This expression equals to the right hand side of when $n\geq 2$. Therefore, the formula is true for $k=n$ .
Appendix 2. The proof of equation when $\vert\lambda\vert \leq 2$ {#appendix-2.-the-proof-of-equation-when-vertlambdavert-leq-2 .unnumbered}
=================================================================
It is easy to see that when $\vert\lambda\vert \leq 2$ equality (\[triv505\]) can be rewritten as $$\begin{gathered}
I^{(k)}(\lambda)=v.p.\int_{-2}^{+2}\frac{1}{(t-\lambda)}
\left( U_{k}(\ts\frac{t}{2})\right)^2 \mu_{0}^{\prime}(t)\,dt=
-\frac{\lambda}{2}\left(U_{k}(\ts\frac{\lambda}{2})\right)^2+L_k,\end{gathered}$$ where $$\label{triv0012}
L_k=\int_{-2}^{+2}\frac{1}{(t-\lambda)}
\left( U_{k}^{2}(\ts\frac{t}{2})-U_{k}^{2}(\ts\frac{\lambda}{2})\right)\mu_{0}^{\prime}(t)\,dt.$$ We want to prove by induction the following formula $$\label{triv0015}
L_{k}(\lambda)=U_{k}(\ts\frac{\lambda}{2})U_{k-1}(\ts\frac{\lambda}{2}),\quad k\geq 0.$$ It is obvious that $ L_0=0.$ We compute $ L_1$: $$\begin{gathered}
L_1=\!\int_{-2}^{+2}\!\frac{1}{(t-\lambda)}\left( U_{1}^{2}(\ts\frac{t}{2})-
U_{1}^{2}(\ts\frac{\lambda}{2})\right) \mu_{0}^{\prime}(t)\,dt=
\!\int_{-2}^{+2}t\mu_{0}^{\prime}(t)\,dt+\lambda\!\int_{-2}^{+2}\mu_{0}^{\prime}(t)\,dt=
\lambda=U_{1}(\ts\frac{\lambda}{2})U_{0}(\ts\frac{\lambda}{2}),\end{gathered}$$ where the last equality follows from .
Using formulas and , we rewrite $U_{k}^{2}(\frac{t}{2})-U_{k}^{2}(\frac{\lambda}{2})$ in the following form $$\begin{gathered}
U_{k}^{2}(\ts\frac{t}{2})-U_{k}^{2}(\ts\frac{\lambda}{2})=
\left(t^{2}U_{k}^{2}(\ts\frac{t}{2})-\lambda^{2}U_{k}^{2}(\ts\frac{\lambda}{2})\right)-\nn\\
\left(2t(-1+U_{k-1}^{2}(\ts\frac{t}{2})+U_{k-2}^{2}(\ts\frac{t}{2}))-
2\lambda U_{k-1}(\ts\frac{\lambda}{2})U_{k-2}(\ts\frac{\lambda}{2})\right)+\nn\\
U_{k-2}^{2}(\ts\frac{t}{2})-U_{k-2}^{2}(\ts\frac{\lambda}{2}).\label{n11}\end{gathered}$$ Substituting into (\[triv0012\]) we get $$\label{triv0013}
L_k=\int_{-2}^{+2}\frac{t^{2}U_{k-1}^{2}(\frac{t}{2})-
\lambda^{2}U_{k-1}^{2}(\frac{\lambda}{2})}{t-\lambda}\mu_{0}^{\prime}(t)\,dt-
2L_{k-1}-L_{k-2}.$$ We write $t^{2}U_{k-1}^{2}(\frac{t}{2})-\lambda^{2}U_{k-1}^{2}(\frac{\lambda}{2})$ in the form $$\label{triv0014}
t^{2}U_{k-1}^{2}(\ts\frac{t}{2})-\lambda^{2}U_{k-1}^{2}(\ts\frac{\lambda}{2})=
(t^{2}U_{k-1}^{2}(\ts\frac{t}{2})-
\lambda^{2}U_{k-1}^{2}(\ts\frac{t}{2}))+
(\lambda^{2}U_{k-1}^{2}(\ts\frac{t}{2})-\lambda^{2}U_{k-1}^{2}(\ts\frac{\lambda}{2})).$$ Substituting (\[triv0014\]) into (\[triv0013\]) we have $$\label{triv0016}
L_k=\lambda+(\lambda^{2}-2)L_{k-1}-L_{k-2},\quad k\geq 2.$$
Above we show that is hold for $k=0$ and $k=1.$ Now we prove it by induction for all $k$. For this end we assume that true for $k=n-1$ and $k=n-2$ and show that it remains true for $k=n$.
From (\[triv0015\]) and (\[triv0016\]) we have $$L_k=\lambda+(\lambda^{2}-2)U_{k-1}(\ts\frac{\lambda}{2})U_{k-2}(\ts\frac{\lambda}{2})-
U_{k-2}(\ts\frac{\lambda}{2})U_{k-3}(\ts\frac{\lambda}{2}).$$ Then for the proof of formula (\[triv0015\]) at $k=n$ it is sufficient to check the validity of the following equality $$\lambda+(\lambda^{2}-2)U_{k-1}(\ts\frac{\lambda}{2})U_{k-2}(\ts\frac{\lambda}{2})-
U_{k-2}(\ts\frac{\lambda}{2})U_{k-3}(\ts\frac{\lambda}{2})=U_{k}(\ts\frac{\lambda}{2})U_{k-1}(\ts\frac{\lambda}{2}).$$ Using recurrent relations (\[triv03\]), we rewrite the previous equality in the form $$\lambda U_{k-1}^{2}(\ts\frac{\lambda}{2})-U_{k-1}(\frac{\ts\lambda}{2})U_{k-2}(\ts\frac{\lambda}{2})=\lambda+
(\lambda^{2}-2)U_{k-1}(\ts\frac{\lambda}{2})U_{k-2}(\ts\frac{\lambda}{2})-
U_{k-2}(\ts\frac{\lambda}{2})U_{k-3}(\ts\frac{\lambda}{2}).$$ It is easy to see that the equality is equivalent to the following equality $$\lambda U_{k-1}^{2}=\lambda+\lambda^{2}U_{k-1}U_{k-2}-\lambda U_{k-2}^{2}.$$ Finally, the last equality is equivalent to the identity (\[triv405\]). Thus as well as are proved.
Appendix 3. The proof of the identity (\[triv405\]) {#appendix-3.-the-proof-of-the-identity-triv405 .unnumbered}
===================================================
We want to prove by induction the following identity $$U_{k}^{2}\big(\ts{\frac{\lambda}{2}}\big)+U_{k-1}^{2}\big(\frac{\lambda}{2}\big)-
\lambda U_{k-1}\big(\frac{\lambda}{2}\big)U_{k}\big(\frac{\lambda}{2}\big)=1,$$ It is obvious that the equality holds for $k=0$. We will show that for any $n\geq 1 $ this equation is valid when $k=n$, if it is true for $k=n-1$. Using recurrent relation (\[triv03\]), we have $$\begin{gathered}
U_{n}^{2}+U_{n-1}^{2}-\lambda U_{n-1}U_{n}=\\
(\lambda U_{n-1}-U_{n-2})^{2}+U_{n-1}^{2}-\lambda U_{n-1}(\lambda U_{n-1}-U_{n-2})=\\
U_{n-1}^{2}+U_{n-2}^{2}-\lambda U_{n-1}U_{n-2}.\end{gathered}$$ The obtained expression is equal to unity according to the inductive assumption. Hence, the formula (\[triv405\]) is proved.
Appendix 4. The proof of the formula {#appendix-4.-the-proof-of-the-formula .unnumbered}
=====================================
Here we construct the polynomials $\{\phi_{n}^{(k)}(\lambda)\}_{n=0}^{\infty}$ related to the Jacobi matrix $\mathcal{J}_{k}^{(\beta)}$. These polynomials (see , ) is determined by following recurrence relations ($n\geq 0$) $$\begin{gathered}
\lambda\phi_{n}^{(k)}(\ts\frac{\lambda}{2})=
\phi_{n+1}^{(k)}(\ts\frac{\lambda}{2})+a_{n}\phi_{n}^{(k)}(\ts\frac{\lambda}{2})+
b_{n-1}\phi_{n-1}^{(k)}(\ts\frac{\lambda}{2}),\label{triv111} \\
\,\phi_{0}^{(k)}=1,\quad a_n =\beta\delta_{n k},\quad b_n =1- \delta_{-1 n}.\nn\end{gathered}$$ From recurrent relations (\[triv03\]) and (\[triv111\]), it follows that $$\label{triv112}
\phi_{0}^{(k)}=U_{0},\quad \phi_{1}^{(k)}(\ts\frac{\lambda}{2})=U_{1}(\ts\frac{\lambda}{2}),\,\ldots\,,\,
\phi_{k}^{(k)}(\ts\frac{\lambda}{2})=U_{k}(\ts\frac{\lambda}{2}).$$ From (\[triv112\]), using the recurrent relations (\[triv03\]) and \[triv111\] one obtains by induction that $$\begin{gathered}
\phi_{k+1}^{(k)}=U_{k+1}-\beta U_{k},\\
\phi_{k+2}^{(k)}=U_{k+2}-\beta U_{k+1}-\beta U_{k-1},...,\\
\phi_{2k+1}^{(k)}=U_{2k+1}-\beta U_{2k}-\beta U_{2k-2}-...-\beta U_{0}.\end{gathered}$$ Next, we have the following relation $$\begin{gathered}
\phi_{2k+2}^{(k)}(\ts\frac{\lambda}{2})=\lambda\phi_{2k+1}^{(k)}(\ts\frac{\lambda}{2})-
\phi_{2k}^{(k)}(\ts\frac{\lambda}{2})=\nn\\
\lambda(U_{2k+1}-\beta U_{2k}-\beta U_{2k-2}-...-\beta U_{0})-
(U_{2k}-\beta U_{2k-1}-\beta U_{2k-3}-...-\beta U_{1}).\label{triv114}\end{gathered}$$ Combining terms in pairs, standing on the same locations in the first and second brackets in the right hand side of equality , and using recurrence relations (\[triv03\]), we get $$\phi_{2k+2}^{(k)}=U_{2k+2}-\beta U_{2k+1}-\beta U_{2k-1}-...-\beta U_{1}.$$ Finally, to check the validity of the formula $$\label{triv115}
\phi_{n}^{(k)}=U_{n}-\beta U_{n-1}-\beta U_{n-3}-...-\beta U_{n-(2k+1)},\qquad
n\geq 2k+3,\quad k\geq 0\,,$$ it is enough to check that the polynomials $\phi_{n}^{(k)}$ defined by equations for $n\geq {2k+3}$ satisfy recurrence relations that for $n\geq k$ have the form $$\phi_{n}^{(k)}(\ts\frac{\lambda}{2})=
\lambda\phi_{n-1}^{(k)}(\ts\frac{\lambda}{2})-\phi_{n-2}^{(k)}(\ts\frac{\lambda}{2}).$$ Indeed, we have for all $n\geq {2k+3}$, $ k\geq 0 $ $$\begin{gathered}
\phi_{n}^{(k)}-\lambda\phi_{n-1}^{(k)}+\phi_{n-2}^{(k)}=\\
U_{n}-\beta U_{n-1}-\beta U_{n-3}-...-\beta U_{n-(2k+1)}-
\lambda(U_{n-1}-\beta U_{n-2}-\beta U_{n-4}-...-\beta U_{n-2k-2})+\\
(U_{n-2}-\beta U_{n-3}-\beta U_{n-5}-...-\beta U_{n-2k-3})=\qquad\qquad\\
(U_{n}-\lambda U_{n-1}+ U_{n-2})-\beta(U_{n-1}-\lambda U_{n-2}+ U_{n-3})-\ldots\qquad\qquad\\
\ldots-\beta(U_{n-(2k+1)}-\lambda U_{n-2(k+1)}+ U_{n-2k-3})=0.\end{gathered}$$ Thus, we proved that for all $k\geq 0 $, the polynomials $\phi_{n}^{(k)}$ can be calculated by the following formulas $$\begin{gathered}
\phi_{0}^{(k)}=U_{0},\quad \phi_{1}^{(k)}=U_{1},\ldots,\phi_{k}^{(k)}=U_{k},\\
\phi_{k+s}^{(k)}=U_{k+s}-\beta U_{k+s-1}-\beta U_{k+s-3}-\ldots-\beta U_{k-s+1},
\qquad 1\leq s\leq k+1,\\
\phi_{n}^{(k)}=U_{n}-\beta U_{n-1}-\beta U_{n-3}-...-\beta U_{n-(2k+1)},
\qquad n\geq 2k+2.\end{gathered}$$
[^1]: This article is extended variant of the manuscript with the same title submitted to the Proceedings of the Days on Diffraction 2016
[^2]: See Appendices 1& 2 for the proof
[^3]: See Appendix 3 for proof
[^4]: See Appendix 4 for the proof
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We trace the development of ideas on dissipative processes in chaotic cosmology and on minisuperspace quantum cosmology from the time Misner proposed them to current research. We show\
1) how the effect of quantum processes like particle creation in the early universe can address the issues of the isotropy and homogeneity of the observed universe,\
2) how viewing minisuperspace as a quantum open system can address the issue of the validity of such approximations customarily adopted in quantum cosmology, and\
3) how invoking statistical processes like decoherence and correlation when considered together can help to establish a theory of quantum fields in curved spacetime as the semiclassical limit of quantum gravity.
author:
- '[*B. L. Hu*]{} [^1]'
- '[*Juan Pablo Paz*]{} [^2]'
- '[*Sukanya Sinha*]{} [^3]'
title: Minisuperspace as a Quantum Open System
---
==c===ł=Ł=ø=Ø==u
c i ł ø u Ł Ø ¶
\#1[\^[\#1]{}]{} \#1[\_[\#1]{}]{} /\#1[/\#1]{}
\#1 \#1[\#1]{} \#1[\#1]{} \#1[\#1]{} \#1[\#1]{} \#1[\#1]{} \#1[\#1|]{} \#1[| \#1]{} \#1\#2[\#1\#2]{} \#1[\#1]{} \#1 \#1
\#1[| \#1|]{} \#1
\#1\#2 \#1\#2[[1 1]{}]{} \#1\#2[[1 1]{}]{} \#1\#2[[1\#2]{}]{} \#1\#2[[\#1\#2]{}]{} \#1\#2\#3[[\^2 \#1\#2 \#3]{}]{}
‘@=11 \#1[@underline\#1 $\@@underline{\hbox{#1}}$]{} ‘@=12
plus 1000pt minus 1000pt \#1 \#1[= to]{} \#1[= to]{}
\#1[Nucl. Phys. \#1]{} \#1[Phys. Lett. \#1]{} \#1[Phys. Rev. Lett. \#1]{} \#1[Phys. Rev.[**A**]{} \#1]{} \#1[Prog. Theor. Phys. \#1]{} Å\#1[Astron. and Astrophys. \#1]{} \#1[$\sp{#1)}$]{} \#1
startsection [section]{}[1]{}[@]{}[-1.5ex plus -.5ex minus -.2ex]{}[1ex plus .2ex]{}[****]{}
============================================================================================
thmcountersep makecaption\#1\#2
tempboxa tempboxa > \#1. \#2
to
PS. @headings[ oddhead evenhead[oddhead]{} oddfootevenfoot[oddfoot]{} ]{} \[section\] \[Theorem\] [Corollary]{} \[Theorem\] [Lemma]{}
[*Dedicated to Professor Misner on the occasion of his sixtieth birthday, June 1992.*]{} To appear in the Proceedings of a Symposium on [*Directions in General Relativity*]{}, College Park, May 1993, Volume 1, edited by B. L. Hu, M. P. Ryan and C. V. Vishveshwara (Cambridge University Press 1993) umdpp 93-60
Introduction
============
In the five years between 1967 and 1972, Charlie Misner made an idelible mark in relativistic cosmology in three aspects.
First he introduced the idea of chaotic cosmology. In contrast to the reigning standard model of Friedmann-Lemaitre-Robertson-Walker universe where isotropy and homogeneity are ‘put in by hand’ from the beginning, chaotic cosmology assumes that the universe can have arbitrary irregularities initially. This is perhaps a more general and philosophically pleasing assumption. To reconcile an irregular early universe with the observed large scale smoothness of the present universe, one has to introduce physical mechanisms to dissipate away the anisotropies and inhomogeneities. This is why dissipative processes are essential to the implementation of the chaotic cosmology program. Misner (1968) was the first to try out this program in a Bianchi type-I universe with the neutrino viscosity at work in the lepton era. Though this specific process was found to be too weak to damp away the shear, the idea remains a very attractive one. As we will see, it can indeed be accomplished with a more powerful process, that of vacuum particle production at the Planck time. The philosophy of chaotic cosmology is similar to that behind the inflationary cosmology of Guth (1981), where the initial conditions are rendered insignificant by the evolutionary process, which, in this case is inflation.
Second, the chaotic cosmology program ushered in serious studies of the dynamics of Bianchi universes. This was exemplified by Misner’s elegant work on the mixmaster universe (1969a). (That was when one of us first entered the scene.) The ingeneous use of pictorial representation of the curvature potentials made the complicated dynamics of these models easy to follow and opened the way to a systematic analaysis of this important class of cosmology. (See Ryan and Shepley 1975.) This development complements and furthers the ongoing program of Lifshitz, Khalatnikov and Belinsky (1963, 1970) where, in seeking the most general cosmological solutions of Einstein’s equations using rigorous applied mathematics techniques (see also Eardley, Liang and Sachs 1971), they found the inhomogeneous mixmaster universe (the ‘generalized’ Kasner solution) as representing a generic behavior near the cosmological singularity. This also paralleled the work of Ellis and MacCallum (1969) who, following Taub (1951) and Heckmann and Schucking (1962), gave a detailed analysis of the group-theoretical structure of the Bianchi cosmologies. Misner’s chaotic cosmology program provided the physical rationale for these studies. It also prepared the ground for minisuperspace quantum cosmology (1969b, 1972).
Third, the establishment of the quantum cosmology program, which includes the application of the ADM quantization (1962) to cosmological spacetimes, and the adaptation of the Wheeler- DeWitt superspace formulation of quantum gravity (Wheeler 1968, De Witt 1967) to minisuperspace cosmology. The physical motivation was to understand ‘the issue of the initial state’ (Wheeler 1964), and, in particular, the quantum effects of gravity on the cosmological singularity. It was Misner and his associates who started the first wave of activity in quantum cosmology (Ryan, 1972; see also Kuchar 1971, Berger 1974, 1975). The second wave, as we know, came with the work of Hartle and Hawking (1983), and Vilenkin (1983) who, while formulating the problem in Euclidean path integral terms, opened the question on the boundary conditions of the universe.
These three directions in cosmology initiated by Misner and his collaborators in the early seventies –chaotic cosmology, dissipative processes and minisuperspace quantum cosmology– have developed in major proportions in the past two decades. A lot of current research work in these areas still carry the clear imprint of his influence. The present authors have had the good fortune to be influenced by his way of thinking. We want to trace out some major developments since Misner wrote these seminal papers and describe the current issues in these areas, with more emphasis on their interconnections from early universe quantum processes to quantum cosmology.
Particle Creation as a Quantum Dissipative Process
--------------------------------------------------
Misner (1968) and Matzner and Misner (1972) showed that neutrino viscosity in the lepton era ($ \approx 1 sec$ from the Big Bang) is in general not strong enough to dissipate away the shear in the Bianchi universes. It was Zel’dovich (1970) who first suggested that vacuum particle creation from the changing gravitational field in anisotropic spacetimes may act as a powerful dissipative source. This created a real possibility for the idea of chaotic cosmology to work, now relying on quantum field processes effective near the Planck time ($10^{-43}$ sec from the Big Bang). One needs a new theory — the theory of quantum fields in curved spacetime. Actually this theory was just beginning to take shape through the work of Parker (1966, 1969) and Sexl and Urbantke (1968) at about the same time when Misner was working on the isotropy of the universe problem (1968). It was later realized that the reason why the Bianchi I universe and not the Robertson-Walker universe produces a copious amount of particles is because it breaks the conformal invariance of the theory (Parker 1973). Zel’dovich used a simple dimensional analysis to explain why the produced particles can strongly influence the dynamics of spacetime at the Planck time. The details are actually more involved than this, because one needs to remove the divergences in the energy-momentum tensor of the matter field before it can be used as the source of the Einstein equation to solve for the metric functions. This is what has been called the ‘backreaction’ problem. Zel’dovich and Starobinsky (1971) were the first to attempt such a calculation for conformal scalar fields in a Bianchi I universe. (See also Lukash and Starobinsky 1974). That was before the basic issues of particle creation processes were fully understood (e.g., ambiguities in the choice of the vacuum in curved space, see, Fulling 1973) and the regularization techniques were perfected. But they managed to show the viability of such processes. It took several years (1974-77) before the common ground between the different ways of regularization (e.g., adiabatic, point-splitting, dimensional, zeta-function) was reached and a firmer foundation for this new theory of semiclassical gravity was established. Hawking’s discovery (1974) of black hole radiance, and the confirmed ‘legality’ of trace anomaly (Capper and Duff 1975, Deser, Duff and Isham 1976) certainly invigorated this endeavour. The theoretical importance of the subject and the infusion of talents from general relativity, field theory, and particle physics have made this field an important component of contemporary theoretical physics (Birrell and Davies 1982). It is, as we shall see, also an intermediate step (the low energy or adiabatic limit) towards a theory of quantum gravity and quantum cosmology.
Vigorous calculation of the backreaction of particle creation with regularized energy momentum sources began with the work of Hu and Parker (1977,1978) using adiabatic regularization methods, and Hartle (1977), Fischetti et al (1979) and Hartle and Hu (1979, 1980), using effective action methods. Their results on the effect of particle creation on anisotropy dissipation confirmed the qualitative estimate of Zel’dovich, and lent strong support for the viability of Misner’s chaotic cosmology program.
Entropy of Quantum Fields and Spacetime?
----------------------------------------
These efforts in quantum field theory in curved spacetime and dissipative processes in the early universe contained the germs for two interesting directions of later development. Both involve quantum field theory and statistical mechanics applied to cosmological problems. They are\
i) Dissipation in quantum fields as a statistical thermodynamic proces s:\
Can one associate a viscosity with vacuum particle creation (Zel’dovich 1979, Hu 1984) like that of neutrino in the lepton era, or leptoquark boson decays in the GUT era? Can one define an entropy of quantum fields? (Hu and Pavon 1986, Hu and Kandrup 1987, Kandrup 1988)? An entropy of spacetime? (Penrose 1979, Hu 1983, Davies, Ford and Page 1986) Can one rephrase the chaotic cosmology idea in terms of some analog of second law of thermodynamics involving both fields and geometry (Hu 1983)? Can one generalize the fluctuation-dissipation relation (FDR) for quantum processes in spacetimes with event horizons (Candelas and Sciama 1977, Mottola 1986) to general dynamical spacetimes? Indeed can one view the backreaction process as a manifestation of a generalized FDR for dynamical systems like quantum fields in cosmological spacetimes (Hu 1989, Hu and Sinha 1993)?
This first direction of research has been pursued in the eighties, with the infusion of ideas and techniques in statistical and thermal field theory. It is intended to be a generalization of thermodynamic ideas successfully applied to black hole mechanics (Davies 1978) to non-equilibrium quantum field systems characteristic of cosmological problems. This was an offshoot of Misner’s interest in dissipative processes in the early universe, only that one now takes on the challenge of doing everything with quantum fields in non-equilibrium conditions. This exercise probes into many deeper issues of thermodynamics and cosmology, such as the statistical nature of gravitational systems as different from ordinary matter, the thermodynamic nature of spacetimes (with or without event horizons), the nature of irreversibility and the cosmological origin of time-asymmetry. We will not belabor this direction of research here, but refer the reader to some recent reviews (Hu 1989, 1991, 1993a, 1993b). As we will see, inquiries into the statistical nature of particle creation and backreaction processes actually have a role to play in the second direction, i.e.,
ii\) Quantum field theory as the semiclassical limit of quantum gravity. How good is modified Einstein’s theory with backreaction from quantum matter fields an approximation to quantum gravity? Under what conditions can the quantum dynamics of spacetime described by a wave function make the transition to the classical picture described by trajectories in superspace? We will use the paradigms of quantum open systems applied to Misner’s minisuperspace models to illustrate these ideas.
From Dissipation to Quantum Open Systems
----------------------------------------
Speaking from the path traversed by one of us, one can identify a few critical nodes in the search for a pathway which connects i) particle creation as a dissipative processes to ii) backreaction and semiclassical gravity to iii) the statistical aspects in the issue of quantum to classical transition which underlies the relation of quantum gravity and semiclassical gravity. (See the Ph.D. thesis of Sinha 1991 for an overview of other major developments and references in these areas.) The (in-out) effective action formalism (Schwinger 1951, DeWitt 1964, 1975) was a correct start, because it gives the rate of particle production and the backreaction equations in a self-consistent manner (Hartle and Hu 1979). However the conditions for calculating the in-out vacuum persistence amplitude are unsuitable for the consideration of backreactions, as the effective equations are acausal and complex. The second major step was the recognition by DeWitt (1986), Jordan (1986), and Calzetta and Hu (1987) that the in-in or closed-time-path formalism of Schwinger (1961) and Keldysh (1964) is the right way to go. Calzetta and Hu (1987) derived the backreaction equations for conformal scalar fields in Bianchi I universe, which provided a beachhead for later discussion of the semiclassical limit of quantum cosmology. They also recognized the capacity of this framework in treating both the quantum and statistical aspects of dynamical fields and indeed used it for a reformulation of non-equilibrium statistical field theory (Calzetta and Hu 1988). The third step in making the connection with statistical mechanics is the construction of the so-called coarse-grained effective action by Hu and Zhang (1990), similar in spirit to the well-known projection operator method in statistical mechanics (Zwanzig 1961).
This is where the connection with ‘quantum open systems’ first enters. The general idea behind the study of such systems is that one starts with a closed system (universe) , splits it into a “relevant" part (system) and an “irrelevant" (environment) part according to the specific physical conditions of the problem (for example, the split between light and heavy modes or slow and fast variables), and follows the behavior only of the relevant variables (now comprising an open system) by integrating out the irrelevant variables. (See Hu 1993a and 1993b for a discussion of the meaningfulness of such splits in the general physical and cosmological context respectively). Thus the coarse-grained effective action generalizes the background field/ fluctuation-field splitting usually assumed when carrying out a quantum loop expansion in the ordinary effective action calculation to that between a system field and an environment field. It takes into account the averaged effect of the environment variables on the system and produces an effective equation of motion for the system variables. This method was used by Hu and Zhang (1990) (see also Hu 1991) to study the effect of coarse-graining the higher modes in stochastic inflation, and by Sinha and Hu (1991) to study the validity of minisuperspace approximations in quantum cosmology. The last step in completing this tour is the recognition by Hu, Paz and Zhang (1992) that the coarse-grained closed-time-path effective action formalism is equivalent to the Feynman-Vernon (1963) influence functional which was popularized by Caldeira and Leggett (1983) in their study of quantum macroscopic dissipative processes. Paz and Sinha (1991, 1992) have used this paradigm for a detailed analysis of the relation of quantum cosmology with semiclassical gravity. We like to illustrate this development with two problems, initially proposed and studied by Misner. One is on the validity of minisuperspace approximation, the other is on the classical limit of quantum cosmology.
Validity of the Minisuperspace Approximation
============================================
In quantum cosmology, the principal object of interest is the wave function of the universe $\Psi(h_{ij}, \phi)$, which is a functional on superspace. It obeys the Wheeler DeWitt equation, which is an infinite-dimensional partial differential equation. Its solution is a formidable task in the general case. To make the problem more tractable, Misner suggested looking at only a finite number of degrees of freedom (usually obtained by imposing a symmetry requirement), and coined the word minisuperspace (Misner 1972) quantization. We note that such a tremendous simplification has its preconditions. Since in the process of this transition one is truncating an infinite number of modes, nonlinear interactions of the modes among themselves and with the minisuperspace degrees of freedom are being ignored. It is also well-known that this truncation violates the uncertainty principle, since it implies setting the amplitudes and momenta of the inhomogeneous modes simultaneously to zero. It is therefore important to understand under what conditions it is reasonable to consider an autonomous evolution of the minisuperspace wave function ignoring the truncated degrees of freedom. Misner was keenly aware of the problems when he introduced this approximation. The first attempt to actually assess the validity of the minisuperspace approximation was made only recently by Kuchar and Ryan (1986, 1989).
Two of us (Sinha and Hu 1991) have tried to address this question in the context of interacting field theory with the help of the coarse-grained effective action method mentioned above. The model considered is that of a self interacting (${\lambda \phi^{4}}$) scalar field coupled to a closed Robertson Walker background spacetime. The “system”(minisuperspace) consists of the scale factor $a$ and the lowest mode of the scalar field, while the “environment" consists of the inhomogeneous modes. [^4] In this model calculation the scalar field is mimicking the inhomogeneous gravitational degrees of freedom in superspace. We are motivated to do this because linearized gravitational perturbations (gravitons) in a special gauge can be shown to be equivalent to a pair of minimally coupled scalar fields (Lifshitz 1946). Quantum cosmology of similar models of gravitational perturbations and scalar fields have been studied by several authors (Halliwell and Hawking 1985, Kiefer 1987, Vachaspati and Vilenkin 1988). Our basic strategy will be to try to derive an “effective" Wheeler-DeWitt equation for the minisuperspace sector which contains the “averaged" effect of the higher modes as a backreaction term. We will then explicitly calculate the backreaction term using the effective action and consequently present a criterion for the validity of the minisuperspace approximation.
Effective Wheeler DeWitt Equation
---------------------------------
The gravitational and matter actions in our model are given by S\_g & = & da\^[2]{}(1 - [a’\^[2]{}a\^2]{})\
S\_m & = & - d\^[4]{}x\[+ m\^2\^2 +[2ł4!]{}\^4 + [R6]{} \^2\] where $a$ is the scale factor of a closed Robertson-Walker universe, a factor $l_p^2 = 2/(3\pi m_p^2) $ is included in the metric for simplification of computations, and $\eta$ is the conformal time. $\Box$ is the Laplace Beltrami operator on the metric $g_{\m\n}$, and $m$ is the mass of the conformally coupled scalar field. Defining a conformally related field $\chi = (al_p)\Phi$ and expanding $\chi$ in scalar spherical harmonics $Q^{k}_{lm}(x)$ on the $S^3$ spatial sections, = [\_[0]{}(t)(2\^2)\^]{} + \_[klm]{}[ ]{}f\_[klm]{} Q\^[k]{}\_[lm]{}(x) where $k=2,3,\ldots \infty, l = 0,1,\ldots k-1, m = -l, -l+1 \ldots
l-1,l$ (henceforth we will use $k$ to denote the set $\{klm\}$). We will make the further assumption that the interactions of orders higher than quadratic of the lowest(minisuperspace) mode $\chi_0$ with the higher modes $f_n$’s ($\sim \chi_0 f_{k}f_{l}f_{m}$) as well as the quartic self interaction of the higher modes ($\sim f_{m}f_{n}f_{k}f_{l}$) are small and can be neglected. With this assumption and with the following redefinitions $ m^{2} \rightarrow {m^{2}/l_p^2} , \l \rightarrow {\l/
2{\pi}^2}$,the matter action can be written as S\_m & = & d{[12]{} - [ł4!]{}\^4 -[12]{}\_[k]{}[f\_k ]{}-[12]{}\_[k]{} [m\^2 a\^2 [f\_k]{}\^2]{}\
\[matter\] & & - [ł4]{}\_[k]{}[[\_0]{}\^2 [f\_k]{}\^2]{} } The Hamiltonian constructed from the action $S = S_g + S_m$ is given by H = -[12]{}[\_a]{}\^[2]{} + [12]{}[\_[\_0]{}]{}\^[2]{} + [12]{}\_[n]{}[\_[f\_n]{}]{}\^2 + V\_[0]{}(a, \_[0]{}) + V(a,\_[0]{}, f\_n) where V\_[0]{}(a,\_[0]{}) = -[12]{}a\^2 + [12]{}m\^[2]{}a\^[2]{}[\_[0]{}]{}\^2 + [ł4!]{}[\_0]{}\^4 and V(a,\_0 ,f\_n) = [ł4]{}\_[k]{}[\_0]{}\^2 [f\_k]{}\^2 + [12]{}\_[k]{}(k\^2 + m\^2 a\^2)[f\_k]{}\^2 where $\pi_a, \pi_{\chi_0},\pi_{f_n}$ are the momenta canonically conjugate to $a, \chi_0, f_n$ respectively. The Wheeler-DeWitt equation for the wave function of the Universe $\Psi$ is obtained from the Hamiltonian constraint by replacing the momenta by operators in the standard way, and is given by (a, \_0, f\_n) = 0 where we choose a factor ordering such that the kinetic term appears as the Laplace-Beltrami operator on superspace.
Writing (a,\_0,f\_n) = \_0 (a, \_0) \_n \_n(a, \_0, f\_n) we would like to obtain from (2.8) an effective Wheeler-DeWitt equation of the form ( H\_0 + H ) \_0 (a, \_0) = 0 where $H_0$ is the part of the Hamiltonian operator in (2.8) independent of $f_n$ and $\Delta H$ represents the influence of the higher modes. By making the assumption that $\Psi_n$ varies slowly with the minisuperspace variables (see Sinha and Hu 1991 for details of this approximation) one can identify H = -\_[n]{} <H\_n> where the expectation value is taken with respect to $\Psi_n$. It is evident that the examination of this term will enable us to comment on the validity of the minisuperspace description. We will then make a further assumption that $\Psi_0$ has a WKB form, i.e, $\Psi_0 = e^{iS(a,\chi_0)}$ which can be used in regions of superspace where the wavefunction oscillates rapidly, such that using eqn. (2.10), $S$ satisfies \^2 + V\_0 = -\_[n]{}< H\_n> \[HJ\] where $\nabla$ is the gradient operator on minisuperspace. The above equation can be regarded as a Hamilton-Jacobi equation with backreaction. It can be also shown that the $\Psi_n$’s satisfy a Schrödinger - like equation with respect to the WKB time. This approximation can therefore be roughly thought of as the semiclassical limit where the minisuperspace modes behave classically, but the higher modes behave quantum mechanically ( for further subtleties regarding the semiclassical limit see Sec. 3). Identifying ${\partial S\over \partial a} = \p_a$ and ${\partial S\over \partial\chi_0} = \pi_{\chi_0}$, and substituting for the canonical momenta in terms of “velocities" $a'$ and $\chi_0'$ Eq. (12) reduces to a’\^[2]{} -[12]{}[’\_0]{}\^2 + V\_0(a,\_0) = -\_n<H\_n> This is the effective Wheeler- De Witt equation in the WKB limit, which we will compare with the backreaction equation derived in the next section, in order to calculate the term on the right hand side.
Backreaction of the Inhomogeneous Modes
---------------------------------------
We would now like to calculate the backreaction term (2.11) explicitly. Since we are making a split of the system from the environment based on the mode decomposition, we should use the coarse-grained effective action where the coarse graining consists of functionally integrating out the higher modes. As we would like to generate vacuum expectation values from the effective action rather than the matrix elements one should use the in-in or closed-time-path (CTP) version of the effective action (Calzetta and Hu 1987) rather than the in-out version. The CTP coarse-grained effective action in our case is given by (for details, see Sinha and Hu 1991) e\^[iS\_[eff]{}( a\^+ ,\_[0]{}\^[+]{},a\^-,\_[0]{}\^[-]{} )]{} = f\_[k]{}\^[+]{} [D]{}f\_[k]{}\^[-]{} e\^[i( S(a\^+ ,[\_0]{}\^+, f\_[k]{}\^[+]{}) - S\^[\*]{}( a\^- ,[\_0]{}\^[-]{} ,f\_[k]{}\^[-]{}) )]{} $S^{*}$ indicates that in this functional integral, $m^{2}$ carries an $i\epsilon$ term. $a^{\pm}, \chi_{0}^{\pm}, f_{k}^{\pm}$ are the fields in the positive (negative) time branch running from $\eta =
\stackrel{-}{+}\infty$ to $\pm\infty$. The path integral is over field configurations that coincide at $t = \infty$. ${\cal D} {f_k}$ symbolizes the functional integration measure over the amplitudes of the higher modes of the scalar field.
We have derived the one loop renormalized coarse-grained effective action given as follows (omitting terms that involve $-$ fields only) S\_[eff]{} & = & S\_[g+]{} + [12]{}d{ [[\^+]{}’]{}\^2 - \^2 [\^+]{}\^2}\
& &-[ł4!]{}d\^4 + [13ł48]{}d+ [116]{}d\^4\
& &+ [132]{} [d\_1 d\_2]{}(\_1) K(\_1 - \_2) (\_2)\
& & +[132]{}[d\_1 d\_2]{}(\_1) |[K]{}(\_1 - \_2)(\_2) where the coupling constants have their renormalized values. $S_{g_+}$ represents the classical gravitational part of the action. $M^{{\pm}^2} = {\tilde{m}}^{{\pm}^2} + \ha \l {\mbox{$\chi_{0}$}}^{{\pm}^2} $, and ${ K}$ and ${\bar{K}}$ are complex nonlocal kernels with explicitly known forms. The effective equations of motion are obtained from this effective action via [^5]
\_ = 0 \_ = 0
Since we are interested in comparing with (2.13) which is equivalent to the $G_{00}$ Einstein equation with backreaction, we need the first integral form of (2.16), which can be derived as & & [12]{}[a’]{}\^2 - [12]{}’\^2 + [12]{} m\^2 a\^2 \^2 - [12]{} a\^2 + [ł4!]{}\^4 -[13ł96]{}\^2 - [1348]{}m\^2 a\^2 - [116]{} M\^2 ln a\
& & + [132]{}d\_1 M\^2()[K]{}(- \_1)M\^2(\_1) = 0 with the assumption of having no quanta of the higher modes in the initial state. ${\cal K} = K + \bar{K}$ and is real and hence the above equation is also manifestly real. Equation (2.17) is then equivalent to the effective $G_{00}$ Einstein equation or the Einstein- Hamilton- Jacobi equation plus backreaction. This in turn can be identified with (2.13), the effective Wheeler- De Witt equation in the WKB limit. Therefore, comparing (2.17) and (2.13) we can identify the backreaction piece in (2.13) as \_[n]{}<H\_n> & = & -[13ł96]{}\^2 - [13ł48]{}m\^2 a\^2 -[116]{}M\^2 ln a\
& & - [132]{}d\_1 M\^2() [K]{}(- \_1)M\^2(\_1) when the boundary conditions on the wave function are appropriate for the $\Psi_n$’s to be in a conformal “in" vacuum state. [^6]
Since equation (2.17) is the “effective" Wheeler- DeWitt equation for the minisuperspace sector within our approximation scheme, the condition for validity of this approximation can be stated as \_[n]{}<H\_n> V\_0 where by the left hand side we mean the regularized value given by (2.18). It was shown that the term in equation (2.18) involving the nonlocal kernel is related to dissipative behavior in closely related models (Calzetta and Hu 1989). This dissipative behavior in turn has been related to particle production by the dynamical background geometry in semiclassical gravity models (Hu, 1989). In our case this can be interpreted as scalar particles in the higher modes being produced as a result of the dynamical evolution of the minisuperspace degrees of freedom generating a backreaction that modifies the minisuperspace evolution. We can therefore think of this term as introducing dissipation in the minisuperspace sector due to interaction with the higher modes that are integrated out. One can justifiably think of autonomous minisuperspace evolution only when this dissipation is small. Since we have used the scalar field modes to simulate higher gravitational modes these considerations can also be directly extended to include gravitons. [^7]
This is an example of how ideas of open systems can be useful in understanding dissipation and backreaction, even in quantum cosmology. One can also use this paradigm to address the problem of quantum to classical transition, specifically the relation of semiclassical gravity with quantum gravity, which we will now address. To do this the formalism will need to be elevated to the level of density matrices.
Semiclassical Limit of Quantum Cosmology
========================================
We now report on the result of some recent work by two of us (Paz and Sinha 1991, 1992) on this problem. Quantum cosmology rests on the rather bold hypothesis that the entire universe can be described quantum mechanically. This pushes us to question the usual Copenhagen interpretational scheme that relies on the existence of an [*a priori*]{} classical external observer/apparatus. Since in the case of quantum cosmology this cannot be assumed, the theory needs to predict the “emergence" of a quasiclassical domain starting from a fundamentally quantum mechanical description. Recently, there has been a lot of interest in this subject, and two basic criteria for classicality have emerged (see Halliwell 1991 and references therein) from these endeavors. The first is [*decoherence*]{} – which requires that quantum interference between distinct alternatives must be suppressed. The second is that of [*correlation*]{}, which requires that the wave function or some distribution constructed from it (e.g. Wigner 1932) predicts correlations between coordinates and momenta in accordance with classical equations of motion. We will study the emergence of a semiclassical limit with the help of a quantum open system paradigm applied in the context of quantum cosmology. The basic mechanism for achieving decoherence and the appearance of correlations is by coarse-graining certain variables acting as an environment coupled to the system of interest.
Reduced Density Matrix
----------------------
We will consider a $D$-dimensional minisuperspace with coordinates $r^m$ ($m=1,\ldots,D$) as our system for which the Hamiltonian can be written as: $$H_g= {1\over{2M}}G^{mm'}p_mp_{m'} + M V(r^m)$$ where $G^{mm'}$ is the (super)metric of the minisuperspace and $M $ is related to the square of the Planck mass. The minisuperspace modes are coupled to other degrees of freedom $\Phi$, such as the modes of a scalar field, or the gravitational wave modes with a Hamiltonian $H_\Phi(\Phi, \pi_\Phi, r^m, p_m)$. These constitute the environment, or the ‘irrelevant’ part in our model. Thus, the wave function of the Universe is a function $\Psi=\Psi(r^m, \Phi)$ which satisfies the Wheeler-De Witt equation (Wheeler 1968, DeWitt 1967) : $$H\Psi= \bigl(H_g + H_\Phi\bigr)\Psi=0
$$ where the momenta are now replaced by operators and $H_{\Phi}$. We use the same factor ordering as before. If one is interested in making predictions only about the behavior of the minisuperspace variables, a suitable quantity for such a coarse-grained description is the reduced density matrix defined as: $$\rho_{red}(r_2,r_1) = \int d\Phi \Psi^{*}(r_1,\Phi) \Psi(r_2,\Phi)$$ We will consider an ansatz for the wave function of the following form: $$\Psi(r,\Phi) = \sum\limits_n e^{iMS_{(n)}(r)}C_{(n)}(r)\chi_{(n)}(r,\Phi).
$$ where $S_n$ is a solution of a Hamilton-Jacobi equation with respect to the $r$ coordinate which is the same as (2.12) without the backreaction term and with the gradient given by $G^{mm'}{{\partial }\over{\partial r^{m'}} }$. The prefactor $C_{(n)}
$is determined through the equation $2G^{mm'} {\partial_mC_{(n)}} \partial_{m'}S_{(n)} + \Box_G S_0 = 0$, and $\chi_{(n)}$ is a solution of a Schrödinger equation with respect to the WKB time ${d\over dt} = G^{mm'}{{\partial S_0}\over{\partial r^{m'}} }
{{\partial}\over{\partial r^m}}$ and with Hamiltonian $H_{\Phi}$. These equations are derived by the well -known order by order expansion in $M^{-1}$, where $M$ acts as a parameter analogous to the mass of the heavy modes (in our case the minisuperspace ones) in the Born-Oppenheimer approximation. The Hamilton-Jacobi equation satisfied by $S_{(n)}$ will have a $D-1$ parameter family of solutions, the subindex $(n)$ labeling such parameters characterizes a particular solution and thus a specific WKB branch.
The reduced density matrix associated with the wave function $(3.4)$ is then given by: $$\rho_{red}(r_2,r_1) =\sum\limits_{n,n'} e^{iM[S_{(n)}(r_1) - S_{(n')}(r_2)]}
C_{(n)}(r_1)C_{(n')}(r_2) I_{n,n'}(r_2,r_1)
$$ where we call $$I_{n,n'}(r_2,r_1)= \int \chi^{*}_{(n')}(r_2,\Phi)\chi_{(n)}(r_1,\Phi) d\Phi
$$ the influence functional. The justification for this name comes from the fact that it has been shown (Paz and Sinha 1991, Kiefer 1991) that it is exactly analogous to the Feynman-Vernon (1963) influence functional in the case in which the environment is initially in a pure state. For models with $r >1$ it can be shown that it is indeed a functional of two histories, which is in turn related to the Gell-Mann Hartle decoherence functional (Gell-Mann and Hartle 1990, Griffith 1984, Omnes 1988). It is the central object of our consideration.
Decoherence and Correlation
---------------------------
We will now study the emergence of classical behavior in the minisuperspace variables as a consequence of their interaction with the environment. As we will show the two basic characteristics of classicality, the issues of correlations and decoherence, are indeed interrelated.
[**i) Decoherence**]{}: Decoherence occurs when there is no interference effect between alternative histories. If each of the diagonal terms in the sum of $(3.4)$ corresponds to a nearly classical set of histories, the interference between them is contained in the non–diagonal terms ($n\neq n'$) of that sum. Thus, the system decoheres if influence functional $I_{n,n'}\propto\delta_{n,n'}$ approximately. [^8] As a consequence the density matrix $(3.4)$ can be considered to describe a “mixture” of non–interfering WKB branches representing distinct Universes.
[**ii) Correlations**]{}: The second criterion for classical behavior is the existence of correlations between the coordinates and momenta which approximately obey the classical equations of motion. To analyze this aspect we will compute the Wigner function associated with the reduced density matrix and examine whether the Wigner function has a peak about a definite set of trajectories in phase space corresponding to the above correlations (Halliwell 1987, Padmandabhan and Singh 1990). Once we have established the decoherence between the WKB branches using criterion i), we look for correlations using the Wigner function [*within*]{} a “decohered WKB branch", i.e, that associated with $(n = n')$. In this sense the two criteria are interrelated, i.e, we [*need*]{} decoherence between the distinct WKB branches to be able to meaningfully predict correlations.
The Wigner function associated with one of the diagonal terms is given by $$W_{(n)}(r,P) = \int\limits_{-\infty}^{+\infty} d\xi^m
C_{(n)}(r_1)C_{(n)}(r_2)
e^{-2i{P_m\xi^m}}
e^{iM[S_{(n)}(r_1) - S_{(n)}(r_2)]}I_{n,n}(r_2,r_1)$$ where $r^m_{1,2} = r^m \pm {\xi^m\over M}$.
The functional $I_{n,n}(r_2,r_1)$ plays the dual role of producing the diagonalization of $\rho_{(n)}$ and affecting the correlations (the phase affects the correlations and the absolute value the diagonalization). [^9] In the language of measurement theory (Wheeler and Zurek 1986) one can say that the environment is “continuously measuring” the minisuperspace variables (Zurek 1981, 1991, Zeh 1986, Kiefer 1987) and that this interaction not only suppresses the $n\neq n'$ terms in the sum of equation $(3.11)$, but also generates a “localization” of the $r$ variables inside each WKB branch. This localization effect is essential in order to obtain a peak in the Wigner function. The form of the peak and the precise location of its center are determined by the form of $I_{n,n}(r_2,r_1)$. To illustrate this better let us assume that the state of the environment is such that the influence functional can be written in the form I\_[(n,n)]{}(r) e\^[i\_m(r) \^m - \^2[\_m]{}[\^m]{}]{} where $\beta$ and $\sigma$ are real and $r = {r_1 + r_2\over 2}$, and we notice that $\sigma$ is related to the degree of diagonalization of the density matrix. The Wigner function for this is computed to be W\_[(n)]{}(r,P) \^2(r) e\^[-[(P\_m - M[Sr\^m]{} -\_m)]{}\^2]{} It is a Gaussian peaked about the classical trajectory $(P_m - M{\partial S\over \partial
r^m}= 0)$ shifted by a backreaction term $\ha\beta_m$, but with a spread characterized by $\sigma$, both of these arising from coarse graining the environment. We also notice a competition between sharp correlations (related to the sharpness of the above peak) and the diagonalization of the density matrix, and this can be formalized (see Paz and Sinha 1991, 1992) in a set of criteria for the emergence of classical behavior through a compromise between decoherence and correlations. It can also be shown that when the influence functional is of the form (3.8) the correlations predicted are exactly those given by the semiclassical Einstein equations. Thus the above arguments can be tied to the emergence of the semiclassical limit of quantum fields in curved space time starting from quantum cosmology. This has been illustrated in Paz and Sinha (1991, 1992) in the context of various specific cosmological models.
Statistical Mechanics and Quantum Cosmology
===========================================
The above two examples give a good illustration of how models of quantum open systems can be used to understand some basic issues of quantum cosmology. The advantage of using statistical mechanical concepts for the study of issues in quantum cosmology and semiclassical gravity has been discussed in general terms by Hu (1991) who emphasized the importance of the interconnection between processes of decoherence, correlation, dissipation, noise and fluctuation, particle creation and backreaction. (see also Hu, Paz and Zhang 1992, 1993, and Gell-Mann and Hartle 1993). In our view, these processes are actually different manifestations of the effect of the environment on the different attributes of the system (the phase information, energy distribution, entropy content, etc). Decoherence is related to particle creation as they both can be related to the Bogolubov coefficients, as is apparent in the example of Calzetta and Mazzitelli (1991) and Paz and Sinha (1992). It is also important to consider the nature of noise and fluctuation for any given environment. In this way their overall effect on the system can be captured more succintly and effectively by general categorical relations like the fluctuation-dissipation relations (see Hu 1989, Hu and Sinha 1993). This also sheds light on how to address questions like defining gravitational entropy and related questions mentioned in the Introduction (see Hu 1993b).
So far one has only succeeded in finding a pathway to show how semiclassical gravity can be deduced from quantum cosmology. One important approximation which makes this transition possible is the assumption of a WKB wave function. To see the true colors of quantum gravity, which is nonlinear and is likely to be also nonlocal, one needs to avoid such simplifications. One should incorporate dynamical fluctuations both in the fields and in the geometry without any background field separation, and deal with nonadiabatic and nonlinear conditions directly. This is a difficult but necessary task. Calzetta (1991) has tackled the anisotropy dissipation problem in quantum cosmology without such an approximation. Recently Calzetta and Hu (1993) have proposed an alternative approach to address the quantum to classical transition issue in terms of correlations between histories. It uses the BBGKY hierarchy truncation scheme to provide a more natural coarse-graining measure which brings about the decoherence of correlation histories. This scheme goes beyond a simple system-environment separation and enables one to deal with nonadiabaticity and nonlinearity directly as in quantum kinetic theory.
[**Acknowledgement**]{}
This essay highlights the work done by Calzetta and the three of us in the years 1987-1992, a good twenty years after Charlie Misner’s seminal works in relativistic cosmology. It is clear from this coarse sampling how much our work is indebted to Misner intellectually. It is also our good fortune to have had personal interactions with Charlie as colleague and friend. We wish him all the best on his sixtieth birthday and look forward to his continuing inspiration for a few more generations of relativists. This work is supported in part by the National Science Foundation under grant PHY91-19726.
[**References**]{} .5cm
A. Anderson, Phys. Rev. [**D42**]{}, 585 (1990)
R. Arnowitt, S. Deser, and C. W. Misner, “The Dynamics of General Relativity.” in [Gravitation: An Introduction to Current Research]{}, ed. L. Witten, pp. 227-65 (Wiley, New York, 1962)
V. A. Belinskii, E. M. Lifschitz and I. M. Khalatnikov, Adv. Phys. [**19**]{}, 525 (1970)
B. K. Berger, Phys. Rev. [**D11**]{}, 2770 (1975)
B. K. Berger, Ann. Phys. (N. Y.) [**83**]{}, 203 (1974)
L. Bianchi, [*Mem. di. Mat. Soc. Ital. Sci.*]{}, [**11**]{}, 267 (1897)
N. D. Birrell and P. C. W. Davies, [*Quantum Fields in Curved Space ,*]{} (Cambridge University Press, Cambridge, 1982).
A. O. Caldeira and A. J. Leggett, Physica [**121A**]{}, 587 (1983)
E. Calzetta, Class. Quant. Grav. [**6**]{}, L227 (1989)
E. Calzetta, Phys. Rev. [**D43**]{}, 2498 (1991)
E. Calzetta and B. L. Hu, Phys. Rev. [**D35**]{}, 495 (1987)
E. Calzetta and B. L. Hu, Phys. Rev. [**D37**]{}, 2838 (1988)
E. Calzetta and B. L. Hu, Phys. Rev. [**D40**]{}, 380 (1989)
E. Calzetta and B. L. Hu, Phys. Rev. [**D40**]{}, 656 (1989)
E. Calzetta and B. L. Hu, “Decoherence of Correlation Histories” in [*Directions in General Relativity*]{} Vol 2 (Brill Festschrift), eds. B. L. Hu, M. P. Ryan and C. V. Vishveshwara (Cambridge Univ. Cambridge, 1993)
E. Calzetta and F. Mazzitelli, Phys. Rev. [**D42**]{}, 4066 (1991)
)
P. Candelas and D. W. Sciama, Phys. Rev. Lett. [**38**]{}, 1372 (1977)
D. M. Capper and M. J. Duff, Nuovo Cimento [**A23**]{}, 173 (1975)
S. Deser, M. J. Duff and C. J. Isham, Nucl. Phys. [**B111**]{}, 45 (1976)
P. C. W. Davies, Rep. Prog. Phys. [**41**]{}, 1313 (1978)
P. C. W. Davies, L. H. Ford and D. N. Page Phys. Rev. [**D34**]{}, 1700 (1986)
B. S. DeWitt, Phys. Rev. [**160**]{}, 1113 (1967)
B. S. DeWitt, Phys. Rep. [**19C**]{}, 297 (1975)
B. S. DeWitt, in [*Quantum Concepts in Space and Time*]{}, ed. R. Penrose and C. J. Isham (Claredon Press, Oxford, 1986)
H. F. Dowker and J. J. Halliwell, Phys. Rev. [**D46**]{}, 1580 (1992)
D. Eardley, E. Liang and R. K. Sachs, J. Math. Phys. [**13**]{}, 99 (1971)
G. F. R. Ellis and M. A. H. MacCallum, [ Commun. Math. Phys.]{} [**12**]{}, 108 (1969)
R. P. Feynman and F. L. Vernon, Ann. Phys. [**24**]{}, 118 (1963)
M. V. Fischetti, J. B. Hartle and B. L. Hu, Phys. Rev. [**D20**]{}, 1757 (1979)
S. A. Fulling, Phys. Rev. [**D7**]{}, 2850 (1973)
M. Gell-Mann and J. B. Hartle, in [*Complexity, Entropy and the Physics of Information*]{}, ed. W. Zurek, Vol. IX (Addison-Wesley, Reading, 1990)
M. Gell-Mann and J. B. Hartle, Phys. Rev. [**D47**]{} (1993)
(1988)
R. Griffiths, J. Stat. Phys. [**36**]{}, 219 (1984)
A. H. Guth, Phys. Rev. [**D23**]{}, 347 (1991)
S. Habib , Phys. Rev. [**D42**]{}, 2566 (1990)
S. Habib and R. Laflamme, Phys. Rev. [**D42**]{}, 4056 (1990)
J. J. Halliwell, Phys. Rev. [**D36**]{}, 3627 (1987)
J. J. Halliwell, Phys. Rev. [**D39**]{}, 2912 (1989)
ys.[**A**]{} (1990)
J. J. Halliwell, Lectures at the 1989 Jerusalem Winter School, in [*Quantum Mechanics and Baby Universes*]{}, eds. S. Coleman, J. Hartle, T. Piran and S. Weinberg (World Scietific, Singapore, 1991)
J. J. Halliwell and S. W. Hawking, Phys. Rev. [**D31**]{}, 1777 (1985)
J. B. Hartle, Phys. Rev. Lett. [**39**]{}, 1373 (1977)
J. B. Hartle, in [*Gravitation in Astrophysics*]{}, NATO Advanced Summer Institute, Cargese, 1986 ed. B. Carter and J. Hartle (NATO ASI Series B: Physics Vol. 156, Plenum, N.Y 1987).
J. B. Hartle and S. W. Hawking, Phys. Rev. [**D28**]{}, 1960 (1983)
J. B. Hartle and B. L. Hu, Phys. Rev. [**D20**]{}, 1772 (1979)
J. B. Hartle and B. L. Hu, Phys. Rev. [**D21**]{}, 2756 (1980)
S. W. Hawking, [*Nature*]{} (London),[**248**]{}, 30 (1974)
O. Heckman and E. L. Schucking, In [*Gravitation: An Introduction to Current Research*]{}, ed. L. Witten (Wiley, New York, 1962)
B. L. Hu, Phys. Lett. [**A90**]{}, 375 (1982)
B. L. Hu, Phys. Lett. [**A97**]{}, 368 (1983)
B. L. Hu, in [*Cosmology of the Early Universe*]{}, ed. L. Z. Fang and R. Ruffini (World Scientific, Singapore, 1984)
B. L. Hu, Physica [**A158**]{}, 399 (1989)
B. L. Hu “Quantum and Statistical Effects in Superspace Cosmology” in [*Quantum Mechanics in Curved Spacetime*]{}, ed. J. Audretsch and V. de Sabbata (Plenum, London 1990)
B. L. Hu, in [*Relativity and Gravitation: Classical and Quantum*]{}, Proc. SILARG VII, Cocoyoc, Mexico 1990, eds. J. C. D’ Olivo et al (World Scientific, Singapore 1991)
B. L. Hu, “Statistical Mechanics and Quantum Cosmology”, in [*Proc. Second International Workshop on Thermal Fields and Their Applications*]{}, eds. H. Ezawa et al (North-Holland, Amsterdam, 1991)
B. L. Hu, “Fluctuation, Dissipation and Irreversibility” in [*The Physical Origin of Time-Asymmetry*]{}, Huelva, Spain, 1991, eds. J. J. Halliwell, J. Perez-Mercader and W. H. Zurek (Cambridge University Press, 1993)
B. L. Hu, “Quantum Statistical Processes in the Early Universe” in [*Quantum Physics and the Universe*]{}, Proc. Waseda Conference, Aug. 1992 ed. Namiki, K. Maeda, et al (Pergamon Press, Tokyo, 1993)
B. L. Hu and H. E. Kandrup, Phys. Rev. [**D35**]{}, 1776 (1987)
B. L. Hu and L. Parker, Phys. Lett. [**63A**]{}, 217 (1977)
B. L. Hu and L. Parker, Phys. Rev. [**D17**]{}, 933 (1978)
B. L. Hu and D. Pavon, Phys. Lett. [**B180**]{}, 329 (1986)
B. L. Hu, J. P. Paz and Y. Zhang, Phys. Rev. [**D45**]{}, 2843 (1992)
B. L. Hu, J. P. Paz and Y. Zhang, Phys. Rev. [**D47**]{} (1993)
B. L. Hu, J. P. Paz and Y. Zhang, “Stochastic Dynamics of Interacting Quantum Fields” Phys. Rev. D (1993)
B. L. Hu and S. Sinha, “Fluctuation-Dissipation Relation in Cosmology” Univ. Maryland preprint (1993)
B. L. Hu and Y. Zhang, “Coarse-Graining, Scaling, and Inflation” Univ. Maryland Preprint 90-186 (1990)
E. Joos and H. D. Zeh, Z. Phys. [**B59**]{}, 223 (1985)
R. D. Jordan, Phys. Rev. [**D33**]{}, 44 (1986)
H. E. Kandrup, Phys. Rev. [**D37**]{}, 3505 (1988)
L. V. Keldysh, Zh. Eksp. Teor. Fiz. [**47**]{}, 1515 (1964) \[Sov. Phys. JEPT [**20**]{}, 1018 (1965)\]
C. Kiefer, Class. Quant. Grav. [**4**]{}, 1369 (1987)
C. Kiefer, Class. Quant. Grav. [**8**]{}, 379 (1991)
K. Kuchar, Phys. Rev. [**D4**]{}, 955 (1971)
K. Kuchar and M. P. Ryan Jr., in [*Proc. of Yamada Conference XIV*]{} ed. H. Sato and T. Nakamura (World Scientific, 1986)
K. Kuchar and M. P. Ryan Jr., Phys. Rev. [**D40**]{}, 3982 (1989)
R. Laflamme and J. Luoko, Phys. Rev. [**D43**]{}, 3317(1991)
and W. Zwerger, Rev. Mod. Phys. 59 No. 1 (1987) 1
E. M. Lifschitz and I. M. Khalatnikov, Adv. Phys. [**12**]{}, 185 (1963)
V. N. Lukash and A. A. Starobinsky, JETP [**39**]{}, 742 (1974)
R. A. Matzner and C. W. Misner, Ap. J. [**171**]{}, 415 (1972)
C. W. Misner, Ap. J. [**151**]{}, 431 (1968)
C. W. Misner, Phys. Rev. Lett. [**22**]{}, 1071 (1969)
C. W. Misner, Phys. Rev. [**186**]{}, 1319 (1969)
C. W. Misner, in [*Magic Without Magic*]{}, ed. J. Klauder (Freeman, San Francisco, 1972)
E. Mottola, Phys. Rev. [**D33**]{}, 2136 (1986)
R. Omnes, J. Stat. Phys. [**53**]{}, 893, 933, 957 (1988); Ann. Phys. (NY) [**201**]{}, 354 (1990): Rev. Mod. Phys. [**64**]{}, 339 (1992)
T. Padmanabhan, Phys. Rev. [**D39**]{}, 2924 (1989)
T. Padmanabhan and T. P. Singh, Class. Quan. Grav. [**7**]{}, 411 (1990)
L. Parker, “The Creation of Particles in an Expanding Universe,” Ph.D. Thesis, Harvard University (unpublished, 1966). L. Parker, [ Phys. Rev. ]{} [**183**]{}, 1057 (1969)
L. Parker, [ Phys. Rev. ]{} [**D7**]{}, 976 (1973)
J. P. Paz, Phys. Rev. [**D41**]{}, 1054 (1990)
J. P. Paz, Phys. Rev. [**D42**]{}, 529 (1990)
.
J. P. Paz and S. Sinha, Phys. Rev. [**D44**]{}, 1038 (1991)
J. P. Paz and S. Sinha, Phys. Rev. [**D45**]{}, 2823 (1992)
J. P. Paz and W. H. Zurek, in preparation (1993)
R. Penrose, “Singularities and Time-Asymmetry” in [*General Relativity: an Einstein Centenary Survey*]{}, eds. S.W. Hawking and W. Israel (Cambridge University Press, Cambridge, 1979)
M. P. Ryan, [*Hamiltonian Cosmology*]{} (Springer, Berlin, 1972)
M. P. Ryan, Jr. and L. C. Shepley, [*Homogeneous Relativistic Cosmologies*]{} (Princeton University Press, Princeton, 1975)
J. Schwinger, [ Phys. Rev.]{}, [**82**]{}, 664 (1951)
J. S. Schwinger, J. Math. Phys. [**2**]{} 407 (1961).
R. U. Sexl and H. K. Urbantke, [ Phys. Rev.]{}, [**179**]{}, 1247 (1969)
S. Sinha, Ph. D. Thesis, University of Maryland (unpublished 1991)
S. Sinha and B. L. Hu, Phys. Rev. [**D44**]{}, 1028 (1991).
A. H. Taub, [ Ann. Math.]{}, [**53**]{}, 472 (1951)
T. Vachaspati and A. Vilenkin, Phys. Rev. [**D37**]{}, 898 (1988)
A. Vilenkin, Phys. Rev. [**D27**]{}, 2848 (1983), [**D30**]{}, 509 (1984); Phys. Lett. [**117B**]{}, 25 (1985)
J. A. Wheeler, in [*Relativity, Groups and Topology*]{}, ed. B. S. De Witt and C. De Witt (Gordon and Breach, New York, 1964)
J. A. Wheeler, in [*Battelle Rencontres*]{} ed. C. DeWitt and J.A Wheeler (Benjamin, N. Y. 1968)
J. A. Wheeler and W. H. Zurek, [*Quantum Theory and Measurement*]{} (Princeton University Press, Princeton 1986)
E. P. Wigner, Phys. Rev. [**40**]{}, 749 (1932)
H. D. Zeh, Phys. Lett. [**A116**]{}, 9 (1986).
Ya. B. Zel’dovich, Pis’ma Zh. Eksp. Teor. Fiz, [**12**]{} ,443 (1970) \[JETP Lett. [**12**]{}, 307(1970)\]
Ya. B. Zel’dovich, in [*Physics of the Expanding Universe*]{}, ed. M. Diemianski (Springer, Berlin, 1979)
Ya. B. Zel’dovich and A. A. Starobinsky, Zh. Teor. Eksp. Fiz. [**61**]{} (1971) 2161 \[JEPT [**34**]{}, 1159 (1972)\]
W. H. Zurek, Phys. Rev. [**D24**]{}, 1516 (1981); [**D26**]{}, 1862 (1982); in [*Frontiers of Nonequilibrium Statistical Physics*]{}, ed. G. T. Moore and M. O. Scully (Plenum, N. Y., 1986); Physics Today [**44**]{}, 36 (1991)
R. Zwanzig, in [*Lectures in Theoretical Physics*]{}, ed. W. E. Britten, B. W. Downes and J. Downes (Interscience, New York, 1961)
[^1]: Department of Physics, University of Maryland, College Park, MD 20742, USA
[^2]: T6 Theoretical Astrophysics, ms 288, LANL, Los Alamos, NM 87545, USA
[^3]: Instituto de Ciencias Nucleares, UNAM, Circuito Exterior, Apartado Postal 70-543, CU Mexico, DF 04510, Mexico.
[^4]: In this example of minisuperspace the scalar field should not be thought of as providing a matter source for the Robertson-Walker background metric, since in that case varying the action with respect to the scale factor will not give the full set of Einstein equations. In particular, the $G_{ij}
= 8 \pi G T_{ij}$ equations that constrain the energy momentum tensor via $T_{ij} = 0$ for $i \neq j$ will be missing.
[^5]: we need to compute only those terms in the effective action that involve the + fields. Terms containing only $-$ fields will not contribute to the equations of motion.
[^6]: Since $\cal K$ is real the backreaction term given above is real and represents a genuine expectation value in the “in" vacuum state rather than an in-out matrix element generated using the in-out effective action.
[^7]: A similar idea has been discussed by Padmanabhan and Singh (1990) in a linearized gravity model where they claim that in order that the minisuperspace approximation be valid, the rate of production of gravitons should be small.
[^8]: for a suggestive argument to justify the choice of $I_{n,n'}(r,r)$ as an indicator of the degree of decoherence between the WKB branches, see Paz and Sinha (1992) Sec. V
[^9]: The diagonalization produced by $I_{n,n}(r_2,r_1)$ has been studied by various authors (Kiefer 1987, Halliwell 1989, Padmandabhan 1989, Laflamme and Louko 1991) and has been identified with decoherence. However, as noted by Paz and Sinha (1992), this term applies more properly to the lack of interference between WKB branches which, as emphasized by Gell-Mann and Hartle (1990), is the more relevant effect. However, it is clear that the diagonalization is an effect that accompanies the former one and has the same origin.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Zheng Yin$^{1}$, Ercai Chen$^{*1,2}$\
1 School of Mathematical Sciences and Institute of Mathematics, Nanjing Normal University,\
Nanjing 210023, Jiangsu, P.R.China\
2 Center of Nonlinear Science, Nanjing University,\
Nanjing 210093, Jiangsu, P.R.China.\
e-mail: [email protected] [email protected]\
title: Large Deviation Properties in Some Nonuniformly Hyperbolic Systems Via Pesin Theory
---
[ This article is devoted to level-1 large deviation properties in some nonuniformly hyperbolic systems via Pesin theory. In particular, our result can be applied to the nonuniformly hyperbolic diffeomorphisms described by Katok and several other classes of diffeomorphisms derived from Anosov systems. ]{}
0.5cm [ Large deviations, Pesin set, weak shadowing property.]{}0.5cm
Introduction
============
The classical theory of large deviations deals with the rates of convergence of sequence of random variables to some limit distribution. We refer to Ellis’ book [@Ell] for a general theory of large deviations and its background in statistical mechanics. We consider large deviations for dynamical systems. Let $(M,d,f)$ be a topological dynamical system and let $\mu$ be a ergodic measure. Given an observable $\varphi\in L^1(\mu)$, let $S_n\varphi(x)=\sum_{i=0}^{n-1}\varphi(f^ix)$. By Birkhoff ergodic theorem, $(1/n)S_n\varphi(x)\to\int\varphi d\mu$ as $n\to\infty$ for almost all $x\in M$. We are interested in the rate of convergence of this time average. More generally, given a reference measure $m$, one interesting question is to study the asymptotic behaviour of $$\begin{aligned}
m\left\{x\in M:\frac{1}{n}S_n\varphi(x)\in E\right\}\end{aligned}$$ for some $E\subseteq \mathbb{R}$. This topic had contributions by many authors in the recent years. We refer the reader to [@AraPac; @Chu; @ComRiv; @EizKifWei; @Kif; @MelNic; @PfiSul; @PolSha; @ReyYou; @Var; @You; @Yur] and the references therein.
The large deviation principle can be divided into two steps, i.e., to obtain the upper bound of $$\begin{aligned}
\limsup\limits_{n\to\infty}\frac{1}{n}\log m\left\{x\in M:\frac{1}{n}S_n\varphi(x)\in E\right\}\end{aligned}$$ and the lower bound of $$\begin{aligned}
\label{equ1.1}
\liminf\limits_{n\to\infty}\frac{1}{n}\log m\left\{x\in M:\frac{1}{n}S_n\varphi(x)\in E\right\}.\end{aligned}$$ The upper bound is usually easier to establish than the lower bound. An effective method in dynamical systems to obtain the lower estimate for (\[equ1.1\]) is “orbit-gluing approach”. For example, Young [@You] formulated quite general large deviation for dynamical systems via specification property. Pfister and Sullivan [@PfiSul] introduced approximate product property to study the large deviations estimates for $\beta$-shifts. Yamamoto [@Yam] considered the large deviation principle for group automorphisms which satisfy the almost specification property. Varandas [@Var] introduced a measure theoretical non-uniform specification to obtain large deviations estimstes for weak Gibbs measures.
Recently, many efforts have been made to extend the theory of large deviations to the scope of nonuniformly hyperbolic systmes. Araújo and Pacífico [@AraPac] gave lower as well as upper bounds in the noninvertible context. Melbourne and Nicol [@MelNic] gave optimal bounds in the Young tower situation (both invertible and noninvertible). Large dviations principles were also obtained by Yuri [@Yur] in the context shifts with countably many symbols and by Comman and Rivera [@ComRiv] for nonuniformly expanding rational maps. More recently, Chung [@Chu] obtained a sufficient condition to hold a full large deviation principle for Markov tower maps induced from return time functions and also gave counterexamples when \[6\]’s conditions do not hold.
Motivated by the work of Young [@You] and Varandas [@Var] our purpose here is to give a contribution for large deviations estimates in some nonuniformly hyperbolic systems. The main tool we use is weak shadowing lemma in Pesin theory. Particularly, our result can be applied (i) to the diffeomorphisms on surfaces, (ii) to the nonuniformly hyperbolic diffeomorphisms described by Katok and several other classes of diffeomorphisms derived from Anosov systems.
This article is organized as follows. In section 2, we provide some notions and results of Pesin theory and state the main result. Section 3 is devoted to the proof of the main result. Examples and applications are given in section 4.
Preliminaries
==============
In this section, we first present some notations to be used in this paper. Then we introduce some notions and results of Pesin theory [@BarPes2; @KatHas; @Pol] and state the main results.
We denote by $\mathscr M(M,f)$, $\mathscr M_{\rm inv}(M,f)$ and $\mathscr M_{\rm erg}(M,f)$ the set of all Borel probability measures, $f$-invariant Borel probability measures and ergodic measures respectively. It is well known that $\mathscr M(M,f)$ equiped with weak\* topology is a compact metrisable space. We will denote by $D$ the metric on $\mathscr M(M,f)$. For an $f$-invariant subset $Z\subset X,$ let $\mathscr M_{\rm inv}(Z,f)$ denote the subset of $\mathscr M_{\rm inv}(M,f)$ for which the measures $\mu$ satisfy $\mu(Z)=1$ and $\mathscr M_{\rm erg}(Z,f)$ denote those which are ergodic. Denote by $C^0(M)$ the space of continuous functions from $M$ to $\mathbb{R}$ with the sup norm. For $\varphi\in C^0(M)$ and $n\geq1$ we denote $\sum_{i=0}^{n-1}\varphi(f^ix)$ by $S_n\varphi(x)$. For every $\epsilon>0$, $n\in \mathbb{N}$ and a point $x\in M$, define $B_n(x,\epsilon)=\{y\in M:d(f^ix,f^iy)<\epsilon,\forall 0\leq i\leq n-1\}$.
Suppose $M$ is a compact connected boundary-less Riemannian $n$-dimension manifold and $f:X\to X$ is a $C^{1+\alpha}$ diffeomorphism. Let $\mu\in\mathscr M_{\rm erg}(Z,f)$ and $Df_x$ denote the tangent map of $f$ at $x\in M.$ We say that $x\in X$ is a regular point of $f$ if there exist $\lambda_1(\mu)>\lambda_2(\mu)>\cdots>\lambda_{\phi(\mu)}(\mu)$ and a decomposition on the tangent space $T_x
M=E_1(x)\oplus\cdots\oplus E_{\phi(\mu)}(x)$ such that $$\begin{aligned}
\lim\limits_{n\to\infty}\frac{1}{n}\log\|(Df^n_x)u\|=\lambda_j(x),\end{aligned}$$ where $0\neq u\in E_j(x), 1\leq j\leq \phi(\mu).$ The number $\lambda_j(x)$ and the space $E_j(x)$ are called the Lyapunov exponents and the eigenspaces of $f$ at the regular point $x,$ respectively. Oseledets theorem [@Ose] say that all regular points forms a Borel set with total measure. For a regular point $x\in M$, we define $$\begin{aligned}
\lambda^+(\mu)=\min\{\lambda_i(\mu)|\lambda_i(\mu)\geq0,1\leq i\leq \phi(\mu)\}\end{aligned}$$ and $$\begin{aligned}
\lambda^-(\mu)=\min\{-\lambda_i(\mu)|\lambda_i(\mu)\leq0,1\leq i\leq \phi(\mu)\}.\end{aligned}$$ We appoint $\min\emptyset=0$. An ergodic measure $\omega$ is hyperbolic if $\lambda^+(\omega)$ and $\lambda^-(\omega)$ are both non-zero.
Given $\beta_1,\beta_2\gg\epsilon>0$ and for all $k\in\mathbb{Z}^+,$ the hyperbolic block $\Lambda_k=\Lambda_k(\beta_1,\beta_2,\epsilon)$ consists of all points $x\in M$ such that there exists a decomposition $T_xM=E_x^s\oplus E_x^u$ satisfying:
- $Df^t(E_x^s)=E^s_{f^tx}$ and $Df^t(E_x^u)=E^u_{f^tx};$
- $\|Df^n|E^s_{f^tx}\|\leq e^{\epsilon k}e^{-(\beta_1-\epsilon)n}e^{\epsilon|t|},\forall t\in\mathbb{Z},n\geq1;$
- $\|Df^{-n}|E^u_{f^tx}\|\leq e^{\epsilon k}e^{-(\beta_2-\epsilon)n}e^{\epsilon|t|},\forall t\in\mathbb{Z},n\geq1;$
- $\tan (\angle(E^s_{f^tx},E^u_{f^tx}))\geq e^{-\epsilon k}e^{-\epsilon|t|},\forall t\in\mathbb{Z}.$
$
\Lambda=\Lambda(\beta_1,\beta_2,\epsilon)=\bigcup\limits_{k=1}^\infty\Lambda_k(\beta_1,\beta_2,\epsilon)
$ is a Pesin set.
The following statements are elementary properties of Pesin blocks (see [@Pol]):
- $\Lambda_1\subseteq\Lambda_2\subseteq\cdots;$
- $f(\Lambda_k)\subseteq\Lambda_{k+1},f^{-1}(\Lambda_k)\subseteq\Lambda_{k+1};$
- $\Lambda_k$ is compact for each $k\geq1$;
- For each $k\geq1$, the splitting $\Lambda_k\ni x\mapsto E_x^s\oplus E_x^u$ is continuous.
The Pesin set $\Lambda(\beta_1,\beta_2,\epsilon)$ is an $f$-invariant set but usually not compact. Given an ergodic measure $\mu\in\mathscr M_{\rm erg}(M,f)$, denote by $\mu|\Lambda_l$ the conditional measure of $\mu$ on $\Lambda_l.$ Let $\widetilde{\Lambda}_l=$ supp$(\mu|\Lambda_l)$ and $\widetilde{\Lambda}_\mu=\bigcup_{l\geq1}\widetilde{\Lambda}_l.$ If $\omega$ is an ergodic hyperbolic measure for $f$ and $\beta_1\leq\lambda^-(\omega)$ and $\beta_2\leq\lambda^+(\omega)$, then $\omega\in \mathscr M_{\rm inv}(\widetilde{\Lambda}_\omega,f)$.
Let $\{\delta_k\}_{k=1}^\infty $ be a sequence of positive real numbers. Let $\{x_n\}_{n=-\infty}^\infty$ be a sequence of points in $\Lambda=\Lambda(\beta_1,\beta_2,\epsilon)$ for which there exists a sequence $\{s_n\}_{n=-\infty}^{\infty}$ of positive integers satisfying: $$\begin{split}
&\text{(a) } x_n\in\Lambda_{s_n},\forall n\in\mathbb{Z};\\
&\text{(b) } |s_n-s_{n-1}|\leq 1, \forall n\in\mathbb{Z};\\
&\text{(c) } d(f(x_n),x_{n+1})\leq\delta_{s_n}, \forall n\in\mathbb{Z},
\end{split}$$ then we call $\{x_n\}_{n=-\infty}^\infty$ a $\{\delta_k\}_{k=1}^\infty$ pseudo-orbit. Given $\eta>0$ a point $x\in M$ is an $\eta$-shadowing point for the $\{\delta_k\}_{k=1}^\infty$ pseudo-orbit if $d(f^n(x),x_n)\leq
\eta\epsilon_{s_n},\forall n\in\mathbb{Z},$ where $\epsilon_k=\epsilon_0e^{-\epsilon k}$ and $\epsilon_0$ is a constant only dependent on the system of $f$.
0.3cm
**Weak shadowing lemma.** [@Hir; @KatHas; @Pol] [*Let $f:M\to M$ be a $C^{1+\alpha}$ diffeomorphism, with a non-empty Pesin set $\Lambda=\Lambda(\beta_1,\beta_2,\epsilon)$ and fixed parameters, $\beta_1,\beta_2\gg \epsilon>0.$ For $\eta>0$ there exists a sequence $\{\delta_k\}$ such that for any $\{\delta_k\}$ pseudo orbit there exists a unique $\eta$-shadowing point.* ]{} 0.3cm
Let $m$ be a finite Borel measure on $M$. We think of $m$ as our reference measure. We define $$\begin{aligned}
\mathcal{V}^+=\big\{\psi\in C^0(M):\exists C,\epsilon>0
\text{ such that } \forall x\in M \text{ and }& \forall n\geq0, \\
&mB_n(x,\epsilon)\leq Ce^{-S_n\psi(x)}\big\}\end{aligned}$$ and $$\begin{aligned}
\mathcal{V}^-=\big\{\psi\in C^0(M):&\exists \text{ arbitrarily small }\epsilon>0
\text{ and } C=C(\epsilon)>0 \\
&\text{ such that } \forall x\in M \text{ and } \forall n\geq0,
mB_n(x,\epsilon)\geq Ce^{-S_n\psi(x)}\big\}.\end{aligned}$$ For $\varphi\in C^0(M)$ and $E\subseteq \mathbb{R}$, we write $$\begin{aligned}
\overline{R}(\varphi,E)=
\limsup_{n\to \infty}\frac{1}{n}\log m\left\{x\in M:\frac{1}{n}S_n\varphi(x)\in E\right\}\end{aligned}$$ and $$\begin{aligned}
\underline{R}(\varphi,E)=
\liminf_{n\to \infty}\frac{1}{n}\log m\left\{x\in M:\frac{1}{n}S_n\varphi(x)\in E\right\}.\end{aligned}$$ Now, we state the main result of this paper as follows:
\[thm2.2\] Let $f:M\to M$ be a $C^{1+\alpha}$ diffeomorphism of a compact Riemannian manifold, with a non-empty Pesin set $\Lambda=\Lambda(\beta_1,\beta_2,\epsilon)$ and fixed parameters, $\beta_1,\beta_2\gg \epsilon>0.$ Then for every $\varphi\in C^0(M)$ and $c\in \mathbb{R}$, the following hold:
1. For $\psi\in \mathcal{V}^+$, we have $$\begin{aligned}
\overline{R}(\varphi,[c,\infty))
\leq\sup\left\{h_\nu(f)-\int\psi d\nu:\nu\in \mathscr M_{\rm inv}(M,f),\int\varphi d\nu\geq c\right\}.\end{aligned}$$
2. For $\psi\in \mathcal{V}^-$ and $\mu\in \mathscr{M}_{erg}(M,f)$, we have $$\begin{aligned}
\underline{R}(\varphi,(c,\infty))
\geq\sup\left\{h_\nu(f)-\int\psi d\nu:\nu\in \mathscr{M}_{inv}(\widetilde{\Lambda}_\mu,f),\int\varphi d\nu> c\right\}.\end{aligned}$$
\[cro2.1\] Let $f:M\to M$ be a $C^{1+\alpha}$ diffeomorphism of a compact Riemannian manifold and let $\omega\in\mathscr M_{\rm erg}(M,f)$ be a hyperbolic measure. If $\beta_1\leq\lambda^-(\omega)$ and $\beta_2\leq\lambda^+(\omega)$, then for every $\varphi\in C^0(M)$ and $c\in \mathbb{R}$, the following hold:
1. For $\psi\in \mathcal{V}^+$, we have $$\begin{aligned}
\overline{R}(\varphi,[c,\infty))
\leq\sup\left\{h_\nu(f)-\int\psi d\nu:\nu\in \mathscr M_{\rm inv}(M,f),\int\varphi d\nu\geq c\right\}.\end{aligned}$$
2. For $\psi\in \mathcal{V}^-$, we have $$\begin{aligned}
\underline{R}(\varphi,(c,\infty))
\geq\sup\left\{h_\nu(f)-\int\psi d\nu:\nu\in \mathscr{M}_{inv}(\widetilde{\Lambda}_\omega,f),\int\varphi d\nu> c\right\},\end{aligned}$$
where $\widetilde{\Lambda}_\omega=\bigcup_{l\geq1}{\rm supp}(\omega|\Lambda_l(\beta_1,\beta_2,\epsilon))$.
Proof of Main Theorem
=====================
In this section, we will verify theorem \[thm2.2\]. The upper bound for the measure of deviation holds for all topological dynamical systems. By [@You], for $\psi\in \mathcal{V}^+$ we have $$\begin{aligned}
\overline{R}(\varphi,[c,\infty))
\leq\sup\left\{h_\nu(f)-\int\psi d\nu:\nu\in \mathscr M_{\rm inv}(M,f),\int\varphi d\nu\geq c\right\}.\end{aligned}$$ To obtain the lower bound estimate we need to construct a suitable pseudo-orbit. Our method is inspired by [@LiaLiaSUnTia] and [@You]. Firstly, we give some important lemmas as follows.
\[lem3.1\][[@Kat2]]{} Let $(X,d)$ be a compact metric space, $f:X\to X$ be a continuous map and $\mu$ be an ergodic invariant measure. For $\epsilon>0$, $\gamma\in (0,1)$ let $N^\mu(n,\epsilon,\gamma)$ denote the minimum number of $\epsilon$-Bowen balls $B_n(x,\epsilon)$, which cover a set of $\mu$-measure at least $1-\gamma$. Then $$\begin{aligned}
h_\mu(f)=\lim_{\epsilon\to0}\liminf_{n\to\infty}\frac{1}{n}\log N^\mu(n,\epsilon,\gamma)=
\lim_{\epsilon\to0}\limsup_{n\to\infty}\frac{1}{n}\log N^\mu(n,\epsilon,\gamma).\end{aligned}$$
[[@Boc]]{}\[lem3.2\] Let $f:M\to M$ be a $C^{1}$ diffeomorphism of a compact Riemannian manifold and $\mu\in\mathscr M_{\rm inv}(M,f)$. Let $\Gamma\subseteq M$ be a measurable set with $\mu(\Gamma)>0$ and let $$\begin{aligned}
\Omega=\bigcup_{n\in \mathbb{Z}}f^n(\Gamma).\end{aligned}$$ Take $\gamma>0$. Then there exists a measurable function $N_0:\Omega\to \mathbb{N}$ such that for a.e.$x\in\Omega$ and every $t\in[0,1]$ there is some $l\in\{0,1,\cdots,n\}$ such that $f^l(x)\in\Gamma$ and $\left|(l/n)-t\right|<\gamma$.
Fix $\psi\in \mathcal{V}^-$. Pick an arbitrary $\nu\in \mathscr{M}_{inv}(\widetilde{\Lambda},f)$ with $\int\varphi d\nu> c$. We will prove that $$\begin{aligned}
\liminf_{n\to \infty}\frac{1}{n}\log\left\{x\in M:\frac{1}{n}S_n\varphi(x)\in E\right\}
\geq h_\nu(f)-\int\psi d\nu-10\gamma-\gamma h_{top}(f).\end{aligned}$$ for any preassigned $0<\gamma<1$. Let $\delta=\frac{1}{5}(\int\varphi d\nu-c)$.
\[lem3.3\] For $\gamma>0,\delta>0$ and $\nu\in\mathscr{M}_{inv}(\widetilde{\Lambda},f)$ there exists a finite convex combination of ergodic probability measures with rational coefficients $\lambda=\sum\limits_{i=1}^{k}a_{i}\mu_{i}$ such that $$\begin{aligned}
\mu_{i}(\widetilde{\Lambda})=1,
\int\varphi d\lambda\geq c+4\delta,\int\psi d\nu\geq\int\psi d\lambda-\gamma \text{ and } h_\lambda(f)\geq h_\nu(f)-\gamma.\end{aligned}$$
We choose $\epsilon_1>0$ sufficiently small such that for any $\mu_1,\mu_2\in\mathscr M_{\rm inv}(M,f)$ with $D(\mu_1,\mu_2)<\epsilon_1$ we have $$\begin{aligned}
\left|\int\varphi d\mu_1-\int\varphi d\mu_2\right|<\delta,
\left|\int\psi d\mu_1-\int\psi d\mu_2\right|<\gamma.\end{aligned}$$ By ergodic decomposition theorem, we can write $\nu=\int_{\mathscr{M}_{erg}(\widetilde{\Lambda},f)}md\tau(m)$. Let $\mathscr{P}=\{P_1,P_2,\cdots,P_k\}$ be a partition of $\mathscr{M}_{erg}(\widetilde{\Lambda},f)$ with diam$\mathscr{P}<\epsilon_1$. Let $a_i=\tau(P_i)$. For every $1\leq i\leq k$ we can choose an ergodic measure $\mu_i\in P_i$ satisfying $h_{\mu_i}(f)\geq h_m(f)-\gamma$ for $\tau$-almost every $m\in P_i$. Let $\lambda=\sum\limits_{i=1}^{k}a_{i}\mu_{i}$. Obviously $\mu_i(\widetilde{\Lambda})=1,1\leq i\leq k$. By ergodic decomposition theorem one can easily prove the remaining inqualities. Since the $a_i$’s can be approximated by rational numbers, the desired results follows.
Since $\mu_i(\widetilde{\Lambda})=1$, we can choose $l\in \mathbb{N}$ such that $\mu_i(\widetilde{\Lambda}_l)>1-\gamma$ for all $1\leq i\leq k$. By lemma \[lem3.1\], we can choose $\epsilon'$ sufficiently small so $$\begin{aligned}
\label{equ3.1}
d(x,y)<\epsilon'\Rightarrow\left|\varphi(x)-\varphi(y)\right|<\delta,\left|\psi(x)-\psi(y)\right|<\gamma\end{aligned}$$ and $$\begin{aligned}
\liminf_{n\to\infty}\frac{1}{n}\log N^{\mu_{i}}(n,4\epsilon',\gamma)>h_{\mu_{i}}(f)-\gamma\end{aligned}$$ for $1\leq i\leq k$. Let $\eta=\frac{\epsilon'}{\epsilon_0},$ it follows from weak shadowing lemma that there is a sequence of numbers $\{\delta_k\}.$ Let $\xi$ be a finite partition of $X$ with diam$(\xi)<\frac{\delta_{l}}{3}$ and $\xi\geq\{\widetilde{\Lambda}_{l}, M\setminus \widetilde{\Lambda}_{l} \}$. For $1\leq i\leq k, n\in\mathbb{N},$ we consider the set $$\begin{aligned}
\Lambda_1^n(\mu_{i})=\left\{x\in\widetilde{\Lambda}_{l}:f^q(x)\in\xi(x){\rm ~for~some~} q\in[n,(1+\gamma)n]\right\}\end{aligned}$$ and $$\label{equ3.2}
\begin{split}
\Lambda_2^n(\mu_{i})
=\bigg\{x\in\widetilde{\Lambda}_{l}:&S_m\psi(x)\leq m(\int \psi d\mu_i+\gamma) \text{ and }\\
&S_m\varphi(x)\geq m(\int \varphi d\mu_{i}-\delta){\rm~for~all~} m\geq n\bigg\},
\end{split}$$ where $\xi(x)$ is the element in $\xi$ containing $x.$ Let $\Lambda^n(\mu_{i})=\Lambda_1^n(\mu_{i})\cap \Lambda_2^n(\mu_{i})$. Then we have the following lemma.
$\lim\limits_{n\to\infty}\mu_{i}(\Lambda^n(\mu_{i}))=\mu_{i}(\widetilde{\Lambda}_{l}),1\leq i\leq k.$
By Birkhoff ergodic theorem, it suffices to prove that $$\begin{aligned}
\lim_{n\to\infty}\mu_{i}(\Lambda_1^n(\mu_{i}))=\mu_{i}(\widetilde{\Lambda}_{l}),1\leq i\leq k.\end{aligned}$$ Let $[(1+\gamma)n]$ be the integer part of $(1+\gamma)n$. Take an element $\xi_k$ of $\xi$ and let $\Gamma=\xi_k$. Applying lemma \[lem3.2\], we can deduce that for $a.e.x\in \xi_k$, there exists a measurable function $N_0$ such that for every $[(1+\gamma)n]\geq N_0(x)$ there is some $q\in\{0,1,\cdots,[(1+\gamma)n]\}$ such that $f^q(x)\in \xi_k$ and $$\begin{aligned}
\frac{q}{[(1+\gamma)n]}-1>-\frac{\gamma}{1+\gamma}.\end{aligned}$$ We obtain $q\geq n-\frac{1}{1+\gamma}$. Since $q\in \mathbb{N}$, we have $q\in [n,(1+\gamma)n]$ and the proof is completed.
By lemma \[lem3.1\], we can take sufficiently large $K\in\mathbb{N}$ such that $$\begin{aligned}
\mu_{i}(\Lambda^n(\mu_{i}))>1-\gamma\end{aligned}$$ for all $n\geq M$ and $1\leq i\leq k.$
Let $$\begin{aligned}
Q(\Lambda^n(\mu_{i}),4\epsilon')&=\inf\{\sharp S:S \text{ is } (n,4\epsilon') \text{ spanning set for } \Lambda^n(\mu_{i}) \},\\
P(\Lambda^n(\mu_{i}),4\epsilon')&=\sup\{\sharp S:S \text{ is } (n,4\epsilon') \text{ separated set for } \Lambda^n(\mu_{i}) \}.\end{aligned}$$ Then for all $n\geq K$ and $1\leq i\leq k$, we have $$\begin{aligned}
P(\Lambda^n(\mu_{i}),4\epsilon')\geq Q(\Lambda^n(\mu_{i}),4\epsilon')\geq N^{\mu_{i}}(n,\epsilon',\gamma).\end{aligned}$$ We obtain $$\begin{aligned}
\liminf_{n\to\infty}\frac{1}{n}\log P(\Lambda^n(\mu_{i}),4\epsilon')
\geq \liminf_{n\to\infty}\frac{1}{n}\log N^{\mu_{i}}(n,4\epsilon',\gamma)
>h_{\mu_{i}}(f)-\gamma.\end{aligned}$$ Thus we can choose $t\in \mathbb{N}$ large enough such that $\exp(t\gamma)>\sharp \xi$ and $$\begin{aligned}
\frac{1}{t}\log P(\Lambda^{t}(\mu_{i}),4\epsilon')>h_{\mu_{i}}(f)-2\gamma.\end{aligned}$$ for $1\leq i\leq k$. Let $S_i$ be a $(t,4\epsilon')$-separated set for $\Lambda^{t}(\mu_{i})$ and $$\begin{aligned}
\# S_i\geq\exp\left(t(h_{\mu_{i}}(f)-3\gamma)\right).\end{aligned}$$ For each $q\in[t,(1+\gamma)t],$ let $$\begin{aligned}
V_q(\mu_i)=\{x\in S_i:f^q(x)\in\xi(x)\}\end{aligned}$$ and let $n_i$ be the value of $q$ which maximizes $\#V_q(\mu_i).$ Obviously, $n_i\geq t\geq\frac{n_i}{1+\gamma}\geq n_i(1-\gamma).$ Since $\exp(t\gamma)\geq t\gamma+1,$ we have that $$\begin{aligned}
\#V_{n_i}(\mu_i)\geq\frac{\#S_i}{t\gamma+1}\geq \exp\left(t(h_{\mu_{i}}(f)-4\gamma)\right).\end{aligned}$$ Consider the element $A_{i}\in\xi$ such that $\#(V_{n_i}(\mu_i)\cap A_{i})$ is maximal. Let $W_{i}=V_{n_i}(\mu_i)\cap A_{i}$. It follows that $$\begin{aligned}
\#W_{i}\geq \frac{1}{\#\xi}\#V_{n_i}(\mu_i)\geq \frac{1}{\#\xi}\exp\left(t(h_{\mu_{i}}(f)-4\gamma)\right).\end{aligned}$$ Since $e^{t\gamma}>\sharp \xi$, $t\geq n_i(1-\gamma)$ and $\#W_i\geq1$, we have $$\begin{aligned}
\#W_{i}\geq \exp\left(n_i(1-\gamma)(h_{\mu_{i}}(f)-5\gamma)\right).\end{aligned}$$ Notice that $A_{i}$ is contained in an open subset $U_i$ with diam$(U_i)\leq3$diam$(\xi).$ By the ergodicity of $\mu,$ for any two measures $\mu_{i_1},\mu_{i_2}$, there exists $s=s(\mu_{i_1},\mu_{i_2})\in\mathbb{N}$ and $y=y(\mu_{i_1},\mu_{i_2})\in U_{i_1}\cap\widetilde{\Lambda}_{l}$ such that $f^s(y)\in U_{i_2}\cap\widetilde{\Lambda}_{l}.$ Letting $C_{i}=\frac{a_{i}}{n_i},$ we can choose an integer $N$ large enough so that $NC_{i}$, $1\leq i\leq k$ are integers and $$\begin{aligned}
N\gamma\geq \sum\limits_{i=1}^{k-1}s(\mu_{i},\mu_{i+1}).\end{aligned}$$ Let $X=\sum\limits_{i=1}^{k-1}s(\mu_{i},\mu_{i+1})$ and $$\begin{aligned}
Y_n=n\sum\limits_{i=1}^{k}Nn_iC_i+X=nN+X,n=1,2,\cdots.\end{aligned}$$ Then we have $$\begin{aligned}
\label{equ3.3}
\frac{X}{N}\leq\gamma, \frac{N}{Y_n}\geq \frac{1}{n+\gamma},\frac{X}{Y_n}\leq \frac{\gamma}{n+\gamma}.\end{aligned}$$ For simplicity of the statement, for $x\in M$ define segments of orbits $$\begin{aligned}
L_{i}(x)&:=(x,f(x),\cdots,f^{n_i-1}(x)), 1\leq i\leq k,\\
\widehat{L}_{i_1,i_2}(x)&:=(x,f(x),\cdots,f^{s(\mu_{i_1},\mu_{i_2})-1}(x)),1\leq i_1,i_2\leq k.\end{aligned}$$
![Quasi-orbit[]{data-label="figure1"}](figure.eps)
For every $n\in \mathbb{N}$, consider the pseudo-orbit with finite length (see Figure \[figure1\]) $$\begin{aligned}
\Big\{ &L_{1}(x(1,1)),L_{1}(x(1,2)), \cdots,L_{1}(x(1,nNC_{1})),\widehat{L}_{1,2}(y(\mu_{1},\mu_{2}));\\
&L_{2}(x(2,1)),L_{2}(x(2,2)), \cdots,L_{2}(x(2,nNC_{2})),\widehat{L}_{2,3}(y(\mu_{2},\mu_{3}));\\
&\vdots\\
&L_{k}(x(k,1)),L_{k}(x(k,2)), \cdots,L_{k}(x(k,nNC_{k}))
\Big\},\end{aligned}$$ where $x(i,j)\in W_{i},1\leq i\leq k,1\leq j\leq nNC_i.$
For $1\leq i\leq k$, $1\leq j\leq nNC_i$, let $$\begin{aligned}
M_1&=0,\\
M_i&=\sum_{l=1}^{i-1}(nNC_ln_l+s(\mu_l,\mu_{l+1})),\\
M_{i,j}&=M_i+(j-1)n_i.\end{aligned}$$ By weak shadowing lemma, there exists a shadowing $z$ for the above pseudo-orbit satisfying $$\begin{aligned}
d(f^{M_{i,j}+q}(z),f^q(x(i,j)))\leq \eta\epsilon_0e^{-\epsilon l}
=\frac{\epsilon'}{\epsilon_0}\epsilon_0e^{-\epsilon l}\leq \epsilon'\end{aligned}$$ for $0\leq q\leq n_i-1$. The shadowing point $z$ can be considered as a map with the variables $x(i,j)$. We define $F_n$ to be the image of the map. More precisely, $$\begin{aligned}
F_n=\Big\{z:z&=z(x(1,1),x(1,2), \cdots,x(1,nNC_{1}),
x(2,1),x(2,2), \cdots,x(2,nNC_{2}),\cdots, \\
&x(k,1),x(k,2), \cdots,x(k,nNC_{k})),x(i,j)\in W_i,1\leq i\leq k,1\leq j\leq nNC_i\Big\}.\end{aligned}$$ Since $W_{i}$ is $(n_i,4\epsilon')$ separated sets, for distinct $x(i,j),x'(i,j)\in W_{i}$ the corresponding shadowing points $z,z'\in F_n$ satisfying $B_{Y_n}(z,\epsilon')\cap B_{Y_n}(z',\epsilon')=\emptyset$. Thus $$\label{equ3.4}
\begin{split}
\#F_n&\geq(\#W_1)^{nNC_1}(\#W_2)^{nNC_2}\cdots(\#W_k)^{nNC_k}\\
&\geq\exp\left\{\sum_{i=1}^k nNn_iC_i(1-\gamma)(h_{\mu_{i}}(f)-5\gamma)\right\}\\
&=\exp \left\{nN(1-\gamma)(h_\lambda(f)-5\gamma)\right\}.
\end{split}$$
\[lem3.5\] For sufficiently large $n$, we have $B_{Y_n}(z,\epsilon')\subseteq \left\{x\in M:\frac{1}{Y_n}S_{Y_n}\varphi(x)>c\right\}$ for all $z\in F_n$.
From (\[equ3.1\]), it suffices to prove that for sufficiently large $n$, $$\begin{aligned}
\frac{1}{Y_n}S_{Y_n}\varphi(z)>c+\delta.\end{aligned}$$ By (\[equ3.1\]), (\[equ3.2\]), $a_i=C_in_i$ and lemma \[lem3.3\], we have $$\begin{aligned}
\frac{1}{Y_n}S_{Y_n}\varphi(z)
&\geq \frac{1}{Y_n}\sum_{i=1}^{k}\sum_{j=1}^{nNC_i}S_{n_i}\varphi(x(i,j))-\frac{nN\delta}{Y_n}-\frac{X\|\varphi\|}{Y_n}\\
&\geq \frac{1}{Y_n}\sum_{i=1}^{k}nNC_in_i(\int\varphi d\mu_i-\delta)-\frac{nN\delta}{Y_n}-\frac{X\|\varphi\|}{Y_n}\\
&= \frac{nN}{Y_n}(\int\varphi d\lambda-\delta)-\frac{nN\delta}{Y_n}-\frac{X\|\varphi\|}{Y_n}\\
&\geq \frac{nN}{Y_n}(c+3\delta)-\frac{nN\delta}{Y_n}-\frac{X\|\varphi\|}{Y_n}.\end{aligned}$$ It follows from (\[equ3.3\]) we obtain $$\begin{aligned}
\lim_{n\to\infty}\frac{nN}{Y_n}=1,\lim_{n\to\infty}\frac{X}{Y_n}=0.\end{aligned}$$ Thus for sufficiently large $n$, we have $$\begin{aligned}
\label{equ3.5}
\frac{1}{Y_n}S_{Y_n}\varphi(z)>c+\frac{3}{2}\delta.\end{aligned}$$ Hence the desired result follows.
\[lem3.6\] For sufficiently large $n$, $$\begin{aligned}
\frac{1}{Y_n}\log m\left\{x\in M:\frac{1}{Y_n}S_{Y_n}\varphi(x)>c\right\}\geq h_\nu(f)-\int \psi d\nu-10\gamma-\gamma h_{top}(f).\end{aligned}$$
By lemma \[lem3.5\], we have $$\begin{aligned}
&\frac{1}{Y_n}\log m\left\{x\in M:\frac{1}{Y_n}S_{Y_n}\varphi(x)>c\right\}\\
\geq &\frac{1}{Y_n}\log\sum_{z\in F_n}mB_{Y_n}(z,\epsilon')
\geq \frac{1}{Y_n}\log\sum_{z\in F_n}C\exp(-S_{Y_n}\psi(z))\\
= &\frac{1}{Y_n}\log\sum_{z\in F_n}\exp(-S_{Y_n}\psi(z))+\frac{1}{Y_n}\log C.\end{aligned}$$ Applying (\[equ3.1\]) and (\[equ3.2\]), we have $$\begin{aligned}
S_{Y_n}\psi(z)<&\sum_{i=1}^{k}\sum_{j=1}^{nNC_i}S_{n_i}\psi(x(i,j))+nN\gamma+X\|\psi\|\\
\leq &\sum_{i=1}^{k}nNC_in_i(\int\psi d\mu_i+\gamma)+nN\gamma+X\|\psi\|\\
= &nN\int\psi d\lambda+2nN\gamma+X\|\psi\|.\end{aligned}$$ It follows from (\[equ3.4\]) lemma \[lem3.3\] that $$\begin{aligned}
&\frac{1}{Y_n}\log m\left\{x\in M:\frac{1}{Y_n}S_{Y_n}\varphi(x)>c\right\}\\
\geq &\frac{1}{Y_n}\log \sum_{z\in F_n}\exp\left\{-nN\int\psi d\lambda-2nN\gamma-X\|\psi\|\right\}+\frac{1}{Y_n}\log C\\
\geq &\frac{1}{Y_n}\log \exp\left\{nN(1-\gamma)(h_\lambda(f)-5\gamma)-nN\int\psi d\lambda-2nN\gamma-X\|\psi\|\right\}
+\frac{1}{Y_n}\log C\\
=&\frac{1}{Y_n}\left\{nN(h_\lambda(f)-\int\psi d\lambda)+nN(5\gamma^2-7\gamma-\gamma h_\lambda(f))-X\|\psi\|\right\}
+\frac{1}{Y_n}\log C\\
\geq&\frac{1}{Y_n}\left\{nN(h_\nu(f)-\int\psi d\nu)+nN(-9\gamma-\gamma h_\lambda(f))-X\|\psi\|\right\}+\frac{1}{Y_n}\log C.\end{aligned}$$ Together with $$\begin{aligned}
\lim_{n\to\infty}\frac{nN}{Y_n}=1,\lim_{n\to\infty}\frac{X}{Y_n}=0,\end{aligned}$$ for sufficiently large $n$, we have $$\begin{aligned}
\frac{1}{Y_n}\log m\left\{x\in M:\frac{1}{Y_n}S_{Y_n}\varphi(x)>c\right\}
\geq h_\nu(f)-\int\psi d\nu-10\gamma-\gamma h_{top}(f).\end{aligned}$$
For any $n\in \mathbb{N}$, let $i_n$ be the unique natural number such that $$\begin{aligned}
Y_{i_n}\leq n<Y_{i_n+1}.\end{aligned}$$ By (\[equ3.5\]), for sufficiently large $n$ we have $$\begin{aligned}
\frac{1}{n}S_{n}\varphi(z)>c+\delta \text{ and }
B_{n}(z,\epsilon')\subseteq \left\{x\in M:\frac{1}{n}S_{n}\varphi(x)>c\right\}\end{aligned}$$ for all $z\in F_{i_n}$. Combining this with previous lemma we obtain $$\begin{aligned}
&\frac{1}{n}\log m\left\{x\in M:\frac{1}{n}S_{n}\varphi(x)>c\right\}\\
\geq &\frac{Y_{i_n}}{n}\cdot\frac{1}{Y_{i_n}}\log\sum_{z\in F_{i_n}} mB(z,\epsilon')\\
\geq &\frac{Y_{i_n}}{n}\left\{h_\nu(f)-\int\psi d\nu-10\gamma-\gamma h_{top}(f)\right\}.\end{aligned}$$ for sufficiently large $n$. Since $\gamma$ was arbitrary, the proof of theorem \[thm2.2\] is completed.
Some Applications
=================
[**Example 1 Diffeomorphisms on surfaces**]{} Let $f:M\to M$ be a $C^{1+\alpha}$ diffeomorphism with $\dim M=2$ and $h_{top}(f)>0$, then there exists a hyperbolic measure $m\in\mathscr M_{\rm erg}(M,f)$ with Lyapunov exponents $\lambda_1>0>\lambda_2$ (see [@Pol]). If $\beta_1=|\lambda_2|$ and $\beta_2=\lambda_1$, then for any $\epsilon>0$ such that $\beta_1,\beta_2>\epsilon$, we have $m(\Lambda(\beta_1,\beta_2,\epsilon))=1$. Let $$\begin{aligned}
\widetilde{\Lambda}=\bigcup_{k=1}^{\infty}\text{supp}(m|\Lambda(\beta_1,\beta_2,\epsilon)),\end{aligned}$$ then for every $\varphi\in C^0(M)$ and $c\in \mathbb{R}$, the following hold:
1. For $\psi\in \mathcal{V}^+$, we have $$\begin{aligned}
\overline{R}(\varphi,[c,\infty))
\leq\sup\left\{h_\nu(f)-\int\psi d\nu:\nu\in \mathscr M_{\rm inv}(M,f),\int\varphi d\nu\geq c\right\}.\end{aligned}$$
2. For $\psi\in \mathcal{V}^-$, we have $$\begin{aligned}
\underline{R}(\varphi,(c,\infty))
\geq\sup\left\{h_\nu(f)-\int\psi d\nu:
\mu\in \mathscr M_{\rm inv}(\widetilde{\Lambda},f),\int\varphi d\nu> c\right\}.\end{aligned}$$
[**Example 2 Nonuniformly hyperbolic systems**]{} In [@Kat1], Katok described a construction of a diffeomorphism on the 2-torus $\mathbb{T}^2$ with nonzero Lyapunov exponents, which is not an Anosov map. Let $f_0$ be a linear automorphism given by the matrix $$\begin{aligned}
A=\begin{pmatrix}
2&1\\
1&1
\end{pmatrix}\end{aligned}$$ with eigenvalues $\lambda^{-1}<1<\lambda$. $f_0$ has a maximal measure $\mu_1$. Let $D_r$ denote the disk of radius $r$ centered at (0,0), where $r>0$ is small, and put coordinates $(s_1,s_2)$ on $D_r$ corresponding to the eigendirections of $A$, i.e, $A(s_1,s_2)=(\lambda s_1,\lambda^{-1}s_2)$. The map $A$ is the time-1 map of the local flow in $D_r$ generated by the following system of differential equations: $$\begin{aligned}
\frac{ds_1}{dt}=s_1\log\lambda, \frac{ds_2}{dt}=-s_2\log\lambda.\end{aligned}$$ The Katok map is obtained from $A$ by slowing down these equations near the origin. It depends upon a real-valued function $\psi$, which is defined on the unit interval $[0,1]$ and has the following properties:
1. $\psi$ is $C^\infty$ except at 0;
2. $\psi(0)=0$ and $\psi(u)=1$ for $u\geq r_0$ where $0<r_0<1$;
3. $\psi^{\prime}(u)>0$ for every $0<u<r_0$;
4. $\int_0^1\frac{du}{\psi(u)}<\infty$.
Fix sufficiently small numbers $r_0<r_1$ and consider the time-1 map $g$ generated by the following system of differential equations in $D_{r_1}$: $$\begin{aligned}
\frac{ds_1}{dt}=s_1\psi(s_1^2+s_2^2)\log\lambda, \frac{ds_2}{dt}=-s_2\psi(s_1^2+s_2^2)\log\lambda.\end{aligned}$$ The map $f$, given as $f(x)=g(x)$ if $x\in D_{r_1}$ and $f(x)=A(x)$ otherwise, defines a homeomorphism of torus, which is a $C^\infty$ diffeomorphism everywhere except for the origin. To provide the differentiability of map $f$, the function $\psi$ must satisfy some extra conditions. Namely, the integral $\int_0^1du/\psi$ must converge “very slowly" near the origin. We refer the smoothness to [@Kat1]. Here $f$ is contained in the $C^0$ closure of Anosov diffeomorphisms and even more there is a homeomorphism $\pi:\mathbb{T}^2\to \mathbb{T}^2$ such that $\pi\circ f_0=f\circ \pi$. Let $\nu_0=\pi_\ast\mu_1$.
In [@LiaLiaSUnTia], the authors proved that there exist $0< \epsilon\ll \beta$ and a neighborhood $U$ of $\nu_0$ in $\mathscr M_{\rm inv}(\mathbb{T}^2,f)$ such that for any ergodic $\nu\in U$ it holds that $\nu\in \mathscr M_{\rm inv}(\widetilde{\Lambda}(\beta,\beta,\epsilon),f)$, where $\widetilde{\Lambda}(\beta,\beta,\epsilon)=\bigcup_{k\geq1}{\rm supp}(\nu_0|\Lambda_k(\beta,\beta,\epsilon))$.
For every $\varphi\in C^0(\mathbb{T}^2)$ and $c\in \mathbb{R}$, the following hold:
1. For $\psi\in \mathcal{V}^+$, we have $$\begin{aligned}
\overline{R}(\varphi,[c,\infty))
\leq\sup\left\{h_\nu(f)-\int\psi d\nu:\nu\in \mathscr M_{\rm inv}(\mathbb{T}^2,f),\int\varphi d\nu\geq c\right\}.\end{aligned}$$
2. For $\psi\in \mathcal{V}^-$, we have $$\begin{aligned}
\underline{R}(\varphi,(c,\infty))
\geq\sup\left\{h_\nu(f)-\int\psi d\nu:
\mu\in \mathscr M_{\rm inv}(\widetilde{\Lambda}(\beta,\beta,\epsilon),f),\int\varphi d\nu> c\right\}.\end{aligned}$$
In [@LiaLiaSUnTia], the authors also studied the structure of Pesin set $\widetilde{\Lambda}$ for the robustly transitive partially hyperbolic diffeomorphisms described by Mañé and the robustly transitive non-partially hyperbolic diffeomorphisms described by Bonatti-Viana. They showed that for the diffeomorphisms derived from Anosov systems $\mathscr M_{\rm inv}(\widetilde{\Lambda},f)$ enjoys many members. So our result is applicable to these maps.
[**Acknowledgements.**]{} The research was supported by the National Basic Research Program of China (Grant No. 2013CB834100) and the National Natural Science Foundation of China (Grant No. 11271191).
[50]{} V. Araujo & M.J. Pacifico, Large deviations for non-uniformly expanding maps, *J. Stat. Phys.* [**125**]{} (2006), 415-457.
L. Barreira & Y. Pesin, Lyapunov exponents and smooth ergodic theory, Univ. Lect. Series, 23, Providence, RI: American Mathematical Society, (2002).
L. Barreira & Y. Pesin, Nonuniform hyperbolicity: Dynamics of systems with nonzero Lyapunov exponents, Cambridge Press, (2007).
J. Bochi, Genericity of zero Lyapunov exponents, *Ergodic Theory Dynam. Systems* [**22**]{} (2002), 1667-1696.
Y.M. Chung, Large deviations on Markov towers, *Nonlinearity* [**24**]{} (2011), 1229-1252.
H. Comman & J. Rivera-Letelier, Large deviations principles for non-uniformly hyperbolic rational maps, *Ann. Inst. H. Poincaré. Anal. Non Linéaire* [**15**]{} (1998), 539-579.
A. Eizenberg, Y. Kifer & B. Weiss, Large deviations for $\mathbb{Z}^d$-actions, *Commun. Math. Phys.* [**164**]{} (1994), 433-454.
R.S. Ellis, Entropy, Large deviations and Statistical Mechanics, Grundlehren der Mathematischen Wissenchaften 271, Spinger-Verlag, New York, (1985).
M. Hirayama, Periodic probability measures are dense in the set of invariant measures, *Discrete Contin. Dyn. Syst.* [**9**]{} (2003), 1185-1192.
A. Katok, Bernoulli Diffeomorphisms on Surfaces, *Annals of Math.* (2) [**110**]{} (1979), 529-547.
A. Katok, Lyapunov exponents, entropy and periodic points for diffeomorphism, *Publ. Math. IHÉS* [**51**]{} (1980), 137-173.
A. Katok & B. Hasselblatt, Introduction to the modern theory of dynamical systems, Encyclopedia of Mathematics and its Applications 54, Cambridge Univ. Press, Cambridge, (1995).
Y. Kifer, Large deviations in dynamical systems and stochastic processes. *Trans. Amer. Math. Soc.* [**321(2)**]{} (1990), 505-524.
C. Liang, G. Liao, W. Sun & X. Tian. Saturated sets for nonuniformly hyperbolic systems, arXiv:1110.6091v1.
I. Melbourne & M. Nicol, Large deviations for nonuniformly hyperbolic systems, *Tran. Amer. Math. Soc.* [**360**]{} (2008), 6661-6676.
V.I. Oseledec, A multiplicative ergodic theorem, *Trans. Mosc. Math. Soc.* [**19**]{} (1969), 197-231.
Y. Pesin, Dimension Theory in Dynamical Systems, Contemporary Views and Applications. Univ. of Chicago Press, (1997).
C.-E. Pfister & W.G. Sullivan, Large deviations estimates for dynamical systems without the specification property. Applications to the $\beta$-shifts, *Nonlinearity* [**18**]{} (2005), 237-261.
M. Pollicott, Lectures on ergodic theory and Pesin theory on compact manifolds, Cambridge Univ. Press, Cambridge, (1993).
M. Pollicott & R. Sharp, Large deviations for intermittent maps, *Nonlinearity* [**22**]{} (2009), 2079-2092.
L. Rey-Bellet & L.-S. Young, Large deviations in non-uniformly hyperbolic dynamical systems, *Ergodic Theory Dynam. Systems* [**28(2)**]{} (2008), 587-612.
P. Varandas, Non-uniform specification and large deviations for weak Gibbs measures, *J. Stat. Phys.* [**146**]{} (2012), 330-358.
K. Yamamoto, On the weaker forms of specification property and their applications, *Proc. Amer. Math. Soc.* [**137**]{} (2009), 3807-3814.
L.-S. Young, Large deviations in dynamical systems, *Trans. Amer. Math. Soc.* [**318**]{} (1990), 525-543.
M, Yuri, Large deviations for countable to one Markov systems, *Commun. Math. Phys.* [**258(2)**]{} (2005), 455-474.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present an accurate analytic fitting formula for the numerical results for the relativistic corrections to the thermal Sunyaev-Zel’dovich effect for clusters of galaxies. The numerical results for the relativistic corrections have been obtained by numerical integration of the collision term of the Boltzmann equation. The fitting is carried out for the ranges $0.02 \leq \theta_{e} \leq 0.05$ and $0 \leq X \leq 20$, where $\theta_{e} \equiv k_{B}T_{e}/m_{e}c^{2}$, $X \equiv \hbar \omega/k_{B}T_{0}$, $T_{e}$ is the electron temperature, $\omega$ is the angular frequency of the photon, and $T_{0}$ is the temperature of the cosmic microwave background radiation. The accuracy of the fitting is generally better than 0.1%. The present analytic fitting formula will be useful for the analyses of the thermal Sunyaev-Zel’dovich effect for high-temperature galaxy clusters.'
author:
- SATOSHI NOZAWA
- NAOKI ITOH AND YOUHEI KAWANA
- YASUHARU KOHYAMA
title: 'RELATIVISTIC CORRECTIONS TO THE SUNYAEV-ZEL’DOVICH EFFECT FOR CLUSTERS OF GALAXIES. IV. ANALYTIC FITTING FORMULA FOR THE NUMERICAL RESULTS'
---
addtoreset[equation]{}[section]{}
AND
INTRODUCTION
============
Compton scattering of the cosmic microwave background (CMB) radiation by hot intracluster gas — the Sunyaev-Zel’dovich effect (Zel’dovich & Sunyaev 1969; Sunyaev & Zel’dovich 1972, 1980a, 1980b, 1981) — provides a useful method to measure the Hubble constant $H_{0}$ (Gunn 1978; Silk & White 1978; Birkinshaw 1979; Cavaliere, Danese, & De Zotti 1979; Birkinshaw, Hughes, & Arnaud 1991; Birkinshaw & Hughes 1994; Myers et al. 1995; Herbig et al. 1995; Jones 1995; Markevitch et al. 1996; Holzapfel et al. 1997; Furuzawa et al. 1998). The original Sunyaev-Zel’dovich formula has been derived from a kinetic equation for the photon distribution function taking into account the Compton scattering by electrons: the Kompaneets equation (Kompaneets 1957; Weymann 1965). The original Kompaneets equation has been derived with a nonrelativistic approximation for the electron. However, recent X-ray observations have revealed the existence of many high-temperature galaxy clusters (David et al. 1993; Arnaud et al. 1994; Markevitch et al. 1994; Markevitch et al. 1996; Holzapfel et al. 1997; Mushotzky & Scharf 1997; Markevitch 1998). In particular, Tucker et al. (1998) reported the discovery of a galaxy cluster with the electron temperature $k_{B} T_{e} = 17.4 \pm 2.5$ keV. Rephaeli and his collaborator (Rephaeli 1995; Rephaeli & Yankovitch 1997) have emphasized the need to take into account the relativistic corrections to the Sunyaev-Zel’dovich effect for clusters of galaxies.
In recent years remarkable progress has been achieved in the theoretical studies of the relativistic corrections to the Sunyaev-Zel’dovich effects for clusters of galaxies. Stebbins (1997) generalized the Kompaneets equation. Itoh, Kohyama, & Nozawa (1998) have adopted a relativistically covariant formalism to describe the Compton scattering process (Berestetskii, Lifshitz, & Pitaevskii 1982; Buchler & Yueh 1976), thereby obtaining higher-order relativistic corrections to the thermal Sunyaev-Zel’dovich effect in the form of the Fokker-Planck expansion. In their derivation, the scheme to conserve the photon number at every stage of the expansion which has been proposed by Challinor & Lasenby (1998) played an essential role. The results of Challinor & Lasenby (1998) are in agreement with those of Itoh, Kohyama, & Nozawa (1998). The latter results include higher-order expansions. Itoh, Kohyama, & Nozawa (1998) have also calculated the collision integral of the Boltzmann equation numerically and have compared the results with those obtained by the Fokker-Planck expansion method. They have confirmed that the Fokker-Planck expansion method gives an excellent result for $k_{B}T_{e} \leq 15$keV, where $T_{e}$ is the electron temperature. For $k_{B}T_{e} \geq 15$keV, however, the Fokker-Planck expansion results show nonnegligible deviations from the results obtained by the numerical integration of the collision term of the Boltzmann equation. Here it should be pointed out that the generalized Kompaneets equation is equivalent to a single-scattering approximation. Thus for high-temperature clusters ($k_{B}T_{e} \geq $15 keV) the relativistic corrections may underestimate the Sunyaev-Zel’dovich effect at high frequencies.
Nozawa, Itoh, & Kohyama (1998b) have extended their method to the case where the galaxy cluster is moving with a peculiar velocity with respect to CMB. They have thereby obtained the relativistic corrections to the kinematical Sunyaev-Zel’dovich effect. Challinor & Lasenby (1999) have confirmed the correctness of the result obtained by Nozawa, Itoh, & Kohyama (1998b). Sazonov & Sunyaev (1998a, b) have calculated the kinematical Sunyaev-Zel’dovich effect by a different method. Their results are in agreement with those of Nozawa, Itoh, & Kohyama (1998b). The latter authors have given the results of the higher-order expansions.
Itoh, Nozawa, & Kohyama (2000) have also applied their method to the calculation of the relativistic corrections to the polarization Sunyaev-Zel’dovich effect (Sunyaev & Zel’dovich 1980b, 1981). They have thereby confirmed the result of Challinor, Ford, & Lasenby (1999) which has been obtained with a completely different method. Recent works on the polarization Sunyaev-Zel’dovich effect include Audit & Simons (1998), Hansen & Lilje (1999), and Sazonov & Sunyaev (1999).
In the present paper we address ourselves to the numerical calculation of the relativistic corrections to the thermal Sunyaev-Zel’dovich effect. As stated above, Itoh, Kohyama, & Nozawa (1998) have carried out the numerical integration of the collision term of the Boltzmann equation. This method produces the exact results without the power series expansion approximation. In view of the recent discovery of an extremely high temperature galaxy cluster with $k_{B}T_{e} = 17.4 \pm 2.5$keV (Tucker et al. 1998), it would be extremely useful to present the results of the numerical integration of the collision term of the Boltzmann equation in the form of an accurate analytic fitting formula.
Sazonov & Sunyaev (1998a, b) have reported the results of the Monte Carlo calculations on the relativistic corrections to the Sunyaev-Zel’dovich effect. In Sazonov & Sunyaev (1998b), a numerical table which summarizes the results of the Monte Carlo calculations has been presented. This table is of great value when one wishes to calculate the relativistic corrections to the Sunyaev-Zel’dovich effect for galaxy clusters of extremely high temperatures. Accurate analytic fitting formulae would be still more convenient to use for the observers who wish to analyze the galaxy clusters with extremely high temperatures. This is the motivation of the present paper. For the analyses of the galaxy clusters with extremely high temperatures, the results of the calculation of the relativistic thermal bremsstrahlung Gaunt factor (Nozawa, Itoh, & Kohyama 1998a) and their accurate analytic fitting formulae (Itoh et al. 2000) will be useful.
The present paper is organized as follows. In $\S$ 2 we give the method of the calculation. In $\S$ 3 we give the analytic fitting formula. Concluding remarks will be given in $\S$ 4.
BOLTZMANN EQUATION
==================
We will formulate the kinetic equation for the photon distribution function using a relativistically covariant formalism (Berestetskii, Lifshitz, & Pitaevskii 1982; Buchler & Yueh 1976). As a reference system, we choose the system which is fixed to the center of mass of the cluster of galaxies. This choice of the reference system affords us to carry out all the calculations in the most straightforward way. We will use the invariant amplitude for the Compton scattering as given by Berestetskii, Lifshitz, & Pitaevskii (1982) and by Buchler & Yueh (1976).
The time evolution of the photon distribution function $n(\omega)$ is written as $$\begin{aligned}
\frac{\partial n(\omega)}{\partial t} & = & -2 \int \frac{d^{3}p}{(2\pi)^{3}} d^{3}p^{\prime} d^{3}k^{\prime} \, W \,
\left\{ n(\omega)[1 + n(\omega^{\prime})] f(E) - n(\omega^{\prime})[1 + n(\omega)] f(E^{\prime}) \right\} \, , \\
W & = & \frac{(e^{2}/4\pi)^{2} \, \overline{X} \, \delta^{4}(p+k-p^{\prime}-k^{\prime})}{2 \omega \omega^{\prime} E E^{\prime}} \, , \\
\overline{X} & = & - \left( \frac{\kappa}{\kappa^{\prime}} + \frac{\kappa^{\prime}}{\kappa} \right) + 4 m^{4} \left( \frac{1}{\kappa} + \frac{1}{\kappa^{\prime}} \right)^{2}
- 4 m^{2} \left( \frac{1}{\kappa} + \frac{1}{\kappa^{\prime}} \right) \, , \\
\kappa & = & - 2 (p \cdot k) \, = \, - 2 \omega E \left( 1 - \frac{\mid \vec{p} \mid}{E} {\rm cos} \alpha \right) \, , \\
\kappa^{\prime} & = & 2 (p \cdot k^{\prime}) \, = \, 2 \omega^{\prime} E \left( 1 - \frac{\mid \vec{p} \mid}{E} {\rm cos} \alpha^{\prime} \right) \, .\end{aligned}$$ In the above $W$ is the transition probability corresponding to the Compton scattering. The four-momenta of the initial electron and photon are $p = (E, \vec{p})$ and $k = (\omega, \vec{k})$, respectively. The four-momenta of the final electron and photon are $p^{\prime} = (E^{\prime}, \vec{p}^{\prime})$ and $k^{\prime} = (\omega^{\prime}, \vec{k}^{\prime})$, respectively. The angles $\alpha$ and $\alpha^{\prime}$ are the angles between $\vec{p}$ and $\vec{k}$, and between $\vec{p}$ and $\vec{k}^{\prime}$, respectively. Throughout this paper, we use the natural unit $\hbar = c = 1$ unit, unless otherwise stated explicitly.
By ignoring the degeneracy effects, we have the relativistic Maxwellian distribution for electrons with temperature $T_{e}$ as follows $$\begin{aligned}
f(E) & = & \left[ e^{\left\{(E - m)-(\mu - m) \right\}/k_{B}T_{e}} \, + \, 1 \right]^{-1} \nonumber \\
& \approx & e^{-\left\{K-(\mu - m)\right\}/k_{B}T_{e}} \, ,\end{aligned}$$ where $K \equiv (E - m)$ is the kinetic energy of the initial electron, and $(\mu - m)$ is the non-relativistic chemical potential of the electron. We now introduce the quantities $$\begin{aligned}
x & \equiv & \frac{\omega}{k_{B}T_{e}} \, , \\
\Delta x & \equiv & \frac{\omega^{\prime} - \omega}{k_{B}T_{e}} \, .\end{aligned}$$ Substituting equations (2.6) – (2.8) into equation (2.1), we obtain $$\frac{\partial n(\omega)}{\partial t} = -2 \int \frac{d^{3}p}{(2\pi)^{3}} d^{3}p^{\prime} d^{3}k^{\prime} \, W \, f(E) \,
\left\{ [1 + n(\omega^{\prime})] n(\omega) - [1 + n(\omega)] n(\omega^{\prime}) e^{ \Delta x } \right\} \, .$$ Equation (2.9) is our basic equation. We will denote the Thomson scattering cross section by $\sigma_{T}$, and the electron number density by $N_{e}$. We will define $$\begin{aligned}
\theta_{e} & \equiv & \frac{k_{B}T_{e}}{m_{e}c^{2}} \, , \\
y & \equiv & \sigma_{T} \int d \ell N_{e} \, ,\end{aligned}$$ where $T_{e}$ is the electron temperature, and the integral in equation (2.11) is over the path length of the galaxy cluster. By introducing the initial photon distribution of the CMB radiation which is assumed to be Planckian with temperature $T_{0}$ $$\begin{aligned}
n_{0} (X) & = & \frac{1}{e^{X} - 1} \, , \\
X & \equiv & \frac{\omega}{k_{B} T_{0}} \, ,\end{aligned}$$ we rewrite equation (2.9) as $$\frac{\Delta n(X)}{n_{0}(X)} \, = \, y \, F(\theta_{e}, X) \, .$$ We obtain the function $F(\theta_{e}, X)$ by numerical integration of the collision term of the Boltzmann equation (2.9). The accuracy of the numerical integration is about $10^{-5}$. We confirm that the condition of the photon number conservation $$\int d X \, X^{2} \, \Delta n(X) \, = \, 0 \,$$ is satisfied with the accuracy better than $10^{-9}$.
We define the distortion of the spectral intensity as $$\begin{aligned}
\Delta I & \equiv & \frac{X^{3}}{e^{X} - 1} \, \frac{\Delta n(X)}{n_{0}(X)} \, \\
& = & y \, \frac{X^{3}}{e^{X} - 1} \, F(\theta_{e}, X) \, .\end{aligned}$$ The graph of $F(\theta_{e},X)$ is shown in Figure 1. The graph of $\Delta I/y$ is shown in Figure 2.
ANALYTIC FITTING FORMULA
========================
We give an accurate analytic fitting formula for the function $F(\theta_{e}, X)$ in equation (2.14) which has been obtained by numerical integration of the collision term of the Boltzmann equation. We will give an analytic fitting formula for the ranges $0.02 \leq \theta_{e} \leq 0.05$, $0 \leq X \leq 20$, which will be sufficient for the analyses of the galaxy clusters. For $\theta_{e} < 0.02$, the results of Itoh, Kohyama, and Nozawa (1998) give sufficiently accurate results (the accuracy is generally better than 1%).
We express the fitting formula for $0.02 \leq \theta_{e} \leq 0.05$ as follows: $$\begin{aligned}
\frac{\Delta n(X)}{n_{0}(X)} & = & y \, F(\theta_{e}, X) \nonumber \\
& = & y \, \left\{ \frac{\theta_{e} X e^{X}}{e^{X}-1} \, \left(
Y_{0} \, + \, \theta_{e} Y_{1} \, + \, \theta_{e}^{2} Y_{2} \, + \, \theta_{e}^{3} Y_{3} \, + \, \theta_{e}^{4} Y_{4} \, \right) \, + R \right\} \, .\end{aligned}$$ The functions $Y_{0}$, $Y_{1}$, $Y_{2}$, $Y_{3}$, and $Y_{4}$ have been obtained by Itoh, Kohyama, and Nozawa (1998) with the Fokker-Planck expansion method, and their explicit expressions have been given.
We define the residual function $R$ in equation (3.1) as follows: $$\begin{aligned}
R & = & \left\{ \begin{array}{ll} 0 \, , & \mbox{for $0 \leq X < 2.5$} \\
\displaystyle{ \sum_{i,j=0}^{10} \, a_{i \, j} \, \Theta_{e}^{i} \, Z^{j} } \, , &
\mbox{for $2.5 \leq X \leq 20.0$ \, , }
\end{array} \right.\end{aligned}$$ where $$\begin{aligned}
\Theta_{e} & \equiv & 25 \left( \theta_{e} \, - \, 0.01 \right) \, \, , \, \, \, \, \, \, 0.02 \leq \theta_{e} \leq 0.05 \, , \\
Z & \equiv & \frac{1}{17.6} \left( X \, - \, 2.4 \right) \, \, , \, \, \, \, \, \, 2.5 \leq X \leq 20.0 \, .\end{aligned}$$ The coefficients $a_{i \, j}$ are presented in TABLE 1. The accuracy of the fitting formula for equation (3.1) is generally better than 0.1% except for a region of $\theta_{e}=0.05$, $X>17$, where the error exceeds 1%.
CONCLUDING REMARKS
==================
We have calculated the relativistic corrections to the thermal Sunyaev-Zel’dovich effect for clusters of galaxies by numerical integration of the collision term of the Boltzmann equation. We have presented an accurate analytic fitting formula for the thermal Sunyaev-Zel’dovich effect. The fitting formula covers all the ranges of the observation of galaxy clusters in the foreseeable future. The accuracy of the fitting is generally better than 0.1%. The present results will be useful for the analyses of the galaxy clusters with extremely high temperatures. For galaxy clusters with relatively low temperatures $\theta_{e} < 0.02$, the Fokker-Planck expansion results of Itoh, Kohyama, & Nozawa (1998) will be sufficiently accurate (the accuracy is generally better than 1%).
We thank Professor Y. Oyanagi for allowing us to use the least square fitting program SALS. We also thank our anonymous referee for many valuable comments which helped us tremendously in revising the manuscript. This work is financially supported in part by the Grant-in-Aid of Japanese Ministry of Education, Science, Sports, and Culture under the contract \#10640289.
Arnaud, K. A., Mushotzky, R. F., Ezawa, H., Fukazawa, Y., Ohashi, T., Bautz, M. W., Crewe, G. B., Gendreau, K. C., Yamashita, K., Kamata, Y., & Akimoto, F. 1994, ApJ, 436, L67 Audit, E., & Simmons, J. F. L. 1999, MNRAS, 305, L27 Berestetskii, V. B., Lifshitz, E. M., & Pitaevskii, L. P. 1982, $Quantum$ $Electrodynamics$ (Oxford: Pergamon) Birkinshaw, M. 1979, MNRAS, 187, 847 Birkinshaw, M., & Hughes, J., P. 1994, ApJ, 420, 33 Birkinshaw, M., Hughes, J. P., & Arnaud, K. A. 1991, ApJ, 379, 466 Buchler, J. R., & Yueh, W. R. 1976, ApJ, 210, 440 Cavaliere, A., Danese, L., & De Zotti, G. 1979, A&A, 75, 322 Challinor, A., Ford, M., & Lasenby, A., 1999, MNRAS in press Challinor, A., & Lasenby, A., 1998, ApJ, 499, 1 Challinor, A., & Lasenby, A., 1999, ApJ, 510, 930 David, L. P., Slyz, A., Jones, C., Forman, W., & Vrtilek, S. D. 1993, ApJ, 412, 479 Furuzawa, A., Tawara, Y., Kunieda, H., Yamashita, K., Sonobe, T., Tanaka, Y., & Mushotzky, R. 1998, ApJ, 504, 35 Gunn, J. E. 1978, in Observational Cosmology, 1, ed. A. Maeder, L. Martinet & G. Tammann (Sauverny: Geneva Obs.) Hansen, E., & Lilje, P. B. 1999, MNRAS, 306, 153 Herbig, T., Lawrence, C. R., Readhead, A. C. S., & Gulkus, S. 1995, ApJ, 449, L5 Holzapfel, W. L. et al. 1997, ApJ, 480, 449 Itoh, N., Kohyama, Y., & Nozawa, S. 1998, ApJ, 502, 7 Itoh, N., Nozawa, S., & Kohyama, Y. 2000, ApJ in press Itoh, N., Sakamoto, T., Kusano, S., Nozawa, S., & Kohyama, Y. 2000, ApJS in press Jones, M. 1995, Astrophys. Lett. Commun., 6, 347 Kompaneets, A. S. 1957, Soviet Physics JETP, 4, 730 Markevitch, M. 1998, ApJ, 504, 27 Markevitch, M., Mushotzky, R., Inoue, H., Yamashita, K., Furuzawa, A., & Tawara, Y. 1996, ApJ, 456, 437 Markevitch, M., Yamashita, K., Furuzawa, A., & Tawara, Y. 1994, ApJ, 436, L71 Myers, S. T., Baker, J. E., Readhead, A. C. S., & Herbig, T. 1995, preprint Mushotzky, R. F., & Scharf, C. A. 1997, ApJ, 482, L13 Nozawa, S., Itoh, N., & Kohyama, Y. 1998a, ApJ, 507, 530 Nozawa, S., Itoh, N., & Kohyama, Y. 1998b, ApJ, 508, 17 Rephaeli, Y. 1995, ApJ, 445, 33 Rephaeli. Y., & Yankovitch, D. 1997, ApJ, 481, L55 Sazonov, S. Y., & Sunyaev, R. A. 1998a, ApJ, 508, 1 Sazonov, S. Y., & Sunyaev, R. A. 1998b, Astronomy Letters 24, 553 Sazonov, S. Y., & Sunyaev, R. A. 1999, MNRAS in press Silk, J. I., & White, S. D. M. 1978, ApJ, 226, L103 Stebbins, A., 1997, preprint astro-ph/9705178 Sunyaev, R. A., & Zel’dovich, Ya. B. 1972, Comments Astrophys. Space Sci., 4, 173 Sunyaev, R. A., & Zel’dovich, Ya. B. 1980a, ARA&A, 18, 537 Sunyaev, R. A., & Zel’dovich, Ya. B. 1980b, MNRAS, 190, 413 Sunyaev, R. A., & Zel’dovich, Ya. B. 1981, Astrophysics and Space Physics Reviews, 1, 1 Tucker, W., Blanco, P., Rappoport, S., David, L., Fabricant, D., Falco, E. E., Forman, W., Dressler, A., & Ramella, M. 1998, ApJ, 496, L5 Weymann, R. 1965, Phys. Fluid, 8, 2112 Zel’dovich, Ya. B., & Sunyaev, R. A. 1969, Astrophys. Space Sci., 4, 301
**Figure Captions**
- Fig.1. The function $F(\theta_{e}, X)$. The dotted, dashed, dash-dotted, and solid curves correspond to the cases for $\theta_{e}=0.02, 0.03, 0.04, 0.05$, respectively.
- Fig.2. The spectral intensity distortion $\Delta I/y$ as a function of $X$. The dotted, dashed, dash-dotted, and solid curves correspond to the cases for $\theta_{e}=0.02, 0.03, 0.04, 0.05$, respectively.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We experimentally characterize the impact response of concentrated suspensions of cornstarch and water. We hypothesize that the dynamically jammed region that propagates ahead of the impactor is responsible for the strong stress response to impact when it spans between solid boundaries. Using surface imaging and particle tracking at the boundary opposite the impactor, we observed that a visible structure and particle flow at the boundary occur with a delay after impact. We show the delay time is about the same time as the the strong stress response, confirming that the strong stress response results from deformation of the dynamically jammed structure once it spans between the impactor and a solid boundary. A characterization of this strong stress response is reported in a companion paper [@MMASB17]. We also elaborate on the structure of the dynamically jammed region once it spans from the impactor to a solid boundary. We observed particle flow in the outer part of the dynamically jammed region at the bottom boundary, with a net transverse displacement of up to about 5% of the impactor displacement, indicating shear at the boundary. Direct imaging of the surface of the outer part of the dynamically jammed region reveals a change in surface structure that appears the same as the result of dilation in other cornstarch suspensions. Imaging also reveals cracks, like a brittle solid. These observations suggest the dynamically jammed structure can temporarily support stress according to an effective modulus, like a soil or dense granular material, along a network of frictional contacts between the impactor and solid boundary.'
author:
- Benjamin Allen
- Benjamin Sokol
- Shomeek Mukhopadhyay
- Rijan Maharjan
- Eric Brown
title: 'System-spanning dynamically jammed region in response to impact of cornstarch and water suspensions'
---
Discontinuous Shear Thickening (DST) suspensions exhibit a remarkable effect in which the suspensions behave like typical liquids at low shear rates, but when sheared faster the resistance to flow can increase discontinuously with shear rate [@Ba89; @BJ14]. This effect has been observed in a large variety of concentrated suspensions of hard, non-attractive particles, and is inferred to be a general feature of such suspensions [@Ba89; @BJ12; @BJ14; @BFOZMBDJ10]. DST suspensions also support large stresses under impact, one example of which is the ability of a person to run on the surface of a pool filled with a suspension of cornstarch and water [@youtube_running; @BJ14]. The impact response of such fluids is of practical interest for impact protection gear because of their strong response during impact while remaining fluid and flexible otherwise [@LWW03; @D3O]. The purpose of this paper is to characterize the internal structure of the suspension that leads to the strong impact response, to aid in developing models. A companion paper focuses on characterizing the stress response [@MMASB17].
Recently it has been found that a ‘dynamically jammed’ region forms ahead of the impactor in the fluid, which moves along with the impactor like a plug [@WJ12]. The dynamically jammed region grows during the impact, with a front propagating away from the impactor [@WJ12; @PJ14; @HPJ16]. There is a sharp velocity gradient at the front, which separates the dynamically jammed region from the surrounding fluid [@PJ14]. In a two-dimensional dry granular experiment the front velocity and width diverge at the same critical packing fraction as the viscosity curve of DST suspensions [@WRVJ13]. While the dynamically jammed region is propagating in the bulk, it is argued to exhibit no significant change in packing fraction [@HPJ16]. Although it is presumed that this dynamically jammed region transmits stress via solid-solid frictional contacts, evidence of this is lacking.
When the dynamically jammed region reaches the boundary, the stress increases beyond the prediction of the added mass effect [@PJ14]. In a companion paper to this one, we report a phenomenological constitutive relation between stress and strain for the impact response [@MMASB17]. We find that when the dynamically jammed region (as defined by its contribution to the added mass effect) propagates to the boundary, the stress increases rapidly beyond the added mass effect, reaching up to the order of MPa [@MMASB17]. It is not yet known what is responsible for the stress scale on the order of MPa, yet it is 3 orders of magnitude larger than steady state shear in rheometer experiments. In rheometer experiments, the scale of the maximum shear stress supported in the shear thickening regime is limited by surface tension at the suspension-air boundary in response to dilation, and transmitted via frictional contacts [@BJ12]. Whether there is any similarity in the force transmission mechanism between steady state shear and impact response remains to be seen.
In this paper, we propose the hypothesis that the dynamically jammed region could support a compressive load that is transmitted via frictional interactions across the system when the dynamically jammed region spans from the impactor to a solid stationary boundary. This assumes that the solid boundaries are much harder and have much more inertial mass than the fluid, so the relatively soft dynamically jammed region will deform as it crashes into the stationary solid boundary. The system-spanning dynamically jammed region could then temporarily support a load as it deforms according to its effective stiffness, perhaps strong enough to support a person running on the surface.
To test this hypothesis and characterize the structure of the dynamically jammed region, we perform impact experiments while imaging and tracking particles at the boundary. In our experiments, the impactor is driven far enough into a suspension to see the dynamically jammed region interacting with the boundary, in contrast to previous experiments which probed mainly the response of the bulk [@WJ12; @WRVJ13; @PJ14; @HPJ16], but not so close to the boundary to be affected by short-range boundary effects (i.e. within $\approx 3$ mm) [@LSZ10]. Our experiments are at $V_I$ faster than quasistatic compression, so that dynamically jammed fronts can exist, but at speeds slow enough that inertial effects [@Bagnold54; @CB13] including added mass [@WJ12] and high Mach number effects [@LPWG10; @POLMFH15; @GKL17]) are negligible. This intermediate velocity regime is where the steady-state DST transition occurs (typically at flow velocities $\stackrel{<}{_\sim} 10^{-2}$ m/s in rheometers [@MB17]), but surprisingly, systematic force measurements have not yet been reported in this regime as far as we know.
The remainder of the paper is organized as follows. The materials and methods of suspension impact experiments are explained in Sections \[sec:materials\] and \[sec:methods\], respectively. In Sec. \[sec:visualization\] we show images at the boundary of the suspension that reveal structural changes in the dynamically jammed region that appear to be a consequence of dilation. In Sec. \[sec:displacement\], we present particle tracking measurements at the boundary to identify where particle flow and shear occurs. In Sec. \[sec:onset\] we compare the timing of the stress increase with that of the onset of motion of tracked particles at the boundary to confirm that the the stress increase is a consequence of the dynamically jammed region spanning between solid boundaries.
Materials {#sec:materials}
=========
The suspensions were made of cornstarch purchased from Carolina Biological Supply, and tap water near room temperature. Weight fractions $\phi$ for cornstarch and water were measured as the weight of the cornstarch divided by the total weight of cornstarch plus water. Weight fractions of cornstarch and water are very sensitive to histories of temperature and humidity [@SF44], so different data sets taken with relative humidity ranging from 8% to 54% are not directly comparable. To avoid misinterpretation from false comparisons, we do not report weight fractions for different experiments. All samples nominally had weight fractions from 0.53 to 0.61, in a range where they all exhibited noticeable shear thickening when stirred by hand. For data sets represented in a single plot, the experiments were taken over a short enough time period to have a humidity standard deviation of 6%. Measurements were made at a temperature of $22.0\pm0.6^{\circ}$C. The density of the suspensions is $1200\pm20$ kg/m$^3$ [@MMASB17].
Samples were initially mixed on a vortex mixer until no dry powder chunks were observed. Before each impact measurement, samples were additionally stirred by slicing through them at least 5 times with a spatula at velocities low enough to avoid significant cracking of the suspension and prevent large air bubbles from being trapped inside the suspension. This additional stirring helps counter any systematic effects of settling or compaction from previous experiments. This procedure produced a level of reproducibility of $\pm 30\%$ in stress measurements, equivalent to what we could achieve by making new samples before each measurement. If instead we did not stir between measurements or we forced air bubbles to get trapped in the suspension, the stress varied by around a factor of 2 from run to run.
Methods {#sec:methods}
=======
We performed experiments to visualize the boundary of the suspension to observe the dynamically jammed region, while simultaneously measuring forces in response to impact. We used a high-speed camera (Phantom M110) to image the suspension boundaries through a transparent acrylic container with a square base, and containing a suspension of cornstarch and water, as shown in Fig. \[fig:apparatus\]. The top, bottom, and side boundaries can all be observed using mirrors. In most experiments (unless otherwise stated), a cylindrical aluminum impactor of diameter $D = 12.7$ mm was pushed into a container with a square base of length $106$ mm, with the suspension filled to a height $H=42$ mm. These dimensions are such that the region of interest below the impactor is far from the sidewalls of the container. The impactor surface unintentionally had a slight wedge shape, which was angled at $4^{\circ}$ relative to the surface. This can be seen to produce some asymmetry in displacements measured in Sec. \[sec:displacement\]. We used an Instron E-1000 dynamic materials tester to push the impactor into the fluid at constant velocity $V_I$, while measuring the normal force on the impactor as a function of depth $z$ from the free surface of the suspension (downward positive). The nominal relative position resolution within each run is 1 $\mu$m. We define $z=0$ and time $t=0$ at the top surface of the suspension, with an uncertainty of 0.5 mm. The impactor started at a height typically $5.0 \pm 0.5$ mm above the suspension surface and was pushed to a final position typically within $10\%$ of the bottom of the container. While the impactor had a set point constant impact velocity $V_I$, it had to accelerate at the beginning and end of the test. This resulted in a standard deviation of the velocity of the impactor of typically 11% while $z>0$. We measured a mean normal stress $\tau$ on the impactor. Calibration of stress measurements and their results are explained in the companion paper [@MMASB17].
Results
=======
Quasi two-dimensional visualization {#sec:visualization}
-----------------------------------
To observe the growth of the dynamically jammed region, we first show a side view of a quasi two-dimensional version of the apparatus in Fig. \[fig:2Dfrontimage\]a. This experiment is designed for qualitative observation only. A solid rectangular impactor of width 57.1 mm and thickness 5 mm was pushed into a suspension at $ V_I=100$ mm/s in an $8$ mm thick cell. We observed a lighter-shaded region, indicative of some kind of change in surface structure, roughly semi-circular in shape, which first appears near the impactor and grows as its front propagates outward faster than the impactor. This front propagation can be observed in Supplementary Video 1 which is played back at 0.1 times real time. This light region appears to be the dynamically jammed region proposed by Waitukaitis & Jaeger [@WJ12], with propagation similar to that found by [@PJ14]. The change in reflected light off the surface indicates a change in surface structure. Cracks can be seen as bright spots due to the image being mostly backlit, which have been observed previously at the surface of DST suspensions [@RMJKS13]. In repeated experiments we observed that the cracks can appear in different places, preferentially starting at the sharp corners of the impactor or at its sides. The cracks never propagate outside of the dynamically jammed region past the propagating front. This indicates that the dynamically jammed region has some characteristics of a brittle solid, as proposed by Waitukaitus & Jaeger [@WJ12] while the outer region remains fluid-like.
The lighter-shaded portion of the dynamic jammed region in Fig. \[fig:2Dfrontimage\] appears matte or rough to the naked eye. The image and video appear very similar to what is seen when a suspension of cornstarch and water dilates under shear (Fig. 11 of [@BJ12]), tension (Fig. 3 of [@SBCB10]), or compression (Fig. 1b of [@RMJKS13]), or when a person steps on the wet sand at the beach that a wave has recently passed by. Dilation is a common result when dense granular flows are sheared, in which the particles push around each other and the packing expands, while the voids between particles enlarge. The rough surface appearance is the result of particles poking through the liquid-air interface of the suspension at the visible surface, while the liquid retreats into the interior to fill the larger voids opened by the dilating particle structure. Cornstarch particles are too small to be seen individually by the naked eye, but scatter reflected light in different directions like a rough surface [@BJ12; @BZFMBDJ11]. The lighter shading in Fig. \[fig:2Dfrontimage\]a is expected as the result of the light reflecting off the surface being mostly indirect, so more light is scattered back to the camera by the rough surface.
To more clearly show the change in surface structure, we took a zoomed-in video of the lighter-shaded portion of the dynamically jammed region in a similar experiment. Snapshots of a 13 mm square region are shown before the impact in Fig. \[fig:2Dfrontimage\]b, and after the impact in Fig. \[fig:2Dfrontimage\]c, so that the latter view is of the dynamically jammed region. The surface of the dynamically jammed region scatters light more diffusively, indicating a rougher surface. Supplementary Video 2 shows a few square centimeter section at the wall in the lighter-shaded region behind the front before, during, and after impact, played back at 0.1 times real time. Initially the quiescent surface has a uniform reflectivity, indicating it is smooth, except for some trapped bubbles air that are on the order of 1 mm in size (some can also be seen in Fig. \[fig:2Dfrontimage\]b), much larger than individual particles. After flow starts (2 seconds into the video), the surface appears rough. During this time, the macroscopic air bubbles disappear, likely as a result of dilation of the particle packing, and liquid can be seen to be drawn away from the wall, resulting in air pockets between the suspension and the wall. The rough appearance of the surface is expected if particles poke through the liquid-air interface as a result of dilation [@BJ12]. Note that the presence of the container wall affects the quantitative values of interfacial tensions, but is not expected to prevent the liquid from drawing away from the surface and being replaced by air (which could be drawn from nearby trapped air bubbles, dissolved gas in the water, or cavitation). After the impactor stops (7 seconds into the video), the dynamically jammed region is seen to retreat and the surface reflectivity becomes more uniform again. Air bubbles are observed to reform out of the rough patches during this relaxation, confirming that there was air in between the suspension and the wall during impact in the rough-looking regions. This surface change which appears to us to be a result of dilation is in contrast to the bulk propagation of the dynamically jammed region, where it is argued to exhibit no decrease in packing fraction [@HPJ16].
Visualization at the boundary of a three-dimensional system
-----------------------------------------------------------
In order to see how this dynamically jammed region behaves in a 3-dimensional system, we imaged a front-lit suspension at the bottom boundary of the square-base container described in Sec. \[sec:methods\] with a high-speed camera at up to 1000 frames per second. An example is shown in Fig. \[fig:surfaceimages\]a at the maximum penetration depth of $z= 39.5$ mm (2.5 mm from the bottom boundary) for $V_I = 396$ mm/s. We observed a localized change in intensity of reflected light on the bottom boundary, in a roughly annular shape centered directly below the impactor, about 3 to 4 times the radius of the impactor. Just outside the impactor radius, this structure appears similar to the observation in Fig. \[fig:2Dfrontimage\], suggesting that this is the dynamically jammed region, and may be the result of dilation. We observed cracks, which appear dark in this case because they do not extend through the entire suspension. They only form if $\tau > 4\cdot 10^6$ Pa, but did not appear systematically. Moreover, the pattern is not always symmetric. Such a stress could correspond to the ultimate strength of a solid-like material. Finally, there is also a central circle about the same size as the impactor with a less noticeable intensity change indicating a different structure in the center.
The time evolution of the bottom boundary image can be seen in Supplementary Video 3, which is played back at 0.05 times real time. The dynamically jammed region appears with a delay after impact, similar to the stress response [@MMASB17]. It appears first directly below the impactor, and grows radially outward. After the impactor stops, the outer edge of the lighter region gradually retreats as the suspension returns to its liquid-like state over about 1 s.
In addition to the observations at the bottom boundary, we also observed a similar change in reflected light intensity on the top surface near the impactor, in agreement with previous observations [@RMJKS13]. At the sides of the container we observed no intensity change in the experiments whose data is presented in this paper and in the companion paper [@MMASB17]. These observations suggest a columnar dynamically jammed structure spanning from the impactor down to the bottom boundary as illustrated in Fig. \[fig:surfaceimages\]b. The column has a nearly circular cross-section at the bottom (Fig. \[fig:surfaceimages\]a), and at the top [@RMJKS13]. We do not have enough information to specify how the diameter of the column varies with depth.
We also performed experiments where the impactor was closer to the sidewall as in Fig. \[fig:2Dfrontimage\]. In this case we observed the same visual change at the sidewall as at the top and bottom boundaries. This is expected as the dynamically jammed region propagates not only below the impactor but also to the sides [@PJ14; @HPJ16]. In these cases we observed that the stress increased sharply with a delay, about the same time as it took to observe the dynamically jammed region at the side boundary, but shorter than it took for the dynamically jammed region to reach the bottom when the impactor was further from the side walls. This indicates that in such geometries the sidewalls may support the load, similar to other dense granular systems [@janssen1895].
Particle tracking {#sec:displacement}
-----------------
To determine how much compression and/or shear is in the dynamically jammed region, we included tracer particles in the suspension during the stress measurements shown in Fig. 2 of [@MMASB17]. In those measurements there was a delay before the stress increased above a weak background, and reached up to the order of MPa. The tracer particles were iron-oxide particles of diameter $a = 0.12 \pm 0.02$ mm. These are heavy and large enough to settle on the bottom boundary and be visible there. On the other hand, they are still small enough to act as tracer particles as it should only take a shear stress of $\sim \rho g a \approx 5$ Pa to overcome friction with the base and move them. This can be achieved even if a tiny fraction of the $\sim$ MPa normal stress on the impactor is transferred into shear stress at the bottom.
We tracked particles using the ImageJ multitracker plugin. To filter out false tracks, usually due to the tracking algorithm picking up different particles at different locations but falsely identifying them as the same particle, we threw out tracked particles that in one frame moved more than a threshold distance, which we varied from 0.25–0.60 mm from run to run. While we set this to ensure eliminating all false tracks, it also resulted in throwing out some real particles corresponding to those that moved the furthest over the course of the experiment, so the following results underestimate the number of particles that moved between 2-6 mm over the course of the experiment.
We calculated the magnitude of the radial component of displacement $d_r$ of each particle as a function of time $t$. From that we calculated the net radial displacement $d_{r,n}$ measured from the beginning of the video (before impact) to the time the impactor reached its maximum depth. An example field map of the net radial displacement $d_{r,n}$ is plotted in Fig. \[fig:displacementfield\] at $V_I=114$ mm/s for the same data set shown in Fig. 2 of [@MMASB17], where the maximum impactor depth was $z=34$ mm. The points are plotted at the starting position of each particle in the x-y plane on the bottom boundary, where the point on the impactor axis is used as the origin of the x-y plane. No significant negative values of $d_r$ were found, indicating the particle flow was moving outward radially from the origin, as might be qualitatively expected in a fluid as material is displaced away from the impactor to create a circulating flow. The field map of $d_{r,n}$ is close to, but not quite, radially symmetric. This asymmetry may be due to the slight wedge shape of the impactor which was angled at $4^{\circ}$ relative to the surface for this set of experiments.
The same data for the net radial displacement $d_{r,n}$ over the course of the impact is plotted as a scatter plot in Fig. \[fig:displacement\_radius\]. A point is shown for each particle tracked as a function of the initial particle radius $r$, where $r$ is measured from the origin on the axis of the impactor. Data for different $V_I$ are shown, and observed to have a similar profile. There is a large scatter in the data, despite the smoothness of the variation between neighboring points in the field map of Fig. \[fig:displacementfield\]. Rather, the scatter in Fig. \[fig:displacement\_radius\] is due to the radial asymmetry of the pattern in Fig. \[fig:displacementfield\]. Note that our filtering method tends to cut out particles with $d_{r,n}$ ranging from 2-6 mm, so this plot may underestimate the number of particles in this range of $d_{r,n}$. In Figs. \[fig:displacementfield\] and \[fig:displacement\_radius\], a dead zone with no significant particle motion is observed directly below the impactor with a radius of 10 to 15 mm, about twice the radius of the impactor. The dead zone is a feature that would not occur in a Newtonian fluid, which would only be expected to have an infinitesimal stagnation point at the center of the image. Rather, such a dead zone is common in granular flows [@ETGZN13]. While the particles may not be moving noticeably in the dead zone, it is still expected that the contact forces fluctuate rapidly and irregularly due to the particle motion at the boundary of the dead zone, transmitting forces in a similar way as the flowing portion of the dynamically jammed region, rather than a static force distribution like statically jammed systems. The sharp increase in $d_{r,n}$ at the edge of the dead zone indicates a strong shear (shear strain is equal to the gradient of displacement) at the interface of the dead zone and the outer portion of the dynamically jammed region, and a well-defined boundary between two flow regimes. This shear profile is in contrast to previous measurements taken at times before the dynamically jammed region spans to the boundaries, where it was found that the low-shear plug-like region is relatively large compared to the region of strong shear, and the shear increases gradually and monotonically moving away from the center of the dynamically jammed region [@WJ12; @PJ14; @HPJ16]. It is unclear yet if this difference in profile is due to the particular geometry and boundary conditions of the different experiments, or is a consequence of the dynamically jammed region colliding with the solid boundary to generate shear-bands.
Figure \[fig:displacement\_time\] shows the time dependence of the radial displacement $\langle d_r\rangle$ for $V_I=114$ mm/s averaged over many particles in different ranges of initial $r$. The bottom surface shown in Fig. \[fig:displacementfield\] is divided into concentric annuli of width $0.7D/2$, and displacements $d_r$ are averaged for all particles that started within each annulus. Time series for each annulus are referred to in the legend by their mean radius $r$. At each radius there is a delay before a significant increase in $\langle d_r \rangle$, similar to the delay in the stress response to impact in the same experiment [@MMASB17]. In Fig. \[fig:displacement\_time\], it can be seen that all of the curves for $r < 8D/2$ increase above the background at the same time, so the delay time is independent of $r$. For larger $r$, this delay time increases with $r$. At the end of each experiment, the impactor decelerated rapidly to a stop. For example in Fig. \[fig:displacement\_time\], this rapid stop is responsible for the kink seen in each curve at $t=0.36$ s. At this time, the load on the impactor reached its peak value, the impactor velocity dropped to 50% of its set point $V_I$, and the impactor was within 8% of its final displacement. After this rapid impactor deceleration, the radial particle displacement $\langle d_r \rangle$ leveled off. This indicates minimal flow after the impactor stopped.
The net transverse displacement $\langle d_{r,n} \rangle$ just outside the dead zone was only about 5% of the displacement of the impactor after the delay time, i.e. after the particles started moving. The strain in the radial direction can be calculated as the gradient in Fig. \[fig:displacementfield\], which yields around 5% compressive strain in the outer part of the dynamically jammed region, but much more significant local strain in shear bands at the boundary of the dead zone and near the wall. Considering that there is compressive strain in both the vertical and radial directions indicates that the dynamically jammed region is not incompressible, and there must be shear strain in the system. A stress transformation suggests the absolute maximum shear strain in a plane extending in the vertical and radial directions is half the larger of the principal strains, corresponding to a shear strain of 44% when the impactor reaches its maximum penetration depth.
Compressive strain and shear can also be observed based on the relative motion seen at the side of the quasi-2-dimensional experiments shown in supplementary video 1, as in [@PJ14]. For example, by tracking features in the central region directly below the impactor, an average compressive strain around of 30% can be seen before the impactor stops. On the sides of the impactor, shear can be observed as the impactor moves down while the cracks remain nearly still. This implies an average shear strain on the sides of about 70%. While we can observe an average strain and shear at the boundaries, we do not have local information about values of shear strain in the interior, which may vary around these averages.
By comparing the displacement measurements from the particle tracking with the boundary visualization in Fig. \[fig:surfaceimages\]a, we can elaborate on the structure of the dynamically jammed region, as illustrated in Fig. \[fig:surfaceimages\]b. The outer part of the dynamically jammed region exhibits particle flow and shear, as indicated by the particle tracking measurements in Fig. \[fig:displacement\_time\]. The shear in the dynamically jammed region can lead to dilation, which can draw liquid from the surface of the dynamically jammed region, which appears to be the case in Figs. \[fig:2Dfrontimage\] and \[fig:surfaceimages\]a. The suggestion of dilation in the dynamically jammed region when it spans between solid boundaries are in contrast to what is claimed for the dynamically jammed region before it spans between solid boundaries [@HPJ16]. In the central portion of the dynamically jammed region, directly in front of the impactor on the opposite boundary and with radius $r \stackrel{<}{_\sim}0.7D$, there is a different type of structure where there is no particle flow, indicated by the particle tracking measurements in Fig. \[fig:displacement\_time\].
Onset time comparison {#sec:onset}
---------------------
In this section we compare the onset time of the particle motion at the bottom boundary (Fig. \[fig:displacement\_time\]) with that of the sharp increase in stress [@MMASB17]. This will test the hypothesis that this strong stress increase is the result of the deformation of the dynamically jammed region once it spans between solid boundaries.
To compare timings between stress and particle displacement, the particle tracking measurements (Sec. \[sec:displacement\]) were taken simultaneously with the stress measurements in Fig. 2 of our companion paper [@MMASB17]. We calibrate the timing between the video and stress measurements by tracking a flag attached to the impactor. We align the time at which the flag has moved 1 pixel in the video with the time that the impactor has moved an equivalent distance.
To determine the onset time for particle tracking, we follow as closely as possible the algorithm for finding the onset of the stress increase above the background in the companion paper [@MMASB17]. We start with the time-dependent mean radial particle displacement $\langle d_r \rangle$ averaged over all particles at all radii from Sec. \[sec:displacement\]. We smooth $\langle d_r \rangle$ uniformly over a range of $\pm0.5$ mm in $z$, just as we did for $\tau$ [@MMASB17] to obtain $\langle d_r \rangle_s$. In Fig. \[fig:onset\] we show representative time series of the simultaneous particle displacement $\langle d_r \rangle_s$ (panel a) and stress $\tau$ (panel b) for $V_I=114$ mm/s. The background from buoyancy and the added mass effect is also shown as the solid line in panel b for the stress data [@MMASB17].
To identify the onset times, we fit each data set in Fig. \[fig:onset\] after the signal first exceeds a threshold value [@MMASB17]. The stress data from the simultaneous experiments is fit as a function of depth $z$ to Eq. 3 of the companion paper [@MMASB17] to include the background from added mass and buoyancy. The particle displacement $\langle d_r \rangle_s$ is fit directly as a function of time, with a threshold value of $5\sigma_d$, where $\sigma_d$ is the standard deviation of $\langle d_r \rangle_s$ from the beginning of recorded data to $t=0$ when the impactor hits the surface. We linearly fit both $\langle d_r \rangle_s$ and $\tau$ starting from their respective threshold values over a fixed range of 3 mm. This fit range is larger than the typically 1 mm range used in the companion paper[@MMASB17], due to the larger noise in the particle tracking data. Each fit is extrapolated to its respective background signal to obtain an onset time $T_d$ for particle displacement or $T_{\tau}$ for stress. For particle displacement, the relevant background is the mean value of $\langle d_r \rangle_s$ for $t<0$. For stress measurements, the fits and extrapolation were done as a function of depth $z$, since that was inferred to be the parameter that the force depends directly on [@MMASB17]. The raw data table with both depth and time data was used as a lookup table to convert the depth at which the extrapolated fit reached the background signal to an onset time $T_{\tau}$. Using the raw data table for this conversion eliminates most of the error from the variation in impact velocity $V_I$ over the course of the experiment.
The solid vertical lines shown in Fig. \[fig:onset\] indicate the best estimate of the onset time $T_d$ or $T_{\tau}$ for each case. The gray bands are the error bars. The uncertainty on $T_{\tau}$ comes from the uncertainty from the added mass effect, the possible range of the contribution of the background after $T_{\tau}$, and the error from the extrapolation of the fit of $d\tau/dz$ to the point of intersection as explained in the companion paper [@MMASB17]. The uncertainty on $T_d$ relative to its zero only includes the errors of the fits of $\langle d_r\rangle_s$ propagated to the point of intersection. We additionally include the relative error between the two timings based on the flag tracking on the error in $T_d$ only. This relative uncertainty on $T_d$ includes the time between frames as a time resolution, plus the time it takes for the impactor to move the flag its first pixel distance to account for the limited spatial resolution of the video, plus an error propagated from the 0.5 mm error in position from smoothing. The 0.5 mm absolute error on the position does not contribute to a relative error on the timing between the particle tracking and stress measurements because it is the same systematic error in both cases, so is not shown in Fig. \[fig:onset\]. The relative timing error is the largest error in Fig. \[fig:onset\], and the total error is 40% of $T_d$ at $V_I=114$ mm/s. Given this error, the delay before the onset of tracked particle motion $T_d$ is still found to be smaller than the delay before the onset of stress $T_{\tau}$ by more than 1 standard deviation.
To obtain more statistics for this comparison, we plot $T_d$ and $T_{\tau}$ at different impact velocities $V_I$ in Fig. \[fig:onset\_vi\]. The errors shown are obtained the same way as those in Fig. \[fig:onset\]. The relative error between the two timings based on the flag tracking is comparable to $T_d$ for larger $V_I$ as the onset times approach the flag timing resolution. For $V_I \ge 396$ mm/s where the absolute timing error is larger than $T_d$ we can only give an upper bound on $T_d$, indicated by the error bars overlapping with $T_d=0$ in Fig. \[fig:onset\_vi\]. The two delay times $T_{\tau}$ and $T_d$ correspond closely to each other over the range measured, with a root-mean-square difference between the onset times of 14% for $V_I \le 114$ mm/s where $T_d$ is resolvable, while the onset times vary over 3 decades. The stress increase usually occurs after particle flow is observed, with the non-overlap of the systematic error bars (i.e. the resolution and the smoothing error) no more than 2 standard deviations of the random error (i.e. the fit errors) when $T_{\tau}$ is smaller than $T_d$, while the non-overlap of the systematic errors is as large as 8 standard deviations of the random error when $T_{\tau}$ is larger than $T_d$. These errors are statistically consistent with the stress increasing at time $T_{\tau}$ equal to or larger than the onset of particle motion at time $T_d$. This close correspondence in delay times, and usually slightly larger $T_{\tau}$ than $T_d$, indicates the large stress increase in the impact response first requires the propagation of a force signal from the impactor to the opposite boundary large enough to initiate particle displacement at the bottom. This close correspondence between stress and particle tracking in Fig. \[fig:onset\_vi\] also confirms that this front is the same as observed in quasi two-dimensional interface tracking [@PJ14].
The small difference in onset times could indicate a return travel time is required for the force signal to travel from the bottom of the container back up to the impactor where the force sensor is located. If this were the case, the relatively small difference between $T_d$ and $T_{\tau}$ would suggest the return travel is faster than that of the downward moving front, which is reasonable since on the return trip the material is already in some way dynamically jammed and would be expected to transmit force more like a solid [@AK06]. However, given the large measurement errors compared to the difference between the two delay times, we do not have the resolution to calculate such a return travel speed with accuracy. At best we can put a lower bound on the return speed that is at least a few times faster than the downward signal propagation.
Conclusions
===========
On impact into a suspension, a dynamically jammed region appears in front of the impactor, and propagates ahead of the impactor (Fig. \[fig:2Dfrontimage\]). This is known to be responsible for the added mass contribution to impact response of suspensions [@WJ12]. We showed that the delay before stress response is consistent with or follows shortly after the onset of particle motion at the bottom boundary (Figs. \[fig:displacement\_time\], \[fig:onset\], and \[fig:onset\_vi\]). This demonstrates that the strong stress response to impact on the order of $10^6$ Pa [@MMASB17] requires deformation of the dynamically jammed region once it spans between solid boundaries.
We observed two different sub-regions of the dynamically jammed region at the bottom boundary. This includes a dead zone about the same size as the impactor cross-section with no particle flow in the central part of the dynamically jammed region (Figs. \[fig:surfaceimages\]a, \[fig:displacementfield\], and \[fig:displacement\_radius\]). In the outer part of the dynamically jammed region at the bottom boundary, we observed particle flow with a net displacement of up to 5% of the impactor displacement (Fig. \[fig:displacement\_radius\]), indicating shear within the dynamically jammed region. We observe a change in surface structure that appears to be the same as the result of dilation in a dense granular suspension (Figs. \[fig:2Dfrontimage\], \[fig:surfaceimages\]a). This dilation is in contrast to the dynamically jammed region as it propagates through the bulk, where it is argued to exhibit no decrease in packing fraction [@HPJ16]. The outer part of the dynamically jammed region also cracks like a brittle solid.
The stress on the order of $10^6$ Pa [@MMASB17] reached in the impact response is so high that it implies that particles have been pushed together beyond the point where lubrication models break down, leading to effectively frictional interactions between neighboring particles where shear stress is proportional to normal stress [@VG88; @MMASB17]. The frictional interactions, fluid-like ability to flow, the appearance of dilation, and cracking, all suggest the dynamically jammed region behaves mechanically much like a soil or dense granular material. The observation of a stress proportional to deformation via an effective modulus is also similar to a soil or dense granular material [@MMASB17].
In soils and other dense granular systems stress is transmitted across the system via frictional interactions, and the scale is determined by the normal stress at the boundary [@LW69], rather than being determined by an intrinsic constitutive rheology in terms of a local shear stress as a function of shear rate and packing fraction. However, for the case of impact, it is not yet clear what is the physical origin for the scale of the normal stress on the order of $10^6$ Pa.
It is interesting to compare and contrast this system with DST in steady-state shear, which is often assumed to be related to the impact response. DST in steady-state shear is triggered by frustration of dilation by a confining stress at a boundary, along with force transmission between solid-solid frictional particle contacts [@BJ12; @SMMD13; @MSMD14; @FMRKMLCHSI13; @Heussinger13; @RBH16; @SGM17]. In steady-state, the structure has time to become well-developed so the stress distribution is more uniform through the suspension, and must be supported at all boundaries, so it is limited by the stiffness of the weakest element in series. In most rheometer experiments, the weakest stiffness comes from surface tension at the suspension-air interface, which limits the maximum normal stress that can be transmitted through the system to about $10^3$ Pa in steady state. In cases where the stress is not limited by the suspension-air interface, the weakest stiffness could be soft walls [@BJ12] or the particle stiffness [@WB09; @BJ12] – the latter case has been observed in steady state flows in simulations with periodic boundary conditions [@OH11; @SMMD13; @SGM17], but not in hard-particle experiments. In contrast, in transient impact, the dynamically jammed region does not have time to propagate to the side wall, and the stress does not have time to become uniform throughout the suspension. Instead, the sides of the columnar dynamically jammed region need support, as illustrated in Fig. \[fig:surfaceimages\]b. The origin of the force that supports the sides of the dynamically jammed region is not yet known.
One possible force that appears in the transient but not the steady state comes from flow through the pores between particles that open up during dilation in the transient. The need to rapidly move liquid as the dynamically jammed region is dilating could introduce significant stresses on the dynamically jammed particle structure. A rough estimate predicts a stress from this pore pressure on the scale of $\tau_p\approx \eta_l \alpha\Delta\phi V_I L/\kappa$, where the viscosity of the interstitial liquid is $\eta_l=9\times10^{-4}$ Pa$\cdot$s, the permeability $\kappa=(1-\phi^3)a^2/180\phi^2$, $\alpha$ is a dimensionless coefficient of order 1, $L$ is the width of the sheared region, and we interpret $\Delta\phi$ as the change in weight fraction due to dilation from the initial value [@JVF16]. If we assume $\alpha=4$ [@JVF16], an estimate for a typical value of $\Delta\phi\approx 0.01$ in a dilating suspension, and $L\approx 1.5$ cm based on the size of the outer sheared region in Fig. \[fig:surfaceimages\]a at $V_I=396$ mm/s, then we obtain $\tau_p \approx 8$ MPa, on the same order of magnitude as the maximum stress observed in our companion paper [@MMASB17]. It remains to be confirmed if this pore pressure is what sets the scale of the stress response to impact.
Acknowledgements
================
We thank Abe Clark, Bob Behringer, Scott Waitukaitis, Ivo Peters, and Heinrich Jaeger for discussions and for sharing their unpublished results. This work was supported by the NSF through grant DMR 1410157.
Supplementary Videos
====================
Supplementary videos may be downloaded from: https://www.eng.yale.edu/brown/publications.html
[37]{}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 [****, ()]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The first dedicated experiment for the rare kaon decay ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ has been performed by the E391a collaboration at the KEK 12-GeV proton synchrotron. A new upper limit of ${2.1 \times 10^{-7}}$ at the 90 % confidence level was set for the branching ratio of the decay ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ using about 10 % of the data collected during the first period of data taking.'
author:
- 'J.K. Ahn$^1$, Y. Akune$^2$, V. Baranov$^3$, M. Doroshenko$^{4, a}$, Y. Fujioka$^2$, Y.B. Hsiung$^5$, T. Inagaki$^6$, S. Ishibashi$^2$, N. Ishihara$^6$, H. Ishii$^7$, T. Iwata$^8$, S. Kobayashi$^2$, S. Komatsu$^7$, T.K. Komatsubara$^6$, A.S. Kurilin$^3$, E. Kuzmin$^3$, A. Lednev$^{9, b}$, H.S. Lee$^1$, S.Y. Lee$^1$, G.Y. Lim$^6$, T. Matsumura$^{10}$, A. Moisseenko$^3$, H. Morii$^{12}$, T. Morimoto$^6$, T. Nakano$^{11}$, N. Nishi$^7$, J. Nix$^{9}$, M. Nomachi$^{7}$, T. Nomura$^{12}$, H. Okuno$^6$, K. Omata$^6$, G.N. Perdue$^{9}$, S. Perov$^3$, S. Podolsky$^{3, c}$, S. Porokhovoy$^3$, K. Sakashita$^7$, N. Sasao$^{12}$, H. Sato$^8$, T. Sato$^6$, M. Sekimoto$^6$, T. Shinkawa$^{10}$, Y. Sugaya$^7$, A. Sugiyama$^2$, T. Sumida$^{12}$, Y. Tajima$^8$, Z. Tsamalaidze$^3$, T. Tsukamoto$^{2, *}$, Y. Wah$^9$, H. Watanabe$^{9, a}$, M. Yamaga$^{6, d}$, T. Yamanaka$^7$, H.Y. Yoshida$^8$, and Y. Yoshimura$^6$\'
title: 'New limit on the ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ decay rate\'
---
The rare decay ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ is a Flavor Changing Neutral Current (FCNC) process from strange to down quarks and is caused by direct CP violation [@Laur; @buras]. The theoretical uncertainty in the branching ratio is only 1-2%, while the branching ratio is predicted to be $(2.8 \pm 0.4) \times 10^{-11}$ [@BGHN06] based on the currently known parameters from other experiments. The decay is considered an ideal process to study the quark flavor physics and a critical test of the standard model as well as a search for new physics beyond it [@isidori]. The present experimental limit is 5.9$\times$10$^{-7}$ at the 90 % confidence level [@ktev]; the Dalitz decay mode ${\pi^0}\to e^+e^-\gamma$ for the final state of ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ was used in the search.
The E391a experiment at the KEK 12-GeV proton synchrotron was proposed to be the first dedicated experiment for the ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ decay and aimed to improve the experimental sensitivity by orders of magnitude, and to verify the experimental method for the next higher sensitivity experiment [@proposal]. The E391a experiment had three data-taking runs in 2004 and 2005. The first data taking (Run-1) was performed from February to June 2004 [@thesis]. In this Letter, we report results obtained from an analysis of about 10 $\%$ of the data collected in Run-1.
The signature of the ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ decay is 2$\gamma$ + [*nothing*]{} in the final state. The energies and hit positions of two photons were measured with an electromagnetic calorimeter. The ${\pi^0}$ vertex, ${Z_{\textrm{vtx}}}$, and its transverse momentum with respect to the beam axis, ${P_T}$, were measured assuming that two photons were produced in a ${\pi^0}\to \gamma\gamma$ decay on the beam axis. One of the crucial tools for this detection method is a neutral beam with small diameter (called “pencil beam”) in order to minimize uncertainty in the ${Z_{\textrm{vtx}}}$ and ${P_T}$ measurements. Good beam collimation with little halo is also important to minimize ${\pi^0}$ production via interactions with detector material. The second tool is a hermetic veto system covering the decay region to reject background decay modes with additional charged particles and photons. High efficiency can be obtained by lowering the energy thresholds [@photonveto; @chargedveto]. The third tool is a decay region with a high vacuum to reduce ${\pi^0}$ produced by neutrons in the beam interacting with the residual gas.
{width="70.00000%"}
Neutral kaons were produced by striking protons from the 12 GeV proton synchrotron on a 60 mm long platinum target. The neutral beam was collimated to a circular shape having 2 mrad of a half cone angle at an angle of 4$^{\circ}$ with respect to the primary proton line [@beamline].
Figure \[fig:det\_all\] shows the E391a detector. The “0 m” in the figure corresponds to 11 m downstream from the target. Detectors were cylindrically assembled along the beam axis, and most of them were installed in a large vacuum vessel to eliminate dead material in front of the detectors. The fiducial region for ${K_L^0}$ decays was from 300 cm to 500 cm.
The electromagnetic calorimeter was placed at the downstream end of the decay region, and was 1.9 m in diameter with a 12 $\times$ 12 cm$^2$ beam hole at the center. It consisted of 576 blocks of undoped CsI crystal, with a size of 7 $\times$ 7 $\times$ 30 cm$^3$ (=16$X_0$) except for the inner 24 crystals that were 5 $\times$ 5 $\times$ 50 cm$^3$ (=27$X_0$) [@csi]. The energy resolution was $\sigma_E/E \simeq 1\% / \sqrt{E} \oplus 1\%$, where $E$ is in GeV, as measured with 25 CsI blocks and a positron beam. The average position resolution was 5 mm for photons. A group of plastic scintillation counters (CV) were placed in front of the CsI calorimeter to veto decays involving charged particles. The decay region was surrounded by two large lead-scintillator sandwich counters, MB and FB. Total thickness of MB (FB) was 13.5 $X_0$ (17.2 $X_0$). Their signal was read through wave-length-shifting fibers by phototubes (PMTs) with high quantum efficiency [@pmt]. The light yield of MB and FB was monitored with an LED calibration system. The gain shift between the on-beam period and the off-beam period was less than 1 % for both MB and FB. Multiple collar-shaped veto counters, CC02 - CC07, were placed along the beam axis to detect photons going through the beam hole. CC02 was a shashlik type lead-scintillator sandwich counter and CC03 was a tungsten-scintillator counter with the layers set parallel to the beam. CC04 and CC05 were lead-scintillator counters with the layers set perpendicular to the beam. CC06 and CC07 consisted of SF-5N lead-glass blocks. Back Anti (BA) was located at the end of the beam in order to detect photons going through the beam hole without being detected by other detectors. BA consisted of lead-scintillator layers and quartz layers, and had 14 $X_0$ in total. In front of BA, a set of 1 mm thick plastic-scintillators, BHCV, was placed to detect charged particles going through the beam hole.
The vacuum tank was divided internally into two regions by a 20 mg/cm$^2$ thick sheet called “membrane” to protect the high vacuum region from out-gassing from detector components. The pressure of the region with detector components was less than 1 Pa, and the pressure in the decay region was $1 \times 10^{-5}$ Pa.
For the CsI calorimeter and all the photon veto detectors, we continually calibrated the energy scale factor using cosmic ray muons and minimum ionization particles in the beam during the operation. We also studied the energy scale factor of the CsI calorimeter using the special data, in which an aluminum(Al) target with a thickness of 5 mm was inserted in the beam at $z = 280.5$ cm (downstream of CC02) in order to produce $\pi^0$ by neutrons in the beam. We adjusted the gain of each CsI crystal so that the invariant mass of two photons agree with the $\pi^0$ mass. This data was also used to check our event reconstruction performance. The more detailed description of the calibration can be found in [@thesis; @csi].
The trigger was designed to accept ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$, ${K_L^0 \rightarrow \pi^0 \pi^0}$ and ${K_L^0 \rightarrow \pi^0 \pi^0 \pi^0}$ decays. The latter two decays were used to study the detector response, and the ${K_L^0 \rightarrow \pi^0 \pi^0}$ decays were used to measure the number of ${K_L^0}$ decays. For triggering purpose, we grouped eight neighboring CsI crystals and defined 72 regions in total. The trigger required that there were two or more such regions with $\ge$ 60 MeV energy deposit in each. The trigger also required no energy deposit in CV and several photon veto detectors. For example, the veto threshold for the total energy deposit in MB was 15 MeV. The trigger rate was 800 events per 2 second beam delivery with a typical intensity of $2.5 \times 10^{12}$ protons on the target. The live time ratio was 78 % with a network distributed data acquisition system with multiple CPUs. The electronics and data acquisition system is briefly described elsewhere [@thesis; @csi].
In the offline analysis, we first looked for photons in the CsI calorimeter. Each cluster of energy deposits was required to have the transverse shower shape consistent with a single electromagnetic shower. The effective energy threshold of each cluster was 10 MeV. We assumed that the two photons came from a ${\pi^0}$ decay, and reconstructed ${Z_{\textrm{vtx}}}$ requiring the two photon invariant mass to have the ${\pi^0}$ mass.
We selected events with exactly two photons hitting the CsI calorimeter and applied selection criteria (cuts) to suppress background events. In Run-1, the downstream membrane was partially hanging in the beam by error, at $z=550$ cm. This produced a large number of background events because neutrons in the beam core struck the membrane and produced secondary ${\pi^0}$’s. If multiple ${\pi^0}$’s were produced at the membrane, and two photons from different ${\pi^0}$’s were detected (“core neutron multi-${\pi^0}$” event), it became a serious background event because we were not able to reconstruct ${Z_{\textrm{vtx}}}$ correctly and these events were distributed in the fiducial region. On the other hand, these events had extra photons in the final state, and thus can be suppressed by detecting those extra photons.
In order to suppress events involving extra photons, we required energy deposit in each photon veto detector to be less than the threshold listed in Table \[tbl:threshold\]. The rejection power of the photon veto was evaluated with four-photon event samples from ${K_L^0 \rightarrow \pi^0 \pi^0}\to 4\gamma$ and ${K_L^0 \rightarrow \pi^0 \pi^0 \pi^0}\to 6\gamma$ with two missing photons. Figure \[fig:g4mass\_plot\] shows the invariant mass of four photons, $M_{4\gamma}$, after applying all the cuts on the photon veto detectors. With all the photon veto cuts, the ratio of the number of ${K_L^0 \rightarrow \pi^0 \pi^0}$ events in $0.45 \le M_{4\gamma} \textrm{(GeV/c$^2$)} \le 0.55$ to the number of ${K_L^0 \rightarrow \pi^0 \pi^0 \pi^0}$ events in $M_{4\gamma} \textrm{(GeV/c$^2$)} \le 0.45$ improved by a factor of 11. This improvement was consistent with the expectation of GEANT-3 based [@ref:geant3] Monte Carlo simulation (MC) within 18 %.
Detector threshold ${\varepsilon_{A}}$ Detector threshold ${\varepsilon_{A}}$
---------- ----------- --------------------- ---------- ----------- ---------------------
CC02 4 MeV 1.0 CC06 5 MeV 0.98
CC03 1.5 MeV 0.98 CC07 50 MeV 0.99
CC04 3 MeV 0.98 FB 2 MeV 0.91
: List of the thresholds applied to the photon veto detectors. $E_Q$ is the total light yield in the BA quartz layers, and $E_S$ is the total energy deposit in the BA scintillator layers. The signal efficiency for a cut $A$, ${\varepsilon_{A}}$, is the ratio of the number of events with all cuts to the number of events with all cuts except for the cut $A$. We estimated ${\varepsilon_{A}}$ using MC ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ events except for the cut on BA. For BA, we first evaluated ${\varepsilon_{A}}$ for ${K_L^0 \rightarrow \pi^0 \pi^0}$ and ${K_L^0 \rightarrow \pi^0 \pi^0 \pi^0}$ decays using real data, which were $0.638 \pm 0.022{_{\textrm{(stat.)}}}$ and $0.658 \pm 0.022{_{\textrm{(stat.)}}}$, respectively, and then assigned the average as ${\varepsilon_{A}}$ for ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ signal. []{data-label="tbl:threshold"}
\
Detector threshold ${\varepsilon_{A}}$
---------- ------------------------------------- ---------------------
CsI 3 MeV for the CsI crystals which do
not belong to the photon clusters 0.78
MB 1 MeV for the inner modules, and
0.5 MeV for the outer modules 0.60
BA 0.5 MIP for $E_{Q}$, and
$E_{Q}/E_{S} \ge 10$ 0.65
: List of the thresholds applied to the photon veto detectors. $E_Q$ is the total light yield in the BA quartz layers, and $E_S$ is the total energy deposit in the BA scintillator layers. The signal efficiency for a cut $A$, ${\varepsilon_{A}}$, is the ratio of the number of events with all cuts to the number of events with all cuts except for the cut $A$. We estimated ${\varepsilon_{A}}$ using MC ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ events except for the cut on BA. For BA, we first evaluated ${\varepsilon_{A}}$ for ${K_L^0 \rightarrow \pi^0 \pi^0}$ and ${K_L^0 \rightarrow \pi^0 \pi^0 \pi^0}$ decays using real data, which were $0.638 \pm 0.022{_{\textrm{(stat.)}}}$ and $0.658 \pm 0.022{_{\textrm{(stat.)}}}$, respectively, and then assigned the average as ${\varepsilon_{A}}$ for ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ signal. []{data-label="tbl:threshold"}
![ Distribution of the invariant mass of four photons, $M_{4\gamma}$, with the cuts on the photon veto detectors and the shower shape of photons in the CsI calorimeter. The dots show data, the open solid histogram shows total MC, the closed solid histogram shows ${K_L^0 \rightarrow \pi^0 \pi^0}$ MC and the hatched histogram shows ${K_L^0 \rightarrow \pi^0 \pi^0 \pi^0}$ MC. []{data-label="fig:g4mass_plot"}](g4-overall-mass-prl.eps){width="50.00000%"}
From the MC study, we found that the “core neutron multi-${\pi^0}$” events had low ${P_T}$ and were populated at the downstream ${Z_{\textrm{vtx}}}$ region. In order to minimize the number of such background events, we used this characteristic and required a parameter, $\alpha \; \equiv \; P_T \textrm{(GeV/c)}- 8.0 \times 10^{-4} \textrm{(GeV/c$\cdot$cm)}
\times {Z_{\textrm{vtx}}}\textrm{(cm)}\quad ({Z_{\textrm{vtx}}}< 525 \textrm{cm})$, to be larger than $-0.225$ GeV/c. Another cut, ${P_T}\ge 0.12$ GeV/c, was applied to suppress ${K_L^0 \rightarrow \gamma \gamma}$ background and $\Lambda \rightarrow \pi^0 \;n$ background, whose maximum $P_T$ is 0.109 GeV/c. The upper boundary on ${P_T}$ was determined to be ${P_T}\le 0.24$ GeV/c from the kinematical limit of the ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ decay ($P_{max} = 0.231$ GeV/c), allowing for the smearing effect due to detector resolutions.
After applying all the selection cuts, we estimated the number of remaining background events in the eight ${P_T}$–${Z_{\textrm{vtx}}}$ regions with the signal regions (c) and (d) as shown in Fig. \[fig:final\_plot\]. Except for the regions (a), (c) and (g), the dominant background source was the “core neutron multi-${\pi^0}$” event. We evaluated the number of “core neutron multi-${\pi^0}$” events using a relational expression with two independent selection cuts: $N_{bkg} = N' \times \textrm{(cut-1 rejection)} \times \textrm{(cut-2 rejection)}$, where “cut-1” is a set of cuts on CV, MB, CC03, CC04, CC06 and CC07, “cut-2” is a set of cuts on the cluster energy and the cluster hit position, and $N'$ is the number of events with all the selection cuts except for the cut-1 and the cut-2. We checked that (i) $(97 \pm 3)$ % of the $N'$ was the “core neutron multi-${\pi^0}$” events even without the cut-1 and cut-2, and (ii) the cut-1 and the cut-2 were independent of each other[^1]. The number of “core neutron multi-${\pi^0}$” background events was $0.0^{+0.7}_{-0.0}$ in region (c) and $1.5 \pm 0.7$ in region (d). The number of background events caused by the halo neutrons interacting with the detector material (CC02) and producing one or more ${\pi^0}$ was $0.9 \pm 0.2 $ in region (a) and $0.04 \pm 0.04$ in region (c). The background events caused by the core neutrons interacting with the membrane and producing $\eta$’s (“core neutron $\eta$” events) was reconstructed around region (c) because the ${\pi^0}$ mass was assumed. For all the events, we re-calculated the decay vertex assuming $\eta$ mass ($Z_{\eta}$) and then rejected events around the membrane in the beam, $525 \le Z_{\eta} \textrm{(cm)} \le 575$. The remaining number of “core neutron $\eta$” events was $0.4 \pm 0.2$, which was the largest component in region (c). The background events from ${K_L^0 \rightarrow \pi^0 \pi^0}$ with two missing photons was evaluated with MC. The number of ${K_L^0 \rightarrow \pi^0 \pi^0}$ background events in the signal region was $0.04 \pm 0.03$, where the error includes the MC statistics and the systematic uncertainties, of which the dominant source was the mismatch between data and MC in the transverse shower shape of photon in the CsI calorimeter. Moreover, the ${K_L^0 \rightarrow \pi^0 \pi^0}$ background events were the largest component in region (g). The total number of background events in the signal region was estimated to be $0.4^{+0.7}_{-0.2}$ in region (c) and $1.5 \pm 0.7$ in region (d).
![ ${Z_{\textrm{vtx}}}$ versus ${P_T}$ with all the event selection cuts. The number of observed (total expected background) events are shown. The expected number of background events was consistent with the observed number of events for all the regions. []{data-label="fig:final_plot"}](z-pt-prl.eps){width="50.00000%"}
We estimated the acceptance of ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ decay to be $(0.657 \pm 0.016) \times 10^{-2}$ based on cut efficiencies evaluated with the real data and MC study. The main components of the acceptance loss were the cuts on MB and BA photon veto detectors. In order to estimate the number of ${K_L^0}$ decays in this search, we analyzed ${K_L^0 \rightarrow \pi^0 \pi^0}$ decays. The invariant mass and the reconstructed decay vertex for ${K_L^0 \rightarrow \pi^0 \pi^0}$ are shown in Fig. \[fig:norm-plot\]. In the ${K_L^0 \rightarrow \pi^0 \pi^0}$ signal region: $0.47 \le M_{4\gamma} \textrm{(GeV/c$^2$)}\le 0.53$, and $300 \le {Z_{\textrm{vtx}}}\textrm{(cm)} \le 500$, there were 2081 ${K_L^0 \rightarrow \pi^0 \pi^0}$ events after subtracting 30 ${K_L^0 \rightarrow \pi^0 \pi^0 \pi^0}$ background events. Based on the MC study, we estimated that the acceptance of ${K_L^0 \rightarrow \pi^0 \pi^0}$ decay was $1.41 \times 10^{-3}$.
![Distribution of the invariant mass (left) and the decay vertex (right) for the ${K_L^0 \rightarrow \pi^0 \pi^0}$ decays. In the top plot, the dots show the data and the histogram shows the MC. The bottom plot shows the ratio of the data to the MC. []{data-label="fig:norm-plot"}](g4-mass-zvtx-prl.eps){width="50.00000%"}
The different final states between the signal and normalization modes caused systematic uncertainties in the single event sensitivity. We assigned the total systematic uncertainty in the single event sensitivity to be $\pm$ 7.0 %. The large sources of systematic uncertainty came from the mismatch between data and MC in the transverse shower shape of the photon (4 %) and the energy distribution in MB (4.2 %).
With the ${K_L^0 \rightarrow \pi^0 \pi^0}$ branching ratio, $(8.83 \pm 0.08) \times 10^{-4}$ [@PDG], we estimated the number of ${K_L^0}$ decays to be ${(1.67 \pm 0.04{_{\textrm{(stat.)}}}) \times 10^9}$. The single event sensitivity was ${(9.11 \pm 0.20{_{\textrm{(stat.)}}}\pm 0.64{_{\textrm{(syst.)}}}) \times 10^{-8}}$. Since we observed no events in the signal region, we set a new upper limit on the branching ratio of ${K_L^0 \rightarrow \pi^0 \nu \bar{\nu}}$ to be $< {2.1 \times 10^{-7}}$ at the 90 % confidence level based on the Poisson statistics. This represents an improvement of a factor of 2.8 over the current limit [@ktev].
We are grateful to the operating crew of the KEK 12-GeV proton synchrotron for their successful beam operation during the experiment. We express our sincere thanks to Professors H. Sugawara, Y. Totsuka, M. Kobayashi and K. Nakamura for their continuous encouragement and support. Thanks are due to A.J. Buras and G. Isidori for useful theoretical discussions and encouragement. This work has been partly supported by a Grant-in-Aid from the MEXT and JSPS in Japan, a grant from NSC in Taiwan and the U.S. Department of Energy.
$^*$Deceased\
$^a$Present address: KEK, Tsukuba, Ibaraki, 305-0801 Japan.\
$^b$Also Institute for High Energy Physics, Protvino, Moscow region, 142281 Russia.\
$^c$Also Scarina Gomel’ State University, Gomel’, BY-246699, Belarus.\
$^d$Present address: Osaka University, Toyonaka, Osaka, 560-0043 Japan.
[99]{} L. S. Littenberg, [*Phys. Rev.*]{} [**D39**]{}, 3322 (1989). A. J. Buras, F. Schwab, and S .Uhling, hep-ph/0405132, and references therein.
D. Bryman [*et al.*]{}, Int. J. Mod. Phys. [**21**]{}, 487 (2006).
A. Alavi-Harati [*et al.*]{}, [*Phys. Rev.*]{} [**D61**]{}, 072006 (2000). T. Inagaki [*et al.*]{}, KEK Internal Report No. 96-13 (1996).
M. Doroshenko, Ph.D. thesis, The Graduate University for Advanced Science, 2005; K. Sakashita, Ph.D. thesis, Osaka University, 2006. S. Ajimura [*et al.*]{}, [*Nucl. Instr. Meth.*]{} [**A 435**]{}, 408 (1999); S. Ajimura [*et al.*]{}, [*Nucl. Instr. Meth.*]{} [**A 552**]{}, 263 (2005). T. Inagaki [*et al.*]{}, [*Nucl. Instr. Meth.*]{} [**A 359**]{}, 478 (1995). H. Watanabe [*et al.*]{}, [*Nucl. Instr. Meth.*]{} [**A545**]{}, 542 (2005). M. Doroshenko [*et al.*]{}, [*Nucl. Instr. Meth.*]{} [**A545**]{}, 278 (2005). M. Itaya [*et al.*]{}, [*Nucl. Instr. Meth.*]{} [**A 522**]{}, 477 (2004).
R. Brun [*et al.*]{}, GEANT 3.21, CERN, Geneva, 1994.
S. Eidelman [*et al.*]{} \[Particle Data Group\], Phys. Lett. B [**592**]{}, 1 (2004); `http://pdg.lbl.gov/` .
[^1]: For each selection cut in the cut-2, we examined the ratio of the number of events passing the cut to the number of events failing the cut. The ratio for the cluster energy cut was $0.79 \pm 0.12$ with the cut-1 and $0.73 \pm 0.03$ without the cut-1. The ratio for the cluster hit position cut was $(5.1 \pm 2.6) \times 10^{-2}$ with the cut-1 and $(5.2 \pm 0.6) \times 10^{-2}$ without the cut-1.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Particle motion at the micro-scale is an incessant tug-of-war between thermal fluctuations and applied forces on one side, and the strong resistance exerted by fluid viscosity on the other. Friction is so strong that completely neglecting inertia – the overdamped approximation – gives an excellent effective description of the actual particle mechanics. In sharp contrast with this result, here we show that the overdamped approximation dramatically fails when thermodynamic quantities such as the entropy production in the environment are considered, in presence of temperature gradients. In the limit of vanishingly small, yet finite inertia, we find that the entropy production is dominated by a contribution that is anomalous, i.e. has no counterpart in the overdamped approximation. This phenomenon, that we call entropic anomaly, is due to a symmetry-breaking that occurs when moving to the small, finite inertia limit. [ Anomalous entropy production is traced back to futile phase-space cyclic trajectories displaying a fast downgradient sweep followed by a slow upgradient return to the original position. ]{}'
author:
- Antonio Celani
- Stefano Bo
- Ralf Eichhorn
- Erik Aurell
title: 'Anomalous thermodynamics at the micro-scale'
---
Life at the micro-scale flows under one law: the fluid gives and the fluid takes away. This arbitrary tyrant lavishly bestows energy to suspended particles through molecular collisions while incessantly draining it from them through friction. The result is the erratic movment of microscopic particles that goes under the name of Brownian motion.
The theory of Brownian motion was developed by Einstein, Smoluchowski and Langevin a little over a century ago [@Duplantier]. A central result of this theory is the [*overdamped*]{} approximation, which says that inertia can be ignored if mass is small, or friction is large. The motion of a Brownian particle, which obeys Newtonian mechanics and is driven by collisions and external forces, is thus reduced to a first-order diffusion equation. The overdamped approximation successfully describes the mechanics on the microscale and is very widely used [@P77; @Dusenbery].
With the advent of micromanipulation it has become possible to measure and control the positions of individual Brownian particles and other small systems [@Ritort06; @LKM10; @BB11; @BAP12]. Thermodynamic concepts such as heat, work and entropy production have hence taken a meaning for single systems [@S05; @Sekimoto]. These developments are the foundation of the new emerging field of stochastic thermodynamics, where the most striking results obtained to date are fluctuation relations, recently reviewed in, e.g., Refs. [@Jarzynski08; @Esposito09; @Jarzynski11]. Here we show that in this setting the overdamped approximation fails, as soon as the temperature field varies in space.
Indeed, although the overdamped approximation correctly yields the trajectories of the Brownian particles in space, it incorrectly estimates the entropy production. This failure is traced back to seemingly innocent corrections that, while having a negligible impact on trajectories, eventually dominate the entropy production in the long run, even in the limit when inertial effects go to zero. As a result, fluctuation relations themselves take a very nontrivial form in the limit of small yet finite inertia.
We dub this phenomenon [*entropic anomaly*]{}, in analogy with similar anomalies encountered in physics. The best known example in classical physics is the viscous dissipative anomaly. The energy dissipation in a fluid flow remains finite even in the limit of arbitrarily small viscosity, whereas it vanishes when viscosity is exactly zero [@Frisch]. The viscous anomaly reflects the loss of time-reversal symmetry of fluid dynamics when going from the inviscid case to the viscous one. Quantum anomalies arise when a current conserved at the classical level (Planck’s constant set to zero) is not conserved anymore at the quantum level (see, e.g., Ref. [@FS04] for a general discussion and Ref. [@CEF01] for a simple example from molecular physics). We show here that the entropic anomaly is associated to the breaking of a symmetry of the zero-inertia, overdamped dynamics, when small finite inertia is considered. The symmetry is the joint reversal of time and particle velocity.
To illustrate our result, let us focus on a minimal thought experiment. A microscopic particle is suspended in a fluid inside a vessel with reflecting walls. The two opposite sides of the vessel are in contact with heat reservoirs at different temperatures. The fluid is motionless and displays by a static, smooth temperature profile $T({\bm x})$ (e.g. a linear one). For the sake of simplicity, no external force is applied to the particle and the friction coefficient is taken independent of particle position. We emphasize that the assumptions on temperature profiles, friction coefficients and absence of external forces are not restrictive, and refer the discussion below and the Supplementary Material for a general discussion.
The motion of the suspended particle is then governed by the Langevin-Kramers equations for position ${\bm X}_t$ and velocity ${\bm V}_t$ [@Dimensions] $$\label{eq:LK}
\begin{array}{lll}
\dot{\bm X}_t &=& {\bm V}_t \\
\dot{\bm V}_t &=& -\gamma {\bm V}_t + \sqrt{2 T({\bm X}_t) \gamma}\, {\bm \eta}_t
\end{array}$$ where ${\bm \eta}_t$ is a Gaussian, zero-mean white noise, i.e. $\langle \eta^i_t\eta^j_{t'} \rangle =\delta^{ij}\delta(t-t')$.
The heat released by the particle to the fluid along a trajectory from time $t'$ to time $t$ reads $Q = -|{\bm V}_t|^2/2 + |{\bm V}_{t'}|^2/2 $. In the case at hand, since there is no force, no work is done on the particle and no potential energy is stored [@Note]. The entropy of the particle is defined as the state function $$\label{eq:Sp}
S_{p}({\bm x},{\bm v},t) = -\ln p ({\bm x},{\bm v},t)$$ where $p$ is the probability density of the particle position and velocity at time $t$ [@S05]. In other words, $p$ is the solution of the Fokker-Planck equation associated to Eq. . The entropy produced by the particle in the environment (the fluid) along a path is the integral of the released heat divided by temperature $$\label{eq:Senv}
S_{\mathit{env}} = - \int_{t'}^t \frac{1}{T({\bm X}_\tau)} {\bm V}_\tau \circ \dot{\bm V}_\tau \,
d\tau$$ where $\circ$ denotes the Stratonovich product (the integral is regularized according to the midpoint rule). The total entropy produced along the process is therefore $$\label{eq:Stot}
S_{tot} =S_p({\bm X}_t,t)-S_p({\bm X}_{t'},t') + S_{env}$$ and is positive on average as prescribed by the second law of thermodynamics [@CG08; @Note2].
We now turn our attention to the overdamped dynamics. It is possible to prove by means of asymptotic techniques (see [@Sekimoto; @B97; @MS00] and the Supplemental Material) that the the spatial trajectory ${\bm X}_t$ in tends – in the limit of small inertia and in the probabilistic sense – to the solution of $$\label{eq:OD}
\begin{array}{lll}
\dot{\bm X}_t & = & \sqrt{2 T({\bm X}_t) / \gamma } \; {\bm \eta}_t
\end{array}$$ The correct interpretation of the equation above is that the product on the right-hand-side has to be taken with the Itō, nonanticipative convention. The velocity follows the local Maxwell-Boltzmann distribution $w({\bm v}|{\bm x})=(2\pi T({\bm x}))^{-3/2} \exp[-|{\bm v}|^2/(2T({\bm x}))]$.
The Fokker-Planck equation associated to Eq. can be interpreted as the mass-conservation equation for a dilute colloidal suspension of noninteracting particles. The flux of particles is ${\bm J} = -T \gamma^{-1} {\bm \nabla} \rho - \gamma^{-1}\rho
{\bm \nabla} T$ and features the contributions of the osmotic force $-T {\bm \nabla} \ln \rho$, and of the thermophoretic force $- {\bm \nabla} T$. For times larger than $L^2\gamma/T$, where $L$ is the size of the vessel that encloses the fluid, the probability density of particle position reaches the equilibrium $\rho_{eq} \propto T^{-1}$ with a zero-flux balance of osmotic and thermophoretic forces.
Stochastic thermodynamics can be formulated for the overdamped dynamics . The particle entropy is $$\label{eq:ODSp}
S^{\mathit{(over)}}_p = -\ln \rho({\bm x},t)$$ where $\rho$ solves the Fokker-Planck equation associated to Eq. . When the particle moves from a region where the probability density $\rho$ is high to one where it is low, its entropy increases. The entropy produced in the environment is the work done by the thermophoretic force on the particle along a trajectory, divided by temperature $$\label{eq:ODSenv}
S^{\mathit{(over)}}_{env} = -\int_{t'}^t \frac{{\bm \nabla} T ({\bm X}_\tau)}{T({\bm X}_\tau)} \circ \dot{\bm X}_\tau d\tau = \ln \left( \frac{T({\bm X}_{t'})}{T({\bm X}_{t})} \right)$$ and it is positive when the particle moves from a hot region into a cold one. When the system reaches equilibrium the total entropy production $$\label{eq:ODStot}
S_{tot}^{(over)}= S^{\mathit{(over)}}_p({\bm X}_t,t) - S^{\mathit{(over)}}_p({\bm X}_{t'},t') +S^{\mathit{(over)}}_{env}$$ tends to zero along any trajectory since these two contributions compensate exactly. The average rate of entropy production is $$\label{eq:ODSave}
\frac{d}{dt} \left\langle S_{\mathit{tot}}^{\mathit{(over)}} \right\rangle =
\int \frac{\left|{\bm \nabla} (\rho T)\right|^2}{\gamma \rho T} d{\bm x} \ge 0$$ where the integral is over the volume of the vessel. It vanishes in equilibrium since $\rho T$ is then a constant. This result coincides with the expression obtained by macroscopic non-equilibrium thermodynamics for the entropy production by a dilute particle suspension [@GM].
We now move to the description of the results of the present work, where the limit of small [*yet finite*]{} inertia is considered. The derivations follow standard asymptotic expansion methods and are detailed in the Supplementary Material. The only assumption is that the length-scale of variation of the temperature be larger than the typical distance travelled during the friction time $\gamma^{-1}$, i.e. the gradients must not be exceedingly large $|{\bm \nabla} T| \ll T \gamma / v \sim \gamma T^{1/2}$ [@Note-BL-ratchet].
The average rate of entropy production of the Langevin-Kramers process , in the limit of vanishing inertia, is $$\label{eq:Save}
\frac{d}{dt}\left\langle S_{\mathit{tot}} \right\rangle = \frac{d}{dt}\left\langle S^{\mathit{(over)}}_{\mathit{tot}} \right\rangle + \frac{5}{6} \left\langle \frac{T}{\gamma} \left(\frac{{\bm \nabla} T}{T} \right)^2 \right\rangle$$ This [*exact asymptotic*]{} expression has to be contrasted with the overdamped [*approximation*]{} . It differs by an additional, positive contribution, which actually controls the asymptotic rate of entropy production since $d\langle S_{\mathit{tot}} ^{\mathit{(over)}}\rangle/dt$ vanishes as equilibrium is approached. This is the most conspicuous effect of the entropic anomaly.
It is indeed possible to isolate the source of the anomaly from the total entropy production , $S_{\mathit{tot}}=S_{\mathit{reg}}+S_{\mathit{anom}}$, where $$S_{\mathit{anom}} = \int_{t'}^t \frac{(5 T({\bm X}_\tau) -\left|{\bm V}_\tau\right|^2)}{2T({\bm X}_\tau)
^2} {\bm V}_\tau \cdot {\bm \nabla} T({\bm X}_\tau) \,d\tau$$ and to show that it arises from the $S_{\mathit{env}}$ contribution. The remainder $S_{\mathit{reg}}$ has a regular limit and tends to the overdamped entropy . In the limit of small inertia the anomalous entropy obeys the identity $$\label{eq:Jarz-anom}
\left\langle \exp\left(-S_{\mathit{anom}}\right) \right\rangle =1\;$$ and its average rate of production is $$\label{eq:Sanomave}
\frac{d}{dt} \left\langle S_{\mathit{anom}} \right\rangle =
\frac{5}{6}\int \rho \frac{ ({\bm \nabla} T)^2 }{\gamma T} d{\bm x}$$ that gives the rightmost term in . It is worth pointing out that the anomalous contribution cannot be eliminated by a suitable redefinition of the overdamped entropy. Indeed, it does not exist any sequential functional of the trajectories that gives the correct limiting statistics of $S_{\mathit{anom}}$ (see Supplemental Material).
\[fig:1\]
What is the interpretation of the anomaly ? Entropy is a measure of irreversibility. It is possible to show in very general terms that the entropy produced in the environment along a trajectory quantifies the probability of the reversed path relative to the forward one (see [@CG08] for a precise statement for diffusion processes). In the Langevin-Kramers case, the reversal consists in the inversion of the arrow of time and of the direction of velocity (see Figure 1) and one has $$\frac{P({\bm X}_{[t,t']},-{\bm V}_{[t,t']})}{P({\bm X}_{[t',t]},{\bm V}_{[t',t]})}= \exp\left(-S_{\mathit{env}} \right)$$ Similarly, for the overdamped dynamics, $$\frac{P^{\mathit(over)}({\bm X}_{[t,t']})}{P^{\mathit(over)}({\bm X}_{[t',t]})}= \exp\left(-S^{\mathit(over)}_{\mathit{env}} \right)$$ From the ratio of these two identities it follows that, in the limit of vanishing inertia, the anomalous entropy gives the relative weight of [*conditional*]{} probabilities $$\label{eq:condit}
\frac{P(-{\bm V}_{[t,t']}|{\bm X}_{[t,t']})}{P({\bm V}_{[t',t]}|{\bm X}_{[t',t]})}= \exp\left(-S_{\mathit{anom}} \right)$$ This result shows that the anomaly emerges from the breaking of time and velocity reversal symmetry at a given spatial position.
The anomaly can be traced back to the deviation of the velocity distribution from the Maxwell-Boltzmann one, $w({\bm v}|{\bm x})$. At equilibrium, the velocity statistics is indeed given by $$\label{eq:maxwdev}
p({\bm v} | {\bm x}) = w({\bm v}|{\bm x}) \left( 1 + \frac{(5 T({\bm x})-|{\bm v}|^2)} {6 \gamma T({\bm x})^2} {\bm v} \cdot {\bm \nabla} T({\bm x}) \right)$$ except for terms of order $\gamma^{-2}$ or higher. This deviation, albeit small, plays a crucial role as it breaks the velocity reversal symmetry along the temperature gradient direction, and is therefore responsible for the anomalous entropy production.
[ Further insight on the interpretation of the anomalous entropy can be gained by the following argument (see Figure \[fig:1\]). At the microscopic level (scales of order of $T^{1/2}/\gamma$, times of order $1/\gamma$) the particle trajectory often “goes round in circles”. Indeed, as seen from eq. , the trajectories with higher probability (positive sign of the correction term) are characterized by a downgradient sweep with speeds larger than the thermal velocity, followed by a slow upgradient motion (red trajectory in Figure \[fig:1\]). Note that the correction does not contribute to the spatial flux of particles as can be directly checked by multiplying by ${\bm v}$ and integrating over velocity. It does not contribute to the mean kinetic heat exchange either, since it is odd in velocity coordinates. During this futile cycling, however, heat is absorbed at higher temperatures and released at smaller ones, thereby producing entropy. In each cycle of duration $\sim \gamma^{-1}$, the amount of entropy produced is approximately given by the heat exchanged $\sim T$, times the difference of inverse temperatures at the extremes of the trajectory, $\sim T^{-2} |\nabla T| (T^{1/2}/\gamma)$. Reversed cycles (such as the one depicted in blue in Figure \[fig:1\]) give a similar contribution with opposite sign. However, forward cycles are more likely with probability $\sim T^{-1/2} |\nabla T|/\gamma$ as follows from eq. . This results in an overall average entropy production rate $\sim \gamma^{-1} T^{-1} |\nabla T|^2$ that we recognize as the anomalous term in eq. . ]{}
The present findings have a much broader range of applicability than the simple example discussed here (see Supplementary material). The addition of potential and nonconservative forces is straightforward since the related heat, work and entropy contributions all have regular limits and are thus correctly described by the overdamped dynamics. A position-dependent friction, which may account for the temperature dependence of viscosity or hydrodynamic effects due to the presence of material boundaries, can also be included with minor changes. Finally, it is also possible to consider time-dependent temperature, friction and forces, provided the variation is not faster than the timescale of overdamped motion. All these modifications do not alter our results. The interpretation is left unchanged as well.
In conclusion, we have shown that thermodynamics at the micro-scale is intrinsically anomalous in the limit of vanishingly small inertia. The anomaly arises from the breaking of time-reversal symmetry that emerges when the particle is subject to a temperature gradient. [ How may the entropic anomaly impact observable phenomena ? The most conspicuous effect should be on the efficiency of thermal stochastic engines in nonuniform temperature environments. Indeed, the intrinsic irreversibility arising from the anomalous contribution – which does not vanish even for quasistatic transformations – should irremediably hamper the ability of the engine of converting absorbed heat into work. Recent advances in experimental techniques of particle confinement, tracking, and heating might pave the way toward an experimental measurement of this effect [@BB11; @BAP12; @LKM10]. ]{}
[99]{} Bertrand Duplantier. Brownian Motion, “Diverse and Undulating”. In [*Einstein, 1905-2005*]{} Poincaré Seminar 2005, (Birkhäuser Verlag, Basel, 2006); [*Progress in Mathematical Physics*]{} 47, 201-293 (2005)
E. M. Purcell. [*American Journal of Physics*]{} 45, 3-11 (1977)
D. B. Dusenbery. [*Living at micro scale*]{}. Harvard University Press, Cambridge, Mass. (2009)
F. Ritort. [*Journal of Physics C*]{}, 18 (2006) R531-R583.
T. Li, S. Kheifets, D. Medellin, and M. G. Raizen. [*Science*]{} 328, 1673-1675 (2010)
V. Blickle and C. Bechinger. [*Nature Physics*]{} 8, 143 - 146 (2011)
Bérut [*et al*]{}. [*Nature*]{} 483, 187-190 (2012)
U. Seifert. [*Phys. Rev. Lett.*]{} 95, 040602 (2005)
K. Sekimoto. [*Stochastic energetics*]{}. Lect. Notes Phys. 799. Springer, Berlin Heidelberg (2010)
C. Jarzynski. [*Eur. Phys. J. B*]{} 64, 331–340 (2008).
M. Esposito, U. Harbola and S. Mukamel. [*Rev. Mod. Phys.*]{}, 81, 1665–1702, (2009).
C. Jarzynski. [*Annual Review of Condensed Matter Physics*]{} 2, 329-351 (2011).
U. Frisch. [*Turbulence*]{}. Cambridge University Press (1996).
K. Fujikawa and H. Suzuki. [*Path Integrals and quantum anomalies (International Series of Monographs on Physics)*]{}. Oxford University Press, USA (2004)
H. E. Camblong [*et al*]{}. [*Phys. Rev. Lett.*]{} 89, 220402 (2001)
The general form of the Langevin-Kramers equation is $\dot{\bm X}_t = {\bm V}_t$, $m\dot{\bm V}_t = {\bm f} -\gamma {\bm v} + \sqrt{kT \gamma} {\bm \eta}_t$, where $m$ is the particle mass, ${\bm f}$ the applied force, $k$ the Boltzmann constant. Equation follows by reduction to non-dimensional form through the rescalings ${\bm X} \to {\bm X}/L$, ${\bm V} \to {\bm V}/\sqrt{kT_0/m}$, ${\bm f} \to {\bm f}/(kT_0/L)$, $t \to t/(L/\sqrt{kT_0/m})$, $\gamma \to \gamma L/\sqrt{kT_0m}$, $T \to T/T_0$, where $L$ is the characteristic length-scale of the problem and $T_0$ is a reference temperature. The overdamped approximation is obtained by setting the inverse of non-dimensional friction to zero, i.e. $\sqrt{kT_0m}/(\gamma L)=0$.
Note that the heat released to the fluid does not alter the temperature profile. This is because the thermal diffusivity of water in standard conditions is six orders of magnitude larger than diffusion coefficient of typical Brownian particles. The fluid recovers the steady state before the particle has travelled any appreciable distance.
R. Chetrite and K. Gawȩdzki. [*Commun. Math. Phys.*]{} 282, 469–518 (2008).
The entropy production within the fluid is constant and independent of particle trajectories, see also [@Note], and can be safely ignored. According to classical non-equilibrium thermodynamics it is $\sigma = \kappa ({\bm \nabla} T)^2/T^2$ for a fluid with thermal conductivity coefficient $\kappa$ (see e.g. [@GM]). Since $\frac{5}{6}\frac{T}{\gamma}\rho$ has the same dimension as $\kappa$ we note that the combined anomaly of a set of non-interacting Brownian particles with number density $\rho$ is actually of the same form as $\sigma$.
S. R. de Groot and P. Mazur. [*Non-equilibrium thermodynamics*]{}. Dover Publications (1984)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We describe how to solve simultaneous [Padé]{}approximations over a power series ring ${{\mathsf{K}}}[[x]]$ for a field ${{\mathsf{K}}}$ using ${O^{\scriptscriptstyle \sim}\!}(n^{\omega - 1} d)$ operations in ${{\mathsf{K}}}$, where $d$ is the sought precision and $n$ is the number of power series to approximate. We develop two algorithms using different approaches. Both algorithms return a reduced sub-bases that generates the complete set of solutions to the input approximations problem that satisfy the given degree constraints. Our results are made possible by recent breakthroughs in fast computations of minimal approximant bases and Hermite [Padé]{}approximations.'
author:
- |
Johan Rosenkilde, né Nielsen\
Technical University of Denmark\
Denmark\
[email protected]
- |
Arne Storjohann\
University of Waterloo\
Canada\
[email protected]
- |
Johan Rosenkilde, né Nielsen\
\
\
Arne Storjohann\
\
\
bibliography:
- 'bibtex.bib'
title: 'Algorithms for Simultaneous [Padé]{}Approximations [^1] '
---
Introduction {#sec:intro}
============
The Simultaneous [Padé]{}approximation problem concerns approximating several power series $S_1,\ldots,S_n \in {{\mathsf{K}}}[[x]]$ with rational functions $\frac {\sigma_1}\lambda,\ldots,\frac {\sigma_n} \lambda$, all sharing the same denominator $\lambda$. In other words, for some $d \in {\mathbb Z\xspace}_{\geq 0}$, we seek $\lambda \in {{\mathsf{K}}}[x]$ of low degree such that each of $${{\textnormal{rem}}}(\lambda S_1,\ x^d) , {{\textnormal{rem}}}(\lambda S_2,\ x^d),\ \ldots,\ {{\textnormal{rem}}}(\lambda S_n,\ x^d)$$ has low degree. The study of Simultaneous [Padé]{}approximations traces back to Hermite’s proof of the transcendence of $e$ [@hermite_sur_1878]. Solving Simultaneous [Padé]{}approximations has numerous applications, such as in coding theory, e.g. [@feng_generalization_1991; @schmidt_collaborative_2009]; or in distributed, reliable computation [@clement_pernet_high_2014]. Many algorithms have been developed for this problem, see e.g. [@beckermann_uniform_1992; @olesh_vector_2006; @sidorenko_linear_2011; @nielsen_generalised_2013] as well as the references therein. Usually one cares about the regime where $d \gg n$. Obtaining $O(n d^2)$ is classical through successive cancellation, see [@beckermann_uniform_1994] or [@feng_generalization_1991] for a Berlekamp–Massey-type variant. Using fast arithmetic, the previous best was ${O^{\scriptscriptstyle \sim}\!}(n^\omega d)$, where $\omega$ is the exponent for matrix multiplication, see \[ssec:cost\]. That can be done by computing a minimal approximant basis with e.g. [@giorgi_complexity_2003; @GuptaSarkarStorjohannValeriote11]; this approach traces back to [@barel_general_1992; @beckermann_uniform_1992]. Another possibility which achieves the same complexity is fast algorithms for solving structured linear systems, e.g. [@bostan_solving_2008]; see [@chowdhury_faster_2015] for a discussion of this approach.
A common description is to require $\deg \lambda < N_0$ for some degree bound $N_0$, and similarly $\deg {{\textnormal{rem}}}(\lambda S_1,\, x^d) < N_i$ for $i = 1,\ldots,n$. The degree bounds could arise naturally from the application, or could be set such that a solution must exist. A natural generalisation is also to replace the $x^d$ moduli with arbitrary $g_1,\ldots,g_n \in {{\mathsf{K}}}[x]$. Formally, for any field ${{\mathsf{K}}}$:
\[prob:sim\_pade\] Given a tuple $({\bm{S}}, {\bm{g}}, {\bm{N}})$ where
- ${\bm{S}} = (S_1,\ldots,S_n) \in {{\mathsf{K}}}[x]^n$ is a sequence of polynomials,
- ${\bm{g}} = (g_1,\ldots,g_n) \in {{\mathsf{K}}}[x]^n$ is a sequence of moduli polynomials with $\deg S_i < \deg g_i$ for $i=1,\ldots,n$,
- and ${\bm{N}} = (N_0,\ldots,N_n)
\in {\mathbb Z\xspace}_{\geq 0}^{n+1}$ are degree bounds satisfying $1\leq N_0 \leq \max_i \deg g_i$ and $N_i \leq \deg g_i$ for $i=1,\ldots,n$,
find, if it exists, a non-zero vector $(\lambda, \phi_1, \ldots, \phi_n)$ such that
1. $\lambda S_i \equiv \phi_i \mod g_i$ for $i = 1,\ldots, n$, and \[p1item1\]
2. $\deg \lambda < N_0$ and $\deg \phi_i < N_i$ for $i=1,\ldots,n$.
We will call any vector $(\lambda, \phi_1, \ldots, \phi_n)$ as above *a solution* to a given Simultaneous [Padé]{}approximation problem. Note that if the $N_i$ are set too low, then it might be the case that no solution exists.
\[ex:simpade\] Consider over ${\mathbb F_{2}\xspace}[x]$ that $g_1 = g_2 = g_3 = x^5$, and ${\bm{S}} = (S_1,S_2,S_3) =
\left(x^{4} + x^{2} + 1,\,x^{4} + 1,\,x^{4} + x^{3} + 1\right)$, with degree bounds ${\bm{N}} = (5, 3, 4, 5)$. Then $\lambda_1 = x^4 + 1$ is a solution, since $\deg \lambda_1 < 5$ and $$\lambda_1 {\bm{S}} \equiv
\left(x^{2} + 1,\ 1,\ x^{3} + 1\right)
\mod x^5 \ .$$ $\lambda_2 = x^{3} + x$ is another solution, since $$\lambda_2 {\bm{S}} \equiv
\left(x,\ x^{3} + x,\ x^4+x^3 + x\right)
\mod x^5 \ .$$ These two solutions are linearly independent over ${\mathbb F_{2}\xspace}[x]$ and span all solutions.
Several previous algorithms for solving \[prob:sim\_pade\] are more ambitious and produce an entire *basis* of solutions that satisfy the first output condition $\lambda S_i \equiv \phi_i \mod g_i$ for $i=1,\ldots,n$, including solutions that do not satisfy the degree bounds stipulated by the second output condition. Our algorithms are slightly more restricted in that we only return the sub-basis that generates the set of solutions that satisfy both output requirements of \[prob:sim\_pade\]. Formally:
\[prob:sim\_pade\_basis\] Given an instance of \[prob:sim\_pade\], find a matrix $A \in {{\mathsf{K}}}[x]^{* \times (n+1)}$ such that:
- Each row of $A$ is a solution to the instance.
- All solutions are in the ${{\mathsf{K}}}[x]$-row space of $A$.
- $A$ is $(-{\bm{N}})$-row reduced[^2].
The last condition ensures that $A$ is minimal, in a sense, according to the degree bounds ${\bm{N}}$, and that we can easily parametrise which linear combinations of the rows of $A$ are solutions. We recall the relevant definitions and lemmas in \[sec:preliminaries\].
We will call such a matrix $A$ a *solution basis*. In the complexities we report here, we cannot afford to compute $A$ explicitly. For example, if all $g_i = x^d$, the number of field elements required to explicitly write down all of the entries of $A$ could be $\Omega(n^2d)$. Instead, we remark that $A$ is completely given by the problem instance as well as the first column of $A$, containing the $\lambda$ polynomials.[^3] Our algorithms will therefore represent $A$ row-wise using the following compact representation.
For a given instance of \[prob:sim\_pade\_basis\], a *solution specification* is a tuple $({\bm{\lambda}},{\bm{\delta}}) \in
{{\mathsf{K}}}[x]^{k \times 1} \times {\mathbb Z\xspace}_{<0}^k$ such that the *completion* of ${\bm{\lambda}}$ is a solution basis, and where ${\bm{\delta}}$ are the $(-{\bm{N}})$-degrees of the rows of $A$.
The *completion* of ${\bm{\lambda}} = (\lambda_1,\ldots,\lambda_k){^\top}$ is the matrix $$\begin{bmatrix}
\lambda_1 & {{\textnormal{rem}}}(\lambda_1 S_1,\ g_1) & \ldots & {{\textnormal{rem}}}(\lambda_1 S_n,\ g_n) \\
\vdots & & \ddots & \vdots \\
\lambda_k & {{\textnormal{rem}}}(\lambda_k S_1,\ g_1) & \ldots & {{\textnormal{rem}}}(\lambda_k S_n,\ g_n) \\
\end{bmatrix}
\ .$$
Note that ${\bm{\delta}}$ will consist of only negative numbers, since any solution ${\bm{v}}$ by definition has $\deg_{-{\bm{N}}} {\bm{v}} < 0$.
A solution specification for the problem in \[ex:simpade\] is $$({\bm{\lambda}}, {\bm{\delta}}) = \big( [x^4 + 1,\ x^3 + x]{^\top},\ (-1, -1) \big) \ .$$ The completion of this is $$A = \begin{bmatrix}
x^4 + 1 & x^{2} + 1 & 1 & x^{3} + 1 \\
x^3 + x & x & x^{3} + x & x^4+x^3 + x
\end{bmatrix}$$ One can verify that $A$ is $(-{\bm{N}})$-row reduced.
We present two algorithms for solving \[prob:sim\_pade\_basis\], both with complexity $O\big(n^{\omega-1}\, {{\mathsf{M}}}(d)\,(\log d)\,(\log
d/n)^2\big)$, where $d = \max_i \deg g_i$ and ${{\mathsf{M}}}(d)$ is the cost of multiplying two polynomials of degree $d$, see \[ssec:cost\]. They both depend crucially on recent developments that allow computing minimal approximant bases of non-square matrices faster than for the square case [@zhou_efficient_2012; @jeannerod_computation_2016]. We remark that from the solution basis, one can also compute the expanded form of one or a few of the solutions in the same complexity, for instance if a single, expanded solution to the simultaneous [Padé]{}problem is needed.
Our first algorithm in \[sec:dual\] assumes $g_i = x^d$ for all $i$ and some $d \in {\mathbb Z\xspace}_{\geq 0}$. It utilises a well-known duality between Simultaneous [Padé]{}approximations and Hermite [Padé]{}approximations, see e.g. [@beckermann_uniform_1992]. The Hermite [Padé]{}problem is immediately solvable by fast minimal approximant basis computation. A remaining step is to efficiently compute a single row of the adjoint of a matrix in Popov form, and this is done by combining partial linearisation [@GuptaSarkarStorjohannValeriote11] and high-order lifting [@storjohann_high-order_2003].
Our second algorithm in \[sec:intersect\] supports arbitrary $g_i$. The algorithm first solves $n$ single-sequence [Padé]{}approximations, each of $S_1,\ldots,S_n$. The solution bases for two problem instances can be combined by computing the intersection of their row spaces; this is handled by a minimal approximant basis computation. A solution basis of the full Simultaneous [Padé]{}problem is then obtained by structuring intersections along a binary tree.
Before we describe our algorithms, we give some preliminary notation and definitions in \[sec:preliminaries\], and in \[sec:subroutines\] we describe some of the computational tools that we employ.
Both our algorithms have been implemented in Sage v. 7.0 [@stein_sagemath_????] (though asymptotically slower alternatives to the computational tools are used). The source code can be downloaded from <http://jsrn.dk/code-for-articles>.
Cost model {#ssec:cost}
----------
We count basic arithmetic operations in ${{\mathsf{K}}}$ on an algebraic RAM. We will state complexity results in terms of an exponent $\omega$ for matrix multiplication, and a function ${{\mathsf{M}}}(\cdot)$ that is a multiplication time for ${{\mathsf{K}}}[x]$ [@von_zur_gathen_modern_2012 Definition 8.26]. Then two $n\times n$ matrices over ${{\mathsf{K}}}$ can be multiplied in $O(n^{\omega})$ operations in ${{\mathsf{K}}}$, and two polynomials in ${{\mathsf{K}}}[x]$ of degree strictly less than $d$ can be multiplied in ${{\mathsf{M}}}(d)$ operations in ${{\mathsf{K}}}$. The best known algorithms allow $\omega < 2.38$ [@coppersmith_matrix_1990; @LeGall14], and we can always take ${{\mathsf{M}}}(d) \in O(n (\log n) (\operatorname{loglog}n))$ [@CantorKaltofen].
In this paper we assume that $\omega > 2$, and that ${{\mathsf{M}}}(d)$ is super-linear while ${{\mathsf{M}}}(d) \in O(d^{\omega-1})$. The assumption ${{\mathsf{M}}}(d) \in O(d^{\omega-1})$ simply stipulates that if fast matrix multiplication techiques are used then fast polynomial multiplication should be used also: for example, $n \, {{\mathsf{M}}}(nd) \in O(n^{\omega} \, {{\mathsf{M}}}(d))$.
Preliminaries {#sec:preliminaries}
=============
Here we gather together some definitions and results regarding row reduced bases, minimal approximant basis, and their shifted variants. For a matrix $A$ we denote by $A_{i,j}$ the entry in row $i$ and column $j$. For a matrix $A$ over ${{\mathsf{K}}}[x]$ we denote by ${{\textnormal{Row}}}(A)$ the ${{\mathsf{K}}}[x]$-linear row space of $A$.
Degrees and shifted degrees
---------------------------
The degree of a nonzero vector ${\bm{v}} \in {{\mathsf{K}}}[x]^{1 \times m}$ or matrix $A \in {{\mathsf{K}}}[x]^{n\times m}$ is denoted by $\deg {\bm{v}}$ or $\deg A$, and is the maximal degree of entries of ${\bm{v}}$ or $A$. If $A$ has no zero rows the [*row degrees*]{} of $A$, denoted by ${{\textnormal{rowdeg}}}\, A$, is the tuple $(d_1,\ldots,d_n)$ with $d_i = \deg
{{\textnormal{row}}}(A,i)$.
The (row-wise) [*leading matrix*]{} of $A$, denoted by ${\rm LM}(A) \in {{\mathsf{K}}}^{n \times m}$, has ${\rm LM}(A)_{i,j}$ equal to the coefficient of $x^{d_i}$ of $A_{i,j}$.
Next we recall [@barel_general_1992; @zhou_efficient_2012; @jeannerod_computation_2016] the shifted variants of the notion of degree, row degrees, and leading matrix. For a [*shift*]{} ${\bm{s}} =
(s_1,\ldots,s_n) \in {\mathbb Z\xspace}^n$, define the $n \times n$ diagonal matrix $x^{{\bm{s}}}$ by $$x^{{\bm{s}}} := \left [ \begin{array}{ccc} x^{s_1}
& & \\
& \ddots & \\ & & x^{s_n} \end{array} \right ].$$ Then the [*${{\bm{s}}}$-degree*]{} of $v$, the [*${{\bm{s}}}$-row degrees*]{} of $A$, and the [*${\bm{s}}$-leading matrix*]{} of $A$, are defined by $\deg_{{\bm{s}}} v := \deg v x^{{\bm{s}}}$, ${{\textnormal{rowdeg}}}_{{\bm{s}}} A := {{\textnormal{rowdeg}}}\, Ax^{{\bm{s}}}$, and ${\rm LM}_{{\bm{s}}}(A) := {\rm
LM}(Ax^{{\bm{s}}})$. Note that we pass over the ring of Laurent polynomials only for convenience; our algorithms will only compute with polynomials. As pointed out in [@jeannerod_computation_2016], up to negation the definition of ${{\bm{s}}}$-degree is equivalent to that used in [@BeckermannLabahnVillard06] and to the notion of [*defect*]{} in [@beckermann_uniform_1994].
For an instance $({\bm{S}}, {\bm{g}}, {\bm{N}})$ of \[prob:sim\_pade\], in the context of defining matrices, we will be using ${\bm{S}}$ and ${\bm{g}}$ as vectors, and by ${ \ifx\null\null \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{\null}} \fi}$ denote the diagonal matrix with the entries of ${\bm{g}}$ on its diagonal.
Row reduced
-----------
Although row reducedness can be defined for matrices of arbitrary shape and rank, it suffices here to consider the case of matrices of full row rank. A matrix $R \in {{\mathsf{K}}}[x]^{n \times m}$ is [*row reduced*]{} if ${\rm LM}(R)$ has full row rank, and [*${\bm{s}}$-row reduced*]{} if ${\rm LM}_{{\bm{s}}}(R)$ has full row rank. Every $A \in {{\mathsf{K}}}[x]^{n \times m}$ of full row rank is left equivalent to a matrix $R \in {{\mathsf{K}}}[x]^{n \times m}$ that is ${{\bm{s}}}$-row reduced. The rows of $R$ give a basis for ${{\textnormal{Row}}}(A)$ that is minimal in the following sense: the list of ${{\bm{s}}}$-degrees of the rows of $R$, when sorted in non-decreasing order, will be lexicographically minimal. An important feature of row reduced matrices is the so-called “predictable degree”-property [@kailath_linear_1980 Theorem 6.3-13]: for any ${\bm{v}} \in {{\mathsf{K}}}[x]^{1 \times n}$, we have $$\deg_{{\bm{s}}}({\bm{v}} R) = \max_{i=1,\ldots,n}( \deg_{{\bm{s}}} {\rm row}(R,i)
+ \deg v_i ) \ .$$
A canonical ${\bm{s}}$-reduced basis is provided by the ${{\bm{s}}}$-Popov form. Although an ${{\bm{s}}}$-Popov form can be defined for a matrix of arbitrary shape and rank, it suffices here to consider the case of a non-singular matrix. The following definition is equivalent to [@jeannerod_computation_2016 Definition 1.2].
\[def:popov\] A non-singular matrix $R \in {{\mathsf{K}}}[x]^{n\times n}$ is in ${{\bm{s}}}$-Popov form if ${\rm LM}_{{\bm{s}}}(R)$ is unit lower triangular and the degrees of off-diagonal entries of $R$ are strictly less than the degree of the diagonal entry in the same column.
Adjoints of row reduced matrices
--------------------------------
For a non-singular matrix $A$ recall that the adjoint of $A$, denoted by ${\rm adj}(A)$, is equal to $(\det A)A^{-1}$, and that entry ${{{\textnormal{adj}}}(A)}_{i,j}{^\top}$ is equal to $(-1)^{i+j}$ times the determinant of the $(n-1) \times (n-1)$ sub-matrix that is obtained from $A$ by deleting row $i$ and column $j$.
\[lem:adjointRowReduced\] Let $A \in {{\mathsf{K}}}[x]^{n \times n}$ be ${\bm{s}}$-row reduced. Then ${{\textnormal{adj}}}(A){^\top}$ is $(-{\bm{s}})$-row reduced with $${{\textnormal{rowdeg}}}_{(-{\bm{s}})} {{\textnormal{adj}}}(A){^\top}=(\eta - s - \eta_1,\ldots , \eta - s -\eta_n) \ ,$$ where ${\bm{\eta}} = {{\textnormal{rowdeg}}}_{{\bm{s}}} A$, $\eta = \sum_i \eta_i$ and $s = \sum_i s_i$.
Since $A$ is ${\bm{s}}$-row reduced then $A x^{{\bm{s}}}$ is row reduced. Note that ${{\textnormal{adj}}}(A x^{{\bm{s}}}){^\top}(A x^{{\bm{s}}}){^\top}= (\det A x^{{\bm{s}}}) I_m$ with $\deg
\det A x^{{\bm{s}}} = \eta$. It follows that row $i$ of ${{\textnormal{adj}}}(A x^{{\bm{s}}}){^\top}$ must have degree at least $\eta - \eta_i$ since $\eta_i$ is the degree of column $i$ of $(A x^{{\bm{s}}}){^\top}$. However, entries in row $i$ of ${{\textnormal{adj}}}(A x^{{\bm{s}}}){^\top}$ are minors of the matrix obtained from $A x^{{\bm{s}}}$ by removing row $i$, hence have degree at most $\eta
- \eta_i$. It follows that the (row-wise) leading coefficient matrix of ${{\textnormal{adj}}}(A x^{{\bm{s}}}){^\top}$ is non-singular, hence ${{\textnormal{adj}}}(A x^{{\bm{s}}}){^\top}$ is row reduced. Since ${{\textnormal{adj}}}(A x^{{\bm{s}}}){^\top}=
(\det x^{{\bm{s}}}) {{\textnormal{adj}}}(A){^\top}x^{-{\bm{s}}}$ we conclude that ${{\textnormal{adj}}}(A){^\top}$ is $(-{\bm{s}})$-row reduced with ${{\textnormal{rowdeg}}}_{(-{\bm{s}})} {{\textnormal{adj}}}(A) =
(\eta - \eta_1 - s, \ldots, \eta - \eta_n - s)$.
Minimal approximant bases
-------------------------
We recall the standard notion of minimal approximant basis, sometimes known as order basis or $\sigma$-basis [@beckermann_uniform_1994]. For a matrix $A \in {{\mathsf{K}}}[x]^{n \times m}$ and order $d \in
{\mathbb Z\xspace}_{\geq 0}$, an *order $d$ approximant* is a vector ${\bm{p}} \in
{{\mathsf{K}}}[x]^{1 \times n}$ such that ${\bm{p}}A \equiv {\bm{0}} \mod x^d.$ An *approximant basis of order $d$* is then a matrix $F \in
{{\mathsf{K}}}[x]^{n \times n}$ which is a basis of all order $d$ approximants. Such a basis always exists and has full rank $n$. For a shift ${\bm{s}} \in {\mathbb Z\xspace}^n$, $F$ is then an *${\bm{s}}$-minimal approximant basis* if it is ${\bm{s}}$-row reduced.
Let ${{\ensuremath{\mathsf{MinBasis}}\xspace}}(d,A,{\bm{s}})$ be a function that returns $(F,{\bm{\delta}})$, where $F$ is an ${\bm{s}}$-minimal approximant basis of $A$ of order $d$, and ${\bm{\delta}} = {{\textnormal{rowdeg}}}_{{\bm{s}}} F$. The next lemma recalls a well known method of constructing minimal approximant bases recursively. Although the output of ${{\ensuremath{\mathsf{MinBasis}}\xspace}}$ may not be unique, the lemma holds for *any* ${\bm{s}}$-minimal approximant basis that ${{\ensuremath{\mathsf{MinBasis}}\xspace}}$ might return.
\[lem:paderec\] Let $A = \left [ \begin{array}{c|c}
A_1 & A_2 \end{array} \right ]$ over ${{\mathsf{K}}}[x]$. If $(F_1, {\bm{\delta}}_1) = {{\ensuremath{\mathsf{MinBasis}}\xspace}}(d,A_1,{\bm{s}})$ and $(F_2,{\bm{\delta}}_2) =
{{\ensuremath{\mathsf{MinBasis}}\xspace}}(d,F_1A_2,{\bm{\delta}}_1)$, then $F_2F_1$ is an ${\bm{s}}$-minimal approximant basis of $A$ of order $d$ with ${\bm{\delta}}_2 = {{\textnormal{rowdeg}}}_{{\bm{s}}} F_2 F_1$.
Sometimes only the [*negative part*]{} of an ${\bm{s}}$-minimal approximant bases is required, the submatrix of the approximant bases consisting of rows with negative ${\bm{s}}$-degree. Let function ${{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d,A,{\bm{s}})$ have the same output as ${{\ensuremath{\mathsf{MinBasis}}\xspace}}$, but with $F$ restricted to the negative part.
\[lem:paderecprune\] \[lem:paderec\] still holds if ${{\ensuremath{\mathsf{MinBasis}}\xspace}}$ is replaced by ${{\ensuremath{\mathsf{NegMinBasis}}\xspace}}$, and “an ${\bm{s}}$-minimal” is replaced with “the negative part of an ${\bm{s}}$-minimal.”
Using for example the algorithm `M-Basis` of [@giorgi_complexity_2003], it is easy to show that any order $d$ approximant basis $G$ for an $A$ of column dimension $m$ has $\det G = x^D$ for some $D \in {\mathbb Z\xspace}_{\geq 0}$ with $D \leq md$.
Many problems of ${{\mathsf{K}}}[x]$ matrices or approximations reduce to the computation of (shifted) minimal approximant bases, see e.g. [@beckermann_uniform_1994; @giorgi_complexity_2003], often resulting in the best known asymptotic complexities for these problems.
Direct solving of Simultaneous [Padé]{}approximations {#sec:direct_solve}
-----------------------------------------------------
Let $({\bm{S}}, {\bm{g}}, {\bm{N}})$ be an instance of \[prob:sim\_pade\_basis\] of size $n$. We recall some known approaches for computing a solution specification using row reduction and minimal approximant basis computation.
### Via reduced basis {#sec:direct_reduced_basis}
Using the predictable degree property it is easy to show that if $R \in {{\mathsf{K}}}[x]^{(n+1)
\times (n+1)}$ is an $(-{\bm{N}})$-reduced basis of $$A =
\left [ \begin{array}{c|c}
1 & {\bm{S}} \\\hline
& { \ifx\null\null \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{\null}} \fi}\end{array} \right]
\in {{\mathsf{K}}}[x]^{(n+1) \times (n+1)},$$ then the sub-matrix of $R$ comprised of the rows with negative $(-{\bm{N}})$-degree form a solution basis. A solution specification $({\bm{\lambda}}, {\bm{\delta}})$ is then a subvector ${\bm{\lambda}}$ of the first column of $R$, with ${\bm{\delta}}$ the corresponding subtuple ${\bm{\delta}}$ of ${{\textnormal{rowdeg}}}_{(- {\bm{N}})} R$.
Mulders and Storjohann [@mulders_lattice_2003] gave an iterative algorithm for performing row reduction by successive cancellation; it is similar to but faster than earlier algorithms [@kailath_linear_1980; @lenstra_factoring_1985]. Generically on input $F \in {{\mathsf{K}}}[x]^{m \times m}$ it has complexity $O(n^3 (\deg F)^2)$. Alekhnovich [@alekhnovich_linear_2005] gave what is essentially a Divide & Conquer variant of Mulders and Storjohann’s algorithm, with complexity ${O^{\scriptscriptstyle \sim}\!}(n^{\omega+1}\deg F)$. Nielsen remarked [@nielsen_generalised_2013] that these algorithms perform fewer iterations when applied to the matrix $A$ above, due to its low *orthogonality defect*: ${\rm OD}(F) = \sum{{\textnormal{rowdeg}}}F - \deg \det F$, resulting in $O(n^2(\deg A)^2)$ respectively ${O^{\scriptscriptstyle \sim}\!}(n^\omega \deg A)$. Nielsen also used the special shape of $A$ to give a variant of the Mulders–Storjohann algorithm that computes coefficients in the working matrix in a lazy manner with a resulting complexity $O(n \,\mathsf{P}(\deg A))$, where $\mathsf P(\deg A) = (\deg A)^2$ when the $g_i$ are all powers of $x$, and $\mathsf P(\deg A) = {{\mathsf{M}}}(\deg A)\deg A$ otherwise.
Giorgi, et al. [@giorgi_complexity_2003] gave a reduction for performing row reduction by computing a minimal approximant basis. For the special matrix $A$, this essentially boils down to the approach described in the following section.
When $n = 1$, the extended Euclidean algorithm on input $S_1$ and $g_1$ can solve the approximation problem by essentially computing the reduced basis of the $2 \times 2$ matrix $A$: each iteration corresponds to a reduced basis for a range of possible shifts [@sugiyama_further_1976; @justesen_complexity_1976; @gustavson_fast_1979]. The complexity of this is $O({{\mathsf{M}}}(\deg g_1) \log \deg g_1)$.
### Via minimal approximant basis
First consider the special case when all $g_i = x^d$ for the same $d$. An approximant ${\bm{v}} = (\lambda, \phi_1, \ldots, \phi_n)$ of order $d$ of $$\begin{aligned}
A &= \left [ \begin{array}{c}
- {\bm{S}} \\
I
\end{array} \right ]
\in {{\mathsf{K}}}[x]^{(n+1) \times n}\end{aligned}$$ clearly satisfies $\lambda S_i \equiv \phi_i \mod x^d$ for $i =
1,\ldots,n$; conversely, any such vector ${\bm{v}}$ satisfying these congruences must be an approximant of $A$ of order $d$. So the negative part of a $(-{\bm{N}})$-minimal approximant basis of $A$ of order $d$ is a solution basis.
In the general case we can reduce to a minimal approximant bases computation as shown by \[alg:simpadedirect\]. Correctness of the algorithm follows from the following result.
\[thm:simpadedirect\] Corresponding to an instance $({\bm{S}}, {\bm{g}}, {\bm{N}})$ of \[prob:sim\_pade\_basis\] of size $n$, define a shift ${\bm{h}}$ and order $d$:
- ${\bm{h}} := -({\bm{N}} \mid N_0 -1, \ldots, N_0-1) \in {\mathbb Z\xspace}^{2n+1}$
- $d := N_0 + \max_i \deg g_i -1$
If $G$ is the negative part of an ${\bm{h}}$-minimal approximant basis of $$H = \left [ \begin{array}{c} -{\bm{S}} \\
I \\
{ \ifx\null\null \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{\null}} \fi}\end{array} \right ] \in {{\mathsf{K}}}[x]^{(2n+1) \times n}$$ of order $d$, then the submatrix of $G$ comprised of the first $n+1$ columns is a solution basis to the problem instance.
An approximant ${\bm{v}} = (\lambda, \phi_1,\ldots,\phi_n,q_1,\ldots, q_n)$ of order $d$ of $H$ clearly satisfies $$\begin{aligned}
\label{eq:lambda}
\lambda S_i = \phi_i + q_ig_i \bmod
x^d\end{aligned}$$ for $i=1,\ldots,n$; conversely, any such vector ${\bm{v}}$ satisfying these congruences must be an approximant of $H$ of order $d$.
Now suppose ${\bm{v}}$ is an order $d$ approximant of $H$ with negative ${\bm{h}}$-degree, so $\deg \lambda \leq N_0-1$, $\deg \phi_i \leq
N_i-1$, and $\deg q_i \leq N_0 - 2$. Since \[prob:sim\_pade\] specifies that $\deg S_i < \deg g_i$ and $N_i \leq \deg g_i$, both $\lambda S_i$ and $q_i g_i$ will have degree bounded by $N_0 + \deg g_i - 2$. Since \[prob:sim\_pade\] specifies that $N_0 \geq 1$, it follows that both the left and right hand sides of (\[eq:lambda\]) have degree bounded by $N_0+ \deg g_i -2$, which is strictly less than $d$. We conclude that $$\begin{aligned}
\label{eq:lambda2}
\lambda S_i = \phi_i + q_i g_i\end{aligned}$$ for $i=1,\ldots,n$. It follows that ${\bm{v}} H = 0$ so ${\bm{v}}$ is in the left kernel of $H$. Moreover, restricting ${\bm{v}}$ to its first $n+1$ entries gives $\bar{{\bm{v}}} := (\lambda, \phi_1,\ldots,\phi_n)$, a solution to the simultaneous [Padé]{}problem with $\deg_{- {\bm{N}}} \bar{{\bm{v}}} = \deg_{{\bm{h}}} {\bm{v}}$. Conversely, if $\bar{{\bm{v}}} = (\lambda, \phi_1,\ldots,\phi_n)$ is a solution to the simultaneous [Padé]{}problem, then the extension ${\bm{v}} = (\lambda, \phi_1,\ldots,\phi_n,q_1,\ldots,q_n)$ with $q_i
= (\lambda S_i - \phi_i)/g_i \in {{\mathsf{K}}}[x]$ for $i=1,\ldots,n$ is an approximant of $H$ of order $d$ with $\deg_{{\bm{h}}} {\bm{v}} =
\deg_{-{\bm{N}}} \bar{{\bm{v}}}$.
Finally, consider that a left kernel basis for $H$ is given by $$K = \left[ \begin{array}{c|c} K_1 & K_2 \end{array} \right ] =
\left [ \begin{array}{cc|c} 1 & {\bm{S}} & \\
& { \ifx\null\null \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{\null}} \fi}& -I \end{array} \right ].$$ We must have $G = M K$ for some polynomial matrix $M$ of full row rank. But then $M K_1$ also has full row rank with ${{\textnormal{rowdeg}}}_{-{\bm{N}}} MK_1
= {{\textnormal{rowdeg}}}_{{\bm{h}}} G$.
${\bm{h}} {\leftarrow}-( {\bm{N}} \mid N_0-1,\ldots,N_0-1) \in {\mathbb Z\xspace}^{2n+1}$ $d {\leftarrow}N_0 + \max_i \deg g_i - 1$ $H =
\left[ \begin{array}{c}
-{\bm{S}} \\
I \\
{ \ifx\null\null \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{\null}} \fi}\end{array} \right]$ $(\left [ \begin{array}{c|c} {\bm{\lambda}} & \ast \end{array}
\right ], {\bm{\delta}}) {\leftarrow}{{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, H, {\bm{h}})$ ${\textbf{return }}({\bm{\lambda}}, {\bm{\delta}})$
[$\mathsf{DirectSimPade}$]{} can be performed in time ${O^{\scriptscriptstyle \sim}\!}(n^{\omega} \deg H) = {O^{\scriptscriptstyle \sim}\!}(n^\omega \max_i \deg g_i)$ using the minimal approximant basis algorithm by Jeannerod, et al. [@jeannerod_computation_2016], see \[sec:subroutines\].
A closely related alternative to [$\mathsf{DirectSimPade}$]{} is the recent algorithm by Neiger [@neiger_fast_2016] for computing solutions to modular equations with general moduli $g_i$. This would give the complexity ${O^{\scriptscriptstyle \sim}\!}(n^{\omega-1} \sum_i \deg g_i) \subset {O^{\scriptscriptstyle \sim}\!}(n^\omega \max_i \deg g_i)$.
All of the above solutions ignore the sparse, simple structure of the input matrices, which is why they do not obtain the improved complexity that we do here.
Computational tools {#sec:subroutines}
===================
The main computational tool we will use is the following very recent result from Jeannerod, Neiger, Schost and Villard [@jeannerod_computation_2016] on minimal approximant basis computation.
\[thm:orderbasis\] There exists an algorithm ${{\ensuremath{\mathsf{PopovBasis}}\xspace}}(d,A, {\bm{s}})$ where the input is an order $d \in {\mathbb Z\xspace}_+$, a polynomial matrix $A \in {{\mathsf{K}}}[x]^{n
\times m}$ of degree at most $d$, and shift ${\bm{s}} \in {\mathbb Z\xspace}^n$, and which returns $(F, {\bm{\delta}})$, where $F$ is an ${\bm{s}}$-minimal approximant basis of $A$ of order $d$, $F$ is in ${\bm{s}}$-Popov form, and ${\bm{\delta}} = {{\textnormal{rowdeg}}}_{{\bm{s}}} F$. [[$\mathsf{PopovBasis}$]{}]{}has complexity $O(n^{\omega-1}\, {{\mathsf{M}}}(\sigma)\, (\log \sigma) \, (\log \sigma /n)^2)$ operations in ${{\mathsf{K}}}$, where $\sigma = md$.
Our next result says that we can quickly compute the first row of ${{\textnormal{adj}}}(F)$ if $F$ is a minimal approximant basis in Popov form. In particular, since $F$ is an approximant basis $\det F = x^D$ for some $D \leq \sigma$, where $\sigma = md$ from \[thm:orderbasis\].
\[thm:fastsolver\] Let $F \in {{\mathsf{K}}}[x]^{n \times n}$ be in Popov form and with $\det F = x^D$ for some $D \in {\mathbb Z\xspace}_{\geq 0}$. Then the first row of ${{\textnormal{adj}}}(F)$ can be computed in $O(n^{\omega-1}\, {{\mathsf{M}}}(D)\, (\log D) \, (\log D/n))$ operations in ${{\mathsf{K}}}$.
Because $F$ is in ${\bm{s}}$-Popov form, $D$ is the sum of the column degrees of $F$. We consider two cases: $D \geq n$ and $D < n$.
First suppose $D
\geq n$. Partial linearisation [@GuptaSarkarStorjohannValeriote11 Corollary 2] can produce from $F$, with no operations in ${{\mathsf{K}}}$, a new matrix $G \in {{\mathsf{K}}}[x]^{\bar n \times \bar n}$ with dimension $\bar{n} < 2n$, $\deg G \leq \lceil D/n\rceil$, $\det G = \det F$, and such that $F^{{-1}}$ is equal to the principal $n \times n$ sub-matrix of $G^{{-1}}$. Let ${\bm{v}} \in {{\mathsf{K}}}[x]^{1 \times \bar{n}}$ be the first row of $x^DI_{\bar{n}}$. Then the first row of ${{\textnormal{adj}}}(F)$ will be the first $n$ entries of the first row of ${\bm{v}}G^{-1}$. High-order $X$-adic lifting [@storjohann_high-order_2003 Algorithm 5] using the modulus $X=(x-1)^{\lceil
D/n \rceil}$ will compute ${\bm{v}}G^{-1}$ in $O\big(n^{\omega}\,
{{\mathsf{M}}}(\lceil D/n \rceil) \,(\log \lceil D/n \rceil)\big)$ operations in ${{\mathsf{K}}}$ [@storjohann_high-order_2003 Corollary 16]. Since $D \geq n$ this cost estimate remains valid if we replace $\lceil D/n \rceil$ with $D/n$. Finally, from the super-linearity assumption on ${{\mathsf{M}}}(\cdot)$ we have $M(D/n) \leq (1/n) {{\mathsf{M}}}(D)$, thus matching our target cost.
Now suppose $D < n$. In this case we can not directly appeal to the partial linearisation technique since the resulting $O(n^{\omega} \lceil D/n\rceil)$ may be asymptotically larger than our target cost. But $D < n$ means that $F$ has — possibly many — columns of degree 0; since $F$ is in Popov form, such columns have a 1 on the matrix’s diagonal and are 0 on the remaining entries. The following describes how to essentially ignore those columns. $D$ is then greater than or equal to the number of remaining columns, thus effectuating the gain from the partial linearisation.
If $n-k$ is the number of such columns in $F$ that means we can find a permutation matrix $P$ such that $$\hat{F} := PFP{^\top}= \left [ \begin{array}{c|c}
F_1 & \\\hline
F_2 & I_{n-k}
\end{array} \right ] \ ,$$ with each column of $F_1$ having degree strictly greater than zero. Let $i$ be the row index of the single 1 in the first column of $P{^\top}$. Since $F^{-1} = P{^\top}\hat{F}^{-1}P$, we have $$\label{first}
{\rm row}({{\textnormal{adj}}}(F),1)P^{-1} = x^D\, {\rm row}(\hat{F}^{-1},i).$$ Considering that $$\hat{F}^{-1} = \left [ \begin{array}{c|c} F_1^{-1} & \\\hline -F_2F_1^{-1} & I_{n-k} \end{array} \right ],$$ it will suffice to compute the first $k$ entries of the vector on the right hand side of (\[first\]). If $i \leq k$ then let ${\bm{v}} \in {{\mathsf{K}}}[x]^{1 \times k}$ be row $i$ of $x^{D}I_k$. Otherwise, if $i>k$ then let ${\bm{v}}$ be row $i-k$ of $-x^{D}F_2$. Then in both cases, ${\bm{v}}F_1^{-1}$ will be equal to the first $k$ entries of the vector on the right hand side of (\[first\]). Like before, high-order lifting combined with partial linearisation will compute this vector in $O\big(k^{\omega}\, {{\mathsf{M}}}(\lceil D/k \rceil)\,(\log \lceil D/k \rceil)
\big)$ operations in ${{\mathsf{K}}}$. Since $D\geq k$ the cost estimate remains valid if $\lceil D/k \rceil$ is replaced with $D/k$.
Reduction to Hermite [PADÉ]{} {#sec:dual}
=============================
In this section we present an algorithm for solving \[prob:sim\_pade\_basis\] when $g_1 = \ldots = g_n = x^d$ for some $d \in {\mathbb Z\xspace}_{\geq 0}$. The algorithm is based on the well-known duality between the Simultaneous [Padé]{}problem and the Hermite [Padé]{}problem, see for example [@beckermann_uniform_1992]. This duality, first observed in a special case [@Mahler68], and then later in the general case [@beckermann_recursiveness_1997], was exploited in [@beckermann_fraction-free_2009] to develop algorithms for the fraction free computation of Simultaneous [Padé]{}approximation. We begin with a technical lemma that is at the heart of this duality.
\[lem:duality\] Let $\hat A, \hat B \in {{\mathsf{K}}}[x]^{(n+1)\times(n+1)}$ be as follows. $$\begin{aligned}
\hat A &=
\left [ \begin{array}{c|c}
x^d & -{\bm{S}} \\\hline
& I
\end{array} \right ]
&&
\hspace*{-1em}\hat B &=
\left [ \begin{array}{c|cccc}
1 & \\{ \hline \\[\dimexpr-\normalbaselineskip+2pt] }
{\bm{S}}{^\top}& x^d I
\end{array} \right ]
\end{aligned}$$ Then $\hat B$ is the adjoint of $\hat A{^\top}$. Furthermore, $\hat A{^\top}$ is an approximant basis for $\hat
B$ of order $d$, and $\hat B{^\top}$ is an approximant basis of $\hat A$ of order $d$.
Direct computation shows that $\hat A{^\top}\hat B = x^d I_m =
\det \hat A{^\top}I_m$, so $\hat B$ is the adjoint of $\hat
A{^\top}$.
Let now $G$ be an approximant basis of $\hat B$. By the above computation the row space of $\hat A{^\top}$ must be a subset of the row space of $G$. But since $G \hat B = (x^dI_m) R$ for some $R \in {{\mathsf{K}}}[x]^{(n+1)\times(n+1)}$, then $\det G = x^d \det R$. Thus $x^d \mid \det G$. But $\det \hat A{^\top}= x^d$, so the row space of $\hat A{^\top}$ can not be smaller than the row space of $G$. That is, $\hat A{^\top}$ is an approximant basis for $B$ of order $d$. Taking the transpose through the argument shows that $\hat
B{^\top}$ is an approximant basis of $\hat B$ of order $d$.
\[thm:dualityMinbasis\] Let $A$ and $B$ be as follows. $$\begin{aligned}
A &= \left [ \begin{array}{cccc}
-{\bm{S}} \\{ \hline \\[\dimexpr-\normalbaselineskip+1pt] }
I
\end{array} \right ] \in {{\mathsf{K}}}[x]^{(n+1) \times (n+1)}
&&&
\hspace*{-1em}B &= \left [ \begin{array}{c}
1 \\ {\bm{S}}
\end{array} \right] \in {{\mathsf{K}}}[x]^{(n+1) \times 1}\end{aligned}$$ If $G$ is an ${\bm{N}}$-minimal approximant basis of $B$ of order $d$ with shift ${\bm{N}} \in {\mathbb Z\xspace}_{\geq 0}^{n+1}$, then ${{\textnormal{adj}}}(G{^\top})$ is a $(-{\bm{N}})$-minimal approximant basis of $A$ of order $d$. Moreover, if ${\bm{\eta}} = {{\textnormal{rowdeg}}}_{{\bm{N}}} G$, then ${{\textnormal{rowdeg}}}_{-{\bm{N}}} {{\textnormal{adj}}}(G) =(\eta - N - \eta_1,\ldots , \eta - N
-\eta_{n+1})$, where $\eta = \sum_i \eta_i$ and $N = \sum_i N_i$.
Introduce $\hat A$ and $\hat B$ as in \[lem:duality\]. Clearly $G$ is also an ${\bm{N}}$-minimal approximant basis of $\hat B$ of order $d$. Likewise, $\hat A$ and $A$ have the same minimal approximant bases for given order and shift.
Assume, without loss of generality, that we have scaled $G$ such that $\det G$ is monic. Since $\hat A{^\top}$ is also an approximant basis for $\hat B$ of order $d$, then $\det G = \det \hat A{^\top}= x^d$. By definition $G\hat B = x^d R$ for some matrix $R \in
{{\mathsf{K}}}[x]^{(n+1)\times(n+1)}$. That means $$\begin{aligned}
x^{2d}((G\hat B){^\top}))^{{-1}}&= x^{2d}((x^d R){^\top})^{{-1}}\ , & \textrm{so} \\
(x^d(G{^\top})^{{-1}})(x^d(\hat B{^\top})^{{-1}}) &= x^{d}(R{^\top})^{{-1}}\ , & \textrm{that is} \\
{{\textnormal{adj}}}(G{^\top}) \hat A &= x^d (R{^\top})^{{-1}}\ .
\end{aligned}$$ Now $\det R = 1$ since $(x^d)^{n+1} \det R = \det(G\hat B) =
x^{d+nd}$, so $(R{^\top})^{{-1}}= {{\textnormal{adj}}}(R{^\top}) \in {{\mathsf{K}}}[x]^{(n+1)
\times (n+1)}$. Therefore ${{\textnormal{adj}}}(G{^\top})$ is an approximant basis of $\hat A$ of order $d$. The theorem now follows from \[lem:adjointRowReduced\] by noting that $G$ is ${\bm{N}}$-row reduced.
We apply \[thm:dualityMinbasis\] to the problem of \[ex:simpade\] with shifts ${\bm{N}} = (5, 3, 4, 5)$. We have $$\begin{aligned}
A &=
\left[\begin{array}{rrr}
x^{4} + x^{2} + 1 & x^{4} + 1 & x^{4} + x^{3} + 1 \\
1 & & \\
& 1 & \\
& & 1
\end{array}\right]
\\
B &= \left[\begin{array}{r}
1 \\ x^4 + x^2 + 1 \\ x^{4} + 1 \\ x^{4} + x^{3} + 1
\end{array}\right]
\end{aligned}$$ An ${\bm{N}}$-minimal approximant basis to order $d = 5$ of $B$ is $$\begin{aligned}
G &=
\left[\begin{array}{rrrr}
x & 0 & x & 0 \\
1 & x^{2} + 1 & 0 & 0 \\
0 & 1 & x^{2} + 1 & 0 \\
0 & x & x + 1 & 1
\end{array}\right] , \textrm{ and}
\\
{{\textnormal{adj}}}(G){^\top}&=
\left[\begin{array}{rrrr}
x^{4} + 1 & x^{2} + 1 & 1 & x^{3} + 1 \\
x & x^{3} + x & x & x^{4} + x \\
x^{3} + x & x & x^{3} + x & x^{4} + x^{3} + x \\
0 & 0 & 0 & x^{5}
\end{array}\right]
\ .
\end{aligned}$$ ${{\textnormal{adj}}}(G){^\top}$ can be confirmed to be an $(-{\bm{N}})$-minimal approximant basis of $A$, since ${{\textnormal{adj}}}(G){^\top}A \equiv 0 \mod x^d$, and since the $(-{\bm{N}})$-leading coefficient matrix of ${{\textnormal{adj}}}(G){^\top}$ has full rank.
Algorithm \[alg:simpadedual\] uses \[thm:dualityMinbasis\] to solve a Simultaneous [Padé]{}approximation by computing a minimal approximant basis of $B$ in Popov form.
$B {\leftarrow}[ 1, S_1, \ldots, S_n ]^T \in {{\mathsf{K}}}[x]^{(n+1) \times 1}$ $G {\leftarrow}{{\ensuremath{\mathsf{PopovBasis}}\xspace}}(d, B, {\bm{N}})$ \[line:simpadedual:basis\] ${\bm{\eta}} {\leftarrow}{{\textnormal{rowdeg}}}_{{\bm{N}}} G$ $\hat{{\bm{\lambda}}} {\leftarrow}$ first column of ${{\textnormal{adj}}}(G{^\top})$ \[line:simpadedual:firstcol\] $\hat{ {\bm{\delta}}} {\leftarrow}(\eta - N -\eta_1, \ldots, \eta - N - \eta_{n+1})$, where $\eta = \sum_i \eta_i$ and $N = \sum_i N_i$ \[line:simpadedual:degrees\] $I {\leftarrow}\{ i \mid \hat{{\bm{\delta}}}_i < 0 \}$, and $k {\leftarrow}|I|$ $({\bm{\lambda}},\ {\bm{\delta}}) {\leftarrow}\big( \hat{{\bm{\lambda}}}_{i \in I},\ (\hat{{\bm{\delta}}}_i)_{i \in I}
\big) \in {{\mathsf{K}}}[x]^{k \times 1} \times {\mathbb Z\xspace}^{k} $ ${\textbf{return }}({\bm{\lambda}}, {\bm{\delta}})$
\[alg:simpadedual\] is correct. The cost of the algorithm is $O(n^{\omega-1}\, {{\mathsf{M}}}(d) (\log d) (\log d/n)^2)$ operations in ${{\mathsf{K}}}$.
Correctness follows from \[thm:dualityMinbasis\]. The complexity estimate is achieved if the algorithms supporting \[thm:orderbasis\] and \[thm:fastsolver\] are used for the computation in lines 2 and 4, respectively.
A Divide & Conquer algorithm {#sec:intersect}
============================
Our second algorithm can handle the full generality of \[prob:sim\_pade\_basis\]. It works by first solving $n$ single [Padé]{}approximations, one for each of the $S_i$ individually, and then intersecting these solutions to form approximations of multiple $S_i$ simultaneously. The intersection is structured in a Divide & Conquer tree, and performed by computing minimal approximant bases. Let $({\bm{S}}, {\bm{g}}, {\bm{N}})$ be an instance of \[prob:sim\_pade\_basis\] of size $n$.
The idea of the intersection algorithm is the following: consider that we have solution specifications for two different Simultaneous [Padé]{}problems, $({\bm{\lambda}}_1, {\bm{\delta}}_1)$ and $({\bm{\lambda}}_2, {\bm{\delta}}_2)$. We then compute an approximant basis $G$ of the following matrix: $$\label{eqn:intersect_R}
R =
\left[\begin{array}{@{}c|c@{}}
1 & 1 \\ \hline
-{\bm{\lambda}}_1 & \\\hline
& -{\bm{\lambda}}_2 \\
\end{array}\right]$$ $G$ then encodes the *intersection* of the ${{\mathsf{K}}}[x]$-linear combinations of the ${\bm{\lambda}}_1$ with the ${{\mathsf{K}}}[x]$-linear combinations of the ${\bm{\lambda}}_2$: any $\lambda \in {{\mathsf{K}}}[x]$ residing in both sets of polynomials will appear as the first entry of a vector in the row space of $G$. We compute $G$ as an ${\bm{r}}$-minimal approximant basis to high enough order, where ${\bm{r}}$ is selected carefully such that the ${\bm{r}}$-degree of any $(\lambda \mid \ldots) \in {{\textnormal{Row}}}(G)$ will equal the $(-{\bm{N}})$-degree of the completion of $\lambda$ according to the combined Simultaneous [Padé]{}problem, whenever this degree is negative. From those rows of $G$ with negative ${\bm{r}}$-degree we then get a solution specification for the combined problem.
Consider again \[ex:simpade\]. We divide the problem into two sub-problems ${\bm{S}}_1 = (S_1, S_2)$, ${\bm{N}}_1 = (5, 3, 4)$, and ${\bm{S}}_2 = (S_3)$ and ${\bm{N}}_2 = (5, 5)$. Note that $N_{1,0} = N_{2,0} = 5$, since this is the degree bound on the sought $\lambda$ for the combined problem. The sub-problems have the following solution specifications and their completions: $$\begin{aligned}
({\bm{\lambda}}_1, {\bm{\delta}}_1) &= \big( [ x^{4} + 1,\ x^3 + x ]{^\top},\ ( -1, -1 ) \big)
\\
A_1 &=
\left(\begin{array}{rrr}
x^{4} + 1 & x^{2} + 1 & 1 \\
x^{3} + x & x & x^{3} + x
\end{array}\right)
\\
({\bm{\lambda}}_2, {\bm{\delta}}_2) &= \big( [ x^2,\ x^3 + x + 1 ]{^\top},\ ( -3, -2 ) \big)
\\
A_2 &=
\left(\begin{array}{rr}
x^{2} & x^{2} \\
x^{3} + x + 1 & x + 1
\end{array}\right)
\end{aligned}$$ We construct $R$ as in , and compute $G$, a minimal approximant basis of $R$ of order $7$ and with shifts ${\bm{r}}= (-5 \mid -1, -1 \mid -3, -2)$ (the $G$ below is actually in ${\bm{r}}$-Popov form): $$G = \left(\begin{array}{rrrrr}
x^{8} & 0 & 0 & 0 & 0 \\
x^{3} + x + 1 & x^{4} + 1 & 1 & 0 & 1 \\
x^{3} + x^{2} + x + 1 & 1 & x + 1 & 1 & 1 \\
x^{4} + x^{3} + x + 1 & 1 & 1 & x^{2} & 1 \\
x^{4} + 1 & 1 & 0 & x + 1 & x + 1
\end{array}\right)$$ $G$ has ${\bm{r}}$-row degree $(3, 3, 0, -1, -1)$. Only rows 4 and 5 have negative ${\bm{r}}$-degree, and their first entries are the linearly independent solutions $x^4 + x^3 + x + 1$ and $x^4 + 1$. Both solutions complete into vectors with $(-{\bm{N}})$-degree -1.
To prove the correctness of the above intuition, we will use \[alg:simpadedirect\] ([$\mathsf{DirectSimPade}$]{}). The following lemma says that to solve two simultaneous [Padé]{}approximations, one can compute a minimal approximant basis of one big matrix $A$ constructed essentially from two of the matrices employed in [$\mathsf{DirectSimPade}$]{}. Afterwards, \[lem:recursive\_R\_solves\] uses this to show that a minimal approximant basis of $R$ in provides the crucial information in a minimal approximant basis of $A$.
${\textbf{return }}{\ensuremath{\mathsf{DirectSimPade}}\xspace}({\bm{S}}, {\bm{g}}, {\bm{N}})$ ${\bm{S}}_1, {\bm{g}}_1 {\leftarrow}$ the first ${\lceil n/2 \rceil}$ elements of ${\bm{S}}, {\bm{g}}$ ${\bm{S}}_2, {\bm{g}}_2 {\leftarrow}$ the last ${\lfloor n/2 \rfloor}$ elements of ${\bm{S}}, {\bm{g}}$ ${\bm{N}}_1
{\leftarrow}(N_0,N_1,\ldots,N_{{\lceil n/2 \rceil}})$ ${\bm{N}}_2
{\leftarrow}(N_0,N_{{\lceil n/2 \rceil}+1},\ldots,N_n)$ $({\bm{\lambda}}_1, {\bm{\delta}}_1) {\leftarrow}{\ensuremath{\mathsf{RecursiveSimPade}}\xspace}\big({\bm{S}}_1, {\bm{g}}_1, {\bm{N}}_1)$ $({\bm{\lambda}}_2, {\bm{\delta}}_2) {\leftarrow}{\ensuremath{\mathsf{RecursiveSimPade}}\xspace}
\big({\bm{S}}_2, {\bm{g}}_2, {\bm{N}}_2)$ ${\bm{r}} {\leftarrow}(-N_0 \mid {\bm{\delta}}_1 \mid {\bm{\delta}}_2)$ $d {\leftarrow}N_0 + \max_i \deg g_i - 1$ \[line:recursive:choosed\] \[lineH\] $R {\leftarrow}\left[\begin{array}{c|c}
1 & 1\\ \hline
-{\bm{\lambda}}_1 & \\\hline
& -{\bm{\lambda}}_2
\end{array}\right]$ $(\left [\begin{array}{c|c} {\bm{\lambda}} & \ast \end{array} \right ],
{\bm{\delta}}) {\leftarrow}{{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, R, {\bm{r}})$ \[calltoNegMin\] ${\textbf{return }}({\bm{\lambda}}, {\bm{\delta}})$
\[lem:recursive\_big\_matrix\] Let $({\bm{S}}_1, {\bm{g}}_1,{\bm{N}}_1)$ and $({\bm{S}}_2, {\bm{g}}_2,{\bm{N}}_2)$ be two instances of \[prob:sim\_pade\_basis\] of lengths $n_1, n_2$ respectively, and where ${\bm{N}}_1 = (N_0 \mid \grave{{\bm{N}}_1})$ and ${\bm{N}}_2 = (N_0 \mid \grave{{\bm{N}}_2})$. Let ${\bm{S}} = ({\bm{S}}_1 \mid {\bm{S}}_2)$, ${\bm{g}} = ({\bm{g}}_1 \mid {\bm{g}}_2)$ and ${\bm{N}} = (N_0 \mid \grave{{\bm{N}}_1} \mid \grave{{\bm{N}}_2})$ be the combined problem having length $n = n_1 + n_2$.
Let ${\bm{h}}_i = (-{\bm{N}}_i \mid N_0 -1 \ldots N_0 -1) \in {\mathbb Z\xspace}^{2n_i+1}$ for $i=1,2$. Let $(F, {\bm{\delta}}) = {{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, A, {\bm{a}})$, where $A$ of dimension $(2n+3) \times (n+2)$ is given as: $$A = \left [ \begin{array}{c|c}
A_1 & A_2
\end{array} \right ]
= \left [ \begin{array}{cc|cc}
& & 1 & 1 \\
-{\bm{S}}_1 & & -1 \\
I & & \\
{ \ifx\null1 \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{1}} \fi} & & \\
& -{\bm{S}}_2 & & -1 \\
& I & \\
& { \ifx\null2 \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{2}} \fi} &
\end{array} \right ] ,$$ with ${\bm{a}} = (- N_0 \mid {\bm{h}}_1 \mid {\bm{h}}_2)$ and $d = N_0 + \max_i \deg g_i - 1$. Then $({\bm{\lambda}}, {\bm{\delta}})$ is a solution specification to $({\bm{S}}, {\bm{g}}, {\bm{N}})$, where ${\bm{\lambda}}$ is the first column of $F$.
Note that the matrix $A$ is right equivalent to the following matrix $B$: $$B := A
\left [ \begin{array}{cccc}
& & I & \\
& & & I \\
1 & & {\bm{S}}_1 & \\
& 1 & & {\bm{S}}_2
\end{array} \right ] =
\left [ \begin{array}{cc|cc}
1 & 1 & - {\bm{S}}_1 & -{\bm{S}}_2 \\
-1 & & & \\
& & I & \\
& & { \ifx\null1 \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{1}} \fi} & \\
& -1 & & \\
& & & I \\
& & & { \ifx\null2 \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{2}} \fi}
\end{array} \right ].$$ Since $F$ is an ${{\bm{a}}}$-minimal approximant of $A$ of order $d$, then it will also be one for $B$. Let $P$ be the permutation matrix that produces the following matrix $C := P B$: $$\setlength{\arraycolsep}{.8\arraycolsep}
C = P B =
\left [ \begin{array}{cc|cc}
1 & 1 & - {\bm{S}}_1 & -{\bm{S}}_2 \\
& & I & \\
& & & I \\
& & { \ifx\null1 \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{1}} \fi} & \\
& & & { \ifx\null2 \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{2}} \fi} \\\hline
-1 & & & \\
& -1 & &
\end{array} \right ] =
\left [ \begin{array}{cc|c}
1 & 1 & - {\bm{S}} \\ & & I \\ & & { \ifx\null\null \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{\null}} \fi}\\\hline
-1 & & \\
& -1 &
\end{array} \right ].$$ Define ${\bm{c}} := {\bm{a}} P^{-1}$, and note that ${\bm{c}} = ({\bm{h}} \mid -N_0, -N_0)$. Since $F = {{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, A, {\bm{a}})$, then $(FP^{-1}, {\bm{\delta}})$ is a valid output of ${{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, C, {\bm{c}})$. Furthermore, since the first column of $P$ is $(1, 0, \ldots, 0)$, the first column of $F$ will be equal to the first column of $FP^{{-1}}$.
We are therefore finished if we can show that if $(F', {\bm{\delta}}')$ is any valid output of ${{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, C, {\bm{c}})$, then the first column of $F'$ together with ${\bm{\delta}}'$ form a solution specification to $({\bm{S}}, {\bm{g}}, {\bm{N}})$.
Consider therefore such an $(F', {\bm{\delta}}')$. By the first two columns of $C$, we must have $F'_{*,1} \equiv F'_{*,2n+2} \equiv F'_{*,2n+3} \mod x^d$, where $F'_{*,i}$ denotes the $i$’th column of $F'$. Since each row of $F'$ have negative ${\bm{c}}$-degree, and since $N_0 < d$, then the congruences must lift to equalities. We can therefore write $F = [ G \mid F'_{*,1} \mid F'_{*,1} ]$ for some $G \in {{\mathsf{K}}}[x]^{k \times (2n+1)}$ for some $k$, and we have ${{\textnormal{rowdeg}}}_{{\bm{h}}} G = {{\textnormal{rowdeg}}}_{{\bm{c}}} F' = {\bm{\delta}}'$.
By the last $n$ columns of $C$, we have $G H \equiv 0 \mod x^d$, where $$H =
\left[\begin{array}{c}
-{\bm{S}} \\
I \\
{ \ifx\null\null \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{\null}} \fi}\end{array}\right] \ .$$ In fact, $(G, {\bm{\delta}}')$ is a valid output for ${{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, H, {\bm{h}})$: for $G$ has full row rank since $F'$ does; $G$ is ${\bm{h}}$-row reduced since $F'$ is ${\bm{c}}$-row reduced; and any negative ${\bm{h}}$-order $d$ approximant of $H$ must clearly be in the span of $G$ since $F'$ is a negative ${\bm{c}}$-minimal approximant basis of $C$.
By the choice of $d$, then \[thm:simpadedirect\] therefore implies that the first column of $G$ together with ${\bm{\delta}}'$ form a solution specification to the problem $({\bm{S}}, {\bm{g}}, {\bm{N}})$. Since the first column of $G$ is also the first column of $F'$, this finishes the proof.
\[lem:recursive\_R\_solves\] In the context of \[lem:recursive\_big\_matrix\], let $({\bm{\lambda}}_1, {\bm{\delta}}_1)$ and $({\bm{\lambda}}_2, {\bm{\delta}}_2)$ be solution specifications to the two sub-problems, and let ${\bm{r}} = (-N_0 \mid {\bm{\delta}}_1 \mid {\bm{\delta}}_2)$. If $([ {\bm{\lambda}} \mid * ], {\bm{\delta}}) = {{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, R, {\bm{r}})$, where ${\bm{\lambda}}$ is a column vector and $$R =
\left[\begin{array}{c|c}
1 & 1\\ \hline
-{\bm{\lambda}}_1 & \\\hline
& -{\bm{\lambda}}_2
\end{array}\right] \ ,$$ then $({\bm{\lambda}}, {\bm{\delta}})$ is a solution specification for the combined problem.
We will prove the lemma by using \[lem:paderecprune\] to relate valid outputs of ${{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, R, {\bm{r}})$ with valid outputs of ${{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, A, {\bm{a}})$ from \[lem:recursive\_big\_matrix\].
For $i=1,2$, since $({\bm{\lambda}}_i, {\bm{\delta}}_i)$ is a solution specification to the $i$’th problem, then by \[thm:simpadedirect\] there is some $G_i \in {{\mathsf{K}}}[x]^{k_i \times 2n_i+1}$ whose first column is ${\bm{\lambda}}_i$ and such that $G_i$ is a valid output of ${{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, H_i, {\bm{h}}_i)$, where $$H_i = \left [ \begin{array}{c}
-{\bm{S}}_i \\
I \\
{ \ifx\nulli \Gamma_{{\bm{g}}} \else \Gamma_{{\bm{g}}_{i}} \fi}
\end{array} \right ] \in {{\mathsf{K}}}[x]^{(2n_i+1) \times n_i} ,$$ and ${\bm{h}}_i$ is as in \[lem:recursive\_big\_matrix\]. Note now that if $$F_1 := \left [\begin{array}{ccc}
1 & & \\
& G_1 \\
& & G_2
\end{array} \right ] \in {{\mathsf{K}}}[x]^{(k_1+k_2+1) \times (2n_1 + 2n_2 + 3)} ,$$ then $(F_1, {\bm{r}})$ is a valid output of ${{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, A_1, {\bm{a}})$: for ${{\textnormal{rowdeg}}}_{{\bm{a}}} F_1$ is clearly ${\bm{r}}$; $F_1$ has full row rank and is ${\bm{r}}$-row reduced; and the rows of $F_1$ must span all ${\bm{a}}$-order $d$ approximants of $A_1$, since the three column “parts” of $F_1$ correspond to the three row parts of $A_1$. .
Note now that $F_1 A_2 = R$. Thus by \[lem:paderecprune\], if $(F_2, {\bm{\delta}}) = {{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, R, {\bm{r}})$, then $(F_2 F_1, {\bm{\delta}})$ is a valid output of ${{\ensuremath{\mathsf{NegMinBasis}}\xspace}}(d, A, {\bm{a}})$. Note that by the shape of $F_1$ then the first column ${\bm{\lambda}}$ of $F_2 F_1$ is the first column of $F_2$. Thus ${\bm{\lambda}}, {\bm{\delta}}$ are exactly as stated in the lemma, and by \[lem:recursive\_big\_matrix\] they must be a solution specification to the combined problem.
\[alg:recsimpade\] is correct. The cost of the algorithm is $O(n^{\omega-1}\, {{\mathsf{M}}}(d) (\log d) (\log d/n)^2)$, $d = \max_i \deg g_i$.
Correctness follows from \[lem:recursive\_R\_solves\]. For complexity, note that the choice of order in \[line:recursive:choosed\] is bounded by $2\max_i \deg g_i$, i.e. twice the value of $d$ of this theorem. So if $T(n)$ is the cost \[alg:recsimpade\] for given $n$ and where the order will be bounded by $O(d)$, then we have the following recursion: $$T(n) = \left \{\begin{array}{ll}
2T(n/2) + P(n) & \textrm{if } n > 1 \\
O({{\mathsf{M}}}(d)\log d) & \textrm{if } n = 1 \textrm{ (see {\Fref{sec:direct_reduced_basis}})}
\end{array}\right . \ ,$$ where $P(n)$ is the cost of line \[calltoNegMin\]. Using algorithm [$\mathsf{PopovBasis}$]{} for the computation of the negative part of the minimal approximant bases we can set $P(n)$ to the target cost. The recursion then implies $T(n) \in O(P(n))$.
[**Acknowledgements.**]{} The authors would like to thank George Labahn for valuable discussions, and for making us aware of the Hermite–Simultaneous [Padé]{}duality. We would also like to thank Vincent Neiger for making preprints of [@jeannerod_computation_2016] available to us. The first author would like to thank the Digiteo Foundation for funding the research visit at Waterloo, during which most of the ideas of this paper were developed.
[^1]: $\copyright$ Johan Rosenkilde, Arne Storjohann. This is the authors’ version of the work. It is posted here for your personal use. Not for redistribution. The definitive version was published in ISSAC ’16, http://dx.doi.org/10.1145/2930889.2930933.
[^2]: The notions $(-{\bm{N}})$-degree, $\deg_{-{\bm{N}}}$ and $(-{\bm{N}})$-row reduced are recalled in \[sec:preliminaries\].
[^3]: The restriction $N_i \leq \deg g_i$ in \[prob:sim\_pade\] ensures that for a given $\lambda$, the only possibilities for the $\phi_i$ in a solution are ${{\textnormal{rem}}}(\lambda S_i, \ g_i)$. In particular, if we allowed $N_i > \deg g_i$ then $(0,\ldots, 0, g_i, 0, \ldots,
0)$ would be a solution which can not be directly reconstructed from its first element.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Cellular cortex, which is a highly viscous thin cytoplasmic layer just below the cell membrane, controls the cell’s mechanical properties, which can be characterized by a hydrodynamic length scale $\ell$. Cells actively regulate $\ell$ via the activity of force generating molecules, such as myosin II. Here we develop a general theory for such systems through coarse-grained hydrodynamic approach including activity in the static description of the system providing an experimentally accessible parameter and elucidate the detailed mechanism of how a living system can actively self-regulate its hydrodynamic length scale, controlling the rigidity of the system. Remarkably, we find that $\ell$, as a function of activity, behaves universally and roughly inversely proportional to the activity of the system. Our theory rationalizes a number of experimental findings on diverse systems and comparison of our theory with existing experimental data show good agreement.'
author:
- Saroj Kumar Nandi
title: 'Activity-dependent self-regulation of viscous length scales in biological systems'
---
Introduction
============
The mechanical properties of a cell are governed by a highly-viscous thin layer of cytoplasm, the cell-cortex, just below the cell membrane. For most practical purposes, cortex can be assumed consisting of actin filaments and myosin motors. Cells must actively regulate the mechanical properties of the cortex that can be characterized by a hydrodynamic length scale, $\ell=\sqrt{\eta/\gamma}$, where $\eta$ is the viscosity of the cortex and $\gamma$ is some friction (e.g., exerted by the cell cytoplasm and cell membrane on the cortex). For example, $\ell$ is much bigger in the quiescent state compared to that when the cell is dividing [@cellbook; @mayer2010; @turlier2014]. Apart from its essential role in the growth and development of living organisms, understanding the detailed behavior of $\ell$ is important from other aspects such as the efficient design of artificial tissues, cellular self-assembly as well as in the field of Biomechanics [@huang1999; @jakab2004; @rumi2007]. It is known that the activity of the motor molecules, like myosin and kinesin, change the property of an assembly of filament-like molecules and motor molecules. A number of recent studies have explored the effect of activity on the properties of such systems [@ahmed2015; @loftus2014; @angelini2011; @koenderink2009; @zhou2009; @fodor2015; @gladrow2016; @battle2016; @sheinman2015; @alvarado2013], for example, myosin II activity softens suspended cells [@chan2015], changes the viscoelastic behavior of an actomyosin network [@humphrey2002] and controls the dynamics in mouse oocytes [@almonacid2015], bacterial cytoplasm (devoid of myosin) gets fluidized by metabolic activity [@parry2014], kinesin motors lead to spontaneous flow in an assembly of microtubules [@sanchez2012] etc. The existing theories of actomyosin cortex have mostly focused on the quiescent state and treated the system as an elastic network [@koenderink2009; @wang2011; @wang2012; @gladrow2016]. In this work, we are interested in the dynamic state when the cell is about to undergo a division [@turlier2014; @sundar2014] and a quantitative description of how activity affects $\ell$ in this regime does not yet exist.
Developing a theory of a highly viscous system, such as cell cortex or a dense assembly of actin and myosin molecules, requires the application of glassy phenomenology [@parry2014; @zhou2009; @nishizawa2017]. Coarse-grained hydrodynamic approaches, including the complexity of the system through the effective parameters of the theory for a minimal description, have been immensely successful in describing active systems [@turlier2014; @sebastian2012; @sundar2014; @sriramrmp; @ranft2010; @cates2015; @jacques15]; we extend such theories to their dense regime. The existing coarse-grained approaches assume a spatial variation in active stress coming from the spatial variation of motor activity and the resulting force transmission [@mayer2010; @sundar2014]. However, these models typically assume an effective viscosity reflecting the internal dissipation and successfully reproduce the the spatiotemporal cortical dynamics during cell division and polarization [@turlier2014; @mayer2010; @sundar2014].
On the other hand, molecular approaches, taking the chemical details of rapid actin turnover through polymerization and depolymerization, their remodeling and force generation mechanism through cross-linking and myosin motors activity [@robin2014; @fritzsche2016; @carlsson2010] have predicted patterns of cortical flow. It remains unclear how to connect this microscopic origin of force generation to large-scale cortical properties. In an important recent work [@william2017], McFadden [*et al*]{} have developed a minimal 2$D$ model of cross-linked actin filaments and myosin motors providing a framework to understand the effects of local microscopic modulation of interactions on the large-scale cellular flows.
In this work, we first develop a generic theory for an assembly of fibrillar and active force generating molecules in the dense high-viscous regime, applicable for the systems of our interest, through the approach of coarse-grained hydrodynamics. We treat the dynamics of the system within the mode-coupling theory (MCT) framework of glassy systems [@das2004; @goetzebook]. In this work, we concentrate on the effect of myosin activity on the hydrodynamic length scale, $\ell$, and show that biological systems can actively self-regulate their length scales controlling their rigidity generalizing recent experimental findings [@humphrey2002; @parry2014; @chan2015; @almonacid2015; @sanchez2012]. We further find that $\ell$ is roughly inversely proportional to activity irrespective of the nature of the system. This is an important, testable prediction of the theory and comparison of our theory with existing experimental data shows good agreement. The rest of the paper is organized as follows: We provide a description of the system of our interest and the simplifying assumptions in Sec. \[description\]. The detailed theory is presented in Sec. \[theory\] followed by the results in Sec. \[results\]. We conclude the paper by discussing our results in Sec. \[disc\]. We have relegated some of the technical details to an Appendix.
![A simplified illustration of the system of our interest. It can be a part of a cell cortex as in [@chan2015; @almonacid2015] or an in vitro assembly of fibrillar molecules like actin or microtubule and force generators like myosin or kinesin [@humphrey2002; @sanchez2012]. The force generating molecules can be either in active or inactive states. When in the active state, as shown at the top in the right, they work both as cross-linkers as well as force generators by walking on two actin filaments. Myosin molecules walk from $-$ to $+$ ends of actin filaments, pulling the filaments towards each other, thus generating force in the network. On the other hand, when myosin molecules are in the inactive state, they act only as cross-linkers. Myosin molecules can change their states from the inactive to the active state through ATP consumption. There are also other cross-linking proteins, like $\alpha$-actinin, in the cortex. The blue patches are representative of focal adhesions (not to the scale, as they are 2-3 orders of magnitude larger than the dimensions of individual proteins). Actin filaments, through cross-linking and attachment with the focal adhesion points create a matrix of inhomogeneous density. All other proteins are treated as passive particles whereas myosin molecules in the active state constitute the active particles in the system.[]{data-label="schematicfluid"}](actincortex_schematic.eps){width="\columnwidth"}
Description of the system {#description}
=========================
For concreteness, we discuss the model in terms of the actomyosin network in the cortex of cells as pictorially illustrated in Fig. \[schematicfluid\], however, the theory is applicable to any assembly of filament-like molecules and the active force generators in the assembly. Cortex is a complicated, dynamic assembly of many proteins; here we consider a simplified picture to enable theoretical development. We set up the problem by first discussing the individual microscopic mechanisms below.
- [**Turnover dynamics of actin filaments:**]{} The actin filaments within cortex are dynamic in nature, constantly being polymerized at one end and depolymerized at the other [@fritzsche2013; @fritzsche2016; @cellbook]. We assume that the turnover dynamics is in a steady state and is not affected by the activity of myosin motor molecules. We thus describe the system through a time-independent inhomogeneous density modulation and a fluctuating part of density that includes the effect of turnover dynamics as well as dynamics of other molecules (Fig. \[schematicfluid\]).
- [**Cross-linkers:**]{} Several types of molecules, such as $\alpha$-actinin as well as myosin molecules act as cross-linkers among actin filaments [@fritzsche2013; @cellbook]. We do not differentiate between the cross-linking abilities of different molecules. Cross-linkers help to entangle the filament-like molecules and is parametrized by the inhomogeneous density within our theory.
- [**Focal adhesions:**]{} The actin filament network is also attached to the focal adhesion points (Fig. \[schematicfluid\]). Since we are not interested in the effects of external medium, we simply assume that ATP concentration does not change the effects of focal adhesions, which simply contribute to the inhomogeneous density within the cortex.
- [**Force generators in the cortex:**]{} Myosin II molecules walk on two actin filaments and generate forces in the cortex. Since we are interested in the general picture here, and not in the detailed chemistry, from now on we denote the force generators simply as myosin. Actin filaments are polar molecules and myosin walks from the negative to the positive ends (Fig. \[schematicfluid\]) thus pulling the two filaments towards each other. Myosin molecules can be in two different states, active and inactive; they generate forces when they are in the active state and act as cross-linkers alone when inactive. We designate the myosin molecules in the active state as the active particles. The myosin molecules can interchange their states of either active or inactive through ATP consumption.
With a fixed amount of ATP supply, we can assume on the average a constant number of myosin molecules in the active state. Thus, the amount of ATP and myosin molecules in the active state are proportional within our theory.
To be specific, force generation in the cortex is proportional to the gradient of myosin density. However, local inhomogeneity of active myosin molecules can also lead to forces. In the experiments, one measure the amount of myosin and their specific state remains unknown [@mayer2010]. We assume the force generation is proportional to the amount of active myosin, and thus, proportional to ATP concentration. There can also be a rearrangement of the polarity of the actin filaments [@daniel2016] or disassembly of actin filaments due to some enzyme [@annececile2012; @haviv2008] that we do not consider in this work for simplicity.
Thus, our theoretical model consists of the following: the filament-like molecules along with the cross-linkers lead to a static inhomogeneous density and there is a time-dependent density fluctuation due to turnover dynamics as well as dynamics of other proteins (that we take as passive particles) in the cortex (Fig. \[schematicfluid\]. ATP concentration controls the fraction of myosin molecules being in the active state, which is proportional to force generation in the system. The cell boundary as well as cytoplasm impart a friction force on the cortex.
Theory
======
Including all the considerations discussed in the previous section, we write the total density of the system as $\rho({\bf r},t)=\rho_0+\delta m({\bf r})+\delta\rho({\bf r},t)$ at position ${{\bf{r}}}$ and time $t$ where $\rho_0$ is a uniform density, $\delta m({\bf r})$, the static inhomogeneous density modulation and $\delta\rho({\bf r},t)$ is the fluctuating part of density. Note that $\rho_0$ has contributions from active and inactive particles as well as the actin filaments. $\delta m({{\bf{r}}})$ has contribution only due to actin filaments and this is small due to the dynamic nature of the filaments. There is some evidence that myosin II acts as actin depolymerization factor when the activity of myosin molecules exceed a certain (unknown) threshold [@haviv2008; @annececile2012]; this may also contribute to reducing $\delta m({{\bf{r}}})$, that leads to fluidization of the assembly. However, we expect this effect to be small and there should always be a static inhomogeneous component of the density in the cell in the time-scale of our interest. Through a detailed microscopic calculation, Wang [*et al*]{} [@wangJCP2011] have also shown that one must keep the inhomogeneous static density component for the description of such systems. Then, we have the continuity equation for density as $$\label{continuity}
{\frac}{{\partial}\rho({\bf r},t)}{{\partial}t}=-\nabla\cdot[\rho({\bf r},t){\bf v}({\bf r},t)]$$ where ${\bf v}({\bf r},t)$ is the velocity of a hydrodynamic volume element of the cortical constituents. The continuity equation for the momentum density, $\rho({{\bf{r}}},t){\bf v}({{\bf{r}}},t)$, is $$\begin{aligned}
\label{ns}
{\frac}{{\partial}(\rho{\bf v})}{{\partial}t}+\nabla\cdot(\rho{\bf v}{\bf v})=&\eta\nabla^2{\bf v}+(\zeta+\eta/3)\nabla\nabla\cdot{\bf v} \nonumber\\
&-\gamma {\bf v}-\rho\nabla{\frac}{\delta F^U}{\delta\rho}+{\bf f},\end{aligned}$$ where $\eta$ and $\zeta$ are shear and bulk viscosities of the fluid, ${\bf f}$, a non-conserving Gaussian white noise, with zero mean and the statistics $\langle{\bf f}_i({{\bf{r}}},t){\bf f}_j({{\bf{r}}}',t')\rangle=-2k_BT[\eta \nabla^2 \mathbb{I}+(\zeta+\eta/3)\nabla \nabla]\delta({{\bf{r}}}-{{\bf{r}}}')\delta(t-t')\delta_{ij}$ where $k_B$ is the Boltzmann constant, $T$, the temperature, $\mathbb{I}$, the identity matrix and $\delta_{ij}$, the Kronecker delta-function. $\gamma$ is a friction coefficient that encodes the friction force exerted by both the cytoplasm as well as the cell membrane on the cortex. $F^U$ is a suitably defined energy functional that includes the effect of matrix density of the system. Note that we use the term ‘matrix’ to refer to the static inhomogeneous density in the system.
Active systems are inherently out of equilibrium; within MCT for active systems, it has been shown that this nonequilibrium nature of the system is manifested through an evolving effective temperature, $T_{eff}(t)$, defined through a generalized fluctuation-dissipation relation as a function of time, $t$ [@noneqmct]. $T_{eff}(t)$ is same as the equilibrium temperature at a very short time and dynamically evolves to a larger value, determined by the parameters of activity, at long times; note that $t$ refers to the time difference of the two-time correlation functions [@noneqmct]. We find that the dynamics of the system can be understood through the long-time limit of $T_{eff}(t\to\infty)\equiv T_{eff}$. A number of other studies have also shown the role of effective temperature in the dynamics of active systems [@shen2004; @lu2006; @wang2011; @wang2013; @loi2011; @loi2011a; @cugliandolo2011; @suma2014; @benisaac2015]. We propose to include activity in the static description of the system and this provides an experimentally accessible parameter for direct comparison with experiments on such systems. We consider a system of $N$ active particles with the inter-particle interaction potential as $\mathcal{V}({\bf r}_1,\ldots,{\bf r}_N)$: $$\mathcal{V}={\frac}{1}{2}\sum_{j,l}v(|{\bf r}_j(t)-{\bf r}_l(t)|)$$ where ${\bf r}_j(t)$ is the position of $j$th particle at time $t$. The density at ${\bf r}$ at time $t$ is defined as $\rho({\bf r},t)=\sum_{j=1}^N\delta({\bf r}-{\bf r}_j(t))$, which we can write in Fourier space as $$\rho_{\bf k}(t)=\sum_{j=1}^N e^{i{\bf k}\cdot{\bf r}_j(t)}$$ where ${\bf k}$ is the wave vector. Following Ref. [@zaccarelli2002] we write $$\begin{aligned}
\ddot{\rho}_{\bf k}(t)=&-\sum_j({\bf k}\cdot\dot{\bf r}_j(t))^2e^{i{\bf k}\cdot{\bf r}_j(t)} \nonumber\\
&-{\frac}{1}{V}\sum_{{\bf k}'}v_{{\bf k}'}({\bf k}\cdot{\bf k}')\rho_{{\bf k}-{\bf k}'}(t)\rho_{{\bf k}'}(t)\end{aligned}$$ where we have set the mass of the particles to unity and $v_{\bf k}$ is the Fourier components of the interaction potential. Following Zwanzig [@zwanzigbook], we write this equation as $$\label{zwanzigform}
\ddot{\rho}_{\bf k}(t)+\Omega_{\bf k}\rho_{\bf k}(t)=\mathcal{F}_{\bf k}(t).$$ The frequency term $\Omega_{\bf k}$ that minimizes the residual forces $\mathcal{F}_{\bf k}(t)$ is given as $$\label{omega}
\Omega_{\bf k}={\frac}{k^2}{S_k^A}k_B(T+T_{e})$$ where we have used the relation $\langle \dot{r}_{i\alpha}\dot{r}_{j\beta}\rangle \sim k_B(T+T_e)\delta_{ij}\delta_{\alpha\beta}$ for an active fluid at an effective temperature $T_{eff}=T+T_e$, where $T_e$ is the excess contribution coming from activity alone and $S_k^A$ is the equal-time two-point correlation function for such a system $$S_k^A={\frac}{1}{N}\langle \rho_k(t)\rho_{-k}(t)\rangle.$$ Then we obtain the residual forces from Eq. (\[zwanzigform\]) as $$\begin{aligned}
\label{residual_f}
\mathcal{F}_{\bf k}(t)=\Omega_k \rho_k(t)-\sum_j({\bf k}\cdot\dot{\bf r}_j(t))^2e^{i{\bf k}\cdot{\bf r}_j(t)} \nonumber\\
-{\frac}{1}{V}\sum_{{\bf k}'}v_{{\bf k}'}({\bf k}\cdot{\bf k}')\rho_{{\bf k}-{\bf k}'}(t)\rho_{{\bf k}'}(t).\end{aligned}$$ The fact that the residual forces are minimized with respect to $\Omega_k$ implies an orthogonality condition $$\langle \rho_{-k}(t)\mathcal{F}_k(t)\rangle=0.$$ Using the detailed form of residual forces from Eq. (\[residual\_f\]), we obtain the equation for $c_k^A$, which is analogous to the direct correlation function of a passive system, $$\begin{aligned}
\label{active_c}
c_k^A=-{\frac}{\beta^A}{k^2N^2S_k^A}\sum_{{\bf k}'}v_{{\bf k}'}({\bf k}\cdot{\bf k}')\langle\rho_{-{\bf k}}(t)\rho_{{\bf k}-{\bf k}'}(t)\rho_{{\bf k}'}(t)\rangle,\end{aligned}$$ where $c_k^A$ is defined through the relation $c_k^A=(1-1/S_k^A)/\rho_0$ with $\rho_0$ being the average density of the system and $$\beta^A={\frac}{1}{k_B(T+T_e)}\simeq {\frac}{1}{k_BT}(1-T_e/T)\simeq {\frac}{1}{k_BT}f(A),$$ where we have written the contribution from activity in $f(A)$ denoting activity as $A$. Note that $\beta^A$, in principle, could depend on wave vector, however, we expect $T_e$ at all wave vectors to be the same as the behavior of the correlation and response functions at all wave vectors within MCT is similar [@noneqmct] though we do not have an [*a priori*]{} justification for this. If the activity is zero, we have $f(A)=1$ and Eq. (\[active\_c\]) gives $c_k$, the direct correlation function of the corresponding passive system. Thus, we can write $$\label{active_c2}
c_k^A=f(A)c_k.$$ An intuitive way to understand Eq. (\[active\_c2\]) is as follows: $c_k$ is related to the inter-atomic interaction potential. Consider one of the particles in the system, this particle interacts with an effective potential obtained from integrating out the other particles. Activity introduces a random motion of the particle, considering the activity of different particles uncorrelated. Since the state of myosin locally depends on the availability of ATP, such an assumption is justified. Thus, activity reduces correlation in the particle motion and $f(A)$ is a monotonically decreasing function of $A$. When the activity is small, we can write $f(A)=e^{-A}$. We note here that it is not clear what should be the actual form of $c^A(r)$. Even in an equilibrium fluid, it is not clear what should enter as $c(r)$ in the Ramakrishnan-Yussouff free energy functional [@ramakrishnan1979] in the dynamical mean-field theory for density waves. Moreover, Eq. (\[active\_c\]), as well as its equilibrium counterpart, is only valid within the random phase approximation [@zaccarelli2002; @saroj2015]. However, this difficulty is not important in the context of the present work as we do not use $c_k^A$ directly within the theory. As we are interested in an assembly of active and passive particles embedded in a static inhomogeneous density background (that we call the matrix), $A$ is directly proportional to the amount of active particles or the amount of ATP [@benisaac2015]. Similar use of effective parameters to describe an active network material, within the random first-order transition theory framework has been demonstrated elsewhere [@wangJCP2011; @wang2012; @wang2013].
The velocity field in a biological system is small and we ignore the second term in Eq. (\[ns\]). We linearize Eqs. (\[continuity\]) and (\[ns\]), take the divergence of the second and eliminate velocity using the first to obtain $$\begin{aligned}
{\frac}{{\partial}\delta\rho({\bf r},t)}{{\partial}t}&=D_L\nabla^2{\frac}{{\partial}\delta\rho({\bf r},t)}{{\partial}t}-{\frac}{\gamma}{\rho_0}{\frac}{{\partial}\delta\rho({\bf r},t)}{{\partial}t} \nonumber\\
&+\nabla\cdot\left[\rho\nabla{\frac}{\delta F^U[\delta\rho({\bf r},t),\delta m]}{\delta\rho({\bf r},t)}\right]-\nabla\cdot{\bf f}\end{aligned}$$ where $D_L=(\zeta+4\eta/3)/\rho_0$. The energy functional $F^U$ for the active system is $$\begin{aligned}
\beta F^U&=\int_{\bf r}\left\{\rho({\bf r},t)\ln{\frac}{\rho({\bf r},t)}{\rho_0}-[\rho({\bf r},t)-\rho_0]\right\} \nonumber\\
& -{\frac}{1}{2}\int_{{\bf r},{\bf r}'}c^A(|{\bf r}-{\bf r}'|)[\rho({\bf r},t)-\rho_0][\rho({\bf r}',t)-\rho_0] \nonumber \\
&+\beta\int_{\bf r}U({\bf r})\rho({\bf r},t)\end{aligned}$$ where $\int_{\bf r}\equiv \int {{\mathrm{d}}{\bf r}}$. The first part is an ideal gas contribution and the second part contains the contribution due to interaction. It is not clear what should be the detailed form of $c^A({{\bf{r}}})$, however, this is not important for the purpose of the present work [^1]. The contribution due to the inhomogeneous density modulation, $\delta m({{\bf{r}}})$, is encoded through a potential (see Appendix for details) $U({\bf r})$, which makes it straightforward to extend the theory for systems under external force, for example, a cell attached to an elastomeric membrane that can be stretched as in Ref. [@jungbauer2008].
To characterize the dynamics of a dense system, we need to look at the two-point correlation functions [@hansenmcdonald]. This is a directly accessible quantity in experiments and all the transport coefficients can be calculated from this function through the application of statistical mechanics. We define the two-point correlation function $S_{k}(t)=\langle\delta\rho_{k}(0)\delta\rho_{-{k}}(t)\rangle$ and the normalized two-point correlation function $\phi_{k}(t)=S_{k}(t)/S_{k}(t=0)$. Then the equation for $\phi_k(t)$ becomes (see Appendix for details) $$\begin{aligned}
{\frac}{{\partial}^2\phi_k(t)}{{\partial}t^2}&+\Gamma_k{\frac}{{\partial}\phi_k(t)}{{\partial}t}+{\frac}{k_BTk^2}{S_k^{A}}\phi_k(t) \nonumber\\
&+\int_0^t\mathcal{M}_k(t-t'){\frac}{{\partial}\phi_k(t')}{{\partial}t'}{\mathrm{d}}t'=0\end{aligned}$$ where $\Gamma_k=D_Lk^2+\gamma/\rho_0$ and $\mathcal{M}_k(t)$ is the memory kernel given as $$\begin{aligned}
\label{memoryk}
\mathcal{M}_k(t)&={\frac}{k_BT\rho_0}{2k^2}\int_{\bf q}[{\bf k}\cdot{\bf q}c_q^A+{\bf k}\cdot({\bf k}-{\bf q})c_{k-q}^A]^2S_{k-q}(t)S_q(t) \nonumber\\
&+{\frac}{k_BT\rho_0}{k^2}\int_{\bf q}\left[{\bf k}\cdot{\bf q}c_q^A+{\frac}{{\bf k}\cdot({\bf k}-{\bf q})}{\rho_0}\right]^2S_{k-q}^bS_q(t)\end{aligned}$$ where $S_k^b=\langle\delta m_k\delta m_{-k}\rangle$. Activity in our model enters through $c_k^A$ within the memory kernel and $S_k^A$ which is defined as $1/(1-\rho_0 c_k^A)$.
The complexity of the full wavevector dependent theory hinders its applicability to a biological system. To render simpler form we take a schematic approximation suppressing the wavevector dependence leaving a single-mode, however, retaining the essential mechanism of the theory [@leutheusser1984; @kirkpatrick1985; @brader2009]. Within this approximation, we write the theory for one wave vector $k_{max}$ that corresponds the first maximum of structure factor [^2]. Ignoring the acceleration term, which is not important at long times and for systems of our interest, we obtain the equation of motion for the normalized two-time correlation function, $\phi(t)\equiv \phi_{k=k_{max}}(t)$, in this limit as $$\label{schematicmct}
(1+{\tilde{\gamma}}){\frac}{{\partial}\phi(t)}{{\partial}t}+\Omega\phi(t)+\int_0^t\mathcal{M}(t-t'){\frac}{{\partial}\phi(t')}{{\partial}t'}{\mathrm{d}}t'=0$$ where $\tilde{\gamma}=\gamma/\rho_0D_Lk_{max}^2$ and we have ignored the activity dependence in the frequency term $\Omega$. The memory kernel becomes $$\label{schematicmemory}
\mathcal{M}(t)\simeq\left(\lambda_1^0-{\alpha}\sqrt{{\lambda}_1^0}\right)\phi(t)+\left(\lambda_2^0-{\alpha}\sqrt{{\lambda}_2^0}\right)\phi(t)^2.$$ The main contribution in the dynamics comes from the parts inside the integrals in Eq. (\[memoryk\]). Therefore, ${\lambda}_1^0$ is proportional to the square of the matrix density (through $S_{k_{max}}^b$) and both ${\lambda}_1^0$ and ${\lambda}_2^0$ are proportional to average density $\rho_0$ and inversely proportional to temperature (see Appendix, Eq. (A8), for their detailed forms). The parameter ${\alpha}$ contains the information of activity of the system and ${\alpha}\propto A $ is proportional to the amount of active myosin or ATP concentration. ${\lambda}_1^0$ contains the information of matrix density as well as the interaction of the fluid with the matrix and ${\lambda}_2^0$ encodes the strength of interaction within the fluid. Both ${\lambda}_1^0$ and ${\lambda}_2^0$ increases as the density of the fluid increases or temperature decreases and only ${\lambda}_1^0$ increases as the matrix density increases.
![We show the decay of the correlation function $\phi(t)$ as a function of time obtained through the numerical solution of Eqs. (\[schematicmct\]) and (\[schematicmemory\]). $\phi(t)$ shows a fast decay to a plateau at short times and then a much slower decay to zero at longer times. Activity, ${\alpha}$, makes the decay of $\phi(t)$ faster. We have used the parameters as $\tilde{\gamma}=1$, ${\lambda}_1^0=0.28$ and ${\lambda}_2^0=3.417$. [**Inset:**]{} We compare our theory with the experimental data of Ref. [@almonacid2015] for the dynamics of cytoplasm in mouse oocytes. The data for $\phi(t)$ is taken from Fig. 4(b) of Ref. [@almonacid2015] and solid lines are fit of our theory showing that activity makes the dynamics faster as $\phi(t)$ decays quicker at larger ${\alpha}$. The parameters ${\lambda}_1^0=0.8$ and ${\lambda}_2^0=1.98$ are same for both the curves.[]{data-label="active_corrfn"}](corrfn_comp.eps){width="\columnwidth"}
Results
=======
The theory, Eqs. (\[schematicmct\]) and (\[schematicmemory\]) for a passive system when ${\alpha}=0$ is well-understood in the literature [@sing1998; @goetze2000; @goetzebook; @das2004]. The correlation function $\phi(t)$ decays to zero quite fast at small density when ${\lambda}_1^0$ and ${\lambda}_2^0$ are small and develops a two-step relaxation scenario at higher density; it first decays to a plateau and then to zero from the plateau at long times. At even larger density, $\phi(t)$ does not decay to zero anymore, showing an ergodicity breaking transition: a discontinuous transition at ${\lambda}_2^0=4$ when ${\lambda}_1^0=0$ and a continuous transition at ${\lambda}_1^0=1$ when ${\lambda}_2^0=0$ [@goetzebook; @das2004]. It is well-known that this transition shifts to higher density (larger ${\lambda}_1^0$ or ${\lambda}_2^0$) in active systems of self-propelled particles [@berthier2013; @mandal2016; @flenner2016; @noneqmct; @viscmct], our theory also has similar feature as can be easily seen from Eq. (\[schematicmemory\]). The ergodicity breaking transition is a failure of the theory whose origin is not yet well-understood [@rfimmct] and all the predictions of the theory break down beyond this point. In the context of the systems of our interest, the transition by itself is not important as living systems can not, in general, go to a state with very high density or low temperature. We expect our theory to work within the regime where biological systems operate.
We set $\tilde{\gamma}$ to unity and show the decay of $\phi(t)$ for ${\lambda}_1^0=0.28$ and ${\lambda}_2^0=3.417$ for different values of the activity parameter ${\alpha}$ in Fig. \[active\_corrfn\]. The correlation function decays faster as the activity becomes stronger, that is at higher values of ${\alpha}$. Ref. [@almonacid2015] looked at the cytoplasmic dynamics for positioning the nucleus in mouse oocytes. We have obtained the data for correlation function, as presented in Fig. 4(b) in [@almonacid2015] and fit our theory with the data with ${\lambda}_1^0=0.8$ and ${\lambda}_2^0=1.98$ (same for both curves) and ${\alpha}=0.01$ and $0.036$ for the two sets of data as shown in the inset of Fig. \[active\_corrfn\]. Note that there is a difference in the scale of measuring times in the theory and the experiment; we have rescaled the time in our calculation by a factor of $0.01$ to match with the experimental data. The important point in this fit is that higher activity fluidizes the cytoplasm and the correlation function decays faster.
![ (a) Different systems are characterized by different values of $({\lambda}_1^0,{\lambda}_2^0)$ leading to distinct values of $\ell_0$. We show the behavior of $\ell$ for many such systems as a function of ${\alpha}$. (b) Since there is a characteristic length scale, $\ell_0$, independent of activity, we expect a scaling for $\ell$ as $\ell=\ell_0^\theta f(\alpha\ell_0^{1/\nu})$ and obtain data collapse for $\theta=1.0$ and $\nu=0.95$ implying $\ell\sim 1/{\alpha}^{0.95}$ in a regime where activity dominates the dynamics. [**Inset:**]{} We obtain the data of length scale from Ref. [@sanchez2012] (see text). Our theory shows $\ell\sim 1/{\alpha}$; therefore we fit the data with $f(x)=a+b/x$. We find excellent agreement between theory and experiment with $a=0.51$ and $b=40.13$.[]{data-label="visclength"}](lengthscale_plot.eps){width="8.6cm"}
We now show how activity controls the hydrodynamic length scale $\ell$ of the system, constituting the main result of this work. We use Kubo formula to obtain the viscosity from the correlation function. The renormalized viscosity of the system is given by $ \eta_R=\int_0^\infty\phi(t'){\mathrm{d}}t'$ [@kubo1966; @hansenmcdonald]. This relation is valid for systems in equilibrium. However, considering that the system is in a state where linear-response theory is valid, we can still use Kubo formula for the viscosity. The hydrodynamic length scale, which determines the rigidity of the system is defined, following Refs. [@mayer2010; @sundar2014], as $\ell=\sqrt{\eta_R/\tilde{\gamma}}\sim \sqrt{\eta_R}$ since we have set $\tilde{\gamma}$ to unity. For a passive system, viscosity, and hence the hydrodynamic length scale, is determined by $({\lambda}_1^0,{\lambda}_2^0)$ alone. However, in an active system viscosity depends on activity; as activity sets in, the renormalized viscosity is different from its passive value.
Different biological systems are characterized by distinct values of $({\lambda}_1^0,{\lambda}_2^0)$ as the number of proteins and molecules attached to the focal adhesion points are different and the densities of the fluid components of the systems also vary. Correspondingly, they can be characterized by distinct $\ell_0 =\sqrt{\eta_R(\alpha=0)}$. We choose a particular set of values for $({\lambda}_1^0,{\lambda}_2^0)$ and denote the system with the corresponding passive length scale $\ell_0$. We show the behavior of $\ell$ as a function of activity for different $\ell_0$ in Fig. \[visclength\](a). $\ell$ saturates at small ${\alpha}$ to a value corresponding to its passive value $\ell_0$ and decreases as ${\alpha}$ increases. Thus, biological systems can actively self-regulate their hydrodynamic length scales, hence, rigidity by changing their activity.
Since there is a characteristic length scale $\ell_0$, independent of activity for each system, we expect a scaling relation [@amitbook] for $\ell$ as $\ell=\ell_0^\theta f(\alpha\ell_0^{1/\nu})$ and obtain scaling collapse for $\theta=1$ and $\nu=0.95$ as shown in Fig. \[visclength\](b). From this we obtain $$\ell\sim 1/{\alpha}$$ in the regime where activity dominates the dynamics. This implies that irrespective of the nature of the system, the viscous length scale $\ell$ follows a simple algebraic law as a function of the activity parameter $\alpha$. As we define $\ell\sim\sqrt{\eta_R}$, the viscosity behaves as $\sim 1/{\alpha}^{1.9}$ as a function of activity in the activity dominated regime.
For a comparison with experiment, we look at the data of velocity correlation function, $\langle V(R)V(0)\rangle$, as a function of distance $R$ as presented in the inset of Fig. 2(c) in [@sanchez2012]. At a certain $R$, $\langle V(R)V(0)\rangle$ changes because of activity. The correlation in a system is controlled by certain length scales. We can assume this change in correlation at a fixed $R=R_0$ varies exponentially: $[\langle V(R)V(0)\rangle]_{R_0} \sim e^{1/\ell}$; this gives $\ell$ up to a constant factor. We have extracted $\ell$ in this way from [@sanchez2012] for $R_0=12$ (the first data point in each curve). The data are presented in terms of ATP concentration; within our theory, this is proportional to the active molecules, designated by the parameter ${\alpha}$. A fit of this data with our theory, presented in the inset of Fig. \[visclength\](b) shows excellent agreement. Ref. [@humphrey2002] has shown that myosin molecules in the active state fluidizes an in vitro actomyosin gel and reduces the local shear relaxation time, [@parry2014] has shown that bacterial cytoplasm becomes solid-like in the dormant state and fluidizes when metabolically active and [@chan2015] has shown that myosin activity softens cells in suspension. Our theory rationalizes these findings in terms of the activity of constituent molecules.
Discussion {#disc}
==========
Our theory is within the framework of mode-coupling theory (MCT) for dense super-cooled systems [@das2004; @goetzebook]. MCT provides a mechanism of the rapid increase of viscosity and relaxation time of systems as temperature decreases or density increases. We have extended this theory to include activity generated by some active motor proteins like myosin or kinesin. Considering detailed microscopic dynamics, Ref. [@gladrow2016] shows that mode-coupling naturally emerges in the dynamics of active networks. MCT shows an ergodicity-breaking transition above certain values of the parameters and the theory fails to describe the system beyond this point [@giulioreview; @goetzebook; @das2004]. This transition is not important for biological systems as living systems cannot become extremely dense; we expect the theory to be valid in the regime where most biological systems operate. Our theory should be viewed as a description, starting from the molecular level, of the mechanism how motor activity changes the dynamics at the macroscopic length scale. Our theory is similar in spirit to the microscopic theory developed by Wolynes [*et al*]{} for such systems [@wangJCP2011; @wang2012; @wang2013].
We have implemented activity in the static description of the system, where activity parameter $\alpha$ is proportional to the ATP concentration. Thus, our theory allows a direct experimentally accessible parameter. The principal result of this work is the behavior of the viscous length scale $\ell$, which is an important quantity as it controls the mechanical properties of a system, as a function of activity; we find that $\ell$ is roughly inversely proportional to the amount of active myosin, generalizing recent experimental findings [@humphrey2002; @parry2014; @almonacid2015; @chan2015; @sanchez2012] on such systems. Ref. [@zhou2009] shows that viscosity of a cell increases with ATP depletion, our theory is in qualitative agreement with this finding. Remarkably, we find universality in the behavior of $\ell$, which varies roughly inversely with $a$ irrespective of the nature of the system. This is an important result readily testable in experiments and simulations. Comparison of our theory with the existing experimental results show good agreement (Fig. 3b). The increased viscosity as well as $\ell^2$ in ATP-depleted nucleoli in [*Xenopus laevis*]{} oocytes [@brangwynne2011] are also in qualitative agreement with our predictions.
The cellular cortex is a complex system, where a large number of active processes control the large-scale behavior. We have considered a simplified picture ignoring a number of such processes. For example, we have neglected the chiral nature of actin-filaments [@sundar2014; @sebastian2013] in our theory, however, we believe, including chirality within the theory is not going to affect the main conclusions of this work. We have treated the effects of cross-linking from non-force-generating molecules, like the $\alpha$-actinin, and that of myosin molecules to be same. It is important to understand how ATP concentration affects the cross-linking that we have modeled through the static inhomogeneous density $\delta m$. Detailed molecular modeling along the lines of Ref. [@william2017] should unravel this aspect.
There may as well have a number of other effects like the organization of the polarity of actin filaments [@daniel2016] or change in the actin polymerization dynamics due to some enzyme that are also going to affect $\ell$. We have not included such dynamics in this work, however, we expect these effects to give higher order corrections in the behavior of $\ell$. The effect of myosin activity where myosin acts as actin depolymerization factor [@haviv2008] when the activity of myosin crosses a certain threshold value is going to fluidize the system. Although we did not look at this effect in detail, this is already included in the coarse-grained description of reducing correlation within our theory. As the length scale directly gives the rigidity of the system, a study along the lines of [@chan2015] or [@humphrey2002] as a function of active myosin molecules of the system provides a straightforward way to further test our theory.
We thank Ludovic Berthier, Anagha Datar, Stephan Grill, Frank J[ü]{}licher, Sriram Ramaswamy, Nir S. Gov, Grzegorz Szamel, Daniel Riveline and Jacques Prost for many important and enlightening discussions and Koshland Foundation for funding through a fellowship.
Some details of the MCT calculation
===================================
Here we first show how the energy functional $F^U$, and hence the force density, can be written in terms of the static inhomogeneous density $\delta m({\bf r})$. The density fluctuates around the static density $m({\bf r})=\rho_0+\delta m({\bf r})$. Minimizing $F^U$ with respect to $m({\bf r})$ and replacing $U({\bf r})$ in terms of $m({\bf r})$, we obtain the force density as $$\begin{aligned}
\rho\nabla{\frac}{\delta \beta F^U}{\delta\rho({\bf r},t)}=&\nabla\int_{{\bf r}'}[\delta({\bf r}-{\bf r}')-\rho_0c^A(|{\bf r}-{\bf r}'|)]\delta\rho({\bf r}',t)\nonumber\\
-&\nabla\int_{{\bf r}'}\delta m({\bf r})c^A(|{\bf r}-{\bf r}'|)\delta\rho({\bf r}',t) \nonumber\\
-&{\frac}{\nabla m({\bf r})}{m({\bf r})}\int_{{\bf r}'}[\delta({\bf r}-{\bf r}')-m({\bf r})c^A(|{\bf r}-{\bf r}'|)]\delta\rho({\bf r}',t)\nonumber\\
-&\delta\rho({\bf r},t)\nabla\int_{{\bf r}'}c^A(|{\bf r}-{\bf r}'|)\delta\rho({\bf r}',t).\end{aligned}$$
Defining Fourier transform of different quantities as $\delta\rho({\bf r},t)=\int_{\bf k}\delta\rho_k(t)e^{i{\bf k}\cdot{\bf r}}$, where we have used the notation $\int_{\bf k}\equiv\int {\frac}{{\mathrm{d}}^dk}{(2\pi)^d}$ in $d$-dimension, we can write the equation of motion for the density fluctuation in Fourier space as $$\begin{aligned}
\label{eqofm1}
\left[ {\frac}{{\partial}^2}{{\partial}t^2}+\Gamma_k{\frac}{{\partial}}{{\partial}t}+{\frac}{k_BTk^2}{S_k^{A}}\right]&\delta\rho_k(t)=-i{\bf k}\cdot[{\bf F}_{\bf k}+{\bf f}_{\bf k}] \nonumber\\
& \hspace{-1cm}=-i{\bf k}\cdot[{\bf F}_{\bf k}^{m\rho}+{\bf F}_{\bf k}^{\rho\rho}+{\bf f_k}],
\end{aligned}$$ where $\Gamma_k=D_Lk^2+\gamma/\rho_0$ and $S_k^{A}=1/(1-\rho_0c^A_k)$. The Force densities are given as $$\begin{aligned}
\label{sm:eqofm2}
{\bf F}_{\bf k}^{m\rho}(t)&=ik_BT\int_{\bf q}[{\bf k}c^A_{\bf q}+({\bf k}-{\bf q})/\rho_0S_{\bf q}^{A}]\delta m_{{\bf k}-{\bf q}}\delta\rho_{\bf q}(t) \\
{\bf F}_{\bf k}^{\rho\rho}(t)&={\frac}{i}{2}k_BT\int_{\bf q}[{\bf q}c^A_{\bf q}+({\bf k}-{\bf q})c^A_{{\bf k}-{\bf q}}]\delta\rho_{\bf q}(t)\delta\rho_{{\bf k}-{\bf q}}(t) \label{sm:eqofm3}.\end{aligned}$$ In this work, we have included activity in $c_k^A$ and $S_k^A$ and ${\bf f}_{\bf k}$ is the bare noise, in contrast to the existing MCT theories on active systems of self-propelled particles [@szamel2016; @noneqmct]. Using the standard prescription of MCT through hydrodynamic approach [@kawasaki2002; @saroj2011] to extract the excess damping due to interactions in terms of a memory kernel given as $$\mathcal{M}_{\bf k}(t)={\frac}{1}{k_BTV}\langle {\bf F}_{\bf k}(0)\cdot{\bf F}_{-{\bf k}}(t)\rangle,$$ where $V$ is the system volume and ${\bf F}_{\bf k}(t)$ is given by Eqs. (\[sm:eqofm2\]) and (\[sm:eqofm3\]). The two-point correlation function is defined as $S_{k}(t)=\langle\delta\rho_{k}(0)\delta\rho_{-{k}}(t)\rangle$. The normalized two-point correlation function is $\phi_{k}(t)=S_{k}(t)/S_{k}(t=0)$ and we obtain the mode-coupling equation as $$\begin{aligned}
\label{sm:mcteq}
{\frac}{{\partial}^2\phi_k(t)}{{\partial}t^2}&+\Gamma_k{\frac}{{\partial}\phi_k(t)}{{\partial}t}+{\frac}{k_BTk^2}{S_k^{A}}\phi_k(t) \nonumber\\
&+\int_0^t \mathcal{M}_k(t-t'){\frac}{{\partial}\phi_k(t')}{{\partial}t'}{\mathrm{d}}t'=0\end{aligned}$$ where $\mathcal{M}_k(t)$ is the memory kernel given as $$\begin{aligned}
\label{sm:memoryk_SM}
\mathcal{M}_k(t)&={\frac}{k_BT\rho_0}{2k^2}\int_{\bf q}[{\bf k}\cdot{\bf q}c_q^A+{\bf k}\cdot({\bf k}-{\bf q})c_{k-q}^A]^2S_{k-q}(t)S_q(t) \nonumber\\
&+{\frac}{k_BT\rho_0}{k^2}\int_{\bf q}\left[{\bf k}\cdot{\bf q}c_q^A+{\frac}{{\bf k}\cdot({\bf k}-{\bf q})}{\rho_0}\right]^2S_{k-q}^bS_q(t)\end{aligned}$$ with $S_k^b=\langle\delta m_k\delta m_{-k}\rangle$. The theory in this form is too complex to be applicable to the systems of our interest as there are too many unknown parameters. Therefore, we take a schematic approximation, which amounts to writing the theory for only one wave vector, $k_{max}$. This approximation provides a simpler form for better understanding and applicability, keeping the key mechanism of the theory intact [@das2004; @goetzebook; @brader2009]. We divide Eq. (\[sm:mcteq\]) by $D_L k^2$ and write $\tilde{\gamma}=\gamma/(\rho_0D_L k^2)$. We expand $c_k^A\sim c_k(1-A)$ (Eq. \[active\_c2\]) and ignore the terms second order in $A$ while completing the squares in Eq. (\[sm:memoryk\_SM\]). The first part of $\mathcal{M}_k(t)$ contains two factors of $\phi(t)\equiv \phi_{k_{max}}(t)$ and the second part contains one $\phi(t)$. The time-independent parts in Eq. (\[sm:memoryk\_SM\]) are defined as ${\lambda}_2^0$ and ${\lambda}_1^0$ respectively: $$\begin{aligned}
{\lambda}_2^0 &\equiv\left[{\frac}{k_BT\rho_0}{2D_Lk^4}\int_{\bf q}[{\bf k}\cdot{\bf q}c_q^A+{\bf k}\cdot({\bf k}-{\bf q})c_{k-q}^A]^2S_{k-q}S_q\right]_{k_{max}}\nonumber\\
{\lambda}_1^0 &\equiv \left[{\frac}{k_BT\rho_0}{D_Lk^4}\int_{\bf q}\left[{\bf k}\cdot{\bf q}c_q^A+{\frac}{{\bf k}\cdot({\bf k}-{\bf q})}{\rho_0}\right]^2S_{k-q}^bS_q\right]_{k_{max}},\end{aligned}$$ where $S_k\equiv S_k(t=0)$ and $[\ldots]_{k_{max}}$ implies that we need to evaluate this term at this wave vector. Then the linear order terms in $A$ in the first and second parts of the memory kernel are proportional to ${\alpha}\sqrt{{\lambda}_2^0}$ and ${\alpha}\sqrt{{\lambda}_1^0}$ respectively where ${\alpha}$ is proportional to $A$. Note that the main contributions in the dynamics come from the terms under the integrals in Eq. (\[sm:memoryk\_SM\]) and not from the prefactors. Thus, we obtain the memory kernel under the schematic approximation as $$\label{sm:schematicmemory}
\mathcal{M}(t)\simeq({\lambda}_1^0-{\alpha}\sqrt{{\lambda}_1^0})\phi(t)+({\lambda}_2^0-{\alpha}\sqrt{{\lambda}_2^0})\phi^2(t)$$ with the mode coupling equation for $\phi(t)$ obtained as $$\label{sm:schematicmct}
(1+{\tilde{\gamma}}){\frac}{{\partial}\phi(t)}{{\partial}t}+\Omega\phi(t)+\int_0^t\mathcal{M}(t-t'){\frac}{{\partial}\phi(t')}{{\partial}t'}{\mathrm{d}}t'=0$$ where we have neglected the acceleration term in Eq. (\[sm:mcteq\]). $\Omega\equiv k_BT/D_L S_k^A$ and we neglect activity dependence in $\Omega$ as it doesn’t affect the dynamics in a strong way. We need to solve Eq. (\[sm:schematicmct\]) along with Eq. (\[sm:schematicmemory\]) numerically where we have set $\Omega$ and $\tilde{\gamma}$ to unity redefining the time scales. This is the reason why the time-scales in our theory and that in Ref. [@almonacid2015] are different and we need to rescale the time-scale in our calculation for comparison with the experimental data.
[^1]: Note that even in a passive system it is not clear what should be the correct form of $c({{\bf{r}}})$ (see Appendix).
[^2]: Within MCT, the relaxation dynamics at all wave vectors are similar [@das2004], therefore, writing the theory for any other wave vector is equivalent.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A new physics scenario shows that four-fermion operators have a strong-coupling UV fixed point, where composite fermions $F$ (bosons $\Pi$) form as bound states of three (two) SM elementary fermions and they couple to their constituents via effective contact interactions at the composite scale $\Lambda \approx {\cal O} $(TeV). We present a phenomenological study to investigate such composite particles at the LHC. Using these contact interactions, we compute the production cross sections and decay widths of composite fermions in the context of the relevant experiments at LHC with $pp$ collisions at $\sqrt{s}=13$ TeV and $\sqrt{s}=14$ TeV. In particular, we focus on the resonant channel $pp\rightarrow e^+F \rightarrow e^+e^- qq''$, whose cross section has been recently limited by the CMS Collaboration. By a simple recasting of this result, we obtain a constraint on the model parameters such that composite fermions of mass $m_F$ below 4.25 TeV are excluded for $\Lambda$ = $m_F$. We further compute 5$\sigma$ contour plots of the statistical significance and highlight the region of parameter space where $F$ can manifest using 3 ab$^{-1}$, expected by the High-Luminosity LHC. It turns out that there is a large portion of the parameter space where $F$ can be discovered and that deserve a dedicated investigation. In addition, we also study the composite boson state $\Pi_0$ with the estimation of branching ratios into two quarks (two jets) ${\cal B}(\Pi_0\rightarrow qq'')$ and into two boosted gauge bosons ${\cal B}(\Pi_0\rightarrow \tilde G \tilde G'')$, from which we obtain the branching ratios of composite-fermion decay into an electron and two boosted gauge bosons ${\cal B}(F\rightarrow e\tilde G\tilde G'')$. Moreover we briefly discuss the possible final states of four jets or one jet with two gauge bosons in LHC $pp$ collision.'
author:
- '<span style="font-variant:small-caps;">R. Leonardi</span>'
- '<span style="font-variant:small-caps;">F. Romeo</span>'
- '<span style="font-variant:small-caps;">H. Sun</span>'
- '<span style="font-variant:small-caps;">A. Gurrola</span>'
- '<span style="font-variant:small-caps;">O. Panella</span>'
- '<span style="font-variant:small-caps;">S.-S. Xue</span>'
title: 'Production at LHC of composite particles from strongly interacting elementary fermions via four-fermion operators of Einstein-Cartan type'
---
Introduction
============
The parity-violating gauge symmetries and spontaneous/explicit breaking of these symmetries for the hierarchy pattern of fermion masses have been at the center of a conceptual elaboration that has played a major role in donating to mankind the beauty of the Standard Model (SM) and possible scenarios beyond SM for fundamental particle physics. A simple description is provided on the one hand by the [*composite*]{} Higgs-boson model or the Nambu-Jona-Lasinio (NJL) model [@njl] with effective four-fermion operators, and on the other by the phenomenological model of the [*elementary*]{} Higgs boson [@higgs]. These two models are effectively equivalent for the SM at low energies. After a great experimental effort for many years, using $pp$ collision data at $\sqrt{s} = 7, 8$ TeV at the Large Hadron Collider (LHC), the ATLAS [@ATLASCollaboration] and CMS [@CMSCollaboration] collaborations have shown the first observations of a 125 GeV scalar particle in the search for the SM Higgs boson [@ATLAS; @CMS]. This far-reaching result begins to shed light on this most elusive and fascinating arena.
Recently, in the Run 2 of the upgraded LHC, studies on $\sqrt{s}$ = 13 TeV $pp$ collision data are performed by ATLAS and CMS to search for new (beyond the SM) resonant and/or non-resonant phenomena [@exp1; @exp2; @exp3; @exp4; @exp5]. These studies are continuously pushing up exclusion bounds on the parameter spaces of many possible scenarios beyond SM [@exp6; @exp7; @exp8; @exp9].
Composite-fermion scenarios have offered a possible solution to the hierarchy pattern of fermion masses [@comp1; @cfmass]. In this context [@comp2; @comp3; @comp4; @comp5; @comp6], SM quarks “$q$” and leptons “$\ell$” are assumed to be bound states of some not yet observed fundamental constituents generically referred as [*preons*]{} and to have an internal substructure and heavy excited states $F$ of masses $m_F^*$ that should manifest themselves at the high energy compositeness scale $\Lambda$. Exchanging preons and/or binding quanta of unknown interactions between them result in effective contact interactions of SM fermions and heavy excited states. While different heavy excited states have been considered in literature [@comp7; @comp8; @comp9], in this article we take as a reference the case of a heavy composite Majorana neutrino, $N_l$, for which the interaction lagrangian would be $(g_*/\Lambda)^2\bar q_L\gamma_\mu q_L \Bar N_L\gamma_\mu \ell_L$. Its theoretical studies and numerical analysis have been carefully elaborated in [@LPF2014; @LARFP2016]. Moreover, an experimental analysis of $\sqrt{s}$ = 13 TeV $pp$ collisions at LHC of the process $pp\rightarrow \ell N_l \rightarrow \ell\ell qq$ of the dilepton (dielectrons or dimuons) plus diquark final states has been carried out by the CMS collaboration [@LARFP2017] excluding the existence of $N_l$ for masses up to $4.60$ $(4.70)$ TeV at $95\%$ confidence level, assuming $m_{N_{l}} = \Lambda$.
Motivated by the theoretical inconsistency [@nogotheorem] of SM bilinear Lagrangian of chiral gauged fermions and quantum-gravity natural regularization, as well as by quadrilinear four-fermion operators of Einstein-Cartan type [@xueqgravity], an alternative physics scenario had been proposed [@xue2016_2; @xue2017] on the basis of SM gauge symmetric four-fermion operators of SM left- and right-handed fermions ($\psi_L,\psi_R$) in the charge sector “$Q_i$” and flavor family “$f$” $$\begin{aligned}
\sum_{f=1,2,3}G\, \Big[\bar\psi^{f}_{_L}\psi^{f}_{_R}\bar\psi^{f}_{_R} \psi^{f}_{_L}\Big]_{Q_i=0,-1,2/3,-1/3}.
\label{q1}\end{aligned}$$ These effective operators are attributed to the new physics at the cutoff $\Lambda_{\rm cut}$, and reduce to the NJL-type operator for the top-quark channel. The effective coupling $G$ (\[q1\]) has two fixed points: the weak-coupling infrared (IR) fixed point and the strong-coupling ultraviolet (UV) fixed point. In the scaling domain of IR fixed point of the weak four-fermion coupling $G$ at the electroweak scale $v\approx 239.5$ MeV, effective Operators (\[q1\]) give rise to SM physics with tightly composite Higgs particle via NJL mechanism, and also offer possible solution to the hierarchy pattern of fermion masses [@xue2016_2; @xuemass]. In the scaling domain of UV fixed point of the strong four-fermion coupling $G$ at the composite scale $\Lambda \approx {\cal O} $ (TeV), composite fermions (bosons) form as bound states of three (two) SM elementary fermions and they couple to their constituents via effective contact interactions [@xue2017; @xue3bound].
In the two previous scenarios, two model-independent properties are experimentally relevant for the study presented below: (i) the existence of contact interactions, in addition to SM gauge interactions, which represents an effective approach for describing the effects of the unknown internal dynamics of compositeness; (ii) the existence of composite fermions or excited states of SM fermions. For more details about the former scenario see Refs. [@comp1; @cfmass; @comp2; @comp3; @comp4; @comp5; @comp6; @comp7; @comp8; @comp9].
In this article we study the latter scenario, focusing on the composite particles arising from four-fermion operators of Einstein-Cartan type, with massive ($m_F$) composite fermions $F^{f}_R\sim \psi^{f}_{_R}(\bar\psi^{f}_{_R} \psi^{f}_{_L})$ (bound states of three SM fermions) and massive ($m_\Pi$) composite bosons $\Pi^{f}\sim (\bar\psi^{f}_{_R} \psi^{f}_{_L})$ (bound states of two SM fermions) forming in the scaling domain of a UV fixed point of the strong four-fermion coupling $G$ at the composite scale $\Lambda \gtrsim 5.14$ TeV and $\Lambda \gtrsim m_F\gtrsim m_\Pi$ [@xue2017; @xuepheno]. The effective coupling between the composite fermion (boson) and its constituents is given by the following contact interaction, which describes composite particle $F^{f}$ ($\Pi^{f}$) production and decay: $$\begin{aligned}
(g_*/\Lambda)^2\bar\psi^{f}_{_L}(\bar\psi^{f}_{_L} \psi^{f}_{_R}) F^{f}_{_R} ~ + ~ {\rm h.c.},
\label{c1}\\
(F_\Pi/\Lambda)^2(\bar\psi^{f}_{_L} \psi^{f}_{_R}) \Pi^{f} ~ + ~ {\rm h.c.},
\label{c1p}\end{aligned}$$ where $(g_*/\Lambda)^2$ is a phenomenological parameter, and one can chose $g_*^2=4\pi$ so that $4\pi/\Lambda^2$ is a geometric cross-section in the order of magnitude of inelastic processes forming composite fermions ( Fig. \[vF\]). Whereas, $(F_\Pi/\Lambda)^2$ is the Yukawa coupling between composite boson (Fig. \[compositeBv\]) and two fermionic constituents, and $(g_*/F_\Pi)^2$ relates to the form factor of composite boson. The composite fermion is in fact a bound state of a SM fermion and composite boson, namely $F^f_R\sim\psi^f_R\Pi^f$. The composite scale $\Lambda > F_\Pi$ can only be experimentally determined like the electroweak scale $v$. The composite-fermion (-boson) mass $m_F, m_\Pi \propto \Lambda$ and the proportionality is of the order of unity.
In the follows we will consider the model in Eq. (\[q1\]) with contact interactions of Eqs. (\[c1\]) and (\[c1p\]) to study composite fermion production and decay at LHC and rely on the aforementioned heavy composite Majorana neutrino experimental studies [@LPF2014; @LARFP2016] for what concerns the determination of constraints on the model parameters. We further compute 5$\sigma$ contour plots of the statistical significance and highlight the region of parameter space where $F$ can manifest using 3 ab$^{-1}$.
The article is arranged as follow. We discuss in Sec. \[contact\] composite fermions’ constituents and effective contact interactions among them. Focusing on the $e^+e^- qq'$ final state in Sec. \[eeqq\_final\], the production cross sections and decay widths of these composite fermions are calculated in Sec. \[xsec\]. In Sec. \[bran\] we present the branching ratios of composite fermions decays in terms of selected parameters: composite scale $\Lambda$, composite particle masses $m_F$ and $m_\Pi$, constrained by the recast of the upper limit on $\sigma(pp \rightarrow eeqq')$ [@LARFP2017] in Sec. \[bound\]. In Sec. \[bound\] we further investigate the region of the paramete space where we expect composite fermion to appear with 3 ab$^{-1}$, which is the statistics expected in the High-Luminosity (HL) LHC. We find out that there is a wide region of model phase space where the composite fermions can be discovered in future searches. We also discuss other channels of composite fermions in Sec. \[OtherChannelsCompositeFermions\] and, in particular, we foresee a new full hadronic final state that, to the best of our knowledge, has not been investigated at the LHC. Finally, we summarize the work with some closing remarks in Sec. \[SummaryAndRemarks\].
Quark-lepton operators and contact interactions {#contact}
===============================================
Composite fermions F
--------------------
To be relevant for possible final states with leptons and quarks in ongoing high-energy experimental searches in $pp$ collisions, we first consider, among four-fermion operators (\[q1\]), the following SM gauge-symmetric and fermion-number conserving four-fermion operators, $$\begin{aligned}
G\left[(\bar\ell^{i}_Le_{R})(\bar d^a_{R}\psi_{Lia})
+(\bar\ell^{i}_L\nu^e_{R})(\bar u^a_{R}\psi_{Lia})\right] + {\rm h.c.},
\label{bhlql}\end{aligned}$$ being the SM doublet $\ell^i_L=(\nu^e_L,e_L)$ and singlet $e_{R}$ with an additional right-handed neutrino $\nu^e_{R}$ for leptons; $\psi_{Lia}=(u_{La},d_{La})$ and $u^a_{R}, d^a_{R}$ for quarks, where the color $a$ and $SU_L(2)$-isospin $i$ indexes are summed over. Eq. (\[bhlql\]) is for the first family only, as a representative of the three fermion families. The SM left- and right-handed fermions are mass eigenstates, their masses are negligible in TeV-energy regime and small mixing among three families encoded in $G$ is also neglected [@xue2016_2].
In Eq. (\[bhlql\]), each four-fermion operator has the two possibilities to form composite fermions, listed in Table \[nuF0\].
[cccc]{} Operator & Composite fermion $F_R$& Composite fermion $\bar F_L$ & Composite boson $\Pi$$(\bar\nu^e_Le_{R})(\bar d^a_{R}u_{La})$& $ E^0_R\sim e_{R}(\bar d^a_{R}u_{La})$ & $\quad \bar E^0_L\sim\bar e_{L}(\bar u^a_{R}d_{La})$& $\Pi^+\sim (\bar d^a_{R}u_{La})$$(\bar e_L\nu^e_{R})(\bar u^a_{R}d_{La})$ & $N^-_R\sim \nu^e_{R}(\bar u^a_{R}d_{La})$ & $\bar N^+_L\sim\bar \nu^e_{L}(\bar d^a_{R}u_{La})$& $\Pi^-\sim (\bar u^a_{R}d_{La})$$(\bar e_Le_{R})(\bar d^a_{R}d_{La})$& $E^-_R\sim e_{R}(\bar d^a_{R}d_{La})$ & $ \bar E^+_L\sim \bar e_{L}(\bar d^a_{L}d_{Ra})$& $\Pi^0_d\sim (\bar d^a_{R}d_{La})$$(\bar\nu^e_L\nu^e_{R})(\bar u^a_{R}u_{La})$&$ N^0_R\sim \nu^e_{R}(\bar u^a_{R}u_{La})$ & $\bar N^0_L\sim\bar\nu^e_{L}(\bar u^a_{L}u_{Ra})$& $\Pi^0_u\sim (\bar u^a_{R}u_{La})$
Up to a form factor, $E$ ($N$) indicates a composite fermion made of an electron (a neutrino) and a color-singlet quark pair, and its superscript for electric charge. There are four independent composite fields $F$: $E^0_R$, $N_R^-$, $E^-_R$, $N_R^0$ and their Hermitian conjugates: $\bar E^0_L=(E^0_R)^\dagger\gamma_0$, $\bar N^+_L=(N^-_R)^\dagger\gamma_0$, $\bar E^+_L=(E^-_R)^\dagger\gamma_0$, $\bar N^0_L=(N^0_R)^\dagger\gamma_0$. They carry SM quantum numbers $t^i_{3L},Y$, and $Q_i=Y+t^i_{3L}$, which are the sum of SM quantum numbers ($t^i_{3L}, Y, Q_i$) of their constituents, i.e., the elementary leptons and quarks in the same SM family [@xue2017], listed in Table \[qnuF0\], so that the contact interactions in Eq. (\[c1\]) are SM gauge symmetric.
[ccccc]{} composite fermions $F_R$ & constituents& charge $Q_i=Y+t^i_{3L}$ & $SU_L(2)$ 3-isospin $t^i_{3L}$& $U_Y(1)$-hypercharge $Y$$E^0_R $& $e_{R}(\bar d^a_{R}u_{La})$ & $0$& $~~1/2$ &$-1/2$$N^-_R$ & $ \nu^e_{R}(\bar u^a_{R}d_{La})$ &$-1$& $-1/2$& $-1/2$$E^-_R$ & $e_{R}(\bar d^a_{R}d_{La})$ & $-1$ &$-1/2$ &$-1/2$$N^0_R$ & $\nu^e_{R}(\bar u^a_{R}u_{La})$&$ 0$ &$~~1/2$ &$ -1/2$
[ccccc]{} composite bosons $\Pi$ & constituents& charge $Q_i=Y+t^i_{3L}$ & $SU_L(2)$ 3-isospin $t^i_{3L}$& $U_Y(1)$-hypercharge $Y$$\Pi^+ $& $(\bar d^a_{R}u_{La})$ & $+1$& $~~1/2$ &$1/2$$\Pi^-$ & $ (\bar u^a_{R}d_{La})$ &$-1$& $-1/2$& $-1/2$$\Pi^0_d$ & $(\bar d^a_{R}d_{La})$ & $0$ &$-1/2$ &$1/2$$\Pi^0_u$ & $(\bar u^a_{R}u_{La})$&$ 0$ &$~~1/2$ &$-1/2$
![ A lepton $\ell$, two quarks $q$ ($u$-type) and $\bar q$ ($d$-type) form a composite fermion $F$ via the contact interaction (dark blob) $P_{L,R}(g_*^2/\Lambda^2)$, where $P_{L,R}=(1\mp\gamma_5)/2$. The thin solid line represents an SM elementary fermion, and the thick double line represents a composite fermion $F$. By a crossing symmetry applied to the lepton line $\ell\rightarrow \ell^\dagger$ (dashed line) the same diagram describes a $2\to2 $ production process $q\bar q \to
\ell^\dagger F$. []{data-label="vF"}](fig1.pdf){height="3.5cm" width="6.9cm"}
![We show the Feynman diagrammatic representation for the contact interaction between the composite boson and its constituent quarks, where the thin solid line represents an SM elementary fermion, double wave line represents a composite boson, and the blob represents an interacting vertex $(F_\Pi/\Lambda)^2P_{L,R}$.[]{data-label="compositeBv"}](fig2){height="3.1cm" width="6.9cm"}
The contact interactions for the production and decay of a composite fermions $F$ are: $$\begin{aligned}
{\mathcal L}^{F}_{\rm CI} = {\mathcal V}_{F} +{\mathcal V}_{F}^\dagger,
\label{cieF}\end{aligned}$$ where $$\begin{aligned}
{\mathcal V}_{\bar E^0} &=& \frac{g_*^2}{\Lambda^2}(\bar E^0_{L} e_{R})(\bar d^a_{R}u_{La}), \quad pp~{\rm or}~ep\rightarrow \bar E^0_{L} e_{R},
\label{cieFf0}\\
{\mathcal V}_{\bar N^+}&=& \frac{g_*^2}{\Lambda^2}(\bar N^+_{L}\nu^e_{R})(\bar u^a_{R}d_{La}), \quad pp~{\rm or}~ep\rightarrow \bar N^+_{L}\nu^e_{R},
\label{cinuFf0}\\
{\mathcal V}_{\bar E^+} &=& \frac{g_*^2}{\Lambda^2}(\bar E^+_{L}e_{R})(\bar d^a_{R}d_{La}),
\quad pp~{\rm or}~ep\rightarrow \bar E^+_{L}e_{R},
\label{cieF0}\\
{\mathcal V}_{\bar N^0} &=& \frac{g_*^2}{\Lambda^2}(\bar N^0_{L}\nu^e_{R})(\bar u^a_{R}u_{La}), \quad pp~{\rm or}~ep\rightarrow \bar N^0_{L}\nu^e_{R},
\label{cinuF0}\end{aligned}$$ and $$\begin{aligned}
{\mathcal V}^\dagger_{E^0} &=& \frac{g_*^2}{\Lambda^2}(\bar e_{L}E^0_{R})(\bar u^a_{R} d_{aL}),\quad E^0_{R}\rightarrow \bar e_{L}(\bar u^a_{R} d_{aL})
\label{dcieFf0}\\
{\mathcal V}^\dagger_{N^-}&=& \frac{g_*^2}{\Lambda^2}
(\bar \nu^e_{L}N^-_R)(\bar d^a_{R}u_{aL}),\quad N^-_R\rightarrow\bar \nu^e_{L}(\bar d^a_{R}u_{aL})
\label{dcinuFf0}\\
{\mathcal V}^\dagger_{E^-} &=& \frac{g_*^2}{\Lambda^2}(\bar e_LE^-_{R})(\bar d^a_{L}d_{Ra}),\quad E^-_{R}\rightarrow \bar e_L(\bar d^a_{L}d_{Ra})
\label{dcieF0}\\
{\mathcal V}^\dagger_{N^0} &=& \frac{g_*^2}{\Lambda^2}(\bar\nu^e_LN^0_{R})(\bar u^a_{L}u_{Ra}),\quad N^0_{R}\rightarrow \bar\nu^e_L(\bar u^a_{L}u_{Ra}).
\label{dcinuF0}\end{aligned}$$ These are relevant contact interactions for phenomenological studies of possible inelastic channels of composite-fermion production and decay in $pp$ or $ep$ collisions.
Composite bosons $\Pi^{0,\pm}$
------------------------------
From the four-fermion interaction in Eq. (\[bhlql\]), it is possible to form composite bosons $$\begin{aligned}
&&\Pi^+=(g^*/F_\Pi)^2(\bar d^a_{R}u_{La}),
\quad \Pi^-= (\Pi^+)^\dagger \label{nB0}\\
&&\Pi^0_d=(g^*/F_\Pi)^2(\bar d^a_{R}d_{La}),\label{dB0}\\
&&\Pi^0_u=(g^*/F_\Pi)^2(\bar u^a_{R}u_{La}),
\label{uB0}\end{aligned}$$ and their Hermitian conjugates. Such normalized composite boson field has the same dimension $[energy]$ of elementary boson field. The composite boson carries the quantum numbers that are the sum over SM quantum numbers of its two constituents, see Table \[qnuP0\]. These are pseudo composite bosons $\Pi^{0,\pm}$, analogous to charged and neutral pions $\pi^{0,\pm}$ in the low-energy QCD.
As shown in Fig. \[compositeBv\], the effective coupling between composite boson and its two constituents can be written as an effective contact interaction, $$\begin{aligned}
{\mathcal L}^{\rm \Pi^\pm}_{\rm CI} &=& g_{_{\rm Y}}(\bar d^a_{R}u_{La})\Pi^- + {\rm h.c.},\label{ciqlB}\\
{\mathcal L}^{\rm \Pi^0_d}_{\rm CI} &=& g_{_{\rm Y}} (\bar d^a_{R} d_{La})\Pi^0_d + {\rm h.c.}
\label{iqlBd}\\
{\mathcal L}^{\rm \Pi^0_u}_{\rm CI} &=& g_{_{\rm Y}}(\bar u^a_{R} u_{Ra})\Pi^0_u + {\rm h.c.}
\label{iqlBu}\end{aligned}$$ where $g_{_{\rm Y}}=(F_\Pi/\Lambda)^2$. Appropriate normalizing the composite boson $\Pi$ with the form factor $(g^*/F_\Pi)^2$ in Eqs. (\[nB0\]-\[uB0\]), the effective contact interaction in Eqs. (\[ciqlB\]-\[iqlBu\]) can be expressed as a dimensionless Yukawa coupling $g_{_{\rm Y}}$, whose value, corresponding to $F_\Pi$ value, can be different for composite bosons in Eqs. (\[nB0\]-\[uB0\]), but we do not consider such difference here.
Contact interaction of composite fermion and boson
--------------------------------------------------
In the view of the composite fermion being a bound state of a composite boson and a SM fermion, using composite-boson fields in Eqs. (\[nB0\]-\[uB0\]), we rewrite ${\mathcal V}^\dagger$ in Eqs. (\[dcieFf0\]-\[dcinuF0\]) as follow, $$\begin{aligned}
{\mathcal V}^\dagger_{E^0} &=& g_{_{\rm Y}}(\bar e_{L}E^0_{R})\Pi^-,\quad E^0_{R}\rightarrow \bar e_{L}\Pi^-
\label{pdcieFf0}\\
{\mathcal V}^\dagger_{N^-}&=& g_{_{\rm Y}}
(\bar \nu^e_{L}N^-_R)\Pi^+,\quad N^-_R\rightarrow\bar \nu^e_{L}\Pi^+
\label{pdcinuFf0}\\
{\mathcal V}^\dagger_{E^-} &=& g_{_{\rm Y}}(\bar e_LE^-_{R})\Pi^0_d,\quad E^-_{R}\rightarrow \bar e_L\Pi^0_d
\label{pdcieF0}\\
{\mathcal V}^\dagger_{N^0} &=& g_{_{\rm Y}}(\bar\nu^e_LN^0_{R})\Pi^0_u,\quad N^0_{R}\rightarrow \bar\nu^e_L\Pi^0_u,
\label{pdcinuF0}\end{aligned}$$ and their Hermitian conjugates ${\mathcal V}$ in Eqs. (\[cieFf0\]-\[cinuF0\]), as shown in Fig.\[compositeFBf\]. These contact interactions in Eqs. (\[pdcieFf0\]-\[pdcinuF0\]) imply that composite fermions $F$: $E^0_R$, $N_R^-$, $E^-_R$, $N_R^0$ can decay into composite bosons $\Pi^\pm$ and $\Pi^0$, which decay then to SM fermions, following the contact interactions in Eqs. (\[ciqlB\]-\[iqlBu\]) at the leading order of tree level. However, we shall consider other decay channels at the next leading order, such as neutral composite boson decay to two SM gauge bosons $\Pi^0_{u,d}\rightarrow \tilde G + \tilde G'$.
![We show the Feynman diagrammatic representation for the contact interaction between the composite fermion and boson, where the thin solid line represents a SM elementary fermion, the double solid line is a composite fermion and the double wave line represents a composite boson and the blob represents an interacting vertex $(F_\Pi/\Lambda)^2P_{L,R}$.[]{data-label="compositeFBf"}](fig3){height="3.5cm" width="6.8cm"}
Contact interaction of $\Pi^0$ composite boson and gauge bosons
---------------------------------------------------------------
Analogously to $\pi^0\rightarrow \gamma\gamma$, the massive $\Pi^0_{u,d}$ composite boson can also decay into two gauge bosons [@xue2017] : $$\begin{aligned}
\Pi^0_{u,d} &\rightarrow& \gamma \gamma, \\
\Pi^0_{u,d} &\rightarrow& \gamma Z^0, \\
\Pi^0_{u,d} &\rightarrow& Z^0 Z^0, \\
\Pi^0_{u,d} &\rightarrow& W^+ W^-,\end{aligned}$$ via the contact interaction $${\mathcal L}^{\Pi^0}_{\tilde G\tilde G'} = \sum_{i=u,d}\frac{gg^\prime N_c}{4\pi^2 F_\Pi}\epsilon_{\mu\nu\rho\sigma}(\partial^\rho A^\mu)(\partial^\sigma A^{\prime\nu})\Pi^0_i
\label{PiGG}$$ where $g$ and $g^\prime$ represent the couplings of gauge bosons $A^\mu$ and $A^{\prime\nu}$ to the SM quarks $u$ and $d$ with different $SU_L(2)$-isospin $i=u,d$. Actually, this effective contact interaction (\[PiGG\]) is an axial anomaly vertex, as a result of a triangle quark loop and standard renormalization procedure in SM.
$e^+e^-qq^\prime$ final state in $pp$ collisions {#eeqq_final}
================================================
In this section we study the processes giving the $e^+e^-qq^\prime$ final state, which we can use to set bounds on the parameters of the model by using the recast of the experimental upper limit on $\sigma (pp \to eeqq^\prime)$ published in [@LARFP2017]. For this purpose, we consider only the case of composite fermions $F=E^0,\bar E^0, E^+,E^-$. The detailed analysis of composite fermions $F=N^0,\bar N^0, N^+,N^-$, giving the $\nu\nu qq^\prime$ final states, will be considered in future.
If the energy $\sqrt{s}$ in the parton center of mass frame is larger than composite fermion masses, the resonant processes described below can occurs. The virtual processes of composite fermions are not considered here. The kinematics of final states is simple in the center of mass frame of $pp$ collisions. In $pp$ collisions at LHC, the $e^+e^-qq^\prime$ final state with this model can be obtained via the production of the composite fermions $E^0, \bar{E}^0, E^-, E^+$ in association with an electron or a positron and the subsequent decay of the composite fermion to a positron or an electron and two quarks: $$\begin{aligned}
pp &\rightarrow& e^+ E^0\rightarrow e^+e^-qq^\prime,\\
pp &\rightarrow& e^- \bar{E^0}\rightarrow e^-e^+qq^\prime,\\
pp &\rightarrow& e^+ E^-\rightarrow e^+e^-qq^\prime,\\
pp &\rightarrow& e^- E^+\rightarrow e^-e^+qq^\prime ,\end{aligned}$$ The quark-family mixing is neglected, so at parton level the previous equations are: $$\begin{aligned}
u\bar{d} &\rightarrow& e^+ E^0\rightarrow e^+e^-u\bar{d},\\
\bar{u}d &\rightarrow& e^- \bar{E^0}\rightarrow e^-e^+\bar{u}d,\\
d\bar{d} &\rightarrow& e^+ E^-\rightarrow e^+e^-d\bar{d},\\
d\bar{d} &\rightarrow& e^- E^+\rightarrow e^-e^+d\bar{d}.\end{aligned}$$ The decay of the composite fermion to a lepton and two quarks can happen directly, via the interactions in Eq. (\[dcieFf0\], \[dcieF0\]), or with the decay of the composite fermion to a lepton and the composite boson, via the interactions in Eq. (\[pdcieFf0\], \[pdcieF0\]), and the subsequent decay of the composite boson to two quarks, via the interactions in Eq. (\[ciqlB\], \[iqlBd\], \[iqlBu\]): $$\begin{aligned}
E^0 &\rightarrow& e^- \Pi^+ \rightarrow e^- u\bar{d},\\
\bar{E^0} &\rightarrow& e^+ \Pi^- \rightarrow e^+ \bar{u}d,\\
E^- &\rightarrow& e^- \Pi^0_d \rightarrow e^- d\bar{d},\\
E^+ &\rightarrow& e^+ \Pi^0_d \rightarrow e^+ d\bar{d} .\end{aligned}$$ The cross sections of these processes are: $$\sigma(pp\rightarrow eF \rightarrow e^+e^-qq^\prime)=\sigma(pp\rightarrow eF)\times{\mathcal B}(F\rightarrow eqq^\prime),$$ where $${\mathcal B}(F\rightarrow eqq^\prime)=\frac{\Gamma_{3-body}(F\rightarrow eqq^\prime)+\Gamma(F\rightarrow e\Pi){\mathcal B}(\Pi\rightarrow qq^\prime)}{\Gamma_{tot}(F)}
\label{Bfeqq}$$ and $$\Gamma_{\rm tot}(F)=\Gamma(F\rightarrow e\Pi)+\Gamma_{3-body}(F\rightarrow eqq^\prime).\label{tfeqq}$$
The total cross section of the $e^+e^-qq^\prime$ channel from the model in $pp$ collisions is approximately given by $$\begin{aligned}
\!\!\!\!\!\!\!\sigma(pp\!\rightarrow\! e^+e^-qq^\prime)\approx&&\, \sigma(pp\rightarrow e^+ E^0)\times {\mathcal B}(E^0 \rightarrow e^+ \bar u d)\nonumber\\
+&&\, \sigma(pp\rightarrow e^- \bar E^0)\times {\mathcal B}(\bar E^0 \rightarrow e^- u \bar d)\nonumber \\
+&&\, \sigma(pp\rightarrow e^+ \bar E^-)\times {\mathcal B}(\bar E^- \rightarrow e^- d \bar d)\nonumber \\
+&&\, \sigma(pp\!\rightarrow\! e^- E^+)\times {\mathcal B}( E^+\! \rightarrow\! e^+ d \bar d).
\label{ppcs}\end{aligned}$$ The calculation of these quantities will be given in the next sections.
 
Cross sections and decay widths {#xsec}
===============================
The partonic cross section of $qq^\prime\rightarrow e F$ is calculated by standard methods via the contact interaction in Eqs. (\[cieF\]-\[cinuF0\]) (all of them give the same result), $$\begin{aligned}
\hat\sigma(\hat{s},m_F)= \frac{1}{3\times64\pi}\left(\frac{g_*^2}{\Lambda^2}\right)^2\frac{(\hat{s}-m_F^2)^2}{m_F^2},
\label{csE0}\end{aligned}$$ where $\sqrt{\hat{s}}$ stands for the parton center-mass-energy of $pp$ collisions in LHC experiments.
We consider the production cross sections for the composite fermions $F$ in $pp$ collisions expected at the CERN LHC collider according to Feynman’s parton model. The QCD factorization theorem allows to obtain any hadronic cross section (e.g. in $pp$ collisions) in terms of a convolution of the hard partonic cross sections $\hat\sigma$, evaluated at the parton center of mass energy $\sqrt{\hat s}=\sqrt{\tau s}$, with the universal parton distribution functions $f_a(x,{\hat{\cal Q}})$ which depend on the parton longitudinal momentum fractions $x$, and on the factorization scale $\hat{\cal Q}$: $$\begin{aligned}
\sigma=
\sum_{ij}\!\!\intop^1_{\frac{m_F^{2}}{s}}\!\!d\tau \!\!\intop_\tau^1 \!\!\frac{dx}{x} f_i(x,\hat {\cal Q}^2)f_j(\frac{\tau}{x},\hat {\cal Q}^2)\hat\sigma(\tau s)\, .
\label{csFf}\end{aligned}$$ The factorization and renormalization scale ${\cal Q}$ is generally fixed at the value of the mass that is being produced. The parametrization of the parton distribution function is NNPDF3.0 [@Ball:2014uwa] and the factorization scale has been chosen as $\hat{\cal Q}=m_F$.
The right panel of Fig. \[ppE0e+\] shows the agreement between analytical calculations based on Eqs. and , for the case of the fermion $E^0$, and the results of simulations with CalcHEP where the model with four-fermion interactions has been implemented. We remark the quite good agreement that validates our model implementation in CalcHEP.
Analytical calculations, in the similar way as the first term in Eq. (5) of Ref. [@LPF2014], yield the width of composite fermion decay to its quark and lepton constituents $$\begin{aligned}
\Gamma_{3-\text{body}}(F \to e qq^\prime )= \left(\frac{g_*^2}{\Lambda^2}\right)^2\frac{m_F^5}{4\times(8\pi)^3}.
\label{decayE0}\end{aligned}$$ Note that at TeV energy scales, composite fermions are massive ($m_F$) Dirac fermions, whereas all SM elementary fermions are treated as massless Dirac fermions of four spinor components, consisting of right- and left-handed Weyl fermions of two spinor components. Alternatively, the decay width $\Gamma_{F}$ has also been evaluated via CalcHEP, and numerical results are completely in agreement with analytical one in Eq. (\[decayE0\]), see left panel of Figure \[E0decaywdth\].
The decay width of the composite fermion to a lepton and a composite boson $\Pi$ can easily be computed from the effective contact lagrangian in Eqs. (\[pdcieFf0\]) and (\[pdcieF0\]): $$\Gamma(F \to e \Pi) = \frac{1}{32\pi}\left(\frac{F_\Pi^2}{\Lambda^2}\right)^2 \, m_F\, \left(1-\frac{m_\Pi^2}{m_F^2}\right)^2.
\label{fep}$$ The decay width of the $\Pi$ boson to two quarks is simply calculated by using the effective contact Lagrangian in Eq. (\[ciqlB\]) and (\[iqlBd\]), $$\Gamma(\Pi\to qq^\prime) = \frac{3}{16\pi}\, \left(\frac{F_\Pi}{\Lambda}\right)^4 \,m_\Pi.
\label{Gamma_pqq}$$ For the $\Pi^+$ and $\Pi^-$ composite bosons this is the only decay channel, therefore composite fermions $E^0$ and $\bar{E^0}$ have ${\mathcal B}(F\rightarrow eqq^\prime)=1$. The $\Pi^0_d$ composite boson, instead, can also decay to two gauge bosons $\tilde G \tilde G'$, according to the contact interaction (\[PiGG\]), the corresponding decay widths are [@xue2017] : $$\begin{aligned}
\Gamma_{\Pi^0_d\rightarrow\gamma\gamma}&=&\left(\frac{5}{9}\right)^2\Gamma, \label{pg}\\
\Gamma_{\Pi^0_d\rightarrow\gamma Z^0}&=&\frac{1}{\sin^2 2\theta_W}\left(\frac{1}{2}-\frac{5}{9}\sin^2 \theta_W\right)^2\Gamma, \label{pgz}\\
\Gamma_{\Pi^0_d\rightarrow Z^0 Z^0}&\!=\!&\left(\frac{1/2\!-\!\sin^2\theta_W\!+\!(5/9)\sin^4\theta_W}{\sin^2 2\theta_W}\right)^{\!2}\Gamma, \label{pzz}\\
\Gamma_{\Pi^0_d\rightarrow W^+ W^-}&=&\left(\frac{1}{8\sin^2\theta_W}\right)^2\Gamma,\label{pww}\end{aligned}$$ where $\theta_W$ is the Weinberg angle, $$\Gamma=\left(\frac{\alpha N_c}{3\pi F_\Pi}\right)^2\frac{m_{\Pi^0_d}^3}{64\pi},
\label{ggrate}$$ and the number of colors $N_c=3$. Total decay rate $\Gamma^{\rm tot}(\Pi^0_d\rightarrow \tilde G\tilde G')$ is the sum over all contributions from Eqs. (\[pg\]-\[pww\]). The total $\Pi^0_d$-decay rate reads $$\Gamma_{\rm tot}(\Pi^0_d)=\Gamma(\Pi^0_d\rightarrow qq')
+\Gamma^{\rm tot}(\Pi^0_d\rightarrow \tilde G\tilde G'),
\label{ptotal}$$ where $\Gamma(\Pi^0_d\rightarrow qq')$ is given by Eq. (\[Gamma\_pqq\]). Based on these results, we calculate the branching ratios of different channels in next section.
![\[BrE\] (Color Online). The branching ratios of the composite fermion $F$ decay to $eqq^\prime$ and to $e~\Pi$. The branching ratio $F\rightarrow eqq^\prime$ is much larger than the branching ratio $F\rightarrow e\Pi$, except in a small regime $0.4 \lesssim m_\Pi/m_F \lesssim 0.9 $, where the branching ratio $F\rightarrow e~\Pi$ is not completely negligible.](Br)
Parameters and Branching ratios {#bran}
===============================
In order to present the branching ratios of different possible channels in terms of parameters of the model, we are bound to discuss physically sensible parameters to explore. This model has four parameters that can be rearranged to three dimensionless parameters for a given $\Lambda$ value: $$(\Lambda, m_F, F_\Pi, m_\Pi) \rightarrow
( m_F/\Lambda,m_\Pi/m_F, F_\Pi/m_\Pi).$$ The ratio $m_F/\Lambda < 1$ ($m_\Pi/\Lambda < 1$) of the composite fermion (boson) mass and the basic composite scale $\Lambda$ gives us an insight into the dynamics of composite fermion (boson) formation. On the other hand, a composite fermion $F$ is composed by a composite boson and an elementary SM fermion, to represent this feature, we adopt the ratio $m_\Pi/m_F < 1$ as a parameter. In addition, considering the parameters $m_\Pi$ and $F_\Pi$ represent the same dynamics of composite boson formation, we approximately adopt $F_\Pi\approx m_\Pi$ so as to reduce the numbers of free parameters at this preliminary stage. As a result, for given $\sqrt{s}$ and $\Lambda$ values, we have two parameters $(m_\Pi/m_F,m_F/\Lambda)$ to represent the results of cross sections, decay rates and branching ratios of various decay channels $E\rightarrow eqq^\prime$, $E\rightarrow e \Pi \rightarrow eqq^\prime$, and $\Pi\rightarrow \tilde G\tilde G^\prime$.
Figure \[BrE\] shows the branching ratios of the composite fermion $E$ decay to $eqq^\prime$, i.e., ${\mathcal B}(E\rightarrow eqq')$ of Eqs. (\[Bfeqq\],\[tfeqq\]), and $E$ decay to $e~\Pi$, $${\mathcal B}(E\rightarrow e\Pi)=\Gamma(E\rightarrow e\Pi)/\Gamma_{\rm tot}(E),
\label{bEep}$$ where $\Gamma(E\rightarrow e\Pi)$ is given by Eq. (\[fep\]). The results show the direct decay channel $E\rightarrow eqq^\prime$ is dominant over the decay channel $E\rightarrow e\Pi$. Note that these branching ratios are independent from $m_F/\Lambda$ for the parameterization $F_\Pi=m_\Pi$.
Figure \[BrPi\] shows for $F_\Pi=m_\Pi$ and two selected $m_F/\Lambda$ values, the branching ratios of the $\Pi^0_d$ decay to two quarks $qq^\prime$, $${\mathcal B}(\Pi^0_d\rightarrow qq^\prime) = \Gamma(\Pi^0_d\rightarrow qq^\prime)/ \Gamma_{\rm tot}(\Pi^0_d)
\label{pqq}$$ and the $\Pi^0_d$ decay to two gauge bosons $\tilde G\tilde G^\prime$, $${\mathcal B}(\Pi^0_d\rightarrow \tilde G\tilde G^\prime)=\Gamma^{\rm tot}(\Pi^0_d\rightarrow \tilde G\tilde G')/\Gamma_{\rm tot}(\Pi^0_d).
\label{pgg}$$ The results show that the decay of $\Pi^0\rightarrow \tilde G\tilde G^\prime$ is not negligible only for small values of both $m_F/\Lambda$ and $m_\Pi/m_F$, see Figure \[BrPi\] left panel.
Figure \[BrFIN\] shows that the branching ratios of the direct $E$ decay to a charged lepton and two quarks $E\rightarrow eqq'$, $${\mathcal B}(E\rightarrow eqq^\prime,{\rm direct})=\Gamma_{\rm 3-body}(E\rightarrow eqq')/\Gamma_{\rm tot}(E),
\label{Eeqqdirect}$$ and indirect $E$ decay $E\rightarrow e\Pi\rightarrow eqq'$, $${\mathcal B}(E\rightarrow e\Pi\rightarrow eqq^\prime)=\frac{
\Gamma(E\rightarrow e\Pi)}{\Gamma_{\rm tot}(E)}{\mathcal B}(\Pi\rightarrow qq'),
\label{Eepqq}$$ and the sum of these two branching ratios gives the total branching ratio ${\mathcal B}(E\rightarrow eqq^\prime)$ of $E$ decay to $eqq^\prime$. In addition, it is also shown in Figure \[BrFIN\] that the branching ratio of the decay channel $E\rightarrow e\Pi^0\rightarrow e\tilde G\tilde G'$, $${\mathcal B}(E\rightarrow e\Pi^0_d\rightarrow e\tilde G\tilde G^\prime)=\frac{
\Gamma(E\rightarrow e\Pi^0_d)}{\Gamma_{\rm tot}(E)}{\mathcal B}(\Pi^0_d\rightarrow \tilde G\tilde G').
\label{Epegg}$$ Despite this decay channel would be peculiar, having a final state signature not typical of the standard model processes, with highly energetically boosted gauge bosons plus an electron, the results show that the branching ratio ${\mathcal B}(E^\pm\rightarrow eqq^\prime)$ is much larger than ${\mathcal B}(E^\pm\rightarrow e\Pi_0\rightarrow e\tilde G\tilde G^\prime)$ in the parameter space we have explored with the aforementioned parameter assumptions.
 
 
Bounds on the model {#bound}
===================
![\[recast\] (Color Online). Recast of the experimental upper limit on $\sigma (pp \to eeqq^\prime)$ published in [@LARFP2017] against the model of composite fermions studied in this article. The dotted line (solid green line) is the 95% C.L. observed (expected) upper limit on $\sigma (pp \to eeqq^\prime)$ as reported in [@LARFP2017]. The solid line (red) is the theoretical expectation from the model described in this work as given by Eq. for the case $\Lambda=m_F$, the dashed lines (orange) are the theoretical expectation from the model for the cases $\Lambda=6,\,9,\,12$ TeV. If $\Lambda=m_F$ one obtains that the composite fermions of this model are excluded up to masses $m^{\rm ex}_F\approx 4.25$ TeV.[]{data-label="1dimlimit"}](recast)
![\[recast2dim\] (Color Online). Recast of the experimental upper limit from [@LARFP2017] (dashed line) and the predicted contour curve at a 5-level statistical significance (solid line) in the 2-dimensional parameter space $(\Lambda,m_F)$. The shaded region denotes unphysical values of the parameters ($\Lambda < m_F$). []{data-label="2dimlimit"}](bounds)
In this section we provide a discussion of the bounds on this model by recasting the 95% confidence level (C.L.) experimental upper limit on $\sigma(pp\rightarrow eeqq^\prime)$ using a recent analysis [@LARFP2017] of 2.3 fb$^{-1}$ data from the 2015 Run II of the LHC by the CMS collaboration with respect to the predictions of the model of composite fermions discussed in this article. Note that both electrons and positrons are collected in the final states of $eeqq^\prime$, electrons and positrons are not distinguished in the data analysis. For the case $m_F=\Lambda$ one obtains that the composite fermions of this model are excluded up to masses $m^{\rm ex }_F\approx 4.25$ TeV. This result is shown in Figure \[recast\], together with the exclusion limits $m^{\rm ex }_F\approx 3.3, 2.4,1.5$ TeV for $\Lambda$ fixed at 6, 9 and 12 TeV. Figure \[2dimlimit\] shows the exclusion curve, lower (dashed) line, in the 2-dimensional parameter space $(\Lambda,m_F)$ for our model obtained via the recasting of the analysis [@LARFP2017] of 2.3 fb$^{-1}$ data from the 2015 Run II of the LHC by the CMS collaboration. Here the regions of the parameter space below the curves are excluded.
We also performed a study about the potential of a dedicated analysis in the High Luminosity LHC (HL-LHC) conditions (center of mass energy of 14 TeV and luminosity of 3 ab$^{-1}$). We used CalcHEP to generate the processes and DELPHES [@Delphes] to simulate the detector effects. In order to separate the signal from the background, we selected events with $pt_{e_1}\geq$ 180 GeV, $pt_{e_2}\geq$ 80 GeV, $pt_{j_1}\geq$ 210 GeV, $m_{ee}\geq$ 300 GeV ($pt$ is the transverse momentum, $e_1$ the leading electron, $e_2$ the subleading electron, $j_1$ the leading jet and $m_{ee}$ the invariant mass of the two electrons). Then we evaluated the reconstruction and selection efficiencies for signal ($\epsilon_s$) and background ($\epsilon_b$) as the ratio of the selected and the total generated events. From these efficiencies, the signal and background cross sections ($\sigma_s$, $\sigma_b$) and the integrated luminosity ($L$), it is possible to evaluate the expected number of events for the signal ($N_s$) and the SM background ($N_b$) and finally the statistical significance ($S$): $$N_s=L\sigma_s\epsilon_s, \quad N_b=L\sigma_b\epsilon_b, \quad S=\frac{N_s}{\sqrt{N_b}}\, .$$ The $S=5$ contour curve is shown by the upper (solid) line in Figure \[2dimlimit\]. It can be used to get indications about the potential for discovery or exclusion with the experiments at the HL-LHC, showing that there is a wide region of the model phase space where the existence of the composite fermions can be investigated; for the case $\Lambda=m_F$ we can reach masses up to $\approx 6.2$ TeV.
Other channels of composite fermions {#OtherChannelsCompositeFermions}
====================================
In this article, we have carried out the analysis of composite fermions $F=E^0,\bar E^0, E^+,E^-$ produced in LHC $pp$ collisions for the final states $eeqq^\prime$ or $ee \tilde G\tilde G^\prime$ . However, the exact same analysis can be done for composite fermions $F=N^0,\bar N^0, N^+,N^-$ for the final state $\nu\nu qq^\prime$ or $\nu\nu \tilde G\tilde G^\prime$, $$pp\rightarrow \nu N \rightarrow \nu\nu qq^\prime, ~~{\rm or}~~\nu\nu \tilde G\tilde G^\prime\label{Nnu}$$ where $\nu\nu$ stands for the SM left-handed neutrino $\nu^e_L$ and/or sterile right-handed neutrino $\nu^e_R$. The latter is a candidate of dark-matter particles, represented by missing energy and momentum in the final states. Substituting $e^+e^-$ by $\nu^e_L\nu^e_R$ in above calculations, we obtain the same results at this preliminary level without turning on SM gauge interactions. For example, analogously to Eqs. (\[Eeqqdirect\],\[Eepqq\],\[Epegg\]), the branching ratios of the direct $N$ decay to a neutrino and two quarks $N\rightarrow \nu qq'$, $${\mathcal B}(N\rightarrow \nu qq^\prime,{\rm direct})
=\Gamma_{\rm 3-body}(N\rightarrow \nu qq')/\Gamma_{\rm tot}(N),
\label{Nnuqqdirect}$$ and indirect $N$ decay $N\rightarrow \nu\Pi^0_u\rightarrow \nu qq'$, $${\mathcal B}(N\rightarrow \nu\Pi^0_u\rightarrow \nu qq^\prime)=\frac{
\Gamma(N\rightarrow \nu\Pi^0_u)}{\Gamma_{\rm tot}(N)}{\mathcal B}(\Pi^0_u\rightarrow qq'),
\label{Nnupqq}$$ and the sum of these two branching ratios gives the total branching ratio ${\mathcal B}(N\rightarrow \nu qq^\prime)$ of $N$ decay to $\nu qq^\prime$. The branching ratio of the decay channel $N\rightarrow \nu\Pi^0_u\rightarrow \nu\tilde G\tilde G'$, $${\mathcal B}(N\rightarrow \nu\Pi^0_u\rightarrow \nu\tilde G\tilde
G^\prime)=\frac{
\Gamma(N\rightarrow \nu\Pi^0_u)}{\Gamma_{\rm tot}(N)}{\mathcal B}(\Pi^0_u\rightarrow \tilde G\tilde G').
\label{Npnvgg}$$ The same numerical results can be found in Fig. (\[BrFIN\]).
In fact, both composite bosons ($\Pi$) and fermions ($F$) have definite SM quantum numbers, so that the Feynman diagrammatic representations of SM perturbative gauge interactions can be easily implemented, see Eqs. (4.8)-(4.11) in Ref. [@xue2017]. However, at the leading order of contact interactions discussed in this article, all gauge interactions are neglected, except the effective contact interaction (\[PiGG\]) of the triangle anomaly, which couples to two SM gauge bosons $\tilde G \tilde G'$. It should be mentioned that these gauge bosons $\tilde G\tilde G'$ in Eq. (\[PiGG\]) can be two gluons that possibly fuse to a Higgs particle in the final states.
Composite fermions $Q$ and four-jet final states
------------------------------------------------
We further consider, among the variants of Eq. (\[q1\]), the following SM gauge-symmetric and fermion-number conserving four-fermion operators of the quark sector, choosing as representative the first family [@xue2016_2; @xue2017], $$\begin{aligned}
G\left[(\bar\psi^{bi}_Ld_{Rb})(\bar d^a_{R}\psi_{Lia})
+(\bar\psi^{bi}_Lu_{Rb})(\bar u^a_{R}\psi_{Lia})\right] + {\rm h.c.}.
\label{qbhlql}\end{aligned}$$ Each four-fermion operator has the two possibilities to form composite fermions, listed in Table \[QF0\].
[cccc]{} Operator & Composite fermion $Q_R$& Composite fermion $\bar Q_L$ & Composite boson $\Pi$$(\bar u_{Lb}d_{Rb})(\bar d^a_{R}u_{La})$& $ D^{2/3}_{Rb}\sim d_{Rb}(\bar d^a_{R}u_{La})$ & $\quad \bar D^{-2/3}_{Lb}\sim \bar d_{Lb}(\bar u^a_{R}d_{La})$ & $\Pi^+\sim (\bar d^a_{R}u_{La})$ $(\bar d_{Lb}u_{Rb})(\bar u^a_{R}d_{La})$ & $U^{-1/3}_{Rb}\sim u_{Rb}(\bar u^a_{R}d_{La})$ & $\bar U^{1/3}_{Lb}\sim \bar u_{Lb}(\bar d^a_{R}u_{La})$ & $\Pi^-\sim (\bar u^a_{R}d_{La})$$(\bar d_{Lb}d_{Rb})(\bar d^a_{R}d_{La})$& $D^{-1/3}_{Rb}\sim d_{Rb}(\bar d^a_{R}d_{La})$ & $ \bar D^{1/3}_{Lb}\sim \bar d_{Lb}(\bar d^a_{L}d_{Ra})$ & $\Pi^0_d\sim (\bar d^a_{R}d_{La})$$(\bar u_{Lb}u_{Rb})(\bar u^a_{R}u_{La})$&$ U^{2/3}_{Rb}\sim u_{Rb}(\bar u^a_{R}u_{La})$ & $\bar U^{-2/3}_{Lb}\sim \bar u_{Lb}(\bar u^a_{L}u_{Ra})$ & $\Pi^0_u\sim (\bar u^a_{R}u_{La})$
Up to a certain form factor, $D$ ($U$) indicates a composite fermion made of a down quark $d$ (an up quark $u$) and a color-singlet quark pair, and its superscript for electric charge. There are four independent composite fields $Q$: $D^{2/3}_{Ra}$, $U_{Ra}^{-1/3}$, $D^{-1/3}_{Ra}$, $U_{Ra}^{2/3}$ and their Hermitian conjugates: $\bar D^{-2/3}_{La}=(D^{2/3}_{Ra})^\dagger\gamma_0$, $\bar U^{1/3}_{La}=(U_{Ra}^{-1/3})^\dagger\gamma_0$, $\bar D^{1/3}_{La}=(D^{-1/3}_{Ra})^\dagger\gamma_0$, $\bar U^{-2/3}_{La}=(U_{Ra}^{2/3})^\dagger\gamma_0$. They carry SM quantum numbers $t^i_{3L},Y$, and $Q_i=Y+t^i_{3L}$, which are the sum of SM quantum numbers ($t^i_{3L}, Y, Q_i$) of their constituents, i.e., the elementary quarks in the same SM family [@xue2017], listed in Table \[qQF0\]. These composite fermions $D$ and $U$ are analogous to the composite fermions $E$ and $N$ that have been previously analyzed.
[ccccc]{} composite fermions $Q_R$ & constituents& charge $Q_i=Y+t^i_{3L}$ & $SU_L(2)$ 3-isospin $t^i_{3L}$& $U_Y(1)$-hypercharge $Y$$D^{2/3}_{Rb} $& $d_{Rb}(\bar d^a_{R}u_{La})$ & $2/3$& $~~1/2$ &$1/6$$U^{-1/3}_{Rb}$ & $ u_{Rb}(\bar u^a_{R}d_{La})$ &$-1/3$& $-1/2$& $1/6$$D^{-1/3}_{Rb}$ & $d_{Rb}(\bar d^a_{R}d_{La})$ & $-1/3$ &$-1/2$ &$1/6$$U^{2/3}_{Ra}$ & $u_{Rb}(\bar u^a_{R}u_{La})$&$ 2/3$ &$~~1/2$ &$ 1/6$
The contact interactions for the production and decay of a composite fermions $Q$ are, $$\begin{aligned}
{\mathcal L}^{Q}_{\rm CI} = {\mathcal V}_{Q} +{\mathcal V}_{Q}^\dagger,
\label{cieQ}\end{aligned}$$ where $$\begin{aligned}
&&{\mathcal V}_{\bar D^{-2/3}} = \frac{g_*^2}{\Lambda^2}(\bar D^{-2/3}_{Lb} d_{Rb})(\bar d^a_{R}u_{La}); \!\!\!\quad\!\!\! pp~\rightarrow \bar D^{-2/3}_{La} d_{Ra},
\label{cieQf0}\\
&&{\mathcal V}_{\bar U^{1/3}}= \frac{g_*^2}{\Lambda^2}(\bar U^{1/3}_{Lb}u_{Rb})(\bar u^a_{R}d_{La}); \quad pp~\rightarrow \bar U^{1/3}_{La}u_{Ra},
\label{cinuQf0}\\
&&{\mathcal V}_{\bar D^{1/3}} = \frac{g_*^2}{\Lambda^2}(\bar D^{1/3}_{Lb}d_{Rb})(\bar d^a_{R}d_{La});
\quad pp~\rightarrow \bar D^{1/3}_{La}d_{Ra},
\label{cieQ0}\\
&&{\mathcal V}_{\bar U^{-2/3}} = \frac{g_*^2}{\Lambda^2}(\bar U^{-2/3}_{Lb}u_{Rb})(\bar u^a_{R}u_{La}); \!\!\!\quad\!\!\!
pp~\rightarrow \bar U^{-2/3}_{Lb}u_{Rb},
\label{cinuQ0}\end{aligned}$$ and $$\begin{aligned}
\!\!&&{\mathcal V}^\dagger_{D^{2/3}} \!=\! \frac{g_*^2}{\Lambda^2}(\bar d_{Lb}D^{2/3}_{Rb})(\bar u^a_{R} d_{La}); \!\!\quad\!\!
D^{2/3}_{Rb}\rightarrow \bar d_{Lb}(\bar u^a_{R} d_{La})
\label{dcieQf0}\\
\!\!\!&&{\mathcal V}^\dagger_{U^{-1/3}}\! = \!\frac{g_*^2}{\Lambda^2}
(\bar u_{Lb}U^{-1/3}_{Rb})(\bar d^a_{R}u_{La});\!\!\!\quad\!\!\! U^{-1/3}_{Rb}\!\!\rightarrow\!\bar u_{Lb}(\bar d^a_{R}u_{La})
\label{dcinuQf0}\\
\!\!\!&&{\mathcal V}^\dagger_{D^{-1/3}}\! = \!\frac{g_*^2}{\Lambda^2}(\bar d_{Lb}D^{-1/3}_{Rb})(\bar d^a_{L}d_{Ra});
\!\!\!\quad\!\!\! D^{-1/3}_{Rb}\!\!\rightarrow\! \bar d_{Lb}(\bar d^a_{L}d_{Ra})
\label{dcieQ0}\\
\!\!&&{\mathcal V}^\dagger_{U^{2/3}} \!=\! \frac{g_*^2}{\Lambda^2}(\bar u_{Lb}U^{2/3}_{Rb})(\bar u^a_{L}u_{Ra});\!\!\quad\!\!
U^{2/3}_{Rb}\rightarrow \bar u_{Lb}(\bar u^a_{L}u_{Ra}).
\label{dcinuQ0}\end{aligned}$$ These are the relevant contact interactions for phenomenological studies of possible inelastic channels of composite-fermion $Q=(D,U)$ production and decay in $pp$ collision. Without considering SM gauge interactions and family mixings, the phenomenological analysis and results are the same as those of composite fermions $(E,N)$, apart from different final states.
Possible $Q$-resonances and four jets final states in $pp$ collisions
---------------------------------------------------------------------
The $pp$ or $ep$ collisions produce a composite fermion $Q$ and a quark $q=u,d$, i.e., the production process $pp\rightarrow \bar Q q $ via the contact interactions in Eq. ${\mathcal V}_Q$ (\[cieQf0\]-\[cinuQ0\]). The composite fermions $Q$ decay to a quark and a pair of quarks, $Q\rightarrow \bar q q\bar q$ via the contact interactions ${\mathcal V}^\dagger_Q$ (\[dcieQf0\]-\[dcinuQ0\]). A composite fermion $Q= D^{2/3}_{Ra}$, $U_{Ra}^{-1/3}$, $D^{-1/3}_{Ra}$, $U_{Ra}^{2/3}$ appears in the $s$-channel as a resonance.
The following final states are foreseen in $pp$ collisions at LHC: $$\begin{aligned}
pp~(u\bar d) &\rightarrow& \bar d D^{2/3}\rightarrow \bar d + d + {\rm two~jets}~(u\bar d ), \label{qud}\\
pp~(d\bar d ) &\rightarrow& \bar d D^{-1/3}\rightarrow d + \bar d + {\rm two~jets}~(d \bar d); \label{qdd}\\
pp~(d\bar u) &\rightarrow& \bar u U^{-1/3} \rightarrow \bar u
+ u + {\rm two~jets}~(d\bar u), \label{qdu}\\
pp~(u\bar u) &\rightarrow& \bar u U^{2/3} \rightarrow \bar u
+ u + {\rm two~jets}~(u\bar u )\label{quu}.\end{aligned}$$ These four-jet events $pp\rightarrow jj + jj$ have a simple, but peculiar kinematic distribution that may be easily identified from background. At the present tree-level approximation of contact interactions, the cross sections, decay rates, kinematics and parameters $(\Lambda,m_F)$ are the same as those of $pp\rightarrow \ell^\dagger \ell + jj$ processes.
Analogously to the branching ratios (\[Epegg\]) and (\[Npnvgg\]) of the decay channels $E\rightarrow e\,\Pi^0_d\rightarrow e\,\tilde G\tilde G'$ and $N\rightarrow \nu\,\Pi^0_u\rightarrow \nu\,\tilde G\tilde G'$, we can obtain the branching ratios of composite fermions $D^{-1/3}$ and $U^{2/3}$ decay into a quark (jet) and two boosted gauge bosons, $$\begin{aligned}
{\mathcal B}(D^{-1/3}\rightarrow d\,\Pi^0_d\rightarrow d\,\tilde G\tilde G')&=&\frac{
\Gamma(D^{-1/3}\rightarrow d\,\Pi^0_d)}{\Gamma_{\rm tot}(D^{-1/3})}\nonumber\\
&\times&{\mathcal B}(\Pi^0_d\rightarrow \tilde G\tilde G');
\label{Dpqgg}\\
{\mathcal B}(U^{2/3}\rightarrow u\,\Pi^0_u\rightarrow u\,\tilde G\tilde G')&=&\frac{
\Gamma(U^{2/3}\rightarrow u\,\Pi^0_u)}{\Gamma_{\rm tot}(U^{2/3})}\nonumber\\
&\times&{\mathcal B}(\Pi^0_u\rightarrow \tilde G\tilde G').
\label{Upqgg}\end{aligned}$$ At the present tree-level approximation of contact interactions, the cross sections, decay rates, branching ratios, kinematics and parameters $(\Lambda,m_F)$ of the composite fermions $D^{-1/3}$ and $U^{2/3}$ production and decay into a jet and two boosted gauge bosons are the same as those of the composite fermions $E^\pm$ and $N$ production and decay into a lepton and two boosted gauge bosons discussed in Eqs. (\[Epegg\]) and (\[Npnvgg\]).
Here again we neglect the small contributions from perturbative SM gauge interactions, and only consider the dominant tree-level contributions of the first-family contact interactions (\[cieQ\]) without any flavor mixing. Other possible channels with final states of gauge and Higgs bosons, as well as heavy quarks [@xue2017; @xuepheno] are expected to have much smaller branching ratios and will be duly discussed and analyzed in future.
Summary and remarks {#SummaryAndRemarks}
===================
In the weak coupling regime the effective four-fermion operators of NJL-type possess an IR-fixed point, rendering the elegant Higgs mechanism of the SM of particle physics at low energies. In the strong coupling regime, on the other end, these operators could possess an UV-fixed point, giving rise to composite fermions/bosons composed by SM fermions and their relevant contact interactions with SM fermions at high energies ${\mathcal O}({\rm TeV})$. Using the first SM family, we study the spectra of composite particles and contact interactions in quark-lepton and quark-quark sectors. The cross sections and decay rates of composite particles are calculated to study their phenomenologies based on the LHC physics from $pp$ collision at high energy TeV scale. In particular, the processes giving $e^+e^- qq'$ final state are analyzed by using the recast of the experimental upper limit on $\sigma(pp\rightarrow eeqq')$ to set bounds on the parameters of composite particles and their contact interactions. We determine that a composite fermion, $(F)$, of mass $m_F$ below 4.25 TeV can be excluded for $\Lambda$ = $m_F$. At the same time, we compute 3$\sigma$ and 5$\sigma$ contour plots of the statistical significance and highlight the phase space in which $F$ can manifest using 3 ab$^{-1}$, foreseen at the high luminosity LHC (HL-LHC). This result shows that there is a vast range of model parameters to which a dedicated search can be sensitive to $F$ composite fermions and we thus encourage such efforts in future investigations at the LHC. Moreover, we further consider other decay channels of composite fermions and, in particular, we find that the case of $Q$-resonances can lead to a four jets final state (a triplet of jets produced in association with another jet). This signature, to the best of our knowledge, has escaped the realm of the searches at the LHC and can offer a new possibility to search for composite fermions and physics beyond the SM. The detailed phenomenology of the $Q$-resonances is the subject of an ongoing work. The phenomenology of two boosted SM gauge bosons Eqs. (\[pgg\],\[Epegg\],\[Npnvgg\],\[Dpqgg\],\[Upqgg\]), and two gluons fusing into a Higgs boson in final states of composite boson and fermion decays will be further studied.
It is an interesting question to see how these phenomenologies can possibly account for some recent results obtained in both space and underground laboratories. The cosmic rays $pp$ collisions produce composite particles $E$ that decay into electrons and positrons. This may explain an excess of cosmic ray electrons and positrons around TeV scale [@changjin2008; @dark2009]. In addition, recent AMS-02 results [@AMS2015] show that at TeV scale the energy-dependent proton flux changes its power-law index. This implies that there would be “excess” TeV protons whose origin could be also explained by the resonance of composite fermions $N$ due to the interactions of dark-matter and normal-matter particles. These composite fermions should appear as resonances by high-energy sterile neutrinos inelastic collisions with nucleons (xenon) at the largest cross-section, then resonances decay and produce some other detectable SM particles in underground laboratories [@pandaX]. Similarly, in the ICECUBE experiment [@icecube], we expect events where the neutrinos change their directions (lower their energies) by their inelastic collisions to form the resonances of composite fermions $N$ at a high energy scale ($\approx$ TeV). Similarly to the analogy between the Higgs mechanism and BCS superconductivity, the composite-particle counterparts in condensed matter physics have been recently discussed [@KX2017].
Acknowledgements
================
The work of Alfredo Gurrola and Francesco Romeo is supported in part by NSF Award PHY-1806612. The work of Hao Sun is supported by the National Natural Science Foundation of China (Grant No.11675033).
[99]{}
Y. Nambu and G. Jona-Lasinio, Phys. Rev. 122 (1961) 345. F. Englert, R. Brout, Phys. Rev. Lett. 13 (1964) 321;\
P. W. Higgs, Phys. Lett. 12 (1964) 132; Phys. Rev. Lett. 13 (1964) 508; Phys. Rev. 145 (1966) 1156;\
G. S. Guralnik, C. R. Hagen, T. W. B. Kibble, Phys. Rev. Lett. 13 (1964) 585; and T. W. B. Kibble, Phys. Rev. 155 (1967) 1554.
ATLAS Collaboration, JINST 3 (2008) S08003
CMS Collaboration, JINST 3 (2008) S08004
ATLAS Collaboration, Phys. Lett. B 716 (2012) 1, and http://atlas.ch/.
CMS Collaboration, Phys. Lett. B 716 (2012) 30-61.
M. Aaboud [*et al.*]{} \[ATLAS Collaboration\], JHEP [**1609**]{}, 001 (2016) doi:10.1007/JHEP09(2016)001 \[arXiv:1606.03833 \[hep-ex\]\].
V. Khachatryan [*et al.*]{} \[CMS Collaboration\], Phys. Rev. Lett. [**117**]{}, no. 5, 051802 (2016) doi:10.1103/PhysRevLett.117.051802 \[arXiv:1606.04093 \[hep-ex\]\]. G. Aad et al. (ATLAS Collaboration), Phys. Rev. D 87, 015010 (2013).
V. Khachatryan [*et al.*]{} \[CMS Collaboration\], Eur. Phys. J. C [**76**]{}, no. 5, 237 (2016) doi:10.1140/epjc/s10052-016-4067-z \[arXiv:1601.06431 \[hep-ex\]\]. G. Aad [*et al.*]{} \[ATLAS Collaboration\], JHEP [**1512**]{}, 055 (2015) doi:10.1007/JHEP12(2015)055 \[arXiv:1506.00962 \[hep-ex\]\]. V. Khachatryan [*et al.*]{} \[CMS Collaboration\], Phys. Rev. D [**91**]{}, no. 5, 052009 (2015) doi:10.1103/PhysRevD.91.052009 \[arXiv:1501.04198 \[hep-ex\]\]. V. Khachatryan [*et al.*]{} \[CMS Collaboration\], JHEP [**1408**]{}, 173 (2014) doi:10.1007/JHEP08(2014)173 \[arXiv:1405.1994 \[hep-ex\]\]. G. Aad [*et al.*]{} \[ATLAS Collaboration\], Phys. Rev. Lett. [**115**]{}, no. 13, 131801 (2015) doi:10.1103/PhysRevLett.115.131801 \[arXiv:1506.01081 \[hep-ex\]\]. V. Khachatryan [*et al.*]{} \[CMS Collaboration\], JHEP [**1506**]{}, 121 (2015) doi:10.1007/JHEP06(2015)121 \[arXiv:1504.03198 \[hep-ex\]\]. J.C. Pati, A. Salam, J.A. Strathdee, Are quarks composite?, Phys. Lett. B 59 (1975) 265, https://doi.org/10.1016/0370-2693(75)90042-8;\
H. Harari, Composite models for quarks and leptons, Phys. Rep. 104 (1984) 159, https://doi.org/10.1016/0370-1573(84)90207-2;\
O.W. Greenberg, C.A. Nelson, Composite models of leptons, Phys. Rev. D 10 (1974) 2567, https://doi.org/10.1103/PhysRevD.10.2567.
P. A. M. Dirac, Scientific American 208, 45 (1963). H. Terazawa, K. Akama, and Y. Chikashige, Phys. Rev. D15, 480 (1977). H. Terazawa, Phys.Rev. D22, 184 (1980). E. Eichten and K. Lane, Physics Letters B 90, 125 (1980). E. Eichten, K. D. Lane, and M. E. Peskin, Phys. Rev. Lett. 50, 811 (1983). H. Terazawa, in Europhysics Topical Conference: Flavor Mixing in Weak Interactions Erice, Italy, March 4-12, 1984 (1984).
N. Cabibbo, L. Maiani, and Y. Srivastava, Phys. Lett. B139, 459 (1984).
U. Baur, M. Spira, and P. M. Zerwas, Phys. Rev. D 42, 815 (1990).
U. Baur, I. Hinchliffe, and D. Zeppenfeld, Int. J. Mod. Phys. A2, 1285 (1987).
R. Leonardi, O. Panella, and L. Fanò, “Doubly charged heavy leptons at LHC via contact interactions”, Phys. Rev. D 90, 035001, (2014).
R. Leonardi, L. Alunni, F. Romeo, L. Fanò, O. Panella, “Hunting for heavy composite Majorana neutrinos at the LHC” Eur. Phys. J. C76 (2016) no.11, 593, arXiv:1510.07988 .
The CMS Collaboration, “Search for a heavy composite Majorana neutrino in the final state with two leptons and two quarks at $\sqrt{s}=13$ TeV”, Phys. Lett. B 775, 315 (2017) arXiv:1706.08578.
H. B. Nielsen and M. Ninomiya, Nucl. Phys. B 185 (1981) 20 \[Erratum ibid. B 195 (1982) 541\]; Phys. Lett. B 105 (1981) 219; Int. J. Mod. Phys. A 6 (1991) 2913.
S.-S. Xue, “Quantum Regge calculus of Einstein-Cartan theory”, Phys. Lett. B 682 (2009) 300,\[arXiv:0902.3407\];\
“Detailed discussions and calculations of quantum Regge calculus of Einstein-Cartan theory”, Phys. Rev. D 82 (2010) 064039 \[arXiv:0912.2435\];\
“The phase structure of Einstein-Cardan theory”, Phys. Lett. B 665 (2008) 54 \[arXiv:0804.4619\];\
“The Phase and Critical Point of Quantum Einstein-Cartan Gravity”, Phys. Lett. B 711 (2012) 404 \[arXiv:1112.1323\].
S.-S. Xue, JHEP 11 (2016) 072, arXiv:1605.01266. For more details, see Refs. [@xuemass].
S.-S. Xue, JHEP 05(2017)146, arXiv:1601.06845. For more details, see Refs. [@xue3bound].
S.-S. Xue, Phys. Rev. D93, 073001 (2016), arXiv:1506.05994; Phys. Lett. B 721 (2013) 347, arXiv:1301.4254; Mod. Phys. Lett. A, Vol. 14 (1999) 2701, [*ibid*]{} Vol.15 (2000) 1089; Phys. Lett. B 398 (1997) 177; [*ibid*]{} B224, (1989) 309; B245, (1990), 565; Z. Phys. C 50, (1991) 145; [*Nuovo Cimento A*]{}, [**105**]{} (1992) 131; [*ibid*]{} [**105**]{} (1992) 1225; Nucl. Phys. B (Proc. Suppl.) 53 (1997) 688, [*ibid*]{} 63A-C (1998) 596, [*ibid*]{} 94 (2001) 781; The proceedings of international workshop “Mass and Mixings of Quarks and Leptons”, Shizuoka, Japan, March 19-21, 1997, Edited by Yoshio Koide\
G. Preparata and S.-S. Xue, Phys. Lett. B 264 (1991) 35, Phys. Lett. B 302 (1993) 442; B325 (1994) 161-165; B377 (1996) 124- 128.
S.-S. Xue, Phys. Lett. B [**381**]{}, 277 (1996), Nucl. Phys. B [**486**]{}, 282 (1997), [*ibid*]{} 580, 365 (2000); Phys. Rev. D 61, 054502 (2000), [*ibid*]{} 64, 094504 (2001); J. Phys. G, Nucl. Part. Phys. 29 (2003) 2381; Phys. Lett. B 706 (2011) 213; [*ibid*]{} 665 (2008) 54; https://arxiv.org/abs/hep-lat/0006024.
S.-S. Xue, Phys. Lett. B737 (2014) 172 (arXiv:1405.1867); [*ibid*]{} B744 (2015) 88 (arXiv:1501.06844); B727 (2013) 308 (arXiv:1308.6486\]).
R. D. Ball [*et al.*]{} \[NNPDF Collaboration\], JHEP [**1504**]{} (2015) 040 doi:10.1007/JHEP04(2015)040 \[arXiv:1410.8849 \[hep-ph\]\]. J. deFavereau et al., JHEP 1402, 057 (2014).
See for example, J. Chang, [*et al.*]{} Nature Vol 456|20 Nov. 2008; Y-Z. Fan, B. Zhang and J. Chang, IJMPD Vol. 19, (2010) 2011.
N. Arkani-Hamed, D. P. Finkbeiner, T. Slatyer, and N. Weiner, Phys. Rev. D 79, 015014 (2009);\
Dmitry Malyshev, Ilias Cholis, and Joseph Gelfand, Phys. Rev. D80, 063005 (2009).
AMS collaboration, Phys. Rev. Lett. 114, 171103 (2015). https://www.lngs.infn.it/en and http://pandax.physics.sjtu.edu.cn/.
https://icecube.wisc.edu/.
H. Kleinert and S.-S. Xue, “Composite Fermions and their pair states in a Strongly-Coupled Fermion Liquid”, Nuclear Physics B 936 (2018) 352–363, https://arxiv.org/abs/1708.04023.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, we first give some new characterizations of Muckenhoupt type weights through establishing the boundedness of maximal operators on the weighted Lorentz and Morrey spaces. Secondly, we establish the boundedness of sublinear operators including many interesting in harmonic analysis and its commutators on the weighted Morrey spaces. Finally, as an application, the boundedness of strongly singular integral operators and commutators with symbols in BMO space are also given.'
address:
- 'Institute of mathematics, Vietnamese Academy of Science and Technology, Hanoi, Vietnam.'
- 'School of Mathematics, Mientrung University of Civil Engineering, Phu Yen, Vietnam.'
- 'School of Mathematics, University of Transport and Communications, Ha Noi, Vietnam.'
author:
- Nguyen Minh Chuong
- Dao Van Duong
- Kieu Huu Dung
title: MAXIMAL OPERATORS AND SINGULAR INTEGRALS ON THE WEIGHTED LORENTZ AND MORREY SPACES
---
[^1]
Introduction {#section1}
============
Let $f$ be a locally integrable function on $\mathbb R^n$. The Hardy-Littlewood maximal operator of $f$ is defined by $$\begin{aligned}
M(f)(x)=\sup\limits_{Q}\frac{1}{|Q|}\int_Q|f(y)|dy,\;\;x\in \mathbb R^n,\end{aligned}$$ where the supremum is taken over all cubes containing $x$. It is well known that the Hardy-Littlewood maximal operator is one of the most important operators and plays a key role in harmonic analysis since maximal operators could control crucial quantitative information concerning the given functions. It is very a powerful tool for solving crucial problems in analysis, for example, applications to differentiation theory, in the theory of singular integral operators and partial differential equations (see [@FS1971], [@St1993], [@T1986] for more details). It is very important to study weighted estimates for maximal operators in harmonic analysis. B. Muckenhoupt [@Mu1972] first discovered the weighted norm inequality for the Hardy-Littlewood maximal operators in the real setting. More precisely, it is proved that for $1<p<\infty$, $$\begin{aligned}
\label{maximal function}
\int\limits_{\mathbb R^n}\left|M(f)(x)\right|^p\omega(x)dx
\leq C \int\limits_{\mathbb R^n}\left|f(x)\right|^p\omega(x)dx,\end{aligned}$$ holds for all $f$ in the weighted Lebesgue space $L^p(\omega(x)dx)$ if and only if $\omega$ belongs to the class of Muckenhoupt weights denoted by $A_p$. Later, Coifman and Fefferman [@CF1974] extended the theory of Muckenhoupt weights to general Calderón-Zygmund operators. They also proved that $A_p$ weights satisfy the crucial reverse Hölder condition. The weighted norm inequalities for the maximal operators are also extended to the vector valued setting by Andersen and John in the work [@AJ1981], and to the Lorentz spaces by Chung, Hunt, and Kurtz in [@CHK1982]. It is well known that the theory of weighted functions plays an important role in the study of boundary value problems on Lipschitz domains, in theory of extrapolation of operators and applications to certain classes of nonlinear partial differential equation. It is also useful to remark that in 2012, Tang [@Ta2012] established the weighted norm inequalities for maximal operators and pseudodifferential operators with smooth symbols associated to the class of new weighted functions $A_p(\varphi)$ (see in Section \[section2\] below for more details) including the Muckenhoupt weighted functions. It should be pointed out that the class of $ A_p(\varphi)$ weights do not satisfy the doubling condition. It is well known that Morrey [@Mo1938] introduced the classical Morrey spaces to study the local behavior of solutions to second order elliptic partial differential equations. Moreover, it is found that many properties of solutions to partial differential equations can be attributed to the boundedness of some operators on Morrey spaces. Also, the Morrey spaces have many important applications to Navier-Stokes and Schrödinger equations, elliptic equations with discontinuous coefficients and potential theory (see, for example, [@Adams1975], [@Caffarelli1988], [@Fan1998], [@Mazzucato03], [@Ruiz1991], [@T1992] and therein references). During last decades, the theory of Morrey spaces has been significantly developed into different contexts, including the study of classical operators of harmonic analysis, for instance, maximal functions, potential operators, singular integrals, pseudodifferential operators, Hausdorff operators and their commutators in generalizations of these spaces (see [@AGL2000], [@Ch2018], [@Guliyev2011], [@G2016], [@KS2009]). Especially, Wang, Zhou and Chen [@WZC2017] recently have established the interesting connection between the $A_p$ weights and Morrey spaces. More precisely, some new characterizations of Muckenhoupt weights are given by replacing the Lebesgue spaces by the Morrey spaces. Motivated by all of the above mentioned facts, the first main of this paper is to give some new characterizations of Muckenhoupt type weights such as $A_p$, $A(p,1)$, and $A_p(\varphi)$ by establishing the boundedness of maximal operators on the weighted Morrey and Lorentz spaces. In particular, we give the weighted norm inequality of weak type for new dyadic maximal operators associated to the $A_p^{\Delta,\eta}(\varphi)$ dyadic weights. The results are given in Section \[section3\] of the paper. The second main of this paper is to study the boundedness of sublinear operators including many interesting operators in harmonic analysis, such as the Calderón-Zygmund operator, Hardy-Littlewood maximal operator, strongly singular integrals, and so on, on the weighted Morrey spaces. Let us first give the definition of sublinear operators with strongly singular kernels. Let the operator $\mathcal{T}$ be well defined on the space of all infinitely differential functions with compact support $C^\infty_c(\mathbb R^n)$. It is said that $\mathcal{T}$ is a strongly singular sublinear operator if it is a linear or sublinear operator and satisfies the size condition as follows $$\begin{aligned}
\label{ineq-sub}
\left|\mathcal{T}f(x)\right|\leq C\int_{\mathbb R^n}\frac{|f(y)|}{|x-y|^{n+\lambda}}dy, \text{\; for\; a.e \;} x\not\in \text{supp}{f},\end{aligned}$$ for all $f\in C^\infty_c(\mathbb R^n)$, where $\lambda$ is a non-negative real number. For a measurable function $b$, the commutator operator $[b, \mathcal{T}]$ is defined as a linear or a sublinear operator such that $$\begin{aligned}
\label{ineq-sub-com}
\left|[b, \mathcal{T}]f(x)\right|\leq C\int_{\mathbb R^n}\frac{|f(y)||b(x)-b(y)|}{|x-y|^{n+\lambda}}dy, \text{\; for\; a.e \;} x\not\in \text{supp}{f},\end{aligned}$$ for every $f\in C^\infty_c(\mathbb R^n)$. For $\lambda\leq 0$, the sublinear operators $\mathcal{T}$ and $[b, \mathcal{T}]$ have been investigated by many authors. For example, see in the works [@Guliyev2011], [@Kokilashvili2016], [@Soria1994] and therein references. In the Section \[section4\] of the paper, we establish the boundedness of sublinear operators $\mathcal{T}$ and $[b, \mathcal{T}]$ for $\lambda\geq 0$ on the weighted Morrey type spaces. As an application, we obtain some new results about boundedness of strongly singular integral operators and their commutators with symbols in BMO space on the weighted Morrey spaces. Moreover, maximal singular integral operators of Andersen and John type are studied on the two weighted Morrey spaces with vector valued functions in Section \[section4\].
Some notations and definitions {#section2}
==============================
Throught the whole paper, we denote by $C$ a positive geometric constant that is independent of the main parameters, but can change from line to line. We also write $a\lesssim b$ to mean that there is a positive constant $C$, independent of the main parameters, such that $a \le Cb$. The symbol $f\simeq g$ means that f is equivalent to g (i.e. $C^{-1} f\leq g \leq Cf)$. As usual, $\omega(\cdot)$ is a non-negative weighted function on $\mathbb{R}^n$. Denote $\omega(B)^{\alpha }=\big(\int_B\omega(x)dx\big)^{\alpha}$, for $\alpha\in\mathbb R$. Remark that if $\omega(x) = x^{\beta}$ for $\beta > -n$, then we have $$\begin{aligned}
\label{ineq-power}
\omega(B_r(0))=\int_{B_r(0)}|x|^{\beta}dx\simeq r^{\beta+n}.\end{aligned}$$ We also denote by $B_r(x_0)=\{x\in\mathbb R^n:|x-x_0|<r\}$ a ball of radius $r$ with center at $x_0$, and let $rB$ define the ball with the same center as $B$ whose radius is $r$ times radius of $B$.
Now, we are in a position to give some notations and definitions of weighted Morrey spaces.
Let $1 \le q < \infty, 0 < \kappa < 1$ and $\omega_1$ and $\omega_2$ be two weighted functions. Then two weighted Morrey space is defined by $$\begin{aligned}
\mathcal{B}^{q,\kappa}_{\omega_1,\omega_2}(\mathbb{R}^n)=\{f\in L^q_{\omega_2,{\rm loc}}(\mathbb{R}^n):\|f\|_{\mathcal{B}^{q,\kappa}_{\omega_1,\omega_2}(\mathbb{R}^n)}<\infty \},\end{aligned}$$ where $$\begin{aligned}
\|f\|_{\mathcal{B}^{q,\kappa}_{\omega_1,\omega_2}(\mathbb{R}^n)}=\sup\limits_{\rm ball\,B} \Big(\frac{1}{\omega_1(B)^{\kappa}}\int_{B}|f(x)|^q\omega_2(x)dx \Big)^{\frac{1}{q}}.\end{aligned}$$
It is easy to see that $\mathcal{B}^{q,\kappa}_{\omega_1,\omega_2}(\mathbb{R}^n)$ is a Banach space. Note that if $\omega_1=\omega, \omega_2=1$, we then write $\mathcal B^{q,\kappa}(\omega,\mathbb R^n):=\mathcal B^{q,\kappa}_{\omega_1,\omega_2}(\mathbb R^n)$. Also, if $\omega_1=\omega_2=\omega$, then we denote $\mathcal B^{q,\kappa}
_\omega(\mathbb R^n):= {\mathcal B}^{q,\kappa}_{\omega_1,\omega_2}(\mathbb R^n)$. In particular, for $\omega=1$ we write $\mathcal B^{q,\kappa}
(\mathbb R^n):=\mathcal B^{q,\kappa}_\omega(\mathbb R^n)$.
Let $1 \le q < \infty, 0 < \kappa < 1$. The local Morrey space is defined by $$\begin{aligned}
\mathcal{B}^{q,\kappa}_{\rm loc}(\mathbb{R}^n)=\{f\in L^q_{{\rm loc}}(\mathbb{R}^n):\|f\|_{\mathcal{B}^{q,\kappa}_{\rm loc}(\mathbb{R}^n)}<\infty \},\end{aligned}$$ where $$\begin{aligned}
\|f\|_{\mathcal{B}^{q,\kappa}_{\rm loc}(\mathbb{R}^n)}=\sup\limits_{x\in\mathbb{R}^n,0<R<1} \Big(\frac{1}{|B_R(x)|^{\kappa}}\int_{B_R(x)}|f(y)|^q dy \Big)^{\frac{1}{q}}.\end{aligned}$$
Note that for $1\leq q\leq p<\infty$, the local Morrey space $\mathcal{B}^{q,1-\frac{q}{p}}_{\rm loc}(\mathbb{R}^n)$ has some important applications to the Navier-Stokes equations and other evolution equations (see in [@Fe1993; @T1992] for more details).
Let $1 \le q < \infty$ and $0 < \kappa < 1$. The weighted inhomogeneous Morrey space is defined by $$\begin{aligned}
{B}^{q,\kappa}_\omega(\mathbb{R}^n)=\{f\in L^q_{{\omega, \rm loc}}(\mathbb{R}^n):\|f\|_{{B}^{q,\kappa}_\omega(\mathbb{R}^n)}<\infty \},\end{aligned}$$ where $$\begin{aligned}
\|f\|_{{B}^{q,\kappa}_{\omega}(\mathbb{R}^n)}=\sup\limits_{x\in\mathbb{R}^n,R\geq 1} \Big(\frac{1}{\omega(B_R(x))^{\kappa}}\int_{B_R(x)}|f(x)|^q\omega(x) dx \Big)^{\frac{1}{q}}.\end{aligned}$$
If $\omega=1$ and $1\leq q\leq p<\infty$, then the inhomogeneous Morrey space ${B}^{q,1-\frac{q}{p}}_\omega(\mathbb{R}^n)$ is introduced by Alvarez, Guzmán-Partida and Lakey (see in [@AGL2000] for more details). Note that $\mathcal{B}^{q,\kappa}_{\rm loc}(\mathbb{R}^n)$ and ${B}^{q,\kappa}_\omega(\mathbb{R}^n)$ are two Banach spaces.
Let $1 \le q < \infty, 0 < \kappa < 1$ and $\omega$ be a weighted function. The weighted Morrey space is defined by $$\begin{aligned}
\mathcal{L}^{q,\kappa}_{\omega}(\mathbb{R}^n)=\{f\in L^q_{\omega,{\rm loc}}(\mathbb{R}^n):\|f\|_{\mathcal{L}^{q,\kappa}_{\omega}(\mathbb{R}^n)}<\infty \},\end{aligned}$$ where $$\begin{aligned}
\|f\|_{\mathcal{L}^{q,\kappa}_{\omega}(\mathbb{R}^n)}=\sup\limits_{\rm cube ~ Q} \Big(\frac{1}{\omega(Q)^k}\int_{Q}|f(x)|^q\omega(x) dx \Big)^{\frac{1}{q}}.\end{aligned}$$
From this, for convenience, we denote $M^p_{q,\omega}(\mathbb{R}^n):=\mathcal{L}^{q,1-\frac{q}{p}}_{\omega}(\mathbb{R}^n)$ for the case $0 < q < p < \infty$.
Let $0 < q \le p < \infty$ and $\omega$ be a weighted function. Then the weighted weak Morrey space is defined by $$\begin{aligned}
WM^{p}_{q,\omega}(\mathbb{R}^n)=\{f\in L^q_{\omega,{\rm loc}}(\mathbb{R}^n):\|f\|_{WM^{p}_{q,\omega}(\mathbb{R}^n)}<\infty \},\end{aligned}$$ where $$\begin{aligned}
\|f\|_{WM^{p}_{q,\omega}(\mathbb{R}^n)}=\sup\limits_{\rm cube ~ Q}\frac{1}{\omega(Q)^{\frac{1}{q}-\frac{1}{p}}}\sup\limits_{\lambda>0}\lambda \Big(\int_{\{x\in Q:|f(x)|>\lambda \}}\omega(x) dx \Big)^{\frac{1}{q}}.\end{aligned}$$
For a measurable function $f$ on $\mathbb{R}^n$, the distribution function of $f$ associated with the measure $\omega(x)dx$ is defined as follows $$\begin{aligned}
d_f(\alpha)=\omega\left(\{x\in \mathbb{R}^n: |f(x)|>\alpha \}\right).\end{aligned}$$
The decreasing rearrangement of $f$ with respect to the measure $\omega(x)dx$ is the function $f^*$ defined on $[0, \infty)$ as follows $$\begin{aligned}
f^*(t)=\inf \{s>0:d_f(s)\le t \}.\end{aligned}$$
(Section 2 in [@CHK1982]). Let $0 < p, q \le \infty$. The weighted Lorentz space $L^{p,q}_\omega (\mathbb{R}^n)$ is defined as the set of all measurable functions $f$ such that $\|f\|_{L^{p,q}_\omega (\mathbb{R}^n)}<\infty$, where $$\begin{aligned}
\|f\|_{L^{p,q}_\omega (\mathbb{R}^n)}=\begin{cases}
\left(\frac{q}{p}\int_0^\infty \left[t^{\frac{1}{p}}f^*(t) \right]^q\frac{dt}{t} \right)^{\frac{1}{q}}, &{\rm if} ~ 0<q<\infty,\\
\sup\limits_{t>0}t^{\frac{1}{p}}f^*(t), &{\rm if}~ q=\infty.\end{cases}\end{aligned}$$
Remark that if either $1 < p < \infty$ and $1 \le q \le \infty$, or $p = q = 1$, or $p = q = \infty$ then $L^{p,q}_\omega (\mathbb{R}^n)$ is a quasi-Banach space. Moreover, there is a constant $C > 0$ such that $$\begin{aligned}
\label{Lor-ineq}
C^{-1}\|f\|_{L^{p,q}_\omega (\mathbb{R}^n)}\le \sup\limits_{\|g\|_{L^{p',q'}_\omega(\mathbb R^n)}\le 1}\left|\int_{\,\mathbb{R}^n}f(x)g(x)\omega(x)dx \right|\le C\|f\|_{L^{p,q}_\omega (\mathbb{R}^n)}.\end{aligned}$$
\[Cor2.3-WZC2017\] [(page 253 in [@H1966] and Corollary 2.3 in [@WZC2017])]{} If $0 < r < q < p < \infty$, $1\leq q_1\leq q_2\leq \infty$ and $\omega$ is a non-negative weighted function on $\mathbb{R}^n$, then there exists a constant $C > 0$ such that $$\begin{aligned}
C\|\cdot\|_{M^p_{r,\omega}(\mathbb{R}^n)}&\le \|\cdot\|_{WM^p_{q,\omega}(\mathbb{R}^n)}\le \|\cdot\|_{M^p_{q,\omega}(\mathbb{R}^n)}\le \|\cdot\|_{WM^p_{p,\omega}(\mathbb{R}^n)}\nonumber
\\
&=\|\cdot\|_{L^{p,\infty}_\omega (\mathbb{R}^n)}\leq \|\cdot\|_{L^{p,q_2}_\omega (\mathbb{R}^n)}\leq \|\cdot\|_{L^{p,q_1}_\omega (\mathbb{R}^n)}.\end{aligned}$$
Next, we present some basic facts on the class of weighted functions $A(p, 1)$ with $1 < p < \infty$. For further information on the weights, the interested readers may refer to the work [@CHK1982]. The weighted function $\omega(x)$ is in $A(p, 1)$ if there exists a positive constant $C$ such that for any cube $Q$, we have $$\begin{aligned}
\|\chi_Q \|_{L^{p,1}_\omega(\mathbb{R}^n)} \|\chi_Q\omega^{-1} \|_{L^{p',\infty}_\omega (\mathbb{R}^n)}\le C|Q|.\end{aligned}$$
\[Lem2.8-CHK1982\] [(Lemma 2.8 in [@CHK1982])]{} For $1\leq p<\infty$, we have $\omega \in A(p, 1)$ if and only if there exists a constant $C$ such that for any cube $Q$ and subset $E \subset Q$, $$\begin{aligned}
\frac{|E|}{|Q|}\le C\left(\frac{\omega(E)}{\omega(Q)} \right)^{\frac{1}{p}}.\end{aligned}$$
Remark that if $\omega\in A(p,1)$ with $1\leq p<\infty$ and $0<\kappa<1$, then $\mathcal{B}^{p,\kappa}_\omega(\mathbb R^n)=\mathcal{L}^{p,\kappa}_{\omega}(\mathbb{R}^n)$ with equivalence of norms.\
Let $1 \le r < \infty$ and $\vec{f}=\{f_k\}$ be a sequence of measurable functions on $\mathbb{R}^n$. We denote $$\begin{aligned}
|\vec{f}(x)|_r=\left(\sum\limits_{k=1}^\infty |f_k(x)|^r \right)^{\frac{1}{r}}.\end{aligned}$$
As usual, the vector-valued space $X(\ell^r,\mathbb{R}^n)$ is defined as the set of all sequences of measurable functions $\vec{f}=\{f_k\}$ such that $$\begin{aligned}
\|\vec{f} \|_{X(\ell^r,\mathbb{R}^n)}=\||\vec{f}(\cdot)|_r \|_X<\infty,\end{aligned}$$ where $X$ is an appropriate Banach space.
Let us recall to define the BMO spaces of John and Nirenberg. For further information on these spaces as well as their deep applications in harmonic analysis, one can see in the famous book of Stein [@St1993].
The bounded mean oscillation space $BMO(\mathbb{R}^n)$ is defined as the set of all functions $b\in L^1_{\rm loc}(\mathbb{R}^n)$ such that $$\begin{aligned}
\|b\|_{BMO(\mathbb{R}^n)}=\sup\limits_{\rm cube ~Q}\frac{1}{|Q|}\int_Q |b(x)-b_Q|dx<\infty,\end{aligned}$$ where $b_Q=\frac{1}{|Q|}\int_Q b(x)dx$.
\[BMO-Lemma\] [([@St1993])]{} If $1<p<\infty$, we then have $$\begin{aligned}
\|b\|_{BMO(\mathbb{R}^n)}\simeq \sup\limits_{\rm cube ~Q}\Big(\frac{1}{|Q|}\int_Q |b(x)-b_Q|^pdx\Big)^{\frac{1}{p}}:=\|b\|_{BMO^p(\mathbb R^n)}.\end{aligned}$$
\[Pro-T1986\] [(Proposition 3.2 in [@T1986])]{} If $b\in BMO(\mathbb R^n)$, then $$|b_{2^{j+1}B}-b_B|\leq 2^n(j+1)\|b\|_{BMO(\mathbb R^n)},\,\textit{\rm for all}\, j\in\mathbb N.$$
Let us recall the definition of $A_p$ weights. For further readings on $A_p$ weights, the reader may find in the interesting book [@Grafakos2008].
Let $1 < p < \infty$. It is said that a weight $\omega \in A_p(\mathbb{R}^n)$ if there exists a constant $C$ such that for all cubes $Q$, $$\begin{aligned}
\left(\frac{1}{|Q|}\int_Q \omega(x)dx \right) \left(\frac{1}{|Q|}\int_Q\omega(x)^{-\frac{1}{p-1}}dx \right)^{p-1}\le C.\end{aligned}$$
A weight $\omega \in A_1(\mathbb{R}^n)$ if there is a constant $C$ such that $$\begin{aligned}
M(\omega)(x)\le C\omega(x), \;{ \rm for \; a. e} ~x\in \mathbb{R}^n.\end{aligned}$$ We denote $A_\infty(\mathbb{R}^n)=\mathop\cup\limits_{1\le p<\infty}A_p(\mathbb{R}^n)$.\
A closing relation to $A_\infty(\mathbb{R}^n)$ is the reverse Hölder condition. If there exist $r > 1$ and a fixed constant $C$ such that $$\begin{aligned}
\Big(\frac{1}{|B|}\int_B \omega(x)^rdx \Big)^{\frac{1}{r}}\le \frac{C}{|B|}\int_B \omega(x) dx,\end{aligned}$$ for all balls $B \subset \mathbb{R}^n$, we then say that $\omega$ satisfies the reverse Hölder condition of order $r$ and write $\omega\in RH_r (\mathbb{R}^n)$. According to Theorem 19 and Corollary 21 in [@IMS2015], $\omega\in A_\infty (\mathbb{R}^n)$ if and only if there exists some $r > 1$ such that $\omega\in RH_r (\mathbb{R}^n)$. Moreover, if $\omega\in RH_r (\mathbb{R}^n),r>1$, then $\omega\in RH_{r+\varepsilon} (\mathbb{R}^n)$ for some $\varepsilon>0$. We thus write $r_\omega = \sup\{r > 1 : \omega\in RH_r (\mathbb{R}^n)\}$ to denote the critical index of $\omega$ for the reverse Hölder condition.
\[rever-Holder\] Let $\omega \in A_p(\mathbb{R}^n) \cap RH_r(\mathbb{R}^n), p \ge 1$ and $r > 1$. Then, there exist two constants $C_1, C_2 > 0$ such that $$\begin{aligned}
C_1\left( \frac{|E|}{|B|}\right)^p\le \frac{\omega(E)}{\omega(B)} \le C_2\left( \frac{|E|}{|B|}\right)^{\frac{r-1}{r}},\end{aligned}$$ for any ball $B$ and for any measurable subset $E$ of $B$.
\[pro2.4DFan\] If $\omega\in A_p(\mathbb R^n)$, $1 \leq p < \infty$, then for any $f\in L^1_{\rm loc}(\mathbb R^n)$ and any ball $B \subset \mathbb R^n$, we have $$\dfrac{1}{|B|}\int_{B}|f(x)|dx\leq C\Big(\dfrac{1}{\omega(B)}\int_{B}|f(x)|^p\omega(x)dx \Big)^{\frac{1}{p}}.$$
Next, we write $\omega\in \Delta_2$, the class of doubling weights, if there exists $D > 0$ such that for any cube $Q$, we have $$\begin{aligned}
\omega(2Q)\le D\omega(Q).\end{aligned}$$ It is known that if $\omega\in A_\infty(\mathbb{R}^n)$ then $\omega\in \Delta$. Now, let us recall the class of $ A_p(\varphi)$ weights proposed by Tang in the work [@Ta2012].
Let $1 < p < \infty$ and $\varphi(t)=(1+t)^{\alpha_0}$ for $\alpha_0>0$ and $t\ge 0$. We say that a weight $\omega\in A_p(\varphi)$ if there exists a constant $C$ such that for all cubes $Q$, $$\begin{aligned}
\Big(\frac{1}{\varphi(|Q|)|Q|}\int_Q \omega(x)dx \Big).\Big(\frac{1}{\varphi(|Q|)|Q|}\int_Q\omega(x)^{-\frac{1}{p-1}}dx \Big)^{p-1}\le C.\end{aligned}$$ A weight $\omega\in A_1(\varphi)$ if there is a constant $C$ such that $$M_\varphi(f)(x)\le C\omega(x), {\rm\; for\; a. e} ~x\in \mathbb{R}^n,$$ where $$\begin{aligned}
M_\varphi(f)(x)=\sup\limits_{x\in {\rm cube }~Q}\frac{1}{\varphi(|Q|)|Q|}\int_Q |f(y)|dy.\end{aligned}$$
Denote $A_\infty(\varphi)=\mathop\cup\limits_{1\le p<\infty} A_p(\varphi)$. It is useful to remark that the $ A_p(\varphi)$ weights do not satisfy the doubling condition. For instance, $\omega(x)=(1+|x|)^{(-n+\eta)}$ for $0\leq\eta\leq n\alpha_0$ is in $A_1(\varphi)$, but not in $A_p$ weights and $\omega(x)dx$ is not a doubling measure. It is also important to see that $M_\varphi$ may be not bounded on the weighted Lebesgue spaces $L^p_\omega(\mathbb R^n)$ for every $\omega\in A_p(\varphi)$. To be more precise, see in Lemma 2.3 in [@Ta2012]. Similarly, in this paper we also introduce a class of dyadic weighted functions associated to the function $\varphi$ as follows.
Let $1<p<\infty$ and $0<\eta<\infty$. A weight $\omega\in A_p^{\Delta,\eta}(\varphi)$ if there exists a constant $C$ such that for all dyadic cubes $Q$, $$\begin{aligned}
\Big(\frac{1}{\varphi(|Q|)^{\eta}|Q|}\int_Q \omega(x)dx \Big).\Big(\frac{1}{\varphi(|Q|)^{\eta}|Q|}\int_Q\omega(x)^{-\frac{1}{p-1}}dx \Big)^{p-1}\le C.\end{aligned}$$
It is obvious that $A_{p_1}^{\Delta,\eta}(\varphi)\subset A_{p_2}^{\Delta,\eta}(\varphi)$ for all $1<p_1<p_2<\infty$. It is also easy to show that $A_p(\mathbb{R}^n)\subset A_p(\varphi)\subset A_p^{\Delta,\eta}(\varphi)$ with $1<p<\infty$ and $0< \eta<\infty$. In particular, $A_1(\mathbb R^n) \subset A_1(\varphi)$.
Next, we give the definitions of the maximal operators $M_\omega$ and $M^\Delta_{\varphi,\eta}$ as follows $$M_\omega(f)(x)=\sup\limits_{x\in\text{ball\;} B}\frac{1}{\omega(5B)}\int_B|f(y)|\omega(y)dy,$$ $$M^\Delta_{\varphi,\eta}(f)(x)=\sup\limits_{x\in {\rm dyadic\,cube}\,Q}\frac{1}{\varphi(|Q|)^\eta |Q|}\int_Q |f(y)|dy, {\;\rm for~all}~0<\eta<\infty.$$ Remark that by the similar arguments to Lemma 2.1 in [@Ta2012], we also have $$M^\Delta_{\varphi,\eta}(f)(x)\lesssim \left(M_\omega(|f|^p)(x)\right)^{\frac{1}{p}}, \;\;x\in\mathbb R^n,$$ where $\omega\in A_{p}^{\Delta,\eta}(\varphi)$ for $0<\eta<\infty$ and $1<p<\infty$. Moreover, we also get the same result as in Lemma 2.3 of the paper [@Ta2012].
\[Lemma-dyadic\] Let $1<p<\infty$ and $\omega\in A_p^{\Delta,\eta}(\varphi)$. Then, for any $p<r<\infty$, we have $$\|M^\Delta_{\varphi,\eta}(f)\|_{L^{r}_\omega(\mathbb R^n)}\leq C \|f\|_{L^{r}_\omega(\mathbb R^n)}.$$
It seems to see that the inequality in Lemma \[Lemma-dyadic\] may be not valid for $r=p$.
\[Theo3.1-AJ1981\] [(Theorem 3.1 in [@AJ1981])]{} If $1 < p < \infty$, then the operator $M$ is bounded from $L^p_\omega(\ell^r, \mathbb{R}^n)$ to itself if and only if $\omega\in A_p$.
\[Theo2.12-CDD2017\] [(Theorem 2.12 in [@CDD2017])]{} If $1 < p < r < \infty$, then the operator $M$ is bounded from $L^{p,1}_\omega(\ell^r, \mathbb{R}^n)$ to $L^{p,\infty}_\omega(\ell^r, \mathbb{R}^n)$ if and only if $\omega\in A(p, 1)$.
\[Lem2.3-Ta2012\] [(Lemma 2.3 in [@Ta2012])]{} If $1 \le p < \infty$, then the operator $M_\varphi$ is bounded from $L^p_\omega(\mathbb{R}^n)$ to $L^{p,\infty}_\omega(\mathbb{R}^n)$ if and only if $\omega\in A_p(\varphi)$.
In 1981, Andersen and John [@AJ1981] established the weighted norm inequalities for vector-valued maximal functions and maximal singular integrals on the space $L^p_\omega(\ell^r,\mathbb R^n)$. Now, let us recall the definition of maximal singular integrals associated to the kernels due to Andersen and John. For more details, see in the work [@AJ1981].
Let $K$ be the kernel such that $$\begin{aligned}
|K(x)|&\le \frac{A}{|x|^n}, |\hat{K}(x)|\le A;\\
|K(x-y)-K(x)|&\le \mu (|y|/|x|)|x|^{-n}, \text{for all} ~|x|\ge 2|y|;\end{aligned}$$ where $A$ is a constant and $\mu$ is non-decreasing on the positive real half-line, $\mu(2t)\le C\mu (t)$ for all $t > 0$, and satisfies the Dini condition $$\begin{aligned}
\int_0^1\frac{\mu(t)}{t}dt<\infty.\end{aligned}$$ Then, the maximal singular integral operator $T^*$ is defined by $$\begin{aligned}
T^*(f)(x)= \mathop{\rm sup}\limits_{\varepsilon >0}\Big|\int_{\,\,|x-y|\geq \varepsilon}K(x-y)f(y)dy\Big|.\end{aligned}$$
If $\{K_k (x)\}$ denote a sequence of singular convolution kernels satisfying the above conditions (2.3)-(2.5) with a uniform constant $A$ and a fixed function $\mu$ not dependent of $k$, then we write $T^*(\vec{f})=\{T^*_k(f_k) \}$, where $T^*_k$ is the operator above corresponding to the kernel $K_k$ .
\[Theo-AJ1981\] [(Theorem 5.2 in [@AJ1981])]{} Let $1 < r < \infty, 1 < p < \infty,$ and suppose $\omega\in A_p$. There exits a constant $C$ such that $$\begin{aligned}
\|T^*(\vec{f}) \|_{L^p_\omega(\ell^r,\mathbb{R}^n)}\le C\|\vec{f} \|_{L^p_\omega(\ell^r,\mathbb{R}^n)}, \text{ for all} ~f\in L^p_\omega(\ell^r,\mathbb{R}^n). \end{aligned}$$
Let $b$ be a measurable function. We denote by $\mathcal{M}_b$ the multiplication operator defined by $\mathcal{M}_bf (x)=b(x) f (x)$ for any measurable function $f$. If $\mathcal{H}$ is a linear or sublinear operator on some measurable function space, the commutator of Coifman-Rochberg-Weiss type formed by $\mathcal{M}_b$ and $\mathcal{H}$ is defined by $[\mathcal{M}_b, \mathcal{H}]f (x)=(\mathcal{M}_b\mathcal{H}-\mathcal{H}\mathcal{M}_b) f (x)$.
The results about the boundedness of maximal operators {#section3}
======================================================
By using Theorem \[Theo3.1-AJ1981\] and estimating as Theorem 1.1 in [@WZC2017], we immediately have the following useful characterization for the Muckenhoupt weights through boundedness of the Hardy-Littlewood maximal operators on the vector valued function spaces.
Let $1 < q < p < \infty, 1 \le r < \infty$. Then, the following statements are equivalent:
1. $\omega\in A_p$;
2. $M$ is a bounded operator from $L^p_\omega(\ell^r,\mathbb{R}^n)$ to $L^{p,\infty}_\omega(\ell^r,\mathbb{R}^n)$;
3. $M$ is a bounded operator from $L^p_\omega(\ell^r,\mathbb{R}^n)$ to $M^{p}_{q,\omega}(\ell^r,\mathbb{R}^n)$;
4. $M$ is a bounded operator from $L^p_\omega(\ell^r,\mathbb{R}^n)$ to $WM^{p}_{q,\omega}(\ell^r,\mathbb{R}^n)$.
Now, we give a new characterization for the class of $A(p,1)$ weights.
Let $1 < q < p < r < \infty$. The following statements are equivalent:
1. $\omega\in A(p,1)$;
2. $M$ is a bounded operator from $L^{p,1}_\omega(\ell^r,\mathbb{R}^n)$ to $L^{p,\infty}_\omega(\ell^r,\mathbb{R}^n)$;
3. $M$ is a bounded operator from $L^{p,1}_\omega(\ell^r,\mathbb{R}^n)$ to $M^{p}_{q,\omega}(\ell^r,\mathbb{R}^n)$;
4. $M$ is a bounded operator from $L^{p,1}_\omega(\ell^r,\mathbb{R}^n)$ to $WM^{p}_{q,\omega}(\ell^r,\mathbb{R}^n)$.
Note that Theorem \[Theo2.12-CDD2017\] follows us to obtain the equivalence of (1) and (2). By Corollary \[Cor2.3-WZC2017\], we immediately have (2) $\Rightarrow$ (3) $\Rightarrow$ (4). Therefore, to complete the proof of the theorem, we need to prove (4) $\Rightarrow$ (1). For any cube $Q$, by the relation (\[Lor-ineq\]), we find a function $f$ such that $\|f\|_{L^{p,1}_\omega(\mathbb{R}^n)}\le 1$ and $$\begin{aligned}
\label{ineq-Ap1}
\int_Q|f(x)|dx\ge \left|\,\int_{\mathbb{R}^n}f(x)\chi_Q\omega^{-1}\omega dx \right|\gtrsim \|\chi_Q\omega^{-1} \|_{L^{p',\infty}_\omega(\mathbb{R}^n)}.\end{aligned}$$ It is obvious that $Q=\{x\in Q: M(f)(x)>\lambda \}$, where $\lambda=\frac{1}{2|Q|}\int_Q|f(x)|dx$. Thus, because $M$ is a bounded operator from $L^{p,1}_\omega(\mathbb{R}^n)$ to $WM^{p}_{q,\omega}(\mathbb{R}^n)$, we have $$\begin{aligned}
\lambda\omega(Q)^{\frac{1}{p}}&=\frac{1}{\omega(Q)^{\frac{1}{q}-\frac{1}{p}}}\lambda\Big(\int_{\{x\in Q: M(f)(x)>\lambda \}}\omega(x)dx \Big)^{\frac{1}{q}}\\
&\le \|M(f) \|_{WM^p_{q,\omega}(\mathbb{R}^n)}\le \|f\|_{L^{p,1}_\omega(\mathbb{R}^n)}\le 1.\end{aligned}$$ As a consequence, by (\[ineq-Ap1\]), we give $$\begin{aligned}
\frac{1}{2|Q|}\|\chi_Q\omega^{-1} \|_{L^{p',\infty}_\omega(\mathbb{R}^n)}.\omega(Q)^{\frac{1}{p}}\lesssim 1.\end{aligned}$$ From this, we have $$\begin{aligned}
\|\chi_Q\|_{L^{p,1}_\omega(\mathbb{R}^n)}\|\chi_Q\omega^{-1} \|_{L^{p',\infty}_\omega(\mathbb{R}^n)}\lesssim |Q|.\end{aligned}$$ This implies that $\omega\in A(p,1)$, and the theorem is completely proved.
Next, we establish the boundedness results for pseudo-differential operators of order $0$ on weighted Lorentz spaces. For $m\in\mathbb R$, we say that the function $a(x,\xi)\in C^{\infty}(\mathbb R^n\times \mathbb R^n)$ is a symbol of order $m$ if it satisfies the following inequality $$|\partial^{\beta}_x\partial^{\alpha}_\xi a(x,\xi)|\leq C_{\alpha,\beta}(1+|\xi|)^{m-|\alpha|},$$ for all multi-indices $\alpha$ and $\beta$, where $C_{\alpha,\beta} > 0$ is independent of $x$ and $\xi$. Then, a pseudo-differential operator is a mapping $f\to T_{a}(f)$ given by $$\begin{aligned}
T_a(f)(x)=\int_{\mathbb R^n} a(x,\xi)\widehat f(\xi)e^{2\pi i x\xi}d\xi.\end{aligned}$$ Remark that $T_a$ is well defined on the space of Schwartz functions $S(\mathbb R^n)$ or on the space of all infinitely differentiable functions with compact support $C^{\infty}_c(\mathbb R^n)$, where $\widehat f$ is the Fourier transform of the function $f$.
\[pseudo1\] Let $1 < q \leq p < \infty$, $\omega\in A_p(\mathbb R^n)$ and $T_a$ be a pseudo-differential operator of order $0$. Then, $T_a$ extends to a bounded operator from $L^{p,q}_\omega(\mathbb R^n)$ to $L^{p,\infty}_\omega(\mathbb R^n)$.
By Theorem 2 and Theorem 4 in [@CHK1982], we have $$\begin{aligned}
\|M(f)\|_{L^{p,\infty}_\omega(\mathbb R^n)}\lesssim \|f\|_{L^{p,q}_\omega(\mathbb R^n)},\,\textit{\rm for all}\,f\in C^{\infty}_c(\mathbb R^n).\nonumber\end{aligned}$$ Next, by estimating as Theorem 2 in [@G2016], we see that $$|T_a(f)(\cdot)|\lesssim M(f)(\cdot),\,\textit{\rm for all}\,f\in C^{\infty}_c(\mathbb R^n).$$ Thus, $$\begin{aligned}
\|T_a(f)\|_{L^{p,\infty}_\omega(\mathbb R^n)}\lesssim \|f\|_{L^{p,q}_\omega(\mathbb R^n)},\,\textit{\rm for all}\,f\in C^{\infty}_c(\mathbb R^n).\nonumber\end{aligned}$$ As mentioned above, since $C^{\infty}_c(\mathbb R^n)$ is dense in $L^{p,q}_\omega(\mathbb R^n)$ (see Corollary 3.2 in [@NTY2004]), we immediately have the desired result.
By using Lemma \[pseudo1\] and Corollary \[Cor2.3-WZC2017\] and applying the Lorentz version Marcinkiewicz interpolation theorem as the proof of Theorem 3 in [@CHK1982], we obtain the following useful result.
\[pseudo2\] Let $1 < p < \infty$, $1<q\leq \infty$, $\omega\in A_p(\mathbb R^n)$ and $T_a$ be a pseudo-differential operator of order $0$. Then, the following statements are true:
1. $T_a$ extends to a bounded operator from $L^{p,q}_\omega(\mathbb R^n)$ to $L^{p,q}_\omega(\mathbb R^n)$;
2. $T_a$ extends to a bounded operator from $L^{p,q}_\omega(\mathbb{R}^n)$ to $M^{p}_{q,\omega}(\mathbb{R}^n)$;
3. $T_a$ extends to a bounded operator from $L^{p,q}_\omega(\mathbb{R}^n)$ to $WM^{p}_{q,\omega}(\mathbb{R}^n)$.
For $1 < p < \infty$, by Lemma \[Lem2.8-CHK1982\], we observe that $\omega\in A(p,1)$ implies $\omega\in \Delta$. Thus, combining with Theorem 3.1 in [@KS2009], we can get the following result.
If $1 < p < \infty, 0 < \kappa < 1, \omega\in A(p,1)$, then the operator $M_\omega$ is bounded on $\mathcal{L}^{q,\kappa}_\omega(\mathbb{R}^n)$.
Similarly to the known characterizations of the $A_p$ weights given in [@WZC2017], we also have another characterizations for the $A_p(\varphi)$ weights as follows.
\[max-phi\] Let either $1 < q < p < \infty$ or $0 < q < p = 1$. Then, the following statements are equivalent:
1. $\omega\in A_p(\varphi)$;
2. $M_\varphi$ is a bounded operator from $L^{p}_\omega(\mathbb{R}^n)$ to $L^{p,\infty}_\omega(\mathbb{R}^n)$;
3. $M_\varphi$ is a bounded operator from $L^{p}_\omega(\mathbb{R}^n)$ to $M^{p}_{q,\omega}(\mathbb{R}^n)$;
4. $M_\varphi$ is a bounded operator from $L^{p}_\omega(\mathbb{R}^n)$ to $WM^{p}_{q,\omega}(\mathbb{R}^n)$.
By Lemma \[Lem2.3-Ta2012\] and Corollary \[Cor2.3-WZC2017\], it is clear that the relation (1) $\Leftrightarrow$ (2) and (2) $\Rightarrow$ (3) $\Rightarrow$ (4). Thus, to complete the proof, we need to prove the (4) $\Rightarrow$ (1). More precisely, it is as the following. In the case $1 < q < p < \infty$, let $Q$ be any cube and take $f_\varepsilon=(\omega+\varepsilon)^{1-p'}\chi_Q$, for all $\varepsilon>0$, where $p'$ is a conjugate real number of $p$, i.e $\frac{1}{p}+\frac{1}{p'}=1$. It immediately follows that $f_\varepsilon\in L^p_\omega(\mathbb{R}^n)$. For any $0 < \lambda < \frac{(\omega+\varepsilon)^{1-p'}(Q)}{\varphi(|Q|)|Q|}$, by letting $x\in Q$, it is clear to see that $$\begin{aligned}
M_\varphi(f_\varepsilon)(x)\ge \frac{1}{\varphi(|Q|)|Q|}\int_Q|f_\varepsilon(y)|dy=\frac{1}{\varphi(|Q|)|Q|}\int_Q (\omega+\varepsilon)^{1-p'}dy>\lambda.\end{aligned}$$ Hence, we obtain $$\begin{aligned}
Q=\{x\in Q: M_\varphi(f_\varepsilon)(x)>\lambda \}.\end{aligned}$$ Consequently, because $M_\varphi$ is a bounded operator from $L^p_\omega(\mathbb{R}^n)$ to $WM^p_{q,\omega}(\mathbb{R}^n)$, we infer $$\begin{aligned}
\label{omegaQ-weak}
\lambda\omega(Q)^{\frac{1}{p}}&=\frac{1}{\omega(Q)^{\frac{1}{q}-\frac{1}{p}}}\lambda\Big(\int_{\{x\in Q: M_\varphi(f_\varepsilon)(x)>\lambda \}}\omega(x)dx \Big)^{\frac{1}{q}}\notag\\
&\le \|M_\varphi(f_\varepsilon) \|_{WM^p_{q,\omega}(\mathbb{R}^n)}\lesssim \|f_\varepsilon\|_{L^{p}_\omega(\mathbb{R}^n)}=\Big(\int_Q (\omega+\varepsilon)^{-p'}\omega(x)dx \Big)^{\frac{1}{p}}.\end{aligned}$$ Thus, by choosing $\lambda=\frac{(\omega+\varepsilon)^{1-p'}(Q)}{2\varphi(|Q|)|Q|}$, we get $$\begin{aligned}
\Big(\frac{1}{\varphi(|Q|)|Q|} \int_Q (\omega+\varepsilon)^{-p'}\omega(x)dx\Big)^p.\Big(\int_Q\omega(x)dx \Big)\lesssim \int_Q (\omega+\varepsilon)^{-p'}\omega(x)dx,\end{aligned}$$ which implies that $$\begin{aligned}
\Big(\frac{1}{\varphi(|Q|)|Q|} \int_Q\omega(x)dx\Big) \Big( \frac{1}{\varphi(|Q|)|Q|} \int_Q (\omega+\varepsilon)^{-p'}\omega(x)dx\Big)^{p-1}\lesssim 1, \end{aligned}$$ for all $\varepsilon>0$. By letting $\varepsilon\to 0^+$ and using dominated convergence theorem of Lebesgue, we obtain $\omega\in A_p(\varphi)$. In the case $0 < q < p = 1$, let us fix $Q$ and take any cube $Q_1 \subset Q$. Thus, we choose $f = \chi_{Q_1}$. For any $0 < \lambda <\frac{|Q_1|}{\varphi(|Q|)|Q|}$, by estimating as (\[omegaQ-weak\]) above, we immediately have $$\begin{aligned}
\lambda\Big( \int_{Q}\omega(x)dx \Big)\le \int_{Q_1}\omega(x)dx.\end{aligned}$$ Next, by choosing $\lambda=\frac{|Q_1|}{2\varphi(|Q|)|Q|}$, we infer $$\begin{aligned}
\frac{1}{|Q|}\int_Q\omega(x)dx\lesssim \frac{1}{|Q_1|}\int_{Q_1}\omega(x)dx, \text{ for any} ~Q_1\subset Q.\end{aligned}$$ Hence, by the definition of operator $M_\varphi$ and the Lebesgue differentiation theorem, it follows that $$\begin{aligned}
M_\varphi(\omega)(x)\lesssim \omega(x), \text{ for \;a.e.}\; x\in\mathbb{R}^n,\end{aligned}$$ which gives $\omega\in A_1(\varphi)$.
In final part of this section, we give the weighted norm inequality of weak type for new dyadic maximal operators $M^\Delta_{\varphi,2\eta}$ on the vector valued Lebesgue spaces with weighted functions in $A_p^{\Delta,\eta}(\varphi)$.
\[max-Delta-2eta\] If $1 < p < r < \infty, \omega \in A_p^{\Delta,\eta}(\varphi)$ for $\eta>0$, then operator $M^\Delta_{\varphi,2\eta}$ is bounded from $L^p_\omega(\ell^r,\mathbb{R}^n)$ to $L^{p,\infty}_\omega(\ell^r,\mathbb{R}^n)$.
Let $\vec{f}\in \vec S$ and $\alpha > 0$, where $\vec S$ the linear space of sequences $\vec f = \{f_k\}$ such that each $f_k(x)$ is a simple function on $\mathbb R^n$ and $f_k(x)\equiv 0$ for all sufficiently large $k$. By using Lemma 2.5 in [@Ta2012], there exists a disjoint union of maximal dyadic cubes $\{Q_j\}$ such that $$\begin{aligned}
\label{ineq-fr}
|\vec{f}(x)|_r \le \alpha,x\notin \Omega=\cup_{j=1}^\infty Q_j;\end{aligned}$$ $$\begin{aligned}
\label{ineq-fr-alpha}
\alpha \le \frac{1}{\varphi(|Q_j|)^\eta|Q_j|}\int_{Q_j}|\vec{f}(x)|_rdx &\le 2^n\varphi(4n).\alpha, \text{for all} ~j\in\mathbb{Z}^+.\end{aligned}$$ Now, we compose $\vec{f}=\vec{f'}+\vec{f^{''}}$, where $\vec{f'}=\{f'_k\},f'_k(x)=f_k(x)\chi_{\mathbb{R}^n\backslash \Omega}(x)$. This gives $$\begin{aligned}
|M^\Delta_{\varphi,2\eta}(\vec{f})(x)|_r\le |M^\Delta_{\varphi,2\eta}(\vec{f'})(x)|_r +|M^\Delta_{\varphi,2\eta}(\vec{f^{''}})(x)|_r.\end{aligned}$$ As a consequence, we need to prove the following two results $$\begin{aligned}
\label{esti-omega1}
\omega\left(\{x\in\mathbb{R}^n: |M^\Delta_{\varphi,2\eta}(\vec{f'})(x)|_r>\alpha \} \right)\lesssim \alpha^{-p}\|\vec{f}\|^p_{L^p_\omega(\ell^r, \mathbb R^n)},\end{aligned}$$ and $$\begin{aligned}
\label{esti-omega2}
\omega\left(\{x\in\mathbb{R}^n: |M^\Delta_{\varphi,2\eta}(\vec{f^{''}})(x)|_r>\alpha \} \right)\lesssim \alpha^{-p}\|\vec{f}\|^p_{L^p_\omega(\ell^r, \mathbb R^n)}.\end{aligned}$$ By Lemma \[Lemma-dyadic\], for $\omega \in A_p^{\Delta,\eta}(\varphi)$ we have $$\int_{\mathbb{R}^n}|M^\Delta_{\varphi,2\eta}(f'_k)(x)|^r\omega(x)dx\leq \int_{\mathbb{R}^n}|M^\Delta_{\varphi,\eta}(f'_k)(x)|^r\omega(x)dx
\lesssim \int_{\mathbb{R}^n}|f'_k(x)|^r\omega(x)dx.$$ This implies that $$\begin{aligned}
\int_{\mathbb{R}^n}|M^\Delta_{\varphi,2\eta}(\vec f')(x)|^r_r\omega(x)dx&=\int_{\mathbb{R}^n}\sum_{k=1}^\infty |M^\Delta_{\varphi,2\eta}(f'_k)(x)|^r\omega(x)dx\\
&=\sum_{k=1}^\infty \int_{\mathbb{R}^n}|M^\Delta_{\varphi,2\eta}(f'_k)(x)|^r\omega(x)dx\\
&\lesssim\sum_{k=1}^\infty \int_{\mathbb{R}^n}|f'_k(x)|^r\omega(x)dx\\
&\lesssim\int_{\mathbb{R}^n}|\vec f'(x)|_r^r\omega(x)dx.\end{aligned}$$ Hence, by the Chebysev inequality, it immediately follows that $$\begin{aligned}
\label{ineq-Chebysev}
\omega\left(\{x\in\mathbb{R}^n: |M^\Delta_{\varphi,2\eta}(\vec{f'})(x)|_r>\alpha \} \right)\lesssim \alpha^{-r}\|\vec{f'}\|^r_{L^r_\omega(\ell^r,\mathbb R^n)}.\end{aligned}$$ On the other hand, by (\[ineq-fr\]), we infer $$\begin{aligned}
|\vec{f'}(x)|_r^r\le \alpha^{r-p}|\vec{f}(x)|_r^p,\end{aligned}$$ which implies that, by (\[ineq-Chebysev\]), the inequality (\[esti-omega1\]) is holded.
It remains only to show that the inequality (\[esti-omega2\]) is true. To estimate the inequality (\[esti-omega2\]), we put $\overline{f}=\{\overline{f}_k \}$ as follows $$\begin{aligned}
\overline{f}_k(x)=\begin{cases}\frac{1}{\varphi(|Q_j|)^\eta|Q_j|}\int_{Q_j}|f_k(y)|dy,& x\in Q_j, j=1,2,...,\\
0,& \text{otherwise}. \end{cases}\end{aligned}$$ Then, we obtain the important inequality as follows $$\begin{aligned}
\label{Fefferman-Stein}
M^\Delta_{\varphi,2\eta}(f^{''}_k)(x)\le M^\Delta_{\varphi,\eta}(\overline{f}_k)(x),x\notin \Omega.\end{aligned}$$ Indeed, let $x\notin \Omega$ and $Q$ be any dyadic cube such that $x \in Q$. Thus, one has $$\begin{aligned}
\int_Q|f^{''}_k(y)|dy=\int_{Q\cap \Omega}|f_k(y)|dy=\sum\limits_{j\in J}\int_{Q\cap Q_j}|f_k(y)|dy,\end{aligned}$$ where $J = \{j \in \mathbb{N} : Q_j \cap Q \ne \emptyset\}$. Since $\{Q_j\}$ and $Q$ are dyadic cubes, and $x \in Q$, we immediately have $J = \{j \in \mathbb{N} : Q_j \subset Q\}$. Hence, we infer $$\begin{aligned}
\label{int_f''}
\int_Q|f^{''}_k(y)|dy=\sum\limits_{j\in J}\int_{Q_j}|f_k(y)|dy.\end{aligned}$$ On the other hand, we get $$\begin{aligned}
\int_{Q_j}\overline{f}_k(y)dy=\int_{Q_j}\Big(\frac{1}{\varphi(|Q_j|)^\eta|Q_j|} \int_{Q_j}|f_k(t)|dt\Big)dy=\frac{1}{\varphi(|Q_j|)^\eta}\int_{Q_j}|f_k(t)|dt.\end{aligned}$$ Therefore, by (\[int\_f”\]), one has $$\begin{aligned}
\frac{1}{\varphi(|Q|)^{2\eta}|Q|}\int_Q|f^{''}_k(y)|dy&=\frac{1}{\varphi(|Q|)^{2\eta}|Q|}\sum\limits_{j\in J}\Big(\varphi(|Q_j|)^\eta\int_{Q_j}\overline{f}_k(y)dy \Big)\nonumber
\\
&=\frac{1}{\varphi(|Q|)^{\eta}|Q|}\sum\limits_{j\in J}\Big(\frac{\varphi(|Q_j|)^\eta}{\varphi(|Q|)^\eta}\int_{Q_j}\overline{f}_k(y)dy \Big)\nonumber
\\
&\le \frac{1}{\varphi(|Q|)^{\eta}|Q|}\int_Q\overline{f}_k(y)dy.\nonumber\end{aligned}$$ This implies that inequality (\[Fefferman-Stein\]) is true.\
Next, for any $x \in \Omega$, there only exists a dyadic cube $Q_j$ such that $x \in Q_j$. Thus, by the Minkowski inequality and (\[ineq-fr-alpha\]), we have $$\begin{aligned}
|\overline{f}(x)|_r&=\Big(\sum\limits_{k=1}^\infty\Big(\frac{1}{\varphi(|Q_j|)^{\eta}|Q_j|}\int_{Q_j}|f_k(y)|dy \Big)^r \Big)^{\frac{1}{r}}\\
&\le \frac{1}{\varphi(|Q_j|)^{\eta}|Q_j|}\int_{Q_j}|\vec{f}(y)|_rdy\le 2^n\varphi(4n).\alpha.\end{aligned}$$ Hence, by using (\[Fefferman-Stein\]) and estimating as (\[ineq-Chebysev\]), it is clear to see that $$\begin{aligned}
&\omega\big(\{x\notin\Omega: |M^\Delta_{\varphi,2\eta}(\vec{f^{''}})(x)|_r>\alpha \} \big)\le \omega\left(\{x\notin\Omega: |M^\Delta_{\varphi,\eta}(\overline{f})(x)|_r>\alpha \} \right)
\\
&\leq \omega\left(\{x\in\mathbb R^n: |M^\Delta_{\varphi,\eta}(\overline{f})(x)|_r>\alpha \} \right)\lesssim \alpha^{-r}\|\overline{f} \|^r_{L^r_\omega(\ell^r,\mathbb R^n)} \lesssim \omega(\Omega),\end{aligned}$$ which leads to $$\begin{aligned}
\label{omega_f"}
&\omega\big(\{x\in\mathbb{R}^n: |M^\Delta_{\varphi,2\eta}(\vec{f^{''}})(x)|_r>\alpha \} \big)\notag\\
&\le \omega(\Omega)+\omega\big(\{x\notin\Omega: |M^\Delta_{\varphi,2\eta}(\vec{f^{''}})(x)|_r>\alpha \}\big)\lesssim \omega(\Omega).\end{aligned}$$ Besides that, by using (\[ineq-fr-alpha\]), the Hölder inequality and $\omega\in A_p^{\Delta,\eta}(\varphi)$, we get $$\begin{aligned}
&\omega(Q_j)\le \alpha^{-p}\Big(\frac{1}{\varphi(|Q_j|)^{\eta}|Q_j|}\int_{Q_j}|\vec{f}(x)|_rdx \Big)^p.\int_{Q_j}\omega(x)dx\\
&\le \alpha^{-p}\Big(\frac{1}{\varphi(|Q_j|)^{\eta}|Q_j|} \Big)^p \Big(\int_{Q_j} |\vec{f}(x)|_r^p\omega(x)dx \Big) \Big(\int_{Q_j}\omega^{-\frac{p'}{p}}(x)dx \Big)^{\frac{p}{p'}}.\int_{Q_j}\omega(x)dx
\\
& \le \alpha^{-p}\Big( \int_{Q_j} |\vec{f}(x)|^p_r\omega(x)dx\Big) \Big(\frac{1}{\varphi(|Q_j|)^{\eta}|Q_j|} \int_{Q_j}\omega(x)dx \Big) \Big(\frac{1}{\varphi(|Q_j|)^{\eta}|Q_j|} \int_{Q_j}\omega(x)^{-\frac{1}{p-1}}dx \Big)^{p-1}\\
&\lesssim \alpha^{-p}\Big( \int_{Q_j} |\vec{f}(x)|^p_r\omega(x)dx\Big), \text{for all} ~j\in\mathbb{N}.\end{aligned}$$ From the above inequality, we infer $$\begin{aligned}
\omega(\Omega)&=\sum\limits_{j=1}^\infty\omega(Q_j)\lesssim \alpha^{-p} \sum\limits_{j=1}^\infty \int_{Q_j} |\vec{f}(x)|^p_r\omega(x)dx=\alpha^{-p}\int_\Omega |\vec{f}(x)|^p_r\omega(x)dx
\\
&\le \alpha^{-p}\int_{\mathbb{R}^n}|\vec{f}(x)|^p_r\omega(x)dx.\end{aligned}$$ As an application, by (\[omega\_f"\]), the proof for the inequality (\[esti-omega2\]) is finished. Finally, since $\vec S$ is dense in $L^p_\omega(\ell^r, \mathbb R^n)$ (see in [@BP1961]), the proof of the theorem is ended.
As a consequence, by combining Theorem \[max-Delta-2eta\], Corollary \[Cor2.3-WZC2017\] and making in the same way as Theorem \[max-phi\], we also obtain a necessary condition and a sufficient condition for the class of $A_p^{\Delta,\eta}(\varphi)$ weights. More precisely, the following is true.
Let $1 < p < r < \infty$ and $\eta >0$. The following statements are true:
1. If $\omega\in A_p^{\Delta,\eta}(\varphi)$, then $M^\Delta_{\varphi,2\eta}$ is a bounded operator from $L^p_\omega(\ell^r,\mathbb{R}^n)$ to $L^{p,\infty}_\omega(\ell^r,\mathbb{R}^n)$, $M^{p}_{q,\omega}(\ell^r,\mathbb{R}^n)$ and $WM^{p}_{q,\omega}(\ell^r,\mathbb{R}^n)$, respectively.
2. If $\omega\notin A_p^{\Delta,\eta}(\varphi)$, then $M^\Delta_{\varphi,\eta}$ is not a bounded operator from $L^p_\omega(\ell^r,\mathbb{R}^n)$ to $WM^{p}_{q,\omega}(\ell^r,\mathbb{R}^n)$.
The results about the boundedness of sublinear operators generated by singular integrals and its commutators {#section4}
============================================================================================================
Let us recall that the two weighted Morrey space $\mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\ell^r,\mathbb R^n)$ with vector-valued functions is defined as the set of all sequences of measurable functions $\vec{f}=\{f_k\}$ such that $$\begin{aligned}
\|\vec{f} \|_{\mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\ell^r,\mathbb R^n)}=\||\vec{f}(\cdot)|_r \|_{\mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\mathbb R^n)}<\infty.\end{aligned}$$
It is not difficult to show that $\mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\ell^r,\mathbb R^n)$ is a Banach space. Our first main result in this section is to give the boundedness of maximal singular integral operators with the kernels proposed by Anderson and John on the space $\mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\ell^r,\mathbb R^n)$. More precisely, we have the following useful result.
Let $1< r<\infty$, $1<p<\infty$, $\omega_1\in A(p,1)$, $\omega_2\in A_p$, $\delta\in (1,r_{\omega_2})$ and $0<\kappa<\frac{\delta-1}{\delta p}$. Then, $T^*$ is a bounded operator on $\mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\ell^r,\mathbb R^n)$.
Let us choose any $\vec f\in \mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\ell^r,\mathbb R^n)$ and ball $B_R(x_0):= B$. Next, we compose $\vec{f}=\vec{f}_1+\vec{f}_2$, where $\vec{f}_1=\{f_{1,k}\}$ such that $f_{1,k}(x)=f_k(x)\chi_{2B}(x)$. This implies that $$\begin{aligned}
\label{J12}
&\frac{1}{\omega_1(B)^{\kappa}}\int_B |T^*(\vec{f})(x)|_r^p\omega_2(x)dx\le \frac{1}{\omega_1(B)^{\kappa}}\int_B |T^*(\vec{f}_1)(x)|_r^p\omega_2(x)dx +\notag\\
&+ \frac{1}{\omega_1(B)^{\kappa}}\int_B |T^*(\vec{f}_2)(x)|_r^p\omega_2(x)dx:=J_1+J_2.\end{aligned}$$ By Theorem \[Theo-AJ1981\] and Lemma \[Lem2.8-CHK1982\], we have $$\begin{aligned}
\label{J1}
J_1&\le \frac{1}{\omega_1(B)^{\kappa}}\int_{\mathbb{R}^n} |T^*(\vec{f}_1)(x)|_r^p\omega_2(x)dx\lesssim \frac{1}{\omega_1(B)^{\kappa}}\int_{2B} |\vec{f}(x)|_r^p\omega_2(x)dx\notag\\
&\le \frac{\omega_1(2B)^{\kappa}}{\omega_1(B)^{\kappa}}\|\vec{f}\|^p_{\mathcal{B}^{p,\kappa}_{\omega_1,\omega_2}(\ell^r,\mathbb{R}^n)}\lesssim \|\vec{f}\|^p_{\mathcal{B}^{p,\kappa}_{\omega_1,\omega_2}(\ell^r,\mathbb{R}^n)}.\end{aligned}$$ Now, for $x \in B$ and $y \in (2B)^c$, it is clear to see that $2R\leq |x_0 - y| \le 2|x - y|$. From this, we get $$|T^*(f_{2,k})(x)|\leq \int_{(2B)^c}\frac{A}{|x-y|^n}|f_k(y)|dy\lesssim \int_{(2B)^c}\frac{1}{|x_0-y|^n}|f_k(y)|dy,$$ for all $k\in\mathbb N$. Hence, by using the Minkowski inequality and the Hölder inequality and assuming $\omega_2\in A_p$, we obtain $$\begin{aligned}
&|T^*(\vec f_2)(x)|_r\lesssim \int_{(2B)^c}\frac{|\vec f(y)|_r}{|x_0-y|^n}dy=\sum\limits_{j=1}^{\infty}\,\int_{2^{j}R\leq |x_0-y|<2^{j+1}R}\frac{|\vec f(y)|_r}{|x_0-y|^n}dy\nonumber
\\
&\lesssim \sum\limits_{j=1}^{\infty}\frac{1}{|2^jB|}\int_{2^{j+1}B}|\vec f(y)|_rdy\leq \sum\limits_{j=1}^{\infty}\frac{1}{|2^jB|}\Big(\int_{2^{j+1}B}|\vec f(y)|_r^p\omega_2(y)dy\Big)^{\frac{1}{p}}\Big(\int_{2^{j+1}B}\omega_2(y)^{1-p'}dy\Big)^{\frac{p-1}{p}}\nonumber
\\
&\lesssim \sum\limits_{j=1}^{\infty}\frac{1}{|2^jB|}\Big(\int_{2^{j+1}B}|\vec f(y)|_r^p\omega_2(y)dy\Big)^{\frac{1}{p}}\frac{|2^{j+1}B|}{\omega_2(2^{j+1}B)^{\frac{1}{p}}}\lesssim \|\vec f\|_{\mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\ell^r,\mathbb R^n)}\sum\limits_{j=1}^{\infty}\frac{\omega_1(2^{j+1}B)^{\frac{\kappa}{p}}}{\omega_2(2^{j+1}B)^{\frac{1}{p}}}.\nonumber\end{aligned}$$ Thus, $$\begin{aligned}
J_2&\lesssim \frac{\omega_2(B)}{\omega_1(B)^{\kappa}} \|\vec f\|^p_{\mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\ell^r,\mathbb R^n)}\Big(\sum\limits_{j=1}^{\infty}\frac{\omega_1(2^{j+1}B)^{\frac{\kappa}{p}}}{\omega_2(2^{j+1}B)^{\frac{1}{p}}}\Big)^{p}=\|\vec f\|^p_{\mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\ell^r,\mathbb R^n)}. \mathcal K^p,\nonumber\end{aligned}$$ where $$\mathcal K= \sum\limits_{j=1}^{\infty}\frac{\omega_1(2^{j+1}B)^{\frac{\kappa}{p}}}{\omega_1(B)^{\frac{\kappa}{p}}}.\frac{\omega_2(B)^{\frac{1}{p}}}{\omega_2(2^{j+1}B)^{\frac{1}{p}}}.$$ Next, by applying Lemma \[Lem2.8-CHK1982\], we have $
\big(\frac{\omega_1(2^{j+1}B)}{\omega_1(B)}\big)^{\frac{\kappa}{p}}\lesssim \big(\frac{|2^{j+1}B|}{|B|}\big)^{\kappa}\lesssim 2^{{(j+1)n\kappa}}.
$ On the other hand, by using Proposition \[rever-Holder\], we infer $$\Big(\frac{\omega_2(B)}{\omega_2(2^{j+1}B)}\Big)^{\frac{1}{p}}\lesssim \Big(\frac{|B|}{|2^{j+1}B|}\Big)^{\frac{(\delta-1)}{\delta.p}}\lesssim 2^{\frac{-(j+1)n(\delta-1)}{\delta p}}.$$ Hence, by $\kappa<\frac{\delta-1}{\delta p}$, one has $
\mathcal K\lesssim \sum\limits_{j=1}^{\infty} 2^{(j+1)n(\kappa-\frac{\delta-1}{\delta p})}<\infty.
$ Thus, $$J_2\lesssim \|\vec f\|^p_{\mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\ell^r,\mathbb R^n)}.$$ Combining this with (\[J12\]) and (\[J1\]) above, we obtain $$\|T^*(\vec f)\|_{\mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\ell^r, \mathbb R^n)}\lesssim \|\vec f\|_{\mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\ell^r, \mathbb R^n)},\,\textit{\rm for all }\, \vec f\in \mathcal B^{p,\kappa}_{\omega_1,\omega_2}(\ell^r, \mathbb R^n),$$ which implies the proof of the theorem is finished.
Our second main result in this section is to establish the boundedness of sublinear operators generated by strongly singular operators on certain weighted Morrey spaces. As an application, we obtain the boundedness of some strongly singular integral operators on the weighted Morrey spaces.
Let us recall the definition of the weighted central Morrey spaces. Let $1 \le q < \infty, 0 < \kappa < 1$ and $\omega$ be a weighted function. Then the weighted central Morrey spaces is defined as the set of all functions in $L^q_{{\rm loc}}(\mathbb{R}^n)$ such that $$\begin{aligned}
\|f\|_{\mathcal{{\mathop B\limits^.}}^{q,\kappa}(\omega, \mathbb{R}^n)}=\sup\limits_{R>0} \Big(\frac{1}{\omega(B_R(0))^{\kappa}}\int_{B_R(0)}|f(x)|^qdx \Big)^{\frac{1}{q}}<\infty.\end{aligned}$$ It is evident that $\mathcal{{\mathop B\limits^.}}^{q,\kappa}(\omega, \mathbb{R}^n)$ is a Banach space. We denote by $\mathfrak{\mathop B\limits^.}^{q,\kappa}(\omega, \mathbb{R}^n)$ the closure of $L^{q}(\mathbb R^n)\cap\mathcal{{\mathop B\limits^.}}^{q,\kappa}(\omega, \mathbb{R}^n)$ with respect to the norm in $\mathcal{{\mathop B\limits^.}}^{q,\kappa}(\omega, \mathbb{R}^n)$. We also recall that the central weighted local Morrey spaces $\mathcal{\mathop B\limits^.}^{q,\kappa}_{\rm loc}(\omega,\mathbb{R}^n)$ as the set of all functions in $L^q_{{\rm loc}}(\mathbb{R}^n)$ such that $$\begin{aligned}
\|f\|_{\mathcal{{\mathop B\limits^.}}^{q,\kappa}_{\rm loc}(\omega, \mathbb{R}^n)}=\sup\limits_{0<R<1} \Big(\frac{1}{\omega(B_R(0))^{\kappa}}\int_{B_R(0)}|f(x)|^qdx \Big)^{\frac{1}{q}}<\infty.\end{aligned}$$
\[Theo-sublinear1\] Let $1<p<\infty$, $\lambda>0$, $0<\kappa<1$, and $\omega(x)=|x|^{\beta}$ for $-n+\frac{\lambda p}{\kappa}<\beta< \frac{\lambda p +(1-\kappa)n}{\kappa}$ and $\kappa_1 \in (0,\kappa-\frac{\lambda p}{n+\beta}]$. Then, the following is true:
[(i)]{} If $\mathcal{T}$ extends to a bounded operator on $L^p(\mathbb R^n)$, then $\mathcal{T}$ can also extend to a bounded operator from $\mathfrak{\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)$ to $\mathcal{{\mathop B\limits^.}}^{p,\kappa_1}_{\rm loc}(\omega, \mathbb{R}^n)$.
[(ii)]{} Let $b \in L^{\eta}_{\rm loc}(\mathbb R^n)\cap BMO(\mathbb R^n)$ with $\eta >p'$. If the commutator $[b, \mathcal{T}]$ extends to a bounded operator on $L^p(\mathbb R^n)$, then it can also extend to a bounded operator from $\mathfrak{\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)$ to $\mathcal{{\mathop B\limits^.}}^{p,\kappa_1}_{\rm loc}(\omega, \mathbb{R}^n)$.
It is sufficient to prove the theorem for all $f\in L^p(\mathbb R^n)\cap\mathcal{{\mathop B\limits^.}}^{p,\kappa}(\omega, \mathbb{R}^n)$.
\(i) By fixing a ball $B_R(x_0):=B$ (for $x_0=0$), with $0<R<1$ and decomposing $f=f_1+f_2$, where $f_1= f.\chi_{2B}$, one has $$\begin{aligned}
\label{I12-strong}
\frac{1}{\omega(B)^{\kappa_1}}\int_{B}|\mathcal{T}(f)(x)|^pdx &\leq \frac{1}{\omega(B)^{\kappa_1}}\int_{B}|\mathcal{T}(f_1)(x)|^pdx +\nonumber
\\
&\,\,\,+\frac{1}{\omega(B)^{\kappa_1}}\int_{B}|\mathcal{T}(f_2)(x)|^pdx:=I_1+I_2.\end{aligned}$$ To estimate $I_1$, by $\mathcal{T}$ extends to a bounded operator on $L^p(\mathbb R^n)$ and the inequality (\[ineq-power\]), we have $$\begin{aligned}
\label{I1-strong}
I_1&\leq \frac{1}{\omega(B)^{\kappa_1}}\int_{\mathbb R^n}|\mathcal{T}(f_1)(x)|^pdx \lesssim \frac{1}{\omega(B)^{\kappa_1}}\int_{2B}|f(x)|^pdx \leq \frac{\omega(2B)^{\kappa}}{\omega(B)^{\kappa_1}}\|f\|^p_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)}\nonumber
\\
&\lesssim R^{(n+\beta)(\kappa-\kappa_1)}\|f\|^p_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)}\leq \|f\|^p_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)}.\end{aligned}$$ On the other hand, by $f_2\in L^{p}(\mathbb R^n)$, one has that $g_m =f.\chi_{(2B)^c\cap {(2mB)}}\to f_2$ in $L^p(\mathbb R^n)$. Thus, by $\mathcal{T}$ bounded on $L^p(\mathbb R^n)$ again, there exists a subsequence $(\mathcal T(g_{m_k}))$-denoted by $(\mathcal T(g_m))$ such that $\mathcal T(g_m)\to \mathcal T(f_2)$ a.e on $\mathbb R^n$. From this, by having $\mathcal{T}$ still satisfies (\[ineq-sub\]) on $L^{p}_{\rm {comp}} (\mathbb R^n)$ and letting $x\in B$ with $m$ large enough, we obtain $$\begin{aligned}
\label{Nakai}
|{\mathcal{T}}{(f_2)}(x)|=\mathop{\rm lim}\limits_{m\to \infty} |{\mathcal{T}}{(g_m)}(x)| \lesssim \mathop{\rm lim}\limits_{m\to \infty} \int_{\mathbb R^n}\frac{|g_m(y)|}{|x-y|^{n+\lambda}}dy= \int_{(2B)^c}\frac{|f(y)|}{|x-y|^{n+\lambda}}dy.\end{aligned}$$ Notice that let $x\in B$ and $y\in (2B)^c$, we have $2R\leq |x_0 - y|\leq 2|x - y|$. This implies that $$\begin{aligned}
\label{Tf2}
|\mathcal{T}(f_{2})(x)|&\lesssim \int_{(2B)^c}\frac{1}{|x_0-y|^{n+\lambda}}|f(y)|dy=\sum\limits_{j=1}^{\infty}\,\int_{2^{j}R\leq |x_0-y|<2^{j+1}R}\frac{|f(y)|}{|x_0-y|^{n+\lambda}}dy\nonumber
\\
&\lesssim \sum\limits_{j=1}^{\infty}\frac{1}{|2^jB|^{(1+\frac{\lambda}{n})}}\int_{2^{j+1}B}|f(y)|dy.\end{aligned}$$ From this, by the Hölder inequality, we deduce $$\begin{aligned}
\label{pre-I2}
|\mathcal{T}(f_{2})(x)|&\lesssim \sum\limits_{j=1}^{\infty}\frac{1}{|2^jB|^{(1+\frac{\lambda}{n})}}\Big(\int_{2^{j+1}B}|f(y)|^pdy\Big)^{\frac{1}{p}}|2^{j+1}B|^{\frac{1}{p'}}.\nonumber
\\
&\lesssim \|f\|_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)}\sum\limits_{j=1}^{\infty}\frac{\omega(2^{j+1}B)^{\frac{\kappa}{p}}|2^{j+1}B|^{\frac{1}{p'}}}{|2^{j}B|^{(1+\frac{\lambda}{n})}}.\end{aligned}$$ As a consequence, by (\[ineq-power\]), we give $$\begin{aligned}
\label{I2-strong}
I_2&\lesssim \frac{|B|}{\omega(B)^{\kappa_1}} \|f\|^p_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)}\Big(\sum\limits_{j=1}^{\infty}\frac{\omega(2^{j+1}B)^{\frac{\kappa}{p}}.|2^{j+1}B|^{\frac{1}{p'}}}{|2^{j}B|^{(1+\frac{\lambda}{n})}}\Big)^{p}\nonumber
\\
&\lesssim \|f\|^p_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega, \mathbb R^n)}\Big(\sum\limits_{j=1}^{\infty}2^{j(\frac{\kappa(n+\beta)-n}{p}-\lambda)}.R^{\frac{(n+\beta)(\kappa-\kappa_1)}{p}-\lambda}\Big)^p\lesssim \|f\|^p_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)}.\end{aligned}$$ Therefore, by (\[I12-strong\]) and (\[I1-strong\]), we immediately have $$\|\mathcal{T}(f)\|_{\mathcal {\mathop B\limits^.}^{p,\kappa_1}_{\rm loc}(\omega,\mathbb R^n)}\lesssim \|f\|_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)},\,\textit{\rm for all}\, f\in L^p(\mathbb R^n)\cap\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n),$$ which gives that the proof of part (i) is ended.
\(ii) As the proof of part (i) above, we also fix a ball $B_R(x_0):=B$ (for $x_0=0$) with $0<R<1$, and write $f=f_1+f_2$ with $f_1= f.\chi_{2B}$. Thus, we get $$\begin{aligned}
\label{K12-strong}
&\frac{1}{\omega(B)^{\kappa_1}}\int_{B}|[b,\mathcal{T}](f)(x)|^pdx \leq \frac{1}{\omega(B)^{\kappa_1}}\int_{B}|[b,\mathcal{T}](f_1)(x)|^pdx +\nonumber
\\
&\,\,\,+\frac{1}{\omega(B)^{\kappa_1}}\int_{B}|[b,\mathcal{T}](f_2)(x)|^pdx:=K_1+K_2.\end{aligned}$$ Next, by $[b, \mathcal{T}]$ extends to a bounded operator on $L^p(\mathbb R^n)$ and the relation (\[ineq-power\]) again, we obtain $$\begin{aligned}
\label{K1-strong}
K_1&\leq \frac{1}{\omega(B)^{\kappa_1}}\int_{\mathbb R^n}|[b,\mathcal{T}](f_1)(x)|^pdx \lesssim \frac{\|b\|^p_{BMO(\mathbb R^n)}}{\omega(B)^{\kappa_1}}\int_{2B}|f(x)|^pdx
\nonumber
\\
&\leq \frac{\omega(2B)^{\kappa}.\|b\|^p_{BMO(\mathbb R^n)}}{\omega(B)^{\kappa_1}}\|f\|^p_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)}\lesssim \|b\|^p_{BMO(\mathbb R^n)}.\|f\|^p_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)}.\end{aligned}$$ Next, by $b\in L^{\eta}_{\rm loc}(\mathbb R^n)$ with $\eta> p'$ and the inequality (\[ineq-sub-com\]), we get $$\begin{aligned}
\left|[b,\mathcal{T}](g)(x)\right|\lesssim \int_{\mathbb R^n}\frac{|g(y)|.|b(x)-b(y)|}{|x-y|^{n+\lambda}}dy, \text{\; a.e \;} x\not\in \text{supp}{(g)},\forall \, g\in L^{p}_{\rm comp}(\mathbb R^n).\nonumber\end{aligned}$$ Thus, by estimating as (\[Nakai\]) above and letting $x\in B$ and $y\in (2B)^c$, we have $$\begin{aligned}
|[b,\mathcal{T}](f_{2})(x)|&\lesssim \int_{(2B)^c}\frac{1}{|x-y|^{n+\lambda}}|f(y)|.|b(x)-b(y)|dy
\nonumber
\\
&\leq \int_{(2B)^c}\frac{1}{|x_0-y|^{n+\lambda}}|f(y)|.|b(x)-b(y)|dy\nonumber
\\
&\leq \Big(\int_{(2B)^c}\frac{|f(y)|}{|x_0-y|^{n+\lambda}}dy\Big)|b(x)-b_B|+\int_{(2B)^c}\frac{|f(y)|.|b_B-b(y)|}{|x_0-y|^{n+\lambda}}dy.\nonumber\end{aligned}$$ This leads to that $$\begin{aligned}
\label{K21-K22}
K_2&\lesssim \frac{1}{\omega(B)^{\kappa_1}}\Big(\int_{(2B)^c}\frac{|f(y)|}{|x_0-y|^{n+\lambda}}dy\Big)^p.\Big(\int_{B}|b(x)-b_B|^pdx\Big)+\nonumber
\\
&\,\,\,+\frac{|B|}{\omega(B)^{\kappa_1}}\Big(\int_{(2B)^c}\frac{|f(y)|.|b_B-b(y)|}{|x_0-y|^{n+\lambda}}dy\Big)^p :=K_{2,1}+K_{2,2}.\end{aligned}$$ For the term $K_{2,1}$, by using (\[Tf2\]), (\[pre-I2\]), (\[I2-strong\]) and Lemma \[BMO-Lemma\], we infer $$\begin{aligned}
\label{K21-strong}
K_{2,1}&\lesssim \frac{|B|}{\omega(B)^{\kappa_1}}\|f\|^p_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)}\Big(\sum\limits_{j=1}^{\infty}\frac{\omega(2^{j+1}B)^{\frac{\kappa}{p}}.|2^{j+1}B|^{\frac{1}{p'}}}{|2^{j}B|^{(1+\frac{\lambda}{n})}}\Big)^{p}\Big(\frac{1}{|B|}\int_{B}|b(x)-b_B|^pdx\Big)\nonumber
\\
&\lesssim \|f\|^p_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)}.\|b\|^p_{BMO(\mathbb R^n)}.\end{aligned}$$ For the term $K_{2,2}$, by the Hölder inequality, we have $$\begin{aligned}
K_{2,2}&\lesssim \frac{|B|}{\omega(B)^{\kappa_1}}\Big(\sum\limits_{j=1}^{\infty}\frac{1}{|2^jB|^{(1+\frac{\lambda}{n})}}\int_{2^{j+1}B}|f(y)|.|b_B-b(y)|dy\Big)^p\nonumber
\\
&\leq\frac{|B|}{\omega(B)^{\kappa_1}}\Big(\sum\limits_{j=1}^{\infty}\frac{1}{|2^jB|^{(1+\frac{\lambda}{n})}}\Big(\int_{2^{j+1}B}|f(y)|^pdy\Big)^{\frac{1}{p}}.\Big(\int_{2^{j+1}B}|b_B-b(y)|^{p'}dy\Big)^{\frac{1}{p'}}\Big)^p\nonumber
\\
&\lesssim \frac{|B|}{\omega(B)^{\kappa_1}}\Big(\sum\limits_{j=1}^{\infty}\frac{\omega(2^{j+1}B)^{\frac{\kappa}{p}}}{|2^jB|^{(1+\frac{\lambda}{n})}}(L_{1,i}+L_{2,i})\Big)^p\|f\|^p_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)},\nonumber\end{aligned}$$ where $L_{1,i}=\Big(\int_{2^{j+1}B}|b(y)-b_{2^{j+1}B}|^{p'}dy\Big)^{\frac{1}{p'}}$ and $L_{2,i}=\Big(\int_{2^{j+1}B}|b_B-b_{2^{j+1}B}|^{p'}dy\Big)^{\frac{1}{p'}}$. On the other hand, by Lemma \[BMO-Lemma\] and Proposition \[Pro-T1986\], we also get $$L_{1,i}\leq \|b\|_{BMO(\mathbb R^n)}.|2^{j+1}B|^{\frac{1}{p'}}$$ and $$L_{2,i}\leq \Big(\int_{2^{j+1}B}\Big(2^n(j+1)\|b\|_{BMO(\mathbb R^n)}\Big)^{p'}dy\Big)^{\frac{1}{p'}}\leq 2^n(j+1).\|b\|_{BMO(\mathbb R^n)}.|2^{j+1}B|^{\frac{1}{p'}}.$$ Thus, by estimating as (\[I2-strong\]) above, we immediately have $$\begin{aligned}
K_{2,2}&\lesssim \frac{|B|}{\omega(B)^{\kappa_1}}\Big(\sum\limits_{j=1}^{\infty}\frac{(j+2).\omega(2^{j+1}B)^{\frac{\kappa}{p}}.|2^{j+1}B|^{\frac{1}{p'}}}{|2^jB|^{(1+\frac{\lambda}{n})}}\Big)^p\|f\|^p_{\mathcal B^{p,\kappa}(\omega,\mathbb R^n)}.\|b\|^p_{BMO(\mathbb R^n)}\nonumber
\\
&\lesssim \Big(\sum\limits_{j=1}^{\infty}(j+2)2^{j(\frac{\kappa(n+\beta)-n}{p}-\lambda)}\Big)^p.\|f\|^p_{\mathcal B^{p,\kappa}(\omega,\mathbb R^n)}.\|b\|^p_{BMO(\mathbb R^n)}\nonumber
\\
&\lesssim \|f\|^p_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)}.\|b\|^p_{BMO(\mathbb R^n)}.\nonumber \end{aligned}$$ From the above estimation, by (\[K12-strong\])-(\[K21-strong\]), we confirm $$\|[b,\mathcal{T}](f)\|_{\mathcal {\mathop B\limits^.}^{p,\kappa_1}_{\rm loc}(\omega,\mathbb R^n)}\lesssim \|b\|_{BMO(\mathbb R^n)}.\|f\|_{\mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)},\,\textit{\rm for all}\, f\in L^p(\mathbb R^n)\cap \mathcal {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n).$$ Therefore, the proof of this theorem is completed.
Now, let us give some applications of Theorem \[Theo-sublinear1\]. Note that Hirschman [@Hirschman1959], Wainger [@Wainger1965], Cho and Yang [@CY2010] studied the strongly singular convolution operators in the context of $L^p(\mathbb R^n)$ spaces defined as follows.
Let $0 < s < \infty$ and $0 < \lambda < \frac{ns}{2}$. The strongly singular integral operator $T^{s,\lambda}$ is defined by $$\begin{aligned}
T^{s,\lambda}(f)(x)=p.v.\int_{\mathbb{R}^n}\frac{e^{i|x-y|^{-s}}}{|x-y|^{n+\lambda}}\chi_{\{|x-y|<1 \}}f(y)dy.\end{aligned}$$
\[Theo-strong\] [(see in [@Fefferman1970; @Hirschman1959; @Wainger1965])]{} Let $0<s<\infty$, $1<p<\infty$, $0<\lambda<\frac{ns}{2}$, $|\frac{1}{p}-\frac{1}{2}|<\frac{1}{2}-\frac{\lambda}{ns}$. Then $T^{s,\lambda}$ extends to a bounded operator from $L^p(\mathbb R^n)$ to itself.
Let $0 < \zeta, s,\lambda < \infty$, and $k$ be an integer with $k\geq 2$. The strongly singular integral operator $T_{\zeta,s,\lambda}$ is defined by $$\begin{aligned}
T_{\zeta,s, \lambda}(f)(x)=p.v.\int_{\mathbb{R}}\frac{e^{i\{\zeta.(x-y)^k+|x-y|^{-s}\}}}{(x-y)|x-y|^{\lambda}}f(y)dy.\end{aligned}$$
\[Theo-strong-CY2010\] [(see in [@CY2010])]{} Let $0 < \zeta, s, \lambda < \infty$, $k\in\mathbb N$ with $k\geq 2$ and $s\geq 2\lambda$. Then $T_{\zeta,s,\lambda}$ extends to a bounded operator from $L^2(\mathbb R)$ to itself.
On the other hand, Li and Lu [@LL2006] also studied the Coifman-Rochberg-Weiss type commutator of strongly singular integral operator defined as follows
Let $0 < s < \infty$ and $0 < \lambda < \frac{ns}{2}$. The Coifman-Rochberg-Weiss type commutator of strongly singular integral operator is defined by $$\label{commuatator-strong}
[b,T^{s,\lambda}](f)(x)=p.v.\int_{\mathbb{R}^n}\frac{e^{i|x-y|^{-s}}}{|x-y|^{n+\lambda}}\chi_{\{|x-y|<1 \}}\big(b(x)-b(y)\big)f(y)dy,$$ where $b$ is locally integrable functions on $\mathbb R^n$.
Moreover, Li and Lu [@LL2006] proved the following interesting result.
\[Theo1.1-LL2006\] [(Theorem 1.1 [@LL2006])]{} Let $0 < s < \infty$, $1<p<\infty$, $0<\lambda<\frac{ns}{2}$, $|\frac{1}{p}-\frac{1}{2}|<\frac{1}{2}-\frac{\lambda}{ns}$ and $b\in BMO(\mathbb R^n)$. Then the commutator $[b, T^{s,\lambda}]$ extends to a bounded operator on $L^p(\mathbb R^n)$.
From Theorem \[Theo-sublinear1\], Theorem \[Theo-strong\], Theorem \[Theo-strong-CY2010\] and Theorem \[Theo1.1-LL2006\], we obtain the useful results as follows.
\[Theo-strong1\] Let $0<s<\infty$, $1<p<\infty$, $0<\lambda<\frac{ns}{2}$, $0<\kappa<1$, $|\frac{1}{p}-\frac{1}{2}|<\frac{1}{2}-\frac{\lambda}{ns}$, $-n+\frac{\lambda p}{\kappa}<
\beta< \frac{\lambda p +(1-\kappa)n}{\kappa}$, $\omega(x)=|x|^{\beta}$ and $\kappa_1\in (0,\kappa-\frac{\lambda p}{n+\beta}]$. Let $b\in L^{\eta}_{\rm loc}(\mathbb R^n)\cap BMO(\mathbb R^n)$ with $\eta>p'$. Then $T^{s,\lambda}$ and $[b, T^{s,\lambda}]$ extend to bounded operators from $\mathfrak {\mathop B\limits^.}^{p,\kappa}(\omega,\mathbb R^n)$ to $\mathcal {\mathop B\limits^.}^{p,\kappa_1}_{\rm loc}(\omega,\mathbb R^n)$.
\[Coro-CY-new\] Let $0<\zeta,\lambda,s<\infty$, $k\in\mathbb N$ with $k\geq 2$, $s\geq 2\lambda$, $0<\kappa<1$, $-1+\frac{2\lambda}{\kappa}<
\beta< \frac{2\lambda +(1-\kappa)}{\kappa}$, $\omega(x)=|x|^{\beta}$ and $\kappa_1\in (0,\kappa-\frac{2\lambda}{1+\beta}]$. Then $T_{\zeta, s,\lambda}$ extends to a bounded operator from $\mathfrak {\mathop B\limits^.}^{2,\kappa}(\omega,\mathbb R)$ to $\mathcal {\mathop B\limits^.}^{2,\kappa_1}_{\rm loc}(\omega,\mathbb R)$.
Remark that in the special case when the weight function in Theorem \[Theo-sublinear1\], Corollary \[Theo-strong1\] and Corollary \[Coro-CY-new\] is a constant function, then we can remove the central condition in the spaces, that is, we may replace $\mathfrak {\mathop B\limits^.}^{p,\kappa}(\mathbb R^n)$ and $\mathcal {\mathop B\limits^.}^{p,\kappa_1}_{\rm loc}(\mathbb R^n)$ by $\mathfrak {M}^{p,\kappa}(\mathbb R^n)$ and $\mathcal { B}^{p,\kappa_1}_{\rm loc}(\mathbb R^n)$, respectively. Here ${\mathfrak{M} }^{p,\kappa}(\mathbb R^n)$ is the closure of $L^{p}(\mathbb R^n)\cap {\mathcal B}^{p,\kappa}(\mathbb R^n)$ in the space ${\mathcal B}^{p,\kappa}( \mathbb R^n)$. Finally, we give the boundedness of sublinear operators in the setting when the weighted function is in the class of Muckenhoupt weights. It is worth pointing out that when $\omega_1=\omega_2=\omega$, then the space $C^\infty_0(\mathbb R^n)$ is contained in ${\mathcal B}^{q,\kappa}_\omega(\mathbb R^n)$. The space ${\mathfrak{B} }^{q,\kappa}_\omega(\mathbb R^n)$ is denoted as the closure of $L^{q}(\mathbb R^n)\cap {\mathcal B}^{q,\kappa}_\omega(\mathbb R^n) $ in the space ${\mathcal B}^{q,\kappa}_\omega(\mathbb R^n)$.
\[Theo-Sublinear2\] Let $1<p<\infty$, $0<\kappa<1$, and $1\leq p^*,\zeta<\infty$, $\omega\in A_{\zeta}$ with the finite critical index $r_\omega$ for the reverse Hölder. Assume that $p > p^{*}\zeta {r^{'}_\omega}, \delta\in (1,r_\omega)$ and $\kappa^*=\frac{p^*(\kappa-1)}{p}+1$. Then, if $\mathcal{T}$ extends to a bounded operator on $L^p(\mathbb R^n)$, then $\mathcal{T}$ can also extend to a bounded operator from ${\mathfrak{B} }^{p,\kappa}_\omega(\mathbb R^n)$ to ${B}^{p^*,\kappa^*}_\omega(\mathbb R^n)$.
Let us fix $f\in L^{p}(\mathbb R^n)\cap {\mathcal B}^{p,\kappa}_\omega(\mathbb R^n)$ and a ball $B_R(x_0):=B$ with $R\geq 1$. From assume that $p > p^*\zeta r^{'}_\omega $, one has $r\in(1, r_\omega)$ satisfying $p = \zeta p^* r'$. Hence, by the Hölder inequality and the reverse Hölder condition, we lead to $$\begin{aligned}
\label{T_p*}
\Big(\int_{B}|\mathcal{T} (f)(x)|^{p*}\omega(x)dx\Big)^{\frac{1}{p^*}}&\leq \Big(\int_{B}|\mathcal{T} (f)(x)|^{\frac{p}{\zeta}}dx\Big)^{\frac{\zeta}{p}}.\Big(\int_B \omega(x)^r dx\Big)^{\frac{1}{rp*}}\nonumber
\\
&\lesssim \Big(\int_{B}|\mathcal{T}(f)(x)|^{\frac{p}{\zeta}}dx\Big)^{\frac{\zeta}{p}}\omega(B)^{\frac{1}{p*}}.|B|^{\frac{-\zeta}{p}}.\end{aligned}$$ Next, we decompose $f=f_1+f_2$ where $f_1=f.\chi_{2B}$. Thus, $$\begin{aligned}
\label{A12}
\Big(\int_{B}|\mathcal{T}(f)(x)|^{\frac{p}{\zeta}}dx\Big)^{\frac{\zeta}{p}}&\lesssim \Big(\int_{B}|\mathcal{T} (f_1)(x)|^{\frac{p}{\zeta}}dx\Big)^{\frac{\zeta}{p}}+ \Big(\int_{B}|\mathcal{T}(f_2)(x)|^{\frac{p}{\zeta}}dx\Big)^{\frac{\zeta}{p}}\nonumber
\\
&:=A_1+A_2.\end{aligned}$$ From assuming that $\mathcal{T}$ extends to a bounded operator on $L^p(\mathbb R^n)$ and using Proposition \[pro2.4DFan\], we get $$\begin{aligned}
\label{A1}
A_1&\leq \Big(\int_{\mathbb R^n}|\mathcal{T}( f_1)(x)|^{\frac{p}{\zeta}}dx\Big)^{\frac{\zeta}{p}}\lesssim \Big(\int_{2B}|f(x)|^{\frac{p}{\zeta}}dx\Big)^{\frac{\zeta}{p}}.\nonumber
\\
&\lesssim \Big(\int_{2B}|f(x)|^{p}\omega(x)dx\Big)^{\frac{1}{p}}\omega(2B)^{\frac{-1}{p}}|2B|^{\frac{\zeta}{p}}\lesssim \|f\|_{\mathcal B^{p,\kappa}_{\omega}(\mathbb R^n)}.\omega(2B)^{\frac{(\kappa-1)}{p}}.|B|^{\frac{\zeta}{p}}.\end{aligned}$$ Next, let us give $x\in B$ and $y\in (2B)^c$. By applying the relation (\[Nakai\]) above and estimating as (\[Tf2\]) and (\[A1\]), we obtain $$\begin{aligned}
&|\mathcal{T}(f_2)(x)|\nonumber
\\
&\lesssim \sum\limits_{j=1}^{\infty}\frac{1}{|2^jB|^{(1+\frac{\lambda}{n})}}\int_{2^{j+1}B}|f(y)|dy\leq \sum\limits_{j=1}^{\infty}\frac{1}{|2^jB|^{(1+\frac{\lambda}{n})}}\Big(\int_{2^{j+1}B}|f(y)|^{\frac{p}{\zeta}}dy\Big)^{\frac{\zeta}{p}}|2^{j+1}B|^{1-\frac{\zeta}{p}} \nonumber
\\
&\leq \sum\limits_{j=1}^{\infty}\frac{1}{|2^jB|^{(1+\frac{\lambda}{n})}}\|f\|_{\mathcal B^{p,\kappa}_{\omega}(\mathbb R^n)}.\omega(2^{j+1}B)^{\frac{(\kappa-1)}{p}}.|2^{j+1}B|\lesssim \|f\|_{\mathcal B^{p,\kappa}_{\omega}(\mathbb R^n)}\sum\limits_{j=1}^{\infty}\frac{\omega(2^{j+1}B)^{\frac{(\kappa-1)}{p}}}{|2^jB|^{\frac{\lambda}{n}}}.\nonumber\end{aligned}$$ Hence, $$\begin{aligned}
\label{A2}
A_2\lesssim \|f\|_{\mathcal B^{p,\kappa}_{\omega}(\mathbb R^n)}\Big(\sum\limits_{j=1}^{\infty}\frac{\omega(2^{j+1}B)^{\frac{(\kappa-1)}{p}}}{|2^jB|^{\frac{\lambda}{n}}}\Big)|B|^{\frac{\zeta}{p}}.\end{aligned}$$ Next, by Proposition \[rever-Holder\] and $\kappa\in (0,1)$, we deduce $$\begin{aligned}
\Big(\frac{\omega(2^{j+1}B)}{\omega(B)}\Big)^{\frac{(\kappa-1)}{p}}\lesssim \Big(\frac{|2^{j+1}B|}{|B|}\Big)^{\frac{(\kappa-1)(\delta-1)}{p\delta}}\lesssim 2^{\frac{jn(\kappa-1)(\delta-1)}{p\delta}}.\nonumber\end{aligned}$$ From this, by using (\[T\_p\*\])-(\[A2\]), $\kappa^*=\frac{p^*(\kappa-1)}{p}+1$ and $R\geq 1$, we get $$\begin{aligned}
&\Big(\frac{1}{\omega(B)^{\kappa^*}}\int_{B}|\mathcal{T}(f)(x)|^{p*}\omega(x)dx\Big)^{\frac{1}{p^*}}\nonumber
\\
&\lesssim \frac{\|f\|_{\mathcal B^{p,\kappa}_{\omega}(\mathbb R^n)}}{\omega(B)^{\frac{\kappa^*}{p*}}}.\Big(\omega(2B)^{\frac{(\kappa-1)}{p}}+\sum\limits_{j=1}^{\infty}2^{-j\lambda}\omega(2^{j+1}B)^{\frac{(\kappa-1)}{p}}\Big).\omega(B)^{\frac{1}{p*}}\nonumber
\\
&\lesssim \|f\|_{\mathcal B^{p,\kappa}_{\omega}(\mathbb R^n)}.\Big(\sum\limits_{j=0}^{\infty}2^{-j\lambda}\Big(\frac{\omega(2^{j+1}B)}{\omega(B)}\Big)^{\frac{(\kappa-1)}{p}}\Big)
\lesssim \|f\|_{\mathcal B^{p,\kappa}_{\omega}(\mathbb R^n)}.\Big(\sum\limits_{j=0}^{\infty}2^{j(\frac{n(\kappa-1)(\delta-1)}{p\delta}-\lambda)}\Big)\nonumber
\\
&\lesssim \|f\|_{\mathcal B^{p,\kappa}_{\omega}(\mathbb R^n)}.\nonumber\end{aligned}$$ Therefore, $$\|\mathcal{T}(f)\|_{ B^{p^*,\kappa^*}_\omega(\mathbb R^n)}\lesssim \|f\|_{\mathcal B^{p,\kappa}_\omega(\mathbb R^n)},\,\textit{\rm for all}\,f\in L^p(\mathbb R^n)\cap {\mathcal B}^{p,\kappa}_\omega(\mathbb R^n).$$ This implies that theorem is proved.
By Theorem \[Theo-Sublinear2\], we obtain the following interesting corollary.
\[Theo-SIO-strong-2\] Let $s,\lambda,\kappa,p$ as Corollary \[Theo-strong1\] and $1\leq p^*,\zeta<\infty$, $\omega\in A_{\zeta}$ with the finite critical index $r_\omega$ for the reverse Hölder. Assume that $p > p^{*}\zeta {r^{'}_\omega}, \delta\in (1,r_\omega)$ and $\kappa^*=\frac{p^*(\kappa-1)}{p}+1$. Then, $T^{s,\lambda}$ can extend to a bounded operator from ${\mathfrak{B} }^{p,\kappa}_\omega(\mathbb R^n)$ to ${B}^{p^*,\kappa^*}_\omega(\mathbb R^n)$.
[79]{}
D. R. Adams, *A note on Riesz potentials*, Duke Math. J. **42** (1975), 765-778.
J. Álvarez, M. Milman, *Vector valued inequalities for strongly Calderón-Zygmund operators*, Rev. Mat. Iberoam. **2** (1986), 405-426.
J. Alvarez, M. Guzmán-Partida and J. Lakey, *Spaces of bounded $\lambda$-central mean oscillation, Morrey spaces, and $\lambda$-central Carleson measures*, Collect. Math. **51** (2000), 1-47.
K. Andersen, R. John, *Weighted inequalities for vector-valued maximal functions and singular integrals*, Studia Math. **69** (1981), no. 1, 19-31.
A. Benedek and R. Panzone, *The space $L^p$ with mixed norm*, Duke Math. J. **28** (1961), 301-324.
L. Caffarelli, *Elliptic second order equations*, Rend. Sem. Mat. Fis. Milano. **58** (1988), 253-284.
H. M. Chung, R. A. Hunt, D. S. Kurtz, *The Hardy-Littlewood maximal function on $L(p, q)$ spaces with weights*, Indiana Univ. Math. J. **31** (1982), no. 1, 109-120.
N. M. Chuong, *Pseudodifferential operators and wavelets over real and p-adic fields*, Springer, 2018.
N. M. Chuong, D. V. Duong and K. H. Dung, *Vector valued maximal Carleson type operators on the weighted Lorentz spaces*, 2017, arxiv:1707.00092v1.
R. Coifman, C. Fefferman, *Weighted norm inequalities for maximal functions and singular integrals*, Studia Math. **51** (1974), no. 3, 241-250.
C. H. Cho, C. W. Yang, *Estimates for oscillatory strongly singular integral operators*, J. Math. Anal. Appl. **362** (2010), 523-533.
D. Fan, S. Lu, and D. Yang, *Regularity in Morrey spaces of strong solutions to nondivergence elliptic equations with VMO coefficients*, Georgian Math. J. **5** (1998), no. 5, 425-440.
C. Fefferman and E. M. Stein, *Some maximal inequalities*, Amer. J. Math. **93** (1971), no. 1, 107-115.
C. Fefferman, *Inequality for strongly singular convolution operators*, Acta Math. **124** (1970), 9-36.
P. Federbush, *Navier and Stokes meet the wavelet*, Comm. Math. Phys. **155** (1993), 219-248.
V. S. Guliyev, S.S. Aliyev, T. Karaman and P. S. Shukurov, *Boundedness of sublinear operators and commutators on generalized Morrey Spaces*, Integr. Equ. Oper. Theory. **71** (2011), 327-355.
F. Gurbuz, *Weighted Morrey and weighted fractional Sobolev-Morrey spaces estimates for a large class of pseudo-differential operators with smooth symbols*, J. Pseudo-Differ. Oper. Appl. **7** (2016), 595-607.
L. Grafakos, *Modern Fourier analysis*, Second Edition, Springer, (2008).
R. A. Hunt, *On L(p,q) spaces*, Enseign. Math. **12** (1966), 249-276.
T. Hytönen, C. Pérez and E. Rela, *Sharp reverse Hölder property for $A_\infty$ weights on spaces of homogeneous type*, J. Funct. Anal. **263** (2012), 3883-3899.
I. I. Hirschman, *On multiplier transformations*, Duke Math. J. **26** (1959), 221-242.
S. Indratno, D. Maldonado and S. Silwal, *A visual formalism for weights satisfying reverse inequalities*, Expo. Math. **33** (2015), 1-29.
M. Kaneko and S. Yano, *Weighted norm inequalities for singular integrals*, J. Math. Soc. Japan . **27** (1975), 570-588.
Y. Komori, S. Shirai, *Weighted Morrey spaces and a singular integral operator*, Math. Nachr. **282** (2009), 219-231.
V. Kokilashvili, A. Meskhi, and H. Rafeiro, *Sublinear operators in generalized weighted Morrey Spaces*, Doklady Mathematics. **94** (2016), 558-560.
G. G. Lorentz, *Some new functional spaces*, Ann. of Math. **51** (1950), no. 1, 37-55.
J. Li and S. Lu, *$L^p$ estimates for multilinear operators of strongly singular integral operators*, Nagoya Math. J. **181** (2006), 41-62.
Y. Lin, *Strongly singular Calderón-Zygmund operator and commutator on Morrey type spaces*, Acta Math. Sin. **23** (2007), 2097-2110.
A. L. Mazzucato, *Besov-Morrey spaces: function space theory and applications to non-linear PDE*, Trans. Amer. Math. Soc. **355** (2003), no. 4, 1297-1364.
B. Muckenhoupt,*Weighted norm inequalities for the Hardy maximal function*, Trans. Amer. Math. Soc. **165** (1972), 207-226.
C. Morrey, *On the solutions of quasi-linear elliptic partial differential equations*, Trans. Amer. Math. Soc. **43** (1938), 126-166.
E. Nakai, N. Tomita and K. Yabuta, *Density of the set of all infinitely differentiable functions with compact support in weighted Sobolev spaces*, Scientiae Math. Jap. Online. **10** (2004), 39-45.
E. Nakai, *Hardy-Littlewood maximal operator, singular integral operators and the Riesz potentials on generalized Morrey spaces*, Math. Nachr. **166** (1994), 95-103.
S. Nakamura and Y. Sawano, *The singular integral operator and its commutator on weighted Morrey spaces*, Collect. Math. **68** (2017), no. 2, 145-174.
C. Pérez, *Endpoints for commutators of singular integral operators*, J. Funct. Anal. **128** (1995), 163-185.
Y. Pan, *Hardy spaces and oscillatory singular integrals*, Rev. Mat. Iberoam. **7** (1991), 55-64.
A. Ruiz and L. Vega, *Unique continuation for Schrödinger operators with potential in Morrey spaces*, Publ. Mat. **35** (1991), no. 1, 291-298. Conference on Mathematical Analysis (El Escorial, 1989).
N. Samko, *Weighted Hardy and singular operators in Morrey spaces*, J. Math. Anal. Appl. **350** (2009), no. 1, 56-72.
Z. Shen, *The periodic Schrödinger operators with potentials in the Morrey class*, J. Funct. Anal. **193** (2002), no. 2, 314-345.
F. Soria, G. Weiss, *A remark on singular integrals and power weights*, Indiana Univ. Math. J. **43** (1994), 187-204.
E. M. Stein, *Harmonic Analysis: Real-Variable Methods, Orthogonality, and Oscilla- tory Integrals*, Princeton University Press, Princeton, 1993.
P. Sjölin, *$L^p$ estimates for strongly singular convolution operators in $\mathbb R^n$*, Ark. Mat. **14** (1976), 59-64.
A. Torchinsky, *Real Variable Methods in Harmonic Analysis*, Academic Press, San Diego, 1986.
L. Tang, *Weighted norm inequalities for pseudodifferential operators with smooth symbols and their commutators*, J. Funct. Anal. **262** (2012), 1603-1629.
M. E. Taylor, *Analysis on Morrey spaces and applications to Navier-Stokes and other evolution equations*, Comm. Partial Differential Equations. **17** (1992), 1407-1456.
D. Wang, J. Zhou, W. Chen, *Another characterizations of Muckenhoupt $A_p$ class*, Acta Math. Scientia. **37** (2017), 1761-1774.
S. Wainger, *Special trigonometric series in k-dimensions*, Mem. Amer. Math. Soc. **56**, 1965.
C. T. Zorko, *Morrey space*, Proc. Amer. Math. Soc. **98** (1986), 586-592.
[^1]: This paper is supported by Vietnam National Foundation for Science and Technology Development (NAFOSTED)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Emission of high energy (HE) photons above 100 MeV that is delayed and lasts much longer than the prompt MeV emission has been detected from several long duration gamma ray bursts (LGRBs) and short hard bursts (SHBs) by the Compton, Fermi and AGILE gamma ray observatories. In this paper we show that the main observed properties of this HE emission are those predicted by the cannonball (CB) model of GRBs: In the CB model all the observed radiations in a GRB are produced by the interaction of a highly relativistic jet of plasmoids (CBs) with the environment. The prompt X-ray and MeV $\gamma$-ray pulses are produced by inverse Compton scattering (ICS) of glory photons -photons scattered/emitted into a cavity created by the wind/ejecta blown from the progenitor star or a companion star long before the GRB- by the thermal electrons in the CBs. A simultaneous optical and high energy emission begins shortly after each MeV pulse when the CB collides with the wind/ejecta, and continues during the deceleration of the CB in the interstellar medium. The optical emission is dominated by synchrotron radiation (SR) from the swept-in and knocked-on electrons which are Fermi accelerated to high energies by the turbulent magnetic fields in the CBs, while ICS of these SR photons dominates the emission of HE photons. The lightcurves of the optical and HE emissions have approximately the same temporal behaviour but have different power-law spectra. The emission of very high energy (VHE) photons above 100 TeV is dominated by the decay of $\pi^0$’s produced in hadronic collisions of Fermi accelerated protons in the CBs. The CB model explains well all the observed radiations, including the high energy radiation from both LGRBs and SHBs as demonstrated here for GRB 090902B and SHB 090510.'
author:
- Shlomo Dado and Arnon Dar
title: High Energy Photons From Gamma Ray Bursts
---
-0.5cm
Introduction
============
During nearly 20 years after the launch of the Compton Gamma Ray Observatory (CGRO), the Burst And Transient Source Experiment (BATSE) on board the CGRO has detected and measured light curves and spectra (Kaneko et al. 2008) in the sub-MeV range of several thousands gamma ray bursts (GRBs). Higher-energy observations with the EGRET instrument aboard CGRO were limited to those GRBs which happened to be in its narrower field of view. Its large calorimeter measured the light-curves and spectra of several GRBs in the 1-200 MeV energy range. Seven GRBs were detected also with the EGRET spark chamber, sensitive in the 30 MeV - 10 GeV energy range. The EGRET detections indicated that the spectrum of bright GRBs extends beyond 1 GeV (Hurley et al. 1994) with no evidence for a spectral cut-off (see, e.g., Dingus 1995, 2001 and references therein). However, a few GRBs, such as 940217 (Hurley et al. 1994) and 941017 (Gonzalez et al. 2003), showed evidence that the high energy component has a slower temporal decay than that of the sub-MeV emission, suggesting that, at least in some cases, it is not a simple extension of the main component, but originates from a different emission mechanism and/or region. This has been confirmed recently by observations of high energy photons in the 30 MeV - 300 GeV range from several GRBs with the AGILE (GRB 080514B: Giuliani et al. 2008, GRB 090510: Giuliani et al. 2009) and the Fermi large area telescope (LAT) (e.g., GRB 080916C: Abdo et al. 2009a, GRB 090902B: Bissaldi et al. 2009 and GRB 090510: Ghirlanda et al. 2009). The arrival times of the high energy photons did not coincide with the times of the brightest peaks seen at hard X-rays and MeV $\gamma$-rays. Also the high energy emission lasted much longer time than that of the prompt keV-MeV emission.
The detection of higher energy gamma rays is affected by pair production in their collisions with the extragalactic infrared background light (Nikishov 1961) resulting in an absorption which is a strong function of redshift and energy. Recent estimates (Primack et al. 2005), validated by HESS observations (Aharonian et al. 2006) predict an optical depth of roughly unity to 500 GeV photons emitted at a redshift $z\!=\!0.2$ and to 10 TeV photons at $z=0.05$. The average redshifts of LGRBs and SHBs are much larger, $z\!=\!2.2$ and $z\!=\!0.5$, respectively. Despite of the strong attenuation of high energy gamma rays in the intergalactic medium (IGM), there have been several claims in the past of detections at the 3 sigma level of TeV gamma-rays from GRBs (see, e.g., Atkins et al. 2003 and references therein). However, more recently, no GRB was conclusively detected in the range 100 GeV to 100 TeV by the ground based water Cherenkov detector Milagro and by the air Cherenkov telescopes MAGIC, Whipple, HESS and VERITAS. Moreover in all previous cases of reported detection of TeV gamma-rays from a GRB, the GRB redshift was not known. If their redshifts are similar to those of ordinary GRBs then their TeV gamma-rays are strongly absorbed by the extragalactic background light (EBL), implying that the TeV emission if detected must be extraordinarily energetic, i.e., with a much larger fluence than that emitted in X-rays and MeV $\gamma$-rays.
Most theoretical models of high energy photon emission in GRBs relied on the standard fireball models of GRBs (for recent reviews see, e.g., Piran 2005; Mészáros 2006; Zhang 2007). In these models, synchrotron radiation of electrons accelerated by ‘internal shocks’ in collisions between conical shells ejected by the GRB’s central engine produces the prompt GRB emission and a blast wave (external shock) driven into the circumburst medium generates their afterglow. The high energy radiation was suggested to be produced either by inverse Compton scattering of the synchrotron radiation (the so called ‘synchrotron self Compton mechanism’) by the shock-accelerated electrons (e.g., Dermer et al. 2000), by the decay of $\pi^0$ photo produced in collisions of shock accelerated hadrons with synchrotron photons in the expanding fireball (Waxman & Bahcall 1997), or by synchrotron radiation from ultra high energy protons (Totani 1998) allegedly accelerated in the GRB fireball (Waxman 1995; Milgrom & Usov 1995; Vietri 1995). All these models that were based on the standard fireball model of GRBs predict simultaneous emissions at all energy bands. This is in conflict with the observed delayed emission of the high energy photons that lasts much longer.
In the cannonball (CB) model (see e.g. Dado, Dar & De Rújula, hereafter DDD, 2009a,b and references therein), GRBs and their afterglows (AGs) are produced by bipolar jets of highly relativistic plasmoids (CBs) ejected in violent stellar processes. The prompt MeV $\gamma$-rays and hard X-rays are produced by inverse Compton scattering (ICS) of glory photons - photons emitted/scattered into a cavity formed by the wind/ejecta puffed by the progenitor or a companion star long before the GRB. Synchrotron radiation (SR) emitted from the electrons of the ionized wind/ejecta that are swept into the CBs and are Fermi accelerated by their turbulent magnetic fields dominates their ‘prompt’ optical emission (e.g., DDD2009a and references therein) which begins when the CBs reach the wind/ejecta. In this paper we show that in the CB model ICS of this SR (e.g., Dado & Dar 2005) dominates the HE emission from long GRBs and SHBs. This HE emission begins simultaneously with the ‘prompt’ optical emission that lags after the prompt X-ray and MeV $\gamma$-ray emission and lasts much longer, has the same lightcurve as the optical emission but with a power-law spectrum that is identical to that of the X-ray emission. It describes well the HE observations, as demonstrated in the paper for GRB 090902B and SHB 090510.
Because of the Klein-Nishina effect, $E^2\,dn/dE$ of ICS of the prompt SR in GRBs peaks near 10 TeV and is cutoff at a few tens of TeV when the rate of synchrotron energy losses by electrons in the CBs exceeds the rate of energy gain by Fermi acceleration. The prompt decay of $\pi^0$’s produced in the collisions between the Fermi accelerated nuclei and the ambient matter in the CBs produces a power-law spectrum that extends to much higher energies where it dominates the HE emission. However, like in blazars, the observed flux of TeV photons from distant GRBs and SHBs is stronglyly suppressed by pair production in collisions with the extragalactic background photons and only relatively nearby GRBs and SHBs might be detected in TeV photons by the large ground based HE gamma ray telescopes such as HESS, MAGIC and VERITAS.
The CB model
============
In the cannonball (CB) model (Dado, Dar & De Rújula (hereafter DDD) 2002; Dar & De Rújula (hereafter DD) 2004; DDD2009a,b and references therein) GRBs and their AGs are produced by bipolar jets of highly relativistic CBs of ordinary matter which are ejected (Shaviv & Dar 1995, Dar & Plaga 1999) in the birth of neutron stars or black holes in core-collapse supernova (SN) explosions (long GRBs) akin to SN 1998bw (Galama et al. 1998), and in the merger of neutron stars and/or phase transition in compact stars (short hard bursts). Their prompt MeV $\gamma$-rays and hard X-rays are produced by the thermal electrons in the CBs’ plasma via inverse Compton scattering (ICS) of glory photons - photons emitted/scattered into a cavity created by the wind/ejecta blown from the progenitor star or a companion star long before the GRB. Slightly later when the CBs encounter the wind/ejecta, and afterwards when the CBs coast through the interstellar medium (ISM) sorounding it, the electrons of the ionized gas in front of them that are swept in and Fermi accelerated by the CBs’ turbulent magnetic fields emit synchrotron radiation (SR) which dominates the ‘prompt’ optical emission and the broad band afterglow emission. ICS of the SR radiation by these electrons and the decay of $\pi^0$’s produced in collision between the swept-in wind and ISM protons and the ambient CB protons produce the ‘prompt’ high energy emission simultaneously with the optical emission. Within the CB model, the burst environment as illustrated in Fig. (\[f0\]) and the above radiation mechanisms, which are summarized in Table \[t1\], suffice to provide a sufficiently accurate description of the observed radiations from GRBs at all times and all detected wavelengths.
ICS of self produced SR
=======================
Self produced SR
----------------
When a CB encounters the wind/ejecta which was blown by the progenitor star long before the GRB, it sweeps in the ionized matter in front of it. The swept in electrons and nuclei that in its rest frame enter it with a Lorentz factor $\gamma$ equal to that of the bulk motion of the CB, are isotropized and Fermi accelerated in the CBs by its turbulent magnetic field. They emit SR with an early-time lightcurve in the observer frame (DDD2009a) : $$F_\nu[t] \propto {e^{-a/t}\,
t^{1-\beta} \over t^2+t_{exp}^2}\, \nu^{-\beta}\!\rightarrow
\! t^{\!-\!(1\!+\!\beta)}\, \nu^{\!-\!\beta}\, ,
\label{SROP}$$ where $t\!=\!T\!-\!T_i$, $T$ is the observer time after trigger and $T_i$ is the observer time when the CB reaches the wind/ejecta. The optical band is initially well below the ‘bend’ frequency (DDD2009a). Consequently, $\beta_O\!\approx\!
0.5$ and the ‘prompt’ optical flare that follows an ICS keV-MeV pulse/flare, decays like $F_\nu \! \propto\!
t^{\!-\!1.5}\,\nu^{\!-\!0.5}\,.$
ICS in the Thomson and Klein-Nishina regimes
--------------------------------------------
The differential cross section for Compton scattering of a photon with energy $E_\gamma$ from an electron at rest was calculated by Klein & Nishina in 1929. In the electron’s rest frame it is given by, $${d\sigma \over d\Omega}={1 \over 2}\, \left ({e^2\over m_e\, c^2}\right)^2
\, (k-k^2\, sin^2\theta+k^3)\,,
\label{KN}$$ where $$k(E_\gamma,\theta)={1\over 1+{E_\gamma \over m_e\, c^2}\,(1-cos\theta)}
\label{KNk}$$ The final energy of the scattered photon is, $k\,E_\gamma\, .$
In the Thomson regime where $E_\gamma\!\ll\! m_e\, c^2$, and then $k\!\approx\! 1$, $d\sigma/ d\Omega\!\propto\!(1\!+\!cos^2\theta)$ and the mean final energy of a photon that suffered an ICS by an electron of energy $E_e$ is $(4/3)\, (E_e/m_e\, c^2)^2\, E_\gamma$. A power-law distribution of Fermi accelerated electrons $dn_e/dE\!
\propto\! E^{-p}$ generates through ICS a power-law distribution of scattered photons with $E^2\, dn_\gamma/dE\!\propto
E^{-(p\!-\!3)/2},$ where $p\!\approx \!2.2$ well below the electron cooling energy and $p\!\approx \!3.2$ well above it.
In the Klein-Nishina (KN) regime where $E_\gamma\!>\! m_e\, c^2$ in the electron rest frame, $k\, E_\gamma \!\sim\!m_e\,c^2$, $\sigma_{KN}\!\propto\!1/E_e$, and the mean energy of a photon that suffered an ICS by an HE electron is approximately $E_e$. In the KN regime, a power-law distribution of electrons $dn_e\!
\propto\! E^{\!-\!(p\!+\!1)}\, dE$ generates by ICS of self produced SR a power-law distribution of scattered photons with $E^2\, dn_\gamma/dE\!\propto
E^{\!-\!(p\!+\!1)/2)}$. In particular, for a distribution of Fermi accelerated elecctrons that suffer a fast radiative cooling, $p\approx 3.2$ and the scattered photons in the lab frame have a power-law distribution, $$E^2\,{ dn_\gamma\over dE}\propto E^2\, {dn_e\over dE}\,
\sigma_{KN} \sim E^{-(p+1)/2}\!\sim\! E^{-1.6}\, .
\label{KNFNU}$$
SSC in GRBs
-----------
Like in blazars, the radiation produced by ICS of self produced SR, the so called ‘SSC emission’, is Doppler boosted by the CB relativistic motion and extends almost to TeV energies above which it is suppressed by the Klein-Nishina effect. Roughly, the SSC energy flux $\nu\, F_\nu$ in the Thomson regime first increases with energy like $E^{(3\!-\!p)/2},$ where $p\!\approx \!2.2$ until $E\!\sim\! E_p\!\approx\!2\, m_e\,c^2\, \gamma\, \delta/3\,(1+z)$, then it changes into a plateau/shallow decrease like $E^{\!-\!(p\!-\!2)/2}$, due to the cooling break in the HE electron distribution, until it enters the Klein-Nishina regime where it decreases like $E^{\!-\!(p\!+\!1)/2}.$ For typical GRBs (DDD2009a) where $\delta\!\approx\!\gamma\!\sim\! 1000$ and $1\!+\!z\!\sim\!3.2$, the peak of the unabsorbed energy flux density is around $E_p\!\sim\! 100$ GeV. In very luminous GRBs, the peak energy $E_p$ of the SSC may approach TeV, which is above the LAT energy range, and then could be detected only in relatively very nearby GRBs where absorption in the IGM by pair-production can be neglected. Beyond the peak-energy, the SSC is cut off at an energy $E_c\!=\! m_e\, c^2\, \gamma_{e,max}\, \gamma/(1\!+\!z)$ by the cut-off in the electron spectrum at $\gamma_{e,max}=\sqrt{6\,e/\sigma_T\,B_{eq}}$ in the CB rest frame when the energy loss-rate by SR exceeds the rate of Fermi acceleration by the strong equipartition magnetic field $B_{eq}\!\approx\! \sqrt{4\,\pi\, n\,m_p\,
c^2}\,\gamma$ generated in the CB by its collision with the wind/ejecta ($n\, m_p $ is the density of the wind and $\sigma_T\,\approx\! 0.665\times 10^{-24}\, {\rm cm^{-2}}$ is the Thomson cross section).
In the CB model each GRB pulse have a sub-MeV energy flux density which is well approximated by an exponentially cutoff power-law $F_E\!\propto\! E^{\!-\!\beta_g}\, e^{-E/E_p(t)}$ where $\beta_g\!\sim\! 0$, $E_p(t)\!\sim E_p(0)\,t_p^2/(t^2\!+\!t_p^2)$, and $t$ is the time after the beginning of the pulse (e.g. DDD2009a and references therein). Towards the end of each ICS pulse simultaneous ‘prompt’ optical and HE emissions begin. They have a power-law spectrum $F_{\nu}\!\propto\!\nu^{-\beta}$, with $\beta\!\sim\!0.5$ in the optical band and $\beta_O\leq \beta\leq \beta_X$ in the LAT band. The time integrated spectrum over a single GRB pulse (or several unresolved GRB pulses) appears to be an exponentially cutoff power-law spectrum with $E_p$ in the keV-MeV range plus a power-law component. The power-law component dominates both at low energies because $\beta_g{\hbox{\rlap{$^<$}$_\sim$}}0$ whereas $\beta_{OX}\!\sim\! 0.8 $, and at high energies because of the exponential cut-off of the spectrum of the prompt MeV emission.
External ICS of prompt GRB photons
==================================
The thermal electrons (and the Fermi accelerated ones) in the CBs that have the CB bulk motion Lorentz factor $\gamma$ in the observer frame can boost by ICS the energy of prompt sub-MeV GRB photons which suffer Compton scattering in the wind and are overtaken shortly by the CBs. However, this mechanism is strongly suppressed by the Klein-Nishina effect for GRB photons whose energy is above $\!\sim\!1$ eV and thus does not contribute effectively to the production of HE photons.
Hadronic production of HE photons by thermal protons
====================================================
The radiative decay of $\pi^0$’s produced in collisions between thermal CB nuclei and wind nuclei produce sub-TeV $\gamma$-rays. Neglecting magnetic deflections of the CB nuclei in the wind, the equivalent isotropic energy of sub-TeV $\gamma$’s produced that way is given by, $$E^{HE}_{iso}\sim 0.08\, E_{CB}\, \sigma_{in}\, N_{wind}\, \delta^2\,,
\label{EISOMIN}$$ where 0.08 is the fraction of the incident kinetic energy of protons that is carried by the $\pi^0$’s produced in HE pp collisions, $\sigma_{in}\!\approx\!40$ mbarn is the HE pp inelastic cross section, $N_{wind}\!\sim \! 10^{21\pm 1}{\rm cm^{-2}}$ is the typical column density of the GRB environment as inferred from the spectral measurement with the Swift X-ray telescope (XRT), $E_{CB}\!\sim 10^{50}$ erg is the canonical kinetic energy of a CB ejected in core collapse supernovae which produce GRBs (DD2004), and $\delta^2$ is the relativistic beaming factor in the CB model. These numbers yield $E^{\pi^0}_{iso}\!\sim\ 3.2\times 10^{50\pm 1}$ erg, which is much smaller than the typically observed equivalent isotropic gamma ray energy of the HE component in GRBs with a detectable HE emission.
Hadronic production of HE photons by Fermi accelerated protons
==============================================================
The wind nuclei which enter a CB are Fermi acceleration through magnetic deflections by the turbulent magnetic fields in the CB. Their total path-length inside a CB before they escape can be much larger than their mean free path for hadronic collisions. In that case, a fraction $\!\sim\! 0.08$ of the CB’s kinetic energy is converted to HE gamma rays through $\pi^0$ production and decay and a similar fraction is converted through gamma ray production in the hadronic showers by radiative decays, SR from charged leptons and ICS of the SR in the CBs . Due to Feynman scaling, these HE $\gamma$-rays have a power-law spectrum with a power-law index equal to that of the Fermi accelerated nuclei, i.e., $dn_\gamma/dE\!\propto\! E^{-2.2}$. The total equivalent isotropic energy is then bounded by a fraction $0.08{\hbox{\rlap{$^<$}$_\sim$}}k {\hbox{\rlap{$^<$}$_\sim$}}0.25$ of the kinetic energy of the jet which is converted into HE $\gamma$-rays that are beamed into a solid angle $\sim \pi\, \delta^{-2}$. This yields the upper limit: $$E^{HE}_{iso}\leq k\, \times 10^{56}\, \left[{E_{CB}\over 10^{50}\,
{\rm erg}}\right]\, \left[{\delta\over 1000}\right]^2\, {\rm erg}.
\label{EISOMAX}$$ In the CB model, the exact value of $k$ and consequently of $E^{HE}_{iso}$ and the early-time lightcurve of the HE ‘hadronic’ $\gamma$ ray emission before the CBs enter the ISM are strongly model-dependent and involve many unknown parameters. However, when the jet enters the ISM, the lightcurve is similar to that of the X-ray afterglow until the jet break/bend (DDD2009a).
Roughly, the same spectra, $E_{iso}$, and lightcurves are expected for the HE $\gamma$-ray emission and the $\nu_\mu$ emission in GRBs from hadronic production and decay of $\pi^0$ and $\pi^{\pm}$, respectively.
Comparison with observations
============================
The rapid localization of a few degrees, which is delivered by the Fermi GBM, is not precise enough for a rapid localization and follow-up by current optical telescopes. The localization of the HE emission detected by the LAT seems good for optical follow-up but, so far, it is delivered with a large delay compared to the GRB itself. Consequently, the coincidence in time between the HE emission and the optical emission and the similarity between their light curves, which are predicted by the CB model, cannot be tested yet in individual GRBs. However, the decay of the prompt optical flares seems to satisfy the CB model prediction (DDD2009) of a power-law decay with an index $\alpha$=1+$\beta_0\!\sim 1.5$. Such a power-law decay seems to agree with the observed power-law decay of the HE emission that was detected by the Fermi LAT in 11 GRBs so far (see, e.g., Ghisellini1, Ghirlanda & Nava 2009). Moreover, the spectrum of the HE emission which was measured by the Fermi LAT seems to be described well by a power-law with a spectral index which satisfies, admittedly within large errors, the relation, $\Gamma_{LAT}\!\approx\!\Gamma_X$, predicteby the CB model, where $\Gamma_X$ is the photon index of the [*late-time X-ray AG*]{}. This is shown in Table \[t2\] and in Fig. \[f1\] where we plotted the ratio $\Gamma_{LAT}/\Gamma_X$ for 7 GRBs that were observed both by the Fermi LAT and by the Swift XRT. Note that the average ratio is consistent with 1 with a small error unlike large error in this ratio for the individual GRBs. Below we compare in detail the CB model predictions and the observations of two representative GRBs with HE emission that was detected by LAT, the long GRB 090902B and SHB 090510.
GRB 090902B
-----------
[**Observations:**]{} The Fermi Gamma-ray Burst Monitor (GBM) triggered on and localized the bright burst GRB 090202B on 2009 September 2 at 11:05:08.31 UT. The burst had a multi-peak structure with the brightest peak around 14 sec after trigger. It was also detected by the INTEGRAL and Suzaku satellites. The sub-MeV ended approximately 25 seconds after trigger. Emission above 100 MeV was detected by the Fermi LAT up to 1k sec after the GBM trigger when the Earth’s limb was starting to enter its field-of-view, with 39 photons above 1 GeV (de Palma et al. 2009). The highest energy photon had $E$= 33.4 +2.7/-3.5 GeV and arrived 82 seconds after the GBM trigger. The emission declined like $t^{\!-\!1.5}$ until the LAT observations were interrupted by entry of the Earth’s limb into its field of view, but analysis of data an hour after trigger, when the source location was again unocculted, showed that any later emission lied below the LAT sensitivity (Bissaldi et al. 2009). At a redshift of $z$=1.822 (Cucchiara et al. 2009), the fluence of $(4.36\pm 0.06)\times 10^{-4}\,
{\rm erg\, cm^{-2}}$ during the first 25 seconds of the prompt emission yields $E_{iso}\!=\!(3.63 \pm 0.05)\times 10^{54}$ erg isotropic equivalent $\gamma$-ray energy in the 10 keV-10 GeV range.
The X-ray follow-up observations with the Swift X-ray telescope (XRT) started only 12.5 hours after the GBM trigger. The measured X-ray spectrum can be fit by an absorbed power-law model with a photon index of $\Gamma\!=\! 2.1\pm 0.3$ and a rest frame column density of $(3.4 \pm
0.9)\times 10^{22}\, {\rm cm^{-2}}$ at $z$=1.822 in addition to the Galactic column density in the direction of the burst (Stratta et al. 2009).
The earliest ground-based optical observations were obtained only $\!\sim\!1.4$ h after trigger by ROTSE-IIIa (Pandey et al. 2009). The burst was later detected also in the optical, infra red and radio.
[**Interpretation:**]{} From the LAT detection of the highest energy photon of 11.16 (+1.48/-0.58) GeV during the prompt phase Bissaldi et al. (2009) inferred a minimum value of the bulk Lorentz factor $\gamma(0)\! \geq\! 1000$ of the source using the flux variability time scale of 53 ms and the constraint that the opacity for $e^{\pm}$ pair-production in the source is less than unity so that such photons can escape outside the source (Fenimore et al. 1993). A similar value was derived for GRB 080916C (Abdo et al. 2009a). [*Such values of the bulk motion Lorentz factor of the jet of CBs were long advocated by the CB model (e.g., DDD2002; DD2004) and were used to explain the typical photon energy and isotropic equivalent energy of long GRBs.*]{} For standard candle GRBs, the largest observed values of $E_{iso}$ and $E_p$ are obtained when the GRBs are viewed from very near axis, i.e., with a viewing angle $\theta^2\!\ll\! 1/\gamma(0)^2$. In that case the Doppler factor of the CBs is $\delta(0)\!\approx \!
2\, \gamma(0)$. In the CB model, the isotropic gamma ray energy of such GRBs is (DD2004): $E_{iso}\!\approx\! 8\times 10^{52}\,(\delta(0)/1000)^3\, N_{CB}$ erg. It yields $E_{iso}\!\geq\! 3.2 \times 10^{54}$ erg for a jet of $N_{CB}$=5 canonical CBs (producing the 5 main peaks in the lightcurve of GRB 090902B) and $\delta(0)\!=\!2\, \gamma(0)\!\geq\! 2000$. This value agrees with the value $E_{iso}\!\approx\!3.63\times 10^{54}$ erg that was inferred by Bissaldi et al. (2009) from the Fermi measurements.
The time-averaged spectrum in the 100-1000 keV during the 25 sec of the sub-MeV GRB prompt phase was well fitted by a power-law with photon index $\Gamma$=0.91+/-0.10 and an exponential cutoff with $E_p$=885+/-0.39 keV (Terada et al. 2009). The photon index $\beta_g$=$\Gamma$-1=-0.09+/-0.10 is consistent with the CB model expectation $\beta_g\!\sim\!0$. The rest frame value of the peak energy $(1\!+\!z)\,E_p\!=\!2550 \pm 112$ keV is also consistent with that expected in the CB model as shown in Fig. (\[f2\]): it satisfies the so called ‘Amati correlation’ which follows from the CB model and was predicted (see e.g., DDD2007 and references therein) long before it was discovered.
The GRB peaks are not well resolved in the Fermi GBM data and do not allow a stringent test of the CB model predictions for the temporal and spectral behaviour of the individual prompt MeV peaks/flares.
Due to insufficient statistics, also the early-time lightcurve of the HE emission, which was measured by the Fermi LAT, was not well resolved into separate peaks. However, as shown in Fig. (\[f3\]), it is well described by Eq. (\[ICSOP\]) assuming a single effective CB with the parameters listed in Table \[t3\]. Note that its predicted decline, until taken over by its shallow decay of the SSC in the ISM is well described by the expected decline, $F_\nu\!\propto\! t^{\!-\!1!\-\!\beta_O}$ (see Eq. (\[ICSOP\]). Also the photon spectral index of the HE emission as measured by Bissaldi et al. 2009, $\Gamma\!=\!1.93\pm 0.03$, as predicted, is equal within errors to the measured spectral index in the X-ray band, $\Gamma_X\!=\!2.1\pm 0.3$ inferred by Stratta et al. (2009) from the Swift XRT data.
No early-time optical data is available to test the CB model prediction that the prompt optical emission and the high energy emission have the same temporal behaviour. In the CB model each prompt MeV peak/flare has a delayed HE peak that decays asymptotically like $t^{\!-\!1\!-\!\beta_O}$ and the time-lag between the prompt MeV emission and the high energy emission in each pulse is of the order of the pulse width, while the bulk of the high energy emission is delayed typically by a time comparable to the duration of the MeV component. There are indications for such correlations in the data obtained by the Fermi LAT and GBM on GRB 090902B and 080916C.
In the CB model, the observed late time behaviour of the lightcurve of the unabsorbed X-ray afterglow is described by a power-law $F_\nu\!\propto\!
t^{\!-\!\alpha_X}\,\nu^{\!-\!\beta_X}$ with $\alpha_X$=$\beta_X$+1/2. A best fit to the Swift XRT data shown in Fig. (\[f4\]) yields $\alpha_X\!=\!1.42\pm 0.1$. This value is consistent within errors with $\alpha_X\!=\!1.6\pm 0.3$ expected from the spectral index $\beta_X\!=\!1.1\pm 0.3$ inferred by Stratta et al. (2009) from the Swift XRT data.
SHB 090510
----------
[**Observations:**]{} SHB 090510 is a short/hard burst at redshift $z$=0.903 (Rau et al. 2009) detected by Fermi (Guiriec et al. 2009), AGILE (Longo et al. 2009), Swift (Hoversten et al. 2009), Konus-Wind (Golenetskii et al. 2009) and Suzaku (Ohmori et al. 2009). The Fermi GBM triggered on a precursor while the main emission episode started $\sim 0.5$ sec after trigger and lasted $\sim 0.2$ sec. Its lightcurve consists of 7 main peaks. The emission detected by the Fermi LAT started 0.65 s after the trigger and lasted $\sim 200$ sec. The photon with the highest energy of $31 \pm 3$ GeV arrived 0.829 sec after trigger (Abdo et al. 2009b). The equivalent isotropic gamma ray energy before the onset of the high energy emission was $E_{iso}$=(3.91 -0.88/+1.91)$\times 10^{52}$ erg. The time integrated photon spectrum measured by AGILE (Giuliani et al. 2009) was fitted with an exponentially cutoff power-law with a photon power-law index $\Gamma$=0.65 (-0.32/+0.28) and a cutoff energy $E_c$=2.8 (-0.6/+0.9) MeV (rest frame peak energy $E_p$=7.19 (-1.54/+2.31) MeV). The time integrated spectrum between 0.5 and 1 sec after trigger was best fitted by adding a second component -a power-law component with a photon index $\Gamma\!=\!1.62 \pm 0.03$- and resulted with $E_p\!=\!3.9 \pm 0.3$ MeV and $E_{iso}\!=\!(1.08 \pm 0.06)\times 10^{53}$ erg (Abdo et al 2009b). The high-energy spectral component accounts for $\sim 37\%$ of the total fluence. The bulk of the photons above 30 MeV arrived $258 \pm 34$ ms later than those below 1 MeV.
[**Interpretation:**]{}
The detection of a 31 GeV photon during the first second sets the highest lower limit on the bulk motion Lorentz factor of the source, $\gamma\!>\!1200$ (Abdo et al. 2009b). This limit is consistent with the typical value $\gamma(0)\!\sim\!1400$ advocated by the CB model for ordinary SHBs (DDD2009b). The large value of $E_p$ probably requires $\gamma(0)\!\sim\!2000$ and a small viewing angle $\theta^2\ll
1/[\gamma(0)]^2$ implying $\delta(0)\!\approx \! 2\, \gamma(0)$ and $E_{iso}$ larger roughly by a factor 23 than $5\times 10^{51}$ erg, the mean $E_{iso}$ of SHBs.
Due to insufficient statistics, the early-time lightcurve which was measured by the Fermi LAT cannot be resolved reliably into separate HE peaks. The lightcurve of the blended peaks inferred from the LAT measurements is well described by Eq. (\[ICSOP\]) for a single effective CB with the parameters listed in Table \[t3\] as shown in Fig. (\[5\]). Also the photon spectral index of the HE emission as measured by Giuliani et al. (2009), $\Gamma\!=\!1.58$ (-0.11,+0.13), and by Bissaldi et al. (2009), $\Gamma\!=\!1.62\pm 0.03$, is roughly within errors that of the spectral index $\Gamma_X\!=\!1.792$ (+0.071/-0.051) inferred by Evans et al. (2009) from the Swift XRT data. No early-time optical data is available to test the CB model prediction that the prompt optical and high energy emissions have the same temporal behaviour. The two spectral components (a prompt MeV cutoff power-law component and a delayed HE power-law component), show a significant temporal correlation (Abdo et al. 2009b) as expected in the CB model. Lag-times within the MeV band and the HE band were not detected, as expected in the CB model (DDD2009b). The bulk of the photons above 30 MeV arrived $258 \pm\ 34$ ms later than those below 1 MeV. This time lag is comparable to the duration of the sub-MeV emission, as expected in the CB model.
In the CB model the asymptotic decline of the X-ray afterglow in $n\!\propto\! 1/(r\!-\!r_c)^2$ density beyond $r_c$ is given by $F_\nu\!\propto\! t^{\!-\!\alpha}\,\nu^{\!-\!\beta}$ with $\alpha$=$\beta$+1=$\Gamma$ (e.g., Eq. 32 in DDD2009b). The Swift XRT lightcurve repository (Evans et al. 2009) reports $\Gamma$=1.792(+0.071/-0.051) from a spectral analysis of the X-ray afterglow, in good agreement with the index $\alpha$=1.89$\!\pm\!$0.06 of the best fit power-law decline beyond 1100 sec after trigger (see Fig. (\[6\])).
Conclusions
===========
In the cannonball (CB) model, high energy emission from GRBs and SHBs is a natural consequence of the model. The dominant leptonic and hadronic emission mechanisms are ICS of SR by Fermi accelerated electrons in the collision of the jet of highly relativistic CBs with the wind/ejecta blown from the progenitor or companion star long before the GRB, and the decay of $\pi^0$ produced in hadronic collisions between the CB nuclei and the nuclei of the hadronic matter (ejecta/wind/ISM) that the jet passes through. The main predictions of the model for the early-time high energy emission are:
- Each prompt keV-MeV pulse is followed by a delayed HE emission which lasts much longer.
- The HE emission coincides in time with the optical emission.
- The light curve of the HE emission is roughly proportional to that of the unextinct optical emission. The decay of both the ‘prompt’ optical and the HE emissions is a power-law with an index $\alpha$=1+$\beta_0\!\sim 1.5$.
- The spectrum of the HE component is a simple power-law with a spectral index approximately equal to that of the X-ray afterglow, i.e., $\Gamma_{LAT}\!\approx\!\Gamma_{X}$.
- The HE emission extends to very high energies, where the observed radiation is strongly attenuated by $e^+e^-$ pair productin in the EBL.
- The equivalent isotropic energy of the HE component in bright GRBs can reach, and even exceed that in the sub-MeV range.
- The neutrino counterpart of the hadronic emission of HE $\gamma$-rays from the most luminous GRBs is barely detectable in $km^3$ underwater/under-ice Cherenkov neutrino telescopes (DD2008).
These predictions are consistent with the present available data on high energy emission from GRBs obtained from the gamma ray satellites and from the large air, ground and underground Cherenkov telescopes. In particular, the main observed properties of the HE emission measured with the Compton, Fermi and AGILE gamma ray satellites are well reproduced by the CB model as demonstrated in this paper for GRB 090902B and SHB 090510.
An observational proof of the ICS origin of the HE gamma ray emission from GRBs requires simultaneous detections of the prompt optical and HE emissions. It is highly desireable that the LAT team improves their automatic analysis, in order to deliver a GRB position within a few seconds after the LAT detection. Even a few tens of seconds will be very useful. This will provide a stringent test of models of HE emission from GRBs (and from other HE transient sources such as blazars, microquasars, pulsars, etc) and help pin down their production mechanism. Extremely optically-luminous GRBs, such as GRB 080319 where the prompt optical emission was resolved into individual flares (Racusin et al. 2008), may show that also the prompt high energy emission consists of HE flares which are associated with and follow promptly each individual keV-MeV pulse, as expected in the CB model.
Abdo, A. A., et al. 2009a, Science, 323, 1688
Abdo, A. A., et al. 2009b, arXiv:0908.1832
Aharonian, F., et al. 2006, MNRAS, 371, 1705
Atkins, R., et al. 2003, ApJ, 583, 824
Bissaldi, E., et al. 2009, arXiv:0909.2470v1
Cucchiara, A., et al. 2009, GCN Circ. 9873
Dado, S. & Dar, A. 2005, ApJ, 627, L109
Dado, S., Dar, A. & De Rujula, A. 2002, A&A, 388, 1079 (DDD2002)
Dado, S., Dar, A. & De Rújula, A. 2007, ApJ. 663, 400 (DDD2007)
Dado, S., Dar, A. & De Rújula, A. 2009a, ApJ, 696, 994 (DDD2009a)
Dado, S., Dar, A. & De Rújula, A. 2009b, ApJ, 693, 311 (DDD2009b)
Dar, A. & Plaga, R. 1999, A&A, 349, 259
Dar, A. & De Rújula, A. 2004, Physics Reports, 405, 203 (DD2004)
Dar, A. & De Rújula, A. 2009, Physics Reports, 466, 179 (DD2008)
de Palma, F., et al. 2009, GCN Circ. 9867
Dermer, C. D., Chiang, J. & Mitman, K. 2000, ApJ, 537, 785
Dingus, B. L., 1995, Astrophys. & Space Sci. 231, 187
Dingus, B. L. 2001, AIP Conf. Proc. 558, 383
Evans, P. A., et al. 2009, MNRAS, submitted, arXiv:0812.3662
Fenimore, E. E., Epstein, R. I. & Ho, C. 1993
Galama, T. J., et al. 1998, Nature, 395, 670
Ghirlanda, G., Ghisellini, G., & Nava, L. 2009, arXiv0909.0016
G. Ghisellini, G., Ghirlanda, G. & Nava, L. 2009, arXiv:0910.2459
Giuliani, A., et al. 2008, A&A, 491, L25
Giuliani, A., et al. 2009, arXiv0908.1908
Golenetskii S., et al. 2009, GCN Circ. 9344
Gonzalez, M. M., et al. 2003, Nature, 424, 749
Guiriec S., et al. 2009, GCN Circ. 9336
Hoversten E. A., et al. 2009, GCN Circ. 9331
Hurley, K., et al. 1994, Nature, 372, 652
Kaneko, Y., et al. 2008, ApJS, 166, 298
Longo F., et al. 2009, GCN Circ. 9343
Mészáros, P. 2006, Rept. Prog. Phys. 69, 2259
Milgrom, M., & Usov, V. 1995, ApJ, 449, L37
Nikishov, A. I. 1961, Zh. Eksp. Teor. Fiz., 41, 549 (English translation:Sov. Phys. JETP 14, 392 \[1962\])
Ohmori N., et al. 2009, GCN Circ. 9355
Pandey, S. B., et al. 2009 GCN Circ. 9878
Piran, T. 2005, Rev. Mod. Phys. 76, 1143
Primack, J. R. et al. 2005, AIP Conf. Proc. 745, 23
Racusin J. L., et al. 2008, Nature, 455, 183
Rau, A., et al. 2009. GCN Circ. 9353
Shaviv, N. & Dar, A. 1995, ApJ, 447, 863
Stratta, G., DElia, V., & Perri, M. 2009, GCN Circ. 9876
Terada, Y., et al. 2009, GCN Circ. 9897
Totani, T. 1998, ApJ, 502, L13
Vietri, M. 1995, ApJ, 453, 883
Waxman, E. 1995, Phys. Rev. Lett. 75, 386
Waxman, E. & Bahcall, J. N. 1997, Phys. Rev. Lett., 78, 2292
Zhang, B. 2007, ChjAA, 7, 1
[llllc]{} Thermal e’s & CB & Glory photons & ICS & keV-MeV\
Fermi accelerated e’s& CB-Wind Collision & CB Magnetic Field & SR & UVOIR\
Fermi accelerated e’s & CB-Wind Collision & Self SR & ICS (SSC) & HE\
Thermal p’s & CB & Wind/Ejecta & $\pi^0$ decay & Sub-TeV\
Fermi accelerated p’s &CB-Wind Collision & CB Nuclei & $\pi^0$ decay& UHE\
\[t1\]
[llllc]{} 080825C & 0 & 200 & $1.96 \pm 0.30$ &\
080916C & 0 & 200 & $2.09 \pm 0.12$ & 2.03 (+0.43, -0.43)\
081024B & 0 & 5 & $1.64 \pm 0.47$ &\
090217 & 0 & 1 & $2.22 \pm 0.40$ &\
090323 & 0 & 400 & $2.05 \pm 0.20$ & 1.96 (+0.24, -0.21)\
090328 & 0 & 100 & $1.76 \pm 0.35$ & 1.82 (+0.23, -0.31)\
090510 & 0 & 7 & $2.15 \pm 0.12$ & 1.79 (+0.071, -0.051)\
090626 & 0 & 600 & $1.70 \pm 0.12$ &\
090902B & 0 & 320 & $2.32 \pm 0.16$ & 2.24 (+0.25, -0.25)\
090926A & 0 & 25 & $2.34 \pm 0.14$ & 2.15 (+0.072, -0.071)\
091003 & 0 & 100 & $1.85 \pm 0.25$ & 1.87 (+0.15, -0.16)\
\[t2\]
[lllllllc]{} GRB 090902B & 0.63 & 6.28 & 0.14& 0.118 $cm^{-2}\, s^{-1}$ & 0.51 & $0.93\pm 0.03$ & $1.10\pm 0.30$\
SHB 090510 & 0.662 & 0.227 & 0.0167 & 36.4 $s^{-1}$ & 0.54 & $0.62\pm 0.03$ & $0.79\pm 0.07$\
\[t3\]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We investigate a $(n+1)$-dimensional generalized Randall-Sundrum model with an anisotropic metric which has three different scale factors. One obtain a positive effective cosmological constant $\Omega_{eff}\sim10^{-124}$ (in Planck unit) which only need a solution $kr\simeq50-80$ without fine tuning, and both the visible and hidden brane tensions are positive which results in the two branes to be stable. Then, we find that the Hubble parameter is seem to be a constant in a large region near its minimum, thus causing the acceleration of the universe. Therefore, the fine tuning problem also can be solved in this model. Meanwhile, the scale of extra dimensions is smaller than the observed scale but greater than the Planck length. This demonstrates that the observed present acceleration of the universe is caused by the extra-dimensional evolution rather than dark energy.
PACS numbers
: 04.50.-h, 11.25.Mj, 98.80.Es
author:
- 'Guang-Zhen Kang'
- 'De-Sheng Zhang'
- Long Du
- Dan Shan
- 'Hong-Shi Zong'
title: 'Cosmic Acceleration Caused by the Extra-Dimensional Evolution in a Generalized Randall-Sundrum Model'
---
Introduction
============
The current cosmic acceleration is an unexpected picture of the universe, revealed by the data sets of last two decades from astrophysics and cosmology [@Riess; @Perlmutter; @Bennett; @Netterfield; @Halverson; @Valent; @Zhang; @Verde; @Simon; @Moresco1; @Moresco2; @Ratsimbazafy; @Stern; @Moresco3]. These data, which coming from the cosmic microwave background radiation, supernovae surveys and baryon acoustic oscillations, etc, indicate that the universe consists of 4% ordinary baryonic matter, 20% dark matter and 76% dark energy [@Peebles]. Dark energy not only has an unknown form of energy but also has not been detected directly. Additionally, dark energy is very similar to the cosmological constant which was proposed by Einstein. In Planck unit, the observed value of the cosmological constant is an extravagantly tiny positive value of order $10^{-124}$. This is the well-known cosmological constant fine tuning problems [@Amendola]. There have been numerous attempts in order to solve this problem, such as quintessence, anthropic principle, $f(R)$ model, etc. [@Peebles88; @Wetterich; @Zlatev; @Caldwell; @Feng; @Sotiriou; @Dvali; @Bousso]. But none of these theories are problem-free. In astrophysics and cosmology, it is still a very important problem.
Another perspective for resolving the above described problem, which seems to be more radical, is the following: The dimensions of our universe must be four? Are there some extra dimensions which are too small to be observed? Does the evolving of these extra dimensions contribute to the current cosmic acceleration? If so, would this help in solving the cosmological constant fine tuning problem? Therefore, we have investigated some higher-dimensional theories [@KK; @NAH1; @NAH2; @RS; @Das; @Antoniadis; @Das1; @Sundrum; @Lykken; @Antoniadis1; @Visinelli; @Vagnozzi; @Paul; @Polchinski]. Among them, the Randall-Sundrum (RS) two-brane model [@RS], which has a natural solution to the hierarchy problem with warped extra dimension, has attracted our attention. The hierarchy problem is essentially a fine tuning problem which can be described as: why there is such a large discrepancy between the electroweak scale/Higgs mass $M_{EW}\sim1$TeV and the Planck mass $M_{pl}\sim10^{16}$TeV? In RS two-brane scenario, our universe is described by a five dimensional line element [@RS] $$\begin{aligned}
ds^{2}=e^{-2\sigma(y)}\eta_{\mu\nu}dx^{\mu}dx^{\nu}+r_{c}^{2}dy^{2},\end{aligned}$$ where $y$ is the extra dimensional coordinate, $r_{c}$ is the extra dimensional compactification radius, $e^{-2\sigma}$ is the well-known warp factor, the term $\sigma=kr_{c}|\phi|$, $k=\sqrt{-\Lambda/24M^{3}}$, $M$ is the five dimensional Planck mass. Then a large hierarchy is generated by the warp factor $e^{-2kr_{c}\pi}$, meanwhile one requires only $kr\approx10$. The cosmological constant fine tuning problem is similar to the hierarchy problem. In RS model, the visible brane is instable caused by the negative brane tension. Furthermore, the cosmological constant on the visible brane is zero which is not consistent with our data sets of last two decades [@Das; @Koley].
The above problems can be solved in a generalized RS braneworld scenario in which $g_{\mu\nu}$ replaces $\eta_{\mu\nu}$ in RS model [@Das]. In this scenario, the tension of the visible brane and the hidden brane can be both positive with a negative induced cosmological constant. It is very interesting because that both branes are stable [@Mitra; @SC1; @SC2; @SC3; @Banerjee; @Kang1]. In order to be consistent with the current constraints, the negative induced cosmological constant $\Omega$ should be transformed into the positive effective cosmological constant $\Omega_{eff}$. This positive $\Omega_{eff}$ can be obtained in a $(n+1)$-dimensional (-d) generalized RS model with two $(n-1)$-branes instead of two 3-branes [@Kang2]. In this model, adopting an anisotropic metric ansatz with two different scale factors, one obtain the positive effective cosmological constant $\Omega_{eff}\sim10^{-124}$ (in Planck unit) which only need a solution $kr\simeq50-80$ without fine tuning. The cosmological constant fine tuning problem can be solved quite well [@Kang2].
But there is no reason to exclude the possibility of the anisotropic metric ansatz with the form of scale factors more than two. In this paper, we investigate a $(n+1)$-dimensional generalized Randall-Sundrum model with an anisotropic metric which has three different scale factors. We obtain that $H_{1}$ has a lower bound $H_{1min}$. Near this minimum value, the Hubble parameter is seem to be a constant in a large region, thus causing the acceleration of the universe. Meanwhile, the scale of extra dimension is smaller than the observed scale but greater than the Planck length. This demonstrates that the observed present acceleration of the universe is caused by the extra-dimensional evolution rather than dark energy. Our work is organized as follows: In Sec. , by considering the two $(n-1)$-branes with the matter field Lagrangian in $(n+1)$-d generalized RS model, the $n$-d Einstein field equations are obtained. In Sec. , we focus on the evolution of $(n+1)$-brane solved from the above field equation with an anisotropic metric ansatz which has three different scale factors. Finally, the summary and conclusion are presented in Sec. .
$(n+1)$-d Generalized Randall-Sundrum Model
===========================================
We consider a $(n+1)$-d generalized RS braneworld model which is consistent with the Ref.[@Kang2]. The action $S_{n+1}$ is: $$\begin{aligned}
S_{n+1}=S_{bulk}+S_{vis}+S_{hid} \label{eq:S},\end{aligned}$$ where $S_{bulk}$ is the bulk action, $S_{vis}$ and $S_{hid}$ are the $(n-1)$-brane visible action and hidden action, respectively: $$\begin{aligned}
S_{bulk}&=&\int d^{n}xdy\sqrt{-G}(M^{n-1}_{n+1}R-\Lambda)\label{eq:S1},\\
S_{vis}&=&\int d^{n}x\sqrt{-g_{vis}}(\mathcal{L}_{vis}-V_{vis}),\\
S_{hid}&=&\int d^{n}x\sqrt{-g_{hid}}(\mathcal{L}_{hid}-V_{hid}),
\label{eq:S2}\end{aligned}$$ where $\Lambda$ denotes a bulk cosmological constant, $M_{n+1}$ is the $(n+1)$-d fundamental mass scale, $G_{AB}$ and $R$ are the $(n+1)$-d metric tensor and Ricci scalar respectively, $\mathcal{L}_{i}$ is the matter field Lagrangian of the visible and hidden branes, $V_{i}$ is the tension of the visible and hidden branes, here $i=hid$ or $vis$. In this $(n+1)$-d generalized RS scenario, the metric takes the form: $$\begin{aligned}
\label{eq:ds2}
ds^{2}=G_{AB}dx^{A}dx^{B}=e^{-2A(y)}g_{ab}dx^{a}dx^{b}+r^{2}dy^{2}.\end{aligned}$$ where $e^{-2A(y)}$ is known as the warp factor, Capital Latin $A,B,...$ indices run over all spacetime coordinate labels, $y$ is the extra dimensional coordinate of length $r$, Lowercase Latin $a,b=0,1,2,\cdot\cdot\cdot,n-1$ which is not include the coordinate $y$, $g_{ab}$ is the $n$-d metric tensor. Variation with respect to the metric $G_{AB}$ and after some easy manipulations, then modulo surface terms, one obtains $$\begin{aligned}
\label{eq:field}
R_{AB}-\frac{1}{2}G_{AB}R=\frac{1}{2M^{n-1}_{n+1}}\{-G_{AB}\Lambda+\sum_{i}[T^{i}_{AB} \nonumber\\
\times\delta(y-y_{i})-G_{ab}\delta_{A}^{a}\delta_{B}^{b}V_{i}\delta(y-y_{i})]\},\end{aligned}$$ where $R_{AB}$ is the $(n+1)$-d Ricci tensor, $T^{i}_{AB}$ is the $(n+1)$-d energy-momentum tensors. Note here the energy-momentum tensor is given by $T^{ia}_{b}=diag[-c_{i},c_{i},\cdot\cdot\cdot,c_{i}]$ [@Kang1; @Kang2], A solution to the Eq. (\[eq:field\]) with the metric tensor Eq. (\[eq:ds2\]) has been derived in Ref. [@Kang2], which reads $$\begin{aligned}
\label{eq:solution-}
A=-\ln[\omega\cosh(k|y|+c_{-})],\end{aligned}$$ where the constant $k\equiv\sqrt{-\Lambda/[M^{n-1}_{n+1}n(n-1)]}\simeq$ Planck mass, $\omega$ is given by $$\begin{aligned}
\omega\equiv\sqrt{\frac{-2\Omega}{(n-1)(n-2)k^{2}}},\end{aligned}$$ the term $c_{-}$ takes the form $$\begin{aligned}
c_{-}\equiv\ln\frac{1-\sqrt{1-\omega^{2}}}{\omega}.\end{aligned}$$ Meanwhile, a $n$-d Einstein field equations can be obtained $$\begin{aligned}
\
\widetilde{R}_{ab}-\frac{1}{2}g_{ab}\widetilde{R}=-\Omega g_{ab}. \label{eq:ndfield}\end{aligned}$$ where $\Omega$ is the induced cosmological constant on the visible brane, $\widetilde{R}$ and $\widetilde{R}_{ab}$ are the $n$-d Ricci scalar and Ricci tensor respectively,
Note that the solution derived above has the negative induced cosmological constant $\Omega$, Here we do not consider the situation for $\Omega>0$, since the tension on the visible brane is negative which results in instability [@Das; @Koley; @Kang1; @Kang2]. Then, an anisotropic metric is assumed to be of the following form [@Middleton; @Kang1; @Kang2]: $$\begin{aligned}
g_{ab}=diag[-1,a_{1}^{2}(t),a_{2}^{2}(t),a_{3}^{2}(t),\cdot\cdot\cdot,a_{n-1}^{2}(t)], \label{eq:gab}\end{aligned}$$ where $a_{i}$ is the scale factor. The case that the scale factors on the visible brane evolve with two different rates has been studied recently [@Kang2]. In this case, we can obtain the positive effective cosmological constant $\Omega_{eff}\simeq{10}^{-124}$ and only requiring $kr\simeq50-80$£¬ where for convenience, the Planck mass has been set to unity. So the cosmological constant fine tuning problem can be solved quite well. Furthermore, the three dimensional (3D) Hubble parameter $H(z)$ is consistent with cosmic chronometers dataset extracted from [@Valent; @Zhang; @Verde; @Moresco1; @Moresco2; @Moresco3; @Ratsimbazafy; @Stern; @Simon]. The observed 3D universe naturally shifts from deceleration expansion to accelerated expansion. It shown that the accelerated expansion of the observed universe is intrinsically an extra-dimensional phenomenon. But, there is no reason to make the scale factor evolve only with two kinds of rates. Therefore, we investigate the case that the scale factors on the visible brane evolve with three different rates.
Anisotropic Evolution of $(n-1)$-Brane
======================================
For the anisotropic metric Eq. (\[eq:gab\]) with three kinds of scale factors and the negative induced cosmological constant $\Omega\sim-10^{-124}$, the field equations (\[eq:ndfield\]) can be written: $$\begin{aligned}
\sum_{i}n_{i}(n_{i}-1)H_{i}^{2}+\sum_{i\neq j}n_{i}n_{j}H_{i}H_{j}=2\Omega, \label{eq:field1}\\
\sum_{i}n_{i}\dot{H}_{i}-\dot{H}_{1}+(\sum_{i}n_{i}H_{i})^{2}-H_{1}\sum_{i}n_{i}H_{i}=2\Omega, \label{eq:field2}\\
\sum_{i}n_{i}\dot{H}_{i}-\dot{H}_{2}+(\sum_{i}n_{i}H_{i})^{2}-H_{2}\sum_{i}n_{i}H_{i}=2\Omega, \label{eq:field3}\\
\sum_{i}n_{i}\dot{H}_{i}-\dot{H}_{3}+(\sum_{i}n_{i}H_{i})^{2}-H_{3}\sum_{i}n_{i}H_{i}=2\Omega, \label{eq:field4}\end{aligned}$$ where $i=1,2,3$, the terms $n_{1}$, $n_{2}$ and $n_{3}$ are the number of dimensions which evolve with three kinds of rates respectively, the Hubble parameter $H\equiv\dot{a}/a$, $\dot{H}_{i}$ is the first time derivative of ${H}_{i}$. Computing the sum of Eqs. (\[eq:field2\]), (\[eq:field3\]) and (\[eq:field4\]), yields a simplified expression for $\sum_{i}n_{i}H_{i}$: $$\begin{aligned}
\sum_{i}n_{i}H_{i}=-\chi_{1}\tan\beta, \label{eq:sumH}\end{aligned}$$ the term $\beta=\chi_{1}t+\theta_{0}$, $\theta_{0}$ is the initial phase angle which is determined by the scale of the formation of the brane, the terms $\chi_{1}$ takes the form $$\begin{aligned}
\chi_{1}=\sqrt{\frac{-2(n-1)\Omega}{n-2}}.\label{eq:chi1}\end{aligned}$$ It is convenient to redefine the sum of the Hubble parameters in the following $$\begin{aligned}
\sum_{i}n_{i}H_{i}\equiv f.\label{eq:f}\end{aligned}$$ Replacing Eqs. (\[eq:sumH\]) and (\[eq:f\]) in Eq. (\[eq:field2\]), then after some easy manipulations, one obtains $$\begin{aligned}
\dot{H}_{1}+H_{1}f=\dot{f}+f^{2}-2\Omega.\label{eq:H1eq}\end{aligned}$$ The solution of the above equation is $$\begin{aligned}
H_{1}=-\frac{\chi_{1}}{n-1}\tan\beta+c\sec\beta\label{eq:3h1},\end{aligned}$$ where $c$ is an integration constant. Eq. (\[eq:field1\]) can be rewritten as $$\begin{aligned}
n_{2}H_{2}^{2}+n_{3}H_{3}^{2}=-n_{1}H_{1}^{2}+f^{2}-2\Omega. \label{eq:3h2h3}\end{aligned}$$ The Eqs. (\[eq:3h1\]) and (\[eq:3h2h3\]) could be combined to give the following equation eliminating $H_{2}$ completely $$\begin{aligned}
(n_{3}+\frac{n_{3}^{2}}{n_{2}})H_{3}^{2}+\frac{2n_{3}}{n_{2}}(n_{1}H_{1}-f)H_{3}+[(\frac{1}{n_{2}}-1)f^2 \nonumber\\
+\frac{n_{1}^{2}}{n_{2}}H_{1}^{2}-\frac{2n_{1}}{n_{2}}fH_{1}+n_{1}H_{1}^{2}+2\Omega]=0. \label{eq:H3eq}\end{aligned}$$ Then the Hubble parameter $H_{3}$ can be obtained $$\begin{aligned}
H_{3}=-\frac{\chi_{1}}{n-1}\tan\beta-\frac{\chi_{3}+n_{1}c}{n_{2}+n_{3}}\sec\beta\label{eq:3h3},\end{aligned}$$ where the terms $\chi_{3}$ is $$\begin{aligned}
\chi_{3}=\sqrt{-2\frac{n_{2}}{n_{3}}(n_{2}+n_{3})\Omega-n_{1}\frac{n_{2}}{n_{3}}(n-1)c^{2}}.\label{eq:p2}\end{aligned}$$ Note here that we have chosen $H_{2}$ is always greater than $H_{3}$. Finally, using Eqs. (\[eq:sumH\]) and (\[eq:3h3\]), one can also obtain the Hubble parameter $H_{3}$ $$\begin{aligned}
H_{2}=-\frac{\chi_{1}}{n-1}\tan\beta+\frac{\chi_{2}-n_{1}c}{n_{2}+n_{3}}\sec\beta\label{eq:3h2},\end{aligned}$$ where the term $\chi_{2}$ is $$\begin{aligned}
\chi_{2}=\sqrt{-2\frac{n_{3}}{n_{2}}(n_{2}+n_{3})\Omega-n_{1}\frac{n_{3}}{n_{2}}(n-1)c^{2}}\label{eq:p1}.\end{aligned}$$ It is easy to see that the integration constant $c$ is constrained by Eqs. (\[eq:p2\]) and (\[eq:p1\]). The constant $c$ must be set to a value less than $c_{max}$ to guarantee that the value in the root is greater than zero, which yields $$\begin{aligned}
c\leq\sqrt{\frac{-2\Omega(n_{2}+n_{3})}{n_{1}(n-1)}}\equiv c_{max},\end{aligned}$$ where $c_{max}$ is the maximum value of $c$. It is evident from Eqs. (\[eq:3h3\]), (\[eq:p2\]), (\[eq:3h2\]) and (\[eq:p1\]) that, if $c=c_{max}$, $H_{2}$ is equal to $H_{3}$. For convenience, we define a parameter $d$ satisfying $$\begin{aligned}
c=dc_{max},\end{aligned}$$ where the value of $d$ is between $0$ and $1$. First, we investigate the effect of parameter $d$ on the Hubble parameters $H$. We choose $n_{1}=3$ which is most in line with the presently observed three dimensional (3D) space. Further setting $n_{2}=1$ and $n_{3}=1$, we plot the Hubble parameters $H_{1}$, $H_{2}$ and $H_{3}$ as a function of $\beta$ in Fig. \[fig:1\] with $d=0$, $d=0.5$ and $d=1$ respectively. In Fig. \[fig:1\](a)-(c), the three curves having same type (color) corresponds to three different value of $d$ respectively. In Fig. \[fig:1\](a), we plot the Hubble parameter $H_{1}$ as a function of $\beta$. When the parameter $d=0$ and $d=0.5$, the Hubble parameters $H_{1}$ monotonically decrease as $\beta$. And when $d>\sqrt{n_{1}/[(n_{2}+n_{3})(n-2)]}\equiv d_{min}=\sqrt{6}/4\approx0.612$ for $n_{1}=3$, $n_{2}=1$ and $n_{3}=1$, the Hubble parameter $H_{1}$ has a minimum $$\begin{aligned}
H_{1min}=\frac{\sqrt{c^2(n-1)^2-\chi_{1}^{2}}}{n-1},\end{aligned}$$ when the term $\beta$ in Eq. (\[eq:3h1\]) takes the form $$\begin{aligned}
\beta_{min}=\arcsin[\frac{\chi_{1}}{c(n-1)}].\end{aligned}$$ If $\beta\leq\beta_{min}$, the Hubble parameters $H_{1}$ is a monotonically decreasing function with time which is not in accordance with the cosmological observations and experiments. As can be seen easily from Fig. \[fig:1\](a), the the Hubble parameters $H_{1}$ has a minimum when $d=d_{min}\approx0.612$. This case may be consistent with the present observations because $H_{1}$ tends to a constant near the minimum, which can lead to accelerated expansion without the contribution of dark energy (or an inflaton field).
![\[fig:2\] The parameter $d_{eq}$ versus the number of extra dimensions $n_{2}$ and $n_{3}$ respectively. The figure on the left is the curve of the parameter $d_{eq}$ with $n_{3}$ when $n_{2}=1, 5, 30$, and the figure on the right depicts the evolution of $d_{eq}$ with $n_{2}$ when $n_{3}=1, 5, 30$.](H123_deq.eps)
Combining Eqs. (\[eq:chi1\]), (\[eq:3h1\]), (\[eq:3h2\]) and (\[eq:p1\]), we obtain $H_{1}=H_{2}$ if $c$ is satisfied by $$\begin{aligned}
c=\sqrt{\frac{-2\Omega n_{3}}{(n-1)(n_{1}+n_{2})}}\equiv c_{eq}.\end{aligned}$$ It is shown that $H_{2}$ tends to $H_{1}$ when $c\rightarrow c_{eq}$. In the case $c=c_{eq}$, the parameter $d_{eq}$ is defined as $$\begin{aligned}
d_{eq}=\frac{c_{eq}}{c_{max}}=\sqrt{\frac{n_{1}n_{3}}{(n_{1}+n_{2})(n_{2}+n_{3})}}.\end{aligned}$$ For $n_{1}=3$, $n_{2}=1$ and $n_{3}=1$, $d_{eq}=\sqrt{6}/4\approx0.612$ Since the extra dimension in our universe is not observed presently, the extra dimension should be very small, which requires that the extra dimensions cannot be too large and the current scale of extra dimensions are still outside the observable range. When $d$ is equal $d_{eq}$, the extra dimension Hubble parameter $H_{2}$ is converted to $H_{1}$. This case is inconsistent with the presently observed 3D space. The Hubble parameter $H_{2}$ of the extra dimensions is plotted in Fig. \[fig:1\](b) with $d=0, 0.5, 1$. It can be easily shown that the Hubble parameter $H_{2}$ has a minimum with $d=0$ and $0.5$, which can lead to accelerated expansion of the extra dimensions. We are not interested in this situation because that it is not in line with the observation. However, the Hubble parameter $H_{2}$ is always negative with $d=1$, which ensures that the extra dimensions always exceeds the observable range. Note here the ratio of $d_{min}$ to $d_{eq}$ is $$\begin{aligned}
\frac{d_{min}}{d_{eq}}=\sqrt{\frac{(n_{1}+n_{2})}{n_{3}(n_{1}+n_{2}+n_{3}-1)}}\leq 1,\end{aligned}$$ where $d_{min}/d_{eq}=1$ if and only if $n_{3}=1$. So we only consider the case $d\gg d_{eq}$ because that we obtain $d\geq d_{min}$ when $d\geq d_{eq}$.
In Fig. \[fig:2\], we have plotted the parameter $d_{eq}$ versus the number of extra dimensions $n_{2}$ and $n_{3}$ respectively. The figure on the left is the curve of the parameter $d_{eq}$ with $n_{3}$ when $n_{2}=1$, $5$, and $30$. It is shown that $d_{eq}$ monotonically increases with $n_{3}$. The parameter $d_{eq}$ tend to $\sqrt{n_{1}/(n_{1}+n_{2})}$ in the $n_{3}\rightarrow\infty$ limit. In particular, $d_{eq}\rightarrow\sqrt{3}/2\approx0.866$ for $n_{2}=1$. The figure on the right depicts the evolution of $d_{eq}$ with $n_{2}$ when $n_{3}=1$, $5$, and $30$. In this case, $d_{eq}$ monotonically decrease with $n_{2}$. The parameter $d_{eq}\rightarrow0$ in the $n_{2}\rightarrow\infty$ limit. To be consistent with observation, we should set $d$ to be greater than $0.866$ and closer to $1$. Further we set the constant $d=0.98$ in the following.
In Fig. \[fig:3\](a)-(c), we plot the Hubble parameters $H_{1}$, $H_{2}$ and $H_{3}$ as a function of $\beta$ with $n_{1}=3$ when $d=0.98$. From top to bottom, the three curves having same type (color) corresponds to $H_{1}$, $H_{2}$ and $H_{3}$ respectively for fixed value of $n_{2}$ and $n_{3}$. In Fig. \[fig:3\](a), we plot the Hubble parameter as a function of $\beta$ at $n_{2}=1$. From top to bottom, the three solid curves are correspond to $H_{1}$, $H_{2}$ and $H_{3}$ at $n_{3}=1$. The three dashed curves ($n_{3}=3$) and the three dotted curves ($n_{3}=5$) are similar to the above case. The Hubble parameter of the extra dimensions are always negative at $n_{3}=1$. With the increase of $n_{3}$, the coordinate of the minimum value of $H_{1}$ tends to $\beta=0$. Meanwhile, $H_{2}$ changes from positive to negative in the region near $\beta\sim-1.5$ and $H_{3}$ is closer to zero. In Fig. \[fig:3\](b) and (c), the Hubble parameters as a function of $\beta$ are shown with $n_{2}=3$ and $n_{2}=5$. $H_{2}$ is always negative with $n_{2}=5$ which is similar to the situation in Fig. (1) of Ref. [@Kang2]. $H_{1}$ has a lower bound $H_{1min}=\sqrt{0.98^{2}c_{max}^{2}-\chi_{1}^{2}/(n-1)^{2}}$ when $\beta_{min}=\arcsin\{\eta_{1}/[0.98(n-1)c_{max}]\}$. In the region near $\beta_{min}$, we get $\Omega_{eff}>0$ is of order $-\Omega$. This situation is similar to the case with two different scale factors, the negative induced cosmological constant $\Omega$ can be transformed into the positive effective cosmological constant $\Omega_{eff}$. It tells us that the observed current cosmic acceleration is intrinsically an extra-dimensional phenomenon rather than dark energy. The cosmological constant fine tuning problem can be solved by this extra-dimensional evolution.
From Eqs. (\[eq:3h1\]), (\[eq:3h3\]) and (\[eq:3h2\]) we can get the scale factors $a_{1}$, $a_{2}$ and $a_{3}$ of the form $$\begin{aligned}
a_{1}=a_{10}|\cos\beta|^{\frac{1}{n-1}}|\sec\beta+\tan\beta|^{\frac{c}{\chi_{1}}},\\
a_{2}=a_{20}|\cos\beta|^{\frac{1}{n-1}}|\sec\beta+\tan\beta|^{\frac{\chi_{2}-n_{1}c}{\chi_{1}(n_{2}+n_{3})}},\\
a_{2}=a_{30}|\cos\beta|^{\frac{1}{n-1}}|\sec\beta+\tan\beta|^{-\frac{\chi_{3}+n_{1}c}{\chi_{1}(n_{2}+n_{3})}},\end{aligned}$$ where $a_{10}$ , $a_{20}$ and $a_{30}$ are the scale factors when the brane forms. Further, the volume of the visible brane is obtain by $$\begin{aligned}
V_{b}=a_{1}^{n_{1}}a_{2}^{n_{2}}a_{3}^{n_{3}}=a_{10}^{n_{1}}a_{20}^{n_{2}}a_{30}^{n_{3}}\cos\beta,\end{aligned}$$ When the brane is just forming, there is no particular reason to make the scale factor different, so we choose $a_{10}=a_{20}=a_{30}$. If the initial scale of the brane is of order $10^{35}$ in planck unit, and consider the presently observed scale of our universe (about of order $10^{61}$), we obtain the scale of extra dimension is at least of order $10^{22}$ with $n_{2}=n_{3}=3$. It is shown that the scale of extra dimension should be much larger than Planck length. This ensures that physics are still valid in the evolution of the visible brane. We obtain $\theta_{0}$ is close to $-\pi/2$ if one want obtain a sufficiently small initial scale. So in the region of $\theta_{0}+\pi/2\ll\eta_{1}t\ll\pi/2$, the Hubble parameters $H_{1}$ is of the form: $$\begin{aligned}
H_{1}&\simeq&[\frac{c}{\chi_{1}}+\frac{1}{n-1}]\frac{1}{t} \nonumber\\
&=&\frac{1}{n-1}[1+d\sqrt{\frac{(n_{2}+n_{3})(n-2)}{n_{1}}}]\frac{1}{t}.\end{aligned}$$ When $n_{2}=n_{3}=1$ and $d=0.98$, $H_{1}$ is about $0.52t$ which is as similar as the radiation dominating eras. In the limit $n_{2}\rightarrow\infty$ (or $n_{3}\rightarrow\infty$), $H_{1}\simeq\sqrt{3}d/3t$.
Summary and Conclusion
======================
In conclusion, we investigate a $(n+1)$-d generalized Randall-Sundrum model with an anisotropic metric which has three different scale factors. In this model, one obtain the positive effective cosmological constant $\Omega_{eff}\sim10^{-124}$ (in Planck unit) which only need a solution $kr\simeq50-80$ without fine tuning. This is consistent with the case with two different scale factors.
In this model, the Hubble parameters $H_{2}$ tends to $H_{1}$ when the integration constant $d$ tends to $d_{eq}$. It is indicate that the Hubble parameter of observable dimensions is related to the value of the integral constant $d$. For convenience, here we have selected the Hubble parameter $H_{1}$ to show the observable dimensions. To be consistent with observation, we should set $d$ to be greater than $0.866$ and closer to $1$. Further setting the constant $d=0.98$, we obtain that $H_{1}$ has a lower bound $H_{1min}=\sqrt{0.98^{2}c_{max}^{2}-\chi_{1}^{2}/(n-1)^{2}}$ when $\beta_{min}=\arcsin\{\eta_{1}/[0.98(n-1)c_{max}]\}$. Meanwhile, the scale of extra dimension is smaller than the observed scale but greater than the Planck length. This demonstrates that the observed current cosmic acceleration is caused by the extra-dimensional evolution rather than dark energy (or an inflaton field).
We wish to acknowledge the supported by the Key Program of National Natural Science Foundation of China (under Grant No. 11535005), the National Natural Science Foundation of China (under Grant No. 11647087), Foundation for Young and Yiddle-Aged Teachers Basic Ability Improvement in Guangxi Universities (under Grant No. 2018KY0326) Special Foundation for Science and Technology Base and Talents in Guangxi (under Grant No. 2018AD19310) China Postdoctoral Science Foundation funded project (under Grant No. 2019M651750), Open project of state key laboratory of solid state microstructure physics (under Grant No. M31037), the Natural Science Foundation of Yangzhou Polytechnic Institute (under Grant No. 201917), and the Natural Science Foundation of Changzhou Institute of Technology (Grant No. YN1509).
[35]{}
A. G. Riess et al., Astron. J. **116** 1009 (1998). S. Perlmutter et al., Astrophys. J. **517**, 565 (1999). C. L. Bennett et al., Astrophys. J. Suppl. Ser. **148**, 1(2003). C. B. Netterfield et al., Astrophys. J. **571**, 604 (2002). N.W. Halverson et al., Astrophys. J. **568**, 38 (2002). A. Gómez-Valent and L. Amendolab, JCAP **04** (2018) 051. C. Zhang, H. Zhang, S. Yuan, T.-J. Zhang and Y.-C. Sun, Res. Astron. Astrophys. **14** (2014) 1221. R. Jiménez, L. Verde, T. Treu and D. Stern, Astrophys. J. **593** (2003) 622. J. Simon, L. Verde and R. Jiménez, Phys. Rev. D **71** (2005) 123001. M. Moresco et al., JCAP **08** (2012) 006. M. Moresco et al., JCAP **05** (2016) 014. A.L. Ratsimbazafy et al., Mon. Not. Roy. Astron. Soc. **467** (2017) 3239. D. Stern, R. Jiménez, L. Verde, M. Kamionkowski and S.A. Stanford, JCAP **02** (2010) 008. M. Moresco, Mon. Not. Roy. Astron. Soc. **450** (2015) L16. P. J. E. Peebles, B. Ratra, Rev. Mod. Phys. **75**, 559 (2003). L. Amendola, S. Tsujikawa, Dark energy: theory and observations, Cambridge University Press (2010). Peebles, P. J. E., and B. Ratra, Astrophys. J. **325** (1988) L17. C. Wetterich, Nucl. Phys. B , 668 (1988). I. Zlatev, L. M. Wang, and P. J. Steinhardt, Phys. Rev. Lett. **82**, 896 (1999). R. R. Caldwell, Phys. Lett. B **545**, 23 (2002). B. Feng, X. L. Wang, and X. M. Zhang, Phys. Lett. B **607**, 35 [2005]{}. T. P. Sotiriou, V. Faraoni, Rev. Mod. Phys. **82**, 451 (2010). G. R. Dvali, G. Gabadadze, and M. Porrati, Phys. Lett. B **485**, 208 (2000). R. Bousso and J. Polchinski, J. High Energy Phys. **0006** 006 (2000). Th. Kaluza, Sitzungseber. Press. Akad. Wiss. Phys. Math. Klasse 996 (1921); O. Klein, Z. Phys. **37**, 895 (1926); Nature (London) **118**, 516 (1926). N. Arkani-Hamed, S. Dimopoulos, and G. Dvali, Phys. Lett. B **429**, 263 (1998). N. Arkani-Hamed, S. Dimopoulos, and G. Dvali, Phys. Rev. D **59**, 086004 (1999). L. Randall and R. Sundrum, Phys. Rev. Lett. **83**, 3370 (1999). S. Das, D. Maity, and S. Sengupta, J. High Energy Phys. **05** 042 (2008) . I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos and G. Dvali, Phys. Lett. **B 436** 257 (1998). A. Das, D. Maity, T. Paul, S. SenGupta, Eur. Phys. J. C **77**, 813 (2017) R. Sundrum, Phys. Rev. D **59**, 085009 (1999). J. Lykken, L. Randall, J. High Energy Phys. **06**, 014 (2000). I. Antoniadis, Phys. Lett. B **246**, 377 (1990). L. Visinelli, N. Bolis, S. Vagnozzi, Phys. Rev. D **97**, 064039 (2018) S. Vagnozzi and L. Visinelli, arXiv:1905.12421. T. Paul and S. SenGupta, arXiv:1808.00172v1. J. Polchinski, String Theory. Vol. 2: Superstring theory and beyond, Cambridge University Press (1998). R.Koley, J. Mitra, and S. SenGupta, Phys. Rev. D **79**, 041902(R) (2009). J. Mitra, T. Paul, S. SenGupta, Eur. Phys. J. C (2017) 77:833. S. Chakraborty, S. SenGupta, Eur. Phys. J. C 75(11), 538 (2015). I. Banerjee, S. Chakraborty, and S. SenGupta, Phys. Rev. D **99**,023515 (2019). S. Chakraborty and S. SenGupta, Phys. Rev. D **92**, 024059 (2015). S. Chakraborty, and S. SenGupta, Eur. Phys. J. C **76**, 552 (2016). G.-Z Kang, D.-S. Zhang, L. Du, J. Xu and H.-S. Zong, Chin. Phys. C [**43(9)**]{} 095101 (2019). G.-Z Kang, D.-S. Zhang, H.-S. Zong, J. Li, arXiv: 1906.05442. C. A. Middleton, and E. Stanley, Phys. Rev. D **84**, 085013 (2011).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, we consider the design of robust linear precoders for MU-MISO systems where users have perfect Channel State Information (CSI) while the BS has partial CSI. In particular, the BS has access to imperfect estimates of the channel vectors, in addition to the covariance matrices of the estimation error vectors. A closed-form expression for the Average Minimum Mean Square Error (AMMSE) is obtained using the second order Taylor Expansion. This approximation is used to formulate two fairness-based robust design problems: a maximum AMMSE-constrained problem and a power-constrained problem. We propose an algorithm based on convex optimization techniques to address the first problem, while the second problem is tackled by exploiting the close relationship between the two problems, in addition to their monotonic natures.'
author:
- |
Department of Electrical and Electronic Engineering, Imperial College London, United Kingdom\
School of Electrical Engineering, Korea University, Seoul, Korea\
Email: {hamdi.joudeh10, b.clerckx}@imperial.ac.uk
bibliography:
- 'References.bib'
title: AMMSE Optimization for Multiuser MISO Systems with Imperfect CSIT and Perfect CSIR
---
AMMSE, Robust Design, Imperfect CSIT.
Introduction
============
The utilization of multiple antennas at the Base Station (BS) combined with simple single-antenna mobile devices could tremendously increase the spectral efficiencies of wireless networks [@Viswanath2003]. However, higher restrictions are imposed, particularly in the Downlink (DL) mode where highly accurate Channel State Information (CSI) is required at the BS. While the ability to provide accurate CSI at the Transmitter (CSIT) remains questionable, considerable work has been done to increase the robustness of transmission schemes initially designed assuming perfect CSI [@Shenouda2008; @Bogale2013; @Joham2010; @Gonzalez-Coma2013; @Vucic2009]. The robust design problem formulation is highly influenced by the nature of the CSI uncertainty, that varies depending on the context in which it occurs. Two main models have emerged to quantify this uncertainty: the stochastic-uncertainty model [@Shenouda2008; @Bogale2013; @Joham2010; @Gonzalez-Coma2013] and the bounded-uncertainty model [@Shenouda2008; @Vucic2009]. In this paper, we consider robust linear precoding design for Multiuser (MU) Multiple-Input Single-Output (MISO) systems where CSIT uncertainty is modeled stochastically. Particularly, the channel estimation error vectors are assumed to have Gaussian distributed entries. The performance metric considered is the Average Mean Square Error (AMSE), i.e. the expectation of the MSE over the distribution of the channel estimation error.
A similar setup was considered in [@Shenouda2008] where the sum AMSE was minimized subject to a total BS power constraint. While this yields an improved overall performance across users, it does not guarantee fairness. In [@Bogale2013], AMSE fairness-based designs for MU Multiple-Input Multiple-Output (MIMO) systems were proposed. However, it is assumed in both [@Shenouda2008] and [@Bogale2013] that CSIT and CSIR have identical uncertainties. While this assumption simplifies the AMSE based transceiver design, it ignores the fact that CSIR is likely to have higher quality. An AMSE UL-DL duality assuming imperfect CSIT and perfect CSIR was established in [@Joham2010]. Based on this duality, an algorithm was proposed to minimize the sum AMSE, where Monte-Carlo integration was applied to calculate expectations that depend on perfect CSI. A similar problem was addressed in [@Bashar2014], where various approximations were used instead of Monte-Carlo integration. *Contribution*: For MU-MISO systems with imperfect CSIT and perfect CSIR, we derive a closed-form expression for a Taylor approximation of the Average Minimum Mean Square Error (AMMSE), i.e. the AMSE obtained when MMSE receivers are applied. This approximation is used to formulate two fairness-based robust design problems:
1. The maximum AMMSE-constrained power minimization problem referred to as $\mathcal{P}$.
2. The power-constrained maximum AMMSE minimization problem referred to as $\mathcal{A}$.
We propose a fast-converging algorithm based on recursive convex optimization, that solves the non-convex problem $\mathcal{P}$. Furthermore, $\mathcal{A}$ is solved by exploiting its relationship with $\mathcal{P}$; an approach inspired by the work in [@Wiesel2006].
Problem $\mathcal{P}$ was addressed in [@Gonzalez-Coma2013] where the authors propose an algorithm based on the ideas in [@Joham2010]. However, this algorithm inherits the shortcoming of the approach in [@Joham2010], i.e. expectations are calculated in each iteration via Monte-Carlo integration. Furthermore, convergence could be very slow due to random initialization and limited per-iteration improvement. Results comparing the algorithm propose in this paper to the algorithm in [@Gonzalez-Coma2013] are given in Section \[Section\_Numerical Results\]. The rest of the paper is organized as follows: the system model is introduced in Section \[Section\_System Model\]. The AMMSE approximation is derived in Section \[Section\_AMSE Approximation\]. Algorithms that solve $\mathcal{P}$ and $\mathcal{A}$ are proposed in Section \[Section\_AMSE Robust Beamforming\]. Simulation results are presented in Section \[Section\_Numerical Results\] and Section \[Section\_Conclusion\] concludes the paper
*Notation*: Boldface uppercase letters denote matrices, boldface lowercase letters denote column vectors and standard letters denote scalars. The superscrips $(\cdot)^{T}$ and $(\cdot)^{H}$ denote transpose and conjugate-transpose (Hermitian) operators, respectively. $\mathrm{tr}(\cdot)$, $\mathrm{rank}(\cdot)$ and $\|\cdot\|$ are the trace, rank and Euclidian norm operators, respectively. $\mathrm{E}_{x}\{\cdot\}$ denotes the expectation w.r.t the random variable $x$. Finally, $\mathbf{X}\succeq0$ denotes that $\mathbf{X}$ is a Hermitian positive semidefinite matrix.
System Model {#Section_System Model}
============
We consider a BS equipped with $N_t\geq K$ antennas serving $K$ active single-antenna users. The vector of zero-mean mutually uncorrelated complex data symbols intended for the $K$ receivers is given as $\mathbf{s}=[s_{1},\ldots,s_{K}]^{T} \in\mathbb{C}^{K}$ where $\mathrm{E}\{\mathbf{s}\mathbf{s}^{H}\}=\mathbf{I}$. $\mathbf{s}$ is linearly precoded into the transmit vector $\mathbf{x}\in\mathbb{C}^{N_{t}}$ given as $$\label{Eq_x}
\mathbf{x}=\mathbf{Ps}=\sum_{i=1}^{K}\mathbf{p}_{i}s_{i}$$ where $\mathbf{P}=[\mathbf{p}_{1},\ldots,\mathbf{p}_{K}]$ is the precoding matrix and $\mathbf{p}_{i}\in\mathbb{C}^{N_{t}}$ is the precoding vector consisting of the beamforming weights for the $i$th user. The total transmit power is denoted as $P_{t}$ from which the transmit power constraint could be written as $\mathrm{E}\{\mathbf{x}^{H}\mathbf{x}\} = \mathrm{tr}(\mathbf{P}^{H}\mathbf{P})\leq P_{t}$. For the $k$th user, the received signal denoted by $y_{k}$ can be written as $$\label{Eq_yk}
y_{k}=\mathbf{h}_{k}^{H}\mathbf{x}+n_{k}$$ where $\mathbf{h}_{k} \in \mathbb{C}^{N_{t}}$ is the narrow-band channel impulse response vector between the $k$th user and the BS. $n_{k} \thicksim \mathcal{CN} ( 0 , \sigma^{2}_{n_{k}} )$ is the Additive White Gaussian Noise (AWGN) at the $k$th user receiver with variance $\sigma_{n_{k}}^{2}$. Throughout the paper, it is assumed that the noise variance is equal across all users i.e. $\sigma_{n_{k}}^{2}=\sigma_{n}^{2}, \ \forall k$. To obtain an estimate of the intended symbol, each user applies a scalar equalizer $g_{k}$ to its received signal such that $\hat{s}_k=g_{k}y_{k}$. The Mean Square Error (MSE) at the output of the $k$th receiver is given as $$\begin{aligned}
% \nonumber to remove numbering (before each equation)
% \nonumber
\varepsilon_{k} =& \ \mathrm{E}_{\mathbf{s},n_{k}}\{|\hat{s}_k - s_k|^{2}\} \\
\label{Eq_MSE_k}
=& \ |g_{k}|^{2}T_{k}-\mathbf{p}_{k}^{H}\mathbf{h}_{k}g_{k}^{H}-g_{k}\mathbf{h}_{k}^{H}\mathbf{p}_{k}+1\end{aligned}$$ where $ T_{k} = \sum_{i=1}^{K} \mathbf{p}_{i}^{H}\mathbf{h}_{k} \mathbf{h}^{H}_{k} \mathbf{p}_{i} + \sigma_{n}^{2}$. The Minimum Mean Square Error (MMSE) receiver can be calculated by setting the first derivative of $\varepsilon_{k}$, with respect to $g_{k}$, to zero. This yields: $g_{k}(\mathbf{h}_{k}) = \mathbf{p}_{k}^{H}\mathbf{h}_{k}T^{-1}_{k}$. The notation $g_{k}(\mathbf{h}_{k})$ is used to emphasise the dependency on the channel state $\mathbf{h}_{k}$. In a block-fading channel (which stays constant over a frame of symbols), the $k$th user can calculate $T_{k}$ empirically as the average received signal plus noise power i.e. $T_{k}=\mathrm{E}_{\mathbf{s},n_{k}}\{|y_{k}|^{2}\}$. Furthermore, the scalar effective channel $\mathbf{p}_{k}^{H}\mathbf{h}_{k}$ could be estimated via DL training, from which $g_{k}(\mathbf{h}_{k})$ is calculated. Plugging $g_{k}(\mathbf{h}_{k})$ into (\[Eq\_MSE\_k\]), the MMSE can be written as $$\begin{aligned}
% \nonumber to remove numbering (before each equation)
% \nonumber
\varepsilon^{\mathrm{MMSE}}_{k} =& \ 1-T^{-1}_{k}R_{k}
%\\
\label{Eq_M_MSE_k}
% =& \ T_{k}^{-1} \bigg( \sum^{K}_{\substack{i=1 \\ i\neq k}} \mathbf{p}_{i}^{H}\mathbf{h}_{k}\mathbf{h}_{k}^{H}\mathbf{p}_{i} + \sigma_{n}^{2} \bigg)\end{aligned}$$ where $R_{k}=\mathbf{p}_{k}^{H}\mathbf{h}_{k}\mathbf{h}_{k}^{H}\mathbf{p}_{k}$.
CSIT Uncertainty
----------------
The channel vector of the $k$th user can be written as $$\label{Eq_hk}
\mathbf{h}_k= \hat{\mathbf{h}}_{k} + \tilde{\mathbf{h}}_{k}$$ where $\hat{\mathbf{h}}_{k}$ and $\tilde{\mathbf{h}}_{k}$ denote the transmitter-side channel estimate and the channel estimation error, respectively. In this work, the channel estimation error is described statistically, i.e. the entries of $\tilde{\mathbf{h}}_{k}$ are i.i.d Zero-Mean Circularly Symmetric Complex Gaussian (ZMCSCG) with $\mathrm{E}\{\tilde{\mathbf{h}}_{k}\tilde{\mathbf{h}}_{k}^{H}\}=\sigma^{2}_{e_{k}}\mathbf{I}$. $\{\sigma^{2}_{e_{k}}\}_{k=1}^{K}$ and $\{\hat{\mathbf{h}}_{k}\}_{k=1}^{K}$ are assumed to be known by the transmitter, i.e. the BS has first and second order statistics of the channel where $\mathbf{h}_{k} \thicksim \mathcal{CN} \big( \hat{\mathbf{h}}_{k} , \mathbf{R}_{e_{k}} \big)$ and $\mathbf{R}_{e_{k}}=\sigma^{2}_{e_{k}}\mathbf{I} $. Furthermore, it is important to note that all derivations in this paper can be extended to arbitrary $\{\mathbf{R}_{e_{k}}\}_{k=1}^{K}$.
AMMSE and its Approximation {#Section_AMSE Approximation}
===========================
Given the statistical nature of the partial CSIT, the optimization of the precoding vectors $\{\mathbf{p}_{k}\}_{k=1}^{K}$ at the BS is carried out in terms of the AMMSE. The AMMSE for the $k$th user is denoted by $\bar{\varepsilon}^{\mathrm{MMSE}}_{k}$ and can be written as $$\label{Eq_AMSE_k}
\bar{\varepsilon}^{\mathrm{MMSE}}_{k} = 1 - \mathrm{E}_{{\mathbf{h}}_{k} | \hat{{\mathbf{h}}}_{k}}\{ T_{k}^{-1}R_{k} \}.$$ For notational brevity, $\bar{\varepsilon}^{\mathrm{MMSE}}_{k}$ will be referred to as $\bar{\varepsilon}_{k}$ in the rest of the paper where the use of MMSE receivers is implicit. Furthermore, $\mathrm{E}_{{\mathbf{h}}_{k} | \hat{{\mathbf{h}}}_{k} }\{\cdot\}$ will be referred to as $\mathrm{E}\{\cdot\}$. The MMSE and the Signal to Interference plus Noise Ratio (SINR) are related such that $\gamma_{k}=\frac{1-\varepsilon_{k}}{\varepsilon_{k}}$, where $\gamma_{k}$ denotes the $k$th user’s SINR. This implies that guaranteeing an AMMSE ensures a minimum average rate [@Gonzalez-Coma2013], and minimizing AMMSE is equivalent to maximizing a lowerbound of the average rate, i.e. $\mathrm{E}\{ \log_{2}(1+\gamma_{k}) \} = \mathrm{E}\{ - \log_{2}(\varepsilon_{k}) \} \geq - \log_{2}( \bar{\varepsilon}_{k})$. Unfortunately, finding an exact closed-form expressions for (\[Eq\_AMSE\_k\]) is not easy. This difficulty can be addressed by following the assumption in [@Shenouda2008], i.e. ignoring the better quality of CSIR and assuming that it is identical to CSIT. A closed-form expression for AMMSE could be obtained and applied to formulate $\mathcal{P}$ and $\mathcal{A}$. However, ignoring CSIR yields a degraded performance as we demonstrate in the next subsection. To account for perfect CSIR, we propose a close-form expression for an approximation of (\[Eq\_AMSE\_k\]).
The Ignorant Approach {#Sub_Section_Ignorant_AMSE}
---------------------
Using the available CSIT, both precoders and receivers are optimized at the BS, which informs each user of its corresponding receiver [@Shenouda2008]. In this case, the $k$th receiver is given as $\hat{g}_{k}(\hat{\mathbf{h}}_{k},\sigma^{2}_{e_{k}})= \mathbf{p}_{k}^{H}\hat{\mathbf{h}}_{k}\bar{T}^{-1}_{k}$ where $$\label{Eq_Tk_bar}
\bar{T}_{k} = \mathrm{E}\{ {T}_{k} \} = \sum_{i=1}^{K} \mathbf{p}_{i}^{H}(\hat{\mathbf{h}}_{k} \hat{\mathbf{h}}^{H}_{k}+\sigma^{2}_{e_{k}}\mathbf{I}) \mathbf{p}_{i} + \sigma_{n}^{2}.$$ $\hat{g}_{k}(\hat{\mathbf{h}}_{k},\sigma^{2}_{e_{k}})$, which is clearly a function of imperfect CSI, is obtained by minimizing the expectation of (\[Eq\_MSE\_k\]), i.e. $\mathrm{E}\{\varepsilon_{k}\}$. The resulting ignorant AMMSE can be written in closed-form as $$\label{Eq_AMSE_k_Ignorant}
\hat{\varepsilon}_{k} = 1-\bar{T}_{k}^{-1}(\mathbf{p}_{k}^{H}\hat{\mathbf{h}}_{k}\hat{\mathbf{h}}_{k}^{H}\mathbf{p}_{k}).$$ \[Remark\_Counter\][Remark]{}
\[Remark\_AMSE\_UB\] [For any given $\{\mathbf{p}_{k}\}_{k=1}^{K}$, we could write $$\begin{aligned}
\nonumber
\bar{\varepsilon}_{k}=& \ 1- \mathrm{E} \big\{T_{k}^{-1} \mathbf{p}_{k}^{H} (\hat{\mathbf{h}}_{k}\hat{\mathbf{h}}_{k}^{H} + \tilde{\mathbf{h}}_{k}\tilde{\mathbf{h}}_{k}^{H} + \hat{\mathbf{h}}_{k}\tilde{\mathbf{h}}_{k}^{H} + \tilde{\mathbf{h}}_{k}\hat{\mathbf{h}}_{k}^{H} )\mathbf{p}_{k} \big\} \\
\label{Eq_AMSE_UB_1}
\leq& \ 1- \mathrm{E} \big\{T_{k}^{-1} \big\} \mathbf{p}_{k}^{H} \hat{\mathbf{h}}_{k} \hat{\mathbf{h}}_{k}^{H} \mathbf{p}_{k}\\
\label{Eq_AMSE_UB_2}
\leq& \ 1- \bar{T}_{k}^{-1} \mathbf{p}_{k}^{H} \hat{\mathbf{h}}_{k} \hat{\mathbf{h}}_{k}^{H} \mathbf{p}_{k} \\
\nonumber
= & \ \hat{\varepsilon}_{k}\end{aligned}$$ where (\[Eq\_AMSE\_UB\_1\]) follows from the non-negativity of the terms $T_{k}^{-1}$, $\mathbf{p}_{k}^{H}\tilde{\mathbf{h}}_{k}\tilde{\mathbf{h}}_{k}^{H}\mathbf{p}_{k}$ and $\mathbf{p}_{k}^{H}(\hat{\mathbf{h}}_{k}\tilde{\mathbf{h}}_{k}^{H}+\tilde{\mathbf{h}}_{k}\hat{\mathbf{h}}_{k}^{H})\mathbf{p}_{k}$, and (\[Eq\_AMSE\_UB\_2\]) follows from Jensen’s inequality. Equality in (\[Eq\_AMSE\_UB\_2\]) holds for $\sigma^{2}_{e_{k}}=0$.]{.nodecor}
\[Remark\_Counter\][Remark]{}
\[Remark\_AMMSE\_Scaling\] [For a set of precoding vectors $\{c\mathbf{p}_{k}\}_{k=1}^{K}$ where $c$ is a positive power-scaling factor, $\{\bar{\varepsilon}_{k}\}_{k}^{K}$ and $\{\hat{\varepsilon}_{k}\}_{k}^{K}$ are monotonically non-increasing in $c$. This is evident from plugging $\{c\mathbf{p}_{k}\}_{k=1}^{K}$ into (\[Eq\_AMSE\_k\]) and (\[Eq\_AMSE\_k\_Ignorant\]).]{.nodecor}
\[Corollary\_Counter\][Corollary]{}
\[Corollary\_Ignorant\_Aware\] [The ignorant approach yields a degraded performance (higher AMMSEs or power) compared to the an aware approach, that takes into account the perfect CSIR. For a power-constrained problem (e.g. $\mathcal{A}$), (\[Eq\_AMSE\_UB\_2\]) holds even if $\{\mathbf{p}_{k}\}_{k=1}^{K}$ were ignorant precoders, i.e. optimally designed w.r.t $\{\hat{\varepsilon}_{k}\}_{k=1}^{K}$. For an AMMSE constrained problem (e.g. $\mathcal{P}$), Remark \[Remark\_AMMSE\_Scaling\] implies that aware-optimization could achieve the same AMMSEs for less power compared to ignorant-optimization by using a down-scaled version of the optimum ignorant precoders. Moreover, even if we assume that perfect CSIR is utilized by users and ignorant AMMSEs are only used as optimization metric at the BS, this corresponds to using upperbounds of the AMMSEs which can be very loose under certain channel conditions [@Bashar2014].]{.nodecor}
AMMSE Taylor Approximation
--------------------------
The expectation of a ratio of two random variables could be approximated using the Taylor expansion [@Rice2009]: \[Lemma\_Counter\][Lemma]{}
\[Lemma\_Taylor\_Expansion\] For two random variables $x$ and $y$ with expectations $\mathrm{E}\{x\}=\bar{x}$ and $\mathrm{E}\{y\}=\bar{y}$, and $y\neq0$, we could write $$\label{Eq_Taylor}
\mathrm{E}\bigg\{\frac{x}{y}\bigg\} \approx \frac{\bar{x}}{\bar{y}} + \sum_{i = 1}^{2(N-1)} (-1)^{i} \frac{\bar{x}\mu_{0,i}+\mu_{1,i}}{\bar{y}^{i+1}}$$ where $N$ is the order of the Taylor expansion and $$\label{Eq_Taylor_Moment}
\mu_{m,n} = \mathrm{E}\{ (x-\bar{x})^{m} (y-\bar{y})^{n} \}.$$
The accuracy of the approximation in (\[Eq\_Taylor\]) increases as $N$ grows larger where $N\rightarrow\infty$ yields infinite accuracy. An approximation of (\[Eq\_AMSE\_k\]) is denoted by $\bar{\varepsilon}_{k}^{(N)}$, where $N$ is the order of the Taylor expansion of $\mathrm{E}\{T_{k}^{-1}R_{k}\}$. To maintain tractability, we consider the second-order Taylor expansion which could be written as [@VanKempen2000] $$\begin{aligned}
\nonumber
\bar{\varepsilon}_{k}^{(2)} =& 1-\bigg(\frac{\bar{R}_{k}}{\bar{T}_{k}} - \frac{\text{cov}\{R_{k},T_{k}\}}{\bar{T}_{k}^{2}} + \frac{\bar{R}_{k} \text{var}\{T_{k}\}}{\bar{T}_{k}^{3}} \bigg) \\
\label{Eq_AMSE_k_Approx_2}
=& 1 \! - \! \frac{\bar{R}_{k}}{\bar{T}_{k}} \! + \! \frac{\mathrm{E} \! \{ \! R_{k} \! T_{k} \! \} \! - \! \bar{R}_{k} \! \bar{T}_{k}}{\bar{T}_{k}^{2}} \! - \! \frac{\bar{R}_{k} \! (\mathrm{E} \! \{ \! T_{k}^{2} \! \} \! - \! \bar{T}_{k}^{2})}{\bar{T}_{k}^{3}}\end{aligned}$$ where $$\label{Eq_Rk_bar}
\bar{R}_{k} = \mathrm{E}\{ {R}_{k} \} = \mathbf{p}_{k}^{H}(\hat{\mathbf{h}}_{k} \hat{\mathbf{h}}^{H}_{k}+\sigma^{2}_{e_{k}}\mathbf{I}) \mathbf{p}_{k}.$$ Each of the terms $\mathrm{E}\{R_{k}T_{k}\}$ and $\mathrm{E}\{T_{k}^{2}\}$ in (\[Eq\_AMSE\_k\_Approx\_2\]) is an expectation of a product of two quadratic forms in a random vector $\mathbf{h}_{k} \thicksim \mathcal{CN} \big( \hat{\mathbf{h}}_{k} , \sigma^{2}_{e_{k}}\mathbf{I} \big)$. Closed-form expressions can be obtained using the following lemma: \[Lemma\_Counter\][Lemma]{}
\[Lemma\_E\_Complex\_Quadratic\_non\_zero\_mean\] For a complex gaussian vector $\mathbf{x}\sim\mathcal{CN}(\hat{\mathbf{x}},\mathbf{C})$ given as $\mathbf{x}=\hat{\mathbf{x}}+\tilde{\mathbf{x}}$ where $\hat{\mathbf{x}}$ is the mean and $\tilde{\mathbf{x}}$ has ZMCSCG entries, and two quadratic forms defined as $Q_{1}=\mathbf{x}^{H}\mathbf{A}\mathbf{x}$ and $Q_{2}=\mathbf{x}^{H}\mathbf{B}\mathbf{x}$ where $\mathbf{A}, \mathbf{B}\succeq 0$, we have $$\begin{gathered}
\label{Eq_E_Q1Q2}
\mathrm{E}\{Q_{1}Q_{2}\} = \hat{\mathbf{x}}^{H}\mathbf{ACB}\hat{\mathbf{x}}+\hat{\mathbf{x}}^{H}\mathbf{BCA}\hat{\mathbf{x}}+\textnormal{tr}(\mathbf{ACBC})\\
+
(\textnormal{tr}(\mathbf{AC})+\hat{\mathbf{x}}^{H}\mathbf{A}\hat{\mathbf{x}})(\textnormal{tr}(\mathbf{BC})+\hat{\mathbf{x}}^{H}\mathbf{B}\hat{\mathbf{x}}).\end{gathered}$$
A sketch of the proof is provided in the Appendix. Using Lemma \[Lemma\_E\_Complex\_Quadratic\_non\_zero\_mean\], we could write $$\begin{aligned}
\nonumber
\mathrm{E}\{R_{k}T_{k}\} =& \ \sigma_{e_{k}}^{2}\hat{\mathbf{h}}_{k}^{H}\mathbf{Q}_{k}\mathbf{Q}\hat{\mathbf{h}}_{k}+
\sigma_{e_{k}}^{2}\hat{\mathbf{h}}_{k}^{H}\mathbf{Q}\mathbf{Q}_{k}\hat{\mathbf{h}}_{k}\\
\label{Eq_E_Rk_Tk}
& + (\sigma_{e_{k}}^{2})^{2}\mathrm{tr}(\mathbf{Q}_{k}\mathbf{Q})+\bar{R}_{k}\bar{T}_{k} \\
\label{Eq_E_Tk2}
\mathrm{E}\{T_{k}^{2}\} =& \ 2\sigma_{e_{k}}^{2}\hat{\mathbf{h}}_{k}^{H}\mathbf{Q}^{2}\hat{\mathbf{h}}_{k}+
(\sigma_{e_{k}}^{2})^{2}\mathrm{tr}(\mathbf{Q}^{2})+\bar{T}_{k}^{2}\end{aligned}$$ where $\mathbf{Q}_{k}=\mathbf{p}_{k}\mathbf{p}_{k}^{H}$ and $\mathbf{Q}=\sum_{i=1}^{K}\mathbf{p}_{i}\mathbf{p}_{i}^{H}$. Before plugging (\[Eq\_E\_Rk\_Tk\]) and (\[Eq\_E\_Tk2\]) back into (\[Eq\_AMSE\_k\_Approx\_2\]), we define $a_{k}$ and $b_{k}$ as follows $$\begin{aligned}
\label{Eq_ak}
a_{k} =& \ \hat{\mathbf{h}}_{k}^{H}(\mathbf{Q}_{k}\mathbf{Q}+\mathbf{Q}\mathbf{Q}_{k})\hat{\mathbf{h}}_{k}
+ \sigma_{e_{k}}^{2}\mathrm{tr}(\mathbf{Q}_{k}\mathbf{Q}), \ \forall k \\
\label{Eq_bk}
b_{k} =& \ 2\hat{\mathbf{h}}_{k}^{H}\mathbf{Q}^{2}\hat{\mathbf{h}}_{k}+
\sigma_{e_{k}}^{2}\mathrm{tr}(\mathbf{Q}^{2}) , \ \forall k\end{aligned}$$ from which (\[Eq\_AMSE\_k\_Approx\_2\]) could be written as $$\begin{aligned}
\label{Eq_AMSE_k_Approx_3}
\bar{\varepsilon}_{k}^{(2)} =&\ 1-\frac{\bar{R}_{k}}{\bar{T}_{k}}
+a_{k}\frac{\sigma_{e_{k}}^{2}}{\bar{T}_{k}^{2}} -
b_{k}\frac{\sigma_{e_{k}}^{2}\bar{R}_{k}}{\bar{T}_{k}^{3}} \\
\label{Eq_AMSE_k_Approx_4}
=&\ 1-\alpha_{k}\frac{\bar{R}_{k}}{\bar{T}_{k}}\end{aligned}$$ where $\alpha_{k}$ is given as $$\label{Eq_Alpha_k}
\alpha_{k} = \ 1 - \frac{\sigma_{e_{k}}^{2}}{\bar{T}_{k}^{2} \bar{R}_{k}}(a_{k}\bar{T}_{k} - b_{k}\bar{R}_{k}).$$ It is clear that (\[Eq\_AMSE\_k\_Approx\_4\]) reduces to $\bar{\varepsilon}_{k}^{(1)}$ if $\alpha_{k}$ was replaced by $1$, while $\hat{\varepsilon}_{k}$ will be obtained if it was replaced by $\big(|\hat{\mathbf{h}}_{k}^{H}\mathbf{p}_{k}|^{2}/\bar{R}_{k}\big)$.
AMMSE Based Robust Beamforming {#Section_AMSE Robust Beamforming}
==============================
In this section, the AMMSE approximation in (\[Eq\_AMSE\_k\_Approx\_4\]) is used to formulate the robust design problems $\mathcal{P}$ and $\mathcal{A}$. The approach followed to solve those problems is based on the one proposed in [@Wiesel2006]. Particularly, the solution of the former problem is based on conic optimization, where the later problem is solved by exploiting the relationship between the two problems, and the monotonic nature of their objective functions.
Power Minimization {#Subsection_Power_Min}
------------------
The power minimization problem is denoted by $\mathcal{P}(\bar{\varepsilon}_{0})$ where $0<\bar{\varepsilon}_{0}<1$ is the given worst AMMSE constraint. $\mathcal{P}(\bar{\varepsilon}_{0})$ could be formulated as $$\begin{aligned}
\nonumber
\mathcal{P}(\bar{\varepsilon}_{0}): \ &\underset{\mathbf{p}_{1},\ldots,\mathbf{p}_{K}}{\min} \ \sum_{i=1}^{K}\|\mathbf{p}_{i}\|^{2} \\
\label{Eq_Opt_P}
& \ \ \ \text{s.t.} \ \ \ 1-\alpha_{k}\frac{\bar{R}_{k}}{\bar{T}_{k}} \leq \bar{\varepsilon}_{0},\ \ \forall k\in\{1,\ldots,K\}.\end{aligned}$$ Introducing the non-negative real-valued slack variable $P_{0}$, and transforming the objective and constraints into the Linear Matrix Inequality (LMI) form, (\[Eq\_Opt\_P\]) could be rewritten as $$\begin{aligned}
\nonumber
\mathcal{P}(\bar{\varepsilon}_{0}): \ &\underset{P_{0},\mathbf{Q}_{1},\ldots,\mathbf{Q}_{K}}{\min} \ P_{0} \\
\nonumber
& \ \ \ \ \text{s.t.} \ \ \ \sum_{i=1}^{K}\mathrm{tr}(\mathbf{Q}_{i}) \leqslant P_{0} ,
\\
\nonumber
\sum_{i=1}^{K}\!
\mathrm{tr} \!
&
\Big(\! \mathbf{Q}_{i} \! \big( \hat{\mathbf{h}}_{k} \! \hat{\mathbf{h}}_{k}^{H} \! + \! \sigma_{e_{k}}^{2} \! \mathbf{I} \! \big) \! \Big)\!
+ \! \sigma_{n}^{2} \! \leq \!
\frac{\alpha_{k}}{1\! - \! \bar{\varepsilon}_{0}}
\mathrm{tr} \! \Big( \! \mathbf{Q}_{k} \! \big( \hat{\mathbf{h}}_{k} \! \hat{\mathbf{h}}_{k}^{H} \! + \! \sigma_{e_{k}}^{2} \! \mathbf{I} \! \big) \! \Big) ,
\\
\nonumber
& \ \ \ \ \ \ \ \ \ \mathbf{Q}_{k}\succeq 0 ,
\\
\label{Eq_Opt_P_2}
& \ \ \ \ \ \ \ \ \ \mathrm{rank}(\mathbf{Q}_{k})= 1 , \ \ \forall k\in\{1,\ldots,K\}.\end{aligned}$$ Assuming that $\mathcal{P}(\bar{\varepsilon}_{0})$ is feasible, finding the optimal $P_{0}$ in (\[Eq\_Opt\_P\_2\]) could be very challenging. This is mainly because the second set of constraints are non-linear due to the presence of $\alpha_{k}$, and the rank constraints are non-convex. To make the problem less complicated, the non-linearity in the second set of constraints could be eliminated by replacing $\{\alpha_{k}\}_{k=1}^{K}$ with fixed, real and non-negative values denoted by $\{\bar{\alpha}_{k}\}_{k=1}^{K}$; e.g. $\bar{\alpha}_{k}$ could be set to $1$ where the AMMSE constraints will be defined in terms of the first-order approximation of $\{\bar{\varepsilon}_{k}\}_{k=1}^{K}$, i.e. $\{\bar{\varepsilon}_{k}^{(1)}\}_{k=1}^{K}$. The linearized problem is denoted by $\mathcal{P}_{l}(\bar{\varepsilon}_{0},\{\bar{\alpha}_{k}\}_{k=1}^{K})$ and takes $1+K$ input arguments. $\mathcal{P}_{l}$ could be made tractable by relaxing the rank constraints. The new linearized and relaxed problem, referred to as $\mathcal{P}_{lr}(\bar{\varepsilon}_{0},\{\bar{\alpha}_{k}\}_{k=1}^{K})$, is convex as it is composed of a linear objective function and a combination of linear and semidefinite constraints. In particular, $\mathcal{P}_{lr}$ is a Semidefinite Program (SDP) which could be solved efficiently using Interior-Point methods [@Boyd2004].
Due to the rank relaxation, the $K$ matrices $\{\mathbf{Q}_{k}\}_{k=1}^{K}$ obtained by solving $\mathcal{P}_{lr}$ will not be rank-1 in general. If they are all rank-1, then the optimum solutions for $\mathcal{P}_{lr}$ and $\mathcal{P}_{l}$ coincide and $\{\mathbf{p}_{k}\}_{k=1}^{K}$ could be obtained directly through eigen decomposition. Otherwise, the power obtained by solving $\mathcal{P}_{lr}$ is a lowerbound for the the optimum objective value of $\mathcal{P}_{l}$. This is due to the fact that relaxation extends the domain of feasible $\{\mathbf{Q}_{k}\}_{k=1}^{K}$ bearing the possibility of a solution with a lower objective compared to the non-relaxed problem.
For each $k$, if $\mathrm{rank}(\mathbf{Q}_{k})>1$, then $\mathbf{p}_{k}$ could be chosen as the principal eigenvector of $\mathbf{Q}_{k}$ or generated using randomization [@Luo2010]. However, it is likely that the resulting beamforming vectors will fail to satisfy the AMMSE constraints. In this case, further optimization is required for power reallocation. In the simulations carried out for this paper, it has been observed that solving $\mathcal{P}_{lr}$ always gives rank-1 solutions. Further investigations regarding this observation is left for future work.
**Initialize**: $n\gets 0$, $P_{0}^{(n)}\gets 0 $, $\bar{\alpha}_k^{(n)}\gets1 \ \forall k $ $n\gets n+1$ $P_{0}^{(n)} \gets \mathcal{P}_{lr}(\bar{\varepsilon}_{0},\{\bar{\alpha}_{k}^{(n-1)}\}_{k=1}^{K})$ $\{\mathbf{Q}_{k}^{(n)}\}_{k=1}^{K} \gets \arg \mathcal{P}_{lr}(\bar{\varepsilon}_{0},\{\bar{\alpha}_{k}^{(n-1)}\}_{k=1}^{K})$ $\mathbf{Q}^{(n)} \gets \sum_{i=1}^{i=K}\mathbf{Q}_{i}^{(n)}$ $\{\bar{\alpha}_{k}^{(n)}\}_{k=1}^{K}$
Going back to the original power optimization problem $\mathcal{P}$, a solution is proposed which involves solving $\mathcal{P}_{lr}$ recursively over multiple iterations where the values of $\{\bar{\alpha}_{k}\}_{k=1}^{K}$ are updated in each iteration. Particularly, in the $n$th iteration, $\mathcal{P}_{lr}(\bar{\varepsilon}_{0},\{\bar{\alpha}_{k}^{(n-1)}\}_{k=1}^{K})$ is optimally solved where $\{\bar{\alpha}_{k}^{(n-1)}\}_{k=1}^{K}$ are obtained using the solution of $\mathcal{P}_{lr}$ in the $(n-1)$th iteration. This is carried out until a desired accuracy (specified by $\epsilon_{P}$) is achieved or a maximum number of iterations $n_{\max}$ is reached. The pseudo-code for this method is shown in Algorithm \[Algthm\_Power\_Opt\]. Although Algorithm \[Algthm\_Power\_Opt\] is not guaranteed to reach a global optimum, simulations show that quick convergence with good performance is achieved.
It is important to highlight that the presence of CSIT uncertainty may impose a feasibility bound on $\bar{\varepsilon}_{0}$. For perfect CSI i.e. $\sigma_{e_{k}}^{2}=0 \ \forall k$, any $\bar{\varepsilon}_{0} \geq 0$ is feasible. This is directly concluded from ${\gamma}_{0} \leq \infty$ in [@Wiesel2006] where ${\gamma}_{0}$ denotes the target SINR. Feasibility for the case where CSIT is imperfect has been addressed in [@Gonzalez-Coma2013a]. Although no closed-form expression has been derived (a QoS region could be obtained through Monte-Carlo integration), it has been observed that $\bar{\varepsilon}_{0}$ is lowerbounded above $0$. This could be confirmed by plugging the scaled precoding vectors $\{c\mathbf{p}_{k}\}_{k=1}^{K}$ into $\bar{\varepsilon}_{k}^{(2)}$ and driving the transmit power up to infinity, i.e. $c\rightarrow\infty$. Regardless of the structure of $\{\mathbf{p}_{k}\}_{k=1}^{K}$, for non-zero fixed ${\{\sigma_{e_{k}}^{2}}\}_{k=1}^{K}$ that do not scale down with increased power, residual interference terms will bound $\bar{\varepsilon}_{k}^{(2)}$ above $0$.
Maximum AMMSE Minimization
--------------------------
The minimization of the maximum AMMSE problem is referred to as $\mathcal{A}(P_{t})$ where $P_{t}$ denotes the total transmission power constraint. This could be written as $$\begin{aligned}
\nonumber
\mathcal{A}(P_{t}): \ &\underset{\mathbf{p}_{1},\ldots,\mathbf{p}_{K}}{\min} \ \underset{k}{\max} \ \ 1-\alpha_{k}\frac{\bar{R}_{k}}{\bar{T}_{k}} \\
\label{Eq_Opt_A}
& \ \ \ \text{s.t.} \ \ \ \ \ \sum_{i=1}^{K}\|\mathbf{p}_{i}\|^{2} \leq P_{t}.\end{aligned}$$ By introducing the non-negative real-valued slack variable $t_{0}$ and adding the constraints $ 1-\alpha_{k}{\bar{T}_{k}}^{-1}{\bar{R}_{k}} \leq t_{0} \ \forall k$, the objective in (\[Eq\_Opt\_A\]) could be written as: $\min \ t_{0}$. Following the formulation of (\[Eq\_Opt\_P\_2\]), (\[Eq\_Opt\_A\]) could be rewritten as $$\begin{aligned}
\nonumber
\mathcal{A}(P_{t}): \ &\underset{t_{0},\mathbf{Q}_{1},\ldots,\mathbf{Q}_{K}}{\min} \ t_{0} \\
\nonumber
& \ \ \ \ \text{s.t.} \ \ \ \sum_{i=1}^{K}\mathrm{tr}(\mathbf{Q}_{i}) \leqslant P_{t} ,
\\
\nonumber
\sum_{i=1}^{K}\!
\mathrm{tr} \!
&
\Big(\! \mathbf{Q}_{i} \! \big( \hat{\mathbf{h}}_{k} \! \hat{\mathbf{h}}_{k}^{H} \! + \! \sigma_{e_{k}}^{2} \! \mathbf{I} \! \big) \! \Big)\!
+ \! \sigma_{n}^{2} \! \leq \!
\frac{\alpha_{k}}{1\! - \! t_{0}}
\mathrm{tr} \! \Big( \! \mathbf{Q}_{k} \! \big( \hat{\mathbf{h}}_{k} \! \hat{\mathbf{h}}_{k}^{H} \! + \! \sigma_{e_{k}}^{2} \! \mathbf{I} \! \big) \! \Big) ,
\\
\nonumber
& \ \ \ \ \ \ \ \ \ \mathbf{Q}_{k}\succeq 0 ,
\\
\label{Eq_Opt_A_2}
& \ \ \ \ \ \ \ \ \ \mathrm{rank}(\mathbf{Q}_{k})= 1 , \ \ \forall k\in\{1,\ldots,K\}.\end{aligned}$$ At a first glance, (\[Eq\_Opt\_A\_2\]) looks similar to (\[Eq\_Opt\_P\_2\]). However, after careful consideration one could see that $t_{0}$ in the second set of constraints in (\[Eq\_Opt\_A\_2\]) is an optimization variable, contrary to $\bar{\varepsilon}_{0}$ in (\[Eq\_Opt\_P\_2\]) which is a given input parameter. Same applies to $P_{t}$ and $P_{0}$ where the former is a power constraint input parameter in (\[Eq\_Opt\_A\_2\]) while the later is the objective variable in (\[Eq\_Opt\_P\_2\]). $\mathcal{A}_{lr}(P_{t},\{\bar{\alpha}_{k}\}_{k=1}^{K})$ denotes a semi-linearized and relaxed version of $\mathcal{A}$ obtained by fixing $\{{\alpha}_{k}\}_{k=1}^{K}$ to $\{\bar{\alpha}_{k}\}_{k=1}^{K}$ and relaxing the rank constraints. However, non-linearity is still present in the constraints of $\mathcal{A}_{lr}$ due to the presence of the objective variable $t_{0}$. This could be tackled by exploiting the relationship between $\mathcal{P}_{lr}$ and $\mathcal{A}_{lr}$, and their monotonic nature. \[Proposition\_Counter\][Proposition]{}
\[Proposition\_P\_A\_Inverse\_Problems\_and\_Monotonicity\] $\mathcal{P}_{lr}$ and $\mathcal{A}_{lr}$ are inverse problems such that: $$\label{Eq_A_P_Inverse_Problems}
\mathcal{A}_{lr}\big(\mathcal{P}(\bar{\varepsilon}_{0},\{\bar{\alpha}_{k}\}_{k=1}^{K}), \{\bar{\alpha}_{k}\}_{k=1}^{K}\big) = \bar{\varepsilon}_{0}$$ $$\label{Eq_P_A_Inverse_Problems}
\mathcal{P}_{lr}\big(\mathcal{A}(P_{t},\{\bar{\alpha}_{k}\}_{k=1}^{K}), \{\bar{\alpha}_{k}\}_{k=1}^{K}\big) = P_{t}.$$ In addition, the optimum objective of $\mathcal{P}_{lr}$ and $\mathcal{A}_{lr}$ are monotonic such that: $$\label{Eq_P_Monotonicity}
\mathcal{P}_{lr}(\bar{\varepsilon}_{0},\{\bar{\alpha}_{k}\}_{k=1}^{K}) > \mathcal{P}_{lr}(\bar{\varepsilon}_{0}^{o},\{\bar{\alpha}_{k}\}_{k=1}^{K}) \Rightarrow \bar{\varepsilon}_{0} < \bar{\varepsilon}_{0}^{o}$$ $$\label{Eq_A_Monotonicity}
\mathcal{A}_{lr}(P_{t},\{\bar{\alpha}_{k}\}_{k=1}^{K}) \leq \mathcal{A}_{lr}(P_{t}^{o},\{\bar{\alpha}_{k}\}_{k=1}^{K}) \Rightarrow P_{t} > P_{t}^{o}$$ where $\bar{\varepsilon}_{0}$ and $\bar{\varepsilon}_{0}^{o}$ are assumed to be feasible.
This could be proved by contradiction [@Wiesel2006]. Due to the lack of space, the reader is referred to the proof of [@Wiesel2006 Theorem 3]. Contrary to [@Wiesel2006 (67)], the left-hand side inequality in (\[Eq\_A\_Monotonicity\]) is not strict. This is due to the possibility that $1-\bar{\alpha}_{k}{\bar{T}_{k}}^{-1}{\bar{R}_{k}}$ will hit a floor at high SNRs, as shown in Section \[Subsection\_Power\_Min\], causing looseness to the power constraint in $\mathcal{A}_{lr}$.
\[Corollary\_Counter\][Corollary]{}
\[Corollary\_A\_Solution\_Bisection\] [Propositions \[Proposition\_P\_A\_Inverse\_Problems\_and\_Monotonicity\] implies that $\mathcal{A}_{lr}(P_{0},\{\bar{\alpha}_{k}\}_{k=1}^{K})$ could be optimally solved by carrying out a one-dimensional bisections search over $\bar{\varepsilon}_{0}$ until the minimum $\bar{\varepsilon}_{0}$ that satisfies $\mathcal{P}_{lr}(\bar{\varepsilon}_{0},\{\bar{\alpha}_{k}\}_{k=1}^{K}) \leq P_{0}$ is found.]{.nodecor}
The pseudo-code for the bisections search method is given in Algorithm \[Algthm\_AMSE\_Opt\_Bisection\] where the value of $\epsilon_{0}$ determines the accuracy of the solution.
**Initialize**: $\bar{\varepsilon}_{\min} \gets 0$, $\bar{\varepsilon}_{\max}\gets 1 $ $\bar{\varepsilon}_{0}\gets (\bar{\varepsilon}_{\min}+\bar{\varepsilon}_{\max})/2$ $\bar{P}_{0} \gets \mathcal{P}_{lr}(\bar{\varepsilon}_{0},\{\bar{\alpha}_{k}\}_{k=1}^{K})$ \[Algthm\_AMSE\_Opt\_Bisection\_Step\_P\]
$\bar{\varepsilon}_{\min} \gets \bar{\varepsilon}_{0}$ $\bar{\varepsilon}_{\max} \gets \bar{\varepsilon}_{0}$ $t_{0} \gets \bar{\varepsilon}_{\max}$ $\{\mathbf{Q}_{k}\}_{k=1}^{K} \gets \arg \mathcal{P}_{lr}(\bar{\varepsilon}_{\max},\{\bar{\alpha}_{k}\}_{k=1}^{K})$
The stopping criteria in the bisection search is expressed in terms of $\bar{\varepsilon}_{\min}$ and $\bar{\varepsilon}_{\max}$ rather than $P_{0}$ (as in [@Wiesel2006]) due to the possibility of power constraint looseness. Furthermore, $t_{0}$ is set to $\bar{\varepsilon}_{\max}$ and the solution is given by $\arg \mathcal{P}_{lr}(\bar{\varepsilon}_{\max},\{\bar{\alpha}_{k}\}_{k=1}^{K})$, as $\bar{\varepsilon}_{\min}$ could be infeasible. Following the same recursive approach used to find a solution for $\mathcal{P}$, a solution for problem $\mathcal{A}$ could be obtained using Algorithm \[Algthm\_AMSE\_Opt\]. As for Algorithm \[Algthm\_Power\_Opt\], quick convergence and good performance could be achieved by Algorithm \[Algthm\_AMSE\_Opt\] despite unguaranteed global optimality.
**Initialize**: $n\gets 0$, $t_{0}^{(n)}\gets 0 $, $\bar{\alpha}_k^{(n)}\gets1 \ \forall k $ $n\gets n+1$ $t_{0}^{(n)} \gets \mathcal{A}_{lr}({P}_{0},\{\bar{\alpha}_{k}^{(n-1)}\}_{k=1}^{K})$ $\{\mathbf{Q}_{k}^{(n)}\}_{k=1}^{K} \gets \arg \mathcal{A}_{lr}({P}_{0},\{\bar{\alpha}_{k}^{(n-1)}\}_{k=1}^{K})$ $\mathbf{Q}^{(n)} \gets \sum_{i=1}^{i=K}\mathbf{Q}_{i}^{(n)}$ $\{\bar{\alpha}_{k}^{(n)}\}_{k=1}^{K}$
Numerical Results {#Section_Numerical Results}
=================
We consider a BS equipped with $N_{t} = 4$ antennas serving $K=4$ single-antenna users. The $k$th user’s CSIT consists of the first and second order statistics of $\mathbf{h}_{k}$, i.e. $\hat{\mathbf{h}}_{k} = \sigma_{c_{k}} [1,e^{j\varphi_{k}},\ldots ,e^{j(N_{t}-1)\varphi_{k}}]^{T} $ with $\varphi_{k} \sim \mathcal{U}(0,2\pi)$, and $\mathbf{R}_{e_{k}}=\sigma^{2}_{e_{k}}\mathbf{I}$. The per-antenna average path gain is defined as $\sigma^{2}_{k} = \sigma^{2}_{c_{k}} + \sigma^{2}_{e_{k}}$. This is similar to the model in [@Gonzalez-Coma2013] with an added feature of controlling individual CSIT qualities, e.g. $\sigma^{2}_{e_{k}}=0$ and $\sigma^{2}_{c_{k}}=0$ correspond to perfect and completely random CSITs for user $k$, respectively. The average noise variance across users is given as $\sigma^{2}_{\text{av}}=\frac{1}{K}\Sigma_{k=1}^{K} \frac{\sigma^{2}_{n}}{ \sigma^{2}_{k}}$ from which the SNR is defined as $\frac{P_{t}}{K \sigma^{2}_{\text{av}}}$ [@Bogale2013]. Throughout the simulations, all users are assumed to have unity path gains except for user-4 whose path gain is $3$ dB lower, i.e. $\sigma^{2}_{{1}} = \sigma^{2}_{{2}} = \sigma^{2}_{{3}} = 1$ and $\sigma^{2}_{{4}} = 0.5$. This is reflected in the CSIT quality (by influencing the UL pilot SNR or the feedback link capacity) yielding $\sigma^{2}_{e_{1}} = \sigma^{2}_{e_{2}} = \sigma^{2}_{e_{3}} = 0.05$ and $\sigma^{2}_{e_{4}}=0.1$. Therefore, user-4 is deemed the *least fortunate*. $\sigma^{2}_{n}$ is fixed to $1$ where the SNR would only vary with $P_{t}$. For all simulation, $\{\hat{\mathbf{h}}_{k}\}_{k}^{K}$ are kept fixed where averaging is carried out over several independent realizations of $\{\tilde{\mathbf{h}}_{k} \}_{k=1}^{K}$. This corresponds to a scenario where the CSIT (and the optimized transmitter) remain unchanged over multiple (time or frequency) channel uses in which the channel response changes. Similar performances are observed for different realizations of $\{\hat{\mathbf{h}}_{k}\}_{k}^{K}$.
Power Minimization {#power-minimization}
------------------
The results obtained from solving $\mathcal{P}(0.25)$ and $\mathcal{P}(0.4)$ using Algorithm \[Algthm\_Power\_Opt\] (SDP-Algorithm) are compared to those obtained using the AO-Algorithm [@Gonzalez-Coma2013 Algorithm 1]. The AO-Algorithm uses $4000$ Monte-Carlo realization and $\epsilon_{P}=10^{-4}$ [@Gonzalez-Coma2013] is used for both algorithms. The AMMSE of the *least fortunate* user and the total required SNR are plotted against the number of iterations in Fig. \[Fig\_P\_AMSE\] and Fig. \[Fig\_Power\], respectively. Since the SDP-Algorithm does not involve Monte-Carlo integration, its AMMSE is obtained by averaging over the same realizations used for the AO-Algorithm. Fig. \[Fig\_P\_AMSE\] shows that both algorithms meet the AMMSE targets with high accuracy (the AMMSE approximation yields an error less than $0.5 \%$ for the SDP-Algorithm). Furthermore, it is evident from Fig. \[Fig\_Power\] that both algorithms yield similar minimized powers. However, the SDP-Algorithm takes significantly less iterations to converge compared to the AO-Algorithm ($4$ vs. $36$ for $\mathcal{P}(0.4)$, and $6$ vs. $91$ for $\mathcal{P}(0.25)$). Contrary to the AO-Algorithm (which is randomly initialized), the first iteration of the SDP-Algorithm solves the problem using the first-order Taylor approximation of AMMSE yielding a good starting point for later iterations and consequently, reducing the total number of iterations.
For each iteration, the SDP-Algorithm has to solve a SDP with a linear objective function, $K$ positive-semidefinite matrix variables of size $N_{t}\times N_{t}$ and $K$ linear inequality constraints. Such problems can be efficiently solved using SDP solvers that apply Interior-Point methods (e.g. [@Sturm1999]) at a worst-case complexity cost that scales with $\mathcal{O}(K^{3.5}N_{t}^{6.5})$ [@Karipidis2008; @Ye1997]. However, the actual runtime complexity scales far slower with $K$ and $N_{t}$. On the other hand, the complexity of the AO-Algorithm mainly comes from the numerical Monte-Carlo integration. In each iteration, multiple operations that scale with $\mathcal{O}( N_{t}^{2}KM)$, $\mathcal{O}( N_{t}K^{2}M)$ and $\mathcal{O}( N_{t}KM)$ are carried out, where $M$ is the number of Monte-Carlo realizations. For a practical system (e.g. $N_{t}\leq 8$), the per-iteration complexity of both algorithms is comparable and the actual runtime complexity for the SDP-Algorithm is significantly smaller.
\[Fig\_Power\_AMSE\]
\[Fig\_AMSE\_Rate\]
Maximum AMMSE Minimizations
---------------------------
The robust CSIR-aware optimization proposed in Algorithm \[Algthm\_AMSE\_Opt\] is compared to the ignorant optimization where the BS uses the available CSIT to jointly design the precoding vectors and the receivers which are forwarded to their corresponding users. Results for the ignorant scheme are obtained using a slightly modified version of Algorithm \[Algthm\_AMSE\_Opt\_Bisection\]. Particularly, in step \[Algthm\_AMSE\_Opt\_Bisection\_Step\_P\], the AMMSE constraints of $\mathcal{P}_{lr}$ are changed to $\hat{\varepsilon}_{k} \leq \bar{\varepsilon}_{0} \ \forall k$, and other changes are made accordingly. The two schemes are compared in terms of the AMMSE and the Average Rate of the *least fortunate* user in Fig. \[Fig\_AMSE\] and Fig. \[Fig\_Rate\], respectively. The gain from the utilization of perfect CSIR grows with SNR as the channel estimation error becomes significant.
Conclusion {#Section_Conclusion}
==========
In this paper, we proposed an AMMSE second order Taylor approximation for linearly-precoded MU-MISO systems with imperfect CSIT and perfect CSIR. This approximation was used to formulate the two fairness-based robust design problems: a power minimization problem ($\mathcal{P}$) and an AMMSE minimization problem ($\mathcal{A}$). Problem $\mathcal{P}$ was solved using a fast-converging algorithm based on recursive convex optimization. On the other hand, problem $\mathcal{A}$ was solved by combining the recursive optimization approach with methods from [@Wiesel2006]. This CSIR-aware robust design was shown to have better performance compared to the ignorant design.
$\mathrm{E}\{Q_{1}Q_{2}\}$ could be written as the sum of 16 terms consisting of linear, quadratic, cubic and quartic forms of $\tilde{\mathbf{x}}$. By extending some of the identities in [@Petersen2008 Ch. 8.2] (given for real Gaussian vectors) to ZMCSCG vectors, each term could be found individually and the expression in (\[Eq\_E\_Q1Q2\]) is obtained.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We study the anisotropy of Ultra-High Energy Cosmic Ray (UHECR) events collected by the Telescope Array (TA) detector in the first 40 months of operation. Following earlier studies, we examine event sets with energy thresholds of $10$ EeV, $40$ EeV, and $57$ EeV. We find that the distributions of the events in right ascension and declination are compatible with an isotropic distribution in all three sets. We then compare with previously reported clustering of the UHECR events at small angular scales. No significant clustering is found in the TA data. We then check the events with $E>57$ EeV for correlations with nearby active galactic nuclei. No significant correlation is found. Finally, we examine all three sets for correlations with the large-scale structure of the Universe. We find that the two higher-energy sets are compatible with both an isotropic distribution and the hypothesis that UHECR sources follow the matter distribution of the Universe (the LSS hypothesis), while the event set with $E>10$ EeV is compatible with isotropy and is not compatible with the LSS hypothesis at 95% CL unless large deflection angles are also assumed. We show that accounting for UHECR deflections in a realistic model of the Galactic magnetic field can make this set compatible with the LSS hypothesis.\
author:
- |
T. Abu-Zayyad$^{1}$, R. Aida$^{2}$, M. Allen$^{1}$, R. Anderson$^{1}$, R. Azuma$^{3}$, E. Barcikowski$^{1}$, J.W. Belz$^{1}$, D.R. Bergman$^{1}$, S.A. Blake$^{1}$, R. Cady$^{1}$, B. G. Cheon$^{4}$, J. Chiba$^{5}$, M. Chikawa$^{6}$, E.J. Cho$^{4}$, W.R. Cho$^{7}$, H. Fujii$^{8}$, T. Fujii$^{9}$, T. Fukuda$^{3}$, M. Fukushima$^{10,11}$, W. Hanlon$^{1}$, K. Hayashi$^{3}$, Y. Hayashi$^{9}$, N. Hayashida$^{10}$, K. Hibino$^{12}$, K. Hiyama$^{10}$, K. Honda$^{2}$, T. Iguchi$^{3}$, D. Ikeda$^{10}$, K. Ikuta$^{2}$, N. Inoue$^{13}$, T. Ishii$^{2}$, R. Ishimori$^{3}$, D. Ivanov$^{1,14}$, S. Iwamoto$^{2}$, C. C. H. Jui$^{1}$, K. Kadota$^{15}$, F. Kakimoto$^{3}$, O. Kalashev$^{16}$, T. Kanbe$^{2}$, K. Kasahara$^{17}$, H. Kawai$^{18}$, S. Kawakami$^{9}$, S. Kawana$^{13}$, E. Kido$^{10}$, H.B. Kim$^{4}$, H.K. Kim$^{7}$, J.H. Kim$^{4}$, J.H. Kim$^{19}$, K. Kitamoto$^{6}$, S. Kitamura$^{3}$, Y. Kitamura$^{3}$, K. Kobayashi$^{5}$, Y. Kobayashi$^{3}$, Y. Kondo$^{10}$, K. Kuramoto$^{9}$, V. Kuzmin$^{16}$, Y.J. Kwon$^{7}$, S.I. Lim$^{20}$, S. Machida$^{3}$, K. Martens$^{11}$, J. Martineau$^{1}$, T. Matsuda$^{8}$, T. Matsuura$^{3}$, T. Matsuyama$^{9}$, J. N. Matthews$^{1}$, M. Minamino$^{9}$, K. Miyata$^{5}$, Y. Murano$^{3}$, I. Myers$^{1}$, K. Nagasawa$^{13}$, S. Nagataki$^{21}$, T. Nakamura$^{22}$, S.W. Nam$^{20}$, T. Nonaka$^{10}$, S. Ogio$^{9}$, M. Ohnishi$^{10}$, H. Ohoka$^{10}$, K. Oki$^{10}$, D. Oku$^{2}$, T. Okuda$^{23}$, A. Oshima$^{9}$, S. Ozawa$^{17}$, I.H. Park$^{20}$, M.S. Pshirkov$^{24}$, D.C. Rodriguez$^{1}$, S.Y. Roh$^{19}$, G. Rubtsov$^{16}$, D. Ryu$^{19}$, H. Sagawa$^{10}$, N. Sakurai$^{9}$, A.L. Sampson$^{1}$, L.M. Scott$^{14}$, P.D. Shah$^{1}$, F. Shibata$^{2}$, T. Shibata$^{10}$, H. Shimodaira$^{10}$, B.K. Shin$^{4}$, J.I. Shin$^{7}$, T. Shirahama$^{13}$, J.D. Smith$^{1}$, P. Sokolsky$^{1}$, T.J. Sonley$^{1}$, R.W. Springer$^{1}$, B.T. Stokes$^{1}$, S.R. Stratton$^{1,14}$, T. Stroman$^{1}$, S. Suzuki$^{8}$, Y. Takahashi$^{10}$, M. Takeda$^{10}$, A. Taketa$^{25}$, M. Takita$^{10}$, Y. Tameda$^{10}$, H. Tanaka$^{9}$, K. Tanaka$^{26}$, M. Tanaka$^{9}$, S.B. Thomas$^{1}$, G.B. Thomson$^{1}$, P. Tinyakov$^{16,24}$, I. Tkachev$^{16}$, H. Tokuno$^{3}$, T. Tomida$^{27}$, S. Troitsky$^{16}$, Y. Tsunesada$^{3}$, K. Tsutsumi$^{3}$, Y. Tsuyuguchi$^{2}$, Y. Uchihori$^{28}$, S. Udo$^{12}$, H. Ukai$^{2}$, G.Vasiloff$^{1}$, Y. Wada$^{13}$, T.Wong$^{1}$, M. Wood$^{1}$, Y. Yamakawa$^{10}$, R. Yamane$^{9}$, H. Yamaoka$^{8}$, K. Yamazaki$^{9}$, J. Yang$^{20}$, Y. Yoneda$^{9}$, S. Yoshida$^{18}$, H. Yoshii$^{29}$, X. Zhou$^{6}$, R.Zollinger$^{1}$, Z. Zundel$^{1}$\
bibliography:
- 'apj-jour.bib'
- 'uhe.bib'
- 'GMF.bib'
title: 'Search for Anisotropy of Ultra-High Energy Cosmic Rays with the Telescope Array Experiment'
---
Introduction {#sec:introduction}
============
One of the keys to understanding the nature of the Ultra-High Energy Cosmic Rays (UHECRs) is their distribution over the sky. This distribution depends on the UHECR sources, as well on the UHECR mass composition and large-scale magnetic fields, both Galactic and extragalactic. Despite significant effort, none of these issues is well understood at present.
Observation of the cutoff in the highest-energy part of the cosmic ray spectrum [@Abbasi:2007sv; @Abraham:2008ru] suggests that the UHECR propagation at high energies is limited by the interaction with the cosmic background radiation (the Greisen-Zatsepin-Kuzmin (GZK) effect [@Greisen:1966jv; @Zatsepin:1966jv]). One therefore expects that the closest sources of UHECRs are situated within the GZK volume of the size $\lsim 100$ Mpc. At these scales the matter distribution in the Universe is inhomogeneous, and so must be the distribution of the UHECR sources. If propagation of UHECRs at these distances is quasi-rectilinear (whether or not this is the case depends on both their composition and the magnetic fields), one generally expects the UHECR flux to be anisotropic, showing variations at large angular scales and possibly point sources.
If UHECR primary particles are protons, as suggested by the composition measurements performed by the High Resolution Fly’s Eye (HiRes) and the Telescope Array (TA) experiments [@Abbasi:2009nf; @Tameda:2010-uhecr2010], the UHECR propagation is, in fact, expected to be quasi-rectilinear. With the existing estimates of the Galactic magnetic field [@Han2006; @Sun:2007mx; @Pshirkov:2011um] and bounds on the extragalactic ones [@Kronberg:1993vk], the deflections of protons should be relatively small. For instance, a random extragalactic field of magnitude 1 nG and correlation length of $\sim 1$ Mpc would deflect a proton of energy $10^{20}$ eV by about $2^\circ$ over a distance of 50 Mpc, while the Galactic field would produce deflections of order $2-4^\circ$ depending on the direction. In this case a sizeable anisotropy may be expected regardless of the density of the UHECR sources down to energies as low as $10^{19}$ eV.
On the contrary, if the composition at highest energies is heavy or predominantly heavy, as the results of the Pierre Auger Observatory (PAO) [@Abraham:2010yv] seem to indicate, the quasi-rectilinear propagation is not expected for the bulk of UHECRs. Some anisotropy at large angles may still arise if the extragalactic fields are sufficiently small and the density of sources is such that only a few nearby ones contribute to the observed flux, but the small-scale anisotropy would be suppressed (for recent analyses see, e.g., @Giacinti:2010dk [@Takami:2012uw]). Thus, the study of the UHECR anisotropy may shed light on both the mass composition and the density of the UHECR sources [@Dubovsky:2000gv; @Yoshiguchi:2002rb; @Yoshiguchi:2004np; @Kachelriess:2004pc].
Numerous attempts at detection of the UHECR anisotropy have been made previously. Early studies indicated clustering of the UHECR events at small angular scales [@Hayashida:1996bc; @Tinyakov:2001ic]. On the basis of small-scale correlations, different classes of putative sources of UHECR were suggested (see, e.g., @Gorbunov:2004bs [@Abbasi:2005qy; @Cronin:2007zz; @Abraham:2007si]). More recently, the Pierre Auger Observatory has claimed correlations of UHECRs with the nearby Active Galactic Nuclei (AGN) [@Cronin:2007zz; @Abraham:2007si] which were not confirmed by observations in the Northern hemisphere [@Abbasi:2008md].
At larger angular scales, evidence for correlations with the supergalactic plane was claimed by @Stanev:1995my, @Glushkov:2001jm, and @Glushkov:2001kb but not confirmed by other authors [@Hayashida:1996bc; @Kewley:1996zt; @Bird:1998nu]. Also, @Kashti:2008bw found the anisotropy in the PAO data which was not confirmed by the HiRes data in the Northern hemisphere [@Abbasi:2010xt].
In this paper we present the anisotropy analysis of UHECR observed by the Surface Detector (SD) of the Telescope Array in the first $40$ months of its operation. TA is a hybrid UHECR detector located in the Northern hemisphere in Utah, USA (39$^\circ$17$'$48$''$ N, 112$^\circ$54$'$31$''$ W) which has been fully operational since March 2008. It consists of 507 scintillator detectors covering the area of approximately 700 km$^2$ (for details see @AbuZayyad:2012kk). The atmosphere over the surface array is viewed by 38 fluorescence telescopes arranged in 3 stations (see @Tokuno:2012mi). The surface detector of TA is the largest in the Northern hemisphere.
In this paper we focus on testing previous observations. Namely, we consider the clustering of the UHECR events at small angular scales (as would be produced by bright point sources), possible correlation with nearby AGN and correlation of the TA events with the large-scale structure (LSS) of the Universe. Following previous analyses, we consider three [*a priori*]{} chosen energy thresholds: $10$ EeV, $40$ EeV, and $57$ EeV. It should be noted that different experiments may have different energy scales due to different systematic errors in the energy determination, which may affect the selection of the events. When referring to the results of other experiments, we assume the energy scales as reported by these experiments. In statistical tests which require a pre-defined confidence level we set the latter to 95%.
The paper is organized as follows. In Section \[sec:data\] we describe the data sets used. In Section \[sec:autoc-funct\] we examine the event sets for a presence of small-scale clustering by studying the UHECR auto-correlation function. In Section \[sec:corr-with-point\] we consider correlations of UHECR events with nearby AGNs. Section \[sec:correlation-with-lss\] describes our search for correlations of the UHECR events with the large-scale structure of the Universe. In Section \[sec:conclusions\] we summarize the results and present conclusions.
Data {#sec:data}
====
Among the existing TA data sets (SD data, Fluorescence Detector (FD) data in mono and stereo modes, and hybrid detector data) the SD data set has by far the largest number of events. The present analysis is based on the data collected in the period 2008.05.11–2011.09.15 (40 months) of operation by the TA surface detector array. Cutting events with zenith angle $> 45^\circ$, the SD data set contains 988 events with energies $> 10$ EeV, 57 events with $E>40$ EeV, and 25 events with $E>57$ EeV. This is the largest UHECR set to date in the Northern hemisphere.
The angular resolution of TA events with $E>10$ EeV, is approximately $1.5^\circ$. This follows from the comparison of the thrown and reconstructed arrival directions of simulated data sets, and is supported by the direct comparison between the SD and FD arrival directions of hybrid events. The energy resolution of the TA surface detector at $E>10$ EeV is better than 20% [@AbuZayyad:2012ru].
The exposure of the TA surface array is calculated by the Monte-Carlo technique with full simulation of the detector, which will be described elsewhere. As follows from the Monte-Carlo simulations, the acceptance of the TA surface detector for $E>10$ EeV and zenith angle cut of $45^\circ$ is close to the geometrical one. For reasons of computational efficiency, in the present analysis aimed at anisotropy at relatively small angular scales we use the geometrical acceptance to generate random event sets.
Figure \[fig:geom-exposure\] shows the comparison between the distributions in declination (left column) and right ascension (right column) of the events simulated with the geometrical exposure (red line) and the data (blue data points) at the energy thresholds of $10$ EeV, $40$ EeV, and $57$ EeV (top, middle, and bottom rows, respectively). The compatibility of expected and observed distributions in all 6 cases was checked by the Kolmogorov-Smirnov (KS) test. The lowest KS probability was $p=0.13$ for the distribution in the right ascension at $E>57$ EeV. Thus, all three sets are compatible with a uniform distribution.
(220,250)(0,0) (0,170)[![\[fig:geom-exposure\] Comparison between the data (blue points) and the sets of $10^4$ events simulated with the geometrical exposure (red histogram) at energies $10$ EeV, $40$ EeV, and $57$ EeV (top, middle, and bottom rows, respectively). Plots show the distribution of events in declination (left column) and right ascension (right column). The compatibility of the two distributions by the Kolmogorov-Smirnov test is given as $P_{\rm
KS}$ in the upper left corner of each plot. ](exposure-hist-E10-dec.eps "fig:"){height="80pt"}]{} (125,170)[![\[fig:geom-exposure\] Comparison between the data (blue points) and the sets of $10^4$ events simulated with the geometrical exposure (red histogram) at energies $10$ EeV, $40$ EeV, and $57$ EeV (top, middle, and bottom rows, respectively). Plots show the distribution of events in declination (left column) and right ascension (right column). The compatibility of the two distributions by the Kolmogorov-Smirnov test is given as $P_{\rm
KS}$ in the upper left corner of each plot. ](exposure-hist-E10-ra.eps "fig:"){height="80pt"}]{} (0,85)[![\[fig:geom-exposure\] Comparison between the data (blue points) and the sets of $10^4$ events simulated with the geometrical exposure (red histogram) at energies $10$ EeV, $40$ EeV, and $57$ EeV (top, middle, and bottom rows, respectively). Plots show the distribution of events in declination (left column) and right ascension (right column). The compatibility of the two distributions by the Kolmogorov-Smirnov test is given as $P_{\rm
KS}$ in the upper left corner of each plot. ](exposure-hist-E40-dec.eps "fig:"){height="80pt"}]{} (125,85)[![\[fig:geom-exposure\] Comparison between the data (blue points) and the sets of $10^4$ events simulated with the geometrical exposure (red histogram) at energies $10$ EeV, $40$ EeV, and $57$ EeV (top, middle, and bottom rows, respectively). Plots show the distribution of events in declination (left column) and right ascension (right column). The compatibility of the two distributions by the Kolmogorov-Smirnov test is given as $P_{\rm
KS}$ in the upper left corner of each plot. ](exposure-hist-E40-ra.eps "fig:"){height="80pt"}]{} (0,0)[![\[fig:geom-exposure\] Comparison between the data (blue points) and the sets of $10^4$ events simulated with the geometrical exposure (red histogram) at energies $10$ EeV, $40$ EeV, and $57$ EeV (top, middle, and bottom rows, respectively). Plots show the distribution of events in declination (left column) and right ascension (right column). The compatibility of the two distributions by the Kolmogorov-Smirnov test is given as $P_{\rm
KS}$ in the upper left corner of each plot. ](exposure-hist-E57-dec.eps "fig:"){height="80pt"}]{} (125,0)[![\[fig:geom-exposure\] Comparison between the data (blue points) and the sets of $10^4$ events simulated with the geometrical exposure (red histogram) at energies $10$ EeV, $40$ EeV, and $57$ EeV (top, middle, and bottom rows, respectively). Plots show the distribution of events in declination (left column) and right ascension (right column). The compatibility of the two distributions by the Kolmogorov-Smirnov test is given as $P_{\rm
KS}$ in the upper left corner of each plot. ](exposure-hist-E57-ra.eps "fig:"){height="80pt"}]{}
Autocorrelation function {#sec:autoc-funct}
========================
The AGASA experiment reported clustering of UHECR events with $E>40$ EeV at the angular scale of $2.5^\circ$ [@Hayashida:1996bc]. Here we repeat this analysis using the TA data set.
The procedure is as follows: for a given angular separation, $\delta$, we count the number of pairs of observed events that are separated by an angular distance less than $\delta$, thus obtaining the data count. We then generate a large number (typically, $10^5$) of Monte-Carlo (MC) event sets each having the same number of events as the real data set. The simulated sets are generated with a uniform distribution according to the TA exposure. In each MC set we count pairs of events in the same way as in the data, which gives the MC count for that set. We then calculate the average MC count for all of the MC sets. This represents the expected number of pairs for the angular scale $\delta$, assuming a uniform cosmic ray distribution. For each value of $\delta$ we then determine the fraction of simulated sets where the number of pairs is greater than or equal to the number of pairs in the data. This gives the $p$-value, $P(\delta)$, that describes how likely the excess of pairs, if found in the data, is to occur as a result of a fluctuation in a random set. Small values of $P(\delta)$, thus indicate a departure from uniformity at the corresponding angular scale.
We first perform a blind test of the AGASA claim. Fixing the energy threshold to $40$ EeV and the separation angle to $\delta = 2.5^\circ$ we find 0 pairs while 1.5 pairs are expected in the case of a uniform distribution. Therefore, there is no excess of small-scale clusters in the TA data.
(220,250)(0,0) (0,170)[![\[fig:auto\] Autocorrelations in the TA data sets at $E>10$ EeV, $E>40$ EeV, and $E>57$ EeV (top, middle, and bottom rows, respectively). [*Left panels:*]{} the number of pairs with angular separations $\delta$ normalized to the area of the angular bin (data points), compared to the expectation for the uniform distribution (shaded histogram). The errors are 1-sigma Poisson errors. [*Right panels:*]{} probability, $P(\delta)$, that the excess of pairs with the angular separation less than $\delta$ occurs as a fluctuation in a uniform distribution. Small $P(\delta)$ indicates a departure from isotropy. ](auto-E10a.eps "fig:"){height="80pt"}]{} (125,170)[![\[fig:auto\] Autocorrelations in the TA data sets at $E>10$ EeV, $E>40$ EeV, and $E>57$ EeV (top, middle, and bottom rows, respectively). [*Left panels:*]{} the number of pairs with angular separations $\delta$ normalized to the area of the angular bin (data points), compared to the expectation for the uniform distribution (shaded histogram). The errors are 1-sigma Poisson errors. [*Right panels:*]{} probability, $P(\delta)$, that the excess of pairs with the angular separation less than $\delta$ occurs as a fluctuation in a uniform distribution. Small $P(\delta)$ indicates a departure from isotropy. ](auto-E10b.eps "fig:"){height="80pt"}]{} (0,85)[![\[fig:auto\] Autocorrelations in the TA data sets at $E>10$ EeV, $E>40$ EeV, and $E>57$ EeV (top, middle, and bottom rows, respectively). [*Left panels:*]{} the number of pairs with angular separations $\delta$ normalized to the area of the angular bin (data points), compared to the expectation for the uniform distribution (shaded histogram). The errors are 1-sigma Poisson errors. [*Right panels:*]{} probability, $P(\delta)$, that the excess of pairs with the angular separation less than $\delta$ occurs as a fluctuation in a uniform distribution. Small $P(\delta)$ indicates a departure from isotropy. ](auto-E40ba.eps "fig:"){height="80pt"}]{} (125,85)[![\[fig:auto\] Autocorrelations in the TA data sets at $E>10$ EeV, $E>40$ EeV, and $E>57$ EeV (top, middle, and bottom rows, respectively). [*Left panels:*]{} the number of pairs with angular separations $\delta$ normalized to the area of the angular bin (data points), compared to the expectation for the uniform distribution (shaded histogram). The errors are 1-sigma Poisson errors. [*Right panels:*]{} probability, $P(\delta)$, that the excess of pairs with the angular separation less than $\delta$ occurs as a fluctuation in a uniform distribution. Small $P(\delta)$ indicates a departure from isotropy. ](auto-E40b.eps "fig:"){height="80pt"}]{} (0,0)[![\[fig:auto\] Autocorrelations in the TA data sets at $E>10$ EeV, $E>40$ EeV, and $E>57$ EeV (top, middle, and bottom rows, respectively). [*Left panels:*]{} the number of pairs with angular separations $\delta$ normalized to the area of the angular bin (data points), compared to the expectation for the uniform distribution (shaded histogram). The errors are 1-sigma Poisson errors. [*Right panels:*]{} probability, $P(\delta)$, that the excess of pairs with the angular separation less than $\delta$ occurs as a fluctuation in a uniform distribution. Small $P(\delta)$ indicates a departure from isotropy. ](auto-E57ba.eps "fig:"){height="80pt"}]{} (125,0)[![\[fig:auto\] Autocorrelations in the TA data sets at $E>10$ EeV, $E>40$ EeV, and $E>57$ EeV (top, middle, and bottom rows, respectively). [*Left panels:*]{} the number of pairs with angular separations $\delta$ normalized to the area of the angular bin (data points), compared to the expectation for the uniform distribution (shaded histogram). The errors are 1-sigma Poisson errors. [*Right panels:*]{} probability, $P(\delta)$, that the excess of pairs with the angular separation less than $\delta$ occurs as a fluctuation in a uniform distribution. Small $P(\delta)$ indicates a departure from isotropy. ](auto-E57b.eps "fig:"){height="80pt"}]{}
We next extend the analysis to all angular scales. No significant excess is found. The results are illustrated in Figure \[fig:auto\] for angles from $0$ to $40^\circ$ and three energy thresholds of $10$ EeV, $40$ EeV, and $57$ EeV as specified on the plots. For each energy threshold, the left panel shows the number of pairs with the angular separations $\delta$ binned in $2^\circ$ bins (data points). The shaded region represents the average number of pairs expected in the case of the uniform distribution. Both the data and the uniform expectation are normalized bin-by-bin to the area of the bin, so that in the case of a uniform full-sky exposure the expectation would be flat. The overall normalization is set in such a way that the expectation in the first bin equals one.
The right panels of Figure \[fig:auto\] show the dependence of the p-value, $P(\delta)$, on the separation angle, $\delta$, for the corresponding energy. Note that $P(\delta)$ is a cumulative quantity since it takes into account all the pairs separated by angles from 0 to $\delta$. For this reason a small, but coherent over several bins, excess at angles from 10 to 20 degrees on the lower left panel of Figure \[fig:auto\] produces a more significant feature in the corresponding $P(\delta)$, lower right panel of Figure \[fig:auto\]. This feature corresponds to the group of events visible on the sky map (see the lower panel of Figure \[fig:p-skymaps\]).
When accessing the significance of departures from isotropy on the basis of $P(\delta)$ represented in Figure \[fig:auto\], one should take into account the fact that the angular scale of the excess is not known in advance. Thus, there is a statistical penalty for choosing this scale [*a posteriori*]{} (see @Tinyakov:2004bb for a detailed discussion). Taking this penalty into account, none of the three examined data sets shows a significant deviation from an isotropic distribution.
Interestingly, although close clusters in the high-energy TA event set are absent, one of the TA events falls within $1.7^\circ$ of a high energy event observed by the Auger Observatory [@Abreu:2010zzj]. Both events have $E>10^{20}$ EeV. The center of the doublet has the Galactic coordinates $l=36^\circ$, $b=-4.3^\circ$.
Correlation with Active Galactic Nuclei {#sec:corr-with-point}
=======================================
The Auger collaboration has reported a correlation [@Cronin:2007zz; @Abraham:2007si] between UHECRs with $E>57$ EeV and the nearby (redshift $z\leq 0.018$ or, equivalently, distance $d<75$ Mpc) Active Galactic Nuclei (AGNs) from the Veron-Cetty & Veron (VCV) catalog [@VeronCetty:2006zz]. The greatest correlation was observed at the angle of $3.1^\circ$. In the control data set, the number of correlating events was 9 out of 13, which corresponds to about 69% of events. The Auger collaboration has recently updated the analysis and found that a smaller fraction of the UHECR events correlates with the same set of AGNs in the latest UHECR data set [@Abreu:2010zzj] than in the original one. Out of 55 events with $E>55$ EeV, 21 were found to correlate with AGNs, which corresponds to a fraction of correlating events equal to 38%. In this section we test the TA data for correlations with AGN.
The set of 472 nearby AGNs used by @Cronin:2007zz contains 7 objects listed at zero redshift, all in the field of view of TA. Of these 7 objects, two are stars, one is a quasar with unknown redshift, one is a Seyfert 2 galaxy, two are spiral galaxies (including the Andromeda galaxy) and one is a dwarf spheroidal galaxy. We exclude these objects from the analysis, which leaves 465 objects in the AGN catalog.
The TA exposure is peaked in the Northern hemisphere, so that the AGNs visible to TA are largely different from those visible to Auger, though there is some overlap. The distribution of nearby AGNs over the sky is not uniform because of the large scale structure (see Section \[sec:correlation-with-lss\] for more detail) and because the VCV catalog is not complete: due to observational bias it tends to contain more objects in the Northern hemisphere. For this reason, a larger fraction of events is expected to correlate with AGNs in the TA data under the assumption that AGNs are sources of the observed UHECRs. Taking into account the distribution of nearby AGNs over the sky and assuming equal AGN luminosities in UHECR, we estimated the correlating fraction will be $\sim 73$% for TA on the basis of the original PAO claim, and $\sim 43$% on the basis of the updated analysis by PAO.
The sky map of TA events with $E>57$ EeV and nearby AGNs from the VCV catalog is represented in Figure \[fig:E57vsAGN\] in Galactic coordinates. The cosmic rays are shown by filled red (correlating events) and empty blue circles (non-correlating events). AGNs are shown by black dots.
(220,150)(0,0) (0,20)[ ![\[fig:E57vsAGN\] Hammer projection of the TA cosmic ray events with $E>57$ EeV and nearby AGNs in the Galactic coordinates. Correlating and non-correlating events are shown by filled red and empty blue circles, respectively. AGNs are represented by black dots. The dashed line shows the boundary of the TA exposure. ](AGN_25.eps "fig:"){height=".16\textheight"} ]{}
Figure \[fig:AGNsignificance\] shows the number of TA events correlating with AGNs as a function of the total number of events with $E>57$ EeV ordered according to arrival time. The black dashed line represents the expected number of random coincidences in case of a uniform distribution calculated via Monte-Carlo simulation. The blue line shows the expected number of correlating events as derived from the original PAO claim. Shaded regions represent 68% and 95% CL deviations from this expectation calculated by the maximum likelihood method of Ref. [@Gorbunov:2005fi]. As is seen from Figure \[fig:AGNsignificance\], present TA data are compatible with both isotropic distribution and the AGN hypothesis.
(220,150)(0,0) (0,10)[ ![\[fig:AGNsignificance\] The number of TA events with $E>57$ EeV correlating with VCV AGNs as a function of the total number of events. The expectation according to the original PAO claim is represented by the blue line together with the 1- and 2-sigma significance bands. The black dashed line shows the expected number of random coincidences. ](AGN-TA-25corr8b1.eps "fig:"){height=".2\textheight"} ]{}
In the full TA SD data set, there are 11 correlating events out of 25 total, while the expected number of random coincidences for this total number of events is 5.9. Making use of the binomial distribution with the probability of a single event to correlate $p_{\rm
iso}=0.24$, one finds that such an excess has probability of $\sim 2\%$ to occur by chance with isotropic distribution of arrival directions.
Correlation with LSS {#sec:correlation-with-lss}
====================
Even though the sources of UHECRs are not known, their distribution in space at large scales must follow that of the ordinary matter. The latter is anisotropic at scales below $\sim 100$ Mpc forming the Large-Scale Structure (LSS) of the Universe that consists of galaxy clusters, filaments and voids. If UHECRs are not strongly deflected on their way to Earth, their distribution over the sky should correlate with the nearby structures, with over-densities corresponding to close clusters and under-densities corresponding to voids.
The amplitude of anisotropy depends on the UHECR propagation length (the larger is the propagation length, the smaller contributions of the local structures and, therefore, the anisotropy) and on the UHECR deflections. In this section the propagation of UHECR is calculated assuming they are protons. However, it should be noted that regardless of whether the UHECR composition is heavy or light, their propagation length changes with energy roughly in the same way and becomes of order several tens of megaparsecs as the energy approaches $10^{20}$ eV. Thus, the most important parameter that determines the amplitude of the anisotropy at a given energy is the typical deflection angle which we denote as $\theta$ (which is, of course, very different for heavy and light composition).
The goal of this analysis is to determine which values of $\theta$ are compatible with the space distribution of the TA events. In principle, this can be done at all energies. To minimize statistical penalties, we limit our analysis to the energy thresholds of $10$ EeV, $40$ EeV, and $57$ EeV.
Statistical Method {#sec:statistical-method}
------------------
To test the compatibility between the observed UHECR distribution over the sky and that expected under the LSS hypothesis (that is, the hypothesis that UHECR sources trace matter distribution in the Universe), we employ the method developed by @Koers:2008ba and used previously in the analysis of the HiRes data [@Abbasi:2010xt]. In this method, one first computes the UHECR flux distribution expected under the LSS hypothesis and then compares it to the observed one by the flux sampling test.
The matter distribution in the nearby Universe may be inferred from the complete galaxy catalogs containing the redshift information. In this work we use the 2MASS Galaxy Redshift Catalog (XSCz)[^1] that is derived from the 2MASS Extended Source Catalog (XSC), with redshifts that have either been spectroscopically measured (for most of the objects) or derived from the 2MASS photometric measurements. This catalog provides the most accurate information about 3D galaxy distribution to date.
For the flux calculations, we use the flux-limited subsample of galaxies with apparent magnitude $m\leq 12.5$. For fainter objects, the completeness of the catalog degrades progressively, while their inclusion does not change the results considerably. We exclude objects closer than $5$ Mpc in order to avoid breaking of the statistical description (if such objects are assumed to be sources of UHECR, they have to be treated individually). We also cut out galaxies at distances further than $250$ Mpc replacing their combined contribution by a uniform flux normalized in such a way that it provides the correct fraction of events as calculated in the approximation of a uniform source distribution. The quantitative justification of these procedures can be found in [@Koers:2009pd]. The resulting catalog contains 106218 galaxies, which is sufficient to accurately describe the flux distribution at angular scales down to $\sim 2^\circ$. The UHECR flux distribution is reconstructed from this flux-limited catalog by the weighting method proposed by @Lynden-Bell:1971 and adapted to flux calculations by [@Koers:2009pd].
The XSCz catalog loses completeness in the band of roughly $\pm 10^\circ$ around the Galactic plane and especially around the Galactic center. The size of this region is not much larger than a typical deflection of a proton even at $57$ EeV, so this gap may be bridged without loss of accuracy. Away from the Galactic center at $|l|>60^\circ$ where only a fraction of the galaxies (the dimmer part) is missing in the catalog, we apply a $l-$ and $b-$dependent weight correction to the remaining galaxies so as to compensate for the missing ones. In the region close to the Galactic center, $|l|<60^\circ$, we extrapolate the flux density from the adjacent regions in a straightforward manner. The latter is not an accurate procedure; however, the Galactic center region overlaps with the TA exposure only slightly, and this inaccuracy is not important for our results as can be checked by excluding this region from the analysis.
When propagating the UHECR primary particles from a source to the Earth, we assume them to be protons and take full account of the attenuation processes. The injection index at the source is taken to be 2.4, which is compatible with the UHECR spectrum observed by HiRes and TA [@AbuZayyad:2012ru] assuming proton composition and the source evolution parameter $m=4$ [@Gelmini:2007jy]. We also assume that the effects of both the Galactic and extragalactic magnetic fields can be approximated by a single parameter, the Gaussian smearing angle $\theta$. We consider $\theta$ a free parameter and vary it in the range $2 - 20^\circ$. In general, the deflections of UHECR in magnetic fields contain both random and regular parts, the latter being due to the regular component of the Galactic magnetic field. The regular deflections are not Gaussian. However, the statistical test we use here is not sensitive to the coherent character of deflections provided they do not exceed $10-20^\circ$ as set by the typical size of the flux variations due to local structures (cf. Figure \[fig:E57skymap\]). So, for the most part of the analysis we will use the Gaussian smearing to represent all the deflections without making the distinction between the regular and random ones. Later, in Section \[sec:acco-galact-magn\] we will discuss, in the case of the lowest energy set and the largest deflections, the effect of explicitly accounting for the regular component of the Galactic magnetic field.
To calculate the expected flux, we assume that UHECR sources follow the space distribution of galaxies. The simplest way to realize this assumption in practice is to assign each galaxy an equal luminosity in UHECR’s. This is a good approximation if the density of the UHECR sources is sufficiently high (so that many sources are present in local structures contributing to the anisotropy). The contribution of each galaxy to the total flux is then calculated taking into account the distance of the source and the corresponding flux attenuation. Individual contributions are smeared with the Gaussian width $\theta$, so that the flux at a given point of the sky is a sum of contributions of all the galaxies within the angular distance of order $\theta$. Further details on the flux calculation can be found in References [@Koers:2008ba; @Koers:2009pd; @Abbasi:2010xt].
Figure \[fig:E57skymap\] shows the flux map calculated by the above procedure for an energy threshold of $57$ EeV and smearing angle $\theta = 6^\circ$, not yet modulated with the TA exposure. Darker regions correspond to higher flux. A band of each color integrates to $1/5$ of the total flux. One can identify the nearby structures which are marked by letters on the plot as explained in the caption.
![Sky map of expected flux at $E>57$ EeV (Galactic coordinates). The smearing angle is $6^\circ$. Letters indicate the nearby structures as follows: [**C**]{}: Centaurus supercluster (60 Mpc); [**Co**]{}: Coma cluster (90 Mpc); [**E**]{}: Eridanus cluster (30 Mpc); [**F**]{}: Fornax cluster (20 Mpc); [**Hy**]{}: Hydra supercluster (50 Mpc); [**N**]{}: Norma supercluster (65 Mpc); [**PI**]{}: Pavo-Indus supercluster (70 Mpc); [**PP**]{}: Perseus-Pisces supercluster (70 Mpc); [**UM**]{}: Ursa Major (20 Mpc); [**V**]{}: Virgo cluster (20 Mpc).[]{data-label="fig:E57skymap"}](skymap-E57-th6-canstellations.eps){width=".95\columnwidth"}
The next step is to compare the calculated flux distribution to the actual distribution of the TA events and determine whether they are statistically compatible. In this work we use the flux sampling test proposed by @Koers:2008ba. The starting point is the map of the expected flux calculated as explained above. One reads off the flux values at positions of the data events. This gives a set of numbers which we refer to as the “data set”. One may say that the cosmic ray events sample the flux map in a particular way that depends on their space distribution. One then generates a large number of Monte-Carlo events which are distributed according to the expected flux and reads off the flux values at their positions. This gives the set of flux values which we refer to as the “MC set”. If the [ *angular*]{} distribution of the data and MC events is the same, so must be the distributions of the flux values in the data and MC sets. These two distributions may be compared by the parameter-free Kolmogorov-Smirnov (KS) test.
The result of the KS test is the $p$-value which shows whether the data and MC flux sets are drawn from the same parent distribution. If this $p$-value is low, the two distributions of flux values are different and, therefore, the angular distributions of data and MC sets are different.
Estimate of Statistical Power of the Flux Sampling Test {#sec:estim-stat-power}
-------------------------------------------------------
An important characteristic of a statistical test is its ability to discriminate between two hypotheses, or the statistical power. For the case at hand, the statistical power is the probability to rule out the LSS hypothesis at 95% CL if the cosmic ray distribution is isotropic. The closer the statistical power to one, the more sensitive is the test. Knowing the statistical power provides an [*a priori*]{} idea of what kind of sensitivity can be reached with the given number of events.
In general, the statistical power increases with smaller smearing angles since this improves the contrast in the flux map. For the same reason, the statistical power increases with energy (the UHECR propagation length becomes shorter and the relative contribution of the local structures is therefore enhanced). Also, the statistical power increases with the number of events.
We have calculated the statistical power of the flux sampling test in case of TA for the three energy thresholds of $10$ EeV, $40$ EeV, and $57$ EeV, and smearing angles varying from $2^\circ$ to $14^\circ$. We have found that for the actual number of events in the TA data set, the statistical power is below 50% for smearing angles $\theta>9^\circ$, $\theta>3^\circ$ and $\theta>4^\circ$ for the above three energy thresholds, respectively.
The case $E>57$ EeV is shown in Figure \[fig:powers\].
![\[fig:powers\] The statistical power of the flux sampling test at $E>57$ EeV as a function of the smearing angle. Different curves correspond to different number of events, as indicated on the plot. The actual number of events in the TA data set with $E>57$ EeV is 25. The gray region shows the expected range of deflections in the Galactic magnetic field in the case of protons. ](powers-E57.eps){width=".9\columnwidth"}
The various curves in the plot correspond to different number of events (note that the actual number of events in the TA data set with $E>57$ EeV is 25). The gray region represents the expected range of deflections in the Galactic magnetic field in the case of protons.
Results {#sec:results}
-------
First, we check the compatibility of the TA event sets with the isotropic distribution. To this end we generate an isotropic flux map modulated with the TA exposure. This map is independent of energy and smearing angle. We then test the compatibility of the TA event set for $E>10$ EeV, $E>40$ EeV, and $E>57$ EeV with this map. The flux sampling test gives the p-values 0.5, 0.9 and 0.6 for the three data sets, respectively. Thus, at all three energy thresholds the data appear to be compatible with an isotropic distribution.
Next, we examine the compatibility of the TA event sets with the LSS hypothesis. Figure \[fig:p-skymaps\] shows the skymaps of the expected flux at energy thresholds of $10$ EeV, $40$ EeV, and $57$ EeV (top to bottom) and the smearing angle of $6^\circ$. The white dots represent the arrival directions of the TA events. The bands are drawn in the same way as in Figure \[fig:E57skymap\], i.e., each band integrates to 1/5 of the total flux. This means, in particular, that if the LSS model were true each band would contain 1/5 of the total number of events in average. Note that the configuration of the bands changes with energy because of the energy dependence of the propagation length.
(200,369)(0,0) (5,246)[![ \[fig:p-skymaps\] The skymaps of the expected flux at energy thresholds of $10$ EeV, $40$ EeV, and $57$ EeV (from top to bottom) in Galactic coordinates with the TA events superimposed (white dots). The smearing angle is $6^\circ$. ](skymap-E10-a24-m4.eps "fig:"){width=".95\columnwidth"}]{} (5,123)[![ \[fig:p-skymaps\] The skymaps of the expected flux at energy thresholds of $10$ EeV, $40$ EeV, and $57$ EeV (from top to bottom) in Galactic coordinates with the TA events superimposed (white dots). The smearing angle is $6^\circ$. ](skymap-E40-a24-m4.eps "fig:"){width=".95\columnwidth"}]{} (5,0)[![ \[fig:p-skymaps\] The skymaps of the expected flux at energy thresholds of $10$ EeV, $40$ EeV, and $57$ EeV (from top to bottom) in Galactic coordinates with the TA events superimposed (white dots). The smearing angle is $6^\circ$. ](skymap-E57-a24-m4.eps "fig:"){width=".95\columnwidth"}]{}
The results of the flux sampling tests are presented in Figure \[fig:p-values\]. The $p$-values are shown as a function of the smearing angle at energy thresholds of $10$ EeV, $40$ EeV, and $57$ EeV.
![ \[fig:p-values\] The results of the statistical test for the compatibility between the data and the LSS hypothesis. The p-values (red points) are shown as a function of the smearing angle $\theta$. Low p-values indicate incompatibility with the LSS model. The horizontal line shows a confidence level of 95%. The three panels correspond to energy thresholds of $10$ EeV, $40$ EeV, and $57$ EeV from top to bottom, as indicated on the plots.](pvals-noB-E10_p2.4_m4.eps "fig:"){width=".9\columnwidth"} ![ \[fig:p-values\] The results of the statistical test for the compatibility between the data and the LSS hypothesis. The p-values (red points) are shown as a function of the smearing angle $\theta$. Low p-values indicate incompatibility with the LSS model. The horizontal line shows a confidence level of 95%. The three panels correspond to energy thresholds of $10$ EeV, $40$ EeV, and $57$ EeV from top to bottom, as indicated on the plots.](pvals-noB-E40_p2.4_m4.eps "fig:"){width=".9\columnwidth"} ![ \[fig:p-values\] The results of the statistical test for the compatibility between the data and the LSS hypothesis. The p-values (red points) are shown as a function of the smearing angle $\theta$. Low p-values indicate incompatibility with the LSS model. The horizontal line shows a confidence level of 95%. The three panels correspond to energy thresholds of $10$ EeV, $40$ EeV, and $57$ EeV from top to bottom, as indicated on the plots.](pvals-noB-E57_p2.4_m4.eps "fig:"){width=".9\columnwidth"}
Each point represents the p-value obtained by the flux sampling test at the corresponding energy threshold and smearing angle.
As one can see from the plots, for $E>40$ EeV and $E>57$ EeV the data are compatible with the structure hypothesis at the 95% C.L. The decrease of the p-values slightly below the 95% C.L. in the case $E>57$ EeV cannot be assigned a real significance in view of the penalty factors for trials (e.g., three energy thresholds).
Although large smearing angles do not have a straightforward physical interpretation in view of the Gaussian approximation used, we have investigated the behavior of the p-values corresponding to the case $E>57$ EeV for larger smearing angles and found that it fluctuates around $p\simeq 0.05$ for angles as large as $\theta \sim 50^\circ$ and then goes to $p\simeq 1$. Such behavior may arise because the flux map for $E>57$ EeV remains anisotropic even for very large smearing angles.
At the energy threshold of $E>10$ EeV the situation is somewhat different. The data are incompatible with the structure model up to angles of order $20^\circ$. In view of the large deflections in magnetic fields at low energies, such behavior is expected. One should be careful, however, with the interpretation of this result. First, Figure \[fig:p-values\] does not include the penalty for the number of trials. Second, at $E>10$ EeV the uncertainties in the flux calculation due to the choice of the model parameters (in particular, the injection index and the evolution parameter) are the largest. Finally, if the smearing angle is attributed to deflections in the magnetic fields, the dominant contribution is likely to come from the regular component of the Galactic magnetic field, as discussed in the next section. Such large and regular deflections require a more accurate modeling, which we attempt in the next section.
Accounting for the Galactic Magnetic Field {#sec:acco-galact-magn}
------------------------------------------
The deviation from the structure model at $E>10$ EeV and small smearing angles is an indication that magnetic field deflections play an important role in the distribution of the UHECR arrival directions. In general, several contributions to the deflections are expected. First, there are deflections produced by intergalactic magnetic fields. These fields are known quite poorly. They are usually thought to obey the upper bound of $B\lsim 10^{-9}$ G with a correlation length, $l\lsim 1$ Mpc [@Kronberg:1993vk]. With these parameters, a proton of energy $10$ EeV coming from $50$ Mpc would be deflected by $\sim 20^\circ$. However, there are indications that the extragalactic magnetic fields may be several orders of magnitude smaller [@Dolag:2004kp] than the upper bound.
Second, UHECRs are deflected in the regular component of the Galactic Magnetic Field (GMF). The regular GMF is known much better than extragalactic fields. It can be inferred, e.g., from the Faraday rotation measures of Galactic and extragalactic radio sources. According to recent studies, a typical deflection of a $10$ EeV proton would be $20-40^\circ$ [@Pshirkov:2011um]. This is comparable or larger than the deflection in the extragalactic field.
Finally, the Galactic field has a random component. Although the amplitude of this component is a few times larger than the regular one, its contribution into the deflections is subdominant (or at most comparable) to that of the regular component [@Tinyakov:2004pw] due to its random character.
From this discussion it is clear that, most likely, the regular part of the magnetic field provides the dominant contribution into the UHECR deflections. At low energies when the magnitude of the deflections becomes large, Gaussian smearing is not a good approximation for such deflections. They have to be taken into account explicitly.
In order to see whether the deflections in the regular GMF can be a reason for the discrepancy between the data and the LSS model we have repeated the analysis of Section \[sec:results\] including the regular GMF. The presence of the regular magnetic field is taken into account by modifying the expected flux distribution. The smearing angle remains a free parameter; it accounts for random deflections in the extragalactic fields and in the random component of the GMF. The statistical test itself remains unchanged.
We adopt the recent GMF model by @Pshirkov:2011um. This model has been obtained by fitting the GMF model parameters to the latest catalog of the Faraday rotation measures of extragalactic sources. In addition to the disk field, this model also contains a toroidal halo field.
Although the fits to the Faraday rotation measures constrain the parameters of the GMF, some combinations of these parameters are constrained rather poorly. In particular, the magnitude of the halo field is degenerate with the halo height above the Galactic disk: making the halo field stronger and simultaneously higher above the disk does not strongly affect the rotation measures. Thus, there remains some freedom in the choice of the GMF parameters. The question is whether this freedom can be used to bring the arrival directions of UHECR into accord with the LSS hypothesis without contradicting the Faraday rotation data.
(210,280)(0,0) (0,150)[![ \[fig:GMFincluded\] [*Upper panel:*]{} The sky map of the expected flux for $E>10$ EeV and smearing angle $6^\circ$ taking into account the GMF (Galactic coordinates). The parameters of GMF are as follows: the magnitude of the halo is 4 $\mu$G and the thickness of the halo is 1.5 kpc. Note the absence of overdensity in the direction of the Virgo cluster. [*Lower panel:*]{} The result of the statistical test of the compatibility between the TA event set with $E>10$ EeV and the LSS hypothesis for different models of GMF: no magnetic field (circles), disk component only (triangles), both disk and halo components (squares). The horizontal line shows the confidence level of 95%. Low p-values indicate incompatibility. ](skymap-E10-a24-m4-GMF.eps "fig:"){width=".95\columnwidth"}]{} (0,0)[![ \[fig:GMFincluded\] [*Upper panel:*]{} The sky map of the expected flux for $E>10$ EeV and smearing angle $6^\circ$ taking into account the GMF (Galactic coordinates). The parameters of GMF are as follows: the magnitude of the halo is 4 $\mu$G and the thickness of the halo is 1.5 kpc. Note the absence of overdensity in the direction of the Virgo cluster. [*Lower panel:*]{} The result of the statistical test of the compatibility between the TA event set with $E>10$ EeV and the LSS hypothesis for different models of GMF: no magnetic field (circles), disk component only (triangles), both disk and halo components (squares). The horizontal line shows the confidence level of 95%. Low p-values indicate incompatibility. ](pvals-comparison-E10_p2.4_m4.eps "fig:"){width=".9\columnwidth"}]{}
We have found that the compatibility with the LSS model cannot be reached without the halo field. When the halo is included, the compatibility with the LSS model is possible, although the required halo field is rather strong (but still compatible with the data on the Faraday rotation measures).
An example of the flux map with the GMF included is shown in the upper panel of Figure \[fig:GMFincluded\]. The flux distribution is calculated for the case $E>10$ EeV and smearing angle of $6^\circ$. The magnetic field parameters are as follows: the magnitude of the halo 4 $\mu$G and the thickness of the halo is 1.5 kpc. Note that after the inclusion of GMF the Virgo region has moved away from the TA field of view, and the expected flux distribution has become closer to the uniform one.
The results of the flux sampling test of the LSS model with the regular GMF included are shown in the lower panel of Figure \[fig:GMFincluded\]. Black squares represent the p-values in the case of the GMF with the parameters described above. For comparison, red circles show the p-values in the absence of GMF (the same as the upper panel of Figure \[fig:p-values\]), while green triangles represent the case of GMF with the disk component only. One can see that the regular GMF can produce deflections that make the data for $E>10$ EeV compatible with the LSS model for all but the smallest smearing angles. Thus, the discrepancy between the LSS hypothesis and the TA data with $E>10$ EeV can, in principle, be explained by the deflections in the regular GMF.
Summary and conclusions {#sec:conclusions}
=======================
In this paper we present a search for anisotropy in the TA data collected over the period of about 40 months, which is the largest UHECR data set to date in the Northern hemisphere. The main focus of this paper is on checking the existing claims: small-scale clustering, correlation with nearby AGNs, and correlation with the LSS.
The results are summarized as follows.
- The TA data show no clustering of the UHECR events at small scales, neither at the angular scale of $2.5^\circ$ in the set with $E>40$ EeV as reported by the AGASA experiment, nor at any angular scale from 0 to 40 degrees in the data sets with $E>10$ EeV, $E>40$ EeV, and $E>57$ EeV. There is a hint of grouping of events at angular scales of $20-30^\circ$ at the highest energies, however the statistical significance of this feature is insufficient for a definite conclusion.
- There is no statistically significant correlation of the TA data with $E>57$ EeV with the positions of nearby AGNs from the VCV catalog using the parameters reported by the PAO (angular scale of $3.1^\circ$ and redshift cut in the VCV catalog $z\leq 0.018$). Out of 25 observed events with $E>57$ EeV, 11 have been found to correlate with positions of nearby AGNs, while 5.9 are expected in average from random coincidences (chance probability of 2%).
- The TA event sets with $E>10$ EeV, $E>40$ EeV, and $E>57$ EeV appear compatible with a uniform distribution according to the flux sampling test. The sets with $E>40$ EeV and $E>57$ EeV are also compatible, at 95% CL, with a model which assumes that sources follow the large-scale structure of the Universe (LSS model). The set with $E>10$ EeV is not compatible, at 95% CL, with the LSS model unless the deflections of these UHECR exceeds $20^\circ$.
- The set with $E>10$ EeV can be made compatible with the LSS model, at smearing angles larger than $\sim 3^\circ$, by including the effect of the regular component of the Galactic magnetic field and assuming a realistic model for the latter. The smearing angle in this case represents the deflections in the random Galactic and extragalactic fields.
From the analysis presented, one concludes that there is no apparent deviation from isotropy in the present TA data. At high energies, this may be merely due to an insufficient number of events. However, if this tendency persists at several times larger statistics, it will be difficult to reconcile with the proton composition of UHECR regardless of the source nature: if the sources within the GZK volume are numerous, they must follow the (inhomogeneous) matter distribution. If the source density is small so that there are only a few within the GZK volume, this very fact will produce anisotropy.
At lower energies the deflections are expected to be large even for protons, which makes the distribution of the events more isotropic. However, the number of events is much larger as well, so that even small deviations from isotropy may become detectable as the statistics increases. The fact that for $E>10$ EeV the distribution of the events is not compatible with the LSS model without assuming a large ($\gsim 20^\circ$) smearing angle may indicate that we are observing the first manifestation of UHECR deflections in the Galactic magnetic fields. The possibility to reconcile the observed UHECR distribution with the LSS model by correcting for the deflections in the realistic model of GMF is in accord with this interpretation.
Acknowledgements
================
The Telescope Array experiment is supported by the Japan Society for the Promotion of Science through Grants-in-Aid for Scientific Research on Specially Promoted Research (21000002) “Extreme Phenomena in the Universe Explored by Highest Energy Cosmic Rays” and for Scientific Research (S) (19104006), and the Inter-University Research Program of the Institute for Cosmic Ray Research; by the U.S. National Science Foundation awards PHY-0307098, PHY-0601915, PHY-0703893, PHY-0758342, PHY-0848320, PHY-1069280, and PHY-1069286 (Utah) and PHY-0649681 (Rutgers); by the National Research Foundation of Korea (2006-0050031, 2007-0056005, 2007-0093860, 2010-0011378, 2010-0028071, 2011-0002617, R32-10130); by the Russian Academy of Sciences, RFBR grants 10-02-01406a and 11-02-01528a (INR), IISN project No. 4.4509.10 and Belgian Science Policy under IUAP VI/11 (ULB); by the Grant-in-Aid for the Scientific Research (S) No. 19104006 by the Japan Society for the Promotion of Science. The foundations of Dr. Ezekiel R. and Edna Wattis Dumke, Willard L. Eccles and the George S. and Dolores Dore Eccles all helped with generous donations. The State of Utah supported the project through its Economic Development Board, and the University of Utah through the Office of the Vice President for Research. The experimental site became available through the cooperation of the Utah School and Institutional Trust Lands Administration (SITLA), U.S. Bureau of Land Management and the U.S. Air Force. We also wish to thank the people and the officials of Millard County, Utah, for their steadfast and warm support. We gratefully acknowledge the contributions from the technical staffs of our home institutions and the University of Utah Center for High Performance Computing (CHPC).
Appendix: List of events with $E>57$ EeV {#sec:evlist}
========================================
In this Appendix we present the list of events with energy $E>57$ EeV and zenith angle $\theta < 45^\circ$ that have been recorded by the surface detector of the Telescope Array from May 11, 2008 to September 15, 2011. During this period, 25 such events were observed. Table \[tab:events\] shows the arrival date and time of these events, the zenith angle $\theta$, energy in units of EeV, and Galactic coordinates $l$ and $b$ in degrees. The angular resolution of these events is $\sim1.5^\circ$, while the energy resolution is better than $20\%$.
date & time (UTC) $\theta$ (deg) $E$ (EeV) $l$ (deg) $b$ (deg)
--------------------- ---------------- ----------- ------------ -----------
2008-06-25 19:45:52 32.8 82.6 178.6 -19.4
2008-07-15 05:26:31 34.4 57.7 90.5 8.0
2008-08-10 12:45:04 38.0 122.6 102.7 -19.2
2008-11-08 14:30:41 15.5 60.0 198.0 43.1
2008-12-30 10:49:32 4.5 59.7 187.0 55.3
2009-01-22 22:54:22 31.3 58.0 89.3 5.2
2009-03-28 04:36:08 34.2 81.2 152.8 22.5
2009-03-29 03:43:34 20.7 75.0 158.1 31.9
2009-05-19 02:19:52 42.5 64.6 25.8 77.3
2009-09-19 08:45:52 34.7 62.0 140.5 8.4
2010-01-08 07:17:31 19.5 57.5 175.6 37.2
2010-01-21 03:53:51 23.4 61.2 149.8 13.1
2010-02-22 07:10:34 14.5 63.5 165.7 42.0
2010-08-29 21:20:45 36.5 69.9 180.3 42.4
2010-08-30 20:50:45 20.0 93.3 98.3 69.7
2010-09-19 07:05:00 23.6 66.8 129.1 -30.6
2010-09-21 20:37:06 21.1 163.0 2.8 76.0
2011-01-05 00:56:23 9.3 67.4 110.0 -30.4
2011-02-28 16:16:26 39.3 137.6 35.5 -5.0
2011-04-17 20:20:29 34.2 74.7 153.7 12.9
2011-07-13 19:12:34 42.6 65.6 132.1 24.7
2011-07-22 22:15:41 11.6 62.2 204.5 64.6
2011-07-24 23:17:22 36.3 61.8 316.5 69.5
2011-07-28 15:21:08 19.6 89.0 147.0 -23.7
2011-08-28 21:14:19 31.6 63.3 215.6 53.2
: \[tab:events\] List of Telescope Array events with $E\ge 57$ EeV and zenith angle $\theta <45^\circ$ recorded from May 11, 2008 to September 15, 2011.
[^1]: We are grateful to T. Jarrett for providing us with the preliminary version of this catalog.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Total radiative thermal neutron-capture $\gamma$-ray cross sections for the $^{182,183,184,186}$W isotopes were measured using guided neutron beams from the Budapest Research Reactor to induce prompt and delayed $\gamma$ rays from elemental and isotopically-enriched tungsten targets. These cross sections were determined from the sum of measured $\gamma$-ray cross sections feeding the ground state from low-lying levels below a cutoff energy, E$_{\rm crit}$, where the level scheme is completely known, and continuum $\gamma$ rays from levels above E$_{\rm crit}$, calculated using the Monte Carlo statistical-decay code DICEBOX. The new cross sections determined in this work for the tungsten nuclides are: $\sigma_{0}(^{182}{\rm W}) = 20.5(14)$ b and $\sigma_{11/2^{+}}(^{183}{\rm W}^{m}, 5.2~{\rm s}) = 0.177(18)$ b; $\sigma_{0}(^{183}{\rm W}) = 9.37(38)$ b and $\sigma_{5^{-}}(^{184}{\rm W}^{m}, 8.33~\mu{\rm s}) = 0.0247(55)$ b; $\sigma_{0}(^{184}{\rm W}) = 1.43(10)$ b and $\sigma_{11/2^{+}}(^{185}{\rm W}^{m}, 1.67~{\rm min}) = 0.0062(16)$ b; and, $\sigma_{0}(^{186}{\rm W}) = 33.33(62)$ b and $\sigma_{9/2^{+}}(^{187}{\rm W}^{m}, 1.38~\mu{\rm s}) = 0.400(16)$ b. These results are consistent with earlier measurements in the literature. The $^{186}$W cross section was also independently confirmed from an activation measurement, following the decay of $^{187}$W, yielding values for $\sigma_{0}(^{186}{\rm W})$ that are consistent with our prompt $\gamma$-ray measurement. The cross-section measurements were found to be insensitive to choice of level density or photon strength model, and only weakly dependent on E$_{\rm crit}$. Total radiative-capture widths calculated with DICEBOX showed much greater model dependence, however, the recommended values could be reproduced with selected model choices. The decay schemes for all tungsten isotopes were improved in these analyses. We were also able to determine new neutron separation energies from our primary $\gamma$-ray measurements for the respective (n,$\gamma$) compounds: $^{183}$W ($S_{\rm n} = 6190.88(6)$ keV); $^{184}$W ($S_{\rm n} = 7411.11(13)$ keV); $^{185}$W ($S_{\rm n} = 5753.74(5)$ keV); and, $^{187}$W ($S_{\rm n} = 5466.62(7)$ keV).'
author:
- 'A. M. Hurst'
- 'R. B. Firestone'
- 'B. W. Sleaford'
- 'N. C. Summers'
- 'Zs. R[é]{}vay'
- 'L. Szentmikl[ó]{}si'
- 'M. S. Basunia'
- 'T. Belgya'
- 'J. E. Escher'
- 'M. Krti[č]{}ka'
bibliography:
- 'w\_prc\_2col\_r.bib'
title: '**An investigation of the tungsten isotopes via thermal neutron capture**'
---
\[sec:level1\]Introduction\
===========================
Neutron-capture decay-scheme data from the Reference Input Parameter Library (RIPL) [@capote:09] are required for nuclear-reaction calculations that are used to generate the Evaluated Nuclear Data File (ENDF) [@chadwick:11]. These data play a valuable role for both nuclear applications and basic research into the statistical properties of the nucleus including level densities and photon strengths. They also provide a wealth of structural information including discrete level spins and parities $J^{\pi}$ and $\gamma$-ray branching ratios. In addition, information on neutron-capture cross sections may also be obtained. Preliminary capture $\gamma$-ray cross sections were previously measured on natural elemental targets and published in the Evaluated Gamma-ray Activation File (EGAF) [@firestone:06]. For many elements only data for the isotopes with the largest cross sections and/or abundances could be obtained with natural targets. This paper describes a new campaign to improve the EGAF database by measuring isotopically-enriched targets.
Traditional methods for determining the total radiative thermal neutron-capture cross section, $\sigma_0$, include neutron-transmission and pile-oscillator measurements, both of which require precise knowledge of the neutron flux, and activation measurements which require an accurate decay-scheme normalization. Large corrections due to epithermal (1 eV to 10 keV), fast ($\gtrsim 10$ keV), and high-energy neutrons ($\gtrsim 1$ MeV) are typically necessary to determine thermal-capture cross sections. In this work we apply a newer method to determine the total radiative thermal neutron-capture cross sections for the tungsten isotopes using partial thermal neutron-capture $\gamma$-ray cross sections, $\sigma_{\gamma}$, measured with a guided thermal-neutron beam, combined with statistical-model calculations to account for unresolved continuum $\gamma$-rays, as described previously for the palladium [@krticka:08], potassium [@firestone:13], and gadolinium isotopes [@choi:13]. The prompt neutron-capture $\gamma$-rays were measured using both isotopically-enriched $^{182,183,186}$W targets and a natural elemental sample, to determine neutron-capture decay schemes for the compound tungsten nuclides $^{183,184,185,187}$W. This information was then used to normalize Monte Carlo simulations for the corresponding neutron-capture decay schemes calculated with the statistical-decay code [DICEBOX]{} [@becvar:98]. The neutron-capture $\gamma$-ray cross sections directly populating the ground state (GS) from low-lying levels were summed with the smaller, calculated, quasi-continuum contribution feeding the GS from higher levels to determine $\sigma_{0}$ for each tungsten isotope. Comparison of the simulated and experimental neutron-capture $\gamma$-ray cross sections populating and depopulating each excited state was also used to improve the tungsten decay schemes with the augmentation of more-complete data: determination of accurate $\gamma$-ray branchings, assessment of multipolarity and $\gamma$-ray mixing ratios ($\delta_{\gamma}$), placements of new $\gamma$-ray transitions, resolution of ambiguous (or tentative) energy-level and $J^{\pi}$ assignments, and neutron-separation energies ($S_{\rm n}$) determined from the observed primary $\gamma$-ray data for $^{183,184,185,187}$W. Also, as a validation of the current approach, the $\gamma$-decay emission probabilities, $P_{\gamma}$, were determined from the activation $\gamma$-ray cross sections corresponding to $^{187}$W $\beta^{-}$ decay. These measurements were found to be consistent with the adopted values, reported in the Evaluated Nuclear Structure Data File (ENSDF) [@ensdf], that are based on the work of Marnada *et al*. [@marnada:99].
Sample Mass \[mg\] $^{182}$W \[%\] $^{183}$W \[%\] $^{184}$W \[%\] $^{186}$W \[%\]
--------------- ------------- ----------------- ----------------- ----------------- -----------------
$^{\rm nat}$W 240 26.50(16) 14.31(4) 30.64(2) 28.43(19)
$^{182}$W 274 92.7(9) 2.0(3) 4.8(9) 0.5(1)
$^{183}$W 180 9.0(8) 74.9(3) 13.7(5) 2.4(3)
$^{186}$W 169 0.35(3) $\sim 0$ $\sim 0$ 99.65(3)
: \[tab:isocomp\] Isotopic composition of natural [@berglund:11] and enriched tungsten samples used in this work. The left-most column refers to the principal-enriched component in the sample.
\[sec:level2\]Experiment and Data Analysis\
===========================================
Isotopically-enriched stable and natural tungsten targets were irradiated with a supermirror-guided [*near-thermal*]{} neutron beam ($T \sim 120$ K; $E_{\rm beam} \sim 4.2$ meV) at the 10-MW Budapest Research Reactor [@rosta:97; @rosta:02]. The isotopic compositions of the enriched samples are shown in Table \[tab:isocomp\] and were determined by comparison with the ratios of peak intensities of strong, well-resolved transitions from the different tungsten isotopes in an elemental sample after accounting for their natural abundances. All enriched samples were oxide powders (WO$_{2}$) that were suspended in the evacuated neutron beam line in Teflon bags. During bombardment the thermal-neutron flux at the Prompt Gamma Activation Analysis (PGAA) target station was approximately $2.3 \times 10^{6}$ ${\rm n} \cdot {\rm cm}^{-2} \cdot {\rm s}^{-1}$. The PGAA facility is located $\sim 35$ m from the reactor wall in a low-background environment. The observed deexcitation $\gamma$ rays from the $^{A}$W(n,$\gamma$)$^{A+1}$W reactions were recorded in a single Compton-suppressed $n$-type high-purity germanium (HPGe) detector with a closed-end coaxial-type geometry, positioned $\sim 23.5$ cm from the target location. The PGAA facility is described in detail in Refs. [@revay:04; @szentmiklosi:10]. Energy and counting-efficiency calibrations of the HPGe detector were accomplished using standard radioactive and reaction sources covering an energy range from approximately $0.05-11$ MeV . The non-linearity and efficiency curves were generated using the $\gamma$-ray spectroscopy software package [HYPERMET-PC]{} [@hypermet], which was also used to perform peak-fitting analysis of the complex capture-$\gamma$ spectra.
Singles $\gamma$-ray data were collected in these (n,$\gamma$) measurements and peak areas for unresolved doublets, and higher-order multiplets, were divided based on branching ratios reported in the ENSDF [@ensdf]. Internal conversion coefficients for all transitions were calculated with the [BRICC]{} calculator, which is based on the Band Raman prescription [@kibedi:08].
\[sec:standard\]Standardization Procedure\
------------------------------------------
Partial neutron-capture $\gamma$-ray cross sections were derived from the measured peak intensities of the tungsten capture-$\gamma$ lines using an internal-standardization procedure where the observed $\gamma$-ray intensities are normalized by scaling to well-known comparator lines [@revay:03]. Here we used tungstic acid (H$_{2}$WO$_{4}$) for standardization [@szentmiklosi:pc] where hydrogen was used as the comparator with $\sigma_{\gamma}(2223~{\rm keV})=0.3326(7)$ b [@revay:03] with a stoichiometric $2:1$ H to W atomic ratio. The cross sections of the standardized tungsten transitions are listed in Table \[tab:tungstic\]. Cross sections for the more intense tungsten $\gamma$-ray transitions were measured with a natural elemental WO$_{2}$ target and then normalized to the standardized, strong, well-resolved cross sections from the standardization measurement using the well-known natural abundances [@berglund:11]. Weaker $\gamma$-ray transitions were measured in irradiations of enriched targets and similarly standardized. Since the tungsten isotopes and the calibration standard cross sections have a pure *1/v* dependence near thermal neutron energies i.e. increasing cross section with lower incident-neutron energy, no correction was necessary for the neutron-beam temperature.
Compound $E_{\gamma}$ \[keV\] $\sigma_{\gamma}$ \[b\]
----------- ---------------------- -------------------------
$^{187}$W 77.39(3) 0.234(4)
$^{187}$W 145.79(3) 1.344(13)
$^{187}$W 273.10(5) 0.380(4)
$^{187}$W 5261.68(6) 0.653(9)
$^{183}$W 6190.78(3) 0.726(10)
: \[tab:tungstic\] Elemental cross sections corresponding to strong lines observed in the tungsten compounds following an internal-standardization (n,$\gamma$) measurement with H$_{2}$WO$_{4}$ [@szentmiklosi:pc] comprising natural elemental tungsten.
\[sec:thickness\]Determination of the Effective Thickness\
----------------------------------------------------------
Since the WO$_{2}$ powders used in these measurements have a density of 10.8 g/cm$^{3}$, the intensity of low energy $\gamma$-rays must be corrected for self attenuation within the sample. To make this correction it is necessary to determine the [*effective*]{} sample thickness and calculate the intensity-attenuation coefficients as a function of $\gamma$-ray energy based on the prescription outlined in Ref. [@hubbel:95] using data from [XMUDAT]{} [@xmudat]. For irregular-shaped targets with non-uniform surfaces, such as the oxide powders used here, it is difficult to measure the sample thickness directly. Thus, to determine the effective WO$_2$ target thicknesses we compared the thin, lower-density (5.6 g/cm$^{3}$), attenuation-corrected tungstic acid target standardization-cross-section data, listed in Table \[tab:tungstic\], to the attenuated cross sections of these same transitions in the WO$_{2}$ targets. We then iteratively varied the sample thickness of the WO$_{2}$ targets until the calculated attenuation converged with the observed values for all transitions. An attenuation correction was then applied to all $\gamma$-rays in the spectrum.
\[sec:level3\]Statistical Model Calculations\
=============================================
The Monte Carlo statistical-decay code [DICEBOX]{} [@becvar:98] was used to simulate the thermal neutron-capture $\gamma$-ray cascade. [DICEBOX]{} assumes a generalization of the extreme statistical model, proposed by Bohr [@bohr:37] in the description of compound-nucleus formation and its subsequent decay. In thermal neutron capture the compound nucleus is formed with an excitation energy slightly above the neutron-separation energy threshold where particle evaporation is negligible. Within this theoretical framework, the [DICEBOX]{} calculation is constrained by the experimental decay scheme known up to a cut-off energy referred to as the critical energy, $E_{\rm crit}$, where all energies, spins and parities, and $\gamma$-ray deexcitations of the levels are regarded as complete and accurate. The code generates a random set of levels between $E_{\rm crit}$ and the neutron-separation energy according to an [*a priori*]{} assumed level density (LD) model $\rho(E,J^{\pi})$. Transitions to and from the quasi continuum to low-lying levels are then determined according to a choice of an [*a priori*]{} assumed photon strength function (PSF), $f^{(XL)}$, where $XL$ denotes the multipolarity of the transition. Selection rules are used to determine allowed transitions between all possible permutations of pairs of initial ($E_{i}$) and final ($E_{f}$) states given by $E_{\gamma} = E_{i} - E_{f}$. The partial radiation widths, $\Gamma_{if}^{XL}$, of the corresponding transition probabilities for non-forbidden transitions are assumed to follow a Porter-Thomas distribution [@porter:56], centered on a mean value according to the expression $$\langle \Gamma_{if}^{(XL)} \rangle = \frac{f^{(XL)}(E_{\gamma}) E_{\gamma}^{2L+1}}{\rho(E_{i}, J_{i}^{\pi_{i}})}.
\label{eq:2}$$ Internal conversion is accounted for using [BRICC]{} [@kibedi:08]. The corresponding simulated decay schemes are called [*nuclear realizations*]{}. Statistical fluctuations in the Porter-Thomas distributions are reflected in the variations between nuclear realizations and provide the uncertainty in the simulation inherent in the Porter-Thomas assumption. In these calculations we performed 50 separate nuclear realizations, with each realization comprising 100,000 capture-state $\gamma$-ray cascades.
The experimental $\gamma$-ray cross sections depopulating the low-lying levels below $E_{\rm crit}$, can then be used to renormalize the simulated population per neutron capture, from [DICEBOX]{}, to absolute cross sections feeding these levels. The total radiative thermal neutron-capture cross section $\sigma_{0}$ is determined as $$\sigma_{0}=\sum \sigma_{\gamma}^{\rm exp}({\rm GS})+\sum \sigma_{\gamma}^{\rm sim}({\rm GS}) = \frac{\sum \sigma_{\gamma}^{\rm exp}({\rm GS})}{1-P({\rm GS})},
\label{eq:3}$$ where $\sum \sigma_{\gamma}^{\rm exp}({\rm GS})$ represents the sum of experimental $\gamma$-ray cross sections feeding the ground state in direct single-step transitions, either via a primary GS transition or secondary transition from a level below $E_{\rm crit}$. The simulated contribution from the quasi continuum above $E_{\rm crit}$ feeding the ground state, $\sum \sigma_{\gamma}^{\rm sim}({\rm GS})$, may also be written as the product of $\sigma_{0}$ and the simulated ground-state population per neutron capture, $P({\rm GS})$, given by [DICEBOX]{} as shown in Equation \[eq:3\].
\[sec:level3.A\]Adopted Models\
===============================
Compound $T$ \[MeV\] $E_{0}$ \[MeV\] $a$ \[${\rm MeV}^{-1}$\] $E_{1}$ \[MeV\] $\Delta$ \[MeV\] $D_{0}$ \[eV\]
----------- ------------- ----------------- -------------------------- ----------------- ------------------ ----------------
$^{183}$W 0.55(2) $-0.92(17)$ 19.22(30) $-0.24(10)$ 0 59.9(61)
$^{184}$W 0.58(2) $-0.64(21)$ 18.76(30) 0.08(14) 0.763 12.0(10)
$^{185}$W 0.56(1) $-1.30(14)$ 19.45(28) $-0.50(8)$ 0 69.9(69)
$^{187}$W 0.57(2) $-1.63(22)$ 19.14(36) $-0.81(13)$ 0 84.8(79)
The simulated population of the levels below $E_{\rm crit}$ depends upon the assumed experimental decay scheme, the capture-state spin composition, $J = 1/2^{+}$ for even-even targets and $J = J_{\rm gs}(\rm target)\pm 1/2$ for odd-odd and odd-$A$ targets, and the choice of adopted phenomenological LD and PSF models.
\[sec:levelLD\]Level Densities\
-------------------------------
The constant temperature formula (CTF) [@gilbert:65] and the back-shifted Fermi gas (BSFG) [@newton:56; @gilbert:65] models were considered in this work. Both models embody a statistical procedure describing the increasing cumulative number density of levels $N(E)$ with increasing excitation energy such that, $$N(E) = \int \rho(E) d(E),
\label{eq:ld_insert1}$$ where $\rho(E)$ represents the level density at an excitation energy $E$. In the CTF model, a constant temperature is assumed over the entire range of nuclear excitation energy that may be explicitly stated as $$\rho(E,J) = \frac{f(J)}{T} \exp{\left( \frac{E-E_{0}}{T} \right)}.
\label{eq:ld_insert2}$$ The nuclear temperature $T$ may be interpreted as the critical temperature necessary for breaking nucleon pairs. The energy backshift related to proton- and neutron-pairing energies is given by $E_{0}$. The temperature and backshift-energy parametrizations used in this work are taken from von Egidy and Bucurescu [@vonegidy:05] and listed in Table \[tab:2\]. A spin-distribution factor $f(J)$ [@gilbert:65] is introduced in Equation \[eq:ld\_insert2\] and assumed to have the separable form of Ref. [@gilbert:65] $$f(J) = \frac{2J+1}{2 \sigma_{c}^{2}} \exp \left( - \frac{(J+1/2)^{2}}{2 \sigma_{c}^{2}} \right),
\label{eq:5}$$ where $\sigma_{c} = 0.98 \cdot A^{0.29}$ denotes the spin cut-off factor [@vonegidy:88].
The BSFG level density model is based on the assumption that the nucleus behaves like a fluid of fermions and may be written as $$\rho(E,J) = f(J) \frac{ \exp (2\sqrt{a(E-E_{1})})}{12\sqrt{2}\sigma_{c}a^{1/4}(E-E_{1})^{5/4} }.
\label{eq:4}$$ Here, the spin cut-off factor $\sigma_{c}$ is defined with an energy dependence given by $$\sigma_{c}^{2} = 0.0146 \cdot A^{5/3} \cdot \frac{ 1+\sqrt{1+4a(E-E_{1})} }{2a}.
\label{eq:6}$$ Since fermions exhibit a tendency to form pairs, the extra amount of energy required to separate them is accounted for by the introduction of the level density parameter, $E_{1}$, in Equation \[eq:4\], above. This parameter corresponds to the back-shift in excitation energy, while $a$ represents the shell-model level density parameter that varies approximately with $0.21\cdot A^{0.87}$ MeV$^{-1}$ [@dobaczewski:01]. As with the CTF, the adopted BSFG parameters used in this work have also been taken from von Egidy and Bucurescu [@vonegidy:05] and are presented in Table \[tab:2\]. In that work, the level density parameters were treated as adjustable and determined by fitting the functional forms of Equations \[eq:ld\_insert2\] and \[eq:4\], above, to experimentally-observed neutron resonance spacings in the region of the capture state above the neutron-separation energy.
\[sec:levelPSF\]Photon Strength Functions\
------------------------------------------
The dominant decay following thermal neutron capture is by $E1$ primary $\gamma$-ray transitions. The $E1$ photon strength is dominated by the low-energy tail of the giant dipole electric resonance (GDER). Theoretical models of the PSF describing the GDER are typically based on parametrizations of the corresponding giant resonance, observed in photonuclear reactions, whose transition probabilities are well described as a function of $\gamma$-ray energy [@krticka:08]. Total photonuclear cross-section data derived from $^{186}$W photoabsorption measurements [@berman:69] can be used to test the validity for a variety of PSFs near the GDER. These data [@berman:69] can be transformed to experimental PSF values $f^{(E1)}(E_{\gamma})$ using the empirical relationship of Ref. [@kopecky:87] $$f^{(E1)}(E_{\gamma}) = \frac{1}{3(\pi \hbar c)^{2}} \cdot \frac{\sigma_{\rm abs}}{E_{\gamma}},
\label{eq:insert2}$$ where the constant $\frac{1}{3(\pi \hbar c)^{2}} = 8.68 \times 10^{-8}$ ${\rm mb} \cdot {\rm MeV}^{-2}$, the photoabsorption cross section $\sigma_{\rm abs}$ is in units of \[mb\], and the $\gamma$-ray energy is in \[MeV\]. The results of this transformation for $^{186}$W are shown in Fig. \[fig:2\].
Isotope Resonance $E_{G_{1}}$ \[MeV\] $\Gamma_{G_{1}}$ \[MeV\] $\sigma_{G_{1}}$ \[mb\] $E_{G_{2}}$ \[MeV\] $\Gamma_{G_{2}}$ \[MeV\] $\sigma_{G_{2}}$ \[mb\]
----------- ----------- --------------------- -------------------------- ------------------------- --------------------- -------------------------- -------------------------
$^{183}$W GDER 12.68 2.71 268.0 14.68 3.62 395.0
GQER 11.10 3.91 4.55 $-$ $-$ $-$
$^{184}$W GDER 12.59 2.29 211.0 14.88 5.18 334.0
GQER 11.08 3.90 4.54 $-$ $-$ $-$
$^{185}$W GDER 12.68 2.71 268.0 14.68 3.62 395.0
GQER 11.06 3.89 4.53 $-$ $-$ $-$
$^{187}$W GDER 12.68 2.71 268.0 14.68 3.62 395.0
GQER 11.02 3.87 4.51 $-$ $-$ $-$
The Brink-Axel (BA) model [@brink:55; @axel:62] and the enhanced generalized Lorentzian (EGLO) model [@kopecky:91; @kopecky:93; @kopecky:98] were used in these calculations to compare with experimental data. The BA model is a form of the standard Lorentzian given by $$f^{(E1)}_{\rm BA}(E_{\gamma}) = \frac{1}{3(\pi \hbar c)^{2}} \cdot \sum_{i=1}^{i=2} \frac{\sigma_{G_{i}} E_{\gamma} \Gamma_{G_{i}}^{2}}{(E_{\gamma}^{2}-E_{G_{i}}^{2})^{2} + E_{\gamma}^{2} \Gamma_{G_{i}}^{2}}.
\label{eq:BA}$$ The resonance shape-driving parameters in Equation \[eq:BA\] are represented by the terms $E_{G_{i}}$ \[MeV\], the centroid of the GDER resonance, $\Gamma_{G_{i}}$ \[MeV\], the width of the resonance, and $\sigma_{G_{i}}$ \[mb\], the cross section of the resonance. The adopted experimental parametrizations for the tungsten isotopes were taken from RIPL [@capote:09] and are listed in Table \[tab:3\]. The corresponding BA PSF based on this parametrization is also shown in Fig. \[fig:2\] where it is compared to the experimental photoabsorption data. Although these data are only available above $E_{\gamma} \gtrsim 9$ MeV, they demonstrate excellent agreement with the Brink hypothesis [@brink:55] in this region.
The EGLO model is derived from the idea of the generalized Lorentzian (GLO) model and was originally proposed by Kopecky and Uhl [@kopecky:90], with the analytic form $$\begin{aligned}
f^{(E1)}_{\rm GLO} (E_{\gamma}, \Theta) &=& \sum_{i=1}^{i=2} \frac{\sigma_{G_{i}} \Gamma_{G_{i}}}{3 (\pi \hbar c)^{2}} \left[ F_{K} \frac{4 \pi^{2} \Theta^{2} \Gamma_{G_{i}}}{E_{G_{i}}^{5}} \right. \nonumber \\
&& \left.+\frac{E_{\gamma} \Gamma_{G_{i}}(E_{\gamma}, \Theta)}{(E^{2}_{\gamma}-E^{2}_{G_{i}})^{2} + E^{2}_{\gamma} \Gamma^{2}_{G_{i}}(E_{\gamma}, \Theta)} \right].
\label{eq:7}\end{aligned}$$ In this model a value of 0.7 has been used for the Fermi-liquid parameter $F_{K}$ [@kadmenski:83]. This factor, together with the remaining terms of the first quotient in the parentheses of Equation \[eq:7\], represent a correction to the Lorentzian function in describing the electric dipole operator in the limit of zero energy (as $E_{\gamma} \rightarrow 0$). This form of the PSF is a violation of the Brink hypothesis since there is an additional dependence on the nuclear temperature $\Theta$, which may be written as a function of excitation energy $$\Theta = \sqrt{(E_{\rm ex} - \Delta)/a},
\label{eq:8}$$ where $E_{\rm ex}$ is the excitation energy of a final state, and $\Delta$ is the pairing energy. The pairing correction has been determined according to the following convention: for even-even nuclei $\Delta = +0.5 \cdot |P_{d}|$ = 0.763 ($^{184}$W); for odd-$A$ nuclei $\Delta = 0$ ($^{183,185,187}$W); and for odd-odd nuclei $\Delta = -0.5 \cdot |P_{d}|$. The deuteron-pairing energy, $P_{d}$ is tabulated in Ref. [@vonegidy:05]. Consequently, GDERs built on excited states may differ vastly in both shape and size to those built on the ground state since the width of the resonance is also a function of the nuclear temperature according to $$\Gamma_{G_{i}} (E_{\gamma}, \Theta) = \frac{\Gamma_{G_{i}}}{E_{G_{i}}^{2}} (E_{\gamma}^{2} + 4 \pi^{2} \Theta^{2}).
\label{eq:9}$$ In the EGLO version of this model, the term $\Gamma_{G_{i}}(E_{\gamma}, \Theta)$ has been modified by an enhancement factor given by an empirical generalization of the width [@kopecky:91; @kopecky:93; @kopecky:98] $$\Gamma_{G_{i}}'(E_{\gamma},\Theta) = \left[ k_{0} + (1-k_{0}) \frac{(E_{\gamma} - E_{0})}{(E_{G_{i}} - E_{0})} \right] \Gamma_{G_{i}}(E_{\gamma},\Theta),
\label{eq:10}$$ where $\Gamma_{G_{i}}'(E_{\gamma}, \Theta)$ is substituted for $\Gamma_{G_{i}}(E_{\gamma}, \Theta)$ in Equation \[eq:7\] to evaluate $f^{(E1)}_{\rm EGLO}(E_{\gamma}, \Theta)$. A fixed value of $E_{0} = 4.5$ MeV has been adopted for the reference-energy [@kopecky:93; @kopecky:98] and is found to have only a weak influence on the overall enhancement. The parameter $k_{0}$ was then varied to optimize agreement with the absorption data of Ref. [@berman:69]. Figure \[fig:2\] shows that for $k_{0} = 3.5$ the EGLO PSF follows closely the experimental data for $E_{\gamma} \lesssim 17$ MeV. Beyond this regime the PSF is heavily damped, however, these $\gamma$-ray energies are not of interest in thermal capture. The GLO model is also plotted in Fig. \[fig:2\] along with an EGLO PSF using the empirically-determined value of $k_{0}$ from the mass-dependent model of Ref. [@kopecky:98] where $k_{0} = 1+ \left[(0.09(A-148) \cdot {\rm exp}(-0.180(A-148))\right]$. The plot illustrates very little difference in overall behavior between the GLO model and EGLO model with the mass-modeled-$k_{0}$ value. Both PSFs fail to reproduce the experimental data at low energy and can only adequately describe the data in the double-humped resonance region.
For the magnetic-dipole transitions, $M1$, a PSF based on the single-particle (SP) model was adopted. The value of $f^{(M1)}_{SP}$ was treated as an adjustable parameter in the [DICEBOX]{} calculations to obtain good agreement between statistical-model predictions and experimental-decay data in addition to the derived value of the total radiative capture width. For the even-odd $^{183,185,187}$W compounds a value of $f^{(M1)}_{SP} = 1 \times 10^{-9}$ MeV$^{-3}$ was used, while a higher value of $f^{(M1)}_{SP} = 3 \times 10^{-9}$ MeV$^{-3}$ was found to reproduce the data better for the even-even $^{184}$W. Other models, such as the scissors [@richter:90] and spin-flip models, were also be considered, however a lack of experimental evidence for a giant dipole magnetic resonance (GDMR) in the tungsten isotopes and the relative insignificance of these transitions in the calculations [@bolinger:70], make the SP model a practical approach.
A giant quadrupole electric resonance (GQER) model has been used to describe the PSF for $E2$ multipoles. This model is represented by a single-humped Lorentzian (cf. the standard Lorentzian in Equation \[eq:BA\]) to describe an isoscalar-isovector quadrupole-type vibration. A global parametrization has been used to determine the set of resonance parameters, listed in Table \[tab:3\]. The following convention was adopted in determining this parametrization: $E_{G} = 63 \cdot A^{-1/3}$ MeV [@speth:81], $\Gamma_{G} = 6.11 - 0.012A$ MeV [@prestwitch:84], and $\sigma_{G} = 1.5 \times 10^{-4} \cdot \frac{Z^{2}E_{G}^{2}A^{-1/3}}{\Gamma_{G}}$ mb [@prestwitch:84]. Quadrupole strength contributes far less than the dipole strengths. Transitions corresponding to higher multipoles, including $M2$, are not considered in modeling capture-state decay in this work.
\[sec:level4\]Results\
======================
Thermal neutron-capture (n,$\gamma$) $\gamma$-ray cross sections depopulating levels in the $^{183,184,185,187}$W compounds, from irradiations of the isotopically-enriched $^{182,183,186}$W targets and a natural tungsten target for $^{184}$W(n,$\gamma$), are discussed below. Only the primary $\gamma$ rays from the capture state or secondary $\gamma$ rays depopulating levels below $E_{\rm crit}$ are included in this paper. The complete decay scheme determined in these measurements will be available in the EGAF database.
$\sigma_{0}$ \[b\] Reference
-------------------- -------------------------------------
[**20.5(14)**]{} [**This work**]{}
19.2(19) H. Pomerance [@pomerance:52]
20.7(5) S. J. Friesenhahn [@friesenhahn:66]
19.6(3) K. Knopf [@knopf:87]
20.0(6) V. Bondarenko [@bondarenko:11]
19.9(3) Atlas [@mughabghab:06]
: \[tab:W182ng\] Summary of $\sigma_{0}$ measurements for $^{182}$W(n,$\gamma$).
All combinations of PSF and LD models described earlier, were used in the [DICEBOX]{} calculations and compared to experimental data by plotting the simulated population against the experimental depopulation for each level below $E_{\rm crit}$ in population-depopulation plots. For model combinations invoking the EGLO PSF we assumed a $k_{0} = 3.5$ enhancement factor. Uncertainties in the population along the vertical axis correspond to Porter-Thomas fluctuations from independent nuclear realizations, while those along the horizontal axis are due to the experimental uncertainty in the measured cross sections depopulating the levels. The vertical axis shows the calculated population per neutron capture to a given level, determined by [DICEBOX]{}, and the experimental depopulation of the corresponding level along the horizontal axis is normalized to the total radiative thermal-capture cross section according to $$P_{L}^{\rm exp} = \sum\limits_{i=1}^{N} \frac{\sigma_{\gamma_{i}}(1+\alpha_{i})}{\sigma_{0}},
\label{eq:14}$$ where $N$ denotes the number of $\gamma$ rays depopulating the level.
The population-depopulation plots compare the intensity balance through all states up to $E_{\rm crit}$. Scatter around the ${\rm population} = {\rm depopulation}$ line is a measure of the quality and completeness of the experimental data and provides a test of the ability of the statistical model to simulate the experimental decay scheme. Model dependence in the population-depopulation plot is indicated by either smooth or spin dependent deviations, and isolated deviations for individual levels are indications of problems with the experimental $J^{\pi}$ assignments or other decay-scheme data.
In this work, we also investigated the parity dependence $\pi (E)$ on the overall LD assuming its separable form $\rho (E,J,\pi) = \rho(E) \cdot f(J) \cdot \pi(E)$. The $\pi(E)$ dependence may be described by a Fermi-Dirac distribution parametrized according to Ref. [@quraishi:03]. In this framework, at large excitation energies $\pi(E) = 0.5$. As $E \rightarrow 0$: $\pi(E) \rightarrow 1$ for even-even nuclei; $\pi(E) \rightarrow 0(1)$ for odd-$A$ nuclei for which the odd nucleon is in an odd-parity (even-parity) orbit; and, $\pi(E) \approx 0.5$ for odd-odd and odd-$A$ nuclei if the Fermi level is occupied by nearly degenerate positive- and negative-parity orbits. Adopting an additional parity dependence in the LD models, $\rho (E,J,\pi) = \rho(E) \cdot f(J) \cdot \pi(E)$, the simulated populations for the odd-A isotopes $^{183,185,187}$W and even-even $^{184}$W were found to yield statistically consistent results with the parity-independent LD models, $\rho (E,J) = \rho(E) \cdot f(J)$; a representative comparison is illustrated in Fig. \[fig:LDPDEP\]. A parity-independent approach was, therefore, considered adequate for modeling the LD in these analyses.
\[sec:level4.A\]$^{182}$W(n,$\gamma$)$^{183}$W\
-----------------------------------------------
A $^{182}$WO$_2$ target was irradiated for a 2.46-h period. The current analysis and previous information in ENSDF [@firestone:92] implies that for $^{183}$W the level scheme is complete up to a level at 485.1 keV and we have set $E_{\rm crit} = 490.0$ keV, which includes an additional level over the value given in RIPL [@capote:09]. A total of 12 levels in $^{183}$W are below $E_{\rm crit}$ with spins ranging from $1/2 \leq J \leq 13/2$, deexcited by 33 $\gamma$ rays and fed by four primary $\gamma$ rays, shown in Table \[tab:183Wg\]. Transition intensities have been corrected for absorption in the source, as discussed earlier. The multipolarities in Table \[tab:183Wg\] are taken from ENSDF [@firestone:92] where available, or assumed based on angular-momentum selection rules, and the conversion coefficients were recalculated with [BRICC]{} [@kibedi:08].
Figure \[w183:PD\] shows the population-depopulation balance for $^{183}$W using the corresponding $\sigma_{\gamma}$ information from Table \[tab:183Wg\] calculated with various LD and PSF models. These plots show little statistical-model dependence in the population of most excited states except for the high-spin $11/2^{+}$, $11/2^{-}$, and $13/2^{+}$ states at 309.5, 475.2, and 485.1 keV, respectively, that appear to be better reproduced using the EGLO PSF. This is also shown in Fig. \[fig:deviation\] where the difference in the [DICEBOX]{}-modeled population ($P_{L}^{\rm sim}$) for a variety of PSF/LD combinations and the experimental depopulation ($P_{L}^{\rm exp}$) is model independent and insensitive to cut-off energies, $E_{\rm c}$, above 300 keV. Figure \[fig:deviation\] shows excellent consistency between the models at each value of $E_{\rm c}$.
The total-capture cross section, $\sigma_{0}$, determined for the different PSF/LD combinations, is also independent of $E_{\rm crit}$ for various model combinations as seen in Fig. \[fig:w183sigma0\]. For $E_{\rm crit} = 100$ keV, with only three low-lying levels, $\sigma_{0}$ remains nearly constant although the systematic uncertainty is larger. This rapid convergence is due to the dominant ground-state feeding from experimental transitions deexciting low-lying levels that dominates the calculation. We adopt the value $\sigma_{0} = 20.5(14)$ b corresponding to the EGLO/CTF combination. Of the $\sim 7$ % uncertainty on our value, the systematic uncertainty from the simulated cross section is 4.3 % and $\gamma$-ray self attenuation accounts for 3.2 %. The statistical and normalization errors are far less significant with each only contributing $\lesssim 2$ %. The result for the total radiative thermal-capture cross section for $^{182}$W(n,$\gamma$)$^{183}$W is consistent with the recommended value of 19.9(3) b [@mughabghab:06] and previous experimental investigations [@pomerance:52; @friesenhahn:66; @knopf:87; @bondarenko:11] listed in Table \[tab:W182ng\].
The choice of PSF and LD combination has a pronounced effect on the calculated capture-state total radiative width. The EGLO/CTF result, $\Gamma_{0} = 0.040(3)$ eV, agrees best with the recommended value of $\langle \Gamma_{0} \rangle = 0.051(4)$ eV. For the EGLO/BSFG and BA/CTF combinations somewhat poorer agreement is obtained with $\Gamma_{0}$ values of 0.071(3) and 0.076(6) eV respectively. The BA/BSFG combination gives much poorer agreement with $\Gamma_{0}=0.138(7)$ eV. Fortunately, the choice of PSF/LD model has only a small effect on the derived cross section.
The $11/2^{+}$ ($T_{1/2}=5.2$ s) isomer at 309.49 keV [@firestone:92] decays by a highly-converted 102.48-keV [@firestone:92] $M2$ transition that was not resolved from the 101.93-keV transition deexciting the 308.95-keV level and the 101.80-keV transition deexciting the 302.35-keV level in $^{187}$W which also contributes to the observed intensity due to a 0.5(1) % $^{186}$W impurity (Table \[tab:isocomp\]) in the measured sample. The total intensity of the triplet is $\sim 15(2)$ % of the 209.69-keV $\gamma$-ray intensity deexciting the 308.95-keV level, which is significantly larger than 7.4(4) % observed from the same level in $^{183}$Ta $\beta^{-}$ decay [@firestone:92]. Assuming the excess intensity, after the additional correction for the $^{186}$W impurity (see Section \[sec:level4.D\]), comes from the isomer transition, we get $\sigma_{\gamma}(102.48)=0.0049(19)$ b. Accounting for internal conversion this gives an experimental depopulation of 0.197(76) b which is consistent with the observed total $\gamma$-ray intensity feeding the metastable isomer, $\sigma_{11/2^{+}}(^{183}{\rm W}^{m}) = 0.177(18)$ b, from the 485.72- and 622.22-keV levels which are deexcited by transitions at 175.89 and 312.72 keV, respectively. The combined intensity of these transitions yields $\sum\sigma_{\gamma}^{\rm exp}(11/2^{+}) = 0.177(18)$ b and the DICEBOX-modeled population of the 309.49-keV isomer is $P(11/2^{+})=0.00154(97)$. The experimental depopulation of the 309.49-keV level is consistent with the simulated population from our [DICEBOX]{} calculations to within 3 $\sigma$ as indicated in the log-log space of Fig. \[w183:PD\]. The current measurement supports the proposed $J^{\pi} = 13/2^{+}$ assignment for the 485.72-keV level that was previously reported in reaction experiments [@saitoh:00]. Our simulations also support the inclusion of a new, highly-converted, 17.2-keV $E2$ transition deexciting the 308.95-keV level with a total intensity of $\sim 180$ mb feeding the 291.72-keV level that improves the agreement between population and depopulation for both levels. The 17.2-keV transition is below the detection threshold of our HPGe detector.
The next level above $E_{\rm crit}$ at 533 keV is reported in ENSDF [@firestone:92] with $J^{\pi} = (1/2,3/2)$. The 533-keV level was only reported as populated by primary $\gamma$-rays in a resonance (n,$\gamma$) experiment [@casten:73] and not seen in our work or later (n,$\gamma$) or reaction experiments. The existence of this level is considered doubtful; certainly the proposed $J^{\pi}$ assignment is highly questionable since these states are expected to be strongly populated in $s$-wave capture on $^{182}$W (see Fig. \[w183:PD\]). Raising the cut-off energy to 625-keV and including the next three levels at 551.1, 595.3, and 622.22 keV leads to poorer agreement in the population-depopulation balance for several levels as shown in Fig. \[w183:PD\](c). We observe the transitions from these three levels, but since the statistical model gives better agreement for $E_{\rm crit} = 490$ keV, it is likely that the decay-scheme information is incomplete between the 490 and 622.22 keV.
\[sec:level4.B\]$^{183}$W(n,$\gamma$)$^{184}$W\
-----------------------------------------------
$\sigma_{0}$ \[b\] Reference
---------------------- -------------------------------------
[**[9.37(38)]{}**]{} [**This work**]{}
10.9(11) H. Pomerance [@pomerance:52]
10.0(3) S. J. Friesenhahn [@friesenhahn:66]
10.5(2) K. Knopf [@knopf:87]
10.4(2) Atlas [@mughabghab:06]
: \[tab:183Wr\] Summary of $^{183}$W(n,$\gamma$) $\sigma_{0}$ measurements.
A $^{183}$WO$_2$ target was irradiated for 2.24 h. Comparison of the DICEBOX-population calculations with the experimental depopulation data for $^{184}$W sets $E_{\rm crit} = 1370.0$ keV. This value is higher than in RIPL where $E_{\rm crit} = 1252.2$ keV and includes 12 levels. There are 18 levels below our cut-off energy including one tentative level assignment. The $^{184}$W decay scheme consists of seven primary $\gamma$-rays and 47 secondary $^{184}$W $\gamma$-rays that are listed in Table \[tab:184Wg\]. The experimental multipolarities and mixing ratios are taken from ENSDF [@baglin:10] where available or assumed based on selection rules. The ground state of the $^{183}$W target nucleus is $J^{\pi} = 1/2^{-}$, allowing $s$-wave neutron capture to populate resonances with $J^{\pi} = 0^{-},1^{-}$. The *Atlas of Neutron Resonances* [@mughabghab:06] indicates that $1^{-}$ capture-states account for 78.3 % of the observed total-capture cross section, 7.4 % is from $0^{-}$ capture states, and the remaining 14.3 % of the cross section is attributed to a negative-parity [*bound*]{} resonance at $E_{0} = -26.58$ eV (with respect to the separation energy) with unknown spin.
The population-depopulation plots in Figs. \[w184:PD\](a) and (b) show that $\sigma_0$ is insensitive to both the $0^{-}/1^{-}$ composition of the capture state and the choice of PSF and LD combinations. Figure \[fig:sigma0Ec\_w184\] shows the dependence of the derived cross section on $E_{\rm c}$. For $E_{\rm c} \leq 900$ keV there are only four levels and $\sigma_0 = 8.65(64)$ b. Adding the level at 903.31 keV, which feeds the ground state with $\sigma_{\gamma} = 1.185(52)$ b, increases the derived cross section significantly, demonstrating the necessity to include as many experimentally known low-lying levels as possible in the simulation. For $E_{\rm crit}$ = 1370.0 keV, with a total of 17 levels (not including the tentative 1282.7-keV level, see later), we get $\sigma_{0} = 9.37(38)$ b, which is comparable at 2 $\sigma$ with the recommended value of 10.4(2) b [@mughabghab:06] and previous measurements shown in Table \[tab:183Wr\]. We also find that the total thermal-capture cross section is statistically insensitive to the $J^{\pi}$ composition of the capture state as illustrated in Fig. \[fig:s0\_varyCS\]. The overall uncertainty on our adopted value for $\sigma_{0}$ of 4.0 % is dominated by the 3.4 % systematic uncertainty in the simulation and the 1.7 % statistical uncertainty. Uncertainties due to $\gamma$-ray self attenuation and normalization are much lower, each contributing $ < 1.0$ %.
The capture-state width, $\Gamma_{0}$, is strongly dependent on the choice of PSF/LD combination, but is only weakly influenced by the capture-state spin composition, as shown in Fig. \[fig:Gamma0\_CS\]: $\Gamma_{0}$ is nearly constant up to $\sim 65$-% $0^{-}$ contribution, and only gradually increases up to $\sim 80~\%$. The EGLO/CTF model combination, with a 78.3-% $1^-$ capture-state composition (Fig. \[w184:PD\](b)), gives $\Gamma_0$=0.066(2) eV, in agreement with the adopted value of 0.073(6) eV [@mughabghab:06]. For the model combinations: EGLO/BSFG, $\Gamma_{0} = 0.129(3)$; BA/CTF, $\Gamma_{0} = 0.121(3)$; and BA/BSFG, $\Gamma_{0}=0.242(6)$; all are substantially higher than the adopted value. The effect of the capture-state composition is most sensitive to the modeled population of the $0^{+}$ and $J \geq 4$ low-lying levels. For $0^{-}$ capture-state compositions of 7.4 % (Fig. \[w184:PD\](a)) and 21.7 % (Fig. \[w184:PD\](b)), the EGLO results give excellent agreement with experiment. If the $0^{-}$ capture-state composition increases to 85 % (Fig. \[w184:PD\](c)), the predicted population of $0^{+}$ and high-spin states is much poorer. The 85-% $0^{-}$ composition also gives $\Gamma_{0}$ values of 0.348(8) for the EGLO/BSFG model combination and 0.178(5) eV for the EGLO/CTF combination that are considerably higher than the adopted value. To determine the most likely $J^{\pi}$ capture-state composition we varied this parameter and calculated the corresponding reduced $\chi^{2}$, using the population-depopulation data for the weakly populated states (circled in Fig. \[w184:PD\]), as $$\chi^{2}/{\rm ndf} = \sum \frac{(P_{L}^{\rm exp} - P_{L}^{\rm sim})^{2}}{(dP_{L}^{\rm sim})^{2}},
\label{eq:chi2_highJ}$$ where $P_{L}^{\rm exp}$ is the expectation value. Figure \[fig:chi2\_JPi\] shows that $\chi^{2}$ approaches 1.0 for capture-state compositions with $J^{\pi}(0^{-}) < 10$ %. Indeed, the simulated populations to these levels is more than $3~\sigma$ away from the expectation value assuming $J^{\pi}(0^{-}) \approx 22$ %. This result implies a likely capture-state composition $J^{\pi}(0^{-}) \lesssim 7$ %, and hence, $J^{\pi} = 1^{-}$ is the most probable assignment for the bound resonance at $-26.58$ eV [@mughabghab:06]. Thus, an overall fractional distribution of $J^{\pi} = 0^{-}(7.4~\%) + 1^{-}(92.6~\%)$ is consistent with the capture-state composition of Ref. [@mughabghab:06].
Our analysis confirms the decay scheme for $^{184}$W reported in ENSDF [@baglin:10] except for the 161.3-keV $\gamma$ ray depopulating the 1282.71-keV $(1,2)^{-}$ level, which we did not observe. This level assignment was tentative and the 161.3-keV $\gamma$-ray was placed twice in the level scheme (also depopulating the $6^{-}$ level at 1446.27 keV). Since this level is expected to be strongly populated, we conclude that it most likely does not exist (or has a considerably different $J^{\pi}$) and have removed it from our analysis. We have also assigned a new $\gamma$ ray at 65.36(19) keV, depopulating the 1360.38-keV level. Another 9.94-keV $\gamma$ ray depopulating the 1294.94-keV level is proposed based on the population-depopulation balance. The 504.03-keV $\gamma$ ray deexciting the 1252.20-keV $8^{+}$ level was not firmly identified although we can set an experimental limit of $\sigma_{\gamma} < 0.16$ mb which is consistent with statistical-model predictions of 0.1(1) mb.
Some $\gamma$ rays from levels below $E_{\rm crit}$ were not observed in our data and their relative cross sections were taken from ENSDF [@baglin:10], normalized to the cross sections of (observed) stronger transitions from those levels, as indicated in Table \[tab:184Wg\]. An unresolved doublet centered at 769 keV $\gamma$-ray deexcites the 1133.85- and 1775.34-keV levels and was resolved using the ENSDF-adopted branching intensities from both levels. Doublets centered around 215 and 996 keV, depopulating levels at 1221.31 and 1360.38 keV, respectively, were also resolved in a similar manner, as indicated in Table \[tab:184Wg\]. The 1285.00-keV level is an 8.33-$\mu$s isomer with $J^{\pi}=5^{-}$, and is populated with a cross section $\sigma_{5^{-}} = 24.7(55)$ mb from beneath $E_{\rm crit}$; transitions from above $E_{\rm crit}$ known to feed the isomer were not observed in this work.
\[sec:level4.C\]$^{184}$W(n,$\gamma$)$^{185}$W\
-----------------------------------------------
A $^{\rm nat}$WO$_{2}$ target was irradiated for 11.52 h. Comparison of the DICEBOX-population calculations with the experimental-depopulation data for the $^{185}$W compound sets $E_{\rm crit} = 392.0$ keV. This value is higher than in RIPL where $E_{\rm crit} = 243.4$ keV which includes eight levels. Table \[tab:185Wg\] lists 11 levels beneath the cut-off energy, deexcited by 25 secondary $\gamma$ rays, and populated by three primary $\gamma$ rays. These data were measured with a natural tungsten sample and supplemented with data from Bondarenko *et al*. [@bondarenko:05] that was renormalized to our cross sections. Ten levels below $E_{\rm crit}$ have negative parity with spins ranging from $1/2^{-}$ to $9/2^{-}$, and there are two positive-parity levels at 197.43 (11/2$^{+}$, $T_{1/2}=1.67$ min) [@wu:05] and 381.70 keV (13/2$^{+}$) [@bondarenko:05] that are high-spin with no $\gamma$ rays observed deexciting them. We have used the total cross section populating the 197.43-keV level from higher-lying levels in $^{185}$W from Ref. [@bondarenko:05], $\sigma_{11/2^{+}} = 6.2(16)$ mb, to determine the $\gamma$-ray cross sections deexciting this isomer. This cross section is substantially larger that than the recommended value, $\sigma_0 = 2(1)$ mb [@mughabghab:06]. The positive-parity levels below $E_{\rm crit}$ play only a small role in our simulations and do not limit the choice of $E_{\rm crit}$. The mixing ratios and multipolarities in Table \[tab:185Wg\] were taken from ENSDF [@wu:05] where available or assumed based on selection rules associated with the $\Delta J$ transitions.
We determined the thermal-capture cross section, $\sigma_{0} = 1.43(10)$ b, for $^{184}$W(n,$\gamma$). The result is largely insensitive with respect to PSF/LD combinations and comparable to the adopted value $\sigma_{0} = 1.7(1)$ b [@mughabghab:06]. Table \[tab:184Wr\] shows the comparison of our value with other reported measurements. For the EGLO/BSFG model combination, shown in Fig. \[fig:sigma0Ec\_w185\], $\sigma_{0}$ is statistically independent of $E_{\rm crit}$. The uncertainty in $\sigma_{0}$ is 7 %. Several low-energy $\gamma$ rays contribute significantly to $\sigma_{0}$ but were not observed by experiment and were, instead, estimated from statistical-model calculations. The systematic uncertainty in the ground-state feeding from the simulation is 4.7 %. A statistical uncertainty of 3.2 % and an uncertainty of 2.4 % in the normalization also contribute. The data from Ref. [@bondarenko:05] were measured with a very thin target so no correction due to $\gamma$-ray self attenuation was required.
The total radiative width of the capture state in $^{185}$W varies widely depending on the choice of PSF/LD models. The EGLO/BSFG combination generates a total width $\Gamma_{0} = 0.052(3)$ eV that is in excellent agreement with the adopted value, $\langle \Gamma_{0} \rangle = 0.052(4)$ eV [@mughabghab:06]. Other combinations show poorer agreement: $\Gamma_{0} = 0.034(3)$ eV for EGLO/CTF; $\Gamma_{0} = 0.069(6)$ eV for BA/CTF; and, $\Gamma_{0} = 0.108(7)$ eV for the BA/BSFG combination.
$\sigma_{0}$ \[b\] Reference
-------------------- -------------------------------------
[**1.43(10)**]{} [**This work**]{}
2.12(42) L. Seren [@seren:47]
1.97(30) H. Pomerance [@pomerance:52]
2.28(23) W. S. Lyon [@lyon:60]
1.70(10) S. J. Friesenhahn [@friesenhahn:66]
1.70(10) K. Knopf [@knopf:87]
1.76(9) V. Bondarenko [@bondarenko:05]
2.40(10) V. A. Anufriev [@anufriev:83]
1.70(10) Atlas [@mughabghab:06]
: \[tab:184Wr\] Summary of $^{184}$W(n,$\gamma$) $\sigma_{0}$ measurements.
Here we report more precise energies for the 301.13 and 332.11-keV levels than are in ENSDF [@wu:05]. No $\gamma$ rays were previously reported deexciting these levels. Our [DICEBOX]{} calculations support the results of Bondarenko *et al*. [@bondarenko:05] where six new $\gamma$ rays were identified depopulating these levels. Two new, low-energy $\gamma$ rays are proposed deexciting levels at 187.88 ($E_{\gamma} \approx 14$ keV) and 390.92 keV ($E_{\gamma} \approx 58$ keV) based on the population-depopulation intensity balance. The $\sim 58$-keV $\gamma$-ray transition is highly
[cccccccccccc]{}
\
$E_{\rm L}$ \[keV\] & $J^{\pi}$ & $E_{\gamma}$ \[keV\] & $\sigma_{\gamma}^{\rm exp}$ \[b\] & $\alpha$ & $XL$ & $E_{\rm L}$ \[keV\] & $J^{\pi}$ & $E_{\gamma}$ \[keV\] & $\sigma_{\gamma}^{\rm exp}$ \[b\] & $\alpha$ & $XL$\
\
$E_{\rm L}$ \[keV\] & $J^{\pi}$ & $E_{\gamma}$ \[keV\] & $\sigma_{\gamma}^{\rm exp}$ \[b\] & $\alpha$ & $XL$ & $E_{\rm L}$ \[keV\] & $J^{\pi}$ & $E_{\gamma}$ \[keV\] & $\sigma_{\gamma}^{\rm exp}$ \[b\] & $\alpha$ & $XL$\
0 & $3/2^{-}$ & & & & & (493.4)& $(9/2^{-})$& (143.2(1))& 0.0222(58)& 1.7 & $[M1]$\
77.29 & $5/2^{-}$ & 77.30(5) & 0.823(14) & 10.17 & $M1+E2$ & 510.00 & $11/2^{-}$& (145.8(1))& 0.0052(21) & 1.62 & $[M1]$\
145.85 & $1/2^{-}$ & 145.84(5) & 4.727(46) & 1.65 & $M1$ & 522.15 & $9/2^{-}$& 171.70(6) & 0.0526(32) & 0.71 & $[M1+E2]$\
201.45 & $7/2^{-}$ & 124.18(5) & 0.282(16) & 2.01 & $M1+E2$ & 538.45 & $11/2^{-}$& 337.18(19) & 0.0096(18) & 0.0604 &$[E2]$\
& & 201.51(5) & 1.515(76) & 0.303 & $[E2]$ & 574.05 & $11/2^{-}$& 209.59(33) & 0.0042(16) & 0.59 & $[M1]$\
204.90 & $3/2^{-}$ & 59.30(5) & 1.048(43) & 3.73 & $M1$ & 597.24 & $11/2^{+}$& - & - & - &\
& & 127.55(5) & 0.646(37) & 1.87 & $M1+E2$ & 613.38 & $9/2^{-}$& (16.20(13))& (0.00293(43)) & 10.08 & $[E1]$\
& & 204.87(5) & 0.666(33) & 0.631 & $[M1]$ & & & 282.86(19) & 0.0054(18) & 0.259 & $[M1]$\
303.35 & $5/2^{-}$ & 98.51(8) & 0.0261(27) & 4.97 & $[M1]$ & & & 310.52(12) & 0.0119(17) & 0.0771 & $[E2]$\
& & 101.80(5) & 0.234(16) & 4.61 & $M1$ & & & 410.8(5)& 0.00031(5) & 0.0944 & $[M1]$\
& & 157.47(5) & 0.1474(81) & 0.713 & $[E2]$ & 640.49 & $5/2^{-}$ & 276.19(6) & 0.0635(48) & 0.109 & $[E2]$\
& & 226.02(5) & 0.379(19) & 0.243 & $M1+E2$ & & & 289.98(6) & 0.300(15) & 0.17 & $M1+E2$\
& & 303.31(6) & 0.248(13) & 0.213 & $[M1]$ & & & 438.91(10) & 0.0174(24) & 0.0794 & $[M1]$\
330.78 & $9/2^{-}$& 129.1(2)& 0.0051(38) & 2.34 & $[M1]$ & & & 563.33(13) & 0.0321(49) & 0.0415 & $[M1]$\
& & 253.51(5)& 0.1268(92) & 0.143 & $[E2]$ & & & 640.55(10) & 0.085(10) & 0.0298 & $[M1]$\
350.43 & $7/2^{-}$ & (19.60(5))& 0.00051(18) & 97.66 & $[M1]$& 710.78 & $13/2^{-}$& (380.0(2))& (0.00030(30)) & 0.0431 & $[E2]$\
& & 148.89(5) & 0.204(11) & 1.55 & $[M1]$ & 727.86 & $11/2^{-}$& 205.7(1)& 0.0016(4) & 0.6166 & $[M1]$\
& & 273.12(5) & 1.337(14) & 0.283 & $[M1]$ & & & 377.0(2)& 0.0014(2) & 0.0444 & $[E2]$\
& & 350.34(9) & 0.0219(25) & 0.0542 & $[E2]$ & 741.08 & $7/2^{+}$& 218.81(7) & 0.0220(27) & 0.0503 & $[E1]$\
364.22 & $9/2^{-}$ & (13.80(4))& 0.00293(22) & 275.2 & $[M1]$& & & 330.97(6) & 0.0775(45) & 0.1682 & $[M1]$\
& & 162.59(12) & 0.0100(15) & 1.2 & $[M1]$ & & & 376.80(5)& 0.184(21) & 0.0134 & $[E1]$\
& & 286.79(7) & 0.0314(26) & 0.0981 & $[E2]$ & & & 390.56(10) & 0.0661(42) & 0.0123 & $[E1]$\
410.06 & $9/2^{+}$& 45.8(3)& 0.301(12) & 0.5941 & $[E1]$ & & & 539.58(14)& 0.0092(21) & 0.00605 &$[E1]$\
432.28 & $7/2^{-}$ & 128.93(6)& 0.1064(82) & 2.34 & $[M1]$ & & & 663.91(8) & 0.0764(62) & 0.00394 & $[E1]$\
& & 227.37(10) & 0.0506(37) & 0.203 & $[E2]$ & 762.15 & $1/2^{-}$& 557.24(5)& 0.572(33)& 0.0427 &$(M1+E2)$\
& & 230.56(14) & 0.0148(33) & 0.453 & $[M1]$ & & & 616.33(5) & 0.304(16) & 0.0329 & $[M1]$\
& & 354.92(7) & 0.1814(95) & 0.14 & $[M1]$ & & & 762.0(5)& 0.0286(60) & 0.0191 & $[M1]$\
& & 432.4(5)& 0.0098(28) & 0.0305 & $[E2]$ & 775.60 & $7/2^{-}$& (135.1(5))& 0.0478(71) & 2.01 & $[M1]$\
& & 253.50(16)& 0.0205(47) & 0.349 & $[M1]$ & & & 659.18(9)& 0.0738(91) & 0.0109 & $[E2]$\
& & 411.28(9) & 0.0164(23) & 0.0944 & $[M1]$ & & & 783.74(13) & 0.0836(82) & 0.0179 & $[M1]$\
782.29 & $1/2^{-}$ & 577.36(5)& 0.921(46) & 0.031 &$(M1+E2)$ & & & 860.77(12) & 0.1058(82) & 0.0141 & $[M1]$\
& & 636.64(35)& 0.0396(42) & 0.0303 & $[M1]$& 863.29 & $5/2^{-}$& 513.0(5)& 0.024(10) & 0.0531 & $[M1]$\
& & 704.9(4)& 0.0138(20) & 0.0094 & $[E2]$ & & & 532.41(7)& 0.039(12) & 0.0180 & $[E2]$\
& & 782.25(5) & 0.606(31) & 0.0179 & $[M1]$ & & & 559.79(9) & 0.0283(32) & 0.0423 & $[M1]$\
797.03 & $11/2^{-}$& 364.7(1)& 0.0013(5) & 0.0482 & $[E2]$ & & & 658.0(3)& 0.0168(84) & 0.0279 & $[M1]$\
& & 466.3(1)& 0.0017(5) & 0.0680 & $[M1]$ & & & 661.9(3)& 0.038(15) & 0.0275 & $[M1]$\
798.22 & $(9/2^{+})$ & - & - & - & - & & & 717.36(14) & 0.0300(64) & 0.00905 & $[E2]$\
803.37 & $3/2^{-}$& 500.02(6)& 0.115(17) & 0.0565 & $(M1)$ & & & 785.73(11) & 0.0850(81) & 0.0177 & $[M1]$\
& & 598.55(15)& 0.0608(88) & 0.0355 & $[M1]$ & & & 862.96(10)& 0.099(11) & 0.014 & $[M1]$\
& & 657.50(7)& 0.320(32) & 0.0279 & $[M1]$ & 866.68 & $3/2^{-}$& 563.51(6)& 0.023(13) & 0.0157 & $[E2]$\
& & 726.03(5) & 0.1118(74) & 0.0216 & $[M1]$ & & & 661.65(7)& 0.068(24) & 0.0275 & $[M1]$\
& & 803.25(8)& 0.1043(70) & 0.0168 & $[M1]$ & & & 789.38(10) & 0.234(52) & 0.00735 & $[E2]$\
809.79 & $(13/2^{-})$ & - & - & - & - & & & 866.37(13) & 0.278(16) & 0.0139 & $[M1]$\
811.7 & $(15/2^{+})$ & - & - & - & - & 881.77 & $5/2^{+}$& 140.47(13) & 0.0260(55) & 1.82 & $[M1]$\
815.51 & $13/2^{+}$& - & - & - & - & & & 449.58(11)& 0.0086(43) & 0.00899 & $[E1]$\
816.26 & $3/2^{-}$ & 176.6(6)& 0.0087(46) & 0.9436 & $[M1]$ & & & 531.29(5)& 0.201(26) & 0.00624 & $[E1]$\
& & 383.87(8) & 0.0217(22) & 0.0422 &$[E2]$& & & 676.79(8) & 0.0475(50) & 0.00379 & $[E1]$\
& & 465.54(8)& 0.0464(34) & 0.0252 & $[E2]$ & & & 679.97(14)& 0.0105(45) & 0.00375 &$[E1]$\
& & 512.52(14)& 0.065(13) & 0.0531 &$[M1]$& & & 803.7(4)& 0.0171(15) & 0.0027 & $[E1]$\
& & 611.34(5)& 0.167(21) & 0.0336 & $(M1)$ & & & 881.58(6) & 0.214(12) & 0.00226 & $[E1]$\
& & 670.37(5) & 0.227(12) & 0.0265 & $[M1]$ & 884.13 & $(5/2^{+})$& 143.15(6)& 0.0414(36) & 1.71 & $[M1]$\
& & 738.84(6) & 0.185(10) & 0.0208 & $[M1]$ & & & 243.63(37)& 0.00296(15) & 0.0381 & $[E1]$\
& & 816.20(20) & 0.436(67) & 0.0161 & $[M1]$ & & & 451.29(19)& 0.0065(21) & 0.0089 & $[E1]$\
840.21 & $1/2^{-}$& 537.21(23) & 0.0114(41) & 0.0176 &$[E2]$& & & 474.02(6) & 0.296(15) & 0.0240 & $[E2]$\
& & 635.37(8)& 0.1059(86) & 0.0304 & $[M1]$ & & & 533.63(6) & 0.0934(64) & 0.00619 & $[E1]$\
& & 694.33(5) & 0.235(13) & 0.0243 & $[M1]$ & 891.93 & $3/2^{-}$& 460.1(8)& 0.0069(25) & 0.0259 & $[E2]$\
& & 762.82(7) & 0.172(14) & 0.00792 & $[E2]$ & & & 541.46(7) & 0.0848(64) & 0.0173 & $[E2]$\
& & 840.17(5) & 0.662(34) & 0.015 & $[M1]$ & & & 588.55(6) & 0.0971(62) & 0.0371 & $[M1]$\
852.41 & $3/2^{-}$ & 502.0(6)& 0.0137(60) & 0.0209 & $[E2]$ & & & 690.15(16)& 0.0082(41) & 0.00985 & $[E2]$\
& & 549.0(5)& 0.0195(80) & 0.0443 & $[M1]$& & & 745.88(5) & 0.236(13) & 0.0203 & $[M1]$\
& & 647.41(8) & 0.1065(73) & 0.029 & $[M1]$ & & & 814.03(19)& 0.122(10) & 0.0162 & $[M1]$\
& & 650.88(14) & 0.0212(41) & 0.0113 & $[E2]$ & & & 891.89(5)& 0.408(22) & 0.0129 & $[M1]$\
& & 706.59(6)& 0.195(16) & 0.0232 & $[M1]$ & 5466.62 & $1/2^{+}$ & 4574.67(7) & 0.397(21) & 0 & $[E1]$\
& & 774.92(6)& 0.128(13) & 0.0184 & $[M1]$ & & & 4585.7(6)& 0.0052(20) & 0 & $[E2]$\
& & 852.18(6) & 0.160(11) & 0.0144 & $[M1]$ & & & 4602.6(15)& 0.024(12) & 0 & $[E1]$\
860.76 & $3/2^{-}$& 428.48(8) & 0.0701(48) & 0.0313 & $[E2]$ & & & 4606.6(11)& 0.0159(60) & 0 & $[E1]$\
& & 655.87(7) & 0.227(14) & 0.0281 & $[M1]$ & & & 4615.3(7)& 0.0052(12) & 0 & $[E1]$\
& & 4626.40(7) & 0.627(33) & 0 & $[E1]$ & & & 5163.5(4)& 0.0135(20) & 0 & $[M2]$\
& & 4650.27(8) & 0.207(12) & 0 & $[E1]$ & & & 5388.85(26)& 0.0143(12) & 0 & $[M2]$\
& & 4662.94(27) & 0.0197(30) & 0 & $[E1]$ & & & 5261.67(9) & 2.297(32) & 0 & $[E1]$\
& & 4684.31(7) & 0.765(40) & 0 & $[E1]$ & & & 5466.47(12) & 0.0675(50) & 0 & $[E1]$\
& & 4704.8(4)& 0.0091(12) & 0 & $[E1]$ & & & 5320.65(8) & 1.625(83) & 0 & $[E1]$\
& & 4826.0(10)& 0.0048(12) & 0 & $[M2]$ & & & & & &\
converted and obscured by a strong tungsten X ray at 57.98 keV, making a $\gamma$ ray of this energy difficult to observe. Both new transitions were assumed to have $M1$ multipolarity. The improvement by including these transitions is shown in Fig. \[w185:PD\]. The $^{185}$W $\gamma$ rays deexciting the first three excited states at 23.55, 65.85, and 93.30 keV were not observed in either this work or that of Bondarenko *et al*. [@bondarenko:05]. The transition cross sections depopulating these levels were determined from the simulated cross section populating those levels, using the EGLO/BSFG model combination and the branching ratios from ENSDF [@wu:05], as shown in Fig. \[w185:PD\]. Our [DICEBOX]{}-simulated population per neutron capture to each of these levels is: 23.55 keV, 0.178(28); 65.85 keV, 0.254(33); and, 93.30 keV, 0.201(32). These values can be compared to those of Bondarenko *et al*. [@bondarenko:05]: 23.55 keV, 0.168(16); 65.85 keV, 0.126(14); and, 93.30 keV, 0.201(17). The difference between simulation and Ref. [@bondarenko:05] for the 65.85-keV level implies there is a substantial contribution from the quasi continuum that is not observed experimentally. Four levels were previously reported with tentative $J^{\pi}$ assignments [@wu:05]. For three of these levels, our simulations are consistent with the assignments of $9/2^{-}$, $7/2^{-}$, and $9/2^{-}$ to the 301.13-, 332.11-, and 390.4-keV levels, respectively. The agreement between modeled population and experimental depopulation by assuming these $J^{\pi}$-level assignments is illustrated in the population-depopulation plot of Fig. \[w185:PD\](b). Those assignments are also consistent with the distorted-wave Born approximation (DWBA) calculations described in Ref. [@bondarenko:05].
\[sec:level4.D\]$^{186}$W(n,$\gamma$)$^{187}$W\
-----------------------------------------------
A $^{186}$WO$_2$ target was irradiated for 2.03 h. Comparison of the DICEBOX-population calculations with the experimental-depopulation data for $^{187}$W sets $E_{\rm crit} = 900.0$ keV. This value is substantially higher than in RIPL where $E_{\rm crit} = 145.9$ keV and includes only three levels. Table \[tab:187Wg\] lists 40 levels below $E_{\rm crit} = 900.0$ keV, deexcited by 121 secondary $\gamma$ rays and populated by 16 primary $\gamma$ rays, with a range of spins from $1/2 \leq J \leq 15/2$. The capture state has $J^{\pi} = 1/2^{+}$. Multipolarities and mixing ratios are taken from ENSDF [@basunia:09] where available or assumed according to $\Delta J$ and $\Delta \pi $ selection rules.
As was the case for the other tungsten isotopes investigated in this study, $\Gamma_{0}$ shows a strong dependence on PSF/LD. The EGLO/BSFG models give $\Gamma_{0} = 0.058(3)$ eV, which compares well with the adopted value of $\langle \Gamma_{0} \rangle = 0.051(5)$ eV [@mughabghab:06]. For the EGLO/CTF combination $\Gamma_{0} = 0.038(2)$ eV, BA/CTF gives $\Gamma_{0} = 0.083(6)$ eV, and BA/BSFG gives $\Gamma_{0} = 0.127(7)$ eV.
A total thermal-capture cross section $\sigma_{0} = 33.33(62)$ b was determined for the $^{186}$W(n,$\gamma$) reaction. Figure \[fig:sigma0Ec\_w187\] shows the stability of this value with increasing cut-off energy, where $\sigma_{0}$ is nearly insensitive to $E_{\rm crit}$ even when as few as three levels are included. For three levels and $E_{\rm crit} = 200$ keV ,we get $\sigma_{0} = 34.7(32)$ b. Adopting $E_{\rm crit} = 900$ keV, with 40 levels in the decay scheme, $\sigma_{0}$ barely changes although the uncertainty is reduced by a factor of five. The overall uncertainty of 1.9 % is dominated by a 1.7 % uncertainty in the simulated cross section with all other errors contributing less than 1 %. In Table \[tab:187Wr\] we compare our result with other measurements in the literature and the value adopted by Mughabghab of $\sigma_{0} = 38.1(5)$ b [@mughabghab:06]. That value was based on an older activation decay-scheme normalization. The literature values in Table \[tab:187Wr\] have been corrected for the decay-scheme normalization from our activation measurement, described in Section \[sec:level4.E\], where possible.
$\sigma_{0}$ \[b\] Reference
-------------------- ----------------------------------------
[**33.33(62)**]{} [**This work (prompt)**]{}
34.2(70) L. Seren [@seren:47]
34.1(27) H. Pomerance [@pomerance:52]
41.3, 51 W. S. Lyon [@lyon:60]
33 J. H. Gillette [@gillette:66]
37.8(12) S. J. Friesenhahn [@friesenhahn:66]
35.4(8) P. P. Damle [@damle:67]
40.0(15) C. H. Hogg [@hogg:70]
33.6(16) G. Gleason [@gleason:77; @exfor]
33.3(11) R. E. Heft [@heft:78]
37.0(30) V. A. Anufriev [@anufriev:81]
38.5(8) K. Knopf [@knopf:87]
34.8(3) M. R. Beitins [@beitins:92]
34.7(15), 37.9(20) S. I. Kafala [@kafala:97]
32.7(10) N. Marnada [@marnada:99]
32.8(10) F. De Corte [@decorte:03-2]
30.6(19) M. Karadag [@karadag:04]
33.4(11) L. Szentmikl[ó]{}si [@szentmiklosi:06]
35.9(11) V. Bondarenko [@bondarenko:08]
38.7(23) M. S. Uddin [@uddin:08]
28.9(18) N. Van Do [@vando:08]
29.8(32) A. El Abd [@elabd:10]
38.1(5) Atlas [@mughabghab:06]
: \[tab:187Wr\] Summary of $^{186}$W(n,$\gamma$) $\sigma_{0}$ measurements.
Figure \[w187:PD\](a) shows excellent agreement between modeled population and experimental depopulation data for all levels except the 364.22-keV level. This level was reported in ENSDF to be deexcited by 162.7- and 286.9-keV $\gamma$ rays [@basunia:09]. The [DICEBOX]{}-simulated population is much larger than the experimentally observed depopulation of this level. Since the experimental data for all other levels compares well with their modeled populations over a range of five orders of magnitude, it is evident that the statistical model is an accurate simulation tool for the $^{187}$W capture-$\gamma$ decay scheme and discrepancies with the experimental intensity suggest incomplete experimental level or transition data. The $J^{\pi} = 9/2^{-}$ assignment is firmly established for this level [@basunia:09], so new $\gamma$ rays depopulating the 364.22-keV level were sought. In Fig. \[w187:PD\](b) we show that including a $\sim14$-keV transition populating the 350.43-keV level considerably improves agreement between experiment and theory. An additional low-energy $\gamma$ ray at 19.6 keV depopulating the 350.43-keV level is also suggested based on the statistical-model calculation. These newly proposed $\gamma$-ray transitions were also inferred from the coincidence data of Bondarenko *et al*. [@bondarenko:08].
In an earlier ENSDF evaluation of $^{187}$W [@firestone:91] two additional levels were reported at 493.41 and 551 keV that were removed in the latest evaluation [@basunia:09]. We see tentative evidence for the 143.2-keV $\gamma$ ray depopulating the 493.41-keV level. The statistical model simulations imply a $J^{\pi} = 9/2^{-}$ assignment for this state. There is insufficient evidence to support a level at around 551 keV, although there is a strong transition at 551.6 keV in the prompt capture-$\gamma$ spectrum. This transition is also present in the delayed $^{187}{\rm W} \rightarrow ^{187}{\rm Re} + \beta^{-}$ beta-decay spectrum and can be attributed to the decay of $^{187}$Re. We propose an additional 135.1-keV $\gamma$ ray depopulating the 775.60-keV level from the observed spectrum and consistency with statistical-model predictions. An additional low-energy transition at 16.20 keV, with likely $E1$ multipolarity, is proposed to depopulate the 613.38-keV level based on statistical-model calculations. The statistical model has also been used to estimate the intensity of the known 380.0-keV transition depopulating the $13/2^{-}$ level at 710.78-keV. A doublet centered on 380.22 keV is observed in our data and we have resolved the intensity of the known 380.0-keV component by determining the intensity limit consistent with model predictions for a transition decaying out of this high-spin state.
The statistical-model simulations were also used to test uncertain $J^{\pi}$ assignments for levels in $^{187}$W. The majority of the tentative $J^{\pi}$ assignments, for energy levels beneath $E_{\rm crit}$, were found to be consistent with the current ENSDF assignments, and 19 $J^{\pi}$ assignments for $^{187}$W [@basunia:09] could be confirmed in our analysis (see Table \[tab:187Wg\]). A recent investigation of the $J^{\pi}$ assignments in $^{187}$W using polarized deuterons incident upon a natural tungsten foil to measure the (d,p) reaction [@bondarenko:08] compared the observed particle angular distribution with DWBA calculations and determined $J$ and $l$-transfer values utilizing the [CHUCK3]{} code [@kunz:CHUCK3]. Our results are consistent with most of the $J^{\pi}$ assignments from (d,p) analysis except for an excited state at 884.13 keV. The (d,p) analysis suggests a value of $J^{\pi} = 7/2^{+}$ for this state, but we find that $J^{\pi} = 5/2^{+}$ is in agreement with our (n,$\gamma$) data, as illustrated in the population-depopulation plots in Fig. \[w187:PD\_2\]. The 884.13-keV state decays by a 474.02-keV transition, an assumed $E2$ quadrupole, to the 1.38-$\mu$s isomer at 410.06 keV, implying a likely $J^{\pi}=9/2^{+}$ assignment for this bandhead. Consequently, all other members of the rotational sequence built on this level will have spin values increased by one unit of angular momentum, as shown in Fig. \[w187:PD\_2\]. The previous $J^{\pi} = (11/2^{+})$ [@basunia:09] assignment for the 410.06-keV isomer was based on the systematics of neighboring odd-$A$ tungsten isotopes. Since only a few DWBA fits have been published, it would be instructive to see how well DWBA calculations for the lower-spin sequence would compare with the (d,p) data, as the shapes of experimental angular distributions are often well described by more than one set calculations, especially where counting statistics may be poor.
$E_{\gamma}$ \[keV\] $\sigma_{\gamma}^{(P)}$ $\sigma_{\gamma}^{(D)}$ $P_{\gamma}$ $P_{\gamma}$ $\sigma_{\gamma}$ $\sigma_{0}$ $\sigma_{\gamma}$ $\sigma_{0}$
---------------------- ------------------------- ------------------------- -------------- -------------- ------------------- -------------- ------------------- --------------
134.34(7) 3.60(12) 3.66(12) 0.110(4) 0.104(2) 3.65(7) 33.2(14) 3.50(2) 31.9(12)
479.47(5) 9.55(16) 9.65(22) 0.289(9) 0.266(4) 9.29(14) 32.1(11) 9.19(9) 31.7(10)
551.22(9) 2.16(19) 2.20(4) 0.0661(17) 0.0614(10) 2.16(4) 32.6(10) 2.14(1) 32.37(85)
617.96(6) 3.12(11) 2.54(5) 0.0762(21) 0.0757(12) 2.66(5) 35.0(11) 2.68(1) 35.18(98)
625.03(10) 0.35(11) 0.419(19) 0.0126(6) 0.0131(2) 0.47(1) 37.2(20) - -
685.74(5) 11.85(21) 11.74(20) 0.352(9) 0.332(5) 11.78(21) 33.5(10) 11.48(6) 32.60(84)
772.99(10) 1.606(95) 1.771(57) 0.053(2) 0.0502(8) 1.75(3) 33.0(13) 1.74(1) 32.8(12)
We did not observe the 45.8(3) keV, presumed $E1$ transition [@basunia:09], deexciting 410.06-keV 1.38-$\mu$s isomer, that was reported by Bondarenko [*et al*]{}. [@bondarenko:08] on the basis of delayed coincidences with the 474.02-keV $\gamma$-ray deexciting the 884.13-keV level. Bondarenko [*et al*]{}. also postulated a second, $\sim 59$-keV transition, based on delayed coincidences with $\gamma$ rays deexciting the 350.43-keV, $7/2^{-}$ level. This transition is of the same energy as the strong tungsten $K_{\alpha_{1}}$ X rays that obscure it in the spectrum. Bondarenko [*et al*]{}. speculated the existence of the 59-keV $\gamma$-ray as unlikely since it required an $M2$ multipolarity assuming an $11/2^{+}$ assignment for the 410.06-keV level. Our new $J^{\pi} = 9/2^{+}$ assignment for the 410.06-keV level implies an acceptable $E1$ transition for this 59-keV $\gamma$ ray. However, the existence of the 59-keV $\gamma$-ray still remains in doubt since the proposed 13.80-keV transition deexciting the 364.22-keV level would also explain the coincidence results. We observed two $\gamma$-rays populating the 410.06-keV isomer from higher levels below $E_{\rm crit}$. The experimental intensity feeding the isomer, $\sum \sigma_{\gamma}^{\rm exp}(9/2^{+};410.06~{\rm keV}) = 0.394(16)$ b, together with the [DICEBOX]{}-modeled contribution from the quasi continuum, $P(9/2^{+};410.06~{\rm keV}) = 0.0145(14)$, yields a radiative thermal-capture cross section for the isomer $\sigma_{9/2^{+}} = 0.400(16)$ b. This lower limit is consistent with our simulated population for $J^{\pi} =9/2^{+}$ and inconsistent with $J^{\pi} = 11/2^{+}$ (Fig. \[w187:PD\_2\]). Based on our analysis we propose new $J^{\pi}$ assignments for the five levels at: 410.06 keV ($9/2^{+}$); 493.4 keV ($9/2^{-}$); 597.24 keV ($11/2^{+}$); 815.51 keV ($13/2^{+}$); and, 884.13 keV ($5/2^{+}$).
\[sec:level4.E\]Activation cross sections for $^{187}$W ($T_{1/2} = 24.000(4)$ h)\
----------------------------------------------------------------------------------
The same $^{186}$W target used in the prompt $\gamma$-ray measurements was later analyzed, offline, to determine the activation cross sections, $\sigma_{\gamma}$, for $\gamma$ rays emitted following $^{187}$W decay. Since this measurement was performed in the same experiment, the decay $\gamma$-ray cross sections could be determined proportionally to the cross sections of the prompt $\gamma$ rays. These activation $\gamma$-ray cross sections, together with their $\gamma$-decay emission probabilities, $P_\gamma$, independently determine the total radiative neutron-capture cross section, $\sigma_{0}$.
The decay $\gamma$ rays were observed in both the prompt spectrum, where the background from prompt $\gamma$ rays was high, and after bombardment, when the background was much lower. To determine the activation $\gamma$-ray cross sections, they must be corrected for saturation during bombardment, decay following bombardment and before counting begins, and decay during the counting interval. The decay $\gamma$ rays, measured in the prompt spectrum, can be corrected with an in-beam saturation factor ($B$) defined as $$B = 1 - \left( \frac{1-{\rm exp}(-\lambda t_{S})}{\lambda t_{S}} \right),
\label{pg2}$$ where $\lambda=\ln(2)/T_{1/2}$ is the decay constant and $t_{S}$ is the irradiation period. This expression is valid assuming a constant neutron flux. Monitoring showed little power variation at the Budapest Research Reactor [@belgya:pc] during our measurements. The corrected activation $\gamma$-ray cross sections, measured in the prompt spectrum, are then given by $$\sigma_{\gamma}^{(P)} = \frac{\sigma_{\gamma}}{B},
\label{pg3}$$ where $\sigma_{\gamma}$ is the uncorrected cross section observed during bombardment.
When the sample is analyzed offline the $\gamma$-ray cross sections in the delayed spectrum must also be corrected for saturation corresponding to in-beam exposure according to the factor $S = 1 - \exp(-\lambda t_{S})$. The decay time $t_D$ following bombardment until analysis commences, introduces a further correction factor $D = \exp(-\lambda t_{D})$. In addition, decay during the counting interval $t_C$ is corrected by a factor $C = [1 - \exp(-\lambda t_{C})]/(\lambda t_{C})$. The overall correction factor accounting for saturation, decay, and counting intervals can then be applied to the cross sections of the decay $\gamma$-rays observed in the delayed spectrum as $$\sigma_{\gamma}^{(D)} = \frac{\sigma_{\gamma}}{S \cdot D \cdot C}.
\label{pg7}$$
In this work the irradiation time was $t_{S} = 7536$ s, and the source decayed for a time $t_{D} = 64859$ s before being counted for $t_{C} = 11645$ s. The activation $\gamma$-ray cross sections for the most intense transitions in the prompt and delayed spectra are shown in Table \[tab:8\]. The prompt and delayed $\gamma$-ray cross sections were consistent. We can then determine the $\gamma$-ray emission probabilities, $P_{\gamma} = \sigma_{\gamma} / \sigma_{0}$, using $\sigma_{0} =33.33(62)$ b from our prompt $\gamma$-ray measurement. These probabilities are also listed in Table \[tab:8\] and are consistent with the $P_{\gamma}$ values from ENSDF [@basunia:09], based on the decay scheme normalization of Marnada *et al*. [@marnada:99]. Using the $P_{\gamma}$ values from our activation data, we can then find independent total radiative thermal neutron-capture cross sections, $\sigma_{0} = \sigma_{\gamma} / P_{\gamma}$, based on the delayed-transition cross sections reported in the activation measurements of Szentmikl[ó]{}si *et al*. [@szentmiklosi:06] and De Corte and Simonits [@decorte:03-2]. In this approach, we find that our prompt measurement, $\sigma_{0} = 33.33(62)$ b, compares well with the weighted average of Szentmikl[ó]{}si *et al*. [@szentmiklosi:06], $\sigma_{0} = 33.4(11)$ b, and also, with that of De Corte and Simonits [@decorte:03-2], $\sigma_{0} = 32.8(10)$.
\[sec:level7\]Neutron separation energies\
==========================================
A byproduct of our analysis is the determination of neutron separation energies, $S_{\rm n}$, for $^{183,184,185,187}$W from the (n,$\gamma$) primary $\gamma$-ray energy measurements and the final-level energies taken from ENSDF. These results, corrected for recoil, are shown in Table \[tab:Sn\] where they are compared with the recommended values of Wang *et al*. [@Wang:AME]. We present more precise determinations of $S_{\rm n}$ for $^{184,185}$W.
Nuclide $E_{\gamma}$ $E_{f}$ $S_{\rm n}$
----------- -------------- -------------- -------------
$^{183}$W 6190.78(6) 0.0 6190.88(6)
6144.28(6) 46.4839(4) 6190.87(6)
Average 6190.88(6)
Adopted 6190.81(5)
$^{184}$W 7410.99(14) 0.0 7411.14(14)
7299.69(16) 111.2174(4) 7411.03(16)
6507.63(16) 903.307(9) 7411.05(16)
6408.60(12) 1002.49(4) 7411.20(13)
6289.51(13) 1121.440(14) 7411.06(13)
Average 7411.11(13)
Adopted 7411.66(25)
$^{185}$W 5753.65(5) 0.0 5753.74(5)
Adopted 5753.71(30)
$^{187}$W 5466.47(12) 0.0 5466.55(12)
5320.65(8) 145.848(9) 5466.57(8)
5261.67(9) 204.902(9) 5466.65(9)
4684.31(7) 782.290(19) 5466.66(7)
4662.94(27) 803.369(22) 5466.37(27)
4650.27(8) 816.256(19) 5466.58(8)
4626.40(7) 840.205(16) 5466.66(7)
4574.67(7) 891.93(4) 5466.66(8)
Average 5466.62(7)
Adopted 5466.79(5)
: \[tab:Sn\] Neutron-separation energies determined from (n,$\gamma$) reactions: $S_{\rm n} = E_{\gamma} + E_{f} + E_{r}$, where $E_{f}$ is the energy of the final level and $E_{r} = E_{\gamma}^{2}/2A$ is the recoil energy. The weighted average for each nuclide is compared to the adopted value of Wang *et al*. [@Wang:AME].
\[sec:level6\]Summary\
======================
The total radiative thermal neutron-capture $\gamma$-ray cross sections, $\sigma_{0}$, for the four major tungsten isotopes are summarized in Table \[tab:summary\]. The cutoff energies, $E_{\rm crit}$, partial $\gamma$-ray cross sections, $\sum \sigma^{\rm exp}_\gamma$, simulated continuum GS feedings, $P({\rm GS})$, and simulated cross sections, $\sum \sigma^{\rm sim}_\gamma$, and an error budget are also given in Table \[tab:summary\]. Our new cutoff energies exceed the RIPL-suggested $E_{\rm crit}$ values [@capote:09] for all isotopes. These analyses have established that $\sigma_{0}$ is nearly independent of the assumed value of $E_{\rm crit}$, which is consistent with our earlier results for the palladium isotopes [@krticka:08].
Several combinations of photon strength function and level density formalisms were compared to the experimental data. Total radiative widths of the capture state were found to be very model dependent. For the compound $^{183,184}$W capture states, we could best reproduce the mean-adopted width $\langle \Gamma_{0} \rangle$ [@mughabghab:06] with the EGLO/CTF model combination. In the cases of $^{185,187}$W, $\Gamma_{0}$ was best reproduced assuming the EGLO/BSFG combination. All combinations involving BA gave much poorer agreement with the adopted $\Gamma_{0}$.
This analysis proposes several changes to the decay schemes for the compound tungsten isotopes $^{183,184,185,187}$W. For $^{183}$W, one new $\gamma$ ray below $E_{\rm crit}$ is proposed, based on statistical-model simulations, and a tentative $J^{\pi}$ assignment is confirmed. The 309.49-keV, 5.2(3)-s, $11/2^{+}$ isomer in $^{183}$W was populated with a cross section of 0.177(18) b. For $^{184}$W, one new $\gamma$ ray was placed in the decay scheme, based on our experiments, an additional low-energy transition is proposed from simulations, and four tentative $J^{\pi}$ assignments are confirmed. Our analysis also indicates that the capture state in $^{184}$W is consistent with the composition $J^{\pi}_{\rm CS} = 1^{-}(\gtrsim 80~\%)$, $J^{\pi}_{\rm CS} = 0^{-}(\lesssim 20~\%)$, which is also consistent with the *Atlas of Neutron Resonances* [@mughabghab:06]. We find $J^{\pi} = 1^{-}$ the most likely assignment for the bound resonance at $-26.58$ eV, implying a likely capture-state spin composition of $J^{\pi}_{\rm CS} = 0^{-}(7.4~\%) + 1^{-}(92.6~\%)$. The 1285.00-keV, 8.33(18)-$\mu$s, $5^-$ isomer in $^{184}$W was populated with a cross section of 0.0246(55) b. In $^{185}$W two new low-energy $\gamma$-ray transitions are proposed based on simulations, and three previous tentative $J^{\pi}$ assignments have been validated. The 197.38-keV, 1.67(3)-min, 11/2$^+$ isomer in $^{185}$W was populated with a cross section of 0.0062(16) b. For $^{187}$W, 19 of the previous $J^{\pi}$ assignments are confirmed and new $J^{\pi}$ assignments are proposed for five levels, including a new $9/2^{+}$ bandhead assignment at 410.06 keV that was previously assigned $(11/2^{+})$. In addition, we reintroduced the 493.4-keV level, from an earlier ENSDF evaluation [@firestone:91], and a new $\gamma$ ray depopulating this level based on tentative evidence in the capture-$\gamma$ spectrum. There is also tentative evidence for a new transition at around 135.1 keV, depopulating the 775.60-keV level. Our $^{187}$W simulations support inclusion of four new low-energy $\gamma$ rays, three of which were previously inferred in the work of Bondarenko [*et al*]{}. [@bondarenko:08]. The 410.06-keV, 1.38(7)-$\mu$s, $11/2^{+}$ isomer in $^{187}$W was populated with a cross section of $0.400(16)$ b. An analysis of the $\beta^{-}$-delayed $\gamma$-ray spectrum provided an independent decay-scheme normalization based on a new set of $P_{\gamma}$ measurements that compare well to the ENSDF decay-scheme normalization [@basunia:09], adopted from the earlier work of Marnada *et al*. [@marnada:99]. Independent values of $\sigma_{0}$, consistent with our prompt measurement, were then determined based on our activation-data decay-scheme normlaization, thus providing confirmation of our approach.
The decay-scheme improvements suggested in this work will be used to improve the ENSDF nuclear-structure evaluations [@ensdf], that contribute to the RIPL nuclear-reaction database [@capote:09]. The new thermal-capture (n,$\gamma$) data will be added to the EGAF database [@firestone:06]. These new data will also be used to help produce a more extensive and complete thermal-capture $\gamma$-ray library for the ENDF [@chadwick:11] neutron-data library. Additional measurements of capture $\gamma$-rays from the rare isotope $^{180}$W(n,$\gamma$) are in progress and will complete our knowledge of the tungsten isotopes and resolve discrepancies in the measured $\sigma_{0}$ for this nucleus.
Acknowledgments {#acknowledgments .unnumbered}
===============
This work was performed under the auspices of the University of California, supported by the Director, Office of Science, Office of Basic Energy Sciences, of the U. S. Department of Energy at the Lawrence Berkeley National Laboratory under Contract DE-AC02-05CH11231, and by the U. S. Department of Energy by the Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. The access to the Budapest PGAA facility was financially supported by the NAP VENEUS08 grant under Contract OMFB-00184/2006. Additional support was received through the research plan MSM 002 162 0859 supplied by the Ministry of Education of the Czech Republic. The operations staff at the Budapest Research Reactor are gratefully acknowledged.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Calculations of annual energy production (AEP) from a wind power plant—whether based on preconstruction or operational data—are critical for wind plant financial transactions. The uncertainty in the AEP calculation is especially important in quantifying risk and is a key factor in determining financing terms. A popular industry practice is to assume that different uncertainty components within an AEP calculation are uncorrelated and can therefore be combined as the sum of their squares. We assess the practical validity of this assumption for operational-based uncertainty, by performing operational AEP estimates for more than 470 wind plants in the United States, mostly in simple terrain. We apply a Monte Carlo approach to quantify uncertainty in five categories: revenue meter data, wind speed data, regression relationship, long-term correction, and future interannual variability. We identify correlations between categories by comparing the results across all 470 wind plants. We observe a positive correlation between interannual variability and the linearized long-term correction; a negative correlation between wind resource interannual variability and linear regression; and a positive correlation between reference wind speed uncertainty and linear regression. Then, we contrast total operational AEP uncertainty values calculated by omitting and considering the correlations between the uncertainty components. We quantify that ignoring these correlations leads to an underestimation of total AEP uncertainty of, on average, 0.1%, and as large as 0.5% for specific sites. Although these are not large increases, these would still impact wind plant financing rates; further, we expect these values to increase for wind plants in complex terrain. Based on these results, we conclude that correlations between the identified uncertainty components should be considered when computing the total AEP uncertainty.'
bibliography:
- 'biblio.bib'
title: 'Operational-Based Annual Energy Production Uncertainty: Are its Components Actually Uncorrelated?'
---
Introduction
============
Calculations of wind plant annual energy production (AEP)—whether based on preconstruction data before a wind power plant is built or on operational data after a wind plant has started its operations—are vital for wind plant financial transactions. Preconstruction estimates of AEP are needed to secure and set the terms for new project financing, whereas operational estimates of long-term AEP are required for important wind plant transactions, such as refinancing, purchasing/selling, and mergers/acquisitions. The need for AEP analyses of wind plants is increasing because global wind capacity increased to 539 GW in 2017, representing 11% and 91% increases over 1-year and 5-year periods, respectively; capacity is expected to increase by another 56%, to 841 GW, by 2022 [@gwec2017]. In the United States, wind plants generated more than 300,000 in 2019, about 7.5% of the total U.S. electricity generation from utility-scale facilities that year, with a 50% increase over a 6-year period [@us2020monthly].
This rapid growth of the wind energy industry is putting an increased spotlight on the accuracy and consistency of AEP calculations. For preconstruction AEP estimates, there has been considerable movement toward standardization. The International Energy Commission (IEC) is currently developing a standard [@iec15] and there have long been guidance and best practices available [@brower2012]. By contrast, long-term operational AEP estimates do not have such extensive guidance or standards. Only limited standards covering operational analyses exist; [@iec12] addresses turbine power curve testing, and [@iec26] addresses the derivation and categorization of availability loss metrics. However, to our knowledge, there are no standards and very limited published guidance on calculating long-term AEP from operational data. Rather, documentation seems to be limited to a consultant report [@lindvall2016], an academic thesis [@khatab2017], and limited conference proceedings [@cameron2012; @lunacek2018].
Documentation and standards for preconstruction AEP methods are of limited use for operational-based AEP methods, given the many differences between the two approaches. In general, operational AEP calculations are simpler than preconstruction estimates because actual measurements of wind plant power production at the revenue meter replace the complicated preconstruction estimate process (e.g., meteorological measurements, wind and wake-flow modeling, turbine performance, estimates of wind plant losses). However, the two methods do share several similarities, including regression relationships between on-site measurements and a long-term wind speed reference, the associated long-term (windiness) correction applied to the on-site measurements, estimates of future interannual variability, and estimates of uncertainty in the resulting AEP calculation. The shared components between operational AEP calculations and preconstruction estimates [@iec15] are listed in Table \[tab:unc\_cat\].
Uncertainty component Description
------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------
On-site measurements Measurement error in met mast wind speeds (preconstruction) or power at the revenue meter (operational)
[Reference wind speed data]{} Measurement or modeling error in measured or modeled long-term reference wind speed data
Losses Error in estimated or reported availability and curtailment losses
Regression Sensitivity in the regression relationship between on-site measurements and reference wind speeds
Long-term (windiness) correction Sensitivity in the long-term correction applied to the regression relationship between on-site measurements and reference wind speeds
Interannual variability of resource Sensitivity in future energy production because of resource variability
The uncertainty values from each component listed in Table \[tab:unc\_cat\] must be combined to produce a total estimate of AEP uncertainty. While general guidelines on how to combine (measurement) uncertainty components exists [@iso1995guide] and can be applied to this task, we found no specific guidance in the literature for combining uncertainty components in an operational AEP estimate. On the other hand, considerable guidance exists for combining preconstruction AEP uncertainties [@lackner2007; @brower2012; @vaisala2014; @kalkan2015; @clifton2016]. In every case, recommended best practices assume that all uncertainties, $\sigma_\text{i}$, are uncorrelated and can therefore be combined using a sum of squares approach to give the total AEP uncertainty, $ \sigma_{\text{tot,uncorr}}$: $$\sigma_{\text{tot,uncorr}} = \sqrt{\sum_\text{i=1}^\text{N} \sigma_\text{i}^2}
\label{unc_unc}$$
To better understand how uncertainties are combined in long-term operational AEP calculations, we reached out to several wind energy consultants who regularly perform these analyses. These conversations revealed that uncertainties in a long-term operational AEP calculation are also assumed uncorrelated and combined using Equation \[unc\_unc\].
Goal of Study
-------------
The purpose of this study is to examine the extent to which the assumption of uncorrelated uncertainties—and, therefore, the combination of those uncertainties through a sum of squares approach—is accurate and appropriate for operational AEP calculations. Specifically, this study aims to identify potential correlations between AEP uncertainty components, using data for over 470 wind plants. While in the analysis we focus on operational AEP calculation, we expect that the results from this analysis—namely, the potential identification of correlated uncertainty components—can be equally relevant for informing and improving preconstruction AEP methods.
In Section 2, we first describe the data sources used in this analysis (wind plant operational data and reanalysis products), the Monte Carlo approach to quantify single uncertainty components in operational AEP, and the approaches used to combine these uncertainty components. Section 3 presents the main results of our analysis in terms of uncertainty contributions and correlation among the different components. We conclude and suggest future work in Section 4.
Data and Methods
================
Wind plant Operational Data and Reanalysis Products
---------------------------------------------------
Operational wind plant energy production data for this analysis are obtained from the publicly available Energy Information Administration (EIA) 923 database [@eia18]. This database provides reporting of monthly net energy production from all power plants in the United States, including wind plants. More than 670 unique wind plants are available from this data set.
Long-term wind speed data (needed to perform the long-term or windiness correction in an AEP estimate) are used from three reanalysis products over the period of January 1997 through December 2017:
- Version 2 of the Modern-Era Retrospective analysis for Research and Applications (MERRA-2, @gelaro2017modern). We specifically use the M2T1NXSLV data product, which provides diagnostic wind speed at 50 m above ground level (AGL), interpolated from the lowest model level output (on average about 32 m AGL), using Monin Obukhov similarity theory. Data are provided at an hourly time resolution.
- The European Reanalysis Interim (ERA-Interim) data set [@dee2011era]. We specifically use output at the 58$^\text{th}$ model level, which on average corresponds to a height of about 72 m AGL. Data are provided at a 6-hourly time resolution.
- The National Centers for Environmental Prediction v2 (NCEP-2) data set [@saha2014ncep]. We specifically use diagnostic wind speed data at 10 m AGL. Data are provided at a 6-hourly time resolution.
The wind speed data are density-corrected at their native time resolutions to correlate more strongly with wind plant power production (i.e., higher-density air in winter produces more power than lower-density air in summer, wind speed being the same): $$U_{\text{dens,corr}} = U \left(\frac{\rho}{\rho_{\text{mean}}} \right) ^{1/3}
\label{den_corr}$$ where $U_{\text{dens,corr}}$ is the density-corrected wind speed, $U$ is the wind speed, $\rho$ is air density (calculated at the same height as wind speed), $\rho_{\text{mean}}$ is the mean density over the entire period of record of the reanalysis product, and the exponent $1/3$ is derived from the basic relationship between wind power and wind speed cubed [@manwell2010wind]. To calculate air density at the same height as wind speed, we first extrapolate the reported surface pressure to the wind speed measurement height, assuming hydrostatic equilibrium [@atmosphere1975iso]: $$p = p_{\text{surf}} \: \exp \left[ \frac{g z}{RT_{\text{avg}}} \right]$$ where $p$ is the pressure at the wind speed measurement height, $p_{\text{surf}}$ is the surface pressure, $g$ is the acceleration caused by gravity, $z$ is the wind speed measurement height, $R$ is the gas constant, and $T_{\text{avg}}$ is the average temperature between the reported value at 2 m AGL and at the wind speed measurement height.
{width="16cm"}
To lessen the impact of limited and/or poor-quality data on the results of our analysis, we filter for wind plants with a moderate-to-strong correlation with all three reanalysis products ($R^2 >$ 0.6). About 25% of the EIA wind plants are discarded with this filter. We also impose a threshold of eight months of wind plant data availability in order to investigate uncertainty as it relates to a low number of data points—but not so low as to make the use of a regression relationship questionable. A total of 472 wind plants are kept for the final analysis, and their locations are shown in Figure \[Map\_Sel\]. Because obtaining an accurate representation of wind data in complex terrain by reanalysis products is challenging [@shravan2009comparision], most of the selected wind plants are located in the Midwest and Southern Plains. Notably, no wind plants in California pass the filtering criteria because they are predominately located in areas with thermally driven wind regimes, such as Tehachapi Pass, where coarse-resolution reanalysis products are poor predictors of wind energy production.
The fundamental step in an AEP calculation involves a regression between density-corrected wind speed (here, from the reanalysis products) and energy production (here, from the EIA 923 database). To investigate whether a simple linear function can be assumed to express the relationship between density-corrected wind speed and wind plant energy production when considering monthly data, we show a scatterplot between MERRA-2 density-corrected monthly wind speed and monthly energy production across all 472 sites in Figure \[linear\_reg\]. For each site, data have been normalized by the respective site mean. We show best-fits using a linear, quadratic, and cubic function, and calculate the mean absolute error (MAE) of each fit.
{width="12cm"}
We find that the difference between the normalized MAE values from the considered functions is less than 0.7%. Therefore, the uncertainty connected with the choice of using a linear regression in the operational AEP methodology at monthly time resolution appears minimal. Moreover, through conversations with wind industry professionals, we found that a linear regression based on monthly data is the standard industry approach when performing bankable[^1] operational AEP analyses.
Operational AEP Methodology
---------------------------
Given the lack of existing guidelines for a standard approach for *operational* AEP calculations, we base our methodology on conversations with four major wind energy consultants who represent most of the operational market share in North America. These conversations overwhelmingly revealed the following characteristics for operational AEP analysis, and we follow the same approach in our analysis:
1. Wind speed data (measured or modeled) are density-corrected at their native time resolution, using Equation \[den\_corr\].
2. Monthly revenue meter data, monthly average availability and curtailment losses, and monthly average wind speeds from a long-term wind resource product are calculated.
3. Monthly revenue meter data are normalized to 30-day months (e.g., for January, the revenue meter values are multiplied by 30/31).
4. Monthly revenue meter data are corrected for monthly availability and curtailment (i.e., monthly gross energy data are calculated).
5. A linear regression between monthly gross energy production and concurrent density-corrected monthly average wind speeds is performed.
6. Long-term density-corrected monthly average wind speed is then calculated for each calendar month (i.e., average January wind speed, average February wind speed, and so forth) with a hindcast approach, using 10–20 years of the available long-term reference monthly wind resource data (reanalysis products, long-term reference measurements, etc.).
7. Slope and intercept values from the regression relationship are then applied to the long-term density-corrected monthly average wind speed data with the long-term or so-called windiness correction. A long-term data set of monthly (January, February, etc.) estimated gross energy production is obtained.
8. The resulting long-term monthly gross energy estimates, which are based on 30-day months, are then denormalized to the actual number of days in each calendar month (e.g., for January, the obtained value is multiplied by 31/30).
9. Long-term estimates of availability and curtailment losses are finally applied to the denormalized long-term monthly gross energy data, leading to a long-term calculation of operational AEP.
In the EIA-923 database, availability and curtailment data are not available. Therefore, in our analysis we omit steps 4 and 9 of the list, and only perform calculations on net energy data.
{width="16cm"}
Monte Carlo Analysis
--------------------
To quantify the impact of the single uncertainty components on the long-term operational AEP estimate obtained using the methodology described in the previous section, we implement a Monte Carlo approach. In general, a Monte Carlo method involves the randomized sampling of inputs to, or calculations within, a method which, when repeated many times, results in a distribution of possible outcomes from which uncertainty can be deduced. This is usually calculated as the standard deviation or the coefficient of variation (i.e. standard deviation normalized by mean) of the resulting distribution [@iso1995guide; @dimitrov2018wind]. Monte Carlo methods have been used in different applications for uncertainty quantification within the wind energy industry, ranging from the prediction of extreme wind speed events [@ishihara2015prediction], to offshore fatigue design [@muller2018application], to economic analysis of the benefits of wind energy projects [@williams2008estimating]. Here, we apply this approach to derive a distribution of long-term operational AEP values from which its uncertainty can be calculated. Using a Monte Carlo approach provides a direct estimate of AEP uncertainty by sampling the relevant parameters connected to the various uncertainty components. By contrast, traditional approaches to assessing uncertainty are often less direct. For example, wind resource IAV is often calculated and then converted to AEP uncertainty through an “energy-velocity” (EV) ratio estimated from the wind and energy data. A Monte Carlo approach avoids this intermediate ratio and any uncertainty or error associated with it.
In our analysis, we separately consider five operational-based uncertainty components so that only the sampling of one parameter is performed in each Monte Carlo configuration. The following uncertainty components are included in our proposed Monte Carlo methodology for long-term operational AEP:
- Revenue meter measurement error. To incorporate this uncertainty component in the Monte Carlo simulation, we sample monthly revenue meter data from a synthesized normal distribution centered on the reported value and $0.5\%$ imposed standard deviation. In fact, a value of 0.5% is consistent with what is typically assumed in the wind energy community as revenue meter uncertainty [@iec60688; @ansic12].
- Reference wind speed data modeling error. Quantifying the uncertainty of the long-term wind resource data used in the operational AEP assessment is challenging because it can vary based on the location, long-term wind speed product used, or instrument from which reference observations are taken. To include this uncertainty component in a systematic way across the 472 locations considered in our analysis, we adopt an ensemble uncertainty approach [@taylor2009wind; @zhang2015comparison] and use as proxy the variability of the wind resource between different reanalysis products. Therefore, at each Monte Carlo iteration at each site, we randomly select wind resource data from one of the three considered reanalysis products.
- Linear regression model uncertainty. We adopt a novel way, directly enabled by the use of Monte Carlo, to incorporate this uncertainty component in the operational AEP assessment. We sample the regression slope and intercept values from a multivariate normal distribution centered on their best-fit values and covariance matrix equal to one of the best-fit parameters. The diagonal terms in the covariance matrix are given by the square of the slope and intercept standard errors. For a regression model between an independent variable, $x$, and a dependent variable, $y$, the standard error of the regression is defined [@iso1995guide] as: $$e_y = \sqrt{\frac{\sum \left(y_i - \hat{y_i} \right)^2}{n-2}},$$ where $\hat{y_i}$ is the regression-predicted value for $y_i$ and $n$ is the number of data points used in the regression. The standard error of the regression slope is: $$e_a = \frac{e_y}{\sum \left(x_i - \overline{x_i} \right)^2},
\label{unc_slope}$$ and the standard error of the intercept is: $$e_b = e_y \: e_a \sqrt{\frac{\sum x_i^2}{n}}.
\label{unc_intercept}$$ $e_a^2$ and $e_b^2$ are the diagonal terms in the covariance matrix of the multivariate normal distribution of regression slope and intercept from which Monte Carlo values are drawn. Slope and intercept values are strongly negatively correlated, which is captured by their covariance when performing the linear regression. The off-diagonal terms in the covariance matrix of the multivariate normal distribution constrain the random sampling of slope and intercept values to avoid sampling unrealistic combinations.
{width="16cm"}
An example of this sampling is shown in Figure \[Ex\_regression\] for two projects of different regression strengths. We sample 500 slope and intercept values from a multivariate normal distribution centered around the best-fit parameters, and with the covariance matrix derived from the standard errors of slope and intercept and their covariance. As shown in Figure 4, the low standard errors found for the leftmost regression relationship constrain the possible slope and intercept values that can be sampled, while the high standard errors in the rightmost regression relationship allow for a much wider sampling.
- Long-term (windiness) correction uncertainty. We incorporate this component by sampling the number of years (randomly picked between 10 and 20) to use as the long-term wind resource data to which the regression coefficients are applied to derive long-term energy production data (the so-called windiness correction).
- Wind resource interannual variability (IAV) uncertainty. We incorporate this uncertainty component in the Monte Carlo method by sampling the long-term (reanalysis) average calendar monthly wind speeds (i.e., average January, average February) used to calculate long-term monthly energy production data. The sampling distribution is normal, centered on the calculated long-term average calendar monthly wind speed, and with a standard deviation equal to the 20-year standard deviation of the long-term average monthly wind speed for each calendar month.
Each of the listed sources of uncertainty corresponds to a Monte Carlo sampling and is highlighted by a probability distribution in the flowchart in Figure \[MC\_process\]. Note that uncertainty components related to availability and curtailment losses are not considered in our approach because the EIA 923 database does not include measurements of these losses. To calculate these uncertainty components at each wind plant, we run the Monte Carlo simulation under five different setups, each of them having only a single sampling performed (i.e., either revenue meter, reference wind speed data, IAV, linear regression, or windiness correction). For each component, we run the Monte Carlo simulation 10,000 times. We quantify the impact of each single uncertainty component on the long-term operational AEP in terms of the coefficient of variation of the distribution of operational AEP resulting from the Monte Carlo simulation run. Convergence of the AEP distribution within 0.5% of the true mean after the 10,000 Monte Carlo runs was verified for all projects, with 95% confidence.
The code used to perform the AEP calculations is published and documented in NREL’s open-source operational assessment software, OpenOA.[^2] Calculations were performed on Eagle, NREL’s high-performance computing cluster. Specifically, each wind plant was assigned a different processor and run in parallel. Given the general simplicity of the AEP method used here, computational requirements were moderate despite the 50,000 simulations (10,000 runs x 5 uncertainty setups) required for each wind plant.
Combination of Uncertainty Components
-------------------------------------
Once the contribution from each uncertainty component to the long-term operational AEP uncertainty has been quantified, the different components need to be combined to obtain the total AEP uncertainty. As stated in the Introduction, it is common practice for wind energy consultants to assume that all uncertainty components are uncorrelated, and combine them using Equation \[unc\_unc\] to obtain $\sigma_{\text{tot,uncorr}}$. To test the validity of this assumption, we apply Equation \[unc\_unc\], in which each of the five considered uncertainty components $\sigma_i$ is quantified as the coefficient of variation of the corresponding operational AEP distribution obtained by running the Monte Carlo simulation with a single sampling performed. We note that the same values of $\sigma_{\text{tot,uncorr}}$ would be obtained by running the Monte Carlo simulation with, at each iteration, all of the five samplings performed, independently from each other.
We contrast the total AEP uncertainty calculated assuming uncorrelated components with what we obtain by taking into account these correlations in the calculation. Following the guidance in @iso1995guide, we combine the various uncertainty components and calculate the total long-term operational AEP uncertainty for each wind plant as: $$\sigma_{\text{tot,corr}} = \sqrt{\sum_\text{i=1}^\text{N} \sigma_\text{i}^2 + 2 \sum_\text{i=1}^\text{N-1} \sum_\text{j=i+1}^\text{N} R_\text{ij} \sigma_\text{i} \sigma_\text{j} }
\label{unc_corr}$$ where, in our analysis, $N = 5$ and $R_\text{ij}$ is the correlation coefficient between each pair of uncertainty components calculated from the results obtained for all 472 wind plants considered in the analysis.
The comparison between $\sigma_{\text{tot,uncorr}}$ and $\sigma_{\text{tot,corr}}$ will give insights into the error arising from ignoring the correlations existing between the various uncertainty components.
Results
=======
Operational-Based AEP Uncertainty Contributions
-----------------------------------------------
Distributions of each uncertainty component, expressed in terms of the percent coefficient of variation of the resulting AEP distributions, across all 472 wind plants are shown in Figure \[unc\_dist\]
![Operational-based AEP uncertainty distributions across projects for the different uncertainty components; mean values across projects are shown in the legend. Uncertainty values are quantified as the percent coefficient of variation of the long-term operational AEP distribution.[]{data-label="unc_dist"}](f05){width="8.3cm"}
Uncertainty connected to wind resource IAV is found to contribute the most (average 4.1% across all wind plants). The uncertainty in the linear regression model has the second-largest contribution (1.5%), followed by the uncertainty of the reference wind speed data (0.8%; here, of the reanalysis products), and revenue meter data (here, imposed at 0.5%). The long-term windiness correction has the smallest uncertainty component (0.4%). Therefore, the number of years used for the long-term windiness correction does not have a large impact on the overall uncertainty in operational AEP, at least for the sampled range of 10–20 years. Using as few as 10 years seems sufficient to give stability to the long-term AEP estimate and adding additional years does not provide a significant reduction in the uncertainty connected with the long-term estimate. As already mentioned in Section 2, these results are obtained for wind plants in mostly simple terrain and with a moderate-to-strong correlation between reanalysis wind resource and wind energy production and, therefore, with an overall low operational AEP uncertainty. We acknowledge that the inclusion of wind plants with a weaker correlation with the reanalysis products would modify the relative contribution of the various uncertainty components (e.g., the importance of the regression uncertainty would increase).
Correlation Between Operational-Based AEP Uncertainty Components
----------------------------------------------------------------
To be able to assess the validity of the uncorrelated assumption when combining different uncertainty components, we assess potential correlations between uncertainty components by analysing the Pearson’s correlation coefficients $R_{ij}$ (needed in Equation \[unc\_corr\] to calculate $\sigma_{\text{tot,corr}}$) from each pair of AEP uncertainty components across the 472 wind plants, and we summarize the results in the correlation matrix in Figure \[unc\_corr\_heatmap\].
{width="12cm"}
To assess which of the obtained correlations have statistical significance, we calculate the $p-$value [@westfall1993resampling] associated with the ten correlation coefficients. The test reveals that for three pairs of uncertainty components, the probability of finding the *observed* not-zero correlation coefficients if the *actual* correlation coefficient were, in fact, zero ($p-$value), is less than $10^{-5}$. Therefore, the following three correlations have strong statistical significance:
- The wind resource IAV and the long-term windiness correction uncertainties are moderately correlated ($R = 0.49$, $p = 1.9 \cdot 10^{-29}$).
- The linear regression and reference wind speed data uncertainties are weakly correlated ($R = 0.35$, $p = 2.5 \cdot 10^{-15}$).
- The wind resource IAV and the linear regression uncertainties appear weakly negatively correlated ($R = -0.21$, $p = 2.6 \cdot 10^{-6}$).
The first correlation noted earlier (wind resource IAV and long-term windiness correction) is explained simply by the fact that both uncertainty components are driven by wind resource variability. At a site with large wind variability, IAV will be large by definition and so will the uncertainty introduced by different lengths of time series used for the long-term AEP calculation.\
The correlation between linear regression and reference wind speed data uncertainties can be justified given the dependence of both these uncertainty components on the number of data points used in the regression between energy production data and concurrent wind speed data (Figure \[regression\_uncertainty\]).
{width="16cm"}
Both the slope and intercept errors (Equations \[unc\_slope\] and \[unc\_intercept\]), from which the linear regression uncertainty depends (as described in Section 2.3), are inversely proportional to the number of data points so that when a regression is performed on few data points, its uncertainty increases. This dependence is exemplified in Figure \[Ex\_regression\], where we have compared the sampling sets of regression lines for two stations in the EIA data set: for these two cases, the standard errors of regression slope and intercept for the station with 8 data points (on the right) are 30–50 times larger than what is found for the station with 90 data points (on the left).
{width="16cm"}
{width="16cm"}
The number of data points used for the regression also has an impact on the reference wind speed data uncertainty. In fact, short periods of wind plant operation record can lead to different interpretations from the reference wind resource data sets used as to whether that short period of record was above, equal to, or below the long-term average resource. Over a longer period of record, these potential discrepancies between different wind resource data sets (in our case, reanalysis products) tend to average out; therefore, leading to a reduced uncertainty. We illustrate this phenomenon by exploring the long-term trend of the reanalysis products for the wind plant with one of the highest reported reference wind speed data uncertainties (EIA ID 60502 reported 3.7% reference wind speed data uncertainty). Figure \[long\_ws\] shows the result. The period of record for wind plant operation (shown by a shaded blue area in Figure \[long\_ws\]) was only 12 months. As shown in the figure, the various reanalysis products have very different interpretations of the wind resource in the short period of record relative to the long-term (ERA-I: 4% above average; MERRA-2: 1% below average; NCEP-2: 1% above average). Consequently, the use of each reanalysis product will lead to different magnitudes (both positive and negative) in the long-term windiness corrections, leading to high uncertainty in the resulting operational AEP calculation. By increasing the period of record (i.e., increasing the number of data points used in the regression), such discrepancies tend to average out. This is illustrated in Figure \[lt\_ratio\], where we show how the period of record to long-term wind speed ratio varies as we extend the period of record by increasing the number of months while keeping December 2017 as the fixed ending time. For short periods of record, there is considerable deviation of this ratio among the different reanalysis products (i.e., the reference wind speed data uncertainty is high). As the length of the period of record increases, this ratio tends to converge to 1.0, and the spread between the three reanalysis products decreases (i.e., the reference wind speed data uncertainty is low).\
Finally, the (weak) negative correlation between linear regression and wind resource IAV uncertainties is linked to the fact that they respond differently to the $R^2$ coefficient between the reanalysis wind speed and the energy production data (Figure \[dep\_r2\]).
{width="16cm"}
Predictably, the linear regression uncertainty is inversely proportional to the coefficient of determination because a stronger correlation between winds and energy production will lead to a reduced uncertainty of the regression between the two variables. On the other hand, wind resource IAV uncertainty shows a positive correlation with the regression $R^2$ coefficient. This dependence can be explained because both quantities are positively correlated with the total variance of wind speed or, equivalently, produced energy. Figure \[dep\_iav\_spread\] shows the relationship between IAV uncertainty and the total sum of squares $SS_{\text{tot, WS}}$ of reanalysis wind speed (here, using MERRA-2 monthly data), which is proportional to the variance of the data: $$SS_{\text{tot, WS}} =\sum_i{(WS_i - \overline{WS})^2.}$$ A positive correlation between IAV uncertainty and $SS_{\text{tot, WS}}$ emerges.
![Relationship between IAV uncertainty and the total sum of squares $SS_{\text{tot, WS}}$ of MERRA-2 wind speed data for the 472 projects considered.[]{data-label="dep_iav_spread"}](f11){width="8.3cm"}
At the same time, the linear regression $R^2$ coefficient also depends on the variance of the produced energy (and, equivalently, of wind speed) as it is defined as: $$R^2 = 1 - \frac{SS_{\text{res}}}{SS_{\text{tot}}} % = 1 - \frac{\sum_i{(y_i - f_i)^2}}{\sum_i{(y_i - \overline{y})^2}}
\label{R2}$$ where $SS_{\text{res}}$ is the total sum of the residuals from the linear regression. Equation \[R2\] shows that when the total sum of squares $SS_{\text{tot}}$ increases, so does $R^2$, thus confirming the positive correlation between $R^2$ and the variance in the data.
Comparison Between Total Operational-Based AEP Uncertainty Under Different Assumptions
--------------------------------------------------------------------------------------
After having revealed the correlations existing between different AEP uncertainty components and explained their sources, we can compare the total operational AEP uncertainty calculated when allowing for these correlations (Equation \[unc\_corr\]) with the total uncertainty calculated with the uncorrelated assumption using the conventional sum of squares approach (Equation \[unc\_unc\]).
{width="16cm"}
Figure \[unc\_histogram\] shows the results of this comparison for the 472 wind plants considered as a scatterplot and also as a histogram of the difference $\sigma_{\text{tot,corr}} - \sigma_{\text{tot,uncorr}}$. A weak bias can be observed with a mean value of $+0.1\%$ in uncertainty difference (and differences up to 0.5% for specific wind plants). In other words, if correlations between the different uncertainty components are ignored in the calculation method, the whole operational AEP uncertainty is then, on average, slightly underestimated.
This difference can be explained by comparing the contributions $R_\text{ij} \overline{\sigma_\text{i}} \overline{\sigma_\text{j}}$ from the various uncertainty pairs in Equation \[unc\_corr\] averaged over the 472 considered wind plants.
{width="16cm"}
Figure \[Pos\_neg\_corr\]a shows the mean magnitude (across all wind plants) of these contributions for all of the considered uncertainty pairs. The negative correlation between IAV and linear regression has the largest single impact because this correlation involves the two largest uncertainty components (Figure \[unc\_dist\]). However, the sum of the contributions from all of the positive correlations exceeds the sum of the contribution from the negatively correlated components (Figure \[Pos\_neg\_corr\]b), thus resulting in the overall average increase in total operational AEP uncertainty when the correlations are taken into account in the calculation.
Financial operations related to wind plants require accurate calculations of the annual energy production (AEP) and its uncertainty prior to the construction of the plant and, often, in the context of its operational analysis. As wind energy penetration increases globally, the need for techniques to accurately assess AEP uncertainty is a priority for the wind energy industry. Typically, current industry practice assumes that uncertainty components in AEP estimates are uncorrelated. However, we have shown that this assumption is not valid for the five components that comprise an operational-based uncertainty. We used a Monte Carlo approach to assess AEP; this provides quantitative insights into aspects of the AEP calculation that drive its uncertainty. We have applied this approach using operational data from 472 wind plants, mostly in simple terrain, across the United States in the EIA-923 database, in order to study potential correlations between uncertainty components. Three pairs of uncertainty components revealed a statistically significant correlation: wind resource interannual variability (IAV) and long-term windiness correction (positive correlation); wind resource IAV and linear regression (negative); and reference wind speed data and linear regression (positive). Wind resource IAV and long-term windiness correction uncertainties are correlated because they both depend on wind resource variability. Wind resource IAV uncertainty is correlated with linear regression uncertainty because they are both inversely proportional to the number of data points in the period of record. Finally, reference wind speed data uncertainty and linear regression uncertainty show a negative correlation because they respond oppositely to the $R^2$ coefficient between the (reanalysis) wind speed and energy production data.
Our results show that ignoring these correlations between uncertainty components causes an underestimation of the total operational AEP uncertainty of, on average, about 0.1%, with peak differences of 0.5% for specific sites. These differences, though not large, would still have a significant impact on increasing wind plant financing rates. Moreover, we expect differences would become even larger for sites characterized by a more complex wind flow. Therefore, our results suggest that correlations between uncertainty components should be taken into account when assessing the total operational AEP uncertainty.
Additional components of uncertainty in an operational AEP were not considered in our study because of limited reporting in the EIA-923 database. These components include reported availability, curtailment uncertainty, and various uncertainties introduced through analyst decision-making (e.g., filtering high-loss months from analysis and regression outlier detection). Future studies could include the impact of these additional sources of uncertainty on the operational AEP assessment. Moreover, our analysis excluded sites, mostly in complex terrain, with a weak correlation between reanalysis wind resource data and wind power production. Future work could explore the magnitude of operational AEP uncertainty and the correlation between its components for such complex flow regimes. Finally, this study focused on correlations between operational AEP uncertainty components. Future work could explore correlations between the numerous preconstruction AEP uncertainty components (e.g., wake loss, wind speed extrapolation, wind flow model).
[^1]: Results are accepted by banks, investors, and so on for use in financing, buying/selling, and acquiring wind plants.
[^2]: https://github.com/NREL/OpenOA
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the dynamical behaviour of the quantum cellular automaton of Refs. [@darianopla; @BDTqcaI], which reproduces the Dirac dynamics in the limit of small wavevectors and masses. We present analytical evaluations along with computer simulations, showing how the automaton exhibits typical Dirac dynamical features, as the Zitterbewegung and the scattering behaviour from potential that gives rise to the so-called Klein paradox. The motivation is to show concretely how pure processing of quantum information can lead to particle mechanics as an emergent feature, an issue that has been the focus of solid-state, optical and atomic-physics quantum simulator.'
author:
- Alessandro
- Giacomo Mauro
- Alessandro
bibliography:
- 'bibliography.bib'
title: 'The Dirac Quantum Cellular Automaton in one dimension: Zitterbewegung and scattering from potential'
---
Introduction
============
The idea of reproducing the evolution of a macroscopic system starting from a simple rule of local interaction among its elementary constituents was first formalized in the pioneering von Neumann’s paper [@neumann1966theory] with the notion of *Cellular Automaton*. The automaton is a regular lattice of cells with a finite number of states, equipped with a rule that updates the cell states from time $t$ to time $t+1$. Such rule must be *local*, namely the state of the $x$ cell at $t+1$ depends only on the states of a finite number of neighboring cells at $t$. Cellular automata have been a popular topic for many years, as a new paradigm for complex systems, and many books have been devoted to the subject (see eg. Refs. [@wolfram2002new; @toffoli1987cellular]). One of the reasons of its first success, which eventually has become its own weakness, is the chaotic behaviour of the automaton dynamics [@sep-cellular-automata].
Differently from classical cellular automata, *Quantum Cellular Automata* (QCA) exhibit a less chaotic behaviour, which makes them predictable for large number of steps [@ambainis2001one]. Here the cells are finite-dimensional quantum systems interacting locally and unitarily. Being locality of interactions an essential ingredient of any physical evolution, QCA have been considered already by Feynman as candidates for simulating physics [@feynman1965quantum; @feynman1982simulating]. More recently QCA earned interest in the quantum information community leading to many results on its mathematical theory [@schumacher2004reversible; @arrighi2011unitarity; @gross2012index], and on their general dynamical features [@ambainis2001one; @knight2004propagating; @valcarcel2010tailoring; @ahlbrecht2011asymptotic; @reitzner2011quantum]. In quantum field theory, after the first appearance of a prototype of QCA in the Feynman chessboard [@feynman1965quantum] for solving the path-integral for the Dirac field, a similar framework has appeared in the work of Nakamura [@nakamura1991nonstandard] motivated by a rigorous formulation of the Feynman path integral, and later in the seminal work of Bialynicki-Birula [@bialynicki1994weyl], as a lattice theory for Weyl, Dirac, and Maxwell fields. Then the possibility of using automata for describing the evolution of relativistic fields emerged in the context of lattice-gas simulations, especially in the work of Meyer [@meyer1996quantum], where a notion of “field automaton” first appeared, and in the papers of Yepez [@Yepez:2006p4406].
More recently QCA have been considered for extending quantum field theory [@darianopla] to the Plank scale. Similar to lattice-gas theories, here the quantum cell corresponds to the evaluation $\psi(x)$ of a quantum field on the site $x$ of a lattice, with the dynamics updated in discrete time steps by a local unitary evolution. However, differently from lattice-gas theory, here the continuum limit is not taken, whereas, instead, the asymptotic large-scale (Fermi) evolution is considered. The main difference is then that Lorentz covariance holds exactly in the relativistic limit of small momentum, whereas generally it is distorted, in a fashion analogous to Refs. [@magueijo2003generalized; @amelino2001planck1; @amelino2001planck]. In this context the one dimensional Dirac automaton has been derived from symmetry principles for the QCA [@BDTqcaI] showing how the usual Dirac dynamics emerges at the Fermi scale, though relativistic covariance and other symmetries are violated at the Planck/ultrarelativistic scale.
In the present paper we analyze in detail the one-particle sector of the automaton of Refs. [@darianopla; @BDTqcaI]. Here, particle states are “smooth” states peaked around a momentum eigenstate of the QCA. We will consider dynamical quantities as the particle position, momentum and velocity, along with their evolution both in the free case and in the presence of a potential, recovering typical features of Dirac quantum field evolution–as [*Zitterbewegung*]{} and [ *Klein paradox*]{}–from the pure quantum information processing of the QCA. Recently there has been a renewed interest in Dirac features in solid-state and atomic physics, which provide a physical hardware to simulate the dynamics. Zitterbewegung can be seen in the response of electrons to external fields [@huang1952zitterbewegung] and can appear for nonrelativistic particles in a crystal [@cannata1991effects; @ferrari1990nonrelativistic; @cannata1990dirac], quasiparticles in superconductors [@lurie1970zitterbewegung] and systems with spin-orbit coupling [@PhysRevLett.95.187203; @PhysRevLett.99.076603]. Proving that the oscillation behavior is not unique to Dirac electrons, but rather is a generic feature of spinor systems with linear dispersion relations, these works opened the way for possible simulation of Zitterbewegung using for example trapped ions [@lamata2007dirac; @gerritsma2010quantum], two-band crystalline structure such as graphene [@cserti2006unified; @rusin2007transient] or semiconductors [@schliemann2005zitterbewegung; @zawadzki2005zitterbewegung; @zawadzki2010nature; @geim2007rise; @zawadzki2011zitterbewegung], ultra cold atoms [@vaishnav2008obserVing], and finally photonic crystals [@PhysRevLett.100.113903]. On the other hand, the Klein paradox (tunneling of relativistic particles) provides insight in the mechanics of relativistic particles propagating through potential barriers, along with vacuum polarization effects, and has been a focus in the hot topic of graphene as a simulator for Dirac equation, as in Ref. [@katsnelson2007graphene], and [@gerritsma2010quantum] for trapped ions. Recently also microfabricated optical waveguide circuits have become an alternative physical simulator for particle dynamics [@sansoni2012two].
After reviewing the Dirac QCA in 1d in Section \[s:DiracQCA\], in Section \[sec:zitterbewegung\] we present the evolution of position and momentum operators for the automaton, showing the Zitterbewegung behaviour produced by the interference between positive and negative frequencies. In Section \[s:barrier\] we modify the QCA in order to insert a potential in the free evolution, and show the automaton dynamics in the presence of a barrier for one particle states.
We end the paper with a summary and some concluding remarks in Section\[s:concl\].
The Dirac Automaton {#s:DiracQCA}
===================
The quantum automaton corresponding to the Dirac equation in 1d, first introduced in [@darianopla], has been derived from the discrete automaton symmetries of parity and time-reversal in Ref. [@BDTqcaI], where also the Dirac equation has been recovered as the large-scale relativistic limit of the automaton. The cell of the quantum automaton is given by the evaluation $\psi(x)$ of the two-component field operator $\psi$, and the unitary evolution of one step of the automaton is given by $$\begin{aligned}
\psi(x)\to U\psi(x),\quad
\psi(x) :=
\left(\begin{array}{c}
\psi_r(x) \\
\psi_l(x)
\end{array}
\right)\quad \end{aligned}$$ where $\psi_l$ and $\psi_r$ denote the [*left*]{} and [*right*]{} mode of the field, whereas the unitary matrix $U$ is given by $$\label{U}
U=\begin{pmatrix}
n S &-im\\
-im & n S^\dag
\end{pmatrix},\\
\quad n^2+m^2=1,$$ with $S$ denoting the shift operator $S f(x) = f(x+1)$. The constants $n$ and $m$ in the last equation can be chosen positive. As shown in Refs. [@darianopla; @BDTqcaI], the parameter $m$ plays the role of an a-dimensional inertial mass, and is bounded by unit. We remark that the automaton description is completely a-dimensional, and a conversion to the usual physical dimensions needs a length, a time and a mass, which one can take as the Planck length $\ell_P$, the Planck time $\tau_P$, and the Planck mass $m_P$, the latter playing the role of the bound for the inertial mass. The maximal speed of propagation of information is one cell per step ($c=\ell_P/\tau_P$ in dimensional units, corresponding to the speed of light). The quantum field can be taken generally as Fermionic, Bosonic, or even Anyonic. However, in the present case it will not be relevant, since we will consider only single-particle states, which span the Hilbert space $\mathbb{C}^2\otimes
l_2(\Z)$, and for which we will use the factorized orthonormal basis $\ket{s}\ket{x}$, where for $\ket{s}$ we consider the canonical basis corresponding to $s=l,r$. These states can be also obtained as $\psi^\dag_s(x)\ket{\Omega}$ upon introducing a vacuum $|\Omega\>$ which is annihilated by the field operator, and invariant under the automaton evolution. Similarly also $N$-particle states with $N>1$ can be obtained by acting with products of $N$ evaluations of the field operator, building up the Fock space in the usual way. Notice that the evolution of the field is restricted to be linear, and there exists a unitary operator $U$ such that the field evolution is given by $V\psi_s(x)V^\dag=U\psi_s(x)$, with $V|\Omega\>=|\Omega\>$, whereas for product of field evaluations the evolution is given by tensor powers of $U$ as $V\psi_{s_1}(x_1) \ldots
\psi_{s_N}(x_N) V^\dag=U^{\otimes N}\psi_{s_1}(x_1)\otimes
\ldots\otimes\psi_{s_N}(x_N)$.
In the $\ket{s}\ket{x}$ representation the unitary matrix $U$ can be written as follows
$$\begin{aligned}
U := \sum_x
\begin{pmatrix}
n \ketbra{x-1}{x} & -im \ketbra{x}{x} \\
-im \ketbra{x}{x} & n \ketbra{x+1}{x}
\end{pmatrix},\end{aligned}$$
describing a [*Quantum Walk*]{} on the Hilbert space $\mathbb{C}^2\otimes l_2(\Z)$ [@ambainis2001one].
Tanks to the translational invariance of $U$, it is convenient to move to the momentum representation $$\begin{aligned}
\label{vk}
\ket{\psi_s}\ket{k} :=\frac{1}{\sqrt{2\pi}} \sum_{x} e^{-ikx} \ket{\psi_s}\ket{x},\quad k\in[-\pi,\pi],\end{aligned}$$ and $U$ becomes $$\begin{aligned}
\label{eq:automaton-Uk}
U=
\int_{\minus \pi}^{\pi} \d k U(k)\otimes\ketbra{k}{k},\:\, {U}(k)=
\begin{pmatrix}
n e^{ik} & -i m\\
-i m & n e^{-ik}
\end{pmatrix}.\end{aligned}$$ Notice that discreteness bound momenta to the Brillouin zone, as in solid-state theory. By diagonalizing the unitary matrix ${U}(k)$ $$\begin{aligned}
\label{eq:eigenstates}
&U(k) \ket{s}_k = e^{-is \omega(k)}\ket{s}_k,\qquad \omega(k) = \arccos(n \cos k)\\\nonumber
&\ket{s}_k:=\tfrac{1}{\sqrt{2}}
\begin{bmatrix}
\sqrt{1-sv(k)}\\s\sqrt{1+sv(k)}
\end{bmatrix}, \quad s=\pm,\quad v(k) := \partial_k \omega(k)\end{aligned}$$ it is easy to evaluate the logarithm of $U$ ($e^{-i H}:= U$) as follows $$\begin{aligned}
\label{eq:hamiltonian}
H = \int_{-\pi}^{\pi} \d k H(k)&\otimes\ketbra{k}{k},\\\nonumber
H(k) &=\omega(k)\left( \ket{+}_k\bra{+}_k -
\ket{-}_k\bra{-}_k\right) \\\nonumber
&=\mathrm{sinc^{-1}}{\omega(k)} (-n\sin k\, \sigma_3+m\sigma_1),\end{aligned}$$ where $\sigma_i$ $i=1,2,3$ denote the usual Pauli matrices.
The function $\omega(k)$ is the dispersion relation of the automaton, which recovers the usual Dirac one $\omega(k)=\sqrt{k^2+m^2}$ in the limit $k,m\ll 1$ and $k/m\gg 1$ as shown in [@BDTqcaI]. This is also clear in Fig. \[fig:disp\] where the dispersion relation as a function of $k$ is reported for four different values of the mass. The derivative $v(k)$ in Eq. (\[eq:eigenstates\]) is then the group velocity of the wavepacket. The $s=+1$ eigenvalues correspond to positive-energy particle states, whereas the negative $s=-1$ eigenvalues correspond to negative-energy anti-particle states.
Notice that the operator $H$ regarded as an Hamiltonian would interpolate the evolution to continuous time as $U(t)\equiv U^t$, which, however, in this context should be considered unphysical [^1]
![The Dirac automaton dispersion relation in Eq. for four different values of the mass: $m=0.1,\,0.2,\,0.4,\,0.8$.[]{data-label="fig:disp"}](DispRel-0_1.pdf "fig:"){width=".21\textwidth"}![The Dirac automaton dispersion relation in Eq. for four different values of the mass: $m=0.1,\,0.2,\,0.4,\,0.8$.[]{data-label="fig:disp"}](DispRel-0_2.pdf "fig:"){width=".21\textwidth"} ![The Dirac automaton dispersion relation in Eq. for four different values of the mass: $m=0.1,\,0.2,\,0.4,\,0.8$.[]{data-label="fig:disp"}](DispRel-0_4.pdf "fig:"){width=".21\textwidth"}![The Dirac automaton dispersion relation in Eq. for four different values of the mass: $m=0.1,\,0.2,\,0.4,\,0.8$.[]{data-label="fig:disp"}](DispRel-0_8.pdf "fig:"){width=".21\textwidth"}
In the following sections we will analyze two typical aspects of the Dirac-field dynamics, namely the Zitterbewegung and the Klein paradox.
Position and Momentum Operators and Zitterbewegung {#sec:zitterbewegung}
==================================================
The QCA (\[U\]) describes very precisely the Dirac field dynamics for customary relativistic wavevectors and energies (consider that e.g. a ultra-high-energy cosmic ray has $k\simeq 10^{-8}$) [@BDTqcaI]. In this section we will show how efficiently it reproduces a typical feature of the one-particle Dirac dynamics, namely the Zitterbewegung.
The Zitterbewegung was first recognized by Schrödinger in 1930 [@schrodinger1930kraftefreie] who noticed that in the Dirac equation describing the free relativistic electron the velocity operator does not commute with the Dirac Hamiltonian: the evolution of the position operator, in addition to the classical motion shows a very fast periodic oscillation with frequency $2mc^2$ and amplitude equal to the Compton wavelength $\hbar/mc$ with $m$ the rest mass of the relativistic particle. This jittering motion first encountered in the Dirac theory of the electron was then shown [@huang1952zitterbewegung] to arise from the interference of states corresponding to the positive and negative energies resulting from the Dirac equation with the trembling disappearing with time [@lock1979zitterbewegung] for a wavepacked particle state. Zitterbewegung oscillations cannot be directly observed by current experimental techniques for a Dirac electron since the amplitude should by very small $\approx 10^{-12}$ m. However, it can be seen in a number of solid-state, atomic-physics, photonic-cristal and optical waveguide simulators, as quoted in the introduction.
The “position” operator $X$ providing the representation $|x\>$ (i.e. such that $X|s\>|x\>=x|s\>|x\>$) is defined as follows $$\begin{aligned}
X=\sum_{x\in\mathbb{Z}}x(I\otimes\ketbra{x}{x}).\end{aligned}$$ Generally $X$ provides the average location of a wavepacket in terms of $\<\psi|X|\psi\>$. The conjugated “momentum” operator is given by $$\begin{aligned}
P=\int_{-\pi}^{\pi}\;\df{k}{2\pi} k(I\otimes\ketbra{k}{k}).\end{aligned}$$
One can verify that $X$ and $P$ obey the usual canonical commutation rule $[X,P]=i$. In the following it will be convenient to work with the continuous time $t$ interpolating exactly the discrete automaton evolution, namely $U^t$. However, all numerical results will be given only for discrete $t$, namely for repeated applications of the automaton unitary $U$ in Eq. (\[U\]).
The time evolution of the position operator ${X}(t)={U}^{ - t} {X}{U}^{t}$ can be more easily computed by integrating the differential equation $A(t) = [H,[H,X(t)]]$ where $H$ was defined in Eq. . We have $$\begin{aligned}
\nonumber
&A(t) = \int_{-\pi}^{\pi} \d k A(k,t) \otimes\ketbra{k}{k} \quad A(k,t) =
e^{2i{H(k)}t}{A}(k)\\
&A(k) = -\frac{2\omega^2}{\sin^2\omega}nm\cos{k}\,
\sigma_2\end{aligned}$$ which leads to $$\begin{aligned}
\label{e:Zx}
{X}(t)={X}(0)+{V} t+{Z}_{{X}}(t)-{Z}_{{X}}(0)\\
{V}(k)=-v(k)^2\sigma_3+v(k)\sqrt{1-v(k)^2}\sigma_1\\
{Z}_{{X}}(k,t)=-\frac{1}{4}{H}^{-2}(k){A}(k,t)\end{aligned}$$ where $V$ is the classical component of the velocity operator which, in the base diagonalizing the Hamiltonian , is $V(k)=v(k)\sigma_3$ and is proportional to the group velocity $v(k)$. Since a generic one-particle state $\ket{\psi}$ is a superposition of a positive and a negative energy state, i.e. $\ket{\psi_+} + \ket{\psi_-}$, the evolution of the mean value of the position operator $X(t)$, can be written as $$\begin{aligned}
\nonumber
x_\psi(t) := \bra{\psi} X(t) \ket{\psi} =
x_{\psi}^+(t) + x_{\psi}^-(t) + x_{\psi}^{\rm{int}}(t)\\\nonumber
x_{\psi}^{\pm}(t) : = \bra{\psi _\pm} X(0) + Vt \ket{\psi_\pm}\\\label{e:zitterbewegung}
x_{\psi}^{\rm{int}}(t) :=
2 \Re [\bra{\psi_+} X(0) - {Z}_{{X}}(0) + {Z}_{{X}}(t) \ket{\psi_-}]\end{aligned}$$ where $\Re$ denotes the real part. The interference between positive and negative frequency is responsible of the $x_{\psi}^{\rm{int}}(t)$. The magnitude of $x_{\psi}^{\rm{int}}(t)$ is bounded by $1/m$ (see appendix \[a:zitterbewegung\]) which in the usual dimensional units correspond to the Compton wavelength $\hbar/ m c$. Moreover the stationary phase approximation shows that for $t \to \infty$ the term $2 \Re [\bra{\psi_+} {Z}_{{X}}(t) \ket{\psi_-}]$, which is responsible of the oscillation, goes to $0$ as $1/\sqrt{t}$ (see appendix \[a:zitterbewegung\]) and only the shift contribution coming from $2 \Re [\bra{\psi_+} X(0) - {Z}_{{X}}(0)
\ket{\psi_-}]$ survives. These results show that $x_{\psi}^{\rm{int}}(t)$ is the automaton analogue of the so-called Zitterbewegung. As already noticed in the introduction this phenomenon was never observed for a free relativistic electron because of the small value of $x_{\psi}^{\rm{int}}(t)$ which is bounded by the electron Compton wave length $10^{-12}$ m in natural units. The results of this section are in agreement with the one for the Hadamard walk [@kurzynski2008relativistic].
In Fig. \[fig:Zitt11-12-21-22\] we have considered the evolution of states with particle and antiparticle components smoothly peaked around some momentum eigenstate, namely $$\begin{aligned}
\label{e:states}
c_+\ket{\psi_+}+c_-\ket{\psi_-},\quad \ket{\psi_\pm}= \int
\df{k}{\sqrt{2\pi}} g_{k_0}(k) \ket{\pm}_k\ket{k}\end{aligned}$$ where $c_+^2 \mkern-6mu+\mkern-6mu c_-^2 \mkern-14mu = \mkern-14mu 1$ and $g_{k_0}$ is a Gaussian peaked around the momentum $k_0$ with width $\sigma$. An easy computation shows that for these states the shift contribution reduces to $2\Re[\bra{\psi}X(0)+
Z_{X}(0)\ket{\psi}]=\Im{(c^{*}_+c_-)}/(2\pi)\int_{-\pi}^{\pi}\,\d{k}
|g_{k_0}(k)|^2 z(k)$ with the function $z(k)=m\cos\omega(k)/\sin^2\omega(k)$ bounded again by the Compton wavelength $1/m$ and the oscillation frequency given by $\omega(0)/\pi$ (see also Fig. \[fig:z-frequency\]).
![(Colors online) Plots of $z(k)$ (left) and $\omega(k)/\pi$ (right) related to the oscillation amplitude and frequency of the position expectation value in Eq. . In both cases the plots are reported for different values of the mass ($m=0.1,\,0.2,\,0.4,\,0.8$ from the top in the figure on the left and from the bottom in the figure on the right).[]{data-label="fig:z-frequency"}](z.pdf "fig:"){width=".23\textwidth"}![(Colors online) Plots of $z(k)$ (left) and $\omega(k)/\pi$ (right) related to the oscillation amplitude and frequency of the position expectation value in Eq. . In both cases the plots are reported for different values of the mass ($m=0.1,\,0.2,\,0.4,\,0.8$ from the top in the figure on the left and from the bottom in the figure on the right).[]{data-label="fig:z-frequency"}](frequency.pdf "fig:"){width=".23\textwidth"}
![ Automaton evolution of a state as in Eq. showing the Zitterbewegung of the position expectation value. [ **Top:**]{} $m=0.15$, $c_+=1/ \sqrt{2}$, $c_-=i/\sqrt{2}$, $k_0=0$, and $\sigma=40^{-1}$. The calculated shift and oscillation frequency are respectively $\bra{\psi} X(0)+ Z_{X}(0)\ket{\psi}=3.2$ and $\omega(0)/\pi=0.05$, accordingly to the simulation. [**Middle:**]{} $m=0.15$, $c_+=1/\sqrt{2}$, $c_-=1/\sqrt{2}$, $k_0=0$, $\sigma=40^{-1}$. The calculated shift and oscillation frequency are $0$ and $0.13$, respectively. [**Bottom:**]{} $m=0.13$, $c_+=\sqrt{2/3}$, $c_-=1/\sqrt{3}$, $k_0=10^{-2}\pi$, $\sigma=40^{-1}$. In this case the particle and antiparticle contribution are not balanced and the average position drift velocity is thus $\bra{\psi _+} V \ket{\psi_+}+\bra{\psi _-} V
\ket{\psi_-}=(|c_+|^2-|c_-|^2)v(k_0)=0.08$, corresponding to an average position $x_{\psi}^+(800) + x_{\psi}^-(800)=464$ (see Eq. ). Notice that for $t \to \infty$ the term $2 \Re [\bra{\psi_+} {Z}_{{X}}(t) \ket{\psi_-}$, which is responsible of the oscillation, goes to $0$.[]{data-label="fig:Zitt11-12-21-22"}](Zitt11.pdf "fig:"){width=".23\textwidth"}![ Automaton evolution of a state as in Eq. showing the Zitterbewegung of the position expectation value. [ **Top:**]{} $m=0.15$, $c_+=1/ \sqrt{2}$, $c_-=i/\sqrt{2}$, $k_0=0$, and $\sigma=40^{-1}$. The calculated shift and oscillation frequency are respectively $\bra{\psi} X(0)+ Z_{X}(0)\ket{\psi}=3.2$ and $\omega(0)/\pi=0.05$, accordingly to the simulation. [**Middle:**]{} $m=0.15$, $c_+=1/\sqrt{2}$, $c_-=1/\sqrt{2}$, $k_0=0$, $\sigma=40^{-1}$. The calculated shift and oscillation frequency are $0$ and $0.13$, respectively. [**Bottom:**]{} $m=0.13$, $c_+=\sqrt{2/3}$, $c_-=1/\sqrt{3}$, $k_0=10^{-2}\pi$, $\sigma=40^{-1}$. In this case the particle and antiparticle contribution are not balanced and the average position drift velocity is thus $\bra{\psi _+} V \ket{\psi_+}+\bra{\psi _-} V
\ket{\psi_-}=(|c_+|^2-|c_-|^2)v(k_0)=0.08$, corresponding to an average position $x_{\psi}^+(800) + x_{\psi}^-(800)=464$ (see Eq. ). Notice that for $t \to \infty$ the term $2 \Re [\bra{\psi_+} {Z}_{{X}}(t) \ket{\psi_-}$, which is responsible of the oscillation, goes to $0$.[]{data-label="fig:Zitt11-12-21-22"}](Zitt12.pdf "fig:"){width=".23\textwidth"} ![ Automaton evolution of a state as in Eq. showing the Zitterbewegung of the position expectation value. [ **Top:**]{} $m=0.15$, $c_+=1/ \sqrt{2}$, $c_-=i/\sqrt{2}$, $k_0=0$, and $\sigma=40^{-1}$. The calculated shift and oscillation frequency are respectively $\bra{\psi} X(0)+ Z_{X}(0)\ket{\psi}=3.2$ and $\omega(0)/\pi=0.05$, accordingly to the simulation. [**Middle:**]{} $m=0.15$, $c_+=1/\sqrt{2}$, $c_-=1/\sqrt{2}$, $k_0=0$, $\sigma=40^{-1}$. The calculated shift and oscillation frequency are $0$ and $0.13$, respectively. [**Bottom:**]{} $m=0.13$, $c_+=\sqrt{2/3}$, $c_-=1/\sqrt{3}$, $k_0=10^{-2}\pi$, $\sigma=40^{-1}$. In this case the particle and antiparticle contribution are not balanced and the average position drift velocity is thus $\bra{\psi _+} V \ket{\psi_+}+\bra{\psi _-} V
\ket{\psi_-}=(|c_+|^2-|c_-|^2)v(k_0)=0.08$, corresponding to an average position $x_{\psi}^+(800) + x_{\psi}^-(800)=464$ (see Eq. ). Notice that for $t \to \infty$ the term $2 \Re [\bra{\psi_+} {Z}_{{X}}(t) \ket{\psi_-}$, which is responsible of the oscillation, goes to $0$.[]{data-label="fig:Zitt11-12-21-22"}](Zitt21.pdf "fig:"){width=".23\textwidth"}![ Automaton evolution of a state as in Eq. showing the Zitterbewegung of the position expectation value. [ **Top:**]{} $m=0.15$, $c_+=1/ \sqrt{2}$, $c_-=i/\sqrt{2}$, $k_0=0$, and $\sigma=40^{-1}$. The calculated shift and oscillation frequency are respectively $\bra{\psi} X(0)+ Z_{X}(0)\ket{\psi}=3.2$ and $\omega(0)/\pi=0.05$, accordingly to the simulation. [**Middle:**]{} $m=0.15$, $c_+=1/\sqrt{2}$, $c_-=1/\sqrt{2}$, $k_0=0$, $\sigma=40^{-1}$. The calculated shift and oscillation frequency are $0$ and $0.13$, respectively. [**Bottom:**]{} $m=0.13$, $c_+=\sqrt{2/3}$, $c_-=1/\sqrt{3}$, $k_0=10^{-2}\pi$, $\sigma=40^{-1}$. In this case the particle and antiparticle contribution are not balanced and the average position drift velocity is thus $\bra{\psi _+} V \ket{\psi_+}+\bra{\psi _-} V
\ket{\psi_-}=(|c_+|^2-|c_-|^2)v(k_0)=0.08$, corresponding to an average position $x_{\psi}^+(800) + x_{\psi}^-(800)=464$ (see Eq. ). Notice that for $t \to \infty$ the term $2 \Re [\bra{\psi_+} {Z}_{{X}}(t) \ket{\psi_-}$, which is responsible of the oscillation, goes to $0$.[]{data-label="fig:Zitt11-12-21-22"}](Zitt22.pdf "fig:"){width=".23\textwidth"} ![ Automaton evolution of a state as in Eq. showing the Zitterbewegung of the position expectation value. [ **Top:**]{} $m=0.15$, $c_+=1/ \sqrt{2}$, $c_-=i/\sqrt{2}$, $k_0=0$, and $\sigma=40^{-1}$. The calculated shift and oscillation frequency are respectively $\bra{\psi} X(0)+ Z_{X}(0)\ket{\psi}=3.2$ and $\omega(0)/\pi=0.05$, accordingly to the simulation. [**Middle:**]{} $m=0.15$, $c_+=1/\sqrt{2}$, $c_-=1/\sqrt{2}$, $k_0=0$, $\sigma=40^{-1}$. The calculated shift and oscillation frequency are $0$ and $0.13$, respectively. [**Bottom:**]{} $m=0.13$, $c_+=\sqrt{2/3}$, $c_-=1/\sqrt{3}$, $k_0=10^{-2}\pi$, $\sigma=40^{-1}$. In this case the particle and antiparticle contribution are not balanced and the average position drift velocity is thus $\bra{\psi _+} V \ket{\psi_+}+\bra{\psi _-} V
\ket{\psi_-}=(|c_+|^2-|c_-|^2)v(k_0)=0.08$, corresponding to an average position $x_{\psi}^+(800) + x_{\psi}^-(800)=464$ (see Eq. ). Notice that for $t \to \infty$ the term $2 \Re [\bra{\psi_+} {Z}_{{X}}(t) \ket{\psi_-}$, which is responsible of the oscillation, goes to $0$.[]{data-label="fig:Zitt11-12-21-22"}](Zitt31.pdf "fig:"){width=".23\textwidth"}![ Automaton evolution of a state as in Eq. showing the Zitterbewegung of the position expectation value. [ **Top:**]{} $m=0.15$, $c_+=1/ \sqrt{2}$, $c_-=i/\sqrt{2}$, $k_0=0$, and $\sigma=40^{-1}$. The calculated shift and oscillation frequency are respectively $\bra{\psi} X(0)+ Z_{X}(0)\ket{\psi}=3.2$ and $\omega(0)/\pi=0.05$, accordingly to the simulation. [**Middle:**]{} $m=0.15$, $c_+=1/\sqrt{2}$, $c_-=1/\sqrt{2}$, $k_0=0$, $\sigma=40^{-1}$. The calculated shift and oscillation frequency are $0$ and $0.13$, respectively. [**Bottom:**]{} $m=0.13$, $c_+=\sqrt{2/3}$, $c_-=1/\sqrt{3}$, $k_0=10^{-2}\pi$, $\sigma=40^{-1}$. In this case the particle and antiparticle contribution are not balanced and the average position drift velocity is thus $\bra{\psi _+} V \ket{\psi_+}+\bra{\psi _-} V
\ket{\psi_-}=(|c_+|^2-|c_-|^2)v(k_0)=0.08$, corresponding to an average position $x_{\psi}^+(800) + x_{\psi}^-(800)=464$ (see Eq. ). Notice that for $t \to \infty$ the term $2 \Re [\bra{\psi_+} {Z}_{{X}}(t) \ket{\psi_-}$, which is responsible of the oscillation, goes to $0$.[]{data-label="fig:Zitt11-12-21-22"}](Zitt32.pdf "fig:"){width=".23\textwidth"}
Evolution with a square potential barrier {#s:barrier}
=========================================
In order to study the scattering with a potential, we modify the automaton adding a position dependent phase representing a square potential barrier, as in Refs. [@kurzynski2008relativistic; @meyer1997quantum]. We will provide explicitly the transmission $T$ and reflection $R$ coefficients as functions of the energy and mass of the incident wavepacket and of the potential barrier’s height. We will find a general behavior independently on the regime, namely on the energy and mass of the incident particle. Increasing the value of the potential barrier beyond a certain threshold a transmitted wave reappears and the reflection coefficient starts decreasing. The width of the $R=1$ region is an increasing function of the mass which is proportional to the gap between positive and negative frequency eigenvalues of the unitary evolution.
For a generic potential $\phi(x)$, the unitary evolution becomes $$\begin{aligned}
U_\phi := \sum_x e^{-i \phi(x)}
\left(
\begin{array}{ll}
n \ketbra{x-1}{x} & -im \ketbra{x}{x} \\
-im \ketbra{x}{x} & n \ketbra{x+1}{x}
\end{array}
\right).\end{aligned}$$ We will analyze the simple case $\phi(x) := \phi \, \theta(x)$ ($\theta(x)$ is the Heaviside step function) that is a potential step which is $0$ for $x <0 $ (region $\mathrm{I}$) and has a constant value $\phi \in
[0, 2\pi]$ for $x \geq 0$ (region $\mathrm{II}$) as illustrated in Fig. \[fig:potential\].
![Schematic of the potential[]{data-label="fig:potential"}](potential6.pdf){width=".30\textwidth"}
![Reflection coefficient as a function of the potential barrier height $\phi$ and of the momentum $k$ of the incident particle state. From the top-left to the bottom-right the reflection coefficient is depicted for different values of the mass: $m=0.1,\,0.2,\,0.4,\,0.8$.[]{data-label="fig:Reflection"}](R-0_1.pdf "fig:"){width=".23\textwidth"}![Reflection coefficient as a function of the potential barrier height $\phi$ and of the momentum $k$ of the incident particle state. From the top-left to the bottom-right the reflection coefficient is depicted for different values of the mass: $m=0.1,\,0.2,\,0.4,\,0.8$.[]{data-label="fig:Reflection"}](R-0_2.pdf "fig:"){width=".23\textwidth"} ![Reflection coefficient as a function of the potential barrier height $\phi$ and of the momentum $k$ of the incident particle state. From the top-left to the bottom-right the reflection coefficient is depicted for different values of the mass: $m=0.1,\,0.2,\,0.4,\,0.8$.[]{data-label="fig:Reflection"}](R-0_4.pdf "fig:"){width=".23\textwidth"}![Reflection coefficient as a function of the potential barrier height $\phi$ and of the momentum $k$ of the incident particle state. From the top-left to the bottom-right the reflection coefficient is depicted for different values of the mass: $m=0.1,\,0.2,\,0.4,\,0.8$.[]{data-label="fig:Reflection"}](R-0_8.pdf "fig:"){width=".23\textwidth"}
![Group velocity of the transmitted wave packet as a function of the potential barrier height $\phi$ and of the momentum $k$ of the incident particle state. From the top-left to the bottom-right the transmitted group velocity for different values of the mass: $m=0.1,\,0.2,\,0.4,\,0.8$.[]{data-label="fig:Momentum"}](Vt-0_1.pdf "fig:"){width=".23\textwidth"}![Group velocity of the transmitted wave packet as a function of the potential barrier height $\phi$ and of the momentum $k$ of the incident particle state. From the top-left to the bottom-right the transmitted group velocity for different values of the mass: $m=0.1,\,0.2,\,0.4,\,0.8$.[]{data-label="fig:Momentum"}](Vt-0_2.pdf "fig:"){width=".23\textwidth"} ![Group velocity of the transmitted wave packet as a function of the potential barrier height $\phi$ and of the momentum $k$ of the incident particle state. From the top-left to the bottom-right the transmitted group velocity for different values of the mass: $m=0.1,\,0.2,\,0.4,\,0.8$.[]{data-label="fig:Momentum"}](Vt-0_4.pdf "fig:"){width=".23\textwidth"}![Group velocity of the transmitted wave packet as a function of the potential barrier height $\phi$ and of the momentum $k$ of the incident particle state. From the top-left to the bottom-right the transmitted group velocity for different values of the mass: $m=0.1,\,0.2,\,0.4,\,0.8$.[]{data-label="fig:Momentum"}](Vt-0_8.pdf "fig:"){width=".23\textwidth"}
Let us now study the eigenvector of $U_\phi$ of the form $$\begin{aligned}
&\ket{\Phi_k} = \Pi_\mathrm{I} \ket{+}_k\ket{k} +\Pi_\mathrm{I} \beta_k
\ket{+}_{-k}\ket{k} + \gamma_k \Pi_\mathrm{II}\ket{+}_{k'}\ket{k'} \\
&\Pi_\mathrm{I} := \sum _{x<0} I \otimes \ketbra{x}{x} \qquad
\Pi_\mathrm{II} := \sum _{x \geq 0} I \otimes \ketbra{x}{x}
\end{aligned}$$ where $\beta_k$, $\gamma_k$ and $k'$ are functions of $k$. The condition that $\ket{\Phi_k}$ is genuinely an eigenstate of $U_\phi$, i.e. $ U_\phi \ket{\Phi_k} = e^{-i \omega(k)}
\ket{\Phi_k}$, implies that $$\begin{aligned}
&\omega(k') = \omega(k) -\phi \label{eq:kprime}\\
&\beta_k=\frac{e^{-ik}\sqrt{(1+ v)(1-v^\prime)}-e^{-ik^\prime}\sqrt{(1-
v)(1+v^\prime)}}{-e^{ik}\sqrt{(1- v)(1-
v^\prime)}+e^{-ik^\prime}\sqrt{(1+v)(1+v^\prime)}} \nonumber \\
&\nonumber
\gamma_k =\frac{2e^{i\xi} (v\cos{k}-i\sin{k})}{-e^{ik}\sqrt{(1- v)(1- v^\prime))}+e^{-ik^\prime}\sqrt{(1+v)(1+v^\prime)}}\end{aligned}$$ with $v:=v(k)$ and $v':=v(k')$ the group velocities of the incident and transmitted wave. Let us now consider the superposition $$\begin{aligned}
\ket{\Psi(0)}:= \int \df{k}{\sqrt{2\pi}} g_{k_0}(k) \ket{\Phi_k}\end{aligned}$$ where $g_{k_0}(k)$ is a function in $C_0^{\infty}[-\pi, \pi]$ which we assumed to be smoothly peaked around $k_0$. The state at time $t$ is then $$\begin{aligned}
\ket{\Psi(t)}:= \int \df{k}{\sqrt{2\pi}} g_{k_0}(k) e^{-i \omega(k)t} \ket{\Phi_k}\end{aligned}$$ and one can verify that for $t \ll 0$ the state is negligible in region $\mathrm{II}$ while the only appreciable contribution in region $\mathrm{II}$ comes from the term $e^{i k_0 x}$ which describes a wavepacket that moves at group velocity $v(k_0)$ and hits the barrier form the left. When $t \gg 0$ the state can be approximated by a superposition of a reflected and a transmitted wavepacket as follows $$\begin{aligned}
\begin{split}
\ket{\Psi(t)} \xrightarrow{ t \gg 0} \beta(k_0) \int \df{k}{\sqrt{2\pi}} g_{k_0}(k) e^{-i \omega(k)t} \ket{+}_{-k}\ket{k} +\\
{}+ \tilde{\gamma}(k_0) e^{-i \phi t} \int \df{k}{\sqrt{2\pi}} \tilde{g}_{k'_0}(k') e^{-i
\omega(k')t} \ket{+}_{k'}\ket{k'}
\end{split}\end{aligned}$$ where we defined $$\begin{aligned}
&k'_0 \mbox{ s.t. } \omega(k'_0) = \omega(k_0) -\phi, \\
&\tilde{\gamma}(k_0) := {\gamma}(k_0)
\sqrt{\frac{v(k'_0)}{v(k_0)}}, \qquad
\tilde{g}_{k'_0}(k') =
\sqrt{\frac{v(k'_0)}{v(k_0)}}{g}_{k'_0}(k') \end{aligned}$$ (one can check $ \int \df{k}{\sqrt{2\pi}}
|\tilde{g}_{k'_0}(k')|^2 = 1$), whose group velocities are $-v(k_0)$ for the reflected wave packet and $v(k^\prime_0)$ for the transmitted wave packet (see Fig. \[fig:Momentum\]).
The probability of finding the particle in the reflected wavepacket is $R = |\beta(k_0)|^2$ (reflection coefficient) while the probability of finding the particle in the transmitted wavepacket is $ T= |\tilde{\gamma}(k_0)|^2$ (trasmission coefficient). The consistency of the result can be verified by checking that $R+T = 1$. For $k \ll m \ll 1 $ (Schröedinger regime) we recover the usual reflection and transmission coefficient for the Schröedinger equation with a potential step. In Fig. \[fig:Reflection\] we plot the reflection coefficient $R$ as a function of $\phi$ and $k$ for different values of the mass $m$. Clearly when $\phi=0$ we have $R = 0$ and increasing $\phi$ while fixing $k$ the value increases up to $R=1$. One notice that when $
\omega(k) - \arccos(n) < \phi < \omega(k) + \arccos(n)$ Eq. has solution for imaginary $k'$ which implies an exponential damping of the transmitted wave and pure reflection. By further increasing the value of $\phi$ beyond the threshold $\omega(k) + \arccos(n)$, Eq. have solution for real $k'$ and negative $\omega(k')$, and then a transmitted wave reappears and the reflection coefficient decreases. This is the so called “Klein paradox” which is originated by the presence of positive and negative frequency eigenvalues of the unitary evolution. The width of the $R=1$ region is an increasing function of the mass equal to $2\arccos(n)$ which is the gap between positive and negative frequency solution solutions (see Fig. \[fig:disp\] ).
![Reflection coefficient for $m=0.4$ and momentum of the incident particle $k_0=2$ as a function of the potential barrier height $\phi$ (section of plots in Figs. \[fig:Reflection\]-\[fig:Momentum\] for $m=0.4$, $k_0=2$).[]{data-label="fig:Section"}](R-m0_4-K2.pdf "fig:"){width=".23\textwidth"}![Reflection coefficient for $m=0.4$ and momentum of the incident particle $k_0=2$ as a function of the potential barrier height $\phi$ (section of plots in Figs. \[fig:Reflection\]-\[fig:Momentum\] for $m=0.4$, $k_0=2$).[]{data-label="fig:Section"}](Vt-m0_4-K2.pdf "fig:"){width=".23\textwidth"}
![Simulations of the Dirac automaton evolution with a square potential barrier. Here the automaton mass is $m=0.2$ while the barrier turns on at $x=140$. In the simulation the incident state is a smooth state of the form $\ket{\psi(0)}= \int
\df{k}{\sqrt{2\pi}} g_{k_0}(k) \ket{+}_k$ peaked around the positive energy eigenstate $\ket{+}_{k_0}$ with $k_0=2$ and with $g_{k_0}$ a Gaussian having width $\sigma=15^{-1}$. The incident group velocity is $v(k_0)=0.90$. The simulation is run for four increasing values of the potential $\phi$. [**Top-Left:**]{} Potential barrier height $\phi=1.42$, reflection coefficient $R=0.25$, velocity of the transmitted particle $v(k^\prime_0)=0.63$. [**Top-Right:**]{} $\phi=1.55$, $R=0.75$, $v(k^\prime_0)=0.1$. [**Bottom-Left:**]{} $\phi=2$, $R=0.1$, $v(k^\prime_0)=0$. [**Bottom-Right:**]{} $\phi=2.4$, $R=0.50$, $v(k^\prime_0)=0.33$.[]{data-label="fig:Simulation"}](R0_25-m0_4-V1_42-k2.pdf "fig:"){width=".23\textwidth"}![Simulations of the Dirac automaton evolution with a square potential barrier. Here the automaton mass is $m=0.2$ while the barrier turns on at $x=140$. In the simulation the incident state is a smooth state of the form $\ket{\psi(0)}= \int
\df{k}{\sqrt{2\pi}} g_{k_0}(k) \ket{+}_k$ peaked around the positive energy eigenstate $\ket{+}_{k_0}$ with $k_0=2$ and with $g_{k_0}$ a Gaussian having width $\sigma=15^{-1}$. The incident group velocity is $v(k_0)=0.90$. The simulation is run for four increasing values of the potential $\phi$. [**Top-Left:**]{} Potential barrier height $\phi=1.42$, reflection coefficient $R=0.25$, velocity of the transmitted particle $v(k^\prime_0)=0.63$. [**Top-Right:**]{} $\phi=1.55$, $R=0.75$, $v(k^\prime_0)=0.1$. [**Bottom-Left:**]{} $\phi=2$, $R=0.1$, $v(k^\prime_0)=0$. [**Bottom-Right:**]{} $\phi=2.4$, $R=0.50$, $v(k^\prime_0)=0.33$.[]{data-label="fig:Simulation"}](R0_75-m0_4-V1_55-k2.pdf "fig:"){width=".23\textwidth"} ![Simulations of the Dirac automaton evolution with a square potential barrier. Here the automaton mass is $m=0.2$ while the barrier turns on at $x=140$. In the simulation the incident state is a smooth state of the form $\ket{\psi(0)}= \int
\df{k}{\sqrt{2\pi}} g_{k_0}(k) \ket{+}_k$ peaked around the positive energy eigenstate $\ket{+}_{k_0}$ with $k_0=2$ and with $g_{k_0}$ a Gaussian having width $\sigma=15^{-1}$. The incident group velocity is $v(k_0)=0.90$. The simulation is run for four increasing values of the potential $\phi$. [**Top-Left:**]{} Potential barrier height $\phi=1.42$, reflection coefficient $R=0.25$, velocity of the transmitted particle $v(k^\prime_0)=0.63$. [**Top-Right:**]{} $\phi=1.55$, $R=0.75$, $v(k^\prime_0)=0.1$. [**Bottom-Left:**]{} $\phi=2$, $R=0.1$, $v(k^\prime_0)=0$. [**Bottom-Right:**]{} $\phi=2.4$, $R=0.50$, $v(k^\prime_0)=0.33$.[]{data-label="fig:Simulation"}](R1-m0_4-V2-k2.pdf "fig:"){width=".23\textwidth"}![Simulations of the Dirac automaton evolution with a square potential barrier. Here the automaton mass is $m=0.2$ while the barrier turns on at $x=140$. In the simulation the incident state is a smooth state of the form $\ket{\psi(0)}= \int
\df{k}{\sqrt{2\pi}} g_{k_0}(k) \ket{+}_k$ peaked around the positive energy eigenstate $\ket{+}_{k_0}$ with $k_0=2$ and with $g_{k_0}$ a Gaussian having width $\sigma=15^{-1}$. The incident group velocity is $v(k_0)=0.90$. The simulation is run for four increasing values of the potential $\phi$. [**Top-Left:**]{} Potential barrier height $\phi=1.42$, reflection coefficient $R=0.25$, velocity of the transmitted particle $v(k^\prime_0)=0.63$. [**Top-Right:**]{} $\phi=1.55$, $R=0.75$, $v(k^\prime_0)=0.1$. [**Bottom-Left:**]{} $\phi=2$, $R=0.1$, $v(k^\prime_0)=0$. [**Bottom-Right:**]{} $\phi=2.4$, $R=0.50$, $v(k^\prime_0)=0.33$.[]{data-label="fig:Simulation"}](R0_5-m0_4-V2_4-k2.pdf "fig:"){width=".23\textwidth"}
In Fig. \[fig:Section\] we plot the reflection $R$ coefficient and the transmitted wave velocity group $v(k_0')$ as a function of the potential barrier height $\phi$ with the incident wave packet having $k_0=2$ and $m=0.4$. From the figure it is clear that after a plateau with $R=1$ the reflection coefficient starts decreasing for higher potentials. In Fig. \[fig:Simulation\] we show the scattering simulation for four increasing values of the potential, say $\phi=1.42,\,1.55,\,2,\,2.4$ (see the caption to figure for the details).
Conclusions {#s:concl}
===========
In this paper we studied the dynamics of the quantum cellular automaton of Refs. [@darianopla; @BDTqcaI], which gives the Dirac dynamics as emergent in the limit of small wavevectors. We presented computer simulations and analytical evaluations, focusing on typical features of the Dirac dynamics, in particular the Zitterbewegung and the scattering from potential. Our automaton covers all regimes of masses and energy-momenta, beyond the same validity range of the Dirac equation, with the possibility of considering arbitrary input states, enabling to investigate and visualize a wide range of fundamental processes. This facts, in addition to the discreteness of the automaton, makes of it the ideal theoretical counterpart for the experimental simulators in the literature. A similar quantum cellular automaton can be also developed in two dimensions [@DP], corresponding to the graphene as quantum simulator.
Bound of the oscillating term and its asymptotic behavior {#a:zitterbewegung}
=========================================================
Here we provide an upper bound for the oscillating term $x_{\psi}^{\rm{int}}(t)$ in Eq. (\[e:zitterbewegung\]) in the position operator evolution derived in Section \[sec:zitterbewegung\] and we derive its behaviour for very long time steps. The jittering of the position expectation value is caused by the operator ${Z}_{{X}}(t)$ which in the base diagonalizing the automaton Hamiltonian $H$ can be written as $$\begin{aligned}
\nonumber
Z_X(t)=\int_{-\pi}^{\pi}\d k e^{2i\omega(k) \sigma_z
t}Z_X(k)\otimes\ketbra{k}{k},\\\nonumber
Z_X(k)=z(k)\sigma_2,\qquad
z(k)=\frac{m\cos{\omega(k)}}{2\sin^2{\omega}(k)}\end{aligned}$$ with $z(k)\in L^1(-\pi,\pi)$ for any $m\neq 0$. By defining $$\begin{aligned}
\nonumber
\ket{\psi_\pm}= \int_{-\pi}^{\pi}\;
\df{k}{\sqrt{2\pi}} g_{\pm}(k) \ket{\pm}_k\ket{k},\qquad g_{\pm}(k)\in C_0^{\infty}[-\pi, \pi]\end{aligned}$$ we have $$\begin{aligned}
\nonumber
2\Re[\bra{\psi_+}Z_{X}{(t)}\ket{\psi_-}]=\int_{-\pi}^{\pi} \df{k}{\pi} z(k)
\Re{\left[i g^*_+(k) g_-(k) e^{2i\omega(k) t}\right]}\end{aligned}$$ Since, for any $m\neq 0$, $\omega(k)$ has three stationary points in $k=0,\,\pm\pi$ ($\omega^{(1)}(0)=\omega^{(1)}(\pm\pi)=0$ and $\omega^{(1)}(k)\neq 0$ elsewhere in the closed interval $k\in[-\pi,\pi]$, with $\omega^{(2)}(0),\omega^{(2)}(\pm\pi)\neq
0$), the stationary phase approximation gives $$\begin{gathered}
\label{e:asymptotic}\nonumber
2\Re[\bra{\psi_+}Z_{X}{(t)}\ket{\psi_-}]\xrightarrow{ t \gg 0}\\\nonumber
\qquad\qquad\sum_{k=0,\pm\pi} z(k)\Re{\left[ig^*_+(k) g_-(k)
e^{2i\omega(k) t}\sqrt{\frac{i}{\pi\omega^{(2)}(k)t}}\;\right]}\end{gathered}$$ showing that the term $2 \Re [\bra{\psi_+} {Z}_{{X}}(t)
\ket{\psi_-}]$, goes to $0$ as $1/\sqrt{t}$.
In order to find an upper bound for $x_{\psi}^{\rm{int}}(t)$ notice that $$\begin{aligned}
\nonumber
| x_{\psi}^{\rm{int}}(t)|&\leq 2 | \bra{\psi_+} X(0) - {Z}_{{X}}(0)
+ {Z}_{{X}}(t) \ket{\psi_-}|\\\nonumber
&\leq 2(|\bra{\psi_+} X(0)
\ket{\psi_-}| + | {Z}_{{X}}(0)| + |{Z}_{{X}}(t)|)
\end{aligned}$$ and, according to the expression of $Z_X(k)$, we get $$\begin{aligned}
\nonumber
| {Z}_{{X}}(0)| + |{Z}_{{X}}(t)| \leq 2| {Z}_{{X}}(0)|\\\nonumber
| {Z}_{{X}}(0)|\leq \max_{k\in[-\pi,\pi]} |z(k)|=z(0)=\frac{\sqrt{1-m^2}}{2m}.\end{aligned}$$ Now defining the $C_0^{\infty}[-\pi, \pi]$ test function $\varphi(k,k')=g^*_+(k) g_-(k')
\bra{+}{}_k \ket{-}_{k'}$, we have $$\begin{aligned}
\nonumber
&|\bra{\psi_+} X(0) \ket{\psi_-}| =\\\nonumber
&\left|\BraKet{
\frac
{\;{\mathop{\!\! \mathrm{d}}}\delta(k-k')}{\;{\mathop{\!\! \mathrm{d}}}(k-k')}
}{\varphi(k,k')}\right|=
\left|\BraKet{\delta(k-k')}
{
\frac
{\;{\mathop{\!\! \mathrm{d}}}\varphi(k,k')}{\;{\mathop{\!\! \mathrm{d}}}(k-k')}
}\right|=\\\nonumber
&\left|\int_{-\pi}^{\pi} \;\df{k}{2\pi}\, \d k' \delta(k-k') g^*_+(k) g_-(k')
\frac{\;{\mathop{\!\! \mathrm{d}}}}{\;{\mathop{\!\! \mathrm{d}}}(k-k')}\bra{+}{}_k \ket{-}_{k'}\right|=\\\nonumber
&\left|\int_{-\pi}^{\pi}\; \df{k}{2\pi} g^*_+(k) g_-(k) f(k) \right|
\leq \max_{k\in[-\pi,\pi]} |f(k)|=f(0)\\\nonumber
&f(k):=\frac{n}{\sin^2{\omega}},\qquad f(0)= \frac{\sqrt{1-m^2}}{m^2}.\end{aligned}$$ which finally gives $$\begin{aligned}
| x_{\psi}^{\rm{int}}(t)| \leq \frac{2}{m}+\frac{2}{m^2}.\end{aligned}$$
[^1]: The interactions between cells in the interpolation interval would be nonlocal, and, in addition, the Hamiltonian would involve distant cells.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this brief review, I summarize the new developments on the description of gluon radiation by energetic quarks traversing a medium as well as the observable consequences in high-energy heavy ion collisions. Information about the initial state is essential for a reliable interpretation of the experimental results and will also be reviewed. Comparison with experimental data from RHIC and expectation for the future LHC will be given.'
address: |
Theory Division, CERN\
CH-1211 Geneva 23, Switzerland\
[email protected]
author:
- 'CARLOS A. SALGADO'
title: |
MEDIUM-INDUCED GLUON RADIATION AND\
JET QUENCHING IN HEAVY ION COLLISIONS.
---
Introduction
============
The experimental program on high energy heavy ion collisions attempts to study the behavior of QCD matter under extreme conditions. The original, and still most important goal, is the creation and characterization of the quark-gluon plasma, a thermalized state of deconfined quarks and gluons that could be the form of matter of the whole Universe only several $\mu s$ after the Big-Bang. The study of high-$p_t$ processes as probes of the produced medium starts with the seminal work of J.D. Bjorken in 1982 [@bjorken]. The idea was that if a medium is produced in a collision, the high-$p_t$ particles produced inside the medium in the initial stage would loss energy (and eventually thermalize) when escaping it. The arguments in [@bjorken] were based on elastic scattering, and the loss turned out to be too small. Later refinements [@wang; @bdmps; @z] propose the medium-induced gluon radiation as the dominant source of energy loss.
Twenty years latter, in the heavy ion collider era started with RHIC, these effects could be measured for the first time. The new experimental facts coming from RHIC when comparing central AuAu with pp collisions are the following: the suppression of particles with high-$p_t$ [@Adler:2003qi; @Adams:2003kv; @Arsene:2003yk] (independent of the particle species for $p_t\gsim 4$ GeV) and the total extinction of the signal associated to a high-$p_t$ particle in the backward hemisphere [@Adler:2002tq] (back-to-back correlations). Together with this, experimental data in dAu collisions find an enhancement of high-$p_t$ particle production [@Arsene:2003yk; @Adler:2003ii; @Adams:2003im; @Back:2003ns] (the so-called Cronin effect) and back-to-back correlations of the same magnitude as the ones measured in pp collisions [@Adams:2003im]. These effects point to a strong interaction of the high-$p_t$ particle with the (dense) produced medium in agreement with the [*jet quenching*]{} scenario. In the following we will present the general formalism, based on collinear factorization, in which most of the present calculations are based as well as some comparison with experimental data.
To leading order in perturbative QCD, high-$p_t$ hadroproduction in proton–proton collisions is described by the factorization formula (see e.g. [@Eskola:2002kv]): $$\begin{aligned}
E\frac{d\sigma^h}{d^3p}&=& K(\sqrt{s})\int dz dx_1 dx_2
\frac{\hat s}{\pi z^2}\delta(\hat s+\hat t+\hat u)\times\nonumber \\
&&\times\sum_{i,j}\, f^A_i(x_1,Q^2)\, f^B_j(x_2,Q^2)\,
%&\times&\frac{\hat s}{\pi}\delta(\hat s+\hat t+\hat u)
\frac{d\sigma^{ij\to k}}{d\hat t} \,
D_{k\to h}(z,Q^2),
\label{eqpqcd}\end{aligned}$$
where $f^A_i$, $f^B_j$ are the proton parton distribution functions (PDF), $d\sigma^{ij\to k} / d \hat t$ is the partonic cross section and $D_{k\to h}(z,Q^2)$ describes the fragmentation of a parton $k$ into the hadron $h$ carrying a fraction $z$ of the momentum. The description of the experimental data by (\[eqpqcd\]) is very reasonable [@Eskola:2002kv; @Vitev:2002aa; @Levai:2003at].
In the case of heavy ion collisions, both the initial and, possibly, the final state are different. Indeed, the nuclear parton distributions are different from those in free protons and the eventually produced medium would modify the fragmentation. The knowledge of the PDF for bounded nucleons gives the baseline for the final state effects which would provide the information about the medium. So, the first goal is to obtain these nuclear PDF from experimental data in different processes.
The review is organized as follows: in next section we give a description of the initial state effects in terms of nuclear modifications of PDF; Section 3 describes the medium-induced gluon radiation spectrum, which is the main part of the present work; in Section 4 some applications are discussed, both for inclusive particle production (where a comparison with RHIC data is possible) as well as for the more differential case of jet observables. In the last two sections we comment on different approaches and give our conclusions.
Initial state effects: shadowing
================================
Nuclear and free proton PDF – $f_i^A$ and $f_i^p$ respectively – are normally related by the ratio $R_i^A$
$$f_i^A(x,Q^2)=R_{i}^A(x,Q^2)f_i^N(x,Q^2)\, .
\label{eqSalgado1}$$
For the corresponding ratio of the structure function $F_2$ several different regions have been measured, as shadowing ($R_{F_2}^A <1$) at small values of $x$, antishadowing ($R_{F_2}^A >1$) for intermediate $x$ and EMC ($R_{F_2}^A <1$) and Fermi motion for large $x$. In this way, a similar structure is expected for the nuclear PDF $f_i^A(x,Q^2)$
The proton PDF are usually obtained in global fits to experimental data in well established DGLAP analysis[@Martin:2001es; @Pumplin:2002vw]. The main difficulty in applying the same method to the nuclear case is the lack of experimental data. In this section we present two sets of nuclear PDF (EKRS [@Eskola:1998df] and HKM [@Hirai:2001np])[^1] and comment about the experimental constraints to the different flavors, specially those for gluons.
EKRS analysis of nuclear PDF
----------------------------
The goal of nuclear DGLAP analyzes is to obtain a set of nuclear PDF following the procedure of the proton case. Namely, fixing the initial parton distributions at a $Q_0^2\gg \Lambda_{QCD}^2$ and evolving them to larger $Q^2$ values by DGLAP equations. The comparison with data would fix the free parameters in the initial condition. In practice, what is usually done is to obtain the initial [*ratios*]{}, $R_i(x,Q^2_0)$ for different partons $i$, and use a known set of proton PDF (as MRST [@Martin:2001es], CTEQ [@Pumplin:2002vw], etc...) to obtain the nuclear PDF. In the EKRS analysis, data on nuclear $F_2$, and DY measured in pA collisions is used. Further constraints are momentum and baryon number sum rules. At the initial scale, the ratios for valence $R_V(x,Q^2_0)$ (same for $u_V$ and $d_V$), sea $R_S(x,Q^2_0)$ (same for $\bar u$, $\bar d$ and $\bar s$) and gluons $R_g(x,Q^2_0)$ are obtained in the following way:
$\bullet$ At large values of $x$ ($x\gsim 0.3$), $R_{F_2}$ data is used to fix the valence quarks ratio $R_V$. Both $R_S$ and $R_g$ are not constrained, so they are assumed to be equal to $R_V$.
$\bullet$ At intermediate values of $x$ ($0.04\lsim x\lsim 0.3$) both DIS and DY data constrain the ratios $R_V$ and $R_S$. Baryon number sum rule imposes also constraints to valence ratio. In this region, the gluon ratio is fixed by momentum sum rule, with the help of NMC data [@Arneodo:1996ru] to fix the value of $x$ where $R_g(x)$=1 (see below). This produces a large gluon antishadowing.
$\bullet$ At small values of $x$ ($x\lsim 0.04$), $R_S=R_{F_2}$ and a saturation of $R_{F_2}$ is assumed for $x\lsim 10^{-3}$; $R_g=R_{F_2}$ is taken for the very small $x$ values – notice that evolution modifies this equality; $R_V$ is fixed by baryon number sum rule.
Once the initial conditions are known, LO-DGLAP evolution is performed, and the parameters of the initial conditions fixed by comparing to data at different values of $Q^2$. The initial conditions obtained by this method are plotted for a Pb nucleus in Fig. \[figSalgado2\] and compared with HKM [@Hirai:2001np]. The main differences come from the fact that HKM do not use data on Drell-Yan [@Alde:im] with nuclei (essential to constrain valence and sea quarks at intermediate $x$) nor the $Q^2$-dependent data measured by NMC [@Arneodo:1996ru].
{width="13cm"}
\[figSalgado2\]
Constraints for gluons
----------------------
At LO, the gluon distribution does not directly contribute to the DIS or DY cross sections. It instead drives the $Q^2-$evolution of all other flavors: at small values of $x$, DGLAP at LO gives $\partial F_2^{p(n)}/\partial \log Q^2
\sim xg(2x,Q^2)$. So, for the ratios, $$%\frac{\partial R_{F_2}^A(x,Q^2)}{\partial \log Q^2}
\partial R_{F_2}^A(x,Q^2)/\partial \log Q^2
\propto
%\frac{10\alpha_s}{27\pi}\frac{xg(2x,Q^2)}{\frac{1}{2}F_2^{\rm D}(x,Q^2)}
\left\{R_g^A(2x,Q^2)-R_{F_2}^A(x,Q^2)\right\} .
\label{eqSalgado4}$$ In order to obtain a positive $\log Q^2$-slope, as measured by the NMC Collaboration [@Arneodo:1996ru] (see Fig. \[figSalgado3\]) a very strong gluon shadowing for $x\gsim 0.01$ is not allowed by (\[eqSalgado4\]). To quantify this statement [@Eskola:2002us], we have applied DGLAP evolution to different initial conditions and compare with the NMC data, the comparison is done in Fig. \[figSalgado3\]. The slopes reflect the gluon distribution, in particular, the negative slopes obtained when taken new HIJING parametrization [@Li:2001xa] indicate that the very strong shadowing for gluons is in disagreement with data. (This is, also, in agreement with the new analysis [@deFlorian:2003qf]).
![[$Q^2$-evolution of the ratios of $F_2$ in Sn and C for different initial conditions EKS98 [@Eskola:1998df] (solid lines), HKM [@Hirai:2001np] (dotted-dashed lines), HPC [@HPC] (dashed lines) and new HIJING [@Li:2001xa] (dotted lines) compared with the NMC results. ]{}[]{data-label="figSalgado3"}](rf2snc.eps "fig:"){width="10cm"} -0.5cm
Summarizing, in order to use the collinear factorization formula (\[eqpqcd\]) a set of nuclear PDF is needed. These nuclear PDF can be constrained by DIS and DY experimental data and evolved by DGLAP equations. In this framework, a strong gluon shadowing for $x\gsim 0.01$ is not supported by present data. EKRS parametrization gives[@Eskola:2002kv], for RHIC at $y\sim 0$, a moderate enhancement in the intermediate region of $p_t$, but interestingly, in agreement with the increase in the $\pi^0$ yield measured in dAu collisions [@Adler:2003ii].
Final-state effects: medium-induced gluon radiation
===================================================
The medium-induced distribution of gluons of energy $\omega$ radiated off an initial hard parton has been computed by several methods and approximations [@bdmps; @z; @glv; @Wiedemann:2000za]. They can be obtained as particular cases of the general $k_t$-differential spectrum [@Wiedemann:2000za; @Salgado:2003gb] $$\begin{aligned}
\omega\frac{dI}{d\omega\, d{\bf k}}
&=& {\alpha_s\, C_R\over (2\pi)^2\, \omega^2}\,
2{\rm Re} \int_{0}^{\infty}\hspace{-0.3cm} dy_l
\int_{y_l}^{\infty} \hspace{-0.3cm} d\bar{y}_l\,
\int d^2{\bf u}\,
e^{-i{\bf k}_t\cdot{\bf u}} \,
e^{ -\frac{1}{2} \int_{\bar{y}_l}^{\infty} d\xi\, n(\xi)\, \sigma({\bf u})}\,
\times
\nonumber\\
&\times&{\partial \over \partial {\bf y}}\cdot
{\partial \over \partial {\bf u}}\,
\int_{{\bf y}={\bf r}(y_l)}^{{\bf u}={\bf r}(\bar{y}_l)}
\hspace{-0.5cm} {\cal D}{\bf r}
\exp\left[ i \int_{y_l}^{\bar{y}_l} \hspace{-0.2cm} d\xi
\frac{\omega}{2} \left(\dot{\bf r}^2
- \frac{n(\xi)\, \sigma({\bf r})}{i\,2\, \omega} \right)
\right]\, .
\label{eqspec}\end{aligned}$$ Here, $C_R$=$C_F$=4/3 for quarks and $C_R$=$C_A$=3 for gluons. Medium properties enter (\[eqspec\]) via the product of the medium density $n(\xi)$ of scattering centers times the dipole cross section $\sigma({\bf r})$ which measures the interaction strength of a single elastic scattering. The solution for a general $n(\xi)\sigma({\bf r})$ is unknown, and two approximations have been studied up to now: the multiple soft scattering limit, $n(\xi)\, \sigma({\bf r})
\approx \frac{1}{2} \hat{q}(\xi)\, {\bf r}^2$, in which the path integrals reduce to those of a harmonic oscillator and can be solved analytically – this is the approximation used by the BDMPS group [@bdmps] and Zakharov [@z]; the single hard scattering limit, which consists in a series expansion in $n(\xi)\sigma({\bf r})$, where $\sigma({\bf r})$ is modeled by a Yukawa potential with Debye screening mass $\mu$ – this is the approximation used by the GLV group [@glv].
Eq. (\[eqspec\]) implies a one-to-one correspondence between the average energy loss of the parent parton, and the transverse momentum broadening of the associated gluon radiation, as argued in Ref. [@Baier:1996sk]. In Fig. \[figdidwdkt\] we present the numerical results of $\omega dI/d\omega\, d{\bf k}$ for quarks in the two approximations.
![[ The gluon energy distribution (\[eqspec\]) as a function of the rescaled gluon energy $\omega/\omega_c$ and the rescaled gluon transverse momentum $\kappa$, see eq. (\[eqvar\]). ($\alpha_s$=1/3 has been taken). ]{}](fig1kt.epsi){width="8cm"}
\[figdidwdkt\]
Most of the qualitative properties of the medium-induced gluon radiation spectrum can be understood by coherence arguments. Let us concentrate on the multiple soft scattering approximation, the same arguments hold for the single hard with the change $\hat q \ \rightarrow\ \mu^2/L$. For a gluon emitted with energy $\omega$ and transverse momentum $k_t$, the phase and the gluon formation time are $$\varphi = \Bigg\langle \frac{k_t^2}{2\omega}\, \Delta z \Bigg\rangle
\Longrightarrow
l_{coh}\sim\frac{\omega}{k_t^2}\, .$$ The medium is characterized by the transport coefficient $\hat q
\simeq\frac{\mu^2}{\lambda}$, giving the average transverse momentum $\mu^2$ transfered from the medium to the gluon per mean free path $\lambda$. So, $k_t^2\sim \mu^2 l_{\rm coh}/\lambda$, when $l_{\rm coh}$ reaches the length of the medium $L$ one has $k_t^2\sim
\hat q L$. So, this is a maximum for the $k_t$ of the emitted gluon. If one defines $$\kappa^2=\frac{k_t^2}{\hat qL} \ , \ \omega_c=\frac{1}{2} \hat qL^2
\hspace{0.5cm} \Bigg[
\kappa^2=\frac{k_t^2}{\mu^2} \ , \ \omega_c=\frac{1}{2} \mu^2L
\ \ {\rm For\ Single\ Hard}\Bigg]\ ,
%\hat q\rightarrow \frac{\mu^2}{L}
\label{eqvar}$$ the phase for $\Delta z=L$ is $\varphi\sim \kappa^2\ \omega_c/\omega$. The radiation can only be formed when $\varphi\gsim 1$, so a suppression of the radiation appears when $\kappa^2\lsim\omega_c/\omega$. The plateau at small values of $\kappa$ for fixed $\omega/\omega_c$ in Figure \[figdidwdkt\] is due to these coherence effects. Moreover, at large values of $\omega\gsim\omega_c$ the spectrum is also suppressed. This is the well known LPM suppression first discussed by the BDMPS group [@bdmps]. These features are characteristic of QCD as the multiple scattering is performed by the (eventually) emitted gluon in the high-energy approximation. In the case of QED, the photon does not interact and the relevant phase contains now the energy of the electron instead of $\omega$.
For practical applications, the $k_t$-integrated spectrum is needed. The kinematical limits $0<k_t<\omega$ are imposed to compute the spectra of Fig. \[figdidw\] for different values of $R=\omega_c L$. A comparison with the BDMPS result is also shown. The origin of R is simple: as $k_t^2$ is limited by $\hat q L$, the upper kinematical limit in the $k_t$-integration cuts the gluon energies $\omega^2\lsim k_t^2 \sim \hat q L$. So, the spectrum is suppressed for $$\left(\frac{\omega}{\omega_c}\right)^{2}\lsim\frac{2}{R}\ .$$ The position of the maxima is in agreement with this estimate. Thus, the suppression in the soft part of the spectra can be understood by formation time arguments. In this way, the fact that the radiation spectrum shows small sensitivity to the infrared region is ground on general arguments rather than on the actual realization of the model. This has important consequences in the experimental observables as we will see. The limit $R\to \infty$ is obtained by integrating the spectrum in $k_t^2<\infty$, $$\lim_{R\to \infty}\,
\omega \frac{dI}{d\omega} = \int_0^\infty dk_t^2
\omega\frac{dI}{d\omega dk_t^2}
= \frac{2\alpha_s C_R}{\pi}\, {\rm Re} \left[
\,\ln \left (%\Bigg \vert
{\cos\sqrt{\frac{\omega_c}{i\omega}}}
%\Bigg \vert \, .
\right )\right ]
\label{eqbdmps}$$ which coincides with the BDMPS result[@bdmps]. The average parton energy loss is, then, $$\langle \Delta E \rangle \equiv
\int_0^\infty d\omega\,
\omega \frac{dI}{d\omega}\hspace{0.2cm}
\xrightarrow[R\to\infty]{}
\hspace{0.2cm}
\frac{\alpha_s C_R}{2}\, \omega_c\, \propto\ \hat q L^2\, .
\label{eqdeltaE}$$ This is the well-known $L^2$ dependence of the average radiative energy loss [@bdmps; @z; @glv; @Wiedemann:2000za; @Salgado:2003gb].
![[ The medium-induced gluon energy distribution $\omega \frac{dI}{d\omega}$ in the multiple soft scattering approximation for different values of the kinematic constraint $R = \omega_c\, L$=1000, 10000 and 40000. ]{}[]{data-label="figdidw"}](didwcomp.ps){width="10cm"}
Expanding medium
----------------
The medium produced in a heavy ion collision is expanding very fast in the longitudinal and probably also in the transverse direction. The expansion is usually parametrized by an exponential decrease of the medium density as $n(\tau)\sim
1/\tau^\alpha$, with $\alpha$=1 for 1-dimensional (Bjorken) expansion. In this case, the transport coefficient changes accordingly as $\hat q(\tau)=\hat q_0 (\tau_0/\tau)^\alpha$ and the corresponding spectrum can be obtained from eq. (\[eqspec\]). It has been found in Ref. [@Salgado:2002cd] that any expanding medium can be related with a equivalent static one with effective transport coefficient $$\bar{\hat{q}} = \frac{2}{L^2} \int_{\xi_0}^{L+\xi_0} d\xi\,
\left(\xi - \xi_0\right)\, \hat{q}(\xi)\, .
\label{eqscal}$$ This result confirms previous relations to the level of the average $\Delta E$ [@Baier:1998yf; @Gyulassy:2000gk] and allows to use the static formulas for any expanding scenario.
Applications
============
Equation (\[eqspec\]) has been calculated for the one-gluon inclusive case. Up to now no progress has been made in computing diagrams with more than one gluon emission, so, for practical applications one usually assumes the independent gluon emission approximation [@Baier:2001yt] $$P_E(\epsilon) = \sum_{n=0}^\infty \frac{1}{n!}
\left[ \prod_{i=1}^n \int d\omega_i \frac{dI(\omega_i)}{d\omega}
\right]
\delta\left(\epsilon -\sum_{i=1}^n {\omega_i \over E} \right)
\exp\left[ - \int d\omega \frac{dI}{d\omega}\right]\ .
\label{eqqw}$$ In the case of small gluon multiplicities, the interference terms are expected to be small [@Baier:2001yt], and (\[eqqw\]) should give a good approximation. For a medium of finite length $L$, there is a finite probability $p_0$ that no energy is lost – no gluon is radiated and the fragmentation is not affected. This discrete contribution decreases with increasing in medium path-length or increasing density of the medium. So, we write $$P_E(\epsilon)=p_0\delta(\epsilon)+p(\epsilon)\, .
\label{eqqw2}$$ In Fig. \[figqw\] we plotted the discrete, $p_0$, and continuous, $p(\epsilon)$, contributions to the [*quenching weights*]{}, $P_E(\epsilon)$ for different values of $R=\omega_c\ L$.
![ The two contributions to the probability (\[eqqw\]) that a parton loses $\Delta E$ of its energy in the medium: Continuous part (left panel) and the discrete probability $p_0$ in (\[eqqw2\]) that the hard parton escapes the medium without interaction (right panel). []{data-label="figqw"}](peps.eps){width="10cm"}
Inclusive particle production
-----------------------------
For high enough $p_t$ of the parton, the hadronization takes place outside the medium. In this case, the medium-modified fragmentation function is usually written as [@Wang:1996yh; @Gyulassy:2001nm; @Salgado:2002cd] $$D_{k\to h}^{(\rm med)}(z,Q^2) = \int_0^1 d\epsilon\, P_E(\epsilon)\,
\frac{1}{1-\epsilon}\, D_{k\to h}(\frac{z}{1-\epsilon},Q^2)\, ,
\label{eqff}$$ where $D_{k\to h}$ is the vacuum fragmentation function. The only effect of the medium in (\[eqff\]) is a shift in the energy of the initial parton given by $P_E(\epsilon)$. Additional (logarithmic) modifications in the $Q^2$-dependence are neglected, as they are subdominant as compared to $\epsilon=\Delta E/E\sim 1/Q$ (we identify $Q$ with the initial transverse energy of the parton $E$). In Fig. \[figff\] the fragmentation functions for different media computed by (\[eqff\]) are compared to the corresponding vacuum case [@Salgado:2002cd].
![Fragmentation functions for quarks and gluons into $\pi$ for media of different $R=\omega_c L$. The vacuum fragmentation functions are taken from [@Kniehl:2001]. []{data-label="figff"}](ff.eps){width="7cm"}
In order to estimate the suppression of produced $\pi$’s we make use of the observation [@Eskola:2002kv] that the partonic cross section and the PDF essentially contributes with $z^6$ to the integral in (\[eqpqcd\]). In Fig. \[figff\] we weight the fragmentation function by this factor, the ratios at the maxima between a medium and the vacuum gives the corresponding suppression of final $\pi$’s. A suppression by a factor of $4\div 5$, as measured at RHIC, can be reached for R=1000 $\div$ 2000 (see also Fig. \[figrhic\]). These values are in agreement[@Salgado:2002cd] with the ones obtained by the GLV group [@Gyulassy:2000gk].
In order to study the sensitivity of these results to the small $\omega$-region, the so-called [*quenching factors*]{} have been introduced in Ref. [@Baier:2001yt], $$\begin{aligned}
Q(p_t)&=&
{{d\sigma^{\rm med}(p_t)/ dp^2_t}\over
{d\sigma^{\rm vac}(p_t)/ dp^2_t}}=
\int d{\Delta E}\, P(\Delta E)\left(
{d\sigma^{\rm vac}(p_t+\Delta E)/ dp^2_t}\over
{d\sigma^{\rm vac}(p_t)/ dp^2_t}\right)\, ,
\label{eqqf}\end{aligned}$$ where the vacuum spectrum is usually taken as $d\sigma^{\rm vac}(p_t)/ dp^2_t\sim p_t^{-n}$. This can be seen as an alternative way of computing the effects of jet quenching. The sensitivity of the results to the infrared region can be studied by cutting-off the spectrum for $\omega\leq\omega_{\rm cut}$ and computing (\[eqqf\]) – see Fig. \[figrhic\]. A strong sensitivity to the small-$\omega$ region appears when the BDMPS spectrum (\[eqbdmps\]) is used [@Baier:2001yt; @Salgado:2003gb]. With the regularization of the small$-\omega$ region due to finite $R=\omega_c
L$, this sensitivity practically disappears [@Salgado:2003gb]. In the RHS of Fig. \[figrhic\] a comparison with PHENIX data of the suppression of $\pi^0$ for the most central AuAu collisions at RHIC is performed. The magnitude and the slope of the effect is in agreement with the data. This is in contrast with previous expectations based on BDMPS spectrum of a much steeper slope, see Fig. \[figrhic\] (LHS). In this way, the results for the multiple soft and the single hard scattering approximations are similar.
![LHS: Quenching factors (\[eqqf\]) computed from the BDMPS spectrum (upper figures) and with finite R (lower figures) applying different cuts to the small-$\omega$ region ($\hat q$=1 GeV$^2$/fm has been taken). RHS: Comparison of the suppression obtained in the multiple soft and in the single hard scattering approximations with the experimental data from PHENIX [@Adler:2003qi] ($\omega_c$=67.5 GeV and R=2000 for both multiple soft and single hard scattering approximations). []{data-label="figrhic"}](qwcomp.eps){width="6cm"}
![LHS: Quenching factors (\[eqqf\]) computed from the BDMPS spectrum (upper figures) and with finite R (lower figures) applying different cuts to the small-$\omega$ region ($\hat q$=1 GeV$^2$/fm has been taken). RHS: Comparison of the suppression obtained in the multiple soft and in the single hard scattering approximations with the experimental data from PHENIX [@Adler:2003qi] ($\omega_c$=67.5 GeV and R=2000 for both multiple soft and single hard scattering approximations). []{data-label="figrhic"}](ffrat.eps){width="6cm"}
Jet shapes
----------
Equation (\[eqspec\]) relates the energy loss of a parton with the transverse momentum broadening of the associated gluon radiation. This dynamics should modify the internal jet substructure from the vacuum case. In order to study these effects, a first attempt to compute jet observables in the presence of a medium has been made in [@Salgado:2003rv]. In the rest of the section we present the medium–modification for two quantities, the fraction of jet energy inside a cone and the gluon multiplicity distribution.
The fraction of the jet energy inside a cone of radius $R=\sqrt{(\Delta \eta)^2 + (\Delta \Phi)^2}$ is $$\rho_{\rm vac}(R) = \frac{1}{N_{\rm jets}} \sum_{\rm jets}
\frac{E_t(R)}{E_t(R=1)}\, .
\label{eq5}$$ In the presence of the medium, this energy is shifted by[@Salgado:2003rv] $$\rho_{\rm med}(R) =
\rho_{\rm vac}(R) - \frac{\Delta E_t(R)}{E_t}
+ \frac{\Delta E}{E_t} \left( 1 - \rho_{\rm vac}(R)\right)\, ,
\label{eq9}$$ where $\Delta E_t(R)$ is the additional (medium) energy radiated outside a cone $\Theta=R$ and $\Delta E(\Theta)=\int \epsilon P(\epsilon,\Theta)d\epsilon$, where the quenching weight is computed by integrating the spectrum (\[eqspec\]) in $\omega\sin\Theta < k_t < \omega$. In Fig. \[figfig2\] we plot the medium-shifted distributions. The shaded area corresponds to the uncertainty in finite quark-energy effects: in the eikonal approximation $P(\epsilon)$ have support in the unphysical region $\epsilon > 1$. To estimate this effect we make the change $P(\epsilon)\,\rightarrow\, P(\epsilon)/\int_0^1 d\epsilon P(\epsilon)$.
=8.5cm
The effect of the medium is very small (at $R$=0.3, it is $\sim$ 5% for a 50 GeV jet and $\sim$ 3% for a 100 GeV jet). The smalleness of this effect could allow for a calibration of the total energy of the jet without tagging in a recoiling hard photon or Z-boson. It also implies that the jet $E_t$ cross section scales with the number of binary collisions. In order to check the sensitivity of our results to the small-energy region, we impose, in analogy to the previous section, low momentum cut-offs which removes gluon emission below 5 GeV. It is interesting that transverse momentum broadening is very weakly affected by these cuts. This is again due to the infrared behavior of the spectrum for small values of $\omega$ – see Fig. \[figdidw\]. A proper substraction of the large background present in heavy ion collisions would benefit from this result.
$k_t$-differential measurements are expected to be more sensitive to medium effects. As an example, the intrajet multiplicity of produced gluons as a function of the transverse (with respect to the jet axis) momentum is plotted in Fig. \[figglmult\]. The medium-induced additional number of gluons with transverse momentum $k_t = \vert {\bf k}\vert$, produced within a subcone of opening angle $\theta_c$, is $$\begin{aligned}
\frac{dN_{\rm med}}{dk_t} = \int_{k_t/\sin\theta_c}^{E_t} d\omega\,
\frac{dI_{\rm med}}{d\omega\, dk_t}\, .
\label{eq8}\end{aligned}$$ For the vacuum we simply assume $dN_{\rm vac}/dk_t\sim 1/k_t\log(E_t\sin\theta_c/k_t)$. In this case, the effect is sizable for transverse momenta of the order of several GeV and could be easily measured experimentally. A more realistic analysis would need of an implementation of the whole fragmentation. However, the origin of the shift is mainly due to the large $k_t\sim
Q_{\rm sat}$ that the gluon obtains from the medium. In this way, we expect this conclusion to be very robust and not depending on the actual realization of the model.
Discussion and other approaches {#secoutros}
===============================
In the previous sections, we have presented the usual framework to compute high-$p_t$ particle production in nuclear collisions. It is based on the collinear factorization, eq. (\[eqpqcd\]), supplemented with nuclear parton distribution functions and final state effects due to medium-induced gluon radiation. This framework has been successfully employed to reproduce experimental data. Let us comment on the differences with other approaches within the same framework, concerning both the initial and the final state.
First, in the initial state intrinsic-$k_t$ and/or Cronin effect due to multiple (elastic) scattering are sometimes introduced (see e.g. [@Vitev:2002aa; @Levai:2003at; @Vitev:2002pf]). The $p\bar p$ data cannot distinguish between these two approaches, however, $hA$ data at energies of several tens of GeV are well described with this mechanism. Its magnitude for central AuAu collisions at RHIC can be as large as a 50% increase. This increase is sometimes compensated by the large shadowing of Ref. [@Li:2001xa], however, as we have seen this strong gluon shadowing is in disagreement with DIS data. The nuclear PDF obtained in a DGLAP analysis result only in a tiny enhancement [@Eskola:2002kv; @deFlorian:2003qf].
Concerning the final state effects, most of the approaches rely on the radiative energy loss and differ only on the approximation used, multiple soft [@bdmps] or single hard scattering [@glv]. We have seen that both approximations give very similar results when the appropriate kinematical limits and correspondingly similar parameters are taken into account [@Salgado:2003gb]. (For an approach based on twist expansion in DIS see Ref. [@Guo:2000nz]). The possibility of collisional energy loss has been also explored [@Mustafa:2003vh] with a reasonable result. As it has been exposed in Section 3, formation time arguments lead to a radiative energy loss which increases as $L^2$, in the case of a collisional energy loss the growth is, however, as $L$. So, the centrality dependence of the effect is expected to be sensitive to this different behavior. Unfortunately, it seems that present data cannot distinguish between a $L$ or $L^2$ behavior [@Choi:2003pq]. Notice that in all these analyzes, hadronization is assumed to take place outside the medium. This could not be the case for the smallest $p_t$ values [@Arleo:2002kh].
On the other hand, an initial state origin for the suppression of high-$p_t$ particle production has been proposed in the framework of the saturation approach [@Kharzeev:2002pc], the origin being the smaller number of initial gluons in the nuclei wave functions. There has been some discussion on whether this removes the Cronin enhancement at intermediate values of $p_t$ or not, but finally the different groups agree [@Albacete:2003iq; @Baier:2003hr; @Kharzeev:2003wz; @Jalilian-Marian:2003mf] in that saturation leads to a suppression for all values of $p_t$. $dAu$ experimental data ruled out this hypothesis as the main source of high-$p_t$ particle suppression at central rapidities. However the prediction is [@Albacete:2003iq] that at higher energies and/or rapidities this mechanism very efficiently suppresses the high-$p_t$ particle yields. The new preliminary data from BRAHMS [@brahmsprel] find a strong reduction of $\pi$’s for $p_t < 2.5$ GeV in the forward direction. From the results in Section 2, a suppression like this seems difficult to accommodate in a DGLAP approach in collinear factorization. If this preliminary data is confirmed it could be the first clear indication of saturation phenomena in nuclear collisions.
Conclusions
===========
In this review we have described the most recent theoretical results relating medium-induced gluon radiation, energy loss and jet broadening. These effects are accessible for the first time in experiments of heavy ion collisions at RHIC. All the experimental data strongly point to a large jet suppression due to interaction with the produced medium. The larger energy of the LHC will allow for a qualitative new regime, where the jets are not completely suppressed and the jet substructure could be measured in the large background environment. This will open a completely new window for the study of the evolution of high energetic particles in a medium.
[**Acknowledgments:**]{} I would like to thank J. Albacete, N. Armesto, K. Eskola, H. Honkanen, V. Kolhinen, A. Kovner, V. Ruuskanen and U. Wiedemann for the very nice and fruitful collaboration which is partially reviewed in this paper. Critical reading of this manuscript by N. Armesto and U. Wiedemann is gratefully acknowledged.
[0]{} J. D. Bjorken, Fermilab-Pub-82/59-THY, Batavia (1982); Erratum, unpublished.
M. Gyulassy and X.-N. Wang, Nucl. Phys. [**B420**]{} (1994) 583;\
X.-N. Wang, M. Gyulassy and M. Plümer, Phys. Rev. [**D51**]{} (1995) 3436.
R. Baier, Yu. L. Dokshitzer, S. Peigné and D. Schiff, Phys. Lett. [**B345**]{} (1995) 277; R. Baier, Yu. L. Dokshitzer, A. H. Mueller, S. Peigné and D. Schiff, Nucl. Phys. [**B483**]{} (1997) 291; R. Baier, Y. L. Dokshitzer, A. H. Mueller and D. Schiff, Nucl. Phys. [**B531**]{} (1998) 403.
B. G. Zakharov, JETP Letters [**63**]{} (1996) 952; ibidem [**65**]{} (1997) 615; Phys. Atom. Nucl. [**61**]{} (1998) 838 \[Yad. Fiz. [**61**]{} (1998) 924\]; JETP Lett. [**70**]{} (1999) 176; ibidem [**73**]{} (2001) 49.
M. Gyulassy, P. Levai and I. Vitev, Phys. Rev. Lett. [**85**]{} (2000) 5535 \[arXiv:nucl-th/0005032\]; Nucl. Phys. [**B594**]{} (2001) 371 \[arXiv:nucl-th/0006010\]; Phys. Rev. [**D66**]{} (2002) 014005 \[arXiv:nucl-th/0201078\].
U. A. Wiedemann, Nucl. Phys.[**B588**]{} (2000) 303 \[arXiv:hep-ph/0005129\]; Nucl. Phys. [**A690**]{} (2001) 731 \[arXiv:hep-ph/0008241\].
C. A. Salgado and U. A. Wiedemann, Phys. Rev. [**D68**]{} (2003) 014008 \[arXiv:hep-ph/0302184\]; Nucl. Phys. [**A715**]{} (2003) 783 \[arXiv:hep-ph/0209025\].
S. S. Adler [*et al.*]{} \[PHENIX Collaboration\], Phys. Rev. Lett. [**91**]{} (2003) 072301 \[arXiv:nucl-ex/0304022\]; D. d’Enterria \[PHENIX Collaboration\], Nucl. Phys. [**A715**]{} (2003) 749 \[arXiv:hep-ex/0209051\].
J. Adams [*et al.*]{} \[STAR Collaboration\], Phys. Rev. Lett. [**91**]{} (2003) 172302 \[arXiv:nucl-ex/0305015\]; G. J. Kunde \[STAR Collaboration\], Nucl. Phys. [**A715**]{} (2003) 189 \[arXiv:nucl-ex/0211018\].
I. Arsene [*et al.*]{} \[BRAHMS Collaboration\], Phys. Rev. Lett. [**91**]{} (2003) 072305 \[arXiv:nucl-ex/0307003\].
C. Adler [*et al.*]{} \[STAR Collaboration\], Phys. Rev. Lett. [**90**]{} (2003) 082302 \[arXiv:nucl-ex/0210033\].
S. S. Adler [*et al.*]{} \[PHENIX Collaboration\], Phys. Rev. Lett. [**91**]{} (2003) 072303 \[arXiv:nucl-ex/0306021\].
J. Adams [*et al.*]{} \[STAR Collaboration\], Phys. Rev. Lett. [**91**]{} (2003) 072304 \[arXiv:nucl-ex/0306024\].
B. B. Back [*et al.*]{} \[PHOBOS Collaboration\], Phys. Rev. Lett. [**91**]{} (2003) 072302 \[arXiv:nucl-ex/0306025\].
K. J. Eskola and H. Honkanen, Nucl. Phys. [**A713**]{} (2003) 167 \[arXiv:hep-ph/0205048\].
I. Vitev, arXiv:hep-ph/0212109.
P. Levai, G. Papp, G. G. Barnafoldi and G. Fai, arXiv:nucl-th/0306019.
A. D. Martin, R. G. Roberts, W. J. Stirling and R. S. Thorne, Eur. Phys. J. [**C23**]{} (2002) 73 \[arXiv:hep-ph/0110215\].
J. Pumplin, D. R. Stump, J. Huston, H. L. Lai, P. Nadolsky and W. K. Tung, JHEP [**0207**]{} (2002) 012 \[arXiv:hep-ph/0201195\].
K. J. Eskola, V. J. Kolhinen and P. V. Ruuskanen, Nucl. Phys. [**B535**]{} (1998) 351 \[arXiv:hep-ph/9802350\]; K. J. Eskola, V. J. Kolhinen and C. A. Salgado, Eur. Phys. J. [**C9**]{} (1999) 61 \[arXiv:hep-ph/9807297\]; K. J. Eskola, H. Honkanen, V. J. Kolhinen, P. V. Ruuskanen and C. A. Salgado, arXiv:hep-ph/0110348.
M. Hirai, S. Kumano and M. Miyama, Phys. Rev. [**D64**]{} (2001) 034003 \[arXiv:hep-ph/0103208\].
D. de Florian and R. Sassot, arXiv:hep-ph/0311227.
L. Frankfurt, V. Guzey, M. McDermott and M. Strikman, JHEP [**0202**]{} (2002) 027 \[arXiv:hep-ph/0201230\].
M. Arneodo [*et al.*]{} \[New Muon Collaboration\], Nucl. Phys. [**B481**]{} (1996) 23.
D. M. Alde [*et al.*]{}, Phys. Rev. Lett. [**64**]{} (1990) 2479.
K. J. Eskola, H. Honkanen, V. J. Kolhinen and C. A. Salgado, Phys. Lett. [**B532**]{} (2002) 222 \[arXiv:hep-ph/0201256\]; arXiv:hep-ph/0205231.
S. y. Li and X. N. Wang, Phys. Lett. [**B527**]{} (2002) 85 \[arXiv:nucl-th/0110075\].
Jan Czyzewski, K.J. Eskola, and J. Qiu, at the III International Workshop on Hard Probes of Dense Matter, ECT$^*$, Trento, June 1995.
R. Baier, Y. L. Dokshitzer, A. H. Mueller, S. Peigne and D. Schiff, Nucl. Phys. [**B484**]{} (1997) 265 \[arXiv:hep-ph/9608322\].
C. A. Salgado and U. A. Wiedemann, Phys. Rev. Lett. [**89**]{} (2002) 092303 \[arXiv:hep-ph/0204221\].
R. Baier, Y. L. Dokshitzer, A. H. Mueller and D. Schiff, Phys. Rev. [**C58**]{} (1998) 1706 \[arXiv:hep-ph/9803473\]. M. Gyulassy, I. Vitev and X. N. Wang, Phys. Rev. Lett. [**86**]{} (2001) 2537.
R. Baier, Y. L. Dokshitzer, A. H. Mueller and D. Schiff, JHEP [**0109**]{} (2001) 033 \[arXiv:hep-ph/0106347\].
X. N. Wang, Z. Huang and I. Sarcevic, Phys. Rev. Lett. [**77**]{} (1996) 231.
M. Gyulassy, P. Levai and I. Vitev, Phys. Lett. [**B538**]{} (2002) 282 \[arXiv:nucl-th/0112071\].
B. A. Kniehl, G. Kramer and B. Pötter, Nucl. Phys. [**B597**]{} (2001) 337
C. A. Salgado and U. A. Wiedemann, arXiv:hep-ph/0310079.
B. Abbott [*et al.*]{} \[D0 Coll.\], FERMILAB-PUB-97-242-E.
I. Vitev and M. Gyulassy, Phys. Rev. Lett. [**89**]{} (2002) 252301 \[arXiv:hep-ph/0209161\].
M. G. Mustafa and M. H. Thoma, arXiv:hep-ph/0311168.
X. f. Guo and X. N. Wang, Phys. Rev. Lett. [**85**]{} (2000) 3591 \[arXiv:hep-ph/0005044\].
S. Y. Choi, M. Drees, B. Gaissmaier and J. Song, arXiv:hep-ph/0310284.
F. Arleo, JHEP [**0211**]{} (2002) 044 \[arXiv:hep-ph/0210104\].
D. Kharzeev, E. Levin and L. McLerran, Phys. Lett. [**B561**]{} (2003) 93 \[arXiv:hep-ph/0210332\].
J. L. Albacete, N. Armesto, A. Kovner, C. A. Salgado and U. A. Wiedemann, arXiv:hep-ph/0307179.
R. Baier, A. Kovner and U. A. Wiedemann, Phys. Rev. [**D68**]{} (2003) 054009 \[arXiv:hep-ph/0305265\].
D. Kharzeev, Y. V. Kovchegov and K. Tuchin, arXiv:hep-ph/0307037.
J. Jalilian-Marian, Y. Nara and R. Venugopalan, arXiv:nucl-th/0307022.
R. Debbe, talk presented at the APS/DNP 2003, Tucson Arizona (2003), http://www4.rcf.bnl.gov/brahms/WWW/presentations.html.
[^1]: By the time this review was finished a new analysis was published [@deFlorian:2003qf]. Quantitative differences appear for the gluons when compared with EKRS but the qualitative features are in agreement with this set. (See Ref. [@deFlorian:2003qf] for further details and the effect of these corrections on the high-$p_t$ $\pi^0$ yields measured at RHIC. See also [@Frankfurt:2002kd] for a related approach).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We investigate optical absorption spectra obtained through time-dependent density functional theory (TD-DFT) based on nonempirical hybrid functionals that are designed to correctly reproduce the dielectric function. The comparison with state-of-the-art $GW$ calculations followed by the solution of the Bethe-Sapeter equation (BSE-$GW$) shows close agreement for both the transition energies and the main features of the spectra. We confront TD-DFT with BSE-$GW$ by focusing on the model dielectric function and the local exchange-correlation kernel. The present TD-DFT approach achieves the accuracy of BSE-$GW$ at a fraction of the computational cost.'
author:
- Alexey Tal
- Peitao Liu
- Georg Kresse
- Alfredo Pasquarello
bibliography:
- 'article.bib'
title: ' Accurate optical spectra through time-dependent density functional theory based on screening-dependent hybrid functionals'
---
Kohn-Sham density functional theory (DFT) is notoriously bad for describing band gaps in semiconductors due to the lack of the derivative discontinuity in semi-local functionals [@Perdew1983; @Sham1983]. Thus, most of the *ab inito* methods for optical absorption calculations based on DFT electronic structures have to address two important problems. First, it is necessary to correct the band gap. Second, the interaction between electrons and holes has to be taken into account. The state-of-the-art approach to improve the band gap is the $GW$ approximation [@Hedin1965; @Hedin1970; @Hybertsen1985; @Aryasetiawan1998], whereas the electron-hole interaction can be included by solving the Bethe-Salpeter equation (BSE) [@Onida2002; @Albrecht1998; @Benedict1998; @Rohlfing1998a; @Hanke1980]. The combined $GW$+BSE approach has been shown to give very accurate results compared to experiment, but the main drawback is the scaling, which makes it computationally very challenging for large size systems.
Gross developed an alternative approach based on the time-dependent electron density, typically referred to as time-dependent density functional theory (TD-DFT) [@Onida2002; @Gross1985; @Runge1984]. In this approach, the time-dependent Kohn-Sham equations include a time-dependent exchange-correlation (xc) potential $v_\textrm{xc}$ and its variation with the time-dependent density, also known as the exchange-correlation kernel $f_\textrm{xc}$. The exact $v_\textrm{xc}$ and $f_\textrm{xc}$ are unknown, but several approximations have been introduced. Local approximations to $f_\textrm{xc}$ lack the correct long wavelength limit, $f_\textrm{xc}(q\rightarrow 0) \propto1/q^2$, responsible for the correct description of the electron-hole interaction. Therefore, local approximations are unable to capture excitonic effects, but can perform well for metallic systems [@Casida1995; @Vasiliev1999; @Rubio1996; @Gavrilenko1996]. The correct asymptotic behavior is recovered in the so-called “nanoquanta” kernel [@Sottile2003; @Marini2003; @Adragna2003] derived from the BSE to capture excitonic effects. Hence this approach produces accurate spectra for solids, but remains computationally as expensive as solving the BSE.
Hybrid functional calculations with non-local Fock exchange can be used to improve band gaps. Moreover, since the long wavelength limit is accounted for, it can be expected that these functionals could be used for calculating optical spectra. Various hybrid functionals have been tested in TD-DFT and it has been shown that a good performance can be achieved in molecules [@Salzner1997; @Laurent2013]. However, a good description in solids requires the consideration of the screening in the exchange interaction [@Bruneval2006; @Botti2007]. Such a screened interaction was found to be crucial for the correct description of optical spectra [@Bruneval2006; @Paier2008].
Several different approximations for the screening of the non-local exchange interaction have been investigated [@Yang2015b; @Refaely-Abramson2015a; @Elliott2019; @Wing2019]. The results suggest that hybrid functionals yield spectra comparable to BSE-$GW$ provided the adopted fraction of Fock exchange accounts for the screening in the long range. In particular, Wing *et al.* obtained good results with screened range-separated hybrid functionals [@Wing2019]. However, the correct screening in the short and medium range was not imposed but rather followed from the empirical setting of the hybrid functional parameters in their TD-DFT approach. For instance, their choice of taking 25% of Fock exchange in the short range does not describe the physically correct behavior of the screening. More importantly, the range separation parameter was empirically tuned so that the calculated band gaps matched the $GW$ ones.
Recently, Chen *et al.* [@Chen2018] developed a nonempirical hybrid functional scheme, in which all the parameters are taken from the static screening without tuning. The method showed very accurate electronic structures and band gaps for a large number of semiconductors and insulators. The advantage of this approach is that it accurately accounts for the wave-vector dependent screening: at short-range the exchange interaction is only weakly screened, whereas in the long-range it is reduced by the static dielectric constant.
In this work, we investigate the performance of hybrid-functional TD-DFT for optical absorption calculations through the comparison with state-of-the-art BSE-$GW$. In the TD-DFT scheme, we employ hybrid functionals that have been designed to reproduce the correct screening properties through a self-consistent procedure [@Chen2018]. We show that this scheme provides absorption spectra with an accuracy comparable to that of BSE-$GW$ without tuning parameters. In particular, we show that equivalent descriptions of the screening in TD-DFT and BSE-$GW$ result in very similar absorption spectra.
Following recent work on dielectric-dependent hybrid functionals (DDH) [@Chen2018; @Cui2018; @Liu2020], we use in this work the explicit form of the exchange-correlation potential given by $$\label{eq:vxc}
\begin{aligned}
V_{\textrm{xc}}\left(\mathbf{r}, \mathbf{r}^{\prime}\right)=
\left[1-\left(1-\epsilon_{\infty}^{-1}\right)\textrm{erf}
\left(\mu\left|\mathbf{r}-\mathbf{r'}\right|\right)
\right]V_{\textrm{x}}^{\textrm{Fock}}\left(\mathbf{r},\mathbf{r'}\right)\\
+\left(1-\epsilon_{\infty}^{-1}\right)
V_{\textrm{x}}^{\textrm{PBE},\textrm{LR}}(\mathbf{r};\mu)
\delta\left(\mathbf{r}-\mathbf{r'}\right)+
V_{\textrm{c}}^{\textrm{PBE}}(\mathbf{r})
\delta\left(\mathbf{r}-\mathbf{r'}\right),
\end{aligned}$$ where $\mu$ is the range-separation parameter, $V_{\textrm{x}}^{\textrm{PBE}}$ and $V_{\textrm{c}}^{\textrm{PBE}}$ are the PBE exchange and correlation potentials [@Perdew1996]. Here, $V_x^{\textrm{Fock}}$ is the Fock exchange operator given by $$V_{\textrm{x}}^{\textrm{Fock}}\left(\mathbf{r,r'}\right)=
-e^{2} \frac{1}{N_\mathbf{k}}\sum_{n\mathbf{k}}
\frac{\psi^*_{n\mathbf{k}}\left(\mathbf{r}^{\prime}\right)
\psi_{n\mathbf{k}}(\mathbf{r})}
{\left|\mathbf{r-r'}\right|},$$ where $\psi_{n\mathbf{k}}$ are one-electron Bloch states, the sum over $\mathbf{k}$ is over $N_\mathbf{k}$ $\mathbf{k}$ points of the Brillouin zone, and the sum over $n$ is over all the occupied bands. In Eq. , the Fock exchange interaction is thus modified by the function $$\label{eq:eps}
c^\text{DDH}_\text{x}\left(\left|\mathbf{r}-\mathbf{r'}\right|\right) =
1-\left(1-\epsilon_{\infty}^{-1}\right)\textrm{erf}
\left(\mu\left|\mathbf{r}-\mathbf{r'}\right|\right).$$ In the $GW$ approximation, the Coulomb interaction is screened by the dielectric function which is a frequency dependent tensor $\epsilon^{-1}(\mathbf{r,r'},\omega)$ [@Aryasetiawan1998]. Thus, $c^\text{DDH}_\text{x}(|\mathbf{r}-\mathbf{r'}|)$ in Eq. corresponds to a model inverse dielectric function $\epsilon^{-1}_\textrm{model}\left(\left|\mathbf{r}-\mathbf{r'}\right|\right)$ that neglects the dynamical screening ($\omega \neq 0$) and the off-diagonal elements.
In the approach of Ref. [@Chen2018], the parameters in Eq. are determined self-consistently. In the long-wave limit, the interaction is set to $1/(\epsilon_\infty r)$, where the dielectric constant $\epsilon_\infty$ is calculated using the random-phase approximation with vertex corrections. The parameter $\mu$ is obtained by fitting the model to the calculated dielectric function. In Fig. \[fig:epsilon\], the model dielectric functions associated with the DDH are compared to the diagonal elements of the dielectric matrix at the $\Gamma$ point at zero frequency. The dielectric matrix is obtained within partially self-consistent $GW$ with vertex corrections \[cf. Supplemental Material (SM) [@Supp]\]. In all the cases, we find the model dielectric function to be in good agreement with the calculated dielectric function.
![Inverse dielectric functions vs. wave vector $G$ for Si, C, SiC, Ar, NaCl and MgO, as calculated in $GW$ and given by the model in Eq. with parameters taken from Ref. [@Chen2018].[]{data-label="fig:epsilon"}](Figure1.pdf){width="8.7cm"}
The excitation spectra in both BSE and TD-DFT are obtained by solving an eigenvalue problem, referred to as the Bethe-Salpeter and Casida equation, respectively [@Onida2002; @Sander2017]: $$\label{eq:casida}
\begin{pmatrix}
\mathbf{A} & \mathbf{B} \\
\mathbf{B^*} & \mathbf{A^*} \\
\end{pmatrix}
\begin{pmatrix}
\mathbf{X} \\
\mathbf{Y} \\
\end{pmatrix}
= \Omega
\begin{pmatrix}
\mathbb{1} & 0 \\
0 & \mathbb{-1} \\
\end{pmatrix}
\begin{pmatrix}
\mathbf{X} \\
\mathbf{Y} \\
\end{pmatrix},$$ where submatrices $\mathbf{A}$ and $\mathbf{B}$ read $$\label{eq:Aterm}
A_{ai,bj}=(\epsilon_a-\epsilon_i)\delta_{i,j}\delta_{a,b} +\bra{ib}K\ket{aj},$$ $$B_{ai,bj}=\bra{ij}K\ket{ab},$$ with the indices $i,j$ and $a,b$ referring to occupied and unoccupied states, respectively. The excitation frequencies of the system are given by $\Omega$. $X$ and $Y$ are the two-body electron-hole eigenstates in the transition basis $\psi_a(\mathbf{r})\psi_i^*(\mathbf{r'})$ and $\psi_i(\mathbf{r})\psi_a^*(\mathbf{r'})$. Matrix $A$ includes two terms, the energy of the direct transition from occupied to unoccupied states and the electron-hole interaction described by the kernel $K$ [@Strinati1984; @Rohlfing1998]. Equation is non-Hermitian, which makes it difficult to solve using standard eigenvalue solvers [@Sander2015; @Maggio2016]. A common practice to avoid this difficulty is to neglect the coupling between excitations and de-excitations by setting $B$ to zero. This approximation is known as the Tamm-Dancoff approximation.
The distinction between BSE-$GW$ and TD-DFT approaches results, on the one hand, from the origin of the one-particle eigenfunctions and energies and, on the other hand, from the type of the interaction kernel $K$. To make the comparison between different methods more transparent, we provide in Fig. \[fig:diagrams\] the Feynman diagrams corresponding to the various irreducible polarizabilities $\tilde{\chi}$ discussed in this work.
![Irreducible polarizabilities $\tilde{\chi}$ in various approximations. The reducible polarizability is obtained from $\chi=\tilde{\chi}+\tilde{\chi}v\chi$. The wiggly line indicates the screened interaction $W$.[]{data-label="fig:diagrams"}](Figure2.pdf){width="7.3cm"}
In BSE-$GW$, the orbitals and energies are derived from a preceding $GW$ calculation and the kernel consists of a Hartree term $V$ and a screened exchange term $W$[@Rohlfing1998]: $$\bra{ib}K\ket{aj}=2\bra{ib}V\ket{aj}-\bra{ib}W\ket{ja}.
\label{eq:bse}$$ The Hartree term describes the bare Coulomb interaction and is the same in all the approximations considered here. It can be included straightforwardly in a two-point formulation involving the polarizability $\chi$. The exchange term, however, requires calculating four-point integrals, which drastically increases the complexity of the problem. The screening of the exchange interaction is determined by the frequency-dependent dielectric function $\epsilon$ obtained from $GW$ and is represented by a vertical wiggly line in the diagrams. However, as it is shown in Refs. [@Rohlfing2000a; @Bechstedt1997; @Marini2003c], the dynamical effects can often be neglected in BSE calculations.
In the TD-DFT approach, the electron energies and wave functions are obtained from a semilocal or hybrid-functional calculation. The interaction kernel consists of three terms, a Hartree and an exchange term like in the BSE, and an additional local exchange-correlation interaction $f^\textrm{loc}_{\textrm{xc}}$[@Casida2012]: $$\bra{ib}K\ket{aj}=2\bra{ib}V\ket{aj}-\bra{ib}c_xv\ket{ja}+\bra{ib}f^\textrm{loc}_\textrm{xc}\ket{aj},
\label{eq:tddft}$$ where $$f_{\textrm{xc}}^\textrm{loc}(\mathbf{r,r'})=
\frac{\delta^{2}\left\{E_{c}^{\textrm{DFT}}+\left(1-c_{\textrm{x}}\right)
E_{x}^{\textrm{DFT}}\right\}}{\delta\rho(\mathbf{r})^2}\delta(\mathbf{r-r'}).
\label{eq:fxc}$$ The screening of the exchange interaction in TD-DFT is provided by a constant $c_\text{x}$ or a function $c_\text{x}(\left|\mathbf{r}-\mathbf{r'}\right|)$, depending on the exchange-correlation functional. In particular, $c_\text{x}=0$ for semilocal DFT functionals. In the case of DDH, the exchange interaction is screened by the model inverse dielectric function $c_\text{x} = c^\text{DDH}_\text{x}(\left|\mathbf{r}-\mathbf{r'}\right|)$ given in Eq. . The local exchange-correlation interaction $f^\textrm{loc}_{\textrm{xc}}$ is derived from the local part of the exchange-correlation potential $V_{\textrm{xc}}$ and is represented in Fig. \[fig:diagrams\] by a dotted line connecting $\chi_0$ and $\tilde{\chi}$. In this work, we refer to this version of TD-DFT as TD-DDH.
![Absorption spectra for Si, C, SiC, Ar, NaCl and MgO calculated with BSE-$GW$ and TD-DDH. Experimentally measured spectra are taken from [@Palik1998] for C, from [@Lautenschlager1987] for Si, from [@Logothetidis1996] for SiC, from [@Saile1976] for Ar, from [@Roessler1968] for NaCl, and from [@Bortz1990] for MgO.[]{data-label="fig:spectra"}](Figure3.pdf){width="8.6cm"}
Next, we focus on the comparison between BSE-$GW$ and TD-DDH. In both schemes, the absorption spectra are obtained from the eigenvalue problem in Eq. (\[eq:casida\]). In particular, the BSE-$GW$ calculations are based on partially self-consistent $GW$ using the “nanoquanta” vertex corrections $f_{\rm xc}$ in the polarizability $\tilde \chi$ [@Note]. These two approaches are tested on a set of materials possessing a wide range of band gaps. The corresponding spectra are given in Fig. \[fig:spectra\]. Our calculations show that both approaches agree well with experiment and that TD-DDH reproduces all the spectral features with the correct oscillator strengths. In the case of C, Si and SiC, the spectra are nearly on top of each other. For Ar, NaCl, and MgO, the relative positions of the main features in the spectra are found to be shifted slightly. In principle, this shift can result from differences in the band structure and in the screening. Our analysis indicates that the dominant effect is due to the energy transition terms in Eq. . From Table \[tab:gaps\], we notice that the calculated band gaps differ by less than 0.15 eV for Si, SiC, and C, but that the disagreement is more substantial for NaCl, Ar, and MgO. Overall, when compared to experimental values corrected for the coupling to phonons, DDH and $GW$ band gaps show mean average errors of 0.11 and 0.22 eV, respectively. These errors are consistent with the current accuracy of [*ab initio*]{} methods [@Shishkin2007; @Chen2015], indicating that the agreement with experiment should be considered excellent for both schemes. As far as the screening is concerned, we show below that the small discrepancies observed in Fig. \[fig:epsilon\] hardly change the spectra.
Si SiC C NaCl Ar MgO
------- -- ---------- ---------- ---------- ---------- ----------- ----------
PBE 0.75 1.35 4.14 5.21 8.70 4.77
DDH 1.31 2.50 5.69 9.13 14.60 8.41
$GW$ 1.41 2.55 5.85 8.86 13.75 8.12
Expt. 1.23[^1] 2.53[^2] 5.85[^3] 9.14[^4] 14.33[^5] 8.36[^6]
: Band gaps (in eV) obtained with the semilocal PBE functional [@Perdew1996], DDH, and $GW$. The experimental values are augmented by theoretical corrections resulting from the coupling to phonons.
\[tab:gaps\]
To compare the screening in BSE-$GW$ and TD-DDH, we show in Fig. \[fig:fxc\] the absorption spectra of diamond calculated in various approximations using the same energies and wave functions, which are taken from a $G_0W_0$ calculation. We start our analysis from a BSE calculation in which the full static inverse dielectric matrix is used ($W^{\rm full}$). In particular, we show that the off-diagonal elements of this matrix barely have any effect on the calculated spectrum ($W^{\rm diag}$). Next, we replace the inverse dielectric function with the model $c^\text{DDH}_\text{x}\left(\left|\mathbf{r}-\mathbf{r'}\right|\right)$ and find no discernible difference in the spectrum ($W^{\rm m}$). This treatment of the screening is equivalent to that of TD-DDH, where the local exchange-correlation kernel is neglected, i.e. $f^\textrm{loc}_\textrm{xc}=0$, also referred to as model BSE (mBSE) [@Bokdam2016]. To restore the full TD-DDH screening, we include the $f^\textrm{loc}_\textrm{xc}$ and still obtain essentially the same spectrum ($W^{\rm m} + f^\textrm{loc}_\textrm{xc}$). Hence, these results indicate that the model screening in TD-DDH gives an accurate description of the screening in $W= \epsilon^{-1} V$ and that the effect of $f^\textrm{loc}_\textrm{xc}$ is negligible for extended systems.
![Optical absorption spectrum of diamond calculated with BSE and full $\epsilon^{-1}$ from $GW$ ($W^\mathrm{full}$), BSE and diagonal $\epsilon^{-1}$ ($W^\mathrm{diag}$), BSE and model $\epsilon^{-1}_\textrm{model}$ ($W^\textrm{m}$), TD-DDH and model $\epsilon^{-1}_\textrm{model}$ with $f^\textrm{loc}_\textrm{xc}$ ($W^\textrm{m}+f^\textrm{loc}_\textrm{xc}$). All spectra are based on energies and wave functions from a $G_0W_0$ calculation. A $6\times6\times6$ [**k**]{}-point grid is used.[]{data-label="fig:fxc"}](Figure4.pdf){width="8cm"}
The numerical complexity of Eq. (\[eq:casida\]) is the same in BSE and TD-DDH. However, the preceding $GW$ calculations required in BSE-$GW$ involve a high computational cost, which scales like $N^4$ in the number of electrons $N$ in most $GW$ implementations instead of like $N^3$ in TD-DFT. Additionally, in the calculation of the Green’s function in $GW$, the convergence with respect to the number of unoccupied states and the number of frequency points has to be controlled carefully, which significantly increases the complexity of the calculations. Note that the static dielectric constant only needs to be determined at the $\Gamma$ point of the Brillouin zone and that it converges quickly with respect to the number of included orbitals. Furthermore, the hybrid-functional approach only requires a model static dielectric function, for which the static limit can be obtained rather efficiently [@Cui2018]. Thus, the hybrid functional approach opens the way to more efficient numerical schemes that can circumvent the calculation of the full dielectric matrix.
In conclusion, we have shown that time-dependent calculations using the parameter-free DDH functional yield optical absorption spectra with an accuracy comparable to BSE-$GW$. The success of this approach originates from the use of a model dielectric function that gives a physically motivated description of the screened exchange interaction over the full spatial range. Notably, the computational complexity of the method is drastically reduced compared to BSE-$GW$, as it eliminates the need for preceding $GW$ calculations. This will allow one to consider larger and more complex systems than hitherto possible.
Support from the Swiss National Science foundation is acknowledged under Grant No. 200020-172524. The calculations have been performed at the Swiss National Supercomputing Centre (CSCS) (grant under project ID s879) and at SCITAS-EPFL.
[^1]: Ref. [@Bludau1974], with a correction of 0.06 eV from Ref. [@Cardona2005].
[^2]: Ref. [@Humphreys1981], with a correction of 0.11 eV from Ref. [@Monserrat2014].
[^3]: Ref. [@Clark1964], with a correction of 0.37 eV from Ref. [@Cardona2005].
[^4]: Ref. [@Roessler1968], with a correction of 0.17 eV from Ref. [@Lambrecht2017].
[^5]: Ref. [@Baldini1962], with a correction of 0.03 eV calculated using the method described in Refs. [@Zacharias2016; @Karsai2018].
[^6]: Ref. [@Hinuma2014], with a correction of 0.53 eV from Ref. [@Nery2018].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We undertake a combinatorial study of the piecewise linear map $g:
{\mathbb{R}}^{2(m+n)} \rightarrow {\mathbb{R}}^{m \times n}$ which assigns to the four vectors $a, A$ in ${\mathbb{R}}^m$ and $b, B$ in ${\mathbb{R}}^n$ the $m \times n$ matrix given by $g_{ij}=\min(a_i+b_j, A_i+B_j)$. This map arises naturally in Pachter and Sturmfels’s work on the tropical geometry of statistical models. The image of $g$ has been a subject of recent interest; it is the positive part of the tropical algebraic variety which parameterizes $n$-tuples of points on a tropical line in $m$-space.
The domains of linearity of $g$ are the regions of the real hyperplane arrangement ${{\cal A}}_{m,n}$, corresponding to the complete bipartite graph $K_{m,n}$. We explain how the images of (some of) the regions provide two polyhedral subdivisions of the image of $g$, one of which is a refinement of the other. The finer subdivision is particularly nice enumeratively: it has $2{m
\choose 2}{n \choose 2}r_{m-2,n-2}$ maximum-dimensional cells, where $r_{m-2,n-2}$ is the number of regions of the arrangement ${{\cal A}}_{m-2,n-2}$.
author:
- 'Federico Ardila [^1]'
title: A tropical morphism related to the hyperplane arrangement of the complete bipartite graph
---
Introduction {#sec:intro}
============
0\. The goal of this paper is to undertake a combinatorial study of the piecewise linear map $g: {\mathbb{R}}^{2(m+n)} \rightarrow {\mathbb{R}}^{m \times
n}$ given by $$g(a,A,b,B)_{ij} = \min(a_i + b_j, A_i + B_j) \qquad \mathrm{for}
\,\,\,\, 1 \leq i \leq m, 1 \leq j \leq n,$$ where $a$ and $A$ denote vectors in ${\mathbb{R}}^m$ and $b$ and $B$ denote vectors in ${\mathbb{R}}^n$.
The paper is organized as follows. In Section \[sec:faces\], we describe and enumerate the domains of linearity of the map $g$; they are precisely the faces of the arrangement ${{\cal A}}_{m,n}$, corresponding to the complete bipartite graph $K_{m,n}$. In Section \[sec:images\] we describe the images $g(R)$, as $R$ ranges over the regions of the arrangement. In Section \[sec:subdivisions\] we show that these images fit together in an unusual way to give the full image ${\mathrm{Im}}(g)$; as a consequence, we obtain two different polyhedral subdivisions of ${\mathrm{Im}}(g)$, one of which is a refinement of the other. The finer subdivision is particularly nice combinatorially; the number of its facets is $2{m \choose 2}{n \choose 2}r_{m-2,n-2}$, where $r_{m-2,n-2}$ is the number of regions of the arrangement ${{\cal A}}_{m-2,n-2}$.
This project was suggested in a recent paper of Pachter and Sturmfels [@Pachter]. It arose in their study of the tropical geometry of statistical models, and provides a new perspective on Develin, Santos and Sturmfels’s study of the image of $g$ [@Develin; @troprank].
To explain their motivation, we start with a brief overview of some recent developments in tropical geometry which led to its consideration.
1\. The *tropical semiring* $({\mathbb{R}}\cup \{\infty\},
{\oplus}, {\odot})$ is the set of real numbers augmented by infinity, together with the operations of *tropical addition and multiplication*, which are defined by $x {\oplus}y = \min(x,y)$ and $x
{\odot}y = x+y.$
*Tropical algebraic geometry* is, roughly speaking, the geometry of the tropical semiring. Let $K = {\mathbb{C}}\{\{t\}\}$ be the (algebraically closed) ring of Puiseux series; its elements are formal power series of the form $f = c_1 t^{a_1} + c_2 t^{a_2} +
\cdots$, where $a_1<a_2<\cdots$ are rational numbers with a common denominator. There is a natural valuation $\deg: K^* \rightarrow
{\mathbb{Q}}$, which sends the non-zero Puiseux series $f$ to its degree $a_1$.
For an ideal $I$ in $K[x_1, \ldots, x_n]$, let $V(I)$ be the corresponding algebraic variety, intersected with the torus $(K^*)^n$. It consists of the $n$-tuples $u(t) = (u_1(t), \ldots,
u_n(t))$ of non-zero Puiseux series such that $f(u(t))=0$ for all $f \in I$. Define $\deg u(t) = (\deg u_1(t), \ldots, \deg
u_n(t))$.
[@Kapranov; @Speyer; @Sturmfels1] For an ideal $I$ in $K[x_1, \ldots, x_n]$, the following subsets of ${\mathbb{R}}^n$ coincide.
1. The topological closure of $\deg V(I)$.
2. The set of $w \in {\mathbb{R}}^n$ such that the initial ideal ${\mathrm{in}}_w(I)$ contains no monomials.
This set is called the *tropical variety* of $I$, and denoted ${\mathrm{trop\,}}V(I)$.
Let $V^+(I)$ be the set of $n$-tuples $u(t)$ in $V(I)$ such that the coefficient of the leading term of each $u_i(t)$ is a positive real number.
[@Lauren] For an ideal $I$ in $K[x_1, \ldots, x_n]$, the following subsets of ${\mathbb{R}}^n$ coincide.
1. The topological closure of $\deg V^+(I)$.
2. The set of $w \in {\mathbb{R}}^n$ such that ${\mathrm{in}}_w(I) \cap {\mathbb{R}}^+[x_1,
\ldots, x_n] = \emptyset$.
This set is called the *positive part of the tropical variety* of $I$, and denoted ${\mathrm{trop^+\,}}V(I)$.
For all $I$, ${\mathrm{trop\,}}V(I)$ is a polyhedral complex, and ${\mathrm{trop^+\,}}V(I)$ is a subcomplex of it.
2\. Let $f=(f_1, \ldots, f_n):{\mathbb{R}}^d \rightarrow {\mathbb{R}}^n$ be a polynomial map. Say that $f$ is *positive* if each polynomial $f_i$ has only positive coefficients. Say that $f$ is *surjectively positive* if, additionally, every point in the image of $f$ whose coordinates are positive has a preimage whose coordinates are positive; *i.e.*, $f({\mathbb{R}}_{>0}^d) = {\mathrm{Im}}f \cap
{\mathbb{R}}_{>0}^n$.
The *tropicalization of $f$* is the piecewise linear map $g:{\mathbb{R}}^d \rightarrow {\mathbb{R}}^n$ obtained from $f$ by replacing every $\times$ with a ${\odot}$ and every $+$ with a ${\oplus}$. Such a map is called a *tropical morphism*. [@Pachter]
The relationship between the positive part of a tropical variety and morphisms is outlined in the following theorem.
[@Pachter; @Lauren] Let $f: {\mathbb{R}}^d \rightarrow {\mathbb{R}}^n$ be a polynomial map, and let $g$ be the tropicalization of $f$. The image ${\mathrm{Im}}(f)$ is an algebraic variety in ${\mathbb{R}}^n$; let $I$ be its corresponding ideal.
1. ${\mathrm{Im}}(g) \subset {\mathrm{trop\,}}V(I)$.
2. If $f$ is positive, ${\mathrm{Im}}(g) \subset {\mathrm{trop^+\,}}V(I)$.
3. If $f$ is surjectively positive, ${\mathrm{Im}}(g) = {\mathrm{trop^+\,}}V(I)$.
3\. Pachter and Sturmfels [@Pachter] used this setup to study the tropical geometry of statistical models. The *naive Bayes model with two features* [@Garcia] is parameterized by the polynomial map $f: {\mathbb{R}}^{2(m+n)} \rightarrow
{\mathbb{R}}^{m \times n}$ which maps an $m \times 2$ matrix and a $2 \times
n$ matrix to their product. Its tropicalization, $g$, is the object of study of this paper.
The image of $f$ consists of the real matrices of rank at most $2$. This is an algebraic variety, whose corresponding prime ideal $I$ is generated by the $3 \times 3$ subdeterminants of an $m
\times n$ matrix.
The tropical variety ${\mathrm{trop\,}}V(I)$ is the set of $m \times n$ real matrices having tropical rank at most $2$, and the positive part ${\mathrm{trop^+\,}}V(I)$ is the set $B_{m,n}$ of $m \times n$ real matrices having Barvinok rank at most $2$. For the relevant definitions and further information, see [@Develin; @troprank; @Pachter].
This model interests us because the map $f$ is surjectively positive:
[@Cohen] Every positive $m \times n$ matrix of rank $2$ can be written as the product of a positive $m \times 2$ matrix and a positive $2
\times n$ matrix.
It follows that ${\mathrm{Im}}(g) = {\mathrm{trop^+\,}}V(I)=B_{m,n}$.
The study of the map $g$ is therefore closely related to Pachter and Sturmfels’s study of the tropical geometry of statistical models, and to Develin, Santos and Sturmfels’s study of the space $B_{m,n}$.
The morphism and its domains of linearity. {#sec:faces}
==========================================
The piecewise linear map $g: {\mathbb{R}}^{2(m+n)} \rightarrow {\mathbb{R}}^{m \times
n}$ which we wish to study is the tropicalization of matrix multiplication; it is given by $$g \left( \begin{bmatrix} & \\ a & A\\ & \end{bmatrix},
\begin{bmatrix} \qquad & b & \qquad \\ \qquad & B & \qquad
\end{bmatrix} \right) = \begin{bmatrix} & \\ a & A\\ &
\end{bmatrix} {\odot}\begin{bmatrix} \qquad & b & \qquad \\ \qquad & B &
\qquad
\end{bmatrix}.$$ Here $a$ and $A$ denote vectors in ${\mathbb{R}}^m$ and $b$ and $B$ denote vectors in ${\mathbb{R}}^n$. Let $x_i = a_i - A_i$ and $y_j = B_j - b_j$. The entry $i,j$ of $g$ is given by $$g_{ij} = \min(a_i + b_j, A_i + B_j) = \left\{ \begin{matrix} a_i +
b_j, & \mbox{if } x_i \leq y_j \\ A_i + B_j, & \mbox{if } x_i >
y_j. \end{matrix} \right.$$
The piecewise linear map $g$ is linear on the faces of the hyperplane arrangement in ${\mathbb{R}}^{2(m+n)}$: $${{\cal A}}_{m,n}: \qquad x_i = y_j \qquad 1 \leq i \leq m, 1 \leq j \leq
n.$$ This is essentially the *graphical arrangement* of the complete bipartite graph $K_{m,n}$; it contains one hyperplane for each edge of $K_{m,n}$.
Therefore, to understand the fibers of the map $g$, we first study the faces of ${{\cal A}}_{m,n}$.
Let us briefly outline the close connection between the arrangement ${{\cal A}}_{m,n}$ and the graph $K_{m,n}$. The matroid $M({{\cal A}}_{m,n})$ associated to the arrangement is isomorphic to the matroid $M(K_{m,n})$ associated to the graph. In other words, a subset of $k$ hyperplanes of ${{\cal A}}_{m,n}$ intersects in codimension $k$ if and only if the corresponding subset of $k$ edges of $K_{m,n}$ does not contain a cycle. Let us call this matroid simply $M_{m,n}$; it encodes much of the combinatorial information of the arrangement of the graph. For more details, see [@Orlik Ch. 2] or [@Oxley Ch.5].
The regions of the arrangement ${{\cal A}}_{m,n}$ are in one-to-one correspondence with the acyclic orientations of the bipartite graph $K_{m,n}$, as follows: Denote the vertices of $K_{m,n}$ by $u_1, \ldots, u_m, v_1, \ldots, v_n$, and consider an acyclic orientation ${{\it o}}$ of $K_{m,n}$. The corresponding region $R({{\it o}})$ consists of the points $(x,y) \in {\mathbb{R}}^{m+n}$ such that $x_i < y_j$ if the edge $u_iv_j$ is directed $u_i \rightarrow v_j$ in ${{\it o}}$, and $x_i > y_j$ otherwise.
The matroid $M_{m,n}$ appeared independently in the work of Martin and Reiner [@Martin]. They used the finite field method of [@mitesis] to compute a generating function for ${\overline{\chi}}_{m,n}(q,t)$, the coboundary polynomial of $M_{m,n}$. This polynomial is a simple transformation of the Tutte polynomial, and it captures much of the interesting enumerative information of the matroid.
[@Martin] If ${\overline{\chi}}_{m,n}(q,t)$ is the coboundary polynomial of $M_{m,n}$, $$1 + q \left( \sum_{(m,n) \in {\mathbb{N}}^2 - \{(0,0)\}} {\overline{\chi}}_{m,n} (q,t)
\frac{x^m}{m!} \frac{y^n}{n!} \right) = \left( \sum_{(m,n) \in
{\mathbb{N}}^2} t^{mn} \frac{x^m}{m!} \frac{y^n}{n!} \right)^q.$$
\[orientations\] [@EC2 Ex. 5.6] Let $r_{m,n}$ be the number of acyclic orientations of $K_{m,n}$ (or equivalently, the number of regions of ${{\cal A}}_{m,n}$). Then $$\sum_{m,n \geq 1} r_{m,n} \frac{x^m}{m!} \frac{y^n}{n!} =
\frac{e^{x+y}}{e^x+e^y-e^{x+y}}.$$
This follows from the formula $r(m,n)=(-1)^{m+n-1}{\overline{\chi}}_{m,n}(-1,0)$ [@Zaslavsky] for the number of regions of a real arrangement.
We now proceed to describe and count the domains of linearity of $g$; that is, the faces of the arrangement ${{\cal A}}_{m,n}$.
\[faces\] Let $f_{k,m,n}$ be the number of $k$-dimensional faces of the arrangement ${{\cal A}}_{m,n}$. Then $$\sum_{k,m,n \geq \, 0} f_{k,m,n} \, t^k \frac{x^m}{m!} \frac
{y^n}{n!} = \frac{1}{e^{-tx} + e^{-ty} - t(e^x-1)(e^y-1) -1}.$$
To describe a face $F$ of the arrangement we must specify, for each hyperplane $x_i = y_j$ $(1 \leq i \leq m, \, 1 \leq j \leq
n)$, whether $F$ is in the halfspace $x_i > y_j$, in the halfspace $x_i<y_j$, or on the hyperplane $x_i=y_j$. Consider the inequalities or equalities $x_i \bigcirc y_j$ that define a face $F$. Each one of them puts a restriction on the relative order of the variables $x_1, \ldots, x_m, y_1, \ldots, y_n$. For a point $(x,y)$ in $F$, the relative order of $x_i$ and $y_j$ is determined for all $i$ and $j$. The relative order of $x_{i_1}$ and $x_{i_2}$ is not always determined: an equality $x_{i_1} =
x_{i_2}$ can only be deduced from two defining equalities of the form $x_{i_1} = y_j$ and $x_{i_2} = y_j$. Similarly, an inequality $x_{i_1} < x_{i_2}$ can only be deduced as a consequence of two defining inequalities or equalities of the form $x_{i_1} \bigcirc
y_j$ and $x_{i_2} \bigcirc y_j$.
The faces of ${{\cal A}}_{m,n}$ are described putting these restrictions together. For example, one face of ${{\cal A}}_{7,8}$ consists of the points $(x,y) \in {\mathbb{R}}^{15}$ such that $$x_1, x_3 < y_2, y_5, y_7 < x_5 < x_2=y_3=y_6 < y_1 < x_6 < x_4 =
x_7 = y_4 < y_8.$$
Each face of ${{\cal A}}_{m,n}$ can be described in a similar way, as a sequence of blocks of variables. The variables $x_{i_1}$ and $x_{i_2}$ are in the same block if the comparisons $x_{i_1}
\bigcirc y_j$ and $x_{i_2} \bigcirc y_j$ yield the same result for each $j$. A similar statement holds for $y_{j_1}$ and $y_{j_2}$. The variables $x_i$ and $y_j$ are in the same block if $x_i = y_j$ in $F$.
Call a block *positive* if it only contains $x_i$s, *negative* if it only contains $y_j$s, and *mixed* otherwise. For a point to belong to the face, the variables within each mixed block are equal. Within an unmixed block, the relative order of the variables is not determined.
It follows that the faces of ${{\cal A}}_{m,n}$ are in one-to-one correspondence with the ordered partitions of the set $\{x_1,
\ldots, x_m, y_1, \ldots, y_n\}$ containing no two consecutive unmixed blocks of the same sign. The dimension of a face is easily determined from the partition: it is equal to the sum of the sizes of the unmixed blocks plus the number of mixed blocks. We can now use the methods of [@EC2 Chapter 5] to compute the desired generating function.
The generating function for non-empty positive blocks is $X(t,x,y)
= \sum_{n \geq 1} t^n \frac{x^n}{n!} = e^{tx}-1$. The generating function for non-empty negative blocks is $Y(t,x,y) = e^{ty}-1$. Therefore, the generating function for partitions of $\{x_1,
\ldots, x_m, y_1, \ldots, y_n\}$ into unmixed blocks of alternating sign is $$\begin{aligned}
Z(t,x,y) &=& (1+X)(1+YX+YXYX+YXYXYX+\cdots)(1+Y) \\ & = &
\frac{1}{e^{-tx} + e^{-ty}-1}.\end{aligned}$$
On the other hand, the generating function for mixed blocks is given by $M(t,x,y) = \sum_{m,n \geq 1} t \frac{x^m}{m!}
\frac{y^n}{n!} = t(e^x-1)(e^y-1)$. The partitions we wish to count are alternating sequences of partitions of the type counted by $Z$ (which may be empty) and mixed blocks. Therefore
$$\begin{aligned}
\sum_{k,m,n \geq \, 0} f_{k,m,n} \, t^k \frac{x^m}{m!} \frac
{y^n}{n!} = Z+ZMZ+ZMZMZ+\cdots,\end{aligned}$$
which is equal to the given expression.
Observe that, under the above correspondence, the regions of ${{\cal A}}_{m,n}$ correspond to the ordered partitions which contain no mixed blocks. These are counted by $Z(t,x,y)$; setting $t=1$ recovers the generating function for $r_{m,n}$.
It will be convenient to label each region $R$ of the arrangement with the permutation $\pi(R)$ of the set $[m,{\overline{n}}] = \{1,2,\ldots,m,
{\overline{1}}, {\overline{2}}, \ldots, {\overline{n}}\}$ obtained by reading the blocks from left to right. The variable $x_i$ corresponds to the letter $i$ (which we call a *positive letter*), and the variable $y_j$ corresponds to the letter ${\overline{j}}$ (which we call a *negative letter*). Within each block, the letters are arranged in increasing order.
For example, the region $$y_2, y_4 < x_3 < y_1 < x_1, x_2 < y_3, y_5$$ of ${{\cal A}}_{3,5}$ is simply denoted by the permutation ${\overline{2}}{\overline{4}}3{\overline{1}}12{\overline{3}}{\overline{5}}$.
This labelling is a bijection between the regions of ${{\cal A}}_{m,n}$ and the permutations of $[m,{\overline{n}}]$ such that any two consecutive letters of the same sign are in increasing order.
The image of $g$ in each region {#sec:images}
===============================
Consider a region $R$ of ${{\cal A}}_{m,n}$. As we mentioned earlier, the restriction of the map $g$ to the region $R$ is a linear function. We now describe the image $g(R)$.
Color each entry of the $m \times n$ matrix $g$ either *black* or *white*: the entry $g_{ij}$ is black if $a_i-A_i > B_j-b_j$ in $R$, and white if $a_i-A_i < B_j-b_j$ in $R$. Permute the rows and columns of $g$ according to the order in which their labels appear in $\pi$. A path $P$ separates the white and black entries: it starts at the northwest corner of the matrix, and takes a step south for each positive letter in $\pi$ and a step east for each negative letter in $\pi$, in the order prescribed by $\pi$.
![The diagram of the region ${\overline{2}}{\overline{4}}3{\overline{1}}12{\overline{3}}{\overline{5}}$.[]{data-label="matriz"}](matriz){width="4in"}
Figure \[matriz\] shows the resulting matrix and path for $\pi =
{\overline{2}}{\overline{4}}3{\overline{1}}12{\overline{3}}{\overline{5}}$. The entry $g_{ij}$ is $A_i + B_j$ if it is below $P$ or $a_i+b_j$ if it is above $P$. We call this picture the *diagram* of $R$ or of $g(R)$.
For $g \in {\mathbb{R}}^{m \times n}$, write $$\Delta_{i_1i_2j_1j_2}(g) = g_{i_1j_1} + g_{i_2j_2} -
g_{i_1j_2}-g_{i_2j_1}.$$ For simplicity, we will omit $g$ from the notation and simply write $\Delta_{i_1i_2j_1j_2}$ for this expression. We will call an equality or inequality of the form $\Delta_{i_1i_2j_1j_2} \bigcirc
\, 0$ a *rectangle relation*. Write $i < {\overline{j}}$ when $i$ appears before ${\overline{j}}$ in $\pi(R)$.
\[im(R)1\] (Version 1.) The image $g(R)$ is an open polytope, described by the rectangle equalities and inequalities that it satisfies. They are the following: $$\begin{aligned}
\label{image}
\Delta_{i_1i_2j_1j_2} &=& 0 \textrm{ if } i_1,i_2<{\overline{j}}_1,{\overline{j}}_2, \\
\Delta_{i_1i_2j_1j_2} &=& 0 \textrm{ if } {\overline{j}}_1, {\overline{j}}_2 < i_i, i_2, \\
\Delta_{i_1i_2j_1j_2} &>& 0 \textrm{ if } i_1<{\overline{j}}_1<i_2<{\overline{j}}_2, \\
\Delta_{i_1i_2j_1j_2} &>& 0 \textrm{ if } {\overline{j}}_1 < i_1<{\overline{j}}_2<i_2, \\
\Delta_{i_1i_2j_1j_2} &>& 0 \textrm{ if } i_1 < {\overline{j}}_1<i<{\overline{j}}_2<i_2 \textrm{ for some } i, \\
\Delta_{i_1i_2j_1j_2} &>& 0 \textrm{ if } {\overline{j}}_1 < i_1<{\overline{j}}<i_2<{\overline{j}}_2
\textrm{ for some ${\overline{j}}$,}\end{aligned}$$ and no others.
In the diagram of $R$, let $\Box_{i_1i_2j_1j_2}$ be the sub-rectangle of the diagram containing rows $i_1$ through $i_2$ and columns ${\overline{j}}_1$ through ${\overline{j}}_2$. (We implicitly assume that $i_1
< i_2$ and ${\overline{j}}_1 < {\overline{j}}_2$ in $\pi(R)$.) Call this sub-rectangle *monochromatic* if all its entries have the same color. Call it *sliced* if it has black and white entries, separated by a single vertical or horizontal line. Call it *jagged* otherwise. These definitions are illustrated in Figure \[rectangles\].
![The three types of rectangle: monochromatic, sliced, and jagged.[]{data-label="rectangles"}](rectangles){width="5in"}
\[im(R)2\] (Version 2.) The image $g(R)$ consists of those $g \in {\mathbb{R}}^{m \times n}$ such that $\Delta_{i_1i_2j_1j_2}$ is:
- equal to zero if $\Box_{i_1i_2j_1j_2}$ is monochromatic,
- positive if $\Box_{i_1i_2j_1j_2}$ is jagged.
(If $\Box_{i_1i_2j_1j_2}$ is sliced, $\Delta_{i_1i_2j_1j_2}$ takes positive and negative values in $R$.)
It is straightforward to verify that the two versions of Proposition \[im(R)1\] are equivalent. Consider a point $(a,A,b,B) \in {\mathbb{R}}^{2(m+n)}$ in $R$. The entry $g_{ij}$ of $g(a,A,b,B)$ is given by $a_i+b_j$ if it is white (*i.e.* if $i<{\overline{j}}$ in $\pi(R)$), or $A_i+B_j$ if it is black (*i.e.* if $i>{\overline{j}}$ in $\pi(R)$). These entries are easily seen to satisfy the given equalities and inequalities.
Conversely, consider a matrix $g \in {\mathbb{R}}^{m \times n}$ which satisfies the given equalities and inequalities. Permute its columns and rows according to the order in which their labels appear in $\pi$, and draw the path $P$.
Choose appropriate values of the $A_i$s and $B_j$s so that $g_{ij}=A_i+B_j$ for all the southernmost and westernmost black entries (the ones surrounded by a box in Figure \[matriz\]). This can be done because the system of equations is independent, and has more unknowns than equations. Make the $A_i$s and $B_j$s which do not appear in this system of equations very large positive numbers. Similarly, choose appropriate values of the $a_i$s and $b_j$s so that $g_{ij}=a_i+b_j$ for all the northernmost and easternmost white entries, and make the other $a_i$s and $b_j$s very large positive numbers.
Since $g$ satisfies the rectangle equalities of $g(R)$, $g_{ij}=A_i+B_j$ for all black entries and $g_{ij}=a_i+b_j$ for all the white entries.
Take a black entry $g_{ij}$. If the northernmost entry on its column or the easternmost entry on its row is black, then $a_i+b_j$ is a very large positive number, larger than $A_i+B_j$. Otherwise, the rectangle determined by it and the northeasternmost corner of the matrix has three white entries and one black entry, and the corresponding rectangle inequality is equivalent to $A_i+B_j < a_i + b_j$. Similarly, $a_i+b_j < A_i + B_j$ for white entries $g_{ij}$. Therefore $(a,A,b,B) \in R$ and $g =
g(a,A,b,B)$.
Let $R$ be a region of ${{\cal A}}_{m,n}$. Then ${\mathrm{span\,}}g(R)$ is the subspace of ${\mathbb{R}}^{m \times n}$ described by the equalities $$g_{i_1j_1} + g_{i_2j_2}= g_{i_1j_2}+g_{i_2j_1}$$ for those $i_1,i_2,j_1,j_2$ such that $g_{i_1j_1}, g_{i_2j_2},
g_{i_1j_2}$ and $g_{i_2j_1}$ have the same color.
For a region $R$ let ${\mathrm{first}}(R)$ be the length of the first block of letters of $\pi(R)$ of the same sign (or equivalently, the number of sources of the corresponding orientation ${{\it o}}(R)$), and let ${\mathrm{last}}(R)$ be the length of the last block of letters of $\pi(R)$ of the same sign (or equivalently, the number of sinks of ${{\it o}}(R)$).
Let $R_1$ be the region where $x_i < y_j$ for all $i$ and $j$, and let $R_2$ be the region where $x_i > y_j$ for all $i$ and $j$.
Let $R$ be a region of ${{\cal A}}_{m,n}$. If $R \neq R_1, R_2$, then $$\dim g(R) = 2m + 2n - 2 - {\mathrm{first}}(R) - {\mathrm{last}}(R).$$ Otherwise, $\dim g(R_1) = \dim g(R_2)= m+n-1$.
The southernmost and westernmost black entries and the northernmost and easternmost white entries linearly generate the remaining ones, and there are no linear relations among them. The number of them is as claimed.
In particular, the regions where $g$ has maximum rank are those corresponding to acyclic orientations of $K_{m,n}$ with a unique source and a unique sink. This maximum rank is equal to $2m+2n-4$.
Two subdivisions of the image of $g$. {#sec:subdivisions}
=====================================
The closures of the regions of the hyperplane arrangement ${{\cal A}}_{m,n}$ give a polyhedral covering of ${\mathbb{R}}^{2(m+n)}$. Their images under the map $g$ give a covering of the full image, ${\mathrm{Im}}(g)$. We now wish to understand how the images $\overline{g(R)}$ of the closures of the regions $R$, which we call the *cells*, fit together.
First notice that if $R$ is a region of ${{\cal A}}_{m,n}$ and $-R$ is its negative (so the permutation $\pi(-R)$ is equal to the permutation $\pi(R)$ reversed), then it follows from Proposition \[im(R)1\] that $\overline{g(R)} = \overline{g(-R)}$. Therefore, we can restrict our attention only to the *positive* regions, where $x_1 < y_1$ (or equivalently, $1 < {\overline{1}}$ in $\pi(R)$).
Also, the following proposition shows that it suffices to study the maximum-dimensional cells.
[@Develin] The image ${\mathrm{Im}}(g)=B_{m,n}$ is pure.
Interestingly, though, our collection of cells *is not* pure-dimensional. For example, the image of the region $1{\overline{1}}{\overline{2}}\ldots{\overline{n}}23\ldots m$ is a subspace which is not contained in any maximum-dimensional cell.
Recall that the cell $\overline{g(R)}$ is maximum-dimensional if and only if the first and the last block of $\pi(R)$ are singletons. Call such a maximum-dimensional cell *large* if the second and the second-to-last blocks of $\pi(R)$ are not singletons, *small* if the second and the second-to-last blocks of $\pi(R)$ are both singletons, and *medium* if one of them is a singleton and the other one is not.
\[subdivisions\] The large cells form a polyhedral subdivision of $B_{m,n}$. The small cells form a finer subdivision of $B_{m,n}$.
The proof is divided into five steps. We start by showing that each small or medium cell is contained in a large cell; therefore, the large cells cover ${\mathrm{Im}}(g)$. Secondly, we show that each large cell is subdivided into small cells; therefore, the small cells cover ${\mathrm{Im}}(g)$ also. The third step is to show that large cells are pairwise interior-disjoint, and so are small cells. Then we prove that the covering of ${\mathrm{Im}}(g)$ with small cells is a subdivision. Finally, we prove that the covering with large cells is also a subdivision.
1\. Take any small or medium cell $\overline{g(R_1)}$; we want to find a large cell containing it. Say $\pi(R_1) =
{\overline{j}}_1i_1{\overline{j}}_2\ldots{\overline{j}}_ri_2\ldots$. Let $R$ be the region with the label $\pi(R) = i_1{\overline{j}}_1{\overline{j}}_2\ldots{\overline{j}}_ri_2\ldots$: we have adjusted the beginning of the permutation so that the second block is not a singleton. This construction is illustrated in Figure \[maximal\]. We will show that $\overline{g(R_1)} \subset
\overline{g(R)}$. If $\overline{g(R_1)}$ is medium, then $\overline{g(R)}$ is large and we are done. If it is small, then $\overline{g(R)}$ is medium; we can then adjust the end of the permutation $\pi(R)$ in the same way, to obtain a large cell containing $\overline{g(R)}$.
![A medium cell and its associated large cell.[]{data-label="maximal"}](maximal){width="5in"}
We need to show that every rectangle relation satisfied by $g(R)$ is also satisfied by $g(R_1)$. This analysis is most easily carried out in terms of Version 2 of Proposition \[im(R)2\]. Notice that $R$ and $R_1$ have exactly the same monochromatic sub-rectangles. Also, the only sub-rectangles which are *sliced* in one diagram and not in the other are $\Box_{i_1ij_1j_s}$ for $2 \leq s \leq r$ and $i \neq i_1$.
It follows that $g(R_1)$ has the same rectangle relations that $g(R)$ has, and the additional relations $\Delta_{i_1ij_1j_s}>0$ for $2\leq s \leq r$ and $i \neq i_1$. The rectangle equalities imply that $\Delta_{i_1ij_1j_s} = \Delta_{i_1i_2j_1j_s}$ for $2
\leq s \leq r$ and $i \neq i_1$. In conclusion, $g(R_1)$ is the subset of $g(R)$ satisfying the extra relations: $$g_{i_1j_1} - g_{i_2j_1} > g_{i_1j_s} - g_{i_2j_s}$$ for $2 \leq s \leq r$.
We conclude that any maximum-dimensional cell is contained in a large cell, so the large cells cover ${\mathrm{Im}}(g)$ by themselves, as desired.
2\. Now, let us describe which medium and small cells are contained in a given large cell. Let $R$ be as defined above, and let $R_s$ be the region with $\pi(R_s) =
{\overline{j}}_si_1{\overline{j}}_1\ldots\widehat{{\overline{j}}_s}\ldots{\overline{j}}_ri_2\ldots$ for $1 \leq s
\leq r$. Imitating the argument above, we conclude that $g(R_s)$ is the subset of $g(R)$ such that $g_{i_1j_s} - g_{i_2j_s}$ is the unique largest element in $\{g_{i_1j_t} - g_{i_2j_t}: 1 \leq t
\leq r\}$. It follows that $\{\overline{g(R_s)}: 1 \leq s \leq
r\}$ is a subdivision of $\overline{g(R)}$.
In general, the above argument shows that the large cell $\overline{g(R)}$ is subdivided into $st$ small cells, where $s$ and $t$ are the lengths of the second and second-to-last blocks of $\pi(R)$. (There are also $s+t$ medium cells in $\overline{g(R)}$. The first $s$ medium cells form a subdivision of $\overline{g(R)}$ and are subdivided into $t$ small cells each, The other $t$ medium cells also form a subdivision of $\overline{g(R)}$ and are subdivided into $s$ small cells each.) In particular, the small cells also cover ${\mathrm{Im}}(g)$.
3\. To show that the two coverings of ${\mathrm{Im}}(g)$ are polyhedral subdivisions, let us start by showing that the interiors of any two large cells are disjoint, and the interiors of any two small cells are also disjoint.
Our strategy is to show that the linear spans of any two large cells (which are $(2m+2n-4)$-dimensional subspaces) are different; therefore the intersection of the two cells cannot be $(2m+2n-4)$-dimensional, and their interiors must be disjoint. We will use the first Corollary to Proposition \[im(R)2\].
Suppose, then, that we know which rectangle equalities an unknown large cell $\overline{g(R)}$ satisfies. We can recover $R$ as follows. Define an equivalence relation $\sim$ on $[m] \times [n]$ by declaring that $(i,j) \sim (i',j')$ if $g_{ij}$ and $g_{i'j'}$ are part of the same rectangle equality, and then taking the transitive closure.
This equivalence relation will have two non-trivial equivalence classes, and several singletons. (The only exceptions are the regions with $L$-shaped diagrams, like $i{\overline{1}}{\overline{2}}\ldots
\widehat{{\overline{j}}}\ldots{\overline{n}}12\ldots \widehat{i} \ldots m {\overline{j}}$. Here there is only one non-trivial equivalence class, and we can immediately recover the diagram, and hence the region, from it.)
In one equivalence class, find an entry $(i,j)$ which appears in a rectangle relation with every other entry in the equivalence class. Then $g_{ij}$ must be, essentially, the southwest or northeast corner of the diagram of $R$ which we are trying to recover. More precisely, $i$ and ${\overline{j}}$ are either in the last positive and first negative block of $\pi(R)$, or in the last positive and first negative block, respectively. Because $\overline{g(R)} = \overline{g(-R)}$, we can assume it is the former.
Let $\pi(R) = (i_0)\, {\overline{j}}\, (i_1\ldots i_r)\, \sigma \,
({\overline{j}}_s\ldots{\overline{j}}_1)\, i\, ({\overline{j}}_0)$ be the (unknown) label of $R$. Here $\sigma$ denotes the segment of $\pi(R)$ which starts at the second negative letter and ends at the second-to-last positive letter. Symbols in parenthesis denote letters which may or may not be in $\pi(R)$.
From Version 1 of Proposition \[im(R)1\], $\Delta_{ii'jj'}=0$ holds in $\overline{g(R)}$ if and only if ${\overline{j}},{\overline{j}}' < i,i'$. For $i', {\overline{j}}'$ in $\sigma$, this holds if and only if ${\overline{j}}'<i'$. This allows us to recover which letters are in $\sigma$, and in which order.
Any positive letters which do not appear in $\sigma$ must appear to the left of it. Their position in $\pi(R)$ is determined by the fact that $\overline{g(R)}$ is large: If there is only one such letter, it must be $i_0$, and $i_1, \ldots, i_r$ do not exist. If there are several such letters, they must be $i_1, \ldots, i_r$, and $i_0$ does not exist. Similarly, we recover the positions of the negative letters which do not appear in $\sigma$. We have recovered the label of $R$, as claimed.
It follows that the interiors of the large cells are disjoint. Since each small cell is in a unique large cell, and each large cell is subdivided into small cells with disjoint interiors, if follows that the interiors of the small cells are disjoint also.
4\. Let us now show that the covering of ${\mathrm{Im}}(g)$ by the small cells is actually a polyhedral subdivision. Consider two small cells $A_1 = \overline{g(R_1)}$ and $A_2 =
\overline{g(R_2)}$. We wish to show that their intersection is a face of $A_1$.
A first description of $A_1 \cap A_2$ is given by the rectangle relations of $A_1$ and $A_2$. We need to find a second description, which only uses defining equalities and inequalities of $A_1$ (corresponding to monochromatic and jagged rectangles in $R_1$), and equalities which define facets of $A_1$ (corresponding to jagged rectangles in $R_1$).
![The position of $\Box_{ii'jj'}$ in the diagrams of two small cells.[]{data-label="twosmall"}](twosmall){width="5in"}
Consider a rectangle relation $\Delta_{ii'jj'}=0$ which holds in $A_1 \cap A_2$. If $\Box_{ii'jj'}$ is monochromatic in $R_1$, the given relation is satisfied in $A_1$. If it is jagged, then the relation defines a facet of $A_1$. Therefore, we can assume that it is sliced. Assume that it is sliced horizontally, so $g_{ij}$ and $g_{ij'}$ are white, and $g_{i'j}$ and $g_{i'j'}$ are black. The sign of $\Delta_{ii'jj'}$ is undetermined in $A_1$, and therefore the equality $\Delta_{ii'jj'}=0$ must hold in $A_2$. The rectangle $\Box_{ii'jj'}$ must then be monochromatic in $R_2$; assume it is black. Also assume that $i \leq i'$ and $j \leq j'$ in $\pi(R_2)$; our arguments extend immediately to the other cases. Figure \[twosmall\] shows the position of $\Box_{ii'jj'}$ in the diagrams of $R_1$ and $R_2$.
Since $A_1$ is small, ${\overline{j}}$ and ${\overline{j}}'$ are not in the first negative block of $\pi(R_1)$. Let ${\overline{j}}_1$ be the first negative letter in $\pi(R_1)$. Then $\Box_{ii'j_1 j}$ is jagged in $R_1$, and $\Delta_{ii'j_1 j} \geq 0$ in $A_1$. If ${\overline{j}}_1 \leq {\overline{j}}$ in $\pi(R_2)$ then $\Box_{ii'j_1 j}$ is monochromatic in $R_2$ and $\Delta_{ii'j_1 j} = 0$ in $A_2$; if ${\overline{j}}_1 > {\overline{j}}$ in $\pi(R_2)$ then $\Box_{ii'j j_1}$ is either monochromatic or jagged in $R_2$ and $\Delta_{ii'j j_1} \geq 0$ in $A_2$. In any case, $\Delta_{ii'j_1j} = 0$ in $A_1 \cap A_2$, and this is a facet equality of $A_1$. Similarly, $\Delta_{ii'j_1j'} = 0$ in $A_1 \cap
A_2$, and this is a facet equality of $A_1$. Therefore the equality $\Delta_{ii'jj'}=0$ is a consequence of two facet equalities of $A_1$.
Exactly the same argument shows that any rectangle inequality satisfied by $A_1 \cap A_2$ is a consequence of the relations of $A_1$ and its facet equalities. It follows that the small cells are actually a polyhedral subdivision of ${\mathrm{Im}}(g)$.
5\. We now use a similar argument to show that the covering of ${\mathrm{Im}}(g)$ into large cells is also a polyhedral subdivision.
Consider two large cells $A_1 = \overline{g(R_1)}$ and $A_2 =
\overline{g(R_2)}$, and a rectangle relation $\Delta_{ii'jj'}=0$ which holds in $A_1 \cap A_2$. As before, assume that $\Box_{ii'jj'}$ is sliced horizontally in $R_1$ and monochromatic black in $R_2$, and that $i \leq i'$ and ${\overline{j}}\leq {\overline{j}}'$ in $\pi(R_2)$.
The argument that we used for small cells carries over to this situation, unless ${\overline{j}}$ and ${\overline{j}}'$ are in the first block of negative letters of $\pi(R_1)$. Since $R_1$ is large, $i$ must then be the unique first positive letter of $\pi(R_1)$. Figure \[twolarge\] shows the position of $\Box_{ii'jj'}$ in the diagrams of $R_1$ and $R_2$ in this case.
![The position of $\Box_{ii'jj'}$ in the diagrams of two large cells.[]{data-label="twolarge"}](twolarge){width="5in"}
Since $A_1$ is large, ${\overline{j}}$ and ${\overline{j}}'$ are not in the last negative block of $\pi(R_1)$. Let ${\overline{j}}_n$ be the last negative letter in $\pi(R_1)$. The rectangles $\Box_{ii'jj_n}$ and $\Box_{ii'j'j_n}$ are jagged in $R_1$. They are either both monochromatic or both jagged in $R_2$. If they are monochromatic, then $\Delta_{ii'jj_n}=0$ and $\Delta_{ii'j'j_n}=0$ in $R_2$, and hence in $R_1 \cap R_2$. These two equalities, which define facets of $A_1$, will imply the desired equality $\Delta_{ii'jj'}=0$. Therefore, assume that the two rectangles are jagged in $R_2$, with $g_{ij_n}$ white.
Now, since $A_2$ is large, $i$ and $i'$ are not in the first positive block of $\pi(R_2)$. Let $i_1$ be the first positive letter in $\pi(R_2)$. The rectangle $\Box_{ii_1jj_n}$ is jagged in $R_1$, so $\Delta_{ii_1jj_n} \geq 0$ in $A_1$. The rectangle $\Box_{i_1ijj_n}$ is jagged in $R_2$, so $\Delta_{i_1ijj_n} \geq
0$ in $A_2$. Therefore $\Delta_{i_1ijj_n} = 0$ in $A_1 \cap A_2$, and this is a facet equality of $A_1$. Similarly, $\Delta_{i_1ij'j_n} = 0$ in $A_1 \cap A_2$, and this is a facet equality of $A_1$.
It follows that $\Delta_{ii_1jj'} = 0$ in $A_1 \cap A_2$. We also have that $\Box_{i_1i'jj'}$ is monochromatic in $R_1$ and $\Delta_{i_1i'jj'} = 0$ holds in $A_1$. The last two equalities imply that $\Delta_{ii'jj'}=0$ in $A_1 \cap A_2$, and this equality is implied by two of the facet equalities and one of the defining equalities of $A_1$. This completes the proof.
Putting together the considerations in the proofs of Proposition \[im(R)1\] and Theorem \[subdivisions\], it is now easy to describe the fiber of a generic point $g$ in the image. Such a point lies in a unique small cell, in two medium cells, and in a unique large cell. Therefore, it has preimages in exactly four regions of ${{\cal A}}_{m,n}$. The diagrams of these four regions are almost equal; they only differ in the color of the northwesternmost and southeasternmost entry; assume for simplicity that they are $g_{11}$ and $g_{mn}$, respectively.
Notice that we can add a constant to the $a_i$s and subtract it from the $b_j$s, or add a constant to the $A_i$s and subtract it from the $B_j$s, without affecting $g(a,A,b,B)$. Therefore, we can focus our attention on the preimages of $g$ with, say, $A_m$ and $b_n$ fixed.
First consider the preimages in the region where $g_{11}$ and $g_{mn}$ are black. With $A_m$ and $b_n$ fixed, forcing $g(a,A,b,B)$ to have the correct border entries determines almost all of $(a,A,b,B)$. Only $a_m$ and $b_1$ are not determined; $a_m$ can take any value larger than $g_{mn}-b_n$ (the value it would have if $g_{mn}$ was white), and $b_1$ can take any value larger than $g_{11}-a_1 = g_{11}-g_{1n}+b_n$ (the value it would have if $g_{11}$ was white). The preimages form a two-dimensional quadrant parallel to the $a_mb_1$ plane of ${\mathbb{R}}^{2(m+n)}$.
The remaining three cells give similar preimages. Putting them all together, we are left with four two-dimensional quadrants parallel to the $a_mb_1$, $b_1B_n$, $B_nA_1$ and $A_1a_m$ planes of ${\mathbb{R}}^{2(m+n)}$. Their common apex is the point with $a_m=g_{mn}-b_n, b_1=g_{11}-a_1, B_n=g_{mn}-A_m, A_1=g_{11}-B_1$.
We have two extra degrees of freedom, given by the choices of $A_m$ and $B_n$. The preimage of a generic point described is a four-dimensional polyhedral complex; this is consistent with the dimension drop from the $(2m+2n)$-dimensional range to the $(2m+2n-4)$-dimensional image.
We now have all the tools to enumerate the cells in the two subdivisions of $B_{m,n}$ that we have constructed.
Let ${\mathrm{small\,}}_{m,n}$ be the number of small cells in $B_{m,n}$. For $m,n \geq 2$, $${\mathrm{small\,}}_{m,n} = 2{m \choose 2}{n \choose 2}r_{m-2,n-2},$$ where $r_{m-2,n-2}$ is the number of regions of the real arrangement ${{\cal A}}_{m-2,n-2}$.
If we remove the top and bottom rows and leftmost and rightmost columns from the diagram of a small cell of $B_{m,n}$, we obtain the diagram of a region of an arrangement combinatorially isomorphic to ${{\cal A}}_{m-2,n-2}$.
To recover the diagram of the cell from the diagram of the region, we need to choose the labels of the top and bottom rows (for which there are $m(m-1)$ options to choose from), and the leftmost and rightmost columns (for which there are $n(n-1)$ options) which we deleted. We also need to extend the path that separates the black and white cells; there is only one way of doing this that gives the diagram of a small cell. Finally, remember that each cell has two diagrams that represent it, which differ by a $180^{\circ}$ rotation and a color switch. The desired result follows.
Let ${\mathrm{large\,}}_{m,n}$ be the number of large cells in $B_{m,n}$. Then $$\sum_{m,n \geq 0} {\mathrm{large\,}}_{m,n}\frac{x^m}{m!}\frac{y^n}{n!} =
\frac12(xX+yY)+ \frac{2XY + X^2(e^x-1) +
Y^2(e^y-1)}{2(e^x+e^y-e^{x+y})},$$ where $\,X=x(e^y-y-1)$ and $\,Y=y(e^x-x-1)$.
Imitate the proof of Proposition \[faces\].
Acknowledgments
===============
I would like to thank Lior Pachter and Bernd Sturmfels for suggesting this project, and Mike Develin for helpful conversations.
[^1]: Mathematical Sciences Research Institute –
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Let $f$ be a smooth Morse function on an infinite dimensional separable Hilbert manifold, all of whose critical points have infinite Morse index and co-index. For any critical point $x$ choose an integer $a(x)$ arbitrarily. Then there exists a Riemannian structure on $M$ such that the corresponding gradient flow of $f$ has the following property: for any pair of critical points $x,y$, the unstable manifold of $x$ and the stable manifold of $y$ have a transverse intersection of dimension $a(x)-a(y)$.'
author:
- Alberto Abbondandolo and Pietro Majer
date: 'March 31, 2004'
title: When the Morse index is infinite
---
Introduction {#introduction .unnumbered}
============
Let $f$ be a smooth Morse function on an infinite dimensional separable Hilbert manifold[^1] $M$. Let us denote by ${\mathrm{crit}}(f)$ the set of its critical points, and let us assume that each $x\in {\mathrm{crit}}(f)$ has finite Morse index $i(x)$. A Riemannian structure $g$ on $M$ determines the vector field $-{\mathrm{grad\,}}f$, whose local flow $\phi_t: M
\rightarrow M$ has the critical points of $f$ as rest points. The Morse condition is translated into the fact that these rest points are hyperbolic. Their unstable and stable manifolds, $$\begin{aligned}
W^u(x;f,g) = W^u(x) = {\left\{{p\in M}\mid{\lim_{t\rightarrow -\infty}
\phi_t(p) = x}\right\}}, \\ W^s(x;f,g) = W^s(x) = {\left\{{p\in
M}\mid{\lim_{t\rightarrow +\infty} \phi_t(p) = x}\right\}},\end{aligned}$$ are submanifolds with $\dim W^u(x) = {\mathrm{codim}}\, W^s(x) =
i(x)$. Moreover for a generic choice of the Riemannian structure $g$, $W^u(x)$ and $W^s(y)$ intersect transversally for every pair of critical points $x,y$. It follows that $W^u(x)\cap W^s(y)$ - if non-empty - is a submanifold of dimension $$\label{launo}
\dim W^u(x) \cap W^s(y) = i(x) - i(y).$$
If the critical points have infinite Morse index and co-index, their unstable and stable manifolds are infinite dimensional and so could be their intersections. However, there are situations in which these intersections are indeed finite dimensional, and one can associate an integer $i_{\mathrm{rel}} (x)$ to each critical point $x$ in such a way that $$\label{ladue}
\dim W^u(x) \cap W^s(y) = i_{\mathrm{rel}}(x) - i_{\mathrm{rel}}(y).$$ This fact allows to develop the analogue of Morse theory for such functions: it is the case of Floer homology[^2] for the Hamiltonian action functional on the space of loops on certain symplectic manifolds [@flo88a; @flo88d; @flo89a], of functionals on Hilbert spaces somehow compatible with either a fixed splitting into two closed linear subspaces [@ama01], or with a closed linear subspace and a fixed flag of finite dimensional linear subspaces [@szu92; @ks97; @gip99], and more generally of functionals on Hilbert manifolds which are compatible with a fixed subbundle of the tangent bundle [@ama03c] or with a polarization [@cjs95].
The aim of this note is to prove the following result:
[Theorem.]{} [*Let $f$ be a smooth Morse function on the Hilbert manifold $M$, all of whose critical points have infinite Morse index and co-index. Let*]{} $$a : {\mathrm{crit}}(f) \rightarrow {\mathbb{Z}}$$ [*be an [*arbitrary*]{} function. Then there exists a Riemannian structure $g$ on $M$ such that for every $x,y\in {\mathrm{crit}}(f)$ the intersection $W^u(x;f,g)\cap W^s(y;f,g)$ is transverse and - if it is non-empty - it is a submanifold of dimension $a(x) - a(y)$.*]{}
[ *Moreover, if ${\left\{{(x_i,y_i)}\mid{i=1,\dots,n}\right\}}$ is a finite set of pairs of critical points such that $a(x_i)>a(y_i)$ and there is a smooth curve $u_i:[0,1]\rightarrow M$, $u_i(0)=x_i$, $u_i(1)=y_i$, $i=1,\dots, n$, such that $Df(u_i)[u_i^{\prime}]<0$ on $]0,1[$, the Riemannian structure $g$ can be chosen in such a way that $W^u(x_i;f,g)\cap
W^s(y_i;f,g)\neq \emptyset$ for every $i=1,\dots,n$.*]{}
[*Finally, if $g_0$ is any Riemannian structure on $M$, the Riemannian structure $g$ can be chosen to be equivalent to $g_0$, meaning that there is $c>0$ for which*]{} $$\frac{1}{c} \, g_0(\xi,\xi) \leq g(\xi,\xi) \leq c \, g_0(\xi,\xi) \quad
\forall \xi \in TM.$$
Therefore the situation is completely different from the case of finite Morse indices, where the identity (\[launo\]) holds for every choice of the Riemannian structure and Morse theory only depends on the pair $(M,f)$. In the case of infinite Morse indices and co-indices no Morse theory based just on the pair $(M,f)$ can possibly exist, and extra structures, such as the ones used in the above mentioned papers, are really needed.
Notice that if the function $a$ is chosen to be constant, the above theorem produces a Riemannian structure $g$ for which the unstable and stable manifolds of two distinct critical points never meet. The statement about the possibility of requiring some intersections to be non-empty says that choosing a non-constant $a$ one can actually produce non-trivial intersections.
Finally, the possibility of choosing $g$ to be equivalent to a given Riemannian structure says that from the point of view of completeness and compactness the Riemannian structure $g$ is not worse than a preferred Riemannian structure $g_0$ we might dispose of: if $(M,g_0)$ is complete so is $(M,g)$, if $f$ satisfies the Palais-Smale condition with respect to $g_0$ it will also satisfy it with respect to $g$.
The idea of the proof is the following. After choosing a suitable Riemannian structure in a neighborhood of the critical points, we can find an integrable subbundle $\mathcal{V}$ of $TM$ having infinite dimension and codimension and such that for every critical point $x$ the negative eigenspace of the Hessian of $f$ at $x$ is a [*compact perturbation*]{} of $\mathcal{V}(x)$, of [*relative dimension*]{} $a(x)$. The existence of $\mathcal{V}$ follows from a well known result of Eells and Elworthy [@ee70] stating in particular that every separable infinite dimensional Hilbert manifold can be smoothly embedded as an open subset of a Hilbert space (hence its tangent bundle has many integrable subbundles) and from Kuiper theorem [@kui65] stating that the general linear group of an infinite dimensional Hilbert space is contractible. The fact that $\mathcal{V}$ is integrable allows us to construct a vector field $X$ on $M$ having $f$ as a Lyapunov function, and whose local flow [*essentially preserves*]{} $\mathcal{V}$. As it is shown in [@ama03c], these facts imply that the unstable and stable manifolds of two rest points $x,y$ of the vector field $X$ have a [*Fredholm intersection*]{} of index $a(x)-a(y)$. The conclusion follows from the fact that $X$ is the negative gradient of $f$ with respect to a suitable Riemannian structure on $M$, which can be perturbed in order to have transverse intersections.
Flows on a Hilbert space which essentially preserve a closed linear subspace
============================================================================
Let $H$ be the infinite dimensional separable real Hilbert space, with inner product $\langle \cdot, \cdot \rangle$ and norm $|\cdot|$. The operator norm will be denoted by $\|\cdot\|$. The set of all closed linear subspaces of $H$ will be denoted by $Gr(H)$, and $Gr_{\infty,\infty}(H)$ will denote the subset of all those subspaces having infinite dimension and codimension.
We start by recalling some basic definitions and some results from [@ama03c]. A pair $(V,W)$ of closed linear subspaces of $H$ is said a [*Fredholm pair*]{} if $\dim V\cap W<\infty$ and ${\mathrm{codim}}(V+W)<\infty$, in which case the number ${\mathrm{ind\,}}(V,W) = \dim
V\cap W - {\mathrm{codim}}(V+W)$ is said the [*Fredholm index*]{} of $(V,W)$. Two submanifolds of $H$ have a [*Fredholm intersection*]{} if at every point of their intersection the pair consisting of their tangent spaces is Fredholm.
Given $V,W$ two closed linear subspaces of the Hilbert space $H$, $V$ is said to be a [*compact perturbation*]{} of $W$ if the difference of the orthogonal projections onto $V$ and onto $W$ is a compact operator, or equivalently if the operators $$\label{ecom}
P_V P_{W^{\perp}} \quad \mbox{and} \quad P_{W} P_{V^{\perp}}$$ are compact. In this case, $(V,W^{\perp})$ is a Fredholm pair and its index is called the [*relative dimension*]{} of $V$ with respect to $W$, and it is denoted by $\dim (V,W)$.
Let $M$ be an open subset of $H$, and let $X$ be a smooth vector field on $M$, having only hyperbolic rest points (this means that the differential of $X$ at every rest point does not have purely imaginary spectrum). Let $W$ be a closed linear subspace of $H$, with orthogonal projector $P=P_W$, and consider the following compatibility conditions between $X$ and $W$:
(C1)
: For every rest point $x$ of $X$, the positive eigenspace of the differential of $X$ at $x$ is a compact perturbation of $W$.
(C2)
: For every $u\in H$, the operator $[DX(u),P]P = (I-P) DX(u)
P$ is compact.
By (C1), the relative dimension of the positive eigenspace of $DX(x)$ with respect to $W$ is a well defined integer, which is said the [*relative Morse index*]{} of $x$, denoted by $i(x,W)$. If $\phi_t$ denotes the local flow determined by the vector field $X$, condition (C2) is equivalent to the fact that $\phi_t$ [*essentially preserves*]{} $W$, in the sense that $D \phi_t(u) W$ is a compact perturbation of $W$, for every $(t,u)$ in the domain of the local flow.
\[lemuno\] The set of smooth vector fields which satisfy (C2) with respect to $W$ is a module over $C^{\infty}(M,{\mathbb{R}})$.
Indeed, the last term in the identity $$[D(\varphi X)(u),P]P = \varphi(u) [DX(u),P]P + D\varphi(u)[P\cdot]
(I-P)X(u)$$ has rank one. The importance of conditions (C1) and (C2) lies in the following fact:
\[pop1\] Assume that the vector field $X$ has only hyperbolic rest points and satisfies (C1), (C2). Then for every pair of rest points $x,y$, the immersed submanifolds $W^u(x)$ and $W^s(y)$ have Fredholm intersection of index $i(x,W) - i(y,W)$.
When the Fredholm index of these intersections is positive, the vector field can be locally perturbed so as to make these intersections transversal at some point. For instance, we have the following:
\[fant\] Let $x,y$ be hyperbolic rest points of a smooth vector field $X:M\rightarrow H$ which satisfies (C1) and (C2). Let $u_0\in W^u(x;X)
\cap W^s(y;X)$ and let $u$ be the integral line of $X$ such that $u(0)=u_0$. If $i(x,W) > i(y,W)$, for every $\epsilon>0$ and every $k\in {\mathbb{N}}$ there exists a smooth vector field $Y:M\rightarrow H$ such that:
1. $Y$ satisfies (C1) and (C2);
2. $Y=X$ on $M\setminus B_{\epsilon}(u_0)$ and on $u({\mathbb{R}})$;
3. $\|Y-X\|_{C^k} < \epsilon$;
4. $W^u(x;Y)$ and $W^s(y;Y)$ meet transversally at $u_0$.
If the vector field $X$ is the negative gradient of a Morse function $f$ with respect to some Riemannian structure $g_0$ on $M$, the rest points of $X$ are exactly the critical points of $f$, and the unstable and the stable manifolds of these points are embedded submanifolds. In this case, we can perturb the Riemannian structure $g_0$ in such a way that the new negative gradient flow of $f$ has the unstable and stable manifolds of pairs of critical points intersecting transversally:
\[pop2\] Assume that $f$ is a smooth Morse function on the open subset $M\subset H$ endowed with the Riemannian structure $g_0$, such that the vector field $-\mathrm{grad}_{g_0}\, f$ satisfies (C1) and (C2) with respect to a closed linear subspace $W$. Let $k\in {\mathbb{N}}$. Then there exists a $C^k$-dense set of smooth Riemannian structures $g$, equivalent to $g_0$, such that $-\mathrm{grad}_g\, f$ satisfies (C1) and (C2) with respect to $W$ and the intersections $W^u(x;f,g)\cap W^s(y;f,g)$ are transverse, for every pair of critical points $x,y$.
Preliminary lemmata
===================
We recall that the support of a diffeomorphism $\phi$ is the closure of the set where $\phi \neq \mathrm{id}$.
\[luno\] Let $A_0\in GL(H)$, $r_1>0$. Then there exists a number $r_0\in
]0,r_1[$ and a smooth diffeomorphism $\phi: H \rightarrow H$ with support in the ball of radius $r_1$, such that $\phi(u)=A_0 u$ for $|u|\leq r_0$.
[[*Proof.*]{}]{}By Kuiper theorem [@kui65], the general linear group $GL(H)$ is connected, so we can find a curve $A\in C^{\infty}({\mathbb{R}},GL(H))$ such that $A(t)=A_0$ for $t\leq
r_1/2$ and $A(t)=I$ for $t\geq r_1$. Since the continuous function $t\mapsto \|A(t)^{-1} A^{\prime}(t)\|$ has support in $]0,r_1]$, it is easy to find a number $r_0\in ]0,r_1[$ and a smooth function $\mu:]0,+\infty[ \rightarrow ]0,+\infty[$ such that $\mu(t) =
t$ for $t\leq r_0$ and $t\geq r_1$, and $$\label{disu}
\mu^{\prime}(t) > \|A(t)^{-1} A^{\prime}(t)\| \mu(t) \quad \forall t>0.$$ The map $\phi:H \rightarrow H$ defined by $$\phi(x) = \begin{cases} \frac{1}{|x|} \mu(|x|) A(|x|) x, & \mbox{if }
x\neq 0, \\ 0, & \mbox{if } x=0, \end{cases}$$ is clearly smooth, coincides with $x \mapsto A_0 x$ for $|x|\leq r_0$ and with the identity for $|x|\geq r_1$.
We claim that $\phi$ is bijective. Indeed, if $v\in H$, $v\neq 0$, the equation $\phi(x)=v$ is equivalent to $$x=tu, \quad \mu(t) A(t) u = v,$$ where $t>0$ and $|u|=1$. By (\[disu\]), $$\begin{aligned}
\frac{d}{dt} \left( \mu(t)^{-2} |A(t)^{-1} v|^2 \right) = -2
\mu(t)^{-2} \left( \frac{\mu^{\prime}(t)}{\mu(t)} |A(t)^{-1} v|^2 +
\langle A(t)^{-1} v,A(t)^{-1} A^{\prime}(t) A(t)^{-1} v \rangle
\right) \\ \leq -2\mu(t)^{-2} \left( \frac{\mu^{\prime}(t)}{\mu(t)} -
\|A(t)^{-1} A^{\prime}(t)\| \right) |A(t)^{-1} v|^2 < 0,\end{aligned}$$ so the function $t\mapsto \mu(t)^{-1} |A(t)^{-1}v|$ is strictly decreasing. Moreover this function maps $]0,+\infty[$ onto $]0,+\infty[$. Therefore there exists a unique $t>0$ such that $\mu(t)^{-1} |A(t)^{-1} v|=1$, and $x=tu=t\mu(t)^{-1} A(t)^{-1} v$ is the unique solution of $\phi(x)=v$.
We claim that $\phi$ is a diffeomorphism. By the inverse mapping theorem it is enough to show that $D\phi(x)$ is invertible for every $x\in H$. This is certainly true if $x=0$, and if $x=tu$ with $t>0$ and $|u|=1$ we have $$D\phi(x) [v] = \frac{\mu(t)}{t} A(t) v \quad \forall v\in \langle u
\rangle^{\perp}.$$ Therefore $D\phi(x)$ is invertible if and only if the vector $\partial_t \phi(tu)|_{t=|x|}$ does not belong to the hyperplane $$D\phi(x) \langle u\rangle^{\perp} = A(t) \langle u \rangle^{\perp} =
\langle {A(t)^*}^{-1} u \rangle^{\perp}.$$ By (\[disu\]) the quantity $$\langle\partial_t \phi(tu) , {A(t)^*}^{-1} u\rangle = \langle
(\mu^{\prime} A + \mu A^{\prime}) u, {A^*}^{-1}u\rangle =
\langle (\mu^{\prime}+ \mu A^{-1}A^{\prime})u,u\rangle \geq
\mu^{\prime} - \mu \|A^{-1} A^{\prime}\|$$ is strictly positive for every $t>0$, hence $\partial_t
\phi(tu)|_{t=|x|}$ does not belong to the hyperplane $D\phi(x) \langle
u\rangle^{\perp}$. [$\Box$ ]{}
The following result is a simple addendum to the Morse Lemma:
\[lue\] Let $f:H\rightarrow {\mathbb{R}}$ be a smooth function such that $f(0)=0$, $Df(0)=0$, and $D^2 f(0) = A$ is invertible. For every $r_1>0$ there exists a smooth diffeomorphism $\phi:H \rightarrow H$ with support in the ball of radius $r_1$, such that $\phi(0)=0$, $D\phi(0)=I$, and $$\label{ml}
f(u) = \frac{1}{2} \langle A \phi(u), \phi(u) \rangle,$$ for $|u|\leq r_0$.
Indeed, Palais’ proof of the Morse Lemma [@pal63] produces a local diffeomorphism at $0$ $\phi$ verifying $\phi(0)=0$, $D\phi(0) = I$, and (\[ml\]). Let $\chi:{\mathbb{R}}\rightarrow $ be a smooth function with compact support and such that $\chi=1$ in a neighborhood of 0. Therefore for $\epsilon>0$ small, $$\phi_{\epsilon}(u) = u + \chi\left( \frac{|u|}{\epsilon} \right)
(\phi(u) - u)$$ defines a smooth map on $H$, which coincides with $\phi$ in a neighborhood of 0, hence satisfying (\[ml\]) therein and $D\phi_{\epsilon}(0)=I$, and which coincides with the identity for $|u|\geq r_1$. Since $$\lim_{\epsilon \rightarrow 0} \|D\phi_{\epsilon} - I\|_{\infty} = 0,$$ $\phi_{\epsilon}$ is a global diffeomorphism for $\epsilon$ small.
The Banach manifolds embedding theorem of Eells and Elworthy [@ee70] implies that every infinite dimensional separable Hilbert manifold has a smooth open embedding into the Hilbert space $H$. Here we can ask that this embedding satisfies also some more conditions:
\[redux\] Let $(M,g_0)$ be a Riemannian Hilbert manifold, let $f:M\rightarrow
{\mathbb{R}}$ be a smooth Morse function whose critical points have infinite Morse index and co-index. Let $V:{\mathrm{crit}}(f) \rightarrow Gr_{\infty,\infty}(H)$ be an arbitrary function. Then there exists a smooth open embedding $\psi:M\rightarrow H$ such that for every $x\in {\mathrm{crit}}(f)$:
1. $D\psi(x):T_x M \rightarrow H$ is an isometry;
2. there exists $r(x)>0$ and a self-adjoint operator $A(x)\in GL(H)$ whose negative eigenspace is $V(x)$, and such that $$f(\psi^{-1}(\psi(x) + u)) = f(x) + \frac{1}{2} \langle A(x) u,u
\rangle,$$ for every $u\in B_{r(x)}(0)$.
[[*Proof.*]{}]{}Let $\psi_0:M \rightarrow H$ be an open embedding, as given by Eells and Elworthy’s theorem [@ee70]. For every $x\in {\mathrm{crit}}(f)$ let $T(x)\in GL(H)$ be such that $T(x) D\psi_0(x)$ is an isometry from $(T_x M,g_0)$ to $H$, and such that $T(x) D\psi_0(x)$ maps the negative eigenspace of the $g_0$-Hessian of $f$ at $x$ onto $V(x)$. The possibility of finding such a $T(x)$ follows from the fact that the unitary group of $H$ acts transitively on the component $Gr_{\infty,\infty}(H)$ of the Grassmannian of $H$. By Lemma \[luno\], there exists a smooth diffeomorphism $\phi_1: H
\rightarrow H$ supported in a neighborhood of $\psi_0({\mathrm{crit}}(f))$ such that $\phi_1(\psi_0(x)+u) = \psi_0(x) + T(x)u$ for every $x\in
{\mathrm{crit}}(f)$ and $|u|$ small. Then $f_1 = f\circ \psi_0^{-1} \circ
\phi_1^{-1}$ is a smooth Morse function on an open subset of $H$, and the negative eigenspace of $A(x) = D^2 f_1(\phi_1(\psi_0(x)))$ is $V(x)$, for every $x\in
{\mathrm{crit}}(f)$. By Lemma \[lue\], we can find a smooth diffeomorphism $\phi_2 : H \rightarrow H$ supported in a neighborhood of ${\mathrm{crit}}(f_1)$ such that for every $x\in {\mathrm{crit}}(f_1)$ there holds $\phi_2(x)=x$, $D\phi_2(x)=I$, and $$f_1(u) = f_1(x) + \frac{1}{2} \langle A(x) \phi_2(u), \phi_2(u)\rangle,$$ for every $u$ close to $x$. The smooth open embedding $\phi_2^{-1} \circ \phi_1 \circ \psi_0$ satisfies all the requirements. [$\Box$ ]{}
Proof of the theorem
====================
Fix some $W\in Gr_{\infty,\infty}(H)$. By Lemma \[redux\] we may assume that $M$ is an open subset of $H$, that the Riemannian structure $g_0$ coincides with $\langle \cdot,
\cdot \rangle$ at every critical point of $f$, and that every $x\in
{\mathrm{crit}}(f)$ has a neighborhood $U_x\subset M$ such that $$\label{mf}
f(u) = f(x) + \frac{1}{2} \langle A(x) (u-x),(u-x) \rangle \quad
\forall u\in U_x,$$ where the negative eigenspace $V^-(A(x))$ of the invertible self-adjoint operator $A(x)=D^2f(x)$ is a compact perturbation of $W$ with $\dim (V^-(A(x)),W) = a(x)$. Up to replacing $g_0$ by a uniformly equivalent metric and up to replacing the neighborhoods $U_x$ by smaller ones, we may assume that $g_0$ coincides with $\langle \cdot,\cdot \rangle$ on the whole $U_x$, for every $x\in {\mathrm{crit}}(f)$.
We denote by $|\cdot|_{g_0(u)}$ the norm associated to the inner product given by $g_0$ at $u$, $|v|_{g_0(u)}^2 = g_0(u)[v,v]$, and by $\|\cdot \|_{g_0(u)}$ the corresponding operator norm on the space of bounded linear operators on $H$, $$\|L\|_{g_0(u)} = \sup_{\substack{v\in H \\ |v|_{g_0(u)}=1}}
|Lv|_{g_0(u)}.$$
We will define a vector field $X$ on $M$, having $f$ as a Lyapunov function, and such that the unstable and stable manifolds of its rest points have the required properties. Then we will find a Riemannian structure $g_1$ for which $X=-\mathrm{grad}_{g_1} f$. Finally we will perturb $g_1$ in order to have a Morse-Smale flow.
The first step of the proof consists in modifying the curves $u_i$ in order to make them integral lines in the intersections $W^u(x_i)\cap W^s(y_i)$, with respect to the vector field $X$ we are defining.
\[step1\] There exist $\theta_0\in [0,1[$, $T>0$ and for every $i=1,\dots,n$ there are curves $v_i\in C^{\infty}({\mathbb{R}},M)\cap C^0(\overline{{\mathbb{R}}},M)$ such that $v_i({\mathbb{R}}) \cap v_j({\mathbb{R}}) = \emptyset$ if $i\neq j$, $$\begin{aligned}
v_i(-\infty)=x_i, \quad v_i(+\infty)=y_i, \quad v_i^{\prime}(t) = -
\mathrm{grad}_{g_0} f(v_i(t)) \mbox{ for } |t|\geq T, \\
v_i([-\infty,-T]) \subset U_{x_i}, \quad v_i([T,+\infty]) \subset U_{y_i},\end{aligned}$$ and $$\label{edun}
|v_i^{\prime}(t) + \mathrm{grad}_{g_0} f(v_i(t))|_{g_0(v_i(t))} \leq
\theta_0 |\mathrm{grad}_{g_0} f(v_i(t))|_{g_0(v_i(t))} \quad \forall
t\in {\mathbb{R}}.$$
[[*Proof.*]{}]{}By modifying the curve $u_i:[0,1]\rightarrow M$ in a neighborhood of $0$ and $1$, we can join it with two negative gradient flow orbits in the unstable manifold of $x_i$ and in the stable manifold of $y_i$, constructing a curve $\tilde{v}_i \in
C^{\infty}({\mathbb{R}},M) \cap C^0(\overline{{\mathbb{R}}},M)$ satisfying all the requirements, with the possible exception of (\[edun\]), which will be achieved by a reparameterization.
The construction of the curve $\tilde{v_i}$ is the following, where we omit the index $i$ in order to simplify the notation. Let $r>0$ be such that $B_r(x)\subset U_{x}$ and $B_r(y) \subset
U_{y}$, so that $f$ has the quadratic form (\[mf\]) on $B_r(x)$ and on $B_r(y)$, and $$\mathrm{grad}_{g_0} f(\xi) = A(x) (\xi - x) \quad \forall \xi
\in B_r(x), \quad \mathrm{grad}_{g_0} f(\xi) = A(y) (\xi - y)
\quad \forall \xi \in B_r(y).$$ Let $\epsilon>0$ be such that $u(\epsilon)\in B_r(x)$ and $u(1-\epsilon) \in B_r(y)$. Since the level sets $\Sigma^- = {\left\{{\xi \in B_r(x)}\mid{f(\xi) =
f(u(\epsilon)) }\right\}}$ and $\Sigma^+ = {\left\{{\xi \in B_r(y)}\mid{f(\xi) =
f(u(1-\epsilon)) }\right\}}$ are connected and meet $x+V^-(A(x))$, respectively $y+V^+(A(y))$, we can find a continuous piecewise smooth curve $w:{\mathbb{R}}\rightarrow M$ such that $w(t) = u(t)$ for $\epsilon \leq t \leq 1-\epsilon$, $w(t)\in \Sigma^-$ for $0\leq t
\leq \epsilon$, $w(t)\in \Sigma^+$ for $1-\epsilon \leq t
\leq 1$, $w(0)\in x + V^-(A(x))$, $w(1)\in y + V^+(A(y))$, and $$w(t) = x + e^{-tA(x)} (w(0)-x) \quad \forall t<0, \quad
w(t) = y + e^{-(t-1)A(y)} (w(1)-y) \quad \forall t>1.$$ The sets $\Sigma^-$ and $\Sigma^+$ are leafs of the codimension one foliation given by the level sets of $f$. Therefore, it is easy to modify $w$ in a neighborhood of $[0,\epsilon]$ and $[1-\epsilon,1]$, obtaining a curve $\tilde{v}$ with the required properties. By slightly perturbing these curves, we may assume that the supports of $\tilde{v}_i$ are pairwise disjoint.
Let $\tau$ be a solution of the ODE $$\tau^{\prime} = - \frac{ \langle \mathrm{grad}_{g_0}
f(\tilde{v}(\tau)) , \tilde{v}^{\prime} (\tau)
\rangle_{g_0(\tilde{v}(\tau))}
}{|\tilde{v}^{\prime}(\tau)|_{g_0(\tilde{v}(\tau))}^2} .$$ Then $\tau^{\prime}(t)>0$ for every $t\in {\mathbb{R}}$, and $\tau^{\prime}=1$ for $|t|$ large, so $\tau:{\mathbb{R}}\rightarrow {\mathbb{R}}$ is a diffeomorphism, and it is a constant shift for $t$ in a neighborhood of $-\infty$ and of $+\infty$. Setting $v(t) =
\tilde{v}(\tau(t))$, we have that $$\begin{aligned}
| v^{\prime} + \mathrm{grad}_{g_0} f(v)|_{g_0(v)} = \left| -
\frac{ \langle \mathrm{grad}_{g_0} f(\tilde{v}(\tau)) ,
\tilde{v}^{\prime} (\tau) \rangle_{g_0(\tilde{v}(\tau))}
}{|\tilde{v}^{\prime}(\tau)|_{g_0(\tilde{v}(\tau))}^2}
\, \tilde{v}^{\prime} (\tau) +
\mathrm{grad}_{g_0} f(\tilde{v}(\tau)) \right| \\
= \min_{\lambda\in {\mathbb{R}}} |\lambda \tilde{v}^{\prime} (\tau) +
\mathrm{grad}_{g_0} f(\tilde{v}(\tau))
|_{g_0(\tilde{v}(\tau))} = |\mathrm{grad}_{g_0}
f(v)|_{g_0(v)} \sin \alpha,\end{aligned}$$ where $\alpha=\alpha(t)$ is the angle between the vector $\tilde{v}^{\prime}(\tau(t))$ and $-\mathrm{grad}_{g_0}
(\tilde{v}(\tau(t)))$, with respect to the inner product $g_0(\tilde{v}(\tau(t)))$. By the properties of $\tilde{v}$, the continuous function $\alpha$ takes values in $[0,\pi/2[$ and it vanishes for $|t|$ large. Hence $\alpha$ is bounded away from $\pi/2$, and (\[edun\]) holds with $\theta_0= \max \sin \alpha <1$. [$\Box$ ]{}
The second step consists in defining the vector field $X$ in a neighborhood of the support of the curves $v_i$:
\[step2\] Let $\theta\in ]\theta_0,1[$. There exist an open neighborhood $U_*$ of $\bigcup_{i=1}^n v_i(\overline{{\mathbb{R}}})$ and a smooth vector field $X_*:U_* \rightarrow H$ such that:
1. $X_*$ satisfies (C1) and (C2) with respect to $W$;
2. $X_*(u) = - \mathrm{grad}_{g_0} f(u)$ in a neighborhood of $\{x_1,\dots,x_n,y_1,\dots,y_n\}$;
3. $v_i^{\prime}(t) = X_*(v_i(t))$ for every $t\in {\mathbb{R}}$ and every $i=1,\dots,n$;
4. $|X_*(u) + \mathrm{grad}_{g_0} f(u)|_{g_0(u)} \leq \theta
|\mathrm{grad} f(u)|_{g_0(u)}$ for every $u\in U_*$;
5. $W^u(x_i;X_*)$ and $W^s(y_i;X_*)$ intersect transversally along the flow line $v_i$, for every $i=1,\dots,n$.
[[*Proof.*]{}]{}We start by defining a smooth vector field $\tilde{X}$ on a neighborhood $\tilde{U}$ of $\bigcup_{i=1}^n v_i(\overline{{\mathbb{R}}})$ satisfying (i), (ii), (iii). For every $i=1,\dots, n$, let $U_i:
[-2T,2T] \rightarrow U(H) \cap (I+\mathcal{L}_c(H))$ be a smooth path in the group of unitary operators on $H$ which are compact perturbations of the identity, such that $$U_i(t) \langle v_i^{\prime}(0) \rangle^{\perp} = \langle v_i^{\prime}
(t) \rangle^{\perp} \quad \forall t\in [-2T,2T].$$ For instance, if $P_i(t)$ denotes the orthogonal projection onto the one-dimensional linear subspace $<v_i^{\prime}(t)>$, one can choose $U_i$ to be the solution of the linear Cauchy problem $$\begin{cases}
U_i^{\prime} (t) = [P_i^{\prime}(t), P_i(t)] U_i(t), \\
U_i(0)=I. \end{cases}$$ See [@kat80], section II - §4.5. If $r>0$ is small enough, the map $$\psi_i : ]-2T,2T[ \times (\langle v_i^{\prime} (0) \rangle^{\perp}
\cap B_r(0) ) \rightarrow H, \quad (t,\xi) \mapsto v_i(t) + U_i(t)\xi,$$ is a diffeomorphism onto a tubular neighborhood $V_i$ of $v_i(]-2T,2T[)$. By choosing a smaller $r$, we may also assume that $V_i\cap {\mathrm{crit}}(f)=\emptyset$ and that $V_i\cap V_j=\emptyset$ if $i\neq j$. The smooth vector field ${\psi_i}_*(\partial/\partial
t)$, $${\psi_i}_*\left( \frac{\partial}{\partial t} \right) (u) =
D\psi_i(\psi_i^{-1} (u)) [(1,0)] \quad \forall u\in V_i,$$ has $v_i|_{]-2T,2T[}$ as an integral flow line and satisfies (C2) with respect $W$, as its differential at every point is a compact operator. Let $V_0\subset \bigcup_{i=1}^n U_{x_i} \cup U_{y_i}$ be an open neighborhood $\bigcup_{i=1}^n v_i(\overline{{\mathbb{R}}} \setminus
]-2T,2T[)$ such that $v_i^{-1}(V_0) \cap [-T,T] = \emptyset$ for every $i=1,\dots,n$. Let $\{\chi_i\}_{i=0}^n$ be a smooth partition of unity subordinated to the open covering $\{V_i\}_{i=0}^n$ of $\tilde{U}:= V_0\cup V_1 \cup \dots \cup
V_n$. By Lemma \[lemuno\], the smooth vector field $$\tilde{X} : \tilde{U} \rightarrow H, \quad \tilde{X} = -\chi_0
\mathrm{grad}_{g_0} f + \sum_{i=1}^n \chi_i {\psi_i}_* \left(
\frac{\partial}{\partial t} \right),$$ satisfies (i), (ii), (iii). By Proposition \[fant\], we can modify $\tilde{X}$ in a small neighborhood of the set $\{v_1(0),\dots,v_n(0)\}$, obtaining a smooth vector field $X_*:\tilde{U} \rightarrow H$ which satisfies also (v). By (ii), (iii), and (\[edun\]), $X_*$ satisfies also (iv) in a smaller neighborhood $U_* \subset \tilde{U}$ of $\bigcup_{i=1}^n v_i (\overline{{\mathbb{R}}})$. [$\Box$ ]{}
The third step consists in defining the vector field $X$ in a neighborhood of the critical points other than $x_i$ and $y_i$. Let $x\in {\mathrm{crit}}(f)\setminus \bigcup_{i=1}^n \{x_i,y_i\}$. We define $X_x$ to be the affine vector field $$X_x(u) = - A(x) (u-x).$$ Its unique rest point is $x$, and the positive eigenspace of $DX_x(x)$ is $V^+(DX_x(x)) = V^+(-A(x)) =
V^-(A(x))$, which was chosen to be a compact perturbation of $W$, so $X_x$ satisfies (C1) with respect to $W$, and the relative Morse index of $x$ is $i(x,W) = a(x)$. Moreover by (\[ecom\]), $$[DX_x(u),P]P = -[A(x),P]P = (P-I)A(x)P = (P-I)A(x) P_{V^+(A(x))} P -
(I-P) P_{V^-(A(x))} A(x) P$$ is compact, so $X_x$ satisfies (C2) with respect to $W$. Since $g_0$ coincides with $\langle \cdot,\cdot \rangle$ on $U_x$, by (\[mf\]) we have $$\label{pg1}
X_x(u) = - \mathrm{grad}_{g_0} f(u) \quad \forall u\in U_x.$$
The fourth step consists in defining the vector field $X$ in a neighborhood of all the remaining points. We shall denote by $M_0$ the complement of $\bigcup_{i=1}^n v_i(\overline{{\mathbb{R}}})$ in $M$. Let $x\in M_0\setminus {\mathrm{crit}}(f) $, and consider the constant vector field $$X_x (u) = - \mathrm{grad}_{g_0} f(x),$$ which trivially satisfies (C2) with respect to $W$. Let $U_x\subset
M_0\setminus {\mathrm{crit}}(f)$ be an open neighborhood of $x$ such that $$\label{pg2}
|X_x (u) + \mathrm{grad}_{g_0} f(u)|_{g_0(u)} \leq \theta\,
|\mathrm{grad}_{g_0} f(u)|_{g_0(u)} \quad \forall u\in U_x.$$
Now we can patch the vector fields $X_x$, $x\in M_0$, and $X_*$ by a partition of unity. Indeed, let $\{\varphi_j\}_{j\in J}$ be a smooth partition of unity subordinated to a locally finite refinement of the open covering $\{U_x\}_{x\in M_0\cup \{*\}}$ of $M$: $\varphi_j$ is non-negative, it has support in $U_{\omega(j)}$, $\omega(j)\in M_0 \cup \{*\}$, every point of $M$ has a neighborhood which intersects finitely many supports of $\varphi_j$’s, and $\sum_{j\in J}
\varphi_j =1$. Consider the smooth vector field on $M$: $$X(x) = \sum_{j\in J} \varphi_j(x) X_{\omega(j)} (x).$$ By Lemma \[lemuno\], $X$ satisfies (C2) with respect to $W$. From (\[pg1\]) and from the fact that the condition in Lemma \[step2\] (iv) and (\[pg2\]) are convex conditions, we deduce $$\label{pg3}
|X (u) + \mathrm{grad}_{g_0} f(u)|_{g_0(u)} \leq
\theta\, |\mathrm{grad}_{g_0} f(u)|_{g_0(u)} \quad
\forall u\in M.$$ In particular, the set of rest points of $X$ is ${\mathrm{crit}}(f)$. By Lemma \[step2\] (ii) and (\[pg1\]), $$\label{enc}
X(u) = - \mathrm{grad}_{g_0} f(u) \quad \forall x\in U_x^{\prime}, \;
\forall x\in {\mathrm{crit}}(f),$$ for some neighborhood $U_x^{\prime}$ of the critical point $x$. So $DX(x)=-A(x)$ for every critical point $x$, hence $X$ satisfies (C1) with respect to $W$, and the relative Morse index of $x$ is $$\label{idi}
i(x,W)= \dim (V^-(A(x)),W) = a(x).$$ By Lemma \[step2\] (iii) and (v), for every $i=1,\dots,n$ the unstable manifold $W^u(x_i;X)$ meets the stable manifold $W^s(y_i;X)$ transversally along the flow line $v_i$ (indeed, by construction $v_i(t)\in {\mathrm{supp\,}}\varphi_j$ if and only if $\omega(j)=*$).
We claim that there exists a Riemannian structure $g_1$ on $M$, uniformly equivalent to $g_0$, such that $X=-\mathrm{grad}_{g_1} f$. The construction of $g_1$ makes use of the following lemma, whose straightforward proof is left to the reader:
Let $x\in M$ and $u,v\in H$ be such that $g_0(u,v)_x > 0$. Then the bounded linear operator $L_x(u,v)$ defined by $$L_x(u,v) w = w - \frac{g_0( w,u )_x}{|u|^2_{g_0(x)}} u +
\frac{g_0(w,v)_x}{g_0(u,v)_x}v, \quad w\in H,$$ is $g_0(x)$-self-adjoint, $g_0(x)$-positive, invertible, satisfies $$L_x(u,v)u=v, \quad L_x(u,u)=I,$$ and $$\label{stm} \begin{split}
\|L_x(u,v)\|_{g_0(x)}
\leq 1 + \frac{|v|^2_{g_0(x)}}{g_0(u,v)_x}, \\
\|L_x(u,v)^{-1}\|_{g_0(x)}
\leq \left( 1+ \frac{|u|_{g_0(x)} |v|_{g_0(x)}}{g_0(u,v)_x}
\right)^2 + \frac{|u|^2_{g_0(x)}}{g_0(u,v)_x}. \end{split}$$
Every Riemannian structure $g_1$ on $M$ can be written as $$g_1(u,v)_x = g_0(G(x)u,v)_x, \quad \forall x\in M,$$ for some smooth map $G:M \rightarrow GL(H)$, such that $G(x)$ is $g_0(x)$-symmetric and $g_0(x)$-positive for every $x$. In this case, the $g_1$-gradient of $f$ is related to the $g_0$-gradient of $f$ by $\mathrm{grad}_{g_0} f = G\,
\mathrm{grad}_{g_1} f$. Moreover, $g_1$ is uniformly equivalent to $g_0$ if and only if $$\sup_{x\in M} \|G(x)\|_{g_0(x)} < +\infty, \quad
\sup_{x\in M} \|G(x)^{-1}\|_{g_0(x)} < +\infty.$$ Then we can define a Riemannian structure $g_1$ by setting $$G(x) = \begin{cases} I & \mbox{for } x\in \bigcup_{y\in {\mathrm{crit}}(f)}
U_y^{\prime}, \\ L_x(X(x),-\mathrm{grad}_{g_0} f(x)) & \mbox{for }
x\in M \setminus {\mathrm{crit}}(f). \end{cases}$$ Indeed, the above formula defines a smooth map because the map $(x,u,v)\mapsto L_x(u,v)$ is smooth on $${\left\{{(x,u,v)\in M \times H \times H}\mid{g_0(u,v)_x >0}\right\}},$$ and because of (\[enc\]) and $L_x(u,u)=I$. Since $L_x(u,v)u=v$, we obtain for $x\in M\setminus {\mathrm{crit}}(f)$, $$G(x)X(x) = L_x (X(x),-\mathrm{grad}_{g_0} f(x)) X(x) = -
\mathrm{grad}_{g_0} f(x) = - G(x) \mathrm{grad}_{g_1} f(x),$$ from which $X = -\mathrm{grad}_{g_1} f$ on $M$. Finally, an easy computation shows that (\[pg3\]) and (\[stm\]) imply $$\|G(x)\|_{g_0(x)} \leq 1 + \frac{1}{1-\theta}, \quad
\|G(x)^{-1}\|_{g_0(x)} \leq \frac{4}{(1-\theta)^2} +
\frac{(1+\theta)^2}{1-\theta}, \quad \forall x\in M,$$ so $g_1$ is uniformly equivalent to $g_0$.
By Proposition \[pop2\], we can perturb $g_1$ obtaining a Riemannian structure $g$ on $M$ such that the vector field $-\mathrm{grad}_g f$ still satisfies (C1) and (C2) with respect to $W$, all the intersections between unstable and stable manifolds are transverse, and the transverse intersections at $v_i$ are preserved, for every $i=1,\dots,n$. Then Proposition \[pop1\] and (\[idi\]) imply that $$\dim W^u(x;f,g) \cap W^s(y;f,g) = i(x,W) - i(y,W) = a(x) - a(y),$$ whenever such intersection is non-empty.
[Flo88b]{}
A. Abbondandolo and P. Majer, *Morse homology on [H]{}ilbert spaces*, Comm. Pure Appl. Math. **54** (2001), 689–760.
[to3em]{}, *A [M]{}orse complex for infinite dimensional manifolds - [P]{}art [I]{}*, e-print [arXiv:math.DS/0309020]{}, 2003.
[to3em]{}, *A [M]{}orse complex for infinite dimensional manifolds - [P]{}art [II]{}*, in preparation, 2004.
R. L. Cohen, J. D. S. Jones, and G. B. Segal, *Floer’s infinite-dimensional [M]{}orse theory and homotopy theory*, The Floer memorial volume (H. Hofer, C. H. Taubes, A. Weinstein, and E. Zehnder, eds.), Birkhäuser, Basel, 1995, pp. 297–325.
J. Eells and K. D. Elworthy, *Open embeddings of certain [B]{}anach manifolds*, Ann. of Math. (2) **91** (1970), 465–485.
A. Floer, *A relative [M]{}orse index for the symplectic action*, Comm. Pure Appl. Math. **41** (1988), 393–407.
[to3em]{}, *The unregularized gradient flow of the symplectic action*, Comm. Pure Appl. Math. **41** (1988), 775–813.
[to3em]{}, *Witten’s complex and infinite-dimensional [M]{}orse theory*, J. Differential Geom. **30** (1989), 207–221.
K. Gȩba, M. Izydorek, and A. Pruszko, *The [C]{}onley index in [H]{}ilbert spaces and its applications*, Studia Math. **134** (1999), 217–233.
T. Kato, *Perturbation theory for linear operators*, Springer, Berlin, 1980.
W. Kryszewski and A. Szulkin, *An infinite-dimensional [M]{}orse theory with applications*, Trans. Amer. Math. Soc. **349** (1997), 3181–3234.
N. H. Kuiper, *The homotopy type of the unitary group of [H]{}ilbert space*, Topology **3** (1965), 19–30.
R. S. Palais, *Morse theory on [H]{}ilbert manifolds*, Topology **2** (1963), 299–340.
A. Szulkin, *Cohomology and [M]{}orse theory for strongly indefinite functionals*, Math. Z. **209** (1992), 375–418.
[^1]: By this we mean a paracompact separable space with a smooth atlas of charts taking values in an infinite dimensional separable real Hilbert space.
[^2]: Actually, in Floer homology one does not consider the gradient flow with respect to a Riemannian metric on a Hilbert manifold, but one uses the gradient equation with respect to an inner product which is not complete on the space where the function is smooth. Therefore, one does not obtain a local flow and there are no stable and unstable manifolds. However, the space of solutions connecting two critical points is a finite dimensional manifold.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'It is known that noisy topological quantum codes are related to random bond Ising models where the order-disorder phase transition in the classical model is mapped to the error threshold of the corresponding topological code. On the other hand, there is a dual mapping between classical spin models and quantum Calderbank-Shor-Stean (CSS) states where the partition function of a classical model defined on a hypergraph $H$ is written as an inner product of a product state and a CSS state on dual hypergraph $\tilde{H}$. It is then interesting to see what is the interpretation of the classical phase transition in the random bond Ising model within the framework of the above duality and whether such an interpretation has any connection to the error threshold of the corresponding topological CSS code. In this paper, we consider the above duality relation specifically for a two-dimensional random bond Ising model. We show that the order parameter of this classical system is mapped to a coherence order parameter in a noisy Toric code model. In particular, a quantum phase transition from a coherent phase to a non-coherent phase occurs when the initial coherent state is affected by two sequences of bit-flip quantum channels where a quenched disorder is induced by measurement of the errors after the first channel. On the other hand, the above transition is directly related to error threshold of the Toric code model. Accordingly, and since the noisy process can be applied to other topological CSS states, we conclude that the dual correspondence can also provide a useful tool for the study of error thresholds in different topological CSS codes.'
author:
- Mohammad Hossein Zarei
- Abolfazl Ramezanpour
title: |
Noisy Toric code and random bond Ising model:\
The error threshold in a dual picture
---
Introduction
============
It has been shown that quantum information theory has many fascinating connections to statistical physics which have led to several cross-fertilizations between these two important fields of physics. During past decades different problems including quantum algorithms for evaluation of partition functions or simulation of classical spin models [@Geraci2008; @Lidar1997; @Geraci2010; @algor; @durmar], mathematical mappings from quantum systems to classical statistical models [@Somma2007; @eis17; @castel2005; @gemma] and applications of percolation theory in quantum error correcting codes [@vodola] have been developed. Of the most interesting connections is a mapping from partition function of classical spin models to quantum stabilizer states [@Nest2007]. Such mappings have recently been simplified by a duality relation based on hypergraphs [@zare18] where partition function of a classical spin model defined on a hypergraph $H$ can be written as an inner product of a product state and a CSS state defined on the dual hypergraph $\tilde{H}$. Using the above dual correspondence, concepts from measurement based quantum computation have been used for considering complexity of classical spin models [@Bravyi2007; @Bombin2008]. In particular, concept of completeness in statistical models [@Nest2008; @Vahid2012b; @Cuevas2009; @xu; @Vahid2012; @yahya] and recently universal models [@science; @cub] have emerged in statistical physics which has opened an important window toward a unification in statistical mechanical models.
On the other hand, one of important and well-established problems in statistical mechanics is the existence of thermal phase transition in classical spin models [@pathr]. In this respect and using the dual correspondence between classical spin models and quantum CSS states, one can expect to find an interpretation of phase transition in classical models for quantum CSS states. For example, it has recently been shown that thermal phase transitions in classical spin models are mapped to topological phase transitions in the corresponding CSS models [@zarei19]. Furthermore, the above mapping has led to insights about noisy topological CSS states where thermal phase transition in a ferromagnetic spin model is mapped to a critical stability against bit-flip noise in the corresponding topological CSS code[@zare18]. In particular, it has been shown that phase transition in two-dimensional (2D) Ising model is mapped to a coherence-noncoherence phase transition in the 2D Toric code model under a bit-flip noise [@zareimon].
In addition to the dual correspondence, connection between thermal phase transition in classical spin models and noisy topological codes has also been considered in a different framework [@Dennis2002; @Katzgraber2009; @q1; @q2; @q3; @q4; @3dcolor], where the error threshold of noisy topological codes is mapped to the transition point of the random bond Ising models on the Nishimori line [@nishi] . Accordingly, it seems that such problem should be related to the dual correspondence between classical spin models and quantum CSS states. If it is correct, it means that the same dual correspondence also relates the error threshold of a topological CSS state defined on a hypergraph $H$ with the phase transition point of a random bond Ising system defined on $\tilde{H}$.
In this paper, we start from the dual correspondence for the 2D random bond Ising model where the partition function is mapped to an inner product of the Toric code state and a product state in which random couplings are encoded. Then, we explore a mapping from this problem to a noisy Toric code model to find a relation to the error threshold problem. To this end, we introduce a quantum formalism for the order parameter of the random bond Ising model. Then, using a mapping from temperature to the probability of noise, we find an important interpretation for the order parameter in a noisy Toric code model. Specifically, we introduce a noisy process on a coherent state in the Toric code where two sequences of bit-flip channels are applied and a quenched disorder is induced by measurement of error syndromes after the first channel. We introduce an order parameter for determining the coherence of an initial state after applying noise, and we show that it is mapped to local magnetization in the random bond Ising model. Therefore, we find a coherence-noncoherence phase transition in the noisy Toric code model corresponding to ferromagnetic-paramagnetic phase transition in the random bond Ising model. Specifically, the phase diagram here is richer than the one that has been derived for ordinary Ising model [@zareimon], in the sense that here the two quenched and annealed noise parameters determine the coherence phase of the model. On the other hand, we show that the above transition is also related to the error threshold of the Toric code model. Since this result is derived by using the dual correspondence and it can be applied to arbitrary random bond Ising systems, we conclude that the same dual correspondence connects the error threshold of topological CSS codes and the transition point of the dual random bond Ising systems. We give examples to show that this is very useful for studying the error threshold of different topological states. The reason is that by using hypergraphs we can easily find the classical spin model corresponding to any quantum CSS state.
The paper is organized as follows. In Sec.(\[sec1\]) we briefly review the quantum formalism of the partition function for Ising model. In Sec.(\[sec2\]), we introduce a quantum formalism for the order parameter of the random bond Ising model. In Sec.(\[sec3\]), we give the main result of the paper where we introduce a mapping to Toric code model under a sequence of bit-flip noises. Specifically, we map the order parameter of the random-bond Ising model to a coherence parameter for the above noisy Toric code model. Finally, in Sec.(\[sec4\]) we explain the connection of our results with the error threshold of the Toric code and we conclude that the duality correspondence between classical spin models and quantum CSS states is helpful for investigation of the error threshold of quantum CSS states.
Quantum formalism for partition function of 2D Ising model {#sec1}
==========================================================
In this section, we give a brief review of the quantum formalism for partition function of the two-dimensional (2D) Ising model. To this end, consider a 2D square lattice where classical spins of $s=\pm 1$ live on vertices (vertex spins). The classical Hamiltonian is defined as follows: $$\label{ran}
H=-\sum_{\la i, j\ra} J_{ij}s_i s_j,$$ where $\la i, j\ra$ refers to the nearest neighbors and $J_{ij}$ refers to the coupling constant corresponding to the interaction of $\la i, j\ra$. All thermodynamic properties of the above model can be derived from the partition function which is obtained by a summation of the Boltzmann weights corresponding to all spin configurations $\mathbf{s}$: $$\label{ranz}
\mathcal{Z}[\mathbf{J}]=\sum_{\mathbf{s}}e^{\beta \sum_{\la i, j\ra} J_{ij}s_i s_j},$$ where $\beta$ is equal to $\frac{1}{k_B T}$ and $k_B$ refers to the Boltzmann constant.
![(Color online) a) Ising spins $s_i$ denoted by black circles live on vertices of a square lattice. Edge spins $S_{ij}=s_i s_j$ denoted by green (light) circles are added on links of the lattice. Product of edge spins belonging to a face $F$ of the lattice should be equal to $1$. b) In a quantum formalism, each constraint corresponding to a face of the lattice is mapped to a face operator $B_F =\prod_{e\in \partial F}Z_e$. It is in fact a stabilizer of the Toric code. There are also another kind of stabilizers $A_v =\prod_{e\in v}X_e$ corresponding to each vertex of the lattice.[]{data-label="k4"}](k4){width="8cm" height="4cm"}
Interestingly, the above partition function can be rewritten in a quantum formalism as follows. We insert new spins $S_e$, (edge spins) denoted by green (light) circles, on the edges of the square lattice, as Fig.(\[k4\]-a) shows. The value of each edge spin $S_e$ is defined as product of the vertex spins living on the two endpoints of the edge $e$, i.e., $S_e =s_i s_j$. Next, we replace the vertex spins in Eq.(\[ranz\]) with edge spins $S_e$. However, we need also to consider the constraints between the edge spins. Specifically, as it is shown in Fig.(\[k4\]-a), corresponding to each face $F$ of the lattice there is a constraint between edge spins in the form of $\prod_{e\in \partial F} S_e =1$ where $e\in \partial F$ refers to an edge around the face $F$. We apply such constraints by delta functions in Eq.(\[ranz\]) and rewrite the partition function in the following form: $$\label{cons0}
\mathcal{Z}[\mathbf{J}]=\sum_{\mathbf{S}}e^{\beta \sum_{e} J_{e}S_e }\prod_{F} \delta(\prod_{e \in \partial F} S_e ,1).$$ Here $\prod_F$ refers to all independent constraints corresponding to faces of the lattice and $J_e=J_{ij}$ where $i, j$ are the two endpoints of $e$. Furthermore, using the following form of the delta function $\delta(\prod_{e \in \partial F} S_e ,1)=\frac{1+\prod_{e \in \partial F} S_e}{2}$, we obtain: $$\label{cons}
\mathcal{Z}[\mathbf{J}]=\sum_{\mathbf{S}}e^{\beta \sum_{e} J_{e}S_e }\prod_{F} \frac{1+\prod_{e \in \partial F} S_e}{2}.$$
Now, we are ready to introduce the quantum formalism of the partition function by replacing edge spins $S_e$ in Eq.(\[cons\]) with Pauli operators $Z_e$. To this end, note that each edge spin $S_e$ takes two values $\pm1$ which can be considered as eigenvalues of a Pauli operator $Z_e$. Let $|0\ra$ and $|1\ra$ refer to eigenstates of the $Z$ operator with eigenvalues $+1$ and $-1$, respectively. It is simple to check that $\sum_{S_e} g(S_e) =2 \la +| g(Z_e)|+\ra$ where $g$ is an arbitrary function and $|+\ra =\frac{1}{\sqrt{2}}(|0\ra +|1\ra )$. Therefore, the partition function in Eq.(\[cons\]) reads as follows: $$\label{qf}
\mathcal{Z}[\mathbf{J}]=2^M ~^{\otimes M}\la +|e^{\beta \sum_{e} J_{e}Z_e }\prod_{F} \frac{1+\prod_{e \in \partial F} Z_e}{2}|+\ra ^{\otimes M},$$ where $M$ refers to the number of edges of the lattice. The right hand side of the above equation is in fact an inner product of the following quantum states: $$|\alpha[\mathbf{J}] \ra =e^{\beta \sum_{e} J_{e}Z_e }|+\ra^{\otimes M}=\frac{1}{\sqrt{2}^M}\otimes_{e}(e^{\beta J_e }|0\ra +e^{-\beta J_e }|1\ra),$$
$$\label{inn}
|G\ra =\prod_{F} \frac{1+B_F}{2}|+\ra ^{\otimes M}.$$
Here $B_F =\prod_{e \in \partial F} Z_e$ which is called a face operator, see Fig.(\[k4\]-b). Finally, the partition function is in the form of $\mathcal{Z}[\mathbf{J}]=2^M \la \alpha[\mathbf{J}] |G\ra$. While $|\alpha \ra$ is simply a product state containing information of the couplings of the Ising model, $|G\ra$ is an entangled state which encodes interaction pattern of the Ising model. Specifically, $|G\ra$ is a Toric code state defined on qubits that live on the edges of the lattice. Note that since $B_F (1+B_F)=(1+B_F )$, it is clear that state $|G\ra$ is stabilized by $B_F$. On the other hand, corresponding to each vertex of the lattice or each face of the dual lattice, there is a vertex operator in the form of $A_v =\prod_{e \in v}X_e$, see Fig.(\[k4\]-b). Here $X_e$ refers to the Pauli operator $\sigma_x$ which is applied to the edge qubit on edge $e$ emanating to vertex $v$, see Fig.(\[k4\]-b). Since each vertex operator $A_v$ has zero or two common qubits with $B_F$ operators, it is concluded that $[A_v ,B_F]=0$. Finally, if we apply $A_v$ to state $|G\ra =\prod_{F} \frac{1+B_F}{2}|+\ra ^{\otimes M}$ it will pass from $\prod_{F} \frac{1+B_F}{2}$ and since $A_v |+\ra ^M =|+\ra^M$, it is concluded that the $A_v$’s are also stabilizers of $|G\ra$. Therefore, state $|G\ra$ can also be written in terms of vertex operators in the following form: $$\label{innx}
|G\ra =\prod_{v} \frac{1+A_v}{2}|0\ra ^{\otimes M}.$$ In this way, the state $|G\ra$ is stabilized by vertex and face operators $A_v$ and $B_F$ which are exactly stabilizers of Toric code state as defined in [@Kitaev2003]. Consequently, the partition function of 2D Ising model has been mapped to inner product of a product state and the Toric code state where couplings of the Ising model are encoded in the product state and interaction pattern of the Ising model is encoded in the Toric code state.
![(Color online) Loop operators $T_z ^1$, $T_z ^2$, $T_x ^1$ and $T_x ^2$ correspond to non-trivial loops around the torus.[]{data-label="k1"}](k1){width="8cm" height="4cm"}
In this way, 2D Ising model is mapped to the Toric code state as an entangled state with topological order which shows a natural robustness against local perturbation [@rob1; @rob2; @zare16; @zareiprb]. In order to understand topological order of this state, let us consider the above mapping for a lattice with a periodic boundary condition, i.e., a torus. With such a topology, it is necessary to reconsider the constraints which have been considered in Eq.(\[cons0\]). In fact, the constraints associated with faces of the lattice are not independent when we consider a periodic boundary condition. As a result, the product of all face operators in Toric code on a torus is equal to Identity operator. Furthermore, there are other constraints which should be added because of the topology of the torus. In particular, consider two non-trivial loops as shown in Fig.(\[k1\]) around the torus in two different directions. Since $Z$-type loop operators, denoted by $T^1 _z$ and $T^2 _z$, corresponding to the above non-trivial loops, can not be constructed by a product of face operators, they should be considered as two independent constraints. Consequently, state $|G\ra$ includes both trivial and non-trivial loop operators. The role of non-trivial loop operators in $|G\ra$ is a property which can not be characterized by any local order parameter. As it is shown in Fig.(\[k1\]), there are two non-trivial loops on the dual lattice and we can define two $X$-type operators corresponding to such loops, denoted by $T^1 _x$ and $T^2 _x$. Since these loop operators cross $Z$-type non-trivial loop operators once, they have an anti-commutation relation in the following form: $$\label{non}
\{T^1 _x , T^2 _z \}=0~~,~~\{T^1 _z , T^2 _x \}=0.$$ Now, if we apply operators $T^1 _x$ and $T^2 _x$ to state $|G\ra$, we obtain the four-fold degenerate ground space of the Toric code: $$\label{deg}
|\psi_{\mu \nu}\ra =(T^1 _x)^\mu (T^2 _x )^\nu |G\ra,$$ with $\mu , \nu =\{0,1\}$. Because of the anticommutation relation(\[non\]), it is concluded that the expectation values of $T^1 _z$ and $T^2 _z$ are different for the above quantum states. Since $T^1 _x$ and $T^2 _x$ are non-local operators, it means that the above quantum states can be distinguished from each other only by non-local order parameters. This property, that there is no local order parameter which can distinguish the above states from each other, is in fact the result of topological order of Toric code states. We should emphasize that here we considered the periodic boundary condition in order to better explain the topological order in the Toric code. However, in the following, we will return to the open boundary condition which is more convenient for our calculations.
![(Color online) A schematic of phase diagram of random bond Ising model where ferromagnetic and paramagnetic phases are separated by a critical curve. There is also a Nishimori line where $p=\frac{e^{-2\beta J}}{1+e^{-2\beta J}}$. []{data-label="schem"}](schem){width="6cm" height="4cm"}
Quantum formalism for the order parameter {#sec2}
=========================================
2D Ising model is a well-known statistical mechanical model which can be used for describing the ferromagnetic-paramagnetic phase transition when all couplings in Eq.(\[ran\]) are fixed to $J$. However, one can consider a quenched disorder for couplings to generate a random bond Ising model where each coupling might be $J$ with a probability of $1-p$ and might be $-J$ with a probability of $p$. The set of all edges $E$ then includes two subsets of $E_1$ and $E_2$ where $E_1 \cup E_2 =E$; all edge couplings in subset $E_1$ are fixed to $+J$ and the couplings in subset $E_2 $ are fixed to $-J$. It has been shown that such a model can still show a ferromagnetic-paramagnetic phase transition. Such phase transition can be characterized by a local order parameter which is the averaged magnetization $m$, see Fig.(\[schem\]) as a schematic of phase diagram of random bond Ising model.
In the previous section we gave a quantum formalism for the partition function of such a model. In this section we are going to find a quantum formalism for the magnetization $m$ in order to consider consequences of phase transition in 2D Ising model in the quantum side. Such a study has been done for the Ising model without quenched disorder where the above phase transition is mapped to a coherent/non-coherent phase transition in the noisy Toric code state [@zareimon]. However, here we are going to consider random-bond Ising model which has a richer phase diagram and the magnetization is a function of the temperature and the probability of disorder $p$.
We start with a quenched configuration of couplings to calculate the magnetization and then we calculate the mean value of this quantity for all realizations of the quenched disorder. For an arbitrary spin, for example $s_n$: $$\label{deg}
m_n[\mathbf{J}]=\la s_n \ra=\frac{1}{\mathcal{Z}[\mathbf{J}]}\sum_{\mathbf{s}}s_n e^{\beta \sum_{\la i,j\ra }J_{ij}s_i s_j },$$
![(Color online) a) String $\gamma_n$ starts from a spin in the boundary and ends in a spin at site $n$. b) The Toric code state corresponding to the Ising model with the above boundary condition will have an open boundary where face operators in the boundary are three-local. To each string $\gamma_n$ we assign a $Z$-type operator $\Gamma_Z =\prod_{i \in \gamma_n }Z_i$.[]{data-label="string"}](string){width="7cm" height="10cm"}
where $J_{ij}=+J~(-J)$ when $i$ and $j$ are endpoints of an edge $e\in E_1 ~(E_2)$. In order to find a quantum formalism for the above relation, let us consider an open boundary condition for the model where spins of the boundary are fixed to $s_0 =+1$, see Fig.(\[string\]). By such boundary condition, we consider a string called $\gamma_n$ which starts from a spin in the boundary and ends in spin $s_n$. Next, since $s_i ^2 =1$, we can rewrite $s_n$ as $s_n s_{n-1}s_{n-1} ... s_1 s_1 s_0$ and therefore, Eq.(\[deg\]) is written in the following form: $$\label{deg2}
m_n[\mathbf{J}]=\frac{\sum_{\mathbf{s}}(s_n s_{n-1})(s_{n-1}s_{n-2})...( s_1 s_0) e^{\beta \sum_{\la i,j\ra }J_{ij}s_i s_j }}{\mathcal{Z}[\mathbf{J}]}.$$ Next, we do the same process done in the previous section for finding the quantum formalism of the partition function. First we replace Ising interactions $s_i s_j$ with edge spins $S_e$ and then these spins are replaced with Pauli operators $Z_e$. Therefore, Eq.(\[deg2\]) is written in the following quantum form: $$\label{deg3}
m_n[\mathbf{J}]=\frac{\la \alpha[\mathbf{J}] |\prod_{e\in \gamma_n}Z_e |G\ra}{\la \alpha[\mathbf{J}] |G\ra},$$ where the operator $\Gamma_Z =\prod_{e\in \gamma_n} Z_e$ refers to a product of $Z$ operators corresponding to all edge qubits belonging to string $\gamma_n$. Furthermore, we note that since here we have an Ising model with open boundary, the state $|G\ra$ in the above relation is also the Toric code state defined on a lattice with open boundary condition. Thus, the face operators in the boundary should be a local operator in the form of a product of three $Z$ operators which is called a three-local operator, see Fig.(\[string\]). The operator $\Gamma_Z$ is in fact the quantum form of the product $(s_n s_{n-1})s_{n-1} ... s_1 (s_1 s_0)$ in Eq.(\[deg2\]). Now we have a quantum formalism for the order parameter of the random bond Ising model with an arbitrary quenched configuration of disorder. We should emphasize that it is necessary to take the mean value $\overline{m_n[\mathbf{J}]}$ for different quenched configurations. Therefore, the mean value $m=(\sum_n \overline{m_n[\mathbf{J}]})/N$ will be a function of $\beta$ and $p$. In the next section, we find a physical meaning for this quantity where we show that it is related to a decoherence problem in the Toric code model.
Mapping to the Toric code model under bit-flip channels {#sec3}
=======================================================
By now, we have only rewritten the partition function as well as the order parameter of the random bond Ising model in a quantum formalism. However, the above quantum formalism seems only a mathematical mapping and we need to introduce a physical meaning for the order parameter in the quantum side. In this section, we show that the order parameter can characterize coherence of a quantum state in the Toric code model which is under two sequences of bit-flip channels. To this end, first we need to introduce the bit-flip channel and consider its effect on the Toric code model. This section involves two subsections where in the first one we introduce the noisy Toric code and in the second one we come back to the quantum formalism of the order parameter and derive the main result of the paper.
Toric code model and bit-flip noise
-----------------------------------
As we briefly mentioned in the first section, Toric code model is defined by two kinds of stabilizer operators $B_F$ and $A_v$. Furthermore, the Toric code state $|G\ra$ is also a ground state of a Hamiltonian in the form of $H=-\sum_F B_F -\sum_v A_v$. Now, suppose that the initial state is the same as the ground state $|G\ra$ and then a Pauli operator $X$ is applied to each qubit with probability $p$. Here we use the same notation of random bond Ising model because we are going to map these problems to each other. Therefore, the initial state will change to the following state after the bit-flip noise: $$\label{bit}
\rho=\sum_{\mathcal{E}} W_{\mathcal{E}}(p) \hat{\mathcal{E}}(X)|G\ra \la G| \hat{\mathcal{E}}(X),$$ where $\mathcal{E}$ refers to a pattern of qubits which are affected by the noise and $\hat{\mathcal{E}}(X)$ refers to a product of $X$ operators which are applied to the qubits belonging to the error pattern $\mathcal{E}$. $W_{\mathcal{E}}(p)$ is the probability that an error patter $\mathcal{E}$ happens. If the number of qubits which have been affected by the noise is equal to $l$ and the total number of qubits is $M$, $W_{\mathcal{E}}(p)$ will be equal to $p^l (1-p)^{M-l}$.
![(Color online) An $X$ operator on a qubit in the Toric code does not commute with two face operators which are neighbors of that qubit. It leads to two excitations in the neighboring faces. A string of $X$ operators also leads to two excitations in the endpoints of that string.[]{data-label="flux"}](flux){width="6cm" height="6cm"}
It is useful to give a geometrical description for the effect of the bit-noise in the Toric code state. As it has been shown in Fig.(\[flux\]), when an $X$ operator is applied to a qubit, it does not commute with two face operators which involve that qubit. Therefore, the system goes to an excited state which can be represented by two excitations, which are called flux anyons, in the corresponding faces. Consequently, the effect of the bit-flip operator on a qubit is geometrically described by a string in the dual lattice whose endpoints are flux anyons. In the same way, the effect of bit-flip noise on other qubits can be described by a pattern of strings whose endpoints show excitations. However, there can also be some closed strings (loops) where the pattern of effect of bit-flip operators is described by loops. Such patterns do not lead to an excitation and the Toric code state remains stable under such patterns. Consequently, the state $\rho$ is a mixture of excited states and the ground state of the Toric code model. In particular, one can compute the fidelity of $\rho$ with the ground state i. e. $F=\la G| \rho |G\ra$ which is also a measure of stability of the Toric code state against the bit-flip noise. This quantity will be in fact equal to total probability that bit-flip noise generates a pattern of loops in the dual lattice.
By the above geometrical description for noise patterns, it seems that some concepts like stability against noise might be related to some geometrical problems. Furthermore, other geometrical problems can emerge when we consider other initial states in Eq.(\[bit\]). For example, we define a specific quantum state as a coherent superposition of the ground state and an excited state of the Toric code model, an excitation which is generated by applying $Z$ operators in the ground state. To this end, as it has been shown in Fig.(\[loop\]), consider a Toric code model with an open boundary condition. Then consider a string $\gamma$ which starts from the boundary and ends to a vertex of the lattice, see Fig.(\[loop\]). Then, we construct the string operator $\Gamma_Z =\prod_{e \in \gamma}Z_e$ corresponding to such a string. This operator commutes with all face operators. It also commutes with all vertex operators except the vertex operator corresponding to the vertex in the endpoint of the string. Therefore, $\Gamma_Z$ leads to an excitation called charge anyon in the Toric code model which is geometrically represented by a string in the lattice. Now, consider a coherent superposition from the ground state and the above excited state in the following form:
![(Color online) Toric code can be defined on a lattice with an open boundary where face operators in the boundary are three-local. A string $\gamma$ which starts from the boundary leads to an excitation in the vertex in the end point of $\gamma$. There are two $X$-type loop operators $L_+$ and $L_-$ which cross the string $\gamma$ for even and odd number of times, respectively.[]{data-label="loop"}](loop){width="6cm" height="6cm"}
$$\label{str}
|\psi_+ \ra =\frac{1}{\sqrt{2}}(|G\ra +\Gamma_Z |G\ra ).$$
Next, we consider the above state as an initial state which has been affected by a bit-flip noise. As we explained, the bit-flip noise can be described by patterns including open strings and loops. Similar to the Toric code state $|G\ra$, the effect of open strings on the state $|\psi_+ \ra$ leads to flux anyons. However, the effect of loops is different here. We divide the noise patterns that correspond to loops to two subsets, see Fig.(\[loop\]). The first subset, denoted by $L_+$, contains loops that cross the string $\gamma$ an even number of times and the second subset, denoted by $L_-$, contains loops that cross the string $\gamma$ for an odd number of times. Since the $X$-type loop operators corresponding to the first subset commute with the operator $\Gamma_Z$, they preserve the state $|\psi_+ \ra$. On the other hand, the $X$-type loop operators corresponding to the second subset anti-commute with the $\Gamma_Z$ and therefore, they convert the state $|\psi _+\ra$ to a new state in the form of $|\psi_- \ra =\frac{1}{\sqrt{2}}(|G\ra -\Gamma_Z |G\ra$. It will be interesting to consider the effect of bit-flip noise only in a subspace generated by $|\psi_+ \ra$ and $| \psi_- \ra$. The initial state is a coherent state in this subspace and after bit-flip channel the quantum state in this subspace is converted to a completely mixed state. It has been shown that there is a quantum phase transition from coherent to the mixed phase in a critical value of the probability $p_{cr}$ [@zareimon]. Such a result has been derived by a mapping from the order parameter of the Ising model to an order parameter in Toric code state which characterizes the coherence in the above subspace. In the next subsection, we consider a different noise process on the $|\psi _+\ra $ where the coherence of the model is mapped to order parameter of random bond Ising model.
Random bond Ising model and transition in coherence
---------------------------------------------------
Let us start again with the initial state $|\psi_+ \ra$. Then, we consider two sequences of bit-flip channels of (\[bit\]) where the probability of the noise is $p$ for the first channel and is $q$ for the second one. We also suppose that after the first quantum channel, a measurement of face operators $B_F$ is done in order to find what excitations have been generated by the noise. Therefore, after measurement we will have a quenched pattern of the error and then the second quantum channel is applied. Consequently, the final quantum state should be in the following form: $$\label{bit1}
\Phi[\boldsymbol\eta] =\sum_{\mathcal{E}} W_{\mathcal{E}}(q) \hat{\mathcal{E}}(X) \hat{\eta}(X) |\psi_+ \ra \la \psi _+| \hat{\eta}(X)\hat{\mathcal{E}}(X),$$ where $\eta$ refers to the error pattern generated by the first channel. Similar to the previous subsection, here we give a geometrical description of the above state. The most important difference is that here we have a combination of two noise patterns related to operators $\hat{\mathcal{E}}(X)$ and $ \hat{\eta}(X)$. However, both the above operators are in the same type and both of them involve patterns of open strings and loops. Therefore, the initial state goes to its excitations if the combination of $\hat{\mathcal{E}}(X)$ and $ \hat{\eta}(X)$ corresponds to a pattern that includes open strings and it remains in the subspace of $|\psi _+\ra$ and $|\psi_ -\ra$ if the above combination corresponds to a pattern of loops. Furthermore, here we can also consider coherence in the above subspace where loops that cross the string $\gamma$ for an odd number of times lead to decoherence in the above subspace. In order to characterize the coherence in the above subspace, we define the following parameter as a measure of coherence: $$\label{coh}
O[\boldsymbol\eta]=\frac{W_+[\boldsymbol\eta] -W_-[\boldsymbol\eta]}{W[\boldsymbol\eta]}.$$ Where $W_+[\boldsymbol\eta]$ is the fidelity of state $\Phi[\boldsymbol\eta]$ with $|\psi_+ \ra$ in the form of $W_+[\boldsymbol\eta] =\la \psi_+ |\Phi[\boldsymbol\eta] |\psi_+ \ra$. Similarly, $W_-[\boldsymbol\eta]$ is the fidelity of $\Phi[\boldsymbol\eta]$ and $|\psi_- \ra$ which is $W_-[\boldsymbol\eta] =\la \psi_- |\Phi[\boldsymbol\eta] |\psi_- \ra$ and $W[\boldsymbol\eta]=W_+[\boldsymbol\eta] + W_-[\boldsymbol\eta]$. By a geometrical description, $W_+[\boldsymbol\eta]$ is interpreted as a probability that the combination of errors $\hat{\mathcal{E}}(X)$ and $ \hat{\eta}(X)$ generates loop configurations which cross string $\gamma$ an even number of times and $W_-[\boldsymbol\eta]$ is the probability that the above combination generates loop configurations which cross $\gamma$ an odd number of times. We should also emphasize that the measurement that we have done after the first channel can lead to different patterns of $\hat{\eta}(X)$ with the corresponding probability. However, we have defined the coherent parameter $O[\boldsymbol\eta]$ for one specific quenched pattern, therefore one should take the mean value $O(p,q)=\overline{O[\boldsymbol\eta]}$ for different realizations of $\hat{\eta}(X)$. Since the probability of generating $\hat{\eta}(X)$ is a function of $p$, the mean value of the coherent parameter is a function of $p$ and $q$. Interestingly, we shall see that the coherence parameter $O(p,q)$ is mapped to the order parameter of random bond Ising model. To show this, consider the quantum form of the local magnetization:
$$\label{deg4}
m_n[\mathbf{J}]=\frac{\la \alpha[\mathbf{J}] |\prod_{e\in \gamma_n}Z_e |G\ra}{\la \alpha[\mathbf{J}] |G\ra}.$$
In this equation, for a quenched configuration of couplings $\mathbf{J}$, there is a product state $|\alpha[\mathbf{J}] \ra =\frac{1}{\sqrt{2}^M}\otimes_{e}(e^{\beta J_e }|0\ra +e^{-\beta J_e}|1\ra)$, where $J_e =+J$ for $e \in E_1$ and $J_e =-J$ for $e \in E_2$. Such a quenched configuration of disorder in couplings can be mapped to a quenched pattern of errors in the Toric code which was denoted by $\eta$. Then, we do a change of variable from $\beta J$ to a parameter $q=\frac{e^{-2\beta J}}{1+e^{-2\beta J}}$. Since $\beta J$ is a quantity between zero to infinity, it is concluded that $0\leq q\leq \frac{1}{2}$. by such a change of variable the state $(e^{\beta J }|0\ra +e^{-\beta J}|1\ra)$ for $e\in E_1$ is converted to $e^{\beta J}~(1+\frac{q}{1-q}X)|0\ra$ and the state $(e^{-\beta J }|0\ra +e^{+\beta J}|1\ra)$ for $e\in E_2$ is converted to $e^{\beta J}~(X+\frac{q}{1-q}1)|0\ra=e^{\beta J}~X(1+\frac{q}{1-q}X)|0\ra$. Therefore, the state $|\alpha \ra$ is written in the following form: $$|\alpha[\mathbf{J}] \ra=e^{M\beta J}\otimes_{e}(1+\frac{q}{1-q}X_e) \otimes_{e\in E_2}X_e |0\ra ^{\otimes M}$$ $$\label{e2}
=\frac{e^{M\beta J}}{(1-q)^{M}} \otimes_{e}((1-q)I+q X_e) \otimes_{e\in E_2}X_e |0\ra ^{\otimes M}.$$ Next, consider the term $\otimes_{e}((1-q)I+q X_e)$ in the above equation. If we expand this product, it generates different patterns of $X$ operators where an $X$ operator appears with probability $q$. Therefore, we have: $$\otimes_{e}((1-q)I+q X_e)=\sum_{\mathcal{E}} W_{\mathcal{E}}(q) \hat{\mathcal{E}}(X),$$ where $W_{\mathcal{E}}(q)$ is the probability of error pattern $\mathcal{E}$ which is defined in Eq.(\[bit\]). On the other hand, the other factor $\otimes_{e\in E_2}X_e$ in Eq.(\[e2\])) is also a quenched pattern of $\eta$ where each qubit has been affected by an $X$ operator with probability $p$. In this way, the state $|\alpha \ra$ includes a superposition of all error patterns generated by the two sequences of bit-flip channels. The quenched error after the first channel is generated by a measurement of the face operators.
Now, we are ready to consider the inner product $\la \alpha[\mathbf{J}] |\prod_{e\in \gamma_n}Z_e |G\ra$ in Eq.(\[deg4\]). As we explained in Sec.(\[sec1\]), since the $A_v$ operators can be represented by loops in the dual lattice, the state $|G\ra=\prod_{v}(1+A_v)|0\ra^{\otimes M}$ includes a superposition of $X$-type loop operators. We divide again all possible loop configurations to two subsets. The first subset called $L_+$ includes loop configurations which cross the string $\gamma_n$ an even number of times and the second subset called $L_-$ includes loop configurations which cross the string $\gamma_n$ for an odd number of times. Therefore, the state $\prod_{e\in \gamma_n}Z_e |G\ra$ will be a superposition of the loop configurations in $L_+$ with a positive weight and the loop configurations in $L_-$ with a negative weight: $$\prod_{e\in \gamma_n}Z_e |G\ra =|G_+\ra -|G_-\ra.$$ Consequently, the inner product $\la \alpha[\mathbf{J}] |\prod_{e\in \gamma_n}Z_e |G\ra$ is equal to $\la \alpha[\mathbf{J}] |G_+\ra -\la \alpha[\mathbf{J}]|G_-\ra$. In this way, we have an inner product of $| \alpha[\mathbf{J}]\ra$, as a superposition of all possible error patterns with the corresponding probabilities, with $|G_+\ra$ and $|G_-\ra$, as two kinds of loop configurations. Therefore, up to an irrelevant factor, $\la \alpha[\mathbf{J}] |G_+\ra$ ($\la \alpha[\mathbf{J}] |G_-\ra$) will be equal to probability that the bit flip noise generates loop configurations which cross the string $\gamma_n$ an even (odd) number of times which is the same as $W_+[\mathbf{J}]$ ($W_-[\mathbf{J}]$). Furthermore, it is clear that denominator in Eq.(\[deg4\]) is also equal to $W_+[\mathbf{J}] +W_-[\mathbf{J}]$ up to an irrelevant factor which is removed from denominator and nominator. Therefore, the order parameter of the random bond Ising model will be in the form of $m_n[\mathbf{J}]=\frac{W_+[\mathbf{J}] -W_-[\mathbf{J}]}{W_+[\mathbf{J}] +W_-[\mathbf{J}]}$ which is the same as definition of the coherence parameter in Eq.(\[coh\]). Consequently, a ferromagnetic paramagnetic phase transition in the 2D random bond Ising model is exactly mapped to a transition from a coherent to a non-coherent phase in the noisy Toric code model. It is also important to compare our result with the result which has already been derived in [@zareimon], where a similar phase transition has been found. We should emphasize that here we have a richer phase diagram with the two quenched and annealed noise parameters which determine the coherence phase of the model. Therefore, there is a coherence region in the phase diagram with a non-zero coherence order parameter. We will return to this point in the conclusion in order to show it can lead to different insights on the noisy Toric code model.
Connection to error correction in the Toric code {#sec4}
================================================
In this section, we want to point out an interesting connection between our problem and quantum error correction in the Toric code. In particular, in [@Dennis2002] authors have considered Toric code model defined on a torus under bit-flip noise where, after measuring error syndromes, errors are corrected by an active process. To this end, one has to apply a string of $X$ operators between two syndromes to generate a loop to correct the error. However, since the Toric code has been defined on a torus, there is a probability that the string applied between two syndromes generates a non-trivial loop around the torus. Since such a non-trivial loop leads to an error in the code space, the Toric code will be non-correctable under such a situation. Therefore, the probability of the noise $p$ should be below a threshold such that the probability of generating a non-trivial loop goes to zero when the lattice size goes to infinity. It technically means the quenched error syndrome and the string of correction are in the same homological class. In the above paper, the authors have shown that the probability of being in the same homological class is mapped to the partition function of the random bond Ising model. Therefore, error threshold of the Toric code is mapped to the critical point of random bond Ising model on Nishimori line where $p=\frac{e^{-2 \beta J}}{1+e^{-2 \beta J}}$. In this respect, it seems that there are similarities with our problem where we have also a mapping to random bond Ising model and $p=q$ is in fact the Nishimori line in our process.
In order to clarify the above connection, let us come back to Eq.(\[bit1\]) and replace the initial state with the ground state of the Toric code model defined on a torus: $$\label{}
\Phi^*[\boldsymbol\eta] =\sum_{\mathcal{E}} W_{\mathcal{E}}(q) \hat{\mathcal{E}}(X) \hat{\eta}(X) |G \ra \la G| \hat{\eta}(X)\hat{\mathcal{E}}(X).$$ In fact, we will have two sequences of bit-flip channels on the Toric code state with a measurement of syndromes after the first channel. In order to be in the Nishimori line, we suppose that for both error patterns $\hat{\mathcal{E}}$ and $\hat{\eta}$, probability of single-qubit noise is equal to $p$. Then, we consider the fidelity of the final state with the initial state in the form of $F^*[\boldsymbol\eta]=\la G|\Phi ^*[\boldsymbol\eta]|G\ra$. Using geometrical interpretation of error patterns, it is concluded that $F^*$ is equal to total probability that error pattern $\hat{\mathcal{E}}$ generates loop configurations when it is added to quenched pattern $\hat{\eta}$. It is indeed equal to the probability that error patterns $\hat{\mathcal{E}}$ and $\hat{\eta}$ are in the same homological class. On the other hand, as it has been shown in [@Dennis2002], this quantity is directly related to correctability of the Toric code under bit-flip noise in the sense that a singularity in this quantity reveals a transition in correctability. Interestingly, $F^*$ is also proportional to partition function of random bond Ising model. To check this, one can start with quantum formalism of the partition function $\mathcal{Z}[\mathbf{J}]=2^M \la \alpha [\mathbf{J}]|G\ra$. Then, similar to the procedure for the order parameter, and by a change of variable in $|\alpha [\mathbf{J}] \ra$ in the form of $q=\frac{e^{-2 \beta J}}{1+e^{-2 \beta J}}$, one can show that $\mathcal{Z}[\mathbf{J}]$ is proportional to $F^*$. Therefore, it is concluded that ferromagnetic-paramagnetic transition in the random bond Ising model on the Nishimori line is mapped to a transition in the correctability of the Toric code.
We note that we derive the above result by using dual correspondence between Toric code and 2D Ising model. Therefore, it seems that such a result can be obtained in other topological CSS codes which are related to classical spin models by the dual correspondence [@zare18]. It means that we have a useful tool for finding error thresholds in topological CSS codes; by using the hypergraph duality which was introduced in [@zare18], one is able to find the classical spin model that is associated to an arbitrary topological CSS state. Then, one should numerically calculate the magnetization of the corresponding random bond spin model to find the transition point on the Nishimori line. In order to clarify this point, let us give some examples as follows.
As it has been shown in [@zare18], using duality of hypergraphs one can show that dual of color code on a D-colex (D dimensional color complex) is an Ising model with (D+1)-body interactions on a D-simplicial lattice. Consequently, in order to find the error threshold of the color code on a D-colex, it is enough to find the transition point in a random bond Ising model with (D+1)-body interaction on the D-simplicial lattice. Interestingly, this problem has recently been studied in [@3dcolor] where the authors have shown that the error threshold of the color code defined on a 3-colex is mapped to the transition point of an Ising system with four-body interactions defined on a 3-simplicial lattice which confirms our statement. For another example, in [@zare18] it has been shown that dual of the Toric code defined on an arbitrary graph in arbitrary dimension is an ordinary Ising model on the same graph. Therefore, transition point of random bond Ising models defined on an arbitrary graph leads to the error threshold of the Toric code on the same graph. We should emphasize that such correspondence has already been introduced for two and three dimensional examples while our result is held for any dimension.
![(Color online) a) X-cube model is defined on a cubic lattice with qubits living on links of the lattice. Corresponding to each cell of the lattice denoted by yellow (light) color, an $X$ type stabilizer is defined by the product of $X$ operators for the 12 qubits belonging to the cubic cell. b) In a dual picture, one should insert spins in the center of each cell. Since each original qubit belongs to four cells of the lattice, each edge of the dual hypergraph involves four spins. []{data-label="xcub"}](xcub){width="6cm" height="11cm"}
Furthermore, we note that the above duality is held for all CSS codes and it is not limited to archetypical examples including Toric codes and color codes. For example, a different quantum CSS code is $X$-cube model which has attracted much attention because of its more exotic properties than topological codes [@xcub0]. This code is defined on a cubic lattice with qubits living on links of the lattice, see Fig.(\[xcub\]-a). $X$-type stabilizers of this code are defined by 12-body interaction terms corresponding to each cubic cell of the lattice. According to [@zare18], we can define a hypergraph $H$ corresponding to this code where each cubic cell is equal to an edge of $H$ connecting the 12 qubits (vertices of $H$) of the cell. In order to find dual of the above hypergraph, we should exchange the role of vertices and edges in $H$. To this end, we insert new vertices in the center of each cubic cell of the lattice. The dual hypergraph $\tilde{H}$ is defined with these new vertices and the set of edges that correspond to the vertices of $H$. Since each vertex of $H$ (each link of the cubic lattice) is a member of four edges of $H$ (four cells of the cubic lattice), we conclude that each edge of $\tilde{H}$ has to involve four vertices corresponding to those cells, see Fig.(\[xcub\]-b). Consequently, the dual hypergraph $\tilde{H}$ can be represented as a three dimensional lattice with vertices living on nodes of the lattice and each plaquette of the lattice corresponds to an edge of $\tilde{H}$. Then we can define a classical spin model corresponding to such a hypergraph where spins live on vertices and there is a four-body interaction for each plaquette of the 3D lattice. Finally, one can study the ferromagnetic phase transition in such a classical model with random couplings to find the error threshold of the $X$-cube model. We should emphasize that the concern of this study is not to find the error threshold for different topological CSS states. The aim of the above examples is to show how the hypergraph duality can be used to find the classical spin models corresponding to different topological CSS states.
Discussion
==========
Although mapping the error threshold of topological codes to the phase transition of random bond Ising models has been established in the past decade, in this paper, we tried to reveal another aspect of those connections by using the quantum formalism of partition functions. First, we introduced a different interpretation for the transition point of the random bond Ising model in the quantum side as a quantum phase transition from a coherent to a non-coherent phase in a noisy Toric code model. Specifically, since there are two quenched and annealed noise parameters $(p,q)$ in the model, we find a richer phase diagram for the coherence phase of the model compared to the homogeneous model. In particular, the phase diagram, which has schematically been plotted in Fig.(\[phase\]), shows that the quenched noise is more destructive to the coherence than the annealed noise, i.e. $p_c < q_c$. Moreover, the maximum level of the quenched noise that the coherence can tolerate is not very sensitive to the strength of the annealed noise up to the Nishimori point $q_{th}(p_c)$. On the other hand, the threshold value of the annealed noise goes abruptly to zero at the Nishimori point $p_c$, which coincides with the error correction threshold.
![(Color online) A schematic of phase diagram for the noisy Toric code model where transition point to the non-coherenct phase denoted by $q_{th}$ decreases by increasing quenched noise $p$ in a sense that corresponding to the transition curve, the $q_{th}$ is a decreasing function of $p$. The line of p=q refers to the Nishimori line and the transition point on the Nishimori line is called Nishimori point which determines the error threshold $q_{th}(p_c)$. []{data-label="phase"}](phase){width="8cm" height="5cm"}
Then, we showed that the problem of finding the error threshold of topological CSS codes is hidden in the quantum formalism of partition functions. Therefore, the duality correspondence between the partition functions of classical spin models and the quantum CSS states can be used to study the error threshold of topological CSS codes. As we explained in a few examples, by using the hypergraph duality correspondence, one would be able to find the classical spin model corresponding to an arbitrary CSS code. Therefore, our result provides a new tool for the study of error threshold in different CSS codes.
Finally, we would like to note that our results reveals the importance of studying different thermodynamic quantities in frame of quantum formalism of partition functions. While most studies had been done on partition functions, we showed that quantum formalism for the order parameter can also provide important insights about quantum stabilizer states. Therefore, it would be interesting to consider quantum formalism for other thermodynamic (well -known) quantities.
Acknowledgement {#acknowledgement .unnumbered}
===============
We would like to thank A. Montakhab for his valuable comments on this paper.
[99]{} J. Geraci, D. A. Lidar, “On the exact evaluation of certain instances of the Potts partition function by quantum computers," Commun. Math. Phys. 279, 735 (2008). D. A. Lidar, O. Biham, “Simulating Ising spin glasses on a quantum computer," Phys. Rev. E 56, 3661 (1997). J. Geraci, D. A. Lidar, “Classical Ising model test for quantum circuits," New J. Phys. 12, 75026 (2010). M. Van den Nest, W. Dür, R. Raussendorf, and H. J. Briegel, “Quantum algorithms for spin models and simulable gate sets for quantum computation." Phys. Rev. A 80, no. 5 (2009): 052334. W. Dür, M. Van den Nest, “Quantum simulation of classical thermal states." Phys. Rev. Lett. 107, no. 17 (2011): 170402. R. D. Somma, C. D. Batista, G. Ortiz, “Quantum approach to classical statistical mechanics," Phys. Rev. Lett. 99, 030603 (2007). C. Gogolin, and J. Eisert, “Equilibration, thermalisation, and the emergence of statistical mechanics in closed quantum systems." Reports on Progress in Physics 79, no. 5 (2016): 056001. C. Castelnovo, C. Chamon, C. Mudry, P. Pujol, “From quantum mechanics to classical statistical physics: Generalized Rokhsar-Kivelson Hamiltonian and the stichastic matrix form decomposition," Ann. Phys. 318(2), 316-344 (2005) G. De las Cuevas, “A quantum information approach to statistical mechanics." Journal of Physics B: Atomic, Molecular and Optical Physics 46.24 (2013): 243001. D. Vodola, D. Amaro, M. A. Martin-Delgado, and M. Müller, “Twins percolation for qubit losses in topological color codes," Phys. Rev. Lett. 121, 060501 (2018). M. Van den Nest, W. Dür, H. J. Briegel, “Classical spin models and the quantum-stabilizer formalism," Phys. Rev. Lett. 98, 117207 (2007). M. H. Zarei, A. Montakhab, “Dual correspondence between classical spin models and quantum CSS states," Phys. Rev. A 98, 012337 (2018). S. Bravyi, R. Raussendorf, “Measurement-based quantum computation with the toric code states," Phys. Rev. A 76, 022304 (2007). H. Bombin, M. A. Martin-Delgado, “Statistical mechanical models and topological color codes," Phys. Rev. A 77, 042322 (2008). M. Van den Nest, W. Dür, H. J. Briegel, “Completeness of the classical 2D Ising model and universal quantum computation," Phys. Rev. Lett. 100, 110501 (2008). V. Karimipour. M. H. Zarei, “Algorithmic proof for the completeness of the two-dimensional Ising model," Phys. Rev. A 86, 052303 (2012). G. De las Cuevas, W. Dür, H. J. Briegel, and M. A. Martin-Delgado, “Unifying all classical spin models in a Lattice Gauge Theory," Phys. Rev. Lett. 102, 230502 (2009). Y. Xu, G. De las Cuevas, W. Dür, H. J. Briegel, M. A. Martin-Delgado, “The U (1) lattice gauge theory universally connects all classical models with continuous variables, including background gravity," J. Stat. Mech. 1102:P02013 (2011). V. Karimipour, M. H. Zarei, “Completeness of classical $\phi^4$ theory on two-dimensional lattices," Phys. Rev. A 85, 032316 (2012). M. H. Zarei, Yahya Khalili, “Systematic study of the completeness of two-dimensional classical $\phi ^4$ theory," Int. J. Quantum Inform. 15, 1750051 (2017). G. De las Cuevas, T. S. Cubitt, “Simple universal models capture all classical spin physics," Science 351.6278 : 1180-1183 (2016). T. S. Cubitt, A. Montanaro, and S. Piddock, “Universal quantum hamiltonians." Proceedings of the National Academy of Sciences 115, no. 38 (2018): 9497-9502. Pathria, R. K. *Statistical Mechanics*, International Series in Natural Philosophy Volume 45, Pergamon Press, Oxford, UK, 1986. M. H. Zarei, A. Montakhab, “classical criticality establishes quantum topological order," arxiv:1907.06216 (2019). M. H. Zarei, A. Montakhab, “Phase transition in a noisy Kitaev Toric code model," Phys. Rev. A 99, 052312 (2019). E. Dennis, A. Kitaev, A. Landahl, and J. Preskill, “Topological quantum memory," J. Math. Phys. 43, 4452 (2002). H. G. Katzgraber, H. Bombin, M. A. Martin-Delgado, “Error threshold for color codes and random three-body Ising models," Phys. Rev. Lett. 103, 090501 (2009). R. S. Andrist, H. G. Katzgraber, H. Bombin, and M. A. Martin-Delgado, “Error tolerance of topological codes with independent bit-flip and measurement errors," Phys. Rev. A 94, 012318 (2016). R. S. Andrist, J. R. Wootton, and H. G. Katzgraber, “Error thresholds for Abelian quantum double models: Increasing the bit-flip stability of topological quantum memory," Phys. Rev. A 91, 042331 (2015). R. S. Andrist, H. Bombin, H. G. Katzgraber and M. A. Martin-Delgado, “Optimal error correction in topological subsystem codes," Phys. Rev. A 85, 050302(R) (2012). H. Bombin, R. S. Andrist, M. Ohzeki, H. G. Katzgraber, and M. A. Martin-Delgado, “Strong Resilience of Topological Codes to Depolarization," Phys. Rev. X 2, 021004 (2012). A. Kubica, M. E. Beverland, F. Brandao, J. Preskill, K. M. Svore, “Three dimensional color code thresholds via statistical mechanical mappings," Phys. Rev. Lett 120, 180501 (2018) H. Nishimori, “Internal energy, specific heat and correlation function of the bond-random ising model," Progress of Theoretical Physics 66(4), 1169-1181 (1981). A. Y. Kitaev, “Fault-tolerant quantum computation by anyons," Ann. Phys. (N.Y.) 303, 2 (2003). S. Trebst, P. Werner, M. Troyer, K. Shtengel, and C. Nayak, “Breakdown of a topological phase: Quantum phase transition in a loop gas model with tension.“ Phys. Rev. Lett. 98, 070602 (2007) S. Dusuel, M. Kamfor, R. Orus, K. P. Schmidt, and J. Vidal, ”Robustness of a perturbed topological phase." Phys. Rev. Lett. 106, 107203, (2011).
M. H. Zarei, “Robustness of topological quantum codes: Ising perturbation." Physical Review A 91, no. 2 (2015): 022319. Mohammad Hossein Zarei, “strong-weak coupling duality between two perturbed quantum many-body systems: CSS codes and Ising-like systems ," Phys. Rev. B 96, 165146 (2017). K. Slagle, Y. B. Kim. “X-cube model on generic lattices: Fracton phases and geometric order," Phys. Rev. B 97.16 (2018): 165106.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: '**Abstract:** A new object, called the velocity tensor, is introduced. It allows to formulate a generally covariant mechanics. Some properties of the velocity tensor are derived.'
author:
- 'E. Kapuścik'
- 'T. Lanczewski'
title: On the Velocity Tensors
---
Introduction
============
In classical mechanics [@1] the velocity $\mathbf{v}\left(
t\right)$ of a material point is defined as $$\label{1}\mathbf{v}\left(t\right) =\frac{d\mathbf{x}\left(
t\right)}{dt},$$ where $\mathbf{x}\left(t\right)$ is the trajectory function of the moving point and $t$ is the time coordinate in the chosen inertial reference frame. In special relativity [@2] the notion of the three-dimensional velocity $\mathbf{v}\left(t\right)$ is generalized to the notion of the four-velocity defined as $$\label{2}u^\mu\left(\tau\right)=\frac{dx^\mu\left(\tau\right)}{d\tau},$$ where the space–time position of the material point is given by four functions $x^\mu \left( \tau \right) (\mu=0,1,2,3; x^0 =ct)$ parameterized by the so-called proper time $$\label{3}d\tau =dt\sqrt{1-\frac{\mathbf{v}^2\left(t\right)}{c^2}}.$$ Due to its dependence on velocity of the moving material point, the notion of the proper time $\tau $ is different for each material point and for non-uniform motions the proper time is not a uniformly changing function of the coordinate time $t$. Moreover, for non-uniform motions the proper time coincides with the coordinate time in continuously changing inertial reference frames (the momentarily rest frames). Only for uniformly moving material points the proper time coincides with the coordinate time in one reference frame (the rest frame of the moving material point). In addition, for many particle systems the trajectories of particles are parametrized by different proper times and it is almost impossible to describe the interaction between particles without the notion of propagating fields. Therefore relativistic mechanics cannot be so well developed as the nonrelativistic mechanics is.
Fortunately, there exists another way of passing from Galilean–Newtonian mechanics to the relativistic one [@3] which is not based on Eq. (\[2\]). Indeed, it is easy to see that rewriting Eq. (\[1\]) in the form $$\label{4}d\mathbf{x}\left(t\right)-\mathbf{v}\left(
t\right)dt=0$$ we can immediately generalize it to a relativistic (as a matter of fact, generally) covariant form $$\label{5}V_\nu ^\mu \left( x\right) dx^\nu =0,$$ where a new mixed tensor field $V_\nu ^\mu \left( x\right) $ is introduced. We shall name this tensor as [*the velocity tensor*]{}.
It is clear that for nontrivial velocity tensors ($V_\nu ^\mu \left(
x\right) \neq \delta _\nu ^\mu $) Eq. (\[5\]) define some submanifolds of the considered space–time. We shall require from the velocity tensors that these submanifolds should always be one dimensional what means that Eq. (\[5\]) must determine some curves interpreted as trajectories of the moving material points.
Form (\[5\]) has the obvious advantage over (\[1\]) and (\[2\]), that it does not use any evolution parameter and therefore it may be applied to systems with arbitrary number of material points by generalizing (\[5\]) to the set of relations $$\label{7}V_{a,\nu }^\mu \left( x_a\right) dx_a^\nu =0,$$ where the index $a$ labels different material points.
At each space–time event the velocity tensors (different for different material points) fix the infinitesimal directions in which any material point located at that event may move. In addition, [*forms (\[5\]) and (\[7\]) are invariant under arbitrary changes of space–time coordinates.*]{} Therefore, they may be used to formulate a generally covariant scheme for classical mechanics.
The aim of the present paper is to describe some interesting properties of the velocity tensors. We shall also provide the explicit construction of the general form of such tensors.
It is clear that velocity tensors are related to the kinematical part of mechanics. We shall also touch the dynamical aspect of mechanics.
General Properties of the Velocity Tensors
==========================================
Equation (\[5\]), in $n$-dimensional space–time, is an eigen equation for the $n\times n$-dimensional matrix $V$ (defined by the velocity tensor) for the eigenvalue $0$, while the infinitesimal displacements $dx^\mu $ in any motion are the eigenvectors of the velocity tensors belonging to this eigenvalue.
Writing the characteristic equation for the general eigenvalue problem $$\label{8}V_\nu ^\mu \left( x\right) dx^\nu =\lambda dx^\mu$$ we get the equation for the possible eigenvalues $\lambda $ $$\label{9}\sum_{j=0}^n\left( -\lambda \right) ^{n-j}Tr_jV\left( x\right) =0,$$ where $Tr_jV\left(x\right)$ denotes the sums of diagonal minors of order $j$ of the matrix $V\left(x\right).$ Obviously, $Tr_1V\left(x\right)$ coincides with the ordinary trace of $V\left(x\right)$ and $Tr_nV\left(
x\right)$ is the determinant of $V\left(x\right)$. For shortness, we also use the convention $$\label{10}Tr_0V\left( x\right) =1$$ for any matrix $V\left( x\right) $.
Due to physical reason we must require that there should be only one eigenvalue equal to $0$. This means that there should be a unique eigenvector for any velocity tensor which fixes the infinitesimal displacements in any motion. The characteristic equation (\[9\]) must be therefore of the form $$\label{11}\lambda ^n=0,$$ from which we get the following conditions for any velocity tensor: $$\label{13}Tr_jV(x)=0$$ for all $j>0.$
Conditions (\[13\]) are generally covariant requirements because all the $Tr_jV,$ being the coefficients in characteristic equation (\[9\]), are invariant under arbitrary similarity matrix transformations and it is well known that for mixed tensors, treated as matrices, the general coordinate transformations locally become the similarity transformations
$$\label{14}V\left( x\right) \rightarrow V^{\prime }\left( x^{\prime }\right)
=S\left( x\right) V\left( x\right) S^{-1}\left( x\right) ,$$
where the matrix elements of $S\left( x\right) $ are given by $$\label{15}S_\nu ^\mu \left( x\right) =\frac{\partial x^{\prime\mu}\left(
x\right) }{\partial x^\nu }$$ for arbitrary changes of space–time coordinates $x^\mu \rightarrow x^{
\prime\mu }\left( x\right) $ .
Conditions (\[13\]) impose $n$ restrictions for the $n^2$ matrix elements of the velocity tensors. Further restrictions come from the requirement that, in each reference frame, from (\[5\]) it should follow that $$\label{16}dx^k=v^k\left( t\right)dt,$$ where $k=1,...,(n-1)$ and $v^k\left( t\right) $ are the components of the standard velocity. This gives us additional $n-1$ restrictions for the matrix elements of the velocity tensor. Finally, we shall require that in $n$-dimensional spacetimes the motions in all $n-k$ subspaces should be described exactly as they were described in the $\left( n-k\right) $-dimensional spacetimes. This means that restricting the motions to subspaces the form of the velocity tensor should reduce to the already established forms of the velocity tensors in the corresponding lower dimensional subspaces. We shall refer to this requirement as to the reduction principle. It is easy to count that such a requirement gives additional $2^{n-1}-2$ conditions for the matrix elements of any velocity tensor. Altogether we are left with $n^2-n-\left( n-1\right) -\left( 2^{n-1}-2\right) =\left( n-1\right)
^2-\left( 2^{n-1}-2\right) $ free parameters of any velocity tensor. These free parameters should represent components of some $\left( n-1\right)$-dimensional vector which will guarantee the covariance of the velocity tensor under space rotations because this is the only simple geometrical interpretation of the remaining constants in the velocity tensors. In this way, we arrive at the equation $$\label{17}\left( n-1\right) ^2-\left( 2^{n-1}-2\right) =n-1.$$ It is surprising that this equation has solution only for $n=2,3$ and $4$. This means that our construction can be performed only in two, three and four-dimensional spacetimes, correspondingly.
General Construction of the Velocity Tensors
============================================
We shall now present a simple method of the construction of all possible velocity tensors.
Let us consider spacetimes for which the passage between inertial reference frames is described by the linear change of coordinates
$$\label{17}x^\mu \rightarrow x^{\prime\mu }=L_\nu ^\mu \left(
\mathbf{u}\right) x^\nu ,$$
where $\mu ,\nu =0,1,2,3$ and $\mathbf{u}$ denote the relative velocity of the two inertial reference frames. From (\[17\]) and the tensor character of the velocity tensor we get the transformation law for it (written in the matrix form) $$\label{18}V\rightarrow V^{\prime }=L\left( \mathbf{u}\right)
VL^{-1}\left( \mathbf{u}\right) =L\left( \mathbf{u}\right)
VL\left( -\mathbf{u}\right) .$$
It is clear that we should look for velocity tensors which are functions of the ordinary velocity of motion. Our basic assumption consists in the requirement that the functional forms of the velocity tensor are the same in each reference frame. This means that $$\label{19}V^{\prime }\left( \mathbf{v}^{\prime }\right) =V\left(
\mathbf{v}^{\prime }\right)$$ because only under such condition in each reference frame we can fulfill conditions (\[16\]). In this way, transformation law (\[18\]) becomes to be a system of functional equations for the matrix elements of the matrix $V$ of the following form: $$\label{20}V\left( \mathbf{v}^{\prime }\right) =L\left(
\mathbf{u}\right) V\left( \mathbf{v}\right) L\left( -
\mathbf{u}\right) ,$$ where $$\label{21}v^{\prime k}=\frac{L_0^k\left( \mathbf{u}\right) +%
\sum_j L_j^k\left( \mathbf{u}\right)
v^j}{L_0^0\left( \mathbf{u}\right) + \sum_j
L_j^0\left( \mathbf{u}\right) v^j}.$$
Taking into account that the particle at rest in the unprimed reference frame moves with the velocity $-\mathbf{u}$ in the primed frame we can rewrite these functional equations in the explicit form: $$\label{22}V\left( \frac{-u^kL_0^0\left( \mathbf{u}\right) +\sum_j
L_j^k\left( \mathbf{u}\right) v^j}{
L_0^0\left( \mathbf{u}\right) +\sum_j
L_j^0\left( \mathbf{u}\right) v^j}\right) =L\left( \mathbf{u}
\right) V\left( \mathbf{v}\right) L\left( -\mathbf{u}\right).$$
The solutions of these equations are obtained by the standard method. We first put $v^k=0$, then change the signs of $u^k$ and finally rename $
\mathbf{u}$ into $\mathbf{v}$. As a result we get $$\label{23}V\left( \mathbf{v}\right) =L\left( -\mathbf{v}%
\right) VL\left( \mathbf{v}\right) ,$$ where on the right-hand side the matrix $V$ has constant matrix elements equal to the elements of $V\left( 0\right) .$ The constant matrix elements of $V$ should be determined by the additional requirements the velocity tensors have to satisfy.
For all dimensions the first column of the velocity tensor $V$ consists of null elements. This follows from the fact that for particles at rest the eigenvector in (\[5\]) is of the form $$\label{24}\left(
\begin{array}{c}
dt \\
0 \\
0 \\
0
\end{array}
\right) .$$
Such eigenvector will satisfy Eq. (\[5\]) only if $V_0^\mu =0.$
Examples
========
Two-Dimensional Space–Time
--------------------------
For $n=2$ from conditions (\[13\]) it follows that
$$\label{25}V=\left(
\begin{array}{cc}
0 & V_1^0 \\
0 & 0
\end{array}
\right) ,$$
where $V_1^0$ is an arbitrary nonzero number. Since Eq.(\[5\]) is homogeneous, this constant can be taken as $1.$
For Galilean space–time $$\label{26}L\left( u\right) =\left(
\begin{array}{cc}
1 & 0 \\
-u & 1
\end{array}
\right)$$ and from (\[23\]) we get $$\label{27}V\left( v\right) =\left(
\begin{array}{cc}
-v & 1 \\
-v^2 & v
\end{array}
\right) .$$
For Lorentz space–time $$\label{28}L\left( u\right) =\frac 1{\sqrt{1-\frac{v^2}{c^2}}}\left(
\begin{array}{cc}
1 & -\frac u{c^2} \\
-u & 1
\end{array}
\right)$$ and from (\[23\]) we get $$\label{29}V\left( v\right) =\frac 1{1-\frac{v^2}{c^2}}\left(
\begin{array}{cc}
-v & 1 \\
-v^2 & v
\end{array}
\right) .$$
Higher-Dimensional Spacetimes
-----------------------------
From the reduction principle and from the form of the velocity tensor in the two-dimensional space–time we immediately get that in all higher dimensional spacetimes the only non-zero components are the $V_k^0$. Therefore the final form of the velocity tensors is $$\label{30}V_\nu ^\mu \left( \mathbf{v}\left( t\right) \right)
=L_0^\mu \left( -\mathbf{v}\left( t\right) \right)
\sum_k V_k^0L_\nu ^k\left( \mathbf{v}\left(
t\right) \right) ,$$ where $\left( V_1^0,V_2^0,....,V_n^0\right) $ are components of a $\left(
n-1\right) $-dimensional vector under rotations in the subspace $\left(
x^1,x^2,....,x^{n-1}\right) .$ Using this form of $V$ and the explicit forms of the Galilean and Lorentz transformations we easily can get the velocity tensor both for the Galilean and Lorentz spacetimes of any dimension.
Dynamics
========
Since our kinematical part of classical mechanics is generally covariant, it is necessary to determine such form of dynamical equations which also will be generally covariant. For this purpose we shall remind that the only generally covariant differential relation which may be reduced to the famous Newton relation $$\label{27}\frac{d\mathbf{p}\left( t\right) }{dt}=\mathbf{F}%
\left( t\right)$$ is of the form $$\label{28}\nabla _\mu \pi ^{\mu \nu }\left( x\right) =F^\nu \left( x\right)
,$$ where $\pi ^{\mu \nu }\left( x\right) $ is some tensorial density and $F^\nu
\left( x\right) $ is a vector density, while $\nabla _\mu $ denotes a corresponding covariant derivative.
Assuming that $\pi ^{\mu \nu }\left( x\right) ,$ like the velocity tensors, is a function of the ordinary velocity, we easily can construct the explicit form of this quantity. This leads, exactly as for the velocity tensors, to the following form of the dynamical tensor: $$\label{29}\pi ^{\mu \nu }\left( \mathbf{v}\left( t\right) \right)
=L_\alpha ^\mu \left( -\mathbf{v}\left( t\right) \right) \pi
^{\alpha \beta }L_\beta ^\nu \left( -\mathbf{v}\left( t\right)
\right) ,$$ where all $\pi ^{\alpha \beta }$ are constants. Since, in contradiction to the velocity tensor, the dynamical tensor $\pi ^{\mu \nu }\left(
\mathbf{v}\right) $ need not to satisfy any additional conditions, we have here to do with $n^2$ arbitrary constants which describe the inertial properties of the considered particles. We may, however, diminish the number of arbitrary constants by requiring the symmetry of $\pi ^{\mu \nu }\left(
\mathbf{v}\right) $ and then only one parameter, the mass of the particle, describes its inertial property. In this case $\pi^{\mu \nu}$ simply is the energy-momentum tensor of the material point. Since the $\pi ^{\mu \nu }\left(\mathbf{v}\left( t\right) \right) $ depends only on the time coordinate, it is clear that Eq. (\[28\]) reduces to (\[27\]).
Conclusions
===========
We have introduced a new mechanical object called the velocity tensor and explicitly constructed the velocity tensors in spacetimes of any dimension. We hope that the notion of the velocity tensor will shed more light on the possible dynamics in general relativity. It also may be useful for relativistic many-body systems.
[9]{} H. Goldstein, [*Classical mechanics*]{} (Addison-Wesley, Reading, 1980)
S. Weinberg, [*Gravitation and Cosmology*]{} (Wiley, New York, 1972)
E. Kapuścik, The Old and New Concepts of Phys., [**4**]{} (2007) 547.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'A. Doron'
- 'T. Levinson'
- 'F. Gorniaczyk'
- 'I. Tamir'
- 'D. Shahar'
title: 'Supplemental Material for Limited role of vortices in transport in highly disordered superconductors near $B_{c2}$'
---
$\boldsymbol{\rho}$ vs. $\boldsymbol{B_{\perp}}$ and vs. $\boldsymbol{B_{||}}$
==============================================================================
In the main-text we study five a:InO films of various thicknesses at $B_{\perp}$ and at $B_{||}$. In Figs. \[RoleOfVortices:Supp:FigS1\]a-e we plot $\rho$ vs. $B_{\perp}$ of these films at various $T$’s. In Figs. \[RoleOfVortices:Supp:FigS1\]f-j we plot $\rho$ vs. $B_{||}$ of these films at various $T$’s.
{height="10"}
In the main-text we discuss the observation that the films become more isotropic with increasing $|B|$. In Figs. \[RoleOfVortices:Supp:FigS2\]a-e we plot $\rho$ vs. $B_{\perp}$ (crosses) and vs. $B_{||}$ (circles) measured at three different $T$’s (0.1K, 0.4K and 1K) of all films. It can be seen that as $|B|$ increases the difference between both $B$ orientation decreases making the samples more isotropic. In the two thinner films (22nm and 26nm), the lowest isotherm of 0.1K displays a reversed anisotropy at high $B$’s where $\rho(B_{||})>\rho(B_{\perp})$. A similar effect was previously reported in a:InO films \[16\] and its origin is not yet understood (we note that Ref. \[18\] does account for this effect).
In the inset of these figures we plot the relative anisotropy defined as $\frac{\rho(B_{\perp})-\rho(B_{||})}{\rho(B_{||})}$ for $T=$0.1K, 0.4K and 1K. It can be seen that the relative anisotropy decays exponentially while approaching $B_{c2}$ (and, as discussed above, can turn negative).
\[h!\] {height="10"}
Mapping of $\boldsymbol{I_{c}}$ between $\boldsymbol{B_{\perp}}$ and $\boldsymbol{B_{||}}$ in the superconducting phase
=======================================================================================================================
In Fig. 3 of the main-text we show that for the 280nm thick sample the linear mapping described by Eq. 1 where the parameters $B_{0}$ and $a$ are extracted from the equilibrium zero bias $\rho$ also maps $J_{c}^{\perp}$ onto $J_{c}^{||}$. In Fig. 4 of the main-text we show that similar linear mappings can also be extracted from zero bias $\rho$ of the 100nm, 57nm, 26nm and 22nm thick samples. In Fig. \[RoleOfVortices:Supp:FigS3\] we show that similarly to the 280nm thick sample also for the thinner, 100nm and 26nm, films the mappings presented in Fig. 4 of the main-text also maps $I_{c}^{\perp}$ onto $I_{c}^{||}$. Unfortunately we did not measure $I_{c}^{||}$’s of the 57nm thick film due to a technical error and the $I-V$’s of the 22nm thick film were very non-linear but did not show discontinuities.
{height="6"}
The mapping parameters of Eq. 1 for different samples are written in Table. \[RoleOfVortices:Table1\]. The column of SC / INS refers to the phase where we conducted the mapping.
\[h!\]
----------------- ------------------ ---------- --------------- -------
Sample name thickness \[nm\] SC / INS $B_{0}$ \[T\] $a$
\[0.5ex\] AD12a 22 SC 4.65 0.71
AD12a 26 SC 5.2 0.64
AD12a 57 SC 2.4 0.857
AD12a 100 SC 2 0.905
AD12a 280 SC 2 0.91
RAM005b 30 INS 2.72 1.02
----------------- ------------------ ---------- --------------- -------
: **Mapping parameters of Eq. 1 for different samples.**[]{data-label="RoleOfVortices:Table1"}
Mapping between $\boldsymbol{B_{\perp}}$ and $\boldsymbol{B_{||}}$ in the insulating phase of a $\boldsymbol{B}$-driven SIT
===========================================================================================================================
In the main-text we discussed the mapping between equilibrium transport at $B_{\perp}$ and $B_{||}$ in the $B$-driven insulating phase of a highly disordered a:InO film (RAM005b). In Fig. \[RoleOfVortices:Supp:FigS4\]a we plot $R(B_{\perp})$ (continuous lines) and $R(B_{||})$ (dashed lines) at different $T$’s. In Fig. \[RoleOfVortices:Supp:FigS4\]b we plot the mapping between $R$ at $B_{||}$ and $B_{\perp}$ where each point on the $B_{||}-B_{\perp}$ plane satisfies $R(B_{||},T)=R(B_{\perp},T)$ The dashed black line is a linear fit.
\[h!\] {height="6"} {height="6"}
In Fig. \[RoleOfVortices:Supp:FigS5\] we plot several $I-V$’s of matching-$B$’s in different regions in the insulating phase.
\[h!\] {height="6"}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We use high resolution 3D SPH simulations to study the evolution of self-gravitating binary protoplanetary disks. Heating by shocks and cooling are included. We consider different orbital separations and masses of the disks and central stars. Isolated massive disks ($M \sim 0.1 M_{\odot}$) fragment into protoplanets as a result of gravitational instability for cooling times comparable to the orbital time. Fragmentation does not occur in binary systems with a separation of about 60 AU. This is because efficient heating owing to strong tidally induced spiral shocks damps any overdensity. The resulting temperatures, above 200 K, would vaporize water ice in the outer disk, posing a problem even for the other model of giant planet formation, core-accretion. Light disks ($M \sim 0.01 M_{\odot}$) do not fragment but remain cold because their low self-gravity inhibits strong shocks. Core accretion would not be hampered in the latter. At separations of about 120 AU the efficiency of fragmentation by disk instability rises and approaches that in isolated systems. If disk instability is the main formation mechanism for giant planets, on going surveys targeting binary systems should find considerably fewer planets in systems with separations below $100$ AU.'
author:
- |
Lucio Mayer $^1$, James Wadsley$^2$, Thomas Quinn$^3$, Joachim Stadel$^1$\
$^1$Institute of Theoretical Physics, University of Zürich, Winterthurerstrasse 190, 8057 Zurich, Switzerland\
$^2$Department of Physics & Astronomy, McMaster University, 1280 Main St. West, Hamilton ON L8S 4M1 Canada\
$^3$Department of Astronomy, University of Washington, Seattle, WA 98195, USA
title: 'Gravitational instability in binary protoplanetary disks; new constraints on giant planet formation'
---
[accretion, accretion discs - planetary systems: protoplanetary discs - planets and satellites: formation- stars:pre-main sequence]{}
Introduction
============
The recent discovery of extrasolar planets (Mayor & Queloz 1995) has ignited renewed interest in models of giant planet formation. In the conventional model, core accretion (Lissauer 1993), it is difficult to grow planets of several Jupiter masses in less than a few million years, the typical disk lifetime estimated from observations (Haisch, Lada & Lada 2001). This problem is exacerbated by the fast inward migration rates produced by the disk-planet interaction as well as by the low accretion rates ensuing once a planet is big enough to open a gap (Nelson et al. 2001; Bate et al. 2003; Nelson & Benz 2003). Consequently, the disk instability model, in which giant planets arise in only a few disk orbital times (less than a thousand years) from the fragmentation of a massive, gravitationally unstable disk (Boss 1997; 2002; Pickett 2000;2003) has gained new attention (Mayer et al. 2002; Rice et al. 2003a,b).
The majority of solar-type stars in the Galaxy belong to double or multiple stellar systems (Duquennoy & Mayor 1991; Eggenberger et al. 2004). Binaries can be formed by the fragmentation of a single bar-unstable molecular cloud core into two distinct objects (Boss 1986; Burkert, Bate & Bodenheimer 1997), from the collision of two dense cores in a giant molecular cloud (Whitworth et al. 1995) or owing to the capture of neighboring stars in dense star forming regions (Bate et al. 2002). Fragmentation is usually considered as the main channel of binary formation and can take place in any type of environment (Horton, Bate & Bonnell 2001). In such scenario two star-disk systems should form if the initial separation is larger than $10$ AU, while at smaller separations a circumbinary disk can arise (Bate 2000). Radial velocity surveys have shown that planets exist in some binary or multiple stellar systems where the stars have separations from 20 to several thousand AU (Eggenberger et al. 2004). Although the samples are still small (20 out of the 120 known extrasolar planets are in binary systems), attempts have been made to compare properties of planets in single and multiple stellar systems (Patience et al. 2003; Udry et al. 2004). The first adaptive optics surveys designed to quantify the relative frequency of planets in single and multiple systems are just starting (Udry et al. 2004). These surveys could offer a new way to test theories of giant planet formation provided that different models yield different predictions as for the effect of a stellar companion.
So far, two works have studied giant planet formation in binary systems. Both focused on the disk instability model and reached opposite conclusions. Nelson (2000) performed 2D SPH simulations of protoplanetary disks that did not form protoplanets in isolation due to quite long cooling times (Johnson & Gammie 2003) and found fragmentation to be even more unlikely in the presence of another disk with identical mass at a mean separation less than 100 AU. In fact the disks were achieving a high stability owing to heating by shocks occurring along the spiral arms. Boss (1998, 2003) performed 3D grid calculations of a disk interacting with a stellar companion also closer than 100 AU and found that giant planet formation by disk instability can be enhanced by the tidal perturbation.
In this Letter we revisit giant plant formation in a binary system using high resolution 3D SPH simulations. In Mayer et al. (2002; 2004), we have shown that a high resolution is crucial in order to decide whether a massive protoplanetary disk will fragment into long-lasting clumps as a result of gravitational instability. Fragmentation requires the disk to cool on a timescale comparable to or shorter than the disk orbital time (Rice et al. 2003a,b; Mayer et al. 2003). Here we consider binary systems of protoplanetary disks which mostly fragment in isolation (Mayer et al. 2003) and investigate whether the tides raised by the companion enhance or suppress fragmentation.
Models and Simulations
======================
The initial conditions comprise two protoplanetary disks, usually of equal mass, orbiting around each other. Each disk orbits a central star as well. In a binary system arising from the fragmentation of a molecular cloud core one expects the two disks to form in the same plane and be rotating in the same direction relative to their orbital motion (e.g. Bate 2000).In this paper we will restrict our investigation to such coplanar prograde configurations (see also Nelson 2000). Table 1 lists the most important parameters of the simulations. The disks extend from 4 to 20 AU and their orbital separation is in the range 60-120 AU (semi-major axis $a=29$ or $58$ AU, eccentricity $e
= 0.14$), corresponding to some of the smallest separations among binary systems with detected giant planets in radial velocity surveys (Eggenberger et al. 2004). Each disk is represented by 200000 SPH particles with equal mass and fixed gravitational softening of $0.06$ AU, while the central star has a softening of 2 AU (see Mayer et al., 2004, hereafter MA04). The initial temperature reaches about $400$ K at the inner boundary and levels off at $65$ K close to 10 AU. The details of the temperature and surface density profiles of the disks can be found in MA04.
=9truecm
Disk models are in equilibrium when evolved for several orbital times in absence of self-gravity (see MA04 for details). Models as light as the minimum mass solar nebula ($0.012 M_{\odot}$) and as massive as the heaviest among T Tauri disks (D’Alessio et al. 2001) are considered. The shape of the profile of the Toomre Q parameter is the same in all disks, while its normalization varies according to the mass of the disk. We recall that a gaseous (nearly) keplerian disk is stable against local axisymmetric perturbations if $Q > 1$, $Q=\Omega c_s/
\pi G \Sigma$, where $\Omega$ is the angular frequency, $c_s$ is the sound speed, $G$ is the gravitational constant and $\Sigma$ is the surface density of the disk. The initial minimum Q parameter (reached near the disk edge) is $Q_{min} \sim 1.4$ or higher (see Table 1), hence close or above the threshold for fragmentation in isolation (MA04; Johnson & Gammie 2003).
The radiative cooling time is taken to be proportional to the local orbital time, following Rice et al. (2003a). Inside 5 AU cooling is switched off in order to maintain temperatures high enough to be comparable to those in protosolar nebula models (e.g. Boss 1998). Cooling is also switched off in regions reaching a density above $10^{-10}$ g/cm$^3$ to account for the local high opacity; the choice of the density threshold is motivated by the simulations of Boss (2002) with radiative transfer in which the temperature of the gas is observed to evolve nearly adiabatically above such densities. We consider cooling times going from $0.3$ to $1.5$ the local orbital time. The jury is still out on whether the cooling times adopted here are credible or excessively short, but recent calculations by Boss (2002) and Johnson & Gammie (2003), which use different approximate treatments of radiative transfer, do find cooling times of this magnitude through a combination of radiative losses and convection (but see Mejia et al. 2003, 2004). In any case, here we are interested in comparing the outcomes of isolated and binary systems for the same choice of the cooling time, in particular when cooling is strong enough to lead to fragmentation in isolated disks. Heating by compressions and shocks is included in the simulations. We usually adopt $\gamma=7/5$, appropriate for molecular hydrogen, with a few simulations having $\gamma = 5/3$. Shocks are modeled using the standard Monaghan viscosity with $\alpha=1$ and $\beta = 2$ plus the Balsara correction term that removes unwanted shear viscosity (see Wadsley, Stadel & Quinn 2003). The full analysis of the runs with isolated disks is carried out elsewhere (Mayer et al. 2003; Mayer et al., in preparation). Here we only report on whether the isolated disks formed clumps or not (see Table 1).
------- ---------- ---------- ---------- ----------- ----- ----------- ------------ ---------- -------------- -------------
Model $M_{dA}$ $M_{dB}$ $M_{*A}$ $M_{*,B}$ $a$ $Q_{min}$ $t_{cool}$ $\gamma$ clumps (bin) clumps (is)
RB1a 0.1 0.1 1 1 29 1.4 0.3 1.4 no yes
RB1b 0.1 0.1 1 1 29 1.4 0.5 1,4 no yes
RB1c 0.1 0.1 1 1 29 1.4 1 1.4 no yes
RB1d 0.1 0.1 1 1 29 1.4 1.5 1.4 no yes
RB1e 0.1 0.1 1 1 29 1.4 0.3 1.66 no yes
RB1f 0.1 0.1 1 1 29 1.4 1 1.66 no yes
RB2a 0.05 0.05 1 1 29 2.8 0.5 1.4 tr no
RB2b 0.05 0.05 1 1 29 2.8 0.3 1.4 yes no
RB3a 0.08 0.08 1 1 29 1.75 0.5 1.4 no yes
RB3b 0.08 0.08 1 1 29 1.75 0.3 1.4 yes yes
RB4a 0.012 0.012 1 1 29 11 0.3 1.4 no no
RB4b 0.012 0.012 1 1 29 11 1.5 1.4 no no
RBm2 0.1 0.05 1 0.5 29 1.4 (2) 0.3 1.4 tr yes
RBwa 0.1 0.1 1 1 58 1.4 0.3 1.4 yes yes
RBwb 0.1 0.1 1 1 58 1.4 0.5 1.4 yes yes
RBwc 0.1 0.1 1 1 58 1.4 1 1.4 tr yes
RBwd 0.1 0.1 1 1 58 1.4 0.5 1.66 tr yes
------- ---------- ---------- ---------- ----------- ----- ----------- ------------ ---------- -------------- -------------
\[t:simul\]
Results
=======
We begin by describing the outcome of the runs with the smallest orbital separations, corresponding to $a=29$ AU (see Table 1). Figure 1 shows the evolution of the disks in one of such simulations. Disk models are typically followed for two orbits. A binary orbit corresponds to about 288 years, or, equivalently, 10 disk orbital times at 10 AU from the disk center. Two calculations were extended further for two more orbits, but, since gravitational instability acts on the disk orbital timescale, fragmentation should occur during the first binary orbit. Disks start at apocenter and develop a strong two-armed spiral pattern after crossing pericenter. The strong non-axisymmetric torques redistribute mass and angular momentum in the disks, which soon develops higher order spiral modes at distances between 12 and 15 AU from their centers. In models that do not fragment (see Table 1) transient, moderately strong high order spiral arms continue to develop at subsequent pericentric passages, but $Q_{min}$ always remains above the threshold for stability (see Figure 2). In models that fragment, $Q_{min}$ drops below 1 (see Figure 2), and clumps appear on the disk side which is further from the other disk, along a strong unwinding trailing spiral arm. On the other side of the disk the developing overdensities are destroyed as tides tear them apart.
We can identify three regimes as for the disk response to the tidal perturbation. Such regimes can be distinguished based on the disk mass. Very light disks, which are completely stable in isolation, develop a clear two-armed spiral pattern but their self-gravity is too low to amplify the waves and sustain the instability (Figure 3). This spiral mode simply evolves periodically with the orbit, strengthening at pericenter and weakening at apocenter. Disks at the high mass end, which are strongly self-gravitating, and, when isolated, undergo fragmentation into protoplanets for cooling times of order the orbital time (Mayer et al. 2003), develop a very strong and complex spiral pattern after the first pericenter passage, with a dominant two-armed mode (see Figure 1). However, such strong amplification of the spiral waves also leads to intense compressional heating along the arms. These spiral shocks increase the temperature of the outer disk by nearly a factor of 3 and bring the system towards stability (Figure 2). This is why the spiral arms fade away considerably during the second orbit (see Figure 1). The suppression of fragmentation by tidally induced shock heating was also advocated by Nelson (2000). In the intermediate mass regime, $M_d =0.05,0.08 M_{\odot}$, self-gravity is non-negligible and spiral instabilities are visibly amplified, but shock heating is mild enough for non-axisymmetric features to last longer and build up more pronounced overdensities (Figure 3). These disks can fragment for the shortest among the cooling times considered here despite the fact that they can avoid fragmentation in isolation. Such short cooling times are probably unrealistic, nevertheless it is interesting that there are cases in which tides can increase the susceptibility to fragmentation as found by Boss (1998).
=6truecm =6truecm
We can ask how realistic are the temperatures seen in our simulations. The high temperatures developing in the outer parts of the most massive disk models are comparable to those in the simulations of Nelson (2000). This author calculated the corresponding radiation flux at far-infrared and radio wavelengths (from 870 $\mu$m to 1.3 cm) assuming that the disk emits like a blackbody and obtained fluxes somewhat lower than observed in a prototype young binary protostellar disk system, L1551 IRS 5 (Bachiller, Tafalla & Cernicharo 1994). Therefore the temperatures seen in our simulations are probably a conservative estimate of those occurring in real binary systems.
Temperature in excesse of 200 K as those obtained here (Figure 3) for the most massive disk models (especially along the spiral shocks) would be enough to vaporize water ice. The latter should contribute almost half of the mass of solid material in a protoplanetary disk (Pollack et al. 1994), and is therefore a fundamental building block of large solid grains and, ultimately, of planetesimals. A reduced growth of rocky planetary embryos could result, and therefore giant planet formation by core accretion could also be less likely in such binary systems relative to isolated system (see also Nelson 2000). Light or intermediate mass disks, instead, maintain outer disk temperatures lower than 100 K between 10 and 20 AU (Figure 2) posing no problem for core accretion.
Many binary systems comprise stars with unequal masses. We simulated the interaction between two disk+star systems with masses differing by a factor of 2 (run Rbm2, see Figure 3). The most massive of the two disks ($0.1 M_{\odot}$), which never fragmented when interacting with an equally massive disk (e.g. run RB1a, RB1b, RB1c), now manages to produce two clumps of roughly one Jupiter mass. Yet these clumps are quickly dissolved as the pressure still overcomes self-gravity. The lighter disk forms strong multi-armed spirals, probably resulting from strong swing amplification (as expected given the lower mass of the central star, see MA04). Although the temperature remains quite low ($Q$ drops close to unity locally), nascent overdensities are apparently sheared away before they are able to fragment because of the strong tidal field of the more massive companion. We note that a $0.1 M_{\odot}$ disk does not produce permanent clumps both while interacting with an identical disk (e.g. run RB1a) and when orbiting a companion system half of its mass as in run RBm2. Coversely, in run RB3b a disk only 20% lighter (see Table 1) does give rise to long-lived clumps while interacting with a system of identical mass (all these runs employ the same cooling times). This suggests that tidally induced spiral shocks become too strong above some critical value of the disk self-gravity, and heating becomes dominant, despite the fact that the strength of the external perturbation changes significantly.
Among the few known binary systems with planets the majority have stars with a projected separation above 100 AU. In runs RBw(a,b,c) we evolved the massive disk models on orbits with a larger semi-major axis, $a = 58$ AU. In all these runs we wintnessed fragmentation (Figure 3), although in run RBwc, in which the cooling time was the longest, clumps did not survive for more than $\sim 3$ disk orbital times. We conclude that at these larger orbital separations disk instability proceeds similarly to the case of isolated disks because tidal forces are considerably weaker. Moreover, the temperatures at such separations are low enough to guarantee the survival of ice grains (Figure 1).
=8truecm
Summary and Discussion
======================
We have shown that fragmentation by disk instability is suppressed in binary systems with orbital separations around 60 AU. This happens because shock heating overwhelms cooling and damps any overdensity. Shock heating is stronger than in isolated disks because the tidally forced spiral arms reach a much greater amplitude relative to spiral arms in isolated disks. This is especially true in the most massive disks since their higher self-gravity amplifies the spiral arms more efficiently. The role of shock heating in disk instability has been already recognized as crucial for understanding if giant planets can form by disk instability (Pickett et al. 2000, 2003). In systems with a separation of 120 AU, disk temperatures remain quite low and fragmentation proceeds more similarly to the isolated disks. The high temperatures ($> 200$ K) developing in massive binary disk systems with separations less than $60$ AU make it hard to form giant planets even by disk core accretion. Intermediate mass systems are those in which both mechanisms are possible if cooling is very efficient whereas in binary disks with small masses, comparable to that of the minimum mass solar nebula model, core-accretion is the only viable mechanism. Models of the core accretion mechanism used to require a disk 3-4 times more massive than the minimum solar nebula in order to form Jupiter in less than 10 million years (Lissauer 1993). However, more recent models that account for orbital migration of rocky cores find formation timescales of a few million years even in a minimum mass solar nebula since the cores feed more efficiently with planetesimals as they migrate in the disk (Rice & Armitage 2003; Alibert, Mordasini & Benz 2004). If core accretion can take place in light disks then giant planets could form regardless of the presence or distance of a companion. This suggests that binarity can be used to probe planet formation models. If the new surveys aimed at quantifying the relative number of giant planets in single and binary systems will find no trend with binary separation disk instability cannot be the main formation mechanism. The opposite might be true if such trend emerges.
Our analysis was restricted to just one type of orbital configuration. Because of the low orbital eccentricity, the external perturbation is nearly continous in amplitude. This could favour high temperatures through nearly continous compressional heating. Although Nelson (2000) did not find any remarkable difference in systems with orbital eccentricities varying by a factor of 3, impulsive tidal perturbations, caused for example by a close fly by of a star or brown dwarf, which would be common in highly dynamical star formation scenarios (e.g. Bate et al. 2002), could produce a different outcome. A strong shock would occur in this case but over time compressional heating would be much lower. We will invesigate such situations in a forthcoming paper. We will also consider a larger variety of initial orbital configurations, for example non-coplanar disks possibly resulting from a capture event. The study of the geometry and relative orbits of debris disks around young binaries will be needed to find out to what extent the simple orbital configurations used in this paper are really representative.
We thank Stephane Udry, Anne Eggenberger and Michel Mayor for stimulating discussions. L.M. thanks the Swiss National Science Foundation for the financial support. T.Q. is supported by the NASA Astrobiology Institute. The simulations were performed on the Zbox supercomputer at the University of Zurich and on LeMieux at the Pittsburgh Supercomputing Center.
Alibert, Y., Mordasini, C., & Benz, W., 2004, A&A, 417, L25
Bachiller, R., Tafalla, M., & Cernichiaro, J., 1994, ApJ, 425, L93
Bate, M.R., 2000, MNRAS, 314, 33
Bate, M.R., Bonnell. I.A., & Bromm, V., 2002, MNRAS, 336, 705
Bate, M.R., Lubow, S.H., Ogilvie, G.I., & Miller, K.A., 2003, MNRAS, in press
Boss, A.P., 1986, ApJS, 62, 519
Boss, A.P. 1997, Science, 276, 1836
Boss, A.P., 1998, DPS meeting, Bulletin of the American Astronomical Society, Vol. 30, p.1057
Boss, A.P., 2002, ApJ, 576, 462
Boss, A.P., 2003, DPS meeting 35, 25.10
Burkert, A., Bate, M.R., & Bodenheimer, P., 1997, MNRAS, 289, 497
D’Alessio, P., Calvet, N., & Hartmann, L., 2001, ApJ, 553, 321
Duquennoy, A., & Mayor, M., 1991, A&A, 248, 485
Eggenberger, A., Udry, S. & Mayor, M., 2004, A&A, in press,astro-ph/0402664
Johnson, B., & Gammie, C.F., 2003, ApJ, 597, 131
Haisch, K. R. jr., Lada, E. A., & Lada, C.J., 2001, AJ, 121, 2065
Horton, A.J., Bate, M.R., & Bonnell, I.A., 2001, MNRAS, 321, 585
Lissauer, J.J., 1993, ARA&A, 31, 129
Mayer, L., Quinn, T., Wadsley, J., & Stadel, J.J., 2002, Science, 298, 1756
Mayer, L., Wadsley, J., Quinn, T. & Stadel, J.J., 2003, Proceedings of the IAP Colloquium “Frontiers on Extrasolar Planets”, in press
Mayer, L., Quinn, T., Wadsley, J., & Stadel, J.J., 2004, ApJ, in press (July 2004 issue)
Mayor, M., & Queloz, D., 1995, Nature, 378, 355
Meija, A.C., Durisen, R.H., & Pickett, B.K., 2003, in ASP Conf. Ser. 294, Scientific Frontiers in
Meija, A.C., 2004, PhD thesis
Nelson, A.F., 2000, ApJ, 537
Nelson, A.F., & Benz, W., submitted to MNRAS, astro-ph/0301619
Nelson, R.P., Papaloizou, J.C.B., Masset, F., & Kley, W. 2000, MNRAS, 318, 18
Patience, J. et al., 2002, ApJ, 581, 654
Pickett, B.K., Durisen, R.H., Cassen, P., & Mejia, A.C. 2000, ApJ, 540, L95
Pickett, B.K., Meija, A., Durisen, R.H., Cassen, P.M., Berry, D.K.,& Link, R.P., 2003, ApJ, 590, 1060
Pollack, J.B., et al., 1994, ApJ, 421, 615.
Rice, W.K.M., Armitage, P.J., Bate, M.R., & Bonnell, I.A., 2003a,MNRAS, 339, 1025
Rice, W.K.M., Armitage, P.J.,Bonnell, I.A, Bate, M.R., Jeffers, S.V., & Vine, S.G., 2003b, MNRAS, 346, L36
Rice, W.K.M., & Armitage, P.J., 2003, ApJ, 598, L55
Udry, S., Eggenberger, A., Beuzit, J.L., Lagrange, A.M., Mayor, M., & Chauvin, G., 2004, Re. Mex. Astron. Astrofis., in press
Wadsley, J., Stadel, J., & Quinn, T.R., New Astronomy, 9, 137
Whitworth, A.P., Chapman, S.J., Bhattal. A.S., Disney, M.J., Pongracic, H., & Turner, J., 1995, MNRAS, 277, 727
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We consider the simple exclusion process with $k$ particles on a segment of length $N$ performing random walks with transition $p>1/2$ to the right and $q=1-p$ to the left. We focus on the case where the asymmetry in the jump rates $b=p-q>0$ vanishes in the limit when $N$ and $k$ tend to infinity, and obtain sharp asymptotics for the mixing times of this sequence of Markov chains in the two cases where the asymmetry is either much larger or much smaller than $(\log k)/N$. We show that in the former case ($b \gg (\log k)/N$), the mixing time corresponds to the time needed to reach macroscopic equilibrium, like for the strongly asymmetric (i.e. constant $b$) case studied in [@LabLac16], while the latter case ($b \ll (\log k)/N$) macroscopic equilibrium is not sufficient for mixing and one must wait till local fluctuations equilibrate, similarly to what happens in the symmetric case worked out in [@Lac16]. In both cases, convergence to equilibrium is abrupt: we have a cutoff phenomenon for the total-variation distance. We present a conjecture for the remaining regime when the asymmetry is of order $(\log k) / N$.
[**MSC 2010 subject classifications**]{}: Primary 60J27; Secondary 37A25, 82C22.\
[**Keywords**]{}: [*Exclusion process; WASEP; Mixing time; Cutoff.*]{}
address:
- 'Université Paris-Dauphine, PSL Research University, Ceremade, CNRS, 75775 Paris Cedex 16, France.'
- 'IMPA, Estrada Dona Castorina 110, Rio de Janeiro, Brasil.'
author:
- Cyril Labbé
- Hubert Lacoin
bibliography:
- 'library.bib'
title: |
Mixing time and cutoff for the weakly asymmetric\
simple exclusion process
---
Introduction
============
The simple exclusion process is a model of statistical mechanics that provides a simplified picture for a gas of interacting particles. Particles move on a lattice, each of them performing a nearest neighbor random walk independently of the others, and interact only via the exclusion rule that prevents any two particles from sharing the same site (when a particle tries to jump on a site which is already occupied, this jump is cancelled).
In spite of its simplicity, this model displays a very rich behavior and has given rise to a rich literature both in theoretical physics and mathematics, see for instance [@KipLan; @Liggett] and references therein.
In the present paper, we study relaxation to equilibrium for a particular instance of the simple exclusion process in which the lattice is a segment of length $N$ and particles feel a bias towards the right that vanishes when $N$ tends to infinity. This setup is often referred to as the Weakly Asymmetric Simple Exclusion Process (WASEP): it interpolates between the symmetric case (SSEP) and the one with a positive constant bias (ASEP).
While convergence to equilibrium for a particle system can be considered on a macroscopic scale via the evolution of the particle density or hydrodynamic profile (see e.g. [@KipLan] and references therein), an alternative and complementary viewpoint (when the system is of finite size) consists in measuring the so-called ${\varepsilon}$-Total Variation Mixing Time [@LevPerWil]. It is defined as the first time at which the total variation distance to the stationary state, starting from the “worst" initial condition, falls below a given threshold $\epsilon$. Compared to the hydrodynamic profile, this provides a much more microscopic information on the particle system.
The problem of mixing time of the simple exclusion process on the segment has been extensively studied both in the symmetric [@Wil04; @Lac16; @Lac162] and the asymmetric setup [@Benjamini; @LabLac16] and it has been proved in [@Lac16] and [@LabLac16] respectively that in both cases, the worst case total variation distance drops abruptly from its maximal value $1$ to $0$, so that the mixing time does not depend at first order on the choice of the threshold ${\varepsilon}$ - a phenomenon known as cutoff and conjectured to hold for a large class of Markov chains as soon as the mixing time is of a larger order than the relaxation time (which is defined as the inverse of the spectral gap of the generator).
However the patterns of convergence to equilibrium in the symmetric and asymmetric cases are very different. Let us for simplicity focus on the case with a density of particles $k=\alpha N$, $\alpha\in (0,1)$. In the symmetric case, the time scale associated with the hydrodynamic profile is $N^2$ and the limit is given by the heat equation [@KOV] (which takes an infinite time to relax to its equilibrium profile which is flat) and microscopic mixing occurs on a larger time scale $N^2\log N$.
In the asymmetric setup the hydrodynamic limit is given by the inviscid Burgers’ equation with a shorter time scale $N$ [@Reza] (see also [@LabbeKPZ; @LabLac16] for adaptations of this result to the segment). The equilibrium profile for this equation is reached after a finite time and in this case, the mixing time is of order $N$ and corresponds exactly to the time at which macroscopic equilibrium is attained.
The aim of this paper is to understand better the role of the asymmetry in mixing and how one interpolates between the symmetric and asymmetric regimes. This leads us to consider a model with an asymmetry that vanishes with the scale of observation, usually referred to as Weakly Asymmetric Exclusion Process (WASEP). While hydrodynamic limit [@Demasi89; @Gartner88; @KipLan] and fluctuations scaling limits [@DG91; @BG97; @LabbeKPZ] for WASEP are now well understood, much less is known about how a weak asymmetry affects the mixing time of the system.
A first step in this direction was made in [@LevPer16]. Therein the order of magnitude for the mixing time was identified for all possible intensities of vanishing bias, but with different constant for the upper and the lower bounds. Three regimes where distinguished (in the case where there is a density of particles):
- When $b_N\le 1/N$, the mixing time remains of the same order as that of the symetric case $N^2\log N$.
- When $1/N\le b_N\le (\log N)/N$, the mixing time is of order $(b_N)^{-2}\log N$.
- When $(\log N)/N\le b_N\le 1$, the mixing time is of order $(b_N)^{-1} N$.
The transition occurring around $b_N\approx N^{-1}$ is the one observed for the hydrodynamic limit: It corresponds to a crossover regime where the limit is given by a viscous Burger’s equation [@Demasi89; @Gartner88; @KipLan] which interpolates between the heat and the inviscid Burgers’ equations. The one occurring for $b_N\approx N^{-1}\log N$ is however not observed in the macrospic profile and is specific to mixing times.
In the present work, we identify the full asymptotic of the mixing time (with the right constant) when the bias is either negligible compared to, or much larger than $\log N / N$ (or $\log k/N$ when the total number of particle is not of order $N$). This implies cutoff in these two regimes. Our result and its proof provide a better understanding of the effect of asymmetry on microscopic mixing: When $b_N\gg N^{-1}\log N$, the pattern of relaxation is identical to that of the fully asymmetric case and microscopic equilibrium is reached exactly when the macroscopic profile hits its equilibrium state. When $b_N\ll N^{-1}\log N$ the pattern of relaxation resembles that of the symmetric case, the mixing time corresponds to the time needed to equilibrate local fluctuations, in particular in the case $(B)$ described above (or more precisely when $1/N\ll b_N\ll (\log N)/N$) this time does not correspond to the time needed to reach macroscopic equilibrium.
We could not prove such a result in the crossover regime $b_N\approx N^{-1}\log N$: In this case the time to reach macroscopic equilibrium and that to equilibrate local fluctuations are of the same order and the two phenomena are difficult to separate. In Section \[conjectos\] we provide a conjecture for the mixing time in this regime in the case of vanishing density. However the techniques developed here are not sufficient to obtain sharp results in this case.
Model and results
=================
Mixing time for the WASEP
-------------------------
Given $N\in \mathbb N$, $k\in {\llbracket}1,N-1 {\rrbracket}$ (we use the notation ${\llbracket}a,b{\rrbracket}=[a,b]\cap {\mathbb{Z}}$) and $p\in(1/2,1]$, the Asymmetric Simple Exclusion Process on ${\llbracket}1, N {\rrbracket}$ with $k$ particles and parameter $p$ is the random process on the state space $${\Omega}_{N,k}^0:=\Big\{ \xi\in\{0,1\}^{N} \ : \ \sum_{x=1}^N \xi(x)=k\Big\},$$ associated with the generator $$\label{defgen}
{\mathcal{L}}_{N,k} f(\xi):= \sum_{y=1}^{N-1} \left( q{\mathbf{1}}_{\{\xi(y)< \xi(y+1)\}}+p{\mathbf{1}}_{\{\xi(y)> \xi(y+1)\}}\right)(f(\xi^y)-f(\xi)),$$ where $q=1-p$ and $$\label{flipz}
\xi^y(x):=\begin{cases} \xi(y+1) \quad &\text{ if } x=y,\\
\xi(y) \quad &\text{ if } x=y+1,\\
\xi(x) \quad& \text{ if } x\notin \{y,y+1\}.
\end{cases}$$ In a more intuitive manner we can materialize the positions of $1$ by particles, and say that the particles perform random walks with jump rates $p$ to the right and $q=1-p$ to the left: These random walks are independent from one another except that any jump that would put a particle at a location already occupied by another particle is cancelled. Having in mind this particle representation, we let for $i\in {\llbracket}1, k{\rrbracket}$, $\xi_i$ denote the position of the $i$-th leftmost particle $$\xi_i:= \min\left\{ y\in {\llbracket}1,N {\rrbracket}\ : \ \sum_{x=1}^y \xi(x)=i \right\}.$$
We let $P^{N,k}_t$ denote the associated semi-group and $(\eta^\xi(t,\cdot))_{t\ge 0}$ denote the trajectory of the Markov chain starting from initial condition $\xi \in {\Omega}_{N,k}^0$. This Markov chain is irreducible, and admits a unique invariant (and reversible) probability measure $\pi_{N,k}$ given by $$\pi_{N,k}(\xi):= \frac{1}{Z_{N,k}}{\lambda}^{-A(\xi)}.$$ where $\lambda = p/q$, $Z_{N,k}:= \sum_{\xi\in {\Omega}_{N,k}^0} {\lambda}^{-A(\xi)}$, and $$\label{def:A}
A(\xi):= \sum_{i=1}^k (N-k+i-\xi_i)\ge 0$$ denotes the minimal number of moves that are necessary to go from $\xi$ to the configuration $\xi^{\min}$ where all the particles are on the right $\xi^{\min}(x):={\mathbf{1}}_{[N-k+1,N]}(x)$ (this terminology is justified by the fact that $\xi^{\min}$ is minimal for the order introduced in Section \[Sec:Prelim\]).
Recall that the total-variation distance between two probability measures defined on the same state-space ${\Omega}$ is defined by $$\|\alpha-{\beta}\|_{TV}=\sup_{A\subset {\Omega}} \alpha(A)-{\beta}(A),$$ where the $\sup$ is taken over all measurable sets $A$.
The mixing time associated to the threshold ${\varepsilon}\in (0,1)$ is defined by $${T_{\rm mix}}^{N,k}({\varepsilon}):=\inf\{ t \ge 0 \ : \ d^{N,k}(t)\le {\varepsilon}\},$$ where $d^{N,k}(t)$ denotes the total-variation distance to equilibrium at time $t$ starting from the worst possible initial condition $$\label{tvdis}
d^{N,k}(t):= \max_{\xi\in {\Omega}_{N,k}^0} \| P^{N,k}_t(\xi, \cdot)-\pi_{N,k}\|_{TV}.$$
We want to study the asymptotic behavior of the mixing time for this system when both the size of the system and the number of particles tend to infinity. A natural case to consider is when there is a non-trivial density of particles, that is $k/N\to \alpha \in (0,1)$, but we decide to also treat the boundary cases of vanishing density ($\alpha=0$) and full density ($\alpha=1$). By symmetry we can restrict to the case when $k=k_N\le N/2$: indeed, permuting the roles played by particles and empty sites boils down to reversing the direction of the asymmetry of the jump rates. Note that we will always impose $k\ge 1$ since when $k=0$ the process is trivial.
The asymptotic behavior of ${T_{\rm mix}}^{N,k}({\varepsilon})$ in the case of constant bias ($p>1/2$ is fixed when $N$ goes to infinity) has been obtained in a previous work.
We have for every ${\varepsilon}>0$, every $\alpha \in [0,1]$ and every sequence $k_N$ such that $k_N/N \to \alpha$
$$\lim_{N\to \infty}\frac{{T_{\rm mix}}^{N,k_N}({\varepsilon})}{N}= \frac{(\sqrt{\alpha}+\sqrt{1-\alpha})^2}{p-q}\;.$$
The result implies in particular that at first order, the mixing time does not depend on ${\varepsilon}\in(0,1)$, meaning that on the appropriate time-scale, for large values of $N$ the distance to equilibrium drops abruptly from $1$ to $0$. This phenomenon is referred to as *cutoff* and was first observed in the context of card shuffling [@AldDia; @DiaSha]. It is known to occur for a large variety of Markov chains, see for instance [@LevPerWil]. In the context of the exclusion process, it has been proved in [@Lac16] that cutoff holds for the Symmetric Simple Exclusion Process (SSEP) which is obtained by setting $p=1/2$ in the generator .
When $p=1/2$, for any sequence $k_N$ that tends to infinity and satisfies $k_N\le N/2$ for all $N$, we have
$$\lim_{N\to \infty}\frac{{T_{\rm mix}}^{N,k_N}({\varepsilon})}{N^2\log k_N}= \frac{1}{\pi^2}.$$
While cutoff occurs in the two cases, it appears to be triggered by different mechanisms. When $p>1/2$, the mixing time is determined by the time needed for the particle density profile to reach its macroscopic equilibrium: After rescaling time and space by $N$, the evolution of the particle density has a non-trivial scaling limit (the inviscid Burgers’ equation with zero-flux boundary conditions), which fixates at time $\frac{(\sqrt{\alpha}+\sqrt{1-\alpha})^2}{p-q}$. The first order asymptotic for the mixing time is thus determined by the time the density profile needs to reach equilibrium.
When $p=1/2$, the right-time scale to observe a macroscopic motion for the particles is $N^2$, and it is worth mentioning that the scaling limit obtained for the particle density (the heat equation) does not fixate in finite time. To reach equilibrium, however, we must wait for a longer time, of order $N^2\log N$, which is the time needed for local fluctuations in the particle density to come to equilibrium.
We are interested in studying the process when the drift tends to zero: this requires to understand the transition between these two patterns of relaxation to equilibrium. Hence we consider $p$ to be a function of $N$ which is such that the bias towards the right $b_N:=p_N-q_N=2p_N-1$ vanishes $$\label{vanishing}
\lim_{N\to \infty} b_N=0.$$ In this regime, the model is sometimes called WASEP for Weakly Asymetric Simple Exclusion Process. Its convergence to equilibrium has already been studied in[@LabbeKPZ; @LevPer16]. In [@LevPer16] the authors identify the order of magnitude of the mixing time as a function of $b_N$ in full generality. However the approach used in [@LevPer16] does not allow to find the exact asymptotic for the mixing time nor to prove cutoff, and does not answer our question concerning the pattern of relaxation to equilibrium.
Results
-------
We identify two main regimes for the pattern of relaxation to equilibrium. The *large bias* regime where $$\label{largebias}
\lim_{N\to \infty} \frac{N b_N}{(\log k_N)\vee 1}=\infty.$$ and the *small bias* regime where $$\label{smallbias}
\begin{cases}
\lim\limits_{N\to \infty} \frac{Nb_N}{\log k_N}=0,\\
\lim\limits_{N\to \infty} k_N=\infty.
\end{cases}$$ We identify the asymptotic expression for the mixing time in both regimes. In the large bias regime we show that the mixing time coincides with the time needed by the particle density to reach equilibrium like in the constant bias case.
\[Th:largebias\]
When holds, and $\lim_{N\to \infty} k_N/N=\alpha\in [0,1]$, we have for every ${\varepsilon}\in (0,1)$ $$\lim_{N\to \infty} \frac{b_N {T_{\rm mix}}^{N,k_N}({\varepsilon})}{N}= \left(\sqrt{\alpha}+\sqrt{1-\alpha}\right)^2.$$
To state our result in the small bias regime, let us introduce the quantity $$\label{thegap}
\operatorname{\mathrm{gap}}_{N}:= (\sqrt{p_N}-\sqrt{q_N})^2+ 4 \sqrt{p_Nq_N} \sin\left( \frac{\pi}{2N}\right)^2,$$ which corresponds to the spectral gap associated with the generator . Notice that it does not depend on the number $k$ of particles in the system. The pattern of relaxation is similar to the one observed in the symmetric case.
\[Th:smallbias\]
When holds, we have
$$\lim_{N\to \infty} \frac{\operatorname{\mathrm{gap}}_{N} {T_{\rm mix}}^{N,k_N}({\varepsilon})}{\log k_N}= \frac12.$$
Using Taylor expansion for $\operatorname{\mathrm{gap}}_N$ we have, whenever $b_N$ tends to zero $$\label{eq:taylorgap}
\operatorname{\mathrm{gap}}_{N}\stackrel{N\to \infty}{\sim} \frac{1}{2}\left( b^2_N+ \left(\frac{\pi}{N}\right)^2 \right).$$ Thus in particular we have $${T_{\rm mix}}^{N,k_N}({\varepsilon})\sim
\begin{cases}
\frac{\log k_N}{b_N^2} & \text{if } 1/N \ll b_N\ll \log k_N/N, \\
\frac{1}{\pi^2} N^2 \log k_N &\text{ if } 0< b_N \ll 1/N, \\
\frac{1}{\pi^2+\beta^2} N^2 \log k_N &\text{ if } b_N\sim \beta/N.
\end{cases}$$
Note that our classification of regimes - does not cover all possible choices of $b_N$. Two cases have been excluded for very different reasons:
- When $b_N=O(N^{-1})$ and $k_N$ is bounded, then we have a system of $k$ diffusive interacting random walks. This system does not exhibit cutoff and has a mixing time of order $N^2$ (The upper bound can actually be deduced from argument presented in Section \[preuv2\] and the lower bound is achieved e.g. by looking at the expectation and variance of the number of particles on right half of the segment, like what is done in [@Morris06 Section 6] ).
- When $b_N$ is of order $\log k_N/N$ the time at which the density profile reaches its equilibrium and the time needed for local fluctuations to reach their equilibrium values are of the same order and we believe that there is an interaction between the two phenomena. We provide a more detailed conjecture in Section \[conjectos\]
We have not included here results concerning the biased card shuffling considered in [@Benjamini; @LabLac16]. Let us mention that while our analysis should also yield optimal bound for the mixing time of this process when holds (i.e. ${T_{\rm mix}}^{k,N}({\varepsilon})\sim 2 N /(p_N-q_N)$), it seems much more difficult to prove the equivalent of Theorem \[Th:smallbias\]. The main reason is that the coupling presented in Section \[Appendix:Coupling\] cannot be extended to a coupling on the permutation process. Building on and adapting the techniques presented in [@Lac16 Section 5] it should a priori be possible to obtain a result concerning the mixing time starting from an extremal condition (the identity or its symmetric), but this is out of the scope of the present paper.
Conjecture in the regime $b_N\asymp \log k_N/N$ {#conjectos}
-----------------------------------------------
Let us here formulate, and heuristically support a conjecture concerning the mixing-time in the crossover regime where $$\label{limbeta}
\lim_{N\to \infty} \frac{b_N N}{\log k_N}={\beta}.$$ for some ${\beta}\in(0,\infty)$. For the ease of exposition, while it should be in principle possible to extend the heuristic to the case of positive density (see Remark \[pluscomplique\] below) we restrict ourselves to the case $\lim_{N\to \infty }k_N/N=0$. The justification we provide for the conjecture might be better understood after a first reading of the entire paper.
\[jecture\] When $b_N$ and $k_N$ display the asymptotic behavior given by , we have for every ${\varepsilon}>0$ $$\lim_{N \to \infty} \frac{{T_{\rm mix}}^{N,k_N}({\varepsilon}) \log k_N}{N^2}= \begin{cases} \frac{2}{{\beta}}+\frac{1}{{\beta}^2}, \quad &\text{ if } {\beta}\le 1/2,\\
\left( \frac{\sqrt{2}+2\sqrt{{\beta}}}{2{\beta}} \right)^2, \quad &\text{ if } {\beta}\ge 1/2.
\end{cases}$$
To motivate this conjecture let us first describe the equilibrium measure and its dependence on ${\beta}$. As we are in the low-density regime, the equilibrium measure is quite close to the product measure one would obtain for the system without exclusion rules: the $k$ particles are approximately IID distributed with the distance from the right extremity being a geometric of parameter $q_N/p_N\approx e^{- \frac{2{\beta}\log k}{N}}$.
Hence the probability of having a particle at site $\lfloor zN \rfloor$ for $z\in [0,1]$ is roughly of order $k^{[1-2{\beta}(1-z)]+o(1)}/N$. Thus, while particles are concentrated near the right extremity at equilibrium, the equilibrium “logarithmic density” of particles exhibit a non trivial profile in the sense that for any $z> 1-(2{\beta})^{-1}$ we have $$\label{liquib}
\lim_{{\varepsilon}\to 0}\lim_{N\to \infty}\frac{\log \left( \sum_{i= (z-{\varepsilon})N}^{(z+{\varepsilon})N} \xi(i) \right) }{\log k_N} \Rightarrow 1-2{\beta}(1-z),$$ where the convergence holds in probability under the equilibrium measure $\pi_{N,k}$ when $N$ tends to infinity and ${\varepsilon}$ tends to zero in that order. The typical distance to zero of the left-most particle at equilibrium is also given by this profile in the sense that it is typically $o(N)$ when ${\beta}\le 1/2$ and of order $N(1-\frac 1 {2{\beta}})$ when ${\beta}\ge 1/2$. While we only give heuristic justification for these statements concerning equilibrium, it is worth mentioning that they can be made rigorous by using the techniques exposed in Section \[Sec:Prelim\].
To estimate the mixing time, we assume that the system gets close to equilibrium once the number of particles on any “mesoscopic” interval of the form $[ (z-{\varepsilon})N, (z+{\varepsilon})N ]$ is close to its equilibrium value. While the mean number of particle is of order $k^{1-2{\beta}(1-z)}$ (cf. ), the typical equilibrium fluctuation around this number should be the given by the square root due to near-independence of different particles and thus be equal to $k^{\frac{1}{2}-{\beta}(1-z)}$.
To estimate the surplus of particles in this interval at time $t N^2 (\log k)^{-1}$ for $t> 1/{\beta}$ (note that $t=1/{\beta}$ is the time of macroscopic equilibrium where most particles are packed on the right), we consider the number of particles that end up there after keeping a constant drift of order $z (\log k) /(N t)$, which is smaller than $b_N$. Neglecting interaction between particles and making a Brownian approximation for the random walk with drift, we obtain that the expected number of particles following this strategy is given by $$k \exp\left(- \log k \frac{({\beta}t-z)^2}{2 t} \right)=k^{1-\frac{({\beta}t-z)^2}{2 t}}.$$
Hence equilibrium should be attained when this becomes negligible with respect to the typical fluctuation $k^{\frac{1}{2}-{\beta}(1-z)}$ for all values of $z$ where we find particles at equilibrium. That is, when the inequality $$\label{fluqueton}
1-\frac{({\beta}t-z)^2}{2 t}< \frac{1}{2}-{\beta}(1-z),$$ is valid for all $z\in[0,1]$ if ${\beta}\le 1/2$ or for all $z\ge 1-\frac{1}{2{\beta}}$ if $\beta\ge 1/2$. A rapid computation show that one only needs to satisfy the condition for the smallest value of $z$ (either $1$ or $1-\frac{1}{2{\beta}}$), which boils down to finding the roots of a degree two polynomial. This yields that we must have $t>t_0$ where $$t_0:=\begin{cases}
\frac{2}{{\beta}}+\frac{1}{{\beta}^2} &\text{ if } {\beta}\le 1/2,\\
\left( \frac{\sqrt{2}+2\sqrt{{\beta}}}{2{\beta}} \right)^2 &\text{ if } {\beta}\ge 1/2.
\end{cases}$$
\[pluscomplique\] Describing the equilibrium “logarithmic profile” of particles when the system has positive density is also possible (note that on the right of $(1-\alpha) N$ it is the density of empty-sites that becomes the quantity of interest). It is thus reasonable to extend the heuristic to that case. However the best strategy to produce a surplus of particle in that case becomes more involved, as the zones with positive density of particles, which are described by the hydrodynamic evolution given in Proposition \[prop:lidro\], play a role in the optimization procedure. For this reason we did not wish to bring the speculation one step further.
Organization of the paper
-------------------------
In the remainder of the paper we drop the subscript $N$ in $k_N$ in order to simplify the notation. The article is organised as follows. In Section \[Sec:Prelim\], we introduce the representation through height functions and collect a few results on the invariant measure, the spectral gap and the hydrodynamic limit of the process. In Sections \[Sec:LBLB\] and \[Sec:UBLB\], we consider the large bias case and prove respectively the lower and upper bounds of Theorem \[Th:largebias\]: While the lower bound essentially follows from the hydrodynamic limit, the upper bound is more involved and is one of the main achievement of this paper. In Sections \[Sec:LBSB\] and \[Sec:UBSB\], we deal with the small bias case and prove respectively the lower and upper bounds of Theorem \[Th:smallbias\]. Here again, the lower bound is relatively short and follows from similar argument as those presented by Wilson [@Wil04] in the symmetric case, while the upper bound relies on a careful analysis of the area between the processes starting from the maximal and minimal configurations and under some grand coupling.
Preliminaries and technical estimates {#Sec:Prelim}
=====================================
Height function ordering and grand coupling
-------------------------------------------
To any configuration of particles $\xi\in\Omega_{N,k}^0$, we can associate a so-called height function $h=h(\xi)$ defined by $h(\xi)(0) = 0$ and $$h(\xi)(x) = \sum_{y=1}^x \big(2\xi(y) - 1\big)\;,\quad x\in {\llbracket}1,N{\rrbracket}\;.$$ For simplicity, we often abbreviate this in $h(x)$. The height function is a lattice path that increases by $1$ from $\ell-1$ to $\ell$ if there is a particle at site $\ell$, and decreases by $1$ otherwise. Its terminal value therefore only depends on $k$ and $N$. The set of height functions obtained from $\Omega_{N,k}^0$ through the above map is denoted $\Omega_{N,k}$.\
The particle dynamics can easily be rephrased in terms of height functions: every upward corner ($h(x)=h(x-1)+1=h(x+1)+1$) flips into a downward corner ($h(x)=h(x-1)-1=h(x+1)-1$) at rate $p$, while the opposite occurs at rate $q$. We denote by $(h^\zeta(t,\cdot),t\ge 0)$ the associated Markov process starting from some initial configuration $\zeta \in \Omega_{N,k}$.\
It will be convenient to denote by $\wedge$ the maximal height function: $$\wedge(x) = x \wedge (2k-x)\;,\quad x\in{\llbracket}1,N{\rrbracket}\;,$$ and by $\vee$ the minimal height function: $$\vee(x) = (-x)\vee(x-2N+2k)\;,\quad x\in{\llbracket}1,N{\rrbracket}\;.$$ Though the dependence on $k$ is implicit in the notations $\wedge,\vee$, this will never raise any confusion as the value $k$ will be clear from the context.
It is possible to construct simultaneously on a same probability space and in a Markovian fashion, the height function processes $(h_t^\zeta,t\ge 0)$ starting from all initial conditions $\zeta\in\cup_k \Omega_{N,k}$ and such that the following monotonicity property is satisfied for all $k$ and all $\zeta,\zeta' \in \Omega_{N,k}$: $$\label{eq:grand}
\zeta \le \zeta' \Rightarrow h_t^\zeta \le h_t^{\zeta'}\;,\quad \forall t\ge 0\;.$$ Here, $\zeta\le \zeta'$ simply means $\zeta(x) \le \zeta'(x)$ for all $x\in{\llbracket}0,N{\rrbracket}$. We call such a construction a monotone Markovian grand coupling, and we denote by ${\mathbb{P}}$ the corresponding probability distribution. The existence of such a grand coupling is classical, see for instance [@LabLac16 Proposition 4]. In a portion of our proof, we require to use a specific grand coupling which is not the one displayed in [@LabLac16] and for this reason we provide an explicit construction in Appendix \[Appendix:Coupling\].
Once a coupling is specified, by enlarging our probability space, one can also define the process $h^{\pi}_t$ which is started from an initial condition sampled from the equilibrium measure $\pi_{N,k}$, independently of $h_t^{\zeta}, \zeta\in {\Omega}_{N,k}$.
Let us end up this section introducing the (less canonical) notation $$\label{eq:strict}
\zeta<\zeta' \quad \Leftrightarrow \quad \left( \zeta\le \zeta' \text{ and } \zeta\ne \zeta' \right).$$ We say that a function $f$ on $\Omega_{N,k}$ is increasing (strictly) if $f(\zeta)\le f(\zeta')$ ($f(\zeta)< f(\zeta')$) whenever $\zeta<\zeta'$. The minimal increment of an increasing function is defined by $$\label{eq:minincr}
\delta_{\min}(f)=\min_{\zeta,\zeta'\in {\Omega}_{N,k}, \zeta<\zeta'} f(\zeta')-f(\zeta).$$
The equilibrium measure in the large-bias case
----------------------------------------------
For $\xi\in {\Omega}_{N,k}^0$ we set $$\label{def1:lknrkn}
\begin{split}
\ell_{N}(\xi)&=\min\{ x\in {\llbracket}1, N{\rrbracket}\ : \ \xi(x)= 1 \},\\
r_{N}(\xi)&=\max\{ x\in {\llbracket}1, N{\rrbracket}\ : \ \xi(x)= 0 \}.
\end{split}$$
A useful observation on the invariant measure is the following. Given $\xi$, we define $\chi(\xi)$ as the sequence of particle spacings: $$\chi_i:= \xi_{i+1}-\xi_i\;,\; \text{ for } i\in {\llbracket}1,k-1{\rrbracket}, \quad \chi_k=N+1-\xi_k.$$ From , under $\pi_{N,k}$ the probability of a given configuration is proportional to $$\label{eq:weights}
{\lambda}^{-\chi_1} {\lambda}^{-2\chi_2} \ldots {\lambda}^{-k \chi_k},$$ In other terms, under the invariant measure the particle spacings $(\chi_i)_{1\le i \le k}$ are distributed like independent geometric variables, with respective parameters ${\lambda}^{-i}$, conditioned to the event $\sum_{i=1}^k \chi_k\le N$.
\[lem:lbeq\] When holds we have for any ${\varepsilon}>0$ $$\begin{split}
\lim_{N\to \infty}& \pi_{N,k}( \ell_N \le (N-k)-{\varepsilon}N)=0,\\
\lim_{N\to \infty}& \pi_{N,k}( r_N \ge (N-k)+{\varepsilon}N)=0,
\end{split}$$
By symmetry it is sufficient to prove the result for $\ell_N$ only, but for all $k\in{\llbracket}1,N-1{\rrbracket}$. Note that there is nothing to prove regarding $\ell_N$ if $\alpha:=\lim_{N\to \infty} k/N =1$, so we assume that $\alpha\in[0,1)$.\
Let $(X_i)_{1\le i \le k}$ be independent geometric variables, with respective parameters ${\lambda}^{-i}$. The sum of their means satisfies (recall that ${\lambda}-1$ is of order $b_N$) $$\begin{aligned}
\sum_{i=1}^k \frac{1}{1-{\lambda}^{-i}} = k+ \sum_{i=1}^k \frac{{\lambda}^{-i}}{1-{\lambda}^{-i}}\le k+ C b_N^{-1} \log \min( k, b_N^{-1})\;,\end{aligned}$$ for some constant $C>0$. The large bias assumption ensures that $b_N^{-1} \log \min( k_N, b_N^{-1})=o(N)$. Hence using the Markov inequality, we obtain that if $(X_i)_{1\le i \le k}$ is a sequence of such geometric variables, and if is satisfied, then for any ${\varepsilon}>0$ $${\mathbb{P}}\Big(\sum_{i=1}^k X_i\ge k+{\varepsilon}N\Big) \le \frac{{\mathbb{E}}\big[\big|\sum_{i=1}^k X_i - k\big|\big]}{{\varepsilon}N}=\frac{{\mathbb{E}}\big[\sum_{i=1}^k X_i - k\big]}{{\varepsilon}N}\;,$$ so that $$\lim_{N\to\infty} {\mathbb{P}}\Big(\sum_{i=1}^k X_i\ge k+{\varepsilon}N\Big) = 0\;.$$ The above inequality for ${\varepsilon}<1-\alpha$ implies that ${\mathbb{P}}(\sum_{i=1}^k X_i\le N)\ge 1/2$ for all $N$ large enough meaning that the conditioning only changes the probability by a factor at most $2$. Then, we can conclude by noticing that $\ell_N=\xi_1= N-\sum_{i=1}^k \chi_k$.
The equilibrium measure in the small-bias case {#Subsec:eqSmallBias}
----------------------------------------------
We aim at showing that with large probability the density of particles everywhere is of order $k^{1+o(1)}/N$. Given $\xi\in {\Omega}^0_{N,k}$ we let $Q_1(\xi)$, resp. $Q_2(\xi)$, denote the largest gap between two consecutive particles, resp. between two consecutive empty sites. $$\begin{split}
Q_1(\xi)&:=\max\{ n\ge 1 \ : \ \exists i\in {\llbracket}0,N-n{\rrbracket}, \ \forall x\in {\llbracket}i+1,i+n{\rrbracket}, \xi(x)=0 \},\\
Q_2(\xi)&:=\max\{ n\ge 1 \ : \ \exists i\in {\llbracket}0,N-n{\rrbracket}, \ \forall x\in {\llbracket}i+1,i+n{\rrbracket}, \xi(x)=1 \},
\end{split}$$ and $Q(\xi)=\max(Q_1(\xi),Q_2(\xi))$.
\[lem:dens\] For all $x\in {\llbracket}1,N{\rrbracket}$, we have $$\label{encadr}
\frac{k}{N} {\lambda}^{x-N} \le \pi_{N,k}(\xi(x) = 1) \le \frac{k}{N} {\lambda}^{x-1}\;.$$ Furthermore, there exists a constant $c>0$ such that for all choices of $N\ge 1$, $u>1$ and $p_N\in (1/2,1]$ and all $k\le N/2$ $$\label{splam}
\pi_{N,k}\left(Q(\xi)\ge \frac{{\lambda}^{N} N u}{k} \right) \le 2k e^{-cu}\;.$$
We set $A_x:= \{\xi(x) = 1\}$, we first prove that for all $y\in {\llbracket}1,N-1{\rrbracket}$ we have $$\label{swiz}
\pi_{N,k}(A_y) \le \pi_{N,k}(A_{y+1}) \le {\lambda}\pi_{N,k}(A_y).$$ We observe that the map $\xi \mapsto \xi^y$ defined in induces a bijection from $A_y$ to $A_{y+1}$ and that for every $\xi\in A^y$, $$\pi_{N,k}(\xi) \le \pi_{N,k}(\xi^y)\le {\lambda}\pi_{N,k}(\xi).$$ The reader can check indeed that $\pi_{N,k}(\xi^y)= {\lambda}\pi_{N,k}(\xi)$ if $\xi\in A_y \setminus A_{y+1}$ and that $\xi^y=\xi$ if $\xi\in A_y \cap A_{y+1}$. The desired inequality is then obtained by summing over $\xi\in A_y$.
By iterating we obtain $${\lambda}^{x-N}\pi_{N,k}(A_1) \le \pi_{N,k}(A_{x}) \le {\lambda}^{x-1} \pi_{N,k}(A_1).$$ By monotony of $\pi_{N,k}(A_y)$ in $y$ and the fact that there are $k$ particles $$N\pi_{N,k}(A_1) \le \sum_{y=1}^N \pi_{N,k}(A_y)=k \le N \pi_{N,k}(A_N),$$ and thus can be deduced.
We pass to the proof of . We can perform the same reasoning as above but limiting ourselves to configurations with no particles in some set $I\subset {\llbracket}1, N{\rrbracket}$. Setting $B_I:=\{ \forall y\in I, \xi(y)=0 \}$ we obtain similarly to (exchanging directly the content of $x$ and $y$ instead of nearest neighbors) that for every $x,y \in {\llbracket}1, N{\rrbracket}\setminus I$ with $x< y$ $$\pi_{N,k}(A_x \cap B_I ) \le \pi_{N,k}(A_y \cap B_I ) \le {\lambda}^{y-x} \pi_{N,k}(A_x \cap B_I ).$$ This allows to deduce that $$\pi_{N,k}(\xi(x)=1 \ | \ \forall y\in I, \xi(y)=0) \ge \frac{k}{N-|I|}{\lambda}^{x-N},$$ and yields by induction $$\pi_{N,k}(\forall x\in I, \ \xi(x)=0)\le \left(1-{\lambda}^{-N} \frac{k}{N}\right)^{|I|} \le \exp\left(-|I| {\lambda}^{-N} \frac{k}{N}\right).$$ Then noticing that $\{Q_1(\xi)\ge 2m\}$ implies that an interval of the type ${\llbracket}mi+1, m(i+1){\rrbracket}$ is empty, a union bound yields that $$\pi_{N,k}(Q_1(\xi)\ge 2m)\le \left\lfloor \frac{N}{m}\right\rfloor \exp\left(-m{\lambda}^{-N} \frac{k}{N}\right).$$ This remains true for $Q_2(\xi)$ upon replacing $k$ by $N-k$, and this concludes the proof of if one choses $m=\frac{{\lambda}^{N} N u}{2k}$ .
Eigenfunctions and contractions {#sec:eigen}
-------------------------------
The exact expression of the principal eigenfunction / eigenvalue has been derived in previous works [@LevPer16; @LabLac16]. It turns out that it can be obtained by applying a discrete Hopf-Cole transform to the generator of our Markov chain. Let us recall some identities in that direction as they will be needed later on; the details can be found in [@LabLac16 Section 3.3]. We set $$\label{Eq:rho} \varrho := \big(\sqrt p - \sqrt q\big)^2 \sim \frac{b_N^2}{2}\;,$$ and we let $a_{N,k}$ be the unique solution of $$\begin{cases}
(\sqrt{pq}\, {\Delta}-\varrho)a(x)=0\;,\quad x\in {\llbracket}1,N-1 {\rrbracket}\;,\\
a(0)=1\;,\quad a(N)= {\lambda}^{\frac{2k-N}{2}} \;,
\end{cases}$$ where ${\Delta}$ denotes the discrete Laplace operator $$\label{laplace}
{\Delta}(f)(x)=f(x+1)+f(x-1)-2f(x), \quad x\in {\llbracket}1,N-1 {\rrbracket}.$$ If $(h^\zeta_t,t\ge 0)$ denotes the height function process starting from some arbitrary initial condition $\zeta\in\Omega_{N,k}$, then the map $$V(t,x) := {\mathbb{E}}[\lambda^{\frac12 h^{\zeta}_t(x)} - a_{N,k}(x)]\;,\quad t\ge 0\;,\quad x\in {\llbracket}0,N{\rrbracket}\;,$$ solves $$\label{Eq:V}
\begin{cases}
\partial_t V(t,x)= (\sqrt{pq}\,{\Delta}- \varrho) V(t,x)\;,\quad x\in{\llbracket}1,N-1 {\rrbracket}\;.\\
V(t,0) = V(t,N) = 0\;.
\end{cases}$$ This allows to identify $N-1$ eigenvalues and eigenfunctions of the generator ${\mathcal{L}}_{N,k}$ of the Markov chain: for every $j\in\{1,\ldots,N-1\}$, the map $$\label{eq:fj}
f^{(j)}_{N,k}(\zeta) = \sum_{x=1}^{N-1} \sin \left(\frac{j x\pi}{N}\right) \left( \frac{\lambda^{\frac12 \zeta(x)} - a_{N,k}(x)}{\lambda - 1} \right)\;,$$ defines an eigenfunction with eigenvalue $$-\gamma_j = -\varrho - 4\sqrt{p_Nq_N} \sin\left( \frac{j \pi}{2N}\right)^2\;.$$ The eigenvalue $\gamma_1$ corresponds to the spectral gap of the generator (this is related to the fact that the corresponding eigenfunction is monotone, see [@LabLac16 Section 3.3] for more details), and for this reason we adopt the notation $$\operatorname{\mathrm{gap}}_N :=\gamma_1= \varrho + 4\sqrt{p_Nq_N} \sin\left( \frac{\pi}{2N}\right)^2.$$ We also set $f_{N,k}:=f^{(1)}_{N,k}(\zeta)$ for the corresponding eigenfunction. Notice that this is a *strictly* increasing function (recall ). An immediate useful consequence of the eigenvalue equation is that $$\label{eq:contract1}
{\mathbb{E}}[f_{N,k}(h^{\zeta'}_t)-f_{N,k}(h^\zeta_t)]= e^{-\operatorname{\mathrm{gap}}_N t} \left( f_{N,k}(\zeta')-f_{N,k}(\zeta) \right).$$
To close this section, let us introduce another function which is not an eigenfunction, but is also strictly increasing and enjoys a similar contraction property $$f^{(0)}_{N,k}(\zeta):=\sum_{x=1}^{N-1} \frac{{\lambda}^{\zeta(x)/2}-a_{N,k}(x)}{{\lambda}-1}.$$ As a direct consequence of at time zero, we have (using the notation introduced in ) $$\begin{gathered}
({\mathcal{L}}_{N,k}f^{(0)}_{N,k})(\zeta)= -\varrho f^{(0)}_{N,k}(\zeta)+\sqrt{pq}\sum_{x=1}^{N-1} \frac{{\Delta}({\lambda}^{\zeta/2}-a_{N,k})(x)}{{\lambda}-1}
\\=-\varrho f^{(0)}_{N,k}(\zeta)-\frac{\sqrt{pq}}{{\lambda}-1}\left[ {\lambda}^{\frac{\zeta(N-1)}2}+{\lambda}^{\frac{\zeta(1)}{2}}-a_{N,k}(N-1)-a_{N,k}(1)
\right].\end{gathered}$$ In particular, we obtain for $\zeta \le \zeta'$ $$\begin{aligned}
&({\mathcal{L}}_{N,k}f^{(0)}_{N,k})(\zeta')- ({\mathcal{L}}_{N,k}f^{(0)}_{N,k})(\zeta)\\&=
-\varrho \left( f^{(0)}_{N,k}(\zeta')-f^{(0)}_{N,k}(\zeta) \right)- \frac{\sqrt{pq}}{{\lambda}-1} \left[{\lambda}^{\frac{\zeta'(N-1)}2}+{\lambda}^{\frac{\zeta'(1)}2}- {\lambda}^{\frac{\zeta(N-1)}2}-
{\lambda}^{\frac{\zeta(1)}2}\right]\\
&\le -\varrho \left( f^{(0)}_{N,k}(\zeta')-f^{(0)}_{N,k}(\zeta) \right).\end{aligned}$$ Considering a monotone coupling between $(h^{\zeta'}_t)_{t\ge 0}$ and $(h^{\zeta}_t)_{t\ge 0}$, we obtain that $$\begin{gathered}
\partial_t {\mathbb{E}}[f^{(0)}_{N,k}(h^{\zeta'}_t)-f^{(0)}_{N,k}(h^\zeta_t)]=
{\mathbb{E}}[({\mathcal{L}}_{N,k}f^{(0)}_{N,k})(h_t^{\zeta'})- ({\mathcal{L}}_{N,k}f^{(0)}_{N,k})(h_t^\zeta)]\\ \le -\varrho {\mathbb{E}}[f^{(0)}_{N,k}(h^{\zeta'}_t)-f^{(0)}_{N,k}(h^\zeta_t)],\end{gathered}$$ and thus $$\label{eq:contract0}
{\mathbb{E}}[f^{(0)}_{N,k}(h^{\zeta'}_t)-f^{(0)}_{N,k}(h^\zeta_t)]\le e^{-\varrho t} \left( f^{(0)}_{N,k}(\zeta')-f^{(0)}_{N,k}(\zeta) \right).$$
The hydrodynamic limit
----------------------
We are interested in the macroscopic evolution of the height function.
For $\alpha\in [0,1]$, we define $\vee_{\alpha} : [0,1] \to {\mathbb{R}}$, $\wedge_{\alpha} : [0,1] \to {\mathbb{R}}$ as $$\vee_{\alpha}(x):= \max(-x,x-2(1-\alpha))\;,\qquad \wedge_{\alpha}(x):= \min(x,2\alpha-x)\;,$$ and we let $g_{\alpha}: {\mathbb{R}}_+\times [0,1]\to {\mathbb{R}}$ be defined as follows $$\begin{split}
g^0_{\alpha}(t,x)&:= \begin{cases} \alpha-\frac{t}{2}-\frac{(x-\alpha)^2}{2t}, \quad &\text{ if } |x-\alpha| \le t, \\
\wedge^{\alpha}(x), \quad & \text{ if } |x-\alpha| \ge t,
\end{cases}\\
g_{\alpha}(t,x)&:= \max(\vee_{\alpha}(x), g^0_{\alpha}(t,x)).
\end{split}$$
\[prop:lidro\] Assume that $Nb_N = N(p_N-q_N) \to\infty$ and that $k_N/N \to \alpha \in (0,1)$. Then, after an appropriate space-time scaling, $h^{\wedge}(\cdot,\cdot)$ converges to $g_\alpha$ in probability as $N\to\infty$. More precisely we have for any ${\varepsilon}>0$, $T>0$, $$\lim_{N\to \infty} {\mathbb{P}}\left[ \sup_{t\le T}\sup_{x\in[0,1]}\left| \frac{1}{N} h\Big( \frac{Nt}{b_N}, Nx \Big)-g_{\alpha}(t,x) \right|\ge {\varepsilon}\right]=0.$$
This is essentially the content of [@LabbeKPZ Th 1.3] where the hydrodynamic limit of the density of particles is shown to be given by the inviscid Burgers’ equation with zero-flux boundary conditions: when starting from the maximal initial condition, this yields (after integrating the density in space) the explicit solution $g_\alpha$.\
Actually the setting of [@LabbeKPZ Th 1.3] is more restrictive as the number of particles is taken to be $k=N/2$ and $p_N-q_N = 1/N^{\alpha}$ with $\alpha \in (0,1)$. However, a careful inspection of the proof shows that we only require $N^{1-\alpha}$ to go to infinity: this corresponds to the assumption $N(p_N-q_N) \to \infty$ which is in force in the statement of the proposition so that the proof carries through *mutatis mutandis*.
Lower bound on the mixing time for large biases {#Sec:LBLB}
===============================================
In the large bias case, the last observable that equilibrates is the position of the leftmost particle. Obtaining a lower bound on the mixing time is thus relatively simple: we have to show that for arbitrary $\delta>0$ at time $$s_{\delta}(N):=[(\sqrt{\alpha}+\sqrt{1-\alpha})^2-\delta]N b_N^{-1},$$ the leftmost particle has not reached its equilibrium position given by Lemma \[lem:lbeq\]. This is achieved by using the hydrodynamic limit for $\alpha>0$, and a simple comparison argument for $\alpha=0$.
When is satisfied, for every $\delta>0$ we have $$\lim_{N\to \infty} \left\| {\mathbb{P}}\left( \ell_N(\eta^{\wedge}_{s_{\delta}(N)}) \in \cdot \right)- \pi_N(\ell_N \in \cdot)\right\|_{TV}=1.$$ As a consequence for all ${\varepsilon}>0$ and $N$ sufficiently large $${T_{\rm mix}}^{N,k}(1-{\varepsilon})\ge s_{\delta}(N).$$
The case $\alpha=0$
-------------------
Given ${\delta}>0$ we want to prove that the system is not mixed at time $s_{\delta}(N):=(1-{\delta})N b_N^{-1}$. We know from Lemma \[lem:lbeq\], that when $\alpha=0$ and holds, at equilibrium we have $$\lim_{N\to \infty}\pi_{N,k}(\ell_N\le (1-{\delta}/2)N)=0.$$ On the other hand observing that the position of the first particle is dominated by a random walk on ${\mathbb{N}}$ with jump rates $p_N$ to the right and $q_N$ to the left, it is standard to check that whenever $\lim_{N\to \infty} b_N N=\infty$ $$\lim_{N\to \infty} {\mathbb{P}}\left( \ell_N(\eta^{\wedge}_{s_\delta})\le (1-{\delta}/2)N \right)=1.$$
The case $\alpha\in (0,1/2]$
----------------------------
Setting $x_\delta:= 1-\alpha-c_\alpha \delta$, for some positive constant $c_\alpha$ sufficiently small, we observe that the hydrodynamic profile at the rescaled time corresponding to $s_\delta$ is above the minimum at $x_{\delta}$ $$g_{\alpha}(x_{\delta}, (\sqrt{\alpha}+\sqrt{1-\alpha})^2-\delta)>\vee_{\alpha}(x_{\delta}).$$ The reader can check that $c_{\alpha}=1/3$ works for all $\alpha\in(0,1/2)$. Thus whenever $\lim_{N\to \infty} b_N N=\infty$ Proposition \[prop:lidro\] yields that $$\lim_{N\to \infty} {\mathbb{P}}\left( \ell_N(\eta^{\wedge}_{s_\delta}) \le (1-\alpha-c_{\alpha}\delta)N \right)=1.$$ On the other hand we know from Lemma \[lem:lbeq\] that when holds, at equilibrium we have for any $\delta>0$ $$\lim_{N\to \infty}\pi_{N,k}(\ell_N\le (1-\alpha-c_{\alpha}\delta)N)=0.$$
Upper bound on the mixing time for large biases {#Sec:UBLB}
===============================================
Let us recall how a grand coupling satisfying the order preservation property is of help to establish an upper bound on the mixing time. Recalling , we have by the triangle inequality $$\label{eq:zone}
d^{N,k}(t)\le \max_{\zeta,\zeta' \in {\Omega}_{N,k}}\|P^{N,k}_{t}(\zeta',\cdot)-P^{N,k}_{t}(\zeta,\cdot)\|_{TV}.$$ On the other hand if ${\mathbb{P}}$ is a monotone grand coupling, one observes that the extremal initial conditions are the last to couple so that one has $$\label{eq:ztwo}
\|P^{N,k}_{t}(\zeta',\cdot)-P^{N,k}_{t}(\zeta,\cdot)\|_{TV}\le {\mathbb{P}}[ h^{\zeta}_t\ne h^{\zeta'}_t] \le {\mathbb{P}}[ h^{\vee}_t\ne h^{\wedge}_t].$$ Hence to establish an upper bound on the mixing time, it suffices to obtain a good control on the merging time $$\label{eq:mergin}
\tau:=\inf\{ t>0 \ : \ h^{\vee}_t= h^{\wedge}_t\}.$$ Let us set for this section $$\label{deltaz}
t_{\delta}(N):=[(\sqrt{\alpha}+\sqrt{1-\alpha})^2+\delta]N b_N^{-1}.$$
\[lapropo\] When is satisfied, for every $\delta>0$, and any monotone grand coupling we have $$\lim_{N\to \infty} {\mathbb{P}}\left( \tau \le t_{\delta}(N)\right)=1.$$ As a consequence for all ${\varepsilon}>0$ and $N$ sufficiently large ${T_{\rm mix}}^{N,k}({\varepsilon})\le t_{\delta}(N)$.
When $\alpha=1/2$, a sharp estimate on $\tau$ can be obtained directly from spectral considerations (Section \[specz\]), but when $\alpha\in [0,1/2)$ we need a refinement of the strategy used in [@LabLac16]: The first step (Proposition \[lerimo\]) is to obtain a control on the position of the leftmost particle which matches the lower bound provided by the hydrodynamic limit. This requires a new proof since the argument used in [@LabLac16] is not sharp enough to cover all biases. The second step is to use contractive functions once the system is at macroscopic equilibrium, this is sufficient to treat most cases. A third and new step is required to treat the case when the bias $b_N$ of order $\log N/ N$ or smaller: as we are working under the assumption we only need to treat this case when $k=N^{o(1)}$. In this third step we use diffusive estimates to control the hitting time of zero for the function $f^{(0)}_{N,k}(h^{\wedge}_t)-f^{(0)}_{N,k}(h^{\vee}_t)$ where $f^{(0)}$ was introduced in Subsection \[sec:eigen\].
The special case $\alpha=1/2$ {#specz}
-----------------------------
In the special case $\alpha=1/2$, a sharp upper-bound can be deduced in a rather direct fashion from spectral estimates repeating the computation performed in [@Wil04 Section 3.2] for the symmetric case. This fact is itself a bit surprising since this method does not yield the correct upper bound in the symetric case nor in the constant bias case.
Recall the definition of $f_{N,k}$ in Equation and below. It being a strictly monotone function and ${\mathbb{P}}$ being a monotone coupling, we obtain using Markov’s inequality (recall ) $$\label{eq:mark}
{\mathbb{P}}(\tau>t)=
{\mathbb{P}}\left[ f_{N,k}(h^{\wedge}_t) > f_{N,k}(h^{\vee}_t)\right] \le
\frac{{\mathbb{E}}\left[ f_{N,k}(h^{\wedge}_t) - f_{N,k}(h^{\vee}_t)\right]}{\delta_{\min}(f_{N,k})}.$$ The expectation decays exponentially with rate $\operatorname{\mathrm{gap}}_N$ and it is not difficult to check that $$\label{delmin}
\delta_{\min}(f_{N,k})\ge {\lambda}^{(k-N)/2} \sin\left(\frac{\pi}{N}\right)\ge N^{-1}{\lambda}^{(k-N)/2}.$$ Hence Equation becomes $$\label{eq:zthree}
{\mathbb{P}}(\tau>t)
\le N {\lambda}^{\frac{N-k}{2}} e^{-\operatorname{\mathrm{gap}}_N t}\left( f_{N,k}(\wedge)-f_{N,k}(\vee) \right)
\le \frac{N^2 {\lambda}^{N/2}}{{\lambda}-1}e^{-\operatorname{\mathrm{gap}}_N t},$$ where in the last inequality we used that $$\sin\left(\frac{x\pi}{N}\right)\left({\lambda}^{\wedge(x)/2}-{\lambda}^{\vee(x)/2}\right)\le {\lambda}^{\wedge(x)/2}\le {\lambda}^{k/2}.$$
Recall that we assume that holds and $b_N$ tends to zero. Recalling , we obtain the following asymptotic equivalent $$\label{eq:lezequiv}
\operatorname{\mathrm{gap}}_N \stackrel{N\to \infty}{\sim} b_N^2/2 \quad \text{ and } \quad \log {\lambda}_N \stackrel{N\to \infty}{\sim} 2b_N$$ Furthermore for $N$ sufficiently large we have $({\lambda}-1)^{-1}\le N$. Hence recalling that $t_\delta=(2+\delta)b^{-1}_N N$ and using in we obtain for all $N$ sufficiently large $${\mathbb{P}}(\tau>t_\delta)\le
\frac{N^2}{{\lambda}-1} \exp\left(\frac{N}{2}\log {\lambda}-\operatorname{\mathrm{gap}}_N t_{{\delta}} \right)
\le N^3 e^{-\frac{\delta}{4}b_N N}.$$ and the left-hand side vanishes when $N$ tends to infinity as a consequence of (recall that as $\alpha=1/2$, $k$ is of order $N$)
The reason why the computation above yields a sharp upper bound is because: (A) The difference of order between $\delta_{\min}(f_{N,k})$ and the typical fluctuation of $f_{N,k}$ at equilibrium is negligible in the computation. (B) Until shortly before the mixing time the quantity $\log \left[f_{N,k}(h^{\wedge}_t) - f_{N,k}(h^{\vee}_t)\right]$ has the same order of magnitude as $\log {\mathbb{E}}\left[f_{N,k}(h^{\wedge}_t) - f_{N,k}(h^{\vee}_t)\right].$ In the case of symmetric exclusion, $(A)$ does not hold, while when the bias is constant, (B) fails to hold. In the weakly asymmetric case, when $\alpha\ne 1/2$, the reader can check by combining Propositions \[prop:lidro\] and \[lerimo\] that (B) holds until time $4\alpha$ (in the macroscopic time-scale) after which $g_{\alpha}(t,\cdot)$ stops to display a local maximum in the interval $(l_{\alpha}(t),r_{\alpha}(t))$ and $f_{N,k}(h^{\wedge}_t) - f_{N,k}(h^{\vee}_t)$ starts to decay much faster than its average.
The case $\alpha \ne1/2$: scaling limit for the boundary processes
------------------------------------------------------------------
In order to obtain a sharp upper-bound for $\alpha \ne 1/2$, we rely on a scaling limit result in order to control the value of $f_{N,k}(h^{\wedge}_t)-f_{N,k}(h^{\vee}_t)$ up to a time close to the mixing time, and then we use the contractive estimate to couple $h^{\wedge}_t$ with $h^{\vee}_t$. Note that Proposition \[prop:lidro\] is not sufficient to estimate $f_{N,k}(h^{\wedge}_t)$: we also need a control on the positions of the left-most particle and right-most empty site in our particle configuration.
In the case when $\alpha=0$ and the bias is of order $\log N/N$ or smaller (this is possible when is satisfied and $k$ grows slower than any power of $N$), we need an additional step, based on diffusion estimates, to couple the two processes. In this last case also, the factor $N^{-1}$ in causes some difficulty. For that reason we use $f^{(0)}_{N,k}$ and instead of $f_{N,k}$ and : observe that $\delta_{\min}(f^{(0)}_{N,k})={\lambda}^{\frac{k-N}{2}}$.
Let us define $[L_N(t),R_N(t)]$ to be the interval on which $h^{\wedge}_t$ and $\vee$ differ. More explicitly, we set $$\begin{split}\label{def:lknrkn}
L_N(t)&:=\max\{x \ : \ h^{\wedge}(t,x)=-x\},\\
R_N(t)&:=\min\{x \ : \ h^{\wedge}(t,x)= x-2(N-k)\},
\end{split}$$ or equivalently $L_N(t):=\ell_N(\eta^{\wedge}_t)-1$ and $R_N(t):=r_N(\eta^{\wedge}_t)$.
We let $\ell_\alpha$ and $r_{\alpha}$ denote the most likely candidates for the scaling limits of $L_N$ and $R_N$ that can be inferred from the hydrodynamic behavior of the system (cf. Proposition \[prop:lidro\]): $$\begin{split}
\ell_\alpha(t)&=\begin{cases} 0 \quad &\text{ if } t\le \alpha\;,\\
(\sqrt{t}-\sqrt{\alpha})^2 \quad &\text{ if } t\in \left(\alpha, (\sqrt{\alpha}+\sqrt{1-\alpha})^2 \right)\;,\\
1-\alpha \quad &\text{ if } t\ge (\sqrt{\alpha}+\sqrt{1-\alpha})^2\;,
\end{cases}
\\
r_{\alpha}(t)&=\begin{cases} 1 \quad &\text{ if } t\le 1-\alpha\;,\\
1-(\sqrt{t}-\sqrt{1-\alpha})^2 \quad &\text{ if } t\in \left(1-\alpha, (\sqrt{\alpha}+\sqrt{1-\alpha})^2 \right)\;,\\
1-\alpha \quad &\text{ if } t\ge (\sqrt{\alpha}+\sqrt{1-\alpha})^2\;.
\end{cases}\end{split}$$ We prove that $\ell_{\alpha}$ and $r_{\alpha}$ are indeed the scaling limits of $L_N$ and $R_N$.
\[lerimo\] If holds and $k_N/N\to \alpha$ then for every $t>0$ we have the following convergences in probability $$\lim_{N\to \infty} \frac{1}{N} L_N\left( b^{-1}_NNt \right)= \ell_\alpha(t),\qquad\lim_{N\to \infty} \frac{1}{N} R_N\left( b_N^{-1} Nt \right)= r_\alpha(t).$$
The assumption is optimal for the above result to hold. To see this, the reader can check that when fails, at equilibrium $\ell_N$ and $r_N$ are typically at a macroscopic distance from $(1-\alpha)N$.
The proof of Proposition \[lerimo\] is presented in the next subsections. Let us now check that it yields the right bound on mixing time. First, notice that the inequalities still hold with $f_{N,k}$ replaced by $f_{N,k}^{(0)}$ since the latter is also a strictly increasing function in the sense of . Next observe that Proposition \[lerimo\] allows an acute control on the quantity $$\frac{f^{(0)}_{N,k}(h^{\wedge}_t)-f^{(0)}_{N,k}(h^{\vee}_t)}{\delta_{\min} (f^{(0)}_{N,k})}\;.$$ We summarize the argument in a lemma.
\[lem:difrence\] Set $D_N(\zeta):= \max\left(|L_N(\zeta)-N+k|, |R_N(\zeta)-N+k|\right)$. We have for $\zeta' \ge \zeta$ $$\frac{f^{(0)}_{N,k}(\zeta')-f^{(0)}_{N,k}(\zeta)}{\delta_{\min} (f^{(0)}_{N,k})}\le Nk {\lambda}^{D_N(\zeta')}.$$
We assume that $\zeta' \ne \zeta$. Then, $$\label{linex}
\frac{\lambda^{\frac12 \zeta'(x)} - \lambda^{\frac12 \zeta(x)}}{{\lambda}-1} = \sum_{n=0}^{\frac{\zeta'(x)-\zeta(x)}{2}-1} \lambda^{\frac12\zeta(x)+n}\\
\le \lambda^{\frac{\zeta'(x)}{2}} \frac{\left(\zeta'(x)-\zeta(x)\right)}{2}.$$ Now for $x\le L_N(\zeta')$ or $x\ge R_N(\zeta')$ we necessarily have $\zeta(x)=\zeta'(x)=\vee(x)$. For $x\in {\llbracket}L_N(\zeta')+1,R_N(\zeta')-1{\rrbracket}$, the fact that $\zeta'$ is $1$-Lipschitz yields $$\zeta'(x)\le k-N+2 D_N(\zeta').$$ Recall that $\delta_{\min} (f^{(0)}_{N,k}) = {\lambda}^{(k-N)/2}$. Hence one obtains from $$\begin{aligned}
\frac{f^{(0)}_{N,k}(\zeta')-f^{(0)}_{N,k}(\zeta)}{\delta_{\min} (f^{(0)}_{N,k})}
&\le
\sum_{x=1}^N \lambda^{\frac{\zeta'(x)-(k-N)}{2}} \frac{\left(\zeta'(x)-\zeta(x)\right)}{2} \\ &\le
\lambda^{D_N(\zeta')} \sum_{x=1}^N \frac{\left(\zeta'(x)-\zeta(x)\right)}{2}\le \lambda^{D_N(\zeta')} Nk.\end{aligned}$$ In the last inequality we simply used that $\zeta'(x)-\zeta(x)\le 2k$ (there are at most $k$ sites where the increment of $\zeta'$ is larger than that of $\zeta$).
We can now apply Proposition \[lerimo\] to obtain an estimate on the mixing time. For convenience we treat the case of smaller bias separately.
### Proof of Proposition \[lapropo\] when $b_N\gg (\log N)/ N$ {#preuv1}
We assume that $$\label{hypo}
\lim_{N\to \infty}(b_N N)/\log N=\infty.$$ We consider first the system at time $t_0(N):= (\sqrt{\alpha}+\sqrt{1-\alpha})^2 N b^{-1}_N$. From Proposition \[lerimo\], we know that at time $t_0$, $L_N$ and $R_N$ are close to their equilibrium positions: we have for $N$ sufficiently large and arbitrary $\delta, {\varepsilon}>0$ $$\begin{gathered}
\label{eq:lax}
{\mathbb{P}}[ L_N(t_0)\ge N-k-(\delta/20) N \ ; \ R_N(t_0)\le N-k+(\delta/20) N ]\\
=: {\mathbb{P}}[{\mathcal{A}}_N] \ge 1-({\varepsilon}/2).\end{gathered}$$ We let ${\mathcal{F}}_t$ denote the canonical filtration associated with the process. For $t\ge t_0$, repeating starting at time $t_0$ for $f^{(0)}_{N,k}$ and combining it with , we obtain that $$\label{eq:lux}
{\mathbb{P}}[ \tau>t \ | \ {\mathcal{F}}_{t_0}]\le e^{-\varrho(t-t_0)}\frac{ f^{(0)}_{N,k}(h^{\wedge}_{t_0})-f^{(0)}_{N,k}(h^{\vee}_{t_0})}{\delta_{\min}(f^{(0)}_{N,k})}\;.$$
Note that on the event ${\mathcal{A}}_N$, we have $D_N(h^{\wedge}_{t_0})\le \delta N/20$. Thus using Lemma \[lem:difrence\] to bound the r.h.s. we obtain $$\label{eq:lox}
{\mathbb{E}}\left[\frac{ f^{(0)}_{N,k}(h^{\wedge}_{t_0})-f^{(0)}_{N,k}(h^{\vee}_{t_0})}{\delta_{\min}(f^{(0)}_{N,k})} \ \Big| \ {\mathcal{A}}_N\right] \le kN{\lambda}^{\frac{\delta N}{20}}.$$ Hence averaging on the event ${\mathcal{A}}_N$ one obtains we obtain $${\mathbb{P}}( \tau>t )\le {\varepsilon}/2+ {\mathbb{P}}[ \tau>t \ | \ {\mathcal{A}}_N]\le {\varepsilon}/2 + e^{-\varrho(t-t_0)}kN{\lambda}^{\frac{\delta N}{20}}.$$ For $t=t_\delta=t_0+ \delta b_N^{-1} N$, replacing $\varrho$ and $\log {\lambda}$ by their equivalents given in and , one can check that for $N$ sufficiently large one has $$\label{oazt}
N k {\lambda}^{\frac{\delta N}{20}} e^{-\varrho (t_{\delta}-t_0)} \le N k e^{-\frac{\delta N b_N}{20}}\le {\varepsilon}/2,$$ where the last inequality is valid for $N$ sufficiently large provided that holds.
### Proof of Proposition \[lapropo\]: the general case {#preuv2}
If we no longer assume that holds, then an additional step is needed in order to conclude: this step relies on diffusion estimates proved in Appendix \[sec:diffu\]. From and , for any ${\varepsilon}, {\delta}>0$ we have for $N$ sufficiently large (recall ) $$\begin{gathered}
{\mathbb{E}}\left[\frac{ f^{(0)}_{N,k}(h^\wedge_{t_{\delta/2}}) - f^{(0)}_{N,k}(h^\vee_{t_{\delta/2}})}{\delta_{\min}(f^{(0)}_{N,k})} \ \Big| \ {\mathcal{A}}_N \right]\\ \le
e^{-\varrho(t_{\delta/2}-t_0)}{\lambda}^{ \frac{\delta N}{20}} k N \le
e^{-\frac{\delta N b_N}{40}} k N\le e^{-\frac{\delta N b_N}{50}} N,\end{gathered}$$ where the second inequality relies on the the asymptotic equivalence in and the last one on .\
Now we can conclude using Proposition \[prop:solskjaer\]-(i) with $a:= 4{\varepsilon}^{-1} e^{-\delta N b_N/50} N$ and $$(M_{s})_{s\ge 0}:= \left(\frac{f^{(0)}_{N,k}(h^\wedge_{t_{\delta/2}+s})-f^{(0)}_{N,k}(h^\vee_{t_{\delta/2}+s})}{\delta_{\min}(f^{(0)}_{N,k})}\right)_{s\ge 0}.$$ Indeed $M_{s}$ is a non-negative supermartingale whose jumps are of size at least $1$ (recall that we have divided the weighted area by $\delta_{\min}(f^{(0)}_{N,k})$ in the definition of $M$). Furthermore, up to the merging time $\tau$, the two interfaces $h^\wedge$ and $h_\vee$ differ on some interval: on this interval $h^\wedge$ makes an upward corner ($\Delta h^\wedge < 0$) and $h^\vee$ makes a downward corner ($\Delta h^\vee > 0$). Consequently, the jump rate of $M$ is at least $1$ up to its hitting time of $0$. From Markov’s inequality we have (recall ) $${\mathbb{P}}[M_0 > a]\le {\mathbb{P}}[{\mathcal{A}}^{{\complement}}_N]+ a^{-1} {\mathbb{E}}[M_0 \ | \ {\mathcal{A}}_N ] \le 3{\varepsilon}/4.$$ Setting $r_{\delta}:= (\delta/2) N b^{-1}_N$ and applying , we have for all $N$ sufficiently large $${\mathbb{P}}[M_{r_{\delta}}>0 \ | \ M_0\le a ]\le 4 a (r_\delta)^{-1/2}\le \frac{16 {\varepsilon}^{-1}}{\sqrt{\delta/2}} \sqrt{N b_N} e^{-\frac{\delta N b_N}{50}}\le {\varepsilon}/4\;,$$ where the last inequality comes from the fact that $Nb_N$ diverges. Hence we conclude by observing that for $N$ sufficiently large $${\mathbb{P}}[\tau>t_{\delta}] ={\mathbb{P}}[M_{r_{\delta}}>0] \le {\mathbb{P}}[M_0 > a]+{\mathbb{P}}[M_{r_{\delta}}>0 \ | \ M_0\le a ]\le {\varepsilon}.$$
An auxiliary model to control the speed of the right-most particle {#sec:auxi}
------------------------------------------------------------------
Our strategy to prove Proposition \[lerimo\] is to compare our particle system with another one on the infinite line, for which a stationary probability exists. We consider $n$ particles performing the exclusion process on the infinite line with jump rate $p$ and $q$ and we add a “slower” $n+1$-th particle on the right to enforce existence of a stationary probability for the particle spacings. To make the system more tractable this extra particle is only allowed to jump to the right (so that it does not feel the influence of the $n$ others). Note that in our application, the number of particles $n$ does not necessarily coincide with $k$.\
The techniques developed in this section present some similarities to those used for the constant bias case in [@LabLac16 Section 6], but also present several improvements, the main conceptual change being the addition of a slow particle instead of modifying the biases in the process. This novelty presents two advantages: Firstly it considerably simplifies the computation since martingale concentration estimates are not needed any more. Secondly this allows to obtain control for the whole large bias regime , something that cannot be achieved even by optimizing all the parameters involved in [@LabLac16 Section 6].
More formally we consider a Markov process $(\hat \eta(t))_{t\ge 0}$ on the state space $$\Theta_n:= \{ \xi\in {\mathbb{Z}}^{n+1} \ : \ \xi_1<\xi_2<{\ifmmode\mathinner{\ldotp\kern-0.2em\ldotp\kern-0.2em\ldotp}\else.\kern-0.13em.\kern-0.13em.\fi}<\xi_{n+1}\}.$$ The coordinate $\hat \eta_i(t)$ denotes the position of the $i$-th leftmost particle at time $t$. The dynamics is defined as follows: the first $n$ particles, $\hat \eta_{i}(t)$, $i\in {\llbracket}1,n{\rrbracket}$ perform an exclusion dynamics with jump rates $p$ to the right and $q$ to the left while the last one $\hat \eta_{n+1}(t)$ can only jump to the right and does so with rate ${\beta}b={\beta}(p-q)$, for some $\beta<1$.
We assume furthermore that initially we have $\hat \eta_{n+1}(0)=0$. The initial position of the other particles is chosen to be random in the following manner. We define $$\label{ladef}
\mu_i:={\beta}+{\lambda}^{-i}(1-{\beta}).$$ and we assume that the spacings $\left( \hat \eta_{i+1}(0)-\hat \eta_{i}(0) \right)_{i=1}^n$ are independent with Geometric distribution $$\label{geom}
{\mathbb{P}}\left[ \hat \eta_{i+1}(0)-\hat \eta_i(0)= m \right]=(1-\mu_i)\mu_i^{m-1}\;,\quad m\ge 1\;.$$
Our aim is to prove the following control on the position of the first particle in this system, uniformly in ${\beta}$ and $n$. In Subsections \[Subsec:alpha0\] and \[Subsec:alphaNon0\], we use this result in order to control the position of $L_N(t)$.
\[deviats\] We have, $$\begin{gathered}
\label{eq:woof}
\lim_{A\to \infty}\sup_{t\ge 1, n\in {\mathbb{N}}, {\beta}\in(0,1)}
{\mathbb{P}}\bigg[\hat \eta_1(t) \le t{\beta}b\\
- A \left( \sqrt{bt}+ \frac{1}{1-{\beta}}\left[n+ b^{-1} \log \min(n,b^{-1}) \right] \right) \bigg]= 0.
\end{gathered}$$
The statement is not hard to prove, the key point is to observe that the distribution of particle spacings is stationary.
\[station\] For all $t\ge 0$, $\left( \hat \eta_{i+1}(t)-\hat \eta_{i}(t) \right)_{i=1}^n$ are independent r.v. with distribution given by .
We use the notation $(m_i)_{i=1}^n\in {\mathbb{N}}^n$ to denote a generic element in the configuration space for the process $\left( \hat \eta_{i+1}(t)-\hat \eta_{i}(t) \right)_{i=1}^n$. We need to show that the measure defined above is stationary.
A measure $\pi$ is stationary if and only if we have
$$\begin{aligned}
& p \pi(m_1+1,{\ifmmode\mathinner{\ldotp\kern-0.2em\ldotp\kern-0.2em\ldotp}\else.\kern-0.13em.\kern-0.13em.\fi},m_n) \\
+\;&\sum_{i=1}^{n-1}\left[p\pi({\ifmmode\mathinner{\ldotp\kern-0.2em\ldotp\kern-0.2em\ldotp}\else.\kern-0.13em.\kern-0.13em.\fi}, m_i-1,m_{i+1}+1,{\ifmmode\mathinner{\ldotp\kern-0.2em\ldotp\kern-0.2em\ldotp}\else.\kern-0.13em.\kern-0.13em.\fi})+q\pi({\ifmmode\mathinner{\ldotp\kern-0.2em\ldotp\kern-0.2em\ldotp}\else.\kern-0.13em.\kern-0.13em.\fi},m_{i-1}+1,m_i-1,{\ifmmode\mathinner{\ldotp\kern-0.2em\ldotp\kern-0.2em\ldotp}\else.\kern-0.13em.\kern-0.13em.\fi})\right]{\mathbf{1}}_{\{m_i\ge 2\}}\\
+\;&q\pi({\ifmmode\mathinner{\ldotp\kern-0.2em\ldotp\kern-0.2em\ldotp}\else.\kern-0.13em.\kern-0.13em.\fi},m_{n-1}+1,m_n-1) {\mathbf{1}}_{\{m_n\ge 2\}} +{\beta}b \pi(m_1,{\ifmmode\mathinner{\ldotp\kern-0.2em\ldotp\kern-0.2em\ldotp}\else.\kern-0.13em.\kern-0.13em.\fi},m_n-1){\mathbf{1}}_{\{m_n\ge 2\}}\\
&=
\pi(m_1,{\ifmmode\mathinner{\ldotp\kern-0.2em\ldotp\kern-0.2em\ldotp}\else.\kern-0.13em.\kern-0.13em.\fi},m_n)\left(q + \sum_{i=1}^{n-1} (p+q) {\mathbf{1}}_{\{m_i\ge 2\}}+ p{\mathbf{1}}_{\{m_n\ge 2\}} + {\beta}b \right),\end{aligned}$$
where in the sums, the dots stand for coordinates that are not modified (and $m_{i-1}$ simply has to be ignored when $i=1$). If we assume that $\pi$ is the product of geometric laws with respective parameters $\mu_i$ (not yet fixed) then the equation above is equivalent to the system $$\label{dasistem}
\begin{cases}
p\mu_1=q+ {\beta}(p-q), &\\
q \frac{\mu_{i-1}}{\mu_i}+p \frac{\mu_{i+1}}{\mu_i}=p+q, &\quad \forall i\in {\llbracket}1, n-1 {\rrbracket},\\
q\frac{\mu_{n-1}}{\mu_n}+\frac{{\beta}(p-q)}{\mu_n}=p.&
\end{cases}$$ where we have taken the convention $\mu_0=1$. One can readily check that $\mu_i$ given by satisfies this equation.
Note that the equations can be obtained directly simply by using the fact that the expected drifts of the particles starting from the geometric distributions are given by $p\mu_i-q \mu_{i-1}$ for the $i$-th particle $i\in {\llbracket}1,n {\rrbracket}$ and ${\beta}(p-q)$ for the $n+1$-th particle, and that stationarity implies that the drifts are all equal. However, the proof is necessary to show that this condition is also a sufficient one.
Starting from stationarity allows us to control the distance between the first and last particle at all time. In particular we have $$\label{staz}\begin{split}
{\mathbb{E}}\left[ \hat \eta_{n+1}(t)-\hat \eta_{1}(t)\right]&={\mathbb{E}}\left[ \hat \eta_{n+1} (0)-\hat \eta_{1}(0)\right]= \sum_{i=1}^n \frac{1}{1-\mu_i}
=\frac{1}{1-{\beta}} \sum_{i=1}^n \frac{1}{1-{\lambda}^{-i}}\\
&\le \frac{1}{1-{\beta}} \left(n+ \frac{C}{{\lambda}-1}\log \left(\min( n, |{\lambda}-1|^{-1} ) \right)\right),
\end{split}$$ for some universal constant $C$. By union bound, the probability in the l.h.s. of is smaller than $$\begin{gathered}
{\mathbb{P}}\left[\hat \eta_{n+1}(t) \le t{\beta}b- A \sqrt{bt}\right]\\+
{\mathbb{P}}\left[\hat \eta_{n+1}(t)
-\hat \eta_1(t)\ge \frac{A}{1-{\beta}}\left[n+b^{-1} \log \min(n,b^{-1})\right]\right].\end{gathered}$$ The first term is small because the expectation and the variance of $\hat \eta_{n+1}(t)$ are equal to $t{\beta}b$. The second can be shown to be going to zero with $A$ using and Markov’s inequality for $\hat \eta_{n+1}(t)-\hat\eta_1(t)$.
Proof of Proposition \[lerimo\] in the case $\alpha=0$ {#Subsec:alpha0}
------------------------------------------------------
We restate and prove the result in this special case (observe that the result for $R_N$ is trivial for $\alpha=0$).
Assume that $\alpha=0$ and holds. We have for any $C>0$ and any ${\varepsilon}>0$ $$\lim_{N\to \infty} \sup_{t\in [0,C b_N^{-1} N]} {\mathbb{P}}\left[ |L_N(t)- b_N t|\ge {\varepsilon}N \right]=0.$$
First let us remark that the convergence $$\lim_{N\to \infty} \sup_{t\in [0,C b_N^{-1} N]} {\mathbb{P}}\left[ L_N(t)\ge b_N t + {\varepsilon}N \right]=0$$ follows from the fact that the first particle is stochastically dominated by a simple random walk with bias $b_N \gg N^{-1}$ on the segment, starting from position $1$.
It remains to prove that $$\label{srup}
\lim_{N\to \infty} \sup_{t\in [0,C b_N^{-1} N]} {\mathbb{P}}\left[ L_N(t)\le b_N t- {\varepsilon}N \right]=0.$$ We provide the details for the most important case $C=1$, and then we briefly explain how to deal with the case $C>1$.\
We couple $\eta^{\wedge}(t)$ with the system $\hat \eta(t)$ of the previous section, choosing $n=k$ and ${\beta}=1-({\varepsilon}/2)$. The coupling is obtained by making the $i$-th particle in both processes try to jump at the same time (for $i\in {\llbracket}1, k {\rrbracket}$) with rate $p$ and $q$, and rejection of the moves occurs as consequences of the exclusion rule or boundary condition (for $\eta^{\wedge}$ only). Initially of course we have $$\label{initial}
\forall i \in {\llbracket}1, n {\rrbracket}, \quad \eta^{\wedge}_{i}(0)\ge \hat \eta_i(0).$$ because of the choice of the initial condition for $\hat \eta$ (recall that by definition $\eta^{\wedge}_i(0)=i$). The boundary at zero, and the presence of one more particle on the right in $\hat \eta$ gives $\eta^\wedge$ only more pushes towards the right, so that the ordering is preserved at least until $\hat \eta_{n+1}$ reaches the right side of the segment and the effect of the other boundary condition starts to be felt: $$\eta^\wedge_{i}(t)\ge \hat \eta_i(t),\quad \forall i \in {\llbracket}1, n {\rrbracket}, \forall t\le {\mathcal{T}}$$ where ${\mathcal{T}}:= \inf \{t \ge 0 \ : \ \hat \eta _{n+1}(t)=N+1 \}$.
Using the assumption , a second moment estimate and the fact that ${\beta}<1$, we have $$\lim_{N\to \infty} {\mathbb{P}}[ {\mathcal{T}}\le b^{-1}_N N] =0,$$ and hence $$\lim_{N\to \infty} \sup_{t\le b^{-1}_N N} {\mathbb{P}}\left[\eta^\wedge_1(t)\le \hat \eta_1(t)\right]=0.$$ Therefore, it suffices to control the probability of $\hat{\eta}_1(t) \le b_N t - {\varepsilon}N$. Observe that the assumptions $(k_N/N)\to 0$ and imply that for any given $A>0$, for all $N$ sufficiently large and for any $t\le b^{-1}_N N$ we have $$\left( \sqrt{b_N t}+ \frac{1}{1-{\beta}}\left[k_N+b_N^{-1}\log\min(k_N , b^{-1}_N) \right] \right)\le {\varepsilon}\frac{N}{2 A}.$$ Furthermore $(1-\beta)b_Nt \le {\varepsilon}N/ 2$ for all $t\in [0,b_N^{-1} N]$. Thus applying Proposition \[deviats\] we obtain that for $N$ sufficiently large $$\begin{gathered}
\sup_{t\in [0,b_N^{-1} N]} {\mathbb{P}}\left[ \hat \eta_1(t)\le b_N t-{\varepsilon}N\right]
\\
\le \sup_{t\in [0,b_N^{-1} N]} {\mathbb{P}}\left[ \hat \eta_1(t)\le {\beta}b_N t- A\left( \sqrt{b_N t}+ \frac{k_N+b_N^{-1}\log\min(k , b^{-1}_N)}{1-{\beta}}\right) \right]\le \delta.\end{gathered}$$ where $\delta$ can be made arbitrarily small by choosing $A$ large. This concludes the proof of for $C=1$. To treat the case $C>1$, it suffices to shift the particle system $\hat{\eta}(0)$ to the left by $\lfloor (C-1)b_N^{-1} N \rfloor$ and to apply the same arguments as before so we omit the details.
Proof of Proposition \[lerimo\] in the case $\alpha\in(0,1)$ {#Subsec:alphaNon0}
------------------------------------------------------------
The roles of $L_N$ and $R_N$ being symmetric, we only need to prove the result for $L_N$ (but we do not assume here that $\alpha\le 1/2)$. A direct consequence of Proposition \[prop:lidro\] is that for all $s\in {\mathbb{R}}$ and ${\varepsilon}>0$ we have $$\lim_{N\to \infty}{\mathbb{P}}\left[ \eta^{\wedge}_1(b_N^{-1} N s ) \ge N\left(\ell_{\alpha}(s)+{\varepsilon}\right)\right] = 0.$$ Hence to conclude we want to prove that $$\label{toprove}
\lim_{N\to \infty}{\mathbb{P}}\left[ \eta^{\wedge}_1(b_N^{-1} N s ) \le N\left(\ell_{\alpha}(s)-{\varepsilon}\right) \right]=0.$$ For the remainder of the proof $s$ and ${\varepsilon}$ are considered as fixed parameters. We set $\delta \in (0,\alpha)$, and $n=\lceil \delta N \rceil$. To prove , we are going to compare $(\eta^{\wedge}_i)_{i=1}^n$ to the particle system considered in Section \[sec:auxi\].
First we observe that as a consequence of Proposition \[prop:lidro\], we have, for any $T>0$ $$\label{camarch}
\lim_{N \to \infty} {\mathbb{P}}\left[ \exists t\in [0, T], \ \eta^{\wedge}_{n+1}(b^{-1}_N Nt)\le N \ell_{\alpha}(t) \right]=0$$ We define the process $\hat \eta$ as in Section \[sec:auxi\] with ${\beta}=1-{\varepsilon}/(2s)$ but with a shifted initial condition. More precisely we set $$\hat \eta_{n+1}(0)= N\left(\ell_{\alpha}(s)-s \right)\le 0,$$ and choose the initial particle spacings to be independent and with geometric distributions given by . As is satisfied, we can couple the two processes in such a way that $$\forall i \in {\llbracket}1, n {\rrbracket}\;, \forall t\le {\mathcal{T}}', \quad \eta^{\wedge}_{i}(t)\ge \hat \eta_i(t),$$ where ${\mathcal{T}}':= \inf\{ t \ : \hat \eta_{n+1}(t)=\eta^{\wedge}_{n+1}(t)\}$. It is a simple exercise to show that for every $T>0$ the position of $\hat \eta_{n+1}$ satisfies the following law of large numbers $$\lim_{N\to \infty}{\mathbb{P}}\left[ \sup_{t\in [0,T]} \left|\frac{\hat \eta_{n+1}(b_N^{-1}N t)}{N}-\left(\ell_{\alpha}(s)-s \right)-{\beta}t\right|\ge \kappa \right]=0,\quad \forall \kappa >0\;,$$ which, combined with , yields $$\lim_{N\to \infty} {\mathbb{P}}[{\mathcal{T}}' \ge b^{-1}_N N s]=1.$$ and thus we only need to prove with $\eta^{\wedge}_1$ replaced by $\hat \eta_{1}$. More precisely we prove that given $\kappa>0$, one can find $\delta$ sufficiently small such that $$\label{provex}
{\mathbb{P}}\left[ \hat\eta_1(b_N^{-1} N s ) \le N\left(\ell_{\alpha}(s)-{\varepsilon}\right) \right]\le \kappa.$$
Using Proposition \[deviats\] for $t=b^{-1}_N N s$ and $A= \delta^{-1/2}$ and taking into account the new initial condition, the probability of the event $$\left\{ \hat \eta_1(b^{-1}_N N s)\le N\left( \ell_{\alpha}(s)-\frac{{\varepsilon}}{2} \right)- \delta^{-1/2}\left( \sqrt{Ns}+ \frac{2s}{{\varepsilon}}[\delta N + b^{-1}_N \log b^{-1}_N] \right) \right\}$$ has a probability which can be made arbitrarily small if $\delta$ is chosen sufficiently small. We can then conclude that holds by observing that for $\delta$ sufficiently small and $N$ sufficiently large $$\delta^{-1/2}\left( \sqrt{Ns}+ \frac{2s}{{\varepsilon}}[\delta N + b^{-1}_N \log b^{-1}_N] \right)\le {\varepsilon}N/2.$$
Lower bound on the mixing time for small biases {#Sec:LBSB}
===============================================
Until the end of the section, we assume that the small bias assumption holds.
Let us set $s_{\delta}(N):= (1-\delta) \log k / (2\operatorname{\mathrm{gap}}_N)$. We show that at time $s_{\delta}$, equilibrium is not reached if one starts from one of the extremal conditions (some moderate efforts allow to replace $\max$ by $\min$ in the statement of the proposition).
\[lbsb\] When assumption holds, we have $$\lim_{N\to \infty} \max_{\zeta\in \{\vee,\wedge\} } \| {\mathbb{P}}( h^{\zeta}_{s_{{\delta}(N)}}\in \cdot )-\pi_{N,k}\|_{TV}=1$$ As a consequence for every ${\varepsilon}\in (0,1)$, ${T_{\rm mix}}^{N,k}({\varepsilon})\ge s_{\delta}(N)$ for $N$ sufficiently large.
The method to obtain a lower bound on the mixing time for small biases is similar to the one used in the symmetric case (see [@Wil04 Section 3.3]), and is based on the control of the two first moments of $f_{N,k}(h^{\wedge}_t)-f_{N,k}(\zeta)$ where $\zeta$ is independent of $h^\wedge_t$ and distributed according to $\pi_{N,k}$: if at time $t$ the mean of $f_{N,k}(h^\wedge_t)-f_{N,k}(\zeta)$ is much larger than its standard deviation, then the system is not at equilibrium (cf [@LevPerWil Proposition 7.12]).
We present estimates for the first two moments that we prove at the end of the section. This first moment bound is elementary.
\[lem:firstmom\] We have $$f_{N,k}(\wedge)-f_{N,k}(\vee)\ge \frac{1}{8} {\lambda}^{(k-N)/2} Nk\;,$$ and as a consequence, for every $t\ge 0$ $$\label{lunoulot}
\max( f_{N,k}(\wedge), -f_{N,k}(\vee)) \ge \frac{1}{16} {\lambda}^{(k-N)/2} Nk\;.$$
The second moment estimates rely on the control of a martingale bracket.
\[Lemma:BoundVar\] For all $t\ge 0$, $N\ge 1$ and all $k\in {\llbracket}1, N/2{\rrbracket}$ we have $${{\rm Var}}(f_{N,k}(h_t^\wedge)\le \frac{k \lambda^k}{2 \operatorname{\mathrm{gap}}_N},$$ The same bound holds for ${{\rm Var}}(f_{N,k}(h_t^\vee))$ and ${{\rm Var}}_{\pi_{N,k}}(f_{N,k})$.
Let us assume for simplicity (recall ) that $$f_{N,k}(\wedge)\ge \frac{1}{16} {\lambda}^{(k-N)/2} Nk.$$ (if not we apply the same proof to $f_{N,k}(\vee)\le -\frac{1}{16} {\lambda}^{(k-N)/2} Nk$). By the material in Section \[sec:eigen\], we have $$\label{cleup}
{\mathbb{E}}\left[ f_{N,k}(h^\wedge_t)\right] \ge \frac{1}{16} e^{-\operatorname{\mathrm{gap}}_N t} {\lambda}^{(k-N)/2} Nk.$$ Applying [@LevPerWil Proposition 7.12] for the probability measures $P_t^{N,k}(\wedge, \cdot)$ and $\pi_{N,k}$ and the function $f_{N,k}$ (recall that $E_{\pi_{N,k}}[f_{N,k}]=0$), we obtain that $$\|P_t^{N,k}(\wedge, \cdot)-\pi_{N,k} \|_{TV} \ge 1-\frac{2 \left({{\rm Var}}(f_{N,k}(h_t^\wedge))+{{\rm Var}}_{\pi_{N,k}}(f_{N,k}) \right)}{{\mathbb{E}}\left[f_{N,k}(h_t^\wedge)\right]^2}\;.$$ Using Lemma \[Lemma:BoundVar\] and , we obtain that $$\frac{{{\rm Var}}(f_{N,k}(h_t^\wedge))+{{\rm Var}}_{\pi_{N,k}}(f_{N,k})}{{\mathbb{E}}\left[f_{N,k}(h_t^\wedge)\right]^2}\le \frac{ 16^2 e^{2\operatorname{\mathrm{gap}}_N t} {\lambda}^N}{\operatorname{\mathrm{gap}}_N N^2 k}.$$ Now if we apply this inequality at time $s_{\delta} = (1-\delta) \log k / (2\operatorname{\mathrm{gap}}_N)$, then we obtain for any given ${\varepsilon}> 0$ and all $N$ sufficiently large $$d(t_1)\ge 1-2\frac{16^2 {\lambda}^N }{k^{\delta}\operatorname{\mathrm{gap}}_N N^2}\ge 1-{\varepsilon}.$$ where we used the small bias assumption . This yields ${T_{\rm mix}}^{N,k}({\varepsilon})\ge s_{\delta}$.
We have $$\begin{aligned}
f_{N,k}(\wedge) - f_{N,k}(\vee) &= \sum_{x=1}^{N-1} \sin \left(\frac{x\pi}{N}\right) \frac{\lambda^{\frac12 \wedge(x)} - \lambda^{\frac12 \vee(x)}}{\lambda-1}\\
&\ge \sum_{x=1}^{N-1} \sin \left(\frac{x\pi}{N}\right) \lambda^{\frac12 \vee(x)} \frac{\wedge(x)-\vee(x)}{2},\end{aligned}$$ where the last inequality is obtained similarly to . Since $\vee(x)\ge k-N$ for all $x$ and $\wedge(x)-\vee(x) \ge k$ for all $x\in \{N/4,\ldots,3N/4\}$, we conclude that $$\sum_{x=1}^{N-1} \sin \left(\frac{x\pi}{N}\right) \lambda^{\frac12 \vee(x)} \frac{\wedge(x)-\vee(x)}{2} \ge \frac{\sqrt 2}{2}{\lambda}^{\frac{k-N}{2}}\frac{N k}{4}.$$
By the material in Section \[sec:eigen\], we know $$M_t := f_{N,k}(h^\wedge_t) e^{\operatorname{\mathrm{gap}}_N t}$$ is a martingale. Its predictable bracket is given by $$\begin{aligned}
\langle M_\cdot \rangle_t &= \int_0^t \sum_{x=1}^{N-1} \lambda^{h_s^\wedge(x)} \sin\Big(\frac{\pi x}{N}\Big)^2 e^{2\operatorname{\mathrm{gap}}_N s}\\
&\quad\times\Big(p_N{\mathbf{1}}_{\{\Delta h_s^\wedge(x) < 0\}} {\lambda}^{-2} + q_N {\mathbf{1}}_{\{\Delta h_s^\wedge(x) > 0\}}\Big) ds\;,\end{aligned}$$ and $M_t^2 - \langle M_\cdot \rangle_t$ is again a martingale. This yields the identity $${{\rm Var}}(f_{N,k}(h_t^\wedge)) = e^{-2\operatorname{\mathrm{gap}}_N t}{\mathbb{E}}\big[\langle M_\cdot \rangle_t\big].$$ To bound the predictable bracket of $M$, let us observe that the number of possible particle transitions to the right and to the left (the number of sites $x$ such that $\Delta h_s^\wedge(x)<0$, resp. $>0$) is bounded by $k$, and that for any $x$ and $\zeta\in {\Omega}_{N,k}$ we have $\lambda^{\zeta(x)} \le \lambda^{k}$. Therefore, we obtain the bound $$\begin{aligned}
{\mathbb{E}}\big[\langle M_\cdot \rangle_t\big] &\le \int_0^t e^{2\operatorname{\mathrm{gap}}_N s} ds\, \lambda^k \sum_x {\mathbf{1}}_{\{\Delta h_s^\wedge(x) \ne 0\}}\le k\lambda^k \frac{e^{2\operatorname{\mathrm{gap}}_N t}}{2\operatorname{\mathrm{gap}}_N}\;,\end{aligned}$$ which yields the asserted bound. The case of $h^{\vee}_t$ is treated in the same manner by symmetry. Since the distribution of $h^\wedge_t$ converges to $\pi_{N,k}$ when $t$ tends to infinity we deduce that $${{\rm Var}}_{\pi_{N,k}}(f_{N,k}) = \lim_{t\rightarrow\infty} {{\rm Var}}(f_{N,k}(h_t^\wedge))\;,$$ which allows to conclude.
Upper bound on the mixing time for small biases {#Sec:UBSB}
===============================================
Until the end of the section we assume that the small bias assumption holds and that the different initial conditions are coupled using the monotone grand coupling ${\mathbb{P}}$ defined in Appendix \[Appendix:Coupling\]. We set for all $\delta > 0$ $$t_{\delta}(N):=(1+\delta)\frac{\log k}{2\operatorname{\mathrm{gap}}_N}.$$ Recall the definition of the merging time $\tau$ from .
\[oldform\] Assume that holds. We have $$\label{laforme}
\lim_{N\to \infty} {\mathbb{P}}[\tau < t_{\delta}(N)]=1.$$ As a consequence, for every ${\varepsilon}>0$ and all $N$ sufficiently large, ${T_{\rm mix}}^{N,k}({\varepsilon})\le t_{\delta}(N)$.
Recall (see the paragraph after ) that $h^{\pi}_t$ denotes the chain with stationary initial condition. For practical reasons, it is simpler to couple two processes when at least one of them is at equilibrium. We thus prove by showing that $\lim_{N\to \infty}{\mathbb{P}}[\tau_i < t_{\delta}(N)]=1$ for $i\in \{1,2\}$ where $$\tau_1:=\inf\{ t>0 \ : \ h^{\wedge}_t=h^{\pi}_t \}\quad \text{ and }\quad \tau_2:=\inf\{ t>0 \ : \ h^{\vee}_t=h^{\pi}_t \}.$$ The argument being completely symmetric, we focus only on $\tau_1$. As in Sections \[preuv1\] and \[preuv2\], we interpret $\tau_1$ as the time at which the weighted area $A_t$ between the maximal and equilibrium interface vanishes $$\label{defAt}
A_{t}:= \frac{f^{(0)}_{N,k}(h^{\wedge}_t)-f^{(0)}_{N,k}(h^{\pi}_t)}{\delta_{\min}(f^{(0)}_{N,k})}
={\lambda}^{\frac{N-k}{2}}\sum_{x=1}^{N-1} \frac{\lambda^{\frac12 h_t^\wedge(x)}-\lambda^{\frac12 h_t^\pi(x)}}{\lambda - 1} \;.$$ A simple computation based on the identity shows that $A$ is a supermartingale.
While in the large bias case (Section \[Sec:UBLB\]) the choice of the grand coupling does not matter, here it is crucial to use a coupling which maximizes in a certain sense the fluctuation of the weighted area $A_t$, so that this process reaches zero as quickly as possible. The coupling defined in Appendix \[Appendix:Coupling\] makes the transitions for the two processes $h^{\wedge}$ and $h^{\pi}$ as independent as possible (some transitions must occur simultaneously for the two processes in order to preserve monotonicity).
We consider $\eta >0$ small and introduce the successive stopping times ${\mathcal{T}}_i$ by setting $${\mathcal{T}}_0 := \inf\big\{t\ge t_{\delta/2}: A_t \le k^{\frac12 - \frac {\delta}5} N\big\}\;,$$ and $${\mathcal{T}}_{i} := \inf\big\{t\ge {\mathcal{T}}_{i-1}: A_t \le k^{\frac12 - i\eta - \frac {\delta}5} N\big\}\;,\quad i\ge 1\;.$$ We also set for coherence ${\mathcal{T}}_{\infty}:=\max(\tau_1,t_{\delta/2})$ the first time at which $A_t$ reaches $0$. Notice that some of these stopping times may be equal to $t_{\delta/2}$.\
Set $T_N:=\min( b^{-2}_N, N^2)$. To prove Proposition \[oldform\], we show first that $A_t$ shrinks to $k^{\frac12-\frac {\delta}5} N$ by time $t_{\delta/2}$ and then that it only needs an extra time $2T_N$ to reach $0$. The second step is performed by controlling each increment ${\Delta}{\mathcal{T}}_i:={\mathcal{T}}_i-{\mathcal{T}}_{i-1}$ separately for each $i$ smaller than some threshold $K:=\lceil 1/(2\eta)\rceil$.
\[newform\] Given $\delta$, if $\eta$ is chosen small enough and $K:=\lceil 1/(2\eta)\rceil$, we have $$\lim_{N\to \infty}{\mathbb{P}}\left( \{ {\mathcal{T}}_{0} = t_{\delta/2}\}\cap \left(\bigcap_{i=1}^K
\{{\Delta}{\mathcal{T}}_{i}\le 2^{-i} T_N\}\right)\cap\{{\mathcal{T}}_{\infty}-{\mathcal{T}}_{K}\le T_N \}\right)=1 \;.$$
Note that on the event defined in the lemma and for all $N$ large enough, we have $$\tau_1\le {\mathcal{T}}_{\infty}\le t_{\delta/2}+2T_N\le t_{\delta}.$$ Hence Proposition \[oldform\] follows as a direct consequence.
The bound on ${\mathcal{T}}_0$ is proved in Section \[s:contract\], while that of on ${\mathcal{T}}_{\infty}-{\mathcal{T}}_{K}$ follows from Lemma \[Lemma:AN\] in Section \[s:dif\], the case of the other increments is more delicate and is detailed in Section \[s:inter\].
Contraction estimates {#s:contract}
---------------------
The approach used in the first step bears some similarity with the one used in Section \[preuv1\], the notable difference being that is not sufficient here and we must work a bit more to show that ${\mathbb{E}}[A_t]$ decays with rate $\operatorname{\mathrm{gap}}_N$.
\[Lemma:T0\] Given $\delta>0$ we have ${\mathbb{P}}\big( {\mathcal{T}}_0 > t_{\delta/2}\big) \to 0$ as $N\to\infty$.
Note that $a(t,x):= {\mathbb{E}}\left[\frac{\lambda^{\frac12 h_t^\wedge(x)}-\lambda^{\frac12 h_t^\pi(x)}}{\lambda - 1} \right]$ is a solution of the equation $$\partial_t a = (\sqrt{pq}\, {\Delta}-\varrho)a\;,$$ with $a(t,0)=a(t,N)=0$. Diagonalising the operator on the right hand side, see Subsection \[sec:eigen\], we get the following bound on the $\ell^2$-norm of the solution: $$\sum_{x=1}^{N-1}a(t,x)^2 \le e^{-2\operatorname{\mathrm{gap}}_N t}\sum_{x=1}^{N-1}a(0,x)^2,$$ and using Cauchy-Schwartz inequality we obtain $${\lambda}^{\frac{k-N}{2}}{\mathbb{E}}[A_t] = \sum_{x=1}^{N-1}a(t,x)\le \sqrt{N} e^{-\operatorname{\mathrm{gap}}_N t}\sqrt{\sum_{x=1}^{N-1}a(0,x)^2}\le 2e^{-\operatorname{\mathrm{gap}}_N t} Nk {\lambda}^{k/2}\;.$$ Since $\lambda^{N/2}$ is, by the small bias assumption, asymptotically smaller than any power of $k$, Markov’s inequality concludes the proof.
Diffusion estimate after time $t_{\delta/2}$ {#s:dif}
--------------------------------------------
Now this part is much more delicate than Section \[preuv2\]. The reason being that since $T_N$ is extremely close to $t_{\delta}$, we need very accurate control on the derivative of the predictable bracket of $A_t$. Our first task is to use Proposition \[prop:solskjaer\] in order to control the increment of the bracket of $A$ in between the ${\mathcal{T}}_i$’s. Let us set $$\Delta_i \langle A\rangle := \langle A_\cdot\rangle_{{\mathcal{T}}_{i}} - \langle A_\cdot\rangle_{{\mathcal{T}}_{i-1}},\qquad\Delta_{\infty} \langle A\rangle := \langle A_\cdot\rangle_{{\mathcal{T}}_{\infty}} - \langle A_\cdot\rangle_{{\mathcal{T}}_{K}}\;,$$
and consider the event $${\mathcal{A}}_N:= \left\{ \forall i \in {\llbracket}1,K {\rrbracket}, \quad \Delta_i \langle A\rangle \le k^{1 - 2(i-1)\eta - \frac {\delta}4} N^2 \right\}\cap
\left\{ \Delta_{\infty} \langle A\rangle \le T_N \right\}$$
\[Lemma:AN\] We have $\lim_{N\to \infty} {\mathbb{P}}[{\mathcal{A}}_N^{{\complement}}]=0$.
We apply Proposition \[prop:solskjaer\]-(ii) to $(A_{t+{\mathcal{T}}_{i-1}})_{t\ge 0}$, with $a=k^{\frac12 - (i-1)\eta - \frac {\delta}5}N$, $b=k^{\frac12 - i\eta - \frac {\delta}5}N$. We obtain that for all $N$ sufficiently large and every $i\le K$ $${\mathbb{P}}[ \Delta_i \langle A\rangle\ge k^{1 - 2(i-1)\eta - \frac {\delta}4} N^2]\le k^{-\delta/100}.$$ Applying the same proposition to $(A_{t+{\mathcal{T}}_{K}})_{t\ge 0}$ with $a=k^{- \frac {\delta}5}N$ and $b=0$, we obtain $${\mathbb{P}}[ \Delta_\infty \langle A\rangle\ge T_N ]\le 8 N k^{-\frac {{\delta}}{5}}(T_N)^{-1/2},$$ and the r.h.s. tends to zero by assumption .
The next step is to compare $\Delta_i \langle A\rangle$ with ${\mathcal{T}}_i-{\mathcal{T}}_{i-1}$. For the last increment this is easy: We have $\partial_t \langle A_\cdot\rangle\ge 1$ for any $t\le {\mathcal{T}}_{\infty}$ (from our construction $A$ changes its value at rate at least $1$, and its minimal increment in absolute value is $1$). We have thus ${\mathcal{T}}_{\infty}-{\mathcal{T}}_K \le {\Delta}_{\infty} \langle A \rangle$, and thus when ${\mathcal{A}}_N$ holds we have $$\label{Eq:Tinf}
{\mathcal{T}}_{\infty}-{\mathcal{T}}_K \le T_N\;.$$
Control of intermediate increments {#s:inter}
----------------------------------
For all other increments we have to use a subtler control of the bracket. Let us set $${\mathbf{H}}(t):= {\lambda}^{\frac{N-k}{2}}\max_{x\in {\llbracket}0,N {\rrbracket}} \frac{{\lambda}^{\frac12 h^{\wedge}_t(x)}- {\lambda}^{\frac12 h^{\vee}_t(x)}}{{\lambda}-1}\;,$$ which corresponds roughly (up to a multiplicative factor ${\lambda}^{N}$) to the maximal height difference $\max h^{\wedge}_t(x)-h^{\vee}_t(x)$ and thus provides a bound for $\max_x h^{\wedge}_t(x)-h^{\pi}_t(x)$.\
Recall $Q(\cdot)$ from Subsection \[Subsec:eqSmallBias\], and set $Q(h_t^\pi) := Q(\eta_t^\pi)$ where $\eta_t^\pi$ is the particle configuration associated with $h_t^\pi$.
\[Lemma:BrackDeriv\] We have $\partial_t \langle A_\cdot\rangle \ge \frac{ A_t }{6 {\mathbf{H}}(t) Q(h_t^{\pi})}$.
As mentioned above, all the jumps of $A_t$ have amplitude larger than or equal to $1$. Moreover, $A_t$ performs a jump whenever $h^{\pi}_t$ performs a transition while $h^{\wedge}_t$ does not, or when the opposite occurs. As any such transition occurs at rate larger than $q_N\ge 1/3$, only considering the transitions for $h^{\pi}_t$ , we obtain the following lower bound for the drift of $\langle A_\cdot\rangle$ (recall ) $$\label{loopz}
\partial_t \langle A_\cdot\rangle \ge \frac{1}{3} \#\{ x\in {\mathcal{C}}_t \ : \ {\Delta}(h^{\pi}_t)(x)\ne 0\}=:\frac{1}{3}\#{\mathcal{D}}_t$$ where $${\mathcal{C}}_t:= \big\{x\in{\llbracket}1,N-1{\rrbracket}\ : \ \exists y\in {\llbracket}x-1,x+1{\rrbracket},\ h^\wedge_t(y) > h^{\pi}_t(y) \big\}\;.$$ Now let ${\llbracket}a, b{\rrbracket}$ be a maximal connected component of ${\mathcal{C}}_t$, we claim that $$\label{zaap}
\#( {\mathcal{D}}_t \cap {\llbracket}a,b {\rrbracket})\ge \max\left( \left\lfloor \frac{b-a}{Q(h^{\pi}_t)}\right\rfloor, 1 \right)
\ge \frac{b-a}{2Q(h^{\pi}_t)}.$$ To check this inequality, notice that $\#( {\mathcal{D}}_t \cap {\llbracket}a,b {\rrbracket})\ge 1$ because $h^{\pi}_t$ cannot be linear on the whole segment ${\llbracket}a, b{\rrbracket}$. On the other hand, considering the particle configuration associated to $h^\pi_t$ and decomposing the segment ${\llbracket}a,b{\rrbracket}$ into maximal connected components containing either only particles or only holes, we see that any two consecutive components corresponds to a point in ${\mathcal{D}}_t$: since $Q(h^{\pi}_t)$ is an upper bound for the size of these components, we deduce that $\#( {\mathcal{D}}_t \cap {\llbracket}a,b {\rrbracket})\ge \left\lfloor \frac{b-a}{Q(h^{\pi}_t)}\right\rfloor$.
Now we observe that $$\label{ziip}
{\lambda}^{\frac{N-k}{2}}\sum_{x=a}^b \frac{{\lambda}^{\frac{h^{\wedge}_t(x)}{2}}-{\lambda}^{\frac{h^{\pi}_t(x)}{2}}}{{\lambda}-1} \le
(b-a) {\mathbf{H}}(t).$$ Combining and and summing over all such intervals ${\llbracket}a,b{\rrbracket}$, we obtain $$A_t\le 2\,\#{\mathcal{D}}_t \, {\mathbf{H}}(t)\,Q(h^{\pi}_t),$$ and allows us to conclude.
The last ingredient needed is then a bound on ${\mathbf{H}}$: The proof of this lemma is postponed to Subsection \[Subsec:Max\]. Recall that $t_0=\log k/(2\operatorname{\mathrm{gap}}_N)$.
\[Prop:BoundMax0\] For any $c >0$ we have $$\lim_{N\to \infty} \sup_{t\ge t_0} {\mathbb{P}}\Big({\mathbf{H}}(t) > k^{\frac12 + c}\Big) = 0\;.$$
By Lemma \[Lemma:T0\], Lemma \[Lemma:AN\] and Equation , we already know that $$\lim_{N\to \infty}{\mathbb{P}}\left( \{ {\mathcal{T}}_{0} \le t_{\delta/2}\}\cap \{{\mathcal{T}}_{\infty}-{\mathcal{T}}_{K}\le T_N \}\right)=1 \;.$$ We define ${\mathcal{H}}_N$ to be the event on which particles are reasonably spread and ${\mathbf{H}}(t)$ is reasonably small for most of the times within the interval $[t_{\delta/2}, t_{\delta/2}+T_N]$, $${\mathcal{H}}_N:= \Big\{ \int^{t_{\delta/2}+T_N}_{t_{\delta/2}} {\mathbf{1}}_{\{\text{ ${\mathbf{H}}(t) \le k^{\frac12 + \frac{\delta}{80}}\} \cap \{Q(h_t^\pi) \le N k^{\frac{\delta}{80}-1}$ } \}} dt \ge T_N (1-2^{-(K+1)}) \Big\}\;.$$ By Markov’s inequality, Proposition \[lem:dens\] and Proposition \[Prop:BoundMax0\], we have $$\lim_{N\to \infty}{\mathbb{P}}({\mathcal{H}}_N) = 1\;.$$ We now work on the event ${\mathcal{H}}_N \cap {\mathcal{A}}_N \cap \{{\mathcal{T}}_0 \le t_{\delta/2}\}$ whose probability tends to $1$ according to Lemmas \[Lemma:T0\], \[Lemma:AN\]. We prove by induction that $\Delta{\mathcal{T}}_j \le 2^{-j} T_N$ for all $j\in {\llbracket}1,K{\rrbracket}$. Let us reason by contradiction and let $i$ be the smallest integer such that $\Delta{\mathcal{T}}_i > 2^{-i} T_N$. We have $$\label{indd}
\Delta_i\langle A\rangle \ge \int_{{\mathcal{T}}_{i-1}}^{{\mathcal{T}}_{i-1}+2^{-i}T_N} \partial_t \langle A_\cdot
\rangle{\mathbf{1}}_{\{\text{ ${\mathcal{H}}(t) \le k^{\frac12 + \frac{\delta}{80}}\}\cap\{Q(h_t^\pi) \le N k^{\frac{\delta}{80}-1}$ } \}} dt$$ Now, Lemma \[Lemma:BrackDeriv\] and the restriction with the indicator function provides a uniform lower bound on $\partial_t \langle A \cdot \rangle$. The assumption $\Delta{\mathcal{T}}_j \le 2^{-j} T_N$ for $j<i$ implies that ${\mathcal{T}}_{i-1}\le t_{\delta/2} + T_N(1-2^{-(i-1)})$, and thus the assumption that ${\mathcal{H}}_N$ holds implies that the indicator in is equal to one on a set of measure at least $2^{-i} - 2^{-(K+1)}\ge 2^{-(K+1)}$. All of this implies that $$\Delta_i\langle A\rangle \ge \frac16 T_N 2^{-(K+1)} k^{1 -i\eta -\frac{\delta}{40} - \frac{\delta}{5}}$$ On the other hand, since we work on ${\mathcal{A}}_N$ we have $\Delta_i\langle A\rangle \le k^{1-2(i-1)\eta-\frac{\delta}{4}}N^2$ so that we get a contradiction as soon as $\eta$ is small enough compared to $\delta$.
Bounding the maximum {#Subsec:Max}
--------------------
Recall the function $a_{N,k}$ defined in Subsection \[sec:eigen\]. Set $$H_1(t,x) := {\lambda}^{\frac{N-k}{2}}\frac{\lambda^{\frac12 h_t^\wedge(x)}-a_{N,k}(x)}{\lambda- 1}\;,\quad H_2(t,x) :={\lambda}^{\frac{N-k}{2}} \frac{\lambda^{\frac12 h_t^{\vee}(x)}-a_{N,k}(x)}{\lambda- 1}\;,$$ so that $$H_1(t,x)-H_2(t,x) = {\lambda}^{\frac{N-k}{2}}\frac{\lambda^{\frac12 h_t^\wedge(x)}-\lambda^{\frac12 h_t^{\vee}(x)}}{\lambda- 1}\;.$$ For every $i=1,2$, we define $${\mathbf{H}}_i(t):= \max_{x\in {\llbracket}0,N{\rrbracket}} |H_i(t,x)|\;.$$ Notice that ${\mathbf{H}}(t) \le {\mathbf{H}}_1(t) + {\mathbf{H}}_2(t)$ so that Proposition \[Prop:BoundMax0\] is a consequence of the following result.
\[Prop:BoundMax\] For any $c >0$, there exists $c' >0$ such that for all $N$ large enough $$\sup_{t\ge t_0} \max_{i\in \{1,2\}} {\mathbb{P}}\Big({\mathbf{H}}_i(t) > k^{\frac12 + c}\Big) \le e^{- k^{c'}}\;.$$
The proof of this bound is split into two lemmas. First, we show that $H_i(t,\cdot)$ can not decrease too much.
\[smsl\] We have for all $N$ sufficiently large, all $x\in {\llbracket}1,N-1{\rrbracket}$, all $t\ge 0$, every $i\in\{1,2\}$ and every $y\ge x$ $$\label{zladiff}
H_i(t,y)-H_i(t,x)\ge -\frac{k^2(y-x)}{4N}.$$
It is of course sufficient to prove that $$H_i(t,x)-H_i(t,x-1)\ge -\frac{k^2}{4N}\;.$$ We have for any $\eta\in {\Omega}^0_{N,k}$, setting $h=h(\eta)$, $$\frac{\lambda^{\frac12 h(x)}-\lambda^{\frac12 h(x-1)}}{\lambda- 1}
=\lambda^{\frac12 (h(x-1)-1)}\left(\eta(x)-\frac{1}{\sqrt{{\lambda}}+1}\right).$$ Note that $a_{N,k}(x)=E_{\pi_{N,k}}[ {\lambda}^{h(x)/2}]$ where $a_{N,k}$ was defined in Section \[sec:eigen\]. Hence using the fact that $\eta^{\wedge}_t(x)\ge 0$, we get (the same holds for $i=2$ and $h^\vee$): $$\begin{gathered}
\label{Eq:H1H1}
H_1(t,x)-H_1(t,x-1)\\
\ge
{\lambda}^{\frac{N-k}{2}}\frac{ E_{\pi_{N,k}}\left[\lambda^{\frac{h(x-1)-1}{2}}\right]-\lambda^{\frac{h^{\wedge}_t(x-1)-1}{2}}}{\sqrt{{\lambda}}+1}
-{\lambda}^{\frac{N-k}{2}} E_{\pi_{N,k}}\left[ \lambda^{\frac{h(x-1)-1}{2}}\eta(x)\right].
\end{gathered}$$ By Proposition \[lem:dens\] and the small bias assumption , we have for all $N$ large enough $${\lambda}^{\frac{N-k}{2}} E_{\pi_{N,k}}\left[ \lambda^{\frac{h(x-1)-1}{2}}\eta(x)\right]\le {\lambda}^{\frac{N}{2}}E_{\pi_{N,k}}[\eta(x)]\le {\lambda}^{2N} \frac{k}{N}\le \frac{k^2}{8N}.$$ Regarding the first term on the r.h.s. of , we simply notice that for $\zeta, \zeta'\in {\Omega}_{N,k}$, we have $\zeta(x)-\zeta'(x)\le 2k$ so that $${\lambda}^{\frac{N-k}{2}}|{\lambda}^{\frac{\zeta(x)}{2}}-{\lambda}^{\frac{\zeta'(x)}{2}}|\le ({\lambda}-1){\lambda}^{\frac{N}{2}} k \le \frac{k^{2}}{8N}\;.$$ This is sufficient to conclude.
Let us introduce the average of $H_i(t,\cdot)$ over a box of size $\ell =\ell_{N,k}= \lceil \frac{N}{k^2}\rceil$ $$\bar{H}_i(t,y) := \frac1{\ell}\sum_{x=(\ell- 1)y+1}^{y\ell} H_i(t,x)\;.$$ As a consequence of Lemma \[smsl\] we have $${\mathbf{H}}_i(t)= \max_{x\in {\llbracket}0,N{\rrbracket}} |H_i(t,x)|\le \max_{y\in {\llbracket}1, N/\ell {\rrbracket}} \big| \bar{H}_i(t,y) \big|+1 \;.$$ The result is of course obvious when $\ell=1$. For $\ell \ge 2$, let us briefly explain why $\max H_i(t,x)\le \max |\bar{H}_i(t,y)|+1$ (the case for $-\min$ follows by symmetry). If $x_{\max}$ is the smallest $x$ at which the $\max$ is attained, we must distinguish two cases
- $x_{\max}> \ell\left( \lfloor N/\ell \rfloor-1\right)+1\ge N-2\ell$, in which case applied for $x_{\max}$ and $N$ implies that $H_i(t,x_{\max})\le 1$,
- $x_{\max}\le \ell\left( \lfloor N/\ell \rfloor-1\right)+1$ in which case one can compare $H_i(t,x_{\max})$ with $\bar{H}_i(t,y)$ for the smallest $y$ such that $x_{\max}\le y(\ell- 1)+1$ using again.
Then, Proposition \[Prop:BoundMax\] is a direct consequence of the following bound on the averages of $H_i$.
For any $a >0$, there exists $a' >0$ such that for all $N$ large enough $$\sup_{t\ge t_0} \max_{i\in\{1,2\}} {\mathbb{P}}\Big( \max_{y\in {\llbracket}1, N/\ell {\rrbracket}} |\bar{H}_i(t,y)| > k^{\frac12 + a}\Big) \le e^{- k^{a'}}\;.$$
We treat in details the bound of $\bar{H}_1$, since the bound of $\bar{H}_2$ follows from the same arguments. Using a decomposition of ${\lambda}^{\frac{k-N}{2}} H_1(t,\cdot)$, which is a solution of , on the basis of eigenfunction of the Laplacian formed by $\sin(i \pi \cdot)$, $i=1,\ldots,N-1$, we obtain the following expression for the mean $${\mathbb{E}}[H_1(t,x)]= {\lambda}^{\frac{N-k}{2}} \sum_{i=1}^{N-1} \frac2{N} e^{-\gamma_i t} f^{(i)}_{N,k}(\wedge) \sin\big( \frac{i \pi x}{N}\big),$$ and the fluctuation around it $$\begin{gathered}
\label{2dterm}
H_1(t,x)- {\mathbb{E}}[H_1(t,x)]\\
={\lambda}^{\frac{N-k}{2}} \sum_{i=1}^{N-1} \frac{2} {N} \big(f^{(i)}_{N,k}(h^\wedge_t)-e^{-\gamma_i t} f^{(i)}_{N,k}(\wedge)\big) \sin\big(\frac{i \pi x}{N}\big).\end{gathered}$$ We bound separately the contributions to $\bar{H}_1$ coming from these two terms. We start with the mean. Since ${\lambda}^{\frac{1}{2}\wedge(y)} \ge a_{N,k}(y) \ge{\lambda}^{\frac{1}{2} \vee(y)}$ for every $y\in{\llbracket}0,N{\rrbracket}$, we have (recall ) $$\begin{aligned}
|f^{(i)}_{N,k}(\wedge)| \le \sum_{y=1}^{N-1} \frac{{\lambda}^{\frac12 \wedge(y)} - a_{N,k}(y)}{{\lambda}-1}
\le \sum_{y=1}^{N-1} \lambda^{\frac12 \wedge(y)}\,\frac{\wedge(y)-\vee(y)}{2}
\le \lambda^{\frac{k}{2}} k N\;.\end{aligned}$$ Since $\lambda^{k/2}$ is negligible compared to any power of $k$, we deduce that for all $a>0$ and all $t\ge t_{0}$ we have for all $N$ large enough $$\sup_{x\in {\llbracket}0,N{\rrbracket}} \left|\sum_{i=1}^{N-1} \frac2{N} e^{-\gamma_i t} f^{(i)}_{N,k}(\wedge) \sin\big(i \pi \frac{x}{N}\big)\right| \le k^{(1+a)/2}
\sum_{i=1}^{N-1} e^{(\gamma_1-\gamma_i) t_0}\;.$$ Notice that there exists $c>0$ such that for all $i\ge 2$ and all $N$ large enough $$\label{Eq:BoundGamma} \gamma_i - \gamma_1 \ge c \frac{i^2}{N^2}\;.$$ In addition, we have $N^2 \operatorname{\mathrm{gap}}_N \ll (\log k)^2$ by the small bias assumption , so that we get for $i\ge 2$ $$e^{(\gamma_1-\gamma_i) t_{0}} \le e^{- c \frac{i^2}{N^2} \frac{\log k}{2 \operatorname{\mathrm{gap}}_N}} \le e^{-c' \frac{i^2}{\log k}} \le e^{-c' \frac{i}{\log k}}\;,$$ so that for all $N$ large enough we have $$\sum_{i=2}^{N-1} e^{(\gamma_1-\gamma_i) t_{0}} \le \sum_{i=2}^{N-1} e^{-c' \frac{i}{\log k}} \le C \log k \;.$$ Recall that $\ell = \lceil N/k^2 \rceil $. Putting everything together and using assumption , we get that given $a>0$ for $N$ sufficiently large and all values of $y$ we have $$\label{stimean}
{\mathbb{E}}[\bar H_1(t,y)] =
\frac{ 2{\lambda}^{\frac{N-k}{2}} }{N\ell} \Big|\sum_{x=\ell(y-1)+1}^{\ell y}\sum_{i=1}^{N-1} e^{-\gamma_i t} f^{(i)}_{N,k}(\wedge)
\sin\big(i \pi \frac{x}{N}\big)\Big|
\le \frac{1}{2} k^{\frac12+a} \;.$$
We turn to the contribution coming from the second term . To that end, we rewrite it in the form $$\bar H_1(s,y)-{\mathbb{E}}\left[ \bar H_1(s,y) \right] = {\lambda}^{\frac{N-k}{2}}\sum_{i=1}^{N-1} \big(f^{(i)}_{N,k}(h_t^\wedge) - e^{-\gamma_i s}f^{(i)}_{N,k}(\wedge)\big) \Phi_{y,i}\;.$$ where (the second expression being obtained by summation by part) $$\begin{gathered}
\Phi_{y,i} = \frac2{N \ell } \sum_{x=y(\ell-1)+1}^{y\ell} \sin\big( \frac{i\pi x}{N}\big)\\
= \frac1 {N \ell \sin \left(\frac{i\pi }{2N}\right)} \left[\cos\left( \frac{[2y(\ell-1)+1]i\pi }{2N}\right)-
\cos\left( \frac{[2y\ell+1]i\pi }{2N}\right) \right]\;.\end{gathered}$$ Note that for all $N\ge1$, $y$ and $i$ we have $$\big|\Phi_{y,i}\big| \le 2\min\left( \frac 1 N, \frac{1}{ i \ell }\right) \;.$$ Now let us fix $t$ and $y$ and introduce the martingale $$N_s^{(t,y)} = {\lambda}^{\frac{N-k}{2}} \sum_{i=1}^{N-1} e^{\gamma_i (s-t)} \big(f^{(i)}_{N,k}(h_s^\wedge) - e^{-\gamma_i s}f^{(i)}_{N,k}(\wedge)\big) \Phi_{y,i}\;,\quad s\in [0,t]\;.$$ which satisfies $$N_0^{(t,y)}=0 \text{ and } N^{(t,y)}_t=\bar H_1(t,y)-{\mathbb{E}}\left[ \bar H_1(t,y) \right].$$ We wish to apply Lemma \[lem:expobd\] to the martingale $N^{(t,y)}$: the maximal jump rate of this process is bounded by $k$ and the maximal amplitude of the jump (cf. the notations introduced in Appendix \[lapC\]) satisfies $$\forall s \in [0,t], \quad S(s)\le {\lambda}^N \sum_{i=1}^{N-1} e^{\gamma_i(s-t)} \big|\Phi_{y,i}\big|\;.$$ Using that as a consequence of we have $\gamma_i\ge c i^2 N^{-2}$ for all $i\ge 1$ for some $c>0$, we deduce that there exist some constants $C, C'>0$ such that $$\begin{aligned}
&\int_0^t S(s)^2 ds\le {\lambda}^{2N} \sum_{i,j=1}^{N-1} \frac{1}{\gamma_i+\gamma_j}|\Phi_{y,i} \Phi_{y,j}|\\
&\le C \lambda^{2N} \Big(\sum_{1\le i \le j \le \frac{N}{\ell}} \frac1{i^2+j^2} + \sum_{1\le i \le \frac{N}{\ell} < j} \frac{N}{j\ell} \frac1{i^2+j^2} + \sum_{\frac{N}{\ell} < i \le j} \frac{N^2}{i j\ell^2} \frac1{i^2+j^2}\Big)\\
&\le C' \lambda^{2N} \log k\;,\end{aligned}$$ Consequently, setting $\gamma= k^{-\frac{1}{2}-2a}$ and using the fact that from we have $\gamma \,C'\lambda^{2N} \log k < 1$ for $N$ sufficiently large, we apply and obtain $$\begin{gathered}
{\mathbb{P}}(N_t^{(t,y)} > \frac 1 2 k^{\frac {1}{2} +a})
\le {\mathbb{E}}[e^{\gamma N_t^{(t,y)}}]e^{-\frac 1 2 \gamma k^{\frac{1}{2}+a}} \\
\le e^{C' e \gamma^2 k {\lambda}^{2N} \log k -\frac 1 2 \gamma k^{\frac{1}{2}+a}}\le e^{-\frac{1}{4}k^{-a}} \;.\end{gathered}$$ A similar same computation for $N_t^{(t,y)} <- \frac 1 2 k^{\frac {1}{2} +a}$ and a union bound yield $$\begin{gathered}
{\mathbb{P}}(\sup_{y \in {\llbracket}1, N/\ell {\rrbracket}} |\bar H_i(t,y)-{\mathbb{E}}[ H_i(t,y)]| > \frac 1 2 k^{\frac{1}{2}+a})\\
={\mathbb{P}}(\sup_{y \in {\llbracket}1, N/\ell {\rrbracket}} |N_t^{(t,y)}| > \frac 1 2 k^{\frac{1}{2}+a})
\le C k^2 e^{-\frac{1}{4}k^{-a}},
\end{gathered}$$ which combined with allows to conclude.
A monotone grand coupling {#Appendix:Coupling}
=========================
The construction below is similar to the one detailed in [@Lac16 Section 8.1] in the symmetric case. We consider a collection of independent Poisson clock processes ${\mathcal{P}}^{(i,\ell)}$ and ${\mathcal{Q}}^{(i,\ell)}$ with rate $p$ and $q$ respectively where $i\in{\llbracket}1 ,N{\rrbracket}$ and $\ell \in {\llbracket}-N,\ldots,N{\rrbracket}$: For each $(i,\ell)$, ${\mathcal{P}}^{(i,\ell)}$ resp. ${\mathcal{Q}}^{(i,\ell)}$ is a random increasing sequence of positive real numbers (or equivalently a random locally finite subset of $(0,\infty)$) whose first term and increments are independent geometric variables of mean $p^{-1}$ resp. $q^{-1}$.
For every $k$ and every $\zeta\in \Omega_{N,k}$, we construct the process $(h_t^{\zeta})_{t\ge 0}$ as follows: The process is càd-làg and may only jump at the times specified by the clock process ${\mathcal{P}}$ and ${\mathcal{Q}}$. We enumerate these Poisson times in increasing order and if $t\in {\mathcal{P}}^{(i,\ell)}$ and if $h^\zeta_{t-}$ displays a local maximum at $i$ and height $\ell$, that is if $$h^\zeta_{t-}(i) = \ell = h^\zeta_{t-}(i-1)+1 = h^\zeta_{t-}(i+1)+1\;,$$ then we flip it downwards to a local minimum by setting , $h^\zeta_t(i) := h^\zeta_{t-}(i) - 2$, and $h^{\zeta}_t(j)=h^{\zeta}_{t-}(j)$ for $j\ne i$. A similar transition occurs if $Q(i,\ell)$ rings and if $h^\zeta_{t-}$ displays local minimum at $i$ and height $\ell$.\
It is simple to check that under this construction, $h^\zeta$ indeed evolves according to the right dynamics, and that monotonicity is preserved.
Diffusion bounds for continuous-time supermartingales {#sec:diffu}
=====================================================
In this section, we assume that $(M_t)_{t\ge 0}$ is a pure-jump supermartingale with bounded jump rate and jump amplitude. This implies in particular that, $M_t$ is square integrable for all $t>0$. With some abuse of notation, we use the notation $\langle M_\cdot \rangle_t$ for the predictable bracket associated with the martingale ${\widetilde}M_t=M_t-A_t$ where $A$ is the compensator of $M$.
\[prop:solskjaer\] Let $(M_t)_{t\ge 0}$ be as above
- Set $\tau= \inf\{ t\ge 0 \ : \ M_t=0\}$. Assume that $M_t$ is non-negative and that, until the absorption time $\tau$, its jump amplitude and jump rate are bounded below by $1$. Then we have for any $a\ge 1$ and all $u>0$ $$\label{lopes}
{\mathbb{P}}[ \tau \ge a^2 u \ | \ M_0\le a ]\le 4 u^{-1/2}.$$
- Given $a \in{\mathbb{R}}$ and $b\le a$, we set $\tau_{b}:=\inf\{ t\ge 0 \ : \ M_t\le b\}$. If the amplitude of the jumps of $(M_t)_{t\ge 0}$ is bounded above by $a-b$, we have for any $u\ge 0$ $${\mathbb{P}}[ \langle M_\cdot \rangle_{\tau_b}\ge (a-b)^2 u \ | \ M_0\le a ]\le 8 u^{-1/2}.$$
The important building block for the proof of the above proposition is the following result.
\[lem:supersub\] Let $(M_t)_{t\ge 0}$ be as above
- If the amplitude of the jumps of $(M_t)_{t\ge 0}$ and the jump rate are bounded below by $1$ then for all ${\lambda}\in (0,1)$, $$\left(e^{-{\lambda}M_t-\frac{{\lambda}^2 t}{4}}\right)_{t\ge 0}$$ is a submartingale
- If the amplitude of the jumps of $M_t$ is bounded above by $a$ then for any ${\lambda}\in (0,a^{-1})$ we have $$\left( \exp\left(-{\lambda}M_t- \frac{{\lambda}^2}{4} \langle M_\cdot \rangle_t\right)\right)_{t\ge 0}$$ is a submartingale.
The result only needs to be proved for $u\ge 4$. Without loss of generality one can assume for the proof of both statements that $P[ M_0\le a ]=1$ and for the second one that $b=0$. We set ${\lambda}=2 a^{-1} u^{-1/2}$.
For $(i)$, a direct application of the Martingale Stopping Theorem to the submartingale of Lemma \[lem:supersub\] yields: $${\mathbb{E}}\left[e^{-\frac{{\lambda}^2\tau}{4}}\right]\ge {\mathbb{E}}[e^{-{\lambda}M_0 }]\ge e^{-{\lambda}a}=e^{-2u^{-1/2}}.$$ On the other hand one has $${\mathbb{E}}\left[e^{-\frac{{\lambda}^2\tau}{4}}\right]\le 1-(1-e^{-\frac{{\lambda}^2}{4} a^2u}){\mathbb{P}}[\tau \ge a^2 u ] \le 1-\frac{1}{2}{\mathbb{P}}[\tau \ge a^2 u].$$ The combination of the two yields $${\mathbb{P}}[\tau \ge a^2 u]\le 2(1-e^{-2u^{-1/2}})\le 4 u^{-1/2}.$$
For $(ii)$, the arguments of the previous case apply almost verbatim if one replaces $\tau$ by $T:=\langle M_\cdot \rangle_{\tau}$. The only thing one has to take into account is that $M_{\tau}$ is not necessarily equal to $0$, but the assumption on the amplitude of jumps yields $M_{\tau}\ge -a$. The Martingale Stopping Theorem gives us $${\mathbb{E}}\left[e^{-\frac{{\lambda}^2 T}{4}}\right]\ge e^{-{\lambda}a } {\mathbb{E}}\left[e^{-{\lambda}M_{\tau}-\frac{{\lambda}^2 T}{4}}\right]\ge e^{-{\lambda}a} {\mathbb{E}}[e^{-{\lambda}M_0}] \ge e^{-4u^{-1/2}}\;.$$ Repeating the rest of the computation yields $${\mathbb{P}}[T \ge a^2 u]\le 2(1-e^{-4u^{-1/2}})\le 8 u^{-1/2}\;.$$
Until the end of the proof, we write ${\mathbb{E}}_t$ for the conditional expectation given $(M_s)_{s\le t}$.\
Case (i). Take ${\lambda}\in (0,1)$. The submartingale identity we need to prove can be written as follows $$\forall s, t\ge 0, \quad \log {\mathbb{E}}_t\Big[e^{-{\lambda}(M_{t+s}-M_t)}\Big]\ge s\frac{{\lambda}^2}{4},$$ Taking derivative, we deduce that it suffices to prove that for all $t,s\ge 0$ we have $$\label{toprov}
\lim_{h\downarrow 0} \frac1{h}{\mathbb{E}}_t\Big[e^{-{\lambda}( M_{t+s+h}-M_t)}-e^{-{\lambda}(M_{t+s}-M_t)}\Big] \ge \frac{{\lambda}^2}{4} {\mathbb{E}}_t\Big[e^{-{\lambda}(M_{t+s}-M_t)}\Big]\;.$$ Notice that for all $x\in {\mathbb{R}}$ $$\label{Eq:ExpoTaylor}
e^{-x}+x-1\ge \frac{\min(1,x^{2})}{4}\;.$$ Thus, using the supermartingale property of $M$ we have for all ${\lambda}\in(0,1)$ $$\begin{aligned}
&{\mathbb{E}}_t\Big[e^{-{\lambda}(M_{t+s+h}-M_t)}-e^{-{\lambda}(M_{t+s}-M_t)}\Big]\\
&= {\mathbb{E}}_t\Big[e^{-{\lambda}(M_{t+s}-M_t)}{\mathbb{E}}_{t+s}\big[e^{-{\lambda}(M_{t+s+h}-M_{t+s})}-1\big]\Big]\\
&\ge {\mathbb{E}}_t\Big[e^{-{\lambda}(M_{t+s}-M_t)}{\mathbb{E}}_{t+s}\big[e^{-{\lambda}(M_{t+s+h}-M_{t+s})}+{\lambda}(M_{t+s+h}-M_{t+s})-1\big]\Big]\\
&\ge \frac{{\lambda}^2}{4}{\mathbb{E}}_t\Big[e^{-{\lambda}(M_{t+s}-M_t)}{\mathbb{E}}_{t+s}\big[\min(1,(M_{t+s+h}-M_{t+s})^2)\big]\Big]\;.\end{aligned}$$ The assumption on the jump rates and the jump amplitudes yield $$\liminf_{h\to 0} \frac1{h}{\mathbb{E}}_{t+s}\big[\min(1,(M_{t+s+h}-M_{t+s})^2)\big]\ge 1,$$ so that Fatou’s Lemma concludes the proof.
Case (ii). We can assume without loss of generality that $a=1$. Here again, taking the derivative of the submartingale identity that we want to establish, it suffices to prove that for all $t,s \ge 0$ we have $$\begin{aligned}
\liminf_{h\downarrow 0} &\frac1{h}{\mathbb{E}}_t\Big[e^{-{\lambda}M_{t+s+h} - \frac{{\lambda}^2}{4}\langle M_\cdot\rangle_{t+s+h}} -e^{-{\lambda}M_{t+s}- \frac{{\lambda}^2}{4}\langle M_\cdot\rangle_{t+s}} \Big]
\ge 0\;.\end{aligned}$$ Taking the conditional expectation w.r.t. $M_{t+s}$, we see that it suffices to prove the existence of some deterministic constant $C>0$ such that $$\label{Eq:subderiv} {\mathbb{E}}_{t+s}\Big[e^{-{\lambda}(M_{t+s+h}-M_{t+s}) - \frac{{\lambda}^2}{4}(\langle M_\cdot\rangle_{t+s+h}-\langle M_\cdot\rangle_{t+s})} -1 \Big] \ge -C h^2\;,$$ for all $h$ small enough.\
Without loss of generality, we can assume that $t+s = 0$ and $M_0=0$. Recall that ${\widetilde}M\ge M$ is the martingale which is obtained by subtracting the (negative) compensator. Thus $$\begin{aligned}
e^{-{\lambda}M_{h} - \frac{{\lambda}^2}{4} \langle M_\cdot\rangle_{h}} -1 &\ge e^{-{\lambda}{\widetilde}{M}_{h} - \frac{{\lambda}^2}{4} \langle M_\cdot\rangle_{h}} -1\\
&\ge \big(1-{\lambda}{\widetilde}M_{h}+ \frac{1}{4}\min(1, {\lambda}^{2} {\widetilde}M_{h}^2)\big) \big(1- \frac{{\lambda}^2}{4} \langle M_\cdot\rangle_{h}\big)-1 \\
&\ge -{\lambda}{\widetilde}M_{h}+ \frac{{\lambda}^2}{4}\big( {\widetilde}M_{h}^2- \langle M_\cdot\rangle_{h}\big)
- \frac{{\lambda}^2}{4}\big({\widetilde}M_{h}^2-{\lambda}^{-2}\big)_+ \\
&\quad+ \frac{{\lambda}^2}{4}\langle M_\cdot\rangle_{h} \big({\lambda}{\widetilde}M_{h} - \frac{1}{4}\min(1, {\lambda}^2 {\widetilde}M_{h}^2) \big)\;,\end{aligned}$$ so that $$\begin{aligned}
{\mathbb{E}}\Big[e^{-{\lambda}M_{h} - \frac{{\lambda}^2}{4} \langle M_\cdot\rangle_{h}} -1\Big] &\ge {\mathbb{E}}\Big[- \frac{{\lambda}^2}{4}\big({\widetilde}M_{h}^2-{\lambda}^{-2}\big)_+\\
&\quad+ \frac{{\lambda}^2}{4}\langle M_\cdot\rangle_{h} \big({\lambda}{\widetilde}M_{h} - \frac{1}{4}\min(1, {\lambda}^2 {\widetilde}M_{h}^2) \big)\Big]\;.\end{aligned}$$ Take ${\lambda}\in (0,1)$. Our assumptions on the increments and jump rates imply that for some constant $C>0$ we have $$\begin{split}
{\mathbb{E}}\left [ \big({\widetilde}M_{h}^2-{\lambda}^{-2}\big)_+ \right]&\le C h^2,\\ \langle M_\cdot\rangle_{h} &\le C h,\\
\max\left( {\mathbb{E}}\left[ |{\widetilde}M_{h}| \right], {\mathbb{E}}\left[ \min(1, {\lambda}^2 {\widetilde}M_{h}^2) \right]\right)&\le Ch,
\end{split}$$ (the compensator being of order $h$ the estimates for ${\widetilde}M$ can be deduced from that for $M$), which allows to conclude that holds.
Exponential moments of continuous-time martingales {#lapC}
==================================================
Let $(M_t)_{t\ge 0}$ be a martingale defined as a function of a continuous time Markov chain on a finite state space $$M_t=f(t,X_t)\;,$$ where $f$ is differentiable in time. We let $B$ denote the maximal jump rate for $X$ and let $S(t)$ denote the maximal amplitude for a jump of $M$ at time $t$: $$S(t):=\max_{\xi\sim \xi'} |f(t,\xi)-f(t,\xi')|\;.$$
\[lem:expobd\] For any ${\lambda}> 0$ we have $${\mathbb{E}}\left[ e^{{\lambda}M_t} \right] \le \exp \left( B\int^t_0 \left[e^{{\lambda}S(s)}-{\lambda}S(s)-1\right]{\,\text{\rm d}}s \right).$$ In particular if ${\lambda}S(t)\le 1$ for all $t\ge 0$ then we have $$\label{Eq:ExpoBdMgale}
{\mathbb{E}}\left[ e^{{\lambda}M_t} \right] \le \exp \left( B e {\lambda}^2 \int^t_0 S^2(s) {\,\text{\rm d}}s \right).$$
We are going to show that for all $t\ge 0$ $$\partial_t \log {\mathbb{E}}\left[ e^{{\lambda}M_t} \right]= \frac{\partial_t {\mathbb{E}}\left[ e^{{\lambda}M_t} \right]}{{\mathbb{E}}\left[ e^{{\lambda}M_t} \right]}\le B [e^{{\lambda}S(t)}-{\lambda}S(t)-1].$$ To that end, it is sufficient to show that almost surely $$\partial_s {\mathbb{E}}\left[ e^{{\lambda}(M_{t+s}-M_t)} - 1 \ | \ {\mathcal{F}}_t \right] |_{s=0} \le B [e^{{\lambda}S(t)}-{\lambda}S(t)-1].$$ We let $\Delta_s M=M_{t+s}-M_t$ denote the martingale increment and as in the previous section write ${\mathbb{E}}_t$ for the conditional expectation w.r.t. $M_t$ . By the martingale property, we have $$\begin{aligned}
{\mathbb{E}}_t \left[ e^{{\lambda}\Delta_s M}-1 \right]&={\mathbb{E}}_t \left[ e^{{\lambda}\Delta_s M}-{\lambda}\Delta_s M-1 \right] \le {\mathbb{E}}_t \left[ e^{{\lambda}|\Delta_s M|}-{\lambda}|\Delta_s M|-1 \right].\end{aligned}$$ Note that $|\Delta_s M|$ is stochastically dominated by $$\left[\max_{u\in[t,t+s]} S(u)\right] {\mathcal{W}}+ s\times \max_{u\in[t,t+s]} \| \partial_u f(u, \cdot)\|_{\infty}$$ where ${\mathcal{W}}$ is a Poisson variable of parameter $B s$. As $S$ is Lipshitz we conclude that $${\mathbb{E}}_t \left[ \frac{e^{{\lambda}|\Delta_s M|}-{\lambda}|\Delta_s M|-1}{s} \right]\le B[e^{{\lambda}S(t)}-{\lambda}S(t)-1]+ c s.$$
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This paper addresses the question of the observations to be performed with the Optical Monitor (OM) of the X-ray Multi-Mirror Satellite (XMM) under several aspects. First, we discuss XMM-OM’s photometric system and its colour transformations towards the standard $U\!BV$ system. Second, we establish a set of procedures to determine the temperature and the amount of interstellar absorption affecting the observed stars. Last, we address the possibility of isolating quasars in multidimensional colour diagrams based on the XMM-OM filter set.'
author:
- 'P. Royer [^1]'
- 'J. Manfroid [^2]'
- 'E. Gosset [^3]'
- 'J.-M. Vreux'
date: 'Received date; accepted date'
title: 'Simulations of Observations with the Optical Monitor of the X-Ray Multi-Mirror Satellite'
---
Introduction
============
Thanks to its unprecedented large X-ray collecting power, the XMM satellite (for X-ray Multi-Mirror) is expected to discover a wealth of new X-ray sources. In order to allow a quick and reliable identification of the sources, as well as to enable multiwavelength monitoring of their variability, the satellite is equipped with an optical complement known as the optical monitor, XMM-OM. This particular element is co-aligned with the three X-ray telescopes and consists of a 30 cm Ritchey-Chrétien telescope coupled with a photon-counting detector. The latter consists of a photocathode followed by three micro-channel plates as intensifier and by a tapered fiber-optic bundle connected to a fast scanning CCD. The usable area is made of 256 by 256 physical pixels but a centroiding process locates the events to 1/8 of a pixel thus mimicking a 2048 by 2048 device. The field of view is 17 arcmin by 17 arcmin with a centroided pixel size of 0.5 arcsec. At the end of the OM exposure, the cumulated image is downloaded to the ground. An engineering mode allows to transmit the whole image but the routine science mode necessitates predefined windowing and/or binning. The telescope was designed so that the limiting magnitude would be no less than 24 when working in unfiltered light. More details can be found in Fordham et al. 1992 and Mason et al. 1996.
XMM-OM holds a 6-filter, UV and optical, photometric system. Three bands of the system were designed to match the Johnson’s $U\!BV$ system (Johnson 1955; Bessel 1990). The rectangular profiles of the latter filters render the colour transformations between the XMM-OM and the Johnson systems quite complicated (see Royer & Manfroid 1996 for a discussion on the untransformability of rectangularly shaped filters), or even not always possible. Such colour transformations will nevertheless remain necessary for those who will have to compare XMM-OM observations with ground-based ones. In this paper, we provide the reader with theoretical estimations of these colour transformations.
The XMM-OM filter set comprises three non-standard filters, exploring a wavelength domain (UV) where no extensive observations have been performed to date. Combined with the poor match between the XMM-OM optical filters and the classical ground-based standard photometric systems, this fact enhances the importance of testing our ability to determine some physical properties of the observed stars directly in the natural XMM-OM colour system. In the present paper, we will show that it is possible to estimate both the temperature and the amount of interstellar absorption (or “reddening”) for the hot stars that will be observed with XMM-OM.
The natural XMM-OM colour system can also be used for other investigations. For example, we have explored the possibilities to discriminate quasars from stars in multidimensional colour spaces based on the XMM-OM photometry. Nevertheless, the goal of this paper is not so much to provide the reader with exact analytical relations for these matters as to give qualitative results that will tell him how to observe and treat the data in order to get the best outcome, and what kind of results are to be expected.
In Sect. 2, we present the tools we used for the synthetic photometry. In Sect. 3, we discuss the colour transformations between the XMM-OM optical filters and the Johnson $U$, $B$, $V$ system. We treat the temperature and the interstellar absorption determination in Sects. 4 and 5 whereas the selection of quasar candidates is addressed in Sect. 6. Sect. 7 outlines the main conclusions of our study.
The basic information and hypotheses
====================================
The filters
-----------
All simulations presented hereafter were established on the basis of synthetic photometry. The filter passbands that we used for the XMM-OM photometric system are extracted from the XMM Users’ Handbook (Dahlem & Schartel 1999). Their transmission curves result from the product of the transmission curves and/or sensitivity curves of all physical devices in the XMM-OM light path. Namely, we have the filters themselves, three aluminized mirrors, one window at the entry of the detector and the detector response curve. All these curves originate from the same reference as above. The resulting passbands are plotted in Fig. 1a and their physical characteristics are summarized in Table 1. Throughout the rest of this paper, small characters refer to the XMM-OM filters and capital letters refer to the standard ones (i.e. the $U$, $B$ and $V$ taken from Bessel 1990).
filter $\lambda_{o}$ $\lambda_{\rm max}$ FWHM
---------------- --------------- --------------------- -------------
[*[uvw2]{}*]{} 2070 2000 $\sim\,500$
[*[uvm2]{}*]{} 2298 2210 439
[*[uvw1]{}*]{} 2905 2680 620
[*[u]{}*]{} 3472 3270 810
[*[b]{}*]{} 4334 3980 976
[*[v]{}*]{} 5407 5230 684
: Main characteristics of the XMM-OM filters. The wavelengths are expressed in Å. $\lambda_0$ is the effective wavelength and $\lambda_{\rm max}$ is the position of the maximal transmission.
The XMM-OM filter set consists of 3 UV and 3 optical filters, further denoted by $uvw2$, $uvm2$, $uvw1$ and $u$, $b$, $v$ respectively, in order of increasing central wavelength. The $u$, $b$ and $v$ filters are intended to match the Johnson $U$, $B$ and $V$ (Bessel 1990) filters respectively, but they possess more or less rectangularly shaped passbands, which greatly affects the quality of the match and of the subsequent colour transformations between the two systems. This is especially true for what concerns the (mis)match between the Johnson $U$ and XMM-OM $u$ filters and thus the $(U-u)$ and $(U-B)$ related colour transformations. The XMM-OM $u$ band is characterized by:
- a strong sensitivity in the 3100-3300 Å region, where the ground-based Johnson’s $U$ is essentially blind;
- a reduced sensitivity in the critical Balmer decrement region (3850-4100 Å), leading to untransformability between both bands for stars showing Balmer jump (see Sects. 3.1.1, 3.1.6, 3.1.7). As we show later, $(U-u)$ can be evaluated thanks to $uvw1$.
Fig. 1b allows comparison between XMM-OM and standard $U$, $B$ and $V$ filters.
The UV XMM-OM filters do not correspond to any pre-existing standard system. No transmission measurements below 1800 Å are given in the XMM flight simulator software files, so we added a point with $\sim0\%$ transmission at 1700 Å in the UV filters.
The stellar spectra
-------------------
The spectra we used for the synthetic photometry are those of the Kurucz’ ATLAS 9 atmosphere models (Kurucz 1992), reddened by the Cardelli interstellar extinction model (Cardelli et al. 1989) with an average reddening law ($R_{\rm V}~=~A_{\rm V}/E(B-V)~=~3.1$) when necessary. Except when otherwise stated, the set of spectra we chose mimics an unreddened solar composition main sequence (MS) since these 55 spectra possess the following characteristics: $A_{\rm V}~=~0.$; $[\rm Fe/H]~=~0.$; $\log~g~=~4.$; $3500\,$K$~\leq~T_{\rm eff}~\leq~35000\,$K. Stars we refer to as [*giants*]{} have $\log~g~=~3.5$ (Allen, 1976), those we refer to as [*metal poor stars*]{} have $[\rm Fe/H]~=~-1.5$, typical for the halo of our Galaxy. The temperature interval between two adjacent Kurucz models varies along the tracks: it is $250~$K below $10000$ K, $500$ K between $10000$ K and $13000$ K, $1000$ K between $13000$ K and $35000$ K, and $2500$ K between $35000$ K and $50000$ K.
Throughout the rest of this paper, the MS spectrum with $T_{\rm eff}~=~9500\,$K was attributed a magnitude of zero in all filters.
All colour transformations established on the basis of the Kurucz spectra were compared with their equivalent obtained with the stellar spectral atlas prepared by Fioc & Rocca-Volmerange (1997, hereafter FRV). This atlas comprises 65 spectra of stars of various luminosity classes (3 supergiants, 30 giants and 32 dwarfs) and spectral types ($T_{\rm eff}~\in~[2500,180000]$ K). It was constructed from observed spectra whenever possible (Gunn & Strycker 1983; Heck et al. 1984), from synthetic spectra otherwise (Kurucz 1992; Clegg & Middlemass 1987).
The stellar locus for the quasar fields
---------------------------------------
In Sect. 6, we analyse the ability of the XMM-OM photometric system to discriminate between quasars and stars. Therefore, we have to delimit both the quasar (see Sect. 2.4) and the stellar loci in the multidimensional colour space defined by the XMM-OM system. The stellar locus is defined by all the stars that are potential contaminants of the quasar candidate population. When looking at high galactic latitudes, the field stars are halo main sequence stars. We modelize their colours by integrating the emergent fluxes from Kurucz model atmospheres. We represent halo main sequence stars by the whole range of $T_{\rm eff}$, $\log~g~=~4$ and $[\rm Fe/H]~=~-1.5$. Due to evolution, only the part for which $T_{\rm eff}~\leq~7000\,$K (corresponding to an early F spectral type) is still populated. These latter stars are the major constituents of the field stellar population. The reason why we consider the hotter stars as well is that their colours are close to those of other families of stars. We will use that property in Sect. 6, and discuss it further there.
Disk main sequence stars are also a possible contaminant in a list of quasar candidates. However, only the cool end ($T_{\rm
eff}~\leq~7000\,$K) could be a problem and is taken into account. We chose to represent them by a $\log~g~=~4$ and a solar metallicity. Indeed, the disk is rather thin: it has a typical scale height of 1 kpc, which corresponds to a distance modulus of 10 magnitudes. This means that disk OBA stars are too bright in apparent magnitudes to be mistaken for quasars. Another possible contaminant is constituted by halo giants. We represent them by Kurucz models with the following parameters: $T_{\rm eff}~\leq~7000\, $K, $\log~g~=~3$ and $[\rm Fe/H]~=~-1.5$.
Some hotter stars are also present at high galactic latitudes as trace constituents but since they are bluer, a property they share with low-redshift quasars, they could constitute a strong contaminant in a quasar candidate list. We consider three families. The first one is the Horizontal-Branch (HB) BA stars. These stars have $T_{\rm
eff}~\geq~10000\, $K (up to $\sim 30000\,$K). On the HB, the gravity is strongly dependent on the effective temperature: we chose the dependency law we derived by averaging the ones quoted by Moehler et al. (1999) and by Conlon et al. (1991). According to Moehler et al. (1999), the best fit of Kurucz models to HB star’s spectra is obtained for metal rich chemical compositions. We adopted $[\rm
Fe/H]~=~0.5$. As noticed by Miller & Mitchell (1988), the population of HB stars corresponds to intrinsically bright objects and is observed to tail-off at faint magnitudes due to the finite dimension of the halo.
A second family is constituted of the subdwarfs of spectral type OB (sd OB). Their evolutionary status is still somewhat uncertain but they are usually associated to the Extended Horizontal Branch and to the evolution thereof (see Caloi, 1989), although some authors refer to some of them as being post-AGB objects. We represent them by Kurucz models with $T_{\rm eff}$ between $20000\,$K and $50000\,$K (see Conlon et al. 1991 but also Table 1 of Lenz et al. 1998), $\log~g~=~5$ and solar metallicity.
Finally, the third family is made of degenerate stars, the so-called white dwarfs. Some models of degenerate stars exist but only a few have their emergent flux published. As a first approximation, degenerate stars are known to have $U\!BV$ colours very similar to black bodies. We therefore computed the colours of black bodies. In any case, this characteristic is perhaps not general and does not apply to the UV part of the spectrum. We finally used the emergent fluxes of the pure-hydrogen atmosphere models for degenerate stars of Koester (1999). We restricted ourselves to effective temperatures ranging from $7000$ K to $80000$ K and to $\log~g~=~8.5$. We also integrated the spectra of the four white-dwarf primary spectrophotometric standards described by Bohlin et al. (1995) and the models of Wesemael et al. (1980). Both works are in good agreement (although not necessarily independent) with Koester’s models.
The quasar spectrum
-------------------
The quasar spectra used in Sect. 6 were derived from an average spectrum build from the composite spectra of Zheng et al. (1997) and Francis et al. (1991). The former was used from 310 to 2000 Å and the latter from 2000 to 6000 Å. The match between both spectra is reasonably good, as shown by Zheng et al. (1997, their Fig. 9). It was performed on the continuum windows identified by these authors between 1400 and 2200 Å. The absorption characteristics of the quasar spectra due to the intervening Ly$\alpha$ clouds were established, between Ly$\alpha$ and the Lyman break, on the basis of the concept of Oke & Korycansky (1982) and of the works presented by Irwin et al. (1991), Zuo & Lu (1993) and Warren et al. (1994). For the region below the Lyman break, three absorption models were computed. The first two were adapted from the works of Møller & Jakobsen (1990), Møller & Warren (1991), Warren et al. (1994) and Giallongo & Trevese (1990). The first one only takes into account the hydrogen clouds with column densities inferior or equal to $10^{17}~{\rm cm}^{-2}$ (sometimes called the Ly$\alpha$ forest clouds). It will further be referred to as model A. The second one additionally includes systems with column densities between $10^{17}$ and $10^{20}~{\rm cm}^{-2}$ (sometimes called the Lyman limit systems). It will further be referred to as model B. As an extreme case, we also considered a model where a strong (10 magnitudes) absorption occurs in the Lyman continuum at a redshift very close to the one of the quasar, absorption which persists in the whole observable Lyman continuum. This model will further be referred to as model C. These three models are strongly inspired by those described by Royer (1994) where full details can be found. To fix ideas, it is worth noticing that the higher normalization of Madau (1995) puts his model between our models B and C.
It must be clear that the quasar spectrum defined above is only representative of an average quasar and that some individual quasar spectra deviate strongly from it. The true population of quasars will exhibit some dispersion around the characteristics of this quasar. It is well known (see e.g. Francis et al. 1991) that the population of quasars displays a variety of power-law flux distributions and that the emission-line equivalent widths vary from one object to the other both in a systematic way (the Baldwin effect) and in a random way. On the basis of the power-law index dispersion reported, e.g. by Francis et al. (1991), we expect this effect to spread the colours of quasars by $\pm~0.3$ mag around our tracks. In addition, particular realizations of the distribution of high column density clouds along some line of sight could induce strong deviations from the mean behaviour.
Colour transformations
======================
Colour transformations between a physical and a standard filter set always are a source of photometric errors (see e.g. Sterken & Manfroid 1992 or Royer & Manfroid 1996 and references therein). In the particular case of the XMM-OM optical filters, the important discrepancies between the standard and non-standard $U$, $B$ and $V$ filters renders the choice of adequate transformations even more critical. In order to avoid additional errors introduced by subsequent colour computations, we established colour transformations for both single filters and colour indices.
-- ------------------------------------------ -- -- -- -- --
*U–u & *B–b & *V–v & *U–B & *B–V & *U–V\
*u–b & 202 & 102 & & 102 & &\
*uvw1–u & 222 & & & & &\
*b–v & & 211 / 3 & 111 / 2 & & 1 &\
*uvw1–b & & & & 102 & &\
*u–v & & & & & & 101 / 1\
*uvw1–v & & & & & & 102 / 1\
************
-- ------------------------------------------ -- -- -- -- --
: Summary of the colour transformations established in this paper. Capital letters refer to Johnson $U$, $B$ and $V$ filters. A one, two or three digit code [$i_1,...,i_n$]{} characterizes the transformation: $n$ is the number of temperature domains that have to be considered; $i_k$ is the order of the transformation in the $k^{\rm th}$ domain (0 means no transformation is possible), e.g. 102 indicates that one has to divide the temperature domain in 3 distinct parts in which respectively linear, no and second order transformations are possible. The numbering of the domains is going from the hottest to the coolest stars.
We restricted ourselves to a limited set of physically meaningful transformations, i.e. between equivalent or neighbouring filters in both filter sets. We nevertheless introduced transformations to the ($U-V$) index in order to get colour transformations for colour indices that avoid the $b$ filter, which is the one for which the XMM-OM detector could be the most easily affected by saturation (XMM Users’ Handbook, eds. Dahlem & Shartel 1999). We also established various transformations in which the $uvw1$ filter supplants the $u$ filter. Indeed, contrary to the $u$ and $U$ filters, the $uvw1$ filter is defined on the short wavelength side of the Balmer jump. This property renders the transformations based on $uvw1$ far better than those based on the neighbouring $u$ filter, i.e. they are more linear and essentially possible on a wider temperature range. As expected (Sect. 2.1), the $uvw1$ band can be used to correct the ($U-u$) index in the critical domain where the Balmer jump is important (see Sect. 3.1.2).
The colour transformations presented below being established on synthetic spectra, it is obvious that they should be refined through actual observations of suitable standard fields with the satellite and from the ground. The main interest in the transformations given here is that they inform us about the kind of relations that are possible, and their validity domain.
The main characteristics of the considered colour transformations are summarized in Table 2. Two of them are illustrated in Figs. 2 and 3. The analytical forms of the transformations are given below, sorted by categories. Though some formal error bars are smaller, we did not indicate uncertainties smaller than 0.001 mag in the relations given below.
Solar composition main sequence stars
-------------------------------------
### ($U-u$) vs ($u-b$)
$4000$ K $\leq~T_{\rm eff}~\leq~5500$ K (K7V - G8V)
($u-b$) $\in~[0.20;1.50]$\
$U-u = 0.049 + 0.002 (u-b) - 0.097 (u-b)^2 \pm 10^{-3}$
$5500$ K $\leq~T_{\rm eff}~\leq~10500$ K (G8V - B9V)
($u-b$) $\in~[-0.15;0.20]$\
No possible transformation.
$10500$ K $\leq~T_{\rm eff}~\leq~35000$ K (B9V - 08V)
($u-b$) $\in~[-1.40;-0.15]$\
$U-u = -0.016 - 0.268 (u-b) - 0.043 (u-b)^2 \pm 10^{-3}$
### ($U-u$) vs ($uvw1-u$)
$3500$ K $\leq~T_{\rm eff}~\leq~7250$ K (M3V - F0V)
($uvw1-u$) $\in~[0.35;1.75]$\
$U-u = -0.009 + 0.212 (uvw1-u) - 0.183 (uvw1-u)^2 \pm 4. 10^{-3}$
$7500$ K $\leq~T_{\rm eff}~\leq~10500$ K (A8V - B9V)
($uvw1-u$) $\in~[-0.10;0.30]$\
$U-u = 0.001 - 0.158 (uvw1-u) + 0.860 (uvw1-u)^2 \pm 2.10^{-3}$
$10500$ K $\leq~T_{\rm eff}~\leq~35000$ K (B9V - O8V)
($uvw1-u$) $\in~[-0.75;-0.10]$\
$U-u = -0.019 - 0.485 (uvw1-u) - 0.128 (uvw1-u)^2 \pm 2.10^{-3}$
### ($B-b$) vs ($b-v$)
$4000$ K $\leq~T_{\rm eff}~\leq~4750$ K (K7V - K3V)
($b-v$) $\in~[1.00;1.30]$\
$B-b = -0.074 - 0.010 (b-v) \pm 10^{-3}$
$4750$ K $\leq~T_{\rm eff}~\leq~10500$ K (K3V - B9V)
($b-v$) $\in~[-0.05;1.00]$\
$B-b = -0.085 (b-v) \pm 10^{-3}$
$10500$ K $\leq~T_{\rm eff}~\leq~35000$ K (B9V - O8V)
($b-v$) $\in~[-0.30;-0.05]$\
$B-b = -0.073 (b-v) -0.065 (b-v)^2 \pm 10^{-3}$
A third order fit can be applied to the whole interval:
$4750$ K $\leq~T_{\rm eff}~\leq~35000$ K (K3V - O8V)
($b-v$) $\in~[-0.30;1.00]$\
$B-b = -0.070 (b-v) - 0.040 (b-v)^2 + 0.026 (b-v)^3 \pm 10^{-3}$
### ($B-b$) vs ($u-b$)
$4250$ K $\leq~T_{\rm eff}~\leq~5500$ K (K6V - G8V)
($u-b$) $\in~[0.20;1.30]$\
$B-b = -0.056 - 0.060 (u-b) + 0.028 (u-b)^2 \pm 10^{-3}$
$5500$ K $\leq~T_{\rm eff}~\leq~10500$ K (G8V - B9V)
($u-b$) $\in~[-0.15;0.20]$\
No possible transformation.
$10500$ K $\leq~T_{\rm eff}~\leq~35000$ K (B9V - O8V)
($u-b$) $\in~[-1.40;-0.15]$\
$B-b = 0.003 - 0.009 (u-b) \pm 10^{-3}$
### ($V-v$) vs ($b-v$)
$4000$ K $\leq~T_{\rm eff}~\leq~5250$ K (K7V - K0V)
($b-v$) $\in~[0.85;1.30]$\
$V-v = 0.061 - 0.136 (b-v) \pm 10^{-3}$
$5250$ K $\leq~T_{\rm eff}~\leq~8750$ K (K0V - A3V)
($b-v$) $\in~[0.07;0.85]$\
$V-v = 0.0025 - 0.068 (b-v) \pm 10^{-3}$
$8750$ K $\leq~T_{\rm eff}~\leq~30000$ K (A3V - B0V)
($b-v$) $\in~[-0.30;0.07]$\
$V-v = -0.001 - 0.033 (b-v) \pm 10^{-3}$
A second order fit can be applied to the whole interval:
$4000$ K $\leq~T_{\rm eff}~\leq~30000$ K (K7V - B0V)
($b-v$) $\in~[-0.30;1.30]$\
$V-v = -0.001 - 0.041 (b-v) - 0.035 (b-v)^2 \pm 2.10^{-3}$
### ($U-B$) vs ($u-b$)
$4250$ K $\leq~T_{\rm eff}~\leq~5500$ K (K6V - G8V)
($u-b$) $\in~[0.20;1.35]$\
$U-B = 0.120 + 1.087 (u-b) - 0.135 (u-b)^2 \pm 2.10^{-3}$
$5500$ K $\leq~T_{\rm eff}~\leq~10500$ K (G8V - B9V)
($u-b$) $\in~[-0.15;0.20]$\
No possible transformation.
$10500$ K $\leq~T-{\rm eff}~\leq~35000$ K (B9V - O8V)
($u-b$) $\in~[-1.4;-0.15]$\
$U-B = 0.003 + 0.819 (u-b) \pm 5.10^{-3}$
### ($U-B$) vs ($uvw1-b$)
$4000$ K $\leq~T_{\rm eff}~\leq~6500$ K (K7V - F5V)
($uvw1-b$) $\in~[0.20;3.20]$\
$U-B = -0.292 + 0.710 (uvw1-b) - 0.051 (uvw1-b)^2 \pm 1.2\,10^{-2}$
$6500$ K $\leq~T_{\rm eff}~\leq~8750$ K (F5V - A3V)
($uvw1-b$) $\in~[0.10;0.20]$\
No possible transformation.
$8750$ K $\leq~T_{\rm eff}~\leq~35000$ K (A3V - O8V)
($uvw1-b$) $\in~[-2.20;0.10]$\
$U-B = 0.002 + 0.528 (uvw1-b) \pm 8.10^{-3}$
### ($B-V$) vs ($b-v$)
$3500$ K $\leq~T_{\rm eff}~\leq~35000$ K (M3V - O8V)
($b-v$) $\in~[-0.30;1.40]$\
$B-V = 0.002 + 0.997 (b-v) \pm 9.10^{-3}$
### ($U-V$) vs ($u-v$)
$4000$ K $\leq~T_{\rm eff}~\leq~6500$ K (K7V - F5V)
($u-v$) $\in~[0.40;2.80]$\
$U-V = 0.133 + 0.947 (u-v) \pm 2.3\, 10^{-2}$
$6500$ K $\leq~T_{\rm eff}~\leq~8500$ K (F5V - A4V)
($u-v$) $\in~[0.20;0.40]$\
No possible transformation, though the deviation from the general shape of the sequence is very slight (see below for further comments).
$8500$ K $\leq~T_{\rm eff}~\leq~35000$ K (A4V - O8V)
($u-v$) $\in~[-1.70;0.20]$\
$U-V = 0.008 + 0.843 (u-v) \pm 8.10^{-3}$
The general shape of our synthetic main sequence is pretty close to a straight line in the ($U-V$) vs ($u-v$) diagram, so that, if precision is not critical (at most 0.1 mag), one can also use the following relation:
$4000$ K $\leq~T_{\rm eff}~\leq~35000$ K (K7V - O8V)
($u-v$) $\in~[-1.70;2.80]$\
$U-V = 0.100 + 0.930 (u-v) \pm 5.7\,10^{-2}$
### ($U-V$) vs ($uvw1-v$)
$4000$ K $\leq~T_{\rm eff}~\leq~7000$ K (K7V - F1V)
($uvw1-v$) $\in~[0.65;4.50]$\
$U-V = -0.201 + 0.786 (uvw1-v) - 0.029 (uvw1-v)^2 \pm 1.9\,10^{-2}$
$7000$ K $\leq~T_{\rm eff}~\leq~7750$ K (F1V - A7V)
($uvw1-v$) $\in~[0.50;0.65]$\
No possible transformation. The problem is however very slight and seems to be due to the Kurucz synthetic spectra rather than to the filters. The best here is to use the general relation proposed below for the whole temperature range.
$7750$ K $\leq~T_{\rm eff}~\leq~35000$ K (A7V - O8V)
($uvw1-v$) $\in~[-2.45;0.50]$\
$U-V = 0.004 + 0.577 (uvw1-v) \pm 9.10^{-3}$
The same kind of considerations as for the ($U-V$) vs\
($u-v$) case holds concerning a general relation for the whole set of temperatures:
$4000$ K $\leq~T_{\rm eff}~\leq~35000$ K (K7V - O8V)
($uvw1-v$) $\in~[-2.45;4.50]$\
$U-V = 0.023 + 0.596 (uvw1-v) \pm 4.10^{-2}$
Giants
------
Whenever a colour transformation is possible for MS stars, it is generally also satisfactorily obeyed by giant stars. The difference essentially lies in the validity range of the colour transformations. Each time a forbidden zone appears in the temperature domain (i.e. each time there is a zero in Table 2), the transformation relative to the hottest stars remains valid for giant stars down to $\sim 250$ K cooler than the lower temperature bound defined for MS stars. Symmetrically, the transformation relative to the coolest giant stars is only valid from stars $\sim 250$ K cooler than the upper bound of the temperature interval defined for the corresponding MS star colour transformation (see Fig. 3a).
When colour transformations are possible for MS stars on the whole temperature domain, they are generally valid for giant stars as well (Fig. 2).
Metal poor stars
----------------
In this case, the situation is worse. Stars still obey the same relations as MS stars when colour transformations exist on the whole temperature domain. Even in other cases, the hottest stars still obey the same relations, but the range on which these transformations remain valid is now considerably diminished: the “hot” relations now hold only for $T_{\rm eff}~\geq~11500$ K and the “cool” ones for $4750$ K $\geq~T_{\rm eff}~\geq~4000$ K. Cool stars sometimes require different relations. This is true for the ($B-b$) vs ($u-b$) as well as for both ($U-u$) colour transformations (Fig. 3a).
Comparison with observed spectra
--------------------------------
As a validity check, all colour transformations presented above were also derived on the basis of the mainly-observed FRV spectra. When excluding stars cooler than $4000$ K, the comparison between transformations established through Kurucz models and through FRV MS and giant star spectra is excellent, as illustrated in Fig. 3b. The latter ones are of course slightly more dispersed, but the dispersion is fully comparable to what could be expected on the basis of the transformations established in Sect. 3.2. Very small differences occur in some particular colour indices and are quite marginal: the ($B-b$) colour index discriminates between the FRV MS and giant stars, but this only happens below $5000$ K; ($V-v$) is decreased by $\sim~0.01$ mag when calculated on the FRV spectra rather than on the Kurucz spectra; the Kurucz and FRV ($U-B$) and ($U-V$) colour indices are slightly discrepant below $5000$ K, as are the ($U-u$) and the ($uvw1-u$) colour indices below $7000$ K.
As a conclusion, before the full inflight calibration on standard fields is performed and reduced, we recommend to use the relations given above to analyse the first data provided by the XMM-OM.
Reddening and temperature determination
=======================================
In order to discriminate between stars having different reddening and temperature characteristics, a combination of colour indices must be found such that temperature tracks (i.e. curves of constant reddening) do not cross each other in the related colour diagram. In the XMM-OM photometric system, we found only one fully suitable pair of such colour indices: ($u-uvw1$) and ($b-v$). The corresponding colour diagram is shown in Fig. 4. This figure is representative of the weakest reddenings, but the diagram keeps its properties up to $A_{\rm
V}~\sim~13$. Only a slight adaptation of the lowest temperature bound ($9000$ K $\rightarrow~11000$ K) is necessary. We did not plot stars below $9000$ K in Fig. 4 because the temperature tracks of MS stars begin to overlay at this temperature, so that nothing can be said for lower temperatures. Uncertainties on the photometry will anyhow probably hamper any temperature determination below $10000$ K (a 0.05 mag photometric error can bring any $10000$ K star on the $9000$ K star locus). Above that value, the precision on the temperature determination depends on the temperature itself, but is typically of a few thousand Kelvins for colour indices accurate to 0.05 mag and for stars in the middle of the temperature domain. Such accuracy allows a reddening determination with a precision of about $\sim~0.3$ magnitudes in $A_{\rm V}$ at any but the “coolest” temperatures ($<~11000$ K).
The situation is slightly worse when one includes giant stars as well since the temperature tracks for the giants at a given reddening cross those of the less reddened MS stars. Even worse, they pass below the locus of the $9000$ K MS stars at $\sim~10500$ K. As any place below this line can be occupied by stars with various combinations of temperature and reddening, the domain in which a temperature determination is possible for giants is restricted to stars above $\sim11000$ K. As one does not know [*a priori*]{} whether or not a star is a giant, this of course also sheds some uncertainty on the temperature and reddening determination for MS stars below $\sim12-13000$ K, where temperature tracks for MS stars and giants begin to significantly differ from each other.
Other discriminant colour indices than those of Fig. 4 exist in the system, but their combination with other ones only allows to determine the temperature and the reddening of the observed stars on much more restricted ranges of reddening and/or temperature. One can nevertheless design another, independent, reddening determination technique. In Fig. 4, reddening and temperature influence both colour indices plotted on the axes. As we will see now, one can nearly decouple these parameters and obtain better precision on the reddening. To do so, we need to define a colour index that is independent of reddening and another that is proportional to it. The former, that we will call $ic_{\rm 0}$, is defined as $$ic_{\rm 0} = (uvw2-uvw1) - \frac{E(uvw2-uvw1)}{E({u-uvw1)}} (u-uvw1)$$ where $E(x-y)$ stands for the colour excess of the ($x-y$) colour index. $ic_{\rm o}$ will remain a reddening free colour index as long as the colour excess ratio remains constant. This property is verified for stars with $T_{\rm eff}~\geq~9\,000$ K and $A_{\rm V}~\leq~5.5$. Nevertheless, since $A_{\rm 2000}~\sim~3~A_{\rm V}$, four magnitudes of absorption in $V$ correspond to more than ten magnitudes of absorption in the $uvm2$ and $uvw2$ filters. Hence, from now on, we will only consider the $A_{\rm V}~\leq~4$ domain. The mean $E(uvw2-uvw1)/E(u-uvw1)$ colour excess ratio over this range is $-2.83$ ($\pm\,0.27$). It is worth to note that $(uvw2-uvw1)$ and $(u-uvw1)$ are, with $(u-b)$ and $(b-v)$, the only pairs of colour indices allowing the definition of a reddening free index over a reasonable range of stellar parameters.
In order to define the reddening dependent index, that we will call $ic_{\rm red}$, the most obvious choice is the empirical $$ic_{\rm red}~=~uvm2~-~(uvw1~+~uvw2)\,/\,2.$$ This index could be used but it is not really independent from $ic_{\rm 0}$ and its reddening dependence can be ameliorated. Indeed, though the $uvm2$ filter stands on the $2175$ Å absorption bump, the $uvw1$ and $uvw2$ filters are not symmetric with respect to it, and of more importance, the $uvw2$ filter is significantly affected by the $2175$ Å absorption bump too. To refine the choice, we explored a large number of UV magnitude combinations and were finally brought to the conclusion that the best choice in terms of simplicity and dynamics of the index (and hence in terms of accuracy) is $$ic_{\rm red}~=~uvm2~-~uvw1$$
There is in fact a wide variety of reddening-dependent indices in the XMM-OM system and, to give just another very simple one, ($uvw2-uvw1$) is nearly as good as ($uvm2-uvw1$). Even ($b-v$) could be used as reddening-dependent index, so that the whole treatment that we carry on here on the UV filters could be performed on the $ubv$ filters as well. Nevertheless, it is important to note that, although the reddening-free index based on the optical filters is much better than the one based on the ultraviolet filters ($E(u-b)/E(b-v)~=~0.80~\pm~0.02$; compare the error bar with the UV case), the dynamic of the ($b-v$) index is smaller than the one of the ($uvm2-uvw1$) index. Consequently, at equivalent accuracy on the photometry, the reddening determined thanks to the UV filters will be more accurate than the one obtained on the basis of the optical photometry (except for temperatures lower than 10000 K or higher than 25000 K). If UV and optical photometries are available, reddening determinations in both domains should of course be used as a quality check. Indeed, as we will see in Sect. 5, the simultaneous use of visible and UV data allows a control of the consistency of the adopted reddening law.
Although lots of reddening dependent indices exist, none is perfectly independent from temperature and from $ic_{\rm 0}$ so that no simple $A_{\rm V}~=~A_{\rm V}(ic_{\rm red})$ relation can be drawn. Instead, one has to perform a bi-dimensional fit over the ($ic_{\rm
0}$,$ic_{\rm red}$) plane (Fig. 5a) to get the $A_{\rm V}~=~A_{\rm
V}(ic_{\rm 0},ic_{\rm red})$ relation. We do not present this fit here since its detailed analytical form can only be usefully obtained through actual in-orbit satellite calibrations. The reddening dependency of $ic_{\rm red}$ is illustrated in Figs. 5a and 5b. The solid line in Fig. 5b represents a section of the colour diagram shown in Fig. 5a at $ic_{\rm 0}~=~0.6$.
This reddening determination method, based on a reddening free and a reddening dependent index, is more accurate than what can be expected with the procedure outlined in Fig. 4 since error bars of 0.2 and 0.05 mag on [$ic_{\rm 0}$]{} and [$ic_{\rm red}$]{} respectively lead to $\leq~0.15$ mag uncertainty on $A_{\rm V}$. On the other hand, one can see by comparing Fig. 5a and Fig. 4 that the uncertainties on $A_{\rm V}$, due to the fact that the temperature tracks for MS and giant stars diverge at the coolest temperatures in the $(u-uvw1)$ vs $(b-v)$ colour diagram, are partly removed here for the weakest reddenings. Nevertheless, the colour diagram shown in Fig. 4 remains necessary since the $A_{\rm V}~=~A_{\rm V}(ic_{\rm
0},ic_{\rm red})$ fit is not valid for all temperatures, so that we need an independent determination of temperature, precisely allowed by Fig. 4.
Non-standard reddening laws
===========================
The reddening determination method presented in the previous section does not take into account the possibility of a non-standard reddening law. We will now address the following question: can we determine whether or not the reddening of a star observed with the XMM-OM is anomalous ? To answer this, we integrated the same spectra with alternative values of $R_{\rm V}$. Instead of $R_{\rm V}~=~3.1$, —the standard value for diffuse interstellar medium and average value for the LMC—, we used $R_{\rm V}~=~2.6$ and $R_{\rm V}~=~5.5$ (observed by Cardelli et al. 1989, in the direction of HD204827 and HD37022 respectively). We also considered the peculiar UV absorption law established by Prévot et al. (1984) for the SMC. This reddening law does not show any absorption bump around $2175$ Å. It is specific to the ultraviolet and was thus only applied to the UV filters.
Figs. 6a and 6b show the resulting $(u-b)$ vs $(b-v)$ and $(u-uvw1)$ vs $(b-v)$ colour diagrams in the $R_{\rm V}~=~5.5$ case. Let us consider a $15000$ K star affected by 2 magnitudes of absorption (in $V$) with a reddening law characterized by such a high value of $R_{\rm V}$ and let us see what will happen if we try to interpret the observed colours assuming a classical reddening law with $R_{\rm V}~=~3.1$. In the $(u-b)$ vs $(b-v)$ diagram, the colours mimic those of a slightly hotter star, affected by hardly more than one magnitude of absorption but in $(u-uvw1)$ vs $(b-v)$, the star falls out of the “authorized” range of colours. This discrepancy will clearly reveal that the assumption was wrong and that the star suffers from a peculiar absorption law. Fig. 6 thus reveals that $(u-uvw1)$, $(b-v)$ and $(u-b)$ can be used to determine $T_{\rm eff}$, $A_{\rm V}$ and $R_{\rm V}$ just as Fig. 4 shows that $(u-uvw1)$ vs $(b-v)$ can be used to determine $T_{\rm eff}$ and $A_{\rm V}$ once $R_{\rm V}$ is known.
Of course, peculiar absorption laws also influence [$ic_{\rm 0}$]{} and [$ic_{\rm red}$]{}. As expected from Fig. 6, the colour excess ratios are reddening law dependent. Considering $0~\leq~A_{\rm V}~\leq~4.$, we found the mean value of ${E(uvw2-uvw1)}/{E(u-uvw1)}$ to vary between $-2.83$ and $-3.68$ for $R_{\rm V}~\in~[2.6;5.5]$ (the value for the Prévot et al. (1984) reddening law is $-3.04$). Consequently, the appropriate ([$ic_{\rm 0}$]{},[$ic_{\rm red}$]{}) has to be used once $R_{\rm V}$ has been determined. Fig. 5b shows the influence of the reddening law variations on the ([$ic_{\rm 0}$]{},[$ic_{\rm red}$]{}) diagram through comparison between $ic_{\rm 0}~=~0.6$ sections performed in the various realizations of that diagram.
As a conclusion, regarding the reddening determination, the main advantages of the XMM-OM UV filters are that
- at equivalent photometric accuracy, they allow a better determination of the amount of interstellar extinction than what is achievable with the optical filters only;
- they allow discrimination between standard and non-standard extinction laws; this is of primary importance since interpretation of X-ray data requires knowledge of the column density along the line of sight, what is often estimated through reddening measurements (e.g. Bohlin et al. 1978).
The XMM-OM multicolour space and quasars
========================================
In the present section, we investigate the ability of the XMM-OM photometric system to segregate quasars from stars on the basis of their colours in the multicolour space definable from the set of the different filters used. Therefore, we integrated both the stellar spectra discussed in Sect. 2.3 and the average quasar spectra discussed in Sect. 2.4. The latter ones have been considered at redshifts from 0.0 to 4.4 by steps of 0.1. We start our analysis with the XMM-OM version of the classical ($U-B$) vs ($B-V$) colour diagram.
The basic ($u-b$) vs ($b-v$) diagram
------------------------------------
The ($U-B$) vs ($B-V$) colour diagram is probably one of the most widely used in astronomy. Therefore, Fig. 7 presents the XMM-OM version (($u-b$) vs ($b-v$)) of this diagram. Fortunately, no huge difference appears between the two versions and the XMM-OM colour diagram retains most of the properties of its classical counterpart.
The locus of the theoretical halo main sequence stars is given (filled circles). The classical potential-well shape of the curve outlining the effect of the Balmer continuum is clearly visible. The part below the turn-off represents the stars that are still on the main sequence in the halo of our Galaxy; it is represented by a bold line. This bold line is the locus of the majority of the field stars which are the objects against which we have to perform the basic discrimination when looking for quasars. The locus of cool giants is also given (open circles) and is very similar to the previously discussed one. On the other hand, the disk main sequence (spectral types FGKM), which could also provide some confusion, is visible below the halo main sequence (asterisks). The HB stars (crosses) have colours very similar to the hot part (astronomically non populated) of the halo main sequence. This is also true for the sdOB stars (plus signs). These objects are classical contaminants of the samples of quasar candidates selected on the basis of the $U/B$ excess. Another contaminant are the degenerate stars. The black body line is also given in Fig. 7 and the Koester’s models whose spectrum we integrated fall close to this line. Fig. 7 also exhibits the track of our average quasars as a function of redshift. For redshifts $z$ $\leq$ 2.1, the average quasars are wandering around $u-b$ = $-1.0$ and $b-v$ = 0.15. This is 0.2 magnitude bluer in $u-b$ than in the case of the standard system (see e.g. Fig. 6 of Cristiani & Vio 1990 and Fig. 9 of Moreau & Reboul 1995). This ability to better detect the bluer objects is essentially due to the comparatively high transmission of the $u$ filter below 3200 Å (see Fig. 1b and Sect. 2.1).
The wandering around the mean place is essentially due to the various emission lines entering and going out of the different filters but also partly to the particular shapes of the top of the transmission curves. The region around this mean place is also where one can find degenerate stars (with typical $T_{\rm eff} \sim 15000$ K) and the presently investigated colour combination is not useful in discriminating between both kind of objects. At $z$ = 2.2, the quasars have left their low-redshift location to move almost parallel to the ($u-b$) axis. This is due to the fact that the Ly$\alpha$ emission line is leaving the $u$ filter to enter the $b$ one and is progressively replaced by the Ly$\alpha$ forest. This occurs 0.07 earlier in redshift than in the Johnson standard system.
The tracks of the average quasars are also given for higher redshifts. It should however be clearly stated that for redshifts $z~\geq~2.9$, both the $u$ and the $b$ filters are mainly sampling the Ly$\alpha$ forest and the related quasar location in the bidimensional (2D) colour diagram is highly model dependent. It is interesting to notice that the model B spectra tend to follow the stellar locus whereas model C quasars stay bluer in $b-v$. In any case, the discrimination is essentially possible for low-redshift ($z~\leq~2.2$) quasars and for high-redshift weakly or strongly absorbed objects.
Fig. 8 gives the distance (in this two-dimensional space) between the model A quasar and the stellar locus as a function of redshift (by steps of 0.1, filled circles). The stellar locus adopted here is the whole halo main sequence. Except for degenerate stars, this locus can be considered as representative of most of the other potential contaminants.
Fig. 8 clearly demonstrates the ability to discriminate between non-degenerate stars and quasars with redshifts $z~\leq~2.2$. In the range $z~\in~[2.2, \sim 3.]$ (or more), quasars have $ubv$ colours very similar to stars and the discrimination power will only improve through the use of additional filters. The apparent improvement for redshifts in the range $z~\in~[3.3,3.6]$ is not present for model B quasars: this pinpoints the model dependent character of this particular result.
At $z~>~3.8$, all three filters are essentially in the Ly$\alpha$ forest and the increase in distance is indicative of a potential discrimination but the latter is bound to be highly dependent on the particular realization of the distribution of the Ly$\alpha$ absorbers both in redshift and in density. The completeness of the related sample will be hard to ascertain.
Adding $uvw1$
-------------
Our simulations concerning the $uvw1$ filter clearly indicate that this filter is roughly as sensitive to the Balmer continuum (not to confuse with the Balmer jump) as the $u$ filter. The ($uvw1-b$) colour index of stars has a behaviour very similar to the ($u-b$) one and the locus of stars in a ($uvw1-b$) vs ($b-v$) diagram is very reminiscent of Fig. 7. The ($uvw1-u$) colour index is much less sensitive to the Balmer continuum. It is interesting to notice that in a 2D ($uvw1-u$) vs ($uvw1-v$) colour diagram, the quasars with $z~\leq~1.5$ are perfectly superimposed on the locus of stars. Therefore, this combination is not interesting for low-redshift quasars but quasars with redshifts between 1.6 and 2.1 are moving away from the stellar locus in the same diagram. This is essentially due to the Ly$\alpha$ emission line leaving $uvw1$ for the $u$ filter and to the Ly$\alpha$ forest becoming dominant in $uvw1$. Nevertheless, full exploitation of this phenomenon requires observations in the $b$ filter. This is particularly striking in the ($uvw1-u$) vs ($u-b$) colour diagram (not shown here).
Fig. 9 gives the 2D ($uvw1-u$) vs ($b-v$) colour diagram. At redshifts $z~\leq~1.5$, quasars are wandering around $uvw1-u$ = $-0.4$ and $b-v$ = 0.15. This is slightly aside the non-degenerate stellar locus and the $uvw1$ filter contributes, although weakly, to the star-quasar separation. For redshifts $1.6~\leq~z~\leq~3.0$, the average quasar joins the stellar locus in this 2D diagram of Fig. 9 but it is known to deviate from the stellar locus in the ($uvw1-u$) vs ($uvw1-v$) colour diagram for $1.6~\leq~z~\leq~2.1$. In Fig. 8 is also given the distance between the quasar and the stellar locus in the three-dimensional space ($uvw1-u$) vs ($u-b$) vs ($b-v$). Increasing the number of dimensions of the space always brings an increase of the distance between objects, although the effect is purely geometrical. To test whether or not the added filter brings a strategical contribution due to its location in the wavelength domain, one has to compare the distances reduced to the lower dimension space. Therefore, in Fig. 8, we compare the two-dimensional true distance (in ($u-b$) vs ($b-v$)) to the reduced 3D distance which is the three-dimensional true distance (in ($uvw1-u$) vs ($u-b$) vs ($b-v$)) multiplied by a $\sqrt{2} / \sqrt{3}$ factor.
From Fig.8, it is absolutely clear that the main contribution of the use of $uvw1$ to the discriminating power of the XMM-OM photometry is essentially located at the redshift range 1.6 to 2.1. It is also interesting to notice that low-redshift quasars are wandering in Fig. 9 slightly aside the black body line. However, the degenerate stars do not follow the black body locus but, rather, are again mixed with low-redshift quasars (a typical effective temperature for a white dwarf in the middle of the low-redshift quasar locus is $12000$ K). This suggests that the discrimination between degenerate stars and quasars is bound to remain poor. Beyond $z~=~3.0$, the model A quasars seem to remain out of the stellar locus, and the model C quasars stay bluer in $b-v$. This again depends on the particular behaviour of the Ly$\alpha$ absorbers in the line of sight of the observed quasar.
Adding $uvw2$
-------------
Filter $uvw2$ could also be used to build-up colour diagrams. However, it should be kept in mind that the XMM-OM is not very sensitive in this passband and the precision of the measurement in $uvw2$ could be markedly worse than in any of the other filters. The use of filter $uvw2$ is illustrated in Fig. 10 where the 2D ($uvw2-uvw1$) vs ($b-v$) colour diagram is given. Similarly to the previous case, low-redshift ($z~\leq~0.6$) quasars are wandering at ($uvw2-uvw1$) = $-0.3$ and, of course, ($b-v$) = 0.15. This is slightly out of the stellar locus. However, at $z~\sim~0.7$, the average quasars progressively become redder in ($uvw2-uvw1$) due as usual to the Ly$\alpha$ emission line going from the first filter to the second. The present colour index is expected to be discriminant when the Ly$\alpha$ line is located in the $uvw1$ filter, i.e.roughly for redshifts between 0.8 and 1.6. This is easily seen in the ($uvw2-uvw1$) vs ($uvw1-u$) colour diagram as well as in the ($uvw2-uvw1$) vs ($uvw1-b$) one; this pinpoints the importance of the joint use of the $u$ filter (or perhaps the $b$ one) along with the pair $uvw2$, $uvw1$. Fig. 8 exhibits the reduced ($\sqrt{2} /
\sqrt{4} = 1 / \sqrt{2}$) four-dimensional distance in the 4D ($uvw2-uvw1$) vs ($uvw1-u$) vs ($u-b$) vs ($b-v$) colour space. It is clear that the contribution of the $uvw2$ filter contrasting with the $uvw1$ one is increasing the discrimination power in the redshift range 0.8 to 1.6. This effect could help in generating quasar samples that are more homogeneous in redshift since the use of the new filters alleviates the well-known bias of $U/B$ selected quasar candidates due to the presence of a strong C [iv]{} line in the $B$ filter (at $z
\sim 1.6 - 1.7$). From Fig. 10, it is again clear that the degenerate stars do not follow the black body line and that they still remain a strong contaminant of the samples of quasars (particularly around $T_{\rm eff} \sim 13000$ K).
General considerations
----------------------
From Fig. 8, one can conclude that the XMM-OM filter set is good at discriminating between non-degenerate stars and quasars at low redshifts ($z~\leq~2.2$). This is particularly true in the range 0.8 to 2.1 where the use of the $uvw1$ and $uvw2$ filters allows a significantly better discrimination that is even able to wash out the decrease in efficiency around $z~\sim~1.6 - 1.7$ sometimes exhibited by traditional ($U-B$) vs ($B-V$) surveys. For very low redshifts ($z~<~0.8$), the advantage of this photometric system is less marked. However, one should not forget that Fig. 8 gives the reduced distance. Indeed, the minimum true distance between the quasar and the stellar locus is, in the 4D space of Sect. 6.3, somewhat larger than 0.35 magnitudes (occuring at $z=0.5$); this already implies a real possibility of segregation. For redshifts between 2.3 and 3.5, the selection is essentially inefficient, as for ground-based surveys neglecting the use of the $R$ and $I$ filters (for example). It is beneficial to recall that XMM-OM was originally designed with a red optical path that has been abandoned in the meantime. Beyond $z~=~3.5$, the average quasar is usually off the stellar locus but this corresponds to the presence of the Ly$\alpha$ forest in most of the filters and is thus again highly dependent on the particular realization of the Ly$\alpha$ absorption (density and actual locations of the strong Ly$\alpha$ absorbers on the line of sight). In addition, the flux below Ly$\alpha$ is comparatively much lower implying a far less precise photometry. Generally, it is clear that the XMM-OM filter set is not adapted to the study of high-redshift quasars: although some of them will be easily spotted, the selection criterion will remain inhomogeneous. On the other hand, the XMM-OM photometry has no discrimination power between degenerate stars and quasars. Particularly for white dwarfs with effective temperatures in the range $12000 {\rm K} - 15000 {\rm K}$, the colours are very similar and the domain in effective temperature is too small to authorize a proper segregation.
As a last point, we would like to recall the existence of the $uvm2$ filter which has already been used to define the $ic_{\rm red}$ index. We found no combination where this filter could be of some help to improve the situation. For example, in a 2D ($uvw2-uvm2$) vs ($uvm2-uvw1$) colour diagram, the quasars are well located on the stellar locus except perhaps for redshifts around $z~\sim 0.8
\rightarrow 1.0$ where they leave the stellar locus but this brings no strong improvement compared to the previously analysed filters.
Conclusions
===========
In the present paper, we discussed the properties of the natural photometric system of the Optical Monitor onboard the X-ray Multi-Mirror satellite. On the basis of numerical simulations, we investigated its transformability to the standard Johnson $U\!BV$ system. We gave the main transformation equations both for individual filters and for colour indices. On the basis of the same technique of simulations, we showed that, for stars with effective temperatures higher than $9000$ K observed with this system, it is possible to determine the temperature and the reddening (due to interstellar extinction) independently, as well as to detect non-standard reddening laws on given lines of sight. Finally, we made a detailed study of the possibilities to select quasar candidates on the basis of their location in the multicolour space definable from this particular photometric system. In particular, the interest of the use of the UV filters has been critically evaluated. The main conclusion of this analysis is that the use of the $uvw1$ and $uvw2$ filters allows to greatly enhance the non-degenerate stars - quasars discrimination in the $[0.8;2.1]$ redshift range. For lower redshifts, the discrimination remains reasonably efficient. The advantage to make use of the XMM-OM photometric system is nevertheless less clear for what concerns the higher redshifts ($z \ge 2.3$) and the discrimination between the quasars and the degenerate stars. The quasar candidate selection could benefit from the addition of criterions based on the X/optical flux ratio, as these are usually larger for QSO/AGN than for stars (Stocke et al. 1991, Schmidt et al. 1998). The full treatment of the quasar candidate selection based on XMM observations is beyond the scope of the present paper, dedicated to the properties of the XMM-OM photometric system, and will be discussed elsewhere.
Allen C. W., 1976, Astrophysical Quantities, Athlone Press London Bessel M.S., 1990, PASP 102, 1181 Bohlin R.C., Savage B.D., Drake J.F., 1978, ApJ 224, 132 Bohlin R.C., Colina L., Finley D.S., 1995, AJ 110, 1316 Caloi V., 1989, A&A 221, 27 Cardelli J.A., Clayton G.C., Mathis J.S., 1989, ApJ 345, 245 Clegg R.E.S., Middlemass D., 1987, MNRAS 228, 759 Conlon E.S., Dufton P.L., Keenan F.P., McCausland R.J.H., 1991, MNRAS 248, 820 Cristiani S., Vio R., 1990, A&A 227, 385 Dahlem M., Schartel N., 1999, XMM Users’ Handbook, Issue 1.1, ESA, Noordwijk Fioc M., Rocca-Volmerange B., 1997, A&A 326, 950 Fordham J.L.A., Bone D.A., Oldfield M.K., Bellis J.G., 1992, in: Morgan B.L. (ed.) Proceedings of the 10th Symposium on Photoelectronic Image Devices, held at Imperial College of Science Technology and Medicine, London, 2-6 September 1991. Institute of Physics Conference Series 121, 105 Francis P.J., Hewett P.C., Foltz C.B., Chaffee F.H., Weymann R.J., Morris S.L., 1991, ApJ 373, 465 Giallongo E., Trevese D., 1990, ApJ 353, 24 Gunn J.E., Stryker L.L., 1983, ApJS 52, 121 Heck A., Egret D., Jaschek M., Jaschek C., 1984, A&AS 57, 213 Irwin M., McMahon R.G., Hazard C., 1991, in: D. Crampton (ed.) The Space Distribution of Quasars, ASP Conf. Ser. 21, 117 Johnson H.L., 1955, Ann. d’Ap. 18, 292 Koester D., 1999, private communication Kurucz R.L., 1992, in: Barbuy B. & Renzini A. (eds.) The stellar populations of galaxies, IAU Symp. 149, Dordrecht, Kluwer Acad. Pub., p.225 Lenz D.D., Newberg H.J., Rosner R., Richards G.T., Stoughton C., 1998, ApJS 119, 121 Madau P., 1995, ApJ 441, 18 Mason K.O., Cropper M.S., Hunt R., Horner S., Priedhorsky W., Ho C., Córdova F.A., Jamar C., Antonello E., 1996, in: Siegmund O.H. and Gummin M.A. (eds.) EUV, X-ray and Gamma-ray instrumentation for astronomy VII, Proc. SPIE Vol 2808, 438 Miller L., Mitchell P.S., 1988, in: Osmer P.S., Porter A.C., Green R.F., Foltz C.B. (eds.) Proceedings of a workshop on optical surveys for quasars, ASP Conf. Ser. 2, 114 Moehler S., Sweigart A.V., Heber U., Landsman W.B., Catelan M., 1999, in: A. Noels et al. (eds.) The Galactic Halo - from Globular Clusters to Field Stars, 35$^{\rm th}$ Li‘ege Int. Astroph. Coll., in press Møller P., Jakobsen P., 1990, A&A 228, 299 Møller P., Warren S., 1991, in: D. Crampton (ed.) The Space Distribution of Quasars, ASP Conf. Ser. 21, 96 Moreau O., Reboul H., 1995, A&AS 111, 169 Oke J.B., Korycansky D.G., 1982, ApJ 255, 11 Prévot M.L., Lequeux J., Maurice E., Prévot L., Rocca-Volmerange B., 1984, A&A 132, 389 Royer P., 1994, Recherche de Candidats Quasars par la Méthode d’Analyse Multicouleurs, Master Thesis, University of Li‘ege Royer P., Manfroid J., 1996, A&AS 119, 569 Schmidt M., Hasinger G., Gunn J., Schneider D., Burg R., Giacconi R., Lehmann M., MacKenty J., Trumper J., Zamorani G., 1998, A&A 329, 495 Sterken C., Manfroid J., 1992, Astronomical Photometry – A Guide. Kluwer Academic Publishers, Dordrecht Stocke J.T., Morris S.L., Gioia I.M., Maccacaro T., Schild R., Wolter A., Fleming T.A., Henry J.P., 1991, ApJS 76, 813 Warren S.J., Hewett P.C., Osmer P.S., 1994, ApJ 421, 412 Wesemael F., Auer L.H., Van Horn H.M., Savedoff M.P., 1980, ApJS 43, 159 Zheng W., Kriss G.A., Telfer R.C., Grimes J.P., Davidsen A.F., 1997, ApJ 475, 469 (erratum: 1998, ApJ 492, 855) Zuo L., Lu L., 1993, ApJ 418, 601
[^1]: Aspirant au Fonds National de la Recherche Scientifique (Belgium)
[^2]: Directeur de Recherche au Fonds National de la Recherche Scientifique (Belgium)
[^3]: Chercheur Qualifié au Fonds National de la Recherche Scientifique (Belgium)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'For many years, multi-object tracking benchmarks have focused on a handful of categories. Motivated primarily by surveillance and self-driving applications, these datasets provide tracks for people, vehicles, and animals, ignoring the vast majority of objects in the world. By contrast, in the related field of object detection, the introduction of large-scale, diverse datasets (e.g., COCO) have fostered significant progress in developing highly robust solutions. To bridge this gap, we introduce a similarly diverse dataset for Tracking Any Object ()[^1]. It consists of high resolution videos, captured in diverse environments, which are half a minute long on average. Importantly, we adopt a bottom-up approach for discovering a large vocabulary of categories, an order of magnitude more than prior tracking benchmarks. To this end, we ask annotators to label objects that move at any point in the video, and give names to them post factum. Our vocabulary is both significantly larger and qualitatively different from existing tracking datasets. To ensure scalability of annotation, we employ a federated approach that focuses manual effort on labeling tracks for those relevant objects in a video (e.g., those that move). We perform an extensive evaluation of state-of-the-art trackers and make a number of important discoveries regarding large-vocabulary tracking in an open-world. In particular, we show that existing single- and multi-object trackers struggle when applied to this scenario in the wild, and that detection-based, multi-object trackers are in fact competitive with user-initialized ones. We hope that our dataset and analysis will boost further progress in the tracking community.'
author:
- Achal Dave
- Tarasha Khurana
- |
Pavel Tokmakov\
Cordelia Schmid
- Deva Ramanan
bibliography:
- 'egbib.bib'
title: |
: A Large-Scale Benchmark for\
Tracking Any Object
---
Introduction
============
A key component in the success of modern object detection methods was the introduction of large-scale, diverse benchmarks, such as MS COCO [@lin2014microsoft] and LVIS [@gupta2019lvis]. By contrast, multi-object tracking datasets tend to be small [@milan2016mot16; @voigtlaender2019mots], biased towards short videos [@yang2019video], and, most importantly, focused on a very small vocabulary of categories [@milan2016mot16; @voigtlaender2019mots; @wen2015ua] (see Table \[tab:dataset\_stats\]). As can be seen from Figure \[fig:cats\], they predominantly target people and vehicles. Due to the lack of proper benchmarks, the community has shifted towards solutions tailored to the few videos used for evaluation. Indeed, Bergmann et al. [@bergmann2019tracking] have recently and convincingly demonstrated that simple baselines perform on par with state-of-the-art (SOTA) multi-object trackers.
In this work we introduce a large-scale benchmark for Tracking Any Object (). Our dataset features high resolution videos captured in diverse environments, which are 30 seconds long on average, and has tracks labeled for object categories. We compare the statistics of to existing multi-object tracking benchmarks in Table \[tab:dataset\_stats\] and Figure \[fig:cats\], and demonstrate that it improves upon them both in terms of complexity and in terms of diversity (see Figure \[fig:tao\] for representative frames from ). Collecting such a dataset presents three main challenges: (1) how to select a large number of diverse, long, high-quality videos; (2) how to define a set of categories covering all the objects that might be of interest for tracking; and (3) how to label tracks for these categories at a realistic cost. Below we summarize our approach for addressing these challenges. A detailed description of dataset collection is provided in Section \[sec:dataset\_collection\].
Existing datasets tend to focus on one or just a few domains when selecting the videos, such as outdoor scenes in MOT [@milan2016mot16], or road scenes in KITTI [@Geiger2012CVPR]. This results in methods that fail when applied in the wild. To avoid this bias, we construct with videos from as many environments as possible. We include indoor videos from Charades [@sigurdsson2016hollywood], movie scenes from AVA [@gu2018ava], outdoor videos from LaSOT [@fan2019lasot], road-scenes from ArgoVerse [@chang2019argoverse], and a diverse sample of videos from HACS [@zhao2019hacs] and YFCC100M [@thomee2015yfcc100m]. We ensure all videos are of high quality, with the smallest dimension larger or equal to 480px, and contain at least 2 moving objects. reports the full statistics of the collected videos, showing that provides an evaluation suite that is significantly larger, longer, and more diverse than prior work. Note that contains fewer training videos than recent tracking datasets, as we intentionally dedicate the majority of videos for in-the-wild [*benchmark*]{} evaluation, the focus of our effort.
\[tab:dataset\_stats\]
Given the selected videos, we must choose [*what*]{} to annotate. Most datasets are constructed with a top-down approach, where categories of interest are pre-defined by benchmark curators. That is, curators first select the subset of categories deemed relevant for the task, and then collect images or videos expressly for these categories [@deng2009imagenet; @lin2014microsoft; @valmadre2018long]. This approach naturally introduces curator bias. An alternative strategy is bottom-up, open-world [*discovery*]{} of what objects are present in the data. Here, the vocabulary emerges post factum [@gu2018ava; @gupta2019lvis; @zhou2017scene], an approach that dates back to LabelMe [@russell2008labelme]. Inspired by this line of work, we devise the following strategy to discover an ontology of objects relevant for tracking: first annotators are asked to label [*all*]{} objects that either move by themselves or are moved by people. They then give names to the labeled objects, resulting in a vocabulary that is not only significantly larger, but is also qualitatively different from that of any existing tracking dataset (see Figure \[fig:cats\]). To facilitate training of object detectors, that can be later used by multi-object trackers on our dataset, we encourage annotators to choose categories that exists in the LVIS dataset [@gupta2019lvis]. If no appropriate category can be found in the LVIS vocabulary, annotators can provide free-form names (see Section \[sec:annotation\_pipeline\] for details).
Exhaustively labeling tracks for such a large collection of objects in long videos is prohibitively expensive. Instead, we extend the federated annotation approach proposed in [@gupta2019lvis] to the tracking domain. In particular, we ask the annotators to label tracks for up to 10 objects in every video. We then separately collect exhaustive labels for every category for a subset of videos, indicating whether all the instances of the category have been labeled in the video. During evaluation of a particular category, we use only videos with exhaustive labels for computing precision and all videos for computing recall. This allows us to reliably measure methods’ performance at a fraction of the cost of exhaustively annotating the videos. We use the LVIS federated mAP metric [@gupta2019lvis] for evaluation, replacing 2D IoU with 3D IoU [@yang2019video]. For detailed comparisons, we further report the standard MOT challenge [@milan2016mot16] metrics in .
{width="\linewidth"}
Equipped with , we set out to answer several questions about the state of the tracking community. In particular, in Section \[sec:analysis\] we report the following discoveries: (1) SOTA trackers struggle to generalize to a large vocabulary of objects, particularly for infrequent object categories in the tail; (2) while trackers work significantly better for the most-explored category of people, tracking people in diverse scenarios (e.g., frequent occlusions or camera motion) remains challenging; (3) when scaled to a large object vocabulary, multi-object trackers become competitive with user-initialized trackers, despite the latter being provided with a ground truth initializations. We hope that these insights will help to define the most promising directions for future research.
Related work {#sec:related}
============
The domain of object tracking is subdivided based on the way the tracks are initialized. Our work falls into the multi-object tracking category, where all the objects out of a fixed vocabulary of classes have to be detected and tracked. Other formulations include user-initialized tracking, and saliency-based tracking. In the remainder of this section we will first review the most relevant benchmarks datasets in each of these areas, and then discuss SOTA methods for multi-object and user-initialized tracking.
Benchmarks {#sec:related_benchmarks}
----------
is the task of tracking an unknown number of objects from a known set of categories. Most MOT benchmarks [@fisher2001context; @Geiger2012CVPR; @milan2016mot16; @wen2015ua] focus on either people or vehicles (see Figure \[fig:cats\]), motivated by surveillance and self-driving applications. Moreover, they tend to include only a few dozen videos, captured in outdoor or road environments, encouraging methods that are overly adapted to the benchmark and do not generalize to different scenarios (see Table \[tab:dataset\_stats\]). In contrast, focuses on diversity both in the category and visual domain distribution, resulting in a realistic benchmark for tracking [*any*]{} object.
Several works have attempted to extend the MOT task to a wider vocabulary of categories. In particular, the [@russakovsky2015imagenet] benchmark provides exhaustive trajectories annotations for objects of 30 categories in 1314 videos. While this dataset is both larger and more diverse that standard MOT benchmarks, videos tend to be relatively short and the categories cover only animals and vehicles. The recent YTVIS dataset [@yang2019video] has the most broad vocabulary to date, covering 40 classes, but the majority of the categories still correspond to people, vehicles and animals. Moreover, the videos are 5 seconds long on average, making the tracking problem considerably easier in many cases. Unlike previous work, we take a bottom-up approach for defining the vocabulary. This results in not only the largest set of categories among MOT datasets to date, but also in a qualitatively different category distribution. In addition, our dataset is over 7 times larger than YTVIS in the number of frames. The recent VidOR dataset [@shang2019annotating] explores Video Object Relations, including tracks for a large vocabulary of objects. But, since ViDOR focuses on relations rather than tracks, object trajectories tend to be missing or incomplete, making it hard to repurpose for tracker benchmarking. In contrast, we ensure TAO maintains high quality for both accuracy and completeness of labels (see Appendix \[sec:appendix-quality\] for a quantitative analysis).
Finally, several recent works have proposed to label masks instead of bounding boxes for benchmarking multi-object tracking [@voigtlaender2019mots; @yang2019video]. In collecting we made a conscious choice to prioritize scale and diversity of the benchmark over pixel-accurate labeling. Instance mask annotations are significantly more expensive to collect than bounding boxes, and we show empirically that tracking at the box level is already a challenging task that current methods fail to solve.
forgoes a fixed vocabulary of categories altogether and instead relies on the user to provide bounding box annotations for the objects that need to be tracked at test time [@fan2019lasot; @huang2018got; @kristan2016novel; @valmadre2018long; @wu2013online]. The benchmarks in this category tend to be larger and more diverse than their MOT counterparts, but most of them still offer a tradeoff between the number of videos in the benchmarks and the average length of the videos (see ). Moreover, even if the task itself is category-agnostic, empirical distribution of categories in the benchmarks tends to be heavily skewed towards a few common objects. We study whether this bias in category selection results in methods failing to generalize to more challenging objects by evaluating state-of-the-art user-initialized trackers on in Section \[sec:results\_user\_init\].
differs from user-initialized tracking in that both the input to the tracker and the output are object masks, not boxes [@perazzi2016benchmark; @xu2018youtube]. As a result, such datasets are a lot more expensive to collect, and videos tend to be extremely short. The main focus of the works in this domain [@caelles2017one; @khoreva2019lucid; @voigtlaender2017online] is on accurate mask propagation, not solving challenging identity association problems, thus their effort is complementary to ours.
is an intriguing direction towards open-world tracking, where the objects of interest are defined not with a fixed vocabulary of categories, or manual annotations, but with bottom-up, motion- [@ochs2013segmentation; @perazzi2016benchmark] or appearance-based [@caelles20192019; @wang2019learning] saliency cues. Our work similarly uses motion-based saliency to define a comprehensive vocabulary of categories, but presents a significantly larger benchmark with class labels for each object, enabling the use and evaluation of large-vocabulary object recognition approaches.
Algorithms {#sec:related_algorithms}
----------
can be categorized into people and multi-category trackers. The former have been mainly developed on the MOT benchmark [@milan2016mot16] and follow the tracking-by-detection paradigm, linking outputs of person detectors in an offline, graph-based framework [@berclaz2006robust; @berclaz2011multiple; @breitenstein2009robust; @ess2008mobile]. These methods mainly differ in the way they define the edge cost in the graph. Classical approaches use overlap between detections in consecutive frames [@jiang2007linear; @pirsiavash2011globally; @zhang2008global]. More recent methods define edge costs based on appearance similarity [@milan2017online; @ristani2018features], or motion-based models [@alahi2016social; @chen2018real; @choi2010multiple; @leal2014learning; @ren2018collaborative; @scovanner2009learning]. Very recently, Bergmann et al. [@bergmann2019tracking] proposed a simple baseline approach that performs on par with SOTA people trackers, which repurposes an object detector’s bounding box regression capability to predict the position of an object in the next frame. Notice that all these methods have been developed and evaluated on the relatively small MOT dataset, which consists of 14 videos captured in very similar environments. By contrast, provides a much richer, more diverse set of videos, encouraging trackers more robust to tracking challenges such as occlusion and camera motion.
The more general multi-object tracking scenario is usually studied using [@russakovsky2015imagenet]. Methods in this group also use offline, graph-based optimization to link frame-level detections into tracks. To define the edge potentials, in addition to bounding box overlap, Feichtenhofer et al. [@feichtenhofer2017detect] propose to use a similarity embedding, which is learned jointly with the detector. Alternatively, Kang et al. [@kang2017object] directly predict short tubelets, and Xiao et al. [@xiao2018video] incorporate a spatio-temporal memory module inside a detector. Inspired by [@bergmann2019tracking], we show that a simple baseline approach, relying on the Viterbi algorithm for linking detections across frames [@feichtenhofer2017detect; @gkioxari2015finding], performs on par with the methods mentioned above on . We then use this baseline for evaluating generic multi-object tracking on in Section \[sec:results\_mot\], and demonstrate that it struggles when faced with a large vocabulary and a diverse data distribution.
tend to rely on a Siamese network architecture that was first introduced for signature verification [@bromley1994signature], and later adapted for tracking [@bertinetto2016fully; @danelljan2017eco; @held2016learning; @tao2016siamese]. They learn a patch-level distance embedding and find the closest patch to the one annotated in the first frame in the following frames. To simplify the matching problem, state-of-the-art approaches limit the search space to the region in which the object was localized in the previous frame. Recently there have been several attempts to introduce some ideas from CNN architectures for object detection into Siamese trackers. In particular, Li et al. [@li2018high] use the similarity map obtained by matching the object template to the test frame as input to an RPN-like module adapted from Faster-RCNN [@ren2015faster]. Later this architecture was extended by introducing hard negative mining and template updating [@zhu2018distractor], as well as mask prediction [@wang2019fast]. In another line of work, Siamese-based trackers have been augmented with a target discrimination module to improve their robustness to distractors [@bhat2019learning; @danelljan2019atom]. We evaluate several state-of-the-art methods in this paradigm for which public implementation is available [@bhat2019learning; @danelljan2019atom; @danelljan2017eco; @li2019siamrpn++; @wang2019fast] on , and demonstrate that they achieve only a moderate improvement over our multi-object tracking baseline, despite being provided with a ground truth initialization for each track (see Section \[sec:results\_user\_init\] for details).
Dataset design {#sec:dataset_design}
==============
Our primary goal in this work is collecting a large-scale dataset of videos with a diverse vocabulary of labeled object tracks for evaluating trackers in the wild. This requires designing a strategy for (1) video collection, (2) vocabulary discovery, (3) scalable annotation, and (4) evaluation. We detail our strategies for (2-4) in this section, and defer the discussion of video collection to Section \[sec:video\_selection\].
Rather than manually defining a set of categories, we discover an object vocabulary from unlabeled videos which span diverse operating domains. Our goal is to focus on *dynamic* objects in the world. Towards this end, we ask annotators to mark all objects that *move* in our collection of videos, without any object vocabulary in mind. We then construct a vocabulary by giving names for all the discovered objects, following the recent trend for open-world dataset collection [@gupta2019lvis; @zhou2017scene]. In particular, annotators are asked to provide a free-form name for every object, but are encouraged to select a category from the LVIS [@gupta2019lvis] vocabulary whenever possible. We detail this process further in Section \[sec:annotation\_categorization\].
Given this vocabulary, one option might be exhaustively labelling all instances of each category in all videos. Unfortunately, exhaustive annotation of a large vocabulary is expensive, even for images, as noted in [@gupta2019lvis]. We choose to use our labeling budget instead on collecting a large-scale, diverse dataset, by extending the federated annotation protocol of [@gupta2019lvis] from image datasets to videos. Rather than labeling every video $v$ with every category $c$, we define three subsets of our dataset for each category: $P_c$, which contains videos where all instances of $c$ are labeled, $N_c$, videos with no instance of $c$ present in the video, and $U_c$, videos where *some* instances of $c$ are annotated. Videos not belonging to any of these subsets are ignored when evaluating category $c$. For each category $c$, we only use videos in $P_c$ and $N_c$ to measure the *precision* of trackers, and videos in $P_c$ and $U_c$ to measure recall. We describe how to define $P_c$, $N_c$, and $U_c$ in Section \[sec:annotation\_federation\].
To collect , we choose to prioritize scale and diversity of the data at the cost of annotation granularity. In particular, we label tracks at 1 frame per second with bounding box labels but don’t annotated segmentation masks. This allows us to label categories in videos at a relatively modest cost. Our decision is motivated by the observation of [@valmadre2018long] that dense frame labeling does not change the relative performance of the methods.
Traditionally, multi-object tracking datasets use either the CLEAR MOT metrics [@bernardin2008evaluating; @Geiger2012CVPR; @milan2016mot16] or a 3D intersection-over-union (IoU) based metric [@russakovsky2015imagenet; @yang2019video]. We report the former in (introducing modifications for large-vocabularies of classes, including multi-class aggregation and federation), but focus our experiments on the latter. To formally define 3D IoU, let $G = \{g_1, \dots, g_T\}$ and $D = \{d_1, \dots, d_T\}$ be a groundtruth and predicted track for a video with $T$ frames. 3D IoU is defined as: $\text{IoU}_{\text{3d}}(D, G) = \frac{\sum_{t=1}^T g_t \cap d_t}{\sum_{t=1}^T g_t \cup d_t}.$ If an object is not present at time $t$, we assign $g_t$ to an empty bounding box, and similarly for a missing detection. We choose 3D IoU (with a threshold of 0.5) as the default metric for , and provide further analysis in .
Similar to standard object detection metrics, (3D) IoU together with (track) confidence can be used to compute mean average precision across categories. For methods that provide a score for each frame in a track, we use the average frame score as the track score. Following [@gupta2019lvis], we measure precision for a category $c$ in video $v$ only if all instances of the category are verified to be labeled in it.
Dataset collection {#sec:dataset_collection}
==================
Video selection {#sec:video_selection}
---------------
Most video datasets focus on one or a few operating domains. For instance, MOT benchmarks [@milan2016mot16] correspond to urban, outdoor scenes featuring crowds of people, whereas AVA [@gu2018ava] is sourced from produced films, typically capturing actors with close shots in carefully staged scenes. As a result, methods developed on any single dataset (and hence domain) fail to generalize in the wild. To avoid this bias, we constructed by selecting videos from a variety of existing video benchmarks to ensure diversity of scenes and objects.
In particular, we used datasets for action recognition, self-driving cars, user-initialized tracking, as well as in-the-wild Flickr videos. In the action recognition domain we selected 3 datasets: Charades [@sigurdsson2016hollywood], AVA [@gu2018ava], and HACS [@zhao2019hacs]. Charades features complex human-human and human-object interactions, but all videos are indoor with limited camera motion. In contrast, AVA has a much wider variety of scenes and cinematographic styles but is scripted. HACS provides unscripted, in-the-wild videos. These action datasets are naturally focused on people and objects with which people interact. To include other animals and vehicles, we also source clips from LaSOT [@fan2019lasot] (a benchmark for user-initialized tracking), BDD [@bdd100k] and ArgoVerse [@chang2019argoverse] (benchmarks for self-driving cars). LaSOT is a diverse collection whereas BDD and ArgoVerse consist entirely of outdoor, urban scenes. Finally we sample in-the-wild videos from the YFCC100M [@thomee2015yfcc100m] Flickr collection.
The videos are automatically filtered to remove short videos and videos with a resolution below 480p. For longer videos, as in AVA, we use [@LokocKSMC19Viret] to extract scenes without shot changes. In addition, we manually reviewed each sampled video to ensure it is high quality: i.e., we removed grainy videos as well as videos with excessive camera motion or shot changes. Finally, to focus on the most challenging tracking scenarios, we only kept videos that contain at least 2 moving objects. The full statistics of the collected videos are provided in Table \[tab:dataset\_stats\]. We point out that many prior video datasets tend to limit one or more quality dimensions (in terms of resolution, length, or number of videos) in order to keep evaluation and processing times manageable. In contrast, we believe that in order to truly enable tracking in the open-world, we need to appropriately scale benchmarks.
{width="\linewidth"}
Annotation pipeline {#sec:annotation_pipeline}
-------------------
Our annotation pipeline is illustrated in Figure \[fig:pipeline\]. We designed it to separate low-level tracking from high-level semantic labeling. As pointed out by others [@barriuso2012notes], semantic labeling can be subtle and error-prone because of ambiguities and corner-cases that arise in category boundaries. By separating tasks into low vs high-level, we are able to take advantage of unskilled annotators for the former and highly-vetted workers for the latter.
### Object mining and tracking. {#sec:annotation_mining}
We combine object mining and track labeling into a single stage of annotation. Given the set of videos described above, we ask annotators to mark [*objects that move at any point in the video*]{}. To avoid overspending our annotation budget on a few crowded videos, we limited the number of labeled object per video to 10. Note that this stage is *category-agnostic*: annotators are not instructed to look for objects from any specific vocabulary, but instead use motion as a [*saliency*]{} cue for mining relevant objects. They are then asked to track these objects throughout the video, and label them with bounding boxes at 1 frame-per-second. Finally, the tracks are verified by one independent annotator. This process is illustrated in Figure \[fig:pipeline\], where we can see that 6 objects are discovered and tracked.
### Object categorization. {#sec:annotation_categorization}
Next, we collected category labels for objects discovered in the previous stage and simultaneously constructed the dataset vocabulary. We focus on the large vocabulary from the LVIS [@gupta2019lvis] object detection dataset, which contains 1,230 synsets discovered in a bottom-up manner similar to ours. Doing so also allows us to make use of LVIS as a training set of relevant object detectors (which we later use within a tracking pipeline to produce strong baselines - ). Because maintaining a mental list of 1,230 categories is challenging even for expert annotators, we use an auto-complete annotation interface to suggests categories from the LVIS vocabulary (Figure \[fig:pipeline\] (b)). The autocomplete interface displays classes with a matching synset (e.g., “person.n.01"), name, synonym, and finally those with a matching definition. Interestingly, we find that some objects discovered in , such as “door” or “marker cap”, do not exist in LVIS. To accommodate such important exceptions, we allow annotators to label objects with free-form text if they do not fit in the LVIS vocabulary.
Overall, annotators labeled objects (%) with LVIS categories, and objects (%) with free-form categories. We use the LVIS categories for MOT experiments (because detectors can be trained on LVIS), but use all categories for user-initialized tracking experiments in .
### Federated “exhaustive” labeling. {#sec:annotation_federation}
Finally, we ask annotators to verify which categories are exhaustively labeled for each video. Specifically, for each category $c$ labeled in video $v$, we ask annotators whether all instances of $c$ are labeled. In , after this stage, annotators marked that ‘person’ is exhaustively labeled, while ‘camel’ is not. Next, we show annotators a sampled subset of categories that are not labeled in the video, and ask them to indicate categories which are absent in the video. In , annotators indicated that ‘bicycle’ and ‘mirror’ are absent.
Dataset splits {#sec:dataset_splits}
--------------
We intend for to be used primarily as an *evaluation* benchmark. We split into three subsets: train, validation and test, containing , and videos respectively. Typically, ‘train’ splits tend to be larger than ‘val’ and ‘test’. We choose to make TAO’s training set small for several reasons. Firstly, the primary goal of TAO is to reliably benchmark trackers in-the-wild. Secondly, most MOT systems are modularly trained using image-based detectors with hyper-parameter tuning of the overall tracking system. In our case, we ensure the train set is sufficiently large for hyper-parameter tuning, and ensure that our large-vocabulary is aligned with large-vocabulary image datasets (e.g., LVIS). This allows us to devote most of our annotation budget for large-scale ‘val’ and held-out ‘test’ sets." We ensure that the videos in train, validation and test are well-separated. As an example, we ensure that each subject in the Charades dataset appears in only one of the train, validation or test sets. We provide further details on split construction in Appendix \[sec:appendix\_splits\].
Analysis of state-of-the-art trackers {#sec:analysis}
=====================================
We now use to analyze how well existing multi- and single-object trackers perform in the wild and when they fail. We tune the hyperparameters of each tracking approach on the ‘train’ set, and report results on the ‘val’ set. To capitalize on existing object detectors, we evaluate using the LVIS categories in . We begin by shortly describing the methods used in our analysis.
Methods {#sec:analysis_methods}
-------
We analyze how well state-of-the-art object detectors perform on our dataset. To this end, we present results using a standard Mask R-CNN [@ren2015faster] detector trained using [@wu2019detectron2] in .
[r]{}[0.5]{}
\[tab:imagenetvid\]
We analyze SOTA multi-object tracking methods on , the largest vocabulary dataset prior to . We first clarify whether such approaches improve detection or tracking. reports the standard Detection mAP and Track mAP. The ‘Detection’ row corresponds to a detection-only baseline widely reported by prior work [@xiao2018video; @feichtenhofer2017detect; @Zhu2017FlowGuidedFA]. D&T [@feichtenhofer2017detect] and STMN [@xiao2018video] are spatiotemporal architectures that produce SOTA improvements of 6-7% in detection mAP over a per-frame detector. However, both D&T and STMN post-process their per-frame outputs using the Viterbi algorithm, which iteratively links and re-weights the confidences of per-frame detections (see [@gkioxari2015finding] for details). [*When the same post-processing is applied to a single-frame detector, one achieves nearly the same performance gain (Table \[tab:imagenetvid\], last row)*]{}.
Our analysis reinforces the bleak view of multi-object tracking progress suggested by [@bergmann2019tracking]: while ever-more complex approaches have been proposed for the task, their improvements are often attributable to simple, baseline strategies. To foster meaningful progress on , we evaluate a number of strong baselines in this work. We evaluate a powerful single-frame detector trained on LVIS [@gupta2019lvis] and COCO [@lin2014microsoft], followed by two linking methods: SORT [@Bewley2016_sort], a simple, online linker initially proposed for tracking people, and the Viterbi post-processing step used by [@feichtenhofer2017detect; @xiao2018video], in .
Detecting and tracking people have been a distinct focus in the multi-object tracking community. compares the above baselines to a recent SOTA people-tracker [@bergmann2019tracking].
We additionally present results using user-initialized trackers. We evaluate several recent methods for which public implementation is available [@bhat2019learning; @danelljan2019atom; @danelljan2017eco; @li2019siamrpn++; @wang2019fast]. Unfortunately, these trackers do not provide a class label for the objects they are tracking, and cannot directly be compared to multi-object trackers. However, these trackers *can* be evaluated with an oracle classifier, allowing us to directly compare their accuracy with the methods that simultaneously detect and track objects.
Finally, to disentangle the complexity of object classification and tracking, we use two oracles. The first, a class oracle, computes the best matching between predicted and groundtruth tracks in each video. Predicted tracks that match to a groundtruth track with 3D IoU $>0.5$ are assigned the category of their matched groundtruth track. Tracks that do not match to a groundtruth track are not modified, and are treated as false positives. This allows us to evaluate the performance of trackers assuming the semantic *classification* task is solved.
The second oracle computes the best possible assignment of per-frame detections to tracks, by comparing them with groundtruth. When doing so, class predictions for each detection are held constant. Any detections that are not matched are discarded. This oracle allows us to analyze the best performance we could expect given a fixed set of detections.
Results {#sec:results_all}
-------
\[sec:results\_detection\] We start by assessing the difficulty of the detection task on TAO. To this end we evaluate the SOTA object detector [@he2017mask] using detection mAP. We train this model on a combination of LVIS and COCO, finding that training on LVIS alone led to a model that struggles to detect people. The final model achieves an mAP of 27.1 on val at IoU 0.5, suggesting that single-frame detection is challenging on .
### Do multi-object trackers generalize to TAO? {#sec:results_mot}
reports results using tracking mAP on . As a sanity check, we first evaluate a per-frame detector by assigning each detection to its own track. As expected, this achieves an mAP of nearly 0 (which isn’t quite 0 due to the presence of short tracks).
Next, we evaluate two multi-object tracking approaches. We compare the SOTA Viterbi linking method to an online SORT tracker [@Bewley2016_sort]. We tune SORT hyperparameters on our diverse ‘train’ set. Appendix \[sec:appendix\_tuning\_sort\] shows that this tuning is key for good accuracy. The offline Viterbi algorithm takes over a month of processing time to run on our ‘train’ set, prohibiting thorough parameter tuning. Instead, we tune a post-processing parameter for Viterbi: the score threshold for reporting a detection at each frame. We detail our tuning procedure in .
Surprisingly, we find that the simpler, online approach of SORT outperforms Viterbi, perhaps because the latter has been heavily tuned for . Because of its scalablity (to many categories and long videos) and relatively better performance, we focus on SORT for the majority of our experiments. However, the performance of both methods remains low, suggesting presents a major challenge for the tracking community, requiring principled novel approaches.
To better understand the nature of the complexity of TAO, we separately measure the challenges of tracking and classification. To this end, we first evaluate the “track” oracle that perfectly links per-frame detections. It achieves a stronger mAP of 31.5, compared to 13.2 for SORT. Interestingly, providing SORT tracks with an oracle class label provides a similar improvement, boosting mAP to 30.2. We posit that these improvements are orthogonal, and verify this by combining them; we link detections with oracle tracks and assign these tracks oracle class labels. This provides the largest delta, dramatically improving mAP to 83.6%. This suggests that [*large-vocabulary tracking requires jointly improving tracking and classification accuracy (e.g., reducing semantic flicker as shown in Fig. \[fig:sort\_qualitative\]).*]{}
[r]{}[0.38]{}
\[tab:tao\_people\]
### How well can we track people? {#sec:results_people}
We now evaluate tracking on one particularly important category: people. Measuring AP for individual categories in a federated dataset can be noisy [@gupta2019lvis], so we emphasize *relative* performance of trackers rather than their absolute AP. We evaluate Tracktor++ [@bergmann2019tracking], the state-of-the-art method designed specifically for people tracking on our dataset, and compare it to the SORT and Viterbi baselines in Table \[tab:tao\_people\]. For fairness, we update Tracktor++ to use the same detector used by our SORT and Viterbi baselines, but only use the ‘person’ predictions from this detector. Additionally, we tune the score threshold for Tracktor++ on our ‘train’ set, but find the method is largely robust to this parameter (see Appendix \[sec:appendix\_tuning\_tracktor\]). We find that Tracktor++ strongly performs other approaches (36.7 AP), while SORT comes in second, modestly outperforming Viterbi (18.6 vs 16.5 AP). It is interesting to note that SORT, which can scale to all object categories, performs noticeably worse on all categories on average (13.2 mAP). shows that this delta between ‘person’ and other classes is even more dramatic using the MOTA metric (6.7 overall vs 54.8 for ‘person’). We attribute the higher accuracy for the ‘person’ category to two factors: (1) a rich history of focused research on this one category, which has led to more accurate detectors and trackers, and (2) more complex categories present significant challenges, such as hand-held objects which undergo repeated occlusions during interactions.
To further investigate Tracktor++’s performance, we evaluate a simpler variant of the method from [@bergmann2019tracking], which does not use appearance-based re-identification nor pixel-level frame alignment. We evaluate this variant on , and find that removing these components reduces AP by over 8 points (from 36.7 to 25.9), suggesting that a majority of improvements over our baselines come from these two components. Our results contrast those of [@bergmann2019tracking], which suggest that re-id and frame alignment are not particularly helpful. *Compared to prior benchmarks, we posit the diversity of results in a challenging testbed for person tracking which encourages trackers robust to occlusion and camera jitter.*
### Do user-initialized trackers generalize better? {#sec:results_user_init}
Next, we present results of recent user-initialized trackers in . For each object in , we provide the user-initialized tracker with a groundtruth box. We consider two strategies for initialization. The standard approach (denoted ‘Init first’) initializes trackers using the first frame an object appears in, and runs trackers for the rest of the video. As the object may be partially occluded in this first frame, we additionally report a variant which initializes trackers using the frame with the largest bounding box (denoted ‘Init biggest’), and runs trackers forwards and backwards in time.
Unlike multi-object trackers, most user-initialized trackers report a bounding box and confidence for objects at each frame, and do not explicitly report when an object is *absent* [@valmadre2018long]. To resolve this, we modify each method to report an object as absent when the confidence drops below a threshold. We tune this threshold on the ‘train’ set in and find that user-initialized trackers are particularly sensitive to this threshold.
--------------------------------- ---------- ------- ------------ --------------
Method Box Init Class Init first Init biggest
SORT
ECO [@danelljan2017eco] 23.7 30.4
SiamMask [@wang2019fast] 30.8 37.0
SiamRPN++ LT [@li2019siamrpn++] 27.2 30.4
SiamRPN++ [@li2019siamrpn++] 29.7 35.9
ATOM\* [@danelljan2019atom] 30.9 **38.6**
DIMP\* [@bhat2019learning] **33.2** 38.5
--------------------------------- ---------- ------- ------------ --------------
: SOTA user-initialized tracking results on ‘val’. Surprisingly, despite using an oracle initial bounding box, these methods provide only modest improvements over a multi-object tracker. Because some user-initialized trackers are trained on videos in TAO, we re-train them on their original train set with videos removed, denoting this with \*.
\[tab:compare\_sot\]
We compare these trackers to SORT, supplying both with a class oracle. As expected, the use of a ground-truth initialization allows the best user-initialized methods to outperform the multi-object tracker. However, even with an oracle box initialization and an oracle classifier, tracking remains challenging on . Indeed, most user-initialized trackers provide at most modest improvements over SORT, despite using an oracle box initialization. The ‘Init biggest’ strategy provides stronger improvements by initializing with easier frames, but this strategy cannot be used in *online* applications, as it requires access to the entire video. notes that user-initialized trackers can accurately track for a few frames after initialization, leading to improvements in MOTA, but provide little benefits in longer-term tracking. We hypothesize that the small improvement of user-initialized trackers over SORT is due to the fact that the former are trained on videos with a small vocabulary of objects with limited occlusions, leading to methods that do not generalize to the most challenging cases in TAO. One goal of user-initialized trackers is open-world tracking of objects without good detectors. ’s large vocabulary allows us to analyze progress towards this goal, indicating that *large-vocabulary multi-object trackers may now address the open-world of objects as well as category-agnostic, user-initialized trackers.*
Discussion
==========
Developing tracking approaches that can be deployed in-the-wild requires being able to reliably measure their performance. With nearly 3,000 videos, provides such a robust evaluation benchmark. Our analysis provides new conclusions about the state of tracking, while further raising a number of important questions to be explored in future work.
User-initialized trackers aim to track *any* object, without requiring category-specific detectors. In this work, we raise a provocative question: with the advent of large vocabulary object detectors [@gupta2019lvis], to what extent can (detection-based) multi-object trackers perform generic tracking [*without*]{} user initialization? , for example, shows that large-vocabulary datasets (such as and LVIS) now allow multi-object trackers to match or outperform user-initialization for a number of categories. Our hope in collecting is to measure progress in tracking in-the-wild. A valid question is whether progress may be better achieved by building trackers for *application-specific* scenarios. An indoor robot, for example, has little need for tracking elephants. However, success in many computer vision fields has been driven by the pursuit of *generic* approaches, that can then be tailored for specific applications. We do not build one class of object detectors for indoor scenes, and another for outdoor scenes, and yet another for surveillance videos. We believe that tracking will similarly benefit from targeting diverse scenarios. Of course, due to its size, also lends itself to use for evaluating trackers for specific scenarios or categories, as in for ‘person.’
Although image-based object detectors have shown significant improvements in recent years, our analysis in suggests that simple post-processing of detection outputs remains a strong baseline for detection in videos. While we do not emphasize it in this work, we note that can also be used to measure progress in video object *detection*, where the goal is not to maintain the identity of objects, but simply to reliably detect them in each frame of a video. The large vocabulary in particularly provides avenues for incorporating temporal information to resolve classification errors, which remain challenging (see ).
We thank Jonathon Luiten and Ross Girshick for detailed feedback on the dataset and manuscript, and Nadine Chang and Kenneth Marino for reviewing early drafts. Annotations for this dataset were provided by Scale.ai. This work was supported in part by the CMU Argo AI Center for Autonomous Vehicle Research, the Inria associate team GAYA, and by the Intelligence Advanced Research Projects Activity (IARPA) via Department of Interior/Interior Business Center (DOI/IBC) contract number D17PC00345. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes not withstanding any copyright annotation theron. Disclaimer: The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied of IARPA, DOI/IBC or the U.S. Government.
**Appendix**
[^1]: <http://taodataset.org/>
| {
"pile_set_name": "ArXiv"
} |
---
abstract: '\[sec\_abstract\] Concept-based video representation has proven to be effective in complex event detection. However, existing methods either manually design concepts or directly adopt concept libraries not specifically designed for events. In this paper, we propose to build Concept Bank, the largest concept library consisting of $4,876$ concepts specifically designed to cover $631$ real-world events. To construct the Concept Bank, we first gather a comprehensive event collection from WikiHow, a collaborative writing project that aims to build the world’s largest manual for any possible How-To event. For each event, we then search Flickr and discover relevant concepts from the tags of the returned images. We train a Multiple Kernel Linear SVM for each discovered concept as a concept detector in Concept Bank. We organize the concepts into a five-layer tree structure, in which the higher-level nodes correspond to the event categories while the leaf nodes are the event-specific concepts discovered for each event. Based on such tree ontology, we develop a semantic matching method to select relevant concepts for each textual event query, and then apply the corresponding concept detectors to generate concept-based video representations. We use TRECVID Multimedia Event Detection 2013 and Columbia Consumer Video open source event definitions and videos as our test sets and show very promising results on two video event detection tasks: event modeling over concept space and zero-shot event retrieval. To the best of our knowledge, this is the largest concept library covering the largest number of real-world events.'
author:
- |
Yin Cui, Dong Liu, Jiawei Chen, Shih-Fu Chang\
Department of Electrical Engineering, Columbia University\
title: Building A Large Concept Bank for Representing Events in Video
---
Introduction {#sec_introduction}
============
Representing and detecting complex events from unconstrained videos remains one of the most challenging problems in computer vision. By definition, an event is a complex activity that involves people interacting with other people and/or objects under certain scene settings. For instance, “changing a vehicle tire” can be defined as a complex event where human, vehicle, tire and tools interact with each other in an outdoor environment. The existing research on video event detection focuses on the use of low-level features combined with sophisticated learning models, and achieves satisfactory performances to some extent [@duan2009domain; @natarajan2012fusion; @tang2012latent]. However, these works fail to provide semantic information in a video event. This hampers high-level event understanding, especially when the number of training videos is small or zero. Therefore, a reasonable and computationally tractable way is to decompose videos depicting an event into a set of atomic concepts [@liu_concept; @yang2012deepnet]. These concepts can be perceived as building blocks of a complex event and are expected to provide a meaningful intermediate level of abstraction towards representing an event rather than low-level visual-audio features directly extracted from videos. In this paper, we focus on such an approach in which the textual title of the event (*e.g.*, “grooming an animal") is used as query to first find concepts relevant to the event, and then we use the selected concepts as representations in supervised event modeling or zero-shot learning.
There are two existing approaches to generate concept based video representation. The first is to manually define suitable concepts for each event, which involves too many human efforts and hence is not feasible for large-scale problems. The second is to directly utilize the existing banks built on certain ontologies. However, the main issue is that such banks are not specifically designed for complex events and do not contain enough relevant concepts. Blindly applying these banks with significant amount of irrelevant concepts will degrade the very purpose of high-level concept-based video representation.
This motivates us to build an event-oriented concept library called “Concept Bank” (CB) for high-level representation of complex events in video. In this paper, we identify three challenges for building such a concept library and make significant contributions in developing corresponding solutions summarized below.
The first challenge is that the library should be **quasi-comprehensive** and cover as many real-world events as possible. To address this challenge, we collect and select events from Wikihow [@wikihow], a collaborative writing project that aims to build the world’s largest how-to manual for any possible event[^1]. In this way, we end up with $631$ events as our event collection. To the best of our knowledge, this is the largest event collection in the literature.
The second is how to **discover concepts** and build their visual models for each event. To solve this, we apply our recently proposed idea on automatic event-driven semantic concept discovery from Web images [@FlickrConcept]. The web is a rich source of information with tremendous images captured for various events and these images are roughly annotated with descriptive tags that indicate the semantics of the image contents. Our intuition is that if we crawl enough number of images by an event query, tags of these images could somehow reveal certain semantics related to the event statistically, and suggest relevant concepts of the event. In light of this, we crawl images and their associated tags from Flickr using event queries extracted from WikiHow, then find the most relevant event-specific concepts from the tags and build visual model for each concept. Finally, after we get a comprehensive concept library for all events, it remains unclear on how to choose the most **relevant concepts** to represent any possible query event. To address this task, we propose a semantic matching method to measure the relevance between each concept and the query event (without using any training videos). We then choose the top ranked concepts to represent the event and get a compact concept based video representation.
Figure \[fig\_overview\] illustrates an overview of how we build the proposed Concept Bank and use it to represent an event video. Experimental results show that the video representation with only few hundred dimensions generated by CB achieve state-of-the-art event detection performance.
![**Building Concept Bank** (in the upper three boxes): We collect a comprehensive event list from WikiHow. For each event, we search Flickr images and discover relevant concepts from their tags [@FlickrConcept]. We train a concept detector for each discovered concept, resulting a large set of event-specific concept detectors. **Video Representation by Concept Bank** (in the lower part): Given a query event, relevant concepts are selected based on semantic matching (Section \[sec\_representation\]). Then the selected concept detectors are applied on frames of a video clip, generating a concept score vector on each frame. Finally, we use average-pooling to aggregate all concept score vectors across the frames.[]{data-label="fig_overview"}](overview.pdf){width="1\columnwidth"}
Related Work {#sec_related_work}
============
Video event detection has became an important research area in computer vision literature. Natarajan *et al.* [@natarajan2012fusion] investigated the multimodal fusion of low-level video features and the extracted videotext information in video content, and achieved good performance on the detection. Ye *et al.* [@bimodal] discovered the bi-modal audio-visual codewords and leveraged the joint patterns across the audio and visual space to boost event detection performance. Duan *et al.* [@duan2009domain] incorporated the web sources videos crawled from Youtube to relieve the insufficiency of the number of training videos of an event, and developed a cross-domain video event detection model. Tang *et al.* [@tang2012latent] developed a large margin framework to exploit the latent temporal structure in successive clips of a long event video. These excellent works focus on modeling events into sophisticated statistical models or fusing mutimodal information. However, none of them can reveal the rich semantics in event videos.
There are some recent works that try to perform event detection with semantic concepts. Yang *et al.* [@yang2012deepnet] applied deep belief nets to group a large number of event video shots into a number of shot clusters, and then treat each cluster center as a data-driven concept. Then each video is mapped onto the cluster centers and encoded into a concept based representation. However, such data-driven concepts do not convey any semantic information and hence cannot be utilized for high-level semantic understanding. Liu *et al.* [@liu_concept] manually defined concepts present in event videos and categorized them into “object”, “scene” and “action”. They annotated the presence of each concept on training videos and then built the individual concept detectors. Nevertheless, as aforementioned, this requires too many manual efforts, and is not applicable for real-world large-scale video event detection tasks. Different from these prior works, we focus on automatically discovering potential concepts present in any possible events with interpretable semantic meaning.
Notably, we applied our previous work on concept discovery from Internet images [@FlickrConcept] as a building component of our concept library construction process. This previous work aims at discovering concepts from a pre-specified set of event definitions within a known domain and studying the large beneficial impacts of such event-specific concepts in event retrieval, zero-example detection, and summarization. In contrast, this paper focuses on discovering semantic concepts from an external event knowledge base, WikiHow, and using its rich ontological hierarchy to organize the large number of concepts learned from the Web. Such ontological structure plays an important role in handling novel events that have not been seen in the learning stage, as confirmed by the significant performance gains reported in the experiments.
There are some existing concept libraries built for different purposes. Object bank [@object_bank] consists of $200$ objects taken from the intersection set of most frequent $1,000$ objects between image datasets LabelMe [@labelme] and ImageNet [@imagenet]. Each object detector is trained with $100\sim 200$ images and their object bounding boxes. Classemes [@classemes] is a concept library comprised of $2,659$ concepts defined from Large Scale Concept Ontology for Multimedia (LSCOM) [@LSCOM]. Each concept detector is trained with $150$ images from *bing.com* search engine using the LP-$\beta$ multiple kernel learning algorithm. Action bank [@action_bank] is built for high-level representation of human activities in video, which contains $205$ template actions. Although these libraries achieve good performances on different tasks, they are not designed for video events. In our work, we build a concept library specifically designed for video events, and the concept ontology and models are automatically discovered from the Web.
Building Concept Bank Ontology {#sec_build}
==============================
In this section, we will introduce the procedure of building Concept Bank ontology, including event collecting and concept discovery for each event.
Collecting Events from WikiHow {#sec_wikihow}
------------------------------
In order to get a quasi-comprehensive list of events, we choose among the events from WikiHow [@wikihow]. WikiHow is a wiki, similar to Wikipedia, in which internet users can read or edit the articles. Currently, WikiHow contains $163,957$ articles that are organized into $20$ categories in a hierarchical structure. The $20$ categories cover the major aspects of human daily life including “Food and Entertaining”, “Pets and Animals", “Sports and Fitness" and so on. Figure \[fig\_wikihow\] illustrates a portion of the hierarchical structure of event category “Sports and Fitness” in WikiHow. Each category contains a number of subcategories (*e.g.*, “Individual Sports" and “Outdoor Recreation") that are instantiations of the higher level event catalog. Subcategories are further divided into various fine-grained events in hierarchy (*e.g.*, “Bicycling", “Mountain Biking" and “Fishing”), each has several articles (*e.g.*, “Jump a Mountain Bike" under “Mountain Biking"), which are not shown in the figure, describing the detailed procedure of each specific event.
\[!ht\] ![ A portion of hierarchical structure for event category “Sports and Fitness” in WikiHow. []{data-label="fig_wikihow"}](wikihow_hie.pdf "fig:"){width="1\columnwidth"}
We find that some events in WikiHow are closely related and are sub-events within an event (*e.g.*, “Mountain Biking” and “Unicycling" under “Bicycling"). Our intuition is that organizing events into such fine granularity will result in a huge event collection with heavy semantic redundancy, which is impractical and unnecessary for our task. Therefore, we only focus on the first three layers in WikiHow (above the bold solid red line in Figure \[fig\_wikihow\]) and treat all events in the third layer as potential events in our library, resulting in $1,257$ event candidates in total.
Furthermore, we notice that some events are not visually detectable. For example, events “getting good grades in college” or “being successful” do not convey consistent visual patterns that can be modeled by computer vision techniques. Therefore, a filtering is performed to remove such undetectable events. To this end, we ask two human labelers to judge whether each of the $1,257$ events corresponds to a visually detectable event. After filtering, we end up with $631$ events which are included as the events in our library.
Concept Discovery for Each Event {#sec_flickr}
--------------------------------
After collecting events from WikiHow, we need to associate a number of concepts to each event. One straightforward way is to use advanced Natural Language Processing technique to extract concepts from article contents on WikiHow. However, besides the complexity of the procedure itself, the resultant concepts do not correspond to any visual patterns, which also brings challenges for training data acquisition. On the other hand, the images on Flickr are associated with tags that describe the semantic of visual content in different events. Therefore, we leverage the tags of Flickr images to discover potential concepts for events.
In this work, we adopt our previous work on event-driven concept discovery from Internet images [@FlickrConcept] to discover concepts for WikiHow events. For each of the $631$ events, we use the article names under this event as textual queries to crawl images and their associated tags from *Flickr.com* image search engine. We only keep color images with resolution higher than $200 \times 200$ pixels. The crawling ends up with $727,910$ images in total and the average number of images is $1,154$ per event. Since some tags are not semantically meaningful words (such as a camera brand name) or do not have consistent visual patterns (such as “biology” or “economy”), directly accepting these tags as concepts may introduce lots of noises and degrade the performance. Therefore, we adopt the following two-phase process to discover concepts that have both semantic meanings and visually consistent patterns from tags.
**Step I : Noisy Tag Cleansing.** Notice that a tag could be either a word or a phrase. So firstly, we convert a tag into lower case and divide the tag into tokens by a tokenizer, then remove non-word tokens and tokens belonging to stop words (like “the”, “and”, or “with”) [@nltk]. We then define a meaningless-word-list including camera names (like iPhone or Nikon) and people names (like Peter or Julia). For a tag, if one of its tokens appears in our meaningless-word-list or does not match any synonyms in WordNet, the tag is regarded as meaningless. During this step, we also bundle together tokens within a WordNet synonym with lemmatization [@nltk]. For example, “run”, “running”, “runs” and “ran” will be regarded as a same token “run". By doing this, only the tags with semantic meaning are retained. Finally, we rank tags within each event in descending order by their frequencies and select top $100$ tags as candidate concepts.
**Step II : Concept Visualness Verification.** If a concept is not visually related, the concept detector will be hard to generalize due to high intra-class variance [@auto_attributes]. Therefore, we use the cross-validation performance as the measure of visualness. Specifically, for a candidate concept, we collect images with the concept as positive samples and randomly select a same number of images which do not contain the concept as negative samples. Then we build a concept detector (see Section \[sec\_classifier\] for details) and calculate the 2-fold cross-validation performance. In this procedure, we adopt Average Precision (AP) as the performance evaluation metric. Finally, only candidate concepts with cross-validation APs higher than $0.8$ are selected as concepts for the event.
In order to get statistically reliable concept detectors, we further do a post processing that removes those concepts contain less than $100$ training images. By doing this, we end up with $4,876$ concepts. Note that the same concept may appear in different events (and are trained with different training images, see Section \[sec\_classifier\]). Therefore, we call our discovered concepts as “event-specific concepts”.
Concept Bank Ontology {#sec_structure}
---------------------
Based on the collected events and concepts, we construct a five-layer tree structured concept ontology that helps event detection (see Section \[sec\_representation\]). In more details, the root layer is a single node covering all events and concepts in the Concept Bank. The second layer consists of $19$ event categories ranging from “Arts and Entertainment” to “Travel”, each corresponds to a category in WikiHow. In the third layer, we have in total $130$ nodes in which each node corresponds to an event subcategory. In the fourth layer, we have $631$ events. Furthermore, each of the $631$ event nodes has several child nodes corresponding to its event-specific concepts. Therefore, the bottom layer of the CB tree ontology contains $4,876$ leaf nodes (concepts). Figure \[fig\_structure\] illustrates a branch of hierarchy structure in CB ontology corresponding to event category “Sports and Fitness”. Due to the space limitation, we remove the root node and only show at most $8$ concepts for each event. In the next section, we will associate a concept detector with each of the leaf nodes.
\[!ht\] ![A branch of the hierarchy structure in Concept Bank ontology corresponding to event category “Sports and Fitness”.[]{data-label="fig_structure"}](structure.pdf "fig:"){width="1\columnwidth"}
Learning Concept Detector {#sec_classifier}
=========================
In this section, we will discuss how to build concept detector for each discovered concept.
Feature Extraction {#sec_feature}
------------------
In order to describe visual content information comprehensively and complementarily, we select $5$ low-level feature descriptors that are proven to be effective in representing visual content. They are respectively **SIFT** [@sift], **GIST** [@gist], **Gabor** [@gabor], **LBP** [@lbp] and **Transformed Color Distribution** [@color]. All descriptors are densely extracted on grids of $20 \times 20$ pixels with $50\%$ overlap from images. For each type of the extracted descriptors, we train a codebook with $1,000$ codewords, and partition each image into $1 \times 1$, $2 \times 2$ and $3 \times 1$ blocks for spatial pyramid matching [@spm]. Finally, we adopt soft quantization [@soft_quantization] to represent each image as a $8,000$ dimensional histogram for each feature. We also adopt these features in generating concept scores for videos.
Training Image Selection {#sec_image_selection}
------------------------
Since images and their tags are crawled from the web, images associated with a concept are noisy and contain some outliers. Therefore, we need to remove the noisy images before training concept detectors. Our intuition is that the majority of images in a concept form a common “visual theme” in the image search result, while distracting images tend to depart from such visual commonality. This motivates us to develop a collective approach to measure the relative confidence of each image in the search result. Specifically, for each concept $c$, we collect a set of images $F=\{1,\ldots,N\}$ that contain the concept as their tag, where $N$ is the total number of images. For image $i$, we extract $M$ features ($M=5$ in this work) and denote its feature vector as $\mathbf{f}_{m}^{i}\in\mathbb{R}^{d_m}$ ($m=1,\ldots,M$) with $d_{m}$ denoting the feature dimensionality of the $m$-th feature. We adopt Kernel Density Estimation (KDE) method [@kde] to estimate the probability of image $i$ belonging to concept $c$ defined as: $$p(c | i) = \frac{1}{M\times N} \sum_{j=1}^{N} \sum_{m = 1}^M \mathcal{G}_m (\mathbf{f}_m^i - \mathbf{f}_{m}^{j}),$$ where $\mathcal{G}_m(\cdot)$ is the Gaussian kernel function of the $m$-th feature defined as: $$\mathcal{G}_m (\mathbf{f}_{m}^i - \mathbf{f}_{m}^j) = \exp (-\frac{\| \mathbf{f}_{m}^i - \mathbf{f}_{m}^j \|^2}{\sigma_m^2}),$$ in which $\|\cdot\|$ denotes the $l_2$ norm, $\sigma_m$ is the kernel radius of the $m$-th feature setting as the mean value of all pairwise distance among the images.
We use $p(c|i)$ as the confidence score of image $i$ belonging to concept $c$ and select $s$ images with highest confidence scores as the positive training samples for each concept. In this work, we set $s=100$ and choose $t=1,000$ images from other concepts as negative training samples. Figure \[fig\_noise\] shows $5$ top selected images for exemplary concepts, from which we can see that the selected images are highly reliable while at the same time achieve reasonable diversity.
\[!ht\] ![Top $5$ selected Web images for $3$ exemplary concepts using our training image selection method.[]{data-label="fig_noise"}](KDE.pdf "fig:"){width="1\columnwidth"}
Since a concept may appear under different events, our method will select different training images for each event-specific concept detector. Therefore, each concept detector will convey context information for its belonging event.
Concept Detector Training
-------------------------
After selecting $s$ positive and $t$ negative training images for each concept, we use Multiple Kernel Linear SVM (MK-LSVM) classifier with L1-Loss to train our concept detector. The dual norm of its objective function is: $$\begin{aligned}
\nonumber
%\vspace{-4mm}
\min_{\boldsymbol{\alpha}, \boldsymbol{\beta}} && \frac{1}{2} \boldsymbol{\alpha}^T \big( \sum_{m=1}^M \beta_m \mathcal{K}_m + \gamma I \big) \boldsymbol{\alpha} - \mathds{1}^T \boldsymbol{\alpha} \\ \nonumber
\mathrm{s.t.} && \boldsymbol{\alpha}^T \boldsymbol{y} = 0,\ 0 \leq \alpha_i \leq C,~~ \ \forall i, \\
&& \sum_{m=1}^M \beta_m = 1,\ \beta_m \geq 0,~~ \ \forall m,
%\vspace{-4mm}\end{aligned}$$ where $\mathbf{y}=[y_1,\ldots,y_{s+t}]^{\top}\in\{0,1\}^{s+t}$ is the label vector with $y_i=1$ if the $i$-th image is positive and $y_i=0$ otherwise, $\boldsymbol{\alpha} = [\alpha_1,\ldots,\alpha_{s+t}]^{\top}\in\mathbb{R}^{s+t}$ is the dual variable vector, $\boldsymbol{\beta}=[\beta_1,\ldots,\beta_M]^{\top}\in\mathbb{R}^{M}$ is the kernel weight vector, $\mathcal{K}_m\in\mathbb{R}^{(s+t)\times(s+t)}$ is the Gram matrix of the $m$-th feature, $I$ is an identity matrix , $\mathds{1}$ is the all-one vector, $\gamma$ and $C$ are model parameters. We use LIBLINEAR [@liblinear] and choose $\gamma = 0.01$ and $C=1$ in training concept detectors.
Representing Videos with Concept Bank {#sec_representation}
=====================================
Given a video clip, we can directly apply all $4,876$ concept detectors in Concept Bank to generate the concept representation. However, as aforementioned, it is desirable to choose the most relevant concepts related to the query event and generate a compact concept representation.
In our work, we turn to use the semantic similarity matching between event and concept names to find relevant concepts for an event query. In order to determine semantic similarity, we adopt ConceptNet [@conceptnet], a semantic network developed by MIT Media Lab containing common sense knowledge for computers to capture human world. We use Divisi, a Python API for ConceptNet, to calculate semantic similarity between two words. The similarity between two phrases are the maximum similarity between all pairs of words in the two phrases.
Denote by $sim(a, b)$ the similarity between two words (phrases) $a$ and $b$ calculated from ConceptNet. For a given event $e$ and a concept $c_0$, we take advantage of the hierarchy in the Concept Bank ontology to calculate their semantic similarity. Specifically, denote $c_1$, $c_2$ and $c_3$ as the three ancestors of concept $c_0$ in Concept Bank ontology residing at event layer, event subcategory layer and event category layer respectively. Then the semantic similarity $S(e,c_0)$ between $e$ and $c_0$ can be calculated as follows: $$\label{Eq:HieSim}
%S(e,c) = sim(e,c)\times sim(e,c_1) \times sim(e,c_2)\times sim(e,c_3),
S(e,c_0) = \prod_{i=0}^3 sim(e, c_i).$$ For instance, for event “grooming a dog”, we calculate its similarity with concept “brush” (similarity value $0.7$) and its three ancestors: , “dog grooming” (similarity value $1$), “dogs” (similarity value $1$), and “pets and animals” (similarity value $0.8$). Then the similarity between event “grooming a dog” and concept “brush” is $0.8 \times 1 \times 1 \times 0.7 = 0.56$.
Eq. (\[Eq:HieSim\]) can be justified as follows. Since the same concept may appear under different events, involving high-level ancestors in the similarity calculation will measure the relevance of each concept based on the coherent similarity of its ancestors with respect to the event query, leading to more precise relevant concept selection. In this way, when answering event query “grooming a dog”, we can effectively select concept “brush” from event “dog grooming” rather than from event “personal makeup” under “Personal Care and Style" even if it also contains concept “brush”.
With semantic similarity matching, we choose the top $n$ most relevant concepts for the query event and use their concept detectors to generate concept based video representation. Specifically, we evenly sample $m$ frames from a video clip. Then we apply the selected concept detectors on each frame and adopt their probabilistic outputs as concept scores. The final representation of the video is the average of concept score vectors across the sampled frames. We choose $n=100$ and $m=20$ in the experiments.
Experimental Evaluations {#sec_experiment}
========================
In this section, we evaluate the effectiveness of the Concept Bank in video event detection. We first introduce the datasets and comparison methods and then present experiment analysis on two event detection tasks. Specifically, we would like to evaluate whether the large number of concepts discovered using WikiHow events can be used to effectively detect events specified in some benchmark datasets. Additionally, we aim at comparing different ways of training the concept detectors and mapping events to concepts.
Dataset and Comparison Methods
------------------------------
**Dataset**. We evaluate our proposed CB representation on two video event detection datasets: (1) **TRECVID 2013 Multimedia Event Detection (MED) dataset**. This is a dataset of $32,744$ videos over $20$ event categories (i.e., the partition used in the Pre-Specified EK$100$ task [@MED]). These $20$ event categories are (from E1 to E20 respectively): “*birthday party*", “*changing a vehicle tire*", “*flash mob gathering*", “*getting a vehicle unstuck*", “*grooming an animal*", “*making a sandwich*", “*parade*", “*parkour*", “*repairing an appliance*", “*working on a sewing project*", “*attempting a bike trick*", “*cleaning an appliance*", “*dog show*", “*giving directions to a location*", “*marriage proposal*", “*renovating a home*", “*rock climbing*", “*town hall meeting*", “*winning a race without a vehicle*", “*working on a metal crafts project*". We follow the standard partition of this dataset, which includes $7,787$ videos in the training set and $24,957$ videos in the test set. Achieving good performance on this dataset is quite challenging because the majority of videos in this dataset are background videos (only $2,000$ videos in training set and $1,489$ videos in test set belong to one of the $20$ events). (2) **Columbia Consumer Video (CCV) dataset**. This is a dataset of $9,317$ YouTube videos over $20$ event categories, where $4,659$ videos are used for training and $4,658$ videos for testing [@ccv]. These $20$ categories are (from E1 to E20 respectively): “*basketball*", “*baseball*", “*soccer*", “*ice skating*", “*skiing*", “*swimming*", “*biking*", “*cat*", “*dog*", “*bird*", “*graduation*", “*birthday*", “*wedding reception*", “*wedding ceremony*", “*wedding dance*", “*music performance*", “*non-music performance*", “*parade*", “*beach*", “*playground*".
**Comparison Methods**. We compare the concept based video representations generated from the following methods. (1) **Classemes** [@classemes], a $2,659$ dimensional concept representation built on the LSCOM concept ontology [@LSCOM]. We generate Classemes feature from each video frame and then average all frame features across a video as the video-level feature. (2) Selected Concepts from Best single Feature (**SCBF**). The concept detector is trained based on the best performing single feature, in which the training images are selected with the method described in Section \[sec\_image\_selection\]. When a query event comes, it uses the semantic matching method in Section \[sec\_representation\] to select relevant concepts for video representation. (3) Selected Concepts from Random Images (**SCRI**). The concept detector is trained by MK-LSVM, but the training images are randomly selected. The relevant concepts are selected using semantic matching. (4) Random Concepts from Selected Images (**RCSI**). The concept detector is trained by MK-LSVM with selected training images. However, concepts for the query event are randomly selected, instead of using semantic matching. (5) Our Selected Concepts from Selected Images (**SCSI**). The concept detector is trained by MK-LSVM with selected training images. The relevant concepts are selected based on semantic matching.
**Evaluation Tasks**. For each method, we evaluate the performance of different representations under two video event detection tasks: Event Detection in Concept Space and Zero-Shot Event Retrieval. In each task, the Average Precision (AP), which approximates the area under precision/recall curve, is used as evaluation metric on each event. We calculate mean Average Precision (mAP) over all event categories as the overall evaluation metric on the dataset.
Event Detection in Concept Space {#sec_ed}
--------------------------------
In this task, all representations are regarded as high-level descriptors in concept space for video event detection. On each dataset, we choose $100$ concepts for each of $20$ events from the Concept Bank and concatenate the concept scores into a $2,000$-dimensional feature vector. A one-vs-all SVM classifier with $\chi^2$ kernel is trained as the event detector. In the test stage, we use SVM probabilistic output as the event confidence score on each test video. Best parameters for SVM are chosen via $2$-fold cross-validation on training set.
The comparison results between the five methods on MED and CCV are given in Figure \[fig\_ed\], from which we have following observations: (1) The proposed SCSI method achieves the best performance by a significant margin compared with other methods, which demonstrates the effectiveness of SCSI in generating reliable concept based video representations. (2) The multiple feature based methods including SCSI and SCRI clearly beat the single feature based method SCBF. This shows the advantages of utilizing multiple features in concept detector training. (3) The SCSI significantly outperforms RCSI, which verifies that our semantic similarity matching method is able to select relevant concepts for event representation. (4) The SCSI has higher performance than SCRI. This is due to the fact that the former leverages more reliable training images in concept detector training than the random images utilized in the latter.
Figure \[fig\_ed\_noc\] illustrates the variation of mAP with different number of concepts. For SCSI, SCRI, SCBF and RCSI, we choose the number of concepts from $10$ to $100$ with step of $10$ for each event, and get concept numbers rang from $200$ to $2,000$. For Classemes, we fix the concept dimension to be $2,659$ (lower dimension of classemes leads to worse performance). As seen, the proposed SCSI method consistently outperforms all other methods when different numbers of concepts are adopted in each of the two datasets. This clearly demonstrates that our method is able to generate accurate concept scores on the videos to facilitate concept based event detection. Furthermore, the event detection performance keeps improving as the number of concepts increases. This shows that event modeling over concept space can benefit from the increasing number of relevant concepts. We also notice that, on both MED and CCV datasets, the proposed SCSI outperforms $2,659$ dimension Classemes even with only $200$ dimensions ($10$ concepts for each of the $20$ events), which again clearly demonstrates the advantages of our CB representation.
[\[fig\_ed:a\] ![ Performance comparison on event detection in concept space task (left: MED; right: CCV). []{data-label="fig_ed"}](AP_MED_ED.pdf "fig:"){width="1\columnwidth"} ]{} [ \[fig\_ed:b\] ![ Performance comparison on event detection in concept space task (left: MED; right: CCV). []{data-label="fig_ed"}](AP_CCV_ED.pdf "fig:"){width="1\columnwidth"} ]{}
[\[fig\_ed\_noc:a\] ![ Event detection performance with different number of concepts (left: MED; right: CCV). []{data-label="fig_ed_noc"}](NOC_MED_ED.pdf "fig:"){width="0.466\columnwidth"} ]{} [ \[fig\_ed\_noc:b\] ![ Event detection performance with different number of concepts (left: MED; right: CCV). []{data-label="fig_ed_noc"}](NOC_CCV_ED.pdf "fig:"){width="0.482\columnwidth"} ]{}
Zero-Shot Event Retrieval {#sec_er}
-------------------------
In this task, we do not use any training videos, but directly use the concept scores on the test video to perform event retrieval. We call this zero-shot event retrieval since the procedure is purely semantic based without using any training examples. Specifically, we rank all test videos based on the detection scores of each selected relevant concept and then fuse the rank lists of all relevant concepts to generate the final rank list. In this work, we adopt the normalized rank fusion method [@rank_fusion] to combine multiple ranking lists. Specifically, for a test set with $n$ videos, the fusion score $R(i)$ for $i$-th video is calculated by $R(i) = \frac{1}{d} \sum_{j=1}^d ( 1 - \frac{r_j}{n} )$, where $r_j \in \{1, 2, \dots, n\}$ is the rank position of $i$-th video in the rank list generated by $j$-th concept and $d$ is the concept number. In this way, we get rid of the influence caused by numeric scale differences of raw concept scores generated from different concept detectors. Finally, we rank test videos based on the fusion scores.
In zero-shot retrieval, since the only available information is the query event name, we do not utilize concept names from other event categories. Therefore, we only choose $100$ relevant concepts selected for the specific query event based on semantic similarity matching described in Section \[sec\_representation\]. To get a fair comparison, we also select $100$ concepts from Classemes using the same method.
Figure \[fig\_sr\] shows the comparison results between the five methods for zero-shot retrieval task and Figure \[fig\_sr\_noc\] illustrates the variation of mAP with different number of concepts. There are two points we want to mention: (1) The performance of zero-shot event retrieval is worse than that of supervised event modeling over concept space (Section \[sec\_ed\]). This is because the latter uses training samples to obtain a more sophisticated event model while the former is merely based on semantic score fusion. Notably, since the concepts are all randomly chosen and irrelevant to the query event, the performance of RCSI degrades a lot and is close to chance in this task. (2) Our SCSI method keeps improving as the number of concepts, whereas the performance of Classemes decreases. This indicates that Classemes does not have enough concepts related to the event, so the increasing number of irrelevant concepts will not help the detection. However, our Concept Bank has a broader coverage thus it can benefit from the increasing number of concepts.
[\[fig\_sr:a\] ![ Performance comparison on zero-shot event retrieval task (left: MED; right: CCV). []{data-label="fig_sr"}](AP_MED_ER.pdf "fig:"){width="1\columnwidth"} ]{} [ \[fig\_sr:b\] ![ Performance comparison on zero-shot event retrieval task (left: MED; right: CCV). []{data-label="fig_sr"}](AP_CCV_ER.pdf "fig:"){width="1\columnwidth"} ]{}
[\[fig\_sr\_noc:a\] ![ Zero-shot event retrieval performance with different number of concepts (left: MED; right: CCV). []{data-label="fig_sr_noc"}](NOC_MED_ER.pdf "fig:"){width="0.474\columnwidth"} ]{} [ \[fig\_sr\_noc:b\] ![ Zero-shot event retrieval performance with different number of concepts (left: MED; right: CCV). []{data-label="fig_sr_noc"}](NOC_CCV_ER.pdf "fig:"){width="0.474\columnwidth"} ]{}
Semantic Recounting in Videos
-----------------------------
For each video of a target event, we rank all the concepts discovered for the event based on their confidence scores and treat the top ranked concepts as the semantic description of the video content. Such a procedure is able to reveal the semantic information contained in a video and is thus called video semantic recounting. Figure \[fig\_recounting\] shows the recounting result on videos from some exemplary events in MED and CCV, in which the top $5$ ranked concepts generated by our method are selected as concepts for each video. As can be seen, these concepts reveal the semantics contained in the videos, which verifies the effectiveness of the our discovered concepts in representing video semantics.
Computational Cost
------------------
The proposed Concept Bank is efficient computationally. We calculate the running time on a 2.8GHz Windows workstation. In average, for a event query, semantic matching needs $78$ seconds with Python. For a video, the average time to generate a $100$ dimension concept score (including feature extraction) is around $110$ seconds with MATLAB.
![ Event Video Recounting Results: each of the $6$ rows shows evenly subsampled frames of an example video and the top $5$ relevant concepts detected in the video. []{data-label="fig_recounting"}](REC.pdf){width="1\columnwidth"}
Conclusions and Future Work {#sec_conclusion}
===========================
We have introduced Concept Bank, a concept library specifically designed for complex event representation in videos. The library consists of $4,876$ concepts organized in a five-layer tree structure, in which the higher-level nodes correspond to the event hierarchies and the leaf nodes are the event-specific concepts. To include as many real-world events in the library as possible, we collect all visually detectable events in WikiHow. For each event, we use its article names in WikiHow to search Flickr and then discover relevant concepts from the tags associated with the crawled images of this event. A Multiple Kennel Linear SVM classifier is then trained for each concept with the crawled images as the atomic concept detector. In addition, we also develop a semantic matching method to determine relevant concepts for a coming event query. Experiments over two event detection tasks verify the effectiveness of the proposed Concept Bank. For future work, we will further extend the Concept Bank with motion concepts, which are trained based on the spatial-temporal features extracted from videos.
[^1]: Although WikiHow only focuses on how to do anything, we realize that it is created based on user’s common interests and thus has good coverage on almost every aspect of human daily life. Advanced event collection techniques can be also used to extract visually detectable events from other knowledge bases, and easily incorporated into our current system.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'One way to detect the presence of new particles in theories beyond the standard model is through their contribution to electroweak loop effects. We comment on the importance of a consistent inclusion of their mixing angles to ensure that the physical requirement of heavy particle decoupling is fulfilled. We illustrate our points by a detailed discussion of the lepton flavor changing effect $\mu \rightarrow e\gamma ,$ investigated recently by Kitano, in the Randall-Sundrum model. Our remarks are equally applicable to models with large compactified dimensions where bulk neutrinos are introduced to account for the observed neutrino oscillations.'
author:
- |
[T. P. Cheng]{}$^{\ast }$ [and]{} [Ling-Fong Li]{}$^{\dagger
} $[ ]{}\
$^{\ast }$[Department of Physics and Astronomy, University of Missouri, St. Louis, MO 63121]{}\
$^{\dagger }$[Department of Physics, Carnegie Mellon University, Pittsburgh, PA 15213]{}
title: '**Heavy particle electroweak loop effects in extra-dimensional models with bulk neutrinos**'
---
Introduction
============
There is considerable interest in the attempts of solving the gauge hierarchy problem in the context of theories invoking extra spatial dimensions. One category of models assumes that the extra dimensions are large. Mass hierarchy results from the large volume effect[@dimo]. Another class makes use of a non-factorizable metric with a warp factor, leading to exponential suppressions of Planck scale masses for the relevant fields which are assumed to reside on the “visible” 3-brane[@RS]. Only gravitons and, in some models, also other fields, can propagate in the extra dimensional space. Such a bulk field will have a tower of Kaluza-Klein states with ever increasing masses. These states often provide us with definitive signatures of these extra dimension theories. If such bulk fields can mix with ordinary SM fields, their presence can in principle be detected through their contributions to the electroweak loop effects. There exist already a substantial body of literature discussing the constraints on such KK states by using the existing, or future, electroweak precision data[@misc]. In this paper we wish to emphasize the importance of a proper and complete inclusion of the mixing angle effects so that the physically sensible requirement of the decoupling theorem can be satisfied.
We believe that the study of any physical phenomena at a given distance scale should not depend sensitively on our knowledge of the physics on much shorter scales, heavy particles should decouple from low-energy processes. Namely, the effects of heavy particles in the virtual intermediate states are suppressed by inverse powers of the heavy particle masses[@app-carr-thm]. This comes about because the relevant amplitudes are reduced by the heavy particle propagators. However, if the heavy mass comes from spontaneous symmetry breaking, the corresponding Yukawa coupling is also large and this can neutralize the large mass of the denominator. This may lead to a violation of the decoupling theorem in the low energy effective theory[@CL91]. For example, in the Standard Model, the $\rho $ parameter grows with $m_{t}^{2}$ and is quite sensitive to the value of $m_{t}.$ In fact, this is one of the clue to the $t$-quark mass before its discovery. On the other hand, if the large mass can be attributed to a gauge invariant mass term, then decoupling should be effective because here one does not need a large Yukawa coupling. Often in a model, particles have a mixture of bare and Yukawa-coupling-induced masses. These decoupling effects may show up as mass-suppressed mixing angles. Masses and mixing angles are often related because they all follow from the same (nondiagonal) mass matrices.
A proto-typical case is the seesaw mechanism for generating neutrino masses [@seesaw]: besides the ordinary light neutrinos with masses $m_{\nu
}\simeq \hat{m}^{2}/\hat{M},$ there is also at least one other superheavy neutrino with a mass $m_{N}\simeq \hat{M}.\;$\[The small mass is assumed to have a magnitude comparable to the masses of ordinary quarks and charged leptons, $\hat{m}\simeq 1\,GeV$, the large mass being an intermediate mass scale below the Planck scale, $\hat{M}\simeq 10^{12}GeV$.\] In the charged weak current, the charged lepton is coupled to the combination $\left( \cos
\theta \left| \nu \right\rangle +\sin \theta \left| N\right\rangle \right) ,$ where $\left| \nu \right\rangle $ stands for some superposition of the light neutrino states with mixing angles that are not necessarily small, but the angle $\theta $ is mass-suppressed $$\theta \simeq \frac{\hat{m}}{\hat{M}}. \label{seesaw-angle}$$ Such a mixing angle simply reflects the property of the mass matrix that, in the $\hat{M}\rightarrow \infty $ limit, the mixing of the singlet neutrino goes to zero. The presence of non-zero neutrino masses naturally leads to flavor violation loop effects such as $\mu \rightarrow e\gamma .\;$Both light and heavy neutrinos contribute, leading to a branching ratio[@CL-ps00] $$B\left( \mu e\gamma \right) =\frac{3\alpha }{8\pi }\zeta ^{2}\theta ^{4}
\label{seesaw-muegamma}$$ where $\alpha $ is the fine structure constant. The factor $\zeta $ being some product of the mixing coefficients among light neutrinos is not expected to be particularly small. Had one not taken into account of the fact that the heavy-light mixing angle $\theta $ is mass-suppressed, one would erroneously conclude that the heavy neutrino did not decouple (and thus giving rise to an unacceptably large branching ratio). But in this representation of the neutrino states, decoupling manifest itself in the form $\theta ^{4}=\left( \hat{m}^{2}/\hat{M}^{2}\right) ^{2}=m_{\nu
}^{2}/m_{N}^{2}$ which yields an immeasurably small branching ratio — because of the superheavy neutrino mass in the denominator, as well as the tiny light neutrino mass in the numerator. The seesaw model of neutrino mass is considered to be an attractive possibility because the presence of such self-consistent features. We suggest that any physically sensible theory containing superheavy particles would have this type of properties that automatically ensures that the heavy states are decoupled in low energy processes.
The mass suppressed mixing angle follows from a special feature of the seesaw neutrino mass matrix — the absence of Majorana mass terms for the left handed neutrino and a superheavy entry for the right handed neutrino mass term. In essence, the reason that the decoupling holds in this case is due to the fact the large mass can be realized by having large bare mass ($\nu _{R}$ being a SM singlet) without having large Yukawa coupling. That mass matrices have the structure which gives rise to decoupling is rather common in models involving heavy particle states. Thus the question of mass suppressed mixing angles is very important in our consideration of heavy particle contribution to low energy loop effects. In this paper we shall illustrate our points in the Randall-Sundrum model[@RS] with bulk neutrinos[@GN]. The investigation of lepton flavor violation loop effects in this context have recently been carried out by Kitano[@kitano]. Here we complete his discussion, in particular with respect the possibility of extracting a meaningful bound on the heavy neutrino mass.
Mixing angles in the RS model with bulk neutrinos
=================================================
The Randal-Sundrum model presupposes a five dimensional spacetime. The extra spatial dimension is taken to be a compactified $S^{1}/Z_{2}$ orbifold with a coordinate $y=r_{c}\phi ,$ with $r_{c}$ being the radius of the compact dimension and the angle $\phi $ having a range of $\left[ -\pi ,\pi \right] $ with opposite sides identified. There are two 3-branes fixed at $\phi =\pi $ (the “visible” brane containing the SM fields) and at $\phi =0$ (the “hidden” brane, also called the Planck brane). The resultant metric is non-factorizable: $$ds^{2}=e^{-2kr_{c}\left| \phi \right| }\eta _{\mu \nu }dx^{\mu }dx^{\nu
}-r_{c}^{2}d\phi ^{2}$$ where $k$, the bulk curvature, has the order of fundamental mass scale $\hat{M}_{5}$ which is comparable to the Planck mass. The exponential warp factor $e^{-2kr_{c}\left| \phi \right| }$ causes a rescaling of the fields, which changes any mass parameter in the fundamental theory ($\simeq $ Planck scale) to an effective mass on the visible brane as $M=e^{-kr_{c}\pi }\hat{M}_{5}$ ($\simeq $ electroweak scale $M_{EW}$). Namely, for a choice of $kr_{c}\approx 12,$ we can have $$\epsilon =e^{-kr_{c}\pi }\approx 10^{-16}\;\;\;\;\;\;\;\;M=\epsilon \hat{M}_{5}\approx 10^{3}GeV.$$
Because this mechanism does not allow any intermediate scale, between the Planck and electroweak, to appear, the seesaw mechanism for generating a naturally small neutrino mass is not applicable in the original RS model. In this connection, Grossman and Neubert[@GN] introduced a bulk fermion field[^1]. They have shown that, for a reasonable range of parameters, the zero mode of such a fermion has a very small wavefunction at the physical brane and the Higgs generated mass can also be naturally suppressed. In this way, neutrino masses that are many orders smaller than $M_{EW}$ can be obtained.
The bulk fermion (with mass $M_{b}$) has the Kaluza-Klein decomposition of $$\Psi _{5}^{L,R}\left( x,\phi \right) =\frac{e^{2kr_{c}\left| \phi \right| }}{\sqrt{r_{c}}}\sum_{n}\hat{f}_{n}^{L,R}\left( \phi \right) \psi
_{n}^{L,R}\left( x\right) \label{KK}$$ where the superscripts $\left( L,R\right) $ signify the chirality states $\Psi _{5}^{L,R}={\frac12}\left( 1\mp \gamma _{5}\right) \Psi _{5},$ and $\left\{ \hat{f}_{n}^{L,R}\left( \phi \right) \right\} $ are the appropriate sets of complete orthonormal functions (in this case some combinations of Bessel functions) normalized so that $\psi _{n}\left( x\right) $ has the canonical scale in four dimensions, $$S_{\psi }=\int d^{4}x\left\{ \bar{\psi}_{n}\left( x\right) i\NEG{\partial}\psi _{n}\left( x\right) -M_{n}\bar{\psi}_{n}\left( x\right) \psi
_{n}\left( x\right) \right\} . \label{dirac}$$ The KK fields $\psi _{n\neq 0}^{L,R}\left( x\right) $ has electroweak scale masses $M_{n}=\epsilon kx_{n}$ with $x_{n}$ (corresponding to zeros of some combinations of the Bessel functions) being of order one. The presence of such states brings hope for experimental searches, or equivalently, for severe constraints by known phenomenology. Our focus in this paper is the proper accounting, in such analyses, of the important effects due to the mixing angles between these heavy states and the SM fields.
Grossman and Neubert[@GN] have shown that bulk fermion zero modes $\left( x_{0}=0\right) $ exist. If we impose the orbifold symmetry $\phi
\rightarrow -\phi ,$ then only one of the chiral zero modes survives. Let it be $\psi _{0}^{R}\left( x\right) ,$ which has a suppressed wavefunction on the visible brane $$\hat{f}_{0}^{R}\left( \phi =\pi \right) \simeq \sqrt{\epsilon kr_{c}}\epsilon ^{\nu -{\frac12}}=O\left( \epsilon ^{\nu }\right) \;\;\;\;\;\;\;\text{with }\nu =\frac{M_{b}}{k}>\frac{1}{2}. \label{n0-wf}$$ where $M_{b}$ is the bulk fermion mass parameter in the original $5$-dimensional Lagrangian. Similarly, orbifold symmetry requires the wavefunctions for the left-handed KK excitations, when evaluated on the visible brane, to vanish $\hat{f}_{n}^{L}\left( \phi =\pi \right) =0$ while those for the right-handed fields have values $$\hat{f}_{n\neq 0}^{R}\left( \phi =\pi \right) \simeq \sqrt{\epsilon kr_{c}}=O\left( \epsilon ^{{\frac12}}\right) . \label{n-wf}$$ Thus, $\left( \hat{f}_{0}^{R}/\hat{f}_{n\neq 0}^{R}\right) _{\phi =\pi
}=O\left( \epsilon ^{\nu -{\frac12}}\right) $ is quite small since $\epsilon $ is tiny and $\nu >\frac{1}{2}.$
Relevant to our discussion of neutrino mass matrix, we shall only display the Yukawa interaction between the SM left-handed lepton doublet $L^{L}=\left( l^{L},\,\nu _{l}^{L}\right) ,$ the right-handed bulk fermion $\Psi _{5}^{R}$, and the Higgs doublet $H=\left( h^{+},h^{0}\right) ,$ with its conjugate being $\tilde{H}=i\sigma _{2}H^{\ast }.\;$Again for simplicity we shall suppress the lepton generation indices $\left( e,\mu ,\tau \right) $ at this stage. $$S_{Y}=-\int d^{4}x\epsilon ^{4}\hat{Y}_{5}\left\{ \bar{L}_{5}^{L}\left(
x\right) \tilde{H}_{5}\left( x\right) \Psi _{5}^{R}\left( x,\pi \right)
+h.c.\right\} \label{5dYukawa}$$ where the factor $\epsilon ^{4}$ originates from the square root of the metric determinant, and $\hat{Y}_{5}$, the fundamental Yukawa coupling, is dimensionful, expected to be somewhat less than $\hat{M}_{5}^{-{\frac12}};$ and the fundamental fields $L_{5}^{L}\left( x\right) $ and $\tilde{H}_{5}\left( x\right) $ can be replaced by the effective fields (which have the canonical normalizations in the four dimensional spacetime): $\epsilon
^{-3/2}L\left( x\right) $ and $\epsilon ^{-1}H\left( x\right) ,$ respectively. After substituting in the KK decomposition of Eq.(\[KK\]), the Yukawa interaction in Eq.(\[5dYukawa\]) has now the form: $$S_{Y}=-\int d^{4}x\left\{ y_{0}\bar{L}^{L}\left( x\right) \tilde{H}\left(
x\right) \psi _{0}^{R}\left( x\right) +\sum_{n=1}y_{n}\bar{L}^{L}\left(
x\right) \tilde{H}\left( x\right) \psi _{n}^{R}\left( x\right) +h.c.\right\}
\label{yukawa}$$ where $y_{n}=\hat{Y}_{5}\hat{f}_{n}^{R}\left( \phi =\pi \right) /\sqrt{\epsilon r_{c}}$. After using the estimates of Eqs.(\[n0-wf\]) and (\[n-wf\]), and $\sqrt{k}\hat{Y}_{5}\lesssim 1$, the four dimensional Yukawa couplings have the size of $$y_{0}\lesssim \epsilon ^{\nu -{\frac12}}\;\;\;\;\;\text{and\ \ \ \ \ }y_{n}\lesssim 1.$$ Spontaneous symmetry breaking due to the Higgs mechanism results in a non-zero vacuum expectation value for the neutral scalar field $\left\langle
h^{0}\right\rangle =v$ of the electroweak scale. Eq.(\[yukawa\]) leads to mass terms : $$m_{0}\bar{\nu}_{l}^{L}\psi _{0}^{R}+\sum_{n=1}m_{n}\bar{\nu}_{l}^{L}\psi
_{n}^{R}$$ with the “Yukawa masses” $$m_{0}=y_{0}v\lesssim \epsilon ^{\nu -{\frac12}}v\ll M_{EW}\;\;\;\;\;\;\text{and\ \ \ \ \ \ }m_{n}=y_{n}v\lesssim M_{EW}.\;$$ Combining with the Dirac mass terms of the KK states $\sum_{n=1}M_{n}\bar{\psi}^{L}\psi _{n}^{R}$, these mass terms can be written in a matrix form $$\left(
\begin{array}{cccc}
\bar{\nu}^{^{\prime }L}, & \bar{\psi}_{1}^{L}, & \bar{\psi}_{2}^{L}, & ...
\end{array}
\right) \left(
\begin{array}{cccc}
m_{0} & m_{1} & m_{2} & .\,. \\
0 & M_{1} & 0 & .\,. \\
0 & 0 & M_{2} & .\,. \\
: & : & : &
\end{array}
\right) \left(
\begin{array}{c}
\psi _{0}^{R} \\
\psi _{1}^{R} \\
\psi _{2}^{R} \\
:
\end{array}
\right) .$$
For simplicity, let us concentrate on the simplest nontrivial case by cutting off the $n>1$ excitations, thus a neutrino mass term of $\bar{\Psi}_{L}^{\prime }\mathbb{M\Psi }_{R}^{\prime },$ where $$\bar{\Psi}_{L}^{\prime }=\left(
\begin{array}{cc}
\bar{\nu}_{l}^{L}, & \bar{\psi}_{1}^{L}
\end{array}
\right) ,\;\;\;\;\;\mathbb{M=}\left(
\begin{array}{cc}
m_{0} & m_{1} \\
0 & M_{1}
\end{array}
\right) ,\;\;\;\;\;\mathbb{\Psi }_{R}^{\prime }=\left(
\begin{array}{c}
\psi _{0}^{R} \\
\psi _{1}^{R}
\end{array}
\right) \label{two-state}$$ with $m_{0}\ll m_{1}\lesssim M_{1.}$ The mass matrix can be diagonalized in terms of the mass eigenstates $\left(
\begin{array}{cc}
\nu , & N
\end{array}
\right) $ by unitary transformation matrices $\mathbb{U}\left( \theta
_{L}\right) $ and $\mathbb{V}\left( \theta _{R}\right) $ acting on the left- and right-handed fields, respectively: $$\begin{aligned}
\mathbb{U}\left(
\begin{array}{c}
\nu \\
N
\end{array}
\right) _{L} &=&\left(
\begin{array}{c}
\nu _{l}^{L} \\
\psi _{1}^{L}
\end{array}
\right) \simeq \left(
\begin{array}{c}
\;\,\nu _{L}+\theta _{L}N_{L} \\
-\theta _{L}\nu _{L}+N_{L}
\end{array}
\right) \notag \\
\mathbb{V}\left(
\begin{array}{c}
\nu \\
N
\end{array}
\right) _{R} &=&\left(
\begin{array}{c}
\psi _{0}^{R} \\
\psi _{1}^{R}
\end{array}
\right) \simeq \left(
\begin{array}{c}
\;\,\nu _{R}+\theta _{R}N_{R} \\
-\theta _{R}\nu _{R}+N_{R}
\end{array}
\right) \label{LRtwostate}\end{aligned}$$ so that $$\mathbb{UMV}^{\dagger }=\mathbb{M}_{diag}=\left(
\begin{array}{cc}
m_{\nu } & 0 \\
0 & m_{N}
\end{array}
\right)$$ with $m_{\nu }\simeq m_{0}$ being very small and $m_{N}\simeq M_{1}$ very large. The mixing angle for the left-handed field $\theta _{L}$ should be fairly small, while $\theta _{R}$ for the right-handed fields is even more suppressed: $$\theta _{L}\simeq \frac{m_{1}}{M_{1}}\lesssim 1\;\;\;\;\text{and}\;\;\;\;\theta _{R}\simeq \frac{m_{0}m_{1}}{M_{1}^{2}}\lesssim O\left(
\epsilon \right) . \label{mix-angles}$$
Next we will examine in some detail how such mixing angles will figure in the constraint that the electroweak loop effects, such as $\mu \rightarrow
e\gamma ,$ will place on the new physics. Obviously for this purpose, we must have at least two distinctive lepton flavors: $\nu _{l}=\nu _{e}$, $\nu
_{\mu }.$ Thus the $m_{0}$ factor in (\[two-state\]) is now a $2\times 2$ non-diagonal mass matrix, whose elements are of same order magnitude as before. The gauge and mass eigenstates in (\[two-state\]) and (\[LRtwostate\]) must be expanded minimally to sets of three states: $$\mathbb{U}\left(
\begin{array}{c}
\nu _{1} \\
\nu _{2} \\
\nu _{3}
\end{array}
\right) _{L}=\left(
\begin{array}{c}
\nu _{e}^{L} \\
\nu _{\mu }^{L} \\
\psi _{1}^{L}
\end{array}
\right) .$$ The mass eigenstates $\left\{ \nu _{i}\right\} $ correspond to two light neutrinos with masses $m_{\nu 1}$ and $m_{\nu 2}$, on the order of zero-mode Yukawa mass $m_{0},$ and one heavy neutrino with $m_{\nu 3}\simeq M_{1}.$ (We have changed the label for the heavy neutrino from $N$ to $\nu _{3}.$) For simplicity, we shall assume that the unitary matrix $\mathbb{U}$ can be parametrized by two mixing angles: one being the rotation angle $\omega $ in the $\left( 1,2\right) $-plane, and the other being $\theta _{L},$ the $\left( 2,3\right) $ light-heavy rotation angle. $$\mathbb{U}_{li}=\left(
\begin{array}{ccc}
\cos \omega & -\cos \theta _{L}\sin \omega & \;\sin \theta _{L}\sin \omega
\\
\sin \omega & \;\cos \theta _{L}\cos \omega & -\sin \theta _{L}\cos \omega
\\
0 & \sin \theta _{L} & \cos \theta _{L}
\end{array}
\right) \label{mixings}$$ We now proceed to the discussion of the $\mu e\gamma $ loop effects.
$\protect\mu \rightarrow e\protect\gamma $ : heavy particle in the gauge boson loop
===================================================================================
The decay amplitude for the $\mu \left( p\right) \rightarrow e\left(
p-q\right) +\gamma \left( q\right) $ can be written as $$T\left( \mu e\gamma \right) =\frac{ie}{16\pi }\varepsilon _{\lambda }^{\ast
}\left( q\right) \bar{u}_{e}\left( p-q\right) \sigma ^{\lambda \rho }q_{\rho
}\left[ A_{+}\left( 1+\gamma _{5}\right) +A_{-}\left( 1-\gamma _{5}\right) \right] u_{\mu }\left( p\right) .$$ The branching ratio (with $m_{e}=0$) can then be expressed in terms of the invariant amplitudes $A_{\pm }$ as $$B\left( \mu e\gamma \right) =\frac{6e^{2}M_{W}^{4}}{g^{4}m_{\mu }^{2}}\left(
\left| A_{+}\right| ^{2}+\left| A_{-}\right| ^{2}\right) ,$$ where $g$ is the weak gauge coupling, $M_{W}$ the weak gauge boson mass.
First we discuss the invariant amplitudes $A_{\pm }^{W}$ coming from the gauge boson loop contribution $\mu ^{-}\rightarrow \left( \nu _{i}W_{\gamma
}^{-}\right) \rightarrow e^{-}$ where the photon is emitted by the charged $W $ boson in the loop (as denoted by the subscript $\gamma $). The gauge boson coupling to the charged lepton and massive neutrinos is $$\mathcal{L}\left( Wl\nu _{i}\right) =\frac{g}{\sqrt{2}}\mathbb{U}_{li}\bar{l}\gamma ^{\alpha }{\frac12}\left( 1-\gamma _{5}\right) \nu _{i}W_{\alpha }^{-}+h.c.,$$ leading (after a detailed calculation[@CL-ps00]) to the amplitudes of $A_{-}^{W}=0$ and $$A_{+}^{W}=\frac{g^{2}m_{\mu }}{8\pi M_{W}^{2}}\sum_{i=1}^{3}\mathbb{U}_{\mu
i}^{\ast }\mathbb{U}_{ei}F\left( \frac{m_{i}^{2}}{M_{W}^{2}}\right) ,
\label{W-amplitude}$$ where the function $$F\left( z\right) =\frac{1}{6\left( 1-z\right) ^{4}}\left(
10-43z+78z^{2}-49z^{3}-18z^{3}\ln z+4z^{4}\right)$$ has limits of $F\left( 0\right) =5/3$ and $F\left( \infty \right) =2/3,$ respectively$.$ In our case we have two light neutrinos $\nu _{1,2}$ (thus $z_{1,2}\simeq 0$) and one heavy one $z_{3}\gg 1,$ resulting in a branching ratio of $$\begin{aligned}
B\left( \mu e\gamma \right) _{W} &=&\frac{3\alpha }{8\pi }\left|
\sum_{i=1}^{3}\mathbb{U}_{\mu i}^{\ast }\mathbb{U}_{ei}F\left( \frac{m_{i}^{2}}{M_{W}^{2}}\right) \right| ^{2} \notag \\
&=&\frac{3\alpha }{8\pi }\left| \mathbb{U}_{\mu 3}^{\ast }\mathbb{U}_{e3}\left[ F\left( \infty \right) -F\left( 0\right) \right] \right| ^{2}=\frac{3\alpha }{8\pi }\zeta ^{2}\theta _{L}^{4}\end{aligned}$$ where $\zeta ={\frac12}\sin 2\omega ,$ as seen in Eq.(\[mixings\])$.$ We have used the orthogonality condition of the mixing matrix $\mathbb{U}$ when going to the second line. Allowing for a large $\mu e$ mixing angle $\omega ,$ the experimental limit[@muegammalimit] of $B\left( \mu e\gamma \right)
\simeq 10^{-11}$ requires a heavy-light angle $\theta _{L}\simeq
m_{1}/M_{1}=O\left( 10^{-2}\right) ,$ which is small, but still plausible as we expect the Yukawa masses $m_{1}$ to be quite bit less than the Dirac (bare) mass $M_{1}$. Thus the measured value begins to give meaningful constraint on the model parameters. Our point is that the significant restriction is on the mixing angle, rather than on the KK masses directly. Note that if we had not taken into account the suppression due to the mixing angles we would get an unacceptable large $B\left( \mu \rightarrow e\gamma
\right) .$ Also in this case the large mass comes from bare mass and not from the large Yukawa coupling and we expect the decoupling to be valid[@CL91]. Indeed $B\left( \mu \rightarrow e\gamma \right) $ vanishes in the limit $M_{1}\rightarrow \infty ,$ after the behavior of the mixing angles is included.
$\protect\mu \rightarrow e\protect\gamma $ : heavy particle in the scalar boson loop
====================================================================================
In the minimal SM with massive neutrinos, the leading $\mu e\gamma $ amplitude comes from the gauge boson loop as discussed in the previous Section. However, for models having more scalars beyond the one Higgs doublet, there could in principle be significant scalar boson loop contribution as well. Even for the minimal SM, it is instructive to consider the scalar boson case separately because the longitudinal gauge boson is simply the (unphysical) Higgs scalar boson. Being proportional to the fermion mass, such Yukawa coupling is the source of the decoupling violation — through the cancellation of the large mass in fermion propagator by the large Yukawa couplings.
The Yukawa interactions of the scalar boson $\phi $ to a charged lepton *l* and a massive neutrino $\nu _{i}$ can be parametrized by the chiral couplings $y_{li}^{\left( \pm \right) }$: $$\mathcal{L}\left( \phi l\nu _{i}\right) =\bar{l}\left[ y_{li}^{\left(
+\right) }\left( 1+\gamma _{5}\right) +y_{li}^{\left( -\right) }\left(
1-\gamma _{5}\right) \right] \nu _{i}\phi _{{}}^{-}+h.c.$$ We have performed a detailed calculation of the scalar boson loop amplitude, $\mu ^{-}\rightarrow \left( \nu _{i}\phi _{\gamma }^{-}\right) \rightarrow
e^{-}$, where the photon is emitted by the charged $\phi $ boson in the intermediate state, and found that, for heavy neutrino intermediate state $\left( m_{i}\gg M_{\phi }\right) ,$ $$A_{+}^{\phi }\left( \nu _{\text{heavy}}\right) =\frac{1}{\pi m_{i}^{2}}\left( \frac{m_{\mu }}{3}y_{ei}^{\left( +\right) \ast }y_{\mu i}^{\left(
-\right) }+m_{i}y_{ei}^{\left( +\right) \ast }y_{\mu i}^{\left( +\right)
}\right) , \label{phi-heavy}$$ and, for light neutrino intermediates state $\left( m_{i}\ll M_{\phi
}\right) $, $$A_{+}^{\phi }\left( \nu _{\text{light}}\right) =\frac{1}{\pi M_{\phi }^{2}}\left( \frac{m_{\mu }}{6}y_{ei}^{\left( +\right) \ast }y_{\mu i}^{\left(
-\right) }+m_{i}y_{ei}^{\left( +\right) \ast }y_{\mu i}^{\left( +\right)
}\right) . \label{phi-light}$$ The other chiral amplitudes $A_{-}^{\phi }$ have similar structures.
Let us first consider the SM case when the scalar is the would-be-Goldstone boson, and becomes the longitudinal gauge boson after spontaneous symmetry breaking. In the renormalizable $R_{\xi }$ gauge, there is a scalar particle with a mass $M_{\phi }=\xi M_{W}.$ The SM fermions obtain their masses through their couplings to the Higgs field, hence the Yukawa couplings are proportional to the fermion masses. If the VEV is written in terms of $g$ and $M_{W}$, we have the explicit form of $$\begin{aligned}
y_{ei}^{\left( +\right) \ast } &=&\frac{gm_{i}}{2\sqrt{2}M_{W}}U_{ei}^{\ast
},\;\;\;\;\;\;\;\;\;y_{ei}^{\left( -\right) \ast }=\frac{-gm_{e}}{2\sqrt{2}M_{W}}U_{ei}^{\ast }, \notag \\
y_{\mu i}^{\left( +\right) } &=&\frac{gm_{\mu }}{2\sqrt{2}M_{W}}U_{\mu
i},\;\;\;\;\;\;\;\;\;y_{\mu i}^{\left( -\right) }=\frac{-gm_{i}}{2\sqrt{2}M_{W}}U_{\mu i}, \label{SMyukawa}\end{aligned}$$ Substituting these relations into Eq.(\[phi-heavy\]), we can check that heavy neutrino $\left( \nu _{3}\right) $ contribution is given by $$A_{+}^{\phi }\left( \nu _{\text{heavy}}\right) =\frac{g^{2}m_{\mu }}{8\pi
M_{W}^{2}}\mathbb{U}_{\mu 3}^{\ast }\mathbb{U}_{e3}\left( \frac{2}{3}\right)$$ in agreement with the result in Eq.(\[W-amplitude\]) with $F\left( \infty
\right) =2/3.$ This shows that the heavy particle non-decoupling contribution to the $\mu e\gamma $ amplitude comes entirely from the Higgs boson loop[@CL91].
For models with non-minimal Higgs structure, we have physical scalar particles with couplings that do not have a simple fermion mass dependence — in fact they are as a rule highly model-dependent. The result of (\[phi-heavy\]) can then be translated, with $m_{\mu }/m_{i}\simeq 0,$ into the branching ratio of $$B\left( \mu e\gamma \right) _{\phi \nu _{i}}=\frac{24\alpha }{\pi g^{4}}\left( \frac{M_{W}^{4}}{m_{\mu }^{2}m_{i}^{2}}\right) \left( y_{ei}^{\left(
+\right) \ast }y_{\mu i}^{\left( +\right) }\right) ^{2}. \label{scalarBR}$$ Clearly a naive assumption of $\left( y_{ei}^{\left( +\right) \ast }y_{\mu
i}^{\left( +\right) }\right) =O\left( 1\right) $ would lead to a meaninglessly weak bound on the heavy neutrino mass[^2] of $m_{i}>10^{7}TeV.\;$Since the generalized Yukawa couplings $y_{li}^{\left( \pm \right) }$ do include small mixing angles, it seems more sensible to use the experimental result[@muegammalimit] of $B\left( \mu e\gamma \right) \lesssim
1.2\times 10^{-11}$ to set a limit on the coupling and mixing angle combination: $$\left( y_{ei}^{\left( +\right) \ast }y_{\mu i}^{\left( +\right) }\right)
\lesssim \frac{10^{-7}}{m_{i}\left( TeV\right) },$$ where we have used the value of Fermi constant $G_{F}=\sqrt{2}g^{2}/8M_{W}^{2}\simeq 10^{-5}/M_{N}^{2}.$
Discussion
==========
We have focused on the Randall-Sundrum version of the extra-dimensional theory. However, our discussion is equally applicable to the original version where the suppression of the bulk field effects (gravitons, singlet-neutrinos, etc.) comes through the large volume of the extra dimensional space. This is the case because the structure of the neutrino mass matrix is very similar in both versions of the theory[@nuLargeVol].
In this paper, we have concentrated on a single bulk neutrino. In principle, there is a whole tower of Kaluza-Klein states. Many authors[@towerSUM] have attempted to sum over the contribution by the whole tower. We have not done so because we do not wish to confuse the issue of decoupling of a single heavy particle with the separate problem of how the sum of this infinite tower should behave. The individual heavy particle contribution is controlled by the heavy-light mixing angle, which is mass-suppressed. If one sums over an infinite number of such small terms, a “non-decoupling” result can be obtained. Clearly, this approach touches upon the difficult issue of convergence of the KK sum, with implications related to the possible presence of new physics at higher scales. Such problems are quite different from the matter of single particle decoupling, which is the focus of this paper.
One of us (T.P.C.) would like to thank Gary Shiu for helpful discussion. L.F.L. acknowledges the support from U.S. Department of Energy (Grant No. DE-FG02-91ER40682).
[99]{} N. Arkani-Hamed, S. Dimopoulos, and G. Dvali *Phys.Lett.* **B 429** (1998) 263 \[hep-ph/9803315\]; I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, and G. Dvali *Phys. Lett.* **B 436** (1998) 257 \[hep-ph/9804398\].
L. Randall and R. Sundrum, *Phys. Rev. Lett.* **83** (1999) 3370 \[hep-ph/9905221\].
See, for example, H. Davoudiasl, J.L. Hewett, and T.G. Rizzo, \[hep-ph/0006041\]: C. Csaki, M.L. Graesser, and G.D. Kribs, \[hep-th/0008151\].
T. Appelquist and J. Carazzone, *Phys. Rev.* **D 11** (1975) 2856.
T.-P. Cheng and L.-F. Li, *Phys. Rev.* **D 44** (1991) 1502; L.-F. Li and T.-P. Cheng, in *Proc.of the Vancouver Meeting*, eds. D. Axen *et al*. (World Scientific, 1992) 801.
T. Yanagida, *Prog. Theor. Phys.* **65** (1978) 66; M. Gell-Mann, P. Ramond, and R. Slansky, in *Supergravity,* eds. P. van Nieuwenhuizen and D. Freedman (North Holland, 1979).
T.-P. Cheng and L.-F. Li, *Gauge Theory of Elementary Particle Physics - Problems and Solutions* (Oxford, 2000), p244 - 250.
Y. Grossman and M. Neubert, *Phys. Lett.* **B 474** (2000) 361 \[hep-ph/9912408\].
R. Kitano, *Phys. Lett.* **B 481** (2000) 39 \[hep-ph/0002279\].
M.L. Brooks *et al.,* \[MEGA Collaboration\], *Phys. Rev. Lett.* **83** (1999) 1521 \[hep-ex/9905013\].
K.R. Dienes, E. Dudas, and T. Gherghetta, *Nucl. Phys* **B 557** (1999) 25 \[hep-ph/9811428\]; N. Arkani-Hamed, S. Dimopoulos, G. Dvali, and J. March-Russell, \[hep-ph/9811448\]; G. Dvali and A.Y. Smirnov, *Nucl. Phys* **B 563** (1999) 63 \[hep-ph/9904211\]; R.N. Mohapatra, S. Nandi, and A. Perez-Lorenzana, *Phys. Lett.* **B 466** (1999) 115 \[hep-ph/9907520\]; K. Agashe and G.H. Wu, \[hep-ph/0010117\]; A. Lukas, P. Ramond, A. Romanino, and G. Ross, \[hep-ph/0011295\].
A. Ioannisian and A. Pilaftsis, *Phys. Rev.* **D 62** (2000) 066001 \[hep-ph/9907522\]; A.E. Faraggi and M. Pospelov, *Phys. Lett.* **B 458** (1999) 237 \[hep-ph/9901299\]; G.C. McLaughlin and J.N. Ng, \[hep-ph/0008209\].
[^1]: Cancellation of parity anomaly requires that there be even number of bulk fermions. Since the presence of multiple bulk fermions should not introduce qualitative changes in our result, we shall ignore such complication and stick with one bulk fermion.
[^2]: In this respect, we differ from the conclusion drawn in Ref. [@kitano] where the scalar contribution to the branching ratio has been estimated to have the mass dependence of $\left( M_{W}/m_{i}\right) ^{4}$, as compared to our result of Eq.(\[scalarBR\]). The author was also silent with regard to the implication of the apparent decoupling violation by the W-loop contribution, as stated in Eq.(\[W-amplitude\]).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We consider the multi-armed bandit setting with a twist. Rather than having just one decision maker deciding which arm to pull in each round, we have $n$ different decision makers (agents). In the simple stochastic setting, we show that a “free-riding” agent observing another “self-reliant” agent can achieve just $O(1)$ regret, as opposed to the regret lower bound of $\Omega (\log t)$ when one decision maker is playing in isolation. This result holds whenever the self-reliant agent’s strategy satisfies either one of two assumptions: (1) each arm is pulled at least $\gamma \ln t$ times in expectation for a constant $\gamma$ that we compute, or (2) the self-reliant agent achieves $o(t)$ realized regret with high probability. Both of these assumptions are satisfied by standard zero-regret algorithms. Under the second assumption, we further show that the free rider only needs to observe the number of times each arm is pulled by the self-reliant agent, and not the rewards realized.
In the linear contextual setting, each arm has a distribution over parameter vectors, each agent has a context vector, and the reward realized when an agent pulls an arm is the inner product of that agent’s context vector with a parameter vector sampled from the pulled arm’s distribution. We show that the free rider can achieve $O(1)$ regret in this setting whenever the free rider’s context is a small (in $L_2$-norm) linear combination of other agents’ contexts and all other agents pull each arm $\Omega (\log t)$ times with high probability. Again, this condition on the self-reliant players is satisfied by standard zero-regret algorithms like UCB. We also prove a number of lower bounds.
author:
- Christopher Jung
- Sampath Kannan
- |
Neil Lutz\
\
Department of Computer and Information Science\
University of Pennsylvania, Philadelphia, PA 19104
bibliography:
- 'freeridebibliography.bib'
title: |
Quantifying the Burden of Exploration\
and the Unfairness of Free Riding
---
Introduction
============
We consider situations where exploitation must be balanced with exploration in order to obtain optimal performance. Typically there is a single decision maker who does this balancing, in order to minimize a quantity called the regret. In this paper we consider settings where there are many agents and ask how a single agent (the *free rider*) can benefit from the exploration of other *self-reliant* agents. For example, competing pharmaceutical companies might be engaged in research for drug discovery. If one of these companies had access to the research findings of its competitors, it might greatly reduce its own exploration cost. Of course, this is an unlikely scenario since intellectual property is jealously guarded by companies, which points to an important consideration in modeling such scenarios: the amount and type of information that one agent is able to gather about the findings of others.
More realistically, and less consequentially, a recommendation system such as Yelp makes user ratings of restaurants publicly available. The assumption underlying such systems is that “the crowd” will explore available options so that we end up with accurate average ratings. Many problems of this sort can be modeled using the formalism of multi-armed bandits. Free riding also arises in online advertising. Each advertiser may be modeled as an agent with a context vector describing its likely customers, and it must choose online niches in which to advertise. A free rider can take advantage of competitors’ exploration of niches by monitoring impressions and clickthroughs of their ads. In fact, there are a number of paid services (WhatRunsWhere, Adbeat, SpyFu, etc.) that facilitate this behavior.
Multi-armed bandit problems model decision making under uncertainty [@Robbins52; @LaiRobbins85; @Bubeck12]. Our focus in this paper will be on the *stochastic bandits model* where there is an unknown reward distribution associated with each arm, and the decision maker has to decide which arm to pull in each round. Her goal is to minimize *regret*, the (expected) difference between the reward of the best arm and the total reward she obtains. In the extension to the *linear contextual bandits model*, each arm $i$ has an unknown parameter vector $\theta_i \in \mathbb{R}^d$ for $i = 1,\ldots,k$, where $k$ is the total number of arms. At round $t$, a *context* $x^t \in \mathbb{R}^d$ arrives. The expected reward for pulling arm $i$ in round $t$ is the inner product $\langle \theta_i , x^t \rangle$.
In the simple stochastic case, there are two types of relevant information: the other agents’ actions and the resulting rewards. In the full-information setting, the free rider has access to both types of information. We also consider a partial-information setting where the free rider can only observe the other agents’ actions. For linear contextual bandits, the full-information setting also includes the context vectors of the other agents.
In our setting, using Yelp as the running example, the $k$ arms correspond to restaurants. Our model differs from standard bandit models in three significant ways. First, there are $n$ decision makers rather than one; in the Yelp example, each decision maker corresponds to a diner. Upon visiting a restaurant, a diner samples from a distribution to determine her dining experience. In the stochastic setting, we assume that all diners have identical criteria for assessing their experiences, meaning that identical samples lead to identical rewards. Second, in the linear contextual setting, the contexts in our model are fixed in time and can be regarded as the *types* of the individual decision makers. Each diner’s context vector represents the weight she assigns to various features (parameters) of a restaurant, such as innovativeness, decor, noise level, suitability for vegetarians, etc. Third, each arm has a distribution over parameter vectors instead of a fixed parameter vector. When a diner visits a restaurant, her reward is determined by taking the inner product of her context with a parameter vector drawn from the restaurant’s distribution, rather than by adding sub-Gaussian noise to the inner product with a fixed parameter vector as in the standard model.
In the standard stochastic or linear contextual bandit setting, a decision-making algorithm is called *zero-regret* if its regret over $t$ rounds is $o(t)$. It is well known that exploration is essential for achieving zero regret [@LaiRobbins85]. One algorithm that achieves the asymptotically optimal regret bound of $O(\log t)$ over $t$ rounds is the so-called Upper Confidence Bound (UCB) algorithm of Lai and Robbins [@LaiRobbins85]. In addition to maintaining a sample mean for each arm, this algorithm maintains confidence intervals around these means, where the width of the confidence interval for arm $i$ drops roughly as $1/\sqrt{n_i}$ where $n_i$ is the number of times arm $i$ has been pulled so far. The UCB algorithm then selects the arm with the highest upper limit to its confidence interval. There are many other zero-regret strategies, such as Thompson sampling [@thompson1933likelihood] or one where an initial round-robin exploration phase is followed by an exploitation phase in which the apparently optimal arm is pulled [@GarivierMS16].
#### Our results:
- In the stochastic setting a free rider can achieve $O(1)$ regret under either of two reasonable assumptions, both of which are satisfied by standard zero-regret algorithms:
- Some self-reliant agent has pulled each arm at least $\gamma \ln t$ times in expectation at all sufficiently large times $t$, where $\gamma$ is a constant derived from our analysis (Theorem \[thm:sampleaugmenting\]).
- Some self-reliant agent is playing a strategy that with high probability achieves $o(t)$ realized regret. In this case, the free rider can achieve $O(1)$ regret even in the partial-information setting (Theorem \[thm:stoch\]). As a corollary, a free rider can achieve $O(1)$ regret whenever a self-reliant agent plays UCB (Corollary \[cor:ucb\]).
- For linear contextual bandits, a free rider can again achieve $O(1)$ regret in the full-information setting under an assumption similar to the first assumption above (Theorem \[thm:fullinfo\]).
- As a way of relating the two assumptions in the first bullet above, we prove that if a self-reliant agent achieves $O(t^{1 - \epsilon })$ regret, then that agent must pull *each* arm $\Omega(\log t)$ times in expectation (Theorem \[thm:expectedcount\]) and with a high probability that depends on $\epsilon$ (Theorem \[thm:highprobcount\]).
- There is a deterministic lower bound of $\Omega (\log t)$ on the number of times a UCB agent must pull each arm in the stochastic case (Theorem \[thm:ucbcount\]).
- To achieve $o(\log t)$ regret in the contextual setting, the free rider must know both the contexts and the observed rewards of the other agents (Theorems \[thm:needcontexts\] and \[thm:needrewards\]).
#### Related work:
This paper asks how and when an agent may avoid doing their “fair share” of exploration. Several recent works have studied how the cost of exploration in multi-armed bandit problems is distributed, from the perspective of algorithmic fairness. Works by Bastani, Bayati, and Khosravi [@bastani2017mostly]; Kannan, Morgenstern, Roth, Waggoner, and Wu [@kannan2018smoothed]; and Raghavan, Slivkins, Vaughan, and Wu [@RaghavanSVW18] show that if the data is sufficiently diverse, e.g., if the contexts are randomly perturbed, then exploration may not be necessary. Celis and Salehi [@celis2017lean] consider a model in both the stochastic and the adversarial setting where each agent in the network plays a certain zero-regret algorithm (UCB in the stochastic setting and EXP3 in the adversarial setting) and study how much information an agent can gather from his neighbors.
There is some discussion in the economics literature of free riding in bandit settings. In the model of Bolton and Harris [@bolton1999strategic], agents choose what fraction of each time unit to devote to a safe action (exploitation) and to a risky action (exploration), and they show that while the attraction of free riding drives agents to select the safe action always, risky action by a agent may enable everyone to converge to the correct posterior belief faster. Keller, Rady, and Cripps [@keller2005strategic] consider a very similar setting where a risky arm will generates positive payoff after an exponentially distributed random time; they characterize unique symmetric equilibrium as well as various asymmetric equilibria. Klein [@klein2013strategic] gives conditions for complete learning in a two-agent, three-armed bandit setting where there are two negatively-correlated risky arms and a safe arm, with further assumptions about their behavior. It is clear that these models do not support having more than two arms (or three in the case of [@klein2013strategic]) and that their goal is maximizing expected reward, not minimizing regret. Moreover, one arm is explicitly designated as the safe arm and the other(s) as risky, *a priori*.
Preliminaries {#sec:prelims}
=============
Stochastic Model {#stochastic-model .unnumbered}
----------------
There are $k$ *arms*, indexed by $[k]=\{1,\ldots,k\}$ and $n$ *players* or *agents*, indexed by $[n]$. Arm $i$ has a *reward distribution* $D_i$ supported on $[-1,1]$ with mean $\mu_i$, and $\mathbf{D}=(D_1,\ldots,D_k)$ is the *reward distribution profile*, or the *stochastic bandit*. The arm with the highest mean reward is denoted by $i^*=\operatorname*{arg\,max}_{i\in[k]}\mu_i$, and we write $\mu^*$ for $\mu_{i^*}$; we assume that $i^*$ is unique. An important parameter is $\Delta=\mu^*-\max_{i\in[k]\setminus\{i^*\}}\mu_{i}$, the *gap* between optimal and suboptimal arms.
In round $t=1,2,\ldots$, each player $p$ selects an arm $i_p^t\in[k]$ and receives a reward $r_p^t\sim D_{i_p^t}$. We write $H^T= ((i^t_p, r^t_p)_{t\in[T]})_{p \in [n]}$ to denote the *history* of all players’ actions and rewards through round $T$. A *policy* or *strategy* for a player $p$ is a function $f_p$ mapping each history to an arm or to a distribution over the arms; a player $p$ with policy $f_p$ who observes history $H^{T}$ will pull arm $f_p(H^{T})$ in round $T+1$. A *policy profile* is a vector $\mathbf{f}=(f_1,\ldots,f_n)$, where each $f_p$ is a policy for player $p$. Notice that a policy profile and a stochastic bandit together determine a distribution on histories. A policy $f_p$ for player $p$ is *self-reliant* if it depends only on $p$’s own observed actions and rewards. In contrast, a *free-riding* policy may use all players’ history.
The *regret* of player $p$ at time $T$ under stochastic bandit $\mathbf{D}$ is $R_p^T(\mathbf{D},\mathbf{f})=T\mu^*-\sum_{t\in[T]}{\mathbb{E}}[r_p^t]$, where the expectation is according to the distribution on histories determined by $\mathbf{D}$ and $\mathbf{f}$.[^1] When it will not introduce ambiguity, we simply write $R^T_p$ or, in single-player settings, $R^T$. We also consider the *realized regret* under a particular history $H^T$, $\hat{R}_p^T(\mathbf{D},\mathbf{f})=T\mu^*-\sum_{t\in[T]} r_p^t$.
For any player $p$, arm $i$, and time $t$, the *sample count* is $N_{p,i}^{t}$, the number of times $i$ has been pulled by the player in the first $t$ rounds, and the *sample mean* is $\mu_{p,i}^{t}$, the average of all of player $p$’s samples of arm $i$ through time $t$.
One well-studied self-reliant policy that achieves logarithmic regret in the stochastic setting is called *$\alpha$-UCB* [@LaiRobbins85], defined by $$\alpha\text{-UCB}(H^t)=\operatorname*{arg\,max}_{i\in[k]}\mu^t_i + \sqrt{\frac{\alpha \ln(t+1)}{2N^{t}_i}}$$ for all histories $H^t$. The parameter $\alpha$ calibrates the balance between exploration and exploitation. For each arm $i$, a player using this policy maintains an *upper confidence bound* on $\mu_i$, and in each round, she pulls the arm with the highest upper confidence bound. The distance from each arm’s sample mean to its upper confidence bound depends its sample count.
Linear Contextual Model {#linear-contextual-model .unnumbered}
-----------------------
The linear contextual model generalizes the stochastic model. Now, each arm $i$ has a *feature distribution* $F_i$ supported on the $d$-dimensional closed unit ball, for some $d\in{\mathbb{N}}$, and $\mathbf{F}=(F_1,\ldots,F_n)$ is the *feature distribution profile* or *contextual bandit*. Each player $p$ has a context $x_p\in{\mathbb{R}}^d$, and $\mathbf{x}=(x_1,\ldots,x_n)$ is the *context profile*. As before, in each round $t$, each player $p$ selects an arm $i_p^t$, but now the reward is given by sampling a feature vector $\theta_p^t\sim F_{i_p^t}$, and taking its inner product with $x_p$, i.e., $r_p^t=\langle \theta_p^t,x_p\rangle$. $D_{p,i}$ is the distribution of rewards from arm $i$ for player $p$, and the mean of this distribution is $\mu_{p,i}={\mathbb{E}}_{\theta_i\sim F_i}[\langle \theta_i, x_p\rangle].$ The optimal arm for player $p$ is $i^*_p=\operatorname*{arg\,max}_{i\in k}\mu_{p,i}$.
Histories, policies, policy profiles, self-reliance, and free riding are defined exactly as in the stochastic setting. The regret of player $p$ through round $T\in{\mathbb{N}}$ under contextual bandit $\mathbf{F}$, context profile $\mathbf{x}$, and policy profile $\mathbf{f}$ is given by $R_p^T(\mathbf{F},\mathbf{x},\mathbf{f})=T\mu_p^*-\sum_{t\in[T]}{\mathbb{E}}[r_p^t]$, where the expectation is taken according to the distribution determined by $\mathbf{F}$, $\mathbf{x}$, and $\mathbf{f}$. Notice that for a self-reliant player $p$ with context $x_p$, the contextual bandit $\mathbf{F}=(F_1,\ldots,F_n)$ is equivalent to the stochastic bandit $\mathbf{D}=(D_{p,1},\ldots,D_{p,n})$.
Technical Lemmas from Other Sources {#technical-lemmas-from-other-sources .unnumbered}
-----------------------------------
Some of our lower bounding arguments will use the following technical lemmas. $\operatorname*{KL}$ denotes the Kullback–Leibler divergence.
\[lem:dd\] Let ${\mathbf{D}}= (D_1,\ldots, D_k)$ and ${\mathbf{D}}'=(D'_1,\ldots,D'_k)$ be stochastic bandits. For any policy $f$ and time $t$, $$\operatorname*{KL}(H^t({\mathbf{D}},f),H^t({\mathbf{D}}',f))=\sum_{i\in[k]}{\mathbb{E}}[N_i^t({\mathbf{D}},f)]\operatorname*{KL}(D_i,D'_i)\,.$$
\[lem:bh\] Let $P$ and $Q$ be probability measures on the same measurable space, and let $A$ be any event. Then, $$P(A) + Q(A^c)\geq\frac12 \exp(-\operatorname*{KL}(P,Q))\,,$$ where $A^c$ is the complement of $A$.
Lower Bounds on Sample Counts {#sec:lowerbound}
=============================
If a self-reliant player has sampled an arm sufficiently many times, then a free rider with full information can use those samples to find a good estimate of that arm’s mean. In this section, we give three lower bounds on the sample counts of each arm.
Theorem \[thm:expectedcount\] shows that if a policy guarantees $O(T^{1-\epsilon})$ regret for some positive $\epsilon$, then every arm must be sampled $\Omega(\log T)$ times in expectation. We prove this using the method of Bubeck, Perchet, and Rigollet [@BuPeRi13], showing via the Bretagnolle-Huber inequality [@BreHub79] that the learner cannot rule out any arm’s optimality without sampling that arm $\Omega(\log T)$ times.
\[thm:expectedcount\] Let $f$ be any self-reliant policy such that $R^T=O(T^{1-\epsilon})$ for all stochastic bandits and some $\epsilon>0$. Then for all stochastic bandits with $\mu^*<1$ and all $i\in[k]$, $f$ satisfies ${\mathbb{E}}[N_i^T]=\Omega(\log T)$.
Let $k\geq 1$ and $\epsilon>0$, and let $f$ be any self-reliant policy satisfying $R^T(\mathbf{D},f)=O(T^{1-\epsilon})$ for all $k$-arm stochastic bandits $\mathbf{D}$. We prove that for all $k$-arm stochastic bandits $\mathbf{D}$ with $\mu^*<1$ and all $i\in[k]$, we have ${\mathbb{E}}[N_i^T(\mathbf{D},f)]=\Omega(\ln T)$.
Fix an arm $i$, and let $X={\mathbb{E}}[N_i^T(\mathbf{D},f)]$. If $i$ is optimal, then $X\geq T-R^T(\mathbf{D},f)/\Delta$ and the theorem holds. Hence, let $i$ be any suboptimal arm, let $\delta=\min\left\{\Delta,\frac{1-\mu^*}{2}\right\}$, let $p=\frac{1-\mu^*-\delta}{1-\mu_i}$, and let ${\mathbf{D}}' = (D_1, \dots, D'_i,\dots,D_k)$, where for all $x\in[-1,1]$, $$D'_i(x) = p \cdot D_i(x) + 1-p\,.$$ Notice that the mean of $D'_i$ is $\mu_i'=\mu^*+\delta$ and that $\operatorname*{KL}(D_i,D'_i)\leq\ln(1/p)$. Now, $$\begin{aligned}
\max\{R^T(\mathbf{D},f),R^T(\mathbf{D}',f)\}&\geq \frac{1}{2}(R^T(\mathbf{D},f)+R^T(\mathbf{D}',f))\\
&\geq\frac{\delta}{2}\sum_{t=1}^T\left(\Pr[i_t(\mathbf{D},f)= i]+\Pr[i_t(\mathbf{D}',f)\neq i]\right)\\
&\geq \frac{\delta}{4}\sum_{t=1}^T\exp(-{\operatorname*{KL}}(H^{t-1}(\mathbf{D},f),H^{t-1}(\mathbf{D}',f)))\\
&\geq \frac{\delta T}{4}\exp(-{\operatorname*{KL}}(H^{T-1}(\mathbf{D},f),H^{T-1}(\mathbf{D}',f)))\,,
\end{aligned}$$ where $\operatorname*{KL}$ denotes Kullback-Leibler divergence and the second-to-last line follows from Lemma \[lem:bh\], the Bretagnolle-Huber inequality [@BreHub79]. By Lemma \[lem:dd\] [@BanditBook], $$\begin{aligned}
{\operatorname*{KL}}(H^{t-1}(\mathbf{D},f),H^{t-1}(\mathbf{D}',f)
&=\operatorname*{KL}(D_i,D'_i)\cdot{\mathbb{E}}[N^T_i({\mathbf{D}},f)]\\
&\leq X\ln(1/p)\,.
\end{aligned}$$ Thus, we have $$\max\{R^T(\mathbf{D},f),R^T(\mathbf{D}',f)\}\geq\frac{\delta T}{4}\exp\left(-X\ln(1/p)\right)\,.$$ It follows that $\exp(-X\ln(1/p))=O(T^{-\epsilon})$ and therefore that $X=\Omega(\ln T)$.
In addition to a bound on the expected sample count, we sometimes need stronger guarantees on the tail of the sample count distribution. In Theorem \[thm:highprobcount\], we use a coupling argument to show that if a policy has regret $O(T^{1-\beta})$ for relatively large $\beta$, then the probability that any arm that is sampled too few times is small.
\[thm:highprobcount\] Let $f$ be any self-reliant policy such that $R^T=O(T^{1-\beta})$ for all stochastic bandits and some $\beta>0$. Then for all stochastic bandits with $\mu^*<1$, all $i \in [k]$, and all $\gamma>0$, $f$ satisfies $$\Pr\left(N^T_i \leq \gamma \ln T\right) = O\big(T^{\gamma c_i-\beta}\big)\,,$$ where $ c_i = \ln\left(\frac{1 - \mu^*}{2(1-\mu_i)}\right)$.
Let $\alpha,\beta,\gamma,t_0>0$, and let $f$ be a self-reliant policy such that for all stochastic bandits ${\mathbf{D}}$ and all $T>t_0$, $R^T({\mathbf{D}},f)\leq \alpha T^{1-\beta}$. Let $t_1\geq t_0$ satisfy $\gamma\ln t_1<t_1/2$. Assume for contradiction that there is some stochastic bandit ${\mathbf{D}}$ with $\mu^*<1$, some arm $i\in[k]$, and some $T>t_1$ such that $$\begin{aligned}
\label{eqn:lowerbound_contradiction}
\Pr\left(N^T_i({\mathbf{D}},f) \leq \gamma \ln T\right) > \frac{C\alpha}{T^{\beta+\gamma \ln p_i}}\,,\end{aligned}$$ where $C=\frac{2}{\min\{\Delta,(1-\mu^*)/2\}}$ and $p_i = \frac{1 - \mu^*}{2(1-\mu_i)}$.
Observe that if $i$ is the optimal arm in ${\mathbf{D}}$, then since $\gamma\ln p_i<0$, we have $$\begin{aligned}
R^T({\mathbf{D}},f)&> \Delta\cdot(T-\gamma\ln T)\cdot \frac{C\alpha}{T^{\beta+\gamma\ln p_i}}\\
&\geq \Delta\cdot(T-\gamma\ln T)\cdot C\alpha T^{-\beta}\\
&\geq \alpha T^{1-\beta}\,,\end{aligned}$$ contradicting the assumption that $R^T({\mathbf{D}},f)\leq \alpha T^{1-\beta}$. Hence, we assume that $i$ is suboptimal.
We now construct a stochastic bandit ${\mathbf{D}}'$ in which $i$ is optimal. Let ${\mathbf{D}}' = (D_1, \dots, D'_i,\dots,D_k)$, where $$D'_i(x) = p_i \cdot D_i(x) + 1-p_i$$ for all $x\in[-1,1]$. Notice that the mean of $D'_i$ is $\mu_i'=\frac{1+\mu^*}{2}$, and that the gap between optimal and suboptimal arms in ${\mathbf{D}}'$ is $\Delta'=\mu_i'-\mu^*=\frac{1-\mu^*}{2}$.
We now use a coupling argument to bound $\Pr\left(N_{i}^T({\mathbf{D}}',f) \leq \gamma \ln T\right)$. Observe that to sample from $D'_i$, one can sample a reward $x \sim D_i$, keep $x$ with probability $p_i$, and otherwise output $1$. Thus, for any history $h$ in which $i$ is pulled exactly $s$ times, $$\Pr\left(H^T({\mathbf{D}}',f)=h\right) \ge p_i^s \cdot \Pr\left(H^T({\mathbf{D}},f)=h\right)\,.$$ By summing over all such histories, we have $$\Pr\left(N^T_i({\mathbf{D}}',f)=s\right)\geq p_i^s\cdot \Pr\left(N^T_i({\mathbf{D}},f)=s\right)\,,$$ and therefore $$\begin{aligned}
\Pr\left(N^T_i({\mathbf{D}}',f)\leq\gamma\ln T\right)&=\sum_{s=0}^{\lfloor\gamma \ln T\rfloor}\Pr\left(N^T_i({\mathbf{D}}',f)=s\right)\\
&\geq \sum_{s=0}^{\lfloor\gamma \ln T\rfloor}p_i^s\cdot \Pr\left(N^T_i({\mathbf{D}},f)=s\right)\\
&\geq p_i^{\gamma\ln T}\sum_{s=0}^{\lfloor\gamma \ln T\rfloor}\Pr\left(N^T_i({\mathbf{D}},f)=s\right)\\
&=T^{\gamma\ln p_i}\cdot\Pr\left(N^T_i({\mathbf{D}},f) \leq \gamma \ln T \right)\,.\end{aligned}$$ Combining this bound with inequality (\[eqn:lowerbound\_contradiction\]) yields $$\begin{aligned}
\Pr\left(N_{i}^T({\mathbf{D}}',f) \leq \gamma \ln(T)\right) &> T^{\gamma \ln p_i} \cdot \frac{C\alpha}{T^{\beta+\gamma \ln p_i}}\\
&=C\alpha T^{-\beta}\,.\end{aligned}$$ Thus, $$\begin{aligned}
R^T({\mathbf{D}}',f) &> C\alpha T^{-\beta} \cdot (T-\gamma\ln T) \cdot \frac{1-\mu^*}{2}\\
&\geq\alpha T^{1-\beta}\,,\end{aligned}$$ which again contradicts the assumed regret bound on $f$.
Finally, we use a delicate inductive argument to prove the following deterministic guarantee on the sample count for each arm when the arms are pulled according to the $\alpha$-UCB policy.
\[thm:ucbcount\] Let $\alpha>0$ and $\eta>2$. There exists a constant $t_0$ such that for all stochastic bandits, all $t\geq t_0$, and all $i\in[k]$, an agent playing the $\alpha$-UCB policy must satisfy $N_i^{t-1}\geq \alpha\ln t/(2\eta^2 k^2)$.
For every $j\in[k]$ and $t\in{\mathbb{N}}$, define the set $$U^t_j=\left\{i\in[k]:N_i^{t-1}\geq\frac{\alpha\ln t}{2\eta^2 j^2}\right\}\,.$$ We claim that for all $j\in[k]$ there is a constant $t_j$ such that for all $t\geq t_j$, $|U^t_j|\geq j$.
We will prove this claim by induction on $j$. For any time $t$, there is clearly some arm $i$ with $N_i^{t-1}\geq\frac{t-1}{k}$, and we can choose $t_1$ such that $\frac{t-1}{k}\geq \frac{\alpha\ln t}{2\eta^2 k^2}$ whenever $t\geq t_1$, so the claim holds for $j=1$.
Now fix $j>1$, and assume that the claim holds for $j-1$. Define a function $g_j:{\mathbb{N}}\to{\mathbb{R}}$ by $$g_j(t)=t-(k-j+1)\frac{\alpha\ln t}{2\eta^2 j^2}\,.$$ We choose $t_j$ sufficiently large such that for all $t\geq t_j$ we have $g_j(t)> t_{j-1}$ and $$\label{eq:gjcond}
\frac{\ln(g_j(t)-1)}{\ln t}> \left(1-\frac{1-2/\eta}{j}\right)^2\,.$$
Assume for contradiction that there is some time $t\geq t_j$ such that $|U^t_j|< j$. Since $U^t_{j-1}\subseteq U^t_j$, the inductive hypothesis then implies that $U^t_j=U^t_{j-1}$. Thus, $|U_j^t|=j-1$, and there are exactly $k-j+1$ arms outside of $U^t_j$. Each one of those arms has been pulled at most $\frac{\alpha\ln t}{2\eta^2 j^2}$ times by round $t-1$, so by the pigeonhole principle there is some $s\in\left[g_j(t)-1,t-1\right]$ such that an arm from $U_j^t$ is pulled in round ${s}$.
Furthermore, inequality (\[eq:gjcond\]) implies $$\frac{\alpha\ln s}{2\eta^2 (j-1)^2}> \frac{\alpha\ln t}{2\eta^2 j^2}\,,$$ which guarantees that $U^s_{j-1}\subseteq U_{j}^t$. Since $s\geq g_j(t_j)-1\geq t_{j-1}$, the inductive hypothesis tells us that $|U_{j-1}^s|\geq {j-1}$, so we have $U_{j-1}^s=U_j^t$, meaning that the arm pulled in round $s$ is also in $U_{j-1}^s$.
Now, $N_i^{s-1}\geq \frac{\alpha\ln s}{2\eta^2 (j-1)^2}$ for all $i\in U_{j-1}^s$, so the upper confidence bound of the arm pulled at time $s$ is at most $$1+\sqrt{\frac{\alpha\ln s}{2\alpha\ln s/(2\eta^2 (j-1)^2)}}=1+\eta\cdot(j-1)\,.$$ The upper confidence bound at time $s$ of any arm in $[k]\setminus U_j^t$ is at least $$-1+\sqrt{\frac{\alpha\ln s}{2\alpha\ln t/(2\eta^2 j^2)}}=-1+\eta j\sqrt{\frac{\ln s}{\ln t}}\,.$$ But since $t\geq t_{j}$ and $s\geq g_j(t_{j})$, inequality (\[eq:gjcond\]) implies $$-1+\eta j\sqrt{\frac{\ln s}{\ln t}}>1+\eta\cdot(j-1)\,.$$ This means that all arms outside of $U_j^t$ have higher upper confidence bounds at time $s$ than the arms in $U_j^t$, contradicting the choice to pull an arm in $U_j^t$ at time $s$.
By induction, we conclude that the claim holds for all $j\in[k]$, and in particular that the theorem holds with $t_0=t_k$.
Free Riding with Stochastic Bandits {#sec:stochastic}
===================================
Full-Information Case for Stochastic Bandits {#full-information-case-for-stochastic-bandits .unnumbered}
--------------------------------------------
Here we describe how a free rider can take advantage of the samples collected by another player $p$ who pulls every arm sufficiently many times in expectation. This free-riding policy, which we call $\textsc{SampleAugmentingMeanGreedy}$, divides time into *epochs* of doubling length. In the $j$^th^ epoch, the free rider checks whether a given player $p$ has observed at least $\gamma j$ samples of each arm $i\in[k]$, where $\gamma$ is an appropriate constant. If all sample counts are sufficient, then the free rider uses $p$’s observed rewards to estimate the mean of each arm, committing to the arm with the maximum estimated mean for the remainder of the epoch. Otherwise, the free rider pulls any under-sampled arms, augmenting all sample counts up to at least $\gamma j$ before proceeding. Doing this allows the free rider to circumvent the logarithmic lower bound on regret and achieve $O(1)$ regret.
\[thm:sampleaugmenting\] Fix a stochastic bandit, and suppose some player $p$ plays a self-reliant policy that satisfies ${\mathbb{E}}[N_i^{t-1}]\geq\gamma\ln t$ for some $\gamma>2/\Delta^2$, all $i\in[k]$, and all sufficiently large $t$. Then a free rider can achieve $O(1)$ regret.
Below, $\hat{\mu}^s_{p,i}$ denotes the average of the first $s$ samples of $D_i$ observed by player $p$.
Let ${\mathbf{D}}$ be a stochastic bandit and let let $f$ be a self-reliant policy. Assume there are constants $\gamma>2/\Delta^2$ and $t_0$ such that ${\mathbb{E}}\left[N_{i}^{t-1}({\mathbf{D}},f)\right]\geq\gamma\ln t$, for all $i\in[k]$ and all $t\geq t_0$. Let $\mathbf{f}=(f_1,\ldots,f_n)$ be a policy profile with $f_1=\textsc{SampleAugmentingMeanGreedy}_{p,\gamma}$ and $f_p=f$, for some player $p\in\{2,\ldots,n\}$. We prove that $R_1^T({\mathbf{D}},\mathbf{f})=O(1)$.
Let $t_0'= 2^{\lceil\log t_0\rceil}$. For all $j\in{\mathbb{N}}$, define $s_j=\gamma j\ln 2$ and $\tilde{i}^*_j = \operatorname*{arg\,max}_i \nu^j_i$. Then, letting $\Delta_i$ denote $\mu^*-\mu_i$, $$\begin{aligned}
R_1^T({\mathbf{D}},\mathbf{f})&=\sum_{i\neq i^*} {\mathbb{E}}\left[N_{1,i}^T({\mathbf{D}},\mathbf{f})\right]\Delta_i\\
&\leq 2t_0'+\sum_{i\neq i^*}\Delta_i\left(\sum_{t=t_0'}^T\Pr[i_1^t=i]\right)\,.
\end{aligned}$$ Now, for each $i\in[k]$, $$\begin{aligned}
\sum_{t=t_0'}^T \Pr[i_1^t=i]&<\sum_{j=\lceil\log t_0\rceil}^{\lceil\log T\rceil} \left(\max\left\{0,s_j-{\mathbb{E}}\left[N_{p,i}^{ 2^j-1}({\mathbf{D}},\mathbf{f})\right]\right\}+\Pr\left[\tilde{i}^*_j=i\right]\cdot 2^{j}\right)\\
&=\sum_{j=\lceil\log t_0\rceil}^{\lceil\log T\rceil} \Pr\left[\tilde{i}^*_j=i\right]\cdot 2^{j}\\
&\leq \sum_{j=\lceil\log t_0\rceil}^{\lceil\log T\rceil}\left( \Pr\left[\nu_i^j-\mu_i>\Delta_i/2\right]+\Pr\left[\mu^*-\nu_{i^*}^j>\Delta_i/2\right]\right)\cdot 2^{j}\\
&\leq 2\sum_{j=0}^{\infty} \exp\left(-2\left( \frac{\Delta_i}{2}\right)^2\gamma j\ln 2\right)\cdot 2^j\\
&= 2\sum_{j=0}^{\infty}2^{(1-\Delta^2\gamma/2)j}\,.
\end{aligned}$$ Since $\Delta^2\gamma/2>1$, this sum converges to a constant, so $R_1^T({\mathbf{D}},\mathbf{f})=O(1)$.
Partial-Information Case for Stochastic Bandits {#partial-information-case-for-stochastic-bandits .unnumbered}
-----------------------------------------------
We now show that a free rider can achieve constant regret by observing a player who plays any policy that is unlikely to pull suboptimal arms too often. This class of policies includes UCB. We consider a specific, natural free-riding policy $\textsc{CountGreedy}_p$, defined by $$\textsc{CountGreedy}_p(H^{t})=\operatorname*{arg\,max}_{i\in[k]} N_{p,i}^{t}\,,$$ which always pulls whichever arm $i$ has been pulled most frequently by player $p$. Notice that this policy does not require the free rider to observe player $p$’s rewards.
One might suspect that it would be sufficient for player $p$’s policy to achieve $R_p^T = o(T)$ in order for the free rider to achieve constant regret under $\textsc{CountGreedy}$, but this turns out not to be the case. It is possible for $p$ to achieve logarithmic regret despite frequently pulling suboptimal arms with non-trivial probability, preventing $\textsc{CountGreedy}$ from achieving constant regret.
For instance, consider the self-reliant policy that, for $j=0, 1, 2\dots$, dictates the following behavior in epochs of tripling length. With probability $1/3^j$, player $p$ “gives up” on rounds $3^j$ to $3^{j+1}-1$, choosing arm $i^t_p=\operatorname*{arg\,min}_i\mu^{3^j-1}_i$. Otherwise, with probability $1-1/3^j$, player $p$ plays $\alpha$-UCB during those rounds. Under this policy, $p$’s regret grows at most logarithmically. Notice that whenever player $p$ gives up, $i_p^{3^j}$ will become her most frequently pulled arm by round $2\cdot 3^j$, so the $\textsc{CountGreedy}_p$-playing free rider will pull this arm at least $3^j$ times before round $3^{j+1}$. It is routine to show that $i_p^{3^j}$ is suboptimal with probability $1-O(1/3^j)$, so the free rider’s regret through round $T$ is at least $R_1^T\geq\sum_{j=0}^{\lfloor \log_3 T\rfloor}(1-O(1/3^j))\cdot\Delta\cdot 3^j
=\Omega(T)$.
Notice that by Theorem \[thm:ucbcount\], the above policy satisfies the conditions of Theorem \[thm:sampleaugmenting\] when $\alpha$ is sufficiently large, and therefore a free rider playing $\textsc{SampleAugmentingMeanGreedy}_p$ would achieve constant regret in this situation. Intuitively, this is because the policy of sometimes giving up on entire epochs is not “rational,” and $\textsc{SampleAugmentingMeanGreedy}$, unlike $\textsc{CountGreedy}$, does not make any implicit assumption of rationality for the self-reliant player.
Since logarithmic regret for player $p$ is not a strong enough assumption, we instead show that if the realized regret $\hat{R}_p^T$ is sublinear with sufficiently high probability, then the free rider achieves constant regret by playing $\textsc{CountGreedy}_p$.
\[thm:stoch\] Fix a stochastic bandit and assume there is some player $p$ such that for all $\epsilon>0$ there exists a $w>1$ satisfying $\Pr(\hat{R}^T_p\geq \epsilon T)=O(T^{-w})$. Then a free rider playing $\textsc{CountGreedy}_p$ achieves $O(1)$ regret.
Let $\mathbf{D}$ be a stochastic bandit, let $p\in\{2,\ldots,n\}$ be a player, and let $\mathbf{f}=(f_1,\ldots,f_n)$ be a policy profile with $f_1=\textsc{CountGreedy}_p$. Assume that for all $\epsilon>0$ there is some $w>1$ satisfying $\Pr(\hat{R}_p^T(\mathbf{D},\mathbf{f}) \geq \epsilon T)=O(T^{-w})$. We prove that $R^T_1(\mathbf{D},\mathbf{f})=O(1)$.
The free rider pulls a suboptimal arm at each time $t$ if and only if $\textsc{CountGreedy}_p(H^{t-1})\neq i^*$, which implies that $N^{t-1}_{p,i^*}\leq\frac{t-1}{2}$ and therefore $\hat{R}_p^T(\mathbf{D},\mathbf{f}) \geq\Delta \frac{t-1}{2}$. Hence, we can bound the free rider’s regret by $$R^T_1(\mathbf{D},\mathbf{f}) \leq 2\sum_{t=0}^{T-1} \Pr(\hat{R}^t_p(\mathbf{D},\mathbf{f}) \geq \Delta t/2)\,.$$ If $w>1$ satisfies $\Pr(\hat{R}^t_p(\mathbf{D},\mathbf{f}) \geq \Delta t/2) = O(t^{-w})$, then we have $$R^T_1(\mathbf{D},\mathbf{f}) \le 2\sum_{t=0}^{T-1} O(t^{-w})=O(1)\,.$$
Audibert, Munos, and Szepesvári [@audibert2009exploration] showed that $\alpha$-UCB satisfies the probability bound of Theorem \[thm:stoch\] in the single-player setting whenever $\alpha>1/2$. Since $\alpha$-UCB is a self-reliant policy, this immediately yields the following corollary.
\[cor:ucb\] If some player $p$’s policy is $\alpha$-UCB for any $\alpha > 1/2$, then a free rider playing $\textsc{CountGreedy}_p$ achieves $O(1)$ regret.
Free Riding with Contextual Bandits {#sec:contextual}
===================================
Theorems \[thm:sampleaugmenting\] and \[thm:stoch\] show that free riding is easy in the stochastic case, in which the reward distribution is identical for all players, but the task is more nuanced when players may have diverse contexts. In the linear contextual setting, different players may have different optimal arms, so a simple free-riding strategy like <span style="font-variant:small-caps;">CountGreedy</span> may fail, even when there are strong regret guarantees for the other players. In fact, as we show in Theorems \[thm:needcontexts\] and \[thm:needrewards\], successful free riding in this setting requires knowledge of both the contexts and the rewards of other players.
Full-Information Cases for Contextual Bandits {#full-information-cases-for-contextual-bandits .unnumbered}
---------------------------------------------
We now consider the full-information setting where the free rider knows other players’ contexts, actions, and rewards. We show that if the free rider’s context is a linear combination of the other players’ contexts — and if other players pull all arms sufficiently many times — then the free rider can aggregate other players’ observations to estimate the means of its own reward distribution profile. In the event that some arm has not been sampled enough by some player, the free rider temporarily acts self-reliantly and chooses arms according to UCB. Under the above assumptions, this free-riding policy, $\textsc{UCBMeanGreedy}$, achieves $O(1)$ regret. Formally, for every history $H^t$, $$\textsc{UCBMeanGreedy}_{\gamma,\mathbf{c}}(H^t)=\begin{cases}
\operatorname*{arg\,max}_i\sum_{p=2}^n c_p\hat{\mu}_{p,i}^{\lceil\gamma j\rceil}&\text{if }S_j\\
2\text{-UCB}\left(\big(i^{2^{j}}_1,r^{2^{j}}_1\big),\ldots,\big(i^{t}_1,r^{t}_1\big)\right)&\text{otherwise}\,,\end{cases}$$ where $j=\lfloor\log t\rfloor$, $S_j$ is the event that $N_{p,i}^{2^{j}-1}\geq\gamma j$ for all $p\in\{2,\ldots,n\}$ and all $i\in[k]$, and $\hat{\mu}^s_{p,i}$ denotes the average of the first $s$ observed samples of $D_{p,i}$. Notice that when applying UCB this policy treats the bandit as stochastic and “starts from scratch” in each epoch, only considering the free rider’s own actions and observed rewards from the current epoch.
\[thm:fullinfo\] Let $\mathbf{x}$ be a context profile and $\mathbf{c}\in{\mathbb{R}}^{n-1}$ be a vector such that $\sum_{p=2}^n c_p x_p=x_1$, Fix a contextual bandit, let $\Delta$ be the gap for player 1, and suppose that $\epsilon>0$ and $\gamma>\frac{8\langle\mathbf{c},\mathbf{c}\rangle\ln 2}{\Delta^2}$ satisfy $$\Pr(N_{p,i}^{t-1}<\gamma \log t) = O((\log t)^{-2-\epsilon})$$ for every player $p\in\{2,\ldots,n\}$ and arm $i\in[k]$. Then a free rider playing $\textsc{UCBMeanGreedy}_{\gamma,\mathbf{c}}$ achieves $O(1)$ regret.
Fix $j\in{\mathbb{N}}$, let *epoch* $j$ be rounds $2^j$ through $2^{j+1}-1$, and let $s_j=\lceil\gamma j\rceil$. For each $i\in[k]$ let $\tilde{\mu}^j_i=\sum_{p=2}^n c_p\hat{\mu}_{p,i}^{s_j}$, and let $\tilde{i}^j=\operatorname*{arg\,max}_{i} \tilde{\mu}^j_i$. We analyze the free rider’s regret incurred during epoch $j$ in each of the following cases: (1) $S_j$ and $\tilde{i}^j = i_1^*$, (2) $S_j$ and $\tilde{i}^j \neq i_1^*$, and (3) $\neg S_j$.
Notice that $\tilde{i}^j$ is defined in the first and second cases. In case (1), the free rider incurs no regret during epoch $j$ because it pulls only the optimal arm $i_1^*$ during that epoch. We now analyze the regret incurred from the other two cases.
For case (2),
$$\Pr(\tilde{i}^j \neq i^*_1)\le \sum_{i\in[k]}\Pr\left(\left|\tilde{\mu}^{j}_i-\mu_{1,i}\right|\geq \frac{\Delta}{2}\right)\,.$$ For fixed $i\in[k]$, letting $\hat{r}^s_{p,i}$ denote the value of the $s$^th^ observed sample of $D_{p,i}$, we have $$\begin{aligned}
\left|\tilde{\mu}^{j}_i-\mu_{1,i}\right|&=\left|\sum_{p=2}^n \frac{c_p}{s_j}\sum_{s=1}^{s_j}\hat{r}^s_{p,i}-\sum_{p=2}^n c_p\mu_{p,i}\right|\\
&=\left|\sum_{p=2}^n\sum_{s=1}^{s_j} X_p^s\right|\,,
\end{aligned}$$ where $X_p^s=\frac{c_p}{s_j}\big(\hat{r}^s_{p,i}-\mu_{p,i}\big)$. Notice that the $X_p^s$ are independent, and each $X_p^s$ is supported on $[-c_p/s_j,c_p/s_j]$ with ${\mathbb{E}}[X_p^s]=0$, so Hoeffding’s lemma gives $${\mathbb{E}}[\exp(\lambda X_p^s)]\leq \exp\left(\frac{\lambda^2c_p^2}{2s_j^2}\right)$$ for all $\lambda\in{\mathbb{R}}$. Let $\lambda=\frac{4j\ln 2}{\Delta}$, and apply a Chernoff bound: $$\begin{aligned}
\Pr\left(\left|\sum_{p=2}^n\sum_{s=1}^{s_j} X_p^s\right|\geq \frac{\Delta}{2}\right)
&\leq 2\exp\left(-\frac{\lambda\Delta}{2}\right){\mathbb{E}}\left[\prod_{p=2}^n\prod_{s=1}^{s_j}\exp(\lambda X_p^s)\right]\\
&=2\exp\left(-\frac{\lambda\Delta}{2}\right)\prod_{p=2}^n\prod_{s=1}^{s_j}{\mathbb{E}}[\exp(\lambda X_p^s)]\\
&\leq 2\exp\left(-\frac{\lambda\Delta}{2}+\sum_{p=2}^n\sum_{s=1}^{s_j}\frac{\lambda^2c_p^2}{2s_j^2}\right)\\
&\leq 2\exp\left(\frac{\lambda}{2}\left(\frac{\lambda}{\gamma j}\langle\mathbf{c},\mathbf{c}\rangle-\Delta\right)\right)\\
&=2\exp\left(2j\ln 2\left(\frac{4\langle\mathbf{c},\mathbf{c}\rangle\ln 2}{\gamma\Delta^2 }-1\right)\right)\,.
\end{aligned}$$ Thus, the contribution of this case to the regret during epoch $j$ is at most $$\begin{aligned}
k \cdot 2^{1+2j\left(\frac{4\langle\mathbf{c},\mathbf{c}\rangle\ln 2}{\gamma\Delta^2 }-1\right)}\cdot 2^{j+1}&=k \cdot 2^{2+j\cdot\left(\frac{8\langle\mathbf{c},\mathbf{c}\rangle\ln 2}{\gamma\Delta^2}-1\right)}\\
&=k\cdot 2^{-\Omega(j)}\,,
\end{aligned}$$ since $\gamma>\frac{8\langle\mathbf{c},\mathbf{c}\rangle\ln 2}{\Delta^2}$.
For the case (3), observe that $$\begin{aligned}
\Pr(\neg S_j)&\le \sum_{p=2}^n \sum_{i=1}^k \Pr(N^{2^j-1}_{p,i} <\gamma j) \\
&= O(nkj^{-2-\epsilon})\,,
\end{aligned}$$ by assumption. In this case, $\textsc{UCBMeanGreedy}_{\gamma,\mathbf{c}}$ resorts to playing $\alpha$-UCB for $2^j$ steps, incurring $O(j)$ regret. Thus, the third case contributes $O(nkj^{-1-\epsilon})$ to the expected regret of epoch $j$. Therefore, for any time horizon $T\in{\mathbb{N}}$, the regret incurred by the free rider is bounded by $$\sum_{j=0}^{\infty}\left(k\cdot 2^{-\Omega(j)}+O(nkj^{-1-\epsilon})\right)\,,$$ which converges to a constant.
Partial-Information Cases for Contextual Bandits {#partial-information-cases-for-contextual-bandits .unnumbered}
------------------------------------------------
Now we consider a situation where player 1 must choose a free-riding policy without knowledge of the other players’ contexts. We show that this restriction can force the free rider to incur logarithmic regret even given knowledge of the other players’ policies, actions, and rewards. Intuitively, this is true because a self-reliant player might behave identically in two different environments, making observations of their behavior useless to the free rider. To prove the theorem, we construct a one-dimensional, two-arm example of two such environments, then appeal to the lower bound technique of Bubeck et al. [@BuPeRi13] to show that the free rider must incur $\Omega(\log T)$ regret when acting self-reliantly.
\[thm:needcontexts\] A free rider without knowledge of the other players’ contexts may be forced to incur $R^T=\Omega(\log T)$ regret, regardless of what self-reliant policies the other players employ.
We prove that there exist a pair of contextual bandits $\mathbf{F}$ and $\mathbf{F}'$ and a pair of two-player context profiles $\mathbf{x}$ and $\mathbf{x}'$ such that, for every time horizon $T$ and every policy profile $\mathbf{f}=(f_1,f_2)$ in which $f_1$ is independent of player 2’s context and $f_2$ is self-reliant, $$\label{eq:partial}
\max\{R^T_1(\mathbf{F},\mathbf{x},\mathbf{f}),R^T_1(\mathbf{F}',\mathbf{x}',\mathbf{f})\}\geq\frac{\ln(T/12)+1}{2}\,.$$ We construct a one-dimensional, two-arm, two-player example. Let $F_1$ be a point mass at $0$; let $F_2$ and $F'_2$ be discrete random variables that take value $1$ with probability $1/3$ and $2/3$, respectively, and value $-1$ otherwise; and let $\mathbf{F}=(F_1,F_2)$ and $\mathbf{F}'=(F_1,F_2')$. Let $\mathbf{f}=(f_1,f_2)$ be any linear contextual bandit policy profile such that $f_2$ is self-reliant, and consider a free-riding player 1.
Let $\mathbf{x}=(1,1)$ and $\mathbf{x}'=(1,-1)$. For $p,i\in[2]$, let $D_{p,i}$ be the reward distribution of arm $i$ for player $p$ under contextual bandit $\mathbf{F}$ and context profile $\mathbf{x}$. Similarly, let $D'_{p,i}$ be the reward distribution of arm $i$ for player $p$ under parameter distribution profile $\mathbf{F}'$ and context profile $\mathbf{x}'$. Observe that $D_{1,1}=D'_{1,1}$, $D_{2,1}=D'_{2,1}$, and $D_{2,2}=D'_{2,2}$, but $D_{1,2}=-D'_{1,2}$.
Informally, the environment $(\mathbf{F},\mathbf{x},\mathbf{x})$ is indistinguishable from $(\mathbf{F}',\mathbf{x}',\mathbf{f})$ from the perspective of player 2. Observing player 2’s actions and rewards will therefore be completely uninformative for player 1, who is ignorant of player 2’s context. Thus, player 1’s task is essentially equivalent to a single-player stochastic bandit problem where the learner must distinguish between reward distribution profiles $(D_{1,1},D_{1,2})$ and $(D'_{1,1},D'_{1,2})$. Bubeck et al. [@BuPeRi13] showed that the latter task requires the learner to experience logarithmic regret. Adapting their proof to the present situation, we can demonstrate that (\[eq:partial\]) holds. Our situation is almost identical to theirs, except for the presence of an uninformative second player, which requires only minor changes to their proof. We include the details here for the sake of completeness:
Let $A=(\mathbf{F},\mathbf{x},\mathbf{f})$ and $B=(\mathbf{F}',\mathbf{x}',\mathbf{f})$ be the two environments. Observe that $$\max\{R_1^T(A),R_1^T(B)\}\geq \frac{{\mathbb{E}}[N_{1,2}^T(A)]}{3}\,,$$ and $$\begin{aligned}
\max\left\{R_1^T(A),R_1^T(B)\right\}&\geq \frac{1}{2}\left(R_1^T(A)+R_1^T(B)\right)\\
&=\frac{1}{6}\sum_{t=1}^T\left(\Pr[i^t_1(A)=1]+\Pr[i^t_1(B)=2]\right)\\
&\geq \frac{1}{12}\sum_{t=1}^T\exp(-{\operatorname*{KL}}(H^t(A),H^t(B)))\\
&\geq \frac{T}{12}\exp(-{\operatorname*{KL}}(H^T(A),H^T(B)))\,,
\end{aligned}$$ where the second-to-last line follows from the Bretagnolle-Huber inequality (Lemma \[lem:bh\]) [@BreHub79].
We now calculate $\operatorname*{KL}((H^T(A),H^T(B))$. Observe that $(i_2^t(A),r_2^t(A))_{t\in[T]}$ and $(i_2^t(B),r_2^t(B))_{t\in[T]}$, player 2’s components of the history in environments $A$ and $B$, are distributed identically. This means that the conclusion of Lemma \[lem:dd\] [@BanditBook] still applies. In particular, by the chain rule for divergence, $$\begin{aligned}
\operatorname*{KL}\left((H^T(A),H^T(B)\right)
&=\sum_{t=1}^T{\mathbb{E}}\left[\operatorname*{KL}\left(\left((i_1^t(A),r_1^t(A))\mid H^{t-1}(A)\right),\left((i_1^t(B),r_1^t(B))\mid H^{t-1}(B)\right)\right)\right]\\
&=\sum_{t=1}^T{\mathbb{E}}[\operatorname*{KL}\left(D_{1,f_1(H^{t-1}(A))},D_{1,f_1(H^{t-1}(B))}\right)]\\
&=\sum_{t=1}^T\Pr[f_1(H^{t-1}(A))=2]\cdot\operatorname*{KL}(D_{2},D'_{2})\\
&=\operatorname*{KL}(D_{2},D'_{2}){\mathbb{E}}[N_{1,2}^T(A)]\\
&={\mathbb{E}}[N_{1,2}^T(A)]/3\,.
\end{aligned}$$ Thus, we have $$\begin{aligned}
\max\{R_1^T(A),R_1^T(B)\}&\geq \frac{1}{2}\left(\frac{{\mathbb{E}}[N_{1,2}^T(A)]}{3}+\frac{T}{12}\exp(-{\mathbb{E}}[N_{1,2}^T(A)]/3)\right)\\
&\geq\frac{1}{6}\min_{x\in[0,T]}\left(x+\frac{Te^{-x/3}}{4}\right)\\
&=\frac{\ln(T/12)+1}{2}\,.
\end{aligned}$$
Similarly, a free rider in the contextual setting needs to know the other players’ rewards; knowing their contexts, policies, and actions may not be sufficient to successfully free ride, even when all other players have low realized regret with high probability and are guaranteed to pull all arms frequently. This is in contrast to the stochastic case, as Theorem \[thm:stoch\] demonstrates. We prove this by describing a self-reliant policy, <span style="font-variant:small-caps;">EpochExploreThenCommit</span>, that again proceeds in doubling epochs. At the beginning of the $j$^th^ epoch, the player samples each arm $\Theta(j)$ times, then commits to the arm with the highest sample mean for the remainder of the epoch. This policy has strong guarantees on sample count and realized regret, but we construct an example where, with constant probability, the sequence of arm pulls is completely uninformative to the free rider.
$t=1$\
\[thm:needrewards\] A free rider without knowledge of the other players’ rewards may be forced to incur $R^T=\Omega(\log T)$ regret, even when all other players satisfy the conditions of Theorems \[thm:sampleaugmenting\] and \[thm:stoch\].
Fix $\gamma\geq 2/\Delta^2$, and let $g_\gamma=\textsc{EpochExploreThenCommit}_{\gamma}$. We prove that $g_\gamma$ satisfies the following three properties:
1. For all contextual bandits $\mathbf{F}$, context profiles $\mathbf{x}$, and policy profiles $\mathbf{f}=(f_1,\ldots,f_n)$ with $f_p=g_\gamma$, there is some $t_0\in\mathbb{N}$ such that $N_{p,i}^T(\mathbf{F},\mathbf{x},\mathbf{f})\geq\gamma\log T$ for all $i\in[k]$ and all $T>t_0$.
2. For all contextual bandits $\mathbf{F}$, context profiles $\mathbf{x}$, policy profiles $\mathbf{f}=(f_1,\ldots,f_n)$ with $f_p=g_\gamma$, and $\epsilon>0$, there is some $w>1$ such that $\Pr\big(\hat{R}^T_p(\mathbf{F},\mathbf{x},\mathbf{f})\geq\epsilon T\big)=O(T^{-w})$.
3. There exist a pair of contextual bandits $\mathbf{F}$ and $\mathbf{F'}$ and a context profile $\mathbf{x}$ such that for all policy profiles $\mathbf{f}=(f_1,g_\gamma,g_\gamma)$ such that $f_1$ is independent of the other players’ observed rewards, $\max\left\{R_1^T(\mathbf{F},\mathbf{x},\mathbf{f}),R_1^T(\mathbf{F'},\mathbf{x},\mathbf{f})\right\}=\Omega(\log T)$.
First, let $j_0\in {\mathbb{N}}$ satisfy $2^{j_0}\geq k\gamma(j_0+2)$, and let $i\in[k]$ be any arm. Then, for all $j\geq j_0$, $f$ satisfies $N_{p,i}^{2^j}(\mathbf{F},\mathbf{x},\mathbf{f})\geq\gamma (j+1)$, i.e., each arm has been pulled at least $\gamma(j+1)$ times at the beginning of the epoch. So for every round $t$ in the $j$^th^ epoch, we have $N_{p,i}^{t}(\mathbf{F},\mathbf{x},\mathbf{f})\geq\gamma(j+1)\geq \gamma\log t$. Hence, $N_{p,i}^{T}(\mathbf{F},\mathbf{x},\mathbf{f})\geq \gamma\log T$ for all $T>2^{j_0}$, so $g_\gamma$ satisfies the first property.
Now, for each $j\in{\mathbb{N}}$, define $\tilde{R}^j=\hat{R}^{2^{j+1}-1}-\hat{R}^{2^j-1}$, the realized regret incurred during epoch $j$ while playing $g_\gamma$. At most $2k\gamma (j+2)$ of this realized regret can come from the exploration phase; any further regret in that epoch can only result from committing to a suboptimal arm. By Hoeffding’s inequality, $$\begin{aligned}
\Pr\left(\tilde{R}^j>2 k\gamma (j+2)\right)&\le \sum_{i \in [k]} \exp\left(-\frac{\gamma (j + 2) (\mu^*-\mu_i)^2}{2}\right) \\
&\le k \cdot \exp\left(-\frac{\gamma (j + 2) \Delta^2}{2}\right)\,.\end{aligned}$$ Notice that $\sum_{j=0}^{\lfloor \log(\epsilon T)\rfloor-2}\tilde{R}^j\leq \sum_{t=1}^{\epsilon T/4} 2=\frac{\epsilon T}{2}$, and that $\lceil\log T\rceil-(\lfloor \log(\epsilon T)\rfloor-2)\leq 4-\log \epsilon$. So $$\begin{aligned}
\Pr\left(\hat{R}^T\geq\epsilon T\right)&\leq\Pr\left(\sum_{j=\lfloor \log(\epsilon T)\rfloor-1}^{\lceil\log T\rceil}\tilde{R}^j\geq\frac{\epsilon T}{2}\right)\\
&\leq \sum_{j=\lfloor \log(\epsilon T)\rfloor-1}^{\lceil\log T\rceil}\Pr\left(\tilde{R}^j\geq \frac{\epsilon T}{8-2\log\epsilon}\right)\,.\end{aligned}$$ For all sufficiently large $T$, $2 k\gamma(\lceil \log T\rceil+2)<\frac{\epsilon T}{8-2\log\epsilon}$, so for $j=\lfloor \log(\epsilon T)\rfloor-1,\ldots,\lceil\log T\rceil$, $$\begin{aligned}
\Pr\left(\tilde{R}^j\geq \frac{\epsilon T}{8-2\log\epsilon}\right)&\leq \Pr\left(\tilde{R}^j>2 k\gamma (j+2)\right)\\
&\leq k \cdot \exp\left(-\frac{\gamma (\lfloor \log(\epsilon T)\rfloor+1) \Delta^2}{2}\right)\,.
\end{aligned}$$ Thus, $$\begin{aligned}
\Pr\left(\hat{R}^T\geq\epsilon T\right)&\leq (4-\log\epsilon)\cdot k \cdot \exp\left(-\frac{\gamma (\lfloor \log(\epsilon T)\rfloor+1) \Delta^2}{2}\right)\\
&=O\left(T^{-\frac{\gamma\Delta^2}{2}}\right)\,,\end{aligned}$$ meaning that $\frac{\gamma\Delta^2}{2}>1$. So $g_\gamma$ satisfies the second property.
Finally, let $\mathbf{x}=((\sqrt{2}/2,\sqrt{2}/2),(1,0),(0,1))$ be a three-player context profile, and let $\mathbf{F}$ be the contextual bandit where for each $i\in[3]$, the feature distribution $F_i$ satisfies $F_i(x_i)=2/3$ and $F_i(-x_i)=1/3$, and $F_4$ is a point mass at $(0,0)$. Define a second contextual bandit $\mathbf{F}'=(F_1',F_2,F_3,F_4)$, where $F_1'(x_1)=1/3$ and $F_1'(-x_1)=2/3$. Let $f_1$ be some policy that is independent of the other players’ observed rewards, and consider the policy profile $\mathbf{f}=(f_1,g_\gamma,g_\gamma)$.
Simulations {#sec:simulation}
===========
![Average realized regret of a $2$-UCB player and a free rider over 100 simulations. There are 10 arms whose reward distributions are Bernoulli, with parameters $0.0, 0.1, \dots, 0.8, 0.9$.[]{data-label="fig:stochastic_bernoulli"}](stochastic_bernoulli.png){width="\columnwidth"}
![Average realized regret of the free rider over 10 simulations. There are 50 players playing $10$-UCB and 30 arms, the dimension of the vectors is 10, and $\Delta=0.1839$.[]{data-label="fig:contextual_big"}](contextual_big.png){width="\columnwidth"}
Finally, we present simulation results for both the stochastic and the contextual cases. In the stochastic case, we have the free rider simply pull the most pulled arm of the self-reliant player, and in the contextual case, we have the free rider pull the arm with the highest sample mean calculated by taking the linear combination of the sample means from the other players. In each of the experiments, we refer to $\max_{i\in[k]}\sum_{t=1}^T r^{t,i}-\sum_{t=1}^T r_p^t$ as the *realized regret* for player $p$, where $r^{t,i}$ is the reward one would have observed by pulling arm $i$ in round $t$. For the stochastic case, we consider situations where the reward distribution $D_i$ of each arm $i$ is Bernoulli, with different parameters $p$. As shown in Figure \[fig:stochastic\_bernoulli\], the realized regret of the free rider flattens out after some constant number of rounds, where this constant depends on the reward distribution profile $\mathbf{D}$.
In the contextual case, the context $x_p$ for each player $p$ and the vector $\mathbf{c}$ (i.e., the coefficients for the linear combination of other players’ contexts that gives the free rider’s context), are all chosen uniformly at randomly from $[-1,1]$. The feature distribution $F_i$ for each arm $i$ is a multi-variate normal distribution with covariance matrix is $0.1\mathcal{I}$, where $\mathcal{I}$ is the identity matrix, and the mean vector is once again chosen by sampling each coordinate uniformly from $[-1,1]$. We normalize $x_p$ and the mean vector of $F_i$ so that the expected reward for each arm $i$ falls within $[-1,1]$ for every player $p$.
Conclusion {#sec:conclusions}
==========
We have demonstrated that in the linear contextual setting, a free rider can successfully shirk the burden of exploration, achieving constant regret by observing other players engaged in standard learning behavior. Furthermore, we have shown that even with partial information and weaker assumptions on the other players’ learning behaviors, the free rider can achieve constant regret in the simple stochastic setting. It would be interesting to examine richer settings. For example, exploring players need not be self-reliant, and both exploring players and free riders could play a range of strategies. As another example, when a free rider in the stochastic setting only sees the actions (and not the rewards) of the self-reliant players and does not know which of them are playing UCB or other zero-regret strategies, can he still achieve constant regret? More realistically, users of a service like Yelp cannot be partitioned into self-reliant public learners and selfish free riders who keep their data private. It would be interesting to explore more nuanced player roles and to characterize the equilibria that arise from their interactions. Such a characterization might also suggest mechanisms for the deterrence of free riding or for incentivizing exploration.
[^1]: Some sources refer to this quantity as *pseudo-regret* and use *regret* to refer to the realized regret.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Most people participate in meetings almost every day, multiple times a day. The study of meetings is important, but also challenging, as it requires an understanding of social signals and complex interpersonal dynamics. Our aim this work is to use a data-driven approach to the science of meetings. We provide tentative evidence that: i) it is possible to automatically detect when during the meeting a key decision is taking place, from analyzing only the local dialogue acts, ii) there are common patterns in the way social dialogue acts are interspersed throughout a meeting, iii) at the time key decisions are made, the amount of time left in the meeting can be predicted from the amount of time that has passed, iv) it is often possible to predict whether a proposal during a meeting will be accepted or rejected based entirely on the language (the set of persuasive words) used by the speaker.'
author:
- Been Kim
- Cynthia Rudin
bibliography:
- 'final1.bib'
title: Learning About Meetings
---
Introduction
============
*“A meeting is indispensable when you don’t want to get anything done.” [@kayser1990mining]*
In the United States alone, an estimated 11 million meetings take place during a typical work day [@meetingNumber]. Managers typically spend between a quarter and three-quarters of their time in meetings [@mackenzie2009time], and approximately 97% of workers have reported in a large-scale study [@hall1994americans] that to do their best work, collaboration is essential. Most of us work directly with others every day, and want to be useful participants in meetings. Meeting analysis (the science of meetings) can potentially help us understand various aspects of meetings, find ways to allow us to be more effective participants in our meetings, and help us to create automated tools for meeting assistance. Meeting dynamics can be complex; often proposals are implicitly communicated and accepted [@Eugenio99theagreement]. Despite the plethora of meetings that happen each day, and despite a body of work on meeting analysis [e.g., see @romano2001meeting for a review], we still cannot claim that we understand the topic of meetings well enough that these studies have led to quantifiable improvements in the overall quality of meetings, nor useful guidelines for more productive meetings. Perhaps if we step back and consider a basic science approach to meeting analysis, we would quantitatively uncover aspects of meetings that could eventually lead to true improvements.
In this work, we develop and use predictive modeling tools and descriptive statistics in order to provide a data-driven approach to the scientific study of meetings. We provide preliminary answers to several key questions: i) Can we automatically detect when the main decisions are made during the meeting, based only on the frequency and type of dialogue acts in a given period of time? In other words, based on the types of utterances people are making, can we determine whether the most important part of the meeting is occurring? ii) Is there any type of pattern of dialogue common to most or all meetings? We would like to know whether there is a “quintessential” pattern of dialogue that we can identify. iii) How long is a meeting going to last, in terms of “wrap-up” time, beyond the time that the main decisions are made? Sometimes this wrap-up time is substantial, and the meeting extends well beyond when the main decisions are made. iv) Can we predict whether a proposal made during a meeting will be accepted or rejected based entirely on the language (the set of persuasive words) used by the speaker? There have been many studies and commentaries focused on “persuasive words,” but do those persuasive words truly correlate with successful outcomes during a meeting? Some of these questions can be answered using current machine learning approaches, but others cannot. For finding patterns in dialogue acts that are common to most/all meetings, we design an algorithm to learn a sparse representation of a meeting as a graph of dialogue acts, which yields insight about the way social dialogue acts intersperse with work-related dialogue acts.
In our study, we use the most extensive annotated corpus of meetings data in existence, originating from the AMI (Augmented multi-party interaction) project [@Mccowan05theami]. This dataset contains a large number of annotated meetings (over 12,000 human-labeled annotations). In total, there are 108,947 dialogue acts in total number of 95 meetings, and 26,825 adjacency pairs (explained below). This corpus is derived from a series of real meetings, controlled in the sense that each meeting has four participants who work as a team to compose a new design for a new remote control. Each participant takes a role, either project manager, marketing expert, industrial designer or interface designer. The participants are given training for their roles at the beginning of the task. Documents used to train participants and annotators are publicly available. The interaction between participants is unstructured, and each person freely chooses their own style of interaction. The length of meetings ranges from approximately 10 minutes to 45 minutes, which overlaps with the most common lengths of meetings [discussed by @romano2001meeting]. Here we provide detail on the annotations provided with the corpus.
*Dialogue Acts*: A dialogue act marks a characteristic of an utterance, representing the intention or the role of the utterance. It is possible to use dialogue act data to predict characteristics of future dialogue [@nagata1994first], or to do automatic dialogue act tagging [@stolcke2000dialogue; @ji2005dialog]. Dialogue acts include questions, statements, suggestions, assessment of suggestions (positive, negative or neutral) and social acts. Each sentence is often divided into pieces and tagged with different dialogues acts.[^1] A sequence of dialogue acts looks like this:\
` A: Suggestion B: Commenting on A’s suggestion C: Asking questions A: Answering C’s question B: Accepting A’s suggestion `
*Decision summary*: A summary of decisions made in a meeting and related dialogue acts.
*Discussion*: A set of dialogue acts and their time stamps that support or reflect decisions described in the decision summary annotation.
*Adjacency pairs*: An adjacency pair encodes the relationship between two dialogue acts. It represents the reaction to the first dialogue act that is expressed within the second dialogue act; e.g., the first dialogue act can be a suggestion and the second dialogue act can be a positive or negative response to it, as demonstrated in the sequence of dialogue acts above.
We note that meetings can have many purposes, [e.g., see @romano2001meeting], however in this work we study only meetings where the purpose is to make a group judgment or decision (rather than for instance, to ensure that everyone understands, or to explore new ideas and concepts).
Our specific goal in this work is to contribute insights to the new scientific field of meeting analysis. We provide tentative answers to the above questions that can be used for constructing other studies. We do not claim that our results definitively answer these questions, only that they yield hypotheses that can be tested more thoroughly through other surveys. Most medical (and more generally, scientific) studies make or verify hypotheses based on a database or patient study. Hypotheses based on these single studies are sometimes first steps in answering important questions - this is a study of that kind, and we believe our results can be tested and built upon.
Our more general goal is to show that ML can be useful for exploratory study of meetings, including and beyond the questions studied here. This paper is not focused on a specific machine learning technique per se, its goal is to show that ML methods can be applied for exploratory scientific analysis of data in this domain, leading towards the eventual goal of increasing meeting productivity. We start with a question for which a simple application of machine learning tools provides direct insight.
Question i: Can we automatically detect when key decisions are made? \[q1\]
===========================================================================
If we can learn the statistics of dialogue acts around the time when important decisions are about to be made or being made, we can potentially detect the critical time window of a meeting. One can imagine using this information in several ways, for instance, to know at what point in the meeting to pay more attention (or to join the meeting in the first place), or to use this as part of a historical meetings database in order to fast forward the meeting’s recording to the most important parts.
In our setup, each feature vector for the detection problem is a time-shifted bag-of-dialogue-acts. The feature vector for a specific timeframe is defined by the counts of different dialogue acts (e.g., one example has 3 suggestions, 2 positive responses and 10 information exchanges). Using this representation means that the results do not depend on specific keywords within the meeting dialogue; this potentially allows our results to hold independently of the specific type and purpose of the meeting being held. Among dialogue acts, we use only a subset that are known to be relevant to the group decision making process [@Eugenio99theagreement], namely: action directive, offer, accept, reject, info-request, and information. Action directives represent all elicit forms of dialogue acts — dialogue acts that require actions from hearers. By limiting to these dialogue acts, we are now working with a total of 53,079 dialogue acts in the corpus.
Using this definition, a timeframe of dialogue becomes one example, represented by a 6-dimensional vector, where each element is the count of a particular dialogue act within the specified timeframe. The labels are 1 if the decisions are annotated as being made in that timeframe of interest and 0 otherwise. We considered timeframes with average size 70 timestamps (roughly 5 minutes). This choice of timeframe comes from the minimum meeting time considered in the research done on meeting profiles by [@panko1995meeting].
We applied supervised and unsupervised classification algorithms to predict whether important decisions are being made within a given block of dialogue. The data were divided into 15 folds, and each fold was used in turn as the test set. The supervised algorithms were SVM with radial basis function (RBF) kernels, SVM with linear kernels, logistic regression, and Naïve Bayes with a Gaussian density assumption on each class and feature. For the unsupervised algorithms, which (purposely) discard the annotations, we used EM with Gaussian Mixture Models and Kmeans. From the resulting two clusters, we chose the better of the two possible labelings. The AUC, precision, recall, and F-measure for each algorithm’s performance on the testing folds are reported in Table \[tab:results\_for\_A\]. The sample mean and sample standard deviation of AUC values over the test folds are reported within the second column of the table.
0.15in
---------------------------------------------------------------------------------------- -- -- -- -- --
**[Method]{} &**[AUC $\pm$ Std. ]{}& **[Precision]{} & **[Recall]{} & **[F-measure]{}\
SVM-Linear& 0.87 $\pm$ 0.05 & 0.89 & 0.87 & 0.88\
Logistic regression& 0.87 $\pm$ 0.05 & 0.70 & 0.56 & 0.62\
SVM-RBF& 0.86 $\pm$ 0.06 & 0.86 & 0.95 & 0.90\
EM-GMM& 0.57 $\pm$ 0.06 & 0.68 & 0.48 & 0.56\
NB-Gaussian & 0.56 $\pm$ 0.10 & 0.88 & 0.81 & 0.84\
Kmeans& 0.48 $\pm$ 0.24 & 0.69 & 0.34 & 0.45\
**********
---------------------------------------------------------------------------------------- -- -- -- -- --
: Results for predicting key decision times\[tab:results\_for\_A\]
-0.1in
The prediction quality, with respect to the AUC, is very similar for SVM-Linear, Logistic Regression, and SVM-RBF. All three methods show high AUC values around 0.86 or 0.87. The three other methods do not perform nearly as well. Naïve Bayes has very strong independence assumptions that are clearly violated here. EM-GMM is a non-convex approach that has a tendency to get stuck in local minima. Both EM-GMM and K-Means are unsupervised, so they do not get the benefit of being trained on human-labeled data. We remark that the predictive performance given by some of these algorithms is quite high given that the imbalance ratio is only 3 to 1 (three “no decision made” examples for each “decision made” example). Logistic regression performed well with respect to the AUC, but not with respect to the other measures. AUC is a rank statistic, whereas the other measures are relative to a decision boundary. Thus, if the decision boundary is in the wrong place, precision, recall, and F-measure suffer, regardless of whether the positives are generally given higher scores than the negatives. (There is an in-depth discussion by [@ErtekinRu11] on logistic regression’s performance with respect to the AUC.) The opposite is true for Naïve Bayes, where the decision boundary seems to be in the right place, leading to good classification performance, but the relative placement of positives and negatives within the classes lead to rank statistics that are not at the level of the other algorithms. This could potentially be due to the choice of scoring measure used for the AUC, as the choice of scoring measure is not unique for Naïve Bayes. We chose $P(y=1)\prod_j \hat{P}(x^j|y=1)$, where the empirical probability is computed over the training set. (Here $y=1$ indicates decisions being made, and $x^j$ is the $j^\textrm{th}$ feature value.) It is possible that even if an example scored highly for being a member of the positive class, it could score even more highly for being a member of the negative class, and thus be classified correctly. Thus, care should be taken in general when judging Naïve Bayes by the AUC; in this case, quantities computed with respect to the decision boundary (true positives, false positives, etc.) are more natural than rank statistics.
------------------------------------------------------------------- -- --
**[Ranking]{} & **[Dialogue Acts]{} & **[$\lambda$ $\pm$ Std.]{}\
1& Information & 0.30 $\pm$0.031\
2& Information Request & 0.11 $\pm$ 0.03\
3& Offer& -0.0076 $\pm$ 0.04\
4& Action-directive& -0.0662 $\pm$0.04\
5& Reject& -0.20 $\pm$0.03\
6& Accept& -0.27 $\pm$0.02\
******
------------------------------------------------------------------- -- --
: Feature ranking using SVM coefficient \[tab:q1ranking\_svm\]
We can use the SVM coefficients to understand the distribution of dialogue acts during the important parts of the meeting. As it turns out, the important parts of the meeting are characterized mostly by information and information request dialogue acts, and very few offers, rejections, or acceptances. This is shown in Table \[tab:q1ranking\_svm\]. We hypothesize that at the important parts of the meeting, when the decisions have been narrowed down and few choices remain, the meeting participants would like to ensure that they have all the relevant information necessary to make the decision, and that the outcome will fit within all of their constraints.
This small experiment has implications for the practical use of machine learning for automated meeting recording and assistance software. First, it is possible to obtain at least 0.87 AUC in detecting the time frame when decisions are made. An algorithm with this level of fidelity could be useful, for instance, in searching through large quantities of meeting data automatically (rather than manually scrolling through each meeting). One could envision having software create an alert that key decisions are being made, so that upper-level management can then choose to join the meeting. Or the software could inform secretarial staff of an estimate for when the meeting is done in order to facilitate schedule planning. (To do this, however, it might be useful to incorporate knowledge from Section \[q3\] about the expected length of the wrap-up time.)
Question ii: Is there a pattern of interactions within a meeting?\[q2\]
=======================================================================
There is often a mixture of social and work-related utterances during a meeting, and it is not obvious how the two interact. In particular, we would like to study the way in which social acts (positive or negative) interact with work-related acts (i.e., acceptance or rejection of proposals). More abstractly, we would like to know if there is a “quintessential representation” of interactions within a meeting, where a representation is a directed graph of dialogue acts. If there is such a representation, we would like to learn it directly from data. To do this, we will present a discrete optimization approach that uses the notion of “template instantiation.” The optimization will be performed via simulated annealing. First, let us formally define the problem of template discovery.
Formalization of the Problem of Template Discovery
--------------------------------------------------
We define a *template* as a graph, where each node takes a value in the set of dialogue acts, and the graph has directed edges from left to right, and additional backwards edges. Formally, let $\Lambda$ be the set of possible dialogue acts ($\Lambda^n$ is a string of length $n$). Specifically, define the set $\mathcal{D}$ as $$\mathcal{D}: ( n \in \{ 1, \cdots, L\} ) \times \Lambda^n \times \Pi_{B,n}$$ so that $\mathcal{D}$ is a set of templates with length of size $n \leq L$, where if the template is of size $n$, there are $n$ dialogue acts. There are forward edges between neighboring nodes (i.e. dialogue acts). $\Pi_{B,n}$ is the set of all possible backward arrows for the template, containing at most $B$ backward arrows. One can represent $\Pi_{B,n}$ as the set of $n \times n$ lower diagonal binary matrices, 1’s indicating backward arrows, with at most $B$ values that are 1. Let $X$ be the set of meetings, which are strings consisting of an arbitrary number of elements of $\Lambda$. Define the loss function $l: \mathcal{D} \times X \to \mathbb{Z_{+}}$ as $$l(t,x) := \displaystyle\min_{t_j \in \textrm{instan}(t)} [\textrm{dist}(t_j, x)]$$ where dist is edit distance, $\textrm{dist}:\Lambda^{n_1} \times \Lambda^{n_2} \to \mathbb{Z_+}$ for strings of lengths $n_1$ and $n_2$. We define $\textrm{instan}(t)$ as the set of *template instantiations* $t_j$ of the template $t$, where a template instantiation is a path through the graph, beginning or ending anywhere within the graph. The path is a sequence of elements from $\Lambda$. Consider for instance the example template at the top of Figure \[template\_enum\]. Two example template instantiations are provided just below that. The edit distance between two strings is the minimum number of insertions, deletions, and/or substitutions to turn one string into the other. Each meeting $x$ is a string of dialogue acts, and each template instantiation $t_j$ is also a string of dialogue acts, so the edit distance is well-defined.
![\[template\_enum\] Example template instantiations. The template is at the top, and two instantiations are below.](backArrowExplain){width="3.2in"}
We assume that we are given meetings $x_1, \cdots, x_m$ drawn independently from an unknown probability distribution $\mu$ over possible meetings $X$. Define empirical risk $R^{\textrm{emp}}(t):\mathcal{D} \to \mathbb{R_+}$ and true risk $R^{\textrm{true}}(t): \mathcal{D} \to \mathbb{R_+}$ as: $$R^{\textrm{emp}}(t):= \displaystyle\frac{1}{m} \displaystyle\sum^m_{i=1} l(t, x_i) \textrm{ and } R^{\textrm{true}}(t) := \mathbb{E}_{x\sim\mu} [ l(t, x) ].$$ We can bound the difference between the empirical risk and true risk for the problem of template discovery as follows:
For all $\delta > 0$, with probability at least $1-\delta$ with respect to the independent random draw of meetings $x \sim \mu$, $x_i \sim \mu$, $i=1,...,m$, and for all templates $t$ in $\mathcal{D}$ with length at most $L$ and with number of backwards arrows at most $B$: $$R^{\text{\rm true}}(t)\leq R^{\text{\rm emp}}(t) + \sqrt{\frac{\log \displaystyle\sum_{n=0}^{L} \left[|\Lambda|^n \displaystyle\sum_{b=0}^{\min(B,n)} {n(n-1)/2 \choose b}\right] + \log \displaystyle\frac{1}{\delta}}{2m}},$$ where $|\Lambda|$ is the number of elements in the set $\Lambda$.
The proof is in the Appendix.
This framework for macro-pattern discovery leads naturally to an algorithm for finding templates in real data, which is to minimize the empirical risk, regularized by the number of backwards arrows and the length of the template.
Macro-Patterns in Meetings
--------------------------
Let us explain the data processing. We selected the annotations that were meaningful in this context, namely: socially positive act, socially negative act, negative assessment act and positive assessment act. This allows us only to focus on assessments (either social or work related), as they may have a generally more powerful effect on the trends in the conversation than other dialogue acts. That is, the assessments create a “macro-pattern” within the meeting that we want to learn. Repeated dialogue acts by the same person were counted as a single dialogue act. The selected data contain 12,309 dialogue acts, on average 130 acts per meeting.
We would like to know if all or most meetings have something in common, in terms of the pattern of dialogue acts throughout the meeting (e.g., a smaller sub-conversation in which certain dialogue acts alternate, followed by a shift to a different sub-conversation, and so on). If this were the case, a template for meetings could be a directed graph, where each meeting approximately follows a (possibly repetitive) path through the graph. Time might loosely follow from left to right. We thus learn the directed graph by optimizing the following regularized risk functional using discrete optimization: $$\begin{aligned}
\label{eq:optfunction}
\lefteqn{F(\textrm{template } t)} \nonumber \\
&=&\frac{1}{m}
\sum_{\textrm{meetings }i} \min_{t_j\in\textrm{instan}(t)}
[\textrm{dist}(t_j, \textrm{meeting } i)]
+ C_1\textrm{length}(t)+C_2\textrm{backw}(t),\end{aligned}$$ which is a regularized version of the empirical risk defined above. Intuitively, (\[eq:optfunction\]) characterizes how well the set of meetings match the template using the first term, and the other two terms are regularization terms that force the template to be sparser and simpler, with most of the edges pointing forwards in time. The smaller templates have the advantage of being more interpretable, also according to the bound above, smaller templates encourage better generalization. The length$(t)$ is the number of nodes in template $t$. The value of backw$(t)$ is the number of backwards directed edges in the template graph. We chose $C_1=1$ and $C_2=0.1$ to force fewer backwards edges than forwards edges, meaning a graph that follows more linearly in time. Template instantiations always follow paths that exist within the template, whereas a real meeting will likely never exactly follow a template (unless the template is overly complex, in which case the algorithm has wildly overfit the data, which is prevented by validation). Our calculation for $\displaystyle\min_{t_j\in\textrm{instan}(t)}[\textrm{dist}(t_j, \textrm{meeting } i)]$ in the first term of (\[eq:optfunction\]) is approximate, in the sense that the min over all templates is calculated over all template instantiations that are approximately the same length as the meeting $i$, rather than over all instantiations. As it is likely the minimum would be achieved at an instantiation with approximately the same length as the meeting, this is a reasonable assumption to make in order to make the calculation more tractable.
We optimize (\[eq:optfunction\]) over templates using simulated annealing, as shown in Algorithm \[alg:sa\]. Simulated annealing probabilistically decides whether it will move to a neighboring state or stay at the current state at each iteration. The neighborhood is defined as a set of templates that are edit distance 1 away from the current template under allowable operations. The allowable operations include insertion of a new node between any two nodes or at the beginning or end of the template, deletion of a node, insertion or deletion of backwards directed edges. For the proposal distribution for simulated annealing, each operation is randomly chosen with uniform probability. The acceptance probability of the new template is 1 if the proposed objective function value is less than the current value. If the proposed value is larger than the current value, the acceptance function accepts with probability $\exp \left( -\Delta F/T \right) $ where $\Delta F$ represents the difference in objective function value, namely the proposed function value minus the current function value, and $T$ is the current temperature (lines \[alg:accept\_jump1\]-\[alg:accept\_jump2\] in Algorithm \[alg:sa\]). The annealing schedule is $T=T_0\cdot 0.95^k$, where $k=800$ is the annealing parameter and $T_0=1000$ is the initial temperature. After every $k$ accepted steps, we restarted the optimization starting at the best point so far, resetting the temperature to its initial value. The maximum iteration number was set at 4000.
\[alg:template\_enum\] \[alg:cal\_edit\_dist1\] \[alg:cal\_edit\_dist2\] \[alg:min\_edit\_dist\] \[alg:costfun\] \[alg:accept\_jump1\] \[alg:accept\_jump2\]
We ran the algorithm starting from 95 different initial conditions, each initial condition corresponding to one of the true meetings in the database. These 95 experiments were performed in order to find a fairly full set of reasonable templates, and to also determine whether the algorithm consistently settled on a small set of templates. The results were highly self-consistent, in the sense that in 98% of the total experimental runs, the algorithm converged to a template that is equivalent to, or a simple instantiation of, the one shown in Figure \[quit\_result\]. This template has a very simple and strong message, which is that the next judgment following a negative assessment is almost never a socially positive act. The converse is also true, that a socially positive act is rarely followed by a negative assessment.
To assess the correctness of this template, we note that of the 1475 times a socially positive act appears, it is adjacent to a negative assessment 141 times. Of the 991 times a negative assessment appears, 132 times a socially positive act is adjacent to it. One can contrast these numbers with the percent of time a positive assessment is adjacent to a socially positive act (68%), though there are generally more positive assessments than either socially positive acts or negative assessments, which intrinsically lowers the probability that a socially positive act would be adjacent to a negative assessment; however, even knowing this, there could be more powerful reasons why socially positive acts are associated with positive assessments rather than negative assessments.
![\[quit\_result\] The template representing the interaction of social and work related acts. Specific template instantiations and how often they occurred in the experiment are also provided.](quint_result){width="3.2in"}
From a social perspective, this result can be viewed as somewhat counterintuitive, as one might imagine wanting to encourage a colleague socially in order to compensate for a negative assessment. In practice, however, sometimes positive social acts can sound disingenuous when accompanying a negative assessment. This can be demonstrated using the following pairs of dialog from within the AMI Corpus (that are not truly side by side): 1) “But I thought it was just completely pointless.” “Superb sketch by the way.” 2) “Brilliantly done.” “It’s gonna inevitably sort of start looking like those group of sort of ugly ones that we saw stacked up.” 3) “It’d be annoying.” “Yeah, it was a pleasure working with you.” 4) “No, a wheel is better.” “All thanks to Iain for the design of that one.” In these cases a positive social act essentially becomes a negative social act.
The algorithm introduced here found a macro pattern of assessments that is prevalent and yields insight into the interaction between social and work related dialogue acts. This demonstrates that machine learning can be used to illuminate fundamental aspects of this domain. This knowledge is general, and could be used, for instance, to help classify assessments when they are not labeled, and to potentially help reconstruct missing/garbled pieces of dialogue. These are necessary elements in building any type of meetings analysis assistant. To our knowledge, the macro pattern we found has not been previously noted in the literature.
Experimental Comparison with Other Methods
------------------------------------------
In this section, we compare our results with other methods, particularly profile Hidden Markov Model (HMM) and Markov chain. We focus on comparison of results in this section, however, further in depth discussion is provided in Section \[relatedwork\_qii\].
Profile HMM is a tool to find and align related sequences, and match a new sequence to known sequences. The structure of a profile HMM has forward arrows between match states, along with delete and insertion nodes that allow any string to fit into the template. Profile HMM shares similar features with our algorithm, in that it finds a common sequence given sequences with different lengths. To create a profile HMM, there are a series of heuristic steps. One way of learning profile HMM, used to produce Table \[tab:profileHMM\], is the following. First, we choose the length of the HMM. Second, we estimate the parameters in the HMM model using pseudo-counts for the prior. Third, we find the most likely string to be emitted from each match state.
As mentioned above, the length of a profile HMM is generally specified by the user. Table \[tab:profileHMM\] shows profile HMMs with a variety of lengths specified. We were not able to recover the full pattern with profile HMM that our method produced, but we were able to see, for instance, that socially positive acts are often next to positive assessments, which is one of the instantiations of the template our method discovered.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --
**[Length specified]{} & **[Result]{}\
3&SP $\rightarrow$ AP $\rightarrow$ AP\
5&SP $\rightarrow$ AP $\rightarrow$ SP $\rightarrow$ AP $\rightarrow$ AP\
10&SP $\rightarrow$ AP $\rightarrow$ SP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP\
20&SP $\rightarrow$ AP $\rightarrow$ SP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP $\rightarrow$ AP\
****
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --
: Profile HMM (SP: Socially positive, AP: Assess positive, AN: Assess negative)\[tab:profileHMM\]
We note that a first-order Markov chain could be used to model transitions between nodes, however, since a Markov chain records all possible transition probabilities, it is not clear in general how to turn this into a template that would yield insight. On the other hand, once we provide a template like the one in Figure \[quit\_result\], it is easy to see the template within the Markov chain. The first-order Markov chain for the meetings data, learned using maximum likelihood estimation, is provided in Figure \[fig:markov\]. This learned Markov chain supports our learned template, as the four highest transition probabilities exactly represent our learned template — with high transition probabilities between socially positive and assess positive, and between assess negative and assess positive. The reason that socially negative did not appear in Figure \[quit\_result\] (as it should not) is because the number of socially negative dialogue acts is much smaller than the number of other dialogue acts (only 1%). This is an important difference between a Markov chain and our algorithm — our algorithm picks the dominant macro pattern, and has the ability to exclude a state (i.e. dialogue act) if there is no strong pattern involving that state. The Markov chain, on the other hand, is required to model transition probabilities between all states that exist within the data. We also note that it is possible for our method to choose forward arrows with very low transition probabilities in the Markov chain, so looking at only the high probability transitions will not suffice for constructing a template that minimizes our risk functional.
![\[fig:markov\] Learned Markov Chain](markov){width="2.5in"}
Question iii: How long is this meeting going to last given that the decision has already been made?\[q3\]
=========================================================================================================
Meetings sometimes last longer than expected. Even when all the decisions seem to be made, it often takes some time to work out the details and formally end the meeting. We would like to know whether it is possible to predict when the meeting is going to be over if we know when the decisions are made.
Figure \[discussion\_patch\](a) displays meetings along the horizontal axis, ordered by total meeting time. The annotated time when key decisions are all made (discussion finishing times) are indicated by blue squares, and the total meeting times are indicated by red squares for each meeting. Figure \[discussion\_patch\](b) shows the time between when key decisions are made on the x-axis, and the wrap-up time on the y-axis (this is the time spent after key decisions are made and before the meeting finishes).
With a simple piecewise linear formula, one can predict with relatively high accuracy what the wrap-up time will be, given the time at which the key decisions are finished being made. Denoting the time to complete key decisions by $x$, the estimated wrap-up time is as follows: If $x\leq$ about 14 minutes, the wrap-up time is about $0.923 + 4.78$ minutes. If $x>$ about 14 minutes, the wrap-up time is $-0.47x + 24.53$ minutes. There are some interesting implications: if the meeting was efficient enough to make all the decisions within 14 minutes, then the team will also be efficient in wrapping up. Once the meeting has gone about 14 minutes without all decisions made, then people also tend to spend more time wrapping up. If the meetings runs very long without decisions being made, then once the decisions are made, the meeting tends to wrap up quickly.
These results are specific to meetings whose length is less than an hour as in the AMI corpus, which, according to @romano2001meeting is true 26% of the time; it would be interesting for future work to see if a piecewise linear model works well for meetings of other lengths, though this hypothesis cannot currently be verified by any database that we are aware of.
Question iv: Do persuasive words exist?\[q4\]
=============================================
A “good” meeting might be one where we have contributed to the team effort. We always want to suggest good ideas and want the team members to accept those ideas. Numerous articles claim that how we package our ideas, and our choice of words, is sometimes as important as the idea itself [@persuasivewords1; @persuasivewords2]. We are interested in understanding this hidden factor in the team’s decision making process. Are there patterns in suggestions that are accepted versus rejected? Can we use this to improve how we present ideas to the team?
To select data for this section, we chose a bag-of-words representation for each “suggestion” dialogue act. We gathered a set of all words that occurred in all suggestions, excluding stop words, leading to a 1,839 dimensional binary vector representation for each suggestion. The labels were determined by the annotations, where accepted suggestions received a $+1$ label, and rejected suggestions received a $-1$ label.
Are published sets of persuasive words really persuasive?\[sec:empiricalPersuasive\]
------------------------------------------------------------------------------------
We would like to know whether the persuasive words of @persuasivewords1 [@persuasivewords2] are truly persuasive. These words are not domain-specific, meaning that they do not depend on the topic of the meeting, and could thus be more generally useful. Note that the data cannot tell us (without a controlled experiment) whether there is a truly causal relationship between using persuasive words and having a proposal accepted; however, we can study correlations, which provide evidence. An important first question is whether the proportion of persuasive words in accepted suggestions differ significantly from the proportion of persuasive words in rejected suggestions. We marked each suggestion as to whether or not one of the words in @persuasivewords1 [@persuasivewords2] appears. We also mark each suggestion as to whether it is an accepted or rejected suggestion. Of the 139 times that a suggestion contained persuasive words from @persuasivewords1 [@persuasivewords2], 134 of these appearances were within accepted suggestions (96% of appearances). Of the 2,185 times a suggestion did not contain any of the words from @persuasivewords1 [@persuasivewords2], 1,981 of these appearances were within accepted suggestions (90% of appearances). Using Fisher’s exact test, the difference in proportions is significant at the 0.01 level (pvalue 0.0037); it appears that persuasive words do appear more often in accepted suggestions than other words do.
On the other hand, we tested each persuasive word individually, to see whether the number of accepted suggestions it appears in is significantly different than the proportion of rejected suggestions it appears in. The difference between the two proportions was not significant at the 0.05 level for any of the persuasive words, again using Fisher’s exact test for differences between proportions. Perhaps this is because each word individually is rare, while the collection is not. This begs the question as to whether there are words that themselves show a significant difference in the proportion of accepted suggestions they appear in. A related question is which words are the most important, in terms of predictive modeling, for distinguishing accepted and rejected suggestions. We will answer both of these questions below, the latter answered first.
Using SVM coefficients to tell us about persuasive words\[svm\_feature\_ranking\]
---------------------------------------------------------------------------------
In this section, we try to discover persuasive words from data alone, and afterwards compare to the published lists of persuasive words. Specifically, we want to know whether we can predict if a suggestion would be accepted or rejected solely based on words that are used, not based on the idea within it. A standard way to do this is to apply a linear SVM with cross-validation for the tradeoff parameter, and examine the values of its coefficients; this provides a ranking of features, and a rough measure of how important each word is to the overall classifier [e.g., see @Guyon02]. The set of 2,324 suggestions with 1,839 features (one feature per unique word) was separated into 5 folds using each in turn as the test set, and the SVM accuracy was 83% $\pm$ 2.1%. Note that the SVM has no prior knowledge of what persuasive words are. Persuasive words identified by the SVM are considered to be words with large absolute coefficient value (normalized) over all 5 folds. These include “things,” “start,” “meeting,” “people,” “yeah.” Many of the persuasive words concerned the topic of marketing (“market,” “presentation,” “gimmick,” “logo”) and non-persuasive words included “buttons,” “speech,” “LCD,” “recognition,” and more often words specific to the topic of creating a remote control (“scroll,” “green”).
The next question is whether these learned persuasive words make sense. To answer this, we checked whether the persuasive words from @persuasivewords1 [@persuasivewords2] had positive or negative SVM coefficients. The overlap between our set of 1,839 words and the persuasive words from @persuasivewords1 [@persuasivewords2] is 29 words. Their average coefficient values over 5 folds are shown in Table \[tab:listofpersWordsCompare\]. The finding is consistent with the Fisher’s exact test result in Section \[sec:empiricalPersuasive\] — not a single word has a large positive coefficient and all words have relatively large standard deviations, showing lack of their individual connection to persuasiveness.
--------------------------------------------------------------------------------------- -- -- --
**[ Words ]{}& **[$\lambda$ $\pm$ Std.]{} & **[Words]{} & **[$\lambda$ $\pm$ Std.]{}\
strength & 0.006 $\pm$ 0.0026 & inspiration & -0.0002 $\pm$ 0.003\
free & 0.0054 $\pm$ 0.0031 & drive & -0.0002 $\pm$ 0.0011\
good & 0.0048 $\pm$ 0.0015 & easy & -0.0004 $\pm$ 0.003\
power & 0.0043 $\pm$ 0.0032 & health & -0.0005 $\pm$ 0.0022\
avoid & 0.0039 $\pm$ 0.0011 & creativity & -0.0016 $\pm$ 0.0021\
offer & 0.0036 $\pm$ 0.0035 & guarantee & -0.0017 $\pm$ 0.001\
save & 0.003 $\pm$ 0.0036 & explore & -0.0017 $\pm$ 0.001\
safe & 0.0028 $\pm$ 0.0015 & safety & -0.0017 $\pm$ 0.001\
energy & 0.0024 $\pm$ 0.0032 & reinvent & -0.0017 $\pm$ 0.0009\
imagine & 0.0024 $\pm$ 0.0016 & approach & -0.0019 $\pm$ 0.003\
important & 0.0023 $\pm$ 0.002 & money & -0.0025 $\pm$ 0.0035\
confidence & 0.0019 $\pm$ 0.0027 & purpose & -0.0027 $\pm$ 0.0008\
wanted & 0.0015 $\pm$ 0.0009 & &\
quick & 0.0015 $\pm$ 0.0009 & &\
memory & 0.0015 $\pm$ 0.0009 & &\
life & 0.0006 $\pm$ 0.0029 & &\
hurry & 0.0005 $\pm$ 0.0029 & &\
********
--------------------------------------------------------------------------------------- -- -- --
: List of persuasive words on which SVM and the articles [@persuasivewords1; @persuasivewords2] agree/disagree \[tab:listofpersWordsCompare\]
In what follows we will explore a different method for feature ranking, where the space is reduced to only words that occur in significantly different proportions between accepted and rejected suggestions before applying an SVM.
Which words are individually persuasive?
----------------------------------------
We have 2,324 suggestions in an 1,839 dimensional space. Fisher’s exact test is based on the hypergeometric distribution and has the benefit that it can be used for features that are rarely present. It considers whether the suggestions containing the feature have significantly different proportions of acceptance and rejection. The selected persuasive words are shown in Table \[tab:persWordsFishers\] with their associated pvalues, where the most significant words are at the top of the list. We can also find non-persuasive words that are significant according to Fisher’s exact test. Together with the persuasive words, this gives us a set of important features to use as a reduced feature space for machine learning. Some of the non-persuasive words include “recognition”, “speech”, “fair”, “selecting”, “flat”, “animals”, “middle”, and “bottom”. The total number of features we used is 244 (where persuasive words and non-persuasive words were selected under the same significance threshold). After this feature reduction, we applied SVM and achieved an accuracy of 87.2% $\pm$ 0.010%, which is higher than before. The SVM coefficients for all of the words that we identified as being persuasive are positive, as shown also in Table \[tab:persWordsFishers\]. These words are thus persuasive both individually and together: they are each individually significant in predicting accepted suggestions, and they have positive predictive coefficients.
--------------------------------------------------------------------------------------------------------------------------------------------- -- -- -- --
**[Words]{}& **[Pvalues]{}& **[Ratio of accepted when appears]{}& **[Ratio of accepted when not appears]{}& **[SVM $\lambda$ $\pm$ Std.]{}\
yeah & 0.012 & 1 (46/46) & 0.90 (2069/2278) & 0.011 $\pm$ 0.0021\
give & 0.020 & 1 (41/41) & 0.90 (2074/2283) & 0.010 $\pm$ 0.0015\
menu & 0.027 & 1 (38/38) & 0.90 (2077/2286) &0.010 $\pm$ 0.0045\
start & 0.043 & 1 (33/33) & 0.90 (2082/2291) &0.012 $\pm$ 0.0020\
meeting & 0.052 & 1 (31/31) & 0.90 (2084/2293) & 0.012 $\pm$ 0.0019\
touch & 0.085 & 1 (26/26) & 0.90 (2089/2298) & 0.0081 $\pm$ 0.00059\
discuss & 0.093 & 1 (25/25) & 0.90 (2090/2299) & 0.0092 $\pm$ 0.0024\
find & 0.093 & 1 (25/25) & 0.90 (2090/2299) & 0.0098 $\pm$ 0.0017\
market & 0.12 & 1 (22/22) & 0.90 (2093/2302) & 0.0094 $\pm$ 0.0023\
yellow & 0.12 & 1 (22/22) & 0.90 (2093/2302) & 0.0076 $\pm$ 0.0018\
work & 0.12 & 1 (22/22) & 0.90 (2093/2302) & 0.0097 $\pm$ 0.0031\
good & 0.13 & 0.97 (37/38) & 0.90 (2078/2286) &0.0029 $\pm$ 0.0035\
fruit & 0.13 & 1 (21/21) & 0.90 (2094/2303) & 0.0065 $\pm$ 0.0032\
logo & 0.15 & 1 (20/20) & 0.90 (2095/2304) & 0.0073 $\pm$ 0.0019\
people & 0.15 & 0.97 (35/36) & 0.90 (2080/2288) & 0.012 $\pm$ 0.0024\
side & 0.16 & 0.97 (34/35) & 0.90 (2081/2289) & 0.0056 $\pm$ 0.0039\
number & 0.16 & 1 (19/19) & 0.90 (2096/2305) & 0.0059 $\pm$ 0.0027\
presentation & 0.18 & 1 (18/18) & 0.90 (2097/2306) &0.0084 $\pm$ 0.0020\
things & 0.19 & 0.95 (45/47) & 0.90 (2070/2277) & 0.012 $\pm$ 0.0026\
chip & 0.20 & 1 (17/17) & 0.90 (2098/2307) & 0.0066 $\pm$ 0.0026\
stick & 0.22 & 1 (16/16) & 0.90 (2099/2308) & 0.0070 $\pm$ 0.0017\
gonna & 0.22 & 0.95 (42/44) & 0.90 (2073/2280) & 0.000063 $\pm$ 0.0057\
information & 0.24 & 1 (15/15) & 0.90 (2100/2309) & 0.0059 $\pm$ 0.0020\
talk & 0.24 & 1 (15/15) & 0.90 (2100/2309) &0.0034 $\pm$ 0.0027\
**********
--------------------------------------------------------------------------------------------------------------------------------------------- -- -- -- --
: List of persuasive words from Fisher’s exact test (higher ranking for smaller pvalues). The second column contains the pvalue from Fisher’s exact test. The third column contains the ratio of accepted proposals when the word appeared. The fourth column contains the ratio of accepted proposals when the word did not appear. The last column contains the SVM coefficients.\[tab:persWordsFishers\]
Studying the most persuasive words from Fisher’s exact test, we find that many of these words are *not* specifically tied to the topic of the meeting (designing a remote control), but seem to be more generally persuasive. In fact, we can make informed hypotheses about *why* these words are persuasive. Some of these observations help us to understand more generally how language is used during meetings. Let us consider the most significant persuasive words as follows:
- *Yeah:* Dialogue segments where the word “yeah” is used include: “or yeah, maybe even just a limited multi-colour so it doesn’t look too childish,” “yeah, if you had one of those, just coming back to your other point about pressing the button and setting off the bleeper in the room,” “Yeah if you are holding it in your hand you could do that.” Judging from these and similar dialogue segments, our hypothesis is that framing a suggestion as an agreement with a previous suggestion increases its chances of being accepted. That is, if the idea comes across as if it were in line with previous thoughts by others, the suggestion has a higher chance of being accepted. This applies either when attributing the full idea to others, or just the line of thought. The case where one attributes their full idea to others in order to increase its chances of acceptance has been considered in popular books [@carnegie2010win].
- *Give:* “Give” is used in at least three ways, the first one occurring the most often: (i) giving with respect to the topic of the meeting, which here is either the customer or the product (“so if you want to give the full freedom to the user,” “You give it the full functions in here,” “We can give them smooth keys”), (ii) giving to the meeting participants (“would give us a little bit of a marketing niche”), and (iii) to indicate that suggestions are based on previous data or knowledge (“given these parameters that we’re just gonna sort of have this kind of uh non-remote remote”, “given speech recognition I think you should go for the less fancy chip”).
- *Menu:* This word seems to be tied to the topic of remote controls (“Um and one other suggestions I’d make is to in is to include in a menu system”), without a general hypothesis for other types of meetings.
- *Start:* Our hypothesis is that the word “start” gives group members the opportunity to agree, where agreement of basic suggestions provides an indication that group members want to be productive during the meeting; e.g., “Shouldn’t we start with the most important parts?” “I will start by the basic one.” This type of agreement may help with alliance building early on in the meeting.
- *Meeting:* The word “meeting” often appears in suggestions about what *not* to discuss: “Or maybe this is something for the next meeting,” “I figure we could get back to it on the next meeting actually,” “We take it to the other meeting.” Our hypothesis is that suggesting that a topic belong to a later meeting may be a way to gently change the topic or move the current meeting along. It can be used instead of a negative assessment of a previous suggestion.
- *Touch:* This seems to be tied to the topic of remote controls, “So we put a touch pad on it,” “We can uh do a touch-pad on our remote.”
- *Discuss:* This word appears mainly in an organizational context for the meeting: “And then we can discuss some more closely,” “I think we shouldn’t discuss any points points that long,” “Maybe we should centralise the discussion here.” It seems that often, people tend to agree with organizational suggestions about the meeting.
- *Find:* Our hypothesis is that “find” is often used in suggestions to gather more information or do more work, and these are suggestions that are often accepted: “we have to find out if it’s possible,” “um and I’m sure we can find more goals for the product we are going to develop,” “but just try to find out what they’re willing to pay for it.”
Having a collection of persuasive words can be immediately useful, assuming that there is a causal relationship between persuasive words and accepted proposals, rather than only a correlation. The hypothesis tested in this work (that persuasive words truly exist) can also be tested for this causal relationship, and it would be very interesting to create a dataset for this purpose. If the hypothesis does hold, it has the potential to allow ideas to be communicated more clearly, and thus to make meetings more efficient overall. Already we have gained some insight for how specific words are used within meetings, and why suggestions containing these words are more likely to be accepted.
Related Work\[related\_work\]
=============================
The closest work to ours is that of the CALO (Cognitive Assistant that Learns and Organizes) project [@calo], which takes a *bottom-up* approach to designing a cognitive assistant that can reason and learn from users using machine learning techniques. Although the CALO project’s focus (e.g., speech recognition, sentence segmentation, dialogue act tagging, topic segmentation, action item detection and decision extraction) does not intersect with our work, it is worthwhile to note that this multi-year project has made a definite step towards improving meetings using machine learning techniques. However, as pointed out by @tur2008calo, a number of challenges (e.g. extracting task descriptions) still exist before this tool can be integrated into daily life. The insights into meetings obtained through our work could be used as part of a *top-down* approach to design such a tool.
We believe our work is the first to take a truly data-driven approach to finding persuasive words; this is the first work to try to prove that a word is persuasive using data.
#### Qualitative Work:
Research on group decision making processes traditionally appears in *qualitative* studies. Works along these lines attempt to understand and model the course of agreement [@Black48], design features to track the course of negotiation [@Eugenio99theagreement], develop theories for group decision making [@davis1973group] and study different group decision-making schemes [@green1980effects].
#### Related Work for Question i:
A few quantitative approaches have attempted to detect when decisions are made. These methods use maximum entropy [@hsueh2008automatic], linear SVMs [@fernandez2008modelling] and directed graphical models [@bui2009extracting]. A common factor of these works is that they require content information (audio and video data; prosodic features, position within the meeting, length in words and duration etc.) that can potentially contain sensitive information about what is being planned. Our approach requires only dialogue acts (i.e. the actual meeting content is not required) to achieve the same goal with relatively high accuracy. One benefit of using only dialogue acts is that the algorithm allows our results to hold independently of the specific type and purpose of the meeting being held and can be used in situations where the meeting contains potentially sensitive information. It is also much simpler logistically to require collecting only dialogue acts.
#### Related Work for Question ii: \[relatedwork\_qii\]
To the best of our knowledge, our work is the first to apply learning techniques to learn dynamical interactions between social and work aspects in meetings. As discussed earlier, one could consider the method we developed generally for learning a template from a sequence of data. Our method has the following characteristics: 1) The loss function considers meetings instantiated from the same template, but possibly with different lengths, to be equally good. For instance: if a template is ‘A’ ‘B’ ‘C’ with a back arrow from ‘C’ to ‘A’, then ‘ABCABC’, ‘BCABCA’, ‘CAB’ are equally good instantiations from this template. 2) An instantiation of the template can start at any position, as demonstrated by the ‘BCABCA’ and ‘CAB’ instances in the template above. 3) By optimizing the loss function we provide in Section \[q2\], the algorithm is able to uncover a template of exactly the desired form, despite the existence of noise. Let us compare this method to other graph learning techniques within related subfields.
*Petri nets* are used (and were designed) to model concurrency and synchronization in distributed systems. In learning Petri nets from data, and generally in learning workflow graphs and workflow nets (see [@van1998application]), all possible transitions in the data must be accounted for within the learned net (e.g., [@agrawal1998mining]), which is the opposite of what we would like to do, which is to create instead a more concise representation that is not necessarily all-inclusive. The kind of templates we discover should be much simpler than Petri nets - for us, there is only sequential and iterative routing, and no parallel routing (no concurrency), there are no “tokens” and thus there is no underlying workflow state, and there are no conditions involved in moving from one place to the next (no conditional routing).
One major difference between our approach to learning macro-patterns and learning probabilistic graphical models such as Markov models is that our approach is deterministic (not probabilistic). We do not model the probabilities of transitions between states, as these probabilities are not of fundamental interest for building the template (but could be calculated afterwards). As in other deterministic methods (e.g., SVM), we aim to directly optimize the quality measure that the macro-pattern will be judged by on the data, in our case involving edit distance. This allows us to handle noisy data, that is, meetings that do not fit precisely into the pattern, without having to include additional nodes that complicate the template. We do not then need a graph that handles all possible transitions and their probabilities (as is required in Markov models or Petri nets). It is certainly possible to model a macro-pattern as a Markov chain, as we demonstrated, but the resulting transition matrix would most likely provide little insight. The full Markov transition matrix would be of size $|\Lambda| \times |\Lambda|$, but distilling this to a compact one dimensional graph with mostly forward arrows would then be an additional task.
Hidden Markov Models aim to infer unobserved states, where in our case, there are no natural unobserved states. Although we could artificially create hidden states, it is more natural to directly model the observed sequence. One exception to this is profile HMM’s, where hidden states are either “match” states, “insert” states, or “delete” states [@eddy1998profile]. Methods for fitting profile HMM’s generally do not create backwards edges, and thus cannot easily accommodate substrings being repeated arbitrarily within the template. Profile HMM’s are designed for accuracy in aligning sequences, but they are not generally designed for conciseness or to minimize, for instance, a count of backwards edges. Methods for learning profile HMMs generally require the user to specify the length of the HMM, whereas we purposely do not do this.
Profile HMM’s generally have a fixed initial state, whereas our method can start anywhere within the template. The price of a fixed initial state and fixed length with no backwards edges is high; for instance for a template ‘A’ ‘B’ ‘C’ with a backwards arrow from ‘C’ to ‘A,’ we can equally well accommodate patterns ‘ABCABCABC’ and ‘BCABCA,’ as well as ‘BC’, all of them being perfect matches to our template. Profile HMM’s, with any fixed initial state, and with no backwards arrows to allow repeats, would require insertions and deletions for each of these patterns, and have difficulty viewing all of these patterns equally. Note that in general, left-to-right HMM’s cannot have backwards loops as our templates do.
One might also think of automata ([@narendra1974learning; @wikiAutomata]) as a way to model meetings. HMM’s are equivalent to probabilistic automata with no final probabilities (see [@dupont2005links]), where above we discussed how our goals do not generally involve hidden states. There are additional characteristics of automata that contrast with our efforts. For instance, probabilistic automata generally have a set of initial states and accepting or final (terminating) states, whereas our meetings can begin an end anywhere in the template, and there is no notion of acceptance or rejection of a meeting to the template. Further, an important characteristic of Probabilistic Deterministic Finite Automata (PDFA) is that for any given string, at most a single path generating the string exists in the automaton, whereas in our work, we do not require this uniqueness ([@guttman2006probabilistic]).
Probabilistic Suffix Automata (PSA) [@ron1996power] are probabilistic automata with variable memory length, that aim to learn an order-$K$ Markov model with $K$ varying in different parts of state space. As with order-1 Markov chains discussed above, is not clear how to construct a concise template of the form we are considering. On the other hand, it would be an interesting extension of our work to keep track of higher order patterns within the template like PSA does.
One could think of our goal in Section \[q2\] as solving a type of case of the consensus string problem [@sim2003consensus; @lyngso2002consensus] with consensus error, but with two major changes: (i) using edit distance between meetings and template instantiations as a metric, rather than between pairs of strings, which allows backwards loops, (ii) encouraging conciseness in the template. Problems encountered when not including these aspects were provided for the ‘ABC’ example above.
The goal of learning metro maps [@shahaf2012trains] is very different than ours, since their goal is to learn a set of special paths through a graph (metro lines) that have specific properties. Temporal LDA [@wang2012tm] is also very different than our work: its goal is to predict the next topic within a stream, whereas our goal is to find a concise representation of a set of strings.
#### Related Work for Question iv:
Identifying characteristics of persuasive speech and discourse has been of great interest in the qualitative research community [@sternthal1978persuasive; @scheidel1967persuasive]. However, there has not been much quantitative research done on this topic. @guerini2008trusting studied the relationship between the choice of words and the reaction they elicit in political speeches based on numerical statistics, including counts of words in each document and how common the words are across all documents (tf-idf). To the best of our knowledge, our work is the first work to apply machine learning techniques to learn persuasive words from free-form conversational meeting data, to rank persuasive words, and to compare resulting persuasive words with qualitative studies to gain insights.
#### Other meeting related work:
Other studies that apply machine learning techniques for meeting related topics (but not specifically related to any of the work in this paper) include meeting summarization [@purver2007detecting], topic segmentation [@galley2003discourse], agreement/disagreement detection [@Hahn06; @Hillard03detectionof; @bousmalis2009spotting], detection of the state of a meeting (discussion, presentation and briefing) [@banerjee2004using; @reiter2005multimodal], and prediction of the roles of meeting participants [@banerjee2004using]. In all of this work, machine learning techniques are used as tools to classify a particular aspect of a meeting with a specific applications in mind. In addition to addressing different aspects of meetings, our work uses machine learning techniques as a way to to learn and study meetings scientifically in an attempt to bridge the gap between qualitative studies on understanding meetings and quantitative application-focused studies.
Conclusion\[conclusion\]
========================
The scientific field of meeting analysis is still in beginning stages; meetings have not yet been well characterized, and this is one of the first works in this new scientific arena. Several hypotheses made in this work *cannot* be tested further with any current available dataset. We hope that by illuminating the potential of fully solving these problems, it will inspire the creation of new meetings corpora. Elaborating further, if we are able to automatically detect when key decisions are made, this could translate directly into a software tool that managers could use to determine when they should join an ongoing meeting of their staff without attending the full meeting. If we know common patterns of dialogue, this might help us to understand social cues better for business settings, and could potentially help reconstruct parts of dialogue that might not have been recorded properly. If we can automatically detect when a meeting’s key decisions are made, and can accurately gauge the meeting wrap-up time, it can give us something immediately valuable, namely an estimated time for the end of the current meeting, which staff can use to plan ahead for the start of the next meeting, or to plan transportation, in a fast-paced corporate culture. If we truly knew which words were persuasive, we could use these words to help convey our ideas in the most favorable light. Before we can do all of this, however, we need to understand the science behind meetings, and make hypotheses that can be tested, which is the goal of this work.
Appendix {#appendix .unnumbered}
========
Proof of Theorem 1. {#proof-of-theorem-1. .unnumbered}
-------------------
We will use Hoeffding’s inequality combined with the union bound to create a uniform generalization bound over all viable templates. The main step in doing this is to count the number of possible viable templates. Let us do this now.
Let $\Lambda$ be the set of possible dialogue acts, and denote $|\Lambda|$ as the number of elements in the set. We will calculate the number of templates that are of size less than or equal to $L$, which is the size of our function class in statistical learning theory.
For a template of exactly length $n$, there are $|\Lambda|^n$ possible assignments of dialogue acts for the templates. Also, for a template of length $n$, there are at most ${n(n-1)/2 \choose B}$ possible assignments of $B$ backward arrows, where $B \leq n$. To see this, consider the set of backwards arrows as represented by an $n\times n$ adjacency matrix, where only the part below the diagonal could be 1. There are $n^2$ total elements in the matrix, $n$ on the diagonal, so $n(n-1)$ off diagonal elements, and $n(n-1)/2$ elements in the lower triangle. If exactly $b$ of these can be 1, the total number of possibilities is at most ${n(n-1)/2 \choose b}$. There can be up to $B$ backward arrows, so the total number of possibilities is at most $$\sum_{b=0}^{\min(B,n)} {n(n-1)/2 \choose b}.$$ Note that this number is an upper bound, as usually we cannot have more than one backwards arrow leaving or entering a node. Finally we could have $n$ anywhere between 0 and $L$ so the final number of possible templates has upper bound: $$\sum_{n=0}^{L} |\Lambda|^n \sum_{b=0}^{\min(B,n)} {n(n-1)/2 \choose b}.$$ Hoeffding’s inequality applies to arbitrary bounded loss functions. This, combined with a union bound over all viable templates, yields the statement of the theorem.
[^1]: More details of definition of dialogue acts can be found in [@Mccowan05theami].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Given a two-dimensional space endowed with a divergence function that is convex in the first argument, continuously differentiable in the second, and satisfies suitable regularity conditions at Voronoi vertices, we show that orphan-freedom (the absence of disconnected Voronoi regions) is sufficient to ensure that Voronoi edges and vertices are also connected, and that the dual is a simple planar graph. We then prove that the straight-edge dual of an orphan-free Voronoi diagram (with sites as the first argument of the divergence) is always an embedded triangulation.
Among the divergences covered by our proofs are Bregman divergences, anisotropic divergences, as well as all distances derived from strictly convex $\mathcal{C}^1$ norms (including the $L_p$ norms with $1< p < \infty$). While Bregman diagrams of the [first kind]{} are simply affine diagrams, and their duals ([weighted]{} Delaunay triangulations) are always embedded, we show that duals of orphan-free Bregman diagrams of the *second kind* are always embedded.
author:
- |
Guillermo D. Canas\
Massachusetts Institute of Technology\
[email protected]
- |
Steven J. Gortler\
Harvard University\
[email protected]
bibliography:
- 'vddw8.bib'
title: 'On the Embeddability of Delaunay Triangulations in Anisotropic, Normed, and Bregman Spaces'
---
Introduction
============
Voronoi diagrams and their dual Delaunay triangulations are fundamental constructions with numerous associated guarantees, and extensive application in practice (for a thorough review consult [@Aurenhammer13] and references therein). At their heart is the use of a distance between points, which in the original version is taken to be Euclidean. This suggests that, by considering distances other than Euclidean, it may be possible to obtain variants which can be well-suited to a wider range of applications.
Attempts in this direction have been met with some success. Power diagrams [@power] generalize Euclidean distance by associating a [bias-term]{} to each site. The duals of these diagrams are guaranteed to be embedded triangulations, in any number of dimensions. Although this is a strict generalization of Euclidean distance, it is a somewhat limited one. The effect of the bias term is to locally enlarge or shrink the region associated to each site, loosely-speaking “equally in every direction". It allows some freedom in choosing local scale, with no preference for specific directions.
Two related, and relatively recent generalizations of Voronoi diagrams and Delaunay triangulations have been proposed, independently, by Labelle and Shewchuk [@LS], and Du and Wang [@DW]. Although their associated anisotropic Voronoi diagrams are, in general, no longer orphan-free (i.e. they may have disconnected Voronoi regions), Labelle and Shewchuk show that a set of sites exists with an orphan-free diagram, whose dual is embedded, in two dimensions. They accomplish this by proposing an iterative site-insertion algorithm that, for any given metric, constructs one such set of sites. Note that this is a property of the output of the algorithm, and not a general condition for obtaining embedded triangulations.
The recent work of [@Bregman] discusses Voronoi diagrams and their duals with respect to Bregman divergences. They show that Bregman Voronoi diagrams of the *first kind* are simply power diagrams, whose duals are known to always be embedded [@powerdiag]. Bregman diagrams of the *second kind* are power diagrams in the dual (gradient) space, but, prior to this work, no results for them were available in the primal space.
In this paper we discuss properties of Voronoi diagrams and Delaunay triangulations for a general class of divergences, including Bregman, quadratic, and all distances derived from strictly convex $\mathcal{C}^1$ norms. We show that, given a divergence $D$ that is convex in the first argument and continuously differentiable in the second, and under a *bounded anisotropy* assumption on the divergence, if a set of sites produces an orphan-free Voronoi diagram with respect to $D$, then its dual is always an embedded triangulation (or an embedded polygonal mesh with convex faces in general), in two dimensions (theorem \[th:main\]). This effectively states that, regardless of the sites’ positions, if the primal is well-behaved, then the dual is also well-behaved. Further, in a way that parallels the ordinary Delaunay case, the dual has no degenerate elements (proposition \[prop:ECB\]), its elements (vertices, edges, faces) are unique (Cor. \[cor:VorI\]), and the dual is guaranteed to cover the convex hull of the sites (theorem \[th:main\]).
Voronoi diagrams with respect to divergences {#sec:setup}
============================================
The class of divergences that we consider in this work are non-negative functions $D:\mathbb{R}^2\times\mathbb{R}^2\rightarrow\mathbb{R}$ which are strictly convex in the first argument and continuously differentiable in the second, and such that ${D\left(x \parallel x\right)}=0$ for all $x\in\mathbb{R}^2$. Following [@Bregman], we let $$\label{eq:defball}
B_1(p; \rho) = \{v\in\mathbb{R}^2 : {D\left(v \parallel p\right)} \le \rho\}, \quad\quad B_2(p;\rho) = \{v\in\mathbb{R}^2 : {D\left(p \parallel v\right)} \le \rho\}$$ be, respectively, balls of the *first* and *second kind*, centered at $p\in\mathbb{R}^2$ of radius $\rho$. Note that balls of the first kind are necessarily convex since ${D\left(\cdot \parallel p\right)}$ is convex. We also assume that $D$ satisfies what we term a *bounded anisotropy* condition, defined in assumption \[ass:BAA\] below.
Given a set ${S}=\{s_1,\dots,s_n\}\subset\mathbb{R}^2$ of $n$ distinct sites on the plane, and a divergence $D:\mathbb{R}^2\times\mathbb{R}^2\rightarrow\mathbb{R}$, the Voronoi regions of the first and second kinds [@Bregman] are: $$\begin{aligned}
\label{eq:defvor1} {{\text{Vor}}^1_i} = \{p\in\mathbb{R}^2 : {D\left(p \parallel s_i\right)} \le {D\left(p \parallel s\right)}, \forall s\in {S}\}, \\
\label{eq:defvor2} {{\text{Vor}}^2_i} = \{p\in\mathbb{R}^2 : {D\left(s_i \parallel p\right)} \le {D\left(s \parallel p\right)}, \forall s\in {S}\}, \end{aligned}$$ respectively, and are indexed by the site its points are closest to. Of course, the two kinds of Voronoi diagrams are different because $D$ is in general not symmetric. In the sequel, and whenever not otherwise specified, we will assume that balls are of the *first kind* (convex), and Voronoi diagrams, and their dual Delaunay triangulations are of the *second kind*. For instance, we will use the convexity of balls (of the first kind) to prove that every face in a Delaunay triangulation (of the second kind) satisfies an *Empty Circum-Ball* property (proposition \[prop:ECB\]) that parallels the empty circumcircle property of Euclidean Delaunay triangulations.
Consider the following definition of Voronoi element:
\[def:VorI\] For each subset $I\subseteq\{1,\dots,n\}$, the set ${\text{Vor}}_I=\cap_{i\in I}{\text{Vor}}^2_i \setminus \cup_{j\not\in I}{\text{Vor}}^2_j$ is a Voronoi element of order $|I|$. Elements of orders $1$, $2$, and $|I|\ge 3$ are denoted regions, edges, and vertices, respectively.
The set of all Voronoi elements ${\text{Vor}}_I$ forms a partition of the plane.
----------------------------------------------- -- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
${D\left(\cdot \parallel \cdot\right)}$ A non-negative divergence strictly convex in its first argument and continuously differentiable in the second.
$D_F(\cdot \parallel \cdot)$ Bregman divergence (section \[sec:DF\]).
$D_f(\cdot \parallel \cdot)$ Csiszár divergence (section \[sec:Df\]).
$D_Q(\cdot \parallel \cdot)$ Quadratic divergence (seciton \[sec:DQ\]).
$\gamma$ Global lower bound on the ratio of eigenvalues of metric $Q$ (quadratic divergence, lemma \[lem:DQgamma\]) or of the Hessian of $F$ (Bregman divergence, lemma \[lem:DFgamma\]).
${S}= \{s_1,\dots,s_n\}$ Set of $n$ sites.
$L_{ij}$ The supporting line of sites $s_i$, $s_j$.
${{conv}\left\{{S}\right\}}$ Convex hull of ${S}$.
$W=\{w_i\in {S}: i=1,\dots,m\}$ Subset of sites on the boundary of ${{conv}\left\{{S}\right\}}$, in clock-wise order.
$B(\cdot\, ;\cdot)$ Convex ball of the first kind (equation ).
$\theta_p(v)$ The ball (of the first kind) $B(v;{D\left(p \parallel v\right)})$ centered at $v$ with $p$ in its boundary.
${\text{Vor}}_i$ Voronoi region of the second kind corresponding to site $s_i$ (equation ).
${\text{Vor}}_I$ Voronoi element of order $|I|=1$ (Voronoi region), $|I|=2$ (Voronoi edge), or $|I|\ge 3$ (Voronoi vertex).
$G=({S},E,F)$ The straight-edge dual triangulation with vertices at the sites.
$B$ The edges in the topological boundary of $G$ (incident to one face).
$\mathcal{B}=(w_i,w_{i\oplus 1})_{i=1}^{|W|}$ The edges in the boundary of ${{conv}\left\{{S}\right\}}$.
$\pi$ Projection from $C(\sigma)$ onto $\partial{{conv}\left\{{S}\right\}}$ (section \[sec:boundary\]).
$\nu_\sigma$ Projection function onto a circle of radius $\sigma$ (section \[sec:boundary\]).
$H^{+}_{ij}, H^{-}_{ij}$ The half-spaces on either side of $L_{ij}$, chosen so $H^{+}_{ij}\cap{S}=\phi$ (fig. \[fig:pinu\]).
$C(\sigma)$ The origin-centered circle of radius $\sigma$ (with respect to the natural metric).
----------------------------------------------- -- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
: Notation[]{data-label="table:notation"}
The following “bounded anisotropy" condition is assumed to hold. It is written in its most general (but very technical) form below, but it becomes much simpler in particular cases, as shown in Section \[sec:summary\]. Typically, it can be rewritten as a simple regularity condition on a symmetric positive definite matrix, such that its ratio of minimum to maximum eigenvalues (a measure of anisotropy) is globally bounded away from zero.
![The *bounded anisotropy asumption* ensures that balls of the first kind are globally well-behaved. []{data-label="fig:gamma"}](boundedanisotropy.pdf){width="2.1in"}
\[ass:BAA\] For every two points $p,q\in\mathbb{R}^2$ with supporting line $L_{pq}$, and every point $r\not\in L_{pq}$, there is a sufficiently large value $\mu > 0$ such that for every point $c\in\mathbb{R}^2$ lying on the same side of $L_{pq}$ as $r$, such that $\|c\| > \mu$, and whose closest point $m$ in $L_{pq}$ lies in the segment $\overline{pq}$, it is ${D\left(r \parallel c\right)} < {D\left(m \parallel c\right)}$.
Note that the condition $\|c\|<\mu$ depends on the (arbitrary) choice of origin. Assumption \[ass:BAA\] is, however, independent of this choice.
Loosely speaking, this condition ensures that balls of the first kind are not just convex, but also “sufficiently round". For instance, it is satisfied by all the $L_p$ distances with $1<p<\infty$, but not for $p=1,\infty$, since (aside from not being strictly convex) the corresponding balls have “kinks".
\[ass:EGA\] For each Voronoi vertex ${\text{Vor}}\{i_1,\dots,i_m\}$ with $m\ge 3$, the gradients $g_j(p)\equiv\nabla_p {D\left(s_{i_j} \parallel p\right)}, j=1,\dots,m$, at $p\in{\text{Vor}}_I$ are distinct and extremal, i.e. they are vertices of the convex hull: ${\text{conv} \{ }g_1(p),\dots,g_m(p)\}$.
In the “typical" case that $m=3$, the above simply means that $g_1,g_2,g_3$ are not colinear. Given two distinct gradients $g_1\ne g_2$, requiring $g_3$ not to be colinear only constraints it to be outside a line. If $D$ is the $L_p$ distance (or any other non-spatially-varying divergence), the extremal gradient assumption can be shown to be always automatically satisfied at Voronoi vertices. Finally, the extremal gradient assumption will be shown to imply that Voronoi vertices are composed of isolated points, and therefore, when satisfied, the assumption only needs to be enforced at a discrete set of points.
Orphan-free Voronoi diagrams and dual triangulations {#sec:simpleplanar}
----------------------------------------------------
As described in the classic survey by Aurenhammer [@Aurenhammer:1991], planar Voronoi diagrams and Delaunay triangulations are duals in a graph theoretical sense. Associated to the ordinary Voronoi diagram is a simple, planar (primal) graph with vertices at points equidistant to three or more sites (Voronoi vertices), and edges composed of line segments equidistant to two sites (Voronoi edges). Because edges are always line segments, the graph is simple (has no multi-edges or self-loops), and this construction provides an embedding of the graph, which must therefore be planar.
For Voronoi diagrams defined by divergences, the situation is markedly different. The incidence relations between Voronoi elements cannot be so easily established. For instance, Voronoi edges may be disconnected and incident to any number of Voronoi vertices. For this reason, we begin our proof by constructing an embedding of a primal graph from the incidence relations of the Voronoi diagram (definition \[def:incidence\]), in a way that generalizes ordinary Voronoi diagrams, and show that this graph is simple and planar (section \[sec:planar\]). This primal graph is then dualized into a simple, planar graph. The dual graph is denoted the Delaunay *triangulation* because, as will be shown, it is composed of convex faces which can be triangulated without breaking any of its important properties, such as embeddability or the *empty circum-ball* property (property \[cor:VorI\]).
The rest of the paper makes heavy use of the following trivial lemmas, which we include here for convenience. The first follows directly from the properties of $D$, while the second is a direct consequence of the strict convexity of ${D\left(\cdot \parallel p\right)}$ and the continuity of $D$ (note that $D$ is globally continuous since it is continuous in the second argument and convex in the first, and therefore it is also continuous in the first argument [@rockafellar1997convex]).
\[lem:site\] Every site $s_i\in{S}$ is an interior point of its corresponding Voronoi region ${\text{Vor}}_i$.
\[lem:midpoint\] Given two sites $s_i,s_j\in {S}$ with supporting line $L_{ij}$, all points $p\in L_{ij}$ that are equidistant to $s_i$ and $s_j$ belong to the segment $\overline{s_i s_j}$. Furthermore, there is always at least one such point.
Summary of results {#sec:summary}
==================
Consider first the special case that all sites in ${S}$ are colinear. The structure of the Voronoi diagram and the Delaunay triangulation is very simple in this case. If we order the sites $s_1,\dots,s_n$ sequentially along their supporting line, lemma \[lem:midpoint\] shows that there must be Delaunay edges between successive sites, while the strict convexity of the balls implies that these are the only edges (all points in $\overline{s_i s_{i\oplus 1}}$ are strictly closer to $s_i,s_{i\oplus 1}$ than to any other site), and that there are no Delaunay faces (since three colinear points cannot be in the boundary of a strictly convex ball). The following proposition does not require assumption \[ass:BAA\] nor \[ass:EGA\].
[For all divergences]{} $D$, the Delaunay triangulation of a set of colinear sites is a chain connecting successive sites $s_i,s_{i\oplus 1}$, $i=1,\dots,n-1$ along their supporting line.
With the colinear site case covered, we assume in the remainder that *not all* sites are colinear, and that $D$ satisfies assumptions \[ass:BAA\] and \[ass:EGA\]. We begin, in section \[sec:planar\], by constructing a primal graph from the incidence relations between Voronoi elements, and dualize it to obtain a simple, planar graph.
\[th:simpleplanar\] The dual of the primal Voronoi graph of an orphan-free Voronoi diagram is a simple, connected, planar graph.
Note that the differentiability of $D$ with respect to the second argument is only used in (a small neighborhood around) Voronoi vertices (a set of isolated points). Everywhere else, it suffices that $D$ is continuous in its second argument.
While this dual graph is an embedded planar graph with curved edges, we then show that it is also an embedded planar graph with vertices at the sites and straight edges.
\[th:main\] The straight-edge dual of a primal Voronoi graph (obtained from an orphan-free Voronoi diagram of a set of sites ${S}$) is embedded with vertices at the sites, has (non-degenerate) strictly convex faces, and covers the convex hull of ${S}$.
As described in Section \[sec:simpleplanar\], lemmas \[lem:regionSC\] and \[lem:SCedges\] can be used in conjunction with theorem \[th:main\] to conclude that orphan-freedom is a sufficient condition for the well-behavedeness of not just the dual, but also of the primal Voronoi diagram. Note that this excludes isolated Voronoi edges (those not incident to any Voronoi vertex), which are shown to be contained in Voronoi regions, and are considered part of their containing regions (section \[sec:propedges\]).
\[cor:VorI\] All the elements of an orphan-free Voronoi diagram are connected, with the exception of isolated Voronoi edges.
Isolated edges are connected components of a Voronoi edge which are incident to a single Voronoi region. Since they do not affect the construction of the primal Voronoi graph, they can be safely discarded, as shown in section \[sec:propedges\].
Perhaps the most fundamental property of the diagrams that we use in the proofs is that every dual face has an “empty" circumscribing *convex* ball. This empty circum-ball (ECB) property is analogous to the empty circumcircle property of ordinary Voronoi diagrams:
\[prop:ECB\] For every dual face with vertices $s_{i_1},\dots,s_{i_k}$ there is a convex ball that circumscribes $s_{i_1},\dots,s_{i_k}$ and contains no site in its interior.
Indeed, since to every dual face $f$ with vertices $s_{i_1},\dots,s_{i_k}$ ($k\ge 3$) corresponds a Voronoi element ${\text{Vor}}_{\{i_1,\dots,i_k\}}$, any point $c\in{\text{Vor}}_{\{i_1,\dots,i_k\}}$ serves as center of an empty circumscribing ball of $f$. To see that this ball must be “empty", note that no site $s'$ can be strictly inside the circumscribing ball (certainly not $s_{i_1},\dots,s_{i_k}$, since they are in the boundary), or $c$ would be closer to $s'$ than to $s_{i_1},\dots,s_{i_k}$, and therefore it would not be $c\in{\text{Vor}}_{\{i_1,\dots,i_k\}}$.
Notice that, although we consider Voronoi diagrams of the second kind, it is the convexity of balls of the *first kind* that establishes the ECB condition. The ECB property is, in general, not satisfied by Delaunay triangulations of the first kind.
![ We prove that the Delaunay triangulation is embedded (theorem \[th:main\]) by showing that its boundary is simple and convex (corollary \[cor:boundary\]), and its interior is a “flat sheet": it has no edge fold-overs (green edge) (lemma \[lem:ef\]). We use a discrete version of the Poincaré-Hopf index theorem (lemma \[lem:ph\]) to prove that an edge fold-over would create a “wrinkle" (circled) somewhere in the triangulation (lemma \[lem:index-1\]), which in turn would force some vertex (blue) to “invade" a face (red) (lemma \[lem:non-negative\]), breaking the face’s empty circum-ball (grey, dotted) condition (proposition \[prop:ECB\]).[]{data-label="fig:outline"}](outline.pdf){width="2.5in"}
After establishing that a Voronoi diagram can be associated with an embedded planar primal graph which can be dualized into a planar dual graph (section \[sec:planar\]), the rest of the paper is concerned with the proof of our main claim (theorem \[th:main\]), whose structure is outlined in figure \[fig:outline\]. The proof of embeddability of the straight-edge dual is divided in two parts. In the first part (section \[sec:boundary\]), we use the bounded anisotropy assumption (assumption \[ass:BAA\]) to show that the boundary of the straight-edge dual Delaunay triangulation (the set of edges shared by only one face) coincides with the boundary of the convex hull of the sites, and therefore is a simple, closed polygonal chain, a fact necessary for the second part of the proof to proceed. Section \[sec:boundary\] is the more technical part of the proof; at its heart it is an application of Brouwer’s fixed point theorem. In section \[sec:interior\], we use the theory of discrete one-forms [@1form] to show that the Delaunay triangulation has no fold-overs (is a “flat sheet") and is therefore a single-cover of the convex hull of ${S}$. Note that these two results, along with the ECB property, mirror similar properties of ordinary Delaunay triangulations.
The above results can be particularized to a number of existing divergences and metrics. We briefly discuss next a few of them, as well as simple conditions for assumption \[ass:BAA\] to hold for some of them (with proofs in Appendix A).
Bregman divergences {#sec:DF}
-------------------
Given a strictly convex, everywhere differentiable function $F:\mathbb{R}^2\rightarrow\mathbb{R}$, the Bregman divergence $$\label{eq:defDF}
D_F(p \parallel q) \equiv F(p) - F(q) - \langle p-q, \nabla F(q)\rangle$$ is the (non-negative) difference between $F(p)$ and the first-order Taylor approximation of $F(p)$ around $q$ (the first order Lagrange remainder). Bregman divergences are widely used in statistics and include the Kullback-Leibler divergence. By the (strict) convexity of $F$, and the definition of $D_F$ it it is clear that, whenever $F$ is twice continuously differentiable, $D_F$ is (strictly) convex in the first argument and continuously differentiable in the second. From the definition of $D_F$, it is clear that Bregman Voronoi diagrams of the first kind are composed of regions $${\text{Vor}}^1_{F,i} = \{ p\in\mathbb{R}^2 : \langle p, \nabla F(s) - \nabla F(s_i)\rangle \le
F(s_i)-\langle s_i, \nabla F(s_i)\rangle +
F(s) - \langle s, \nabla F(s)\rangle , \forall s\in {S}\},$$ which are intersections of half-spaces of the form $\{p\in\mathbb{R}^2 : \langle p, a\rangle \le b\}$. Furthermore, Bregman Voronoi diagrams of the first kind are simply power diagrams [@Bregman], and thus their dual Delaunay triangulations of the first kind are always embedded [@powerdiag; @DMG]. On the other hand, Bregman diagrams of the *second* kind can be shown to be affine diagrams only in the dual (gradient) space [@Bregman]. In the original space, the cells ${\text{Vor}}^2_{F,i}$ are not simple intersections of half-spaces and, in general, they have curved boundaries. Prior to this work, no guarantees concerning Bregman Delaunay triangulations of the second kind were available.
\[lem:DFgamma\] If $F\in\mathcal{C}^2$ and there is $\gamma > 0$ such that the Hessian of $F$ has ratio of eigenvalues bounded by $\lambda_{\text{min}}/\lambda_{\text{max}}\ge \gamma$, then assumption \[ass:BAA\] holds.
Quadratic divergences {#sec:DQ}
---------------------
As is well known, the approximation efficiency of a piecewise-linear function supported on a triangulation can be greatly improved by adapting the shape and orientation of its elements to the target function [@triangle; @DAzevedo; @DBLP:conf/imr/CanasG06]. An effective way to construct such anisotropic triangulations is to dualize a Voronoi diagram derived from an anisotropic divergence [@LS; @DW].
By considering a $\mathcal{C}^1$ metric (in coordinates: a function $Q:\mathbb{R}^2\rightarrow\mathbb{R}^{2\times 2}$ that is symmetric, positive definite), we define the quadratic divergence as: $$\label{eq:defDQ}
D_Q(p \parallel q) \equiv \left[ (p-q)^t Q(q) (p-q) \right]^{1/2},$$ which is clearly strictly convex in the first argument and continuously differentiable in the second. Voronoi diagrams and Delaunay triangulations with respect to $D_Q$, of the first and seconds kinds, have been considered in the literature. The diagram and the dual triangulation of the *first kind* were proposed by Labelle and Shewchuk [@LS], while those of the second kind were discussed by Du and Wang [@DW]. While the work of Du and Wang does not provide theoretical guarantees, that of Labelle and Shewchuk provides an algorithm that is guaranteed to output *a* set of sites for which the Voronoi diagram of the *first kind* is orphan-free, and whose corresponding Delaunay triangulation is embedded.
\[lem:DQgamma\] If there is $\gamma > 0$ such that $Q$ has ratio of eigenvalues bounded by $\lambda_{\text{min}}/\lambda_{\text{max}}\ge \gamma$, then assumption \[ass:BAA\] holds.
Note that the above condition on the bounded anisotropy of $Q$ may commonly hold in practice, for instance if the metric is sampled on a compact domain and continuously extended to the plane by reusing sampled values only.
In the case of quadratic divergences, there already exists sufficient conditions to generate orphan-free Voronoi diagrams. In particular, it has been shown that if $\sigma$ is a bound on a certain measure of variation of $Q$, then any (asymmetric) $\epsilon$-net with respect to $D_Q$ that satisfies $\epsilon\sigma \le 0.098$ (corresponding to a roughly $10\%$ variation of eigenvalues between Voronoi-adjacent sites) is guaranteed to be orphan-free [@avd].
Normed spaces {#sec:Lp}
-------------
Our results also cover all normed spaces with a continuously differentiable, strictly convex norm, including the $L_p$ spaces, but excluding the cases $p=1$ and $p=\infty$.
\[lem:Lpgamma\] Distances derived from strictly convex $\mathcal{C}^1$ norms satisfy assumption \[ass:BAA\].
Csiszár f-divergences {#sec:Df}
---------------------
Given a convex real function $f$ with $f(1)=0$ and two measures $\rho,\mu$ over a probability space $\Omega$, Csiszár’s f-divergence [@CsiszarTutorial] is $$\label{eq:defDf}
D_f(\rho\parallel \mu) \equiv \int_{\Omega}d\mu\, f\left(\frac{d\rho}{d\mu}\right)$$ where $\rho$ is absolutely continuous with respect to $\mu$, and therefore has a Radon-Nikodym derivative $d\rho/d\mu$.
If $f$ is strictly convex, then the f-divergence is strictly convex in the first argument and continuously differentiable in the second (in this case it is also jointly convex). For instance, the strictly convex function $f:x\mapsto \left(\sqrt x - 1\right)^2$ generates the Hellinger distance. F-divergences are functions of measures, and thus often in practice restricted to the probability simplex.
The limitation of our work to two dimensions implies that results for f-divergences are limited to probability measures supported on just three atoms. Their applicability is thus somewhat limited, and are only included for completeness.
Primal Voronoi diagram and dual Delaunay triangulation {#sec:planar}
======================================================
In this section we use the definition of Voronoi diagram (definition \[def:VorI\]) to construct an embedded simple planar graph whose incidence relations match those of the Voronoi diagram. We then dualize this graph to obtain an embedded simple planar graph with vertices at the sites and curved edges. Section \[sec:dual\] will then show that the dual graph is also embedded when replacing curved edges by straight segments. Recall that we have assumed that not all sites are colinear (the colinear case is described in section \[sec:summary\]).
Assumptions {#sec:assumptions}
-----------
We begin by making the following two technical assumptions. [**Path-connectedness.**]{} Assume that all connected components of Voronoi elements are also path-connected. In fact, given the assumption below, as well as assumptions \[ass:BAA\] and \[ass:EGA\], we only need to further assume that connected components of Voronoi *edges* are path-connected. Indeed, Voronoi regions are open and Voronoi vertices will be shown to be composed of isolated points, and therefore their connected components are automatically path-connected [@munkres2000topology p. 158].
[**Boundaries of Voronoi regions.**]{} Further assume that the boundary of bounded, simply-connected Voronoi regions are simple, closed (Jordan) curves. For unbounded regions $U$, we assume that they can be first mapped through a continuous transformation $T:U\rightarrow U'$ onto a bounded set $U'$, for instance through an appropriate M[ö]{}bius transformation. Bounded simply-connected sets whose boundary is a Jordan curve are those that are uniformly connected *im kleinen* [@Moore1918][^1].
Properties of Voronoi elements {#sec:properties}
------------------------------
Before constructing an appropriate primal graph from the connectivity relations of the Voronoi diagram, we first establish some relevant properties of the diagram’s elements.
We say that Voronoi element ${\text{Vor}}_I$ is incident to Voronoi element ${\text{Vor}}_J$ (denoted ${\text{Vor}}_I\rightsquigarrow{\text{Vor}}_J$) if their closures overlap and $\overline{{\text{Vor}}_I}\cap\overline{{\text{Vor}}_J}\subseteq\overline{{\text{Vor}}_J}$.
From this incidence relation we build a primal Voronoi graph, whose dual is the Delaunay triangulation with respect to $D$. Since “planar graphs, and graphs embeddable on the sphere are one and the same” [@bondy2008graph p. 247], we consider incidence relations on the Riemann sphere (by stereographically projecting the plane onto $\mathbb{S}^2$), where the added vertex at infinity is defined to be incident to unbounded elements on the plane. Geometric constructions will, however, typically be carried out on the plane for convenience.
### Incident elements {#sec:incidence}
![A portion of a Voronoi diagram, with highlighted incidence relations between Voronoi elements. The incidence relation (definition \[def:incidence\]) forms a directed acyclic graph. []{data-label="fig:incidence"}](incidence.pdf){width="2.5in"}
Consider the following definition of incidence between Voronoi regions (or between connected components of Voronoi regions):
\[def:incidence\] Given $I,J\subseteq S$, we say that ${\text{Vor}}_I$ is incident to ${\text{Vor}}_J$ (written ${\text{Vor}}_I \rightsquigarrow {\text{Vor}}_J$) iff $\overline{{\text{Vor}}_I}\cap\overline{{\text{Vor}}_J}\ne\phi$ *and* $I\subset J$.
By orphan-freedom, and lemma \[lem:connectededges\], both Voronoi regions and edges are connected (except for isolated edges, which are defined in section \[sec:propedges\]). For simplicity, in the sequel we refer to connected components of Voronoi vertices simply as “Voronoi vertices", except for the statement of lemma \[lem:vertexincidence\], which makes this distinction explicit.
Note that this definition and the one in section \[sec:properties\] are equivalent since, for distinct sets $I\ne J$, and by the continuity of $D$, $\overline{{\text{Vor}}_I}\cap\overline{{\text{Vor}}_J}\subseteq\overline{{\text{Vor}}_J}$ implies $I\subset J$ (and viceversa).
Given the following substitution rules: $$\begin{aligned}
A,B\rightsquigarrow C ~&\Rightarrow~ A\rightsquigarrow C \text{ and } B\rightsquigarrow C \\
A \rightsquigarrow B,C ~&\Rightarrow~ A\rightsquigarrow B \text{ and } A\rightsquigarrow C, \end{aligned}$$ the following are the incidence relations depicted in figure \[fig:incidence\]: $$\begin{aligned}
{\text{Vor}}_i,{\text{Vor}}_j &\rightsquigarrow& {\text{Vor}}_{ij},{\text{Vor}}_{ijk} \\
{\text{Vor}}_j,{\text{Vor}}_k &\rightsquigarrow& {\text{Vor}}_{jk},{\text{Vor}}_{ijk} \\
{\text{Vor}}_k,{\text{Vor}}_i &\rightsquigarrow& {\text{Vor}}_{ki},{\text{Vor}}_{ijk} \\
{\text{Vor}}_{ij},{\text{Vor}}_{jk},{\text{Vor}}_{ki} &\rightsquigarrow& {\text{Vor}}_{ijk}, \end{aligned}$$ where we often write ${\text{Vor}}_{ij}$ instead of ${\text{Vor}}_{\{i,j\}}$ for simplicity.
\[prop:boundaryincidence\] All points in the boundary of a Voronoi element ${\text{Vor}}_I$ belong to either ${\text{Vor}}_I$, or to an element that ${\text{Vor}}_I$ is incident to.
Let $p\in\partial{\text{Vor}}_I$, and $J$ be the set of sites that $p$ is equidistant to. Since $p\in\partial{\text{Vor}}_I$, by the continuity of $D$, $p$ is equidistant to all sites in $I$, and therefore $I\subseteq J$. The property follows from the definition of incidence.
From the properties of strict set containment, it follows that the incidence relation $\rightsquigarrow$ forms a directed acyclic graph (a cycle ${\text{Vor}}_I\rightsquigarrow{\text{Vor}}_J\rightsquigarrow{\text{Vor}}_K\rightsquigarrow{\text{Vor}}_I$ would imply $I\subset I$, a contradiction).
From property \[prop:boundaryincidence\] it follows that closed Voronoi elements are those with zero out-degree in the incidence graph (e.g. ${\text{Vor}}_{ijk}$ in figure \[fig:incidence\]), and that open Voronoi elements (i.e. Voronoi regions) are those with zero in-degree (e.g. ${\text{Vor}}_i,{\text{Vor}}_j,{\text{Vor}}_k$ in figure \[fig:incidence\]).
### Properties of Voronoi vertices
The main properties at Voronoi vertices are derived from the two assumptions in section \[sec:setup\]. Assumptions \[ass:BAA\] and \[ass:EGA\] are useful when deriving properties of the vertex at infinity, and bounded vertices (all other vertices), respectively.
Given the set negated gradients $g_1,\dots,g_m$ at a bounded vertex point (eq. \[eq:neggrad\]), by assumption \[ass:EGA\] they are distinct vertices of their convex hull. It is then possible to define “outward" vectors $d_1,\dots,d_m$ (eq. \[eq:dkdef\]) such that eq. \[eq:extremal1\] holds. This is because, for each $k=1,\dots,m$, eq. \[eq:extremal1\] simply requires all gradients other than $g_k$ to be below the (red dotted) line orthogonal to $d_k$ passing through $g_k$ (as shown in fig. \[fig:EGA.a\] for $d_1$), which is possible because $g_1,\dots,g_m$ are the distinct vertices of ${\text{conv} \{ }g_1,\dots,g_m\}$.
Figure \[fig:EGA.b\] shows that eq. \[eq:extremal2\] holds for the same reason as above. Given two gradients that are adjacent vertices of ${\text{conv} \{ }g_1,\dots,g_m\}$ (for instance $g_1,g_2$), eq. \[eq:extremal2\] (in this case with $k=1, k\oplus 1=2$) is possible whenever all gradients different from $g_1,g_2$ are simultaneously below two lines, both passing through $g_1$ and orthogonal to $d_1$ and $d_2$ (the gray area). This holds because the outward directions $d_k$ can be chosen to form an obtuse angle with both segments $g_k,g_{k\oplus 1}$ and $g_k,g_{k\ominus 1}$. The same argument applies to eq. \[eq:extremal3\].
\[lem:vertexincidence\] A Voronoi vertex ${\text{Vor}}_I$ is a collection of discrete points, at each of which there is an ordered set of indices $i_1,\dots,i_m$ such that $I=\{i_i,\dots,i_m\}$ and the following incidence relations hold: $$\begin{aligned}
{\text{Vor}}_{i_1},\dots,{\text{Vor}}_{i_m} &\rightsquigarrow& {\text{Vor}}_I ~~~~~~~~~~~~~ \text{(region-vertex incidence)}\\
{\text{Vor}}_{\{i_1,i_2\}},{\text{Vor}}_{\{i_2,i_3\}},\dots,{\text{Vor}}_{\{i_m,i_1\}} &\rightsquigarrow& {\text{Vor}}_I ~~~~~~~~~~~~~ \text{(edge-vertex incidence)}.
\end{aligned}$$ Additionally, if an edge ${\text{Vor}}_{jk}$ is incident to a vertex ${\text{Vor}}_I$, then ${\text{Vor}}_j,{\text{Vor}}_k \rightsquigarrow {\text{Vor}}_{jk}$.\
If ${\text{Vor}}_I$ is the vertex at infinity (${\text{Vor}}_\infty$), then $i_1,\dots,i_m$ are the indices of the sites in the boundary of the convex hull ${{conv}\left\{{S}\right\}}$, in either clockwise or counter-clockwise order.
[**\[Bounded vertices, ${\text{Vor}}_I$\]**]{}. Let ${\text{Vor}}_I$ be a Voronoi vertex not at the point at infinity and $v$ be a point in ${\text{Vor}}_I$. By the extremal gradient assumption (assumption \[ass:EGA\]), the negated gradients $$\label{eq:neggrad}
g_k\equiv -\nabla_p {D\left(s_{i_k} \parallel p\right)}\big|_v ~,~~ i_k \in I$$ are distinct vertices of their convex hull. Let $i_1,\dots,i_m$ be the indices in $I$ ordered (for instance clockwise) around $\partial {\text{conv} \{ }g_1,\dots,g_m\}$, as shown in figure \[fig:EGA.a\].
Since $g_k$, with $k=1,\dots,m$ are distinct vertices of their convex hull, it is easy to show that there are direction (unit) vectors $d_k$, with $k=1,\dots,m$, such that for all $k$ it holds: $$\label{eq:extremal1}
\left< g_k - g_j, d_k \right> > 0 ~,~~ j\in\{1,\dots,m\}, j \ne k.$$ For instance $$\label{eq:dkdef}
d_k\equiv \frac{(g_k-g_{k\oplus 1})+(g_k-g_{k\ominus 1})}{\|g_k-g_{k\oplus 1}+g_k-g_{k\ominus 1}\|}.$$
By the multivariate version of Taylor’s theorem [@konigsberger2006analysis p. 68], for each $k,j$, and $p\in\mathbb{R}^2$, we may write: $$-{D\left(s_k \parallel p\right)} + {D\left(s_j \parallel p\right)} = \left<g_k - g_j, p-v\right> + o\left(\|p-v\|\right).$$ For each $k=1,\dots,m$, and $j=1,\dots,m$ with $j\ne k$, let $p-v = \mu d_k$, with $\mu > 0$, and let $\alpha_{k,j} \equiv \left<g_k - g_j, d_k\right> / 2 > 0$. It then follows that: $$\left[-{D\left(s_k \parallel v+\mu d_k\right)} + {D\left(s_j \parallel v+\mu d_k\right)}\right]/\mu = 2\alpha_{k,j} + f(\mu),$$ where $\lim_{\mu\rightarrow 0} f(\mu)=0$. Note that, crucially, $f$ depends on $\mu$ but not on the direction $d_k$.
Since $f(\mu)\rightarrow 0$ with $\mu\rightarrow 0$, we can pick constants $\varepsilon_{k,j}>0$ sufficiently small so that for all $\mu < \varepsilon_{k,j}$ it holds $|f(\mu)| < \alpha_{k,j}$, and therefore $\left[-{D\left(s_{i_k} \parallel v+\mu d_k\right)} + {D\left(s_{i_j} \parallel v+\mu d_k\right)}\right]/\mu > \alpha_{k,j}$. Let $\varepsilon >0$ be the minimum of all $\varepsilon_{k,j}$, with $j,k=1,\dots,m$, and $j\ne k$.
Since $v$ is strictly closest to sites $s_{i_1},\dots,s_{i_m}$, let $\delta$ be small enough so all points $p\in\mathbb{R}^2$ with $\|p-v\|<\delta$ are closest only to sites in $s_{i_1},\dots,s_{i_m}$ (which is possible since $D$ is continuous). Consider the set of points in a small circle of radius $0 < \mu < \min\{\delta,\varepsilon\}$ around $v$. From the above, we have that at the point $v + \mu d_k$, it holds: $$\left[-{D\left(s_k \parallel v+\mu d_k\right)} + {D\left(s_j \parallel v+\mu d_k\right)}\right]/\mu > \alpha_{k,j} > 0 ~,~~ j\in\{1,\dots,m\}, j \ne k,$$ from which it follows that $v+\mu d_k$ is strictly closer to $s_{i_k}$ than to any other site. Since this is true for all $k=1,\dots,m$ and for all sufficiently small $0<\mu<\min\{\delta,\varepsilon\}$, the incidence relations $${\text{Vor}}_{i_1},\dots,{\text{Vor}}_{i_m} \rightsquigarrow {\text{Vor}}_I$$ follow.
Because $g_1,\dots,g_m$ are vertices of ${\text{conv} \{ }g_1,\dots,g_m\}$, it is clear, as shown in figure \[fig:EGA.b\], that for each $k=1,\dots,m$ there are constants $\beta_{k,j},\beta_{k\oplus 1,j} > 0$, with $j\ne k$ and $j\ne k\oplus 1$, such that, for every unit vector $d_{k,k\oplus 1}$ intermediate between $d_k$ and $d_{k\oplus 1}$, it holds: $$\begin{aligned}
\label{eq:extremal2} \left< g_k - g_j, d_{k,k\oplus 1} \right> &> 2\beta_{k,j} > 0 ~,~~~~~ j\ne k, j \ne k\oplus 1 \\
\label{eq:extremal3} \left< g_{k\oplus 1} - g_j, d_{k,k\oplus 1} \right> &> 2\beta_{k \oplus 1,j} > 0 ~,~~ j\ne k, j \ne k\oplus 1.
\end{aligned}$$ Let $\xi_k > 0$ be small enough such that for all $0 < \mu < \xi$, it holds $f(\mu) < \min\{\min_j \beta_{k,j}, \min_j \beta_{k\oplus 1,j}\}$. Let $\xi\equiv \min_k \xi_k$, then for all $0 < \mu < \min\{\delta,\varepsilon,\xi\}$, and every point $v+\mu d_k$ it holds: $$\begin{aligned}
-{D\left(s_k \parallel v+\mu d_{k,k\oplus 1}\right)} + {D\left(s_j \parallel v+\mu d_{k,k\oplus 1}\right)} &> \beta_{k,j} > 0 ~,~~~~~ j\ne k, j \ne k\oplus 1 \\
-{D\left(s_{k\oplus 1} \parallel v+\mu d_{k,k\oplus 1}\right)} + {D\left(s_j \parallel v+\mu d_{k,k\oplus 1}\right)} &> \beta_{k\oplus 1,j} > 0 ~,~~~~~ j\ne k, j \ne k\oplus 1,
\end{aligned}$$ and therefore $v+\mu d_{k,k\oplus 1}$ is closest to either $s_k,s_{k\oplus 1}$, or to both. For each such $\mu$, and for each $k=1,\dots,m$, by the intermediate value theorem, there is a direction vector $d$ between $d_k,d_{k\oplus 1}$ such that $v+\mu d$ is in ${\text{Vor}}_{k,k\oplus 1}$. Note that, by the above construction, for every such sufficiently small $\mu$, ${\text{Vor}}_{k,k\oplus 1}$, with $k=1,\dots,m$, are the only Voronoi edges inside the ball of radius $\mu$ around $v$. From this it directly follows that:
1. since all points $v+\mu d$, with unit vector $d$ and sufficiently small $\mu$ have been shown to be in a Voronoi region or edge, $v$ is an isolated point of ${\text{Vor}}_I$; since $v$ is a generic point of ${\text{Vor}}_I$, it follows that ${\text{Vor}}_I$ is composed of isolated points;
2. it holds ${\text{Vor}}_{\{i_1,i_2\}},{\text{Vor}}_{\{i_2,i_3\}},\dots,{\text{Vor}}_{\{i_m,i_1\}} \rightsquigarrow {\text{Vor}}_I $; and
3. if a Voronoi edge ${\text{Vor}}_{jk}$ is incident to ${\text{Vor}}_I$, then ${\text{Vor}}_j,{\text{Vor}}_k\rightsquigarrow {\text{Vor}}_{jk}$, since the only edges incident to ${\text{Vor}}_I$ are ${\text{Vor}}_{i_k,i_{k\oplus 1}}$, with $k=1,\dots,m$.
[**\[Vertex at infinity, ${\text{Vor}}_\infty$\]**]{}. Incidence to the vertex at infinity is dealt with in section \[sec:boundary\], where lemma \[lem:VW\] shows that the only unbounded elements are of the form ${\text{Vor}}_I$ where all $i_k\in I$ are vertices of ${{conv}\left\{{S}\right\}}$, and lemmas \[boundary\_easy\] and \[lem:hard\] show that, if $s_{i_1},\dots,s_{i_m}$ are the vertices on the boundary of ${{conv}\left\{{S}\right\}}$ (whether on an edge or vertex of $\partial{{conv}\left\{{S}\right\}}$), ordered around $\partial{{conv}\left\{{S}\right\}}$, then ${\text{Vor}}_{i_1},\dots,{\text{Vor}}_{i_m}$ and ${\text{Vor}}_{i_1,i_2},\dots,{\text{Vor}}_{i_m,i_1}$ are the only unbounded elements (and therefore incident to ${\text{Vor}}_\infty$). In this sense we can say that the vertex at infinity ${\text{Vor}}_\infty$ is the Voronoi vertex ${\text{Vor}}_{i_1,\dots,i_m}$. The proofs in section \[sec:boundary\] show that points $p$ in any circle of sufficiently large radius are incident only to sites in $s_{i_1},\dots,s_{i_m}$, that $p$ cannot be incident to more than two sites simultaneously (lemma \[lem:contrad\]), and therefore $p$ cannot belong to a Voronoi vertex, and finally that $p$ can *only* be simultaneously closest to two consecutive sites of the form $s_{i_k},s_{i_{k\oplus 1}}$ (page ). Note that the relevant proofs of section \[text:boundary\] use the bounded anisotropy assumption (assumption \[ass:BAA\]), but do not use any result from this section.
From the proof of lemma \[lem:vertexincidence\], it is clear that the bounded anisotropy assumption (assumption \[ass:BAA\]) is constructed so that lemma \[lem:vertexincidence\] holds for the vertex at infinity, while the extremal gradient assumption (assumption \[ass:EGA\]) is meant to ensure that lemma \[lem:vertexincidence\] holds for regular (bounded) vertices.
### Properties of Voronoi edges {#sec:propedges}
We begin by considering (isolated) Voronoi edges that are bounded and not incident to any Voronoi vertex. Since, as will be shown in lemma \[lem:SCedges\], Voronoi edges are simply connected, it is easy to see that for any Voronoi edge ${\text{Vor}}_{ij}$ that is not incident to any bounded Voronoi vertex, it can only be ${\text{Vor}}_i\rightsquigarrow{\text{Vor}}_{ij}$ or ${\text{Vor}}_j\rightsquigarrow{\text{Vor}}_{ij}$, and ${\text{Vor}}_{ij}$ cannot be involved in any other incidence relation. To see this, first note that an isolated component of ${\text{Vor}}_{ij}$ has, by definition, zero out-degree, and therefore it is closed. Because ${\text{Vor}}_{ij}$ is not incident to the vertex at infinity, it is bounded. Since ${\text{Vor}}_{ij}\rightsquigarrow{\text{Vor}}_{kl}$ implies that their common boundary belongs to vertex ${\text{Vor}}_{ijkl}$ (where it may be $k=l$), ${\text{Vor}}_{ij}$ is not incident to any Voronoi edge. ${\text{Vor}}_{ij}$ cannot be incident to a region ${\text{Vor}}_k$ with $k\notin\{i,j\}$, or else their common boundary would belong to vertex ${\text{Vor}}_{ijk}$. Finally, we show that it cannot be both ${\text{Vor}}_i\rightsquigarrow{\text{Vor}}_{ij}$ and ${\text{Vor}}_j\rightsquigarrow{\text{Vor}}_{ij}$. Because ${\text{Vor}}_{ij}$ is closed, simply connected, and bounded, by the continuity of $D$, we can consider a sufficiently small $\varepsilon > 0$ such that every $\varepsilon$-offset of its outer boundary cannot be closest to any site $s_k$ with $k\notin\{i,j\}$. If ${\text{Vor}}_i,{\text{Vor}}_j\rightsquigarrow{\text{Vor}}_{ij}$, then there must be $0 < \mu < \varepsilon$ such that the $\mu$-offset $\nu_\mu$ of ${\text{Vor}}_{ij}$’s outer boundary has at least one point closest to $s_i$, and one point closest to $s_j$, and therefore, by continuity of $D$, at least one point equally close to $s_i,s_j$. Since all points in $\nu_\mu$ are closest to $s_i,s_j$ only, then $\nu_\mu$ has been shown to have a point in ${\text{Vor}}_{ij}$, contradicting the fact that $\nu_\mu$ is a $\mu$-offset of ${\text{Vor}}_{ij}$’s outer boundary, and therefore outside ${\text{Vor}}_{ij}$. Let ${\text{Vor}}_{ij}$ be an bounded isolated Voronoi edge such that ${\text{Vor}}_i\rightsquigarrow{\text{Vor}}_{ij}$. Because they are not incident to any Voronoi vertex, bounded isolated edges will not be considered part of the primal Voronoi graph. For simplicity, we consider all points of an isolated edge ${\text{Vor}}_{ij}$ to be part of its containing Voronoi region (say ${\text{Vor}}_i$), and therefore to be (by definition) strictly closer to $s_i$ than to any other site. This is not just a simplification (which does not affect the final Voronoi graph), but will allow us to prove that Voronoi regions are simply connected.
We begin by proving the following technical lemma.
\[lem:RSC\] Let the boundary $\partial R$ of $R\subset\mathbb{R}^2$ be a simple, closed path, and ${\text{Vor}}_I$ be a Voronoi element of an orphan-free diagram. If $\partial R\subseteq{\text{Vor}}_I$, then $R\subseteq{\text{Vor}}_I$.
Let $I=i_1,\dots,i_m$, and $\gamma\equiv\partial R$. We begin by showing that $R$ does not contain any site $s_i$ whenever $m>1$ or $i\ne i_1$. Let $m=1$, and $s_i\in R$ with $i\ne i_1$, as in figure \[fig:RSC.a\]. Let $r$ be the ray starting from $s_i$ in the direction of $s_i-s_{i_1}$ (note that $s_{i_1}$ may be inside or outside $R$). Since $r$ is unbounded and $R$ is bounded, then part of $r$ is outside $R$ and, by the Jordan curve theorem, it must intersect $\gamma$ at some point $q_{i_1}$. Since $\gamma\subset{\text{Vor}}_{i_1}$, $q_{i_1}$ is closest to $s_{i_1}$, while $s_i$ is closest to $s_i$ (since ${D\left(s_i \parallel s_i\right)}=0$ and ${D\left(\cdot \parallel s_i\right)}$ is non-negative and convex). By the continuity of $D$, there is an intermediate point $q_{i,i_1}$ between $s_i$ and $q_{i_1}$ that is equidistant to $s_{i_1}$ and $s_i$, contradicting lemma \[lem:midpoint\].
Let $m>1$, and let $s_i$ be any site (figure \[fig:RSC.b\]). Pick $j\ne i$ among $j\in\{i_1,\dots,i_m\}$, which is always possible because $m>1$. The argument is identical in this case, except that, because $\gamma\subset{\text{Vor}}_I$, then $q_j\in\gamma$ is closest and equidistant to $\{i_1,\dots,i_m\}$, and therefore closer to $s_i$ than to $s_j$, and the same argument holds.
[**\[Voronoi regions\]**]{}. We now prove that no point $p\in R$ belongs to a Voronoi region ${\text{Vor}}_i$ if $m>1$ or $i\ne i_1$. Let $p\in R$ belong to ${\text{Vor}}_i$, with $m>1$ or $i\ne i_1$, we show that this leads to a contradiction.
We first show that ${\text{Vor}}_i\subset R$. Assume otherwise. Since ${\text{Vor}}_i$ is open and connected (by the orphan-freedom property), it is path connected. Let $\Gamma\subset{\text{Vor}}_i$ be a simple path from $p$ to a point $q\in{\text{Vor}}_i$ outside $R$. By the Jordan curve theorem, $\Gamma\subset{\text{Vor}}_i$ intersects $\gamma\subset{\text{Vor}}_I$, which leads to a contradiction whenever $m>1$ or $i\ne i_1$.
Since ${\text{Vor}}_i\subset R$ and, by lemma \[lem:site\], $s_i\in{\text{Vor}}_i$, then $s_i\in R$, contradicting the fact that $R$ does not contain any site $s_i$ if $m>1$ or $i\ne i_1$.
[**\[Voronoi vertices\]**]{}. If $R$ contains a point $p$ that belongs to a Voronoi vertex ${\text{Vor}}_J$ with $J\ne I$, then $p$ must be in the interior of $R$, since its boundary $\gamma$ is in ${\text{Vor}}_I$. By lemma \[lem:vertexincidence\], $p$ is incident to ${\text{Vor}}_{j_1},\dots,{\text{Vor}}_{j_k}$, where $J=j_1,\dots,j_k$ and $k\ge 3$. Since $p$ is in the interior of $R$, then there are points $p_{j_i},\dots,p_{j_k}\in R$ that belong to ${\text{Vor}}_{j_1},\dots,{\text{Vor}}_{j_k}$, respectively. If $m>1$, then this contradicts the fact that $R$ does not have any point in a Voronoi region. If $m=1$, since $k\ge 3$, then one of $j_1,\dots,j_k$ must be different from $i_1$, contradicting the fact that $R$ does not have any point in a Voronoi region different from ${\text{Vor}}_{i_1}$.
[**\[Voronoi edges\]**]{}. Let ${\text{Vor}}'_{ij}$ be a connected component of a Voronoi edge, with $\{i,j\}\ne I$. If some point $p\in{\text{Vor}}'_{ij}$ is in $R$, then ${\text{Vor}}'_{ij}\subset R$, or else since, by the assumption in section \[sec:assumptions\], ${\text{Vor}}'_{ij}$ is path connected, there would be a path $\Gamma\subset{\text{Vor}}'_{ij}$ connecting $p$ to a point of ${\text{Vor}}'_{ij}$ outside $R$. By the Jordan curve theorem $\Gamma\subset{\text{Vor}}'_{ij}$ would intersect $\gamma\subset{\text{Vor}}_I$, a contradiction.
Since we have already discarded isolated Voronoi edges that are not incident to any Voronoi vertex, a Voronoi edge is always incident to a Voronoi vertex and, since ${\text{Vor}}'_{ij}$ is in the interior of $R$, then its incident Voronoi vertex is in $R$, a contradiction.
Finally, since we have shown that there cannot be any Voronoi vertices, edges, or regions ${\text{Vor}}_J$ with $J\ne I$ in $R$, then it must be $R\subset{\text{Vor}}_I$.
\[lem:connectededges\] Voronoi edges of an orphan-free diagram are connected.
Let ${\text{Vor}}^1_{ij},{\text{Vor}}^2_{ij}$ be two disconnected pieces of a Voronoi edge ${\text{Vor}}_{ij}$, as shown in figure \[fig:connectededges.a\]. Since we have discarded (bounded) isolated edges, we assume that ${\text{Vor}}_{ij}$ is incident to at least one vertex, and therefore by lemma \[lem:vertexincidence\], it is ${\text{Vor}}_i,{\text{Vor}}_j\rightsquigarrow{\text{Vor}}^1_{ij}$ and ${\text{Vor}}_i,{\text{Vor}}_j\rightsquigarrow{\text{Vor}}^2_{ij}$.
Since ${\text{Vor}}_i,{\text{Vor}}_j$ are incident to both ${\text{Vor}}^1_{ij},{\text{Vor}}^2_{ij}$, the boundaries of ${\text{Vor}}_i$ and ${\text{Vor}}_{ij}^1$ overlap (and likewise $\partial{\text{Vor}}_i \cap \partial{\text{Vor}}_{ij}^2, \partial{\text{Vor}}_j \cap \partial{\text{Vor}}_{ij}^1, \partial{\text{Vor}}_j \cap \partial{\text{Vor}}_{ij}^2 \ne \phi$). Let $p^1_i\in\partial{\text{Vor}}_i\cap\partial{\text{Vor}}^1_{ij}$ be a point in the common boundary between ${\text{Vor}}_i$ and ${\text{Vor}}^1_{ij}$, and $p^1_j\in\partial{\text{Vor}}_j\cap\partial{\text{Vor}}^1_{ij}$ be a point in the common boundary between ${\text{Vor}}_j$ and ${\text{Vor}}^1_{ij}$, and define the points $p^2_i,p^2_j$ analogously. Since ${\text{Vor}}^1_{ij},{\text{Vor}}^2_{ij}$ are disjoint, it holds $p^1_i\ne p^2_i$ and $p^1_j\ne p^2_j$, and therefore by lemma \[lem:regionpath\] there are non-crossing simple paths $\gamma^1_i,\gamma^2_i\subset{\text{Vor}}_i$ from $s_i\in{S}$ to $p^1_i,p^2_i$, respectively, and non-crossing simple paths $\gamma^1_j,\gamma^2_j\subset{\text{Vor}}_j$ from $s_j\in{S}$ to $p_{j,1},p_{j,2}$, respectively. Additionally, since by the assumption in section \[sec:assumptions\] ${\text{Vor}}^1_{ij},{\text{Vor}}^2_{ij}$ are path connected, there are simple paths $\gamma^1_{ij}\subset{\text{Vor}}^1_{ij}$ and $\gamma^2_{ij}\subset{\text{Vor}}^2_{ij}$ connecting $p^1_i$ to $p^1_j$, and $p^2_i$ to $p^2_j$, respectively. Let $\gamma^1$ be the concatenation of paths $\gamma^1_i,\gamma^1_{ij},\gamma^1_j$, and $\gamma^2$ be the concatenation of paths $\gamma^2_i,\gamma^2_{ij},\gamma^2_j$. By construction, and since ${\text{Vor}}^1_{ij},{\text{Vor}}^2_{ij}$ are disjoint, the simple paths $\gamma^1,\gamma^2$ only meet at their endpoints $s_i,s_j$.
Let $\gamma$ be the simple closed curve resulting from concatenating $\gamma^1,\gamma^2$. By the Jordan curve theorem, $\gamma$ divides the plane into an interior ($Int$) and exterior regions, bounded by $\gamma$. We first show that $Int$ does not contain any sites (other than $s_i,s_j$).
[**\[$Int$ contains no sites\]**]{}. We first divide $Int$ in three parts, as shown in figure \[fig:connectededges.a\]:
1. the region $U$ bounded by $\overline{s_i p^1_i}$, $\gamma^1_{ij}$, and $\overline{p^1_j s_j}$,
2. the region $L$ bounded by $\overline{s_i p^2_i}$, $\gamma^2_{ij}$, and $\overline{p^2_j s_j}$,
3. and $R\equiv Int \setminus \left(U \cup L\right)$.
We begin by observing that if $w_{ij}\in {\text{Vor}}_{ij}$, then the triangle $\triangle{s_i w_{ij} s_j}$ cannot contain any site (other than $s_i,s_j$) because 1) $w_{ij}$ is closest and equidistant to $s_i,s_j$, and 2) the ball of the first kind $\theta_{w_{ij}}(s_i)$ centered at $w_{ij}$ with $s_i,s_j$ in its boundary (see table \[table:notation\]) is convex and therefore contains $\triangle{s_i w_{ij} s_j}$. Since the sides of $\triangle{s_i w_{ij} s_j}$ are line segments, and $\theta_{w_{ij}}(s_i)$ is strictly convex, the only points of $\triangle{s_i w_{ij} s_j}$ touching the boundary of $\partial\theta_{w_{ij}}(s_i)$ are $s_i,s_j$, and therefore a site at any other point in $\triangle{s_i w_{ij} s_j}$ would be strictly closer to $w_{ij}$ than $s_i,s_j$, a contradiction.
Since $U$ can be written as the union of triangles with vertices $s_i,w_{ij},s_j$ with $w_{ij}\in\gamma^1_{ij}\subset{\text{Vor}}^1_{ij}\subset{\text{Vor}}_{ij}$, then $U$ does not contain any site other than $s_i,s_j$. An analogous argument proves that $L$ does not contain any site other than $s_i,s_j$.
We split the remaining region $R$ into four parts $R_{1,i},R_{1,j},R_{2,i},R_{2,j}$. Let $R_{1,i}$ be the part of $R$ bounded by the segment $\overline{s_i p^1_i}$ and the curve $\gamma^1_i$. Let $R'_{1,i} \equiv \displaystyle{\cup_{r\in \gamma^1_i} \overline{s_i r}}$ be the union of segments connecting $s_i$ to points in $\gamma^1_i$. Clearly, it is $R_{1,i}\subset R'_{1,i}$. We show that $R'_{1,i}$ cannot contain any site other than $s_i$, and thus the same is true of $R_{1,i}$.
Let $z\in R'_{1,i}$ be a site, and let $r\in\gamma^1$ be the point such that $z\in\overline{s_i r}$. Because $r\in\gamma^1_{i}\subset\overline{{\text{Vor}}_i}$, $r$ is closest and equidistant to $s_i$ (and possibly also to $s_j$), that is: ${D\left(s_i \parallel r\right)} \le {D\left(s_k \parallel r\right)}$ for all $k=1,\dots,n$. Since $z\in\overline{s_i r}$ and $z\ne s_i$, we can write $z=\lambda s_i + (1-\lambda) r$, with $0 \le \lambda < 1$, and therefore by the strict convexity of ${D\left(\cdot \parallel r\right)}$ it holds: $${D\left(z \parallel r\right)} = {D\left(\lambda s_i + (1-\lambda) r \parallel r\right)} < \lambda{D\left(s_i \parallel r\right)} + (1-\lambda){D\left(r \parallel r\right)} = \lambda{D\left(s_i \parallel r\right)} < {D\left(s_i \parallel r\right)},$$ where the last equality follows from ${D\left(r \parallel r\right)}=0$, and the last inequality follows from the non-negativity of $D$. This shows that the site $z$ is *strictly* closer to $r$ than $s_i$, a contradiction. Therefore there are no sites in $R'_{1,i}$, and thus no sites in $R_{1,i}\subseteq R'_{1,i}$ either. Applying an identical argument to $R_{1,j},R_{2,i},R_{2,j}$ shows that $R$ cannot contain any sites other than $s_i,s_j$.
[**\[Points in $Int$ can only be closest to $s_i$ and/or $s_j$\]**]{}. We begin by showing that there is no point $p\in Int$ that is *strictly* closer to a site $s_k\notin \{s_i,s_j\}$ than to any other site ($p\in{\text{Vor}}_k$). If $p\in Int$ is closest to $s_k\notin\{s_i,s_j\}$, then we first show that ${\text{Vor}}_k$ is wholly contained in $Int$. Assume otherwise, and pick a point $q\in{\text{Vor}}_k$ outside $Int$. Since Voronoi regions are path-connected, let $\Gamma_{pq}\subset{\text{Vor}}_k$ be a path connecting $p,q$. By the Jordan curve theorem, $\Gamma_{pq}$ crosses the boundary $\gamma\subset {\text{Vor}}_i\cup {\text{Vor}}_j\cup {\text{Vor}}_{ij}$, contradicting the fact that $\Gamma_{pq}\subset{\text{Vor}}_k$. Since ${\text{Vor}}_k$ is completely inside $Int$ then, by lemma \[lem:site\], it is $s_k\in{\text{Vor}}_k\subset Int$, contradicting the fact the $Int$ contains no sites other than $s_i,s_j$, and therefore ${\text{Vor}}_k\cap Int=\phi$ with $k\notin\{i,j\}$. We now show that no point $p\in Int$ can be closest to $s_k\notin\{s_i,s_j\}$, even if it is also simultaneously closest to $s_i$ and/or $s_j$. Since $p$ is closest to $s_k$, and the boundary of $Int$ is $\gamma\subset {\text{Vor}}_i\cup{\text{Vor}}_j\cup{\text{Vor}}_{ij}$, then $p$ belongs to the interior of $Int$. By definition, $p$ belongs to a Voronoi edge or vertex. If it belongs to a Voronoi vertex and is closest to $s_k\in{S}$ then, by lemma \[lem:vertexincidence\], and since Voronoi vertices are composed of isolated points, $p$ is incident to ${\text{Vor}}_k$, a contradiction since ${\text{Vor}}_k\cap Int=\phi$ whenever $k\notin\{i,j\}$. Therefore $Int$ does not contain any Voronoi vertices.
Finally, we show that no point $p\in Int$ can be closest to a site $s_k\notin\{s_i,s_j\}$ and belong to a Voronoi edge ${\text{Vor}}_E$. Since $p$ is in the interior of $Int$, the connected component ${\text{Vor}}'_E$ of ${\text{Vor}}_E$ that $p$ belongs to must be fully contained in $Int$, or else by the Jordan curve theorem ${\text{Vor}}'_E$ would be separated by the boundary $\gamma\subset {\text{Vor}}_i\cup{\text{Vor}}_j\cup{\text{Vor}}_{ij}$ of $Int$. Since we have discarded connected components of Voronoi edges not incident to any Voronoi vertex, then ${\text{Vor}}'_E$ is incident to some vertex ${\text{Vor}}_I$. Since ${\text{Vor}}'_E$ is in the interior of $Int$, then ${\text{Vor}}_I$ must be contained in $Int$. As we have shown above, $Int$ does not contain any Voronoi vertex, and therefore $p$ cannot be closest to $s_k\notin\{s_i,s_j\}$.
[**\[${\text{Vor}}_{ij}$ is connected\]**]{}. Finally, we show that there is a path in ${\text{Vor}}_{ij}$ connecting ${\text{Vor}}^1_{ij}$ to ${\text{Vor}}^2_{ij}$, and therefore ${\text{Vor}}_{ij}$ is connected. Recall that all points in $Int$ can only be closest to $s_i$ and/or $s_j$, that $\gamma^1,\gamma^2$ are simple paths from $s_i$ to $s_j$, and that, by construction, they do not meet except at their endpoints. Clearly, $\gamma^1,\gamma^2$ are path homotopic [@munkres2000topology p. 323], for instance via the straight-line homotopy.
We begin by constructing a path homotopy $F$ between $\gamma^1$ and $\gamma^2$ (a continuous function $F:[0,1]\times [0,1]\rightarrow \mathbb{R}^2$ such that $F(\cdot,0)=\gamma^1(\cdot)$ and $F(\cdot,1)=\gamma^2(\cdot)$) contained in $Int$. Since $\gamma$ is a Jordan curve, and $Int$ is simply connected, by Carathéodory’s theorem [@conformal], there is a homeomorphism $h$ from $\overline{Int}$ to the closed unit disk $D_2$ that maps $\gamma$ to the unit circle. Since $\gamma^1,\gamma^2\subset\gamma$ and $D_2$ is convex, the straight-line homotopy $F'$ between $h(\gamma^1)$ and $h(\gamma^2)$ is contained in $D_2$. We can now inversely map this homotopy through $h^{-1}$ to obtain a path homotopy $F=h^{-1}\circ F'$ between $\gamma^1$ and $\gamma^2$ which is contained in $Int$ (i.e. $F(\cdot,\alpha)\subset\overline{Int}$ with $0\le\alpha\le 1$).
Since every path $F(\cdot,\alpha)$ with $0\le\alpha\le 1$ starts at $s_i$ and ends at $s_j$, and $D$ is continuous, there is $0< t_\alpha < 1$ such that $F(t_\alpha,\alpha)\in\overline{Int}$ is equidistant to $s_i,s_j$. Since we have shown above that all points in $\overline{Int}$ are closest to $s_i$ and/or $s_j$, then $F(t_\alpha,\alpha)\in{\text{Vor}}_{ij}$ for $0\le\alpha\le 1$. By the continuity of $D$ and $F$, is it possible to choose $t_\alpha$ to be continuous with $\alpha$, and such that the path $\Phi:[0,1]\rightarrow\mathbb{R}^2$ with $\Phi(\alpha)=F(t_\alpha,\alpha)$ is $\Phi([0,1])\subset\overline{Int}\cap{\text{Vor}}_{ij}$. Since the path $\Phi$ is defined to start at ${\text{Vor}}^1_{ij}$ and end at ${\text{Vor}}^2_{ij}$, then ${\text{Vor}}^1_{ij}$ and ${\text{Vor}}^2_{ij}$ are connected, and therefore ${\text{Vor}}_{ij}$ must be connected.
\[lem:SCedges\] Voronoi edges of orphan-free diagrams are simply connected.
Recall that, by the assumption in section \[sec:assumptions\], connected Voronoi edges are also path connected. Let ${\text{Vor}}_{ij}$ be a Voronoi edge, and $\gamma\subset{\text{Vor}}_{ij}$ be a simple path not contractible to a point. By the Jordan curve theorem, $\gamma$ divides the plane into an exterior (unbounded), and an interior (bounded) region $R$. By lemma \[lem:RSC\], $R\subset{\text{Vor}}_{ij}$, and therefore $\gamma$ is contractible to a point.
### Properties of Voronoi regions
\[lem:regionSC\] Voronoi regions of orphan-free diagrams are simply connected.
Let ${\text{Vor}}_i$ be a Vornoi region, which must be connected since the diagram is orphan-free. Since ${\text{Vor}}_i$ is open, it is path connected [@munkres2000topology p. 158].
Assume that ${\text{Vor}}_i$ is not simply connected, and therefore has a closed simple path $\gamma\subset{\text{Vor}}_i$ that is not contractible to a point. By the Jordan curve theorem the path $\gamma$ separates the plane into an exterior and an interior region $R$. By lemma \[lem:RSC\], $R\subset{\text{Vor}}_{i}$, and therefore $\gamma$ is contractible to a point.
\[lem:regionpath\] For every Voronoi region ${\text{Vor}}_i$ of an orphan-free Voronoi diagram, there is a collection of simple paths connecting the site $s_i$ to each point in the boundary of ${\text{Vor}}_i$, such that:
1. all paths are contained in $\overline{{\text{Vor}}_i}$,
2. paths intersect the boundary $\partial{\text{Vor}}_i$ only at the final endpoint, and
3. two paths meet only at the starting point $s_i$.
By the assumption in section \[sec:assumptions\], the boundary of Voronoi regions are simple closed paths. Since a Voronoi region ${\text{Vor}}_i$ is also simply connected (lemma \[lem:regionSC\]), we may use Carathéodory’s theorem [@conformal] to map $\overline{{\text{Vor}}_i}$ to the closed unit disk $D_2$ through a homeomorphism $h$ that maps the boundary $\partial{\text{Vor}}_i$ to the unit circle. Since, by lemma \[lem:site\], $s_i$ is an interior point of ${\text{Vor}}_i$, then $s'_i\equiv h(s_i)$ is an interior point of $D_2$. We now simply construct a set of straight paths from $s'_i$ to each point in the unit circle. These paths are contained in $D_2$, and meet only at the starting point. We map them back through $h^{-1}$ to obtain the desired set of paths.
Voronoi edges are incident to two and only two Voronoi vertices
---------------------------------------------------------------
\[lem:val\_ge2\] No Voronoi edge is incident to just one Voronoi vertex.
Let ${\text{Vor}}_{ij}$ be a Voronoi edge incident to just one Voronoi vertex ${\text{Vor}}_I$. By lemma \[lem:vertexincidence\], it is ${\text{Vor}}_i\rightsquigarrow{\text{Vor}}_{ij}$, and therefore ${\text{Vor}}_{ij}$ has a common boundary with ${\text{Vor}}_i$. Recall from property \[prop:boundaryincidence\] that the boundary $\partial{\text{Vor}}_i$ belongs to Voronoi edges and vertices to which ${\text{Vor}}_i$ is incident. Since, by lemma \[lem:vertexincidence\], Voronoi vertices are isolated points, and two Voronoi edges ${\text{Vor}}_{ij},{\text{Vor}}_{kl}$ can only meet at a Voronoi vertex ${\text{Vor}}_I$ (with $\{i,j,k,l\}\subset I$), we can enumerate an alternating sequence of Voronoi edges and vertices $\left[\dots,{\text{Vor}}_{ij},{\text{Vor}}_I,{\text{Vor}}_{kl},{\text{Vor}}_K,\dots\right]$ in clockwise order around $\partial{\text{Vor}}_i$, in which every edge is incident to the previous and next vertices in the sequence. Therefore, a Voronoi edge can only be incident to one Voronoi vertex if the sequence is $\left[{\text{Vor}}_{ij},{\text{Vor}}_I\right]$.
If ${\text{Vor}}_I$ is not the vertex at infinity, then we can show that the above is not possible with an argument identical to the proof of lemma \[lem:RSC\] (figure \[fig:RSC\]). Note that ${\text{Vor}}_{ij}\rightsquigarrow{\text{Vor}}_I$ implies $\{i,j\}\subset I$, and therefore all points in $\partial{\text{Vor}}_i$ are equidistant to $s_i,s_j$. Let $\gamma\equiv\partial{\text{Vor}}_i$, and consider the ray $r$ from $s_i$ in the direction $s_i-s_j$ which, since $r$ is unbounded and ${\text{Vor}}_i$ is bounded (since it is not incident to ${\text{Vor}}_\infty$), it must cross $\gamma$ at some point $q$. Since $q\in\gamma$, $q$ is equidistant to $s_i,s_j$, contradicting lemma \[lem:midpoint\].
If ${\text{Vor}}_\infty$ is the vertex at infinity, then ${\text{Vor}}_{ij}$ is not incident to any Voronoi vertex, and is unbounded. Therefore, ${\text{Vor}}_{ij}$ does not cross any Voronoi edge, or else ${\text{Vor}}_{ij}$ would be incident to their intersection point (a Voronoi vertex). Recall from lemma \[lem:midpoint\] that ${\text{Vor}}_{ij}$ can never intersect the supporting line $L_{ij}$ of $s_i,s_j$ outside the segment $\overline{s_i,s_j}$. Let $L^i_{ij}$ ($L^j_{ij}$) be the ray starting at $s_i$ ($s_j$) with direction $s_i-s_j$ ($s_j-s_i$), as shown in figure \[fig:connectededges.b\]. It can be easily shown that every point in $L^i_{ij}$ ($L^j_{ij}$) is strictly closer to $s_i$ ($s_j$) than to $s_j$ ($s_i$). Since, regardless of the choice of origin, *every* origin-centered circle $C(\sigma)$ of sufficiently large radius $\sigma$ intersects $L_{ij}$ at exactly one point $p_i$ in $L^i_{ij}$, and one point $p_j$ in $L^j_{ij}$, the following holds. Let $L_{ij}$ divide $\mathbb{R}^2$ into two half spaces $H^{+}_{ij},H^{-}_{ij}$, and let $C^{+}(\sigma)\equiv C(\sigma)\cap H^{+}_{ij}$ and $C^{-}(\sigma)\equiv C(\sigma)\cap H^{-}_{ij}$. Since $p_i$ ($p_j$) is closer to $s_i$ ($s_j$) than to $s_j$ ($s_i$), and $p_i,p_j$ are the endpoints of $C^{+}(\sigma), C^{-}(\sigma)$, by the continuity of $D$, there are points $p^{+}_{ij}\in C^{+}_{ij}$ and $p^{-}_{ij}\in C^{-}_{ij}$ equidistant to $s_i,s_j$. Since ${\text{Vor}}_{ij}$ does not intersect any Voronoi element, then $p^{+}_{ij},p^{-}_{ij}$ are also closest to $s_i,s_j$. Because this holds for all sufficiently large $\sigma$, then both ${\text{Vor}}_{ij}\cap H^{+}_{ij}$ and ${\text{Vor}}_{ij}\cap H^{-}_{ij}$ are unbounded, contradicting lemma \[lem:contrad\], which states that every point $p^{-}_{ij}\in H^{-}_{ij}$ that is sufficiently far from the origin and equidistant to $s_i,s_j$ (and therefore its closest point in $L_{ij}$ lies in $\overline{s_i,s_j}$) is closer to a site in $S\setminus\{s_i,s_j\}$ than to $s_i,s_j$.
\[lem:closureedge\] Let ${\text{Vor}}_{ij}$ be a Voronoi edge. For every $r\in{\text{Vor}}_{ij}$ and $q\in\overline{{\text{Vor}}_{ij}}$ there is a simple path $\gamma:[0,1]\rightarrow\overline{{\text{Vor}}_{ij}}$ such that $\gamma(0)= r$, $\gamma(1)= q$, and $\gamma((0,1))\subset{\text{Vor}}_{ij}$.
[[**\[Case $q\in{\text{Vor}}_{ij}$\]**]{}]{}. Recall that connected components Voronoi edges are assumed to be path-connected (section \[sec:assumptions\]). Since Voronoi edges are connected (lemma \[lem:connectededges\]), they are path-connected. Therefore, if $q\in{\text{Vor}}_{ij}$, there is always a path $\gamma:[0,1]\rightarrow{\text{Vor}}_{ij}$ connecting $r,q$.\
[[**\[Case $q\in\partial{\text{Vor}}_{ij}\setminus{\text{Vor}}_{ij}$\]**]{}]{}. In this case, by property \[prop:boundaryincidence\], $q$ must belong to a Voronoi element of higher order than ${\text{Vor}}_{ij}$ (a Voronoi vertex ${\text{Vor}}_I$), to which ${\text{Vor}}_{ij}$ is incident (with ${\text{Vor}}_{ij}\rightsquigarrow{\text{Vor}}_I$). Since, by lemma \[lem:vertexincidence\], Voronoi vertices are composed of isolated points, then $q$ is a connected component of ${\text{Vor}}_I$ (possibly the vertex at infinity). Consider separately whether $q$ is the vertex at infinity.\
[[**\[Case $q\in\partial{\text{Vor}}_{ij}\setminus{\text{Vor}}_{ij}$ and $q$ is not the vertex at infinity\]**]{}]{}. Recall that the proof of lemma \[lem:vertexincidence\] defines an ordering of $I=i_1,\dots,i_m$, and a set of associated direction vectors $d_1,\dots,d_m$. Let $g_k=\nabla_p{D\left(s_i \parallel p\right)}\big|_{p=q}$, with $k=1,\dots,m$, and let $d_{ij}$ be the unit vector orthogonal to $g_i-g_j$ in the direction outgoing from ${\text{conv} \{ }-g_1,\dots,-g_m\}$ (which exists since, by assumption \[ass:EGA\], it is $g_i\neq g_j$). We assume, without loss of generality, that the coordinate representation of $d_{ij}$ is $[\left(d_{ij}\right)_x, \left(d_{ij}\right)_y] = [1,0]$. Since $D\in\mathcal{C}^1$ and $g_i\ne g_j$, by the implicit function theorem, there is an open $L_2$ ball $B_2(q;\xi)$ around $q$ in which the implicit equation $ {D\left(s_i \parallel p\right)} = {D\left(s_j \parallel p\right)} $ can be written as $y=f(x)$, with $f'(0)=0$, as shown in figure \[fig:closureedge.a\].
Since ${\text{Vor}}_{ij}$ is incident to ${\text{Vor}}_I$ at $q$, there is $k\in\{1,\dots,m\}$ such that $i=i_k, j=i_{k\oplus 1}$. Choose $0 < \mu < \xi$ to be sufficiently small for the conditions of the proof of lemma \[lem:vertexincidence\] to apply (in particular $\mu < \min\{\delta,\varepsilon,\xi\}$, as defined in the proof). Let $W$ be a circular wedge contained in the $L_2$ ball $B_2(q;\mu)$, and bounded by the rays $q+\mu d_k$ and $q+\mu d_{k\oplus 1}$ which, aside from $q$, only contains points strictly closer to $\{s_i,s_j\}$ than to all other sites. From the definition of $d_k,d_{k\oplus 1}$ it is clear that the segment $\overline{q, q+\mu d_{ij}}$ is contained in $W$.
Since $\mu<\xi$, and inside $W$ all points with the exception of $q$ are closest only to $s_i,s_j$, the implicit equation $ {D\left(s_i \parallel p\right)} = {D\left(s_j \parallel p\right)} $ represents the set of points in $W\cap {\text{Vor}}_{ij}$. Since $ {D\left(s_i \parallel p\right)} = {D\left(s_j \parallel p\right)} $ can be written in coordinates as $y=f(x)$ inside $W$, it is clear that, inside $W$, ${\text{Vor}}_{ij}$ is a simple curve, and that this is the only part of ${\text{Vor}}_{ij}$ incident to $q$. Given $r\in{\text{Vor}}_{ij}$, find any point $v\in W\cap{\text{Vor}}_{ij}$ that is closer to $q$ than $r$. Because $v\in{\text{Vor}}_{ij}$, there is a simple path $\gamma_1\subset{\text{Vor}}_{ij}$ connecting $r$ to $v$ and, because $v$ is in $W$, there is also a simple path $\gamma_2\subset{\text{Vor}}_{ij}$ from $v$ to $q$ (part of the curve $y=f(x)$ of figure \[fig:closureedge\]). Finally, because $v$ is closer to $q$ than $r$ is, the paths $\gamma_1$ and $\gamma_2$ do not cross, and therefore the concatenation of $\gamma_1$ and $\gamma_2$ meets the requirements of the lemma.
[[**\[Case $q\in\partial{\text{Vor}}_{ij}\setminus{\text{Vor}}_{ij}$ and $q$ is the vertex at infinity\]**]{}]{}. Since ${\text{Vor}}_{ij}\rightsquigarrow{\text{Vor}}_{\infty}$ then, by definition, ${\text{Vor}}_{ij}$ is unbounded. Let $r_0\equiv r$ and, for each $k\in\mathbb{N}$, let $r_k\in{\text{Vor}}_{ij}$ be at distance $\|r-r_k\|_2=k$. One can always find such a sequence of points because ${\text{Vor}}_{ij}$ is unbounded and path-connected (if there is no $r_k\in{\text{Vor}}_{ij}$ at distance $\|r-r_k\|_2=k$ then the circle with center at $r$ and radius $k$ would disconnect ${\text{Vor}}_{ij}$). Let $\gamma'_k:[0,1]\rightarrow{\text{Vor}}_{ij}$ be paths connecting $r_{k-1}$ to $r_k$, and $\gamma':\mathbb{R}^+\rightarrow{\text{Vor}}_{ij}$ be the concatenation of $\gamma'_1,\gamma'_2,\dots$, where $\gamma'(k+t)\equiv\gamma'_k(t)$, with $k\in\mathbb{N}$ and $t\in[0,1]$.
Define $\gamma:[0,1]\rightarrow{\text{Vor}}_{ij}\cup{\text{Vor}}_\infty$ as $\gamma(t) \equiv \gamma'(1/(1-t))$. Consider $\gamma$ on the Riemann sphere, transformed through a stereographic projection. Since $\gamma'$ is continuous and $\gamma$ has an accumulation point at the point at infinity (north pole on the sphere), it is continuous on the sphere. If $\gamma$ is not simple, it can be appropriately cut and reparametrized until it is (i.e. by tracing the path and, upon arrival to a point $c$ where the path crosses itself, cutting out the next portion up to the highest $t$ for which $\gamma(t)=c$, and proceeding this way to the end of the path).
Note that for lemma \[lem:closureedge\] to hold it is crucial that edges ${\text{Vor}}_{ij}$ are incident to vertices ${\text{Vor}}_I$ as a curve arriving at $v\in{\text{Vor}}_I$ from a single direction, as illustrated in figure \[fig:closureedge.a\]. To see that assumption \[ass:EGA\] is required, consider figure \[fig:closureedge.b\], which depicts an edge ${\text{Vor}}_{ij}$ incident to two vertices ${\text{Vor}}_I,{\text{Vor}}_J$ which do not satisfy assumption \[ass:EGA\], in which *every* path connecting the two disks passes through either ${\text{Vor}}_I$ or ${\text{Vor}}_J$, and therefore for which lemma \[lem:closureedge\] does not hold.
\[lem:tree\] In an orphan-free diagram, for every Voronoi edge ${\text{Vor}}_{ij}$ that is incident to Voronoi vertices ${\text{Vor}}_{I_1},\dots,{\text{Vor}}_{I_m}$, there is an embedded tree graph in $\overline{{\text{Vor}}_{ij}}$ whose leafs are ${\text{Vor}}_{I_1},\dots,{\text{Vor}}_{I_m}$.
Unless otherwise specified, we assume in this proof that all paths are simple, contained in $\overline{{\text{Vor}}_{ij}}$, parametrized over the unit interval $[0,1]$, and that, using lemma \[lem:closureedge\], there is a path connecting any two points in $\overline{{\text{Vor}}_{ij}}$ that does not intersect a Voronoi vertex (expect perhaps at the endpoints). We use throughout the fact that Voronoi edges are path connected (lemma \[lem:SCedges\] and section \[sec:assumptions\]).
If $m=1$, pick a point $r\in{\text{Vor}}_{ij}$ as root and, using lemma \[lem:regionpath\], consider a simple path $\gamma_{r,1}\subset\overline{{\text{Vor}}_{ij}}$ connecting $r$ to ${\text{Vor}}_{I_1}$, then the tree with vertex set $V=\{r,{\text{Vor}}_{I_1}\}$, and edge set $E=\{\gamma_{r,1}\}$ meets the requirements of the lemma.
For each $k\ge 2$, assume that there is an embedded tree graph $T_{k-1}\subset\overline{{\text{Vor}}_{ij}}$ with ${\text{Vor}}_{I_1},\dots,{\text{Vor}}_{I_{k-1}}$ as leafs. We construct a new embedded tree $T_k$ as follows (figure \[fig:tree\]). Let $r\in{\text{Vor}}_{ij}$ be the root of $T_{k-1}$, and let $\gamma$ be a simple path connecting $r$ to ${\text{Vor}}_{I_k}$ which, making use of lemma \[lem:closureedge\], is chosen such that it does not intersect any Voronoi vertex (other than the final endpoint). Let $$t_k \equiv \max \{t\in [0,1] ~:~ \gamma(t)\in T_{k-1} \},$$ which always exists because $T_{k-1}$ is closed and $\gamma(0)=r\in T_{k-1}$. Let $v_k\equiv \gamma(t_k)$ be the “last" point along $\gamma$ that belongs to $T_{k-1}$. Because $\gamma(1)={\text{Vor}}_{I_k}\notin T_{k-1}$ then it must be $t_k <1$. Additionally, $v_k$ cannot be a Voronoi vertex, since $\gamma$ doesn’t intersect Voronoi vertices except at the final endpoint ${\text{Vor}}_{I_k}$.
Let $\gamma_k$ be the path $\{\gamma(t) ~:~ t\in [t_k,1]\}$, that is, the part of $\gamma$ from $v_k$ to ${\text{Vor}}_{I_k}$. We construct a new tree graph $T_k\subset\overline{{\text{Vor}}_{ij}}$ as follows. Begin by setting $T_k$ equal to $T_{k-1}$. We then insert a new vertex ${\text{Vor}}_{I_k}$ into $T_k$. Next, we proceed differently depending on whether $v_k\in T_{k-1}$ is a vertex, or it belongs to an edge of $T_{k-1}$ (note that, since $v_k$ is not a Voronoi vertex, it cannot be a leaf vertex of $T_{k-1}$).
If $v_k$ is an internal vertex of $T_{k-1}$, as in figure \[fig:tree.a\], then we add a new edge $\gamma_k$ to $T_k$ connecting vertices $v_k$ and ${\text{Vor}}_{I_k}$. Since, by construction, $\gamma_k$ does not cross any edge in $T_k$, the tree graph remains embedded.
If, on the other hand, $v_k$ belongs to an edge $e$ of $T_{k-1}$ connecting vertices $v_1,v_2$, as shown in figure \[fig:tree.b\], then:
1. we insert a new (internal) vertex $v_k$ into $T_k$;
2. we split $e$ into two edges: $e_1$ and $e_2$, connecting $v_1,v_k$, and $v_k,v_2$, respectively;
3. we insert a new edge $\gamma_k$ connecting vertices $v_k$ and ${\text{Vor}}_{I_k}$.
Note that the edge $e$ is split into two edges that represent the same set of points, and therefore, since $\gamma_k$ didn’t cross any edges of $T_{k-1}$, then $\gamma_k$ does not cross any edge of $T_k$. Hence, since $T_{k-1}$ is an embedded tree graph, the new tree $T_k$ is also embedded and has ${\text{Vor}}_{I_1},\dots,{\text{Vor}}_{I_k}$ as leafs.
The lemma follows by induction on $m$.
![By assuming that a Voronoi edge ${\text{Vor}}_{ij}$ is incident to three Voronoi vertices ${\text{Vor}}_{I_1},{\text{Vor}}_{I_2},{\text{Vor}}_{I_3}$, we can construct a planar embedding of the non-planar graph $K_{3,3}$, a contradiction. The more general figure \[fig:planarity.a\] further illustrates the proof of lemma \[lem:val\_le2\]. []{data-label="fig:K33"}](K3_graph2.pdf){width="3.5in"}
The final lemma of this section can be used in conjunction with lemma \[lem:val\_ge2\] to establish that Voronoi edges are incident to exactly two Voronoi vertices. We sketch here the argument that shows that a Voronoi edge ${\text{Vor}}_{ij}$ cannot be incident to three vertices ${\text{Vor}}_{I_1},{\text{Vor}}_{I_2},{\text{Vor}}_{I_3}$ (figure \[fig:K33\]). The general case in the proof of lemma \[lem:val\_le2\] follows a similar argument. We first use lemma \[lem:tree\] to build a tree inside ${\text{Vor}}_{ij}$ with leafs at ${\text{Vor}}_{I_1},{\text{Vor}}_{I_2},{\text{Vor}}_{I_3}$, and show that it can be collapsed into a star-graph with a vertex $r\in{\text{Vor}}_{ij}$, and non-crossing edges $(r,{\text{Vor}}_{I_1}), (r,{\text{Vor}}_{I_2}), (r,{\text{Vor}}_{I_3})$, as shown in the figure. The incidence rules of lemma \[lem:vertexincidence\], as well as lemma \[lem:regionpath\] allows us to construct six non-crossing edges from $s_i$ and $s_j$, to ${\text{Vor}}_{I_1},{\text{Vor}}_{I_2},{\text{Vor}}_{I_3}$, respectively. We have just constructed an embedding of a graph which can be easily shown to be the non-planar graph $K_{3,3}$, thereby reaching a contradiction.
\[lem:val\_le2\] Voronoi edges of an orphan-free diagram are incident to no more than two Voronoi vertices.
Let ${\text{Vor}}_{ij}$ be a Voronoi edge incident to Voronoi vertices ${\text{Vor}}_{I_1},\dots,{\text{Vor}}_{I_m}$. Since ${\text{Vor}}_{ij} \rightsquigarrow {\text{Vor}}_{I_1},\dots,{\text{Vor}}_{I_m}$, and Voronoi vertices are of higher order ($|I_k| > 2$) than Voronoi edges, by the definition of incidence (definition \[def:incidence\]), it is $\{i,j\} \subset {\text{Vor}}_{I_k}$, with $k=1,\dots,m$. We prove the lemma on the sphere $\mathbb{S}^2$, where any of the Voronoi vertices may be the vertex at infinity. Note also that some of the sets $I_k$ with $k=1,\dots,m$ may be equal, since Voronoi vertices have not yet been shown to be connected.
By lemma \[lem:vertexincidence\], the vertices ${\text{Vor}}_{I_k}$ are isolated points (possibly the point at infinity), and ${\text{Vor}}_i,{\text{Vor}}_j\rightsquigarrow {\text{Vor}}_{I_1},{\text{Vor}}_{I_2},\dots,{\text{Vor}}_{I_m}$. We begin by assuming that $m>2$, and build an embedded planar graph $G_m$ (figure \[fig:planarity.a\]). We then show that $G_m$ can only be planar if $m\le 2$, reaching a contradiction.
By lemma \[lem:tree\], there is an embedded tree graph $T_m\subset\overline{{\text{Vor}}_{ij}}$ with ${\text{Vor}}_{I_1},\dots,{\text{Vor}}_{I_m}$ as leafs. We begin by setting $G_m$ equal to $T_m$. We then insert the vertices $s_i$ and $s_j$ in $G_m$ (as shown in figure \[fig:planarity.a\]). Since ${\text{Vor}}_i,{\text{Vor}}_j \rightsquigarrow {\text{Vor}}_{I_1},{\text{Vor}}_{I_2},\dots,{\text{Vor}}_{I_m}$, by lemma \[lem:regionpath\], there are non-crossing paths $\gamma_{i,k}\subset\overline{{\text{Vor}}_i}$, with $k=1,\dots,m$, connecting $s_i$ to ${\text{Vor}}_{I_k}$ and non-crossing paths $\gamma_{j,k}\subset\overline{{\text{Vor}}_j}$, with $k=1,\dots,m$, connecting $s_j$ to ${\text{Vor}}_{I_k}$. We insert the above paths $\gamma_{i,k},\gamma_{j,k}$, $k=1,\dots,m$, as edges of $G_m$. Aside from all paths $\gamma_{i,k}$ ($\gamma_{j,k}$) only crossing at their starting point, all paths $\gamma_{i,k}$ ($\gamma_{j,k}$) are, by lemma \[lem:regionpath\], contained (except for their final endpoint) in the interior of ${\text{Vor}}_i$ (${\text{Vor}}_j$), and therefore they can only cross an edge of $T_m\subset\overline{{\text{Vor}}_{ij}}$ at an endpoint. $G_m$ is therefore embedded in $\mathbb{S}^2$, and so it is a planar graph.
Recall that the minors of a graph are obtained by erasing vertices, erasing edges, or contracting edges, and that minors of planar graphs are themselves planar [@bondy2008graph p. 269]. We now construct an appropriate minor $G'_m$ of the planar graph $G_m$, shown in figure \[fig:planarity.b\], and prove that it is non-planar whenever $m > 2$, creating a contradiction.
Clearly, every tree $T_m$ satisfying the conditions of lemma \[lem:tree\] has a minor $T'_m$ directly connecting the root to each leaf ${\text{Vor}}_{I_k}$, $k=1,\dots,m$ (see figure \[fig:planarity.b\]), which is obtained by successively contracting every edge of $T_m$ that connects two internal vertices. We apply the same sequence of edge contractions to obtain $G'_m$ from $G_m$, as shown in figure \[fig:planarity\].
Let $r'$ be the root of $T'_m$, and $\gamma_{r',k}$ be edges from $r'$ to ${\text{Vor}}_{I_k}$, with $k=1,\dots,m$. The minor $G'_m$ has vertex set $$V=\{s_i,s_j,r', ~{\text{Vor}}_{I_1},\dots,{\text{Vor}}_{I_m}\},$$ and edge set $$E=\{ \gamma_{i,1},\dots,\gamma_{i,m}, ~ \gamma_{j,1},\dots,\gamma_{j,m}, ~ \gamma_{r',1},\dots,\gamma_{r',m}\},$$ and therefore $G'_m$ has $v=m+3$ vertices and $e=3m$ edges. Since (as is easily verified) every cycle in $G'_m$ has length four or more, and $G'_m$ is planar, then it holds $2e \ge 4f$, where $f$ is the number of faces. Using Euler’s identity for planar graphs, $v-e+f=2$ [@bondy2008graph], and the fact that $2e \ge 4f$, $v=m+3$, and $e=3m$, it follows that $m \le 2$, and therefore $G'_m$ is not planar whenever $m> 2$ (for instance, $G'_3$ is the utility graph $K_{3,3}$).
Since $m> 2$ leads to a contradiction, it follows that every Voronoi edge is incident to at most two Voronoi vertices.
Primal Voronoi graph and dual Delaunay triangulation {#primaldual}
----------------------------------------------------
We use the results in this section to construct a graph from the incidence relations of an orphan-free Voronoi diagram, and dualize it into a planar embedded graph.
Let the *primal Voronoi graph* $\tilde{P}=(\tilde{P}_V, \tilde{P}_E)$ of an orphan-free Voronoi diagram be defined as follows. The vertices $\tilde{P}_V$ are the connected components of Voronoi vertices. Since, by lemma \[lem:vertexincidence\], Voronoi vertices are composed of isolated points, then $\tilde{P}_V$ is a collection of isolated points. By lemmas \[lem:val\_ge2\] and \[lem:val\_le2\], Voronoi edges that are incident to some Voronoi vertex are incident to exactly two Voronoi vertices. For each Voronoi edge ${\text{Vor}}_{ij}$ incident to some Voronoi vertex, we include in $\tilde{P}_E$ an edge connecting the vertices in $\tilde{P}_V$ corresponding to the connected components of Voronoi vertices that ${\text{Vor}}_{ij}$ is incident to. By lemma \[lem:closureedge\], for each such Voronoi edge ${\text{Vor}}_{ij}$ there is a simple path in ${\text{Vor}}_{ij}$ connecting the two Voronoi vertices incident to ${\text{Vor}}_{ij}$, and therefore $\tilde{P}$ is an embedded planar graph.
[**Theorem \[th:simpleplanar\]**]{}. *Let $\tilde{G}=(\tilde{V}, \tilde{E})$ be the dual of the primal Voronoi graph corresponding to an orphan-free Voronoi diagram, then $\tilde{G}$ is a simple, connected, planar graph.*
The dual graph $\tilde{G}$ is constructed by dualizing $\tilde{P}$ and using the natural embedding described in [@bondy2008graph p. 252], in which dual vertices are placed inside primal faces (at the sites in this case), and dual edges cross once their corresponding primal edges. From this construction, $\tilde{G}$ is an embedded planar graph [@bondy2008graph p. 252], and is connected by virtue of being the dual of a planar graph [@bondy2008graph p. 253].
We show that $\tilde{G}$ is simple (edges have multiplicity one, and there are no loops: edges incident to the same vertex). Edges of $\tilde{G}$ are one-to-one with edges of $\tilde{P}$. In turn, edges of $\tilde{P}$ correspond to Voronoi edges, and these are, by lemma \[lem:connectededges\], connected. Therefore the edges of $\tilde{G}$ have multiplicity one.
Since loops and cut edges (those whose removal disconnects the graph) are duals of each other [@bondy2008graph p. 252], we now show that $\tilde{P}$ has no cut edges, and therefore $\tilde{G}$ has no loops.
By [@bondy2008graph p. 86], an edge of $\tilde{P}$ is a cut edge iff it belongs to no cycle of $\tilde{P}$. To every edge of $\tilde{P}$ corresponds an Voronoi edge ${\text{Vor}}_{ij}$ that is incident to two Voronoi vertices. By lemma \[lem:vertexincidence\], ${\text{Vor}}_{ij}$ is incident to at least one Voronoi region ${\text{Vor}}_i$. We next show that the Voronoi elements in the boundary of every Voronoi region ${\text{Vor}}_i$ form a cycle, and therefore ${\text{Vor}}_{ij}$ belongs to a cycle, so it cannot be a cut edge.
Clearly, the boundary $\partial {\text{Vor}}_i$ of ${\text{Vor}}_i$ is composed of Voronoi edges and Voronoi vertices, since ${\text{Vor}}_i\rightsquigarrow{\text{Vor}}_j$ is not possible because $\{i\}\not\subset\{j\}$ (see section \[sec:incidence\]). Let $C=\left[{\text{Vor}}_{i,j_1},{\text{Vor}}_{i,j_1,j_2}, {\text{Vor}}_{i,j_2}, \dots\right]$ be the sequence of elements around the boundary of ${\text{Vor}}_i$, with ${\text{Vor}}_{ij}\in C$. We show that $C$ is a cycle.
[**\[$C$ has no repeated Voronoi vertices\]**]{}. By the assumption of section \[sec:assumptions\], Voronoi regions have boundaries that are simple closed curves (in $\mathbb{S}^2$). Note that, because vertices are isolated points, there are no repeated vertices in $C$ since the boundary of ${\text{Vor}}_i$ is a simple curve. [**\[$C$ has no repeated Voronoi edges\]**]{}.\
Let ${\text{Vor}}_{ij}$ appear twice in $C$ as $\left[\dots, {\text{Vor}}_I, {\text{Vor}}_{ij}, {\text{Vor}}_J, \dots, {\text{Vor}}_{ij}, \dots\right]$, where ${\text{Vor}}_I,{\text{Vor}}_J$ are Voronoi vertices, as in figure \[fig:planar.a\]. Let $p,q$ be two points in each of the two common boundaries between ${\text{Vor}}_i$ and ${\text{Vor}}_{ij}$. By lemma \[lem:regionpath\], there are simple paths $\gamma_{i,p},\gamma_{i,q}\subset{\text{Vor}}_i$ from $s_i$ to $p,q$, respectively, which only meet at the initial endpoint (figure \[fig:planar.a\]). Since ${\text{Vor}}_{ij}$ is simply connected, we can consider a simple path $\gamma_{ij}\subset{\text{Vor}}_{ij}$ connecting $p,q$. Let $\gamma$ be the simple closed path obtained by concatenating $\gamma_{i,p},\gamma_{ij},\gamma_{i,q}$ which, by the Jordan curve theorem divides the plane into a bounded region $Int$, and an unbounded region. Since it must be ${\text{Vor}}_I\in Int$ or ${\text{Vor}}_J\in Int$, assume without loss of generality that ${\text{Vor}}_I\in Int$, and note that it cannot be ${\text{Vor}}_I={\text{Vor}}_\infty\in Int$, since $Int$ is bounded. We show that ${\text{Vor}}_I\in Int$ is not possible, and therefore that $C$ has no repeated elements.
Let ${\text{Vor}}_I\in Int$ and let $k\in I$ be $k\ne i,j$, which always exists because $|I|\ge 3$. By lemma \[lem:vertexincidence\], there is a point $p\in{\text{Vor}}_k\cap Int$. Since ${\text{Vor}}_k$ is path connected, and the boundary of $Int$ is $\gamma\subset{\text{Vor}}_{i}\cup{\text{Vor}}_{ij}$, then ${\text{Vor}}_k\subset Int$, and therefore $s_k\in Int$. We show that $Int$ cannot contain any sites other than $s_i$, reaching a contradiction.
Recall that the boundary $\gamma$ of $Int$ is the concatenation of $\gamma_{i,p}\subset{\text{Vor}}_i$, $\gamma_{ij}\subset{\text{Vor}}_{ij}$, and $\gamma_{i,q}\subset{\text{Vor}}_i$, and that $s_i\in\gamma$, as in figure \[fig:planar.b\]. Let $Int'$ be the union of segments from $s_i$ to every point in $\gamma$: $$Int'\equiv \displaystyle{\left(\cup_{r\in\gamma_{i,p}} \overline{s_i, r}\right) \cup
\left(\cup_{r\in\gamma_{ij}} \overline{s_i, r}\right) \cup
\left(\cup_{r\in\gamma_{i,q}}\overline{s_i,r}\right)}.$$ Since it is clearly $Int\subset Int'$, it suffices to show that $Int'$ does not contain any site $s_k$ different from $s_i$. Every segment of the form $\overline{s_i,r}$ with $r\in\gamma_{i,p}\subset{\text{Vor}}_i$ or $r\in\gamma_{i,q}\subset{\text{Vor}}_i$ cannot contain a site $s_k$ or else, by the convexity of $D$, $r$ would be closer to $s_k$ than to $s_i$. Similarly, every segment of the form $\overline{s_i,r}$ with $r\in\gamma_{ij}$ cannot contain a site $s_k$, or else by the convexity of $D$, $r$ would be closer to $s_k$ than to $s_i,s_j$.
Since every Voronoi edge ${\text{Vor}}_{ij}$ is part of a cycle, it cannot be a cut edge, and therefore its dual has no loops.
Embeddability of the Delaunay triangulation {#sec:dual}
===========================================
Let $\tilde{G}=(\tilde{V}, \tilde{E})$ be the dual of the primal Voronoi graph corresponding to an orphan-free Voronoi diagram, as defined in section \[sec:planar\]. By theorem \[th:simpleplanar\], $\tilde{G}$ is simple and planar with vertices at the sites. Let $G=({S},E,F)$ be the planar graph obtained by replacing curved edges by straight segments. Recall from section \[sec:planar\] that, while Voronoi regions and edges are connected, Voronoi vertices may have multiple connected components, and therefore $G$ can have duplicate faces in $F$. We only show after this section that faces have multiplicity one by virtue of $G$ being embedded.
[**Faces with more than three vertices**]{}. Every face $f\in F$ is dual to a Voronoi element ${\text{Vor}}_I$ of order $|I|=k\ge 3$, to which corresponds (proposition \[prop:ECB\]) a convex ball $B(c;\rho)$, with $c\in{\text{Vor}}_I$, that circumscribes the sites $(s_i)_{i\in I}$ incident to $f$. Due to the planarity of $G$, we can assume the sites $(s_i)_{i\in I}$ to be ordered around $f$. In order to find whether a point $p\in\mathbb{R}^2$ belongs to $f$, we simply triangulate $f$ in a fan arrangement: $\tau_1=\{s_{i_1},s_{i_2},s_{i_k}\}; \tau_2=\{s_{i_2},s_{i_3},s_{i_k}\}; \dots$, and consider that $p\in f$ iff it lies in any of the resulting $\tau_j$. Note that this arrangement does not interfere with the original edges in $E$ (other than creating new ones), all new edges are incident to two faces (they are not in the topological boundary of $G$), and most importantly, every $\tau_j$, with $j=1,\dots,k-2$ satisfies the empty circum-ball property with the same [witness]{} ball $B(c;\rho)$ as $f$. We assume in the sequel that $G$ has been triangulated in this way. The fact that this triangulated $G$ will be shown to be embedded will imply that every face $f$ is in fact convex.
For convenience in the remainder of this section we name $W=\{w_i\in {S}: i=1,\dots,m\}$ the sites that are part of the boundary of the convex hull ${{conv}\left\{{S}\right\}}$, and order them in clock-wise order around ${{conv}\left\{{S}\right\}}$.
Boundary {#sec:boundary}
--------
In this section, we assume that the divergence $D$ satisfies the bounded anisotropy assumption \[ass:BAA\], and conclude that the boundary of the dual triangulation of an orphan-free diagram is the same as the boundary of the convex hull of the sites (and in particular it is simple and closed).
The vertices in the *topological boundary* of $G$ are those whose corresponding primal regions are unbounded, while topological boundary edges are those connecting topological boundary vertices. For convenience, we call $B\subseteq E$ the set of topological boundary edges of $G$.
The boundary $\mathcal{B}$ of the convex hull is a simple circular chain $\mathcal{B} = \{(w_i, w_{i\oplus 1}) : i=1,\dots,m\}$. We prove that it is $B=\mathcal{B}$ (loosely speaking: the topological, and geometric boundaries of $G$ are the same and coincide with the boundary of ${{conv}\left\{{S}\right\}}$), which implies that $G$ covers the convex hull of the sites, and its topological boundary edges form a simple, closed polygonal chain. All the proofs of this section are in Appendix B.
\[boundary\_easy\] To every topological boundary edge of $G$ corresponds a segment in the boundary of ${{conv}\left\{{S}\right\}}$.
We now turn to the converse claim: that to every segment in $\mathcal{B}$ corresponds one in $B$. Since $B$ is the set of boundary edges of $G$, whose corresponding primal edges are unbounded, the claim is equivalent to proving that, to every segment in $\mathcal{B}$ corresponds a boundary edge $(w_i,w_j)\in E$ of $G$ whose corresponding primal edge ${\text{Vor}}_{ij}$ is unbounded.
The proof proceeds as follows. First, assume without loss of generality that the origin is in the interior of ${{conv}\left\{{S}\right\}}$. Let $C(\sigma)=\{x\in\mathbb{R}^2 : \|x\|=\sigma\}$ be an origin-centered circle of radius $\sigma$ large enough so that lemmas \[lem:VW\] and \[lem:contrad\] hold in $C(\sigma)$. We define two functions: $$\begin{aligned}
\label{eq:pinu}
&\nu_\sigma &: \partial{{conv}\left\{{S}\right\}}\rightarrow C(\sigma), ~~~~~~~~~~~~~\nu_\sigma: r\mapsto \sigma \cdot r / \|r\|, \\
&\pi &: C(\sigma)\rightarrow\partial{{conv}\left\{{S}\right\}}, $$ $\nu_\sigma$ simply projects points in the boundary of ${{conv}\left\{{S}\right\}}$ out to their closest point in $C(\sigma)$ (using the natural metric; note that $\sigma$ can always be chosen large enough so this projection is unique). $\pi$ is constructed as follows.
![The construction of the projection function $\pi:C(\sigma)\rightarrow \partial{{conv}\left\{{S}\right\}}$. Note that in this case the region-to-site-index function is simply $j(i)=i$, but this cannot be assumed in general. []{data-label="fig:pi"}](pi.pdf){width="2.7in"}
\[text:boundary\] Consider the situation illustrated in figure \[fig:pi\]. By lemma \[lem:VW\], all points in $C(\sigma)$ are closer to $W$ than to any interior site ${S}\setminus W$. We split $C(\sigma)$ into a sequence $(C_j)$ of *connected* parts closest to the same boundary site $w_{i(j)}$ (the function $i(\cdot)$ is used to map part indices to the index of their closest site). By the convexity of balls, adjacent regions *must* be closest to (circularly) consecutive sites in $W$ (e.g. if regions $C_1,C_2$ had $i(1)=1$ and $i(2)=3$, by the continuity of $D$, the point $p$ where $C_1,C_2$ meet would be closest to $w_1,w_3$; however, since the sites $w_i$ are in cyclic order around $\partial{{conv}\left\{{S}\right\}}$, $p$ would be closer to $w_2$ than to $w_1,w_3$, a contradiction). Pick one point $p_j$ for each region $C_j$, and let $\pi(p_j) \equiv w_{i(j)}$. For each pair of consecutive regions $C_j, C_{j\oplus 1}$ meeting at $p_{j,j\oplus 1}$, let $\pi(p_{j,j\oplus 1}) \equiv (w_{i(j)} + w_{i(j\oplus 1)})/2$ (the midpoint of two consecutive boundary sites). The remaining values of $\pi$ are filled using simple linear interpolation. By construction, the following holds:
\[prop:pi\] $i\cdot$ $\pi:C(\sigma)\rightarrow\partial{{conv}\left\{{S}\right\}}$ is continuous.\
$ii\cdot$ Given $p\in C(\sigma)$ and consecutive boundary sites $w_i, w_{i\oplus 1}$, then $p\in{\text{Vor}}_{i, i\oplus 1}$ iff $\pi(p)=(w_i + w_{i\oplus 1})/2$.
By the convexity of ${{conv}\left\{{S}\right\}}$, $\nu_\sigma$ is continuous in $\partial{{conv}\left\{{S}\right\}}$. Note that, because ${{conv}\left\{{S}\right\}}$ is assumed to contain the origin then, as shown in figure \[fig:pinu\], $\nu_\sigma$ projects every point $\pi(p)\in (w_i,w_j)$ lying on a segment of $\partial{{conv}\left\{{S}\right\}}$, *outwards* from the convex hull (and on the *empty* side of $(w_i,w_j)$); that is, so that $\nu_\sigma(\pi(p))\in H^{+}_{ij}\cap C(\sigma)$ (i.e. $\nu_\sigma(\pi(p))$ is in the empty half-space of $(w_i,w_j)$).
![The construction for the proof of lemma \[lem:hard\].[]{data-label="fig:pinu"}](pinu.eps){width="2.7in"}
The claim now reduces to showing that for each segment $(w_i,w_j)$ of $\partial{{conv}\left\{{S}\right\}}$, and for *every* sufficiently large $\sigma$, there is $p\in C(\sigma)$ with $p\in{\text{Vor}}_{ij}$ (i.e. $\pi(p)=(w_i+w_j)/2$). Since this implies that ${\text{Vor}}_{ij}$ is unbounded, it means that the corresponding edge $(w_i,w_j)$ is in $B$ (the topological boundary of $G$).
The proof is by contradiction. Lemma \[lem:Sn\] uses Brouwer’s fixed point theorem to show that, for every segment $(w_i,w_j)$ of $\mathcal{B}$, if there were no $p\in C(\sigma)$ closest to $w_i,w_j$, then the function $\nu_\sigma\circ\pi:C(\sigma)\rightarrow C(\sigma)$ (in fact a slightly different but related function) would have a point $q\in C(\sigma)$ such that $\nu_\sigma(\pi(q))=-q$, that is, such that $q$ is “behind" the segment $(w_i,w_j)\in\partial{{conv}\left\{{S}\right\}}$ to which it is closest ($q\in H^{-}_{ij}$). On the other hand, lemma \[lem:contrad\] shows that, for all sufficiently large circles $C$, no point $q\in C(\sigma)$ can be closest to a segment $(w_i,w_j)\in\partial{{conv}\left\{{S}\right\}}$ it is *behind* of, creating a contradiction.
The next Lemma is used to create a contradiction, and relies on assumption \[ass:BAA\]. Lemma \[lem:Sn\] is the key lemma in this section, and is a simple application of Brouwer’s fixed point theorem.
\[lem:contrad\] There is $\rho > 0$ such that, for any segment $(w_i,w_j)\in\mathcal{B}$ with supporting line $L_{ij}$, every $p\in H^{-}_{ij}$ with $\|p\| > \rho$ whose closest point in $L_{ij}$ belongs to $\overline{w_i w_j}$ is closer to a site in ${S}\setminus\{w_i,w_j\}$ than to $L_{ij}$.
\[lem:Sn\] Every continuous function $F:\mathbb{S}^n\rightarrow\mathbb{S}^n$ that is not onto has a fixed point.
\[lem:hard\] To every segment in the boundary of ${{conv}\left\{{S}\right\}}$ corresponds a boundary edge of $G$.
Finally, since we have shown that the topological boundary of the dual triangulation is the same as the boundary of the convex hull of the sites, we can conclude that:
\[cor:boundary\] The topological boundary of the dual of an orphan-free Voronoi diagram is the boundary of the convex hull ${{conv}\left\{{S}\right\}}$, and is therefore simple and closed.
Interior {#sec:interior}
--------
This section concludes the proof of Theorem \[th:main\] by showing that, if the topological boundary of $G$ is simple and closed, then $G$ must be embedded. The main argument in the proof uses proposition \[prop:ECB\] and \[cor:boundary\], as well as the theory of discrete one-forms on graphs, to show that there are no “edge fold-overs" in $G$ (edges whose two incident faces are on the same side of its supporting line), and uses this to conclude that the interior of $G$ is a single “flat sheet", and therefore it is embedded.
The following definition, from [@1form], assumes that, for each edge $(s_i,s_j)\in E$ of $G$, we distinguish the two opposing half-edges $(s_i,s_j)$ and $(s_j,s_i)$.
\[def:1form\] A non-vanishing (discrete) one-form $\xi$ is an assignment of a real value $\xi_{ij} \neq 0$ to each half edge $(s_i,s_j)$ in $G$, such that $\xi_{ji} = -\xi_{ij}$.
We can construct a non-vanishing one-form over $G$ as follows. Given some unit direction vector $n\in\mathbb{S}^1$ (in coordinates $n=\left[n_1,n_2\right]^t$), we assign a real value $z(v) = n^t v$ to each vertex $v$ in $G$, and define $\xi_{ij} \equiv z(s_i) - z(s_j)$, which clearly satisfies $\xi_{ji} = -\xi_{ij}$. The one-form, denoted by $\xi^n$, is non-vanishing if, for all edges $(s_i,s_j)\in E$, it is $\xi_{ij} = n^t (s_i - s_j) \neq 0$, that is, if $n$ is not orthogonal to any edge. The set of edges has finite cardinality $|E| \le |{S}| (|{S}|-1)/2$, so *almost all* directions $n\in\mathbb{S}^1$ generate a non-vanishing one-form $\xi^n$.
Since $G=({S},E,F)$ is a planar graph with a well-defined face structure, there is, for each face $f\in F$, a cyclically ordered set $\partial f$ of half-edges around the face. Likewise, for each vertex $v\in {S}$, the set $\delta v$ of cyclically ordered (oriented) half-edges emanating from each vertex is well-defined.
Given non-vanishing one-form $\xi$, the index of vertex $v$ with respect to $\xi$ is $$\mathbf{ind}_{_{\xi}}(v) \equiv 1 - \mathbf{sc}_{_{\xi}}(v) / 2,$$ where $\mathbf{sc}_{_{\xi}}(v)$ is the number of sign changes of $\xi$ when visiting the half-edges of $\delta v$ in order. The index of face $f$ is $$\mathbf{ind}_{_{\xi}}(f) \equiv 1 - \mathbf{sc}_{_{\xi}}(f)/2$$ where $\mathbf{sc}_{_{\xi}}(f)$ is the number of sign changes of $\xi$ as one visits the half-edges of $\partial f$ in order.
Note that, by definition, it is always $\mathbf{ind}_{_{\xi^n}}(v) \le 1$. A discrete analog of the Poincaré-Hopf index theorem relates the two indices above:
\[lem:ph\] For any non-vanishing one-form $\xi$, it is $$\displaystyle{\sum_{v\in {S}} \mathbf{ind}_{_{\xi}}(v) + \sum_{f\in F} \mathbf{ind}_{_{\xi}}(f) = 2}$$
Note that this follows from Theorem 3.5 of [@1form] because the unbounded, outside face, which is not in $G$, is assumed in this section to be closed and simple (corollary \[cor:boundary\]), and therefore has null index. Note that the machinery from [@1form] to deal with degenerate cases isn’t needed here because vertices, by definition, cannot coincide (${S}$ is not a multiset). All proofs in this section, except for that of theorem \[th:main\], are in Appendix C.
The one-forms $\xi^n$ constructed above satisfy the following property:
\[lem:non-negative\] Given a non-vanishing one-form $\xi^n$, the sum of indices of interior vertices (${S}\setminus W$) of $G$ is non-negative.
The next two lemmas relate the presence of edge fold-overs and the ECB property (proposition \[prop:ECB\]) to the indices of vertices in $G$.
\[lem:index-1\] If $G$ has an edge fold-over, then there is $n\in\mathbb{S}^1$ and non-vanishing one-form $\xi^n$ such that $\mathbf{ind}_{_{\xi^n}}(v) < 0$ for some interior vertex $v\in {S}\setminus W$.
\[lem:index1\] Given $n\in\mathbb{S}^1$ and non-vanishing one-form $\xi^n$, if $G$ has an interior vertex $v\in {S}\setminus W$ with index $\mathbf{ind}_{_{\xi^n}}(v)=1$, then there is a face $f$ of $G$ that does not satisfy the empty circum-ball property (proposition \[prop:ECB\]).
The above provides the necessary tools to prove the following key lemma.
\[lem:ef\] $G$ has no edge fold-overs.
Finally, the absence of edge fold-overs, together with a simple and closed boundary, is sufficient to show that $G$ is embedded.
\[lem:interior\] If its (topological) boundary is simple and closed, then the straight-line dual of an orphan-free Voronoi diagram, with vertices at the sites, is an embedded triangulation.
Proof-of-concept implementation {#sec:implementation}
===============================
Though not aiming for an efficient implementation, we tested a simple proof-of-concept that constructs anisotropic Voronoi diagrams (using a quadratic divergence $D_Q$ of the type discussed in section \[sec:DQ\]) and their duals (figure \[fig:test\]). A closed-form metric, which has bounded ratio of eigenvalues (and therefore by lemma \[lem:DQgamma\] satisfies assumption \[ass:BAA\]), is discretized on a fine regular grid, and linearly interpolated inside grid elements, resulting in a continuous metric. The sites are generated randomly (figures \[fig:img\_a\] and \[fig:img\_b\]), or using a combination of random, and equispaced points forming an (asymmetric) $\epsilon$-net [@avd] (remaining figures).
The primal diagram was obtained using front propagation from the sites outwards, until fronts meet at Voronoi edges. The runtime is proportional to the grid size, since every grid-vertex is visited exactly six times (equal to their valence), and so linear in the resolution of the sampled divergence $D_Q$.
The implementation does not guarantee the correctness of the diagram unless it *is* orphan-free, and serves to verify the claims of the paper since well-behave-ness of the dual is predicated on that of the primal.
The two main claims of the paper (that orphan-freedom is sufficient to ensure well-behavedeness of both the dual and the primal) are clearly illustrated in these examples. In all examples, the dual covers the convex hull of the vertices (corollary \[cor:boundary\]), is a single cover, embedded with straight edges without edge crossings (lemma \[lem:interior\]), and has no degenerate faces (since, by proposition \[prop:ECB\], the vertices of a face lie on the boundary of a strictly convex ball). By focusing on the primal diagrams (second and fourth column), further claims in the paper become apparent, namely that Voronoi regions (Voronoi elements of order one according to definition \[def:VorI\]) are simply connected (lemma \[lem:regionSC\]), and Voronoi edges (order two), and vertices (order three or higher) are connected (corollary \[cor:VorI\]).
Conclusion and open problems
============================
We studied the properties of duals of orphan-free Voronoi diagrams with respect to divergences, for the purposes of constructing triangulations on the plane. The main result (Theorems \[th:main\]) is that the dual, with straight edges and vertices at the sites, is embedded and covers the convex hull of the sites, mirroring similar results for ordinary Voronoi diagrams and their duals. Additionally, the primal is composed of connected elements (corollary \[cor:VorI\]).
![ The main proof of this paper does not work as is in higher dimensions. This arrangement of tetrahedra is not embedded: the red tetrahedra has been “inverted" (the green dotted edge is *behind* the solid blue edge), “invading" the two front tetrahedra (closest to the viewer), as well as the two back tetrahedra (farthest from the viewer). However, it does not violate the ECB condition (proposition \[prop:ECB\]). []{data-label="fig:3dbreaks"}](3dbreaks.pdf){width="5.6cm"}
Perhaps the most important outstanding question is whether these results extend to higher dimensions. The proofs in Secs. \[sec:boundary\] and \[sec:interior\], except for lemma \[lem:regionSC\], can be trivially extended to n dimensions. Section \[sec:boundary\] has been written only for the two-dimensional case, but a similar construction, and the same argument would work in higher dimensions (lemma \[lem:Sn\] being a hint of this). It is the argument in section \[sec:interior\], and described in figure \[fig:outline\], that becomes problematic. While the ECB property is shown to be sufficient to prevent fold-overs in the triangulation, it is not sufficient in higher dimensions. In particular, fixing the boundary to be simple and convex, there are simple arrangements of tetrahedra in $\mathbb{R}^3$ that contain face fold-overs but do not break the ECB property. In particular, the arrangement of tetrahedra of figure \[fig:3dbreaks\] is not embedded: the red tetrahedra has been “inverted" (indicated by the green dotted edge being *behind* the solid blue edge); its interior overlaps that of the two front tetrahedra (closest to the viewer), as well as the two back tetrahedra (those farthest from the viewer). However, this arrangement does not break the ECB condition (proposition \[prop:ECB\], which holds in any dimension), and therefore the same argument used in this work would not create a contradiction in higher dimensions.
Appendix A: Bounded anisotropy condition {#app:gamma .unnumbered}
========================================
[**lemma \[lem:DFgamma\]**]{} (Bounded anisotropy for Bregman divergences). *If $F\in\mathcal{C}^2$ and there is $\gamma > 0$ such that the Hessian of $F$ has ratio of eigenvalues bounded by $\lambda_{\text{min}}/\lambda_{\text{max}}\ge \gamma$, then assumption \[ass:BAA\] holds.*
Consider the situation described in figure \[fig:gamma\], in a coordinate system with the y-axis along $L_{pq}$.
Let $\rho\equiv D_F(m\parallel c) = F(m)-F(c)-\langle m-c,\nabla F(c)\rangle$. Because $F\in\mathcal{C}^2$ and the ball $B(c;\rho)$ it tangent to the y-axis at $m$, it is $$\begin{aligned}
0 &=& \langle \hat{y}, \nabla_x {D\left(x \parallel c\right)}\bigg|_{x=m}\rangle = \langle \hat{y}, \nabla_x \left[ F(x) - F(c) - \langle x-c, \nabla F(c) \rangle \right]\bigg|_{x=m} \rangle \\
&=& \langle \hat{y}, \nabla F(m) - \nabla F(c) \rangle.\end{aligned}$$ Since $D_F(m\parallel c) - D_F(r\parallel c) = F(m)-F(r) - \langle m-r,\nabla F(c)\rangle$, we can obtain the value of $D_F(r\parallel c)$ by integration from $m$, first along the y-axis from $(m_x,m_y)$ to $(m_x,r_y)$, then along the x-axis from $(m_x,r_y)$ to $(r_x,r_y)$.
Let $\delta_x\equiv r_x-m_x$ and $\delta_y\equiv r_y-m_y$, and assume that $\delta_x> 0$ and $\delta_y\ge 0$ without loss of generality, since $r\not\in L_{pq}$, $r$ is on the same side of $L_{pq}$ as $c$, and we have freedom in choosing the sign of the axis.
For assumption \[ass:BAA\] to hold it must be $D_F(r\parallel c) < D_F(m\parallel c)$. This holds whenever $$\delta_x |\nabla F(m)-\nabla F(c)| >
\lambda_{\text{max}} \delta_x^2 / 2 + \lambda_{\text{max}}\delta_y^2 / 2 + \lambda_{\text{max}} \delta_x\delta_y$$ or equivalently $$\lambda_{\text{min}}\delta_x \|m - c\| >
\lambda_{\text{max}} \delta_x^2 / 2 + \lambda_{\text{max}}\delta_y^2 / 2 + \lambda_{\text{max}} \delta_x\delta_y$$ which reduces to $$\|m-c\| > \gamma^{-1} \left( \delta_x^2/2 + \delta_y^2/2 + \delta_x\delta_y\right)/\delta_x$$ and is always satisfied whenever $\|c\| \ge \max\{\|p\|,\|q\|\} + \gamma^{-1} \left( \delta_x^2/2 + \delta_y^2/2 + \delta_x\delta_y\right)/\delta_x$. Note that this bound is finite because $\gamma>0$ and $\delta_x>0$, $\delta_y\ge 0$.
[**lemma \[lem:DQgamma\]**]{} (Bounded anisotropy for quadratic divergences). *If there is $\gamma > 0$ such that $Q$ has ratio of eigenvalues bounded by $\lambda_{\text{min}}/\lambda_{\text{max}}\ge \gamma$, then assumption \[ass:BAA\] holds.*
The proof of this lemma can be reduced to that of lemma \[lem:DFgamma\]. Given $c\in\mathbb{R}^2$, we let $$F(\cdot)\equiv D^2_Q(\cdot\parallel c)=(\cdot-c)^t Q(c) (\cdot-c)/2,$$ whose Hessian is $H_F\equiv Q(c)$. Since $Q$ has eigenvalues bounded from below by $\gamma$, the conditions of the proof of lemma \[lem:DFgamma\] hold. Note that this definition of $F(\cdot)$ is *per choice of* $c$, and therefore we are not defining a real Bregman divergence this way, but simply choosing a different $F$ for each $c$ as to satisfy the conditions of the proof.
[**Lemma \[lem:Lpgamma\]**]{} (Bounded anisotropy for normed spaces) *Distances derived from strictly convex $\mathcal{C}^1$ norms satisfy assumption \[ass:BAA\].*
Let $\|\cdot\|_K$ be a strictly convex $\mathcal{C}^1$ norm, whose unit ball is the symmetric convex body $K$. Let $p,q\in\mathbb{R}^2$ with supporting line $L_{pq}$, and $r\notin L_{pq}$ be given. For any $c\notin L_{pq}$ with closest point $m\in\overline{pq}$ in $L_{pq}$, define $\pi_{m,c} \equiv (c-m) / \|c-m\|_K$. Defining $m$ to be the origin, let $T_{\pi_{m,c}}$ be a linear transformation that maps the $L_{pq}$ direction into the ${y}$-axis, and $\pi_{m,c}$ into the $x$ axis. The fact that $c\notin L_{pq}$ implies that $T_{\pi_{m,c}}$ is non-singular. Choose the sign of the $y$-axis so that $\lambda_{max}(T_{\pi_{m,c}}) \ge \lambda_{min}(T_{\pi_{m,c}}) > 0$ are the maximum and minimum eigenvalues of $T$, respectively.
Consider the following statements:
1. For all pairs $(m,c)$, there is a sufficiently large $\mu_{\pi_{m,c}} > 0$ such that whenever $\|m-c\| > \mu_{\pi_{m,c}}$ then $\|r-c\|_K < \|m-c\|_K$.
2. For all pairs $(m,c)$, there is a sufficiently large $\mu^T_{\pi_{m,c}} > 0$ such that whenever $\|T(m) - T(c)\| > \mu^T_{\pi_{m,c}}$ then $\|r-c\|_K < \|m-c\|_K$.
[**\[Reducing assumption \[ass:BAA\] to statement (i)\]**]{}. Given (i), and since both $\overline{pq}$ and $\partial K$ are compact, we can define: $$\mu \equiv \max\{ \|p\|, \|q\| \} + \displaystyle{\max_{\substack{ \|\pi_{m,c}\|_{_K} = 1 \\ m\in \overline{pq} }} \mu_{\pi_{m,c}}, }$$ from which it follows that whenever $\|c\| > \mu$, it holds $$\|m - c\| \ge \|c\| - \|m\| \ge \|c\| - \max\{ \|p\|, \|q\| \} >
\displaystyle{\max_{\substack{ \|\pi_{\tilde{m},\tilde{c}}\|_K =
1 \\ \tilde{m}\in \overline{pq} }} \mu_{\pi_{\tilde{m},\tilde{c}}}} \ge \mu_{\pi_{m,c}},$$ and therefore $\|r-c\|_K < \|m-c\|_K$, thereby satisfying assumption \[ass:BAA\]. [**\[Reducing statement (i) to statement (ii)\]**]{}. Assume (ii) is true and let $\mu_{\pi_{m,c}} \equiv \lambda_{max}(T^{-1}) \mu^T_{\pi_{m,c}}$. Whenever $\|m-c\| > \mu_{\pi_{m,c}}$, it holds: $$\begin{aligned}
\|T(m) - T(c)\| &=& \|T(m-c)\| \ge \lambda_{min}(T) \|m-c\| > \lambda_{min}(T)\mu_{\pi_{m,c}} \\
&=& \lambda_{min}(T) \lambda_{max}(T^{-1}) \mu^T_{\pi_{m,c}} = \mu^T_{\pi_{m,c}},\end{aligned}$$ and therefore by (ii) it is $\|r-c\|_K < \|m-c\|_K$.
[**\[Proof of statement (ii)\]**]{}. Consider the situation depicted in figure \[fig:norm.a\], which shows a portion of the plane transformed by $T$. Given $c'=m+\pi_{m,c}$, consider the set of points at distance $\|m-c'\|_K$ from $c'$ (red line). First note that, because we have temporarily chosen $m$ as the origin, then $m=T(m)$, and $T(m)+\alpha [T(c')-T(m)] = \alpha T(c')$. Because $\|\cdot\|_K$, there is an open interval $(-\varepsilon,\varepsilon)$ and a function $f\in\mathcal{C}^1$ such that $(f(y),y)$, with $y\in(-\varepsilon,\varepsilon)$ are the coordinates of the points (in $T$-space) at distance $\|m-c'\|_K$ from $c'$.
Because $m$ is the point closest to $c'$ in $L_{pq}$, then, in $T$-space, $(f(y),y)$ is tangent to the $y$-axis at $y=0$, and therefore $f(0)=f'(0)=0$, from which it follows that $$\displaystyle{\lim_{y\rightarrow 0} (f(y)-f(0)) / (y-0)} = \displaystyle{\lim_{y\rightarrow 0} f(y)/y} = 0.$$
By a simple calculation, it is simpe to show that moving $T(c')$ further down along the $x$ axis to $\alpha T(c')$ (figure \[fig:norm.b\]), scales the red curve of figure \[fig:norm.a\] by a factor $\alpha$, so that it becomes $(\alpha f(y\alpha), y)$ with $y\in(-\alpha\varepsilon, \alpha\varepsilon)$.
Given $T(r)=(r^T_x,r^T_y)$ in coordinates, with $r^T_x > 0$ and $r^T_y \ge 0$, without loss of generality, then, from the figure and the expression for the curve $(\alpha f(y\alpha), y)$, it is clear that it is possible to choose $\alpha$ large enough so that $T(r)$ is below the curve $(\alpha f(y\alpha), y)$, and therefore $r$ is closer (with respect to $\|\cdot\|_K$) to $\alpha T(c')$ then $T(m)$. By setting $\mu^T_{\pi_{m,c}} \equiv \alpha \|T(c')-T(m)\|$, statement (ii) follows.
In particular, we simply choose $\alpha$ large enough so that
- $[\alpha T(c')]_x > r^T_x$;
- $\alpha T(c')$ is far enough from $T(m)$ so the line $\overline{[\alpha T(c')] T(r)}$ crosses the $y$-axis between $(-\alpha\varepsilon, \alpha\varepsilon)$, which is clearly possible for sufficiently large $\alpha$;
- if $r^T_x > 0$, then it is a simple calculation to show that we can ensure that $T(r)$ is “below" the curve $(\alpha f(y\alpha), y)$ as follows: 1) choose a small enough $\delta$ such that $f(\delta)/\delta < r^T_x/r^T_y$, which is always possible because $\lim_{y\rightarrow 0} f(y)/y = 0$, and 2) enforcing $\alpha > r^T_y / \delta$.
Appendix B: dual triangulation (boundary) {#app:boundary .unnumbered}
=========================================
Let $s_i,s_j\in{S}$ be two sites, we denote by $H^{+}_{ij},H^{-}_{ij}$ the two open half-spaces on either side of their supporting line $L_{ij}$. The set $\{H^{+}_{ij},H^{-}_{ij},L_{ij}\}$ is therefore a disjoint partition of $\mathbb{R}^2$. Whenever the two sites we consider are on the boundary of ${{conv}\left\{{S}\right\}}$, they are denoted by $w_i,w_j\in W\subseteq{S}$, and we always choose $H^{+}_{ij}$ to be the “empty" half space of the two (such that $H^{+}_{ij} \cap {S}=\phi$).
\[lem:halfspace\] Given a Voronoi edge ${\text{Vor}}_{ij}$ corresponding to neighboring sites $s_i,s_j\in{S}$, if ${\text{Vor}}_{ij}\cap H^{+}_{ij}$ (${\text{Vor}}_{ij}\cap H^{-}_{ij}$) is unbounded, then it is $ H^{+}_{ij}\cap {S}=\phi$ ($ H^{-}_{ij}\cap {S}=\phi$), where $H^{+}_{ij},H^{-}_{ij}$ are open half spaces on either side of the supporting line of $s_i,s_j$.
Assume $w\in H^{+}_{ij}\cap {S}$. Since ${\text{Vor}}_{ij}\cap H^{+}_{ij}$ is unbounded, we choose $p\in {\text{Vor}}_{ij}\cap H^{+}_{ij}$ of sufficiently large norm, so that assumption \[ass:BAA\] implies that ${D\left(w \parallel p\right)} < {D\left((w_i+w_j)/2 \parallel p\right)}$. By the convexity of ${D\left(\cdot \parallel p\right)}$, this means that ${D\left(w \parallel p\right)} < {D\left((s_i+s_j)/2 \parallel p\right)} <
{D\left(s_i \parallel p\right)}={D\left(s_j \parallel p\right)}$. Since $p$ is closer to $w$ than to $s_i,s_j$, it is $p\notin {\text{Vor}}_{ij}$, a contradiction.
[**Lemma \[boundary\_easy\]**]{} *To every topological boundary edge of $G$ corresponds a segment in the boundary of ${{conv}\left\{{S}\right\}}$.*
By the definition of $G$, to every boundary edge $(s_i,s_j)\in B$ corresponds a primal edge ${\text{Vor}}_{ij}$ that is unbounded. Consider the two open half-planes $H^{+}_{ij}$ and $H^{-}_{ij}$ on either side of the supporting line $L_{ij}$ of $s_i,s_j$. We split ${\text{Vor}}_{ij}$ in three parts: ${\text{Vor}}_{ij}\cap H^{+}_{ij}$, ${\text{Vor}}_{ij}\cap L_{ij}$, and ${\text{Vor}}_{ij}\cap H^{-}_{ij}$, at least one which must be unbounded.
Since, by lemma \[lem:midpoint\], it is ${\text{Vor}}_{ij}\cap L_{ij} \in \overline{s_i s_j}$ (and therefore bounded), then it must be that either ${\text{Vor}}_{ij}\cap H^{+}_{ij}$ or ${\text{Vor}}_{ij}\cap H^{-}_{ij}$ are unbounded. By lemma \[lem:halfspace\], they cannot both be, or else $H^{+}_{ij}\cap {S}=\phi$ and $H^{-}_{ij}\cap {S}=\phi$, and therefore all sites would be in $L_{ij}$ (all colinear). Assume w.l.o.g. that ${\text{Vor}}_{ij}\cap H^{+}_{ij}$ is unbounded.
By lemma \[lem:halfspace\], ${\text{Vor}}_{ij}\cap H^{+}_{ij}$ unbounded implies $H^{+}_{ij}\cap {S}=\phi$, and so $s_i,s_j$ must lie in the boundary of ${{conv}\left\{{S}\right\}}$ ($s_i,s_j\in W$ and $\overline{s_i s_j}\subseteq\partial{{conv}\left\{{S}\right\}}$). It only remains to show that $s_i,s_j$ are consecutive in the sequence $(w_i :
i=1,\dots,m)$. We prove this by contradiction. If they were not, then since $\overline{s_i s_j}\subseteq\partial{{conv}\left\{{S}\right\}}$, there must be some site $w\in \overline{s_i s_j}$, $w\neq s_i,s_j$. However, this is not possible. To see this, simply pick some point $p\in {\text{Vor}}_{ij}$, by definition closest to $s_i,s_j$; by the convexity of ${D\left(\cdot \parallel p\right)}$, it must be ${D\left(w \parallel p\right)} < {D\left(s_i \parallel p\right)}={D\left(s_j \parallel p\right)}$, a contradiction.
Since $s_i,s_j$ are consecutive vertices in $(w_i : i=1,\dots,m)$, then $(s_i,s_j)\in\mathcal{B}$.
\[fig:VW\]
\[lem:VW\] There is $\rho$ such that all $p\in\mathbb{R}^2$ with $\|p\| > \rho$ are closer to $W$ than to ${S}\setminus W$.
Let $W\subseteq S$ be the sites that lie in the boundary of the convex hull ${{conv}\left\{{S}\right\}}$. We prove that there is a sufficiently large value $\rho$ such that all $p\in\mathbb{R}^2$ with $\|p\|>\rho$ are strictly closer (in the sense of $D$) to $W$ than to the remaining sites $S\setminus W$.
Pick any pair of consecutive sites $w_i,w_{i\oplus 1}$ along the boundary of ${{conv}\left\{{S}\right\}}$, and any third site $v$ from $S\setminus W$. We first show that there are values $\rho_{w_i,w_{i\oplus 1},v}$ such that all $p\in\mathbb{R}^2$ with $\|p\|>\rho_{w_i,w_{i\oplus 1},v}$ are strictly closer to $w_i,w_{i\oplus 1}$ than to $v$. By letting $$\rho \equiv \displaystyle{\max_{\substack{w_i,w_{i\oplus 1} \in W \\ v \in S\setminus W}}
\rho_{w_i,w_{i\oplus 1},v}},$$ lemma 6.5 of \[ass:BAA\]. For a triple $w_i,w_{i\oplus 1},v$ with $w_i,w_{i\oplus 1}\in W$ consecutive vertices in $\partial{{conv}\left\{{S}\right\}}$, and $v\in S\setminus W$, we consider the supporting line $l_{i,{i\oplus 1}}$ of $w_i,w_{i\oplus 1}$, which divides space into two half-spaces $H^{+}_{i,{i\oplus 1}}$ and $H^{-}_{i,{i\oplus 1}}$, where we pick $H^{-}_{i,{i\oplus 1}}$ so as to contain $v$. Note that $v\notin l_{i,{i\oplus 1}}$ or else it would be $v\in W$.
First, let $\rho_{w_i,w_{i\oplus 1},v}$ be large enough so that all $p\in\mathbb{R}^2$ with $\|p\| > \rho$ are outside the convex hull ${{conv}\left\{{S}\right\}}$. The divide the proof in three cases, depending on whether point $p$ belongs to $l_{i,{i\oplus 1}}$, $H^{-}_{i,{i\oplus 1}}$, or $H^{+}_{i,{i\oplus 1}}$, respectively. Each case will result in a different constraint $\|p\| > \rho^{o}_{i,i\oplus 1, v}$, $\|p\| > \rho^{-}_{i,i\oplus 1, v}$, $\|p\| > \rho^{+}_{i,i\oplus 1, v}$, $$\rho_{i,i\oplus 1, v} \equiv \max\{\rho^{o}_{i,i\oplus 1, v}, \rho^{-}_{i,i\oplus 1, v}, \rho^{+}_{i,i\oplus 1, v} \}.$$
[**\[Case $p\in l_{i,{i\oplus 1}}$\]**]{}. Since not all sites are colinear, then it is $|W| > 2$, and therefore the segment $\overline{w_{i\oplus 1} w_{i\oplus 2}}$ is different from $\overline{w_i,w_{i\oplus 1}}$. If $p\in l_{i,{i\oplus 1}}$, we can consider the “next" segment $\overline{w_{i\oplus 1} w_{i\oplus 2}}$, for which, since $p$ is outside ${{conv}\left\{{S}\right\}}$, it must hold $p\in H^{+}_{w_{i\oplus 1} w_{i\oplus 2}, v}$, and therefore we can simply let $\rho^{o}_{i,i\oplus 1, v} \equiv \max_{v\in S\setminus W} \rho^{+}_{i\oplus 1, i\oplus 2, v}$. Note that there is no circular dependency in this definition, since we can resolve it by simply letting $$\rho^{o}_{i,i\oplus 1, v} \equiv \displaystyle{\max_{w_i\in W |||} \rho^{+}_{i, i\oplus 1, v}}.$$
[**\[Case $p\in H^{-}_{i,{i\oplus 1}}$\]**]{}. By assumption \[ass:BAA\], there is $ \rho^{-}_{w_i,w_{i\oplus 1},v} > 0$ such that all $p\in\mathbb{R}^2$ with $\|p\| > \rho^{-}_{w_i,w_{i\oplus 1},v}$ are closer to $v$ than to $w_i$ or $w_{i\oplus 1}$. [**\[Case $p\in H^{+}_{i,{i\oplus 1}}$\]**]{}. Let $p\in H^{+}_{i,{i\oplus 1}}$, and consider the ball $B$ of the first kind centered at $p$ with radius ${D\left(v \parallel p\right)}$. Since $B$ is convex, we can find a line $l_v$ passing through $v$ that “separates" $B$, that is, $B$ lies in the half-space $H^{+}_{l_v}$ associated to $l_v$. It follows that $v$ is the closest point to $p$ in the line $l_v$. Note that, because $v\in H^{-}_{i,{i\oplus 1}}$, and $l_v$ passes through $v$, then it must be either $w_i \in H^{+}_{l_v}$ or $w_{i\oplus 1} \in H^{+}_{l_v}$ (otherwise, if $w_i,w_{i\oplus 1}\in H^{-}_{l_v}$ then it would be $v\in H^{+}_{i,{i\oplus 1}}$, a contradiction). Without loss of generality, let $w_i$ be in $H^{+}_{l_v}$.
Pick two point $a,b$ along $l_v$ such that $v$ is between $a$ and $b$. We are now ready to apply assumption \[ass:BAA\], using the substitution $p=a$, $q=b$, $r=w_i$, and $c=p$, from which it follows that there is a sufficiently large $\rho_{w_i,w{i\oplus 1},v} > 0$ such that if $p\in H^{+}_{i,{i\oplus 1}}$ and $\|p\| > \rho_{w_i,w{i\oplus 1},v}$, then $p$ is closer (in the sense of $D$) to $w_i$ than to $v$.
[**lemma \[lem:contrad\]**]{} *There is $\rho > 0$ such that, for any segment $(w_i,w_j)\in\mathcal{B}$ with supporting line $L_{ij}$, every $p\in H^{-}_{ij}$ with $\|p\| > \rho$ whose closest point in $L_{ij}$ belongs to $\overline{w_i w_j}$ is closer to a site in ${S}\setminus\{w_i,w_j\}$ than to $L_{ij}$.*
For each edge $(w_i,w_j)\in\mathcal{B}$ with supporting line $L_{ij}$, pick a site $v\in{S}\setminus\{w_i,w_j\}$ that isn’t in $L_{ij}$ (which always exists since not all sites are colinear). By assumption \[ass:BAA\], there is a sufficiently large $\rho_{ij}$ such that every point $p\in H^{-}_{ij}$ whose closest point $m_p$ to $L_{ij}$ satisfies $m_p\in\overline{w_i w_j}$ is closer to $v$ than to $m_p$. Since $p\in H^{-}_{ij}$, then $p$ is closer to $m_p$ than to either $w_i,w_j$, and thus $p$ is closer to $v$ than to either $w_i,w_j$.
Letting $\rho$ be the maximum of $\rho_{ij}$ over all edges $(w_i,w_j)\in\mathcal{B}$ completes the proof.
[**lemma \[lem:Sn\]**]{} *Every continuous function $F:\mathbb{S}^n\rightarrow\mathbb{S}^n$ that is not onto has a fixed point.*
Assume $F$ misses $p\in\mathbb{S}^n$, and let $\gamma:\mathbb{S}^n\setminus\{p\}\rightarrow D^n$ be a diffeomorphism between the punctured sphere and the open unit disk. Since $\gamma\circ F$ is continuous and $\mathbb{S}^n$ is compact, then the set $C = (\gamma\circ F) (\mathbb{S}^n)\subset D^n$ is compact.
The function $g:C\rightarrow C$ with $g = \gamma \circ F\circ\gamma^{-1}$ is continuous and therefore, by Brouwer’s fixed point theorem [@Milnor], has a fixed point $x\in C$. The fact that $(\gamma \circ F\circ \gamma^{-1}) (x) = x$ implies $F(\gamma^{-1}(x)) = \gamma^{-1}(x)$ and thus $\gamma^{-1}(x)\in\mathbb{S}^n$ is a fixed point of $F$.
[**lemma \[lem:hard\]**]{} ($B\supseteq\mathcal{B}$) *To every segment in the boundary of ${{conv}\left\{{S}\right\}}$ corresponds a boundary edge of $G$.*
Let $(w_i,w_j)\in\mathcal{B}$ be a segment in the boundary of ${{conv}\left\{{S}\right\}}$, as shown in figures \[fig:pinu\] and \[fig:pi\]. Pick a sufficiently large $\rho > \max_{v\in {S}}\|v\|$ such that every $p$ with $\|p\| > \rho$ is outside ${{conv}\left\{{S}\right\}}$ and such that Lemmas \[lem:VW\] and \[lem:contrad\] hold. For any $\sigma > \rho$, if $A:C(\sigma)\rightarrow C(\sigma)$ is the antipodal map $A(p) = -p$, then, by continuity of $\pi$ (property \[prop:pi\].i) and by the continuity of $\nu_\sigma$, the function $A\circ\nu_\sigma\circ\pi:C(\sigma)\rightarrow C(\sigma)$ is continuous. By lemma \[lem:VW\] and property \[prop:pi\].ii, if for some $p_{ij}\in C(\sigma)$ it is $\pi(p_{ij})= (w_i+w_j)/2$ with $(w_i,w_j)\in\mathcal{B}$, then $p_{ij}$ is (strictly) closest to $w_i,w_j$, and therefore belongs to the primal edge ${\text{Vor}}_{ij}$, which implies that $(w_i,w_j)\in B$.
Showing that $\mathcal{B}\subseteq B$ now reduces to showing that for all $(w_i,w_j)\in\mathcal{B}$, for all $\sigma > \rho$, there is $p_{ij}\in
C(\sigma)$ such that $\pi(p_{ij})= (w_i+w_j)/2$.
Assume otherwise. The function $A\circ\nu_\sigma\circ\pi$ is not onto and therefore, by lemma \[lem:Sn\] (and using the fact that $C(\sigma)$ is isomorphic to $\mathcal{S}^1$), it must have a fixed point $q$.
Since $(A\circ\nu_\sigma\circ\pi)(q) = q$ then $(\nu_\sigma\circ\pi)(q) = -q$. Since $\pi(q)$ is the closest point to $q$ in $\partial{{conv}\left\{{S}\right\}}$, there is a segment $(w_k,w_l)\in\mathcal{B}$ such that $\pi(q)\in\overline{w_k w_l}$. Consider two open half spaces $H^{+}_{kl}$ and $H^{-}_{kl}$ on either side of the supporting line of $w_k,w_l$. Since not all sites are colinear, we can choose these half spaces so that $H^{+}_{kl}\cap {S}=\phi$ and $H^{-}_{kl}\cap {S}\neq \phi$. By the definition of $\nu_\sigma$, and recalling that the chosen origin of $\mathbb{R}^2$ is in the interior $\mathbf{int\ }{{{conv}\left\{{S}\right\}}}$ of the convex hull, it is $\nu_\sigma(\pi(q))\in H^{+}_{kl}$, and $q=-\nu_\sigma(\pi(q))\in H^{-}_{kl}$. To see this note that the outward-facing normal $n(\pi(q))$ is defined so that $\pi(q)+n(\pi(q))\in H^{+}_{kl}$ and so $\nu_\sigma(\pi(q)) = \sigma\cdot n(\pi(q)) / \|n(\pi(q))\| \in H^{+}_{kl}$. On the other hand, since the origin is in $\mathbf{int\ }{{{conv}\left\{{S}\right\}}}$, the fact that $\nu_\sigma(\pi(q))\in H^{+}_{kl}$ implies $q=-\nu_\sigma(\pi(q))\in H^{-}_{kl}$.
Since $\rho$ was chosen sufficiently large for lemma \[lem:contrad\] to hold, and $q\in H^{-}_{kl}$, $q$ is closer to some site $v\in
{S}\setminus\{w_k,w_l\}$ than to $\overline{w_k w_l}$. Since $v\in{{conv}\left\{{S}\right\}}$, this contradicts the fact that $\pi(q)\in\overline{w_k w_l}$ is the closest point to $q$ in ${{conv}\left\{{S}\right\}}$.
Appendix C: dual triangulation (interior) {#app:interior .unnumbered}
=========================================
[**lemma \[lem:non-negative\]**]{} *Given a non-vanishing one-form $\xi^n$, the sum of indices of interior vertices (${S}\setminus W$) of $G$ is non-negative.*
Given non-vanishing $\xi^n$, the index of a face $f$ is $\mathbf{ind}_{_{\xi^n}}(f) \le 0$. To see this, assume otherwise: a face with vertices $v_1,\dots,v_m$ around it, and index one satisfies, by the definition of index and of $\xi^n$, $n^t v_1 < \dots < n^t v_m < n^t v_1$ (or $n^t v_1 > \dots > n^t v_m > n^t v_1$), a contradiction.
Because, by corollary \[cor:boundary\], the boundary edges of $G$ form a convex, simple polygonal chain, then, given any non-vanishing $\xi^n$, all the boundary vertices have index zero, except for the “topmost" ($\underset{v\in {S}}{\mathbf{argmax\ }} \xi^n(v)$) and “bottommost" ($\underset{v\in {S}}{\mathbf{argmin\ }} \xi^n(v)$) vertices, which have index one (note that the topmost and bottommost vertices are unique because $n$ is chosen not to be orthogonal to any edge in the triangulation).
Since face indices are non-positive, and the sum of indices of boundary vertices is two then, by lemma \[lem:ph\], the sum of indices of [interior]{} vertices must be non-negative.
[**lemma \[lem:index-1\]**]{} *If $G$ has an edge fold-over, then there is $n\in\mathbb{S}^1$ and non-vanishing one-form $\xi^n$ such that $\mathbf{ind}_{_{\xi^n}}(v) < 0$ for some interior vertex $v\in {S}\setminus W$.*
If edge fold-over $e=(v,w)$ is a non-boundary edge, then at least one of its incident vertices, say $v$ is an interior vertex $v\in
{S}\setminus W$. Consider the two faces $f_1,f_2$ incident to $e$, which, by definition of edge fold-over, are on the same side of its supporting line, and the two edges $e_1,e_2$ in $f_1,f_2$ respectively, incident to $v$. Taking the half-line $h$ from $v$ towards $w$ as reference, consider the (open) set $L_i\subset\mathbb{S}^1$ of directions ranging from $h$ to $e_i$. The set $L=L_1\cap L_2$ is not empty since, by proposition \[prop:ECB\], $f_1,f_2$ are not degenerate, and therefore neither $e_1,e_2$ are parallel to $h$. $L$ is also uncountable, since it is a range of the form $$L = \{n\in\mathbb{S}^1 :
{n_{\perp}}^t {h} < 0 \wedge
{n_{\perp}}^t {e}_1 > 0 \wedge
{n_{\perp}}^t {e}_2 > 0\}$$ where ${h},{e}_i$ are the direction vectors of $h,e_i$, and ${n_\perp}$ is one of the two orthogonal directions to $n$, chosen to fit the definition.
Because $L$ is not empty, and is uncountable, and because the set of edges $E$ is finite, then there is always some direction $n\in L$ that is not orthogonal to any edge in $E$. Pick any such $n$. We prove that the non-vanishing one-form $\xi^n$ is such that $\mathbf{ind}_{_{\xi^n}}(v)<0$.
The (cyclic) sequence of oriented half-edges [around]{} $v$ is, without loss of generality, $\mathcal{S}=\left[(v,v_1);(v,w);(v,v_2);\dots\right]$, and therefore the values of the one-form around $v$ are $[\xi^n(v_1)-\xi^n(v)$, $\xi^n(w)-\xi^n(v)$, $\xi^n(v_2)-\xi^n(v)$, $\dots]$. By the definition of $n$, it is $\xi^n(v_1)<\xi^n(v)$, $\xi^n(w)>\xi^n(v)$, and $\xi^n(v_2)<\xi^n(v)$, and therefore the number of sign changes in the subsequence $\mathcal{S}'=[(v,v_1);(v,w);(v,v_2)]$ is four. Since the number of sign changes in the full sequence $\mathcal{S}$ cannot be less than that of its subsequence $\mathcal{S}'$, it is $\mathbf{sc}_{_{\xi^n}}(v)>4$ and therefore $\mathbf{ind}_{_{\xi^n}}(v)=1 - \mathbf{sc}_{_{\xi^n}}(v)/2 <0$.
[**lemma \[lem:index1\]**]{} *Given $n\in\mathbb{S}^1$ and non-vanishing one-form $\xi^n$, if $G$ has an interior vertex $v\in {S}\setminus W$ with index $\mathbf{ind}_{_{\xi^n}}(v)=1$, then there is a face $f$ of $G$ that does not satisfy the empty circum-ball property (proposition \[prop:ECB\]).*
We must prove that there is a face $f$ all of whose circumscribing balls contain some vertex in its interior.
Consider the vertex $v\in {S}\setminus W$ with $\mathbf{ind}_{_{\xi^n}}(v)=1-\mathbf{sc}_{_{\xi^n}}(v)=1$, and thus with $\mathbf{sc}_{_{\xi^n}}(v)=0$. If $\left[u_1,u_2,\dots,u_m\right]$ is the cyclic sequence of vertices neighboring $v$, then $\mathbf{sc}_{_{\xi^n}}(v)=0$ implies either $\xi^n(u_i)>\xi^n(v)$, $i=1,\dots,m$, or $\xi^n(u_i)<\xi^n(v)$, $i=1,\dots,m$. Assume the former w.l.o.g. The line $l=\{x\in\mathbb{R}^2 : n^t x = n^t v\}$, passing through $v$, strictly separates $v$ from the convex hull of its neighbors. Consider the mesh $G'$, with the same structure as $G$ but in which all the incident faces to $v$ are eliminated. We show that, in $G'$, the face count of $v$ (the number of faces in which $v$ lies) is at least one. Since $l$ separates $v$ from its neighbors, it also separates all the faces incident to $v$ from $v$ (except for $v$ itself, which lies on $l$). Pick any direction $d\in\mathcal{S}^1$ with $n^t{d} < 0$. The half-line $h$ starting at $v$ with direction $d$ does not intersect any face in $G$ that is incident to $v$. Since there is only a finite number of edges and vertices, it is always possible to choose $h$ not to contain any vertex other than $v$, and not to be parallel to any edge. Since ${{conv}\left\{{S}\right\}}$ is bounded and $h$ isn’t, there is some point $x\in h$ outside ${{conv}\left\{{S}\right\}}$, whose face count must be zero. Moving from $x$ toward $v$, $h$ crosses $\partial{{conv}\left\{{S}\right\}}$ only once (since ${{conv}\left\{{S}\right\}}$ is convex), incrementing the face count to one. Because every interior edge is incident to exactly two faces, every subsequent edge cross (which is transversal because $h$ is not parallel to any edge) modifies the face count by either zero, two, or minus two. Since the face count cannot be negative, and it is one at $h\cap \partial{{conv}\left\{{S}\right\}}$, then it must be at least one at $v$. Since $G'$ does not contain any face incident to $v$, this implies that there is some face $f$ not incident to $v$ such that $v\in f$. We prove that the face $f$ above cannot satisfy the ECB property. Since $v$ is in $f$ but is not incident to it, and $f$ is convex then, by Carathéodory’s theorem [@matousek2002lectures], $v$ can be written as a convex combination $v=\lambda_1 u_1+\lambda_2 u_2+\lambda_3 u_3$, $\sum_{i=1}^3\lambda_i=1$, $\lambda_i\in(0,1)$ of vertices $u_1,u_2,u_3$ incident to $f$ (note that this is slightly more general than required since we have already made sure in the beginning of section \[sec:dual\] that $f$ is a triangle). Given a ball circumscribing the vertices incident to $f$, because it is strictly convex, and $u_1,u_2,u_3$ lie in its boundary, then any convex combination of them with $\lambda_i\in(0,1)$ must be in the interior of the circumscribing ball, and therefore $f$ does not satisfy the ECB property.
[**lemma \[lem:ef\]**]{} *$G$ has no edge fold-overs.*
Assume $G$ has an edge fold-over. By lemma \[lem:index-1\], there is a non-vanishing one-form $\xi^n$ such that some interior vertex $v\in {S}\setminus W$ has $\mathbf{ind}_{_{\xi^n}}(v)<0$. Since, by lemma \[lem:non-negative\], the sum of indices of interior vertices is non-negative, then there must be at least one interior vertex $u\in {S}\setminus W$ with positive index $\mathbf{ind}_{_{\xi^n}}(u)=1$. In that case, by lemma \[lem:index1\], there is a face of $G$ that does not satisfy the ECB property, raising a contradiction. Therefore $G$ has no edge fold-overs.
[**lemma \[lem:interior\]**]{} *If its (topological) boundary is simple and closed, then the straight-line dual of an orphan-free Voronoi diagram, with vertices at the sites, is an embedded triangulation.*
Given a point $p\in\mathbf{int\ }{{{conv}\left\{{S}\right\}}}$ in the interior of the convex hull of ${S}$, we show that its *face count* (the number of straight-edge faces that contain it) is one. Consider a line $l$ passing through $x$ that does not pass through any vertex of $G$, and is not parallel to any (straight) edge. It is always possible to find such a line since the set of vertices and edges is finite. Because the line is unbounded and ${{conv}\left\{{S}\right\}}$ is bounded, there is a point $x\in l$ that is outside ${{conv}\left\{{S}\right\}}$. At this point clearly the face count is zero. Moving from $x$ toward $p$, $l$ crosses the boundary of ${{conv}\left\{{S}\right\}}$ (and therefore, by corollary \[cor:boundary\], the boundary of $G$) only once, since it is a simple convex polygonal chain, incrementing the face count by one. At every edge crossing (which is transversal by the choice of line), the face count remains one since, by lemma \[lem:ef\] there are no edge fold-overs, and thus every non-boundary edge is incident to two faces that lie on either side of its supporting plane. Therefore the face count at $p$ must be one. Since every point inside ${{conv}\left\{{S}\right\}}$ is covered once by faces in $G$, and the boundaries of $G$ and ${{conv}\left\{{S}\right\}}$ coincide, then $G$ is a single-cover of ${{conv}\left\{{S}\right\}}$. Because two straight edges that cross at a non-vertex always generate points with face count higher than one, then the edges of $G$ can only meet at vertices, and therefore $G$ is embedded.
[^1]: A space $M$ is uniformly connected *im kleinen* if for every $\varepsilon>0$ there is $\delta_\varepsilon>0$ such that for every pair of points $p,q\in M$ with $\|p-q\|_2<\delta_\varepsilon$ there is a connected subset $V\subseteq M$ with $p,q\in V$ and $V\subseteq B_2(p;\varepsilon)$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The Schröder-Bernstein Theorem for sets is well known. The question of whether two subisomorphic algebraic structures are isomorphic to each other, is of interest. An $R$-module $M$ is said to satisfy the Schröder-Bernstein (or SB) property if any pair of direct summands of $M$ are isomorphic provided that each one is isomorphic to a direct summand of the other. A ring $R$ (with an involution $\star$) is called a Baer (Baer $\star$-)ring if the right annihilator of every nonempty subset of $R$ is generated by an idempotent (a projection). It is clear that every Baer $\star$-ring is a Baer ring. Kaplansky showed that Baer $\star$-rings satisfy the SB property. This motivated us to investigate whether any Baer ring satisfies the SB property. In this paper we carry out a study of this question and investigate when two subisomorphic Baer modules are isomorphic. Besides, we study extending modules which satisfy the SB property. We characterize a commutative domain $R$ over which any pair of subisomorphic extending modules are isomorphic.'
address:
- 'Department of Mathematics, College of Sciences, Persian Gulf University, Bushehr, Iran, 751-6913817.'
- 'Department of Mathematics, The Ohio State University, Lima, Oh 45804, USA.'
author:
- Najmeh Dehghani
- 'S. Tariq Rizvi'
title: When mutually subisomorphic Baer modules are isomorphic
---
Introduction
============
The famous Schröder-Bernstein Theorem states that any two sets with one to one maps into each other are isomorphic. The question of whether two subisomorphic algebraic structures are isomorphic to each other has been of interest to a number of researchers. Various analogues of the Schröder-Bernstein Theorem have been appeared for categories of associative rings, categories of functors and categories of $R$-modules [@bumby], [@cantor2], [@S.B], [@Ashish], [@soo.lee], [@Rizvi.Muller], [@rososhek2], [@rososhek], [@cantor1],[@Functor], [@vasconcelos] and [@correct]. Bumby in 1965 [@bumby], showed that any two injective modules which are subisomorphic to each other are isomorphic. Müller and Rizvi in 1983 [@Rizvi.Muller], extended Bumby’s result for the class of continuous modules which are a generalization of injective modules.
[[@Rizvi.Muller Proposition 10]]{}\[M. R22\] Let $M$ and $N$ be continuous $R$-modules. If $M$ and $N$ are subisomorphic to each other, then $M\simeq N$.
They constructed an example which shows that in the above theorem both $N$ and $K$ cannot be quasi-continuous. For abelian groups, Kaplansky in 1954 [@infinite; @abelian; @groups p.12], posed the following question, also known as Kaplansky’s First Test Problem: “If $G$ and $H$ are abelian groups such that each one is isomorphic to a direct summand of the other, are $G$ and $H$ necessarily isomorphic?" Negative answers have been given to this question by several authors [@solution; @of; @kaplansky], [@the; @kaplansky; @test; @problem] and [@I.kaplansky]. Besides Kaplansky in 1968 [@Kaplansky Theorem 41], showed that every Baer $\star$-ring satisfies this analogue of the Schröder-Bernstein Theorem. Recall that a ring $R$ with an involution $\star$ is called a [*Baer $\star$-ring*]{} if the right annihilator of every nonempty subset of $R$ is generated by a projection $e$ (the idempotent $e$ of the $\star$-ring $R$ is called a [*projection*]{} if $e^{\star}= e$). In particular he proved the following result:
[@Kaplansky Theorem 41]\[Kap.Baer\] Let $R$ be a Baer $\star$-ring and $e$, $f$ be projections in $R$. If $eR$ is isomorphic to a direct summand of $fR$ and $fR$ is isomorphic to a direct summand of $eR$ then $eR$ is isomorphic to $fR$.
An $R$-module $M$ is called to [*satisfy the Schröder-Bernstein property (or SB property)*]{} if any two d-subisomorphic direct summands of $M$ are isomorphic (the $R$-modules $N$ and $K$ are called [*d-subisomorphic to each other*]{} whenever $N$ is isomorphic to a direct summand of $K$ and $K$ is isomorphic to a direct summand of $N$). Modules which satisfy the SB property was introduced and studied in [@S.B]. (For convenience, we have modified the notation in [@S.B] from “S-B" to “SB"). Moreover a subclass $\mathcal{C}$ of $R$-modules is called to [*satisfy the SB property*]{} provided that any pair of members are isomorphic whenever they are d-subisomorphic to each other. Modules that satisfy the SB property, have been also studied in [@Ashish]. By Kaplansky’s Theorem, every Baer $\star$-ring satisfies the SB property.\
Kaplansky in 1968 [@Kaplansky], introduced the notion of Baer ring. Recall that a ring $R$ is called [*Baer*]{} if the right annihilator of any nonempty subset of $R$ is generated, as a right ideal, by an idempotent. It is easy to observe that the Baer property is left and right symmetric. The notion of Baer ring was extended to a general module theoretic, introducing a Baer module. An $R$-module $M$ is called [*Baer*]{} if for all $N\leq M$, ann$_{S}(N)$ is a direct summand of $S$ where $S={\rm End}_{R}(M)$ [@Extension Chapter 4]. Clearly $R$ is a Baer ring if $R_R$ is Baer. Every nonsingular extending modules is Baer [@Extension Theorem 4.1.15].
Now what Kaplansky proved for Baer $\star$-rings (Theorem \[Kap.Baer\]), motivated us to ask “[*when any pair of subisomorphic or d-subisomorphic Baer modules are isomorphic to each other*]{}".\
In Section 2, first we give some examples to show that subisomorphic Baer modules are not necessarily isomorphic to each other (Examples \[1.1\] and \[Baer.subisomorphic\]). We note that in a Baer $\star$-ring $R$, the set of all projections forms a complete lattice and this is the main point in the proof of Theorem \[Kap.Baer\]. Here, we give an example to show that for Baer rings this is not the case in general (Example \[triangular metrix ring\]). Besides, we show that every Baer (or Rickart) module with only countably many direct summands, satisfies the SB property (Theorem \[Baer.Countable\]). Moreover, we show that duo Baer rings and reduced Baer rings satisfy the SB property (Theorem \[reduced\] and Corollary \[duo.Baer.S.B\]).\
We also investigate rings over which any pair of subisomorphic Baer modules are isomorphic. For instance, if $R$ is a right nonsingular ring with finite uniform dimension then any two subisomorphic Baer $R$-module are isomorphic if and only if $R$ is a semisimple Artinian ring (Corollary \[nonsingular\]). Rings over which every Baer module is injective are precisely semisimple Artinian rings (Theorem \[all Baer is inj\]). Moreover, we investigate when two extending modules which are subisomorphic to each other are isomorphic. It is proved that the study of the SB property for the class of extending modules can be reduced to the study of such modules when they are singular and nonsingular (Theorem \[separate extendings\]). We characterize commutative domains over which any pair of subisomorphic (torsion free) extending modules are isomorphic (Proposition \[12\] and Corollary \[16\]).
Throughout this paper, rings have nonzero identity unless otherwise stated. All modules are right and unital. Let $M$ be an $R$-module. The notations $N\subseteq M$, $N\leq M$, $N\leq_{ess}M$, or $N\leq_{\oplus} M$ mean that $N$ is a subset, a submodule, an essential submodule, or a direct summand of $M$, respectively. End$_{R}(M)$ is the ring of $R$-endomorphisms of $M$ and E$(M)$ denotes the injective hull of $M$. The notations $M^{(A)}$ and $M^{A}$ mean $\oplus_{i\in A}M_{i}$ and $\prod_{i\in A} M_{i}$, respectively, where $A$ is an index set and each $M_i\simeq M$. The annihilator of an element $m\in M$ will be denoted by ann$_{R}(m)$. ${\rm J}(R)$ stands for the Jacobson radical of a ring $R$. For any $n\in\Bbb{N}$, ${\rm M}_{n}(R)$ stands for $n\times n$ matrix ring over a ring $R$. The singular submodule of $M$ is denoted by ${\rm Z}(M)$ and the second singular submodule of $M$ is denoted by ${\rm Z}_{2}(M)$. The cardinal number of a set $X$ is denoted by $|X|$ and the cardinal number of the set $\mathbb{N}$ of natural numbers is customarily denoted by $\aleph_{\circ}$. For other terminology and results, we refer the reader to [@Extension], [@Kaplansky] and [@Muller].
**On Schröder-Bernstein property for Baer modules**
===================================================
A ring $R$ with an involution $\star$ is called [*a Baer $\star$-ring*]{} if the right annihilator of every nonempty subset of $R$ is generated by an idempotent. We remind that an $R$-module $M$ is called [*a Baer module*]{} if ann$_{S}(N)$ is a direct summand of $S$ where $S={\rm End}_{R}(M)$ and $N\leq M$ (or equivalently, for every left ideal $I$ of $S$, ann$_{M}(I)$ is a direct summand of $M$). A ring $R$ is called [*Baer*]{} if $R_R$ is Baer. In a commutative domain $R$, every right ideal is Baer, as a module over $R$.\
By Kaplansky’s Theorem, it is known that every Baer $\star$-ring satisfies the SB property. Since Baer $\star$-rings are Baer rings, it is natural to ask whether Baer rings do satisfy the SB property. So we will be concerned with the question of when any two Baer modules which are subisomorphic or direct summand subisomorphic to each other are necessarily isomorphic.\
Consider the following conditions on an $R$-module $M$:\
($C_1$) Every submodule of $M$ is essential in a direct summand of $M$.\
($C_2$) Every submodule isomorphic to a direct summand of $M$ is a direct summand.\
($C_3$) The sum of two independent direct summands of $M$ is again a direct summand.
The module $M$ is called [*continuous*]{} if it has C$_1$ and C$_2$, [*quasi-continuous*]{} if it has C$_1$ and C$_3$ and [*extending*]{} if it has C$_1$, respectively [@Muller Chapter 2]. It is well known that $C_2\Rightarrow C_3$. The following implications hold:
Injective $\Rightarrow$ Quasi-injective $\Rightarrow$ Continuous $\Rightarrow$ Quasi-continuous $\Rightarrow$ Extending.
Recall from [@Extension 1.1.8], that the singular submodule Z$(M)$ of an $R$-module $M$ is defined by Z$(M) = \{m \in M \ |\ mI = 0$ for some essential right ideal $I$ of $R \}$ and the submodule Z$_2(M)$ satisfying Z$(M/{\rm
Z}(M))$ = Z$_2(M)/{\rm Z}(M)$ is called [*the second singular submodule of $M$*]{}. The module $M$ is called [*singular*]{} ([*nonsingular*]{}) if Z$(M)=M$ (Z$(M)=0$). A ring $R$ is called [*right nonsingular*]{} if it is nonsingular as a right $R$-module. An $R$-module $M$ is called [*$\mathcal{K}$-nonsingular*]{} if for any $\varphi\in S={\rm End}_{R}(M)$, ${\rm Ker}\varphi\leq_{ess}M$ implies $\varphi=0$. Every nonsingular module is $\mathcal{K}$-nonsingular. We begin with a result from [@Extension] for latter uses.
[@Extension Lemma 4.1.16]\[extending nonsingular\] Every $\mathcal{K}$-nonsingular extending module is Baer. In particular, every nonsingular extending module is Baer.
In the following, we give some examples to show that any two subisomorphic Baer modules are not isomorphic in general.
\[1.1\] Let $R$ be a commutative domain and $I$ be any non principal ideal of $R$. Clearly $R$, $I$ are Baer $R$-modules and subisomorphic to each other however $R\not\simeq I$.
In the following we show that even if $N$ and $K$ are Baer $R$-modules with the stronger condition: “$N$ is isomorphic to a submodule of $K$ and $K$ is isomorphic to a direct summand of $N$" then $N$ is not isomorphic to $K$ in general.
\[Baer.subisomorphic\] Let $N=\Bbb{Q}^{(\Bbb{N})}\oplus \Bbb{Z}$ and $K=\Bbb{Q}^{(\Bbb{N})}$. Thus the $\Bbb{Z}$-modules $N$ and $K$ are nonsingular extending [@Muller p. 19]. Therefore by Theorem \[extending nonsingular\], $N$ and $K$ are Baer. Moreover, it is clear that $K\leq_{\oplus} N$ and $N$ is isomorphic to a submodule of $K$, however, $N$ is not isomorphic to $K$.
We recall that an $R$-module $M$ is called [*Rickart*]{} if ${\rm Ker}f={\rm ann}_{M}(f)\leq_{\oplus}M$ where $f\in {\rm End}_{R}(M)$. A ring $R$ is called [*right (left) Rickart*]{} if $R_R$ (${}_{R}R$) is Rickart. Clearly every Baer module is Rickart. For more details see [@Extension Chapter 3].\
We recall that a ring $R$ is called [(*von-Neumann) regular*]{} provided that for each $r\in R$, $r\in rRr$. It is well known that regular rings $R$ are precisely the ones whose every principal (finitely generated) right ideals are direct summands. The following result was shown in [@Rangaswamy]:
[@Rangaswamy Theorem 4]\[Rangaswamy\] Let $M$ be an $R$-module and $S={\rm End}_{R}(M)$. Then $S$ is a regular ring if and only if for each $\varphi\in S$, ${\rm Ker}\varphi$ and ${\rm Im}\varphi$ are direct summands of $M$.
By the above theorem, any $R$-module $M$, with the regular endomorphism ring is a Rickart module.\
Following [@Muller Chapter 4], an $R$-module $M$ is said to have [*$D_{2}$ property*]{} whenever $M/N$ is isomorphic to a direct summand of $M$ implies that $N$ is a direct summand of $M$ where $N\leq M$. It is well known that every quasi-projective module has ${\rm D}_{2}$ property. Moreover, every Rickart module has ${\rm D}_2$ property [@dual; @rickart Theorem 1.5].\
Following [@S.B], an $R$-module $M$ is said to [*satisfy the Schröder-Bernstein property (or SB property)*]{} if any two d-subisomorphic direct summands of $M$ are isomorphic (the $R$-modules $N$ and $K$ are called [*d-subisomorphic to each other*]{} whenever $N$ is isomorphic to a direct summand of $K$ and $K$ is isomorphic to a direct summand of $N$). A ring $R$ is said to satisfy the SB property if $R_R$ has the SB property. We note that the notion of the SB property for rings is right and left symmetric [@S.B Theorem 2.4 (c)].\
In the following lemma, we show that Rickart modules $N$ and $K$ are d-subisomorphic to each other if and only if they are epimorphic images of each other.
\[dual S.B and Baer\] Let $N$ and $K$ be Rickart modules. Then $N$ and $K$ are d-subisomorphic to each other if and only if there are $R$-epimorphisms $N\rightarrow K$ and $K\rightarrow N$.
We note that every Rickart module has D$_{2}$ property [@dual; @rickart Theorem 1.5] and for modules $M_1$ and $M_2$ with D$_2$ property, $M_1$ and $M_2$ are d-subisomorphic to each other if and only if they are epimorphic images of each other [@DSB Lemma 2.4].
In the next example, we show that two Rickart modules which are d-subisomorphic to each other are not isomorphic in general.
\[2\] Suppose that $V$ is an infinite dimensional vector space over a field $F$ with $S= {\rm End}_{F}(V)$. Let $\beta=\{v_{i}\}_{i\in I}$ be a basis for $V_{F}$ and $R:=\{(f,g)\in S\times S \ | $ rank$(f-g)< \infty \}$. Clearly $R$ is a subring of $S\times S$. We note that $R$ is a regular ring and so by Theorem \[Rangaswamy\], $R_R$ is Rickart. There exist idempotents $e$ and $ g$ in $R$ such that $eR$ and $gR$ are d-subisomorphic to each other however $eR$ is not isomorphic to $gR$ (see [@S.B Example 2.2] for more details). Since every direct summand of a Rickart module has the property [@Extension Proposition 4.5.4], $eR$ and $gR$ are Rickart $R$-module. Therefore the Rickart module $R_R$ does not satisfy the SB property.
Regarding examples \[1.1\], \[Baer.subisomorphic\], \[2\], and Theorem \[Kap.Baer\] about Baer $\star$-rings, it is natural to ask the question: [*“does any Baer module satisfy the SB property?"*]{}\
It is clear that any Baer module satisfies the SB property if and only if any pair of Baer modules which are d-subisomorphic to each other are isomorphic. In order to answer this question, we note that the main point in the proof of Theorem \[Kap.Baer\], is that the set of all projections in a Baer $\star$-ring forms a complete lattice under $``\leq"$ (if $e, f$ are idempotents in a ring $R$, we write $e\leq f$ in case $ef=fe=e$, i.e., $e\in fRf$). While in a Baer ring, the set of all right ideals generated by idempotents forms a complete lattice [@Extension Theorem 3.1.23]. In the next example we show that the set of all idempotents in a Baer ring is not a complete lattice in general.
\[triangular metrix ring\] Suppose that $R={\rm M}_{3}(\Bbb{Z})$ is the $3\times 3$ matrix ring over $\Bbb{Z}$. Then $R$ is a Baer ring [@Extension Theorem 6.1.4]. Let $e_{x}=\left[\begin{array}{ccc} 1 &0&x \\0&0&0\\0&0&0
\end{array}\right]$ where $x\in\Bbb{Z}$. Clearly each $e_{x}$ is an idempotent in $R$. We show that the set E $=\{e_{x} \ | \ x\in \Bbb{Z}\}$ has no supremum in the lattice of all idempotents in $R$. To see this, let Sup$(E)=f$ where $f=\left[\begin{array}{ccc} a_{1} &b_{1}&c_{1} \\ a_{2}& b_{2}& c_{2}\\a_{3}& b_{3}&c_{3}
\end{array}\right]$ be an idempotent in $R$. Thus $e_{x}\leq f$ for every $x\in\Bbb{Z}$ and so $fe_{x}=e_{x}f=e_{x}$. This shows that $a_{1}=c_{3}=1$, $b_{1}=c_{1}=a_{2}=a_{3}=b_{3}=0$ and $f=\left[\begin{array}{ccc} 1 &0&0 \\0&b_{2}&c_{2}\\0&0&1
\end{array}\right]$. Since $f^{2}=f$, $b_{2}^{2}=b_{2}$ and $b_{2}c_{2}+c_{2}=c_{2}$. Therefore either $b_{2}=0$ or $b_{2}=1$. If $b_{2}=1$ then $c_{2}=0$. Thus either $f=\left[\begin{array}{ccc} 1 &0&0 \\0&0&c_{2}\\0&0&1
\end{array}\right]$ or $f=\left[\begin{array}{ccc} 1 &0&0 \\0&1&0\\0&0&1
\end{array}\right]$. Let $f_{1}:=\left[\begin{array}{ccc} 1 &0&0 \\0&1&0\\0&0&1
\end{array}\right]$ and $f_{2}:=\left[\begin{array}{ccc} 1 &0&0 \\0&0&c_{2}\\0&0&1
\end{array}\right]$. Suppose that $f=f_{2}$. In this case, let $g=\left[\begin{array}{ccc} 1 &0&0 \\0&0&\alpha \\0&0&1
\end{array}\right]$ where $c_{2}\neq\alpha\in\Bbb{Z}$. Clearly for every $x\in\Bbb{Z}$, $e_{x}\leq g$. Thus $f_{2}\leq g$ and then $c_{2}=\alpha$, a contradiction. Otherwise, $f=f_{1}$. Since for every $x\in\Bbb{Z}$, $e_{x}\leq f_{2}$, we have $f_{1}\leq f_{2}$, a contradiction. Thus $E$ has no supremum between the set of all idempotents in $R$. Hence the set of all idempotents of $R$ is not a complete lattice.
By the following result from [@S.B], any module with idempotents in its endomorphism ring forming a complete lattice has the SB property.
\[complete lattice\][@S.B Theorem 2.23] Let $M$ be an $R$-module and $S={\rm End}_{R}(M)$. If the set of all idempotents in $S$ is a complete lattice with respect to the ordering $e\leq f$ then $M$ satisfies the SB property.
In what follows, we show that under some certain conditions, any Baer module satisfies the SB property.
\[reduced\] Let $N$ and $K$ be two Baer $R$-modules which are d-subisomorphic to each other. If for all idempotents $g, h\in {\rm End}_{R}(N)$, $gh=0$ implies $hg=0$, then $N\simeq K$. In particular every reduced Baer ring satisfies the SB property.
Without loss of generality, we may assume that $K\leq_{\oplus} N$ and $N$ is isomorphic to a direct summand of $K$. By Theorem \[complete lattice\], it is enough to show that the set of all idempotents in $S=$ End$_{R}(N)$ forms a complete lattice. Since $N$ is Baer, $S$ is a Baer ring [@Extension Theorem 4.2.8]. Therefore the set of all direct summands in $S_S$ forms a complete lattice [@Extension Theorem 3.1.23]. Now let $\{e_{i}\}_{i\in I}$ be an arbitrary family of idempotents in $S$. Thus there exists an idempotent $e\in S$ such that Sup$\{e_{i}S\}_{i\in I}=eS$. Thus each $e_{i}S\leq eS$. Hence every $e_{i}=ee_{i}$ and so $(1-e)e_{i}=0$. Now by our assumption on $S$, for all $i\in I$, $e_{i}(1-e)=0$ and so $e_{i}\leq e$. Now let $g$ be an arbitrary idempotent of $S$ such that every $e_{i}\leq g$. Hence $e_{i}S\leq gS$ and so $eS\leq gS$. Thus $(1-g)e=0$ and so $e(1-g)=0$. Thus $e\leq g$. Therefore Sup$\{e_{i}\}_{i\in I}=e$. Similarly every family of idempotents in $S$ has an infimum. The last statement is now clear.
An $R$-module $M$ is called [*a duo module* ]{} if every submodule of $M$ is fully invariant. A ring $R$ is called [*a right (or left) duo ring*]{} if $R_R$ (or ${}_{R}R$) is duo. For more details, see [@duomodule].
\[duo.Baer.S.B\] Every duo Baer module satisfies the SB property. In particular every right (or left) duo Baer ring satisfies the SB property.
Let $R$ be a ring and $M$ be a duo Baer $R$-module. Suppose that $e, f\in S=$ End$_{R}(M$) are idempotents such that $ef=0$. By our assumption, $e(M)$ is a fully invariant submodule of $M$. Thus $f(e(M))\leq e(M)$ and so $0=efe=fe$. An application of Theorem \[reduced\] yields the result. The last statement is now clear.
\[omm.Baer\] Every commutative Baer ring satisfies the SB property.
It follows from Corollary \[duo.Baer.S.B\].
By the next result from [@S.B], every module with ascending chain condition on its direct summands satisfies the SB property.
[@S.B Theorem 2.16]\[sim.direct summand\] Let $M$ be an $R$-module with the following condition:\
for every descending chain $M_{1}\geq M_{2}\geq...$ of direct summands of $M$, there exists $n\geq 1$ such that $M_{n}\simeq M_{n+1}$.\
Then $M$ satisfies the SB property. In particular, every module with descending (ascending) chain condition on direct summands satisfies the SB property.
\[Baer.Countable\] If $M$ is a Baer $R$-module with only countably many direct summands, then $M$ satisfies the SB property.
Let $M$ be a Baer $R$-module with only countably many direct summands. Thus $S=$ End$_R(M)$ is a Baer ring [@Extension Theorem 4.2.8] and it has only countably many idempotents. This implies that $S$ has no infinite set of orthogonal idempotents [@Extension Theorem 3.1.11]. Therefore $M$ has descending chain condition on its direct summands and so by Theorem \[sim.direct summand\], $M$ satisfies the SB property.
In the following some applications of our results are indicated by characterizing rings over which any pair of subisomorphic Baer modules are isomorphic.
\[Baer. subisomorphic\] Let $R$ be a ring and any pair of subisomorphic Baer $R$-modules are isomorphic. Then every nonsingular extending $R$-module is injective.
Suppose that any two subisomorphic Baer $R$-modules are isomorphic and $M$ is any nonsingular extending $R$-module. Let $K=$E$(M)^{\Bbb{N}}\oplus M$ and $N=$ E$(M)^{\Bbb{N}}$. The $R$-module $N$ is extending and nonsingular module. By Theorem \[extending nonsingular\], $N$ is a Baer module. Moreover by , $K$ is also a Baer module. Clearly $N$ is a direct summand of $K$ and $K$ is isomorphic to a submodule of $N$. Thus by our assumption, $N\simeq K$ and so $M$ is an injective $R$-module.
We recall that an $R$-module $M$ [*has finite unifrom dimension*]{} if there exist uniform submodules $U_{1}, U_{2},..., U_{n}$ of $M$ such that $\oplus_{i=1}^{n}U_{i}\leq_{ess}M$. It is well known that $M$ has finite uniform dimension if and only if $M$ does not contain any infinite direct sums. Clearly every uniform module has finite uniform dimension.
\[Z2\] Let $R$ be a ring such that $R_R$ has finite uniform dimension. Consider the following conditions:\
[(a)]{} Any pair of subisomorphic Baer $R$-modules are isomorphic;\
[(b)]{} Any pair of Baer $R$-modules $N$ and $K$ are isomorphic provided that $N$ is isomorphic to a submodule of $K$ and $K$ is isomorphic to a direct summand of $N$;\
[(c)]{} $R/{\rm Z}_{2}(R)$ is a semisimple Artinian ring.\
Then (a) $\Rightarrow$ (b) $\Rightarrow$ (c).
\(a) $\Rightarrow$ (b). This is clear.\
(b) $\Rightarrow$ (c). Let $I=$ ${\rm Z}_{2}(R_R)$. Since $R$ has finite uniform dimension and $I$ is a closed submodule of $R_{R}$, it is well known that $R/I$ has finite uniform dimension as an $R/I$ module. In addition, Z$((R/I)_{R})=0$, then $R/I$ is a right nonsingular ring. We also note that if $M$ is an $R/I$-module, then $M$ is Baer as $R/I$-module if and only if it is Baer as $R$-module. It is routine to see that the condition (b) holds for $R/I$-modules. Thus without loss of generality we may assume that $R$ is a right nonsingular ring. By our assumption on $R$, there exist uniform right ideals $U_{1}, U_2,...,U_{n}$ ($n\geq 1$) of $R$, such that $U_{1}\oplus U_{2}\oplus...\oplus U_{n}\leq_{ess} R_R$. Since each $U_i$ is a nonsingular extending $R$-module, by the proof of Proposition \[Baer. subisomorphic\] and the assumption (b), each $U_{i}$ must be injective. Hence $U_{1}\oplus U_{2}\oplus...\oplus U_{n}$ is an injective $R$-module and so $R=U_{1}\oplus U_{2}\oplus...\oplus U_{n}$ is a right self-injective ring. Therefore Z$(R_R)=$ J$(R)$ and $R/$J$(R)$ is a regular ring [@Muller Proposition 3.15]. Since Z$(R_R)=0$, $R$ is a regular ring. Therefore each $U_{i}$ is a simple $R$-module and so $R$ is a semisimple Artinian ring, as desired.
\[nonsingular\] Let $R$ be a right nonsingular ring and $R_R$ has finite uniform dimension. The following conditions are equivalent:\
[(a)]{} Any pair of subisomorphic Baer $R$-modules are isomorphic;\
[(b)]{} Any pair of Baer $R$-modules $N$ and $K$ are isomorphic provided that $N$ is isomorphic to a submodule of $K$ and $K$ is isomorphic to a direct summand of $N$;\
[(c)]{} $R$ is a semisimple Artinian ring.
It follows from Theorem \[Z2\] and the fact that any pair of semisimple subisomorphic modules are isomorphic.
We remind that by Bumby’s Theorem, any two subisomorphic injective modules are isomorphic to each other [@bumby Theorem]. Therefore if every Baer module is injective then any two subisomorphic Baer modules are isomorphic to each other. So in the following, we investigate the stronger case: “when every Baer $R$-module is injective".\
We recall that a ring $R$ is called [*a right V-ring*]{} if every simple $R$-module is injective.
\[all Baer is inj\] Let $R$ be a ring. The following are equivalent:\
[(a)]{} Every Baer $R$-module is injective;\
[(b)]{} $R$ is a semisimple Artinian ring.
\(a) $\Rightarrow$ (b). First we claim that $R$ is a right Noetherian right V-ring provided that every semisimple $R$-module is injective. To see this, assume that every semisimple $R$-module is injective. Clearly $R$ is a right V-ring. It is well known that a ring $R$ is right Noetherian if and only if any arbitrary direct sum of cocyclic injective $R$-modules is injective. We note that cocyclic injective $R$-modules are precisely injective envelope of simple $R$-modules. Let $\{T_{i}\}_{i\in I}$ be a family of simple $R$-modules. Since $R$ is a right V-ring, $\oplus_{i\in I}$E$(T_{i})=\oplus_{i\in I}T_{i}$ is semisimple and so by our assumption is an injective $R$-module. Thus $R$ is right Notherian.\
Now suppose that every Baer $R$-module is injective. Thus every semisimple $R$-module is injective and then by the above note, $R$ is a right Noetherian right V-ring. Therefore $R$ is a semiprime right Goldie ring and so $R$ is right nonsingular. Since every Baer $R$-module is injective, then any pair of subisomorphic Baer $R$-modules are isomorphic to each other. Now by Corollary \[nonsingular\], $R$ is a semisimple Artinian ring.\
(b) $\Rightarrow$ (a). It is obvious.
By Theorem \[extending nonsingular\], every extending nonsingular module is Baer. Therefore the question “[*when any pair of subisomorphic extending modules are isomorphic to each other*]{}" is natural. In Example \[Baer.subisomorphic\], we found two subisomorphic nonsingular extending $\Bbb{Z}$-modules $N$ and $K$ which are not isomorphic to each other. While over a right Noetherian ring $R$, any mutually d-subisomorphic extending $R$-modules are isomorphic to each other [@S.B Theorem 2.12].\
In the next, we show that the study of subisomorphic extending modules leads to the study of subisomorphic singular extending modules and subisomorphic nonsingular extending modules.
\[separate extendings\] Any two (d-)subisomorphic extending $R$-modules are isomorphic to each other if and only if any pair of (d-)subisomorphic extending singular $R$-modules are isomorphic to each other and any pair of (d-)subisomorphic extending nonsingular $R$-modules are isomorphic.
First we note that $M$ is an extending $R$-module if and only if $M={\rm Z}_{2}(M)\oplus M_{1}$ where ${\rm {Z}}_{2}(M)$ and $M_{1}$ are extending submodules and ${\rm Z}_{2}(M)$ is $M_1$-injective. Suppose that any pair of subisomorphic singular extending $R$-modules are isomorphic to each other and it does hold true for the class of nonsingular extending $R$-modules. Let $X$ and $Y$ be subisomorphic extending $R$-modules. Without loss of generality, we may assume that $Y\leq X$ and $\theta:X\rightarrow Y$ is an $R$-monomorphism. Moreover, $X={\rm Z}_{2}(X)\oplus X_{1}$ and $Y={\rm Z}_{2}(Y)\oplus Y_1$ where $X_{1}\leq X$ and $Y_1 \leq Y$. It is clear that ${\rm Z}_{2}(Y)\leq {Z_{2}}_(X)$ and $\theta({\rm Z}_{2}(X))\leq {\rm Z}_{2}(Y)$. Therefore by our assumption, ${\rm Z}_{2}(X)\simeq {\rm Z}_{2}(Y)$. On the other hand, since $Y_{1} \cap {\rm Z}_{2}(X)=0$ and $\theta(X_{1}) \cap {\rm Z}_{2}(Y)=0$, then $Y_{1}\leq X_1$ and $X_{1}\simeq \theta(X_{1})\leq Y_{1}$. Thus $X_{1}$and $Y_{1}$ are nonsingular extending $R$-modules which are subisomorphic to each other. Therefore by our assumption $X_{1}\simeq Y_{1}$. Hence $X\simeq Y$. The converse is clear.\
The proof of the case of d-subisomorphism is similar.
\[extenging.nonsingular.charac\] Let $R$ be a ring. The following statements are equivalent:\
[(a)]{} For any $R$-module $Y$ and any nonsingular extending (injective) $R$-module $X$, if $X$ and $Y$ are subiomorphic to each other then $X\simeq Y$;\
[(b)]{} Any pair of nonsingular subisomorphic $R$-modules are isomorphic to each other;\
[(c)]{} $R/{\rm Z}_{2}(R)$ is a semisimple Artinian ring.
First we note that $R/{\rm Z}_{2}(R)$ is a semisimple Artinian ring if and only if every nonsingular $R$-module is injective [@S.B Theorem 3.5].\
(a) $\Rightarrow$ (c). Let (a) holds true. Suppose that $M$ is any nonsingular $R$-module and $Y:={{\rm E}(M)}^{\Bbb{N}}\oplus M$ and $X:={{\rm E}(M)}^{\Bbb{N}}$. It is clear that $X$, $Y$ are subisomorphic nonsingular $R$-modules and $X$ is injective (and so extending). Thus by our assumption, $X\simeq Y$ and then $M$ is an injective $R$-module. Therefore every nonsingular $R$-module is injective and so $R/{\rm Z}_{2}(R)$ is a semisimple Artinian ring.\
(c) $\Rightarrow$ (b). It follows by the above note and Bumby’s Theorem.\
(b) $\Rightarrow$ (a). We note that the class of nonsingular modules is closed under taking submodules.
In [@S.B Theorem 4.7], it has been shown that [“*over a commutative domain $R$, any two subisomorphic uniform $R$-modules are isomorphic to each other if and only if $R$ is a PID"*]{}. We note that every uniform module is extending. In the following, we characterize a commutative domain $R$ over which any pair of subisomorphic extending $R$-modules are isomorphic to each other. First we include some results from [@extending; @over; @comm.; @domain] and [@structure] that characterize the structure of extending modules over Dedekind domains.\
Let $M$ be an $R$-module. We recall that $M$ is said to be [*prime*]{} if $M$ is fully faithful as module over $R/{\rm ann}_{R}(M)$. It is routine to check that the annihilator of every prime $R$-module $M$ is a prime ideal. [*An associated prime of $M$*]{} is any prime ideal $P$ of $R$ which equals the annihilator of some prime submodule $N$ of $M$. The set of all associated primes of $M$ is denoted by [*Ass$(M)$*]{}. For any $R$-module $M$, ${\rm Ass}(M)\neq \emptyset$ provided that $R$ is right Noetherian. It is well known that over a right Noetherian ring $R$, any uniform module has unique associated prime. For more details see [@Goodearl Chapters 3 and 5].
[@structure Theorem 18]\[structure\] Let $R$ be a commutative Noetherian ring. An $R$-module $M$ is extending if and only if $M=\oplus_{P}M(P)$ (unique up to isomorphism) where $M(P)$ has associated prime $P$, is extending and is $M(Q)$-injective for all $Q\neq P$.
[@structure Corollary 23]\[21\] Let $M$ be a torsion module over a Dedekind-domain $R$. Then $M$ is extending if and only if for each nonzero prime ideal $P$, either $M(P)$ is injective or $M(P)$ is a direct sum of copies $R/P^{n}$ or $R/P^{n+1}$ for some $n\in \Bbb{N}$.
[@extending; @over; @comm.; @domain Theorem 14]\[30\] Let $M$ be a torsion free module over a Dedekind domain $R$. Then $M$ is extending if and only if $M=F\oplus E$ where $E$ is an injective $R$-module and $F\simeq \oplus_{i=1}^{n}NI_{i}$ where $N$ is a proper $R$-submodule of the quotient field $K$ and each $I_{i}$ is a fractional ideal of $R$.
In the next Theorem, we investigate when over a commutative Noetherian ring $R$, two subisomorphic extending $R$-modules are isomorphic.
\[10\] Let $R$ be a commutative Noetherian ring. The following statements are equivalent:\
[(a)]{} Any pair of subisomorphic extending $R$-modules are isomorphic;\
[(b)]{} For any prime ideal $P$ of $R$, any pair of subisomorphic extending $R$-modules with associated prime $P$ are isomorphic to each other.
\(a) $\Rightarrow$ (b). This is clear.\
(b) $\Rightarrow$ (a). Assume that $X$ and $Y$ are subisomorphic extending $R$-modules and $f:X\rightarrow Y$, $g:Y\rightarrow X$ are $R$-monomorphisms. By Theorem \[structure\], $X$ and $Y$ have the decompositions $X=\oplus_{P}X(P)$ and $Y=\oplus_{P}Y(P)$ where $P$ runs over all nonzero prime ideals, $X(P)$ and $Y(P)$ are extending submodules of $X$ and $Y$ respectively, with associated primes $P$. We claim that for any prime ideal $P$, $f(X(P))\cap \oplus_{P\neq Q}Y(Q)=0$. Let $L:=f(X(P))\cap \oplus_{P\neq Q}Y(Q)\neq 0$. Over a right Noetherian ring $R$, any nonzero $R$-module has an associated prime [@Goodearl Proposition 3.12]. Therefore $\emptyset\neq {\rm Ass}(L)\subseteq {\rm Ass}(X(P))=\{P\}$. Thus ${\rm Ass}(L)=\{P\}$. On the other hand $\{P\}={\rm Ass}(L)\subseteq {\rm Ass}(\oplus_{P\neq Q}Y(Q))=\{Q \ | \ Q$ is a prime ideal of $R$ and $Q\neq P\}$, a contradiction. Therefore $f(X(P))\cap \oplus_{P\neq Q}Y(Q)=0$ and then $f(X(P))\subseteq Y(P)$. Similarly, $g(Y(P))\subseteq X(P)$. Thus for any prime ideal $P$, $X(P)$ and $Y(P)$ are extending subisomorphic $R$-modules with associated primes $P$. Hence by our assumption, for any prime ideal $P$, $X(P)\simeq Y(P)$ and so $X$ is isomorphic to $Y$.
\[12\] let $R$ be a commutative domain. The following are equivalent:\
[(a)]{} Any pair of subisomorphic extending $R$-modules are isomorphic.\
[(b)]{} Any pair of subisomorphic torsion free extending $R$-modules are isomorphic.\
[(c)]{} $R$ is a field.
\(a) $\Rightarrow$ (b) and (c) $\Rightarrow$ (a) are clear.\
(b) $\Rightarrow$ (c). Since $R$ is a commutative domain, any nonzero ideal of $R$ is uniform and so extending. Let $I$ be any nonzero ideal of $R$. Therefore $R$ and $I$ are two extending torsion free $R$-modules which are subisomorphic to each other. Hence by our assumption, $R\simeq I$. Thus $R$ is a PID. Now let $X=Q^{\Bbb{N}}\oplus R$ and $Y=Q^{\Bbb{N}}$ where $Q=Q(R)$ is the quotient field of $R$. By Theorem \[30\], $X$ and $Y$ are extending. It is clear that $X$ and $Y$ are also torsion free $R$-modules which are subisomorphic to each other. Therefore by our assumption, $X\simeq Y$. Thus $R_R$ should be injective and then $R=Q$ is a field.
In the following example we show that two mutually subisomorphic torsion extending $\Bbb{Z}$-modules are not isomorphic to each other in general.
\[13\] Suppose that $X=\Bbb{Z}_{p}\oplus \Bbb{Z}_{p^{2}}^{(\Bbb{N})}$ and $Y=\Bbb{Z}_{p}^{(2)}\oplus \Bbb{Z}_{p^{2}}^{(\Bbb{N})}$ where $p$ is a prime number. We note that $X$ and $Y$ are torsion and also by Theorem \[21\], are extending $\Bbb{Z}$-modules. It is obvious that $X$ and $Y$ are subisomorphic to each other. However, $X$ is not isomorphic to $Y$. To see this, it is routine to check that $X$ has only one simple submodule $\Bbb{Z}_{p}\oplus 0$ which is not contained in any proper cyclic submodule, however $Y$ has two simple submodules $\Bbb{Z}_{p}\oplus 0$ and $0\oplus \Bbb{Z}_{p}$ which are not contained in any proper cyclic submodule. Therefore two subisomorphic extending torsion modules are not necessarily isomorphic to each other even over a PID.
Now, in the next result we investigate when over a commutative principal ideal ring $R$, mutually subisomorphic torsion extending $R$-modules are isomorphic to each other. We recall that a commutative ring $R$ [*has zero dimensional*]{} whenever every prime ideal of $R$ is maximal and is denoted by dim$(R)=0$.
\[15\] Let $R$ be a commutative principal ideal ring. If any pair of subisomorphic extending torsion $R$-modules are isomorphic then ${\rm dim}(R)=0$.
First we prove the case that $R$ is a PID. Assume that $R$ is a PID such that any pair of subisomorphic extending torsion $R$-modules are isomorphic to each other. We show that $R$ is a field. Let $Q$ be any nonzero prime ideal of $R$, $X=R/Q\oplus ({R/Q^{2}})^{(\Bbb{N})}$ and $Y=(R/Q)^{(2)}\oplus ({R/Q^{2}})^{(\Bbb{N})}$. Clearly $X$ and $Y$ are torsion $R$-modules. Since $R$ is a PID, then $R/Q$ is isomorphic to a submodule of $R/Q^{2}$ and so $X$ and $Y$ are subisomorphic to each other. Moreover, by Theorem \[21\], $X$ and $Y$ are extending $R$-modules. Therefore by our assumption $X\simeq Y$, a contradiction. This proves that $R$ has no nonzero prime ideal. Hence the zero ideal of $R$ is maximal and so $R$ is a field, as desired.\
Now we prove the case that $R$ is a principal ideal ring. Let $P$ be any prime ideal of $R$ and $M$ be any $R/P$-module. It is routine to see that ${\rm Z}(M_{R/P})\subseteq {\rm Z}(M_{R})$. Therefore if $M_{R/P}$ is torsion then $M_R$ is so. Besides, $M_R$ is extending if and only if $M_{R/P}$ is extending. Since $R/P$ is a PID and any pair of subisomorphic extending torsion $R/P$-module are isomorphic to each other, by the above note, $R/P$ is a field. Hence $P$ is a maximal ideal of $R$.
\[16\] Let $R$ be a PID. Then any pair of subisomorphic extending torsion $R$-modules are isomorphic if and only if $R$ is a field.
It follows from Theorem \[15\].
We present in the following, a result from [@S.B] which will be used in this paper. Let $M$ be an $R$-module. A family $\{M_{i}\}_{i\in\Lambda}$ of submodules of $M$ is called [*a local direct summand*]{} of $M$, if $\sum_{i\in\Lambda} M_{i}$ is direct and $\sum_{i\in F}M_{i}$ is a direct summand of $M$ for every finite subset $F\subseteq \Lambda$. Further, if $\sum_{i\in\Lambda} M_{i}$ is a direct summand of $M$, we say that the [*local direct summand is a direct summand*]{} [@Muller Definition 2.15].
\[sum of chain of direct summand\][@S.B Theorem 2.10] Let $N$ and $K$ be $R$-modules which are subisomorphic to direct summands of each other. If every local direct summand of $N$ is a direct summand then $N\simeq K$.
Let $M$ be an $R$-module. $M$ is called [*a dual Baer module*]{} if for any $R$-submodule $N$ of $M$, the right ideal D$(N)=\{f\in S \ | $ Im$f\subseteq N\}$ of $S$ is generated by an idempotent in $S$ where $S=$ End$_{R}(M)$ [@dual; @baer; @tutu]. The module $M$ is called [*dual Rickart*]{} if for every $R$-homomorphism $f:M\rightarrow M$, Im$f$ is a direct summand of $M$. It has been shown that $M$ is dual Baer if and only if it is dual Rickart and the sum of every family of direct summands of $M$ is a direct summand [@dual; @rickart Theorem 1.7]. In the following, we show that subisomorphisms between a dual Rickart and a dual Baer module leads to an isomorphism between them. First we prove the following basic Lemma which should be compared with Lemma \[dual S.B and Baer\].
\[40\] Let $N$ and $K$ be dual Rickart modules. Then $N$ and $K$ are subisomorphic to each other if and only if they are d-subisomorphic to each other.
We note that every dual Rickart module has C$_{2}$ property [@dual; @rickart Proposition 2.21] and for modules $N$ and $K$ with C$_{2}$ property the notions of subisomorphism and d-subisomorphism coincide.
\[d.rickar and baer\] Let $N$ be a dual Baer $R$-module and $K$ be a dual Rickart $R$-module. If $N$ and $K$ are subisomorphic to each other then $N\simeq K$.
Let $N$ be dual Baer, $K$ be dual Rickart and $N$, $K$ be subisomorphic to each other. By Lemma \[40\], $N$ and $K$ are d-subisomorphic to each other. Since $N$ is dual Baer, the sum of every family of direct summands of $N$ is a direct summand [@dual; @baer; @tutu Theorem 2.1]. Therefore if $\{N_i\}_{i\in I}$ is a family of submodules of $N$ which is a local direct summand of $N$ then each $N_{i}$ is a direct summand of $N$. Thus $\sum_{i\in I} N_{i}$ is a direct summand of $N$. Hence every local direct summand of $N$ is a direct summand. Now the result follows from Theorem \[sum of chain of direct summand\].
The following example shows that the hypothesis of Proposition \[d.rickar and baer\], cannot be weakened any further such as to assume that both $N$ and $K$ are dual Rickart.
\[1\] First we note that a ring $R$ is regular if and only if $R_R$ is dual Rickart. Moreover every direct summand of a dual Rickart module is dual Rickart [@dual; @baer; @tutu Remark 2.2 and Proposition 2.8]. Now let $R$ be the regular ring of Example \[2\]. Therein we found two non-isomorphic direct summands $eR$ and $gR$ such that they are d-subisomorphic to each other, where $e$ and $g$ are idempotents in $R$. By the above note, $eR$ and $gR$ are also dual Rickart $R$-modules.
Here the following natural question arises: “are any pair of dual Baer modules which are epimorphic images of each other, isomorphic?" We end the paper with giving an example showing that the answer of this question is negative. The following example can be also compared with Examples \[1.1\] and \[Baer.subisomorphic\].
\[3\] We show that two dual Baer modules which are epimorphic images of each other are not isomorphic in general. To see it, first we note that a ring $R$ is right hereditary right Noetherian if and only if every injective $R$-module is dual Baer [@dual; @baer; @tutu Theorem 2.3]. Now let $X=\Bbb{Q}^{\Bbb{N}}$ and $Y=\Bbb{Q}^{\Bbb{N}}\oplus \Bbb{Z}_{p^{\infty}}$ where $p$ is a prime integer number. By the above note $X$ and $Y$ are dual Baer $\Bbb{Z}$-modules. Clearly there are epimorphisms $X\rightarrow Y$ and $Y\rightarrow X$while $X$ is not isomorphic to $Y$.
[**Acknowledgments.**]{} The authors express their gratitude to Mathematics Research Institute, the College of Arts and Sciences, the Office of International Affairs, OSU-Columbus and OSU-Lima, for their partial support of this research work.
[60]{}
G. F. Birkenmeier, J. K. Park, S. T. Rizvi, [*Extensions of Rings and Modules*]{}, Springer New York Heidelberg Dordrecht London (2013).
R. T. Bumby, “Modules which are isomorphic to submodules of each other", [*Arch. Math.,*]{} 16 (1965) 184-185.
I. G. Connell, “Some ring theoretic Schröder-Bernstein theorems", [*Trans. Amer. Math. Soc.,*]{} 132 (2) (1968) 335-351.
P. Crawley, “Solution of Kaplansky’s Test Problems for primary abelian groups", [*J. Algebra*]{}, 2 (1965) 413-431.
N. Dehghani, F. E. Azmy, S. T. Rizvi, “On the Schröder-Bernstein property for Modules", [*J. Pure Appl. Algebra,*]{} 223 (1) (2019) 422-438.
N. Dehghani, S. T. Rizvi, “When modules which are mutually epimorphic are isomorphic", Submitted.
P. C. Eklof, S. Shelah, “The Kaplansky Test Problems for $\Bbb{N}_{1}$-separable groups", [*Proc. Amer. Math. Soc*]{}., 126 (7) (1998) 1901-1907.
K. R. Goodearl, R. B. Warfield, [*An introduction to non commutative Noetherian rings*]{}, Second edition. London Mathematical Society Student Texts, 61. Cambridge University Press, Cambridge, (2004).
P. A. Guil-Asensio, B. Kalebogaz, A. K. Srivastava, “The Schröder-Bernstein problem for modules", [*J. Algebra*]{}, 498 (2018) 153-164.
M. A. Kamal, B. J. Müller, “Extending modules over commutative domains", [*Osaka J. Math*]{}., 25 (1988) 531-538.
M. A. Kamal, B. J. Müller, “The structure of extending modules over Noetherian rings", [*Osaka J. Math.*]{}, 25 (1988) 539-551.
I. Kaplansky, [*Infinite Abelian Groups*]{}, University of Michigan Press, Ann Arbor (1954).
I. Kaplansky, [*Rings of Operators*]{}, Benjamin, New York (1968).
D. S. Lee, C. Park, “Notes on correct modules", [*Comm. Korean Math. Soc.,*]{} 11 (2) (1966) 205-301.
G. Lee, S. T. Rizvi, C. S. Roman, “Dual Rickart Modules", [*Comm. Algebra*]{}, 39 (11) (2011) 4036-4058.
S. H, Mohamed, B. J. Müller, [*Continuous and Discrete Modules*]{}, London Math. Soc., LN 147, Cambridge Univ. Press, (1990).
B. J. Müller, S. T. Rizvi, “On injective and quasi-continuous modules", [*J. Pure Appl. Algebra*]{}, 28 (1983) 197-210.
A. C. Özcan, A. Harmanci, P. F. Smith, “Duo modules", [*Glasgow Math. J.,*]{} 48 (2006) 533-545.
S. K. Rososhek, “Correctness of modules", [*Iz. Vuz. Math.,*]{} 22 (10) (1978) 77-82. S. K. Rososhek, “Purely correct modules", [*Uspekhi Mat. Nauk,*]{} 33 (3) (1978) 176.
D. K. Tütüncü, R. Tribak, “On dual Baer modules", [*Glasg. Math. J.,*]{} 52 (2) (2010) 261-269.
E. Sasiada, “Negative solution of I. Kaplansky’s first test problem for abelian groups and a problem of K. Borsuk concerning cohomology groups", [*Bull. Acad. Polon. Sci. Sér. Sci. Math. Astronom. Phys.,* ]{} 9 (1961) 331-334.
A. d. S. Napoli, D. Mundici, M. N. Praha, “A Cantor-Bernstein Theorem for $\sigma$-complete MV-algebras", [*Czech. Math. J*]{}., 53 (2) (2003) 437-447.
K. M. Rangaswamy, “Abelian groups with endomorphic images of special types", [*J. Algebra*]{}, 6 (1967) 271-280.
V. Trnkova, V. Koubek, “The cantor-Bernstein Theorem for functors", [*Comment. Math. Univ. Carol*]{}., 14 (1) (1973) 197-204.
W. V. Vasconcelos, “Injective endomorphisms of finitely generated modules", [*Proc. Amer. Math. Soc.,*]{} 25 (4) (1970) 900-901.
R. Wisbauer, “Correct classes of modules", [*Algebra Discrete Math*]{}., 4 (2004) 106-118.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We derived the evolution of the energy deposition in the intergalactic medium (IGM) by different decaying (or annihilating) dark matter (DM) candidates. Heavy annihilating DM particles (with mass larger than a few GeV) have no influence on reionization and heating, even if we assume that all the energy emitted by annihilations is absorbed by the IGM. In the case of lighter particles, the impact on reionization and heating depends on the efficiency of energy absorption by the IGM. We calculated the fraction of energy produced by decays and annihilations which is effectively absorbed by the IGM. We found that this fraction is generally high at very high redshift ($\gg{}100$), but drops at more recent epochs.'
address:
- 'SISSA/ISAS, Via Beirut 2-4, Trieste I-34014, Italy , '
- |
Kapteyn Astronomical Institute, University of Groningen, Postbus 800, 9700 AV,\
Groningen, The Netherlands,
author:
- MICHELA MAPELLI
- EMANUELE RIPAMONTI
title: Impact of dark matter on reionization and heating
---
Introduction {#intro}
============
The reionization and heating history of the intergalactic medium (IGM) has not yet been fully understood. In particular, the nature of the sources of these processes is mostly unclear. An important contribution might arise from the first stars; but other more exotic sources (e.g. intermediate mass black holes, decaying dark matter particles, etc.) have been proposed as well.[@CF] In this proceeding, we focus on the possible role of dark matter (DM) decays and annihilations on reionization and heating of the IGM.
The energy absorbed fraction
============================
The rate of energy released by DM decays/annihilations and absorbed (per baryon) by the IGM can be expressed as $$\label{eq:fabs}
\epsilon{}(z)=f_{\rm abs}(z)\,{}\dot{n}_{\rm DM}(z)\,{}m_{\rm DM}c^2,$$ where $f_{\rm abs}(z)$ is the fraction of energy absorbed by the IGM, $m_{\rm DM}$ is the mass of the DM particle and $\dot{n}_{\rm DM}(z)$ is the decrease rate of the number of DM particles per baryon. The expression of $\dot{n}_{\rm DM}(z)$ depends on the density of DM particles and on the decays lifetime or on the thermally averaged annihilation cross-section, in the case of decays and annihilations, respectively.[@RMF06]
The most crucial parameter in equation (\[eq:fabs\]) is $f_{\rm abs}(z)$. Previous studies[@MFP06] assumed a complete and immediate energy absorption by the IGM (i.e. $f_{\rm abs}(z)=1$), deriving upper limits to the effects of DM on reionization and heating. Annihilating heavy DM particles ($m_{\rm DM}c^2{}\gtrsim{}$ GeV) have no effects on such processes, even assuming $f_{\rm abs}(z)=1$, because of the very small allowed interacting rate. Instead, under the assumption of complete absorption, lighter particles ($m_{\rm DM}c^2{}\lesssim{}$100 MeV) could be important sources of partial early reionization and heating[@MFP06].
$f_{\rm abs}(z)$ can be hardly calculated in the case of heavy particles ($m_{\rm DM}c^2{}\gg$100 MeV), because of the uncertainties in modeling the cascade associated with massive product particles[^1]. On the other hand, for relatively light ($m_{\rm DM}c^2\lesssim{}100$ MeV) DM candidates it is possible to derive the correct behaviour of the absorbed fraction[@RMF06] $f_{\rm abs}(z)$. In fact, the possible decay/annihilation products of these light particles are only photons, electron-positron pairs, or neutrinos (which are assumed to have negligible interactions with matter). For photons the effects of Compton scattering and photo-ionization must be considered; for pairs, the relevant processes are inverse Compton scattering, collisional ionizations, and positron annihilations.
If the decay products are both photons and pairs, $f_{\rm abs}(z)$ is found to be high at early epochs ($z\gg{}100$), when photo-ionization (considering photons) and inverse Compton scattering or/and positron annihilations (considering pairs) are efficient processes. In both cases, $f_{\rm abs}(z)$ drops at lower redshifts. As an example, in Fig. 1 we show the behaviour of $f_{\rm abs}(z)$ in the case of sterile neutrinos which decay into photons (left panel) and light dark matter (LDM) particles which annihilate producing pairs (right panel). Similar results are obtained for decaying LDM particles[@RMF06].
\#1
(\#1)
\[fig:fig1\]
Effects of decays and annihilations on reionization and heating
===============================================================
We can now calculate how the behaviour of $f_{\rm abs}(z)$ affects the impact of DM decays/annihilations on reionization and heating[@RMF06]. As an example, in Fig. 2 we again consider the case of sterile neutrinos decaying into photons (left panel) and of LDM particles annihilating into pairs (right panel). The thin and thick lines in both panels refer to the case where we assume complete absorption and to the case in which $f_{\rm abs}(z)$ is the same as shown in Fig. 1, respectively. The effect of sterile neutrino decays on reionization and heating is strongly suppressed, if we consider the correct $f_{\rm abs}(z)$. A similar result can be found for LDM decays[@RMF06]. Also in the case of LDM annihilations this suppression is apparent, even if less dramatic.
\[fig:fig2\]
In summary, we can conclude that the correct calculation of $f_{\rm abs}$ is crucial[@RMF06] : if the correct values of $f_{\rm abs}(z)$ are taken into account, the impact of DM decays and annihilations on reionization and heating is almost negligible, a factor $\sim{}2$ to 1000 lower than previous estimates based on the hypothesis of complete and immediate absorption.
Acknowledgments {#acknowledgments .unnumbered}
===============
MM acknowledges the organizers of the Eleventh Marcel Grossmann meeting for the MGF grant. The authors thank P. L. Biermann for inviting them to the meeting. ER acknowledges support from NWO grant 436016.
[00]{}
B. Ciardi and A. Ferrara, [*Space Science Reviews*]{} [**116**]{} 625 (2005), and references therein. E. Ripamonti, M. Mapelli and A. Ferrara, [*MNRAS*]{} [**374**]{} 1067 (2007). M. Mapelli, A. Ferrara and E. Pierpaoli, [*MNRAS*]{} [**369**]{} 1719 (2006), and references therein.
[^1]: Modeling the cascade is not only difficult, but also not necessary for our purposes, since the effects of massive particles on reionization are negligible[@MFP06], even if we assume $f_{\rm abs}(z)=1$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This note provides a detailed description and derivation of the domain decomposition algorithm that appears in previous works by the author. Given a large re-estimation problem, domain decomposition provides an iterative method for assembling Boltzmann distributions associated to small subproblems into an approximation of the Bayesian posterior of the whole problem. The algorithm is amenable to using Boltzmann sampling to approximate these Boltzmann distributions. In previous work, we have shown the capability of heuristic versions of this algorithm to solve LDPC decoding and circuit fault diagnosis problems too large to fit on quantum annealing hardware used for sampling. Here, we rigorously prove soundness of the method.'
author:
- Brad Lackey
title: A belief propagation algorithm based on domain decomposition
---
Introduction
============
In our previous work [@bian2014discrete; @bian2016mapping], we explored methods to decompose constrained optimization problems into subproblems that could be solved on quantum annealing hardware, where the individual solutions can then be reassembled into a solution of the whole problem. We have used the term *domain decomposition* to represent any of a number of heuristic belief propagation-like algorithms to do this. In [@bian2014discrete], we created a variant of the min-sum algorithm to solve LDPC decoding problems with more variables than qubits available on the available hardware. In [@bian2016mapping], we created a sum-product style algorithm that solved fault diagnosis problems too large to fit on the hardware directly.
While ultimately successful in solving these problems, the versions of the domain decomposition method used in those works was heuristic: there was no guarantee that the object formed by reassembling the solutions on each subproblem was in any way related to the original problem. The goal of this note is provide a mathematical proof of the soundness of a specific domain decomposition algorithm (Algorithm \[algorithm\]) formed by balancing “free energy flow” between subproblems. In §[2]{} we review the well known relationship between the Boltzmann distribution, Helmholtz free energy, and belief propagation algorithm. In §[3]{} we develop the notion of a regional approximate free energy analogous to the Bethe free energy. In §[4]{} we examine the free energy of a single region and show that when the free energy between regions is balanced, a critical point for the regional approximate free energy is obtained (Theorem \[theorem:reconstitute\]). We then given an explicit statement for the domain decomposition algorithm (Algorithm \[algorithm\]) and prove that a stationary point for this algorithm is such a critical point (Theorem \[theorem:soundness\]).
Like belief propagation there is no guarantee that the domain decomposition algorithm converges. In the case of belief propagation failure to converge is often attributed to a lack of convexity of the Bethe approximate free energy stemming from loops in the factor graph. Several sufficient conditions for convergence have been based on this observation [@tatikonda2002loopy; @heskes2003stable; @heskes2004uniqueness; @mooij2007sufficient; @watanabe2009graph; @watanabe2011uniqueness], and other forms of generalized belief propagation algorithms have been developed from convex bounds or treating convex and concave regions separately [@heskes2002approximate; @teh2002unified; @yuille2002cccp; @heskes2003approximate; @globerson2007approximate]. While the algorithm we propose here is a significant departure from these methods, it is possible that one or more of these techniques could apply. We leave such explorations for future work.
Algorithm \[algorithm\] relies on obtaining marginals of Boltzmann distributions along certain “boundary” variables. While the intent was to empirically estimate these using a dedicated device that produces Boltzmann samples, these marginals could also be computed by any number of other means such as simulated annealing or even belief propagation. It would be interesting to see if domain decomposition using—say—simulated annealing to compute these marginals could outperform belief propagation by carefully designing regions that internalize many of the loops in the factor graph, and so reduce the nonconvexity of the free energy approximation.
Free energy and belief propagation
==================================
Belief propagation [@pearl2014probabilistic] is an iterative method to approximate posterior marginal distributions of a constrained optimization problem given observations. The most common version of belief propagation, the sum-product algorithm, can derived in a straightforward manner by carefully tracking conditional probabilities and making suitable independence assumptions; for a concise presentation see [@kschischang2001factor]. Such algorithms, including various variants of belief propagation, can be viewed as a type of generalized distributive law on semirings [@aji2000generalized].
Our starting point is the variational approach to Bayesian re-estimation, [@yedidia2000generalized; @aji2001generalized; @yedidia2001bethe; @pakzad2002belief; @yedidia2003understanding; @yedidia2005constructing]: the Bayesian posterior distribution is a critical point of a Helmholtz free energy constructed from the given problem. Specifically, given a space $\Omega$ of configurations, let us write our target probability as a Gibbs state based on the product of problem dependent factors $$\label{equation:ansatz}
p_0(x) = \frac{1}{Z} \prod_{\alpha=1}^m f_\alpha(x) = \frac{1}{Z(T)} e^{-\sum_{\alpha=1}^m E_\alpha(x)/kT},$$ where each factor term $f_\alpha(x) \propto e^{-E_\alpha(x)/kT}$ contributes an energy $E_\alpha(x)$ to the total energy $E(x) = \sum_{\alpha=1}^m E_\alpha(x)$. The Helmholtz free energy of an ensemble with probability distribution $p$ is $$A(p) = U(p) - TH(p) = \sum_{x\in\Omega} p(x)E(x) + kT\sum_{x\in\Omega} p(x)\log p(x).$$ From (\[equation:ansatz\]) above one can write $$E(x) = \sum_{\alpha=1}^m E_\alpha(x) = -kT(\log p_0(x) + \log Z(T)).$$ Inserting this into the above equation, $$\begin{aligned}
A(p) &=& \sum_{x\in\Omega} \left( -kT p(x) \log p_0(x) + kT p(x) \log p(x) \right) - kT \log Z(T)\\
&=& -kT\log Z(T) + kT\sum_x p(x)\log\left(\frac{p(x)}{p_0(x)}\right)\\
&=& -kT\log Z(T) + kTD_{KL}(p \|p_0).\end{aligned}$$ The Kullback-Leibler divergence has $D_{KL}(p \|p_0) \geq 0$, with equality if and only $p = p_0$, and therefore the posterior is the global minimum of this Helmholtz free energy.
Decomposing the Helmholtz free energy as $$A(p) = U(p) - TH(p) = \sum_{x\in\Omega}\sum_{\alpha=1}^m p(x)E_\alpha(x) + kT\sum_{x\in\Omega} p(x)\log p(x),$$ one sees each factor contributes an energy term to the internal energy. Let us assume that the energies are “local.” We do not want to be too formal about this; simply, we assume the configurations in our space all have the form $x = (x_1, \dots, x_n)$ and that each $E_\alpha$ does not depend on all the variables $x_1,\dots,x_n$, but rather only a few of them. Let us introduce the notation
- $j \prec \alpha$ to mean $E_\alpha$ (through $f_\alpha$) depends nontrivially on $x_j$,
- $x_\alpha = (x_j\::\: j \prec \alpha)$ for the support of $E_\alpha$, and
- $\Omega_\alpha$ for the domain of the vector $x_\alpha$.
The marginal distribution with respect to one of these supports is $b_\alpha(x_\alpha) = \sum_{x\setminus x_\alpha} p(x)$ (the notation $b_\alpha$ refers to the “belief” and hence “belief propagation”). The internal energy then has the form $$\sum_{x\in\Omega} p(x) E_\alpha(x) = \sum_{x_\alpha\in \Omega_\alpha} b_\alpha(x_\alpha) E_\alpha(x_\alpha).$$ That is, the internal energy is linear in the sense that $$U(p) = \sum_{x\in\Omega}\sum_{\alpha=1}^m p(x)E_\alpha(x) = \sum_{\alpha=1}^m \sum_{x_\alpha\in \Omega_\alpha} b_\alpha(x_\alpha) E_\alpha(x_\alpha) = \sum_{\alpha=1}^m U_\alpha(b_\alpha).$$
At this point it is natural to disassociate the beliefs $b_\alpha$ with the marginals of $p$ and pose a local free energy associated to each factor (for which we will reuse the name $b_\alpha$ for its argument) $$A_\alpha(b_\alpha) = U_\alpha(b_\alpha) - TH(b_\alpha) = \sum_{x_\alpha} b_\alpha(x_\alpha) E_\alpha(x_\alpha) + kT\sum_{x_\alpha} b_\alpha(x_\alpha)\log b_\alpha(x_\alpha).$$ The main problem lies in “localizing” the entropy term. The sum of the factors’ local free energies does not recover the Helmholtz free energy partially because of the nonlinearity inherent in the entropy, but mostly because one has grossly over-counted entropy contributions from factors sharing common variables. To illustrate this by a simple example, suppose $\Omega = \Omega_1\times\Omega_2\times\Omega_3$ and $p(x) = p(x_1,x_2,x_3)$ is uniform; form the marginals $$b_1(x_2,x_3) = \sum_{x_1\in\Omega_1} p(x_1,x_2,x_3) \text{ and } b_2(x_1,x_3) = \sum_{x_2\in\Omega_2} p(x_1,x_2,x_3),$$ which are also uniform. Then $$H(p) = k\log |\Omega| = k(\log |\Omega_1| + \log |\Omega_2| + \log |\Omega_3|).$$ and $$H(b_1) = k(\log |\Omega_2| + \log |\Omega_3|) \text{ and } H(b_2) = k(\log |\Omega_1| + \log |\Omega_3|).$$ Therefore $H(b_1) + H(b_2)$ over-counts the entropy by $k\log |\Omega_3|$ simply because $x_3$ in the support of each belief.
The Bethe approximation overcomes this failure by correcting the entropy count at each variable. For each variable $x_j$ let us write $C_j = \#\{ \alpha : j \prec \alpha \}$, the number of factors involving this variable. Removing extra entropy through over counting gives $$\begin{aligned}
\nonumber
A_\text{Bethe} &=& \sum_{\alpha=1}^m\left[\,\sum_{x_\alpha\in\Omega_\alpha} b_\alpha(x_\alpha)E_\alpha(x_\alpha) + kT\sum_{x_\alpha\in\Omega_\alpha} b_\alpha(x_\alpha)\log b_\alpha(x_\alpha)\,\right]\\\label{eqn:Bethe}
&& \qquad +\ kT \sum_{j=1}^n \left((1-C_j)\cdot\sum_{x_j\in\Omega_j} b_j(x_j) \log b_j(x_j)\right).\end{aligned}$$ There is some freedom to use different weights to correct the entropy contributions, which leads to variants of the sum-product algorithm [@wiegerinck2003fractional; @weller2015bethe]. We will only consider the Bethe approximation as given above.
Here, the “ensemble” of the Bethe approximate free energy is a collection of beliefs $\{b_\alpha, b_j\}$, one for each factor and variable. The target posterior distribution minimizes the Helmholtz free energy, so it is not unreasonable to attempt to approximate this posterior with the minimum of the Bethe approximation. However, the minimum of the Helmholtz free energy is taken over global probability distributions, while the Bethe approximation is a function of disjoint beliefs. To rectify this, one adds consistency constraints on the beliefs so to make them marginals of a single distribution:
- $\sum_{x_\alpha} b_\alpha(x_\alpha) = 1$ and $\sum_{x_j} b_j(x_j) = 1$, and
- whenever $j \prec \alpha$ we require $\sum_{x_\alpha\setminus x_j} b_\alpha(x_\alpha) = b_j(x_j)$.
Enforcing these conditions with Lagrange multipliers produces a constrained Bethe approximate free energy: $$\begin{aligned}
\tilde{A}_\text{Bethe} &=& \sum_{\alpha=1}^m\left[\,\sum_{x_\alpha\in\Omega_\alpha} b_\alpha(x_\alpha)E_\alpha(x_\alpha) + kT\sum_{x_\alpha\in\Omega_\alpha} b_\alpha(x_\alpha)\log b_\alpha(x_\alpha)\,\right]\\
&& \quad +\ kT \sum_{j=1}^n \left((1-C_j)\cdot\sum_{x_j\in\Omega_j} b_j(x_j) \log b_j(x_j)\right) \\
&& \quad +\ \sum_{\alpha=1}^m \lambda_\alpha \left(\sum_{x_\alpha\in\Omega_\alpha} b_\alpha(x_\alpha) - 1\right)
+ \sum_{j=1}^n \lambda_j \left(\sum_{x_j\in\Omega_j} b_j(x_j) - 1\right)\\
&& \quad +\ \sum_{\alpha=1}^m \sum_{j\prec\alpha} \sum_{x_j\in\Omega_j}\lambda_{j\alpha}(x_j) \left(\sum_{x_\alpha \setminus x_j} b_\alpha(x_\alpha) - b_j(x_j)\right).\end{aligned}$$
We find relations at interior critical points of $\tilde{A}_\text{Bethe}$ by setting various derivatives to zero. The derivatives with respect to the multipliers simply recover the constraints when set to zero. The two nontrivial types of derivatives are $$\begin{aligned}
\frac{\partial \tilde{A}_\text{Bethe}}{\partial b_\alpha(x_\alpha)} &=& E_\alpha(x_\alpha) +kT(\log b_\alpha(x_\alpha) + 1) + \lambda_\alpha + \sum_{j\prec\alpha} \lambda_{j\alpha}(x_j), \text{ and}\\
\frac{\partial \tilde{A}_\text{Bethe}}{\partial b_j(x_j)} &=& kT(1-C_j) (\log b_j(x_j) +1) + \lambda_j - \sum_{\alpha \succ j} \lambda_{j\alpha}(x_j).\end{aligned}$$ Setting the first of these to zero produces the equation $$b_\alpha(x_\alpha)= e^{-(1+\lambda_\alpha/kT)} e^{-E_\alpha(x_\alpha)/kT} \prod_{j\prec\alpha}e^{-\lambda_{j\alpha}(x_j)/kT}.$$ In particular, at a critical point the multiplier $\lambda_\alpha$ can be selected, and is completely determined by, the normalization constraint $\sum_{x_\alpha} b_\alpha(x_\alpha) = 1$. So, we are free to work with $b_\alpha$ unnormalized, for which we have $$\label{equation:BPfactor}
b_\alpha(x_\alpha) \propto f_\alpha(x_\alpha) \cdot \prod_{j\prec\alpha}e^{-\lambda_{j\alpha}(x_j)/kT}$$ at any interior critical point. Similarly, the second type of derivative above produces the relation $$\label{equation:BPvariable}
b_j(x_j) \propto \prod_{\alpha \succ j} e^{-\lambda_{j\alpha}(x_j)/kT(C_j-1)}$$ at an interior critical point, where now $\lambda_j$ is selected, and determined by, the normalization of $b_j$.
The remaining constraints are the consistency of the marginals. Computing the marginal $\sum_{x_\alpha\setminus x_j} b_\alpha(x_\alpha) = b_j(x_j)$ using (\[equation:BPfactor\]) one finds the additional relations on each $b_j$: $$b_j(x_j) \propto e^{-\lambda_{j\alpha}(x_j)/kT}\cdot \sum_{x_\alpha\setminus x_j} f_\alpha(x_\alpha) \prod_{r\prec\alpha\setminus j} e^{-\lambda_{r\alpha}(x_j)/kT}.$$ We define the function $$M_{\alpha \to j}(x_j) \propto \sum_{x_\alpha\setminus x_j} f_\alpha(x_\alpha) \prod_{r\prec\alpha\setminus j} e^{-\lambda_{r\alpha}(x_j)/kT},$$ with the condition $\sum_{x_j} M_{\alpha \to j}(x_j) = 1$. One can simplify the above relation to $$\label{equation:BPinter}
b_j(x_j) \propto e^{-\lambda_{j\alpha}(x_j)/kT} \cdot M_{\alpha \to j}(x_j).$$
Note the trivial equation $$\prod_{\beta \succ j\::\: \beta \not= \alpha} b_j(x_j) = b_j(x_j)^{C_j-1},$$ which is just a restatement of the definition of $C_j$. Now, we evaluate the left side of this equation using (\[equation:BPinter\]) and the right side using (\[equation:BPvariable\]). This results in the relation $$\prod_{\beta \succ j\::\: \beta \not= \alpha} e^{-\lambda_{j\beta}(x_j)/kT} \cdot M_{\beta \to j}(x_j) \propto \prod_{\beta \succ j} e^{-\lambda_{j\beta}(x_j)/kT}.$$ Canceling common terms from both side leaves $$\label{equation:BPrel}
e^{-\lambda_{j\alpha}(x_j)/kT} \propto \prod_{\beta \succ j\::\: \beta \not= \alpha} M_{\beta \to j}(x_j).$$ Now we define the function $$M_{j\to \alpha}(x_j) \propto e^{-\lambda_{j\alpha}(x_j)/kT},$$ where as before we require $M_{j\to \alpha}(x_j)$ to be a probability distribution.
Suppose that for each variable $x_j$ and factor $f_\alpha$ with $j \prec \alpha$ one has two probability distributions, $M_{\alpha \to j}(x_j)$ and $M_{j \to \alpha}(x_j)$, which jointly satisfy $$\begin{aligned}
M_{\alpha \to j}(x_j) &\propto& \sum_{x_\alpha\setminus x_j} f_\alpha(x_\alpha) \prod_{k\prec\alpha\setminus j} M_{k \to \alpha}(x_k), \text{ and}\\
M_{j\to \alpha}(x_j) &\propto& \prod_{\beta \succ j\::\: \beta \not= \alpha} M_{\beta \to j}(x_j).\end{aligned}$$ Then the probability distributions $b_\alpha(x_\alpha)$ and $b_j(x_j)$ satisfying $$\begin{aligned}
b_\alpha(x_\alpha) &\propto& f_\alpha(x_\alpha) \cdot \prod_{j \prec \alpha} M_{j \to \alpha}(x_j), \text{ and}\\
b_j(x_j) &\propto& \prod_{\alpha \succ j} M_{\alpha\to j}(x_j),\end{aligned}$$ are critical points of the Bethe approximate free energy with $$\sum_{x_\alpha\setminus x_j} b_\alpha(x_\alpha) = b_j(x_j), \text{ whenever $j \prec \alpha$.}$$
This result hands us the sum-product algorithm. We initialize distributions $M^{(0)}_{\alpha \to j}(x_j)$ and $M^{(0)}_{j \to \alpha}(x_j)$ in a reasonable way (which will be problem dependent) and iteratively redefine these as in the proposition: $$\begin{aligned}
M^{(t+1)}_{\alpha \to j}(x_j) &\propto& \sum_{x_\alpha\setminus x_j} f_\alpha(x_\alpha) \prod_{k\prec\alpha\setminus j} M^{(t)}_{k \to \alpha}(x_k), \text{ and}\\
M^{(t+1)}_{j\to \alpha}(x_j) &\propto& \prod_{\beta \succ j\::\: \beta \not= \alpha} M^{(t+1)}_{\beta \to j}(x_j).\end{aligned}$$ The proposition shows that if this converges to a stationary point, that point is a critical point of the Bethe approximate free energy and so may be taken as an approximation of the Bayesian posterior.
In practice, one has a class of factors that depend only a single variable $f_\alpha = f_\alpha(x_j)$; these typically arise as prior probabilities in a re-estimation problem. Let us write $\alpha = \{j\}$ in this case. Since $\{j\}\setminus j$ is vacuous, we have $$M_{\{j\}\to j}^{(t)}(x_j) \propto f_{\{j\}}(x_j)$$ for all $t$ and hence there is no need to compute these. One can then simplify $$M_{j\to\alpha}^{(t+1)} \propto f_{\{j\}}(x_j) \cdot \prod_{\beta\succ j\::\:\beta\not=\alpha,\{j\}}M^{(t+1)}_{\beta \to j}(x_j).$$ Since $j\not\prec \{k\}$ for $j \not= k$, one can also eliminate all factors arising from priors and use this last rule as the update rule at such nodes.
In most problems, one is primarily interested in the marginals $b_j(x_j)$, as these are the posterior probabilities one wishes to compute by re-estimation. From (\[equation:BPvariable\]) these satisfy $$b_j(x_j) \propto \prod_{\alpha\succ j} e^{-\lambda_{j\alpha}(x_j)/kT(C_j-1)} \propto e^{-(\sum_{\alpha\succ j}(\lambda_{j\alpha}(x_j)/(C_j-1)))/kT}.$$ That is, $b_j$ is the Gibbs state associated to the energy spectrum given by normalized Lagrange multipliers $\lambda_{j\alpha}(x_j)/(C_j-1)$. In particular, the posterior marginal is itself the minimum of the Helmholtz free energy $$A_j(b_j) = \sum_{\alpha\succ j} \sum_{x_j\in\Omega_j} b_j(x_j)\lambda_{j\alpha}(x_j)/(C_j-1) + kT\sum_{x_j\in\Omega_j}b_j(x_j)\log b_j(x_j).$$
Regional approximate free energies
==================================
The presentation of the Bethe approximate free energy of the previous section was a purposefully unusual. Let us return to this construction, but rather than focus on the energies of single factors let us construct “regions” containing multiple factors. We introduce the notation $f_R(x) = \prod_{\alpha \in R} f_\alpha(x)$ for a region of factors $R$. Without loss of generality, we will assume that to every variable $x_j$ there is a factor $f_j(x_j)$. As noted at the end of the previous section, we can treat priors differently and so not include them in any of the regions. To be precise, all the factors that are not priors are partitioned into regions and the prior factors are treated individually. We hasten to indicate that unlike the usual approach to regional belief propagation algorithms [@yedidia2000generalized; @aji2001generalized; @pakzad2002belief], we do not use a hierarchy of regions and the Kukuchi free energy approximation. Here we simply organize our factors in a different way and use the Bethe approximation above.
We extend the notation to $x_R$, meaning the collection of variables for which $f_R$ depends nontrivially; we also write $x_j \in x_R$ or $j \prec R$ when $x_j$ is one of these variables. Each region has internal energy $$U_R(b_R) = \sum_{x_R} b_R(x_R)E_R(x_R),$$ where we have defined $E_R(x) = \sum_{\alpha \in R} E_\alpha(x)$. Similarly, if $E_j(x_j)$ is the energy associated to the prior $f_j(x_j)$, the internal energy at the variable $x_j$ is $$U_j(b_j) = \sum_{x_j} b_j(x_j)E_j(x_j).$$ When $b_R$ and $b_j$ are obtained by marginalizing a global probability $p$, the internal energy of the whole system is $$U(p) = \sum_R U_R(b_R) + \sum_j U_j(b_j).$$
Exactly at in the Bethe method, define the weight $$C_j = \#\{R \::\: R \succ j\},$$ and the constrained regional approximate free energy: $$\begin{aligned}
\tilde{A}_\text{regional} &=& \sum_R\left(\sum_{x_R} b_R(x_R) E_R(x_R) + kT \sum_{x_R} b_R(x_R)\log b_R(x_R)\right)\\
&&\quad +\ \sum_j\left(\sum_{x_j} b_j(x_j) E_j(x_j) - kT(C_j-1) \sum_{x_j} b_j(x_j)\log b_j(x_j)\right)\\
&&\quad +\ \sum_R \lambda_R \left(\sum_{x_R} b_R(x_R) - 1\right) + \sum_j \lambda_j \left(\sum_{x_j} b_j(x_j) - 1 \right)\\
&&\quad +\ \sum_R \sum_{j\prec R} \sum_{x_j} \lambda_{jR}(x_j) \left( \sum_{x_R\setminus x_j} b_R(x_R) - b_j(x_j) \right).\end{aligned}$$
By a similar argument, setting the derivative with respect to $b_R(x_R)$ to zero yields $$\label{equation:GBPfactor}
b_R(x_R) \propto f_R(x_R) \cdot \prod_{j\prec R}e^{-\lambda_{jR}(x_j)/kT}.$$
One finds the analysis of the derivative with respect to $b_j(x_j)$ splits into two cases. First if $C_j = 1$ (when the variable is internal to one region) one obtains $$\frac{\partial\tilde{A}_\text{regional}}{\partial b_j(x_j)} = E_j(x_j) + \lambda_j - \lambda_{jR}(x_j),$$ where $R$ is the region that contains $j$. By setting this to zero and exponentiating, we find $$e^{-\lambda_{jR}(x_j)/kT} \propto f_j(x_j).$$ However for variables contained in multiple regions (which we call *boundary* variables) one finds $$\frac{\partial \tilde{A}_\text{regional}}{\partial b_j(x_j)} = E_j(x_j) - (C_j-1)kT (\log b_j(x_j) + 1) + \lambda_j - \lambda_{jR}(x_j).$$ Setting this derivative to zero produces the relation $$\label{equation:GBPvariable}
b_j(x_j) \propto (f_j(x_j))^{-1/(C_j-1)}\cdot \prod_{R\succ j} e^{-\lambda_{jR}(x_j)/kT(C_j-1)}.$$
The variables forming the support of a region $R$ divide into boundary variables (denoted $\partial R$) and the remainder which we call interior (denoted $R^\circ$). Computing the marginal of the belief in (\[equation:GBPfactor\]) at a boundary variable $j \in \partial R$ yields the relation $$b_j(x_j) \propto e^{-\lambda_{jR}(x_j)/kT} \sum_{x_R\setminus x_j} f_{R}(x_R)
\prod_{r \prec R\setminus j} e^{-\lambda_{rR}(x_r)/kT}.$$ Note that in this product, $r \prec R\setminus j$ runs over all variables. For variables in $R^\circ$ we know $e^{-\lambda_{rR}(x_r)/kT} = f_r(x_r)$. Based on this we define the *augmented* regional factor as the usual factor for that region times the priors for all its internal variables: $$\tilde{f}_R(x_R) = f_R(x_R)\cdot \prod_{r \prec R^\circ} f_r(x_R).$$ As $j \prec \partial R$, $$b_j(x_j) \propto e^{-\lambda_{jR}(x_j)/kT} \sum_{x_R\setminus x_j} \tilde{f}_{R}(x_R)
\prod_{r \prec \partial R\setminus j} e^{-\lambda_{rR}(x_r)/kT}.$$ This leads us to define $$M_{R\to j}(x_j) = \sum_{x_R\setminus x_j} \tilde{f}_{R}(x_R) \prod_{r \prec \partial R\setminus j} e^{-\lambda_{rR}(x_r)/kT}$$ so that $$\label{equation:GBPinter}
b_j(x_j) \propto e^{-\lambda_{jR}(x_j)/kT} M_{R\to j}(x_j).$$
Except for the fact that one deals only with messages to and from boundary variables, the arguments of the previous section apply. For a fixed region $R$ one has $$\prod_{S \succ j \::\: S \not= R} b_j(x_j) = b_j(x_j)^{C_j-1}.$$ Evaluating the left and right sides with equations (\[equation:GBPinter\]) and (\[equation:GBPvariable\]) respectively gives the relation $$\prod_{S\succ j \::\: S \not= R} e^{-\lambda_{jS}(x_j)/kT} M_{S\to j}(x_j) \propto
f_j(x_j)^{-1} \cdot \prod_{S\succ j} e^{-\lambda_{jS}(x_j)/kT}.$$ Canceling common terms produces the analogous result: $$M_{j \to R}(x_j) \propto e^{-\lambda_{jR}(x_j)/kT} \propto f_j(x_j)\cdot \prod_{S\succ j \::\: S \not= R} M_{R\to j}(x_j).$$
\[proposition:regional-BP\] Suppose that for each region $R$ and variable $x_j$ with $j \prec \partial R$ one has two probability distributions, $M_{R \to j}(x_j)$ and $M_{j \to R}(x_j)$, which jointly satisfy $$\begin{aligned}
M_{R \to j}(x_j) &\propto& \sum_{x_R\setminus x_j} \tilde{f}_{R}(x_R) \prod_{r\prec \partial R\setminus j} M_{r \to R}(x_r), \text{ and}\\
M_{j\to R}(x_j) &\propto& f_j(x_j)\cdot\prod_{S \succ j\::\: S \not= R} M_{S \to j}(x_j).\end{aligned}$$ Then the probability distributions $b_R(x_R)$ and $b_j(x_j)$ satisfying $$\begin{aligned}
b_R(x_R) &\propto& \tilde{f}_R(x_R) \cdot \prod_{j \prec \partial R} M_{j \to R}(x_j), \text{ and}\\
b_j(x_j) &\propto& f_j(x_j)\cdot \prod_{R \succ j} M_{R\to j}(x_j),\end{aligned}$$ are critical points of the regional approximate free energy with $$\sum_{x_R\setminus x_j} b_R(x_R) = b_j(x_j), \text{ whenever $j \prec \partial R$.}$$
Note that this proposition yields a generalized belief propagation algorithm. However this algorithm becomes impractical as the size of the regions increases. With large regions, there is potentially far fewer variables $x_j$ for which we need to compute messages. However the marginalization over $x_R\setminus x_j$ becomes intractable.
Modified free energy of a single region
=======================================
In the previous section we developed a regional belief propagation algorithm that rapidly becomes inefficient as the size of regions get large. However, if we assume we have access to a device that samples from the Boltzmann distribution, can this capability aid us in computing—or at least in approximating—these marginals?
Given a regional decomposition as in the last section, the Helmholtz free energy of the factors of a single region $R$ is given by $$A_R(b_R) = \sum_{x_R} b_R(x_R) \left(E_R(x_R) + \sum_{j\prec R} E_j(x_j)\right) + kT \sum_{x_R} b_R(x_R) \log b_R(x_R).$$ Unless $R$ is isolated from the rest of the system one would not expect that minimizing $A_R$ would produce a result related to our desired posterior distribution. Nonetheless, we argue as follows. At the minimum of the free energy, the outgoing message $M_{j\to S}(x_j) \propto e^{-kT\lambda_{jS}(x_j)}$ represents the flow out of variable $j$ into region $S$. The (change in) free energy associated to this is $$A_j^{S}(b_j) = \sum_{x_j} b_j(x_j) \lambda_{jS}(x_j) + kT b_j(x_j)\log b_j(x_j).$$ At each variable in $\partial R$ we compensate for this expected flow. This produces the *modified free energy* of the region $R$, defined as $$A_R(b_R) - \sum_{j\in\partial R}\left( \sum_{S\succ j\::\:S\not= R} A_j^{S}(b_j)\right).$$ As with the Bethe approximation, we add appropriate marginalization constraints, to obtain the *constrained modified free energy* $$\begin{aligned}
\tilde{A}_R &=& \sum_{x_R} b_R(x_R) \tilde{E}_R(x_R) + kT \sum_{x_R} b_R(x_R) \log b_R(x_R)\\
&&\quad +\ \sum_{j\prec \partial R}\sum_{x_j} b_j(x_j) V^R_j(x_j) - kT(C_j-1) \sum_{x_j} b_j(x_j) \log b_j(x_j)\\
&&\quad +\ \lambda_R\left(\sum_{x_R} b_R(x_R) - 1 \right) + \sum_{j\prec \partial R} \lambda^R_j \left(\sum_{x_j} b_j(x_j) - 1\right)\\
&&\quad +\ \sum_{j\prec \partial R} \sum_{x_j}\lambda_{jR}(x_j) \left( \sum_{x_R\setminus x_j} b_R(x_R) - b_j(x_j)\right),\end{aligned}$$ where $$\tilde{E}_R(x_R) = E_R(x_R) + \sum_{j \prec R\setminus\partial R} E_j(x_j)$$ and for $j\in\partial R$, $$V_j^R(x_j) = E_j(x_j) - \sum_{S \succ j \::\: S\not= R} \lambda_{jS}(x_j).$$ Note the the definition of $V_j^R$ involves the Lagrange multipliers of the constrained modified free energy of adjacent regions.
Adding these local “corrective” potentials $V_j^R$ could be viewed as a form of hybrid mean field approach [@riegler2013merging], however the values of the potential are unknown. Nonetheless, like the cavity method [@mooij2007loop], if the corrective potentials are set as indicated above then the minima of each regional free energy do minimize the global approximate free energy.
\[theorem:reconstitute\] A joint interior critical point of each modified regional free energy defines a critical point of the regional approximate free energy.
Just as above we compute an interior critical point of each region’s constrained modified free energy, yielding $$\begin{aligned}
\tilde{E}_R(x_R) + kT(\log b_R(x_R) + 1) + \lambda_R + \sum_{j\prec \partial R} \lambda_{jR}(x_j) &=& 0,\\
V_j^R(x_j) - (C_j-1)kT(\log b_j(x_j) + 1) + \lambda^R_j - \lambda_{jR}(x_j) &=& 0.\end{aligned}$$ Now using $V_j^R(x_j) = E_j(x_j) - \sum_{S \succ j \::\: S\not= R} \lambda_{jS}(x_j)$ these become $$\begin{aligned}
b_R(x_R) &\propto& \tilde{f}_R(x_R)\cdot \prod_{j\prec \partial R} e^{-\lambda_{jR}(x_j)/kT}\\
b_j(x_j) &\propto & f(x_j)^{1/(1-C_j)}\cdot \prod_{S\succ j} e^{-\lambda_{jS}(x_r)/kT(C_j-1)}.\end{aligned}$$ These are precisely the relations of a critical point of the regional approximate free energy.
Finally we can tackle the question of how to utilize a modest sized device or method that can produce Boltzmann samples. *Suppose that*: given an arbitrary selection of corrective potentials $\{V_j^R(x_j)\}_{j \prec \partial R}$, we have a black box that produces an interior minimum $b_R(x_R)$ of each region’s modified regional free energy. *Then*: we instantiate Algorithm \[algorithm\] below. Note that in this algorithm it is only the marginals of the Boltzmann distribution $b_R(x_R)$ that are required. With the ability to do Boltzmann sampling from physical hardware (or by other means), we approximate these marginals in each region by drawing a sufficiently large number of samples and estimating the marginal probabilities empirically. Specifically, we decompose our large re-estimation problem into regions small enough for our Boltzmann sampler to handle, and use the sampler on each region to approximate the marginals of Algorithm \[algorithm\]. By Theorem \[theorem:soundness\] below, if the algorithm converges then it recovers the approximate posterior.
Initialize each $F_{j\to R}(x_j)$ appropriately (e.g. to a uniform distribution) for each $j,R$ with $j \prec \partial R$
The numerical results presented in [@bian2014discrete; @bian2016mapping] indicate that these approximations can be sufficient to solve problems too large to be sampled directly. In [@bian2014discrete] we used a variant of Algorithm \[algorithm\] based on the min-sum method to solve 1000 variable LDPC decoding problems on a 504 qubit D-Wave quantum annealer for which belief propagation failed to converge. In [@bian2016mapping] we implemented a slight variant of Algorithm \[algorithm\] to solve hardware fault diagnosis problems requiring up to seven regions to completely embed on a D-Wave 2X architecture.
If Algorithm \[algorithm\] converges then the marginal at any boundary variable $b_j(x_j) = \sum_{x_R\setminus x_j} b_R(x_R)$ is independent of the region $R$ used to compute it.
For any boundary variable $x_j$ and region $R \succ j$, if the algorithm has converged then steps 5 and 6 of the algorithm show $$\sum_{x_R\setminus x_j} b_R(x_R) \propto F_{j\to R}(x_j)\cdot F_{R\to j}(x_j) \propto f_j(x_j)\cdot \prod_{S\succ j} F_{S\to j}(x_j),$$ which is independent of $R$.
\[theorem:soundness\] If Algorithm \[algorithm\] converges then the collection of minima of each region’s modified regional free energy $b_R(x_R)$ from step 4, and their marginals $b_j(x_j) = \sum_{x_R\setminus x_j} b_R(x_R)$, produces a critical point of the approximate regional free energy of the whole system.
For an arbitrary selection of corrective potentials $V_j^R(x_j)$, the critical point of the constrained modified free energy satisfies $$\begin{aligned}
b_R(x_R) &\propto& \tilde{f}_R(x_R)\cdot \prod_{j\prec \partial R} e^{-\lambda_{jR}(x_j)/kT}\\
b_j^{(R)}(x_j) &\propto & e^{V_j^R(x_j)/kT(C_j-1)} \cdot e^{-\lambda_{jR}(x_j)/kT(C_j-1)}.\end{aligned}$$ Therefore at a critical point, $$b_R(x_R) \propto \tilde{f}_R(x_R)\cdot \prod_{j\prec \partial R} b^{(R)}_j(x_j)^{(C_j-1)} e^{-V_j^R(x_j)/kT}.$$ Define $$\begin{aligned}
M_{j\to R}(x_j) &\propto& b^{(R)}_j(x_j)^{(C_j-1)} e^{-V_j^R(x_j)/kT}\\
M_{R \to j}(x_j) &\propto& \sum_{x_R\setminus x_j} \tilde{f}_R(x_R)\cdot \prod_{k\prec \partial R\setminus j} M_{k\to R}(x_k).\end{aligned}$$ Note then than $$b_j^{(R)}(x_j) = \sum_{x_R\setminus x_j} b_R(x_R) \propto M_{j\to R}(x_j) \cdot M_{R\to j}(x_j).$$
Now suppose that we are at a stationary point of the algorithm. Then from the lemma, $$b_j(x_j)^{(C_j-1)} \propto \prod_{S\succ j\::\:S \not= R} F_{j\to S}(x_j)\cdot F_{S\to j}(x_j) \propto (f_j(x_j))^{-1}\cdot \prod_{S\succ j} F_{j\to S}(x_j).$$ Also, $$e^{-V_j^R(x_j)/kT} = f_j(x_j) \cdot \prod_{S\succ j\::\: S \not=R} (F_{j\to S}(x_j))^{-1}.$$ Thus, $$M_{j\to R}(x_j) \propto b^{(R)}_j(x_j)^{(C_j-1)} e^{-V_j^R(x_j)/kT} \propto F_{j\to R}(x_j).$$ But since both these are probabilities, this proportionality is in fact an equality. Also by the lemma, $$F_{j\to R}(x_j)\cdot F_{R\to j}(x_j) \propto b_j(x_j) \propto M_{j\to R}(x_j) \cdot M_{R\to j}(x_j),$$ and so $$M_{R\to j}(x_j) = F_{R\to j}(x_j).$$ Therefore, we must have $$M_{j\to R}(x_j) \propto f_j(x_j) \cdot \prod_{S \succ j\::\:S\not=R} M_{S\to j}(x_j)$$ since this relation is satisfied by the $F$-messages. The result then follows from Proposition \[proposition:regional-BP\].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We investigate the effective interaction between two like-charged rods in the regime of large coupling parameters using both Molecular Dynamics simulation techniques and the recently introduced strong-coupling theory. We obtain attraction between the rods for elevated Manning parameters accompanied by an equilibrium surface-to-surface separation of the order of the Gouy-Chapman length. A continuous unbinding between the rods is predicted at a threshold Manning parameter $\xi_c = 2/3$.'
author:
- 'A. Naji[^1]'
- 'A. Arnold'
- 'C. Holm'
- 'R. R. Netz ($^{\ast \ast}$)'
title: 'Attraction and unbinding of like-charged rods [^2]'
---
Introduction {#sec:intro}
============
In recent years, both experiments [@Bloom] and numerical simulations [@Guld86; @Gron97; @Gron98; @Wu; @AllahyarovPRL; @Linse; @Stevens99; @Messina; @Deserno03] showed that like-charged macroions can attract each other via effective forces of electrostatic origin; one prominent example is the formation of dense packages of DNA molecules (DNA condensates) [@Bloom]. These observations indicate like-charge attraction in the presence of multivalent counterions, at low temperatures or for large surface charge density on macroions, [*i.e.*]{} when electrostatic correlations between charged particles become increasingly important. The standard mean-field Poisson-Boltzmann (PB) theory predicts repulsion between like-charged objects [@VO]. In contrast, incorporation of electrostatic correlations generates attractive interactions in agreement with a variety of experimental and numerical results [@Kjellander84; @Attard; @Oosawa; @Stevens90; @Barbosa00; @Rouzina96; @Kornyshev; @Shklo; @Arenzon99; @Netz01; @Andre; @Naji]. The importance of electrostatic correlations can be quantified by means of the [*coupling parameter*]{}, $\Xi=2\pi q^3 \ell_{\ab{B}}^2 \sigma_{\ab{s}}$, which depends on the charge valency of counterions, $q$, the surface charge density of macroions, $\sigma_{\ab{s}}$, and the Bjerrum length, $\ell_{\ab{B}}=e^2/(4\pi \varepsilon\varepsilon_0 k_{\ab{B}} T)$ (associated with a medium of dielectric constant $\varepsilon$ and at temperature $T$). The PB theory is valid in the limit of vanishingly small coupling parameter $\Xi\rightarrow 0$ [@Netz01; @Andre; @Yoram04], while non-mean-field features emerge in the opposite limit of large coupling, $\Xi\gg1$, and are typically accompanied by the formation of strongly-correlated counterion layers. For charged [*curved*]{} surfaces, one has to consider the entropy-driven counterion-condensation process as well. For rod-like macroions, as considered in this paper, counterion condensation is controlled by the so-called [*Manning parameter*]{}, $\xi=q\ell_{\ab{B}} \tau$ [@Manning69], where $\tau$ stands for the single-rod linear charge density (in units of the elementary charge $e$). For very small Manning parameter, $\xi$, counterions de-condense from the rods leading to a bare electrostatic repulsion between them [@Note00]. For sufficiently large $\xi$, on the other hand, a certain fraction of counterions remains bound to the rods and attraction becomes possible for moderate to large couplings. We are interested in the regime of Manning parameters, $\xi$, and coupling parameters, $\Xi$, where an effective rod-rod attraction is present.
Several mechanisms for the attraction between like-charged rods have been considered in the literature, including covalence-like binding [@Manning97], Gaussian-fluctuation correlations [@Oosawa], and structural correlations [@Kornyshev; @Shklo; @Arenzon99]. The two latter approaches yield attraction based on correlated fluctuations of condensed counterions on opposite rods and short-ranged correlations due to the ground-state configuration of the system, respectively. Conflicting predictions for the threshold value of Manning parameter, above which attraction between two rods is possible, have been obtained: The analysis of Ray and Manning [@Manning97], based on the standard counterion-condensation model [@Manning69], leads to attraction for $\xi>1/2$. In contrast, an attraction regime of $\xi>2$ was proposed by the counterion-condensation theory of Arenzon [*et al.*]{} [@Arenzon99]. Recent numerical simulations [@Gron97; @Deserno03], on the other hand, reveal attraction already for Manning parameters of the order of $\xi\approx 1$ and for moderate coupling parameters, though did not specifically consider the threshold value of $\xi$. Recently, a systematic treatment of correlations in highly-coupled systems has been put forward [@Netz01], that employs a perturbative scheme (virial expansion) in terms of $1/\Xi$. This scheme leads to the asymptotic strong-coupling (SC) theory, which becomes exact in the limit which is complementary to the mean-field regime, [*i.e.*]{} when $\Xi\rightarrow \infty$. The SC theory has been used to study the interaction between planar charged walls [@Netz01] and shows quantitative agreement with Monte-Carlo simulations for moderate to large coupling parameters [@Andre]. The SC mechanism of like-charge attraction qualitatively agrees with the structural-correlation scenario [@Rouzina96; @Kornyshev; @Shklo; @Arenzon99] for large macroion charges [@Netz01; @Andre; @Yoram04; @Naji]. In this letter, we study the effective interaction in a system of two like-charged rods using both Molecular Dynamics (MD) simulation methods and the SC theory. The nature of electrostatic correlations in such a system has been studied in our previous MD simulations [@Deserno03], and exhibits a competition between electrostatic and excluded-volume interactions. Here, we focus on the strong-coupling characteristics of attraction between like-charged rods. In the following, these numerical results will be compared with the predictions of the SC theory [@Naji].
Strong-coupling theory
======================
In the limit of large coupling parameter $\Xi\gg 1$, the canonical free energy of a system of fixed macroions with their neutralizing counterions, ${\mathcal F}$, admits a large-coupling expansion as ${\mathcal F}={\mathcal F}_1/\Xi+{\mathcal F}_2/\Xi^2+\ldots$ [@Naji], where the coefficients ${\mathcal F}_1, {\mathcal F}_2,\ldots$ are expressed in terms of weighted integrals of Mayer functions over counterionic degrees of freedom, which are convergent for systems of counterions at charged macroscopic objects [@Netz01]. The first term in the above expression generates the leading non-vanishing contribution to statistical observables such as effective interaction between macroions for $\Xi\gg 1$; hence ${\mathcal F}_{\ab{SC}}={\mathcal F}_1/\Xi$ is referred to as the SC free energy of the system and ${\mathcal F}_1$ as the rescaled SC free energy. The SC free energy takes a very simple form in terms of the one-particle interaction energies between counterions and macroions (see, [*e.g.*]{}, eq. (\[eq:SCfree\]) below), and the mutual interaction between counterions themselves enters only in the sub-leading corrections [@Netz01; @Naji]. Physically, this reflects the fact that for $\Xi\rightarrow \infty$, counterions at charged objects are surrounded by an extremely large correlation hole and thus, the single-particle term becomes the dominant contribution [@Netz01; @Andre; @Yoram04; @Naji] (see also the discussion).
Let us consider two infinitely long similar rods of length $H$ and radius $R$ that are located parallel at axial separation $D$ in a rectangular box of lateral extension $L$ containing also neutralizing counterions of valency $q$ (fig. \[fig:fig1\]a). The rods are impenetrable to counterions and no additional salt is present. It is convenient to establish a dimensionless formalism for such systems by rescaling all length scales by the Gouy-Chapman length $\mu=1/(2\pi q \ell_{\ab{B}} \sigma_{\ab{s}})$, such as ${\tilde x}=x/\mu$, [*etc.*]{} [@Netz01], where $\sigma_{\ab{s}}=\tau/(2\pi R)$ stands for the surface charge density of the rods. Clearly, the rod radius in rescaled units, ${\tilde R}=R/\mu$, coincides with Manning parameter, [*i.e.*]{} ${\tilde R}=\xi=q \ell_{\ab{B}} \tau$. (We may occasionally use $\xi$ or ${\tilde R}$ to denote the Manning parameter.) The rescaled rod length, ${\tilde H}$, is related to the total number of counterions, $N$, via the global electroneutrality condition as ${\tilde H}=N \Xi/\xi$. The rescaled SC free energy of this system per rescaled unit length follows as [@Naji] $$\frac{{\mathcal F}_1}{\tilde H}=
-2 \xi^2\ln {\tilde D}
-2\xi \ln\left[\int_V \upd{\tilde x}\upd{\tilde y}\,\,
{\tilde r}_1^{-2\xi}{\tilde r}_2^{-2\xi}\right].
\label{eq:SCfree}$$ The integral runs over the volume $V$ available for counterions within the confining box, excluding the rods. We defined ${\tilde r}_{1,2}=[({\tilde x}\pm{\tilde D}/2)^2+{\tilde y}^2]^{1/2}$ as the radial distances from the rods axes. The first term in eq. (\[eq:SCfree\]) is nothing but the bare electrostatic repulsion between the rods. The second term involves the leading (energetic and entropic) contributions from counterions, which leads to a counterion-mediated attraction between sufficiently highly charged rods [@Note0]. This term also reflects the de-condensation process of counterions at low Manning parameters: For $\xi<1/2$, the counterionic integral in eq. (\[eq:SCfree\]) diverges in the infinite-volume limit ${\tilde L}\rightarrow \infty$, hence the distribution of counterions around the rods and the counterion-mediated force vanish, [*i.e.*]{} the rods purely repel each other [@Note00]. For large Manning parameter $\xi\gg 1$, on the other hand, the rescaled SC free energy, ${\mathcal F}_1$, shows a long-ranged attraction and a pronounced global minimum at a small axial separation ${\tilde D}_\ast\approx 2{\tilde R}$, which is nothing but the equilibrium axial separation of the rods [@Naji]. For $\xi\gg 1$, the counterionic distribution becomes strongly localized in a narrow region between the rods, indicating that the SC attraction is accompanied by strong accumulation of counterions in the intervening region. This allows for a saddle-point calculation of the counterionic integral in eq. (\[eq:SCfree\]), giving the approximate form of ${\mathcal F}_1$ in the vicinity of its local minimum as [@Naji] $${\mathcal F}_1/{\tilde H} \approx 6 \xi^2\ln {\tilde D}
-2 \xi \ln ({\tilde D}-2{\tilde R}).
\label{eq:SCfree_Rlarge}$$ The equilibrium surface-to-surface distance of the rods for large $\xi$ is obtained by minimizing expression (\[eq:SCfree\_Rlarge\]) as ${\tilde \Delta}_\ast\equiv {\tilde D}_\ast-2{\tilde R}\approx 2/3+{\mathcal O}(1/\xi)$, which corresponds, in real units, to a surface separation of the order of the Gouy-Chapman length, $\mu$ [@Note2]. Figure \[fig:fig1\]b shows the global behavior of the rescaled equilibrium surface-to-surface distance, ${\tilde \Delta}_\ast$, obtained by numerical minimization of the full SC free energy, eq. (\[eq:SCfree\]), as a function of $\xi$ for several box sizes. For large $\xi$, attraction is dominant, the rods form a closely-packed bound state, and the volume of the bounding box is irrelevant due to strong condensation of counterions. However, for decreasing $\xi$, attraction is weakened and rods eventually unbind in a continuous fashion when ${\tilde L}\rightarrow \infty$. For two unconfined rods (solid curve in fig. \[fig:fig1\]b), the unbinding transition occurs at $\xi_c=2/3$ and exhibits an asymptotic power-law behavior as a function of the reduced Manning parameter as ${\tilde \Delta}_\ast\sim (\xi -2/3)^{-\alpha}$, where $\alpha\approx 3/2$ [@Naji]. Note that the predicted onset of rod-rod attraction, $\xi_c=2/3$, is somewhat larger than the counterion-condensation threshold for the combined system of two rods, $\xi=1/2$ [@Manning97], and smaller than the condensation threshold for a single rod, $\xi =1$ [@Manning69]. This reflects the subtle crossover from the case of two coupled rods (when the distance between them is small) to the case of two decoupled rods as the axial distance diverges when the unbinding threshold is approached. The predicted attractive force between rods, $F_{\ab{rods}}= -(k_{\ab{B}}T)\partial {\mathcal F}_{\ab{SC}}/\partial D$ (in actual units), is found from eq. (\[eq:SCfree\_Rlarge\]) to be inversely proportional to the axial distance, $D$. This force increases and tends to a temperature-independent value upon increasing $\xi$ (or decreasing temperature) [@Gron97], which is nothing but the [*energetic attraction*]{} mediated by counterions sandwiched between closely-packed rods, [*i.e.*]{} $F_{\ab{rods}}=-3e^2\tau^2/(2\pi \varepsilon \varepsilon_0 D)$ [@Naji]. The SC attraction for large $\xi$, thus, originates from the energetic contributions induced by the ground-state configuration of counterions and in this respect, agrees qualitatively with the low-temperature picture for like-charge attraction [@Rouzina96; @Kornyshev; @Shklo; @Arenzon99]. Though for the specific case of two charged rods, the quantitative predictions of the SC theory for the (energetic) inter-rod force at large $\xi$, and for the attraction threshold differ from the results of the model studied in ref. [@Arenzon99], which incorporates a discrete charge pattern for the rods and obtains attraction for $\xi>2$ [@Arenzon99].
Simulation method
=================
To study the interaction between two like-charged rods numerically, we have performed extensive Molecular Dynamics simulations making use of a Langevin thermostat to drive the system into the canonical state. The geometry of the simulated system is similar to what we have sketched in fig. \[fig:fig1\]a; the outer confining box is here chosen to be cylindrical. Other parameters are defined in the same way unless explicitly mentioned. Charged particles interact via bare Coulombic interaction and excluded-volume interactions. We apply periodic boundary conditions in the direction parallel to the rods axes. This leads to summation of infinite series for Coulomb interaction over all periodic images, which is handled numerically using the MMM1D summation method [@mmm2d]. As we are only interested in examining electrostatic aspects of the effective rod-rod interaction, excluded-volume interactions are considered only between rods and counterions and not between counterions themselves. (Extended numerical results with excluded-volume interaction between counterions will be presented elsewhere [@Arnold03].) We employ a shifted Lennard-Jones potential as $$V_{\ab{LJ}}(r) =
\begin{cases}
4\,\epsilon
\left[\left(\frac{a}{r-R_0}\right)^{12} -
\left(\frac{a}{r-R_0}\right)^6 +
\frac{1}{4} \right] &\quad : \quad (r-R_0) < 2^{1/6}a \\
0 &\quad : \quad \text{otherwise}
\end{cases} ,
\label{eq:LJ}$$ where $R_0$ is an offset used to control the radius of the rods, $a$ defines our basic length scale in the simulations, and $\epsilon\approx 1 k_{\ab{B}}T$. Therefore, $R=R_0+a$ may be regarded as the effective rod radius and is directly compared with the SC result (with hard-core rods). This is justified since in the simulations only a negligible fraction of counterions penetrates the above potential to reach radial distances smaller than $R$. In the simulations, the diameter of the (cylindrical) outer box is chosen as $8D$. For the final comparison (fig. \[fig:fig2\]), the theoretical curves are also calculated using a similar constraint, though for simplicity and as explained before, calculations have been done for a square box of edge size $L=8D$. As we have explicitly checked, the results are insensitive to small changes in the box size for the considered range of Manning parameter (see also fig. \[fig:fig1\]b), and the SC predictions can thus be compared with the simulations. To obtain the location of the zero-force (equilibrium) point in the force curves, we have employed a bisection algorithm followed by a linear regression. The error bars are determined from the error of the linear regression using the error propagation method.
Results and discussion
======================
To clearly separate the attraction and repulsion regimes of two like-charged rods numerically, we study the rescaled equilibrium surface-to-surface separation of rods, ${\tilde \Delta}_\ast$, as a function of Manning parameter, $\xi$, in our simulations. The results will then be compared with the SC prediction (fig. \[fig:fig1\]b). To this end, we fix the actual surface-to-surface distance between the rods, $\Delta=D-2R$, and their linear charge density, $\tau$, together with the counterion valency, $q$, but vary the Bjerrum length, $\ell_{\ab{B}}$, and the actual rod radius, $R$. By doing so, the Gouy-Chapman length associated with this system, $\mu=R/(\ell_{\ab{B}} q \tau)$, is varied, which then allows for determining the equilibrium surface-to-surface separation [*in rescaled units*]{}, ${\tilde \Delta}_\ast=\Delta_\ast/\mu$. The results will also depend upon the electrostatic coupling parameter, $\Xi=2\pi q^3 \ell_{\ab{B}}^2 \sigma_{\ab{s}}$, which is finite in the simulations and may be written as $$\Xi=\xi\,{\tilde \Delta}\,\gamma_{\ab{RB}},
\label{eq:coupling}$$ where $\gamma_{\ab{RB}}=q/(\tau \Delta)$ is a dimensionless parameter referred to here as the [*Rouzina-Bloomfield parameter*]{} (see the discussion below). The simulation results for the equilibrium surface-to-surface separation of the rods are shown in fig. \[fig:fig2\] for different $\gamma_{\ab{RB}}$ ranging from 3 to 60. The data have been obtained using various combinations of fixed $\tau a=1.0, 0.33, 0.10$, $q=1, 3, 5, 10$ and $\Delta/a=0.5, 1.0, 2.0, 2.5$. For a given set of system parameters ($\tau$, $q$ and $\Delta$) chosen in the simulations, $\gamma_{\ab{RB}}$ is fixed and identifies a single curve. As seen, upon increasing the Rouzina-Bloomfield parameter, $\gamma_{\ab{RB}}$, the equilibrium separation decreases indicative of a stronger attraction operating between the rods, and at the same time, the agreement between simulation data and the SC prediction (solid curve in fig. \[fig:fig2\]) becomes progressively better. The agreement is quantitative for large $\gamma_{\ab{RB}}$, [*i.e.*]{} $\gamma_{\ab{RB}}=50$ and 60. (The data set with $\gamma_{\ab{RB}}=60$ is obtained using $q=3$, $\tau a=0.1$ and $\Delta/a=0.5$, and data with $\gamma_{\ab{RB}}=50$ are obtained using $q=5$, $\tau a=0.1$ and $\Delta/a=1.0$.) The observed trend for increasing $\gamma_{\ab{RB}}$ is associated with the increase of the coupling parameter, $\Xi$, in the simulated system (see eq. (\[eq:coupling\])), which eventually exhibits the strong-coupling regime. (For instance, for a moderate Manning parameter of $\xi=3.0$, $\Xi$ increases from 18 for cross symbols up to about 100 for filled diamonds.)
It is noteworthy that $\gamma_{\ab{RB}}$, as defined above, can be expressed as $\gamma_{\ab{RB}}=\delta/\Delta$, where $\delta=q/\tau$ is the projected distance between counterions along a single rod as follows from the local electroneutrality condition. Note that $\delta$ roughly gives the correlation hole size around counterions at charged surfaces [@AllahyarovPRL; @Linse; @Deserno03; @Andre; @Yoram04]. The appearance of strong attractive forces for $\gamma_{\ab{RB}}>1$ was first addressed by Rouzina and Bloomfield [@Rouzina96] for a system of two planar charged walls. Analytical and numerical results for planar walls [@Netz01; @Andre; @Yoram04] have in fact shown that the [*relative*]{} magnitude of higher-corrections to the asymptotic SC theory decreases with $\gamma_{\ab{RB}}$, and the SC regime at finite $\Xi$ is characterized by the Rouzina-Bloomfield criterion $\gamma_{\ab{RB}}>1$. Physically, this corresponds to a large correlation hole size around counterions at macroion surfaces as compared with the macroions surface separation, [*i.e.*]{} $\delta>\Delta$, leading to a dominant contribution from single (isolated) counterions [@Netz01; @Andre; @Yoram04], which is formally obtained within the SC scheme [@Netz01]. It becomes exceedingly difficult to perform systematic analytic calculations to examine higher-order corrections to SC predictions given the geometry of the two-rod system. Nonetheless, the present numerical results clearly indicate a qualitatively similar trend for increasing $\gamma_{\ab{RB}}$ in this system and that, the SC regime is characterized by Rouzina-Bloomfield criterion. Due to convergence limitations, our numerical investigation so far has been limited to the range of $\xi>0.8$ and thus do not cover the close vicinity of the unbinding transition. (It becomes more difficult to obtain good data as the rods equilibrium distance rapidly increases for small $\xi$.) However, the excellent convergence of the simulation results to the asymptotic SC prediction suggests an attraction threshold of about $\xi_c = 2/3$ for two rods in the strong-coupling limit as obtained using the SC theory. Note that for small Manning parameters where counterions de-condense ($\xi\leq 1/2$), electrostatic correlations are suppressed due to entropic dilution of the counterionic cloud. Physically, this regime does not exhibit strong energetic coupling and higher-order corrections to the SC theory may become important. Nevertheless as we showed, the SC theory captures the de-condensation process and leads to a qualitatively consistent picture for the whole range of Manning parameters [@Note00]. (Note that the predicted onset of de-condensation at $\xi=1/2$ and also the bare macroionic repulsion for small $\xi$ [*quantitatively*]{} agree with previous findings [@Manning97].) How and whether the predicted attraction threshold ($\xi_c = 2/3$) and the unbinding behavior is influenced by the de-condensation process of counterions at small $\xi$ remains to be clarified.
In summary, we have investigated the attraction and repulsion regimes of two like-charged rods in terms of Manning parameter, $\xi$, and the coupling parameter, $\Xi$, by combining numerical and analytical approaches. For large $\xi$, the rods form a closely-packed bound state of small surface-to-surface separation of about the Gouy-Chapman length, $\mu$ [@Note2]. The attraction is weakened and the rods unbind for decreasing $\xi$. The attraction threshold is found at $\xi_c=2/3$, and the unbinding of rods proceeds in a continuous fashion characterized by a power-law for the rods surface separation, $\Delta_\ast\sim (\xi-2/3)^{-\alpha}$, with an exponent $\alpha\approx 3/2$. The predicted attraction regime for large $\xi$ is accessible experimentally using, [*e.g.*]{}, DNA molecules ($\tau\, e\approx 6\,e/\un{nm}$) in the presence of multivalent counterions such as spermidine ($q=3$), which yields $\xi\approx 12$ and $\Xi\approx 80$. For these values, our results predict attraction with an equilibrium surface separation $\Delta_\ast\sim \mu$, where for the DNA-spermidine system $\mu\sim 1$Å[@Note2]. It should be noted, however, that for DNA-like molecules (large $\tau$), other factors such as excluded-volume interaction between counterions [@Deserno03; @Arnold03] and the helical structure of DNA [@Kornyshev; @Shklo] may become important and contribute additional components to the total effective force. The predicted continuous unbinding occurs at quite small values of Manning parameter. Experimentally, this phenomenon could be studied with weakly charged stiff polymers, such as poly-phenylene with a suitably chosen small density of charged side-chains. The effects of salt can be qualitatively accounted for by associating the bounding box size with the screening length: Thus only close to the unbinding do we expect addition of salt to matter. The effects of finite chain stiffness and the bundling of many chains constitute interesting applications for the future. A.N. and R.R.N. acknowledge funds from DFG German-French Network. C.H. and A.A. acknowledge funds from the Zentrum für Multifunktionelle Werkstoffe und Miniaturisierte Funktionseinheiten, grant BMBF 03N 6500, DFG grant Ho 1108/11-1, and SFB 625.
[1]{}
V. Bloomfield, Biopolymers [**31**]{}, 1471 (1991); H. Strey et al., Curr. Opin. Struct. Biol. [**8**]{}, 309 (1998); M. Olvera de la Cruz et al., J. Chem. Phys. [**103**]{}, 5781 (1995); J. X. Tang et al., Ber. Bunsen-Ges. Phys. Chem. [**100**]{}, 1 (1996); J. C. Butler et al., Phys. Rev. Lett. [**91**]{}, 028301 (2003).
L. Guldbrand, B. Jönsson, H. Wennerström, P. Linse, J. Chem. Phys. [**80**]{}, 2221 (1984); L. Guldbrand, L. G. Nilsson, and L. Nordenskiöld, J. Chem. Phys. [**85**]{}, 6686 (1986); A. P. Lyubartsev and L. Nordenskiöld, J. Phys. Chem. [**99**]{}, 10373 (1995).
N. Gr[ø]{}nbech-Jensen, R. Mashl, R. Bruinsma, and W. Gelbart, Phys. Rev. Lett. [**78**]{}, 2477 (1997).
N. Gr[ø]{}nbech-Jensen, K. M. Beardmore, P. Pincus, Physica A [**261**]{}, 74 (1998).
J. W. Wu, D. Bratko, and J. M. Prausnitz, Proc. Natl. Acad. Sci. USA [**95**]{}, 15169 (1998).
E. Allahyarov, I. D’Amico, and H. Löwen, Phys. Rev. Lett. [**81**]{}, 1334 (1998).
P. Linse and V. Lobaskin, Phys. Rev. Lett. [**83**]{}, 4208 (1999); J. Chem. Phys. [**112**]{}, 3917 (2000).
M. J. Stevens, Phys. Rev. Lett. [**82**]{}, 101 (1999).
R. Messina, C. Holm, and K. Kremer, Phys. Rev. Lett. [**85**]{}, 872 (2000).
M. Deserno, A. Arnold, and C. Holm, Macromolecules [**36**]{}, 249 (2003).
E. J. Verwey and J. T. G. Overbeek, [*Theory of the stability of lyophobic colloids*]{} (Elsevier, Amsterdam, 1948); J. C. Neu, Phys. Rev. Lett. [**82**]{}, 1072 (1999).
R. Kjellander and S. Mar[č]{}elja, J. Chem. Phys. [**82**]{}, 2122 (1985).
P. Attard, D. J. Mitchell, and B. W. Ninham, J. Chem. Phys. [**88**]{}, 4987 (1988); R. Podgornik, J. Phys. A: Math. Gen. [**23**]{}, 275 (1990); P. A. Pincus and S. A. Safran, Europhys. Lett. [**42**]{}, 103 (1998); D. B. Lukatsky and S. A. Safran, Phys. Rev. E [**60**]{}, 5848 (1999); M. Kardar and R. Golestanian, Rev. Mod. Phys. [**71**]{}, 1233 (1999); A. W. C. Lau and P. Pincus, Phys. Rev. E [**66**]{}, 041501 (2002); A. W. C. Lau, D. Levine, and P. Pincus, Phys. Rev. Lett. [**84**]{}, 4116 (2000).
F. Oosawa, Biopolymers [**6**]{}, 1633 (1968); J.-L. Barrat and J.-F. Joanny, Adv. Chem. Phys. [**XCIV**]{}, 1 (1996); B.-Y. Ha and A. J. Liu, Phys. Rev. Lett. [**79**]{}, 1289 (1997); R. Podgornik and V. A. Parsegian, Phys. Rev. Lett. [**80**]{}, 1560 (1998).
M. J. Stevens and M. O. Robbins, Europhys. Lett. [**12**]{}, 81 (1990).
M. C. Barbosa, M. Deserno, and C. Holm, Europhys. Lett. [**52**]{}, 80 (2000).
I. Rouzina and V. A. Bloomfield, J. Phys. Chem. [**100**]{}, 9977 (1996).
A. Kornyshev and S. Leikin, J. Chem. Phys. [**107**]{}, 3656 (1997); Phys. Rev. Lett. [**82**]{}, 4138 (1999).
B. I. Shklovskii, Phys. Rev. Lett. [**82**]{}, 3268 (1999); A. Yu. Grosberg, T. T. Nguyen, and B. I. Shklovskii, Rev. Mod. Phys. [**74**]{}, 329 (2002).
J. J. Arenzon, J. F. Stilck, and Y. Levin, Eur. Phys. J. B [**12**]{}, 79 (1999); Y. Levin, Rep. Prog. Phys. [**65**]{}, 1577 (2002).
R. R. Netz, Eur. Phys. J. E [**5**]{}, 557 (2001).
A. G. Moreira and R. R. Netz, Phys. Rev. Lett [**87**]{}, 078301 (2001); Eur. Phys. J. E [**8**]{}, 33 (2002).
Y. Burak, D. Andelman and H. Orland, Phys. Rev. E [**70**]{}, 016102 (2004).
G. S. Manning, J. Chem. Phys. [**51**]{}, 924 (1969).
The de-condensation process at small $\xi$ occurs in the whole range of coupling parameters and does not exhibit energetic strong-coupling characteristics (Naji and Netz, preprint (2004)).
J. Ray and G. S. Manning, Macromolecules [**30**]{}, 5739 (1997); Langmuir [**10**]{}, 2450 (1994).
A. Naji and R. R. Netz, Eur. Phys. J. E [**13**]{}, 43 (2004).
A. Arnold and C. Holm, to be published.
Note that the translational entropy of the rods is irrelevant for large rod length.
A. Arnold and C. Holm, Comp. Phys. Comm. [**148**]{}, 327 (2002).
The SC results hold for arbitrary counterion size, when hard-core interactions are taken into account. In this case, $R$ measures the distance of closest approach between rods and counterions, and the equilibrium surface separation of rods (in actual units) is found to increase by an amount equal to the counterion diameter [@Naji].
[^1]: Present address: Physics Dept., Technical University of Munich, 85748 Garching, Germany.
[^2]: Revised version published in Europhys. Lett. [**67**]{}, 130 (2004).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We show that deformations of a surjective morphism onto a Fano manifold of Picard number 1 are unobstructed and rigid modulo the automorphisms of the target, if the variety of minimal rational tangents of the Fano manifold is non-linear or finite. The condition on the variety of minimal rational tangents holds for practically all known examples of Fano manifolds of Picard number 1, except the projective space. When the variety of minimal rational tangents is non-linear, the proof is based on an earlier result of N. Mok and the author on the birationality of the tangent map. When the varieties of minimal rational tangents of the Fano manifold is finite, the key idea is to factorize the given surjective morphism, after some transformation, through a universal morphism associated to the minimal rational curves.'
---
¶[[**P**]{}]{} Ø[[O]{}]{}
[**Unobstructedness of deformations of holomorphic maps** ]{}\
[**onto Fano manifolds of Picard number 1**]{}
[^1]
Key words: deformation of holomorphic maps, Fano manifolds, variety of minimal rational tangents
2000MSC: 14J45, 32H02
Introduction
============
We will work over the complex numbers. A variety or (a manifold) will be assumed to be irreducible except when we say ‘the variety of minimal rational tangents’, which may have finitely many components. See Section 2 for the definition. For a complex manifold $Y$, $T(Y)$ denotes its tangent bundle and $T_y(Y)$ denotes the tangent space at a point $y \in Y$.
For two projective varieties $X$ and $Y$, denote by ${\rm Hom}^s(Y, X)$ the space of surjective holomorphic maps $Y \to X$ and by ${\rm Aut}_o(X)$ the identity component of the group of biregular automorphisms of $X$. In \[HM3\] and \[HM4\], the following result was proved.
Theorem 1.1 was first proved for the rational homogeneous space $X=G/P$ in \[HM1\]. It was proved when the variety of minimal rational tangents has non-degenerate Gauss map in \[HM2\]. This was surpassed by \[HM4\] which proves it when the variety of minimal rational tangents is non-linear. The proofs in these three papers are of the same nature. The proof when the variety of minimal rational tangents is finite is quite different and appeared in \[HM3\].
The condition that the variety of minimal rational tangents is non-linear or finite holds for practically all known examples of Fano manifolds of Picard number 1, except the projective space. In fact, we have the following non-linearity conjecture:
There are some partial results toward Conjecture 1.2. For example, it was proved for Fano manifolds of index $\geq \frac{\dim X
+3}{2}$ in \[H1, Corollary 2.3\].
For the projective space, the assertion in Theorem 1.1 certainly does not hold. In this sense, Theorem 1.1 is a reasonably satisfactory result, except that it does not say whether ${\rm Hom}^s(Y, X)$ is reduced. In other words, it does not address the unobstructedness of infinitesimal deformations. The goal of this paper is precisely to remedy this. Our main result is the following, which also gives an alternative proof of Theorem 1.1.
Note that $H^0(Y, f^*T(X))$ is the Zariski tangent space to ${\rm
Hom}^s(Y, X)$ at $[f]$ and $H^0(X, T(X))$ is the Zariski tangent space to ${\rm Aut}_o(X)$ at the identity. Thus the identity $$H^0(Y, f^* T(X)) = f^* H^0(X, T(X))$$ implies that the natural morphism ${\rm Aut}_o(X) \to {\rm Hom}^s(Y, X)$ sending each $g\in
{\rm Aut}_o(X) $ to $ g \circ f \in {\rm Hom}^s(Y, X)$ is bijective, implying the last sentence of Theorem 1.3.
The proof of Theorem 1.3 when the variety of minimal rational tangents is non-linear is rather simple modulo the main result of \[HM4\] on the tangent map. In retrospect, this proof is the culmination of successive refinements of the arguments in \[HM2\] and \[HM4\]. The final formulation is much simpler than the old proofs and will be given in Section 2.
The difficult case is when the variety of minimal rational tangents is finite. The key idea of the proof in that case is to show that, after a certain transformation, the morphism $f: Y \to X$ can be factorized through the universal morphism for the family of minimal rational curves. This factorization is established in Section 4. Combining this with an idea from \[H1\] on the behavior of minimal rational curves near the branch locus of $f$ explained in Section 5, the proof is completed in Section 6 by using an argument in \[H2\].
Proof of Theorem 1.3 when the variety of minimal rational tangents of $X$ is non-linear
========================================================================================
Throughout this paper, we will denote by $X$ a Fano manifold of Picard number 1. We refer the readers to \[K\] for basics on the space of rational curves on $X$. An irreducible component ${\cal K}$ of the space of rational curves on $X$ is called a [*minimal component*]{} if for a general point $x \in X$, the subscheme ${\cal K}_x$ of ${\cal K}$ consisting of members passing through $x$ is non-empty and complete. In this case, the subvariety ${\cal C}_x$ of the projectivized tangent space ${\P}T_x(X)$ consisting of the tangent directions at $x$ of members of ${\cal K}_x$ is called the [*variety of minimal rational tangents*]{} at $x$ (see \[HM4\] for more details). We say that the variety of minimal rational tangents of $X$ is [*non-linear*]{} if $\dim {\cal C}_x
>0$ and some component of ${\cal C}_x$ is not a linear subspace in $\P T_x(X)$. Otherwise, we say that the variety of minimal rational tangents is linear.
For a general member $C$ of $\K$, the normalization $\nu: \P_1 \to C
\subset X$ is an immersion and $$\nu^*T(X) = \O(2) \oplus \O(1)^p
\oplus \O^q$$ where $p = \dim \C_x$ for a general $x \in X$ and $p+q
= \dim X -1$. Denote by $H^0(C, T^*(X))$ the vector space $$H^0(\P_1, \nu^*T^*(X)) = H^0(\P_1, \O^q).$$ For a non-singular point $x \in C$, denote by $H^0(C, T^*(X))_x \subset T_x^*(X)$ the $q$-dimensional subspace of the cotangent space at $x$ given by evaluating the elements of $H^0(C, T^*(X))$ at the point $x$.
. From the irreducibility of $\K$, it suffices to show that the component $\C_1$ of $\C_x$ corresponding to $\K_1$ is a linear subspace. For a general member $C$ of ${\cal K}_1$, $x$ is a non-singular point of $C$. Denote by $$\P H^0(C, T^*(X))_x \subset \P T_x^*(X)$$ the projectivization of $H^0(C, T^*(X))_x$. The closure of the union of $\P H^0(C,
T^*(X))_x$ as $C$ varies over general points of ${\cal K}_1$, is the dual variety of ${\cal C}_1 \subset \P T_x(X)$ by \[HR, Corollary 2.2\]. Thus the existence of $v$ implies that the dual variety of ${\cal C}_1$ is linearly degenerate in ${\bf P}T^*_x(X)$, i.e., ${\cal C}_1$ is a cone. Thus Proposition 2.1 follows from \[HM4, Proposition 13\], which says that ${\cal C}_1$ cannot be a cone unless it is a linear subspace. $\Box$
The next proposition is \[HM2, Lemma 4.2\].
. Let $\nu: \hat{C} \to C$ be the normalization of $C$ and let $\varphi \in H^0(\hat{C}, \nu^*T^*(X))$ be a section of the cotangent bundle of $X$ on $\hat{C}$. Let $\nu': \hat{C}' \to C'$ be the normalization of $C'$ and $\hat{f}: \hat{C}' \to \hat{C}$ be the lifting of $f$. Let $\varphi' \in H^0(\hat{C}',
(\nu \circ \hat{f})^*T^*(X))$ be the section induced by $\varphi$ and $\hat{\sigma} \in H^0( \hat{C}', ( f \circ \nu')^* T(X))$ be the section induced by $\sigma$. Since $ \nu \circ \hat{f} = f
\circ \nu'$, the pairing $\varphi'(\hat{\sigma})$ is a holomorphic function on $\hat{C}'$, hence is constant. It follows that $\varphi'(\sigma_{y_1} ) = \varphi'(\sigma_{y_2}).$ Thus $\sigma_{y_1} - \sigma_{y_2}$ is annihilates the evaluation of $\varphi$ at $x$. $\Box$
Now we can prove Theorem 1.3 when the variety of minimal rational tangents of $X$ is non-linear.
. Fix an element $\sigma \in H^0(Y, f^*T(X)) \setminus
f^* H^0(X, T(X)).$ For each $y \in Y$, let $\sigma_y \in
T_{f(y)}(X)$ be the corresponding tangent vector of $X$. Associated to $\sigma$, we have the projective subvariety $\Sigma
\subset T(X)$ defined by $$\Sigma := \{ \sigma_y \in T_{f(y)} (X), y \in Y\}.$$ Since $\sigma \not\in f^*H^0(X, T(X))$, the natural projection $\pi:
\Sigma \to X$ is a finite morphism of degree $>1$ and $\sigma$ induces a natural section $\sigma'$ of $\pi^*T(X)$ with $\sigma'
\not\in \pi^* H^0(X, T(X))$. Thus replacing $Y$ by $\Sigma$, we may assume that $f: Y \to X$ is a finite morphism and for any $x \in X$, $$\sigma_{y_1} \neq \sigma_{y_2} \mbox{ as vectors in } T_x(X)
\mbox{ for each } y_1 \neq y_2 \in f^{-1}(x).$$
Let $x$ be a general point of $X$ and $\K_1$ be an irreducible component of $\K_x$. By Proposition 2.2, there exist two distinct points $y_1, y_2 \in f^{-1}(x)$ such that for each general member $C$ of $\K_1$, there exists an irreducible component $C'$ of $f^{-1}(C)$ with $\{ y_1, y_2 \} \subset C'$. Then by Proposition 2.3, $H^0(C, T^*(X))_x$ is annihilated by $\sigma_{y_1}-
\sigma_{y_2}$ for all general members $C$ of $\K_1$. Applying Proposition 2.1 with $v= \sigma_{y_1} - \sigma_{y_2}$, we conclude that the variety of minimal rational tangents of $X$ is linear. $\Box$
Free curves with trivial normal bundle
======================================
It is convenient to introduce the following notion. Let $Y$ be a projective manifold of dimension $n$ and $C \subset Y$ be an irreducible curve. We say that $C$ is a [*free curve with trivial normal bundle*]{} if the following holds.
\(i) Under the normalization $\nu: \hat{C} \to C$, we have an exact sequence of vector bundles on $\hat{C}$ $$0 \longrightarrow T(\hat{C}) \longrightarrow \nu^* T(Y)
\longrightarrow N_C \longrightarrow 0$$ where the second arrow is the differential of $\nu: \hat{C} \to Y$ and $N_C$ is a trivial bundle of rank $=(n-1)$ on $\hat{C}$.
\(ii) Deformations of $C$ with constant geometric genus cover an open subset of $X$.
The germ of the space of deformations of $C$ with constant geometric genus must have dimension $\geq n-1$. The Zariski tangent space to this space at the point corresponding to $C$ is $H^0(\hat{C}, N_C)$, which has dimension $n-1$ from the triviality of the normal bundle. Thus the germ of this space of deformations of $C$, which we denote by ${\cal M}_C$, is non-singular. The following is obvious from the deformation theory of submanifolds.
From now throughout the rest of this paper, we will fix a Fano manifold $X$ of Picard number 1 and a minimal component $\K$ such that the variety of minimal rational tangents at a general point is finite. Then a general member $C$ of $\K$ is a free curve with trivial normal bundle and the germ ${\cal M}_C$ can be realized by an open neighborhood of $[C] \in \K$. By desingularizing the universal family over $\K$ (see \[K, II.2.12\] for the definition of the universal family), we have the following. The proof, which is quite standard, will be omitted.
Let us denote by $T^{\rho} \subset T(\rho^{-1}(Z_o))$ the relative tangent bundle of $\rho$ over $\rho^{-1}(Z_o)$. Let ${\cal C}
\subset {\bf P}T(X)$ be the closure of the union of ${\cal C}_x$’s for general points $x \in X$. Let $\hat{\C} \subset T(X)$ be the cone over $\C$. Denote by $O \subset T(X)$ the zero section and by $\pi: T(X) \to X$ the natural projection. The following is immediate.
. That $C$ intersects the branch divisor transversally is obvious from Proposition 3.2 (b). The fact that $C'$ is a free curve with trivial normal bundle is precisely \[HM3, Proposition 6\]. The canonical isomorphisms and the equivalence of germs are obvious from the isomorphism of two trivial vector bundles $
N_{C'} \cong \hat{f}^* N_C$ induced by the differential $df: T(Y)
\to \hat{f}^*T(X)$. $\Box$
. This is a consequence of Proposition 2.3. Since $C$ has trivial normal bundle, $H^0(C, T^*(X))_x$ is the conormal space of $C$ at $x$. Thus $\sigma_{y_1} - \sigma_{y_2} \in T_x(C)$. $\Box$
Factorization through $\mu$
===========================
In the setting of Theorem 1.3, given a section $\sigma \in H^0(Y,
f^*T(X))$, the values of $\sigma$ define a projective variety in $T(X)$ dominant over $X$, as explained in the proof of Proposition 2.4. In fact, Theorem 1.3 is equivalent to the statement that a projective variety in $T(X)$ dominant over $X$ must be a section of $T(X)$. In other words, we have to prove that there do not exist projective varieties of $T(X)$ which have degree $>1$ over $X$. The goal of this section is to show that given a projective variety $\Sigma \subset T(X)$ of degree $>1$ over $X$, the difference transform of $\Sigma$ contains an irreducible component that has very special properties with respect to the morphisms $\mu, \rho$ of Proposition 3.2. It should be mentioned that all the propositions proved from now on, except Proposition 5.1, are under the assumption of the existence of $\Sigma $ of degree $>1$, which will lead eventually to contradiction. In this sense all these propositions are of hypothetical nature.
. For a general $\zeta \in Z_o$, there exists an irreducible component $P'_{\zeta}$ of $\pi^{-1}(P_{\zeta}) \cap
\Sigma$ such that the projection $P'_{\zeta} \to P_{\zeta}$ is finite of degree $>1$ by Proposition 2.2. Thus we can choose two $a_1 \neq a_2$ on $P'_{\zeta}$ over $x \in P_{\zeta}$. The point $(a_1, a_2) \in \Sigma \times \Sigma$ lies in $\Sigma \times_X
\Sigma.$ From the generality of the choice of $\zeta$ and $x$, there is a unique component $\Sigma^{\sharp}$ containing $(a_1, a_2)$. Certainly, $\Sigma^{\sharp}$ satisfies the required property from the irreducibility of $Z_o$. $\Box$
. We will apply Proposition 3.5 with $Y = \Sigma, f=
\pi|_{\Sigma}$ and $C = P_{\zeta}$. There is a tautological section $\sigma \in H^0(Y, f^*T(X))$ defined by $$\sigma_a = a \in T_x(X) \; \mbox{ for each } a \in \Sigma \cap
T_x(X).$$ By Proposition 3.5, $$a_1 - a_2 \; \in \;
T_x(P_{\zeta}) \; \subset \; \hat{\C}.$$ As $\zeta$ varies over general points of $Z_o$, the element $a_1-a_2$ varies over an open subset in the irreducible $\delta( \Sigma^{\sharp})$. It follows that $\delta(\Sigma^{\sharp}) \subset \hat{\C}$. Since $a_1 \neq
a_2$, $\delta(\Sigma^{\sharp})$ is not contained in the zero section $O$. The dominant rational map $\chi^{\sharp}$ is certainly generically finite. $\Box$
. Choose a desingularization $\alpha: \tilde{\Sigma}
\to \delta(\Sigma^{\sharp})$ which eliminates the indeterminacy of the generically finite rational map $\chi^{\sharp}$ such that $
\chi^{\sharp} \circ \alpha$ defines a generically finite morphism $g: \tilde{\Sigma} \to X'$. Denote by $\tau$ the natural projection $\delta(\Sigma^{\sharp}) \to X$. Then $\tau \circ
\alpha = \mu \circ g$. Since $\delta(\Sigma^{\sharp}) \subset
T(X)$, there exists a tautological section $\kappa \in
H^0(\delta(\Sigma^{\sharp}), \tau^* T(X))$ defined by $\kappa(a) =
a \in T_{\tau(a)}(X)$ for each $a \in \delta(\Sigma^{\sharp})$. Let $$\theta \in H^0( \tilde{\Sigma}, (\mu \circ g)^* T(X)) = H^0(
\tilde{\Sigma}, (\tau \circ \alpha)^* T(X))$$ be the pull-back of $\kappa$ by $\alpha$. Then $\theta$ satisfies property (1), because $\alpha$ is birational and the tautological section $\kappa$ satisfies an analog of (1). It satisfies property (2) from $\delta (\Sigma^{\sharp}) \subset
\hat{\C}.$ $\Box$
Univalence of $\K$ on the branch divisor of $\mu$
=================================================
In the setting of Proposition 3.2, we say that $\K$ is [*univalent on an irreducible hypersurface*]{} $B \subset X$ if (i) there exists only one irreducible component $E$ of $\mu^{-1}(B)$ that is dominant over both $Z$ and $B$, and (ii) the morphism $\mu|_E: E \to B$ is birational. This is equivalent to saying that at a general point $z \in B$, there exists exactly one member $C$ of $\K$ passing through $z$ with $C \not\subset B$ and $C$ is non-singular at $z$.
The following is essentially the same as \[H1, Proposition 3.2\].
. By assumption, there exist a union $D$ of components of $\mu^{-1}(B)$ each of which is dominant over $Z$ and $B$, and the morphism $\mu|_D: D \to B$ has degree $>1$. Let $y_1, y_2$ be two distinct points of $\mu^{-1}(x) \cap D$. Since $x$ is general, both $\rho(y_1)$ and $\rho(y_2)$ lie in $Z_o$. There exist open neighborhoods $W_1 \subset \rho^{-1}(Z_o)$ of $y_1$, $W_2 \subset \rho^{-1}(Z_o)$ of $y_2$ and $W_0 \subset W$ of $x$ with the following properties
\(1) $\mu(W_1) = \mu(W_2) = W_0$,
\(2) $\mu|_{W_1}$ and $\mu|_{W_2}$ are biholomorphic,
\(3) $W_1 \cap D$ and $W_2 \cap D$ are non-singular and transversal to the fibers of $\rho$.
There exist open neighborhood $W_1' \subset W_1$ of $y_1$ and $W'_2
\subset W_2$ of $y_2$ such that for any $y \in W_1'$ (resp. $y \in
W_2'$) $\rho^{-1}(\rho(y)) \cap W_1'$ (resp. $\rho^{-1}(\rho(y))
\cap W_2'$) is connected. Let $y$ be a general point in $ \mu(W_1')
\cap \mu(W_2')$. Let $y_1' = W_1' \cap \mu^{-1}(y)$ and $y_2' = W_2'
\cap \mu^{-1}(y).$ Then $\zeta_1 := \rho(y_1')$ and $\zeta_2 :=
\rho(y_2')$ give the desired two distinct points. $\Box$
The idea of the proof of the following proposition is the same as that of \[H1, Proposition 3.3\].
. Suppose that $\K$ is not univalent on some component $B$. Set $Y:= \tilde{\Sigma}$ and $f= \mu \circ g$. Then $B$ is a component of the branch divisor of $f: Y \to X$. Let $R \subset Y$ be an irreducible component of the ramification divisor of $f$ such that $B =f(R)$. Let $z \in R$ be a general point and let $r$ be the local sheeting number of $f$ at $z$. We can choose a holomorphic coordinate neighborhood $V$ of $z$ with coordinates $(w_1, \ldots, w_n)$ at $z$ and a holomorphic coordinate neighborhood $W$ of $f(z)$ with coordinates $(z_1, \ldots, z_n)$ such that $B \cap W$ is defined by $z_n=0$ and $f$ is given by $$z_1 = w_1, \ldots, z_{n-1} = w_{n-1}, z_n = w_n^r.$$ Let $x \in W\setminus B$ and $\zeta_1, \zeta_2 \in Z_o$ be as in Proposition 5.1. Setting $C_1=P_{\zeta_1}$ (resp. $C_2
=P_{\zeta_2}$), an easy coordinate computation in the above coordinate systems (see e.g. \[HM3, p.636, Lemma1\]) shows that there exists a unique irreducible component $C'_1$ (resp. $C'_2$) of $f^{-1}(C_1)$ (resp. $f^{-1}(C_2)$) intersecting $V$ such that an irreducible component of $C'_1 \cap V$ (resp. $C'_2 \cap V$) contains $f^{-1}(x) \cap V$. In particular, $C'_1 \cap C'_2$ contains the $r$ distinct points $f^{-1}(x) \cap V$. Let $y_1 \neq
y_2$ be two distinct points in $f^{-1}(x) \cap V \cap C'_1 \cap
C'_2$. Applying Proposition 3.5 to $C'_1$ and $C_1$, $$\theta_{y_1} - \theta_{y_2} \in T_{x_o}(C_1).$$ Applying Proposition 3.5 to $C'_2$ and $C_2$, $$\theta_{y_1} - \theta_{y_2} \in
T_{x_o}(C_2).$$ Since $T_{x_o}(C_1) \cap T_{x_o}(C_2) =0$, we get $\theta_{y_1} = \theta_{y_2}$, a contradiction to Proposition 4.3 (1). $\Box$
. Suppose not. Since $C'$ is a free curve with trivial normal bundle by Proposition 3.4, we may assume that $C'$ intersects $D$ at a general point $x'$ of $D$. Then through a general point $x'$ of $D$, we have two distinct curves, $C'$ and a fiber of $\rho$, neither of which are contained in $D$. Since $\mu$ is unramified at $x'$ by Proposition 3.2 (b), the images of these curves under $\mu$ are of the form $P_{\zeta_1}, P_{\zeta_2}$ with $\zeta_1 \neq \zeta_2$. Since these two curves pass through $x =
\mu(x')$, which is a general point of $B$, ${\cal K}$ is not univalent on $B$, a contradiction to Proposition 5.2. $\Box$
Completion of the proof of Theorem 1.3
======================================
In this section, we will complete the proof of Theorem 1.3. The strategy is to establish some analogs of \[H2, Section 5\].
. By Proposition 3.5, $C^{\flat}$ determines a unique element in $H^0(\hat{C}, N_C)$. By the isomorphism in Proposition 3.4, this determines an element $\vartheta_{C'} \in H^0(\hat{C}',
N_{C'}).$ It cannot be zero because of Proposition 4.3 (2) and the generality of $C$. $\Box$
The next proposition is an analog of \[H2, Lemma 5.5\], although their proofs are of different nature.
. For a deformation $[C'_s] \in {\cal M}_{C'}$ of $C'$, we get a deformation $[C^{\flat}_s] \in {\cal M}_{C^{\flat}}$ with $C^{\flat}_s \subset g^{-1}(C'_s).$ By applying Proposition 6.1 to $C^{\flat}_s$, we get an element $\vartheta_{C'_s} \in H^0(C'_s,
N_{C'_s}).$ Thus the choice of $C^{\flat}$ determines a germ of holomorphic vector fields $\vartheta$ on ${\cal M}_{C'}$. By Proposition 4.3 (2), this is a germ of non-vanishing vector fields. Let $$\{ C'_t, t \in \Delta, C'=C'_0\} = \{ [C'_t] \in {\cal M}_{C'}, t \in \Delta \}$$ be a local analytic arc integrating the vector field $\vartheta$. (i) and (ii) are obvious from the local equivalence ${\cal M}_C \cong {\cal M}_{C'}$ in Proposition 3.4. It suffices to check (iii). Let $x \in C'$ be a general point and $y \in C^{\flat}$ be the point over $x$. Then the germ of holomorphic vector fields defined by $\theta$ at $y$ induces a germ $\theta'$ of holomorphic vector fields in a neighborhood of $x$. Applying 3.1, we see that the integral curve of $\theta'$ through $x$, $\{x_t \in X', t \in \Delta\}$ with $x =
x_0$, satisfies $x_t \in C'_t$ up to reparametrization. Since $\theta'$ is a section of $T^{\rho}$ by Proposition 4.3 (2), $x_t
\in \rho^{-1}(\rho(x))$. It follows that $$\;\; \rho^{-1}(z) \cap C'_t \neq \emptyset \mbox{ for general
} z \in \rho(C') \mbox{ and each } t \in \Delta.$$ This implies (iii). $\Box$
The proof of the next proposition is, modulo Proposition 5.3 and Proposition 6.2, almost verbatim that of \[H2, Proposition 5.3\]. Since the terms and the notation are slightly different, we reproduce the proof for the reader’s convenience.
. Suppose not. Then $h$ is unramified over $\rho(C') \cap Z_o$. Let us use the deformation $C_t$ constructed in Proposition 6.2. By the generality of $C$, we may assume that for each $t \in
\Delta$ the holomorphic map $$h_t: \hat{C}'_t \rightarrow
\widehat{\rho(C'_t)} = \widehat{\rho(C')}, \;\; h_0 =h,$$ which is the lift of $\rho|_{C'_t}$ to the normalization, is unramified over $\rho(C') \cap Z_o$. Since $h_t$ is a continuous family of coverings of the Riemann surface $\widehat{\rho(C')}$ with fixed branch locus, we can find a biholomorphic map $$(\clubsuit) \;\;\; \psi_t: \hat{C}' \rightarrow \hat{C}'_t, \;\; \psi_0 = {\rm Id}_{\hat{C}'} \mbox{ with } h = h_t \circ
\psi_t,$$ which depends holomorphically on $t$ (e.g. \[S, p. 32, Corollary 1\].).
By Proposition 6.2 (ii), there are at least two distinct points in $\hat{C}_t$, say $a_t \neq b_t \in \hat{C}_t$, such that the corresponding points in $C_t$ lie in the branch divisor of $\mu$ in $X$. Let $\{ 0, \infty\} \subset \P_1$ be two distinct points on the projective line. We can choose a family of biholomorphic maps $\{ \sigma_t: \hat{C}_t \rightarrow \P_1, \; t \in \Delta \}$ such that $\sigma_t(a_t) = 0$ and $ \sigma_t(b_t) = \infty$ for each $t \in \Delta$. Denote by $\mu_t: \hat{C}'_t \rightarrow \hat{C}_t$ the lift of $\mu|_{C'_t}$ to the normalization. Then $$\{ \varphi_t: \hat{C}' \longrightarrow
\P_1, \;\;\; \varphi_t := \sigma_t \circ \mu_t \circ \psi_t, \; t
\in \Delta \}$$ is a family of meromorphic functions on the compact Riemann surface $\hat{C}'$.
By Proposition 5.3, for each component $E$ of the branch divisor of $\mu$, the intersection of $C'_t$ with $\mu^{-1}(E)$ has a fixed image in $\rho(C')=\rho(C'_t)$, independent of $t \in \Delta$. This implies that there is a finite subset $Q \subset
\widehat{\rho(C')}$, independent of $t$, such that $$\mu_t^{-1}(a_t) \cup \mu_t^{-1}(b_t) \subset h_t^{-1}(Q)$$ for any $t
\in \Delta$. Then $$\varphi_t^{-1}(0) = \psi^{-1}_t \circ
\mu_t^{-1} \circ \sigma_t^{-1}(0) = \psi^{-1}_t(\mu_t^{-1}(a_t))
\subset \psi_t^{-1}(h_t^{-1}(Q))$$ for all $t \in \Delta$. Since $\psi_t^{-1}(h_t^{-1}(Q))= h^{-1}(Q)$ by the choice of $\psi_t$ in $(\clubsuit)$, $ \varphi_t^{-1}(0) \subset h^{-1}(Q)$ for any $t
\in \Delta$. Consequently, $\varphi_t^{-1}(0) =
\varphi_0^{-1}(0)$ for all $t \in \Delta$. By the same argument we get $\varphi_t^{-1}(\infty) = \varphi_0^{-1}(\infty)$ for all $t
\in \Delta$. In other words, the family of meromorphic functions $\varphi_t$ have the same zeroes and the same poles on the Riemann surface $\hat{C}'$. It follows that for any $w_1, w_2 \in
\hat{C}'$ and any $t \in \Delta$, $$(\diamondsuit) \;\;\;\;
\varphi_t(w_1) = \varphi_t(w_2) \mbox{ if and only if }
\varphi_0(w_1) = \varphi_0(w_2).$$
Since $\mu|_{C'}$ is finite of degree $>1$ by our assumption, we can choose two points $\alpha \neq \beta \in \hat{C}'$ such that $\varphi_0(\alpha) = \varphi_0(\beta).$ Furthermore, denoting by $\bar{\alpha}\in \rho(C')$ (resp. $\bar{\beta} \in \rho(C')$) the point corresponding to $h_0(\alpha) \in \widehat{\rho(C')}$ (resp. $h_0(\beta)\in \widehat{\rho(C')}$) under the normalization, we may assume that $$(\heartsuit) \;\;\; \bar{\alpha} \mbox{ and }
\bar{\beta} \mbox{ are two distinct points in } Z_o.$$ From $(\diamondsuit)$, we have $\varphi_t(\alpha) = \varphi_t(\beta)$ for all $t \in \Delta$. Since $\varphi_t = \sigma_t \circ \mu_t
\circ \psi_t$ and $\sigma_t$ is biholomorphic, we see that $$(\spadesuit) \;\;\;\; \mu_t \circ \psi_t(\alpha) = \mu_t \circ
\psi_t(\beta) \mbox{ for all } t \in \Delta.$$
Denote by $$\alpha_t \in C'_t \subset X' \;\;\; \mbox{ (resp. }
\beta_t \in C'_t \subset X')$$ the point corresponding to $\psi_t(\alpha) \in \hat{C}'_t$ (resp. $\psi_t(\beta) \in \hat{C}'_t$) under the normalization. Then the locus $$A := \{ \alpha_t \in X', t \in \Delta \} \;\;\; \mbox{ (resp.
} B := \{ \beta_t \in X', t \in \Delta \})$$ covers a non-empty open subset in the fibre $\rho^{-1}(\bar{\alpha})$ (resp. $\rho^{-1}(\bar{\beta})$). Thus $\mu(A)$ (resp. $\mu(B)$) covers a non-empty open subset in $$P_{\bar{\alpha}} := \mu(\rho^{-1}(\bar{\alpha})) \;\;\; \mbox{
(resp. } P_{\bar{\beta}} :=\mu( \rho^{-1}(\bar{\beta} ))).$$ Since $\mu(A)$ (resp. $\mu(B)$) is the locus of points corresponding to $\mu_t \circ \psi_t (\alpha)$ (resp. $\mu_t \circ \psi_t (\beta)$) by the normalization $\hat{C}_t \rightarrow C_t$, the equality $(\spadesuit)$ above implies that $\mu(A) = \mu(B)$. Consequently, $$P_{\bar{\alpha}}= P_{\bar{\beta}},$$ a contradiction to Proposition 3.2 via $(\heartsuit)$. $\Box$
Now we are ready to finish the proof of Theorem 1.3 as follows.
. As explained at the beginning of Section 4, we may assume the situation of Proposition 4.3 and get a contradiction. From Proposition 6.3, let $z \in C'$ be the image of a ramification point of $h$ such that $\rho(z) \in Z_o$. Then a component of the germ of $C'$ at $z$ must be tangent to $T^{\rho}_z$. Choose $C^{\flat}$ as in Proposition 6.1. The value of $\theta$ at a point of $C^{\flat}$ over $z$ determines $\theta_z
\in T_z(X')$ which is in $T^{\rho}_z$ by Proposition 4.3 (2). Thus $\theta_z$ is tangent to a component of the germ of $C'$ at $z$. This means that the non-zero element $\vartheta_{C'} \in
H^0(\hat{C}', N_{C'})$ in Proposition 6.1 vanishes at $z$, a contradiction to the triviality of $N_{C'}$. $\Box$
An essential idea for this work was obtained during my visit to Fudan University in December, 2007. I would like to thank Ngaiming Mok and Yuxin Dong for the invitation and the hospitality.
\[H1\] Hwang, J.-M.: Deformation of holomorphic maps onto Fano manifolds of second and fourth Betti numbers 1. Ann. Inst. Fourier [**57**]{} (2007) 815-823
\[H2\] Hwang, J.-M.: Base manifolds for fibrations of projective irreducible symplectic manifolds. Invent. math. [**174**]{} (2008) 625-644
\[HM1\] Hwang, J.-M. and Mok, N.: Varieties of minimal rational tangents on uniruled manifolds. in [*Several Complex Variables*]{}, ed. by M. Schneider and Y.-T. Siu, MSRI Publications 37, Cambridge University Press (2000) 351-389
\[HM2\] Hwang, J.-M. and Mok, N.: Cartan-Fubini type extension of holomorphic maps for Fano manifolds of Picard number 1. Journal Math. Pures Appl. [**80**]{} (2001) 563-575
\[HM3\] Hwang, J.-M. and Mok, N.: Finite morphisms onto Fano manifolds of Picard number 1 which have rational curves with trivial normal bundles. J. Alg. Geom. [**12**]{} (2003) 627-651
\[HM4\] Hwang, J.-M. and Mok, N.: Birationality of the tangent map for minimal rational curves. Asian J. Math. [**8**]{}, [*Special issue dedicated to Yum-Tong Siu*]{}, (2004) 51-64
\[HR\] Hwang, J.-M. and Ramanan, S.: Hecke curves and Hitchin discriminant. Ann. scient. Ec. Norm. Sup. [**37**]{} (2004) 801-817
\[K\] Kollár, J.: Rational curves on algebraic varieties. Erg. d. Math. 3 Folge, Band 32. Springer Verlag (1996)
\[S\] Shokurov, V. V.: Riemann surfaces and algebraic curves. In [*Algebraic curves, algebraic manifolds and schemes*]{}, Springer Verlag (1998)
Jun-Muk Hwang
Korea Institute for Advanced Study
207-43 Cheongnyangni-dong
Seoul 130-722, Korea
[email protected]
[^1]: This work was supported by the SRC Program ASARC funded by the Korea government.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'It is known that the vacuum polarization of zero-point field arises around a conical singularity generated by an infinite, straight cosmic string. In this paper we study quantum electromagnetic corrections to the gravitational Aharonov-Bohm effect around a cosmic string. We find the scattering amplitude from a conical defect for charged Klein-Gordon field.'
author:
- |
Kiyoshi Shiraishi\
Akita Junior College, Shimokitade-Sakura, Akita-shi,\
Akita 010, Japan
date: 'Journal of the Korean Physical Society. [**25**]{} (1992) pp. 192–195 '
title: 'Quantum Corrections to Scattering Amplitude in Conical Space-time '
---
Introduction
============
In the past few years, quantum theory on a cone has attracted much attention. One of the motivations, besides theoretical interest in $2+1$ dimensional field theory, is the application to quantum field theory around cosmic strings. An idealized cosmic string, which is infinitely thin, can be represented by a conical singularity of space-time. Quantum effect around cosmic strings has been studied by many authors including Dowker [@1], Frolov [@2], Linet [@3] and Smith.[@4]
The scattering by a cosmic string or a conical singularity has also been investigated recently.[@5] This non-trivial scattering is often dubbed as the gravitational Aharonov-Bohm (AB) effect.
It is interesting to consider “radiative” corrections to the gravitational AB scattering, which arise due to quantum fluctuations of all interacting field around cosmic strings.
In the present paper, we consider vacuum fluctuations of the electromagnetic field. Then, charged particles scattered by a cosmic string undergo quatum corrections even if the cosmic string has no charge and no classical field strength.
The organization of this paper is as follows: In section 2, the vacuum polarization of gauge field is obtained. In section 3, we derive the quantum-mechanically corrected Klein-Gordon equation for a charged scalar field. The scattering amplitude is calculated in section 4. Section 5 is devoted to conclusion.
The vacuum polarization of electromagnetic field
================================================
In this section, we compute the vacuum value of $\langle A_\mu A^\mu\rangle$ around a infinite straight cosmic string, where $A_\mu$ is the electromagnetic field. Connection to an effective wave equation for a charged scalar field is presented in the succeeding section. We first anticipate that $\langle A_\mu
A^\mu\rangle\propto r^{-2}$, since there is no dimensional quantity other than the distance from the string.
A comment on gauge invariance is in order. Classically the presence of the cosmic string does not break the gauge invariance. At the one-loop level, however, one can see the breakdown of gauge Symmetry near the cosmic string, due to the mass-like term $e^2\langle\phi^2\rangle A_\mu A^\mu$ in the scalar QED; here the vacuum expectation value $\langle\phi^2\rangle$ depends on the distance from the cosmic string. Thus the non-zero expectation value of $A_\mu A^\mu$ does not give rise to a real conflict. Note also that we often encounter the expression $\langle
A_\mu A^\mu\rangle$ in the discussion on the Coleman-Weinberg potential (or radiative gauge-symmetry breaking) in scalar QED at zero and finite temperature.[@12]
We assume the following line element around the idealized cosmic string which lies along the $z$ axis: $$ds^2=-dt^2+dz^2+dr^2+(r^2/\nu^2)d\theta^2\,,
\label{2.1}$$ where $\nu$ is related to the mass density of the string $\mu$ by $\nu^{-1}=1-4\pi G\mu$. Working in the Coulomb gauge, we can denote the gauge field in the form of normal-mode expansion, such as $$A_i=\sum_{s=1,2} a^s A^s_i(r,\theta) e^{ik_s z-i\omega_s t}+h.c.
\label{2.2}$$ Mode functions $A^s_i$ are the solutions for the Maxwell equations. In the coordinate system represented by eq. (\[2.1\]), we find the following set of mode functions: $$\begin{aligned}
A^1_z&=&\{i(l/\omega) J_{\nu n}(lr) \sin n\theta\,,
i(l/\omega) J_{\nu n}(lr) \cos n\theta\}\,,
\label{2.3-1} \\
A^1_r&=&\{i(k/2\omega) (J_{\nu n+1}(lr)-J_{\nu n-1}(lr)) \sin
n\theta\,, \nonumber \\
& &i(k/2\omega) (J_{\nu n+1}(lr)-J_{\nu n-1}(lr)) \cos
n\theta\}\,,
\label{2.3-2}\\
A^1_\theta&=&\{-(k/2\omega)(r/\nu)(J_{\nu n+1}(lr)+J_{\nu n-1}(lr)) \sin
n\theta\,,\nonumber \\
& &-(k/2\omega)(r/\nu)(J_{\nu n+1}(lr)+J_{\nu n-1}(lr)) \cos
n\theta\}\,,
\label{2.3-3}\\
A^2_z&=&0\,,
\label{2.3-4} \\
A^2_r&=&\{(k/2)(J_{\nu n+1}(lr)+J_{\nu n-1}(lr)) \sin
n\theta\,,\nonumber \\
& &(k/2)(J_{\nu n+1}(lr)+J_{\nu n-1}(lr)) \cos n\theta\}\,,
\label{2.3-5}\\
A^2_\theta&=&\{-(k/2)(r/\nu)J_{\nu n+1}(lr)-J_{\nu n-1}(lr)) \sin
n\theta\,,\nonumber \\
& & -(k/2)(r/\nu)(J_{\nu n+1}(lr)-J_{\nu n-1}(lr)) \cos n\theta\}\,,
\label{2.3-6}\end{aligned}$$ where $J_m$ is the Bessel function, $n$ is positive integer and $\omega^2=k^2+l^2$.
Using the mode functions, we can perform the calculation of $\langle A_i A^i\rangle$ by mode-summation. After regularization, we find that a finite portion of $\langle A_i A^i\rangle$ is given by $$\langle A_i A^i\rangle=\frac{\nu^2-1}{24\pi^2r^2}\,.
\label{2.4}$$
The quantity has been regularized so as to become zero if $\nu=1$. (Note that this is twice of the real-scalar contribution.[@4]) The mode expansion is also useful to investigale properties of charged fields around a cosmic string.[@6]
In the next section, we consider the modification of field equation due to the vacuum field $\langle A_i A^i\rangle$.
Modified Klein-Gordon equation for a charged scalar
===================================================
Suppose that a minimally-coupled scalar field $\phi$ is governed by Klein-Gordon equation, where the space-time derivative is replaced by the covariant derivative including the gauge field.
In this paper, we assume that the cosmic string has no classical electromagnetic field. Although general cosmic strings in GUTs have fluxes in their core, their fluxes are not magnetic fluxes we know, but are associated with gauge fields of broken symmetry. For a superconducting cosmic string [@7] the analysis ought to be done by some other ways, since the metric is more or less deformed by electric currents. Therefore, we treat the case with normal cosmic strings throughout this paper.
However, we take into account the vacuum expcctation value of the electromagnetic field. Only the second moment $\langle A_i
A^i\rangle$ appears and the field equation is expressed as $$\Box\phi-e^2\langle A_i A^i\rangle\phi-m^2\phi=0\,,
\label{3.1}$$ where $e$ is the coupling constant and $m$ is the mass. The term $e^2\langle A_i A^i\rangle$ behaves as a potential term.
If we use the metric (\[2.1\]) and substitute (\[2.2\]) into eq. (\[3.1\]), we obtain $$\frac{1}{r}\frac{\partial}{\partial
r}r\frac{\partial\phi}{\partial
r}+\frac{\nu^2}{r^2}\frac{\partial^2\phi}{\partial\theta^2}-
\frac{\gamma^2}{r^2}\phi-m^2\phi=0\,,
\label{3.2}$$ where $\gamma^2=e^2(\nu^2-1)/24\pi^2$.
A solution for the modified equation (\[3.2\]) is written in the form $$\phi(t,z,r,\theta)=J_{\alpha_n}(lr) e^{ikz-i\omega t+in\theta}\,,
\label{3.3}$$ with $\alpha_n=(\nu^2n^2+\gamma^2)^{1/2}$ and $\omega^2=l^2+k^2+m^2$.
In the following section, we study the scattering problem by use of this solution.
We set $m=0$ in the following analysis in this paper.
Scattering amplitude in the presence of the quantum potential
=============================================================
To treat the scattering by an infinite string, we consider two-dimensional scattering, taking the $z$-component of the momentum ($k$ in (\[3.3\])) to be zero. We have set $m=0$.
We first review the scattering in two dimensions.[@5; @8] The wave function behaves asymptotically, $$\phi\stackrel{r\rightarrow \infty}{\longrightarrow}e^{ilr\cos\theta}
+e^{i\pi/4}\frac{f(\theta)}{r^{1/2}}e^{ilr}\,,
\label{4.1}$$ when a target sits at the origin. Here $f(\theta)$ is a scattering amplitude, and a differential cross section (for a cosmic string, per unit length) is given by $|f(\theta)|^2$. The phase in front of $f(\theta)$ is chosen to simplify the expression of the optical theorem in two dimensions [@5; @8] (see later). In order to find the scattering amplitude, we compare the asymptotic form of the solution of the wave equation with (\[4.1\]). We determine $f$ by matching the coefficient of the in-going “spherical” wave ($\propto e^{-ilr}$).
The asymptotic form of the plane wave going in the $x$-direction is given by $$e^{ilr\cos\theta}\stackrel{r\rightarrow\infty}{\longrightarrow}
(2\pi lr)^{-1/2}e^{i\pi/4}\sum_{n=-\infty}^\infty
\{-i e^{ilr}+(-1)^n e^{-ilr}\} e^{in\theta}\,.
\label{4.2}$$ Further if we assume that we can take the following asymptotic behavior of the solution for the wave equation: $$\phi\stackrel{r\rightarrow\infty}{\longrightarrow}
(2\pi lr)^{-1/2}e^{i\pi/4}\sum_{n=-\infty}^\infty
\{-i e^{ilr+2i\delta_n}+(-1)^n e^{-ilr}\} e^{in\theta}\,,
\label{4.3}$$ then we find $$f(\theta)=-i(2\pi l)^{-1/2}\sum_{n=-\infty}^\infty
(e^{2i\delta_n}-1) e^{in\theta}\,,
\label{4.4}$$ where $\delta_n$ is called as a phase shift.
Now, we will turn back to the present problem. The wave equation to be considered is eq. (\[3.2\]) with $m=0$. We regard $\gamma$ as an independent parameter of $\nu$ for a later use. The mode solution (\[3.3\]) behaves asymptotically as $$\phi_n\stackrel{r\rightarrow\infty}{\longrightarrow}
(2\pi lr)^{-1/2}e^{i\pi/4}A_n
\{-i e^{ilr-i\pi(\alpha_n-n)}+(-1)^n e^{-ilr}\} e^{in\theta}\,,
\label{4.5}$$ where $A_n=(-1)^n e^{-in\alpha_n/2}$. Thus one can find $$\delta_n=-(\pi/2)(\alpha_n-n)=-(\pi/2)\{(\nu^2n^2+\gamma^2)^{1/2}-n\}\,.
\label{4.6}$$
If $\gamma^2$ is independent of $\nu$, $\delta_n$ is negative when $\nu=1$ and $\gamma^2>0$. Therefore we can say that the “potential” $\gamma^2/r^2$ gives rise to a repulsive force. As studied and stated in ref. [@5], the scattering amplitude for $\nu\ne 1$ contains delta functions. Namely, for $\gamma^2=0$, one can find [@5] $$\begin{aligned}
& &(2\pi
l)^{1/2}f(\theta)=
\frac{\sin\{(\nu-1)\pi\}}{\cos\{(\nu-1)\pi\}-\cos\theta}
-i\pi
\sum_n\{\delta(\theta+(\nu-1)\pi-2\pi n)\nonumber \\
& &+\delta(\theta-(\nu-1)\pi-2\pi
n)-2\delta(\theta-2\pi n)\}\,.
\label{4.7}\end{aligned}$$ This peculiarity originates from the conical space generates “long-range force” in some naive sense. We discuss the amplitude in which such divergences are removed.[@5]
We will calculate $$\begin{aligned}
(2\pi l)^{1/2}{\rm
Re}f(\theta)&=&\sum_n\sin(n\theta-\pi(\alpha_n-n))\,,
\label{4.8-1}\\
(2\pi l)^{1/2}{\rm
Im}f(\theta)&=&-\sum_n\cos(n\theta-\pi(\alpha_n-n))\,,
\label{4.8-2}\end{aligned}$$ where $\theta+(\nu-1)\pi\ne 2\pi n$ and $\theta-(\nu-1)\pi\ne 2\pi n$ ($n$: integer). We first compute the scattering amplitude for $\nu=1$ and $\gamma\ne 0$ (we take $\gamma$ as an independent parameter.). This is used for a check of general results. The result obtained by numerical calculation is shown in Fig. 1. We examine some approximation schemes.
![The scattering amplitude for $\nu=1$ and $\gamma\ne 0$. (a) $(2\pi l)^{1/2}{\rm Re} f(\theta)$, (b) $(2\pi l)^{1/2}{\rm Im}
f(\theta)$. The solid line stands for $\gamma^2=0.2$ and the broken line for $\gamma^2=0.5$.[]{data-label="f1"}](jkps1.eps "fig:"){width="6cm"} ![The scattering amplitude for $\nu=1$ and $\gamma\ne 0$. (a) $(2\pi l)^{1/2}{\rm Re} f(\theta)$, (b) $(2\pi l)^{1/2}{\rm Im}
f(\theta)$. The solid line stands for $\gamma^2=0.2$ and the broken line for $\gamma^2=0.5$.[]{data-label="f1"}](jkps2.eps "fig:"){width="6cm"}
The $S$-wave ($n=0$) contribution in the formula (\[4.4\]) gives $$\begin{aligned}
(2\pi l)^{1/2}{\rm
Re}f_0&=&-2\sin(\pi\gamma/2)\cos(\pi\gamma/2)\,,
\label{4.9-1}\\
(2\pi l)^{1/2}{\rm
Im}f_0&=&2\sin^2(\pi\gamma/2)\,,
\label{4.9-2}\end{aligned}$$ and these seem to be a good estimation as the lowest order.
The optical theorem $${\rm Im} f(0)=(l/8\pi)^{1/2}\int_0^{2\pi}|f(\theta)|^2d\theta\,,
\label{4.10}$$ is applicable for $\nu=1$ and one can confirm this at the order in using the $S$-wave result (\[4.9-1\], \[4.9-2\]). Note that the amplitude does not contain divergences for $\nu=1$ (for instance, delta functions cancel one another in (\[4.7\]) when $\nu=1$.).
Although the potential has a short-range nature, the Born approximation is not applicable because the singularity near the origin is too strong.
Next we consider the case for a cosmic string. Here we recover the relation $\gamma^2=e^2(\nu2-1)/24\pi^2$. We define $\Delta f$ by the difference in the scattering amplitudes for $\gamma^2=e^2(\nu2-1)/24\pi^2$ and $\gamma^2=0$; i.e., the contribution of (\[4.7\]) is subtracted. The numerical results are given in Fig. 2. The value of ${\rm Im} \Delta f$ in the vicinity of $(\nu-1)\pi$ is not indicated because of the poor convergence in the sum of the oscillating series. The value seems almost constant.
![The difference in the scattering amplitudes for $\gamma^2=e^2
(\nu^2-I)/24\pi^2$ and $\gamma^2=0$ in the case of $\nu-1=0.1$. We set $e^2/4\pi=1/137$. The solid line stands for $(2\pi l)^{1/2}{\rm Re}
\Delta f$ and the broken line for $(2\pi l)^{1/2}{\rm Im}
\Delta f$.[]{data-label="f2"}](jkps3.eps){width="6cm"}
The real part of $\Delta f$ exhibits a logarithmical divergence at $\theta=(\nu-1)\pi$. The behavior of the amplitude is approximately given by: $$\begin{aligned}
& &(2\pi l)^{1/2}{\rm Re} \Delta
f=-2\sin(\pi\gamma/2)\cos(\pi\gamma/2)
\nonumber \\
& &\qquad+\frac{\pi\gamma^2}{2\nu}\ln
\left|4 \sin\frac{\theta+(\nu-1)\pi}{2}
\sin\frac{\theta-(\nu-1)\pi}{2}\right| +O(\gamma^4)\,,
\label{4.11-1}\end{aligned}$$ $$\begin{aligned}
& &(2\pi l)^{1/2}{\rm Im} \Delta
f=2\sin^2(\pi\gamma/2)
\nonumber \\
& &\qquad+\frac{\pi\gamma^2}{2\nu}
\left\{\nu-1-
\left\{
\begin{array}{cc}
1 & (\theta<(\nu-1)\pi)\\
0 & ((\nu-1)\pi<\theta)
\end{array}
\right\}
\right\}
+O(\gamma^4)\,.
\label{4.11-2}\end{aligned}$$ For a realistic cosmic string, $\nu-1$ may be too small for us to detect the deviation in the scattering amplitude unless the strength of the coupling become sufficiently large.
Conclusion
==========
In this paper we have derived the lowest-order vacuum quantum space electromagnetic corrections to the scattering amplitude in a conical background space-time. Inclusion of the quantum effect of non-Abelian gauge fields around a cosmic string is an important extension of the present work: it may be connected with non-Abelian AB effect [@9] and the baryon decay/genesis mediated by cosmic strings.[@10] It is also important to treat the effect of self-interaction in a conical space for the Yang-Mills case. (Self-interacting scalar fields around a cosmic string have been studied in ref. [@11].).
[99]{} J. S. Dowker, Phys. Rev. [**D36**]{} (1987) 3742. V. P. Frolov and E. M. Serebriany, Phys. Rev. [**D35**]{} (1987) 3779. B. Linet, Phys. Rev. [**D35**]{} (1987) 536. A. G. Smith, in [*The formation and evolution of cosmic strings*]{}, eds. by G. Gibbons, S. Hawking and T. Vachaspati (Cambridge Univ., Cambridge, 1990).
G. ’t Hooft, Commun. Math. Phys. [**117**]{} (1988) 685; S. Deser and R. Jackiw, Commun. Math. Phys. [**118**]{} (1988) 495; P. de Sousa Gerbert and R Jackiw, Commun. Math. Phys. [**124**]{} (1989) 229; G. W. Gibbons, F. Ruiz and T. Vachaspati, Commun. Math. Phys. [**127**]{} (1990) 295. K. Shiraishi, in preparation. E. Witten, Nucl. Phys. [**B249**]{} (1985) 557. I. R. Lapidus, Am. J. Phys. [**50**]{} (1982) 45; P. A, Maurone and T. K. Lim. Am. J. Phys. [**51**]{} (1983) 856; J. F. Perez and F. A. B. Coutinho, Am. J. Phys. [**59**]{} (1991) 52; G. T. Barton, Am. J. Phys. [**51**]{} (1983) 420. P. A. Horvathy, Phys. Rev. [**D33**]{} (1986) 407; M. G. Alford, J. March-Russell and F. Wilczek, Nucl. Phys. [**B337**]{} (1990) 695; M. G. Alford, A. Benson, S. Coleman, J. March-Russell and F. Wilczek, Phys. Rev. Lett. [**64**]{} (1990) 1632; Nucl. Phys. [**B349**]{} (1991) 414; F. Wilczek and Y.-S. Wu, Phys. Rev. Lett. [**65**]{} (1990) 13; M. Bucher, Nucl. Phys. [**B350**]{} (1991) 163. R. H. Brandenberger and L. Perivolaropoulos, Phys. Lett. [**B208**]{} (1988) 396; R H. Brandenberger, A.-C. Davis and A. M. Matheson, Phys. Lett. [**B218**]{} (1989) 304; A.-C. Davis and W. Perkins, Phys. Lett. [**B228**]{} (1989) 37; R. H. Brandenberger, A.-C. Davis and A. M. Matheson, Nucl. Phys. [**B307**]{} (1988) 909; R. Gregory, A.-C. Davis and R. H. Brandenberger, Nucl. Phys. [**B323**]{} (1989) 187; L. Perivolaropoulos. A. Matheson, A.-C. Davis and R. H. Brandenberger, Phys. Lett. [**B245**]{} (1990) 556; A. Matheson, L. Perivolaropoulos, W. Perkins, A.-C. Davis and R. H. Brandenberger, Phys. Lett. [**B248**]{} (1990) 263; A. Matheson, Mod. Phys. Lett. [**A6**]{} (1991) 769; W. B. Perkins, L. Perivolaropoulos, A.-C. Davis, R. H. Brandenberger and A. Matheson, Nucl. Phys. [**B353**]{} (1991) 237; M. G. Alford, J. March-Russell and F. Wilczek, Nucl. Phys. [**328**]{} (1989) 140. I. H. Russell and D. J. Toms, Class. Q. Grav. [**6**]{} (1989) 1343; K. Shiraishi and S. Hirenzaki, Class. Q. Grav. [**9**]{} (1992) 2277. A. D. Linde, Rep. Prog. Phys. [**42**]{} (1979) 389, and references there in.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Haver’s near-selection theorem deals with approximate selections of Hausdorff continuous CE-valued mappings defined on $\sigma$-compact metrizable $C$-spaces. In the present paper, we extend this theorem to all paracompact $C$-spaces. The technique developed to achieve this generalisation is based on oriented simplicial complexes. This approach makes not only a considerable simplification in the proof but is also successful in generalising the special case of compact metric $C$-spaces to all paracompact finite $C$-spaces.'
address: 'Department of Mathematics, Faculty of Science, University of Malta, Msida MSD 2080, Malta'
author:
- Valentin Gutev
title: 'Near-selection theorems for $C$-spaces'
---
Introduction
============
A space $X$ has property $C$, or is a *$C$-space*, if for any sequence $\{\mathscr{U} _n:n<\omega\}$ of open covers of $X$ there exists a sequence $\{\mathscr{V} _n:n<\omega\}$ of pairwise disjoint families of open subsets of $X$ such that each $\mathscr{V} _n$ refines $\mathscr{U} _n$ and $\bigcup_{n<\omega}\mathscr{V} _n$ is a cover of $X$. The $C$-space property was originally defined by W.Haver [@haver:1974] for compact metric spaces, subsequently Addis and Gresham [@addis-gresham:78] reformulated Haver’s definition for arbitrary spaces. It should be remarked that a $C$-space is paracompact if and only if it is countably paracompact and normal, see e.g. [@MR2352366 Proposition 1.3]. Every finite-dimensional paracompact space, as well as every countable-dimensional metrizable space, is a $C$-space [@addis-gresham:78], but there exists a compact metric $C$-space which is not countable-dimensional [@pol:81].
For spaces (sets) $X$ and $Y$, we write $\varphi:X\operatorname{\leadsto}Y$ to designate that $\varphi$ is a *set-valued* (or *multi-valued*) *mapping* from $X$ to the nonempty subsets of $Y$. For a metric space $(Y,\rho)$ and $\varepsilon>0$, we will use $\mathbf{O}_\varepsilon(y)$ for the open $\varepsilon$-ball centred at a point $y\in Y$; and $\mathbf{O}_\varepsilon(S)=\bigcup_{q\in S}\mathbf{O}_\varepsilon(q)$, whenever $S\subset Y$. In these terms, a set-valued mapping $\varphi:X\operatorname{\leadsto}Y$ is *$\rho$-continuous* (or *Hausdorff continuous*) if for every $\varepsilon >0$, each point $x\in X$ has a neighbourhood $V$ such that $$\label{eq:Finite-C-v2:3}
\varphi(x)\subset \mathbf{O}_\varepsilon(\varphi(p))\ \text{and}\
\varphi(p) \subset \mathbf{O}_\varepsilon(\varphi(x)),\quad
\text{for every $p\in V$.}$$ A map $f:X\to Y$ is an *$\varepsilon$-selection* for a mapping $\varphi :X\operatorname{\leadsto}Y$ if $f(x)\in \mathbf{O}_\varepsilon\left(\varphi(x)\right)$, for every $x\in X$. More generally, for a function ${\varepsilon:X\to (0,+\infty)}$, we say that $f:X\to Y$ is an *$\varepsilon$-selection* for $\varphi$ if $f(x)\in \mathbf{O}_{\varepsilon(x)}\left(\varphi(x)\right)$, for every $x\in X$. Finally, following Haver [@MR503224], a subset $S\subset Y$ is called *CE* if for every $\varepsilon>0$ there exists $\delta>0$ such that $\mathbf{O}_\delta(S)$ is contractible in $\mathbf{O}_\varepsilon(S)$. The following theorem was proved by Haver in [@MR503224 Theorem 1].
\[theorem-Selections-IDS:1\] Let $(Y,\rho)$ be a metric space and $X=\bigcup_{n<\omega} X_n$ be a metrizable space with each $X_n$ being a compact $C$-space. Then each $\rho$-continuous CE-valued mapping $\varphi:X\operatorname{\leadsto}Y$ has a continuous $\varepsilon$-selection, for every continuous function $\varepsilon:X\to (0,1]$.
By a *simplicial complex* we mean a collection $\Sigma$ of nonempty finite subsets of a set $S$ such that $\tau\in \Sigma$, whenever ${\varnothing}\neq \tau\subset \sigma\in \Sigma$. The set $\bigcup\Sigma $ is the *vertex set* of $\Sigma$, while each element of $\Sigma$ is called a *simplex*. The *$k$-skeleton* $\Sigma^k$ of $\Sigma$ ($k\geq 0$) is the simplicial complex $\Sigma^{k}=\{\sigma\in \Sigma:\operatorname{Card}(\sigma)\leq k+1\}$, where $\operatorname{Card}(\sigma)$ is the cardinality of $\sigma$. In the sequel, for simplicity, we will identify the vertex set of $\Sigma$ with its $0$-skeleton $\Sigma^0$.
The vertex set $\Sigma^0$ of a simplicial complex $\Sigma$ can be embedded as a linearly independent subset of some linear normed space. Then to any simplex $\sigma\in \Sigma$ we may associate the *geometric simplex* $|\sigma|$ which is the convex hull of $\sigma$. The *geometric realisation* of $\Sigma$ is the set $|\Sigma|=\bigcup\{|\sigma|:\sigma\in \Sigma\}$. As a topological space, we will always consider $|\Sigma|$ endowed with the *Whitehead topology* [@MR1576810; @MR0030759]. This is the topology in which a subset $U\subset |\Sigma|$ is open if and only if $U\cap |\sigma|$ is open in $|\sigma|$, for every $\sigma\in \Sigma$. The following special case of Theorem \[theorem-Selections-IDS:1\] was obtained by Haver in [@haver:1974 Proposition 3].
\[theorem-Selections-IDS-v4:1\] Let $X$ be a compact metric $C$-space, $(Y,\rho)$ be a metric space and $\varphi:X\operatorname{\leadsto}Y$ be a $\rho$-continuous mapping such that for every $x\in X$ and $\varepsilon>0$, there exists $\delta>0$ with the property that for every finite simplicial complex $\Sigma$, every continuous map $g:|\Sigma|\to \mathbf{O}_\delta(\varphi(x))$ is null-homotopic in $\mathbf{O}_\varepsilon(\varphi(x))$. Then $\varphi$ has a continuous $\varepsilon$-selection, for every $\varepsilon>0$.
In the present paper, we show that Theorem \[theorem-Selections-IDS:1\] is valid for all paracompact $C$-spaces, while Theorem \[theorem-Selections-IDS-v4:1\] — for all paracompact finite $C$-spaces. In fact, both these generalisations are obtained with an almost identical proof. Namely, in the next section, we consider an abstract embedding relation $S{\lhook\joinrel\xrightarrow{*}}T$ of subsets $S\subset T\subset Y$ of a space $Y$. The only requirement we place on this relation is to be *monotone* in the sense that $A\subset S{\lhook\joinrel\xrightarrow{*}} T\subset B$ implies $A{\lhook\joinrel\xrightarrow{*}} B$. In this general setting, for a metric space $(Y,\rho)$, we consider *uniformly $UV^{*}$* subsets defined by “$\mathbf{O}_\delta(S){\lhook\joinrel\xrightarrow{*}} \mathbf{O}_\varepsilon(S)$”, i.e.precisely as the properties of the values of $\varphi$ in Theorems \[theorem-Selections-IDS:1\] and \[theorem-Selections-IDS-v4:1\], but now using the relation ${\lhook\joinrel\xrightarrow{*}}$. Next, we show that any $\rho$-continuous mapping $\varphi:X\operatorname{\leadsto}Y$ with uniformly $UV^*$-values, admits a special sequences $\{\mathscr{U}_n:n<\omega\}$ of open covers of $X$ approximating the values of $\varphi$, see Theorem \[theorem-Near-sel-v1:1\]. This is essentially the only place where $\rho$-continuity of $\varphi$ is used. Section \[sec:orient-simpl-compl\] contains two constructions of continuous maps from the geometric realisation of oriented simplicial complexes, see Theorems \[theorem-Near-sel-v7:1\] and \[theorem-Near-sel-v9:1\]. Finally, in Section \[sec:appr-select-c\], we apply Theorem \[theorem-Near-sel-v7:1\] to nerves of covers to obtain with ease our generalisation of Theorem \[theorem-Selections-IDS:1\] (see Theorem \[theorem-Finite-C-v6:1\]). The same approach, but now using Theorem \[theorem-Near-sel-v9:1\], gives the generalisation of Theorem \[theorem-Selections-IDS-v4:1\] (see Theorem \[theorem-Finite-C-v5:1\]). The last section of the paper contains two applications of these theorems dealing with near-selections for Hausdorff continuous mappings whose values are absolute retracts, see Corollaries \[corollary-Near-sel-vgg:1\] and \[corollary-Near-sel-vgg:2\].
Monotone Embedding Relations {#sec:monot-embedd-relat}
============================
Let $S{\lhook\joinrel\xrightarrow{*}}T$ be an embedding relation defined on subsets $S,T\subset Y$ of a space $Y$. In this, we will always assume that $S\subset T$, so that the relation ${\lhook\joinrel\xrightarrow{*}}$ places some extra properties on the inclusion $S\subset T$. Such a relation will be called *monotone* if $A\subset S{\lhook\joinrel\xrightarrow{*}} T\subset B$ implies $A{\lhook\joinrel\xrightarrow{*}} B$. Various examples of monotone embedding relations will be given in the next sections. In fact, several $UV$-properties can be expressed in these terms by saying that a subset $S\subset Y$ is $UV^*$ if every neighbourhood $U$ of $S$ contains a neighbourhood $V$ of $S$ such that $V{\lhook\joinrel\xrightarrow{*}} U$.
In what follows, all covers are assumed to be of *nonempty* sets. Also, in all statements in this section, $(Y,\rho)$ is a fixed metric space and ${\lhook\joinrel\xrightarrow{*}}$ is a monotone embedding relation on the subsets of $Y$. In this general setting, we shall say that a subset $S\subset Y$ is *uniformly $UV^*$* if for every $\varepsilon>0$ there exists $\delta=\delta(\varepsilon)>0$ such that $\mathbf{O}_\delta(S){\lhook\joinrel\xrightarrow{*}} \mathbf{O}_\varepsilon(S)$. Clearly, we may always pick $\delta$ such that $\delta(\varepsilon)\leq \varepsilon$, which is in good accord with the requirement that $\mathbf{O}_\delta(S)\subset \mathbf{O}_\varepsilon(S)$.
The main goal of this section is the following approximative representation of $\rho$-continuous mappings with uniformly $UV^*$-values.
\[theorem-Near-sel-v1:1\] Let $X$ be a paracompact space and $\varphi:X\operatorname{\leadsto}Y$ be a $\rho$-continuous mapping such that each $\varphi(x)$, $x\in X$, is uniformly $UV^*$. Then for every continuous function $\varepsilon:X\to (0,+\infty)$ there exists a sequence $\mathscr{U}_n$, $n<\omega$, of open covers of $X$ and mappings $\Phi_n,\Psi_n:\mathscr{U}_n\operatorname{\leadsto}Y$ such that for every $U\in \mathscr{U}_n$,
1. \[item:Near-sel-v3:1\] $\Phi_n(U){\lhook\joinrel\xrightarrow{*}}\Psi_n(U)\subset
\mathbf{O}_{\varepsilon(p)}(\varphi(p))$, whenever $p\in U$
2. \[item:Near-sel-v3:2\] $\Psi_k(V)\subset \Phi_n(U)$, whenever $k>n$ and $V\in \mathscr{U}_k$ with $V\cap U\neq {\varnothing}$.
To prepare for the proof, we proceed with several observations.
\[proposition-Near-sel-v1:1\] Let $\varepsilon:X\to (0,+\infty)$ be a continuous function on a space $X$. Then there exists an open cover $\mathscr{V}$ of $X$ and a function $\delta:\mathscr{V}\to (0,+\infty)$ such that $\delta(V)\leq \varepsilon(p)$, for every $p\in V\in \mathscr{V}$.
Since $\varepsilon$ is continuous, each point $x\in X$ is contained in an open set $V_x$ such that $\varepsilon(p)>\frac{\varepsilon(x)}2=\delta_x$, for every $p\in V_x$. Take $\mathscr{V}=\{V_x:x\in X\}$, and for each $V\in \mathscr{V}$ set $\delta(V)=\delta_x$, where $\delta_x$ corresponds to some $x\in X$ with $V=V_{x}$.
Let $\mathscr{F}(Y)$ be the collection of all nonempty closed subsets of $Y$. A natural topology on $\mathscr{F}(Y)$ is the Hausdorff one $\tau_{H(\rho)}$ generated by the Hausdorff distance $H(\rho)$ associated to $\rho$. Recall that the *Hausdorff distance* is defined by $$\begin{aligned}
\label{eq:Near-sel-v4:1}
H(\rho)(S,T)&=\inf\big\{\varepsilon>0: S\subset
\mathbf{O}_\varepsilon(T)\ \text{and}\ T\subset
\mathbf{O}_\varepsilon(S)\big\}\\
&=\sup\left\{\rho(S,y)+\rho(y,T): y\in S\cup
T\right\},\qquad S,T\in \mathscr{F}(Y).\nonumber\end{aligned}$$ The Hausdorff topology $\tau_{H(\rho)}$ is always metrizable even though the distance $H(\rho)$ may assume infinite values. Moreover, given any nonempty set $S\subset Y$, we have that $\rho(y,S)=\rho\left(y,\overline{S}\right)$, for every $y\in Y$. So, we may consider $H(\rho)(S,T)$ defined on all nonempty subsets of $Y$. Then, according to and , a mapping $\varphi:X\operatorname{\leadsto}Y$ is $\rho$-continuous if and only if it is continuous with respect to $H(\rho)$ as a usual map, namely if for every $\varepsilon>0$, each point $x\in X$ is contained in an open set $V$ such that $H(\rho)(\varphi(p),\varphi(x))<\varepsilon$ for every $p\in V$.
\[proposition-Finite-C-v4:1\] Let $\varphi:X\operatorname{\leadsto}Y$ be a $\rho$-continuous mapping on a space $X$, and $x\in X$ be such that $\varphi(x)$ is uniformly $UV^*$. Then for every $\varepsilon>0$ there exists $\delta_x\in (0,\varepsilon)$ and an open set $U_x\subset X$ containing $x$ such that for every $p,q\in U_x$, $$H(\rho)(\varphi(p),\varphi(q))<\frac{\delta_x}2\quad
\text{and}\quad \mathbf{O}_{\delta_x}(\varphi(p)){\lhook\joinrel\xrightarrow{*}}
\mathbf{O}_\varepsilon(\varphi(p)).$$
Take $\delta_x\in \left(0,\frac\varepsilon4\right)$ with $\mathbf{O}_{2\delta_x}(\varphi(x)) {\lhook\joinrel\xrightarrow{*}}
\mathbf{O}_{\frac\varepsilon2}(\varphi(x))$, and set $$U_x=\left\{p\in X:
H(\rho)(\varphi(p),\varphi(x))<\frac{\delta_x}4\right\}.$$ Then $U_x$ is an open set such that $H(\rho)(\varphi(p),\varphi(q))<\frac{\delta_x}2$, for every $p,q\in U_x$. Moreover, $p\in U_x$ implies $\mathbf{O}_{\delta_x}(\varphi(p)){\lhook\joinrel\xrightarrow{*}}
\mathbf{O}_\varepsilon(\varphi(p))$ because ${\lhook\joinrel\xrightarrow{*}}$ is monotone and $$\mathbf{O}_{\delta_x}(\varphi(p))\subset
\mathbf{O}_{2\delta_x}(\varphi(x)) {\lhook\joinrel\xrightarrow{*}}
\mathbf{O}_{\frac\varepsilon2}(\varphi(x)) \subset
\mathbf{O}_\varepsilon(\varphi(p)).\qedhere$$
\[proposition-Finite-C-v4:3\] Let $X$ be a paracompact space and $\varphi:X\operatorname{\leadsto}Y$ be a $\rho$-continuous mapping such that each $\varphi(x)$, $x\in X$, is uniformly $UV^*$. Suppose that $\mathscr{U}_1$ is a locally finite open cover of $X$ and $\delta_1:\mathscr{U}_1\to (0,+\infty)$. Then there exists an open locally finite refinement $\mathscr{U}_2$ of $\mathscr{U}_1$ and functions $\delta_2,\varepsilon_2:\mathscr{U}_2\to (0,+\infty)$ with $\delta_2\leq \varepsilon_2$, such that
1. $\mathbf{O}_{\delta_2(U)}(\varphi(p)){\lhook\joinrel\xrightarrow{*}}
\mathbf{O}_{\varepsilon_2(U)}(\varphi(p))$, for every $p\in U\in \mathscr{U}_2$,
2. $H(\rho)(\varphi(p),\varphi(q))<\frac{\delta_2(U)}2$, for every $p,q\in U\in \mathscr{U}_2$,
3. $\varepsilon_2(U_2)\leq \frac{\delta_1(U_1)}3$, whenever $U_i\in \mathscr{U}_i$, $i=1,2$, with $U_1\cap U_2\neq {\varnothing}$.
Since $\mathscr{U}_1$ is locally finite, the paracompact space $X$ has a locally finite open cover $\mathscr{V}$ such that each family $\mathscr{U}_1^V=\left\{U\in \mathscr{U}_1: U\cap \overline{V}\neq
{\varnothing}\right\}$, $V\in \mathscr{V}$, is finite. For convenience, set $$\label{eq:Near-sel-vgg:1}
\varepsilon(V)= \min_{U\in
\mathscr{U}_1^V}\frac{\delta_1(U)}3>0,\quad \text{whenever $V\in
\mathscr{V}$.}$$ Then by Proposition \[proposition-Finite-C-v4:1\], for each $V\in \mathscr{V}$ there exists an open and locally finite in $\overline{V}$ cover $\mathscr{W}_V$ of the paracompact space $\overline{V}$ and a map $\delta_V:\mathscr{W}_V\to (0,\varepsilon(V))$ such that $\mathscr{W}_V$ refines $\mathscr{U}_1^{V}$ and for every $p,q\in W\in \mathscr{W}_V$, $$\label{eq:Near-sel-vgg:3}
H(\rho)(\varphi(p),\varphi(q))<\frac{\delta_V(W)}2\quad
\text{and}\quad \mathbf{O}_{\delta_V(W)}(\varphi(p)){\lhook\joinrel\xrightarrow{*}}
\mathbf{O}_{\varepsilon(V)}(\varphi(p)).$$
Finally, we can take $\mathscr{U}_2=\left\{W\cap V: W\in \mathscr{W}_V\ \text{and}\ V\in
\mathscr{V}\right\}$, which is a locally finite open cover of $X$ because so is $\mathscr{V}$ and each $\mathscr{W}_V$, $V\in \mathscr{V}$, is locally finite in $\overline{V}$. To define the required maps $\delta_2,\varepsilon_2:\mathscr{U}_2\to (0,+\infty)$, for each $U\in \mathscr{U}_2$ fix elements $\lambda(U)\in \mathscr{V}$ and $\mu(U)\in \mathscr{W}_{\lambda(U)}$ with $U=\mu(U)\cap \lambda(U)$, and set $$\label{eq:Near-sel-vgg:4}
\delta_2(U)=\delta_{\lambda(U)}(\mu(U))\quad \text{and}\quad
\varepsilon_2(U)= \varepsilon(\lambda(U)),\ \text{
$U\in \mathscr{U}_2$.}$$
If $U_1\cap U_2\neq {\varnothing}$ for some $U_i\in \mathscr{U}_i$, $i=1,2$, then $U_1\cap \lambda(U_2)\neq {\varnothing}$ and, therefore, $U_1\in \mathscr{U}_1^{\lambda(U_2)}$. According to , we get that $\varepsilon_2(U_2)=\varepsilon(\lambda(U_2)) \leq
\frac{\delta_1(U_1)}3$. The rest of the properties are evident from and . The proof is complete.
We are now ready for the proof of Theorem \[theorem-Near-sel-v1:1\].
Let $\varphi$ and $\varepsilon$ be as in this theorem. Since $X$ is a paracompact space, by Proposition \[proposition-Near-sel-v1:1\], it has an open locally finite cover $\mathscr{V}$ and a function $\delta:\mathscr{V}\to (0,+\infty)$ such that $$\label{eq:Near-sel-v1:1}
\delta(V)\leq \varepsilon(p),\quad \text{for every
$p\in V\in \mathscr{V}$.}$$ Hence, by Proposition \[proposition-Finite-C-v4:3\] (applied with $\mathscr{U}_1=\mathscr{V}$ and $\delta_1=\delta$), $X$ has an open locally finite cover $\mathscr{U}_0$ which refines $\mathscr{V}$, and maps $\delta_0,\varepsilon_0: \mathscr{U}_0\to (0,+\infty)$ with $\delta_0\leq \varepsilon_0$, such that for every $p,q\in U\in \mathscr{U}_0$, $$\label{eq:Near-sel-v1:2}
\begin{cases}
\mathbf{O}_{\delta_0(U)}(\varphi(p)){\lhook\joinrel\xrightarrow{*}}
\mathbf{O}_{\varepsilon_0(U)}(\varphi(p)),\\[3pt]
H(\rho)(\varphi(p),\varphi(q))<\frac{\delta_0(U)}2,\\
\varepsilon_0(U)\leq \frac{\varepsilon(p)}2.\\
\end{cases}$$ To see the last property in , take $p\in U\in \mathscr{U}_0$ and $V\in \mathscr{V}$ with $U\subset
V$. Then by and Proposition \[proposition-Finite-C-v4:3\], $\varepsilon_0(U)\leq \frac{\delta(V)}3\leq
\frac{\varepsilon(p)}3\leq \frac{\varepsilon(p)}2$.
Using Proposition \[proposition-Finite-C-v4:3\], the construction can be carried on by induction. Thus, there exists a sequence $\mathscr{U}_n$, $n<\omega$, of locally finite open covers of $X$ and functions $\delta_n,\varepsilon_n:\mathscr{U}_n\to (0,+\infty)$ with $\delta_n\leq \varepsilon_n$, such that each $\mathscr{U}_{n+1}$ refines $\mathscr{U}_n$ and the following properties hold for each $p,q\in U\in \mathscr{U}_{n}$: $$\label{eq:Near-sel-v2:1}
\begin{cases}
\mathbf{O}_{\delta_n(U)}(\varphi(p)){\lhook\joinrel\xrightarrow{*}}
\mathbf{O}_{\varepsilon_n(U)}(\varphi(p)),\\[3pt]
H(\rho)(\varphi(p),\varphi(q))<\frac{\delta_n(U)}2,\\
\varepsilon_{n+1}(V)\leq \frac{\delta_n(U)}3,\ \text{whenever
$V\in \mathscr{U}_{n+1}$ with $V\cap U\neq {\varnothing}$.}
\end{cases}$$
To finish the proof, for each $n<\omega$ take a map $\pi_n:\mathscr{U}_n\to X$ with $\pi_n(U)\in U$, $U\in \mathscr{U}_n$. Next, define $\Phi_n,\Psi_n:\mathscr{U}_n\operatorname{\leadsto}Y$ by $$\label{eq:Near-sel-v5:1}
\Phi_n(U)=\mathbf{O}_{\delta_n(U)}(\varphi(\pi_n(U)))\quad
\text{and}\quad
\Psi_n(U)=\mathbf{O}_{\varepsilon_n(U)}(\varphi(\pi_n(U))),\ U\in
\mathscr{U}_n.$$
To see \[item:Near-sel-v3:2\], take $k>n$, $U\in \mathscr{U}_n$ and $V_k\in \mathscr{U}_{k}$ with $V_k\cap U\neq {\varnothing}$. Since each $\mathscr{U}_{i+1}$ refines $\mathscr{U}_i$, $i<\omega$, there are elements $V_i\in \mathscr{U}_i$, $n+1\leq i< k$, such that $V_{i+1}\subset V_{i}$ for every $n+1\leq i< k$. Take a point $q\in V_k\cap U\subset V_{n+1}\cap U$. Since $\pi_k(V_k),q\in V_{n+1}$ and $q,\pi_n(U)\in U$, it follows from that $$\begin{aligned}
H(\rho)(\varphi(\pi_{k}(V_k)),
&\varphi(\pi_n(U)))\leq\\
&\leq H(\rho)(\varphi(\pi_{k}(V_k)),\varphi(q))+
H(\rho)(\varphi(q),\varphi(\pi_n(U)))\\
&<
\frac{\delta_{n+1}(V_{n+1})}2+\frac{\delta_n(U)}2
\leq
\frac12 \frac{\delta_{n}(U)}3+ \frac{\delta_{n}(U)}2=
\frac{2\delta_{n}(U)}3.
\end{aligned}$$ For the same reason, $\varepsilon_k(V_k)\leq \dots \leq \varepsilon_{n+1}(V_{n+1})\leq
\frac{\delta_n(U)}3$. Accordingly, $$\begin{aligned}
\Psi_{k}(V_k)=\mathbf{O}_{\varepsilon_{k}(V_k)}(\varphi(\pi_{k}(V_k)))
&\subset
\mathbf{O}_{\frac{\delta_n(U)}3}(\varphi(\pi_{k}(V_{k})))\\
&\subset
\mathbf{O}_{\frac{\delta_n(U)}3+
\frac{2\delta_n(U)}3}(\varphi(\pi_n(U)))=\Phi_n(U).
\end{aligned}$$
Finally, we show that \[item:Near-sel-v3:1\] holds as well. The “${\lhook\joinrel\xrightarrow{*}}$”-embedding property in \[item:Near-sel-v3:1\] is evident from and . As for the inclusion $\Psi_n(U)\subset \mathbf{O}_{\varepsilon(p)}(\varphi(p))$, where ${p\in U\in \mathscr{U}_n}$, it follows from the special case of $n=0$. Indeed, for $n>0$, take $U_0\in \mathscr{U}_0$ with $U\subset U_0$, which is possible because $\mathscr{U}_n$ refines $\mathscr{U}_0$. Then according to \[item:Near-sel-v3:2\], $\Psi_n(U)\subset \Phi_0(U_0)$ because $U\cap U_0\neq
{\varnothing}$. Thus, by and , we also have that $$\begin{aligned}
\Psi_n(U)\subset
\Phi_0(U_0)&=
\mathbf{O}_{\delta_0(U_0)}(\varphi(\pi_0(U_0)))\\
&\subset \mathbf{O}_{2\delta_0(U_0)}(\varphi(p))\subset
\mathbf{O}_{2\varepsilon_0(U_0)}(\varphi(p))\subset
\mathbf{O}_{\varepsilon(p)}(\varphi(p)).\qedhere
\end{aligned}$$
Oriented Simplicial Complexes {#sec:orient-simpl-compl}
=============================
As stated in the Introduction, we identify the vertices of a simplicial complex $\Sigma$ with its $0$-skeleton $\Sigma^0$. A simplicial complex $\Sigma$ is called *oriented* if there exists a partial order $\leq$ on its vertex set $\Sigma^0$ such that each simplex of $\Sigma$ is linearly ordered with respect to $\leq$.
The present section deals with two constructions of continuous maps on oriented simplicial complexes. In the first construction, for a space $Y$ and subsets ${S\subset T\subset Y}$, we write $S{\lhook\joinrel\xrightarrow{\infty}} T$ if $S$ is contractible in $T$. Moreover, for mappings $\Phi,\Psi:Z\operatorname{\leadsto}Y$, we will use $\Phi{\lhook\joinrel\xrightarrow{\infty}}\Psi$ to express that $\Phi(z){\lhook\joinrel\xrightarrow{\infty}} \Psi(z)$, for every $z\in Z$.
\[theorem-Near-sel-v7:1\] Let $\Sigma$ be an oriented simplicial complex with respect to some partial order $\leq$ on $\Sigma^0$, and $\Phi,\Psi:\Sigma^0\operatorname{\leadsto}Y$ be mappings into a space $Y$ such that $\Phi{\lhook\joinrel\xrightarrow{\infty}} \Psi$ and $\Psi(v)\subset \Phi(u)$, whenever $u,v\in \Sigma^0$ with $u<v$. Then there exists a continuous map $h:|\Sigma|\to Y$ such that $$\label{eq:Near-sel-v7:1}
h(|\sigma|)\subset \Psi(\min \sigma),\quad\text{for every
$\sigma\in \Sigma$.}$$
The proof of Theorem \[theorem-Near-sel-v7:1\] is based on the following considerations. The *cone* $Z*v$ over a space $Z$ with a vertex $v$ is the quotient space of $Z\times[0,1]$ obtained by identifying all points of $Z\times\{1\}$ into a single point $v$. The following representation of $Z*v$ will play an important role in our proof. $$\label{eq:Near-sel-v7:2}
Z*v=\big\{t v + (1-t) z: z\in Z\ \text{and}\ t\in [0,1]\big\}.$$ This representation is unique for each point $q\in Z*v$ with $q\neq v$, and is in good accord with the fact that $|\sigma|=|\tau|*v$, whenever $\sigma$ is an abstract simplex, $v\in \sigma$ and $\tau=\sigma\setminus\{v\}\neq {\varnothing}$ is the face opposite to the vertex $v$. Finally, for a space $T$ and $S\subset T$, let us recall that a homotopy $H:S\times [0,1]\to T$ is a contraction of $S$ into a point $p\in T$ if $$\label{eq:Near-sel-v7:4}
H(y,0)=y\ \ \text{and}\ \
H(y,1)=p,\quad \text{for every $y\in S$.}$$
\[proposition-Near-sel-v7:2\] Let $H:S\times [0,1]\to T$ be a contraction of a subset $S\subset T$ into a point $p\in T$ of a space $T$. Then for a space $Z$, each continuous map $g:Z\to S$ can be extended to a continuous map $h:Z*v\to T$ such that $$\label{eq:Near-sel-v7:3}
h\big(tv+(1-t)z\big)=H(g(z),t),\quad \text{for every $z\in Z$ and
$t\in [0,1]$.}$$
Define $h:Z*v\to T$ as in , which is possible because $H(g(z),1)=p$ for every $z\in Z$, see and . This $h$ is as required because $H(g(z),0)=g(z)$, for every $z\in Z$.
We apply the above construction to show the special case of Theorem \[theorem-Near-sel-v7:1\] in the setting of an oriented simplex $\sigma$ by a linear order $\leq$ on its vertices. For convenience, such a simplex will be denoted by $(\sigma,\leq)$.
\[proposition-Near-sel-v8:1\] Let $(\sigma,\leq)$ be an oriented simplex and $\Phi,\Psi:\sigma\operatorname{\leadsto}Y$ be mappings into a space $Y$ such that $\Phi{\lhook\joinrel\xrightarrow{\infty}}\Psi$ and $\Psi(v)\subset \Phi(u)$, for every $u,v\in \sigma$ with $u<v$. For every $v\in \sigma$, fix a contraction $H_v:\Phi(v)\times[0,1]\to \Psi(v)$ into some point $p_v\in \Psi(v)$. Then there exists a unique continuous map $h:|\sigma|\to Y$ such that
1. \[item:Near-sel-v8:1\] $h(v)=p_v$, $v\in \sigma$,
2. \[item:Near-sel-v8:2\] $h\big(tv+(1-t)z\big)=H_v(h(z),t)$, $z\in |\tau|$ and $t\in [0,1]$, whenever $\tau\subset \sigma$ is a face and $v\in \sigma$ with $v< u$ for every $u\in \tau$.
Suppose that $\sigma=\{v_0,\dots, v_{n+1}\}$, where $v_0<v_1<\dots <v_{n+1}$, and consider the maximal chain $\tau_{n+1}\subset \tau_n\subset\dots \subset \tau_0=\sigma$ of faces of $\sigma$ defined by $\tau_{n+1}=\{v_{n+1}\}$ and $\tau_k=\tau_{k+1}\cup\{v_k\}$, $k\leq n$. We will construct the required map $h:|\sigma|\to Y$ inductively on the faces $|\tau_k|$, $k\leq n+1$, of this chain. To this end, set $S_k=\Phi(v_k)$, $T_k=\Psi(v_k)$, $H_k=H_{v_k}$ and $p_k=p_{v_k}$, $k\leq n+1$. Next, define $h_{n+1}:|\tau_{n+1}|\to T_{n+1}$ by $h_{n+1}(v_{n+1})=p_{n+1}$. Since $|\tau_n|=|\tau_{n+1}|*v_n$ and $T_{n+1}=\Psi(v_{n+1})\subset \Phi(v_n)=S_n$, by Proposition \[proposition-Near-sel-v7:2\], $h_{n+1}$ has a continuous extension $h_n:|\tau_n|\to T_n$ such that $$\begin{aligned}
h_n\big(tv_n+(1-t)z\big)&=H_n(h_{n+1}(z),t)\\
&=H_n(h_n(z),t),\quad\text{$z\in
|\tau_{n+1}|$ and $t\in [0,1]$.}
\end{aligned}$$ It is also evident that $h_n(v_n)=H_n(p_{n+1},1)=p_n$. The construction can be carried on by a finite induction to get a continuous map $h=h_0:|\tau_0|=|\sigma|\to T_0$ satisfying \[item:Near-sel-v8:1\] and such that for every $0\leq k\leq n$, $$\label{eq:Near-sel-v8:1}
h\big(tv_k+(1-t)z\big)=H_{k}(h(z),t),\quad \text{$z\in
|\tau_{k+1}|$ and $t\in [0,1]$.}$$ This $h$ is as required. Indeed, take any continuous map $g:|\sigma|\to Y$ satisfying \[item:Near-sel-v8:1\] and . Then $g(v_{n+1})=h(v_{n+1})$ and by , we get that ${g\operatorname{\upharpoonright}|\tau_n|=h\operatorname{\upharpoonright}|\tau_n|}$. Inductively, this implies that $g=g\operatorname{\upharpoonright}|\tau_0|=h\operatorname{\upharpoonright}|\tau_0|=h$.
To see finally that \[item:Near-sel-v8:2\] is equivalent to , take a face $\tau\subset \sigma$ and a vertex $v\in \sigma$ with $v<u$, for every $u\in \tau$. Then $v=v_k$ for some $0\leq k\leq n$, and $\tau\subset \tau_{k+1}$. If $z\in |\tau|\subset |\tau_{k+1}|$ and $t\in [0,1]$, it follows from that $$h\big(tv+(1-t)z\big)=h\big(tv_k+(1-t)z\big)=
H_k(h(z),t)=H_v(h(z),t).\qedhere$$
Let $\Sigma$ be an oriented simplicial complex with respect to some partial order $\leq$ on $\Sigma^0$, and $\Phi,\Psi:\Sigma^0\operatorname{\leadsto}Y$ be as in Theorem \[theorem-Near-sel-v7:1\]. For each vertex ${v\in \Sigma^0}$, fix a contraction $H_v:\Phi(v)\times[0,1]\to \Psi(v)$ into some point ${p_v\in \Psi(v)}$. Since each simplex $\sigma\in \Sigma$ is oriented with respect to $\leq$, by Proposition \[proposition-Near-sel-v8:1\] applied with $\Phi\operatorname{\upharpoonright}\sigma$, $\Psi\operatorname{\upharpoonright}\sigma$ and the fixed contractions $H_v$, $v\in \sigma$, there exists a unique continuous map $h_\sigma:|\sigma|\to Y$ satisfying both properties \[item:Near-sel-v8:1\] and \[item:Near-sel-v8:2\] in that proposition. Suppose that $\sigma_1,\sigma_2\in \Sigma$ are simplices which have a common face $\tau=\sigma_1\cap
\sigma_2$. Then the linear order $\leq$ on $\tau$ is the same in each one of the oriented simplices $(\sigma_i,\leq)$, $i=1,2$. Hence, by Proposition \[proposition-Near-sel-v8:1\], we get that $h_{\sigma_1}\operatorname{\upharpoonright}|\tau|= h_\tau=h_{\sigma_2}\operatorname{\upharpoonright}|\tau|$. Accordingly, we may define a map $h:|\Sigma|\to Y$ by $h\operatorname{\upharpoonright}|\sigma|=h_\sigma$, $\sigma\in \Sigma$. This map $h$ is as required. Indeed, it is continuous because so is each $h_\sigma$, $\sigma\in \Sigma$. To see that $h$ also has the property in , take a simplex $\sigma\in \Sigma$, and set $v=\min\sigma$. If $\sigma=\{v\}$, then by \[item:Near-sel-v8:1\] of Proposition \[proposition-Near-sel-v8:1\], $h(|\sigma|)=h(v)=h_\sigma(v)=p_v\in \Psi(v)$. Otherwise, if $\tau=\sigma\setminus\{v\}\neq {\varnothing}$, it follows from \[item:Near-sel-v8:2\] of Proposition \[proposition-Near-sel-v8:1\] that for every $z\in |\tau|$ and $t\in [0,1]$, $$h\big(tv+(1-t)z\big)= h_\sigma\big(tv
+(1-t)z\big)=H_v(h_\sigma(z),t)\in \Psi(v).$$ Accordingly, $h(|\sigma|)\subset \Psi(v)$ because $|\sigma|=|\tau|*v$.
Recall that the $k$-skeleton $\Sigma^k$, $k\geq 0$, of a simplicial complex $\Sigma$ is the simplicial subcomplex $\Sigma^{k}=\{\sigma\in \Sigma:\operatorname{Card}(\sigma)\leq k+1\}$. A simplicial complex $\Sigma$ is called *finite-dimensional* if $\Sigma=\Sigma^k$ for some ${k\geq 0}$, and we say that $\Sigma$ is *$k$-dimensional* if $k\geq0$ is the smallest integer with this property, i.e. for which $\Sigma=\Sigma^k$. If $\Sigma$ is an $(n+1)$-dimensional simplicial complex which is oriented with respect to some partial order $\leq$ on $\Sigma^0$, then we may consider the chain of subcomplexes $\Sigma_{n+1}\subset \Sigma_n\subset \dots \subset \Sigma_0=\Sigma$ defined by $$\label{eq:Near-sel-v9:2}
\begin{cases}
\Sigma_k=(\Sigma_k)^{n+1-k}, &\text{and}\\
\min\sigma\notin (\Sigma_{k+1})^0, &\text{whenever
$\sigma\in \Sigma_k\setminus \Sigma_{k+1}$.}
\end{cases}$$ Namely, let $V_0$ be the minimal elements of $\Sigma^0$, i.e. $V_0$ consists of all $v\in \Sigma^0$ such that $v=u$ whenever $u\in
\Sigma^0$ with $u\leq v$. Next, let $V_1$ be the minimal elements of $\Sigma^0\setminus V_0$. Thus, inductively, for each $k\leq n$, we may define the set $V_{k+1}$ which consists of the minimal elements of $\Sigma^0\setminus (V_0\cup\dots\cup V_k)$. Since $\Sigma$ is $(n+1)$-dimensional, each simplex of $\Sigma$ has at most $n+2$ elements. Therefore, $\Sigma^0= V_0\cup \dots \cup V_{n+1}$ and we may now define the required subcomplexes by $$\label{eq:Near-sel-vgg:2}
\Sigma_k=\left\{\sigma\in \Sigma: \sigma\subset V_k\cup\dots \cup
V_{n+1}\right\},\quad k\leq n+1.$$
Our second construction deals with finite-dimensional oriented simplicial complexes by relaxing the requirement in Theorem \[theorem-Near-sel-v7:1\] that $\Phi{\lhook\joinrel\xrightarrow{\infty}}\Psi$. To this end, for a space $Y$ and $S\subset T\subset Y$, we will write that $S{\lhook\joinrel\xrightarrow{\omega}} T$ if for each finite-dimensional simplicial complex $\Sigma$, any continuous map $g:|\Sigma|\to S$ is null-homotopic in $T$. Just like before, for mappings $\Phi,\Psi:Z\operatorname{\leadsto}Y$, we will use $\Phi{\lhook\joinrel\xrightarrow{\omega}}\Psi$ to designate that $\Phi(z){\lhook\joinrel\xrightarrow{\omega}} \Psi(z)$, for every $z\in Z$.
For a simplicial complex $\Sigma$ and a point $v$ with $v\notin |\Sigma|$, the *cone* on $\Sigma$ with a vertex $v$, is the simplicial complex defined by $$\label{eq:Near-sel-v9:3}
\Sigma*v=\Sigma\cup\left\{\sigma\cup \{v\}:
\sigma\in\Sigma\right\}\cup \{v\}.$$ According to , $|\Sigma*v|$ is the cone $|\Sigma|*v$ over the geometric realisation $|\Sigma|$ of $\Sigma$. Moreover, for a space $T$ and $S\subset T$, we have that $S{\lhook\joinrel\xrightarrow{\omega}} T$ iff for each finite-dimensional simplicial complex $\Sigma$ and $v\notin |\Sigma|$, any continuous map $g:|\Sigma|\to S$ can be extended to a continuous map $h:|\Sigma|*v\to T$ on the cone $|\Sigma|*v=|\Sigma*v|$, compare with Proposition \[proposition-Near-sel-v7:2\]. Thus, we also have the following theorem.
\[theorem-Near-sel-v9:1\] Let $\Sigma$ be a finite-dimensional simplicial complex which is oriented with respect to some partial order $\leq$ on $\Sigma^0$, and $\Phi,\Psi:\Sigma^0\operatorname{\leadsto}Y$ be mappings into a space $Y$ such that $\Phi{\lhook\joinrel\xrightarrow{\omega}} \Psi$ and $\Psi(v)\subset \Phi(u)$, whenever $u,v\in \Sigma^0$ with $u<v$. Then there exists a continuous map $h:|\Sigma|\to Y$ such that $$\label{eq:Near-sel-v9:1}
h(|\sigma|)\subset \Psi(\min \sigma),\quad\text{for every
$\sigma\in \Sigma$.}$$
Suppose that $\Sigma$ is $(n+1)$-dimensional, and consider the chain of subcomplexes $\Sigma_{n+1}\subset \Sigma_n\subset\dots \subset \Sigma_0=\Sigma$ with the properties in , see also . We are going to construct continuous maps $h_k:|\Sigma_k|\to Y$, $k\leq n+1$, satisfying and such that each $h_k$ is an extension of $h_{k+1}$. Then we can take $h=h_0$ because $\Sigma_0=\Sigma$. To this end, using that $\Sigma_{n+1}=(\Sigma_{n+1})^0$ is $0$-dimensional, we may define $h_{n+1}:\Sigma_{n+1}\to Y$ by $h_{n+1}(v)\in \Psi(v)$, for every $v\in \Sigma_{n+1}$. Evidently, this $h_{n+1}$ satisfies with respect to any singleton (simplex) of $\Sigma_{n+1}$.
Suppose that $h_{k+1}:\big|\Sigma_{k+1}\big|\to Y$ is a continuous map satisfying , and let us show how to extend it to a continuous map $h_{k}:\big|\Sigma_{k}\big|\to Y$ preserving this property. So, take any vertex $v\in \Sigma^0$ with $\{v\}\in \Sigma_k\setminus \Sigma_{k+1}$, and consider the cone $\Omega_v* v\subset \Sigma_k$ associated to the subcomplex $\Omega_v=\big\{\sigma\in \Sigma_{k+1}: \sigma\cup\{v\}\in
\Sigma_{k}\big\}$, see . If $\Omega_v={\varnothing}$, then $\Omega_v*v$ is the singleton $\{v\}$, and we define $h_v:|\Omega_v*v|\to Y$ by $h_v(v)\in
\Psi(v)$. Suppose that $\Omega_v\neq {\varnothing}$, and take any simplex $\sigma\in \Omega_v\subset \Sigma_{k+1}$. According to , $v<\min\sigma$ and, therefore, $\Psi(\min\sigma)\subset \Phi(v)$. Since $h_{k+1}$ satisfies , this implies that $h_{k+1}(|\sigma|)\subset \Phi(v)$. Thus, $h_{k+1}(|\Omega_v|)\subset \Phi(v)$ and using that $\Phi(v){\lhook\joinrel\xrightarrow{\omega}} \Psi(v)$, there exists a continuous extension $h_v:|\Omega_v*v|\to \Psi(v)$ of $h_{k+1}\operatorname{\upharpoonright}|\Omega_v|$. Evidently, $h_v$ satisfies with respect to the simplicial complex $\Omega_v*v$. Finally, we may define the required continuous extension $h_k:|\Sigma_k|\to Y$ of $h_{k+1}$ by letting $h_k\operatorname{\upharpoonright}|\Sigma_{k+1}|=h_{k+1}$ and $h_k\operatorname{\upharpoonright}|\Omega_v*v|=h_v$, for every vertex $v\in \Sigma^0$ with $\{v\}\in \Sigma_k\setminus \Sigma_{k+1}$.
\[remark-Near-sel-v10:1\] The embedding relation ${\lhook\joinrel\xrightarrow{\omega}}$ in Theorem \[theorem-Near-sel-v9:1\] was used only with subcomplexes of $\Sigma$. Hence, this theorem remains valid if the property “$\Phi{\lhook\joinrel\xrightarrow{\omega}} \Psi$” is relaxed to the one that for each simplicial subcomplex $\Omega\subset \Sigma$ and $v\in \Sigma^0$, any continuous map $g:|\Omega|\to \Phi(v)$ is null-homotopic in $\Psi(v)$.
Approximate Selections and $C$-spaces {#sec:appr-select-c}
=====================================
If a paracompact space is a countable union of closed $C$-subspaces, then it is itself a $C$-space [@gutev-valov:00 Theorem 4.1], see also [@addis-gresham:78 Theorem 2.7]. Accordingly, the space $X$ in Theorem \[theorem-Selections-IDS:1\] is a $C$-space. In this section, we show that Theorem \[theorem-Selections-IDS:1\] remains valid for all paracompact $C$-spaces. It is based on the relation “$S{\lhook\joinrel\xrightarrow{\infty}}T$” defined in the previous section, which is clearly a monotone embedding relation. In its terms, a subset $S$ of a metric space $(Y,\rho)$ is *CE* iff it is *uniformly $UV^\infty$* (see Section \[sec:monot-embedd-relat\]), namely if for every $\varepsilon>0$ there exists $\delta\in (0,\varepsilon]$ such that $\mathbf{O}_\delta(S){\lhook\joinrel\xrightarrow{\infty}} \mathbf{O}_\varepsilon(S)$.
\[theorem-Finite-C-v6:1\] Let $X$ be a paracompact $C$-space, $(Y,\rho)$ be a metric space and $\varphi:X\operatorname{\leadsto}Y$ be a $\rho$-continuous mapping such that each $\varphi(x)$, $x\in X$, is uniformly $UV^\infty$. Then $\varphi$ has a continuous $\varepsilon$-selection, for every continuous function ${\varepsilon:X\to (0,+\infty)}$.
For a continuous function $\varepsilon:X\to (0,+\infty)$, take open covers $\mathscr{U}_n$, $n<\omega$, of $X$ and mappings $\Phi_n,\Psi_n:\mathscr{U}_n\operatorname{\leadsto}Y$ as those in Theorem \[theorem-Near-sel-v1:1\] (applied with the relation ${\lhook\joinrel\xrightarrow{\infty}}$ and the function $\varepsilon$). Then, since $X$ is a $C$-space, there exists a sequence $\mathscr{V}_{n}$, $n<\omega$, of pairwise disjoint families of nonempty open subsets of $X$ such that each $\mathscr{V}_n$, $n<\omega$, refines $\mathscr{U}_n$, and $\mathscr{V}=\bigcup_{n<\omega}\mathscr{V}_n$ is a cover for $X$. Moreover, we assume that the families $\mathscr{V}_n$, $n<\omega$, have no common elements.
Let $\mathscr{N}(\mathscr{V})$ be the *nerve* of $\mathscr{V}$, i.e. the simplicial complex defined by $$\label{eq:Near-sel-v10:1}
\mathscr{N}(\mathscr{V})= \left\{\sigma\subset \mathscr{V}:\sigma\
\text{is finite and}\ \bigcap\sigma\neq{\varnothing}\right\}.$$ Then there exists a natural orientation of $\mathscr{N}(\mathscr{V})$ generated by the families $\mathscr{V}_n$, $n<\omega$. Namely, define a partial order $\leq$ on $\mathscr{V}$ by $V<W$, whenever $V\in \mathscr{V}_n$ and $W\in \mathscr{V}_k$ are such that $V\cap W\neq {\varnothing}$ and $n<k$. Evidently, by , this partial order makes $\mathscr{N}(\mathscr{V})$ an oriented simplicial complex.
Since each $\mathscr{V}_n$ refines $\mathscr{U}_n$, there are maps $r_n:\mathscr{V}_n\to \mathscr{U}_n$, $n<\omega$, such that $V\subset r_n(V)$, for every $V\in \mathscr{V}_n$. We may now define mappings $\Phi,\Psi:\mathscr{V}\operatorname{\leadsto}Y$ by $\Phi\operatorname{\upharpoonright}\mathscr{V}_n=\Phi_n\circ r_n$ and $\Psi\operatorname{\upharpoonright}\mathscr{V}_n=\Psi_n\circ r_n$, $n<\omega$. According to Theorem \[theorem-Near-sel-v1:1\] and the definition of the partial order $\leq$ on $\mathscr{V}$, for every $V\in \mathscr{V}$ we have that $$\label{eq:Near-sel-v4:2}
\begin{cases}
\Phi(V){\lhook\joinrel\xrightarrow{\infty}}\Psi(V)\subset
\mathbf{O}_{\varepsilon(p)}(\varphi(p)),\
\text{whenever $p\in V$, and }\\
\Psi(V)\subset \Phi(U),\ \text{whenever $U\in \mathscr{V}$ with
$U<V$ and $U\cap V\neq {\varnothing}$.}
\end{cases}$$ Thus, by Theorem \[theorem-Near-sel-v7:1\], there exists a continuous map ${h:\big|\mathscr{N}(\mathscr{V})\big|\to Y}$ satisfying with respect to this orientation of $\mathscr{N}(\mathscr{V})$. Finally, since $X$ is paracompact, the cover $\mathscr{V}$ has a canonical map $g:X\to |\mathscr{N}(\mathscr{V})|$. In other words, see [@gutev:2018a Proposition 2.5], $g$ is a continuous map such that $g(x)\in |\Sigma_\mathscr{V}(x)|$, $x\in X$, where $\Sigma_\mathscr{V}(x)=\big\{\sigma\in \mathscr{N}(\mathscr{V}):
x\in \bigcap\sigma\big\}$. The composite map $f=h\circ g$
&()\
[X]{} & Y
is now a continuous $\varepsilon$-selection for $\varphi$. Indeed, take $x\in X$ and a simplex ${\sigma\in \Sigma_\mathscr{V}(x)}$ with $g(x)\in |\sigma|$. Then $x\in \min \sigma$ and by and , $$f(x)=h(g(x))\in h(|\sigma|)\subset
\Psi(\min \sigma)\subset
\mathbf{O}_{\varepsilon(x)}(\varphi(x)).\qedhere$$
Approximate Selections and Finite $C$-spaces {#sec:appr-select-finite}
============================================
Finite $C$-spaces were defined by Borst [@MR2280911] for separable metrizable spaces, subsequently the definition was extended by Valov [@valov:00] for arbitrary spaces. For simplicity, we will consider these spaces in the realm of normal spaces. In this setting, a (normal) space $X$ is called a *finite $C$-space* if for any sequence $\{\mathscr{U} _n:n<\omega\}$ of finite open covers of $X$ there exists a finite sequence $\{\mathscr{V} _n:n\leq k\}$ of pairwise disjoint open families in $X$ such that each $\mathscr{V} _n$ refines $\mathscr{U} _n$ and $\bigcup_{n\leq k}\mathscr{V} _n$ is a cover of $X$. The following characterisation of paracompact finite $C$-spaces was obtained Valov in [@valov:00 Theorem 2.4], it brings the defining property of these spaces closer to that of paracompact $C$-spaces.
\[theorem-Selections-IDS-v2:2\] A paracompact space $X$ is a finite $C$-space if and only if for any sequence $\{\mathscr{U} _n:n<\omega\}$ of open covers of $X$ there exists a finite sequence $\{\mathscr{V} _n:n\leq k\}$ of discrete open families in $X$ such that each $\mathscr{V} _n$ refines $\mathscr{U} _n$ and $\bigcup_{n\leq k}\mathscr{V} _n$ is a cover of $X$.
Regarding the proper place of finite $C$-spaces, it was shown by Valov in [@valov:00 Proposition 2.2] that a space $X$ is a finite $C$-space if and only if its Čech-Stone compactification $\beta X$ is a $C$-space. Accordingly, each compact $C$-space is a finite $C$-space. On the other hand, there are $C$-spaces whose Čech-Stone compactification is not a $C$-space, see e.g.[@MR2080284 Remark 3.7].
In this section, we extend Theorem \[theorem-Selections-IDS-v4:1\] to all paracompact finite $C$-spaces. It is based on the relation “$S{\lhook\joinrel\xrightarrow{\omega}} T$” defined in Section \[sec:orient-simpl-compl\], which is another example of a monotone embedding relation. In terms of this relation, we shall say that a subset $S$ of a metric space $(Y,\rho)$ is *uniformly $UV^\omega$* if for each $\varepsilon>0$ there exists $\delta\in (0,\varepsilon]$ such that $\mathbf{O}_\delta(S){\lhook\joinrel\xrightarrow{\omega}} \mathbf{O}_\varepsilon(S)$.
\[theorem-Finite-C-v5:1\] Let $X$ be a paracompact finite $C$-space, $(Y,\rho)$ be a metric space and $\varphi:X\operatorname{\leadsto}Y$ be a $\rho$-continuous mapping such that each $\varphi(x)$, $x\in X$, is uniformly $UV^\omega$. Then $\varphi$ has a continuous $\varepsilon$-selection, for every continuous function ${\varepsilon:X\to (0,+\infty)}$.
We proceed as in the proof of Theorem \[theorem-Finite-C-v6:1\], but now using Theorem \[theorem-Near-sel-v9:1\] instead of Theorem \[theorem-Near-sel-v7:1\]. Namely, for a continuous function $\varepsilon:X\to (0,+\infty)$, take open covers $\mathscr{U}_n$, $n<\omega$, of $X$ and mappings $\Phi_n,\Psi_n:\mathscr{U}_n\operatorname{\leadsto}Y$ as those in Theorem \[theorem-Near-sel-v1:1\] (applied with the embedding relation ${\lhook\joinrel\xrightarrow{\omega}}$ and the function $\varepsilon$). Since $X$ is a paracompact finite $C$-space, by Theorem \[theorem-Selections-IDS-v2:2\], there exists a finite sequence $\mathscr{V}_{n}$, $n\leq k$, of discrete families of nonempty open subsets of $X$ such that each $\mathscr{V}_n$, $n\leq k$, refines $\mathscr{U}_n$, and $\mathscr{V}=\bigcup_{n\leq k}\mathscr{V}_n$ is a cover for $X$. Next, precisely as in the proof of Theorem \[theorem-Finite-C-v6:1\], we define a partial order on $\mathscr{V}$ and mappings $\Phi,\Psi:\mathscr{V}\operatorname{\leadsto}Y$ such that for every $V\in \mathscr{V}$, $$\label{eq:Near-sel-v10:2}
\begin{cases}
\Phi(V){\lhook\joinrel\xrightarrow{\omega}}\Psi(V)\subset
\mathbf{O}_{\varepsilon(p)}(\varphi(p)),\
\text{whenever $p\in V$, and }\\
\Psi(V)\subset \Phi(U),\ \text{whenever $U\in \mathscr{V}$ with
$U<V$ and $U\cap V\neq {\varnothing}$.}
\end{cases}$$ Thus, by Theorem \[theorem-Near-sel-v9:1\], there exists a continuous map $h:|\mathscr{N}(\mathscr{V})|\to Y$ satisfying with respect to the oriented simplicial complex $\mathscr{N}(\mathscr{V})$.
We may now finish the proof precisely as before. Namely, since $X$ is paracompact, the cover $\mathscr{V}$ admits a continuous map $g:X\to |\mathscr{N}(\mathscr{V})|$ such that $g(x)\in |\sigma_x|$, $x\in X$, where $\sigma_x=\{V\in \mathscr{V}: x\in V\}\in \mathscr{N}(\mathscr{V})$, see [@dowker:47]. Then the composite map $f=h\circ g$ is a continuous $\varepsilon$-selection for $\varphi$. Indeed, $x\in \min \sigma_x$ and by and , we get that $f(x)=h(g(x))\in h(|\sigma_x|)\subset \Psi(\min \sigma_x)\subset
\mathbf{O}_{\varepsilon(x)}(\varphi(x))$.
\[remark-Near-sel-vgg:2\] For an infinite cardinal $\tau$, we shall say that a subset $S$ of a metric space $(Y,\rho)$ is *uniformly $\tau$-$UV^\omega$* if for each $\varepsilon>0$ there exists $\delta\in(0,\varepsilon]$ such that for every finite-dimensional simplicial complex $\Sigma$ with $\operatorname{Card}(\Sigma^0)<\tau$, every continuous map $g:|\Sigma|\to \mathbf{O}_\delta(S)$ is null-homotopic in $\mathbf{O}_\varepsilon(S)$. Also, let us recall that the *degree of compactness* of space $X$ is the least infinite cardinal $\kappa(X)$ such that every open cover of $X$ has an open refinement of cardinality less than ${\kappa(X)}$, see [@choban-mihaylova-nedev:08]. Using the same proof, Theorem \[theorem-Finite-C-v5:1\] remains valid if each $\varphi(x)$, $x\in X$, is only assumed to be uniformly $\kappa(X)$-$UV^\omega$. Indeed, if $\mathscr{V}_n$, $n\leq k$, are as in that proof, then ${\operatorname{Card}(\mathscr{V})<\kappa(X)}$ because each family $\mathscr{V}_n$, $n\leq k$, is discrete. Hence, according to Remark \[remark-Near-sel-v10:1\], we can still apply Theorem \[theorem-Finite-C-v6:1\] to get a continuous map $h:|\mathscr{N}(\mathscr{V})|\to Y$ satisfying with respect to the oriented simplicial complex $\mathscr{N}(\mathscr{V})$. Having already the map $h$, we can finish the proof precisely as before. One benefit of this refined version of Theorem \[theorem-Finite-C-v5:1\] is that it implies Theorem \[theorem-Selections-IDS-v4:1\]. Namely, for a compact space $X$ and a metric space $(Y,\rho)$, a mapping $\varphi:X\operatorname{\leadsto}Y$ has uniformly $\kappa(X)$-$UV^\omega$-values precisely when its values are as in Theorem \[theorem-Selections-IDS-v4:1\].
Approximate Selections and Absolute Retracts
============================================
Here, by an A(N)R we mean a metrizable space which is an *Absolute* (*Neighbourhood*) *Retract* for the metrizable spaces. A closed subset $Y\subset E$ of a metric space $(E,\rho)$ is a *uniform neighbourhood retract* of $E$ if to every $\varepsilon > 0$ there corresponds $\eta(\varepsilon) > 0$ such that there exists a retraction $r:\mathbf{O}_{\eta(+\infty)}(Y)=
\bigcup_{\varepsilon>0}\mathbf{O}_{\eta(\varepsilon)}(Y)\to Y$ with $\rho(z,r(z)) < \varepsilon$, whenever $z\in \mathbf{O}_{\eta(\varepsilon)}(Y)$. If one can take $\eta(+\infty) = +\infty$ (so that the domain of $r$ is always $E$), then $Y$ is called a *uniform retract* of $E$. A metric space $(Y,\rho)$ is called a *uniform* A(N)R if it is a uniform (neighbourhood) retract of any metric space containing $(Y,\rho)$ isometrically as a closed subset. Uniform ANR’s and AR’s were considered by Toruńczyk [@MR365471; @MR445503] and Michael [@michael:79], see also Dugundji [@dugundji:58]. It should be remarked that these uniform retracts are slightly different from those considered by Isbell [@MR141074; @isbell:64]. We now have the following application of Theorem \[theorem-Finite-C-v6:1\].
\[corollary-Near-sel-vgg:1\] Let $X$ be a paracompact $C$-space, $(Y,\rho)$ be a metric space which is a uniform ANR, and $\varphi:X\to \mathscr{F}(Y)$ be a $\rho$-continuous mapping such that each $\varphi(x)$, $x\in X$, is a uniform retract of $Y$. Then $\varphi$ has a continuous $\varepsilon$-selection, for every continuous function ${\varepsilon:X\to (0,+\infty)}$.
According to Theorem \[theorem-Finite-C-v6:1\], it suffices to show that each $\varphi(x)$, $x\in X$, is uniformly $UV^\infty$. To this end, take a point $p\in X$, and let $r:\mathbf{O}_{\eta(+\infty)}(\varphi(p))\to \varphi(p)$ be the corresponding retraction, where $\eta(\varepsilon)$ is as in the definition of a uniform retract for this subset $\varphi(p)\subset Y$. Moreover, since $(Y,\rho)$ is a uniform ANR, to each $\varepsilon>0$ there exists $\gamma(\varepsilon)>0$ such that any two $\gamma(\varepsilon)$-close continuous maps from an arbitrary space to $Y$ are $\varepsilon$-homotopic, see e.g. [@MR3099433 Theorem 6.8.6]. Finally, set $\delta(\varepsilon)= \eta(\gamma(\varepsilon))$, and let us show that this works. So, let $g=r\operatorname{\upharpoonright}\mathbf{O}_{\delta(\varepsilon)}(\varphi(p))$ and $f:\mathbf{O}_{\delta(\varepsilon)}(\varphi(p))\to
\mathbf{O}_{\delta(\varepsilon)}(\varphi(p))$ be the identity of $\mathbf{O}_{\delta(\varepsilon)}(\varphi(p))$. Since $\delta(\varepsilon)=\eta(\gamma(\varepsilon))$, we have that $\rho(f(z),g(z))= \rho(z,r(z))< \gamma(\varepsilon)$, for every $z\in \mathbf{O}_{\delta(\varepsilon)}(\varphi(p))$. Hence, $f$ and $g$ are $\gamma(\varepsilon)$-close and by the choice of $\gamma(\varepsilon)$, they are also $\varepsilon$-homotopic. On the other hand, $\varphi(p)$ is contractible being an AR, see [@dugundji:58 Theorem 12.4]. Therefore, $g$ is null-homotopic in $\varphi(p)$. Thus, the map $f$, i.e. the identity of $\mathbf{O}_{\varepsilon}(\varphi(p))$, is null-homotopic as well.
In case the domain of $\varphi$ is a paracompact finite $C$-space, we may relax both properties that each $\varphi(x)$, $x\in X$, is a uniform ANR and a contractible set (being an AR). In fact, the contractibility can be also relaxed in Corollary \[corollary-Near-sel-vgg:1\]. Namely, for a subset $S\subset T$ of a space $T$ and $k\geq 0$, we write that $S{\lhook\joinrel\xrightarrow{k}} T$ if every continuous map of the $k$-sphere in $S$ can be extended to a continuous map of the $(k+1)$-ball in $T$. A space $S$ is called *aspherical*, or $C^\omega$, if $S{\lhook\joinrel\xrightarrow{k}} S$, for every $k\geq 0$. The following property describing aspherical spaces follows easily from the defining relation that $S{\lhook\joinrel\xrightarrow{k}}S$, for all $k\geq 0$.
\[proposition-Finite-C-v2:1\] A space $S$ is $C^\omega$ if and only if for every simplicial complex $\Sigma$, each continuous map $g:|\Sigma|\to S$ is null-homotopic.
The importance of aspherical spaces is that a metrizable space is an AR iff it is an aspherical ANR, [@dugundji:58 Theorem 12.4].
Evidently, a compact metric space is a uniform A(N)R iff it is an A(N)R. Following Lefschetz [@MR0007094], a compact metric space $(Y,\rho)$ is called $LC^*$ if for every $\varepsilon>0$ there exists $\eta(\varepsilon)>0$ such that if $\Sigma$ is a finite simplicial complex and $K\subset \Sigma$ is a subcomplex containing the vertex set $\Sigma^0$, then every continuous map $g:|K|\to Y$ with $\operatorname{diam}(g(|K\cap \sigma|))<\eta(\varepsilon)$ for every $\sigma\in \Sigma$, can be extended to a continuous map $f:|\Sigma|\to Y$ such that $\operatorname{diam}(f(|\sigma|))<\varepsilon$, $\sigma\in \Sigma$. It was shown by Lefschetz [@MR0007094 Theorem 6.6] that a compact metric space $(Y,\rho)$ is a (uniform) ANR if and only if it is $LC^*$, while $(Y,\rho)$ is an AR iff it is both $LC^*$ and $C^\omega$. In [@michael:79 Theorem 7.1], Michael extended this characterisation to arbitrary uniform ANR’s by showing that a metric space $(Y,\rho)$ is a uniform ANR if and only if it has the above Lefschetz property for arbitrary simplicial complexes $\Sigma$ rather than finite. Here, we are interested in another extension of the $LC^*$ property to arbitrary subsets of metric spaces. Namely, we shall say that a subset $S\subset Y$ of a metric space $(Y,\rho)$ is *uniformly $LC^*$* if for every $\varepsilon>0$ there exists $\eta(\varepsilon)>0$ such that if $\Sigma$ is a finite-dimensional simplicial complex and $K\subset \Sigma$ is a subcomplex containing the vertex set $\Sigma^0$, then each continuous map $g:|K|\to S$ with $\operatorname{diam}(g(|K\cap \sigma|))<\eta(\varepsilon)$ for every $\sigma\in \Sigma$, can be extended to a continuous map $f:|\Sigma|\to S$ such that $\operatorname{diam}(f(|\sigma|))<\varepsilon$, $\sigma\in \Sigma$. The following property of uniformly $LC^*$ sets was essentially established by Michael in [@michael:56b Lemma 11.1].
\[proposition-Near-sel-vgg:1\] Let $S\subset Y$ be a uniformly $LC^*$ subset of a metric space $(Y,\rho)$. Then for every $\varepsilon>0$ there exists $\gamma(\varepsilon)>0$ such that if $Z$ is a finite-dimensional paracompact space and $g:Z\to \mathbf{O}_{\gamma(\varepsilon)}(S)$ is a continuous map, then there exists a continuous map $h:Z\to S$ such that $h(z)\in \mathbf{O}_\varepsilon(g(z))$, for every $z\in Z$.
We repeat the proof of [@michael:56b Lemma 11.1]. Namely, let $\gamma(\varepsilon)=\frac14\eta\left(\frac{2\varepsilon}3\right)$, where $\eta(\varepsilon)\leq \varepsilon$ is as in the definition of uniformly $LC^*$ of $S$. Take a finite-dimensional paracompact space $Z$ and a continuous map $g:Z\to \mathbf{O}_{\gamma(\varepsilon)}(S)$. Then there exists a locally finite open cover $\mathscr{U}$ of $Z$ such that its nerve $\mathscr{N}(\mathscr{U})$ is finite-dimensional (see [@dowker:47 Theorem 3.5]) and $\operatorname{diam}(g(U))<\gamma(\varepsilon)$, for every $U\in \mathscr{U}$. Next, identifying the $0$-skeleton of $\mathscr{N}(\mathscr{U})$ with $\mathscr{U}$, define a (continuous) map $\ell:\mathscr{U}\to S$ such that $\ell(U)\in \mathbf{O}_{\gamma(\varepsilon)}(g(U))$, for every $U\in \mathscr{U}$. Hence, we also have that $$\label{eq:Near-sel-vgg:5}
\rho(\ell(U),g(z))<2\gamma(\varepsilon)\leq
\frac\varepsilon3,\quad \text{for every
$z\in U\in \mathscr{U}$.}$$ Thus, for $U,V\in \mathscr{U}$ with $U\cap V\neq {\varnothing}$, we get that $\rho(\ell(U),\ell(V))<4\gamma(\varepsilon)=
\eta\left(\frac{2\varepsilon}3\right)$. Since $\mathscr{N}(\mathscr{U})$ is finite-dimensional and $\eta\left(\frac{2\varepsilon}3\right)$ is as in the definition of uniformly $LC^*$ of $S$, the map $\ell$ can be extended to a continuous map $f:|\mathscr{N}(\mathscr{U})|\to S$ such that ${\operatorname{diam}(f(|\sigma|))<\frac{2\varepsilon}3}$, for every $\sigma\in \mathscr{N}(\mathscr{U})$. Finally, let $\pi:Z\to |\mathscr{N}(\mathscr{U})|$ be a canonical map and $h:Z\to S$ be the composite map $h=f\circ \pi$. If $z\in Z$, then $z\in \bigcap \sigma$ for some $\sigma\in
\mathscr{N}(\mathscr{U})$. Take any $U\in \sigma\subset
|\sigma|$. Since $\pi(z),U\in |\sigma|$ and $f(U)=\ell(U)$, it follows from that $$\rho(h(z),g(z))\leq \rho(f(\pi(z)),f(U))+ \rho(\ell(U),g(z))
<\frac{2\varepsilon}3 +\frac{\varepsilon}3
=\varepsilon.\qedhere$$
We may now apply Theorem \[theorem-Finite-C-v5:1\] to get the following further consequence.
\[corollary-Near-sel-vgg:2\] Let $(Y,\rho)$ be a metric space which is a uniform ANR, $X$ be a paracompact finite $C$-space and $\varphi:X\operatorname{\leadsto}Y$ be a $\rho$-continuous mapping such that each $\varphi(x)$, $x\in X$, is both $C^\omega$ and uniformly $LC^*$. Then $\varphi$ has a continuous $\varepsilon$-selection, for every continuous function $\varepsilon:X\to (0,+\infty)$.
According to Theorem \[theorem-Finite-C-v5:1\], it suffices to show that each $\varphi(x)$, $x\in X$, is uniformly $UV^\omega$. To this end, take a point $p\in X$, and let $\gamma(\varepsilon)$ be as in Proposition \[proposition-Near-sel-vgg:1\]. Since $(Y,\rho)$ is a uniform ANR, as in the proof of Corollary \[corollary-Near-sel-vgg:1\], to each $\varepsilon>0$ there exists $\eta(\varepsilon)>0$ such that any two $\eta(\varepsilon)$-close continuous maps from an arbitrary space to $Y$ are $\varepsilon$-homotopic. Finally, take $\delta(\varepsilon)= \gamma(\eta(\varepsilon))$, and let us show that this works. So, let $g:|\Sigma|\to \mathbf{O}_{\delta(\varepsilon)}(\varphi(p))$ be a continuous map for some finite-dimensional simplicial complex $\Sigma$. Then by Proposition \[proposition-Near-sel-vgg:1\], there exists a continuous map ${h:|\Sigma|\to \varphi(p)}$ such that $\rho(h(z),g(z))<\eta(\varepsilon)$, for all $z\in
|\Sigma|$. Accordingly, $g$ and $h$ are $\eta(\varepsilon)$-close and by the choice of $\eta(\varepsilon)$, the maps $g$ and $h$ are $\varepsilon$-homotopic. On the other hand, by Proposition \[proposition-Finite-C-v2:1\], $h:|\Sigma|\to \varphi(p)$ is null-homotopic in $\varphi(p)$ because $\varphi(p)$ is aspherical. This implies that $g$ is null-homotopic in $\mathbf{O}_{\varepsilon}(\varphi(p))$, which shows that $\varphi(p)$ is uniformly $UV^\omega$.
[10]{}
D. Addis and J. Gresham, *A class of infinite-dimensional spaces. [I]{}. [D]{}imension theory and [A]{}lexandroff’s problem*, Fund. Math. **101** (1978), no. 3, 195–205.
P. Borst, *Some remarks concerning [C]{}-spaces*, Topology Appl. **154** (2007), no. 3, 665–674.
M. Choban, E. Mihaylova, and S. Nedev, *On selections and classes of spaces*, Topology Appl. **155** (2008), 797–804.
C. H. Dowker, *Mappings theorems for non-compact spaces*, Amer. J. Math. **69** (1947), 200–242.
J. Dugundji, *Absolute neighborhood retracts and local connectedness in arbitrary metric spaces*, Comp. Math. **13** (1958), 229–246.
V. V. Fedorchuk, *Weakly infinite-dimensional spaces*, Uspekhi Mat. Nauk **62** (2007), no. 2(374), 109–164; translation in Russian Math. Surveys **62** (2007), no. 2, 323–374.
V. Gutev, *Constructing selections stepwise over skeletons of nerves of covers*, Serdica Math. J. **44** (2018), 137–154.
V. Gutev and V. Valov, *Continuous selections and [$C$]{}-spaces*, Proc. Amer. Math. Soc. **130** (2002), 233–242.
W. Haver, *A covering property for metric spaces*, Topology [C]{}onference ([V]{}irginia [P]{}olytech. [I]{}nst. and [S]{}tate [U]{}niv., [B]{}lacksburg, [V]{}a., 1973), Springer, Berlin, 1974, pp. 108–113. Lecture Notes in Math., Vol. 375.
, *A near-selection theorem*, General Topology Appl. **9** (1978), no. 2, 117–124.
J. Isbell, *Uniform neighborhood retracts*, Pacific J. Math. **11** (1961), 609–648.
, *Uniform spaces*, Amer. Math. Soc. Providence, 1964.
T. Kimura and C. Komoda, *Spaces having a compactification which is a [$C$]{}-space*, Topology Appl. **143** (2004), no. 1-3, 87–92.
S. Lefschetz, *Topics in [T]{}opology*, Annals of Mathematics Studies, no. 10, Princeton University Press, Princeton, N. J., 1942.
E. Michael, *Continuous selections [II]{}*, Ann. of Math. **64** (1956), 562–580.
, *Uniform [AR]{}’s and [ANR]{}’s*, Composition Math. **39** (1979), no. 2, 129–139.
R. Pol, *A weakly infinite-dimensional compactum which is not countable-dimensional*, Proc. Amer. Math. Soc. **82** (1981), no. 4, 634–636.
K. Sakai, *Geometric aspects of general topology*, Springer Monographs in Mathematics, Springer, Tokyo, 2013.
H. Toruńczyk, *Absolute retracts as factors of normed linear spaces*, Fund. Math. **86** (1974), 53–67.
, *On factors of the [H]{}ilbert cube and metric properties of retractions*, Bull. Acad. Polon. Sci. Sér. Math. Astronom. Phys. **24** (1976), no. 9, 757–765.
V. Valov, *Continuous selections and finite [$C$]{}-spaces*, Set-Valued Anal. **10** (2002), no. 1, 37–51.
J. H. C. Whitehead, *Simplicial [S]{}paces, [N]{}uclei and m-[G]{}roups*, Proc. London Math. Soc. (2) **45** (1939), no. 4, 243–327.
, *Combinatorial homotopy. [I]{}*, Bull. Amer. Math. Soc. **55** (1949), 213–245.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the problem of coupling a stochastic block model with a planted bisection to a uniform random graph having the same average degree. Focusing on the regime where the average degree is a constant relative to the number of vertices $n$, we show that the distance to which the models can be coupled undergoes a phase transition from $O(\sqrt{n})$ to $\Omega(n)$ as the planted bisection in the block model varies. This settles half of a conjecture of Bollobás and Riordan and has some implications for sparse graph limit theory. In particular, for certain ranges of parameters, a block model and the corresponding uniform model produce samples which must converge to the same limit point. This implies that any notion of convergence for sequences of graphs with $\Theta(n)$ edges which allows for samples from a limit object to converge back to the limit itself must identify these models.'
author:
- |
James Hirst\
Department of Mathematics, MIT
bibliography:
- 'couplings.bib'
title: Optimal couplings between sparse block models
---
Introduction
============
The idea of putting a metric on the space of graphs has generated a significant amount of fruitful research in the last decade. In particular, this has led to an extremely rich theory of dense graph limits, a line of work stemming from [@BCLSV08] which remains active today. Here, the word *dense* refers to the scaling in the number of edges: a dense sequence of graphs with $n$ vertices should have $\Theta(n^2)$ edges, and any set of $o(n^2)$ edges doesn’t contribute to the limit of the sequence (they become measure 0 sets in the limit). The metric that best captures similarity between dense graphs is the *cut distance*, which, informally, is a measure of how well one can align two graphs to have a small number of edges across every cut. Closeness in cut distance turns out to be equivalent to a number of natural metrics including subgraph counts and free energies. For an excellent exposition of these results, we refer the reader to [@Lov12].
Modulo some technical conditions, a suitably re-normalized version of the cut distance is also a valid and interesting metric for graphs with with $o(n^2)$ edges. In [@BR09], the authors show that for graph sequences in which every $o(n)$ vertices of a graph span $o(n)$ edges (forbidding graphs like a clique on $o(n)$ vertices), many of the nice properties of the cut distance continue to hold. The more recent work of [@BCCZ19] uses a relaxation of this condition based on boundedness in an $\ell_p$ sense (ordinary graphs can be thought of as bounded in $\ell_\infty$). However, in both of these results is an implicit requirement that graphs have unbounded degrees as $n$ grows, i.e., $\omega(n)$ edges.
An attempt at treating the $\Theta(n)$ sparsity regime is given in [@BR11]. One of the main difficulties here is that the random graph $G(n,c/n)$ concentrates much more weakly than usual. Indeed, the degree of a vertex has mean and variance of the same order $c$, and many key quantities have tails which decay only exponentially in $n$, rather than super-exponentially. As such, basic results which follow from a union bound for denser graphs break down. In this setting, the cut distance turns out to be much too strong of a metric: even a sequence of i.i.d. graphs from $G(n,c/n)$ fails to be a Cauchy sequence. On the other hand, local metrics based on subgraph counts are in some sense too weak, as they fail to distinguish between a random graph and a random bipartite graph of the same average degree.
The most basic object in dense graph limit theory is the graphon, a symmetric and measurable function $W:[0,1]^2 \to [0,1]$. Given a graphon $W$, there is a natural way to sample an $n$-vertex graph $G_n$: choose $x_1,\ldots,x_n$ uniformly from $[0,1]$ and independently place each edge $(i,j)$ with probability $W(x_i,x_j)$. This produces a graph with $\Theta(n^2)$ edges (assuming $W \neq 0$), and such models are well studied. For instance a basic result from graph limit theory says that the sequence $\{G_n\}$ converges to $W$ (under the cut distance) almost surely.
With a slight modification, we can use this construction to sample graphs with $\Theta(n)$ edges. To avoid confusion, we will call a symmetric and measurable function $\kappa:[0,1]^2 \to \R$ a *kernel* to distinguish it from a graphon which typically maps to $[0,1]$. Given a bounded kernel $\kappa:[0,1]^2 \to \R$, again take $x_1,\ldots,x_n$ as i.i.d. $U(0,1)$ but now place edge $(i,j)$ with probability $\kappa(x_i,x_j)/n$ (projecting this value into $[0,1]$ as required). Denote this random graph model by $G(n,\kappa)$.
Can we still say in some sense that the (sparsely) sampled sequence $\{G_n\}$ converges to $\kappa$? We will see momentarily that there is a natural barrier to this which is not present in the dense case. Given $n$-vertex graphs $G$ and $H$ on the same vertex set, we define the *edit distance* between $G$ and $H$ to be the minimum number of edges which need to be changed to turn $H$ into an isomorphic copy of $G$, and we write $$\d_1(G,H) = \min_{G' \cong G} \left|E(G') \triangle E(H) \right|.$$ As in the dense case, it is reasonable to assume that any notion of convergence for sparse graphs should be invariant up to edit distance $o(n)$. Following $\cite{BR11}$, we will call the models $G(n,\kappa_1)$ and $G(n,\kappa_2)$ (or sometimes just $\kappa_1$ and $\kappa_2$) *essentially equivalent* if there is a coupling $\mu$ between $G(n,\kappa_1)$ and $G(n,\kappa_2)$ such that $\E_\mu \d_1(G,H) = o(n)$, where $(G,H) \sim \mu$. We will call them *essentially different* otherwise. Returning to our question above, if we can find $\kappa_1 \neq \kappa_2$ which are essentially equivalent, then samples from $G(n,\kappa_1)$ and $G(n,\kappa_2)$ must converge in some sense to both $\kappa_1$ and $\kappa_2$.
Consider the following example. Let $\kappa_1$ be identically $c$, and take $\kappa_2$ to have value $c+\delta$ on $[0,1/2]\times[1/2,1]$ and $c-\delta$ elsewhere. Henceforth we will set $\Q = G(n,\kappa_1)$ and $\P = G(n,\kappa_2)$, suppressing the dependence on $n$, $c$, and $\delta$ wherever it avoids confusion. In [@BR11], the authors give a simple example to show that when $c \geq \delta$ and $0 \leq \delta \leq 1$, $\P$ and $\Q$ are in fact essentially equivalent. The following extension of this is half of Conjecture 6.3 from [@BR11].
\[conj:br\] Let $0 \leq \delta \leq c$. If $\delta < \sqrt{c}$, then $\P$ and $\Q$ are essentially equivalent.
Note that here the underlying kernels are truly different: One cannot compose $\kappa_1$ and $\kappa_2$ with (potentially different) measure preserving maps to make them equal almost everywhere. This makes the idea that $\P$ and $\Q$ can none the less be essentially equivalent somewhat surprising at first glace.
Our main result is a proof of Conjecture \[conj:br\]. In fact, we prove an even stronger result: when $\delta < \sqrt{c}$, the two models can be coupled to an expected edit distance of $O(\sqrt{n})$.
\[thm:ub\] Let $0 \leq \delta \leq c$ and denote by $\Pi(\P,\Q)$ the space of coupling measures between $\P$ and $\Q$. If $\delta < \sqrt{c}$, then there exists a coupling $\mu \in \Pi(\P_n,\Q_n)$ such that $$\E_{\mu}\left[\d_1(G,H)\right] = O(\sqrt{n}).$$
We also give a lower bound on the expected edit distance under any coupling.
\[thm:lb\] Let $0 \leq \delta \leq c$. For any coupling $\mu \in \Pi(\P,\Q)$, $$\E_{\mu}\left[\d_1(G,H)\right] = \Omega\left(\frac{c^{\log(n)^{1/3}}}{\log(n)^{1/3}}\right).$$
Preliminaries
=============
Optimal transport
-----------------
The proof strategy for both Theorem \[thm:ub\] and Theorem \[thm:lb\] is to obtain bounds on the quantity $$\inf_\mu \E_{\mu}\left[\d_1(G,H)\right],$$ where the infimum is taken over $\mu \in \Pi(\P,\Q)$. This is actually a well known type of problem called an optimal transport problem (for the cost $d_1$) and as such we will make use of a number of tools from the area. For an excellent survey of optimal transport, see [@San15]. The main item we will make use of is the following duality formula.
\[prop:dual\] We have the equality $$\label{eq:dual}
\inf_\mu \E_\mu \lb \d_1(G,H) \rb = \sup_f \left| \E_P f - \E_Q f \right|,$$ where $\mu \in \Pi(\P,\Q)$, $(G,H) \sim \mu$, and $f$ is 1-Lipschitz in the edit distance.
This is in fact a special case of Monge-Kantorovich duality for cost functions which satisfy the triangle inequality. Note that Proposition \[prop:dual\] immediately implies the useful inequality $$\inf_\mu \E_\mu \lb \d_1(G,H) \rb \geq \sup_f \left| \E_P f - \E_Q f \right|,$$ which is often referred to as *weak duality*.
It is not hard to see that the infimum in is always attained: the space $\Pi(\P,\Q)$ is compact for the weak convergence in duality with the space of continuous functions. Hence, by the definition of weak convergence, the functional sending $\mu$ to $\E_\mu \lb d_1(G,H)\rb$ is continuous. This is enough to guarantee the existence of a minimizer $\mu^*$ (Weierstrass). As it turns out, the supremum is also attained, but the proof is somewhat more involved (see Section 1.6 in [@San15]).
\[prop:inf\] There exists a coupling $\mu^* \in \Pi(\P,\Q)$ and a 1-Lipschitz function $f^*$ such that $$\E_{\mu^*} \lb \d_1(G^*,H^*) \rb = \left| \E_P f^* - \E_Q f^* \right|,$$ where $(G^*,H^*) \sim \mu^*$.
Stochastic block models
-----------------------
The study of sparse random graphs has also seen some attention from a very different group, motivated by the problem of detecting hidden communities planted in a random model. The stochastic block model for two balanced communities is usually defined as follows. Take parameters $a > b > 0$ (for a ferromagnetic block model) and assign to each vertex $i$ of an $n$ vertex graph a random spin $\sigma_i \in \{-1,1\}$. Then, connect edges $i$ and $j$ with probability $a/n$ if $\sigma_i \sigma_j = 1$, and with probability $b/n$ otherwise, independently for every edge. Notice that samples from the stochastic block model are identical to samples from $\P$, with $a = c+\delta$ and $b = c-\delta$.
Much of the previous work on such models is algorithmic, but some recent theoretical results have been obtained which are extremely relevant to the problem at hand. Of particular interest is [@MNS15], in which the authors settle a conjecture from [@DKMZ11] and give an information-theoretic lower bound for the detection problem. This lower bound comes in the form of a contiguity result.
\[prop:cont\] When $\delta < \sqrt{c}$, the models $\P_n$ and $\Q_n$ are mutually contiguous, i.e., for a sequence of events $A_n$ we have $$\P_n(A_n) \to 0 \iff \Q_n(A_n) \to 0.$$
To be precise, contiguity is a notion that should be applied to sequences of probabilities, and so when we say the models $\P_n$ and $\Q_n$ are mutually contiguous, we really mean that the sequences $\{\P_n\}$ and $\{\Q_n\}$ are.
An easy but important corollary which shall be our main point of contact with Theorem \[prop:cont\] is that it is impossible to consistently distinguish between samples from $\P$ and $\Q$ below the contiguity threshold.
\[cor:est\] Generate a random pair $(\sigma,G)$, where $\sigma \in \{0,1\}$ and $G$ is an $n$-vertex graph as follows. Take $\sigma \sim \ber(1/2)$ and sample $G$ from $\P$ if $\sigma = 1$, sampling $G$ from $\Q$ otherwise. When $\delta < \sqrt{c}$, there is no estimator $\pi$ taking an $n$-vertex graph to $\{0,1\}$ such that $$\label{eq:est}
\Pr\lb \pi(G) = \sigma \rb = 1 - o(1).$$
Given such a $\pi$, take $A_n = \{G: \pi(G) = 1\}$. Then $\P_n(A_n) \to 1$ while $\Q_n(A_n) \to 0$.
We often call an estimator $\pi$ satisfying an estimator for the block model or simply an estimator for $\P$. It is known that such estimators exist for all $\delta > \sqrt{c}$ and can even be efficiently computed (see [@MNS15] and [@MNS18]).
Proofs
======
At a high level, the proof strategy for Theorem \[thm:ub\] is to use Monge-Kantorovich duality along with an upper bound for $$\sup_f \left| \E_P f - \E_Q f \right|.$$ If $\left| \E_P f - \E_Q f \right|$ is too large for some $f$, then $f$ can be used as an estimator for $\P$ by thresholding. In order to make this work, we need to know that $f$ is sufficiently concentrated around its mean.
One can try to apply Azuma-Hoeffding using the vertex exposure martingale, but the martingale differences can only be bounded by the maximum degree of the graph which is a bit less than $\log(n)$ under both $\P$ and $\Q$. This gives a tail bound of the form $$\Pr \lb|f - \E f| > t \rb < 2\exp\left(-\frac{2t^2}{n\log(n)^2}\right)$$ at best, losing a $\log(n)^2$ factor in the variance. Rather than looking for tails with the correct variance, we will instead use the following inequality which bounds the variance directly.
\[prop:es\] Let $X_1,\ldots,X_n$ be independent random variables and set $X = (X_1,\ldots,X_n)$. Denote by $X^{(i)}$ the random variable obtained from $X$ by re-sampling $X_i$. Then we have $$\var(f(X)) \leq \frac{1}{2}\sum_{i=1}^n\E\left[\left(f(X) - f(X^{(i)})\right)^2\right].$$
With Efron-Stein, we can prove concentration for Lipschitz functions for samples from any bounded kernel.
\[lem:conc\] Let $\kappa$ be a bounded kernel with $d = \sup \kappa$, take $G_n \sim G(n,\kappa)$, and let $f: \Gc_n \to \R$ be 1-Lipschitz in the edit distance. Then we have $$\var(f(G_n)) = O(n).$$
The proof is a direct application of Efron-Stein. With $m = \binom{n}{2}$, $f(G_n)$ is a function of $n + m$ independent $U(0,1)$ random variables, $n$ type (or vertex) variables and $m$ edge variables. Notice that for a type variable $i$, $f(X) - f(X^{(i)})$ is stochastically dominated by a $\bin(n,d/n)$ random variable since $f$ is Lipschitz. So we can bound the marginal variance as $$\E\left[\left(f(X) - f(X^{(i)})\right)^2\right] \leq \var(\bin(n,d/n)) = d\left(1-\frac{d}{n}\right) \leq d.$$ Similarly for an edge variable $j$, $f(X) - f(X^{(j)})$ is stochastically dominated by a $\ber(d/n)$ random variable which gives $$\E\left[\left(f(X) - f(X^{(j)})\right)^2\right] \leq \var(\ber(d/n)) = \frac{d}{n}\left(1-\frac{d}{n}\right) \leq \frac{d}{n}.$$ Putting these together yields $$\var(f(G_n)) \leq \frac{1}{2}\left(dn + \frac{d}{n}m\right) \leq dn.$$
We’re now ready to prove Theorem \[thm:ub\].
Suppose that in we have $$\sup_f\left| \E_\P f - \E_\Q f \right| = \omega(\sqrt{n}).$$ By Proposition \[prop:inf\] there exists a 1-Lipschitz function $f$ such that $$\left| \E_\P f - \E_\Q f \right| = \omega(\sqrt{n}) = \alpha(n)\sqrt{n},$$ where $\alpha(n)$ is a function with $\lim_{n\to\infty}\alpha(n) = \infty$. Using Lemma \[lem:conc\] and Chebyshev’s inequality we have $$\label{eq:conc}
\pr \left[ |f - \E f| > \sqrt{\alpha(n)}\sqrt{(c+
\delta)n}\right] \leq \frac{1}{\alpha(n)} \to 0$$ for both $\P$ and $\Q$. Now, one of the terms $\E_\P f$ and $\E_\Q f$ must be at least the difference $\alpha(n)\sqrt{n}$. For every $n$, define the estimator $\pi$ as follows. If $E_{\P} f > \E_{\Q} f$ then let $$\pi(G) = \left\{ \begin{array}{cc}
1 & \text{ if} \ f(G) \geq \E_{\P} f - \frac{\alpha(n)\sqrt{n}}{2} \\
0 & \text{otherwise}
\end{array}\right..$$ Similarly if $\E_{\Q} f > \E_{\P} f$ let $$\pi(G) = \left\{ \begin{array}{cc}
0 & \text{ if} \ f(G) \geq \E_{\Q} f - \frac{\alpha(n)\sqrt{n}}{2} \\
1 & \text{otherwise}
\end{array}\right..$$ It follows from that $\pi$ is an estimator for $\P$ which is impossible by Corollary \[cor:est\].
Returning to we get the upper bound $$\inf_\mu \E_\mu \lb \d_1(G,H) \rb = O(\sqrt{n}).$$ Again referring to Proposition \[prop:inf\], there exists a coupling $\mu\in \Pi(\P,\Q)$ such that $$\E_\mu \lb \d_1(G,H) \rb = O(\sqrt{n}).$$
We should note that it is important to make the distinction between $\E_\P f > \E_\Q f$ and $\E_\Q f > \E_\P f$ for every $n$, as even though $f$ concentrates around its mean for a fixed $n$, the values of $\E_\P f$ and $\E_\Q f$ might not be converging as $n$ grows.
Let us now turn to the proof of Theorem \[thm:lb\].
From weak duality, for any 1-Lipschitz function $f$ we have the inequality $$\inf_\mu \E_\mu \lb \d_1(G,H) \rb \geq \left| \E_\P f - \E_\Q f \right|.$$ Given a graph $G$ and a parameter $k\in \N$, define $Y_k$ to be the maximum number of $k$-cycles in a disjoint cycle packing of $G$. Set $f(G) = Y_k$, noting that $f$ is certainly 1-Lipschitz, as an edge can contribute to a single cycle in any disjoint packing.
To remains to control $\E_\P f$ and $\E_\Q f$. To this end, define $X_k$ to be simply the number of (potentially overlapping) $k$-cycles in $G$, counted as ordered sets of $k$ vertices, up to automorphism. Here, we have $$\E_\Q X_k = \frac{n(n-1)\cdots(n-k+1)}{2k}\left(\frac{c}{n}\right)^k = \frac{c^k}{2k} + o(1),$$ where the second equality holds for $k = o(\sqrt{n})$. Indeed, there are $\sim n^k/2k$ ways to choose the vertices along with an ordering for the cycle modulo automorphism, and a probability $(c/n)^k$ of the edges appearing in sequence. Extending this result to $\P$, it is shown in [@MNS15] (Section 3) that $$\E_\P X_k = \frac{c^k + \delta^k}{2k} + o(1),$$ again for $k = o(\sqrt{n})$.
Consider the set $\Hc$ of all graphs $H$ which can be obtained as the union of two distinct, non-disjoint $k$-cycles. Note that every graph $H \in \Hc$ has $m \leq 2k$ vertices and at least $m+1$ edges, since it cannot be a cycle itself. For every such $H$, define $X_H$ to be the number of injective homomorphisms from $H$ into $G$ . We can easily compute $$\E_\Q X_H \leq n^m \left(\frac{c}{n}\right)^{m+1} \leq \frac{c^{2k+1}}{n}.$$
Define $Z_k$ by $$Z_k = \sum_{H \in \Hc} X_H.$$ Relating $X_k$ and $Y_k$ through $Z_k$, we have the inequalities $$X_k \geq Y_k \geq X_k - (2k)^k Z_k,$$ since every embedding of a graph $H \in \Hc$ into $G$ contributes at most $(2k)^k$ $k$-cycles to $X_k$, and removing all such edges yields a graph with $X_k = Y_k$. A graph with $2k$ vertices has at most $(2k)^2/2 = 2k^2$ edges so, counting over all possible labelled graphs, we get $|\Hc| \leq 2^{2k^2}$, where graphs on $<2k$ vertices are identified with graphs on $2k$ vertices with isolated vertices. This gives $$\E_\Q Z_k \leq \frac{2^{2k^2}c^{2k+1}}{n}.$$ Taking $k = \log(n)^{1/3}$ (in fact, any $k = \log(n)^{\alpha}$, $\alpha < 1/2$ will work), the numerator is $o(n)$ and so $\E_\Q Z_k = o(1)$. Since the number of copies of any graph in $\P$ is stochastically dominated by the number of copies in a $G(n,(c+\delta)/n)$, $\E_\P Z_k = o(1)$ as well. Putting everything together, we end up with $$\left| \E_\P f - \E_\Q f \right| \geq \frac{\delta^{k}}{2k} - o(1) = \Omega\left(\frac{c^{\log(n)^{1/3}}}{\log(n)^{1/3}}\right).$$
As noted, taking $k = \log(n)^{\alpha}$ for any $\alpha < 1/2$ improves the bound in Theorem \[thm:lb\] to $$\frac{c^{\log(n)^{\alpha}}}{\log(n)^{\alpha}},$$ but the value of this function remains somewhere between $\plog(n)$ and $\poly(n)$ for all $\alpha < 1/2$ and so this is not a particularly meaningful improvement.
Concluding remarks {#sec:rem}
==================
As mentioned, Conjecture \[conj:br\] is only half of the conjecture made in [@BR11]. In addition, the authors conjecture that for $\delta > \sqrt{c}$ and $c > 1$ (to ensure a giant component), the models $\P$ and $\Q$ are essentially different. This means that it should be impossible to couple $\P$ and $\Q$ to an expected edit distance of $o(n)$. Now if we allow $\delta > A\sqrt{c}$ for some sufficiently large constant $A$, then this is certainly the case. This is because the minimum bisection in a graph $G \sim \Q$ is of the order $cn/4 - \Theta(\sqrt{c}n)$ with high probability, whereas $G \sim \P$ almost always has a bisection with $cn/4 - \Theta(\delta n/4)$ edges. Plugging the minimum bisection into gives the desired $\Theta(n)$ lower bound.
It is not clear at all that we should be able to take $A = 1$ in the above. One might hope that for $\delta >\sqrt{c}$, the minimum bisections of $\P$ and $\Q$ would differ by $\Theta(n)$. One of the main factors mitigating progress in this area is the complexity of even computing the minimum bisection of a graph $G \sim \Q$. Writing $\m(G)$ for the value of the minimum bisection of $G$, the authors of [@DMS17] recently determined the second order growth of $m(G)$ through the formula $$\label{eq:minb}
\E_\Q\lb \m(G)\rb = n\left(\frac{c}{4} + \frac{\Pn^*}{2}\sqrt{c} + o(\sqrt{c})\right) + o(n).$$ The constant $\Pn^* \approx 0.7632$ here is the ground state energy of the Sherrington-Kirkpatrick spin glass model, and has no known simple closed form. A similar formula for $\E_\P\lb \m(G) \rb$ is given in [@Sen18] with an even more mysterious constant $\Cn^*$. It would be a significant breakthrough to pinpoint the value of $\delta$ for which the minimum bisections in $\P$ and $\Q$ transition from having a difference $O(\sqrt{n})$ to $\Theta(n)$. It could even be that two separate phase transitions occur.
The work of [@JMRT16] uses some strong heuristic methods from statistical physics to make predictions in this direction. Although non-rigorous, these heuristics (including the so-called cavity method) have predicted a number of results which have since been proven rigorously (like Theorem \[prop:cont\] above). They study an SDP relaxation of minimum bisection and predict that it detects communities down to $\delta = \sqrt{c}(1 + (8c)^{-1} + O(c^{-2}))$. This lower bound is at least partially vindicated by a matching upper bound from $\cite{MS16}$, who show that the SDP based estimator in fact does detect communities at $\delta = \sqrt{c} + \epsilon$ for any $\epsilon$, provided $c = c(\epsilon)$ can be taken sufficiently large. It seems as though minimum bisection should not do any worse than the SDP relaxation, which would allow us to take $A = 1 + o_c(1)$ above, but this is not a proof by any means.
Ignoring the question of bisections, it may well be that the models $\P$ and $\Q$ do in fact become essentially different at the contiguity threshold $\delta = \sqrt{c}$. Our choice of $f$ in the proof of Theorem \[thm:lb\] was based on the simple fact that $\P$ is biased towards having more $k$-cycles than $\Q$ on average. Taking $k$ larger gives a better lower bound, but one cannot take $k$ past $o(\sqrt{\log(n)})$ without overlapping cycles contributing to the expectation. To normalize $f$ into a Lipschitz function would then require an analysis of this overlap and seems difficult. A radically different choice of $f$ may be able to give improved lower bounds. For $\delta < \sqrt{c}$, it seems as though $\Theta(\sqrt{n})$ should be the correct answer: If not, then the minimum bisections in $\P$ and $\Q$ are actually mixed to their second order terms, which would be somewhat surprising.
As a final remark, we should note that the notion of being essentially different is robust under $o(n)$ adversarial edits to our samples. If $\P$ and $\Q$ become essentially different when $\delta > \sqrt{c}$, then Monge-Kantorovich duality along with Lemma \[lem:conc\] implies the existence of a *robust* estimator for $\P$. As far as we are aware, no estimator which achieves the $\delta > \sqrt{c}$ threshold is known to be robust to $o(n)$ edits. For instance, one can remove all cycles of length $k = o(\log(n))$ with $o(n)$ edits, breaking any current cycle-based estimators which only count short cycles.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We develop a method to study coupled dynamics of gauge-invariant variables, constructed out of metric and gauge field fluctuations on the background of a AdS$_5$ Reissner-Nordström black brane. Using this method, we compute the numerical spectrum of quasinormal modes associated with fluctuations of spin 0, 1 and 2, non-perturbatively in $\mu/T$. We also analytically compute the spectrum of hydrodynamic excitations in the small chemical potential limit. Then, by studying the spectral curve at complex momenta in every spin channel, we numerically find points at which hydrodynamic and non-hydrodynamic poles collide. We discuss the relation between such collision points and the convergence radius of the hydrodynamic derivative expansion. Specifically in the spin 0 channel, we find that within the range $1.1\lesssim \mu/T\lesssim 2$, the radius of convergence of the hydrodynamic sound mode is set by the absolute value of the complex momentum corresponding to the point at which the sound pole collides with the hydrodynamic diffusion pole. It shows that in holographic systems at finite chemical potential, the convergence of the hydrodynamic derivative expansion in the mentioned range is fully controlled by hydrodynamic information. As the last result, we explicitly show that the relevant information about quantum chaos in our system can be extracted from the pole-skipping points of energy density response function. We find a threshold value for $\mu/T$, lower than which the pole-skipping points can be computed perturbatively in a derivative expansion.'
author:
- |
Navid Abbasi$^{a}$[^1], Sara Tahery$^{b}$[^2]\
\
\
\
\
title: '[Complexified quasinormal modes and the pole-skipping in a holographic system at finite chemical potential ]{}'
---
Introduction
============
Low energy dynamics of many body systems near thermal equilibrium can be effectively described in terms of IR variables. These variables themselves are of two types. First are those corresponding to fast thermalizing excitations. These are in correspondence with non-conserved quantities and swallowed by thermal vacuum, very soon after being excited. Second are those corresponding to conserved quantities whose excitations cannot be locally thermalized. They are relaxed just through the transport in long wavelength compared to local thermalization time scale.
In fact the first-type degrees of freedom are in local equilibrium defined by the second-type ones. Hydrodynamics is an effective theory which describes the dynamics of the second-type modes, namely slow modes, in a many-body system. Hydrodynamic equations are then the conservation equations of stress tensor $T^{\mu\nu}$ and charge current $J^{\mu}$ in the system. The main idea of hydrodynamics is that, in local equilibrium, each of these quantities are given in a perturbative expansion of local thermodynamic variables and their derivatives. Such perturbative expressions are called the hydrodynamic constitutive relations. Using the symmetry considerations, general form of the constitutive relations can be fixed up to a set of unknown coefficients, which are referred to as the transport coefficients. While hydrodynamics is a universal regime in thermal systems, the associated transport coefficients depend on the underlying microscopic theory by which the system is described.
Considering hydrodynamics as a classical field theory, one can find simply the response functions of the conserved quantities (see refs. [@Kovtun:2012rj; @Glorioso:2018wxw]); the corresponding poles are then the so-called *hydrodynamic modes*. Hydrodynamic modes are the longest lived modes around thermal equilibrium in the system and are gapless, i.e. $\omega(\textbf{q}\rightarrow 0)=0$. In contrast, the excitations corresponding to the first-type degrees of freedom, mentioned in the first paragraph, are short-lived in the sense that $\omega(\textbf{q}\rightarrow 0)\ne0$. They are called *non-hydrodynamic modes*.
Hydrodynamic modes can be in general found from the linearized hydrodynamic equations perturbatively, order by order in a derivative expansion (see ref. [@Kovtun:2012rj] for a comprehensive review). However, without knowing the microscopic of system, the result will be restricted to general perturbative expressions containing unknown transport coefficients. Moreover, even by having the microscopic theory, computing the transport coefficients in general is limited to lower orders in derivative expansion and also to the perturbative regime of the underlying field theory [@Jeon:1995zm].
When the microscopic theory has a holographic dual [@Maldacena:1997re; @Witten:1998qj], however, extracting information will be relatively straightforward not only in the hydrodynamic limit, but also beyond that. In fact holography allows to know the microscopic of certain strongly coupled field theories via studying their gravitational dual in a one-higher-dimensional AdS space time. In the last two decades, by analyzing perturbations around specific AdS black brane solutions, significant information about the strongly coupled $\mathcal{N}=4$ SYM plasma has been found: transport coefficients of the first order hydrodynamics specifically $\eta/s$ [@Policastro:2002se], conjecturing a lower bound for $\eta/s$ [@Kovtun:2004de], the shear viscosity of R-charged plasma [@Son:2006em; @Mas:2006dy], location of sound poles [@Policastro:2002tn], the leading ’t Hooft coupling correction to $\eta/s$ [@Buchel:2004di], the spectrum of quasinormal modes [@Kovtun:2005ev], the second order transport coefficients [@Baier:2007ix; @Bhattacharyya:2008jc], the ’t Hooft coupling correction to the second order transport coefficients [@Grozdanov:2014kva], anomalous transport coefficients [@Erdmenger:2008rm; @Banerjee:2008th], the third order transport coefficients for linear [@Bu:2014ena] and full non-linear [@Grozdanov:2015kqa] hydrodynamics and many other related issues.
At first sight, higher order terms in the hydrodynamic constitutive relations seem just to improve the accuracy of the derivative expansion. However, as it was found in refs. [@Heller:2013fn], large order behavior of the hydrodynamic expansion might have some information about non-hydrodynamic modes as well. By numerically computing energy density for a holographic boost-invariant flow, up to terms with 240 derivatives, authors of ref. [@Heller:2013fn] found the radius of convergence of the hydrodynamic derivative expansion to be zero. Let us emphasize that this result was associated with an expansion over the inverse proper time in position space. Then by using Borel resummation techniques, they found the frequency corresponding with the leading singularity in Borel-transformed hydrodynamic stress tensor to be exactly the same as frequency of the lowest-lying non-hydrodynamic mode, found in refs. [@Starinets:2002br; @Janik:2006gp] [^3].
In a recent paper [@Withers:2018srf], a different approach has been proposed to extract information from the large order behavior of the hydrodynamic derivative expansion. Instead of working with full non-linear hydrodynamics to study a highly symmetric flow in position space [@Heller:2013fn], the choice of ref. [@Withers:2018srf] is to work with linear hydrodynamics in the complex $q$ plane. The focus of this reference is to study the dispersion relation of the hydrodynamic shear mode at large orders in derivative expansion, in a holographic model in $2+1$ dimensions with finite chemical potential. Considering the multi-sheet structure of exact $\omega_{\text{shear}}(\textbf{q})$, the closest non-analytic points on the longest-lived sheet are found to be as $q=\pm i q_{*}$ [@Withers:2018srf], where $q=|\textbf{q}|$. The author of ref. [@Withers:2018srf] then argues that $q_{*}$ sets a finite radius of convergence for the hydrodynamic expansion of the shear mode [^4]. As explicitly mentioned in ref. [@Withers:2018srf], the physical origin of $q_*$ is the collision between a hydrodynamic and non-hydrodynamic mode on the imaginary $q$ axis.
Much more recently, it was shown that the same result could be found from analytic structure of spectral curves in classical hydrodynamics[^5]. In ref. [@Grozdanov:2019kge], the above-mentioned collision points were found from the associated spectral curves in a holographic neutral fluid in 3+1 dimensions, for both sound and shear hydrodynamic modes. In the mentioned reference, such collision points are called level-crossing points. In fact the theory of [@Grozdanov:2019kge; @Grozdanov:2019uhi] is the general theory for studying spectral curves and level-crossing.
Let us denote that in [@Withers:2018srf], the relation between $q_{*}$ and convergence radius of the shear mode has been numerically confirmed at a specific fixed value of $\mu/T$. To investigate how actually $q_{*}$ may vary with $\mu/T$ and also to study the convergence of other modes, in this paper we follow the issue in a holographic model at finite chemical potential in $3+1$ dimensions. In the gravity side we consider a AdS$_5$ Reissner-Nordström black brane. In the bulk of AdS$_5$, such black brane is identified with the temperature $T$ as well as a parameter $Q$, both related to $\mathcal{N}=4$ SYM boundary theory, where $Q$ is a monotonically increasing function of $\mu/T$.
The main difficulty comes from the fact that in our model parturbations of gravity and gauge field in the bulk are coupled. One may think of finding master fields and then deriving decoupled equations governing their dynamics [@Kodama:2003kk][^6]. But we choose to work with coupled equations! The reason is related to the numerical method that we want to adopt. We construct the generalized version of Frobenius expansion used in ref. [@Kovtun:2005ev] to find the quasinoraml modes associated with *coupled* differential equations. The advantage of working with coupled equations then is that it lets each of our results at $Q=0$ be comparable with some counterpart result within refs. [@Kovtun:2005ev] or [@Grozdanov:2019kge]. Let us denote that our numerical method works very well in the range $0\le Q\lesssim0.88$, or equivalently for $0\le \frac{\mu}{T} \lesssim 4$. However, in our plots we will demonstrate the results within the range $0\le Q\le0.8$.
We first construct gauge-invariant variables out of the bulk field perturbations. The perturbations, and consequently the gauge invariant variables, can be classified according to their transformation properties under $SO(2)$ group corresponding to the isotropy of the transverse plane perpendicular to $\textbf{q}$. This is actually the subject of . In we develop a new method to find the quasinormal modes of coupled differential equations in the bulk. In we will derive coupled equations governing dynamics of gauge-invariant variables in each of the spin channels, on the AdS$_5$ RN black brane background. Our equations can be regarded as non-trivial generalizations of decoupled equations associated with an AdS$_5$ Schwarzschild black brane accompanied by a probe gauge field [@Kovtun:2005ev].
In each of the spin channels, we find the spectrum of quasinormal modes by considering variations of both ${{\mathfrak{q}}}=q/(2\pi T)$ and $Q$. Spin 0 and spin 1 spectra, each turns out to be a superposition of two types of poles. In the spin 0 channel, poles are in correspondence with fluctuations of scalar components of stress tensor and charge current. The hydrodynamic modes associated with former fluctuations are the sound modes while that of related to the latter fluctuations is the diffusion mode. In the spin 1 channel, poles correspond to fluctuations of vector components of stress tensor and charge current. In this case, only one hydrodynamic mode does exist which is related to the stress tensor fluctuations. That is actually the shear mode. Finally, from the single dynamical equation of the spin 2 channel, we find poles corresponding to fluctuations of the tensor components of stress tensor. As expected, there is no any hydrodynamic mode in this channel.
The spectrum of quasinormal modes at finite chemical potential have been already computed in AdS$_4$ RN model [@Brattan:2010pq; @Edalati:2010hk; @Edalati:2010pn]. But in holographic models in 3+1 dimensions, well-known results are just limited to the spin 1 and spin 2 channels [@Maeda:2006by; @Janiszewski:2015ura]. To best of our knowledge our study is the first computation of quasinormal modes in the spin 0 channel of AdS$_5$ RN model. In addition, by analytically solving the coupled dynamical equations in the hydrodynamic limit, we find the spectrum of hydrodynamic modes as well. Again, this is the first analytic computation of the hydrodynamic modes from AdS$_5$ RN black brane. However it should be noted that we will find analytic solutions in the bulk perturbatively in $\mu/T$. We then use fluid/gravity [@Erdmenger:2008rm; @Banerjee:2008th] to confirm our results by explicit computation of the hydrodynamic modes at finite $\mu/T$.
In the second part of the paper, in , we study quasinormal modes at complex momenta. We find radii of convergence of the derivative expansions corresponding to dispersion relations of all three hydrodynamic modes, separately, within the range $0\le Q\le 0.8$. For the shear mode, the level crossing between hydrodynamic and non-hydrodynamic poles in the spin 1 channel sets the convergence radius of the dispersion relation. It turns out that the radius of convergence monotonically increases when $Q$ increases from $0$ to $0.8$.
Our surprising results are mostly related to the spin 0 channel. We motivate that for each of the sound and diffusion modes, one has to find the critical points of spectral curve lying on their own branch of Puiseux series [^7]. Doing so, for the diffusion case we show that within the whole range of $Q$, the level crossing happens between the diffusion pole and gapped poles associated with the charge current spectrum. But sound modes are found to collide with various types of poles, depending on the value of $\mu/T$. Their first collisions are actually with either gapped poles of the stress tesnor spectrum or gapped poles of the charge current spectrum or even with the diffusion pole! The latter is a novel aspect of level crossing phenomenon, specific to holographic systems at finite chemical potential [^8]. It can be regraded as a counterexample for the statement that finite radius of convergence of hydrodynamic derivative expansion is determined by the interplay between hydrodynamic and non-hydrodynamic modes. In other words, we numerically find that in a specific range of $\mu/T$, the convergence radius of derivative expansion is fully determined by hydrodynamic information.
In the last part of the paper, in , we study one another aspect of quasinormal modes. Following recent studies on the relation between hydrodynamics and quantum chaos in maximally chaotic systems [@Grozdanov:2017ajz; @Blake:2017ris; @Blake:2018leo], we will show that the pole-skipping point of energy density response function in our system precisely coincides with the chaos point of the system. This result is found without making any assumption about the value of $\mu/T$. Thus it provides a new support for the hydrodynamic origin of quantum chaos [@Blake:2017ris]. Then we focus on the spin 0 channel and numerically find the dispersion relation of sound and diffusion poles at purely imaginary momenta. For a typical value of $\mu/T$, we show that the above-mentioned pole-skipping points lie actually on the sound curve.
In the very last part, we discuss the possibility of finding the chaos point by using the derivative expansion. Exploiting the spin 0 channel results of , we will find that the chaos point does not always lie within the domain of validity of hydrodynamics. We find a critical value $Q_c$, beyond which the chaos points lie outside of convergence domain of the hydrodynamic derivative expansion. It simply means that for this range of $Q$, the pole-skipping point of energy density response function has to be found non-perturbativley.
Finally in , we briefly review our results and discuss possible follow-up directions.
Gauge invariant variables {#Gauge_section}
=========================
The back ground solution on which we would like to find the quasinormal modes can be written in general as $$\label{}
\begin{split}
ds^2&=\, a(r)\left(-f(r)dt^2+\frac{}{}\sum_{i=1}^{p}dx_i^2\right)+b(r)dr^2,\\
A&=\,\tilde{q} \,c(r)\, dt,
\end{split}$$ where $\tilde{q}=q_0/r_h^p$ is the electric charge density on the horizon. $r_h$ is the radial location of the horizon: $f(r_h)=0$. We take the fluctuations of metric and gauge field to be $h_{\mu \nu}(r)e^{- i \omega t+ i q z}$ and $a_{\mu}(r)e^{- i \omega t+ i q z}$, respectively, where $z=x^p$. We focus on $p=3$ case. We then find the specific combinations of perturbations which are invariant under the simultaneous general diffeomorphism, denoted by $\xi^{\mu}$, and gauge transformations $\phi$ in the bulk $$\label{diff_gauge}
\begin{split}
h_{\mu\nu}&\rightarrow h_{\mu\nu}-\nabla_{\mu}\xi_{\nu}-\nabla_{\nu}\xi_{\mu},\\
a_{\mu}& \rightarrow a_{\mu}-\partial_{\mu} \phi-\xi^{\lambda}\nabla_{\lambda}a_{\mu}-a_{\lambda}\nabla_{\mu}\xi^{\lambda}.
\end{split}$$ It is convenient to arrange perturbations $h_{\mu\nu}$ and $a_{\mu}$ according to representations of $SO(2)$ group associated with the two dimensional plane perpendicular to $q$ (see Appendix \[diff\_gauge\_App\]). We now start to find gauge invariant variables associated with different spin channels.
Spin $0$ channel {#spin-0-channel .unnumbered}
----------------
In this channel six perturbations $G_{i}\in\{a_t, a_z, h_{tt}, h_{tz}, h_{zz}, h\}$, where $h=\frac{1}{2}(h_{xx}+h_{yy})$, are coupled to each other. Any first order gauge invariant variable, denoted by $G_{\text{inv}}$, is naturally a linear combination of $G_i$’s: $$\label{G_inv}
G_{\text{inv}}=\sum_{i=1} ^{6}\alpha_i G_i.$$ On the other hand, there are in general four diffeomorphism and gauge parameter functions in this channel: $\zeta_{i}\in\{\phi, \xi_t, \xi_z, \xi_r\}$. The task is to find $a_i$’s such that $G_{\text{inv}}$ to be independent of $\zeta_i$’s. Under the transformations , $G_i$’s are transformed as $$\label{}
G_{\text{inv}} \rightarrow G_{\text{inv}}+\sum_{i=1} ^{4}\beta_i \zeta_i,$$ with $\beta_i$ being a linear combination of $\alpha_i$’s. Gauge invariance demands $\beta_i$’s must vanish. So we find four equations among six parameters $\alpha_i$. As a result, two of the parameters $\alpha_i$ remain free; it simply shows that in the sound channel we deal with two gauge invariant variables. Let us now find them in details.
From , the above mentioned two gauge invariant variables can be found either by $\bullet$ finding an appropriate combination of its first four lines; or by $\bullet$ finding an appropriate combination of its last three lines.
In the first case, we take $h_{zz}+ \ell_1 h_{tt}+ \ell_2 h_{tz}+ \ell_3 h$ and then demand its diff-gauge transformed be independent of $\xi_{t}$, $\xi_{z}$ and $\xi_{r}$. We find $$\label{}
\ell_1=\frac{q^2}{\omega^2}\,\,\,\,\,\ell_2=\frac{2 q }{\omega},\,\,\,\,\,\,\ell_3=\frac{1}{2}\left(-1+\frac{q^2 f(r)}{\omega^2}+\frac{q^2 a(r)f'(r)}{\omega^2 a'(r)}\right).$$ In the second case, we take $a_{z}+ e_1 a_{t}+ e_2 h$ combination. Again, demanding its diff-gauge transformed be independent of $\xi_r$ and $\xi_t$ fixes the coefficients $e_1$ and $e_2$: $$\label{}
e_1=\frac{q}{\omega},\,\,\,\,\,e_2=-\frac{q}{ 2\omega}\,\frac{ \tilde{q}\,c'(r)}{a'(r)}.$$ In summary, the gauge-invariant variables in the spin $0$ channel can be written as $$\label{Z_2}\boxed{
\begin{split}
Z_0&={{\mathfrak{q}}}^2 f H_{tt}+2\, {{\mathfrak{w}}}\,{{\mathfrak{q}}}H_{tz}+{{\mathfrak{w}}}^2 H_{zz}+{{\mathfrak{q}}}^2 f\left(1+\frac{a f'}{a' f}-\frac{{{\mathfrak{w}}}^2}{{{\mathfrak{q}}}^2 f}\right)H\\
E_{z}&={{\mathfrak{q}}}A_t+{{\mathfrak{w}}}A_z-{{\mathfrak{q}}}\left(\tilde{q}\,\frac{a \,c'}{a'\,c_h}\right)\,H
\end{split}}$$ where $A_{t}=a_{t}/c_h$, $A_{z}=a_{z}/c_h$, $H_{tt}=h_{tt}/a f$, $H_{tz}=h_{tz}/a$, $H_{ij}=h_{ij}/a \,\,(i,j\ne t)$ and $H=h/(p-1) a$. We have defined $c_h=c(r_h)$. In the above equations, $0$ denotes the spin. At this point one could continue to work with $h_{\mu\nu}$’s and construct $Z_0$ in terms of them. But an overall rescaling factor $1/a$, giving $Z_0$ in terms of $H_{\mu\nu}$’s, causes $Z_0$ to asymptote to a finite value on the boundary. Let us emphasize that $a_{\mu}$ fields have already finite limit at the boundary and the rescaling factor $c_{h}$ is nothing but a normaloization constant.
Spin $1$ channel {#spin-1-channel .unnumbered}
-----------------
From , it is obvious that $E_{\alpha}=i \omega A_{\alpha}$ ($\alpha= x, y $) are gauge invariant. In addition, two another gauge-invariant variables can be constructed by demanding $h_{z\alpha}+ \kappa h_{t \alpha}$ be independent of diff-gauge parameter functions. Therefore there are four gauge invariant variables in this channel that due to the $SO(2)$ symmetry, we need to only work with two of them $$\label{Z_1}\boxed{
\begin{split}
Z_{1}&= {{\mathfrak{w}}}\, H_{z x}\,+\,{{\mathfrak{q}}}\,H_{t x}\\
E_{x}&= {{\mathfrak{w}}}\,A_{x}
\end{split}}$$ where $A_{\alpha}=a_{\alpha}/c_h$ and $H_{t\alpha}=h_{t\alpha}/a$.
Spin $2$ channel {#spin-2-channel .unnumbered}
-----------------
From one easily finds that there are two gauge invariant variables in this channel $$\label{gauge_inv_Spin_2}\boxed{
\begin{split}
Z_{2}&= H_{xy}\\
W_{2}&=H_{xx}-H_{yy}
\end{split}}$$
Quasinormal modes from coupled differential equations {#method}
======================================================
As we will see later, except for the spin 2 channel, the dynamics in each of the other two channels is governed by a specific set of coupled differential equations. In this section we construct a method to find, in general, the associated quasinoaml modes from such coupled equations. Let us take the two gauge invariant variables in these channels as $Z(x^{\mu},u)$ and $E(x^{\mu},u)$. The radial coordinate $u$ is related to $r$ through $r\sim1/\sqrt{u}$ with the horizon located at $u=1$. We then make Fourier transformation in the boundary directions $$\label{}
(Z(x^{\mu},u), \,E(x^{\mu},u))=\,\int\frac{d^dk}{(2\pi)^d}\,e^{i k x}(Z(u),E(u)).$$ In each of the channels of spin 0 and 1, upon imposing the ingoing boundary condition at the horizon, the solution to Fourier component fields can be formally written as $$\label{ingoing_coupled_sol}
\begin{split}
Z(u)&= C_{Z}\,g_{Z}(u)+\,C_{E}\,y_{Z}(u),\\
E(u)&= C_{Z}\,g_{E}(u)+\,C_{E}\,y_{E}(u).
\end{split}$$ The pair of coefficients $C_Z$ and $C_E$ are fixed by the values of $Z(u)$ and $E(u)$ at the horizon, so $y_Z(1)=\,g_{E}(1)=0$. Moreover, both functions $y_Z(u)$ and $g_{E}(u)$ must vanish at $Q=0$, because we know at $Q=0$ the two equations in decouple.
In order to find associated quasinormal modes, we need to know the near boundary expansion of bulk fields. Considering the conformal dimensions of $Z$ and $E$, one finds $$\label{near_bdy_exp}
\begin{split}
u\rightarrow 0:\,\,\,\,\,\,\,\,\,\,\,\,\,Z(u)&= \left(c_1+\frac{}{}\cdots\right)+\,u^2\big(c_2+\cdots\big),\\
E(u)&=\left( c_3+\frac{}{}\cdots\right)+\,u\big(c_4+\cdots\big),
\end{split}$$ where $$\label{c_1_c_3}
c_1=\,C_{Z}\,g_{Z}(0)+\,C_{E}\,y_{Z}(0),\,\,\,\,\,\,\,\,c_3=\, C_{Z}\,g_{E}(0)+\,C_{E}\,y_{E}(0).$$ Using these solutions, the bilinear boundary bulk action takes the following form [^9]: $$\label{bilinear_bdy_action}
\begin{split}
S \sim &\lim_{u\rightarrow 0}\int_{\omega, \textbf{q}}\begin{pmatrix}
Z'_{k}(u)&E'_{k}(u)
\end{pmatrix}\begin{pmatrix}
A(u)&B(u)\\
B^*(u)& D(u)
\end{pmatrix}\begin{pmatrix}
Z_{-k}(u)\\
E_{-k}(u)
\end{pmatrix}+\,\text{contact\,\,terms},\\
\sim& \lim_{u\rightarrow 0}\int_{\omega, \textbf{q}}\begin{pmatrix}
Z_k(u)&E_k(u)
\end{pmatrix}\begin{pmatrix}
2u\frac{c_2}{c_1}A(u)&2u\frac{c_2}{c_1}B(u)\\
\frac{c_4}{c_3}B^*(u)& \frac{c_4}{c_3}D(u)
\end{pmatrix}\begin{pmatrix}
Z_{-k}(u)\\
E_{-k}(u)
\end{pmatrix}+\,\text{contact\,\,terms},
\end{split}$$ with $k=(\omega, \textbf{q})$ and “contact terms” are finite parts of the boundary counter-term. Needless to say that $2\,u\,A(u)$, 2$u\,B(u)$, $B^*(U)$ and $D(u)$ all go to finite values when $u\rightarrow 0$: $$\label{limits}
\lim_{u\rightarrow 0}2u\,A(u)=\,a,\,\,\lim_{u\rightarrow 0}2u\,B(u)=\,b,\,\,\,\lim_{u\rightarrow 0}B^*(u)=\,c,\,\,\,\lim_{u\rightarrow 0}D(u)=\,d.$$ By making an appropriate unitary transformation $U$, the middle matrix in above is simply diagnolized: $$\label{diagonalized}
\begin{split}
S\sim &\int_{\omega, \textbf{q}}\begin{pmatrix}
Z_{k}(0)&E_{k}(0)
\end{pmatrix}U\begin{pmatrix}
G_1&0\\
0& G_2
\end{pmatrix}U^{\dagger}\begin{pmatrix}
Z_{-k}(0)\\
E_{-k}(0)
\end{pmatrix}+\,\text{contact\,\,terms}\\
\sim& \int_{\omega, \textbf{q}}\begin{pmatrix}
\mathcal{Z}_{k}&\mathcal{E}_{k}
\end{pmatrix}\begin{pmatrix}
G_1&0\\
0& G_2
\end{pmatrix}\begin{pmatrix}
\mathcal{Z}_{-k}\\
\mathcal{E}_{-k}
\end{pmatrix}+\,\text{contact\,\,terms}.
\end{split}$$ The transformed variables at $u=0$ are then given by $$\label{new_Z_E}
\begin{split}
\mathcal{Z}_{k}=&\,U^{\dagger}_{11}Z_k(0)+\,U^{\dagger}_{12}E_{k}(0),\\
\mathcal{E}_{k}=&\,U^{\dagger}_{21}Z_k(0)+\,U^{\dagger}_{22}E_{k}(0).
\end{split}$$ The holographic AdS/CFT duality [@Witten:1998qj] implies that $\mathcal{Z}$ and $\mathcal{E}$ couple to specific operators $\mathcal{O}_1$ and $\mathcal{O}_2$ in the boundary theory. It is then clear that $G_1$ and $G_2$ in are the retarded Green’s functions of these operators. Using , and , one finds $$\label{}
\langle \mathcal{O}_1 \mathcal{O}_1 \rangle_{R}\sim G_1\sim \frac{\mathcal{K}_{-}(c_1,c_2,c_3,c_4)}{c_1 c_3}+\,\text{c.t.} ,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,
\langle \mathcal{O}_2 \mathcal{O}_2 \rangle_{R}\sim G_2 \sim\frac{\mathcal{K}_{+}(c_1,c_2,c_3,c_4)}{c_1 c_3}+\,\text{c.t.}$$ where “c.t.” stands for the contact term contributions [@Policastro:2002tn] and $$\label{}
\mathcal{K}_{\pm}=\,\frac{1}{2}\left(a\, c_2 c_3+d\, c_1c_4\pm \sqrt{(a\, c_2 c_3 + d\, c_1 c_4)^2-4 c_1 c_2 c_3 c_4(a d - b c )}\right).$$ By definition, the quasinormaol modes of bulk perturbations correspond to the poles of retarded Green’s functions [@Nunez:2003eq; @Birmingham:2001pj; @Horowitz:1999jd]. In our present case, these poles are roots of $c_1\,c_3$. Using , the corresponding roots are then found from $$\label{general_Dirichlet}
\begin{split}
C_{Z}\,g_{Z}(0)+\frac{}{}C_{E}\,y_{Z}(0)=&\,0,\\
C_{Z}\,g_{E}(0)+\frac{}{}C_{E}\,y_{E}(0)=&\,0.
\end{split}$$ Let us recall that each of $g_{Z}(0)$, $y_Z(0)$, $\cdots$ is a complicated function of $\omega$ and $\boldsymbol{q}$. In order for leads to non-trivial relations between $\omega$ and $\boldsymbol{q}$, it is required that $$\label{det_quasi}
\det\begin{pmatrix}
g_{Z}(0)&y_{Z}(0)\\
g_{E}(0)& y_{E}(0)
\end{pmatrix}=\,0.$$ This is our first result in this paper; the equation from which, we can find the quasinormal modes of coupled perturbations $Z$ and $E$.
Let us summarize. In order to find the quasinoraml modes of coupled perturbations $Z$ and $E$ in the spin 0 and 1 channels, we first construct the corresponding coupled differential equations. We should find the solutions to them which are ingoing at the horizon. Such solutions can be formally written in the form of , up to two arbitrary normalization constants $C_Z$ and $C_E$. Having specified the functions $g_Z$, $g_E$, $y_Z$ and $y_E$, then equation determines the spectrum of quasinormal modes in the associated channel.
Quasinormal spectrum and hydrodynamic modes in $\mathcal{N}=4$ SYM theory at finite chemical potential {#RN_equations}
======================================================================================================
Our system of interest is holographically described by dynamics of metric and a $U(1)$ gauge field in the bulk of AdS. The corresponding action is given by $$S = \frac{1}{16 \pi G_5} \int d^5 x \,\,\sqrt{-g} \left(R + \frac{12}{L^2} - F^{M N} F_{M N}\right)+ S_{bdy}.$$ where $S_{bdy}$ is the boundary counter term. The equations of motion are given by: $$\label{EoM}
\begin{split}
G_{\mu \nu}-6 g_{\mu \nu}+2\left(F_{\mu \rho} F^{\rho}_{\,\,\nu}+\,\frac{1}{4}F^{\alpha \beta}F_{\alpha \beta}\,\,g_{\mu \nu}\right)=\,0,\,\,\,\,\,\,\,\,\,
\nabla_{\mu}F^{\mu \nu}=\,0.
\end{split}$$ We work in the unite where $L=1$. The solution in the Poincare patch is written as it follows $$\label{back_ground}
ds^2=r^2\left(-f(r)dt^2+dx^2+dy^2+\frac{}{}dz^2\right)+\frac{1}{r^2 f(r)}dr^2,\,\,\,\,\,\,\, A=-\frac{\sqrt{3}q_b}{2 r^2}dt,$$ with $ f(r)=1-\frac{m}{r^4}+\frac{q_b^2}{r^6}$. Parameters $m$ and $q$ are two constants. It is convenient to re-scale the quantities with the radius of the outer horizon, $R$, namely the largest positive root of $f(r)=0$. We may write $$\label{rescale}
\rho\equiv\frac{r}{R},\,\,\,\,\,\,\,\,M=\frac{m}{R^4},\,\,\,\,\,\,\,\,Q=\frac{q_b}{R^3},\,\,\,\,\,\,\,\,Q^2=M-1.$$ In the rescaled coordinates, the outer horizon locates at $\rho=1$ while the boundary is identified with $\rho\rightarrow \infty$. For further requirements, we need to work in a system with finite domain of the radial coordinate; for this purpose, we make the change $\rho=1/\sqrt{u}$. The function $f$ then takes the following form $$f(u)=1-(Q^2+1)u^2+Q^2 u^3.$$ Using this together with , the Hawking temperature $T$ and the chemical potential $\mu$ of the boundary theory are found to be $$\label{mu_T}
\begin{split}
T=&\,\frac{1}{4\pi}\left(-2R\, f'_u\big|_{u=1}\right)=\,\frac{R}{2\pi}\,(2-Q^2),\\
\mu=&\,A_{t}\big|_{\infty}-A_{t}\big|_{u=1}=\,\frac{ \sqrt{3} q}{2R^2}=\,\frac{\sqrt{3}}{2}Q R.
\end{split}$$ From the expressions , one can derive $Q$ as a function of $\mu/T$: $$\label{Q__mu_T}
Q=\,\frac{\sqrt{2}\,a\, (\mu/T)}{1+\sqrt{1+a^2\,(\mu/T)^2}},\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,a=\left(\frac{8}{3\pi^2}\right)^{1/2}.$$ This relation simply shows that there is a one-to-one map between $\mu/T$ and $Q$. Thus in the following, we take $T$ and $Q$ as the two independent thermodynamic variables. Then the bulk metric and gauge field can be finally written in $(t,x,y,z, u)$ coordinates as the following: $$\label{Metric_Gauge_u_coord}
\begin{split}
ds^2&=\frac{1}{u}\left(\frac{2\pi T}{2-Q^2}\right)^2\left(-f(u)dt^2+\frac{}{}dx^2+dy^2+dz^2\right)+\frac{1}{4u^2 f(u)}du^2,\\
A&=-\frac{\sqrt{3}\pi T Q}{2-Q^2}\,u\,dt.
\end{split}$$
Spin $0$ Channel
----------------
We turn on the set of perturbations $\{h_{tt}, h_{tz}, h_{zz}, h, A_t, A_z\}$ in the radial gauge. On the RN background solution , the gauge invariant variables take the following form $$\label{Z_0_E_0_on_shell}
\begin{split}
Z_0&={{\mathfrak{q}}}^2 f H_{tt}+ 2 {{\mathfrak{q}}}\,{{\mathfrak{w}}}H_{tz}+{{\mathfrak{w}}}^2 H_{zz}+\,{{\mathfrak{q}}}^2\left(2-Q^2 u^3-\frac{{{\mathfrak{w}}}^2}{{{\mathfrak{q}}}^2}-f\right)H,\\
E_z&={{\mathfrak{q}}}A_t+ {{\mathfrak{w}}}A_z+{{\mathfrak{q}}}\,Q\,H.
\end{split}$$ Note that on the background solution , $\tilde{q}=Q$. To find the coupled dynamical equations of the above two variables, firstly it is required the perturbations $\{h_{tt}, h_{tz}, h_{zz}, h, a_t, a_z\}$ in the spin $0$ equations to be replaced with $\{a f H_{tt}, a H_{tz}, a H_{zz}, 2 a H, c_h A_t, c_h A_z\}$ (see and explanations given below that). Doing so, we find equations of the latter “six” perturbations. The difficult task is to eliminate all these perturbations in favor of the two gauge invariant variables $Z_0$ and $E_z$. After long computations, which are not shown here, we arrive at the two following coupled differential equations $$\label{Sound_channel_diff}
\begin{split}
Z_0''&+\,{{\textswab{a}}}_{1}\,Z_0'+\,{{\textswab{a}}}_{2}\,Z_0+\,{{\textswab{b}}}_{3}\,E_z+\,{{\textswab{b}}}_{4}\,E_z'=0,\\
E_z''&+\,{{\textswab{b}}}_{1}\,E_z'+\,{{\textswab{b}}}_{2}\,E_z+\,{{\textswab{a}}}_{3}\,Z_0+\,{{\textswab{a}}}_{4}\,Z_0'=0.
\end{split}$$ We have put $\textswab{a}_i$ coefficients in front of $Z_0$ and $Z_0'$ and have done the same for $\textswab{b}_i$ coefficients with $E_z$ and $E_z'$. Considering ${{\mathfrak{w}}}=\omega/2\pi T$ and ${{\mathfrak{q}}}=k/2\pi T$, the coefficients are found to be as $$\label{Z_0_coef}
\begin{split}
{{\textswab{a}}}_{1}&=\frac{(Q^2u^3-2){{\mathfrak{w}}}^2\mathcal{D}+f({{\mathfrak{q}}}^4(Q^4u^6-4)+{{\mathfrak{q}}}^2{{\mathfrak{w}}}^2(2-Q^2u^3)+3{{\mathfrak{w}}}^4)+ {{\mathfrak{q}}}^4(4+2Q^2u^3-3f)f^2}{u f({{\mathfrak{w}}}^2-{{\mathfrak{q}}}^2 f)\left(\mathcal{D}-{{\mathfrak{q}}}^2 f\right)},\\
{{\textswab{a}}}_{2}&=\frac{\tilde{Q}^2u \,{{\mathfrak{w}}}^2\,\mathcal{D}+{{\mathfrak{q}}}^2 f\left((Q^2u^3-2)(8-{{\mathfrak{q}}}^2\tilde{Q}^2u+8Q^2u^3)-4\tilde{Q}^2u {{\mathfrak{w}}}^2+(32+{{\mathfrak{q}}}^2\tilde{Q}^2u+8Q^2 u^3-16f)f)\right)}{4 u^2 f^2\, \left(\mathcal{D}-{{\mathfrak{q}}}^2 f\right)},\\
{{\textswab{a}}}_{3}&=\,-\,\frac{2\,{{\mathfrak{q}}}\,Q\,(1+Q^2 u^3-f)}{u f (\mathcal{D}-{{\mathfrak{q}}}^2 f)},\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,
{{\textswab{a}}}_{4}=\frac{{{\mathfrak{q}}}\, Q\,({{\mathfrak{q}}}^2(2+Q^2u^3)-{{\mathfrak{w}}}^2-{{\mathfrak{q}}}^2 f)}{(\mathcal{D}-{{\mathfrak{q}}}^2 f)({{\mathfrak{q}}}^2 f-{{\mathfrak{w}}}^2)}
\end{split}$$ and $$\label{E_z_coef}
\begin{split}
{{\textswab{b}}}_{1}&=\frac{2 {{\mathfrak{q}}}^2Q^2 u^2 f-{{\mathfrak{w}}}^2f'}{f({{\mathfrak{q}}}^2 f-{{\mathfrak{w}}}^2)},\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\\
\\
{{\textswab{b}}}_{2}&=\frac{-\tilde{Q}^2{{\mathfrak{w}}}^4 \mathcal{D}+2{{\mathfrak{q}}}^2 {{\mathfrak{w}}}^2 f({{\mathfrak{q}}}^2 \tilde{Q}^2+6 Q^2 u^2)(Q^2 u^3-2)+{{\mathfrak{w}}}^4f(7 {{\mathfrak{q}}}^2 \tilde{Q}^2+36 Q^2 u^2)}{4 u f^2(\mathcal{D}-{{\mathfrak{q}}}^2 f)({{\mathfrak{q}}}^2 f -{{\mathfrak{w}}}^2)}\\
&\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,+\frac{{{\mathfrak{q}}}^2 f^2(48 {{\mathfrak{q}}}^2 Q^2 u^2-{{\mathfrak{q}}}^4\tilde{Q}^2(Q^2 u^3-2)-5({{\mathfrak{q}}}^2\tilde{Q}^2+12Q^2 u^2){{\mathfrak{w}}}^2)+{{\mathfrak{q}}}^6\tilde{Q}^2f^3}{4 u f^2(\mathcal{D}-{{\mathfrak{q}}}^2 f)({{\mathfrak{q}}}^2 f -{{\mathfrak{w}}}^2)},\\
{{\textswab{b}}}_{3}&=\,-\frac{6\,{{\mathfrak{q}}}\, Q\bigg((Q^2u^3-2){{\mathfrak{w}}}^2\mathcal{D}+(2{{\mathfrak{q}}}^4(Q^2u^3-2)+{{\mathfrak{q}}}^2{{\mathfrak{w}}}^2(2-Q^2u^3)+6{{\mathfrak{w}}}^4)f+(4{{\mathfrak{q}}}^4-6{{\mathfrak{q}}}^2{{\mathfrak{w}}}^2)f^2\bigg)}{f\,({{\mathfrak{q}}}^2 f-{{\mathfrak{w}}}^2)\,\left(\mathcal{D}-\,{{\mathfrak{q}}}^2 f\right)},\\
{{\textswab{b}}}_{4}&=\frac{2 \,{{\mathfrak{q}}}\, Q\,u\,\left(\mathcal{D}-{{\mathfrak{q}}}^2 f\right)}{({{\mathfrak{w}}}^2-{{\mathfrak{q}}}^2 f)}.
\end{split}$$ We have defined $\mathcal{D}=({{\mathfrak{q}}}^2(Q^2u^3-2)+3 {{\mathfrak{w}}}^2)$ and $\tilde{Q}=Q^2-2$. At this point it should be noted that when $Q=0$, the above equations reduce exactly to the pair of decoupled equations (4.5b) and (4.35) in ref. [@Kovtun:2005ev]. Finding the quasinormal modes as well as hydrodynamic modes from these equations is the subject of following subsections.
### Quasinormal modes {#quasinormal-modes .unnumbered}
The analytic solution to equations is unknown; so to find the associated spectrum of quasinormal modes, one may proceed with numerically solving them. To this end, we combine the method developed in together with the Frobenius expansions of $Z_0$ and $E_z$ in the bulk (see Appendix \[Frobenius\] for more details). In what follows, the corresponding numerical results will be given.
In Fig.\[quasi\_EzZ0\_Q\_12\_fig\], the typical arrangement of poles has been demonstrated for two cases in this channel. In the *left panel,* we have shown the quasinormal modes associated with ${{\mathfrak{q}}}=1$ at $Q=0.5$. As can be seen, we have splitted them into two sets, denoted by dots and stars. The idea for such spitting comes from the knowledge about the arrangement of poles in the sound channel as well as in the $R-$current channel on the AdS-Schwarzschild background [@Kovtun:2005ev]. In the latter case $R-$current fluctuations decouple from the spin 0 fluctuations of stress tensor and one finds two separated spectra for their associated quasinormal modes. Superposition of those two spectra is what to which we have compared our plots to distinguish between stars and dots.
![Left panel: Stars represent poles of the charge current Green’s function. Dots correspond with poles of energy density Green’s function. Right panel: As ${{\mathfrak{q}}}$ decreases, all poles stay at a finite distance from the real axis, except for the one marked with a large star and the two marked with large dots. Large dots therefore manifest the existence of two sound modes and the large dot corresponds with the existence of a diffusive $U(1)$ charge mode in the boundary $\mathcal{N}=4$ SYM theory at finite chemical potential.[]{data-label="quasi_EzZ0_Q_12_fig"}](sound_quasi_q_1_Q_12.pdf "fig:"){width="42.00000%"} ![Left panel: Stars represent poles of the charge current Green’s function. Dots correspond with poles of energy density Green’s function. Right panel: As ${{\mathfrak{q}}}$ decreases, all poles stay at a finite distance from the real axis, except for the one marked with a large star and the two marked with large dots. Large dots therefore manifest the existence of two sound modes and the large dot corresponds with the existence of a diffusive $U(1)$ charge mode in the boundary $\mathcal{N}=4$ SYM theory at finite chemical potential.[]{data-label="quasi_EzZ0_Q_12_fig"}](sound_quasi_q_compare.pdf "fig:"){width="42.00000%"}
In the *right panel* of Fig.\[quasi\_EzZ0\_Q\_12\_fig\], we have compared the left panel spectrum with the one corresponding to the same $Q$ but at a smaller ${{\mathfrak{q}}}$. One clearly notices that as ${{\mathfrak{q}}}$ decreases, all complex poles move away from the horizontal axis, but the two large dots and the large star move towards the origin. This simply shows that the spectral curve of the spin 0 fluctuations includes three branches of Puiseux series passing through the origin of complex plane. In the next subsection, we explicitly derive the equation of these branches in the vicinity of origin. These three branches correspond to three hydrodynamic modes: *two sound modes* together with *one diffusion mode*. It should be also noted that the poles other than these three are all gapped modes.
{width="45.00000%"}
So far we have just talked about the quasinormal modes at a fixed finite value of $Q$. In Fig.\[quasi\_EzZ0\_Q\_12\_fig\_color\], we have depicted part of the spectrum of quasinormal modes at the fixed momentum ${{\mathfrak{q}}}=1$ for several values of $Q$ within the range $0\le Q \le 0.8$. One observes that by approaching towards extremal limit[^10], even at a finite fixed value of momentum, e.g. at ${{\mathfrak{q}}}=1$, the non-hydrodynamic (gapped) modes move away from the origin. As mentioned in the Introduction, our numerical method works well within the range $0\le Q\le0.88$. It would be interesting to try other numerical methods to find the extrapolation of colorful paths depicted in Fig.\[quasi\_EzZ0\_Q\_12\_fig\_color\] at larger values of $Q$, specifically when $Q\rightarrow\sqrt{2}$ or equivalently when $T\rightarrow 0$.
### Hydrodynamic limit {#hydrodynamic-limit .unnumbered}
Although the complete analytic solution to equations is unknown, one can analytically solve them in the hydrodynamics limit. At first sight, equations may seem impossible to become decoupled. But as we will show, at small $\mu/T$ limit, namely when $Q\ll1$, they decouple in the hydrodynamic limit. What we are going to do is to perturbatively solve in the hydrodynamic expansion and also in $Q$ expansion.
Demanding the solutions be ingoing at the horizon, the near horizon behavior of $Z_0$ and $E_z$ are fixed as follows: $$\label{N_H_sound}
u\rightarrow 1\,\,\,\,\,\,\,\,\,\,\,\,\,\,Z_0(u), E_z(u)\,\sim\,\left(1-(1+Q^2)u^2+\frac{}{}Q^2 u^3\right)^{-i {{\mathfrak{w}}}/2}.$$ In order to enter the hydrodynamic expansion, we apply the rescaling ${{\mathfrak{w}}}\rightarrow \epsilon \,{{\mathfrak{w}}}$ and ${{\mathfrak{q}}}\rightarrow \epsilon \,{{\mathfrak{q}}}$ to the dynamical equations as well as to . A quick look at the the coefficients and reveals that $$\begin{aligned}
\textswab{a}_{1}, \textswab{a}_{2} \sim O(\epsilon^0),& &\,\,\,\,\,\,\,\textswab{b}_{3}, \textswab{b}_{4} \sim O(\epsilon),\\
\textswab{b}_{1}, \textswab{b}_{2} \sim O(\epsilon^0),& &\,\,\,\,\,\,\,\textswab{a}_{3}, \textswab{a}_{4} \sim O(\epsilon^{-1}).\end{aligned}$$ Such scaling simply means that in each of the equations , the function $Z_0$ must be one order higher in $\epsilon$ expansion, than the function $E_z$. Thus the appropriate ansatz for the functions $Z_0(u)$ and $E_z(u)$ in the double expansion over $\epsilon$ and $Q$ is given by $$\label{Z_hydro}
\begin{split}
Z_0(u) &=\left(1-(1+Q^2)u^2+\frac{}{}Q^2 u^3\right)^{-i {{\mathfrak{w}}}/2}\, \sum_{k=0} {{\mathfrak{w}}}^{k+1}\left(Z_{0}^{k,0}(u)+ Q Z_{0}^{k,1}(u)+ Q^2 Z_{0}^{k,2}(u)\right),\\
E_z(u) &=\left(1-(1+Q^2)u^2+\frac{}{}Q^2 u^3\right)^{-i {{\mathfrak{w}}}/2}\, \sum_{k=0} {{\mathfrak{w}}}^{k}\left(E_{z}^{k,0}(u)+ Q\frac{}{} E_{z}^{k,1}(u)+ Q^2 E_{z}^{k,2}(u)\right).
\end{split}$$ Substituting the above expressions into equations and expanding over $\epsilon$ and $Q$, we obtain a set of second order ordinary differential equations for the functions $Z_{0}^{n,m}$ and $E_z^{n,m}$.
Starting from the lowest order in $\epsilon$ and $Q$, one firstly finds $Z_{0}^{0,0}(u)$ from the first line in up to an unknown coefficient. Regularity at $u=1$ together with fixing its value at the same point, namely $Z_{0}^{0,0}(1)=C_1$, picks out a unique regular solution to $Z_{0}^{0,0}(u)$. Using this solution, then the next function that can be found is $E_{z}^{0,0}(u)$; from the second equation in . Again, regularity at $u=1$ and demanding $E_z^{0,0}(1)=C_2$ fix the solution.
In the Appendix \[App\_sol\_HYDRO\_SOUND\], we have listed the corresponding perturbative solutions at higher orders, according to the ordering we have found them through the perturbation theory. It should be noted that at every order of perturbation, one of the two boundary conditions is regularity at $u=1$ and the second one is the the normalization of the solution at the same point: $$\label{Dirichlete_sound}
\begin{split}
Z^{0,0}(1)=&\,C_1, \,\,\,\,\,\,\,E_z^{0,0}(1)=C_2,\\
\text{and}\,\,\,\,\,\,\,\,\,\, Z^{m,n}(1)=&\,0, \,\,\,\,\,\,\,\,\,E_z^{m,n}(1)=0:\,\,\,\,\,\,m^2+n^2\ne 0.
\end{split}$$ Quasinormal modes are, by definition [@Nunez:2003eq], the modes obtained upon applying the Dirichlet boundary condition to . One thus writes $$\begin{aligned}
Z_0(0)=0,\,\,\,\,\,E_z(0)=0.
\end{aligned}$$ According to our earlier discussions these two equations are coupled. By explicit computations (Appendix \[App\_sol\_HYDRO\_SOUND\]) we find[^11] $$\label{Coupled_Dirichete}
\begin{split}
Z_0(0)=0&:\,\,\,\,\,\,\,\,\,m_{11}({{\mathfrak{w}}}, {{\mathfrak{q}}}; Q) \,C_1+m_{12}({{\mathfrak{w}}}, {{\mathfrak{q}}}; Q)\, C_2=0,\\
E_z(0)=0&:\,\,\,\,\,\,\,\,\,m_{21}({{\mathfrak{w}}}, {{\mathfrak{q}}}; Q)\, C_1+m_{22}({{\mathfrak{w}}}, {{\mathfrak{q}}}; Q)\, C_2=0,
\end{split}$$ where $C_1$ and $C_2$ are the normalization coefficients defined in . The $m_{ij}$ coefficients, up to first order in derivatives, are found to be $$\label{m_ij_Spin_0}
\begin{split}
m_{11}&=(1-3\textswab{y}^2)\,{{\mathfrak{w}}}-2 i\, {{\mathfrak{w}}}^2+\bigg(\frac{5-6\textswab{y}^2}{3\textswab{y}^2-2}\,{{\mathfrak{w}}}+\frac{8-2\textswab{y}^2(\log 8+1)+3\textswab{y}^4(1-3\textswab{y}^2)(3-\log 8)}{2\textswab{y}^2(3\textswab{y}^2-2)}\,{{\mathfrak{w}}}^2\bigg)\,Q^2,\\
m_{12}&=\bigg(-\frac{4\sqrt{3}}{\textswab{y}}\,{{\mathfrak{w}}}+\frac{4i\,(-2+\textswab{y}^2(\log 8-3))}{\sqrt{3}\textswab{y}^3}\,{{\mathfrak{w}}}^2\bigg)\,Q,\\
m_{21}&=\bigg(-\frac{\sqrt{3} \,\textswab{y}}{4}-\frac{i\sqrt{3}\,(-1+\textswab{y}^2(\log 2-1))}{4 \textswab{y}}\,{{\mathfrak{w}}}\bigg)\,Q,\\
m_{22}&=1+i \left(\frac{1}{\textswab{y}^2}-\log 2\right)\,{{\mathfrak{w}}}+\bigg(\frac{3}{2}+\frac{1}{2}i\, \left(\frac{1}{\textswab{y}^2}-\log 2\right){{\mathfrak{w}}}\bigg)\,Q^2
\end{split}$$ where $\textswab{y}={{\mathfrak{w}}}/{{\mathfrak{q}}}$. Obviously, in the limit $Q=0$, the two non-diagonal coefficients $m_{12}$ and $m_{21}$ vanish and one is left with two decoupled equations in ; $C_1\,m_{11}=0$ and $C_2\,m_{22}=0$. Requiring $C_1, \,C_2\ne0$, these equations then give the well-known dispersion relations, namely equations (4.43) and (4.16) in [@Kovtun:2005ev], respectively. When $Q\ne0$, however, equations are coupled and have a non-trivial set of solutions if and only if one demands $$\label{m_ij_Spin_0_matrix}
\det\begin{pmatrix}
m_{11}&m_{12}\\
m_{21}& m_{22}
\end{pmatrix}=\,0.$$ Solving the recent equation to second order in ${{\mathfrak{q}}}$ and $Q$, we find the dispersion of the spin 0 hydrodynamic excitations in a holographic charged fluid as it follows $$\label{hydro_mode_holog_spin_0}
\begin{split}
{{\mathfrak{w}}}^{\pm}_{\text{sound}}=&\,\pm\frac{1}{\sqrt{3}}{{\mathfrak{q}}}-\frac{i}{6}\left(2-3 \frac{}{}Q^2\right){{\mathfrak{q}}}^2,\\
{{\mathfrak{w}}}_{\text{diffusion}}=&\,-i\left(1- \frac{}{}Q^2\right){{\mathfrak{q}}}^2,\\
{{\mathfrak{w}}}_{\text{non-hydro}}=&-\frac{i}{\log 2}(1+4 Q^2)\,+i {{\mathfrak{q}}}^2 \left(1-\frac{1+ \log 2}{\log 2}Q^2\right).
\end{split}$$ The expression in the first line of is the dispersion relation of the two hydrodynamic sound modes in the charged fluid. The second line is showing the hydrodynamic diffusion of the $U(1)$ charge. Finally the expression given in the third line does obviously correspond to a gapped mode which lies beyond the regime of hydrodynamics. To best of our knowledge, this is the first computation of hydrodynamic modes in a AdS RN background.[^12]
Spin $1$ Channel
----------------
We turn on the set of perturbations $ \{h_{tx}, h_{zx}, h, A_x\}$ in the radial gauge. As mentioned in , the gauge invariant variables associated with this channel are identified with $Z_1$ and $E_x$ given by . Just like in the spin 0 channel, the difficult part of the computation here is to find dynamical equations governing dynamics of these variables. Combining all spin 1 components of and performing long computations, which are not shown here, we have eliminated $ \{h_{tx}, h_{zx}, h, A_x\}$ in favor of $Z_1$ and $E_x$. Eventually we have arrived at the following two coupled dynamical equations $$\label{Shear_channel_diff}
\begin{split}
0=&\,Z_1''+\,\frac{(Q^2 u^3-2){{\mathfrak{w}}}^2+{{\mathfrak{w}}}^2f + {{\mathfrak{q}}}^2 f^2}{u f({{\mathfrak{w}}}^2-{{\mathfrak{q}}}^2 f)}\,Z_1'+\,\frac{\tilde{Q}^2({{\mathfrak{w}}}^2-{{\mathfrak{q}}}^2 f)}{4 u f^2}\,Z_1+\,\frac{3 Q \,{{\mathfrak{q}}}\, {{\mathfrak{w}}}\,\,u\, f'}{({{\mathfrak{w}}}^2-{{\mathfrak{q}}}^2 f)\,f}\,E_x+\,\frac{3 Q {{\mathfrak{q}}}u }{{{\mathfrak{w}}}}\,E_x',\\
0=&\,E_x''+\,\frac{f'}{f}\,E_x'+\,\frac{\tilde{Q}^2({{\mathfrak{w}}}^2-{{\mathfrak{q}}}^2 f)^2-12 \,Q^2 \,{{\mathfrak{w}}}^2\,u^2 f}{4 \,u f^2\,({{\mathfrak{w}}}^2-{{\mathfrak{q}}}^2 f)}\,E_x-\frac{ Q \,{{\mathfrak{q}}}\, {{\mathfrak{w}}}}{({{\mathfrak{w}}}^2-{{\mathfrak{q}}}^2 f)}\,Z_1'.
\end{split}$$ As it must be, one can readily check that when $Q=0$, the above equations reduce exactly to the pair of decoupled equations (4.5a) and (4.26) in ref. [@Kovtun:2005ev]. In the following two subsections, we proceed with numerically and analytically solving the above equations, respectively.
### Quasinormal modes {#quasinormal-modes-1 .unnumbered}
The analytic solution to coupled equations is unknown. However, just like what was done in the spin 0 channel, we can use the Frobenius expansions of $Z_1$ and $E_x$ to find the corresponding quasinormal modes via the method developed in .
Firstly in order to get familiar with the typical arrangement of quasinormal modes in the spin 1 channel, in the *left* panel of Fig.\[quasi\_ExZ1\_Q12\_fig\] we have shown the spectrum at ${{\mathfrak{q}}}=1$ for $Q=0.5$. We have splitted the poles into two sets, denoted by stars and dots. The idea of such splitting originates from the corresponding spectra of quasinormal modes associated with decoupled variables $Z_1$ and $E_x$ on an AdS-Schwarzschild background. One naturally concludes that dots in the above figures correspond to the poles of spin 1 fluctuations of stress tensor, e.g. $T_{xz}$. The stars then identify the poles of spin 1 fluctuations of boundary charge current, e.g. $J_{x}$.
![Left panel: Stars are poles of the transverse momentum density Green’s function. Dots correspond with the poles of the transverse charge current Green’s function. Right panel: As ${{\mathfrak{q}}}$ decreases, all poles stay at a finite distance from the real axis, except for the one marked with a large dot. The latter manifests the existence of a diffusive shear mode in the boundary $\mathcal{N}=4$ SYM theory at finite chemical potential. []{data-label="quasi_ExZ1_Q12_fig"}](Shear_quasi_q_1_Q_12.pdf "fig:"){width="42.00000%"} ![Left panel: Stars are poles of the transverse momentum density Green’s function. Dots correspond with the poles of the transverse charge current Green’s function. Right panel: As ${{\mathfrak{q}}}$ decreases, all poles stay at a finite distance from the real axis, except for the one marked with a large dot. The latter manifests the existence of a diffusive shear mode in the boundary $\mathcal{N}=4$ SYM theory at finite chemical potential. []{data-label="quasi_ExZ1_Q12_fig"}](Shear_quasi_q_compare.pdf "fig:"){width="42.00000%"}
In the *right* panel of Fig.\[quasi\_ExZ1\_Q12\_fig\] we have compared the left panel spectrum with the spectrum associated for the same $Q$ but at a smaller ${{\mathfrak{q}}}$. As it can be seen, when ${{\mathfrak{q}}}$ decreases, all complex modes move away from the real axis. At the same time, the large dot mode becomes closer and closer to the origin. It simply shows that there exits one branch of Puiseux series associated with the spectral function of spin 1 channel that passes through the origin. In other words, there is only one gapless mode in this channel which is actually the hydrodynamic *shear mode*. In next subsection we explicitly derive the dispersion relation of this mode.
Our discussion on quasinormal modes in this channel has been so far limited to the case with a fixed value of $Q$. In Fig.\[quasi\_ExZ1\_Q12\_fig\_color\], we have demonstrated part of the spectrum of quasinormal modes at the fixed momentum ${{\mathfrak{q}}}=1$ for several values of $Q$ within the range $0\le Q \le 0.8$. One observes that by approaching towards extremality, the gapped modes move away from the real axis, while the gapless mode becomes close to the origin. It would be interesting to study the spectrum in the extrapolation region $0.9 < Q \le \sqrt{2}$ to see what the fate of this mode in the extremal limit will be [@navid:2020].
{width="45.00000%"}
### Hydrodynamic limit {#hydrodynamic-limit-1 .unnumbered}
In this subsection we are going to find an analytic solution to the dynamical equations in the hydrodynamic limit ${{\mathfrak{w}}}({{\mathfrak{q}}}\rightarrow0)=\,0$. As mentioned in [@Kovtun:2005ev], it turns out that the appropriate rescaling in this channel is ${{\mathfrak{w}}}\rightarrow \epsilon^2 {{\mathfrak{w}}}$ and ${{\mathfrak{q}}}\rightarrow \epsilon {{\mathfrak{q}}}$ with $\epsilon \ll1$. But even in this limit, equations are coupled; then as was the case in the spin 0 channel, we proceed with perturbatively expanding equations over $Q$ as well. This expansion for $Q\ll1$ together with the derivative expansion over $\epsilon$ make it possible to analytically find $Z_1$ and $E_x$.
To specify the general form of the solutions, it is required to investigate the behavior of coefficients in under the above-mentioned hydro rescaling. Doing so we find that the appropriate ansatz for the functions $Z_1$ and $E_x$ is given by: $$\label{Shear_hydro}
\begin{split}
Z_1(u) &=\left(1-(1+Q^2)u^2+\frac{}{}Q^2 u^3\right)^{-i {{\mathfrak{w}}}/2}\, \sum_{k=0} {{\mathfrak{w}}}^{k}\left(Z_{1}^{k,0}(u)+ Q Z_{1}^{k,1}(u)+ Q^2 Z_{1}^{k,2}(u)\right),\\
E_x(u) &=\left(1-(1+Q^2)u^2+\frac{}{}Q^2 u^3\right)^{-i {{\mathfrak{w}}}/2}\, \sum_{k=1} \left(\frac{{{\mathfrak{w}}}}{{{\mathfrak{q}}}}\right)^{k}\left(E_{x}^{k,0}(u)+ Q\frac{}{} E_{x}^{k,1}(u)+ Q^2 E_{x}^{k,2}(u)\right).
\end{split}$$ Thus the main task is to find the functions $Z_{1}^{k,m}$ and $E_x^{k,m}$. It should be noted that the ingoing boundary condition at the horizon has been already fulfilled in . The remaining boundary conditions are then given by $$\label{Dirichlete_shear}
\begin{split}
Z_1^{0,0}(1)=&\,C_3, \,\,\,\,\,\,\,E_x^{0,0}(1)=C_4,\\
\text{and}\,\,\,\,\,\,\,\,\,\, Z_1^{m,n}(1)=&\,0, \,\,\,\,\,\,\,\,\,E_x^{m,n}(1)=0:\,\,\,\,\,\,m^2+n^2\ne 0.
\end{split}$$ Here $C_3$ and $C_4$ are the values of $Z_1$ and $E_x$ at the horizon. The solution functions that obey the above boundary conditions have been given in the Appendix \[App\_sol\_HYDRO\_SOUND\]. The hydrodynamic modes in this channel are then found by applying the Dirichlet condition to the solutions evaluated at $u=0$ (see ). The latter can be formally written as $$\label{Coupled_Dirichete_Shear}
\begin{split}
Z_1(0)=0&:\,\,\,\,\,\,\,\,\,s_{11}({{\mathfrak{w}}}, {{\mathfrak{q}}}; Q) \,C_3+s_{12}({{\mathfrak{w}}}, {{\mathfrak{q}}}; Q)\, C_4=0,\\
E_x(0)=0&:\,\,\,\,\,\,\,\,\,s_{21}({{\mathfrak{w}}}, {{\mathfrak{q}}}; Q)\, C_3+s_{22}({{\mathfrak{w}}}, {{\mathfrak{q}}}; Q)\, C_4=0.
\end{split}$$ To leading order in $\epsilon$ and second order in $Q$, we find $$\label{Dirichlete_Spin_1}
\begin{pmatrix}
s_{11}&s_{12}\\
s_{21}& s_{22}
\end{pmatrix}=\, \begin{pmatrix}
1+ \frac{i {{\mathfrak{q}}}^2}{2 {{\mathfrak{w}}}}&\frac{3Q}{2}\\
\frac{i {{\mathfrak{q}}}^2 Q}{2{{\mathfrak{w}}}}& \left(1+\frac{3Q^2}{2}\right)
\end{pmatrix}.$$
It is obvious that at $Q=0$ limit, equations decouple and one finds ${{\mathfrak{w}}}=-i {{\mathfrak{q}}}^2/2$, the well-known shear mode in a holographic neutral fluid [@Kovtun:2005ev]. When $Q\ne0$, equations are coupled and have non-trivial solutions if and only if $det(s_{ij})=0$; this gives $$\label{hydro_mode_holog_spin_1}
{{\mathfrak{w}}}=- \frac{i}{4}\left(2-3 Q^2\right) {{\mathfrak{q}}}^2,$$ which shows how the density of $U(1)$ charge perturbatively modifies the dispersion of the shear mode in the system.
Spin $2$ Channel
----------------
As discussed in , by turning on metric perturbations $ \{h_{xy}, h_{xx}, h_{yy}\}$ in the radial gauge, two spin 2 gauge invariant variables $Z_2$ and $W_2$ are excited (see ). We find that on the RN background solution , these quantities commonly obey the following equation $$\label{Spin_2_dynamics}
Z_2''+\,\left(\frac{f'}{f}-\frac{1}{u}\right)\,Z_2'+\,\frac{\tilde{Q}^2\,u ({{\mathfrak{w}}}^2- {{\mathfrak{q}}}^2 f)+f\big(12(1-f)+6Q^2 u^3+10 u f'- 4 u^2 f''\big)}{4 u^2 f^2}\,Z_2=\,0.$$ With no need to the follow , we take the following Frobenius expansion $$Z_2(u)=(1-(1+Q^2)u^2+Q^2 u^3)^{-i {{\mathfrak{w}}}/2}\sum_{n=0}^{\infty} c_n({{\mathfrak{w}}}, {{\mathfrak{q}}})(1-u)^n.$$ Substituting it into , we find coefficients $c_n$ all in terms of $c_0$. Then by applying the Dirichlet boundary condition we arrive at the spectral curve of fluctuations in this channel: $$Z_2(0)=\sum_{n=0}^{\infty} c_n({{\mathfrak{w}}}, {{\mathfrak{q}}})=\,0.$$ By keeping sufficient number of terms in the sum, one can numerically find the spectrum of quasinormal modes.
![Quasinormal modes associated with $Q=0.5$ at two different values of ${{\mathfrak{q}}}$, in spin 1 channel. At each value of ${{\mathfrak{q}}}$, we have shown five lowest quasinormal modes. As ${{\mathfrak{q}}}$ decreases, all poles stay at finite distances from the real axis. The latter manifests the non-existence of any spin 2 hydrodynamic mode in $\mathcal{N}=4$ SYM theory at finite chemical potential.[]{data-label="quasi_Exy"}](Spin2_two_qs.pdf){width="60.00000%"}
{width="45.00000%"}
In Fig.\[quasi\_Exy\], we have demonstrated the spectra associated with $Q=0.5$ for two values of ${{\mathfrak{q}}}$. For both cases, we have shown five lowest-lying quasinormal modes. When ${{\mathfrak{q}}}$ decreases, all complex modes move away from the real axis. At the same time, the purely imaginary mode becomes closer to the origin, however, it never reaches the origin. This observations is enough to conclude that there is no any branch of poles passing through the origin when ${{\mathfrak{q}}}\rightarrow 0$. In other words, spin 2 channel of fluctuations contains only gapped excitations.
Our discussion on quasinormal modes in this channel has been so far limited to the case with a fixed value of $Q$. In Fig.\[quasi\_Exy\_color\], we have demonstrated part of the spectrum of quasinormal modes at the fixed momentum ${{\mathfrak{q}}}=1$ for various values of $Q$ within the range $0.45\le Q \le 0.8$. One observes that by approaching towards extremality, the gapped modes move away from the real axis, while the purely imaginary mode becomes close to the origin. It would be interesting to study the spectrum extrapolating region $0.9 < Q \le \sqrt{2}$ to see what the fate of this mode in the extremal limit will be.
Complex life of quasinormal modes and the radius of convergence of the hydrodynamic derivative expansion {#complex_life}
========================================================================================================
Let us recall that in every channel of our study, the Dirichlet boundary condition at $u=0$ gives an algebraic equation between ${{\mathfrak{w}}}$ and ${{\mathfrak{q}}}$. Such an equation specifies the spectral curve of the collective excitations in the associated channel. For the spin $n$ channel, it can be formally written as[^13] $$\label{Spectral_Spin_n}
\boldsymbol{F}_n({{\mathfrak{q}}}^2, {{\mathfrak{w}}})=0.$$ It is clear that both non-hydrodynamic and possible hydrodynamic modes are encoded in this equation. Hydrodynamic modes, ${{\mathfrak{w}}}({{\mathfrak{q}}}^2\rightarrow 0)= 0$, are expected just to live in the vicinity of origin $(0,0)$. The small-${{\mathfrak{q}}}$ expansion of ${{\mathfrak{w}}}({{\mathfrak{q}}}^2)$ then can be found by using the theorem of Puiseux. The Puiseux analysis implies that the domain of convergence of Puiseux series centered at the origin is the circle whose radius is set by the distance from the origin, to the nearest critical point of the associated spectral curve [@Grozdanov:2019kge]. The critical points of the spectral curve, themselves, can be found by solving the following set of equations: $$\label{Critical}
\boldsymbol{F}_n( {{\mathfrak{q}}}_c^2, {{\mathfrak{w}}}_c)=0,\,\,\,\,\,\,\,\frac{\partial\boldsymbol{F}_n( {{\mathfrak{q}}}_c^2, {{\mathfrak{w}}}_c)}{\partial {{\mathfrak{w}}}}=\,0.$$ In [@Grozdanov:2019kge], it has been shown that the critical points obtained from these equations are exactly the level-crossing points of the complexified quasinormal modes in the associated channel.
The focus of [@Grozdanov:2019kge] is to study the quasinormal modes of a holographic *neutral* fluid. However, when fluid carries a conserved charge as well, the arrangement of complexified quasinormal modes may change significantly. Specifically in the spin 0 channel, in addition to the sound branches of the Puiseux series passing through the origin, the diffusion branch passes through the same point, too. It may possibly give rise to emergence of new critical points, due to crossing between sound and diffusion branches. As we explicitly show in the next subsection, such critical points will appear when the parameter $Q$ exceeds a specific threshold.
Spin 0 channel
--------------
In we formally argued how to find the quasinormal modes of coupled variables in the bulk. For the present spin channel, the spectral equation, namely $\boldsymbol{F}_\text{0}( {{\mathfrak{q}}}^2, {{\mathfrak{w}}}; Q)=0$, was already solved numerically in previous section and the corresponding spectrum of quasinormal modes typically was shown in Fig.\[quasi\_EzZ0\_Q\_12\_fig\]. From the arrangement of poles in that figure one notices that $\boldsymbol{F}_\text{0}( {{\mathfrak{q}}}^2, {{\mathfrak{w}}}; Q)=0$ has three eigen frequencies in the vicinity of $(0,0)$. Assuming the analyticity of $\boldsymbol{F}_0$ at $(0,0)$, then the implicit function theorem gives these three branches by three Puiseux series as it follows $$\label{Puiseux}
\begin{split}
{{\mathfrak{w}}}^{\pm}_{\text{sound}}=&\,-i\sum_{n=1}^{\infty}a_n e^{\pm \frac{i \pi n}{2}}{{\mathfrak{q}}}^n=\,\pm a_1 {{\mathfrak{q}}}+\, i a_2{{\mathfrak{q}}}^2+\cdots\\
{{\mathfrak{w}}}_{\text{diffusion}}=&\,-i\sum_{n=1}^{\infty}c_n {{\mathfrak{q}}}^{2n}=\,- i c_1 {{\mathfrak{q}}}^2+\cdots .
\end{split}$$ Following refs. [@Withers:2018srf; @Grozdanov:2019kge], what we refer to as the hydrodynamic derivative expansion is the form of the above expansions in the momentum space.
Let us recall that the two sound branches are located symmetrically with respect to imaginary axis in the complex ${{\mathfrak{w}}}$ plane. Thus, in half of the ${{\mathfrak{w}}}$ plane, for instance, where $\text{Re} \, {{\mathfrak{w}}}>0$ there are exactly two branches passing through the origin; ${{\mathfrak{w}}}^{+}_{\text{sound}}$ together with ${{\mathfrak{w}}}_{\text{diffusion}}$. By naively applying the statement of [@Grozdanov:2019kge] to the present case, one may conclude that the distance between $(0,0)$ and the critical point of $\boldsymbol{F}_0$, the nearest to origin, identifies the radius of convergence of the derivative expansion in this channel. But an immediate follow-up question is: to which series given in such radius corresponds?
Needless to say that ${{\mathfrak{w}}}^{+}_{\text{sound}}$ and ${{\mathfrak{w}}}_{\text{diffusion}}$ may have different radii of convergence. In fact *the radius of convergence of ${{\mathfrak{w}}}^{+}_{\text{sound}}$ (${{\mathfrak{w}}}_{\text{diffusion}}$) is identified with the distance between origin and the nearest critical point located on the sound (diffusion) branch of Puiseux series*.
![ The radius of convergence of the derivative expansion versus $Q$. As $Q$ increases, the radius of convergence of ${{\mathfrak{w}}}_{\text{shear}}$ (red curve) monotonically increases. At the same time, domain of convergence of ${{\mathfrak{w}}}_{\text{sound}}$ non-trivially changes. As discussed in the text, it can be studied in three different intervals: $0\le Q\le 0.386$, $0.386\le Q \le 610$ and $0.610\le Q\le 0.8$. The intersection point of blue curve with the vertical axis, related to the sound mode in the $\mathcal{N}=4$ SYM theory in the vanishing $\mu$ limit, was found in [@Grozdanov:2019kge].[]{data-label="q_c_Sound_diiusion"}](q_c_Q_sound.pdf){width="70.00000%"}
Thus in addition to find the nearest critical points, via solving , one has to be careful about positioning of them on the considered branch of Puiseux series.
We have numerically found the nearest critical points to the origin, on both sound and diffusion branches of Puiseux series, for several values of $Q$, within the range $0\le Q\le 0.8$. The result has been shown in Fig.\[q\_c\_Sound\_diiusion\]. The typical behavior of the diffusion mode, shown by the red curve, seems to be qualitatively the same for the whole range of $Q$. The situation for sound mode, i.e. the blue curve, however, is more complicated. It should be studied in three different intervals; $(i)\, 0\le Q \le 0.386$, $(ii) \,0.386 \le Q \le 0.610$ and $(iii) \,0.610\le Q \le 0.8$.
To explore more on the relation between radius of convergence of the derivative expansion and $Q$ we now start to study the complex life of quasinormal modes. We assume ${{\mathfrak{q}}}^2$ to be a complex number ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta }$. Then $\theta=0$ simply corresponds to poles with real ${{\mathfrak{q}}}^2$, already shown in Fig.\[quasi\_EzZ0\_Q\_12\_fig\]. As before, we show them by (large) dots and (large) stars in upcoming figures. Then we let the phase $\theta$ change from $0$ to $2\pi$. For a given $|{{\mathfrak{q}}}^2|$, such change of $\theta$ corresponds to moving dots and stars along some trajectories in the complex ${{\mathfrak{w}}}$ plane. The interaction of poles via crossing of their trajectories is the main issue that we will discuss in details for each of the three intervals mentioned above, separately.
$\boldsymbol{(i)\,\, 0\le Q \le 0.386}$: In this interval we choose to show the results associated with $Q=0.3$. See Fig.\[Complex\_Sound\_Chennel\]. It is clear that by increasing $|{{\mathfrak{q}}}^2|$, trajectories of poles become more complicated. Let us firstly consider the highest large dots. These pols actually lie on the two sound branches of the Puiseux series near the origin. When $|{{\mathfrak{q}}}_c^2|\approx 2.32$, their trajectory collides with that of (dot) gapped mode poles, at two points (marked by black dots in the bottom row plots). One then concludes that at $Q=0.3$ the radius of convergence of the derivative expansion for the ${{\mathfrak{w}}}^{\pm}_{\text{sound}}$ is $|{{\mathfrak{q}}}^{\text{sound}}_c|\approx(2.32)^{1/2}\approx 1.52$.
![Poles of the retarded two-point function in the spin 0 channel at $Q=0.3$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. While at $|{{\mathfrak{q}}}^2|=0.60$ (top left panel) all poles follow closed orbits, at $|{{\mathfrak{q}}}^2|=1.00$ (top middle panel), the hydrodynamic diffusion pole and the two highest star gapped poles follow open orbits. It simply means that the dispersion relation ${{\mathfrak{w}}}_{\text{diffusion}}({{\mathfrak{q}}}^2)$ has branch point singularities in the complex momentum squared plane at $(0.60)^{1/2}<|{{\mathfrak{q}}}_c|<(1.00)^{1/2}$. It is clear that the point $(Q=0.5, |{{\mathfrak{q}}}_c|)$ then lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.50$ (top right panel) the orbits of the hydrodynamic sound pole and the two nearest dot poles are still closed. However, by further increasing $|{{\mathfrak{q}}}^2|$, their associated trajectories come close to each other. Finally they collide at the positions marked by black dots in the bottom row plots. The collision points are identified with critical value of momentum $|{{\mathfrak{q}}}_c^2|=2.32$. It is clear that the point $(Q=0.3, (2.32)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. After the collision, for instance at $|{{\mathfrak{q}}}^2|=2.34$, the orbits of sound pole and the two nearest dot gapped poles are no longer closed: four of them exchange their positions as the phase $\theta$ increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TT-}$**crossing**. []{data-label="Complex_Sound_Chennel"}](sound_3_10_first.pdf "fig:"){width="35.00000%"}![Poles of the retarded two-point function in the spin 0 channel at $Q=0.3$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. While at $|{{\mathfrak{q}}}^2|=0.60$ (top left panel) all poles follow closed orbits, at $|{{\mathfrak{q}}}^2|=1.00$ (top middle panel), the hydrodynamic diffusion pole and the two highest star gapped poles follow open orbits. It simply means that the dispersion relation ${{\mathfrak{w}}}_{\text{diffusion}}({{\mathfrak{q}}}^2)$ has branch point singularities in the complex momentum squared plane at $(0.60)^{1/2}<|{{\mathfrak{q}}}_c|<(1.00)^{1/2}$. It is clear that the point $(Q=0.5, |{{\mathfrak{q}}}_c|)$ then lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.50$ (top right panel) the orbits of the hydrodynamic sound pole and the two nearest dot poles are still closed. However, by further increasing $|{{\mathfrak{q}}}^2|$, their associated trajectories come close to each other. Finally they collide at the positions marked by black dots in the bottom row plots. The collision points are identified with critical value of momentum $|{{\mathfrak{q}}}_c^2|=2.32$. It is clear that the point $(Q=0.3, (2.32)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. After the collision, for instance at $|{{\mathfrak{q}}}^2|=2.34$, the orbits of sound pole and the two nearest dot gapped poles are no longer closed: four of them exchange their positions as the phase $\theta$ increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TT-}$**crossing**. []{data-label="Complex_Sound_Chennel"}](sound_3_10_second.pdf "fig:"){width="35.00000%"}![Poles of the retarded two-point function in the spin 0 channel at $Q=0.3$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. While at $|{{\mathfrak{q}}}^2|=0.60$ (top left panel) all poles follow closed orbits, at $|{{\mathfrak{q}}}^2|=1.00$ (top middle panel), the hydrodynamic diffusion pole and the two highest star gapped poles follow open orbits. It simply means that the dispersion relation ${{\mathfrak{w}}}_{\text{diffusion}}({{\mathfrak{q}}}^2)$ has branch point singularities in the complex momentum squared plane at $(0.60)^{1/2}<|{{\mathfrak{q}}}_c|<(1.00)^{1/2}$. It is clear that the point $(Q=0.5, |{{\mathfrak{q}}}_c|)$ then lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.50$ (top right panel) the orbits of the hydrodynamic sound pole and the two nearest dot poles are still closed. However, by further increasing $|{{\mathfrak{q}}}^2|$, their associated trajectories come close to each other. Finally they collide at the positions marked by black dots in the bottom row plots. The collision points are identified with critical value of momentum $|{{\mathfrak{q}}}_c^2|=2.32$. It is clear that the point $(Q=0.3, (2.32)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. After the collision, for instance at $|{{\mathfrak{q}}}^2|=2.34$, the orbits of sound pole and the two nearest dot gapped poles are no longer closed: four of them exchange their positions as the phase $\theta$ increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TT-}$**crossing**. []{data-label="Complex_Sound_Chennel"}](sound_3_10_third.pdf "fig:"){width="35.00000%"} ![Poles of the retarded two-point function in the spin 0 channel at $Q=0.3$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. While at $|{{\mathfrak{q}}}^2|=0.60$ (top left panel) all poles follow closed orbits, at $|{{\mathfrak{q}}}^2|=1.00$ (top middle panel), the hydrodynamic diffusion pole and the two highest star gapped poles follow open orbits. It simply means that the dispersion relation ${{\mathfrak{w}}}_{\text{diffusion}}({{\mathfrak{q}}}^2)$ has branch point singularities in the complex momentum squared plane at $(0.60)^{1/2}<|{{\mathfrak{q}}}_c|<(1.00)^{1/2}$. It is clear that the point $(Q=0.5, |{{\mathfrak{q}}}_c|)$ then lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.50$ (top right panel) the orbits of the hydrodynamic sound pole and the two nearest dot poles are still closed. However, by further increasing $|{{\mathfrak{q}}}^2|$, their associated trajectories come close to each other. Finally they collide at the positions marked by black dots in the bottom row plots. The collision points are identified with critical value of momentum $|{{\mathfrak{q}}}_c^2|=2.32$. It is clear that the point $(Q=0.3, (2.32)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. After the collision, for instance at $|{{\mathfrak{q}}}^2|=2.34$, the orbits of sound pole and the two nearest dot gapped poles are no longer closed: four of them exchange their positions as the phase $\theta$ increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TT-}$**crossing**. []{data-label="Complex_Sound_Chennel"}](sound_3_10_before.pdf "fig:"){width="42.00000%"}![Poles of the retarded two-point function in the spin 0 channel at $Q=0.3$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. While at $|{{\mathfrak{q}}}^2|=0.60$ (top left panel) all poles follow closed orbits, at $|{{\mathfrak{q}}}^2|=1.00$ (top middle panel), the hydrodynamic diffusion pole and the two highest star gapped poles follow open orbits. It simply means that the dispersion relation ${{\mathfrak{w}}}_{\text{diffusion}}({{\mathfrak{q}}}^2)$ has branch point singularities in the complex momentum squared plane at $(0.60)^{1/2}<|{{\mathfrak{q}}}_c|<(1.00)^{1/2}$. It is clear that the point $(Q=0.5, |{{\mathfrak{q}}}_c|)$ then lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.50$ (top right panel) the orbits of the hydrodynamic sound pole and the two nearest dot poles are still closed. However, by further increasing $|{{\mathfrak{q}}}^2|$, their associated trajectories come close to each other. Finally they collide at the positions marked by black dots in the bottom row plots. The collision points are identified with critical value of momentum $|{{\mathfrak{q}}}_c^2|=2.32$. It is clear that the point $(Q=0.3, (2.32)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. After the collision, for instance at $|{{\mathfrak{q}}}^2|=2.34$, the orbits of sound pole and the two nearest dot gapped poles are no longer closed: four of them exchange their positions as the phase $\theta$ increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TT-}$**crossing**. []{data-label="Complex_Sound_Chennel"}](sound_3_10_after.pdf "fig:"){width="42.00000%"}
Since the above-mentioned collision happens between two (dot) poles which both belong to the stress tensor spectrum, we call the crossing of the associated trajectories the $\boldsymbol{TT}-$**crossing**. As a result, within the range $0\le Q \le 0.386$, this is $TT-$crossing which determines the radius of convergence of the derivative expansion for the sound branch. As pointed out in Fig.\[q\_c\_Sound\_diiusion\], the AdS-Schwarzschild case with $Q=0$, studied in [@Grozdanov:2019kge], falls into the same range.
![Poles of the retarded two-point function in the spin 0 channel at $Q=0.5$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. While at $|{{\mathfrak{q}}}^2|=0.75$ (top left panel) all poles follow closed orbits, at $|{{\mathfrak{q}}}^2|=1.50$ (top middle panel), the hydrodynamic diffusion pole and the two highest star gapped poles follow open orbits. It simply means that the dispersion relation of ${{\mathfrak{w}}}_{\text{diffusion}}({{\mathfrak{q}}})$ has branch point singularities in the complex momentum squared plane at $(0.75)^{1/2}<|{{\mathfrak{q}}}_c|<(1.50)^{1/2}$. It is clear that the point $(Q=0.5, |{{\mathfrak{q}}}_c|)$ lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.85$ (top right panel) the orbits of the hydrodynamic sound pole and the two nearest dot poles are still closed. But interestingly, the trajectory of the diffusion pole seems going to become closed again. At $|{{\mathfrak{q}}}^2|=2.02$ (bottom left panel), the orbit of diffusion pole is already closed, however it nearly collide the left sound pole at a position marked by black dot on the vertical axis. The collision point is identified with critical value of momentum $|{{\mathfrak{q}}}_c^2|=2.04$. As expected, the point $(Q=0.5, |{{\mathfrak{q}}}_c|=(2.04)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. After the collision, for instance at $|{{\mathfrak{q}}}^2|=2.06$, the orbits of sound poles and the diffusion pole are no longer closed. They exchange their positions cyclically as the phase increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TJ-}$**crossing**.[]{data-label="Complex_Sound_Chennel_1_2"}](sound_1_2_first.pdf "fig:"){width="35.00000%"}![Poles of the retarded two-point function in the spin 0 channel at $Q=0.5$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. While at $|{{\mathfrak{q}}}^2|=0.75$ (top left panel) all poles follow closed orbits, at $|{{\mathfrak{q}}}^2|=1.50$ (top middle panel), the hydrodynamic diffusion pole and the two highest star gapped poles follow open orbits. It simply means that the dispersion relation of ${{\mathfrak{w}}}_{\text{diffusion}}({{\mathfrak{q}}})$ has branch point singularities in the complex momentum squared plane at $(0.75)^{1/2}<|{{\mathfrak{q}}}_c|<(1.50)^{1/2}$. It is clear that the point $(Q=0.5, |{{\mathfrak{q}}}_c|)$ lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.85$ (top right panel) the orbits of the hydrodynamic sound pole and the two nearest dot poles are still closed. But interestingly, the trajectory of the diffusion pole seems going to become closed again. At $|{{\mathfrak{q}}}^2|=2.02$ (bottom left panel), the orbit of diffusion pole is already closed, however it nearly collide the left sound pole at a position marked by black dot on the vertical axis. The collision point is identified with critical value of momentum $|{{\mathfrak{q}}}_c^2|=2.04$. As expected, the point $(Q=0.5, |{{\mathfrak{q}}}_c|=(2.04)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. After the collision, for instance at $|{{\mathfrak{q}}}^2|=2.06$, the orbits of sound poles and the diffusion pole are no longer closed. They exchange their positions cyclically as the phase increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TJ-}$**crossing**.[]{data-label="Complex_Sound_Chennel_1_2"}](sound_1_2_second.pdf "fig:"){width="35.00000%"}![Poles of the retarded two-point function in the spin 0 channel at $Q=0.5$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. While at $|{{\mathfrak{q}}}^2|=0.75$ (top left panel) all poles follow closed orbits, at $|{{\mathfrak{q}}}^2|=1.50$ (top middle panel), the hydrodynamic diffusion pole and the two highest star gapped poles follow open orbits. It simply means that the dispersion relation of ${{\mathfrak{w}}}_{\text{diffusion}}({{\mathfrak{q}}})$ has branch point singularities in the complex momentum squared plane at $(0.75)^{1/2}<|{{\mathfrak{q}}}_c|<(1.50)^{1/2}$. It is clear that the point $(Q=0.5, |{{\mathfrak{q}}}_c|)$ lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.85$ (top right panel) the orbits of the hydrodynamic sound pole and the two nearest dot poles are still closed. But interestingly, the trajectory of the diffusion pole seems going to become closed again. At $|{{\mathfrak{q}}}^2|=2.02$ (bottom left panel), the orbit of diffusion pole is already closed, however it nearly collide the left sound pole at a position marked by black dot on the vertical axis. The collision point is identified with critical value of momentum $|{{\mathfrak{q}}}_c^2|=2.04$. As expected, the point $(Q=0.5, |{{\mathfrak{q}}}_c|=(2.04)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. After the collision, for instance at $|{{\mathfrak{q}}}^2|=2.06$, the orbits of sound poles and the diffusion pole are no longer closed. They exchange their positions cyclically as the phase increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TJ-}$**crossing**.[]{data-label="Complex_Sound_Chennel_1_2"}](sound_1_2_third.pdf "fig:"){width="35.00000%"} ![Poles of the retarded two-point function in the spin 0 channel at $Q=0.5$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. While at $|{{\mathfrak{q}}}^2|=0.75$ (top left panel) all poles follow closed orbits, at $|{{\mathfrak{q}}}^2|=1.50$ (top middle panel), the hydrodynamic diffusion pole and the two highest star gapped poles follow open orbits. It simply means that the dispersion relation of ${{\mathfrak{w}}}_{\text{diffusion}}({{\mathfrak{q}}})$ has branch point singularities in the complex momentum squared plane at $(0.75)^{1/2}<|{{\mathfrak{q}}}_c|<(1.50)^{1/2}$. It is clear that the point $(Q=0.5, |{{\mathfrak{q}}}_c|)$ lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.85$ (top right panel) the orbits of the hydrodynamic sound pole and the two nearest dot poles are still closed. But interestingly, the trajectory of the diffusion pole seems going to become closed again. At $|{{\mathfrak{q}}}^2|=2.02$ (bottom left panel), the orbit of diffusion pole is already closed, however it nearly collide the left sound pole at a position marked by black dot on the vertical axis. The collision point is identified with critical value of momentum $|{{\mathfrak{q}}}_c^2|=2.04$. As expected, the point $(Q=0.5, |{{\mathfrak{q}}}_c|=(2.04)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. After the collision, for instance at $|{{\mathfrak{q}}}^2|=2.06$, the orbits of sound poles and the diffusion pole are no longer closed. They exchange their positions cyclically as the phase increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TJ-}$**crossing**.[]{data-label="Complex_Sound_Chennel_1_2"}](sound_1_2_before.pdf "fig:"){width="42.00000%"}![Poles of the retarded two-point function in the spin 0 channel at $Q=0.5$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. While at $|{{\mathfrak{q}}}^2|=0.75$ (top left panel) all poles follow closed orbits, at $|{{\mathfrak{q}}}^2|=1.50$ (top middle panel), the hydrodynamic diffusion pole and the two highest star gapped poles follow open orbits. It simply means that the dispersion relation of ${{\mathfrak{w}}}_{\text{diffusion}}({{\mathfrak{q}}})$ has branch point singularities in the complex momentum squared plane at $(0.75)^{1/2}<|{{\mathfrak{q}}}_c|<(1.50)^{1/2}$. It is clear that the point $(Q=0.5, |{{\mathfrak{q}}}_c|)$ lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.85$ (top right panel) the orbits of the hydrodynamic sound pole and the two nearest dot poles are still closed. But interestingly, the trajectory of the diffusion pole seems going to become closed again. At $|{{\mathfrak{q}}}^2|=2.02$ (bottom left panel), the orbit of diffusion pole is already closed, however it nearly collide the left sound pole at a position marked by black dot on the vertical axis. The collision point is identified with critical value of momentum $|{{\mathfrak{q}}}_c^2|=2.04$. As expected, the point $(Q=0.5, |{{\mathfrak{q}}}_c|=(2.04)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. After the collision, for instance at $|{{\mathfrak{q}}}^2|=2.06$, the orbits of sound poles and the diffusion pole are no longer closed. They exchange their positions cyclically as the phase increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TJ-}$**crossing**.[]{data-label="Complex_Sound_Chennel_1_2"}](sound_1_2_after.pdf "fig:"){width="42.00000%"}
$\boldsymbol{(ii)\,\, 0.386\le Q \le 0.610}$: In this interval we choose to show the results associated with $Q=0.5$. See Fig.\[Complex\_Sound\_Chennel\_1\_2\]. By increasing $|{{\mathfrak{q}}}|^2$, the poles tend to collide. Interestingly, it turns out that the first collision of the sound poles (the highest dots) is with the other hydro pole in this channel, namely the diffusion pole denoted by a star on imaginary axis. The collision point is identified with critical value of momentum $|{{\mathfrak{q}}}_c|^2\approx 2.04$. One then concludes that at $Q=0.5$ the radius of convergence of the derivative expansion for the sound branch of Puiseux series is $|{{\mathfrak{q}}}^{\text{sound}}_c|\approx(2.04)^{1/2}\approx 1.43$. Let us emphasize that while one of the two colliding poles comes from the stress tensor spectrum, the other one belongs to the spectrum of charge current. Therefore it is reasonable to call such crossing of trajectories the $\boldsymbol{TJ-}$**crossing**.
![Poles of the retarded two-point function in the spin 0 channel at $Q=0.7$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. At $|{{\mathfrak{q}}}^2|=0.75$ (top left panel) all poles follow closed orbits. By increasing $|{{\mathfrak{q}}}^2|$, the trajectory of the hydrodynamic diffusion pole comes close to those of the nearest star gapped poles. At $|{{\mathfrak{q}}}^2|=1.48$ (top middle panel) they nearly collide at the positions marked by black dots. The actual collision point is identified with the critical value of the momentum $|{{\mathfrak{q}}}_{c}^{2}|\approx 1.50$. It is clear that the point $(Q=0.7, |{{\mathfrak{q}}}_c|=(1.50)^{1/2})$ lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.52$ (top right panel), the orbits of the diffusion pole and the two highest star gapped poles are no longer closed; these poles cyclically exchanges their positions as the phase $\theta$ increases from $0$ to $2\pi$. This is actually the manifestation of $\boldsymbol{JJ-}$**crossing**. By further increasing $|{{\mathfrak{q}}}^2|$, the trajectory of sound poles come close to the that of the nearest gapped pole on the imaginary axis. At $|{{\mathfrak{q}}}^2|=1.93$ (bottom left panel) they nearly collide at the positions marked by black dots. The actual collision point is identified with the critical value of the momentum $|{{\mathfrak{q}}}_{c}^{2}|\approx 1.95$. As expected, the point $(Q=0.7, |{{\mathfrak{q}}}_c|=(1.95)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.96$ (bottom right panel) they exchange their positions cyclically as the phase increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TJ-}$**crossing**.[]{data-label="Complex_Sound_Chennel_7_10"}](sound_7_10_first.pdf "fig:"){width="35.00000%"}![Poles of the retarded two-point function in the spin 0 channel at $Q=0.7$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. At $|{{\mathfrak{q}}}^2|=0.75$ (top left panel) all poles follow closed orbits. By increasing $|{{\mathfrak{q}}}^2|$, the trajectory of the hydrodynamic diffusion pole comes close to those of the nearest star gapped poles. At $|{{\mathfrak{q}}}^2|=1.48$ (top middle panel) they nearly collide at the positions marked by black dots. The actual collision point is identified with the critical value of the momentum $|{{\mathfrak{q}}}_{c}^{2}|\approx 1.50$. It is clear that the point $(Q=0.7, |{{\mathfrak{q}}}_c|=(1.50)^{1/2})$ lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.52$ (top right panel), the orbits of the diffusion pole and the two highest star gapped poles are no longer closed; these poles cyclically exchanges their positions as the phase $\theta$ increases from $0$ to $2\pi$. This is actually the manifestation of $\boldsymbol{JJ-}$**crossing**. By further increasing $|{{\mathfrak{q}}}^2|$, the trajectory of sound poles come close to the that of the nearest gapped pole on the imaginary axis. At $|{{\mathfrak{q}}}^2|=1.93$ (bottom left panel) they nearly collide at the positions marked by black dots. The actual collision point is identified with the critical value of the momentum $|{{\mathfrak{q}}}_{c}^{2}|\approx 1.95$. As expected, the point $(Q=0.7, |{{\mathfrak{q}}}_c|=(1.95)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.96$ (bottom right panel) they exchange their positions cyclically as the phase increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TJ-}$**crossing**.[]{data-label="Complex_Sound_Chennel_7_10"}](sound_7_10_second.pdf "fig:"){width="35.00000%"}![Poles of the retarded two-point function in the spin 0 channel at $Q=0.7$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. At $|{{\mathfrak{q}}}^2|=0.75$ (top left panel) all poles follow closed orbits. By increasing $|{{\mathfrak{q}}}^2|$, the trajectory of the hydrodynamic diffusion pole comes close to those of the nearest star gapped poles. At $|{{\mathfrak{q}}}^2|=1.48$ (top middle panel) they nearly collide at the positions marked by black dots. The actual collision point is identified with the critical value of the momentum $|{{\mathfrak{q}}}_{c}^{2}|\approx 1.50$. It is clear that the point $(Q=0.7, |{{\mathfrak{q}}}_c|=(1.50)^{1/2})$ lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.52$ (top right panel), the orbits of the diffusion pole and the two highest star gapped poles are no longer closed; these poles cyclically exchanges their positions as the phase $\theta$ increases from $0$ to $2\pi$. This is actually the manifestation of $\boldsymbol{JJ-}$**crossing**. By further increasing $|{{\mathfrak{q}}}^2|$, the trajectory of sound poles come close to the that of the nearest gapped pole on the imaginary axis. At $|{{\mathfrak{q}}}^2|=1.93$ (bottom left panel) they nearly collide at the positions marked by black dots. The actual collision point is identified with the critical value of the momentum $|{{\mathfrak{q}}}_{c}^{2}|\approx 1.95$. As expected, the point $(Q=0.7, |{{\mathfrak{q}}}_c|=(1.95)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.96$ (bottom right panel) they exchange their positions cyclically as the phase increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TJ-}$**crossing**.[]{data-label="Complex_Sound_Chennel_7_10"}](sound_7_10_third.pdf "fig:"){width="35.00000%"} ![Poles of the retarded two-point function in the spin 0 channel at $Q=0.7$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. At $|{{\mathfrak{q}}}^2|=0.75$ (top left panel) all poles follow closed orbits. By increasing $|{{\mathfrak{q}}}^2|$, the trajectory of the hydrodynamic diffusion pole comes close to those of the nearest star gapped poles. At $|{{\mathfrak{q}}}^2|=1.48$ (top middle panel) they nearly collide at the positions marked by black dots. The actual collision point is identified with the critical value of the momentum $|{{\mathfrak{q}}}_{c}^{2}|\approx 1.50$. It is clear that the point $(Q=0.7, |{{\mathfrak{q}}}_c|=(1.50)^{1/2})$ lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.52$ (top right panel), the orbits of the diffusion pole and the two highest star gapped poles are no longer closed; these poles cyclically exchanges their positions as the phase $\theta$ increases from $0$ to $2\pi$. This is actually the manifestation of $\boldsymbol{JJ-}$**crossing**. By further increasing $|{{\mathfrak{q}}}^2|$, the trajectory of sound poles come close to the that of the nearest gapped pole on the imaginary axis. At $|{{\mathfrak{q}}}^2|=1.93$ (bottom left panel) they nearly collide at the positions marked by black dots. The actual collision point is identified with the critical value of the momentum $|{{\mathfrak{q}}}_{c}^{2}|\approx 1.95$. As expected, the point $(Q=0.7, |{{\mathfrak{q}}}_c|=(1.95)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.96$ (bottom right panel) they exchange their positions cyclically as the phase increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TJ-}$**crossing**.[]{data-label="Complex_Sound_Chennel_7_10"}](sound_7_10_before.pdf "fig:"){width="42.00000%"}![Poles of the retarded two-point function in the spin 0 channel at $Q=0.7$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. At $|{{\mathfrak{q}}}^2|=0.75$ (top left panel) all poles follow closed orbits. By increasing $|{{\mathfrak{q}}}^2|$, the trajectory of the hydrodynamic diffusion pole comes close to those of the nearest star gapped poles. At $|{{\mathfrak{q}}}^2|=1.48$ (top middle panel) they nearly collide at the positions marked by black dots. The actual collision point is identified with the critical value of the momentum $|{{\mathfrak{q}}}_{c}^{2}|\approx 1.50$. It is clear that the point $(Q=0.7, |{{\mathfrak{q}}}_c|=(1.50)^{1/2})$ lies on the red curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.52$ (top right panel), the orbits of the diffusion pole and the two highest star gapped poles are no longer closed; these poles cyclically exchanges their positions as the phase $\theta$ increases from $0$ to $2\pi$. This is actually the manifestation of $\boldsymbol{JJ-}$**crossing**. By further increasing $|{{\mathfrak{q}}}^2|$, the trajectory of sound poles come close to the that of the nearest gapped pole on the imaginary axis. At $|{{\mathfrak{q}}}^2|=1.93$ (bottom left panel) they nearly collide at the positions marked by black dots. The actual collision point is identified with the critical value of the momentum $|{{\mathfrak{q}}}_{c}^{2}|\approx 1.95$. As expected, the point $(Q=0.7, |{{\mathfrak{q}}}_c|=(1.95)^{1/2})$ lies on the blue curve in Fig.\[q\_c\_Sound\_diiusion\]. At $|{{\mathfrak{q}}}^2|=1.96$ (bottom right panel) they exchange their positions cyclically as the phase increases from $0$ to $2\pi$. This is the manifestation of the $\boldsymbol{TJ-}$**crossing**.[]{data-label="Complex_Sound_Chennel_7_10"}](sound_7_10_after.pdf "fig:"){width="42.00000%"}
$\boldsymbol{(iii)\,\, 0.610\le Q \le 0.8}$: In this interval we choose to show the results associated with $Q=0.7$. See Fig.\[Complex\_Sound\_Chennel\_7\_10\]. At this value of $Q$, specifically, we illustrate crossings associated with critical points of both sound and diffusion branches of Puiseux series. It turns out that by increasing $|{{\mathfrak{q}}}^2|$, the first collision of diffusion pole would be with the nearest star gapped poles (see the black dots in top middle and top right panels). This occurs at $|{{\mathfrak{q}}}_c^2|\approx 1.50$. Since both colliding poles are star poles, we call such crossing of trajectories the $\boldsymbol{JJ-}$**crossing**.The first collision of the sound poles, however, will occur at a larger value of momentum. As can be seen in the bottom panels of Fig.\[Complex\_Sound\_Chennel\_7\_10\], at $|{{\mathfrak{q}}}_c^2|\approx 1.95$ sound poles collide with the nearest purely imaginary gapped pole. Since a dot pole collides with a star pole, such crossing of trajectories is a $TJ-$crossing.
As a result one concludes that at $Q=0.7$, the diffusion dispersion relation ${{\mathfrak{w}}}_{\text{diffusion}}({{\mathfrak{q}}}^2)$ converges for $|{{\mathfrak{q}}}|<(1.50)^{1/2}$ while the convergence of ${{\mathfrak{w}}}_{\text{sound}}({{\mathfrak{q}}})$ is for $|{{\mathfrak{q}}}|<(1.95)^{1/2}$.
Before ending this subsection let us give two comments regarding our results. *Firstly*, as mentioned implicitly earlier and also can be seen by the behavior of red curve in Fig.\[q\_c\_Sound\_diiusion\], the critical point of the diffusion branch, not only at $Q=0.7$ but also in the whole range of $0\le Q\le 0.8$ is of $JJ-$crossing type. *Secondly*, one may wonder if $TJ-$crossings associated with intervals $(ii)$ and $(iii)$ are the same. In fact within $(ii)$ the sound pole collides with the other hydro pole, namely the diffusion pole; while within $(iii)$ it collides with a gapped pole belonging to the spectrum of charge current. The $TJ-$crossing within the interval $(ii)$ is then quite novel in the sense that it implies convergence of the derivative expansion for $ 0.386\le Q \le 0.610$ is fully controlled by the hydrodynamic information. This is a new aspect of level-crossing specific to systems at finite chemical potential. In previous studies at vanishing $\mu$, level-crossing was found as the result of interplay between hydrodynamic and non-hydrodynamic poles.
Spin 1 channel
--------------
The typical spectrum of quasinormal modes in this channel was already shown in Fig.\[quasi\_ExZ1\_Q12\_fig\]. The spectrum includes poles associated with transverse component of stress tensor (represented by dots) as well as those associated with transverse component of charge current (represented by stars). We argued that there would exist only one branch of Puiseux series passing through $(0,0)$, namely the dispersion relation of the shear mode.
In this subsection we investigate how the radius of convergence of the derivative expansion for shear mode changes with $Q$. To this end, we numerically find the spin 1 spectrum of quasinormal modes at complex momenta. The result has been given in Fig.\[converegence\_shear\]. The smoothness of plot suggests that for the whole range of $0\le Q\le 0.8$, only one specific type of level-crossing corresponds to the critical point of ${{\mathfrak{w}}}_{\text{shear}}({{\mathfrak{q}}}^2)$.
{width="40.00000%"}
![Poles of the retarded two-point function in the spin 0 channel at $Q=0.3$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. At $|{{\mathfrak{q}}}^2|=0.25$ (top left panel) all poles follow closed orbits. By increasing $|{{\mathfrak{q}}}^2|$, firstly the two highest star gapped poles collide. After the collision, for instance at $|{{\mathfrak{q}}}^2|=0.75$ (top middle panel) their orbits are no longer closed; they exchange their positions cyclically as the phase $\theta$ increases from $0$ to $2\pi$. At $|{{\mathfrak{q}}}^2|=1.4$ (top right panel), the orbits of dot poles are still closed. By further increasing $|{{\mathfrak{q}}}^2|$, the trajectory of shear pole comes close top those of nearest dot gapped poles; at $|{{\mathfrak{q}}}^2|=2.69$, they nearly collide at the positions marked by black dots (bottom middle panel). Just after the collision, at $|{{\mathfrak{q}}}^2|=2.71$ (bottom right panel) the orbits of shear pole and the two highest dot gapped pole are no longer closed; three of them exchange their positions cyclically as $\theta$ increases from $0$ to $2\pi$. This is the manifestation of the level-crossing.[]{data-label="Complex_Shear_Chennel_3_10"}](Shear_310_first.pdf "fig:"){width="35.00000%"}![Poles of the retarded two-point function in the spin 0 channel at $Q=0.3$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. At $|{{\mathfrak{q}}}^2|=0.25$ (top left panel) all poles follow closed orbits. By increasing $|{{\mathfrak{q}}}^2|$, firstly the two highest star gapped poles collide. After the collision, for instance at $|{{\mathfrak{q}}}^2|=0.75$ (top middle panel) their orbits are no longer closed; they exchange their positions cyclically as the phase $\theta$ increases from $0$ to $2\pi$. At $|{{\mathfrak{q}}}^2|=1.4$ (top right panel), the orbits of dot poles are still closed. By further increasing $|{{\mathfrak{q}}}^2|$, the trajectory of shear pole comes close top those of nearest dot gapped poles; at $|{{\mathfrak{q}}}^2|=2.69$, they nearly collide at the positions marked by black dots (bottom middle panel). Just after the collision, at $|{{\mathfrak{q}}}^2|=2.71$ (bottom right panel) the orbits of shear pole and the two highest dot gapped pole are no longer closed; three of them exchange their positions cyclically as $\theta$ increases from $0$ to $2\pi$. This is the manifestation of the level-crossing.[]{data-label="Complex_Shear_Chennel_3_10"}](Shear_310_second.pdf "fig:"){width="35.00000%"}![Poles of the retarded two-point function in the spin 0 channel at $Q=0.3$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. At $|{{\mathfrak{q}}}^2|=0.25$ (top left panel) all poles follow closed orbits. By increasing $|{{\mathfrak{q}}}^2|$, firstly the two highest star gapped poles collide. After the collision, for instance at $|{{\mathfrak{q}}}^2|=0.75$ (top middle panel) their orbits are no longer closed; they exchange their positions cyclically as the phase $\theta$ increases from $0$ to $2\pi$. At $|{{\mathfrak{q}}}^2|=1.4$ (top right panel), the orbits of dot poles are still closed. By further increasing $|{{\mathfrak{q}}}^2|$, the trajectory of shear pole comes close top those of nearest dot gapped poles; at $|{{\mathfrak{q}}}^2|=2.69$, they nearly collide at the positions marked by black dots (bottom middle panel). Just after the collision, at $|{{\mathfrak{q}}}^2|=2.71$ (bottom right panel) the orbits of shear pole and the two highest dot gapped pole are no longer closed; three of them exchange their positions cyclically as $\theta$ increases from $0$ to $2\pi$. This is the manifestation of the level-crossing.[]{data-label="Complex_Shear_Chennel_3_10"}](Shear_310_third.pdf "fig:"){width="35.00000%"} ![Poles of the retarded two-point function in the spin 0 channel at $Q=0.3$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. At $|{{\mathfrak{q}}}^2|=0.25$ (top left panel) all poles follow closed orbits. By increasing $|{{\mathfrak{q}}}^2|$, firstly the two highest star gapped poles collide. After the collision, for instance at $|{{\mathfrak{q}}}^2|=0.75$ (top middle panel) their orbits are no longer closed; they exchange their positions cyclically as the phase $\theta$ increases from $0$ to $2\pi$. At $|{{\mathfrak{q}}}^2|=1.4$ (top right panel), the orbits of dot poles are still closed. By further increasing $|{{\mathfrak{q}}}^2|$, the trajectory of shear pole comes close top those of nearest dot gapped poles; at $|{{\mathfrak{q}}}^2|=2.69$, they nearly collide at the positions marked by black dots (bottom middle panel). Just after the collision, at $|{{\mathfrak{q}}}^2|=2.71$ (bottom right panel) the orbits of shear pole and the two highest dot gapped pole are no longer closed; three of them exchange their positions cyclically as $\theta$ increases from $0$ to $2\pi$. This is the manifestation of the level-crossing.[]{data-label="Complex_Shear_Chennel_3_10"}](Shear_310_fourth.pdf "fig:"){width="35.00000%"}![Poles of the retarded two-point function in the spin 0 channel at $Q=0.3$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. At $|{{\mathfrak{q}}}^2|=0.25$ (top left panel) all poles follow closed orbits. By increasing $|{{\mathfrak{q}}}^2|$, firstly the two highest star gapped poles collide. After the collision, for instance at $|{{\mathfrak{q}}}^2|=0.75$ (top middle panel) their orbits are no longer closed; they exchange their positions cyclically as the phase $\theta$ increases from $0$ to $2\pi$. At $|{{\mathfrak{q}}}^2|=1.4$ (top right panel), the orbits of dot poles are still closed. By further increasing $|{{\mathfrak{q}}}^2|$, the trajectory of shear pole comes close top those of nearest dot gapped poles; at $|{{\mathfrak{q}}}^2|=2.69$, they nearly collide at the positions marked by black dots (bottom middle panel). Just after the collision, at $|{{\mathfrak{q}}}^2|=2.71$ (bottom right panel) the orbits of shear pole and the two highest dot gapped pole are no longer closed; three of them exchange their positions cyclically as $\theta$ increases from $0$ to $2\pi$. This is the manifestation of the level-crossing.[]{data-label="Complex_Shear_Chennel_3_10"}](Shear_310_before.pdf "fig:"){width="35.00000%"}![Poles of the retarded two-point function in the spin 0 channel at $Q=0.3$, in the complex ${{\mathfrak{w}}}-$plane, at various values of the complexified momentum ${{\mathfrak{q}}}^2=|{{\mathfrak{q}}}^2|e^{i \theta}$. Large dots and large stars correspond to the poles with purely real momentum (i.e. at $\theta=0$). As $\theta$ increases from $0$ to $2\pi$, each pole moves counter-clockwise following the trajectory whose color changes continuously from blue to red. At $|{{\mathfrak{q}}}^2|=0.25$ (top left panel) all poles follow closed orbits. By increasing $|{{\mathfrak{q}}}^2|$, firstly the two highest star gapped poles collide. After the collision, for instance at $|{{\mathfrak{q}}}^2|=0.75$ (top middle panel) their orbits are no longer closed; they exchange their positions cyclically as the phase $\theta$ increases from $0$ to $2\pi$. At $|{{\mathfrak{q}}}^2|=1.4$ (top right panel), the orbits of dot poles are still closed. By further increasing $|{{\mathfrak{q}}}^2|$, the trajectory of shear pole comes close top those of nearest dot gapped poles; at $|{{\mathfrak{q}}}^2|=2.69$, they nearly collide at the positions marked by black dots (bottom middle panel). Just after the collision, at $|{{\mathfrak{q}}}^2|=2.71$ (bottom right panel) the orbits of shear pole and the two highest dot gapped pole are no longer closed; three of them exchange their positions cyclically as $\theta$ increases from $0$ to $2\pi$. This is the manifestation of the level-crossing.[]{data-label="Complex_Shear_Chennel_3_10"}](Shear_310_after.pdf "fig:"){width="35.00000%"}
To be more precise, in Fig.\[Complex\_Shear\_Chennel\_3\_10\], we have shown the situation of complexified poles at several values of $|{{\mathfrak{q}}}^2|$ associated with $Q=0.3$. As it is seen, by increasing $|{{\mathfrak{q}}}^2|$, firstly the highest star gapped poles collide. It does actually occur at some critical value of $0.25<|{{\mathfrak{q}}}_c^2|<0.75$. By further increasing (the top right panel and bottom left panel), trajectories of the highest dot gapped poles, which are still closed, come close to that of the diffusion pole, namely the single dot pole on the imaginary axis. The first critical point of the spectral curve on the shear branch of Puiseux series turns out to be at $|{{\mathfrak{q}}}_{c}^2|\sim2.70$. We have demonstrated the situation of poles just before and just after the collision in the bottom middle and bottom right panels, respectively. We then conclude that at $Q=0.3$, the dispersion relation of shear mode converges for $|{{\mathfrak{q}}}|<|{{\mathfrak{q}}}_{c}|\sim(2.70)^{1/2}\sim 1.64$.
We have checked that the crossings between trajectories at complex momenta are qualitatively independent of the value of $Q$. Thus in contrast to spin 0 channel, in the current channel the level-crossing corresponding to radius of convergence of the derivative expansion is only of one type. Since colliding poles both belong to spectrum of (vector components of) stress tensor fluctuations, we call such type of crossing point as the $TT-$crossing.
Let us recall one of the reasons for which we started the present work. As mentioned in the Introduction, ref. [@Withers:2018srf] finds the radius of convergence of the shear mode at a fixed value of $\mu/T$ in a $2+1$ dimensional system. Our result shows that at least for the range of $\mu/T$ studied in the current paper, radius of convergence of the shear mode in our $3+1$ dimensional system increases by $\mu/T$. Additionally as has been pointed out to in the figure, at $\mu=0$ we find a finite radius of convergence in full agreement with [@Grozdanov:2019kge].
Before ending this section we briefly review the results in the following table.
\[table one\]
Spin Type of mode Range of $Q$ Type of level-crossing Type of colliding poles
------ -------------- ------------------------ ------------------------ -------------------------
$0\le Q \le 0.386$ $TT$ hydro with non-hydro
Sound $0.386\le Q \le 0.610$ $TJ$ hydro with hydro
$0.610\le Q \le 0.8$ $TJ$ hydro with non-hydro
diffusion $0\le Q\le 0.8$ $JJ$ hydro with non-hydro
1 shear $0\le Q\le 0.8$ $TT$ hydro with non-hydro
: We have classified all different types of level-crossing in terms of spin, value of $Q$ and type of colliding poles. For example, the first row of the table is saying that sound mode is an eigen frequency for the spectral curve of spin 0 fluctuations. When $0\le Q \le 0.386$, derivative expansion associated with dispersion relation of sound converges; the radius of convergence is identified with the collision of a hydrodynamic mode, actually the sound pole, belonging to the spectrum of stress tensor $T$, and a non-hydrodynamic mode, which is actually gapped, belonging to the same spectrum, namely $T$. []{data-label="tdpvfd"}
Let us also denote that we did not study spin 2 channel in this section. The reason is that its corresponding spectral curve does not have any branch passing through the origin (see Fig.\[quasi\_Exy\]). The latter is equivalent to say that there is no any spin 2 hydrodynamic mode.
Quasinormal modes and the chaos point {#chaos}
=====================================
Earlier than establishing the relation of quasinormal modes at complex momenta with convergence of the derivative expansion, they were found to be related with quantum chaos. The first observation of such relation turns to ref. [@Grozdanov:2017ajz]. Before recalling the main idea of the latter reference, let us recall that the exponential decrease of the out-of-time-order-correlator (OTOC) $$\label{OTOC_0}
\langle \,V(t,\vec{x})W(0,0)V(t,\vec{x})W(0,0)\,\rangle_{\beta}=1-\frac{1}{N} \,e^{\lambda\left( (t-t_*)-\frac{x}{v_B}\right)}+\cdots$$ in a thermal large-$N$ system is the manifestation of quantum chaos [@LArkin; @Shenker:2013pqb; @kitaev:2014bwb; @Roberts:2014isa; @Shenker:2014cwa; @Maldacena:2015waa; @Davison:2016ngz; @Kukuljan:2017xag; @Ling:2016ibq; @Blake:2017qgd]. Here $V$ and $W$ are two generic few-body operators, $\lambda$ is the quantum Lyapunov exponent and $t_*$ is scrambling time. $v_{B}$ is the butterfly velocity. Equation can be also written in the form of a plane wave $$\label{OTOC}
\langle \,V(t,\vec{x})W(0,0)V(t,\vec{x})W(0,0)\,\rangle_{\beta}=1-\frac{1}{N} \,e^{-i \omega (t-t_*)+ i q |\vec{x}|}+\cdots$$ with purely imaginary values for both momentum and frequency $$\label{chaos_p}
\omega_{ch}= i \lambda,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,q_{ch}=i \frac{\lambda}{v_{B}}.$$ The point $(\omega_{ch}, q_{ch})$ is called the chaos point[^14].
In holography, is mapped onto the back reaction of an small amount of energy thrown towards the horizon of a two-sided black hole [@Shenker:2013pqb]. The resultant deformed geometry is described by a Dray-’t Hooft shock wave [@Dry]. But as has been explained in [@Dry2], the same geometry can be found from linearized Einstein equations too. In addition, shock wave solution deals with dynamics of energy-momentum in the boundary theory. Then one may conclude that the same information about quantum chaos and scrambling can be extracted from studying the linear perturbations of metric around the horizon. It was actually shown to be the case firstly in [@Grozdanov:2017ajz] and then in [@Blake:2018leo]. The statement is that at point $(\omega_*, q_*)=(i \lambda, i \lambda/v_B)$, one specific component of Einstein equations at the horizon becomes trivial. In the dual boundary theory, it corresponds to multi-valudeness of energy density response function at $(\omega_*, q_*)$. This point is called the pole-skipping point [@Blake:2018leo]. More precisely, this point lies on the analytically continued dispersion relation of the sound mode in the boundary theory [@Grozdanov:2017ajz]. The coincidence of $(\omega_{ch}, q_{ch})$ with $(\omega_*, q_*)$ can be regraded as the existence of a direct link between hydrodynamics and quantum chaos, at least in holographic systems[^15].
In the following two subsections we investigate on the relation between hydrodynamics and quantum chaos in our holographic model.
Chaos point from shock wave computations
----------------------------------------
In this subsection we analytically compute the chaos point in a system dual to a AdS$_5$ RN black brane. We exploit the result of ref. [@Blake:2017qgd]. In this reference, based on the shock wave propagation picture in the bulk of AdS$_5$ [@Shenker:2013pqb], the butterfly velocities for an anisotropic Q-lattice have been computed. Considering the background as $$\label{Blake_metric}
ds^2=\,-F(r)dt^2+\frac{dr^2}{F(r)}+h_{T}(r)\,(dx^2+dy^2)+h_{L}(r) \,dz^2,$$ it was shown that butterfly velocities in longitudinal, $L$, and transverses, $T$, directions are given as the following $$\label{Blake_butterfly}
v_{L}=\frac{2 \pi T}{\sqrt{h_L} m}\bigg|_{r_h},\,\,\,\,\,\,\,\,v_{T}=\frac{2 \pi T}{\sqrt{h_T} m}\bigg|_{r_h},\,\,\,\,\,\,\,\text{with}\,\,\,m^2=\,\pi T\left(\frac{2h'_T h_L+h'_l h_T}{h_Th_L}\right)\bigg|_{r_h}.$$ In our case $h_{T}(r)=h_L(r)=\,r^2$ and so $m=6\pi T/r_h$. The isotropic butterfly velocity then reads $$\boxed{
v_{B}=\sqrt{\frac{2-Q^2}{3}}}.$$ This result is an exact expression in the whole range of $Q$, including $Q=0$ case corresponding to AdS$_5$ Schwarzschild black brane [@Shenker:2013pqb], as well as $Q=\sqrt{2}$ case corresponding to the extremal AdS$_5$ RN black brane. Considering the fact that all systems with gravity dual saturate the chaos bound [@Maldacena:2015waa], the chaos point in our system is found to be given by $$\label{chaos_point}
({{\mathfrak{w}}}_{ch}, {{\mathfrak{q}}}_{ch})=\,\left(\frac{\omega_{ch}}{2\pi T}, \frac{q_{ch}}{2\pi T}\right)=\,\left(i ,\,\pm i \,\sqrt{\frac{3}{2-Q^2}}\,\right).$$ In next subsections, we find pole-skipping points and discuss their relation with .
Pole-skipping
-------------
As discussed earlier, pole-skipping points of energy density response function can be found from a near horizon analysis of perturbations associated with bulk fields. To this end, it is convenient to work with the ingoing Eddington-Finkelstein coordinates. In this system of coordinates, bulk solutions take the following general form $$\begin{aligned}
ds^2&=&2 dv dr- f(r)dv^2+ h(r)\left(dx^2+dy^2+\frac{}{}dz^2\right),\\
A&=&-\frac{\sqrt{3}q_b}{2 r^2}\left(dv-\frac{dr}{f(r)}\right),\end{aligned}$$ where $v$ is the ingoing time coordinate. Because of the relation between energy dynamics and quantum chaos in maximally chaotic systems [@Grozdanov:2017ajz; @Blake:2018leo], it is natural to look for pole-skipping points in spin 0 channel of fluctuations. But as was recently shown, the same information about chaos can be extracted from both spin 1 and spin 2 channels too, at least in a holographic system with vanishing $\mu$ [@Grozdanov:2019uhi]. Thus for completeness, we study pole-skipping in the spin 1 channel as well.
To clarify the notation, let us mention that we take metric and gauge field perturbations as $\delta g_{\mu\nu}(r,v,\vec{x})=\delta g_{\mu\nu}(r)e^{-i\omega v+i q z }$ and $\delta A_{\mu}(r,v,\vec{x})=\delta A_{\mu}(r)e^{-i\omega v+i q z}$. The near horizon expansions of these perturbations are given by $$\label{}
\delta g_{\mu\nu}(r)=\,\sum_{n=0}^{\infty}\delta g_{\mu \nu}^{(0)}\,(r-r_h)^n,\,\,\,\,\,\,\delta A_{\mu}(r)=\,\sum_{n=0}^{\infty}\delta A_{\mu}^{(0)}\,(r-r_h)^n.$$
### Spin 0 Channel {#spin-0-channel-3 .unnumbered}
By taking the momentum along the third boundary direction, spin 0 perturbations in the bulk are $\delta g_{vv}$, $\delta g_{vr}$, $\delta g_{rr}$, $\delta g_{vz}$, $\delta g_{rz}$, $\delta g_{xx}+\,\delta g_{yy}$, $\delta g_{zz}$, $\delta A_{r}$, $\delta A_v$ and $\delta A_z$. Following [@Blake:2018leo], we expand $E_{vv}$ component of Einstein equations about the horizon $r=r_h$; to linear order in perturbations, we arrive at $$\label{E_vv}
\left(q^2-\frac{3}{2}ih'(r_h)\omega\right)\delta g_{vv}^{(0)}- i\left (i \omega +\frac{1}{2}f'(r_h)\right)\left( 2q \delta g_{vz}^{(0)}+\frac{}{}\omega \delta g_{ii}^{(0)}\right)=\,0.$$ Using metric functions given around , equation at the horizon $r_h=R$ becomes $$E^{(0)}_{vv}\equiv\,\left(q^2-3 i R\,\omega\right)\delta g_{vv}^{(0)}- i \left(i \omega +\frac{}{}R(2-Q^2)\right)\left( 2q \delta g_{vz}^{(0)}+\frac{}{}\omega \delta g_{ii}^{(0)}\right)=\,0.$$ It is clear that this equation trivially holds at $$\label{pole_skipping_point}
({{\mathfrak{w}}}_*, {{\mathfrak{q}}}_*)=\,\left(\frac{\omega_*}{2\pi T}, \frac{q_*}{2\pi T}\right)=\,\left(i ,\,\pm i \,\sqrt{\frac{3}{2-Q^2}}\,\right).$$ Therefore, at exactly the above points the rank of matrix $E^{(0)}_{\mu \nu}$ decreases by one. The latter is equivalent to say that line of poles of energy density response function in the boundary suddenly skips at . Thus the point is nothing but the pole-skipping point of energy density response function [^16]. In the following, we will confirm it numerically.
![Left panel: Three branches of Puiseux series at purely imaginary momenta, associated with $Q=0.4$ in spin 0 channel. Obviously the points lie on the sound branches, manifesting the hydrodynamic origin of quantum chaos. Right panel: Comparison between the left panel dispersion relations with those of associated with a AdS$_5$ Schwarzschild black brane ($Q=0$). []{data-label="imaginary_dispersion_Spin_0"}](Chaos_Sound.pdf "fig:"){width="48.00000%"}![Left panel: Three branches of Puiseux series at purely imaginary momenta, associated with $Q=0.4$ in spin 0 channel. Obviously the points lie on the sound branches, manifesting the hydrodynamic origin of quantum chaos. Right panel: Comparison between the left panel dispersion relations with those of associated with a AdS$_5$ Schwarzschild black brane ($Q=0$). []{data-label="imaginary_dispersion_Spin_0"}](Chaos_Sound_Q_0_410.pdf "fig:"){width="42.00000%"}
More concretely, the above-mentioned pole-skipping points lie on a hydrodynamic dispersion relation associated with the energy density response function [@Blake:2017ris]. Let us recall that energy dynamics is related to the spin 0 channel. Hydrodynamic poles in this channel are the sound and the diffusion modes. However, according to , these modes become purely imaginary at purely imaginary values of ${{\mathfrak{q}}}$. One then takes a fixed value of $Q$ and numerically finds frequencies of these modes at several purely imaginary momenta other than the momentum of pole-skipping points. Then, interpolation between the resultant points gives the dispersion relations we look for.
In Fig.\[imaginary\_dispersion\_Spin\_0\], we have shown the results related to $Q=0.4$. In the left panel, we have shown that how pole-skipping points $$\label{pole_skipping_4_10}
({{\mathfrak{w}}}_*, {{\mathfrak{q}}}_*)=\,\left(i ,\,\pm i \,\sqrt{\frac{3}{2-Q^2}}\,\right)\bigg|_{Q=0.4}\approx\,(i,\,\pm1.27\,i).$$ lie on the sound branches. The same can be done for other values of $Q$. As a result, we understand that each of the chaos points in our system does always coincide with a point at which pole of the hydrodynamic sound mode is skipped. This result completes our discussion about the hydrodynamic origin of quantum chaos in a holographic system at finite chemical potential. In the right panel of Fig.\[imaginary\_dispersion\_Spin\_0\], we have compared our results with those associated with AdS$_5$ Schwarzschild case with $Q=0$ [@Grozdanov:2019uhi].
At this point one may ask: Does pole-skipping happen within the domain of validity of the hydrodynamic approximation? To answer this question, in Fig.\[copara\_chaos\_critical\], we compare the absolute value of momentum at the pole-skipping point, namely $|{{\mathfrak{q}}}_{*}|$, with the radius of convergence of the hydrodynamic dispersion relation $|{{\mathfrak{q}}}_c|$. It is seen that when $Q$ exceeds the critical value $Q_c\approx0.668$, $|{{\mathfrak{q}}}_{*}|$ lies outside of the domain of convergence of the derivative expansion. Is it physically reasonable?
Recalling $|{{\mathfrak{q}}}_{*}|=|{{\mathfrak{q}}}_{ch}|$, one notices that when $Q>Q_c$ the chaos point lies outside the regime of validity of the hydrodynamic derivative expansion too. This is exactly one of the reasons for which a quantum theory of hydrodynamics was constructed in ref. [@Blake:2017ris]. As explicitly mentioned by its authors, to capture the exponential behavior of , they constructed an effective hydrodynamic theory non-perturbativley in derivatives[^17]. Then our result can be interpreted as it follows. When $Q<Q_c$, the derivative expansion is sufficient to see pole-skipping while for the range $Q>Q_c$ a non-perturbative treatment is required to find it from energy density response function [@Blake:2017ris].
![Comparison between the convergence radius of the derivative expansion associated with hydrodynamic sound mode (denoted by blue curve, already given in Fig.\[q\_c\_Sound\_diiusion\]) and the absolute value of momentum corresponding to pole-skipping points (denoted by green curve). When $Q<Q_c$, pole-skipping and the chaos points lie within the domain of convergence of classical hydrodynamics. Thus they can be found perturbatively in a derivative expansion from energy density response function. However, when $Q>Q_c$, extracting any information about quantum chaos from energy density response function requires non-perturbative computations [@Blake:2017ris].[]{data-label="copara_chaos_critical"}](q_c_q_star_q_sound.pdf){width="60.00000%"}
A very explicit example of pole-skipping outside the regime of convergence of hydrodynamics was found in the self-dual graviton-axion model [@Grozdanov:2019uhi]. While the derivative expansion associated with dispersion relation of gapless mode in this theory diverges at the pole-skipping point, it was shown that the Borel resummation of the series converges at the same point. Such Borel resummation is in fact one kind of non-perturbative methods which we already argued should be used in the range $Q>Q_c$ [^18].
### Spin 1 Channel {#spin-1-channel-3 .unnumbered}
Spin 1 perturbations in the bulk are $\delta g_{vx}$, $\delta g_{rx}$, $\delta g_{rr}$, $\delta g_{zx}$, $\delta A_{r}$ and $\delta A_x$. Now the question is if there exists a combination of spin 1 components of dynamical equations at the horizon, namely $E^{(0)}_{\mu x}=0$ and $E^{(0)}_{x}=0$, which trivially holds at . In the vanishing $\mu$ limit, Maxwell’s equations decouple and one finds $E^{(0)}_{vx}+i \sqrt{2/3} E^{(0)}_{xz}$ to be the desired combination [@Grozdanov:2019uhi]. This expression becomes trivial at $(-{{\mathfrak{w}}}_*, \mp i {{\mathfrak{q}}}_*)$.
In our case with $\mu\ne0$, however, Maxwell’s equations are coupled to Einstein equations. One finds that the simplest possible combination close to what we want is $$\label{pole_skipping_eq_shear}
\left(E^{(0)}_{vx}+i\sqrt{\frac{2}{3}}\,\frac{1+Q^2}{\sqrt{1-Q^2/2}}\,\,E^{(0)}_{xz}+\frac{\sqrt{3}}{2}R\,Q\, E^{(0)}_{x}\right)\bigg|_{(-{{\mathfrak{w}}}_*, \mp i {{\mathfrak{q}}}_*)}=\,\frac{3}{2}Q^2\,\left(-3 \delta g_{vx}^{(0)}+i \sqrt{6-3Q^2} \delta g_{xz}^{(0)}\right).$$ Although at $Q=0$ this gives the decoupled equation $E^{(0)}_{vx}+i \sqrt{2/3} E^{(0)}_{xz}=0$, at a general value of $Q$ the left hand side combination is still coupled to metric perturbations of the right hand side [^19]. Thus at $Q\ne0$ we do not expect to see any pole-skipping in this channe, at least in the upper half of the complex frequency plane [^20]. We do not prove this statement explicitly but we just show that $(-{{\mathfrak{w}}}_*, \mp i {{\mathfrak{q}}}_*)$ do not correspond to pole-skipping of the shear dispersion relation. In Fig.\[Shear\_pole\_skipping\] it has been demonstrated for $Q=0.5$. The point
![In a neutral fluid, $Q=0$, hydrodynamic branch of poles in spin 1 channel has a pole-skipping point denoted by blue dot. This point turns out to be $(- {{\mathfrak{w}}}_*, \mp i{{\mathfrak{q}}}_*)$ and so includes the same information about the chaos point $({{\mathfrak{w}}}_*, {{\mathfrak{q}}}_*)$. However, when $Q\ne0$, it will no longer be the case. As an example, black dot denotes the point $(- {{\mathfrak{w}}}_*, \mp i{{\mathfrak{q}}}_*)$ evaluated at $Q=0.5$; it obviously does not lie on the line of poles associated with $Q=0.5$.[]{data-label="Shear_pole_skipping"}](Chaos_Shear_Orange.pdf){width="60.00000%"}
$$\label{pole_skipping_Shear}
(- {{\mathfrak{w}}}_*, \mp i{{\mathfrak{q}}}_*)=\,\left(-i , \,\sqrt{\frac{3}{2-Q^2}}\,\right)\bigg|_{Q=0.5}\approx\,(-i,\,1.31).$$
does not lie on the dispersion relation of the shear mode associated with $Q=0.5$.
Again, we emphasize that it is not in contradiction with statements of [@Blake:2017ris]. Even if the point was really a pole-skipping point in spin 1 channel, it would not have anything to do to with “hydrodynamic origin of quantum chaos”. The reason is that the points would lie in the lower half of the complex frequency plane while those being responsible for exponential growth of OTOC have $\text{Im}\, {{\mathfrak{w}}}>0$ and are located in the upper half plane.
Before ending this section let us denote that we intentionally do not study possible pole-skipping points in the spin 2 channel. Looking at eq. (4.53) in ref. [@Grozdanov:2019uhi], we understand that pole-skipping in this channel cannot happen in the upper half plane. As a result, existence or non-existence of such points is not related to quantum chaos at all and so will be unrelated to the scope of our discussion.
Review, Conclusion and Outlook {#conclusion}
==============================
In this paper we explored three aspects of quasinormal modes in a holographic system at finite chemical potential: $\textcircled{1}$ The dependence of the quasinormal mode spectra and also the hydrodynamic excitations on the value of $\mu/T$ (or equivalently on $Q$), at different channels of spin. $\textcircled{2}$ The collisions of hydrodynamic and non-hydrodynamic poles at complex momenta, and specifically changes in the pattern of collisions when $\mu/T$ varies. $\textcircled{3}$ The relation between pole-skipping of energy density response function and hydrodynamics in the system and also extraction of chaos information from energy dynamics by use of the derivative expansion.
Let us emphasize that the necessary tool for studying all above aspects was initially provided in . In fact besides all analytical and numerical methods and techniques used in the paper, it was equation by use of which we were able to produce most of our results, i.e. those associated with spin 0 and 1 channels.
In part $\textcircled{1}$, among all other results we think that finding analytic expressions for hydrodynamic excitations is of more importance. To best of our knowledge, hydrodynamic modes and were never found in the literature.
In part $\textcircled{2}$, our main results are definitely related to spin 0 channel. We have shown that convergence radius of the derivative expansion for the sound mode non-trivially depend on the value of $\mu/T$. In the vanishing $\mu$ limit, it has been shown that convergence of the derivative expansion both in momentum [@Grozdanov:2019kge; @Withers:2018srf] and in position space [@Heller:2020uuy] corresponds to the collision of hydrodynamic and non-hydrodynamic modes. However, for $\mu\ne0$ we have found that within the range $0.386\le Q\le0.610$ or equivalently $1.1\lesssim \mu/T\lesssim 2$, this is the collision between two hydrodynamic modes, namely sound and diffusion modes, which determines the radius of convergence for the sound dispersion relation. Then it would be interesting to investigate how to extract information about the non-hydrodynamic sectors from the large order behavior of derivative expansion, in the mentioned range. We leave study on this issue to a future work.
In part $\textcircled{3}$, we provided a new evidence for the hydrodynamic origin of quantum chaos. As we showed, for all values of $\mu/T$, the chaos points found from shock wave computations precisely coincide with the pole-skipping points of energy density response function. However comparison between momentum of the pole-skipping points with the radius of converegnece of sound mode leads to an interesting result. Let us recall that quantum chaos seems not to be captured by classical hydrodynamics. The reason is that hydrodynamics is applicable for time intervals $\Delta t$ much larger than $1/T$, while in a maximally chaotic system the exponential behavior in [@Maldacena:2015waa] can be captured when $\Delta t\sim 1/T$ [@Blake:2017ris]. It shows why a to-all-order non-perturbative theory of hydrodynamics is needed to describe the chaos. On the other hand, we have found that in the range $0\le \mu/T\lesssim 2.1$, chaos information can be extracted perturbatively in the standard hydrodynamic derivative expansion. This results is simply the consequence of our another result saying that in the mentioned range, pole-skipping happens in the domain of convergence of the hydrodynamic derivative expansion.
It should be noted that the discussion of previous paragraph associated with part $\textcircled{3}$, might implicitly confirm the result shown in Fig.\[q\_c\_Sound\_diiusion\] of part $\textcircled{2}$. Let us assume that the blue curve in that figure was monotonically increasing by $Q$. As a result, in Fig.\[copara\_chaos\_critical\], the green curve would totally be located below the blue one. Since an increase in $Q$ at a fixed $\mu$ is equivalent to a decrease in $T$, then one would conclude that even as $T\rightarrow 0$, where the derivative expansion definitely breaks down, the chaos point could be still found perturbatively in a derivative expansion. This is absolutely wrong. Consequently, the quantum nature of chaos forces the blue curve in Fig.\[q\_c\_Sound\_diiusion\] not to be always upper than the green one.
Based on the above discussion, it becomes more important to study the quasinormal modes at higher values of $Q$ than those studied in this paper [@navid:2020], specifically in low temperature regime which is in correspondence with a near extremal black hole [@Moitra:2020dal].
Acknowledgment {#acknowledgment .unnumbered}
==============
N.A. would like to thank Javad Tabatabaei for collaboration in the early stages of this work. N.A. would like to thank Casey Cartwright, Ali Davody and Matthias Kaminski for useful discussions and specially appreciate Matthias Kaminski’s suggestions on the draft. We are particularly grateful to Sa$\check{\text{s}}$o Grozdanov for reading the draft and providing valuable comments. We would also like to thank Marco Ruggieri, Gianluca Giuliani, Bonan Zhang and Shen-Song Wan for discussions on related topics. N.A. acknowledges Hong-Fei Zhang and Wen-Hui Long for their supports. The work of S.T. was supported in part by National Natural Science Foundation of China, under Grant No. 11575254. We thank Benjamin Withers for correcting our over-interpretation of his results in the first version of this paper.
Diffeomorphism and gauge transformations {#diff_gauge_App}
========================================
In the spin $0$ channel, all diff-gauge transformations are given as the following $$\label{gauge_trans_Spin_0}
\begin{split}
h_{tt}&\rightarrow h_{tt}+ 2 i \omega \,\xi_t(r)+\frac{\big(f(r)a(r)\big)'}{b(r)}\xi_r(r),\\
h_{tz}&\rightarrow h_{tz}-i q \, \xi_t(r)+ i \omega\, \xi_z(r),\\
h_{zz}&\rightarrow h_{zz}-2 i q \, \xi_z(r) -\frac{a'(r)}{b(r)} \xi_r(r),\\
h&\rightarrow h -\frac{a'(r)}{b(r)} \xi_r(r),\\
a_{t}&\rightarrow a_t+ i \omega\, \phi(r)-i \omega\, \frac{\tilde{q}\, c(r)}{a(r)f(r)} \xi_t(r)-\frac{\tilde{q}\,c'(r)}{b(r)} \xi_r(r),\\
a_{z}&\rightarrow a_z- i q\, \phi(r)+i q\, \frac{ \tilde{q}\,c(r)}{a(r)f(r)} \xi_t(r),
\end{split}$$ where $h=\sum_{\alpha}h_{\alpha \alpha}$ and $\alpha\in \{x,y\}$. The transformations in the spin $1$ channel are $$\label{gauge_trans_Spin_1}
\begin{split}
h_{t\alpha}&\rightarrow h_{t\alpha}+ i \omega \,\xi_{\alpha}(r),\\
h_{z\alpha}&\rightarrow h_{z\alpha}- i q \,\xi_{\alpha}(r),\\
a_{\alpha}&\rightarrow a_{\alpha}.
\end{split}$$ Finally in the tensor channel: $$\label{gauge_trans_Spin_2}
h_{\alpha\beta}-\frac{1}{2}h \delta_{\alpha \beta}\rightarrow h_{\alpha\beta}-\frac{1}{2}h \delta_{\alpha \beta}.$$
Frobenius solution {#Frobenius}
===================
In order to find the quasinormal modes in channel of spin 0 and 1, one has to find $Z$ and $E$ from two specific coupled second order differential equations. Following the explanations given in [@Kovtun:2005ev], the solutions obeying the ingoing boundary condition at the horizon can be represented as $$\begin{split}
Z(u)=&(1-(1+Q^2)u^2+Q^2 u^3)^{-i {{\mathfrak{w}}}/2}\sum_{n=0}^{\infty} a_n({{\mathfrak{w}}}, {{\mathfrak{q}}})(1-u)^n,\\
E(u)=&(1-(1+Q^2)u^2+Q^2 u^3)^{-i {{\mathfrak{w}}}/2}\sum_{n=0}^{\infty} b_n({{\mathfrak{w}}}, {{\mathfrak{q}}})(1-u)^n.
\end{split}$$ Substituting the above two solutions into the corresponding coupled differential equations, one finds the coefficients $a_n$ and $b_n$ of the series expansion, all in terms of $a_0$ and $b_0$. Then by imposing the Dirichlet boundary condition at $u=0$, $$Z(0)=\sum_{n=0}^{\infty} a_n({{\mathfrak{w}}}, {{\mathfrak{q}}})=\,0,\,\,\,\,\,\,\,\,\,
E(0)=\sum_{n=0}^{\infty} b_n({{\mathfrak{w}}}, {{\mathfrak{q}}})=\,0$$ one arrives at a system of coupled algebraic equations which can be formally written as $$\begin{split}
\mathcal{C}_{11}({{\mathfrak{w}}}, {{\mathfrak{q}}}) a_0+\,\mathcal{C}_{12}({{\mathfrak{w}}}, {{\mathfrak{q}}}) b_0=&\,0,\\
\mathcal{C}_{21}({{\mathfrak{w}}}, {{\mathfrak{q}}}) a_0+\,\mathcal{C}_{22}({{\mathfrak{w}}}, {{\mathfrak{q}}}) b_0=&\,0.
\end{split}$$ Quasinormal spectrum then can be determined by demanding the above equations have non-trivial solutions for $({{\mathfrak{w}}}, {{\mathfrak{q}}})$: $$\label{detreminant_numeric}
\mathcal{C}_{11}({{\mathfrak{w}}}, {{\mathfrak{q}}})\,\,\mathcal{C}_{22}({{\mathfrak{w}}}, {{\mathfrak{q}}})-\,\mathcal{C}_{12}({{\mathfrak{w}}}, {{\mathfrak{q}}}) ,\mathcal{C}_{21}({{\mathfrak{w}}}, {{\mathfrak{q}}})=\,0$$ and solving equation numerically; the latter is done by taking a sufficiently large but finite number of terms, $n$, in the sum.
Comparison with explicit hydrodynamic computations
==================================================
In this section we reproduce the hydrodynamic excitations found from the study of perturbations, given in and , via explicit hydrodynamic computations on the boundary. The hydrodynamic regime of a holographic charged fluid has been studied in the context of fluid/gravity duality [@Banerjee:2008th; @Erdmenger:2008rm] and also in [@Son:2006em]. Following [@Banerjee:2008th] we write the associated constitutive relations up to first order in the derivative expansion, in 4-dimensions, as the following $$\begin{split}
T_{\mu\nu}=&\,p(\eta_{\mu\nu}+4u_{\mu}u_{\nu})-2 \eta \sigma_{\mu\nu},\\
j_{\mu}=&\,n \,u_{\mu}-\mathfrak{D} \left(P_{\mu}^{\nu}\partial_{\nu}n+ \frac{}{}3 (u^{\lambda}\partial_{\lambda}u_{\mu})n\right),
\end{split}$$ with $P_{\mu\nu}=\eta_{\mu\nu}+u_{\mu}u_{\nu}$. The coefficients found in [@Banerjee:2008th; @Erdmenger:2008rm] can be written as $$p=\frac{(1+Q^2)}{16\pi G_5}\left(\frac{2\pi T}{2-Q^2}\right)^4,\,\,\,\,\,\,\,\,\,\,\eta=\frac{1}{16\pi G_5}\left(\frac{2\pi T}{2-Q^2}\right)^3$$ and $$n=\frac{\sqrt{3}Q}{16\pi G_5}\left(\frac{2\pi T}{2-Q^2}\right)^3,\,\,\,\,\,\,\,\,\,\,\mathfrak{D}=\frac{4-Q^4}{8\pi T(1+Q^2)}.$$ Now by perturbing the equilibrium state of the fluid $$u^{\mu}(x,t)=\big(1,\delta v_{x}(x,t),\delta v_{y}(x,t),\delta v_{z}(x,t)\big),\,\,\,\,T(x)=T+\delta T(x,t),\,\,\,\,\,Q(x,t)=Q+\delta Q(x,t)$$ we search for the plane wave solutions $e^{-i \omega t+i q x_3}$ from the hydro equations of motion $$\partial_{\nu}T^{\mu\nu}=\,0,\,\,\,\,\,\,\,\partial_{\mu} j^{\mu}=0.$$ To first order in perturbations, these equations take the following form $$\label{linear_eq}
\begin{split}
0=&\left({{\mathfrak{w}}}+\frac{i }{4 }\frac{2-Q^2}{1+Q^2}{{\mathfrak{q}}}^2\right)\delta v_{\perp}\,\,\,\,\,\,\,\,\,\,\,(\perp\in \{x,y\}),\\
0=&\,2(-2+Q^2)\bigg( T{{\mathfrak{q}}}\delta v_z-3 {{\mathfrak{w}}}\delta T\bigg)+9T Q\frac{2+Q^2}{1+Q^2} \,{{\mathfrak{w}}}\,\delta Q,\\
0=&\,\frac{2}{9}(-2+Q^2)T \bigg({{\mathfrak{w}}}+\frac{i }{3} \frac{2-Q^2}{1+Q^2}{{\mathfrak{q}}}^2\bigg)\delta v_z+2 (2-Q^2){{\mathfrak{q}}}\,\delta T+3 T Q\frac{2+Q^2}{1+Q^2}{{\mathfrak{q}}}\,\delta Q,\\
0=&\,T Q (-2+Q^2) \left(1+\frac{3i}{4}\frac{4-Q^4}{1+Q^2}{{\mathfrak{w}}}\right) {{\mathfrak{q}}}\, \delta v_z+3 Q(2-Q^2)\left({{\mathfrak{w}}}+\frac{i}{4}\frac{4-Q^4}{1+Q^2}{{\mathfrak{q}}}^2\right)\delta T\\
&\,\,\,\,\,\,\,+ T (2+5Q^2)\left({{\mathfrak{w}}}+\frac{i}{4}\frac{4-Q^4}{1+Q^2}{{\mathfrak{q}}}^2\right)\delta Q.
\end{split}$$ The corresponding eigen modes of these equations, namely the hydrodynamic modes, can be expressed in the representations of earlierly discussed $SO(2)$ group. In the spin 0 channel, we find the following three modes $$\begin{split}
{{\mathfrak{w}}}_{1,2}=&\pm\frac{1}{\sqrt{3}}{{\mathfrak{q}}}-\frac{i}{6}\left(\frac{2-Q^2}{1+Q^2}\right)\,{{\mathfrak{q}}}^2,\,\,\,\,\,\,\,\\
{{\mathfrak{w}}}_3=&-\frac{i}{4}\left(\frac{4-Q^4}{1+Q^2}\right){{\mathfrak{q}}}^2.
\end{split}$$ The first two ones are the *sound* modes and the third mode is the diffusive *R-charge* mode. These excitations are in fact the eigen modes of the three last equations in . It is obvious that at the limit $Q\rightarrow 0$, and up to the order $Q^2$, the above expressions reduce to those in .
From the first equation in , we find the spin 1 modes as it follows $${{\mathfrak{w}}}_{4,5}=-\frac{i}{4}\left(\frac{2-Q^2}{1+Q^2}\right){{\mathfrak{q}}}^2$$ which are the transverse *shear* modes. At the limit $Q\rightarrow 0$, and up to the order $Q^2$, this expression becomes the one found in .
Perturbative solution of equations and {#App_sol_HYDRO_SOUND}
=======================================
We list the solution functions $Z_{0}^{m.n}(u)$ and $E_{z}^{m,n}(u)$ according to ordering we find them through the perturbative solving the coupled equations : ($\textswab{y}={{\mathfrak{w}}}/{{\mathfrak{q}}}$) $$\label{}
\begin{split}
Z_{0}^{0,0}(u)&=\,C_1\,(1- 3 \textswab{y}^2+ u^2),\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,E_{z}^{0,0}(u)=C_2,\\
Z_{0}^{0,1}(u)&=\,-\frac{3 }{\textswab{y}}C_2\,(u^2-1),\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,E_{z}^{0,1}(u)=\,\textswab{y}\,C_1(u-1),\\
Z_{0}^{0,2}(u)&=\,\frac{1}{3\textswab{y}^2-2}C_1\left[5-9u^2+4u^3-\frac{}{}6\textswab{y}^2(1-2u^2+u^3)\right],\\
E_{z}^{0,2}(u)&=-\frac{3}{2}C_2\,(u-1),\\
Z_{0}^{1,0}(u)&=\,2 i \,C_1\,(u^2-1),\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,
E_{z}^{1,0}(u)=-\frac{i}{\textswab{y}^2}C_2\,\left[u-1+\textswab{y}^2\log\frac{u+1}{2}\right],\\\nonumber
Z_{0}^{1,1}(u)&=\,-\frac{i}{ \textswab{y}^3}C_2(u-1)\,\bigg[3\textswab{y}^2(u+1)\log \frac{u+1}{2}+2+3\textswab{y}^2-2u(u-1) \bigg], \,\,\,\,\,\,\,\,\,\,\,\,\,\\
E_{z}^{1,1}(u)&=\frac{i}{\textswab{y}}C_1\,\left[(\textswab{y}^2+1)(u-1)-\textswab{y}^2\log\frac{u+1}{2}\right],\nonumber
\end{split}$$ and $$\label{}
\begin{split}
Z_{0}^{1,2}(u)&=\,i C\bigg[-\frac{3}{2}(1+3\textswab{y}^2-3u^2)\log\frac{1+u}{2}\\\nonumber
& -\frac{u-1}{2(2\textswab{y}^2-3)}\bigg(8(u^2-u-1)+2\textswab{y}^2(u^2+4u+1)-3\textswab{y}^4(3+7u^2)+27\textswab{y}^6\bigg)\bigg],\\
E_{z}^{1,2}(u)&=-i C \left[(5+3 u) \log\frac{1+u}{2}+\frac{(u-1)(1+u+\textswab{y}^2 u)}{2\textswab{y}^2(1+u)}\right].
\end{split}$$ The perturbative solutions to are given by $$\label{}
\begin{split}
Z_{1}^{0,0}(u)&=\,C_3\,\left(1+\frac{i {{\mathfrak{q}}}^2}{2 {{\mathfrak{w}}}} (1-u^2)\right),\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,E_{x}^{1,0}(u)=C_4,\\
Z_{1}^{0,1}(u)&=\,\frac{3 }{2}C_4\,(1-u^2),\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,E_{x}^{1,1}(u)=\,i C_3\frac{{{\mathfrak{q}}}^2}{2 {{\mathfrak{w}}}}(1-u),\\
Z_{1}^{0,2}(u)&=\,iC_3 \frac{{{\mathfrak{q}}}^2}{2{{\mathfrak{w}}}}u^2(u-1),\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,E_{x}^{1,2}(u)=\,\frac{3}{2}C_4(u-1).
\end{split}$$
[10]{}
P. Kovtun, “Lectures on hydrodynamic fluctuations in relativistic theories,” J. Phys. A **45**, 473001 (2012) \[arXiv:1205.5040 \[hep-th\]\]. H. Liu and P. Glorioso, “Lectures on non-equilibrium effective field theories and fluctuating hydrodynamics,” PoS TASI [**2017**]{}, 008 (2018) \[arXiv:1805.09331 \[hep-th\]\].
S. Jeon and L. G. Yaffe, “From quantum field theory to hydrodynamics: Transport coefficients and effective kinetic theory,” Phys. Rev. D **53**, 5799-5809 (1996) \[arXiv:hep-ph/9512263 \[hep-ph\]\]. J. M. Maldacena, “The Large N limit of superconformal field theories and supergravity,” Int. J. Theor. Phys. **38**, 1113-1133 (1999) \[arXiv:hep-th/9711200 \[hep-th\]\]. E. Witten, “Anti-de Sitter space and holography,” Adv. Theor. Math. Phys. [**2**]{}, 253 (1998) \[hep-th/9802150\].
G. Policastro, D. T. Son and A. O. Starinets, “From AdS / CFT correspondence to hydrodynamics,” JHEP [**0209**]{}, 043 (2002) \[hep-th/0205052\]. P. Kovtun, D. T. Son and A. O. Starinets, “Viscosity in strongly interacting quantum field theories from black hole physics,” Phys. Rev. Lett. **94**, 111601 (2005) \[arXiv:hep-th/0405231 \[hep-th\]\]. D. T. Son and A. O. Starinets, “Hydrodynamics of r-charged black holes,” JHEP [**0603**]{}, 052 (2006) \[hep-th/0601157\].
J. Mas, JHEP **03**, 016 (2006) doi:10.1088/1126-6708/2006/03/016 \[arXiv:hep-th/0601144 \[hep-th\]\].
G. Policastro, D. T. Son and A. O. Starinets, “From AdS / CFT correspondence to hydrodynamics. 2. Sound waves,” JHEP [**0212**]{}, 054 (2002) \[hep-th/0210220\].
A. Buchel, J. T. Liu and A. O. Starinets, “Coupling constant dependence of the shear viscosity in N=4 supersymmetric Yang-Mills theory,” Nucl. Phys. B **707**, 56-68 (2005) \[arXiv:hep-th/0406264 \[hep-th\]\]. P. K. Kovtun and A. O. Starinets, “Quasinormal modes and holography,” Phys. Rev. D [**72**]{}, 086009 (2005) \[hep-th/0506184\].
R. Baier, P. Romatschke, D. T. Son, A. O. Starinets and M. A. Stephanov, “Relativistic viscous hydrodynamics, conformal invariance, and holography,” JHEP **04**, 100 (2008) \[arXiv:0712.2451 \[hep-th\]\].
S. Bhattacharyya, V. E. Hubeny, S. Minwalla and M. Rangamani, “Nonlinear Fluid Dynamics from Gravity,” JHEP **02**, 045 (2008) \[arXiv:0712.2456 \[hep-th\]\]. S. Grozdanov and A. O. Starinets, “On the universal identity in second order hydrodynamics,” JHEP **03**, 007 (2015) doi:10.1007/JHEP03(2015)007 \[arXiv:1412.5685 \[hep-th\]\].
J. Erdmenger, M. Haack, M. Kaminski and A. Yarom, “Fluid dynamics of R-charged black holes,” JHEP [**0901**]{}, 055 (2009) 5doi:10.1088/1126-6708/2009/01/055 \[arXiv:0809.2488 \[hep-th\]\].
N. Banerjee, J. Bhattacharya, S. Bhattacharyya, S. Dutta, R. Loganayagam, and P. Surowka, “Hydrodynamics from charged black branes,” JHEP [**1101**]{}, 094 (2011), \[arXiv:0809.2596 \[hep-th\]\].
Y. Bu and M. Lublinsky, “Linearized fluid/gravity correspondence: from shear viscosity to all order hydrodynamics,” JHEP **11**, 064 (2014) \[arXiv:1409.3095 \[hep-th\]\].
S. Grozdanov and N. Kaplis, “Constructing higher-order hydrodynamics: The third order,” Phys. Rev. D **93**, no.6, 066012 (2016) \[arXiv:1507.02461 \[hep-th\]\].
M. P. Heller, R. A. Janik and P. Witaszczyk, “Hydrodynamic Gradient Expansion in Gauge Theory Plasmas,” Phys. Rev. Lett. **110**, no.21, 211602 (2013) \[arXiv:1302.0697 \[hep-th\]\]. A. O. Starinets, “Quasinormal modes of near extremal black branes,” Phys. Rev. D **66**, 124013 (2002) \[arXiv:hep-th/0207133 \[hep-th\]\].
R. A. Janik and R. B. Peschanski, “Gauge/gravity duality and thermalization of a boost-invariant perfect fluid,” Phys. Rev. D **74**, 046007 (2006) \[arXiv:hep-th/0606149 \[hep-th\]\]. A. Buchel, M. P. Heller and J. Noronha, “Entropy Production, Hydrodynamics, and Resurgence in the Primordial Quark-Gluon Plasma from Holography,” Phys. Rev. D **94**, no.10, 106011 (2016) \[arXiv:1603.05344 \[hep-th\]\]. M. Baggioli and A. Buchel, “Holographic Viscoelastic Hydrodynamics,” JHEP **03**, 146 (2019) \[arXiv:1805.06756 \[hep-th\]\].
B. Withers, “Short-lived modes from hydrodynamic dispersion relations,” JHEP **06**, 059 (2018) \[arXiv:1803.08058 \[hep-th\]\]. M. P. Heller, A. Serantes, M. Spaliński, V. Svensson and B. Withers, “The hydrodynamic gradient expansion in linear response theory,” \[arXiv:2007.05524 \[hep-th\]\].
S. Grozdanov, P. K. Kovtun, A. O. Starinets and P. Tadić, “Convergence of the Gradient Expansion in Hydrodynamics,” Phys. Rev. Lett. [**122**]{}, no. 25, 251601 (2019) \[arXiv:1904.01018 \[hep-th\]\]. S. Grozdanov, P. K. Kovtun, A. O. Starinets and P. Tadić, “The complex life of hydrodynamic modes,” JHEP [**1911**]{}, 097 (2019) \[arXiv:1904.12862 \[hep-th\]\]. H. Kodama and A. Ishibashi, “Master equations for perturbations of generalized static black holes with charge in higher dimensions,” Prog. Theor. Phys. **111** (2004), 29-73 \[arXiv:hep-th/0308128 \[hep-th\]\].
X. H. Ge, Y. Matsuo, F. W. Shu, S. J. Sin and T. Tsukioka, “Density Dependence of Transport Coefficients from Holographic Hydrodynamics,” Prog. Theor. Phys. **120**, 833-863 (2008) doi:10.1143/PTP.120.833 \[arXiv:0806.4460 \[hep-th\]\]. Y. Matsuo, S. J. Sin, S. Takeuchi, T. Tsukioka and C. M. Yoo, “Sound Modes in Holographic Hydrodynamics for Charged AdS Black Hole,” Nucl. Phys. B **820** (2009), 593-619 \[arXiv:0901.0610 \[hep-th\]\].
A. Jansen, A. Rostworowski and M. Rutkowski, “Master equations and stability of Einstein-Maxwell-scalar black holes,” JHEP **12** (2019), 036 \[arXiv:1909.04049 \[hep-th\]\].
D. K. Brattan and S. A. Gentle, “Shear channel correlators from hot charged black holes,” JHEP **04**, 082 (2011) \[arXiv:1012.1280 \[hep-th\]\].
R. A. Davison and A. O. Starinets, “Holographic zero sound at finite temperature,” Phys. Rev. D **85**, 026004 (2012) \[arXiv:1109.6343 \[hep-th\]\]. D. K. Brattan and S. A. Gentle, “Shear channel correlators from hot charged black holes,” JHEP **04**, 082 (2011) \[arXiv:1012.1280 \[hep-th\]\].
M. Edalati, J. I. Jottar and R. G. Leigh, “Shear Modes, Criticality and Extremal Black Holes,” JHEP **04**, 075 (2010) \[arXiv:1001.0779 \[hep-th\]\]. M. Edalati, J. I. Jottar and R. G. Leigh, “Holography and the sound of criticality,” JHEP **10**, 058 (2010) \[arXiv:1005.4075 \[hep-th\]\].
K. Maeda, M. Natsuume and T. Okamura, “Viscosity of gauge theory plasma with a chemical potential from AdS/CFT,” Phys. Rev. D **73**, 066013 (2006) \[arXiv:hep-th/0602010 \[hep-th\]\].
S. Janiszewski and M. Kaminski, “Quasinormal modes of magnetic and electric black branes versus far from equilibrium anisotropic fluids,” Phys. Rev. D **93** (2016) no.2, 025006 \[arXiv:1508.06993 \[hep-th\]\].
I. Novak, J. Sonner and B. Withers, “Overcoming obstacles in nonequilibrium holography,” Phys. Rev. D **98**, no.8, 086023 (2018) \[arXiv:1806.08655 \[hep-th\]\]. S. Grozdanov, K. Schalm and V. Scopelliti, “Black hole scrambling from hydrodynamics,” Phys. Rev. Lett. **120** (2018) no.23, 231601 \[arXiv:1710.00921 \[hep-th\]\]. M. Blake, H. Lee and H. Liu, “A quantum hydrodynamical description for scrambling and many-body chaos,” JHEP [**1810**]{}, 127 (2018) \[arXiv:1801.00010 \[hep-th\]\]. M. Blake, R. A. Davison, S. Grozdanov and H. Liu, “Many-body chaos and energy dynamics in holography,” JHEP [**1810**]{}, 035 (2018) \[arXiv:1809.01169 \[hep-th\]\]. M. Kaminski, K. Landsteiner, J. Mas, J. P. Shock and J. Tarrio, “Holographic Operator Mixing and Quasinormal Modes on the Brane,” JHEP **02** (2010), 021 \[arXiv:0911.3610 \[hep-th\]\]. R. A. Janik, J. Jankowski and H. Soltanpanahi, “Quasinormal modes and the phase structure of strongly coupled matter,” JHEP **06**, 047 (2016) \[arXiv:1603.05950 \[hep-th\]\]. A. Nunez and A. O. Starinets, “AdS / CFT correspondence, quasinormal modes, and thermal correlators in N=4 SYM,” Phys. Rev. D [**67**]{}, 124013 (2003) \[hep-th/0302026\]. D. Birmingham, I. Sachs and S. N. Solodukhin, “Conformal field theory interpretation of black hole quasinormal modes,” Phys. Rev. Lett. [**88**]{}, 151301 (2002) \[hep-th/0112055\]. G. T. Horowitz and V. E. Hubeny, “Quasinormal modes of AdS black holes and the approach to thermal equilibrium,” Phys. Rev. D [**62**]{}, 024027 (2000) \[hep-th/9909056\].
N. Abbasi “Work in progress."
A. I. Larkin and Y. N. Ovchinnikov, “Quasiclassical Method in the Theory of Superconductivity,” Soviet Journal of Experimental and Theoretical Physics 28, 1200 (1969).
S. H. Shenker and D. Stanford, “Black holes and the butterfly effect,” JHEP [**1403**]{}, 067 (2014) \[arXiv:1306.0622 \[hep-th\]\].
A. Kitaev, “talk given at Fundamental Physics Prize Symposium, Nov. 10, 2014. and Stanford SITP seminars, Nov. 11 and Dec. 18, 2014.” (2014).
D. A. Roberts, D. Stanford and L. Susskind, “Localized shocks,” JHEP [**1503**]{}, 051 (2015) \[arXiv:1409.8180 \[hep-th\]\].
S. H. Shenker and D. Stanford, “Stringy effects in scrambling,” JHEP [**1505**]{}, 132 (2015) \[arXiv:1412.6087 \[hep-th\]\]. J. Maldacena, S. H. Shenker and D. Stanford, “A bound on chaos,” JHEP [**1608**]{}, 106 (2016) \[arXiv:1503.01409 \[hep-th\]\].
R. A. Davison, W. Fu, A. Georges, Y. Gu, K. Jensen and S. Sachdev, “Thermoelectric transport in disordered metals without quasiparticles: The Sachdev-Ye-Kitaev models and holography,” Phys. Rev. B [**95**]{}, no. 15, 155131 (2017) \[arXiv:1612.00849 \[cond-mat.str-el\]\]. I. Kukuljan, S. Grozdanov and T. Prosen, “Weak Quantum Chaos,” Phys. Rev. B [**96**]{}, no. 6, 060301 (2017) \[arXiv:1701.09147 \[cond-mat.stat-mech\]\]. Y. Ling, P. Liu and J. P. Wu, “Holographic Butterfly Effect at Quantum Critical Points,” JHEP [**1710**]{}, 025 (2017) \[arXiv:1610.02669 \[hep-th\]\]. M. Blake, R. A. Davison and S. Sachdev, “Thermal diffusivity and chaos in metals without quasiparticles,” Phys. Rev. D [**96**]{}, no. 10, 106008 (2017) \[arXiv:1705.07896 \[hep-th\]\].
T. Dray and G. ’t Hooft, “Effect Of Spherical Shells Of Matter On The Schwarzschild Black Hole," Commun. Math. Phys. 99, 613 (1985).
T. Dray and G. ’t Hooft, “The Gravitational Shock Wave of a Massless Particle," Nuclear Physics B253 (1985) 173-188.
F. M. Haehl and M. Rozali, “Effective Field Theory for Chaotic CFTs,” JHEP [**1810**]{}, 118 (2018) \[arXiv:1808.02898 \[hep-th\]\]. F. M. Haehl, W. Reeves and M. Rozali, “Reparametrization modes, shadow operators, and quantum chaos in higher-dimensional CFTs,” arXiv:1909.05847 \[hep-th\].
S. Grozdanov, “On the connection between hydrodynamics and quantum chaos in holographic theories with stringy corrections,” JHEP [**1901**]{}, 048 (2019) \[arXiv:1811.09641 \[hep-th\]\].
M. Blake, R. A. Davison and D. Vegh, “Horizon constraints on holographic Green’s functions,” arXiv:1904.12883 \[hep-th\].
M. Natsuume and T. Okamura, “Holographic chaos, pole-skipping, and regularity,” arXiv:1905.12014 \[hep-th\].
M. Natsuume and T. Okamura, “Nonuniqueness of Green’s functions at special points,” arXiv:1905.12015 \[hep-th\]. M. Natsuume and T. Okamura, “Pole-skipping with finite-coupling corrections,” arXiv:1909.09168 \[hep-th\].
X. Wu, “Higher curvature corrections to pole-skipping,” arXiv:1909.10223 \[hep-th\].
Y. Ahn, V. Jahnke, H. S. Jeong and K. Y. Kim, “Scrambling in Hyperbolic Black Holes: shock waves and pole-skipping,” arXiv:1907.08030 \[hep-th\].
W. Li, S. Lin and J. Mei, “Thermal diffusion and quantum chaos in neutral magnetized plasma,” Phys. Rev. D [**100**]{}, no. 4, 046012 (2019) \[arXiv:1905.07684 \[hep-th\]\].
N. Ceplak, K. Ramdial and D. Vegh, “Fermionic pole-skipping in holography,” arXiv:1910.02975 \[hep-th\].
S. Das, B. Ezhuthachan and A. Kundu, “Real Time Dynamics in Low Point Correlators,” arXiv:1907.08763 \[hep-th\]. N. Abbasi and J. Tabatabaei, “Quantum chaos, pole-skipping and hydrodynamics in a holographic system with chiral anomaly,” JHEP **03**, 050 (2020) \[arXiv:1910.13696 \[hep-th\]\].
Y. Liu and A. Raju, “Quantum Chaos in Topologically Massive Gravity,” \[arXiv:2005.08508 \[hep-th\]\].
Y. Ahn, V. Jahnke, H. S. Jeong, K. Y. Kim, K. S. Lee and M. Nishida, “Pole-skipping of scalar and vector fields in hyperbolic space: conformal blocks and holography,” \[arXiv:2006.00974 \[hep-th\]\].
U. Moitra, S. K. Sake and S. P. Trivedi, “Near-Extremal Fluid Mechanics,” \[arXiv:2005.00016 \[hep-th\]\].
[^1]: [email protected]
[^2]: [email protected]
[^3]: The same treatment with large order hydrodynamic gradients in potion space has been applied to FLRW universe in ref. [@Buchel:2016cbj] and to viscoelastic media in ref. [@Baggioli:2018bfa].
[^4]: This result seems to be in contrast to that of [@Heller:2013fn]. However in a very recent paper [@Heller:2020uuy], it was shown that hydrodynamic derivative expansion converges in the position space as well, if initial data have support in momentum space not exceeding a critical value.
[^5]: In the present discussion we do not consider hydrodynamic fluctuations. See refs. [@Kovtun:2012rj] and [@Glorioso:2018wxw] for inclusion of fluctuations in classical hydrodynamics.
[^6]: The master equations for AdS$_5$ RN black brane have been found in [@Ge:2008ak; @Matsuo:2009yu]. See also [@Jansen:2019wag] for the case of Einstein-Maxwell-Dilaton black branes.
[^7]: Every branch of the spectral curve at the vicinity of origin corresponds to a specific hydrodynamic mode ${{\mathfrak{w}}}({{\mathfrak{q}}}\rightarrow 0)=0$. The small-${{\mathfrak{q}}}$ expansion of each mode is given by a Puiseux series.
[^8]: In a different context, collision between hydrodynamic sound mode and the hydrodynamic diffusion mode was already observed in [@Novak:2018pnv].
[^9]: At this point we partly adopt the idea of ref. [@Kaminski:2009dh] to write the boundary action as the first line of . In this reference, a framework for calculating holographic Green’s functions from this bilinear action has been developed. However, what we are going to do is different from the framework of this reference. See also [@Janik:2016btb] for yet another treatment with coupled dynamics in the bulk.
[^10]: The extremal limit is identified with $Q=\sqrt{2}$.
[^11]: These equations are analogue of the formal equations ; however, here are specifically associated with spin 0 channel and with the hydrodynamic limit as well.
[^12]: In [@Matsuo:2009yu], using the master field method, only the first term of sound mode, namely ${{\mathfrak{w}}}=1/\sqrt3{{\mathfrak{q}}}$ has been found. Authors of [@Matsuo:2009yu] have also computed the first order hydrodynamic transport coefficients via using the relevant Kubo formulas.
[^13]: It should be noted that we did not explicitly write down the spectral curve equations anywhere in this paper. However it would be useful to note that in the hydrodynamic limit, the perturbative spectral curve of the spin 0 channel, to first order in gradients and second order in $Q$, is obtained by substituting into . In the spin 1 channel, it is simply given by the determinant of .
[^14]: Since throughout this paper subscript $c$ is used for frequencies and momenta associated with critical points of spectral curves, we then unconventionally use the subscript $ch$ to label the chaos point.
[^15]: Pole-skipping has been also derived as a general prediction of effective field theory in maximally chaotic systems in ref. [@Blake:2017ris]. This phenomenon has been explicitly shown to happen in 2-dim CFT at large central charge [@Haehl:2018izb] and recently in higher dimensions [@Haehl:2019eae]. See [@Grozdanov:2018kkt] for considering the stringy corrections.
[^16]: It has been recently shown that such behavior may also happen when studying perturbations of probe bulk fields other than metric itself [@Blake:2019otz]. However the resultant pole-skipping points in such cases lie totally in the lower half of $\text{Im} {{\mathfrak{w}}}- \text{Im} {{\mathfrak{q}}}$ plane. See also [@Natsuume:2019vcv; @Wu:2019esr; @Ahn:2019rnq; @Li:2019bgc; @Natsuume:2019sfp; @Natsuume:2019xcy; @Ceplak:2019ymw; @Das:2019tga; @Abbasi:2019rhy; @Liu:2020yaf; @Ahn:2020bks].
[^17]: At this point it is worth mentioning that by “hydrodynamic origin of quantum chaos” posed in [@Blake:2017ris], they actually mean a quantum hydrodynamic origin. However, what we have dealt with in this paper have been about classical hydrodynamics.
[^18]: We thank Sašo Grozdanov for bringing this example to our attention.
[^19]: One another decoupling case is $Q=\sqrt{2}$, i.e. the extremal case, in which $E_{xz}^{(0)}=0$ decouples from the rest of equations.
[^20]: We thank Sa$\check{\text{s}}$o Grozdanov for pointing this out to us.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The classical boundary-value problem of the Einstein field equations is studied with an arbitrary cosmological constant, in the case of a compact ($S^{3}$) boundary given a biaxial Bianchi-IX positive-definite three-metric, specified by two radii $(a,b).$ For the simplest, four-ball, topology of the manifold with this boundary, the regular classical solutions are found within the family of Taub-NUT-(anti)de Sitter metrics with self-dual Weyl curvature. For arbitrary choice of positive radii $(a,b),$ we find that there are three solutions for the infilling geometry of this type. We obtain exact solutions for them and for their Euclidean actions. The case of negative cosmological constant is investigated further. For reasonable squashing of the three-sphere, all three infilling solutions have real-valued actions which possess a “cusp catastrophe” structure with a non-self-intersecting “catastrophe manifold” implying that the dominant contribution comes from the unique real positive-definite solution on the ball. The positive-definite solution exists even for larger deformations of the three-sphere, as long as a certain inequality between $a$ and $b$ holds. The action of this solution is proportional to $-a^{3}$ for large $a\, (\sim b)$ and hence larger radii are favoured. The same boundary-value problem with more complicated interior topology containing a “bolt” is investigated in a forthcoming paper.'
author:
- |
\
M M Akbar[^1] & P D D’Eath[^2]\
\
Department of Applied Mathematics and Theoretical Physics,\
Centre for Mathematical Sciences,\
University of Cambridge,\
Wilberforce Road,\
Cambridge CB3 0WA,\
U.K.
title: '[**[Classical Boundary-value Problem in Riemannian Quantum Gravity and Self-dual Taub-NUT-(anti)de Sitter Geometries]{}**]{}'
---
DAMTP-2002-26
Introduction
============
In quantum gravity, as treated by the combined approaches of the Dirac canonical quantization and its dual, the Feynman path integral [@fm], one studies (for example) the amplitude for $n$ disconnected compact three-surfaces to have given spatial three-metrics, respectively $h^1_{ij}, h^2_{ij}, ...,h^n_{ij}.$ The amplitude is given formally by a path integral. In the simplest case $n = 1$, with only one connected three-surface, which could then be regarded as a spatial cross-section of a cosmological model, the amplitude is given by the ‘no-boundary’ or ‘Hartle-Hawking’ state [@HH] $${\Psi}_{HH}(h_{ij})
={\int}{\cal D}g_{{\mu}{\nu}}{\exp}(-I_{E}[g_{{\mu}{\nu}}]), \label{1}$$ where the integral is over Riemannian four-geometries $g_{{\mu}{\nu}}$ on compact manifolds-with-boundary, where the three-metric induced on the boundary agrees with the prescribed $h_{ij}$ above. Here $I_{E}$ denotes the Euclidean action [@Gibbons:1976ue] of the four-dimensional configuration, as in Eq.(\[29\]) below.
Naturally, in considering semi-classical approximations to this integral, one is led first to study the classical Riemannian boundary-value problem: namely, to find one (or more) Riemannian solutions to the classical Einstein field equations, possibly including a cosmological constant ${\Lambda},$ obeying $$R_{{\mu}{\nu}}={\Lambda}g_{{\mu}{\nu}}, \label{2}$$ which are smooth on the interior manifold, and which agree with the spatial three-metric $h_{ij}$ at the boundary. Below, we shall see examples in which (different) such classical solutions exist, for a class of choices of the three-metric $h_{ij}$ on a boundary diffeomorphic to the three-sphere $S^{3}$, on a manifold-with-boundary with the simplest possibility of a four-ball topology.
Here, the boundary-value problem is studied within the class of biaxial Bianchi-IX models [@EGH; @GP1], which may be written locally in the form $$ds^2 = dr^2 + a^2(r)({\sigma}_1^{~2}+{\sigma}_2^{~2})
+ b^2(r){\sigma}_3^{~2}. \label{3}$$ Here ${\sigma}_i,{\;}i=1,2,3,$ denote the left-invariant one-forms on $S^{3}$ (see, for example, [@GP1]). The boundary 3-metric at a given value $r_{0}$ of $r$ is then a Berger three-sphere [@ped; @sak], with intrinsic three-metric $$ds^2 = a^2(r_{0})({\sigma}_1^{~2}+{\sigma}_2^{~2})
+ b^2(r_{0}){\sigma}_3^{~2}. \label{4}$$ Subject to condition (\[2\]), the metrics (\[3\]) are known in a closed form – they are the well-known family of Riemannian Taub-NUT-(anti)de Sitter metrics [@Carter; @EGH; @GP1]. For the topologically simplest case that we are studying in this paper, we insist that the $S^{3}$ with the given intrinsic metric (\[4\]) bound a four-ball with smooth four-metric; this corresponds to requiring the Taub-NUT-(anti)de Sitter metrics to close with a regular “nut” and to have half-flat Weyl curvature. We find that the problem can be translated into an algebraic system of degree three which can be solved exactly to find the possible infilling geometries of this type and their actions for *any* boundary data $a,b$. Depending on $a,b,$ one may therefore have three real roots, or one real root together with one complex conjugate pair, for this third-degree equation. A similar study was carried out in [@Louko], for large three-volume and small anisotropy, assuming a positive cosmological constant. However, we have been able to find all complex- and real-valued self-dual Taub-NUT-(anti)de Sitter solutions on the 4-ball and their actions for arbitrary values of $(a,b)$ for both positive and negative cosmological constants. The case of zero cosmological constant can be obtained by taking the cosmological constant to zero (equivalently) from either the positive or the negative direction.
Self-dual Riemannian Einstein spaces with a negative cosmological constant and of biaxial Bianchi-IX type have also been studied, for example, by Pedersen [@ped], in connection with the conformal boundary-value problem with a three-sphere (topologically) at infinity. The more general case of generic (non-biaxial) Bianchi-IX models has been treated in this context by Hitchin [@Hit] and by Tod [@Tod].
Since a strictly negative cosmological constant ${\Lambda}$ is needed to have any hope of incorporating gauge theories of matter with local supersymmetry in a four-dimensional field theory [@PD; @WB], we have given a more detailed analysis for $\Lambda<0$; the real and complex solutions have been classified completely in terms of the boundary values $a,b$, and the numerical behaviour of the solutions and Euclidean action $I_{E}$ has been investigated in greater detail.
Einstein Biaxial Bianchi-IX Metrics of Riemannian Signature
===========================================================
The general form of a biaxial Bianchi-IX four-metric is given by Eq.(\[3\]). Such metrics are invariant under the group action of $SU(2) \times U(1)$, whose Lie algebra is isomorphic to that of $U(2)$. When one further imposes the Einstein equations with a $\Lambda$ term, one arrives at the two-parameter Taub-NUT-(anti)de Sitter family [@Carter; @EGH; @GP1] : $$ds^2=\frac{\rho^{2} - L^{2}}{\Delta} d\rho^{2}+ \frac{4
L^{2}\Delta}{\rho^{2}-L^{2}}(d\psi+\cos \theta\,
d\phi)^{2}+(\rho^{2}-L^{2})(d\theta^{2}+\sin^{2}\theta \,d\phi^{2}), \label{Taub}\label{5}$$ where $$\Delta=\rho^{2}-2M\rho +L^{2}+\Lambda \left( L^{4}+2
L^{2}\rho^{2}-\frac{1}{3}\rho^{4}\right).$$ Here $L$ and $M$ are the two parameters and $0\leq\theta\leq\pi$, $0\leq\phi\leq2\pi$, $0\leq\psi\leq4\pi/k$ ($k$ is a natural number). When $k=1$, the surfaces of constant $\rho$ are topologically $S^{3}$. The general form (\[5\]), however, is only valid for a coordinate patch for which $\Delta \ne
0$. In general $\Delta$ will have four roots. At the roots the metric degenerates to that of a round $S^{2}$, and each such root therefore corresponds to a two-dimensional set of fixed points of the Killing vector field $\partial/{\partial \psi}$. However, if a root occurs at $\rho=|L|$, the fixed-point set is zero-dimensional (as the two-sphere then collapses to a point). Such two- and zero-dimensional fixed point sets have been given the names “bolts” and “nuts” respectively [@GH]. The coordinate $\rho$ ranges continuously from a root of $\Delta$ until it encounters another root of $\Delta$, if there is any; otherwise $\rho$ ranges from the root to infinity. In general, the bolts of the above Taub-NUT family of metrics are not regular points of the metric. For them to be regular, the metric has to close smoothly at the bolts, for which the condition is [@Page]: $$\frac{d}{d\rho}\left(\frac{{\Delta}}{\rho^{2}-L^{2}}
\right)_{(\rho=\rho_{bolt})}=\frac{1}{2kL}.\label{7}$$ For finite $L\ne 0$ and $k=1$, condition (\[7\]) leads to the self-dual Taub-NUT-(anti)de Sitter metric (see below) and the Taub-Bolt-(anti)de Sitter metric [@EGH; @GP2], which reduce to the Euclidean Taub-NUT [@Hawk] and the Taub-Bolt metrics [@Page] respectively for $\Lambda=0$. For the limiting cases $L \rightarrow \infty$ and $L=0$, for which (\[5\]) is not well-defined, one can obtain regular solutions by suitable coordinate transformations and assigning correct periodicities to the coordinate parametrizing the $S^1$ fibre. For example, for $\Lambda=0$, the Eguchi-Hanson metric [@EH] and the Schwarzschild metric can be obtained from (\[5\]) in these limits [@Page]. In the next section we will encounter more examples of Bianchi-IX metrics that arise at these limits as we discuss the self-dual Taub-NUT-(anti)de Sitter solutions. For a recent discussion on obtaining different biaxial Bianchi-IX metrics as various limits of (\[5\]) see [@CGP].
Self-dual Weyl tensor and Regularity
------------------------------------
It will become evident below that the regular, positive-definite geometries of this type on a four-ball interior are Weyl half-flat, that is, having either self-dual or anti-self-dual Weyl curvature [@AHS; @GP2] which requires: $$M=\pm L(1+\frac{4}{3}\Lambda L^{2}) \label{8}$$ For either sign, such metrics are often called (by physicists) half-flat – a name which we will be using in this paper.
It is easy to check that, when the metric has a nut – or equivalently $\Delta$ has a (double) root at $\rho= L$ (or $\rho= -L$) – then the two parameters $L$ and $M$ are related by $$M= \pm L(1+\frac{4}{3}\Lambda L^{2}),\label{9}$$ which is precisely the condition above for self-duality of the Weyl tensor. One can further check that Eq.(\[7\]) is automatically satisfied at the nut of the metric (\[5\]). Therefore, within the Taub-NUT-(anti)de Sitter family, self-duality of the Weyl tensor will imply regularity at the origin. Without loss of generality, we will work with the positive sign of Eq.(\[8\]) (and (\[9\])). The condition (\[8\]) of half-flatness (or condition (\[9\]) of nut-regularity) implies that $\Delta$ simplifies: $$\Delta=(\rho - L)^{2}-\frac{1}{3}\Lambda(\rho+3 L)(\rho-L)^{3}.$$ One can note that the terms involving $\Lambda$ vanish as $\rho \rightarrow L$, and that the metric near the origin is that of the $\Lambda=0$ case, namely Hawking’s (Taub-NUT) solution [@Hawk]: $$ds^2=\left(\frac{\rho + L}{\rho - L}\right) d\rho^{2}+4 L^{2}\left(\frac{\rho - L}{\rho + L}\right) (d\psi+\cos \theta
d\phi)^{2}+(\rho^{2}-L^{2})(d\theta^{2}+\sin\theta^{2} d\phi^{2})\label{11}$$ Apart from the double root at $\rho=L$, $\Delta$ has two more roots or “bolts” at $\rho=\pm
\sqrt{(4 L^{2}+3 /\Lambda)}-L$. Beyond these two roots $\Delta$ is negative. So, the permissible range of $\rho$ is from $L$ to $\sqrt{(4 L^{2}+3 /\Lambda)}-L$ (when $L$ is positive) or from $L$ to $-\sqrt{(4
L^{2}+3 /\Lambda)}-L$ (when $L$ is negative). For a complete nonsingular metric, we have to check whether such points are regular, i.e., whether $$\left(\frac{2}{3}\frac{(L-\rho)(\rho^{2}+3L\rho+4 L^{2})+3 L}{(\rho+L)^{2}}\right)_{(\rho=\rho_{bolt})}=\frac{1}{2L}$$ which is identically satisfied at the nut (at $\rho=L$) as we have seen already. However, for $\rho=
\sqrt{(4 L^{2}+3/ \Lambda)}-L$, which is positive definite and greater than $L$, this would require: $$\frac{4}{3} \Lambda L- \frac{2}{3} \sqrt{(4\Lambda^{2} L^{2}+3
\Lambda)}=\frac{1}{2L}$$ which can only happen for $L$ negative ($=-\frac{1}{8}\sqrt{6/\Lambda}$) and hence we find a contradiction. (In other words, this would give a regular bolt at $\rho=\frac{7}{8}\sqrt{6/\Lambda}$ – on the “other side” of $\rho=L$ which is negative, and hence cannot be reached continuously starting from $L$.) The same argument applies for the other root at $\rho=
-\sqrt{(4 L^{2}+3/ \Lambda)}-L$. Therefore, for positive cosmological constant, the half-flat (equivalently, regular-nut) solution, has a singular bolt at finite $\rho$ which exists and cannot be made regular for any finite value of the parameter $L\ne 0$. However, in the limit $L \rightarrow 0$, one can obtain the standard round metric on $S^{4}$ by suitable coordinate transformations; this is regular everywhere and has a regular nut at each of the two poles. For the $L \rightarrow
\infty$ limit, on the other hand, one can obtain the regular Fubini-Study metric on $\bbbc P^{2}$. This is regular everywhere and has a regular nut at the origin and a regular bolt at infinity [@GP1].
The case with a negative cosmological constant is different. Writing $\Lambda= - \lambda $ ($\lambda $ is positive), one has $$\Delta=(\rho - L)^{2}+\frac{1}{3}\lambda(\rho+3 L)(\rho-L)^{3}.$$ $\Delta$ now has two roots at at $\rho=L$ and two others at $\rho=\pm
\sqrt{(4 L^{2}-3 \lambda)}-L$, which are beyond the permissible range of $\rho \ge L$ ($L$ positive) and $\rho \le L$ ($L$ negative). So, for a fixed negative $\Lambda$, the one-parameter family of half-flat solutions of the Taub-NUT-anti de Sitter type are necessarily regular for $\rho \in [L,\infty)$ for any finite, non-zero value of $L$. Using similar coordinate transformations as in the case of positive cosmological constant, one can obtain the canonical metric on $H^4$ and the Bergman metric on $\overline{\bbbc P^{2}}$ in the limits of $L \rightarrow 0$ and $L
\rightarrow \infty$ respectively, both of which are nut-regular at the origin.
So, for both $\Lambda >0$ and $\Lambda < 0$, in this half-flat Taub-NUT-(anti) de Sitter family any hypersurface of constant $\rho$ (which is a Berger sphere with a given 3-metric) bounds a four-ball with a regular 4-metric. The singular bolt at $\rho=
\pm \sqrt{(4 L^{2}+3/ \Lambda)}-L$, in the case of $\Lambda>0$, poses no problem as the Berger sphere lies between the (regular) nut and the (singular) bolt when we are interested in filling the Berger sphere with positive-definite regular metrics. It remains to see which classical 4-metrics of Taub-NUT-(anti)de Sitter type can be given on the four-ball inside a given Berger sphere. In other words, how many members in this one-parameter ($L$) family of metrics are there for which a given Berger sphere is a possible hypersurface at some constant $\rho$?
In addressing this question we can include complex solutions in the interior of the 3-sphere. For this we let $\rho$ and $L$ and hence the 4-metric be in general complex-valued as long as the 4-metric induces the prescribed positive-definite 3-metric on the boundary and is non-singular in the interior.[^3] We will return to this issue in the next section. Real positive-definite infilling 4-metrics therefore form a subclass of the complex-valued solutions to this boundary-value problem and do not necessarily exist for arbitrary boundary data. However, it is possible to find the necessary and sufficient condition on the boundary data for the existence of real positive-definite solutions as we will see in Sections 4 and 5.
Infilling Geometries and their Action
=====================================
Following the discussion above, the problem of finding classical infilling regular/(anti)self-dual Einstein metrics of Taub-NUT-(anti)de Sitter type on the four-ball bounded by a typical Berger sphere with two radii $(a,b)$ can be translated into solving the following two equations in $\rho$ and $L$: $$a^{2}-\rho^{2}+L^{2}=0, \label{15}$$ and $$b^{2}(\rho^{2}-L^{2})-4\,{L}^{2}\left (\left (\rho-L\right )^{2}-\frac{1}{3}\,\Lambda\,\left (\rho+3
\,L\right )\left (\rho-L\right )^{3}\right )=0. \label{16}$$ These are polynomial equations of degree two and six in the variables $\rho$ and $L$ and hence, by Bézout’s Theorem (see, for example, [@Reid]), would intersect at at most twelve points in $\bbbc^{2}$. However, trying to solve Eq.(\[15\])-(\[16\]) explicitly for $\rho$ and $L$ is not very easy and not to be recommended. However, there is a more “symmetric” approach, which will enable us to find explicit solutions, and first of all enables us to count the number of infilling geometries.
Number of Infilling Geometries
------------------------------
[**[Theorem:]{}**]{} *For any boundary data $(a,b)$, where $a$ and $b$ are the radii of two equal and one unequal axes of a Berger 3-sphere (squashed 3-sphere with two axes equal) respectively, there are precisely three (modulo “orientation”) self-dual Taub-NUT-(anti)de Sitter geometries which contain the 3-sphere as their boundary.[^4]*\
[**[Proof:]{}**]{} Note that the system of equations (\[15\])-(\[16\]) admits the discrete symmetry $(\rho, L)
\leftrightarrow (-\rho, -L)$. (This is what we mean by “orientation”). Make the substitution: $$\begin{array}{rcl}
x&=&\rho+L\\
y&=&\rho-L.
\end{array} \label{17}$$ The problem now reduces to solving $$a^{2}=xy \label{18}$$ and $$b^{2}=\frac{y}{3x}\left (x-y\right )^{2}\left (3-2\,\Lambda\,xy+\Lambda\,{
y}^{2}\right ).\label{19}$$ Note that this preserves the discrete symmetry, $(\rho, L)
\leftrightarrow (-\rho, -L)$ in $(x,y) \leftrightarrow
(-x,-y)$. Substitution of $x$ in Eq.(\[19\]) now gives the univariate equation: $$\Lambda y^{6}+(3-4 a^{2} \Lambda) y^4 + (5 a^{4} \Lambda -6 a^{2}) y^2
-(3a^{2}b^{2}-3a^{4}+ 2\,a^{6}\Lambda)=0 \label{20}$$ which is a cubic equation in $y^{2}$. The six solutions for $y$ therefore will appear in pairs of opposite signs. Since $a^{2}$ is positive, this implies that the six corresponding solutions of $(x,y)$ are of the form $(x_{1},y_{1})$, $(x_{2},y_{2})$, $(x_{3},y_{3})$ and $(-x_{1},-y_{1})$, $(-x_{2},-y_{2})$, $(-x_{3},-y_{3})$.
Since the set of solutions have the symmetry $(x_{i},y_i{}) \rightarrow
(-x_{i},-y_{i})$, by applying the transformation $(x,y) \rightarrow
(-x,-y)$, we would obtain no new solutions for $(x,y)$ (hence for $(\rho,L)$) and would reproduce the same set. Therefore there are six points in $\bbbc^{2}$ where the two polynomials meet, i.e., six solutions for $(\rho, L)$ which are related by the reflection symmetry $(\rho, L)
\leftrightarrow (-\rho, -L)$. Hence the number of geometries modulo orientation for any given boundary data $(a,\,b)$ is three. $\Box$
Real/Complex Roots and Real Infilling Metrics
---------------------------------------------
It is more convenient to rewrite Eq.(\[20\]) for $z=y^{2}$ in the form $$\Lambda z^{3}+(3-4 A \Lambda) z^2 + (5 A^{2} \Lambda -6 A) z
-(3AB-3A^{2}+ 2\,A^{3}\Lambda)=0 \label{21}$$ where we have denoted $(a^{2},b^{2})$ by $(A,B)$. It is easy to see that the real solutions for $(x,y)$ (and hence for $(\rho,L)$) come from the [*[positive]{}*]{} roots of Eq.(\[21\]). Depending on the boundary data $(A,B)$, the number of positive solutions for $z$ can range from zero to three. However, a positive solution of $z$ and, hence a real solution $(\tilde{\rho},\tilde{L})$, would not necessarily correspond to a [*[real positive-definite infilling geometry]{}*]{}. For it to qualify as a regular real solution in the interior, the metric should be well-defined for $\rho$ taking values on the real line from $\tilde{L}$ to $\tilde{\rho}$. As we discussed in Section $2.1$, this can happen if and only if $\tilde{\rho}$ and $\tilde{L}$ have the same sign and $|\tilde{\rho}|>|\tilde{L}|$. The latter condition is automatically satisfied as a consequence of $a^2$ being positive irrespective of whether the signs of $\tilde{\rho}$ and $\tilde{L}$ are similar or not. If $\tilde{\rho}$ and $\tilde{L}$ have opposite signs, the metric will not be positive definite for values of $\rho$ within the interval $(\tilde{L},-\tilde{L})$ on the real line, and will be singular at $\rho=-\tilde{L}$. Therefore, it would not qualify as a real positive-definite solution. Such solutions for $(\tilde{\rho},\tilde{L})$ should be considered to correspond to complex metrics where $\rho$ is generally complex-valued in the interior and real-valued on the given $S^{3}$ boundary and at $\tilde{L}$. (One can choose contours for $\rho$ in the complex plane which avoid the singularity on the real line at $-\tilde{L}$ and ensure regularity at $\tilde{L}$; see discussions in [@Louko].) We will discuss the dependence of positivity (and complexity) of $z$ on the boundary data $(A,B)$ in detail for $\Lambda<0$ in Section $5$ to explore the structure of (real) solutions and their actions and shall return to this issue. We will show that for a given boundary data $(A,B)$ the real regular infilling solution is unique and exists when a certain inequality holds between $A$ and $B$.
For negative or complex roots of Eq.(\[21\]), the interior metrics are necessarily complex-valued.
Explicit Solutions
------------------
We now write down explicit solutions for our boundary-value problem. Rewrite Eq.(\[21\]) with the conformal rescaling $$\begin{array}{rcl}
A |\Lambda| &\rightarrow& A,\\
B |\Lambda| &\rightarrow& B,\\
z |\Lambda| &\rightarrow& z,\\
\end{array}$$ so that $A,B$ are now dimensionless. For positive cosmological constant, this gives $$f(z):= z^{3}+(3-4A) z^2 - (5 A^{2} -6 A) z +(3AB-3A^{2}+2 A^{3})=0,\label{23}$$ and, for negative cosmological constant, $$g(z):= z^{3}-(4A +3) z^2 + (5 A^{2} +6 A) z +(3AB-3A^{2}-2 A^{3})=0\label{24}.$$ Since we will be more interested in the case of negative cosmological constant, we only write down the solutions of $g(z)=0$. These are $$z_{1}=\frac{1}{6}\,\sqrt [3]{Q}+{\frac {\frac{2}{3}\,{A}^{2}+4\,A+1}{\sqrt [3]{Q}}}+1+
\frac{4}{3}\,A,\label{sol1}$$ $$z_{2}=-\frac{\sqrt [3]{Q}}{12}-{\frac {\frac{1}{3}\,{A}^{2}+2A+3}{\sqrt [3]{Q}}}+
1+\frac{4}{3}\,A+\frac{i}{2}\,\sqrt {3}\left (\frac{1}{6}\,\sqrt [3]{Q}-{\frac {\frac{2}{3}\,{A}
^{2}+6\,A+6}{\sqrt [3]{Q}}}\right ),\label{sol2}$$ and $$z_{3}=z_{2}^{*},\label{sol3}$$ where $$Q=72\,{A}^{2}+8\,{A}^{3}+216\,A-324\,AB+216+36\,\sqrt {AB\left (81\,AB-108-108
\,A-36\,{A}^{2}-4\,{A}^{3}\right )}$$\
The solutions of $f(z)=0$ can be found by just flipping the signs of $A$ and $B$. It is important to note here that, the appearance of explicit factors of $i$ in the expressions (\[sol1\])-(\[sol3\]) for the roots may not be an accurate guide as to whether a root is real or complex. In Section $4$, we will see how to analyse the roots of $g(z)$ (in a way that would enable one to determine whether the roots are complex or real, for different values of $(A,
B)$) without invoking the explicit solutions (\[sol1\])-(\[sol3\]). However, we do need the explicit solutions in order to calculate the action for such geometries.
Action Calculation
------------------
The Euclidean action is given by [@Gibbons:1976ue]: $$8 \pi \,G \,I_{E} = -\int_{\a} d^4 x \sqrt{g} \,\left(\frac{1}{2} R -
\Lambda\right)- \int_{\partial \a} d^3x \sqrt{h}\, K \label{29}$$ where $\a$ is the Riemannian 4-manifold with boundary $\partial
\a$, and the 4-geometry $g_{\mu\nu}$ agrees with the specified three-metric $h_{ij}$ on the boundary, i.e., $g_{{ij}|_{\partial \a}}=h_{ij}$. $K$ is the trace of the extrinsic curvature $K_{ij}$ of the boundary.
For a classical solution, obeying the Euler-Lagrange equations of the action (\[29\]) (the Einstein equations $R_{\mu\nu}= \Lambda
g_{\mu\nu}$), one has $$8 \pi \,G \,I_{E} = -{\Lambda}\int_{\a} d^4 x \sqrt{g} - \int_{\partial \a} d^3x \sqrt{h}\, K.$$ Now $\int_{\a} d^4 x \sqrt{g}$ is just the volume of $\a$, which for the self-dual Taub-NUT-(anti)de Sitter metric gives $$-{\Lambda}\int_{\a} d^4 x \sqrt{g}=-\frac{32{\Lambda}}{3}\pi^2\,L\,(\rho-L)^2\,(\rho+2L)=-\frac{8{\Lambda}}{3}\,{\pi}^{2}\left
({y}^{2}-A\right )\left ({y}^{2}-3\,A\right ). \label{31}$$ The trace of the extrinsic curvature is $(2\frac{a'}{a}+\frac{b'}{b})/N$ for any metric of the form $$ds^2 = N(r)^{2}dr^2 + a^2(r)({\sigma}_1^{~2}+{\sigma}_2^{~2})
+ b^2(r){\sigma}_3^{~2}$$ giving $$- \int_{\partial \a} d^3x \sqrt{h}\,
K= -\frac{1}{N}\left(2\frac{a'}{a}+\frac{b'}{b}\right)(16 \pi^{2} a^{2}b).$$ For the self-dual Taub-NUT-(anti)de Sitter metric the surface term in Eq.(\[29\]) is therefore $$- \int_{\partial \a} d^3x \sqrt{h}\,
K=\frac{8}{3}{\pi}^{2}\,{\frac {\left ({y}^{2}-A\right )\left (\Lambda\,{y}^{4}+\Lambda\,A{y}^{2}+3\,{y
}^{2}+9\,A-8\,\Lambda\,{A}^{2}\right )}{A}}.$$ The total action is then: $$8 \pi \,G \,I_{E}=\frac{8}{3}\,{\frac {{\pi}^{2}\left ({y}^{2}-A\right )\left (-5\,\Lambda\,{A}
^{2}+9\,A+3\,{y}^{2}+\Lambda\,{y}^{4}\right )}{A}}$$ Using Eq.(\[21\]), $I_{E}$ can be further simplified to give: $$8 \pi \,G \,I_{E}=\frac{8}{3}\,{\pi}^{2}\left (7\,\Lambda\,{A}^{2}-12\,A+12\,z+3\,\Lambda\,{z}^{
2}-10\,\Lambda\,Az+3\,B\right ) \label{36}$$ which in terms of the scaled variables gives, in the case $\Lambda>0$ $$8 \pi \,G \, (\Lambda I_{E})=\frac{8}{3}\,{\pi}^{2}\left (7\,{A}^{2}-12\,A+12\,z+3\,{z}^{
2}-10\,Az+3\,B\right ) \label{37}$$ In the case of a negative cosmological constant, this is just: $$8 \pi \,G \, (\lambda I_{E})=-\frac{8}{3}\,{\pi}^{2}\left (7\,{A}^{2}+12\,A-12\,z+3\,{z}^{
2}-10\,Az-3\,B\right ) \label{38}$$ Clearly, the substitutions (\[17\]) have led to considerable simplifications. The action is a quadratic function of $z$ – the solution of the third-degree equation – and hence would, in general, be complex (real) for $z$ complex (real). Corresponding to every solution of Eq.(\[23\]) or Eq.(\[24\]), one can calculate the action merely by substitution in Eq.(\[37\]) or Eq.(\[38\]) – one does *not* need to find $y$ (or $x$) at all to evaluate the action. As we discussed in Section $3.2$, not all positive solutions for $z$ correspond to real positive-definite solutions in the interior. Therefore all real- and complex-valued solutions in the interior corresponding to the positive solutions for $z$ have real-valued Euclidean actions. This is because the two limits (here $\tilde{\rho}$ and $\tilde{L}$) of the integral (\[31\]) of $\rho$ lie on the real-line. Interestingly, for negative solutions for $z$, which correspond to $\tilde{\rho}$ and $\tilde{L}$ being purely imaginary, the actions are real as well. This is not surprising given that the middle expression of (\[31\]) is a quartic polynomial in $\rho$ and $L$. Only the complex solutions for $z$ would in general have complex actions.
Vacuum Case
===========
Before embarking on solutions with non-zero cosmological constant, it is desirable to understand the $\Lambda=0$ case. This will also provide us with an example and illustrate the utility of the method developed so far. For this case, the regular Taub-NUT metrics are given by the Hawking solution (\[11\]). The corresponding (quadratic) polynomial equation and actions are obtained by simply setting $\Lambda=0$ in Eq.(\[21\]) and Eq.(\[36\]) respectively. The two solutions and their actions are $$z=A\pm\sqrt{AB} \label{solvac}$$ and $$I_{E}= \frac{\pi}{G} (B \pm 4 \sqrt{AB}). \label{actvac}$$ Following the discussions in Section $3.2$, it is easy to check that the negative-sign solution of (\[solvac\]) corresponds to a real positive-definite metric on the four-ball provided that $$a>b. \label{vac}$$ This can be confirmed by solving for $\rho$ and $L$ directly (which is possible in this case). For the negative and positive signs of Eq.(\[solvac\]) they are $$\rho=\frac{a}{2}\frac{(2a-b)}{\sqrt{a^2-ab}},\,\,\,\,\,L=\frac{b}{2}\sqrt{\frac{a(a-b)}{a-b}},$$ and $$\rho=\frac{a}{2}\frac{(2a+b)}{\sqrt{a^2+ab}},\,\,\,\,\,L=-\frac{b}{2}\sqrt{\frac{a(a+b)}{a+b}}.$$ (Two other solutions are obtained by changing orientation.) Note that, in contrary to the statement made in [@Louko], positive-definite real infilling solutions do not exist for arbitrary boundary data although the actions (\[actvac\]) are real-valued for any value of $A$ and $B$. The issue of positive-definite real solutions on the four-ball will be made clearer in the following sections where we discuss the case of non-zero cosmological constant (which includes $\Lambda=0$ as a special case).
Solutions for $\Lambda < 0$ and their Actions
=============================================
We now treat the case of a negative cosmological constant in greater detail and systematically discuss the structure of the solutions of $$g(z):= z^{3}-(4A +3) z^2 + (5 A^{2} +6 A) z +(3AB-3A^{2}-2 A^{3})=0 \label{44}$$ as functions of boundary data $(A,B)$. Recall that $(\rho,L)$ can have a real solution only when $z$ is positive. Therefore, by studying the solutions of Eq.(\[44\]), we are able to see the real and complex solutions of this boundary-value problem quite readily. However, before doing so we discuss the possibility of real positive-definite infilling solutions.
Real, Positive-definite Infilling Solutions
-------------------------------------------
For a given boundary data $(A,B)$, let the infilling solutions be denoted by $(\tilde{\rho},\tilde{L})$. Recall from our discussion in Section $3.2$ that a real positive-definite infilling solution exists if $\tilde{\rho}$ and $\tilde{L}$ have like signs (and if $|\tilde{\rho|} >
|\tilde{L}|$ which holds automatically). Assuming without any loss of generality, that both $\tilde{\rho}$ and $\tilde{L}$ are positive, the equivalent requirement in the variables $x$, $y$ is $x > y$. It then follows from Eq.(\[18\]) that the corresponding requirement on the possible solutions for $z \equiv y^2$ is the following: $$z < a^2. \label{neweq1}$$ Therefore, only positive roots of Eq.(\[23\]) or Eq.(\[24\]) within the interval $ (0,A)$ give positive-definite real solutions on the four-ball which are regular everywhere including the origin, i.e., at $\rho= \tilde{L}$. Note that the requirement (\[neweq1\]) is independent of the size or sign of the cosmological constant.
We now analyse the case of negative cosmological constant and find the necessary and sufficient conditions on the boundary data for roots of Eq.(\[44\]) to satisfy the inequality (\[neweq1\]).\
[**[Lemma:]{}**]{} *There is a unique real, positive-definite self-dual Taub-NUT-anti de Sitter solution on the four-ball bounded by a given Berger-sphere of radii $(a,b)$, if and only if* $$b^2< \frac{1}{3}a^2\,(2a^2|\Lambda|+3). \label{cond}$$ [**[Proof:]{}**]{} Recall that, for any function $F(x)$, if $F(a)$ and $F(b)$ have unlike signs, then an odd number of roots of $F(x)=0$ lie between $a$ and $b$ (see, for example, [@BC]). Depending on the boundary data there are two distinct possibilities for the sign of the constant term in Eq.(\[44\]). For $B<
\frac{1}{3}A(2A+3)$, $g(0)$ is negative and $g(A)=3AB$ and is strictly positive. Therefore, for $B<
\frac{1}{3}A(2A+3)$, there will be either one root or three roots in the interval $(0,A)$. By direct differentiation, one can show that $g'(z)$ and $g''(z)$ are strictly non-negative within the interval $[0,A]$. It therefore follows from Fourier’s theorem that the number of roots in the interval $(0,A)$ is one.[^5] If, on the other hand, $B\ge
\frac{1}{3}A(2A+3)$, there will be no root within the interval $(0,A)$. $\Box$
Note that (\[cond\]) gives (\[vac\]) in the $|\Lambda|=0$ limit. There are certain particular features of the coefficients of $g(z)$ that simplified our analysis. Firstly, it is only in the constant ($z^{0}$) term that $B$ appears. Also, and more importantly, except for the constant term the coefficients of $g(z)$ are either positive or negative definite for arbitrary $(A,B)$. (In the case of a positive $\Lambda$, we are not fortunate enough to have this particular advantage – the corresponding analysis is therefore just a step lengthier, although the methodology developed in this section equally applies. Possible positive-definite real solutions are unique as in the case of negative cosmological constant.)
Structure of Roots
------------------
We now discuss how the roots of Eq.(\[44\]) change as the boundary data is varied. Since for real (complex) roots the action is real (complex) this would enable us to study the real (complex) actions. In fact, as we will see below, regions of physical interest are covered by real solutions of Eq.(\[44\]).
As is apparent from the previous discussions, we must look carefully at the two different cases, for which the constant term is positive or negative. The condition determining the reality of the roots of $g(z)$ is following (see, for example, [@AS]): $$\begin{array}{rcl}
B&<&\frac {4}{81}\,\frac{(A+3)^{3}}{A}\;\;\;\;\; \;\;\;\;\;
{\rm (three\;\;real\;\;roots)}\\
B&=&\frac {4}{81}\,\frac{(A+3)^{3}}{A}\;\;\;\;\; \;\;\;\;\;
{\rm (three\;\;real,\;\;at\;\;least\;\;two\;\; equal)}\\
B&>&\frac {4}{81}\,\frac{(A+3)^{3}}{A}\;\;\;\;\; \;\;\;\;\;
{\rm (one\;\; real\;\;two\;\;complex\;\;roots)}
\end{array} \label{realcon}$$ However, for the discussion below one further needs to know Descartes’ Rule [@BC]: *the equation $g(z)=0$ can not have more positive roots than $g(z)$ has changes of sign, or more negative roots than $g(-z)$ has changes of sign*. (It is also convenient to remember the elementary fact that the product of the roots of a polynomial equation of odd degree is minus the constant term.)
*Case $1$: $B< \frac{1}{3} A(3+2A)$* {#case-1-b-frac13-a32a .unnumbered}
------------------------------------
In this case $(3AB-3A^{2}-2 A^{3})$ is negative and hence $g(z)$ has three changes of sign and $g(-z)$ has none. Therefore, there will be either $(1)$ three positive roots of $g(z)$ or $(2)$ two complex and one positive roots, depending on which condition of (\[realcon\]) $(A,B)$ satisfies. One of the positive roots in the former case and the positive root in the latter case will correspond to real positive-definite regular solutions in the interior.
*Case $2$: $B> \frac{1}{3}
A(3+2A)$* {#case-2-b-frac13a32a .unnumbered}
--------------------------
In this case, $(3AB-3A^{2}-2 A^{3})$ is positive and hence $g(z)$ has two changes of sign and $g(-z)$ has one. So $g(z)$ has either $(1)$ two positive roots and one negative or $(2)$ two complex and one negative, depending on which condition of (\[realcon\]) $(A,B)$ satisfies.
*Case $3$: $B= \frac{1}{3} A(3+2A)$* {#case-3-b-frac13-a32a .unnumbered}
-------------------------------------
For this special case $(3AB-3A^{2}-2 A^{3})=0$ and hence the valid roots are the solutions of $$z^{2}-(4A + 3) z + (5 A^{2} +6 A) =0$$ which will have either two positive or two complex roots. (The other root is $z=0$ which is not physically meaningful.) However, neither of the positive roots, which exist if $A\le \frac{3}{2}$, will be less than $A$ and hence there is no real infilling solution in the interior.\
We now combine Descartes’ rule with the conditions of (\[realcon\]) (Fig. 1). We plot the
curves $B= \frac{1}{3} A(3+2A)$ (green line) and $B=\frac {4}{81}\,\frac{(A+3)^{3}}{A}$ (blue line). For physical interest, we have also plotted the line $A=B$ which represents isotropy. The two curves divides the $(B,A)$ plane in four regions.
Region I {#region-i .unnumbered}
--------
This is the region bounded between the two curves on the left-hand side of their intersection point, with the green line included. One root is negative and two others positive. The positive roots become equal on the blue line and the negative root become zero on the green line. None of the positive roots corresponds to a real infilling solution in the interior of the Berger sphere.
Region II {#region-ii .unnumbered}
---------
This is the region above the intersection point of the two curves. So there is one pair of complex roots and a negative root. On the green line there are only a pair of complex roots, the real one being zero.
Region III {#region-iii .unnumbered}
-----------
Two roots are complex and one is real and positive. The positive root corresponds to a real positive-definite infilling solution in the interior. As above, on the green line, the real root is zero.
Region IV {#region-iv .unnumbered}
----------
This is the region bounded by the two lines below their point of intersection, with the blue line included. All roots are positive and one of them corresponds to a real infilling solution. Two roots are equal on the blue line.
Numerical Study
---------------
In the previous section we have studied the general behaviour of the roots systematically. Although $B$ appears only in the constant term of $g(z)$, its role is rather crucial in determining the sign and reality of the roots. In this section we will study the roots of $g(z)$ and their corresponding actions numerically, and demonstrate an interesting connection with catastrophe theory.
First we study the roots and actions for fixed values of $A$, as functions of $B$. From the analysis in the previous section (and from Fig. 1), one would expect to study three distinct generic cases, determined by the value of $A$ less than, equal to and greater than its value where the two curves meet, i.e., at $A=3/2$ (and $B=3$).
Case 1: $A<3/2$ {#case-1-a32 .unnumbered}
---------------
For a fixed $A<3/2$, by varying $B$ one moves continuously from region IV (three positive roots) to the green line where one of the roots become zero (two still being positive), then into region I, where two roots continue to be positive, but the other is now negative, and then to the blue line where the positive roots become equal, and then to region II where they turn complex (the other still being negative).
The behaviour of the three roots and the corresponding Euclidean actions are plotted in Figs. 2 and 3 (as long as they remain real), as functions of $B$ for $A=0.2$ and $A=1$ respectively. (The same colour is used to show the correspondence.)
[{width="45.00000%"}]{}
[{width="45.00000%"}]{}
Case 2: $A=3/2$ {#case-2-a32 .unnumbered}
---------------
For the special value of $A=3/2$, all roots are positive in region IV, where, on the point that the green and blue lines meet two become equal and the other becomes zero. On entering region II the two equal roots become a complex-conjugate pair and the other turns negative. The behaviour of the three roots is therefore similar to the previous behaviour, except that they all turn complex and become negative at the same value ($B=3$) (Fig. $4$). The corresponding actions also have a structure similar to those in $A<3/2$.
[{width="45.00000%"}]{}
Case 3: $A>3/2$ {#case-3-a32 .unnumbered}
----------------
For $A>3/2$ one starts in region IV with three positive roots, two of which become equal on the blue line and then turn complex and remain so in region II. The other root remains positive in region III and becomes zero on the green line to become negative in region II. See Fig. $5$ ($A=3$) and Fig. $6$ ($A=100$).
[![Real Roots and corresponding Actions as functions of $B$ when $A>3/2$ ($A=3$)[]{data-label="..."}](root4.eps "fig:"){width="45.00000%"}]{}
[![Real Roots and corresponding actions as functions of $B$ for large $A$($A=100$)[]{data-label="..."}](root5.eps "fig:"){width="45.00000%"}]{}
The “Catastrophe Manifold” of $I_{E}$ {#the-catastrophe-manifold-of-i_e .unnumbered}
-------------------------------------
One can see that certain patterns emerge for both $g(z)$ and $I_{E}$. For fixed values of $A$, the real solutions of $g(z)$ form a two-fold pattern as functions of $B$. One can check that this occurs for all $A$, large and small. The upper and lower folds turn over at $B=\frac{4}{81} \frac{(A+3)^3}{A}$ and at $B=0$ respectively. This is easily understood with the help of Fig. $1$: the curves (of two roots) in the upper fold meet when the two roots become equal at $B=\frac{4}{81}
\frac{(A+3)^3}{A}$, i.e., before entering the combined region of II and III where they turn complex. On the other hand, on $B=0$, $g(z)$ has a double root equal to $A$ (the other root is $2A$) – therefore the lower folds turn over around the $B=0$ line. The surface $g(z)=0$ thus formed by placing such images successively (Fig. $11$) is similar in structure to a cusp catastrophe manifold familiar in dynamical systems driven by a quartic potential with two control parameters (see, for example, [@PS]). The minima of a quartic potential occur when its derivative (a polynomial of degree three) is set to zero and hence the catastrophe manifold represents the equilibrium points of the system. The “catastrophe map” is then the part of projection of the catastrophe manifold onto the plane of the control variables bounded between the lines along which the folds turn over – in our case this is the region bounded between $B=0$ and $B=\frac{4}{81}
\frac{(A+3)^3}{A}$ curves in Fig. $1$, i.e., the union of region I and IV. However, one may wonder why a cusp does not appear in this case. This is because of the (particular) way in which $A$ and $B$ combine in the coefficients of $g(z)$ and also because they are constrained to be positive – both facts are dictated by the physical configuration. One can obtain a cusp, however, by working with the new variable $D(=AB)$ instead of $B$. A cusp will appear at $A=-3$, which is obviously outside the range of our physical interest.
It is not difficult to see that $I_{E}$ has the same catastrophe map, namely the union of region I and IV. However, the more important observation is that the relative orientation of the folds of $I_{E}$ is the same as that of the corresponding roots of $g(z)$, i.e., they do not cross each other. This pattern persists even when one gets very close to $A=0$ as shown in Fig. $7$. For very small values of $A$ the green and blue lines nearly overlap and coincide completely for $A=0$. They therefore meet the red line at infinity. For higher values of $A$ one gets a persistent behaviour as in Figs. $8$-$10$. Therefore the “catastrophe manifold” of $I_{E}$ is “diffeomorphic” to the one found above for $z$, i.e., the surface $I_{E}$ is obtainable from that of $z$ by a deformation which preserves the catastrophe map. This is not automatic or obvious given the form of $I_{E}$ which is a quadratic function of $z$. Note, however, that the surface of $I_{E}$ is not smooth where the upper fold occurs.
The two-dimensional catastrophe map in a dynamical system with a quartic potential demarcates the regions of three stable minima and one stable minimum. In our case they indicate a demarcation between the regions where there are three real $I_{E}$ and one real $I_{E}$ for the boundary-value problem.
[![Actions as functions of $B$ ($A=25\times 10^{-6},\,25\times
10^{-4},3\times 10^{-2}$)[]{data-label="..."}](CN1.eps "fig:"){width="28.00000%"}]{}
[![Actions as functions of $B$ ($A=3\times 10^{-1},\,8\times 10^{-1},\,1.2$)[]{data-label="..."}](CN4.eps "fig:"){width="28.00000%"}]{}
[![Actions as functions of $B$ ($A=2.0,\,5.0,\,15.0$)[]{data-label="..."}](CN7.eps "fig:"){width="28.00000%"}]{}
[![Actions as functions of $B$ ($A=1500,\,15000,\,150000$)[]{data-label="..."}](CN10.eps "fig:"){width="28.00000%"}]{}
Large Radii and Small Anisotropy
--------------------------------
The observations that the space of $I_{E}$ is diffeomorphic to that of $z$ and that it does not intersect itself have immediate physical implications. This means that the dominant contribution will always come from [*[one]{}*]{} solution, namely the one represented by the green curve in Figs. $2$-$10$ (and the meshed surface in Fig. $11$). Note that this is the solution which gives a positive-definite infilling solution as long as $b^2< \frac{1}{3}a^2\,(2a^2+3)$. One can verify from Figs. $2$-$6$ that this solution always takes values within the interval $(0,A)$.
In cosmology, one is more interested in regions where $b$ is not greatly different from $a$ and usually when they are both large. All three roots are positive in this latter region. Taking $A=B$, the roots of $g(z)$ are $$\begin{array}{rcl}
z_{1}&=&A+\frac{3}{2}-\frac{1}{2}\,\sqrt {12\,A+9},\\
z_{2}&=&2\,A,\\
z_{3}&=&A+\frac{3}{2}+\frac{1}{2}\,\sqrt {12\,A+9}.\\
\end{array}$$ It is easy to check that only $z_{1} < A$ and, hence, $z_{1}$ corresponds to the positive-definite real infilling solution in the interior. The dominant contribution to the path integral (\[1\]) will therefore come from $z_{1}$ (which can also be checked explicitly in this case). The action of this solution is given by $$8 \pi \,G \, (\lambda I_{E})= -\frac{4}{3}\,{\pi}^{2}\left (-9+3\,\sqrt {12\,A+9}+4\,A\sqrt {12\,A+9}\right
)\,\, \sim\,\, -\frac{32}{{\sqrt{3}}}\,
{\pi}^{2}\,A^{\frac{3}{2}},$$ so becoming more negative as $A$ grows. Actions for $z_{2}$ and $z_{3}$ are positive and become more positive as $A$ grows, as can be checked explicitly by direct substitution.
Conclusion
==========
We have shown that, for a given boundary which is a Berger sphere (a squashed $S^{3}$ with two axes equal), there are in general three distinct ways in which one can fill in with a self-dual Taub-NUT-(anti)de Sitter metric. With suitable choice of variables, the problem of finding explicit solutions for such infilling metrics can be translated into a univariate algebraic equation of degree three and hence can be solved exactly. The Euclidean action $I_{E}$ is a quadratic function of the solutions of this third-degree equation and hence, corresponding to every solution of this equation, the action can be found purely in terms of the boundary data – the two radii $(a,b)$ of the Berger sphere, for both positive and negative cosmological constants. The positive and negative roots of this equation correspond to metrics for which actions are real – only complex solutions correspond to complex-valued actions in general.
In the case of a negative cosmological constant, we have further discussed systematically the ranges of $a$ and $b$ where the solutions of the algebraic system lead to real- and complex-valued solutions in the interior and have studied the structure of the three roots. We found that one of the roots corresponds to a positive-definite infilling solution if and only if $b^2<
\frac{1}{3}a^2\,(2a^2+3)$. For small squashing, i.e., when $a$ and $b$ are of the same order and not too small, all three roots are positive. When $a$ and $b$ are exactly equal, this holds for small radius as well. We therefore investigated the roots and their corresponding actions numerically in this range (i.e., until two of them turn complex) as functions of the boundary variables $a$, $b$. We found that both the roots and the actions have structures similar to those of the cusp catastrophe in dynamical systems. The “catastrophe manifold” of $I_{E}$ does not intersect itself which implies that the the dominant contribution will come from the positive-definite infilling solution. Further, the classical actions for large values of the radii $(a,b)$ in their isotropic limit (of particular interest for cosmology) have been discussed. The (dominant) contribution coming from the positive-definite solution has an action proportional to $-a^3$, thereby favouring large radii.
Acknowledgements {#acknowledgements .unnumbered}
----------------
We would like to thank Gary Gibbons, Stefano Kovacs, Alexei Kovalev, Jorma Louko, Henrik Pedersen, Nick Shepherd-Barron and Galliano Valent for helpful discussions and comments. MMA was supported by awards from the Cambridge Commonwealth Trust and the Overseas Research Scheme.
[99]{} M. Abramowitz and I. A. Stegun (1965). *Handbook of Mathematical Functions* (Dover, New York).
M. F. Atiyah, N. J. Hitchin and I. M. Singer, “Self-duality in four-dimensional Riemannian geometry,” Proc. Roy. Soc. Lond. A [**362**]{} (1978) 425. S. Barnard, and J. M. Child (1936). *Higher Algebra* (Macmillan, India).
B. Carter, “Hamilton-Jacobi And Schrödinger Separable Solutions Of Einstein’s Equations,” Commun. Math. Phys. [**10**]{} (1968) 280. M. Cvetič, G. W. Gibbons, H. Lu and C. N. Pope, “Bianchi IX Self-dual Einstein Metrics and Singular G(2) Manifolds,” arXiv:hep-th/0206151. P. D. D’Eath (1996). *Supersymmetric Quantum Cosmology* (Cambridge University Press)
T. Eguchi, P. B. Gilkey and A. J. Hanson, “Gravitation, gauge theories and differential geometry,” Phys. Rep. [**66**]{} (1980) 213.
T. Eguchi and A. J. Hanson, “Asymptotically Flat Self-dual Solutions to Euclidean Gravity,” Phys. Lett. B [**74**]{} (1978) 249.
R. P. Feynman and A. R. Hibbs (1965). *Quantum Mechanics and Path Integrals* (McGraw-Hill, New York, 1965).
G. W. Gibbons and S. W. Hawking, “Action Integrals And Partition Functions In Quantum Gravity,” Phys. Rev. D [**15**]{} (1977) 2752. G. W. Gibbons and S. W. Hawking, “Classification of Gravitational Instanton Symmetries,” Commun. Math. Phys. [**66**]{} (1979) 291. G. W. Gibbons and C. N. Pope, “$\bbbc P^{2}$ as a Gravitational Instanton,” Commun. Math. Phys. [**61**]{} (1978) 239. G. W. Gibbons and C. N. Pope, “The Positive Action Conjecture and Asymptotically Euclidean Metrics in Quantum Gravity,” Commun. Math. Phys. [**66**]{} (1979) 267. J. B. Hartle and S. W. Hawking, “Wave Function of the Universe,” Phys. Rev. D [**28**]{} (1983) 2960. S. W. Hawking, “Gravitational Instantons,” Phys. Lett. A [**60**]{} (1977) 81. N. J. Hitchin, “Twistor Spaces, Einstein Metrics and Isomonodromic Deformations,” J. Differential Geom. [**42**]{} (1995), no. 1, 30–112.
L. G. Jensen, J. Louko and P. J. Ruback, “Biaxial Bianchi Type IX Quantum Cosmology,” Nucl. Phys. B [**351**]{} (1991) 662. D. N. Page, “Taub-Nut Instanton with an Horizon,” Phys. Lett. B [**78**]{} (1978) 249. H. Pedersen, “Einstein metrics, spinning top motions and monopoles,” Math. Ann. [**274**]{} (1986) 35.
T. Poston and I. Stewart (1978). *Catastrophe Theory and its Applications* (Pitman, London).
M. Reid (1988). *Undergraduate Algebraic Geometry* (Cambridge University Press).
T. Sakai, “Cut loci of Berger’s spheres”, Hokkaido Math. J. [**[10]{}**]{}(1981), no. 1, 143-155.
K. P. Tod, “Self-dual Einstein metrics from the Painlevé VI equation,” Phys. Lett. A [**190**]{} (1994) 221.
J. Wess and J. Bagger (1992). *Supersymmetry and Supergravity* (Princeton University Press).
[^1]: Email:[email protected]
[^2]: E-mail:[email protected]
[^3]: For a detailed discussion on such complex-valued metrics on real manifolds see [@Louko] and references therein.
[^4]: As we are dealing with finite $L \ne 0$ for which (\[5\]) is well-defined, self-dual metrics of biaxial Bianchi-IX type (admitting $S^3$-foliations) arising as limiting cases ($L \rightarrow 0$ and $L \rightarrow \infty$) of the self-dual Taub-NUT-(anti)de Sitter – as discussed in Section $2$ – are precluded from the considerations here. Depending on two radii of the Berger sphere, possible infilling solutions of such metrics can easily be included when one considers the larger class of (self-dual) Bianchi-IX type metrics on the four-ball.
[^5]: Fourier’s theorem: If $F(x)$ be a polynomial of degree $n$ and $F_{1},F_{2},...,F_{n}$ are its successive derivatives, the number of real roots $R$ which lie between two real numbers $p$ and $q$ ($p<q$) are such that $R\le N-N'$, where $N$ and $N'$ ($N\ge N'$) respectively denote the number of changes of sign in the sequence $f_{1},f_{2},...f_{n}$, when $x=p$ and when $x=q$. Also $((N-N')-R)$ is an even number or zero. See, for example, [@BC].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The wide spread of location-based social networks brings about a huge volume of user check-in data, which facilitates the recommendation of points of interest (POIs). Recent advances on distributed representation shed light on learning low dimensional dense vectors to alleviate the data sparsity problem. Current studies on representation learning for POI recommendation embed both users and POIs in a common latent space, and users’ preference is inferred based on the distance/similarity between a user and a POI. Such an approach is not in accordance with the semantics of users and POIs as they are inherently different objects. In this paper, we present a novel spatiotemporal aware (STA) representation, which models the spatial and temporal information as *a relationship connecting users and POIs*. Our model generalizes the recent advances in knowledge graph embedding. The basic idea is that the embedding of a $<$time, location$>$ pair corresponds to a translation from embeddings of users to POIs. Since the POI embedding should be close to the user embedding plus the relationship vector, the recommendation can be performed by selecting the top-*k* POIs similar to the translated POI, which are all of the same type of objects. We conduct extensive experiments on two real-world datasets. The results demonstrate that our STA model achieves the state-of-the-art performance in terms of high recommendation accuracy, robustness to data sparsity and effectiveness in handling cold start problem.'
author:
- |
Bei Liu[$~^{1}$]{}, Tieyun Qian[$~^{1}$]{}, Bing Liu[$~^{2}$]{}, Liang Hong[$~^{3}$]{}, Zhenni You[$~^{1}$]{}, Yuxiang Li[$~^{1}$]{}\
[$~^{1}$]{} State Key Laboratory of Software Engineering, Wuhan University, China\
[$~^{2}$]{} Department of Computer Science, University of Illinois at Chicago, USA\
[$~^{3}$]{} School of Information Management, Wuhan University, China\
{qty, beiliu}@whu.edu.cn, [email protected], {hong, znyou, liyux}@whu.edu.cn\
title: 'Learning Spatiotemporal-Aware Representation for POI Recommendation'
---
Introduction
============
Location-based social networks (LBSN), such as Foursquare, Yelp, and Facebook Places, are becoming pervasive in our daily lives. Users on LBSN like to share their experiences with their friends for points of interest (POIs), e.g., restaurants and museums. The providers of location-based services have collected a huge amount of users’ check-in data, which facilitates the recommendation of POIs to unvisited users. The POI recommendation is of high value to both the users and companies, and thus has attracted much attention from researchers in recent years [@Cheng_tist16; @Zhu_kdd15; @Chen_aaai15; @Gao_aaai15].
Most existing studies mainly focused on leveraging spatial information due to the well-known strong correlation between users’ activities and geographical distance [@Zheng_www09; @Ye_gis10; @Cho_kdd11]. For example, Ye et al. proposed a Bayesian collaborative filtering (CF) algorithm to explore the geographical influence. Cheng et al. captured the geographical influence by modeling the probability of a user’s check-in on a location as a multi-center Gaussian model and then combined it into a generalized matrix factorization model. Lian et al. adopted a weighted matrix factorization framework to incorporate the spatial clustering phenomenon.
Similar to the geo-spatial information, time is another important factor in POI recommendation. Ye et al., found the periodic temporal property that people usually go to restaurants at around noon and visit clubs at night. Yuan et al., developed a CF based model to integrate temporal cyclic patterns. Cheng et al. [@Cheng_ijcai13] explored the temporal sequential patterns for personalized POI recommendation by using the transition probability of two successive check-ins of a user.
Existing studies has exploited spatial or temporal influences mainly using CF [@Ye_sigir11; @Yuan_sigir13] and Markov transition approaches [@Cheng_ijcai13]. Due to the sparsity of users’ check-in records, it is hard to find similar users or calculates transition probability. Although matrix factorization (MF) methods are effective in dealing with the sparsity in user-POI matrix [@Cheng_aaai12; @Lian_kdd14], they do not consider the current location of the user. More importantly, while time and location together play a critical role in determining users’ activities in LBSNs, rare work has modeled their joint effects. Considering only one factor will deteriorate the predictive accuracy. For instance, a student may go to a school cafeteria or to a food court in a mall at lunch time depending on he/she is on campus or outside. It is not suggested for a system to recommend the same restaurant to a user at the same time but different location. This example shows the ineffectiveness when using one type of information but ignoring the other. However, taking both time and location into consideration exaggerates the data sparsity.
In this paper, we propose a novel spatiotemporal aware (STA) model, which captures the joint effects of spatial and temporal information. Our model has the following distinct characteristics.
- STA takes location and time as a whole to determine the users’ choice of POIs.
- STA embeds a spatiotemporal pair $<$time, location$>$ as *a relationship* connecting users and POIs.
By considering the time and location at the same time, our model can be successfully applied to real-time POI recommendation. Furthermore, distributed representations of STA are very effective in solving the problem of data sparsity.
Two recent works [@Feng_ijcai15; @Xie_cikm16] also exploited the power of distributed representation for alleviating data sparsity. The personalized ranking metric embedding (PRME) by Feng et al. projected each POI and each user into a latent space, and then recommended a POI $v$ to a user $u$ at location $l$ based on the Euclidean distance between the POI and the user $\parallel\vec{u}-\vec{v}\parallel^2$ and that between the POI and the location $\parallel\vec{l}-\vec{v}\parallel^2$. Xie et al. proposed a graph based embedding model (GE) by embedding graphs into a shared low dimensional space, and then computed the similarity between a user $u$’s query $q$ at current time $t$ and location $l$ and a POI $v$ using an inner product, $S(q,v)=\vec{u}^T\cdot\vec{v}+\vec{t}^T\cdot\vec{v}+\vec{l}^T\cdot\vec{v}$. While PRME, especially GE, shows significant improvements over many other baselines, these two methods have the drawback that they embed both users and POIs in a common latent space, and users’ preference is inferred based on the distance/similarity between a user and a POI. Such an approach is unnatural since users and POIs are inherently different objects. In contrast, our STA model generalizes recent advances in knowledge graph embedding [@Lin_aaai15]. A user $u$ reaches an interested POI $v_q$ via an edge $tl$ denoting the $<$time, location$>$ pair, i.e., $\vec{u} + \vec{tl} \approx \vec{v_q}$. With this transformation, we can do recommendation for $u$ by selecting the top-*k* POIs similar to POI $v_q$, which are all of the same type of objects with similar semantics.
Problem Definition and Preliminary
==================================
Definition 1. (***POI***) A POI *v* is defined as a unique identifier representing one specific position (e.g., a cafe or a hotel), and *V* is a set of POIs, i.e., $V=\{v|v=(pid, position)\}$.
Definition 2. (***Check-in Activity***) A check-in activity is a quadruple (u, t, l, v), which means a user *u* visits a POI *v* in location *l* at time *t*.
Definition 3. (***Spatiotemporal pattern***) A spatiotemporal pattern, denoted as *tl*, is a combination of a time slot *t* and a location *l* like $<$11 a.m., Los Angeles$>$. For ease of presentation, we summarize the notations in Table \[tbl:note\]. The POI recommendation problem investigated in this paper has the same settings as that in [@Xie_cikm16]. The formal problem definition is as follows.
[c|c]{} Variable & Interpretation\
*u*, *v* & the user *u* and POI *v*\
*t* & the time slot discretized from timestamp\
*l* & the location mapped from (longitude, latitude)\
*tl* & the spatiotemporal pattern $<$t, l$>$\
$\vec{u}$,$\vec{tl}$,$\vec{v}$ & embeddings of *u*, (*t,l*), and *v*\
$u_q$, $t_q$, $l_q$ & query user $u_q$, his/her current time $t_q$ and location $l_q$\
$v_q$ & the potential POI that query user $u_q$ is interested in\
**Problem Definition** (Location-based Recommendation) Given a dataset $D=\{d|d=(u,t,l,v)\}$ recording a set of users’ activities, and a query $q=(u_q, t_q, l_q)$, we aim to recommend top-*k* POIs in *V* that the query user $u_q$ would be interested in.
**Preliminary - KG Embedding** The knowledge graph (KG) is a directed graph whose nodes and edges describing *entities* and their *relations* of the form (*head, relation, tail*), denoted as (*h, r, t*). The goal of knowledge graph embedding is to learn a continuous vector space where the embeddings of entity and relation can preserve certain information of the graph. Bordes et al. presented a simple yet effective approach TransE to learn vector embeddings for both entities and relations in KG. The basic idea is that the relationship between entities corresponds to a translation the embeddings of entities, namely, $\vec{h}+\vec{r}\approx\vec{t}$ when (*h ,r, t*) exits in graph. Later, a model named TransH [@Wang_aaai14] was proposed to enable an entity to have distinct representations when it is involved in different relations.
Both TransE and TransH project all entities and relations into the same space. However, some entities may have multiple aspects and relations focusing on different aspects of the entities. Such entities are close in the entity space when they are similar, but they should be far away from each other in the relation space if they are strongly different in some specific aspects. To address this issue, Lin et al. presented a TransR model to project two entities *h* and *r* of (*h,r,t*) into a *r*-relation space as $h_r$ and $t_r$ with operation $M_r$, such that $\vec{h_r}+\vec{r}\approx\vec{t_r}$ holds in the relation space.
Our Proposed Framework
======================
We seek to learn the representations with the following characteristics.
- Spatiotemporal awareness - Location and time together play a crucial role when a user selects a POI; they should not be separated into individual ones.
- Semantics consistency - All the POIs, either the query user’s interested POI $v_q$ or all existing POIs $v \in V$, should come from a consistent semantic space.
In order to satisfy the first requirement, we combine each time slot and location as a spatiotemporal pattern $<$*t*, *l*$>$, and convert the quadruples $(u,t,l,v) \in D$ into triples ($u$, $<$$t$, $l$$>$, $v$) in $D'$. We then learn representations for users, spatiotemporal patterns, and POIs from the converted set $D'$ to meet the second condition, using the translation technique originated from knowledge graph embedding.
STA model
---------
For the location-based recommendation problem, we focus on the connections between users and POIs corresponding to the spatiotemporal relations. Intuitively, if a POI *v* is often visited by similar users in location *l* at time *t*, the probability of a query user $u_q$ visiting *v* with the same spatiotemporal relation will be high. On the other hand, users similar in the entity space may visit different POIs under distinct temporal and geographic conditions. In order to capture the strong correlations of users and POIs to the spatiotemporal patterns, we generalize the TransR technique [@Lin_aaai15] to fit the POI recommendation task. The basic idea is that a user $u$ will reach an interested POI $v_q$ via a translation edge $tl$, i.e., $\vec{u} + \vec{tl} \approx \vec{v_q}$. Fig. \[fig:relation\] illustrates the impacts of *tl* patterns.
![Impacts of spatiotemporal patterns[]{data-label="fig:relation"}](relation3.eps){width="68.00000%" height="4.2cm"}
In Fig. \[fig:relation\], suppose $u_1$, $u_q$, and $u_2$ are three university students, $u_1$ and $u_q$ taking same courses, and $u_2$ and $u_q$ sharing the dormitory. Given two patterns $tl_1=<12 a.m., campus>$ and $tl_2=<8 p.m., dormitory>$, the query user $u_q$ will be translated into two POIs $v_{q1}$ and $v_{q2}$, hence we should recommend for $u_q$ the POI $v_1$ in the left lower sub-figure and $v_2$ in the right lower sub-figure, which are the close neighbor of $v_{q1}$ and $v_{q2}$, respectively. The different recommending results $v_1$ and $v_2$ are caused by the effects of different spatiotemporal relations $tl_1$ and $tl_2$.
We now introduce the detail for STA. For each triple ($u$, $<$$t$, $l$$>$, $v$) in $D'$, the user $u$, the spatiotemporal pair $<$$t$, $l$$>$ ($tl$ in short), and POI $v$ corresponds to the head entity *h*, the relationship edge *r* and the tail entity *t* in TransR, respectively. Their embeddings are set as $\vec{u}$, $\vec{v}$ $\in$ $\Re^d$, and $\vec{tl}$ $\in$ $\Re^m$. For each spatiotemporal pair $tl$, we set a projection matrix $M_{tl} \in \Re^{d\times m}$ to project a user embedding $\vec{u}$ and a POI embedding $\vec{v}$ in the original entity space to $\vec{u_{tl}}=\vec{u}M_{tl}$ and $\vec{v_{tl}}=\vec{v}M_{tl}$ in the relation space, such that $\vec{u_{tl}} + \vec{tl} \approx \vec{v_{tl}}$. This indicates that a POI embedding $\vec{v_{tl}}$ should be the nearest neighbor of $\vec{u_{tl}} + \vec{tl}$. Hence the the score function can be defined as: $$\begin{aligned}
\label{score}
\small
\nonumber s_{tl}(u, v) = \parallel \vec{u_{tl}} + \vec{tl} - \vec{v_{tl}} \parallel _2^2\\
\nonumber s.t. \parallel \vec{u} \parallel _2 \leq 1, \parallel \vec{v} \parallel _2 \leq 1, \parallel \vec{tl} \parallel _2 \leq 1, \\
\parallel \vec{u_{tl}} \parallel _2 \leq 1, \parallel \vec{v_{tl}} \parallel _2 \leq 1\end{aligned}$$
Given the score function defined in Eq. \[score\] for a triple ($u$, $tl$, $v$), the entire objective function for training is as follows. $$\label{obj}
\small
L = \sum_{(u,tl,v) \in T} \sum_{(u',tl,v') \in T'} max(0, s_{tl}(u, v) + \gamma - s_{tl}(u', v')),$$ where max(*a*,*b*) is used to get the maximum between *a* and *b*, $\gamma$ is the margin, *T* and *T’* are the sets of correct and corrupted triples, respectively. The corrupted triples are generated by replacing the head and tail entities in correct triples using the same sampling method as that in [@Wang_aaai14].
We adopt stochastic gradient descent (SGD) (in mini-batch mode) to minimize the objective function in Eq. \[obj\]. A small set of triplets, is sampled from the training data. For each such triplet, we sample its corresponding incorrect triplets. All the correct and incorrect triples are put into a mini-batch. We compute the gradient and update the parameters after each mini-batch. When the iteration reaches a predefined number, we learn all the embedding for users, POIs, and spatiotemporal patterns.
Recommendation Using STA
------------------------
Once we have learned the embeddings, given a query user $u_q$ with the query time $t_q$ and location $l_q$, i.e., *q* = ($u_q$, $t_q$, $l_q$), we first combine $t_q$ and $l_q$ as a spatiotemporal pattern $tl_q$, and then we can get the potential POI $v_q$ using Eq. \[querypoi\]. $$\label{querypoi}
\small
%\vec{v_{tlq}} = \vec{u_{tlq}} + \vec{tl_q}
\vec{v_q} = \vec{u_q} M_{tl} + \vec{tl_q}$$
The learned POI embedding $v_q$ naturally reflects the user’s preference, because it encodes the users’ past activities in $\vec{u_q}$. It also captures the geographic and temporal influence in $\vec{tl_q}$.
For each POI $v \in V$, we compute its distance to the POI $v_q$ in the normed linear space as defined in Eq. \[simpoi\], and then select the *k* POIs with the smallest ranking scores as recommendations. $$\label{simpoi}
\small
%d(v_{tl}, v_{tlq}) = \parallel \vec{v_{tl}} - \vec{v_{tlq}} \parallel_1
d(v, v_q) = \parallel \vec{v} M_{tl} - \vec{v_q} \parallel_1$$
We would like to emphasize our differences in computing $v_q$ and recommending POIs from those in [@Lin_aaai15; @Xie_cikm16]. First, we can find an explicit POI $v_q$ directly from the latent space through the translation of the embedding of the spatiotemporal pattern on the user’s embedding, while others compute an implicit $v_q$ by its distance/similarity to user $u_q$. Second, since the embeddings for POIs in *V* are also from the same space, we can choose the ones which are the closest neighbors of $v_q$ in this space. This indicates that our recommended POIs are semantically consistent with the query user’s interested POI $v_q$.
Dealing with Cold Start POIs
----------------------------
Considering the cold start POIs, which contain geographic and content information like tags but do not have any check-ins [@Xie_cikm16], we can simply extend our model to include the POI-POI relationship through the translation of content patterns. We call this model STA-C. The rationale is that, if two POIs share a common tag or location, there will be a high degree of similarity between them, and their vector representations should be close to each other. Based on this observation, we define the score function as following:
$$\label{scorecold}
\small
\begin{aligned}
\begin{split}
s_{tlw}(u, v, s) & = s_{tl}(u, v) + s_{wl}(v, s) \\
%& = s_{tl}(u, v) + s_{w}(v, s) \\
& = \parallel \vec{u_{tl}} + \vec{tl} - \vec{v_{tl}} \parallel _2^2 + \parallel \vec{v_{wl}} + \vec{wl} - \vec{s_{wl}} \parallel _2^2,
\end{split}
\end{aligned}$$
where *s* is a POI sharing at least one $<$word, location$>$ pair with POI *v*, and the objective function for cold start POIs is defined as:
$$\label{objcold}
\small
\begin{aligned}
\begin{split}
LC = & \sum_{(u,tl,v) \in T} \sum_{(u',tl,v') \in T'} max(0, s_{tl}(u, v) + \gamma - s_{tl}(u', v')) + \\
& \sum_{(v,wl,s) \in W} \sum_{(v',wl,s') \in W'} max(0, s_{wl}(v, s) + \gamma - s_{wl}(v', s'))
\end{split}
\end{aligned}$$
We once again use stochastic gradient descent to minimize the objective function *LC* in Eq. \[objcold\]. The only difference is the sampling procedure. For STA-C, since we have two types of edges, we sample the triplets (u, tl, v) and (v, wl, s) and their corresponding incorrect triples alternatively to update the model.
Our STA-C model proposed for dealing with cold start POIs can also be applied to the normal POI recommendation problem. However, it requires that those POIs should contain content information. For the recommendation on datasets like Gowalla, STA-C is not valid. Hence we only treat it as an extended model. Please also note that, it is STA-C that uses the same information as GE does. Our standard STA model, on the other hand, uses less information than GE because it does not include the contents of POIs.
Experimental Evaluation
=======================
In this section, we first introduce the experimental setup and then compare our experimental results with those of baselines. Finally we show the performance of our method for addressing the data sparsity and cold start problem.
Experimental Setup
------------------
**Datasets** We evaluate our methods on two real-life LBSN datasets: Foursquare and Gowalla. A number of researchers have conducted experiments on data collected from these two social networks [@Yuan_sigir13; @Chen_aaai15; @Gao_aaai15; @Xie_cikm16; @Yin_tkde16]. However, many of them are collected from various regions or in different time spans. For a fair comparison with GE, we use the publicly available version [^1] provided by the authors of [@Xie_cikm16].
The two datasets have different scales such as geographic ranges, the number of users, POIs, and check-ins. Hence they are good for examining the performance of algorithms on various data types. Their statistics are listed in Table \[tbl:dataset\].
[ccc]{} & **Foursquare** & **Gowalla**\
\# of users & 114,508 & 107,092\
\# of POIs & 62,462 & 1,280,969\
\# of Check-ins & 1,434,668 & 6,442,892\
\#std time slots & 24 & 24\
\# of locations & 5,846 & 200\
\# of $<$t, l$>$ patterns & 28,868 & 3,636\
Each check-in is stored as user-ID, POI-ID, POI-location in the form of latitude and longitude, check-in timestamp, and POI-content (only for Foursquare). In order to get the spatiotemporal patterns $<$t, l$>$ in Table \[tbl:dataset\], we use the same discretized method as that in [@Xie_cikm16], i.e., dividing time into 24 time slots which correspond to 24 hours, and the whole geographical space into a set of regions according to 5,846 administrative divisions (for Foursquare) and 200 regions clustered by a standard *k*-means method (for Gowalla). We finally get 28,868 and 3,636 $<$t, l$>$ pairs on Foursquare and Gowalla, respectively.
**Baselines - {GE, STA-E, STA-H}** We use GE, the state-of-the-art location based recommendation approach in [@Xie_cikm16], as our baseline. GE adopts a graph-based embedding framework. It learns the embeddings based on the POI-POI, POI-Time, POI-Location, and POI-Words graphs. By integrating the sequential, geographical, temporal cyclic, and semantic effect into a shared space, GE effectively overcomes the data sparsity problem and reaches the best performance so far.
We do not compare our method with other existing approaches because, GE has already significantly outperformed a number of baselines including JIM [@Yin_cikm15joint], PRME [@Feng_ijcai15], and Geo-SAGE [@Wang_kdd15Geo]. We thus only show our improvements over GE.
Also note that although we choose the TransR technique in knowledge graph embedding to materialize our STA model, the essential of our proposed framework is the translation of $<$time, location$>$ pairs in the embedding space. This indicates that we do not rely on a specific translation model. Hence we can use TransE [@Bordes_ml14] and TransH [@Wang_aaai14] to realize STA. We denote the resulting methods as STA-E and STA-H baselines, respectively.
**Settings** We first organize the quadruples (u, v, t, l) in each dataset by users to get each user’s profile $D_u$. We then rank the records in $D_u$ according to the check-in timestamps, and finally divide these ordered records into two parts: the first 80% as the training data, and the rest 20% data as the test data. Moreover, the last 10% check-in records in the training data are used as a validation set for tuning the hyper-parameters. We use the accuracy@*k* (*k* = {1, 5, 10, 15, 20}) as our evaluation metric. All these settings, as well as the computation approach to accuracy@*k*, are same as those in [@Xie_cikm16].
We use the default settings in the original TransR [@Lin_aaai15] as the parameter settings for our STA model. Specifically, we set the learning rate $\lambda = 0.0001$, the margin $\gamma = 2$, the mini-batch size $B=4800$, and the embedding dimensions $m = d = 100$, and we traverse over all the training data for 1000 rounds.
Comparison with baselines
-------------------------
For a fair comparison, we implement GE using the same LINE software provided by the authors of [@Tang_www15] on our data divisions. All the parameters for GE are same as those in [@Xie_cikm16]. We find a slightly difference (less than 1% in accuracy) between the original results and those by our implemented GE. This is understandable and acceptable considering the randomness when sampling negative edges in LINE and initiating the centers of clusters of regions. All parameters for STA-E and STA-H use the default settings in [@Bordes_ml14] and [@Wang_aaai14]. We present the comparison results on Foursquare and Gowalla in Fig. \[fig:comp\] (a) and (b), respectively.
{width="82.00000%" height="5.2cm"}
From Fig. \[fig:comp\] (a), it is clear that all our proposed STA-style models significantly outperform GE. For instance, the accuracy@1 for STA, STA-H, and STA-E is 0.307, 0.280, 0.255, respectively, much better than 0.225 for GE. Similar results can be observed in Fig. \[fig:comp\] (b) on Gowalla dataset. This clearly demonstrates the effectiveness of our translation based framework.
While STA shows drastic improvement over GE for all *k*s on Foursquare, the trend is not that obvious on Gowalla when *k* = 15, 20. This is because there is a much smaller number of relations in Gowalla than that in Foursquare. As shown in Table \[tbl:dataset\], Gowalla only has 3,636 relation patterns ($<$t, l$>$ pairs) while Foursquare has 28,868 pairs. Hence the learnt embeddings for entities and relations are worse than those on Foursquare, and incur the less accurate results when *k* is large.
Besides the improvement over GE, STA outperforms STA-H and STA-E as well. The reason is that TransR can differentiate the entities in the transformed relation space. Nevertheless, we see a less significant enhancement of STA over STA-H on Gowalla. This also conforms to the characteristics of the data: the graph of Gowalla is much larger but has less *tl* relation edges than that of Foursquare, and the advantage of TransR over TransE is not obvious on such a dataset.
Effects of Model Parameters
---------------------------
The effects of embedding dimension *d* on Foursquare and Gowalla are shown in Table \[tab:dimf\] and Table \[tab:dimw\], respectively.
1 5 10 15 20
----- ------- ------- ------- ------- -------
70 0.281 0.376 0.409 0.433 0.451
80 0.294 0.384 0.417 0.445 0.462
90 0.300 0.390 0.425 0.459 0.476
100 0.307 0.393 0.434 0.461 0.483
110 0.311 0.407 0.439 0.463 0.486
120 0.312 0.407 0.439 0.464 0.486
: Effects of Dimensionality on Foursquare
\[tab:dimf\]
1 5 10 15 20
----- ------- ------- ------- ------- -------
70 0.355 0.432 0.474 0.503 0.527
80 0.358 0.436 0.478 0.508 0.530
90 0.359 0.439 0.482 0.509 0.535
100 0.361 0.445 0.486 0.511 0.539
110 0.361 0.445 0.488 0.513 0.540
120 0.361 0.445 0.488 0.513 0.540
: Effects of Dimensionality on Gowalla
\[tab:dimw\]
We can see that the experimental results are not very sensitive to the dimension *d*. With an increasing number of dimension, the accuracy on Gowalla is almost unchanged, i.e., the improvement is less than 1% in nearly all cases. The accuracy on Foursquare is slightly enhanced with a large dimension *d*, and finally it becomes stable.
To investigate the effects of time interval, we divide timestamps by three methods, i.e., splitting time into 24, 7, and 2 time slots, corresponding to the daily, weekly, and weekday/weekend patterns, respectively. Figure \[fig:time\] shows the effects of various time intervals.
{width="82.00000%" height="5.5cm"}
We observe that the impact of the daily patterns is the most significant on both datasets. In addition, the results for different patterns vary widely, suggesting a good strategy for dividing the time slot is important.
Sensitivity to Data Sparsity
----------------------------
To investigate the sensitivity to data sparsity of STA and GE, we conduct extensive experiments to evaluate the performance on two datasets by reducing training data. More precisely, we keep the testing dataset unchanged and reduce the training data randomly by a ratio of 5% to 20% stepped by 5. Due to the space limitation, we only present the results by reducing 20% training data Table \[tab:20less\]. The trends with other ratios are all alike.
We have the following important notes for Table \[tab:20less\].
- With the reduction of training data, the accuracy values for STA and GE both decrease. However, STA always achieves the best results at different *k* values on two datasets.
- The reduction of accuracy of our STA model is much smaller than that of GE. For instance, the accuracy@1 of GE decreases from 0.225 to 0.154, showing a 31.69% drop. In contrast, our STA model only has a 20.00% change. This strongly suggests that our model is more robust to the data sparsity.
- The declination of accuracy on Foursquare is more obvious than on Gowalla. The reason may be that Foursquare is much sparser in users’ check-ins than Gowalla, hence reducing the training data has a greater impact on Foursquare.
Test for Cold Start Problem
---------------------------
In this experiment, we further compare the effectiveness of our extended STA-C model with GE when addressing the cold-start problem. The cold start POIs are defined as those visited by less than 5 users [@Yin_tkde16]. To test the performance of cold start POI recommendations, we select users who have at least one cold-start check-in as test users. For each test user, we choose his/her check-in records associated with cold-start POIs as test data and the remains as training data. Since there is no content information for POIs in Gowalla, we conduct experiments, just as GE did, only on Foursquare. The results are shown in Fig. \[fig:cold\].
![Test for Cold Start Problem on Foursquare[]{data-label="fig:cold"}](cold1.eps){width="56.00000%" height="3.6cm"}
From Fig. \[fig:cold\], it is clear that our proposed STA-C model consistently beats GE when recommending cold start POIs. The superior performance of STA-C model is due to the translation of content and geography information *wl* from an ordinary POI *v* to a cold start POI $v_c$. As long as there is an existing *v* sharing one $<$word, location$>$ pair with $v_c$, our STA-C model can get a translation for $v_c$. In contrast, GE utilizes the bipartite graphs of POI-Word and POI-Location. The weight of an edge in the graph is calculated by a TF-IDF value of the word or the frequency of a location. The edge weight is proportional to the probability of edge sampling. Since there are few check-in records for cold start POIs, a $v_c$-word and $v_c$-location edge has an extremely rare chance to be selected and updated. Consequently, the learnt embedding for $v_c$ will be poor and further deteriorates the recommendation accuracy.
Conclusion
==========
We present a novel spatiotemporal aware model STA for learning representations of users, spatiotemporal patterns, and POIs. The basic idea is to capture the geographic and temporal effects using a $<$time, location$>$ pair, and then model it as a translation connecting users and POIs. We realize STA using the knowledge graph embedding technique. Our method has two distinguished advantages. 1) We learn a joint representation for spatiotemporal patterns whose components contribute together to a user’s choice in POIs. 2) The translation mechanism enables the learnt POI embeddings to be in the same semantic space with that of the query POI.
We conduct extensive experiments on two real-life datasets. Our results show that STA achieves the state-of-the-art performance in recommendation accuracy. It also significantly outperforms the baselines in terms of the effectiveness in addressing both the data sparsity and cold start problems.
Acknowledge {#acknowledge .unnumbered}
===========
The work described in this paper has been supported in part by the NSFC project (61572376).
Antoine Bordes, Xavier Glorot, Jason Weston, and Yoshua Bengio. A semantic matching energy function for learning with multi-relational data. , 94(2):233–259, 2014.
Xuefeng Chen, Yifeng Zeng, Gao Cong, Shengchao Qin, Yanping Xiang, and Yuanshun Dai. On information coverage for location category based point-of-interest recommendation. In [*Proc. of AAAI*]{}, page 37¨C43, 2015.
Chen Cheng, Haiqin Yang, Irwin King, and Michael R Lyu. Fused matrix factorization with geographical and social influence in location-based social networks. In [*Proc. of AAAI*]{}, pages 17–23, 2012.
Chen Cheng, Haiqin Yang, Michael R. Lyu, and Irwin King. Where you like to go next: Successive point-of-interest recommendation. In [*Proceedings of IJCAI*]{}, pages 2605–2611, 2013.
Chen Cheng, Haiqin Yang, Irwin King, and Michael R Lyu. A unified point-of-interest recommendation framework in location-based social networks. , 8(1):1–21, 10 2016.
Eunjoon Cho, Seth A. Myers, and Jure Leskovec. Friendship and mobility: user movement in location-based social networks. In [*Proceedings of SIGKDD*]{}, page 1082¨C1090, 2011.
Shanshan Feng, Xutao Li, Yifeng Zeng, Gao Cong, Yeow Meng Chee, and Quan Yuan. Personalized ranking metric embedding for next new poi recommendation. In [*Proc. of 24th IJCAI*]{}, pages 2069–2075, 2015.
Huiji Gao, Jiliang Tang, Xia Hu, , and Huan Liu. Content-aware point of interest recommendation on location-based social networks. In [*Proceedings of 29th AAAI*]{}, pages 1721–1727, 2015.
Defu Lian, Cong Zhao, Xing Xie, Guangzhong Sun, Enhong Chen, and Yong Rui. Geomf: joint geographical modeling and matrix factorization for point-of-interest recommendation. In [*Proceedings of SIGKDD*]{}, pages 831–840, 2014.
Yankai Lin, Zhiyuan Liu, Maosong Sun, Yang Liu, and Xuan Zhu. Learning entity and relation embeddings for knowledge graph completion. In [*Proc. of 29th AAAI*]{}, pages 2181–2187, 2015.
Jian Tang, Meng Qu, Mingzhe Wang, Ming Zhang, Jun Yan, and Qiaozhu Mei. Line: Large-scale information network embedding. In [*Proceedings of WWW*]{}, pages 1067–1077, 2015.
Zhen Wang, Jianwen Zhang, Jianlin Feng, and Zheng Chen. Knowledge graph embedding by translating on hyperplanes. In [*Proc. of 28th AAAI*]{}, pages 1112–1119, 2014.
Weiqing Wang, Hongzhi Yin, Ling Chen, Yizhou Sun, Shazia Sadiq, and Xiaofang Zhou. Geo-sage: A geographical sparse additive generative model for spatial item recommendation. In [*Proceedings of SIGKDD*]{}, pages 1255–1264, 2015.
Min Xie, Hongzhi Yin, Hao Wang, Fanjiang Xu, Weitong Chen, and Sen Wang. Learning graph-based poi embedding for location-based recommendation. In [*Proc. of CIKM*]{}, pages 15–24, 2016.
Mao Ye, Peifeng Yin, and Wang Chien Lee. Location recommendation for location-based social networks. In [*Proc. of ACM SIGSPATIAL*]{}, pages 458–461, 2010.
Mao Ye, Krzysztof Janowicz, and Wang Chien Lee. What you are is when you are: the temporal dimension of feature types in location-based social networks. In [*Proc. of ACM SIGSPATIAL*]{}, pages 102–111, 2011.
Mao Ye, Peifeng Yin, Wang-Chien Lee, and Dik-Lun Lee. Exploiting geographical influence for collaborative point-of-interest recommendation. In [*Proceedings of SIGIR*]{}, pages 325–334, 2011.
Hongzhi Yin, Xiaofang Zhou, Yingxia Shao, Hao Wang, and Shazia Sadiq. Joint modeling of user check-in behaviors for point-of-interest recommendation. In [*Proceedings of CIKM*]{}, pages 1631–1640, 2015.
Hongzhi Yin, Xiaofang Zhou, Bin Cui, Hao Wang, Kai Zheng, and Nguyen Quoc Viet Hung. Adapting to user interest drift for poi recommendation. , 28(10):2566–2581, 2016.
Quan Yuan, Gao Cong, Zongyang Ma, Aixin Sun, and Nadia Magnenat Thalmann. Time-aware point-of-interest recommendation. In [*Proc. of SIGIR*]{}, pages 363–372, 2013.
Yu Zheng, Lizhu Zhang, Xing Xie, and Wei-Ying Ma. Mining interesting locations and travel sequences from gps trajectories. In [*Proc. of WWW*]{}, pages 791–800, 2009.
Wen-Yuan Zhu, Wen-Chih Peng, Ling-Jyh Chen, Kai Zheng, and Xiaofang Zhou. Modeling user mobility for location promotion in location-based social networks. In [*Proceedings of SIGKDD*]{}, pages 1573–1582, 2015.
[^1]: https:/sites.google.com/site/dbhongzhi
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Let $K$ be the fraction field of a Henselian discrete valuation ring with algebraically closed residue field $k$. In this article we give a sufficient criterion for a projective variety over such a field to have index $1$.'
address:
- 'BCAM - Basque Centre for Applied Mathematics, Alameda de Mazarredo 14, 48009 Bilbao, Spain'
- 'Instituto de Matemática Pura e Aplicada, Estr. Dona Castorina, 110 - Jardim Botânico, Rio de Janeiro - RJ, 22460-320, Brazil'
author:
- Ananyo Dan
- Inder Kaur
title: 'Examples of varieties with index one on $C_1$-fields'
---
Introduction
============
A field $K$ is called $C_1$ if any degree $d$ polynomial in $n$ variables with $n>d$ has a non-trivial solution. The $C_{1}$ conjecture due to Lang, Manin and Kollár states that every separably rationally connected variety over a $C_{1}$ field has a rational point. The conjecture has already been proven for several $C_1$-fields (see [@ind] for a complete discussion). However it is still open in the case when $K$ is the fraction field of a Henselian discrete valuation ring of characteristic $0$ with algebraically closed residue field of characteristic $p>0$. Recently, the conjecture was shown to hold trivially for certain rationally connected varieties over such fields (see [@ind2]).
It is natural to ask whether a similar conjecture holds if we replace the condition for a rational point by the condition of index one and weaken the condition on rational connectedness. Recall that the *index* of a variety $X$, denoted ${\mathrm{ind}}(X)$, is the gcd of the set of degrees of zero dimensional cycles on $X$. In [@esn2015 Corollary $2.5$], Esnault, Levine and Wittenberg prove that if $X$ is a smooth, projective variety over the fraction field of a Henselian discrete valuation ring with algebraically closed residue field of characteristic $0$, then ${\mathrm{ind}}(X)$ divides the Euler characteristic of the structure sheaf of $X$. Using this they prove that, in the case $X$ is a rationally connected variety over such a field, we have ${\mathrm{ind}}(X)=1$ (see [@esn2015 Corollary $3$]). Since the Euler characteristic of $\operatorname{\mathcal{O}}_X$ is one if and only if $X$ has arithmetic genus $0$, this gives a positive answer to the modified conjecture only for very few choices of $X$.
In this article we study a weaker notion of index, which we call the linear index. Let $K$ be a field of characteristic zero and $X$ a projective $K$-variety. We define the *linear index* of $X$, denoted ${\mathrm{ind}}_{{\mathrm{lin}}}(X)$, to be the gcd of the Euler characteristic of the set of line bundles on $X$. The definition of linear index is inspired by Kollár’s definition of elw-index, denoted ${\mathrm{elw}}(X)$, as given in [@koll], which is the gcd of the Euler characteristic of all coherent sheaves on $X$. An advantage of using ${\mathrm{ind}}_{{\mathrm{lin}}}(X)$ over ${\mathrm{elw}}(X)$ is that it is much easier to compute the index of $X$ using the former notion. In particular, it is extremely hard to enumerate the set of all coherent sheaves on a variety (even fixing Hilbert polynomial is not sufficient to guarantee boundedness of families of coherent sheaves, see [@huy]). In comparison, the set of all invertible sheaves is given by the Picard group, which is of finite rank in numerous examples. Moreover, the Euler characteristic of an invertible sheaf is significantly easier to compute than that of a general coherent sheaf (see Riemann Roch formula for coherent sheaves [@baum]). As a result, we are able to give a simple combinatorial criterion under which a variety has index $1$. More precisely,
\[thm1\] Fix an ample line bundle $H$ on $X$ such that if there exists an invertible sheaf $H_0$ on $X$ with $H_0^{\otimes n} \cong H$ then $n$ must be $1$ or $-1$. Suppose that $H^1(\mathcal{O}_X)=0$, ${\mathrm{Pic}}(X_{{\overline{K}}})$ is of rank $r$, generated by ${\mathcal{L}}_1, ..., {\mathcal{L}}_{r-1}$ and ${\mathcal{L}}_r:=H_{{\overline{K}}}=H \otimes_K {\overline{K}}$ satisfying the following conditions:
1. the ideal $(\deg({\mathcal{L}}_1),\deg({\mathcal{L}}_2),...,\deg({\mathcal{L}}_r))$ in ${\mathbb{Z}}$ generated by $\deg({\mathcal{L}}_i)$ for $i=1,...,r$ coincides with the ideal $(1)$, where degree of the invertible sheaves are taken with respect to $H_{{\overline{K}}}$ (see [@huy Definition $1.2.11$]),
2. for any $r \times r$-matrix $A=(a_{i,j})$ with integral entries $a_{i,j}$, $a_{r,k}=0$ for all $k<r$, $a_{r,r}=1$, $A \not={\mathrm{Id}}$ and $A^t={\mathrm{Id}}$ for some $t>0$, we have $\sum_j a_{ij} \deg({\mathcal{L}}_j)\not= \deg({\mathcal{L}}_i)$ for some $i>0$.
Then, each ${\mathcal{L}}_i$ is $G$-invariant and ${\mathrm{gcd}}\{\chi({\mathcal{L}}_i(n))| i=1,...,r \mbox{ and } n \in {\mathbb{Z}}\}=1$. Moreover, if $K$ is a $C_1$-field, then ${\mathrm{ind}}_{{\mathrm{lin}}}(X)={\mathrm{ind}}(X)=1 \, \mbox{ if } {\mathrm{char}}(k)=0 \mbox{ and }$ prime-to-p part of ${\mathrm{ind}}(X)$ and ${\mathrm{ind}}_{{\mathrm{lin}}}(X)$ equals $1$ if ${\mathrm{char}}(k)=p>0.$
By prime-to-p part of $N$ we mean the largest divisor of $N$ which is prime to $p$.
We use the criterion in Theorem \[thm1\] to give examples of non-rationally connected varieties having index one over a $C_1$-field (see Example \[elw8\] and Remark \[elw11\]). In Example \[elw02\], we give examples in the case $K$ is not a $C_1$-field.
*Acknowledgements* The first author is supported by ERCEA Consolidator Grant $615655$-NMST and also by the Basque Government through the BERC $2014-2017$ program and by Spanish Ministry of Economy and Competitiveness MINECO: BCAM Severo Ochoa excellence accreditation SEV-$2013-0323$. The second author is funded by a fellowship from CNPq Brazil. A part of this work was done when she was visiting ICTP. She warmly thanks ICTP, the Simons Associateship and Prof. C. Araujo for making this possible. We also thank the referee for several helpful suggestions.
Index of varieties
==================
\[e1\] Let $K$ be a field of characteristic $0$ and $X$ be a projective $K$-variety.
Given a smooth, quasi-projective $K$-variety $Y$, we define the associated (cohomological) *Brauer group* ${\mathrm{Br}}(Y)=H^2_{\mbox{\'{e}t}}(Y, {\mathbb{G}}_m)$.
If $K$ is the maximal unramified extension of a complete field, then $K$ is a $C_1$-field (see [@lang1]). Recall, for any $C_1$-field $K$, we have ${\mathrm{Br}}(K)=0$ (see [@ser §X.$7$]).
One can check the following elementary properties of ${\mathrm{ind}}_{{\mathrm{lin}}}(X)$:
1. If $X$ is a projective $K$-curve containing a $K$-rational point, then ${\mathrm{ind}}_{{\mathrm{lin}}}(X)=1$.
2. This is not true in higher dimension. If $K={\mathbb{C}}$ and $X$ is a very general smooth, projective quartic surface in $\operatorname{\mathbb{P}^3}$ (by Noether-Lefschetz theorem, a very general quartic has Picard rank one), then ${\mathrm{ind}}_{{\mathrm{lin}}}(X)$ is divisible by $2$ (use Riemann-Roch theorem).
3. For any $X$, ${\mathrm{ind}}_{{\mathrm{lin}}}(X)$ divides the gcd of the set of Euler characteristics of $H^{\otimes a}$ as $a$ varies over ${\mathbb{Z}}$. In particular, if $X$ is an odd degree surface in ${\mathbb{P}}^3_K$, then ${\mathrm{ind}}_{{\mathrm{lin}}}(X)=1$.
\[elw01\] Suppose $K$ is the quotient field of a Henselian discrete valuation ring $R$ with algebraically closed residue field $k$. We then have
1. if ${\mathrm{char}}(k)=0$, then ${\mathrm{ind}}(X)$ divides ${\mathrm{ind}}_{{\mathrm{lin}}}(X)$,
2. if ${\mathrm{char}}(k)=p>0$, then the prime-to-$p$ part of ${\mathrm{ind}}(X)$ divides that of ${\mathrm{ind}}_{{\mathrm{lin}}}(X)$,
The proof follows easily from [@esn2015 Theorem $3.2$].
Denote by $G$ the absolute Galois group ${\mathrm{Gal}}({\overline{K}}/K)$. An invertible sheaf ${\mathcal{L}}_{{\overline{K}}}$ on $X_{{\overline{K}}}:=X \times_K \operatorname{\mathrm{Spec}}({\overline{K}})$ is called $G$-*invariant* if for any $\sigma \in G$ and the induced morphism $\sigma:X_{{\overline{K}}} \to X_{{\overline{K}}}$, we have $\sigma^*{\mathcal{L}}_{{\overline{K}}} \cong {\mathcal{L}}_{{\overline{K}}}$.
\[elw9\] Let $H$ be an ample divisor on $X$ such that if there exists an invertible sheaf $H_0$ on $X$ with $H_0^{\otimes n} \cong H$, then $n=1$ or $-1$. Suppose that $H^1(\mathcal{O}_X)=0$, ${\mathrm{Pic}}(X_{{\overline{K}}})$ is of rank $r$, generated by ${\mathcal{L}}_1, ..., {\mathcal{L}}_{r-1}$ and ${\mathcal{L}}_r:=H_{{\overline{K}}}=H \otimes_K {\overline{K}}$ satisfying the following conditions:
1. the ideal $(\deg({\mathcal{L}}_1),\deg({\mathcal{L}}_2),...,\deg({\mathcal{L}}_r))$ in ${\mathbb{Z}}$ generated by $\deg({\mathcal{L}}_i)$ for $i=1,...,r$ coincides with the ideal $(1)$, where $\deg({\mathcal{L}}_i)$ is with respect to $H_{{\overline{K}}}$,
2. for any $r \times r$-matrix $A=(a_{i,j})$ with integral entries $a_{i,j}$, $a_{r,k}=0$ for all $k<r$, $a_{r,r}=1$, $A \not={\mathrm{Id}}$ and $A^t={\mathrm{Id}}$ for some $t>0$, we have $\sum\limits_j a_{ij} \deg({\mathcal{L}}_j)\not= \deg({\mathcal{L}}_i)$ for some $i>0$.
Then, each ${\mathcal{L}}_i$ is $G$-invariant and ${\mathrm{gcd}}\{\chi({\mathcal{L}}_i(n))| i=1,...,r \mbox{ and } n \in {\mathbb{Z}}\}=1$, where ${\mathcal{L}}_i(n):={\mathcal{L}}_i \otimes H_{_{{\overline{K}}}}^{\otimes n}$.
By [@ega43 Théorème $8.5.2$], there exists a finite field extension $K'$ of $K$ and an invertible sheaf ${\mathcal{L}}'_i$ on $X_{K'}=X \times_K \operatorname{\mathrm{Spec}}(K')$ such that ${\mathcal{L}}_i \cong {\mathcal{L}}'_i \otimes_{K'} {\overline{K}}$ i.e., the pull-back of ${\mathcal{L}}'_i$ to $X_{{\overline{K}}}$ is isomorphic to ${\mathcal{L}}_i$. Without loss of generality (replace $K'$ by the smallest Galois extension of $K$ containing $K'$), we can assume that $K'$ is a finite Galois extension of $K$. Denote by ${\mathcal{L}}'_r:=H_{K'}=H \otimes_K K'$. Let $\sigma \in {\mathrm{Gal}}(K'/K)$ and $\sigma:X_{K'} \to X_{K'}$ the induced morphism. Suppose that $$\sigma^*{\mathcal{L}}'_i \cong \bigotimes\limits_{j=1}^r ({\mathcal{L}}'_j)^{\otimes a_{i,j}} \, \mbox{ for some integer } a_{i,j}, i<r.$$ As $H$ comes from $X$, we have $\sigma^* {\mathcal{L}}'_r \cong {\mathcal{L}}'_r$. Suppose that $\sigma^*{\mathcal{L}}'_i \not\cong {\mathcal{L}}'_i$ for some $i>0$. Then there exists $j \not= i$ such that $a_{i,j} \not= 0$. In particular, the matrix $A:=(a_{i,j})$ is not the identity matrix. Note that, $a_{r,j}=0$ for all $j<r$ and $a_{r,r}=1$. Since $\sigma$ is of finite order, there exists an integer $b$ such that $${\mathcal{L}}'_i \cong (\sigma^*)^b ({\mathcal{L}}'_i)= \bigotimes\limits_{j=1}^r ({\mathcal{L}}'_j)^{\otimes b_{i,j}} \, \mbox{ where } A^b=(b_{i,j}), i=1,...,r, \, j=1,...,r.$$ In other words, $A^b={\mathrm{Id}}$. Since the Hilbert function of ${\mathcal{L}}'_i$ is the same as that of $\sigma^*{\mathcal{L}}'_i$, we conclude that $\deg({\mathcal{L}}_i')=\sum_j a_{i,j} \deg({\mathcal{L}}'_j)$. But, this contradicts our assumption $(2)$. Hence, $\sigma^*{\mathcal{L}}'_i \cong {\mathcal{L}}'_i$ for all $i=1,...,r$. In other words, each ${\mathcal{L}}_i$ is $G$-invariant.
We now prove that ${\mathrm{gcd}}\{\chi({\mathcal{L}}_i(n))|i=1,...,r \mbox{ and } n \in {\mathbb{Z}}\}=1$. Denote by $P_i(t)$ (resp. $P_0(t)$) the Hilbert polynomial of the invertible sheaf ${\mathcal{L}}_i$ (resp. $\operatorname{\mathcal{O}}_{X_{{\overline{K}}}}$) for $i=1,...,r$. Note that the leading coefficient of $Q_i(t):=P_i(t)-P_0(t)$ is $\deg({\mathcal{L}}_i)/(d-1)!$, where $d=\dim X$ (see [@huy Definition $1.2.11$]). We claim that ${\mathrm{gcd}}\{Q_i(n)| n \in {\mathbb{Z}}\}$ divides $\deg({\mathcal{L}}_i)$ for each $i=1,...,r$. Indeed, denote by $D^1Q_i(t):=Q_i(t+1)-Q_i(t)$ and recursively, $D^jQ_i(t):=D^{j-1}Q_i(t+1)-D^{j-1}Q_i(t)$. Note that, $D^jQ_i(t)$ is of degree $d-1-j$ with leading coefficient $(d-1)(d-2)...(d-j) \deg({\mathcal{L}}_i)/(d-1)!$ for all $j \ge 1$. Thus, $$D^{d-1}Q_i(t)=(d-1)!\deg({\mathcal{L}}_i)/(d-1)!=\deg({\mathcal{L}}_i).$$ It follows immediately, ${\mathrm{gcd}}\{Q_i(n)| n \in {\mathbb{Z}}\}$ divides $\deg({\mathcal{L}}_i)$. This proves the claim. Now, $${\mathrm{gcd}}\{\chi({\mathcal{L}}_i(n))|i=1,...,r \mbox{ and } n \in {\mathbb{Z}}\}\, \mbox{ divides } {\mathrm{gcd}}\{\chi({\mathcal{L}}_i(n))-\chi(\operatorname{\mathcal{O}}_{X_{{\overline{K}}}}(n)) | i=1,...,r \mbox{ and } n \in {\mathbb{Z}}\}$$ which is equal to ${\mathrm{gcd}}\{Q_i(n)|n \in {\mathbb{Z}}, \, i=1,...,r\}.$ Since the ${\mathrm{gcd}}\{Q_i(n)|n \in {\mathbb{Z}}, \, i=1,...,r\}$ divides the generator of the ideal $(\deg({\mathcal{L}}_1),\deg({\mathcal{L}}_2), ..., \deg({\mathcal{L}}_r))=(1)$, we conclude that $${\mathrm{gcd}}\{\chi({\mathcal{L}}_i(n))|i=1,...,r \mbox{ and } n \in {\mathbb{Z}}\} = 1.$$ This proves the theorem.
\[elw6\] Suppose $K$ is the quotient field of a Henselian discrete valuation ring $R$ with algebraically closed residue field $k$. If $X$ satisfies the hypothesis of Theorem \[elw9\], then ${\mathrm{ind}}(X)=1 \mbox{ if } {\mathrm{char}}(k)=0 \mbox{ and } \mbox{ prime-to-p part of } {\mathrm{ind}}(X) \mbox{ equals } 1 \mbox{ if } {\mathrm{char}}(k)=p>0.$
Recall the Brauer-Picard exact sequence: $$\label{elw13}
0 \to {\mathrm{Pic}}(X) \to {\mathrm{Pic}}(X_{{\overline{K}}})^G \xrightarrow{{\mathrm{br}}_X} {\mathrm{Br}}(K) \to {\mathrm{Br}}(X).$$ Note that, in this case ${\mathrm{Br}}(K)=0$. Hence, every $G$-invariant invertible sheaf on $X_{{\overline{K}}}$ descends to an invertible sheaf on $X$. By Theorem \[elw9\], this implies $${\mathrm{ind}}_{{\mathrm{lin}}}(X)={\mathrm{gcd}}\{\chi({\mathcal{L}}_i(n))|i=1,...,r \mbox{ and } n \in {\mathbb{Z}}\} = 1.$$ The corollary then follows directly from Lemma \[elw01\].
The corollary gives numerous examples of smooth, projective varieties with index $1$.
\[elw8\] Suppose $K$ is the quotient field of a Henselian discrete valuation ring $R$ with algebraically closed residue field $k$. Let $X$ be a smooth, projective variety with $\deg(H_{{\overline{K}}})>2$, $H^1(\mathcal{O}_X)=0$, ${\mathrm{Pic}}(X_{{\overline{K}}})$ is of rank $2$ and there exists an invertible sheaf ${\mathcal{L}}_0$ of degree coprime to $\deg(H_{{\overline{K}}})$ (for example, see Remark \[elw11\] below). Theorem \[elw9\] implies that every invertible sheaf on $X_{{\overline{K}}}$ is $G$-invariant and Corollary \[elw6\] implies that $${\mathrm{ind}}(X)={\mathrm{ind}}_{{\mathrm{lin}}}(X)=1.$$ Indeed, we simply need to check that the two conditions in Theorem \[elw9\] are satisfied. Let ${\mathcal{L}}_1$ and ${\mathcal{L}}_2:=H_{{\overline{K}}}$ be the generators of ${\mathrm{Pic}}(X_{{\overline{K}}})$. Since ${\mathcal{L}}_0$ is a linear combination of ${\mathcal{L}}_1$ and ${\mathcal{L}}_2$ and ${\mathrm{gcd}}(\deg({\mathcal{L}}_0),\deg(H_{{\overline{K}}}))=1$, we have ${\mathrm{gcd}}(\deg({\mathcal{L}}_1),\deg(H_{{\overline{K}}}))=1$. In other words, the ideal $(\deg({\mathcal{L}}_1), \deg({\mathcal{L}}_2))=1$ i.e., condition $(1)$ of Theorem \[elw9\] is satisfied. Let $$A = \left( \begin{array}{cc}
a_0 & a_1 \\
0 & 1 \end{array} \right)$$ be a matrix with integral entries. Note that, for any integer $b>0$ $$A^b = \left( \begin{array}{cc}
a_0^b & a_1(a_0^{b-1}+a_0^{b-2}+...+1) \\
0 & 1 \end{array} \right)$$ Then, $A^b={\mathrm{Id}}$ if and only if $a_0^b-1=0=a_1(a_0^{b-1}+a_0^{b-2}+...+1)$. If $a_1 \not=0$ then $a_0=-1$. Since $\deg({\mathcal{L}}_2)>2$ is coprime to $\deg({\mathcal{L}}_1)$, we have $2\deg({\mathcal{L}}_1) \not= a_1\deg({\mathcal{L}}_2)$ for any integer $a_1$. Thus condition $(2)$ of Theorem \[elw9\] is satisfied. Hence, by Theorem \[elw9\] and Corollary \[elw6\], we conclude that ${\mathcal{L}}_1$ is $G$-invariant and ${\mathrm{ind}}(X)={\mathrm{ind}}_{{\mathrm{lin}}}(X)=1.$
\[elw11\] Fix coordinates $X_0, X_1, X_2, X_3$ on ${\mathbb{P}}^3_{{\mathbb{Q}}}$. Take any $d \ge 4$ and $F_1, F_2$ two homogeneous polynomials of degree $d$ in variables $X_i$ and coefficients in ${\mathbb{Q}}$. It is easy to check that for general $F_1, F_2$, the surface defined by $F:=F_1X_1+F_2X_2$ is smooth and ${\mathrm{rk}}({\mathrm{Pic}}(X)) = 2$ (the hyperplane section and the line defined by $X_1, X_2$ generate ${\mathrm{Pic}}(X)$). For any prime $p$, take $K={\mathbb{Q}}_p^{ur}$, the maximal unramified extension of ${\mathbb{Q}}_p$. Note that $K$ is a $C_1$-field, hence ${\mathrm{Br}}(K)=0$. Let $X$ be the surface in ${\mathbb{P}}^3_K$ defined by $F$. By [@R3 Ex. III.$5.5$], we have $H^1(\operatorname{\mathcal{O}}_X)=0$. Then Example \[elw8\] implies that ${\mathrm{ind}}_{{\mathrm{lin}}}(X)=1={\mathrm{ind}}(X)$. One can similarly construct numerous examples of surfaces in ${\mathbb{P}}^3_K$ of any degree, satisfying the conditions of Theorem \[elw9\], arising from the theory of Noether-Lefschetz locus (see [@v2; @v3]), thereby having index $1$.
We now give some examples in the case ${\mathrm{Br}}(K) \not= 0$, in particular $K$ is not a $C_1$-field.
\[elw02\] Let $K={\mathbb{R}}$ and $G:={\mathrm{Gal}}({\overline{K}}/K)$ the absolute Galois group. Recall, ${\mathrm{Br}}({\mathbb{R}})={\mathbb{Z}}/2{\mathbb{Z}}$. Denote by ${\mathrm{br}}_X: {\mathrm{Pic}}(X_{{\overline{K}}})^G \to {\mathrm{Br}}({\mathbb{R}})$ as in the Brauer-Picard exact sequence .
1. We first consider the case, when ${\mathrm{br}}_X$ is the zero map. Let $X$ be the smooth, projective surface in $\mathbb{P}^3_{{\mathbb{R}}}$ defined by the equation $X_0^2+X_1^2+X_2^2-X_3^2=0$, where $X_i$ are the coordinates of $\operatorname{\mathbb{P}^3}$ for $i=0,...,3$. Note that, $X_{{\mathbb{C}}}:=X \times_{{\mathbb{R}}} \operatorname{\mathrm{Spec}}({\mathbb{C}})$ contains the two lines $L_1:=Z(X_0-iX_1, X_2-X_3)$ and $L_2:=Z(X_0+iX_1, X_2-X_3)$. The element of the absolute Galois group $G$ sending $i$ to $-i$ interchanges $L_1$ and $L_2$. Since ${\mathrm{Pic}}(X_{{\mathbb{C}}})={\mathbb{Z}}^{\oplus 2}$, generated by $L_1$ and $L_2$, we conclude that the $G$-invariant subgroup of ${\mathrm{Pic}}(X_{{\mathbb{C}}})$ is generated as a ${\mathbb{Z}}$-module by $L_1+L_2$, which is linearly equivalent to the hyperplane section $H_{{\mathbb{C}}}:=H \otimes_{{\mathbb{R}}} {\mathbb{C}}$. Therefore, in this case ${\mathrm{Pic}}(X_{{\mathbb{C}}})^G$ consists of points corresponding to multiples of the invertible sheaf $H_{{\mathbb{C}}}$. Since $H_{{\mathbb{C}}}$ comes from $X$, the exactness of implies that ${\mathrm{br}}_X$ is the zero map. It is easy to check that ${\mathrm{ind}}_{{\mathrm{lin}}}(X)=1$ (Euler characteristic of $\operatorname{\mathcal{O}}_X$ is $1$). Note that, ${\mathrm{ind}}(X)=1$ as $X$ contains ${\mathbb{R}}$-rational points.
2. We now consider the case, when ${\mathrm{br}}_X$ is non-trivial (equivalently surjective). Let $X$ be the ${\mathbb{R}}$-plane conic defined by $X_0^2+X_1^2+X_2^2=0$, where $X_i$ are the coordinates of ${\mathbb{P}}^2_{{\mathbb{R}}}$, for $i=0,1,2$. In this case, $h^1(\operatorname{\mathcal{O}}_X)=0$, i.e., ${\mathrm{Pic}}^0(X)=0$. This implies, there exists an unique invertible sheaf ${\mathcal{L}}_{{\overline{K}}}$ on $X_{{\overline{K}}}$ of degree $1$, hence it is $G$-invariant. If ${\mathrm{br}}_X({\mathcal{L}}_{{\overline{K}}})=0$, then by the exact sequence there exists an invertible sheaf ${\mathcal{L}}$ on $X$ such that ${\mathcal{L}}_{{\overline{K}}} \cong {\mathcal{L}} \otimes_K {\overline{K}}$. Since $\deg({\mathcal{L}})=1$, the Riemann-Roch theorem would then imply that $X$ contains a rational point, which gives us a contradiction. Hence, ${\mathrm{br}}_X$ is non-trivial. Using [@koll ($1.3$)], we have ${\mathrm{ind}}_{{\mathrm{lin}}}(X)$ is the gcd of ${\mathrm{ind}}(X)$ and $1-\rho_a(X)$. Since ${\mathrm{ind}}(X)=2$, we have ${\mathrm{ind}}_{{\mathrm{lin}}}(X)={\mathrm{ind}}(X)=2$ (observe $\rho_a(X)=1$).
[ELW15]{}
P. Baum, W. Fulton, and R. MacPherson. Riemann-roch for singular varieties. , 45:101–145, 1975.
H. Esnault, M. Levine, and O. Wittenberg. Index of varieties over [H]{}enselian fields and [E]{}uler characteristic of coherent sheaves. , 24(4):693–718, 2015.
A. Grothendieck. l[é]{}ments de g[é]{}om[é]{}trie alg[é]{}brique (r[é]{}dig[é]{}s avec la collaboration de [J]{}ean [D]{}ieudonn[é]{}): [IV]{}. [é]{}tude locale des sch[é]{}mas et des morphismes de sch[é]{}mas, troisi[è]{}me partie. , 28:5–255, 1966.
R. Hartshorne. . Graduate text in Mathematics. Springer-Verlag, 2010.
D. Huybrechts and M. Lehn. . Springer, 2010.
I. Kaur. . Ph. d. thesis, Freie University Berlin, 2016.
I. Kaur. A pathological case of the [$C_1$]{} conjecture in mixed characteristic. , 2018.
J. Koll[á]{}r. Esnault-[L]{}evine-[W]{}ittenberg indices. , 2013.
S. Lang. On quasi algebraic closure. , pages 373–390, 1952.
J. P. Serre. , volume 67. Springer Science & Business Media, 2013.
C. Voisin. Une précision concernant le théorème de [N]{}oether. , 280(4):605–611, 1988.
C. Voisin. Composantes de petite codimension du lieu de [N]{}oether-[L]{}efschetz. , 64(4):515–526, 1989.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The first measurement of the helicity dependence of the photoproduction cross section of single neutral pions off protons is reported for photon energies from 600 to 2300MeV, covering nearly the full solid angle. The data are compared to predictions from the SAID, MAID, and BnGa partial wave analyses. Strikingly large differences between data and predictions are observed which are traced to differences in the helicity amplitudes of well known and established resonances. Precise values for the helicity amplitudes of several resonances are reported.'
author:
- 'M.Gottschall'
- 'A.V.Anisovich'
- 'B.Bantes'
- 'D.Bayadilov'
- 'R.Beck'
- 'M.Bichow'
- 'S.Böse'
- 'K.-Th.Brinkmann$^,$'
- 'Th.Challand'
- 'V.Crede'
- 'F.Dietz'
- 'H.Dutz'
- 'H.Eberhardt'
- 'D.Elsner'
- 'R.Ewald'
- 'K.Fornet-Ponse'
- 'St.Friedrich'
- 'F.Frommberger'
- 'Ch.Funke'
- 'A.Gridnev'
- 'M.Grüner'
- 'E.Gutz$^,$'
- 'Ch.Hammann'
- 'J.Hannappel'
- 'J.Hartmann'
- 'W.Hillert'
- 'Ph.Hoffmeister'
- 'Ch.Honisch'
- 'I.Jaegle'
- 'D.Kaiser'
- 'H.Kalinowsky'
- 'S.Kammer'
- 'I.Keshelashvili'
- 'F.Klein'
- 'E.Klempt'
- 'K.Koop'
- 'B.Krusche'
- 'M.Kube'
- 'M.Lang'
- 'I.Lopatin'
- 'Y.Maghrbi'
- 'K.Makonyi'
- 'V.Metag'
- 'W.Meyer'
- 'J.Müller'
- 'M.Nanova'
- 'V.Nikonov$^,$'
- 'R.Novotny'
- 'D.Piontek'
- 'G.Reicherz'
- 'T.Rostomyan'
- 'A.Sarantsev'
- 'St.Schaepe'
- 'Ch.Schmidt'
- 'H.Schmieden'
- 'R.Schmitz'
- 'T.Seifen'
- 'V.Sokhoyan'
- 'A.Thiel'
- 'U.Thoma'
- 'M.Urban'
- 'H.vanPee'
- 'D.Walther'
- 'Ch.Wendel'
- 'U.Wiedner'
- 'A.Wilson'
- 'A.Winnebeck.'
title: 'First measurement of the helicity asymmetry for $\gamma p\rightarrow p\pi^0$ in the resonance region'
---
0.1cm
The spin structure of the proton has been of topical interest since the discovery that the quark spins constitute an unexpectedly small fraction of the proton spin [@CERN-EP-87-230]. Deep inelastic scattering experiments have revealed gluonic contributions to the proton spin to be consistent with zero, at least within the admittedly large errors [@Franco:2012mi]. At low energies, a different aspect of the proton spin structure is tested by a comparison of the helicity dependence of the total $\gamma p$ cross section integrated over all photon energies, $\int_0^\infty \mathrm
dE_\gamma\,(\sigma_{3/2} -\sigma_{1/2})/E_\gamma$, with the proton magnetic moment [@arXiv:nucl-ex/0603021], a relation which is known as Gerasimov-Drell-Hearn (GDH) sum rule [@48796; @SLAC-PUB-0187]. The subscripts denote the total helicity, $h$=$1/2$ for photon and proton spin anti-aligned, $h$=$3/2$ for both spins aligned. The GDH integral sums over all energies and all final states. A breakdown of the GDH integral into exclusive final states can provide a link between inclusive properties of the proton like its magnetic moment and the contributions of specific nucleon resonances to the GDH integral.
In this letter, we report the first measurement of the double polarization observable $$E=
\frac{\left(\frac{\mathrm d\sigma}{\mathrm d\Omega}\right)_{1/2}-
\left(\frac{\mathrm d\sigma}{\mathrm d\Omega}\right)_{3/2}}
{\left(\frac{\mathrm d\sigma}{\mathrm d\Omega}\right)_{1/2}+
\left(\frac{\mathrm d\sigma}{\mathrm d\Omega}\right)_{3/2}}
\label{E}$$ for the exclusive reaction $\gamma p\to p\pi^0$ in the energy range from 600 to 2300MeV, and compare the results with predictions of well established partial wave analyses (PWA) like SAID [@SAID-SN11; @SAID-CM12], MAID [@Drechsel:2007if], and BnGa [@Anisovich:2011fc]. The double polarization observable $G$ [@Thiel:2012yj] – governing the correlation between linearly polarized photons and longitudinally polarized target protons – revealed remarkable differences in the predictions of the three partial wave analysis groups even in the well-studied 2$^{\rm nd}$ resonance region around $E_\gamma=$750MeV [@Comment]. Here, we extend the covered energy regime to 2.3GeV for the double polarization observable $E$. So far, data on $E$ were published up to 780MeV for a limited angular range [@Ahrens:2002; @Ahrens:2004]; further data from CLAS exist and were reported at a conference [@Iwamoto:2012zza].
------------------------------------------------------ ----- ----------------------------------------------------------
{width="48.50000%"} {width="48.90000%"}
------------------------------------------------------ ----- ----------------------------------------------------------
The experiment was carried out using the tagged photon beam of the ELectron Stretcher Accelerator ELSA at Bonn [@Hillert:2006yb]. Photons with circular polarization [@Olsen:1959zz], $$P_{\odot}=\frac{4x-x^2}{4-4x+3x^2}P_{\rm e^-}\quad\mbox{with~}x=\frac{E_{\gamma}}{E_{\rm e^-}},$$ were produced by scattering a 2.4GeV beam of longitudinally polarized electrons with polarization $P_{\rm e^-}$ off a bremsstrahlung target. The electrons were deflected by a magnet into a tagging hodoscope which defines the energy of the bremsstrahlung photons. The electron polarization ($P_{\rm e^-}$$\approx$0.60) was monitored in a M[ø]{}ller polarimeter [@kammer]. The photon beam impinged on a butanol ($\rm C_4H_9OH$) target, which was polarized by dynamic nucleon polarization [@Dutz:2004eb]. An average polarization of ${P_T}$$\approx$0.71 was obtained.
Particles produced by photoproduction off the target were detected in a 4$\pi$ geometry using several sub-detectors. The target was surrounded by an inner scintillating fiber detector designed to detect charged particles [@Suft:2005cq], and by the CsI(Tl) Crystal Barrel detector [@CB]. A forward calorimeter of 90 CsI(Tl) crystals and a TAPS wall of 216 BaF$_2$ crystals [@TAPS] completed the calorimeter setup in the forward direction. For charged particle detection, both forward calorimeters were equipped with plastic scintillation counters in front of the crystals. With this setup, the four-momenta of neutral mesons decaying into photons could be determined by measuring their energies and directions from the target center to the shower center. The reaction $\gamma p\to
p\pi^0\to p\gamma\gamma$ was reconstructed by a series of kinematic cuts. Events with exactly two photons and a proton were used. The latter was demanded in order to suppress background from reactions off neutrons. For the proton candidate, only a hit in one of the charged particle detectors was required to include also low energy protons, which did not reach the calorimeter. The invariant mass of the two photons and the missing mass was calculated and a [$\pm2\sigma$]{} cut around the $\pi^0$ and the proton mass was applied. The azimuthal angle between the direction of proton and meson was asked to be 180$^{\circ}$ within a $\pm2\sigma$ window (coplanarity). To remove untagged events originating from photons below the tagging threshold (due to random coincidences), the beam photon energy was also calculated from the reaction and compared to the measured photon energy. Finally, a time coincidence was required between the tagger hit and the reaction products and random time background was subtracted.
The helicity-dependent differential cross section for photoproduction of neutral pions off longitudinally polarized protons can be written as $$\frac{\mathrm d\sigma}{\mathrm d\Omega} =
\frac{\mathrm d\sigma_0}{\mathrm d\Omega}\,(1\pm P_TP_{\odot}{E})\,.
\label{sigma}$$ When a butanol ($\rm C_4H_9OH$) target is used, unpolarized $(u)$ free protons $(f)$ as well as nucleons bound $(b)$ in carbon or oxygen contribute to the count rate, in addition to the polarized $(p)$ free protons. For the same number of beam photons for both photon helicities, and under the assumption that (unpolarized) nucleons bound in carbon have the same response to impinging photons as nucleons bound in oxygen, we get for beam and target polarized in the same direction the yield $N_{3/2}$= $N^{f,p}_{3/2}$+$N^{f,u}$+$N^b$ and in opposite direction $N_{1/2}$=$N^{f,p}_{1/2}$+$N^{f,u}$+$N^b$, which leads to [$${E}=
\frac{N_{1/2}-N_{3/2}}{N_{1/2}+N_{3/2}}\cdot \frac{1}{d}\cdot
\frac{1}{P_{\odot}P_T}\,.
\label{counts}$$]{} The bound nucleons are taken into account by the dilution factor $d=\frac{N^f}{N^f+N^b}$. It was determined by an additional measurement using a carbon foam target with approximately the same density, size, and environment as the carbon and oxygen part of the butanol target. The fraction of carbon (and oxygen) in butanol was then determined using two alternative methods [@Diss]: by fitting a) the missing mass and b) the coplanarity distribution outside of the signal of the free protons. The resulting scaling factor $s(E_{\gamma})$ for each energy bin was then used to determine [$$d(E_{\gamma},\theta_{\pi})= \frac{N_{\rm
C_4H_9OH}(E_{\gamma},\theta_{\pi})-s(E_{\gamma})\cdot
N_{C}(E_{\gamma},\theta_{\pi})}{N_{\rm
C_4H_9OH}(E_{\gamma},\theta_{\pi})}\,.$$]{} Fig. \[pic:Sigma\] shows on the left the total cross section $\sigma_0$ and its decomposition into the two helicity components $\sigma_0$=$1/2 (\sigma_{1/2}$+$\sigma_{3/2}$). The latter cross sections were calculated from the values for $E$ reported below - and extrapolated into the unmeasured angular bins by the BnGa-PWA - and the total cross section $\sigma_0$ (from the BnGa-PWA). The figure reveals important details of the spin structure of the photo-excitation of the proton; $\sigma_{1/2}$ and $\sigma_{3/2}$ evidence quite different structures. Both cross sections show a peak at $W$$\approx$1500MeV in $\sigma_{1/2}$ and, slightly shifted, at $W$$\approx$1520MeV in $\sigma_{3/2}$. The latter peak can be assigned to the N(1520)$3/2^-$ and its large helicity coupling to $h$=$3/2$. The lower mass peak cannot be solely due to the N(1535)$1/2^-$: contributions from other resonances or from background amplitudes, e.g. in the $1/2^-$ partial wave, must be significant. The N(1680)$5/2^+$ contributes nearly only to $\sigma_{3/2}$, where a clear peaking structure is observed. A broad structure at $W$$\approx$1900MeV is very visible in the $\sigma_{3/2}$ excitation curve and seems to be absent in $\sigma_{1/2}$. Photons in the $E_{\gamma}\approx$1500MeV range may excite some of the resonances with spin $J=1/2,\cdots,7/2$ in the $W$=1900-2000MeV mass region. Their contribution is also well visible in the cross section difference $\sigma_{3/2}$-$\sigma_{1/2}$=$\,-E\cdot 2\sigma$ in Fig. \[pic:Sigma\] on the right. Obviously, they are preferentially excited by the $A_{3/2}$ helicity amplitude.
The helicity asymmetry $E$ in eq.(\[counts\]) is a function of the pion production angle $\theta_{\pi}$. Fig. \[pic:results\] shows selected results. There are very significant changes in these distributions as a function of energy. Obviously, the quantity $E$ is very sensitive to the contributions from baryon resonances. The lowest incident photon energy bin covers the part of the region with positive $E$. The shape indicates a strong contribution from the $J^P=1/2^-$ and $J^P=3/2^-$ partial waves, with N(1535)$1/2^-$ and N(1520)$3/2^-$ being the dominant resonances in this energy range. The asymmetry in the angular distribution reflects the presence of several weakly contributing resonances, like N(1440)$1/2^+$ and $\Delta(1232)3/2^+$. The next bin, $E_{\gamma}$=960-1100MeV, covers N(1680)$5/2^+$. Together with the $J^P=1/2^-$ partial wave, it produces a W-shaped angular distribution. The strong forward-backward asymmetry signals contributions from other partial waves, among which the $D_{33}$ partial wave including the $\Delta(1700)3/2^-$ plays an important role. We have split the 4$^{\rm th}$ resonance region from 1350 to 1650MeV into three 100MeV slices. The basic structure remains over the full energy range, which suggests a sizeable contribution from a $J^P=7/2^+$ resonance, which would lead to a three minima structure in the angular distribution. There are only relatively small differences when the three energy bins are compared: $\Delta(1950)7/2^+$ seems to be the dominant contribution. The highest mass bin exhibits strong structures but no clear pattern; several resonances seem to make comparable contributions to the data.
![\[pic:results\]The helicity asymmetry $E$ as a function of $\cos\theta_{\pi}$ for selected $E_{\gamma}$ bins. PWA predictions: black dashed curve: BnGa2011-02, blue solid curve: MAID, red solid curve: CM12, red dashed curve: SN11. Fit to the data points (BnGa2011-02E): black solid curve.](E_new_fig2.eps){width="46.00000%" height="0.495\textheight"}
MAID2007 SN11 CM12 BnGa2011 BnGa2011E MAID2007 SN11 CM12 BnGa2011 BnGa2011E
---------------- ---------- ----------- ----------- ------------ ------------ ---------------- ---------- ----------- ------------ ------------ ------------
N(1440) N(1520)
M 1440 1485 1485 1430$\pm$8 1430$\pm$8 1530 1515 1515$\pm$3 1517$\pm$3 1516$\pm$2
$\Gamma$ 350 284 284 365$\pm$35 360$\pm$30 130 104 104$\pm$5 114$\pm$5 113$\pm$5
BR(N$\pi$) 70 79 79 62$\pm$3 62$\pm$3 60 63 63$\pm$3 62$\pm$3 62$\pm$2
A$_{1/2}$ -61 -58$\pm$1 -56$\pm$1 -61$\pm$6 -62$\pm$8 -27 -16$\pm$2 -19$\pm$2 -22$\pm$4 -20$\pm$3
A$_{3/2}$ 161 156$\pm$2 141$\pm$2 131$\pm$10 131$\pm$7
N(1535) N(1650)
M 1535 1547 1547 1519$\pm$5 1518$\pm$4 1690 1635 1635 1651$\pm$6 1651$\pm$6
$\Gamma$ 100 188 188 128$\pm$14 125$\pm$10 100 115 115 104$\pm$10 102$\pm$10
BR(N$\pi$) 40 36 36 54$\pm$5 55$\pm$5 85 100 100 51$\pm$4 50$\pm$4
A$_{1/2}$ 66 99$\pm$2 128$\pm$4 105$\pm$10 105$\pm$9 33 65$\pm$25 55$\pm$30 33$\pm$7 33$\pm$7
$\Delta$(1620) $\Delta$(1950)
M 1620 1615 1615 1600$\pm$8 1598$\pm$6 1945 1921 1921 1915$\pm$6 1915$\pm$5
$\Gamma$ 150 147 147 130$\pm$11 130$\pm$8 280 271 271 246$\pm$10 249$\pm$8
BR(N$\pi$) 25 32 32 28$\pm$3 28$\pm$3 40 47 47 45$\pm$2 46$\pm$2
A$_{1/2}$ 66 64$\pm$2 29$\pm$3 52$\pm$5 52$\pm$5 -94 -71$\pm$2 -83$\pm$4 -71$\pm$4 -70$\pm$5
A$_{3/2}$ -121 -92$\pm$2 -96$\pm$4 -94$\pm$5 -93$\pm$5
Next we compare the data in Figs. \[pic:Sigma\] and \[pic:results\] with predictions from SAID (SN11 and CM12 [@SAID-SN11; @SAID-CM12]), MAID [@Drechsel:2007if] and BnGa2011-02 [@Anisovich:2011fc]. In addition, a fit to the data within the BnGa-PWA (BnGa2011-02E) is shown. First we note that all PWAs give a reasonable description of $\sigma^{\gamma p\to p\pi^0}$. But we find large and unexpected discrepancies already at rather low energies, in the region of the four-star resonances N(1440)$1/2^+$, N(1535)$1/2^-$ and N(1520)$3/2^-$. In the $\Delta(1950)7/2^+$ region, MAID and SAID show a striking enhancement at $\cos\theta_{\pi}\approx -0.45$. In the MAID online version, the enhancement disappears when the helicity couplings of the N(1520)$3/2^-$ are reduced by a factor of 4. The N(1520)$3/2^-$, however, is the only resonance in this partial wave; hence, such a large effect, which is presumably due to interference with other partial waves, is unexpected. In SN11, this structure is smaller than in MAID, and is further reduced in CM12. The origin of the structure, also visible in the SAID prediction, remains unclear. None of the fits reproduces the data over the full angular range. However, BnGa2011-02 required only a slight adjustment of free parameters to get close to the data points (solution BnGa2011-02E). For examples see Table \[tab:nstars\]. It compares selected results on low-mass excitations of the nucleon for MAID and for the two latest solutions from SAID and BnGa, respectively. One formal difference is remarkable between MAID and SAID on the one hand, and BnGa on the other hand: masses, widths and branching ratios are fixed from $\pi$N elastic scattering in MAID and SAID. This approach excludes the possibility of photoproduction providing more than the photo-couplings of resonances. Results of the multichannel BnGa-PWA demonstrate however the importance of including information from photon induced and especially double polarization measurements for the determination of nucleon resonance parameters in general.
Finally, we come back to the GDH integral. The contribution of single $\pi^0$ production to the GDH integral is evaluated. The total GDH integral up to 2.9GeV was determined to $I^{\rm
2.9}_{\text{tot}}=226\pm5^{\text{stat}}\pm12^{\text{syst}}$$\mu$b [@Ahrens:2000bc; @Ahrens:2001qt; @Dutz:2003mm; @Dutz:2004zz; @Dutz:2005ns]; the results were summarized in [@arXiv:nucl-ex/0603021]. Here, a contribution of $27.5$$\mu$b had been subtracted to account for the description of the very low energy region by MAID [@Drechsel:2007if]. Restricted to a maximum energy of 2.3GeV, the total integral results in $227\pm5^{\text{stat}}$$\mu$b and we find $159\pm8^{\text{stat}}$$\mu$b for the contribution of single $\pi^0$ production. Up to 0.6GeV photon energy, this contribution amounted to $156\pm8^{\text{stat}}$$\mu$b: above $\Delta(1232)$, there is practically no contribution of single $\pi^0$ production to the GDH integral.
Summarizing, we have reported the first measurement of the helicity dependent photoproduction cross section for photons at $E_{\gamma}$=0.6-2.3GeV with nearly full solid angular coverage. The observable $E$ is shown to be highly sensitive to the contributions from $s$-channel resonances. Even after many years of studying the simplest photoproduction reaction, $\gamma p\to
p\pi^0$, the new data reveal very significant discrepancies in comparison with model predictions. As expected, it is obvious that a fit to differential cross sections and to single polarization observables alone is not sufficient to arrive at unambiguous solutions.
We thank the technical staff of ELSA and the participating institutions for their invaluable contributions to the success of the experiment. We acknowledge support from the *Deutsche Forschungsgemeinschaft* (SFB/TR16) and *Schweizerischer Nationalfonds*.
[99]{} J. Ashman [*et al.*]{}, Phys. Lett. B [**206**]{}, 364 (1988). C. Adolph [*et al.*]{}, Phys. Rev. D [**87**]{}, 052018 (2013). K. Helbing, Prog. Part. Nucl. Phys. [**57**]{}, 405 (2006). S. D. Drell and A. C. Hearn, Phys. Rev. Lett. [**16**]{}, 908 (1966). S. B. Gerasimov, Sov. J. Nucl. Phys. [**2**]{}, 430 (1966) \[Yad. Fiz. [**2**]{}, 598 (1965)\]. R. L. Workman [*et al.*]{}, Phys. Rev. C [**85**]{}, 025201 (2012). R. L. Workman [*et al.*]{}, Phys. Rev. C [**86**]{}, 015202 (2012). D. Drechsel, S. S. Kamalov and L. Tiator, Eur. Phys. J. A [**34**]{}, 69 (2007). A. V. Anisovich [*et al.*]{}, Eur. Phys. J. A [**48**]{}, 15 (2012). A. Thiel [*et al.*]{}, Phys. Rev. Lett. [**109**]{}, 102001 (2012). R. L. Workman [*et al.*]{}, Phys. Rev. Lett. [**110**]{}, 169101 (2013); A. Thiel [*et al.*]{}, Phys. Rev. Lett. [**110**]{}, 169102 (2013). J. Ahrens [*et al.*]{}, Phys. Rev. Lett. [**88**]{}, 232002 (2002). J. Ahrens [*et al.*]{}, Eur. Phys. J. A [**21**]{}, 323 (2004). H. Iwamoto, PhD thesis, GWU, 2011; and AIP Conf. Proc. [**1432**]{}, 275 (2012). W. Hillert, Eur. Phys. J. A [**28S1**]{}, 139 (2006). H. Olsen and L. C. Maximon, Phys. Rev. [**114**]{}, 887 (1959). S. Kammer, PhD thesis, Universität Bonn, 2010. H. Dutz, Nucl. Instrum. Meth. A [**526**]{}, 117 (2004). G. Suft [*et al.*]{}, Nucl. Instrum. Meth. A [**538**]{}, 416 (2005). E. Aker [*et al.*]{}, Nucl. Instrum. Meth. A [**321**]{}, 69 (1992). R. Novotny, IEEE Trans. Nucl. Sci. [**38**]{}, 379 (1991). M. Gottschall, Ph.D thesis, Universität Bonn, 2013. J. Ahrens [*et al.*]{}, Phys. Rev. Lett. [**84**]{}, 5950 (2000). J. Ahrens [*et al.*]{}, Phys. Rev. Lett. [**87**]{}, 022003 (2001). H. Dutz [*et al.*]{}, Phys. Rev. Lett. [**91**]{}, 192001 (2003). H. Dutz [*et al.*]{}, Phys. Rev. Lett. [**93**]{}, 032003 (2004). H. Dutz [*et al.*]{}, Phys. Rev. Lett. [**94**]{}, 162001 (2005). H. van Pee [*et al.*]{}, Eur. Phys. J. A [**31**]{}, 61 (2007).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The primary $\gamma$-ray spectra of $^{166}$Er and $^{167}$Er are deduced from the ($^3$He,$\alpha \gamma$) and ($^3$He,$^3$He’$\gamma$) reaction, respectively, enabling a simultaneous extraction of the level density and the $\gamma$-ray strength function. Entropy, temperature and heat capacity are deduced from the level density within the micro-canonical and the canonical ensemble, displaying signals of a phase-like transition from the pair-correlated ground state to an uncorrelated state at $T_c\sim 0.5$ MeV. The $\gamma$-ray strength function displays a bump around $E_{\gamma}\sim 3$ MeV, interpreted as the pygmy resonance.'
address:
- 'Department of Physics, University of Oslo, N-0316 Oslo, Norway'
- 'Frank Laboratory of Neutron Physics, Joint Institute of Nuclear Research, 141980 Dubna, Moscow reg., Russia'
author:
- 'E. Melby, M. Guttormsen, J. Rekstad, A. Schiller, and S. Siem'
- 'A. Voinov'
title: 'Thermal and electromagnetic properties of $^{166}$Er and $^{167}$Er'
---
[2]{}
Introduction
============
The energy distribution of primary $\gamma$-rays provides information on both the level density and the $\gamma$-ray strength function. The present aim is to study the level density and $\gamma$-strength function of the nuclei $^{166,167}$Er and to compare with other nuclei in this mass region. In addition, thermodynamic and electromagnetic properties of the two nuclei will be investigated.
Most of the experimental information on the level density of rare earth nuclei originates from the neutron-resonance spacing at the neutron-separation energy [@ilj92] and direct counting of levels in the vicinity of the ground state [@toi]. Experimental knowledge of the $\gamma$-strength function is mainly based on the study of photo-absorption cross sections [@AogN] and radiative neutron capture [@becvar].
A new method [@andreas:nim] derives the level density and $\gamma$-strength function simultaneously from a set of primary $\gamma$ spectra. This technique has proven to give a valuable supplement to the more traditional methods for level density extraction, and to our knowledge, it represents the least model-dependent method to obtain the $\gamma$ strength function over a wide energy region below the neutron-separation energy.
The nuclear level density is closely related to the thermodynamic properties of nuclei, and can therefore be utilized to deduce e.g. entropy, temperature and heat capacity of nuclei. The thermodynamic observables may reveal phase transitions.
The $\gamma$-strength function is a measure for the average electromagnetic properties of nuclei and has a fundamental importance for the understanding of nuclear structure and reactions involving $\gamma$ rays [@kop]. At $\sim 3$ MeV of $\gamma$ energy, a bump is observed in the $\gamma$-strength function of rare earth nuclei from ($^3$He,$\alpha$) experiments [@voinov], and is probably of the same origin as the pygmy resonance found in $(n,\gamma)$ reactions [@iga].
Section II describes the experimental methods. In Sect. III the experimental level density and $\gamma$-strength function of $^{166}$Er and $^{167}$Er are obtained. Section IV examines thermodynamic properties within the micro-canonical and the canonical ensemble, while electromagnetic properties of the two nuclei are discussed in Sect. V. Concluding remarks are given in Sect. VI.
Experimental method and techniques
==================================
The experiment was carried out at the Oslo Cyclotron Laboratory. The reactions employed were the neutron pick-up ($^3$He,$\alpha \gamma$) and the inelastic scattering ($^3$He,$^3$He’$\gamma$) with a beam energy of $45$ MeV, populating $^{166}$Er and $^{167}$Er with high excitation energy and low spin. The $Q$-value of the ($^3$He,$\alpha$)-reaction is $14.142$ MeV. The self-supporting $^{167}$Er target was isotopically enriched to $95.6$% and had a thickness of $1.5$ mg/cm$^2$.
The charged particles and $\gamma$ rays were recorded with the detector array CACTUS [@mg90], which contains 8 Si particle telescopes and 27 NaI $\gamma$-ray detectors. Each telescope is placed at an angle of 45$^{\circ}$ relative to the beam axis, and comprises one Si front and one Si(Li) back detector with thickness $140$ and $3000$ [$\mu$]{}m, respectively. The NaI $\gamma$-detector array, having a resolution of $\sim 6$ % at $\gamma$ energy $E_{\gamma} = 1$ MeV and a total efficiency of $\sim15$ %, surrounds the target and particle detectors. In addition, two Ge detectors were used to monitor the spin distribution and selectivity of the reactions.
The excitation energy of the resulting $^{166}$Er and $^{167}$Er nuclei are determined by means of reaction kinematics of the ejectile. By setting proper gates in the particle spectra, each coincident $\gamma$ ray can be assigned to a $\gamma$ cascade originating from a specific excitation energy. The data are sorted into a matrix of $(E,E_{\gamma})$ energy pairs, $E$ being the excitation energy. Examples of the recorded $\gamma$ spectra, the so called raw spectra, from two excitation energies are shown in the left panel of Figs. \[fig:adata\] and \[fig:tdata\] for $^{166}$Er and $^{167}$Er, respectively. Note that the statistics of $^{167}$Er is about twice as good as for $^{166}$Er, since the ($^3$He,$^3$He’) reaction has a higher cross section than the ($^3$He,$\alpha$) reaction.
In order to determine the true $\gamma$-energy distribution, the $\gamma$ spectra are corrected for the response of the NaI detectors with the unfolding procedure of Ref. [@mg96]. Unfolded $\gamma$ spectra are shown in the central panel of Figs. \[fig:adata\] and \[fig:tdata\] for the two nuclei.
The now corrected $(E,E_{\gamma})$ matrix comprises the $\gamma$-energy distribution of the total $\gamma$ cascade, and make it possible to derive the primary $\gamma$ matrix according to the subtraction technique of Refs. [@mg87; @lisa]. Primary $\gamma$ spectra can be seen in the right panel of Figs. \[fig:adata\] and \[fig:tdata\].
The method of extracting the primary $\gamma$ spectra is based on the assumption that the decay properties of a bin of excited states are independent of whether the states are directly populated through the nuclear reaction or from de-excitation from higher excited states. This is believed to be approximately fulfilled because the long life time of excited states give the nucleus time to thermalize prior to the $\gamma$ decay.
Extraction of level density and $\gamma$-ray strength function
==============================================================
The primary $\gamma$ matrix provides information on both the level density and the $\gamma$-ray strength function, enabling a simultaneous determination of the two functions. The fundamental assumption behind the extraction procedure is the Brink-Axel hypothesis [@brink; @axel], where the probability of $\gamma$-decay in the statistical regime, represented by the primary $\gamma$ matrix $P(E_i, E_{\gamma})$, can be expressed simply as a product of the final-state level density $\rho(E-E_{\gamma})$ and a $\gamma$-energy dependent factor $F(E_{\gamma})$ $$P(E, E_{\gamma}) \propto F(E_{\gamma}) \rho(E-E_{\gamma}).
\label{eq:ab}$$ The $\gamma$-energy dependent factor is proportional to $\sum_{XL}E_{\gamma}^{2L + 1}f_{XL}(E_{\gamma})$, where $f_{XL}(E_{\gamma})$ is the $\gamma$-ray strength function for the multipolarity $XL$.
During the last few years a method for simultaneous deduction of level density and $\gamma$-strength function from primary $\gamma$ spectra has been developed at the Oslo Cyclotron Laboratory [@andreas:nim; @lisa; @trine]. In the previously published results on $^{166}$Er [@elin], the level densities and $\gamma$-strength functions were still unnormalized. The procedure is now fully replaced by an iteration technique deriving the level density and the $\gamma$-energy dependent factor by a $\chi^2$-fit to the experimental primary $\gamma$ spectra [@andreas:nim]. This technique is described in detail in Ref. [@andreas:nim], and only an outline of the normalization procedures is given here. It is important to note that the extracted experimental quantity $\rho$ is in fact the density of levels accessible to the nuclear system in the $\gamma$-decay process. This quantity is interpreted as the level density, but may be influenced by selection rules in the $\gamma$ decay. For a thermalized nucleus in the continuum, the interpretation is approximately valid [@andreas61].
Equation (\[eq:ab\]) has an infinite number of solutions. It can be shown [@andreas:nim] that all equally good solutions of Eq. (\[eq:ab\]) can be obtained by the transformations $$\begin{aligned}
\tilde{\rho}(E-E_{\gamma}) &=& A \exp[\alpha(E-E_{\gamma})] \rho(E-E_{\gamma}) ,
\label{eq:array}
\\
\tilde{F}(E_{\gamma}) &=& B \exp(\alpha E_{\gamma}) F(E_{\gamma})
\label{eq:array2}\end{aligned}$$ of any particular solution ($\rho,F$). Consequently, neither the slope nor the absolute value of the two functions can be obtained through the iteration procedure, but the three variables $A$, $B$ and $\alpha$ of Eqs. (\[eq:array\]) and (\[eq:array2\]) have to be determined independently to give the best physical solution of the level density and $\gamma$-strength function.
The level density
-----------------
The parameters $A$ and $\alpha$ of Eq. (\[eq:array\]) can be determined by fitting the level density from the iteration procedure to the number of known discrete levels [@toi] at low excitation energy and to the level density estimated from neutron-resonance spacing data at high excitation energy [@ilj92]. This normalization procedure is shown for $^{167}$Er in the right panel of Fig. \[fig:normrho\]. The deduced level density is fitted to the discrete levels (between the arrows in the upper panel of Fig. \[fig:normrho\]) as far up in energy as we can assume that all levels are known. At high excitation energies the deduced level density is fitted (between the arrows in the lower panel of Fig. \[fig:normrho\]) to a Fermi-gas approximation of the level density (line) forced to pass through the level-density estimate at the neutron-separation energy (filled square) obtained from the neutron-resonance spacing data.
Unfortunately, $^{165}$Er is an unstable nucleus ($T_{1/2}=10.36$ h), making the neutron capture $^{165}$Er(n,$\gamma$)$^{166}$Er difficult to investigate. Thus, there is no neutron-resonance data available for $^{166}$Er. In this case the level density at the neutron-separation energy is estimated with the Fermi-gas expression $$\rho= \frac{\exp(2 \sqrt{aU})}{12 \sqrt{2}a^{1/4}U^{5/4} \sigma},$$ where the level density parameter $a=0.21A^{0.87}$, the shifted excitation energy $U=E-C_1-\Delta$, and the spin-cutoff parameter $\sigma$ is defined by $\sigma^2=0.0888\sqrt{aU}A^{2/3}$, while $A$ is the mass number, $\Delta$ the pairing-gap parameter and $C_1=-6.6A^{-0.32}$. All parameters are calculated according to Ref. [@egidy], which provides a level density in acceptable agreement with the experimental level density at the neutron-separation energy for $^{167}$Er. Also the nuclei $^{161,162}$Dy and $^{171,172}$Yb show conformity between the experimental and theoretical level density at the neutron-separation energy [@magne62]. The left panel of Fig. \[fig:normrho\] shows the normalization of $^{166}$Er, where the open square represents the theoretically calculated level density at the neutron-separation energy with an error of $80$% [@egidy].
The normalized level density of $^{166}$Er is compared to the level densities of $^{162}$Dy and $^{172}$Yb [@voinov] in the left panel of Fig. \[fig:rhosmnl\]. The three level-density functions coincide well at high excitation energies, indicating that the chosen value for the level density at the neutron-separation energy for $^{166}$Er is reasonable. For excitation energies below $1.9$ MeV the experimental level density of the erbium nucleus is seen to be lower than for the two others. In this energy region most levels in the three nuclei are believed to be known. From counting of known levels [@toi], it is found that between $1.1$ and $1.9$ MeV, the level density of $^{162}$Dy and $^{172}$Yb indeed exceed the level density of $^{166}$Er with the same degree as observed experimentally.
The right panel of Fig. \[fig:rhosmnl\] compares the level density of $^{167}$Er with the level densities of $^{161}$Dy and $^{171}$Yb [@voinov]. The two latter nuclei were populated through ($^3$He,$\alpha$) reactions. The $^{171}$Yb-level density is seen to be lower than the two others below $1.1$ MeV. According to [@toi], this agrees with reality below $900$ keV. Above this limit the level densities from counting of known levels start to decrease, making it evident that not all levels are known.
In Ref. [@andreas61] the level density of $^{162}$Dy extracted from two different reaction mechanisms are compared. The primary $\gamma$ spectra from the ($^3$He,$\alpha$) pick-up reaction and the inelastic ($^3$He,$^3$He’) reaction are found to provide level densities with very good agreement except below $\sim 1$ MeV of excitation energy. There, the level density obtained from the inelastic $^3$He scattering overestimates the number of levels by a factor of $\sim 3$, because this reaction populates more collective excitations with similar structure as the ground-state band. This causes a relatively large $\gamma$-transition rate from the direct populated states to the ground band, which in turn gives an overestimation of the level density there. The same is observed in the level density of $^{172}$Yb deduced from the inelastic reaction [@andreasdr]. This effect of overestimation can not be ruled out in the level density of $^{167}$Er derived from the ($^3$He,$^3$He’) reaction, however, the experimental data fit the known level density at low excitation energy very well, see Fig. \[fig:normrho\]. A difference between these ($^3$He,$^3$He’) experiments is the spin of the target nuclei. The $^{167}$Er target has a spin of $7/2$ making excitations to a large number of states probable, with a resulting fragmentation of collective strength and a lower probability for subsequent $\gamma$ transitions directly into the ground-state band than in the even-even nuclei. Thus, the accessible states in the $\gamma$ decay seem to approximates the level density quite well.
The $\gamma$-ray strength function
----------------------------------
Blatt and Weisskopf [@blatt] suggested the ratio of the partial radiative width $\Gamma_i(E_{\gamma})$ and the level spacing of the initial states $i$ with equal spin and parity $D_i$ to describe the $\gamma$ decay in the continuum. The corresponding definition of the $\gamma$-ray strength function is given by $f_{XL}=\Gamma_i(E_{\gamma})/(E_{\gamma} ^{2L+1}D_i)$, where $X$ denotes the electric or magnetic character, and $L$ defines the multipolarity of the $\gamma$ transition.
After the normalization of the level density, the parameter $\alpha$ of Eq. (\[eq:array2\]) is already fixed, and the slope, i.e. the exponential $\exp(\alpha E_{\gamma})$, is included in the $\gamma$-energy dependent functions shown for $^{166}$Er and $^{167}$Er in Fig. \[fig:f\]. Still, the normalization constant $B$ of Eq. (\[eq:array2\]) remains to be determined.
The $\gamma$-energy dependent factor $F(E_{\gamma})$ is proportional to the sum of $E_{\gamma}^{2L+1} f_{XL}(E_{\gamma})$ for all possibilities of $X$ and $L$. We assume that the $\gamma$ decay in the continuum of nuclei with low spin is mainly governed by electric and magnetic dipole radiation and that the accessible levels have equal numbers of positive and negative parity states. Thus, the observed $F$ can be expressed by a sum of the E1 and M1 $\gamma$-strength functions only: $$BF(E_{\gamma})=
[f_{\mathrm{E1}}(E_{\gamma})+f_{\mathrm{M1}}(E_{\gamma})] E_{\gamma}^3.
\label{eq:prim1}$$ The average total radiative width of neutron resonances $\langle\Gamma_\gamma\rangle$ [@kop] with excitation energy equal to the neutron-separation energy $S_n$, spin $I$, and parity $\pi$ $$\begin{aligned}
\langle\Gamma_\gamma\rangle=\frac{1}{\rho(S_n, I, \pi)} \sum_{XL}\sum_{I_f, \pi_f}&&\int_0^{S_n}{\mathrm{d}}E_{\gamma} E_{\gamma}^{2L+1} f_{XL}(E_{\gamma})
\nonumber\\
&&\rho(S_n-E_{\gamma}, I_f, \pi_f),
\label{eq:norm}\end{aligned}$$ can be written in terms of $F$ by means of Eq. (\[eq:prim1\]). With the experimental level density $\rho$ already normalized, the normalization constant $B$ can be deduced. Assuming $s$-neutron capture, $I,\pi$ take the values $I_t\pm \frac{1}{2},\pi_t$, where $I_t,\pi_t$ are the spin and parity of the target nucleus in the $(n,\gamma)$ reaction. The experimental value of the total radiative width is the average over the possible spins of the compound state ($I_t\pm \frac{1}{2}$). Since we expect the $\gamma$ decay to be governed by dipole transitions, the second sum is restricted to possible final states with spin $I_f$ and parity $\pi_f$ accessible by dipole radiation.
The methodical difficulties in the primary $\gamma$-ray extraction, prevents determination of the functions $F(E_{\gamma})$ and $\rho(E)$ in the interval $E_\gamma<1$ MeV and $E > S_n-1$ MeV, respectively. In addition, the data at the highest $\gamma$-energies, above $E_{\gamma} \sim S_n-1$ MeV, suffer from poor statistics. Therefore, extrapolations of $F$ and $\rho$ were necessary in order to calculate the integral of Eq. (\[eq:norm\]). For the level density a Fermi-gas extrapolation is used (see Fig. \[fig:normrho\]), and for the $\gamma$-energy dependent factor a pure exponential of the form $\exp(bE_{\gamma})$ is utilized (see Fig. \[fig:f\]). The contribution of the extrapolation to the total radiative width in Eq. (\[eq:norm\]) does not exceed $15$%, thus the errors due to a possibly poor extrapolation are expected to be of minor importance [@voinov].
Values for the neutron-resonance radiative width are given in [@tecdoc], and for $^{167}$Er we find $\langle\Gamma_\gamma\rangle= 92(8)$ meV. Since the radiative width of $^{166}$Er is unknown, its value is taken as the average of $\langle\Gamma_\gamma\rangle$ of the neighbouring isotopes. This is justified by the general slow variance in the $\langle\Gamma_\gamma\rangle$ values of neighbouring isotopes for other elements. From interpolation, we thus adopt the value $\langle\Gamma_\gamma\rangle= 90(20)$ meV for $^{166}$Er. The normalized $\gamma$-strength functions of $^{166}$Er and $^{167}$Er are compared to the strength functions of $^{162}$Dy and $^{172}$Yb, and $^{161}$Dy and $^{171}$Yb, respectively, in Fig. \[fig:sfsmnl\]. All strength functions are increasing smoothly with $\gamma$ energy except for a bump around $E_{\gamma}\sim3$ MeV. The $\gamma$-strength functions of neighbouring isotopes display a striking qualitative similarity. This can be expected for nuclei with approximately the same charge distribution. The location of the bump can however be seen to move towards higher $\gamma$ energies with increasing mass number.
Thermodynamic properties
========================
For a statistical description of hot nuclei, the micro-canonical thermodynamics is the proper theory. Within this frame, the system is isolated, giving a well defined energy. However, the canonical ensemble, permitting heat exchange, and the grand-canonical ensemble, which in addition allows particle exchange, are often used due to mathematical difficulties with detailed calculations within the micro-canonical ensemble. In this work, both the micro-canonical and the canonical ensemble will be utilized to discuss thermodynamic properties of the $^{166,167}$Er nuclei experimentally.
In particular we will focus on different ways, within the two ensembles, to obtain experimental values of the critical temperature for the pair-breaking process and the general quenching of pair correlations. The micro-canonical ensemble gives a detailed description of the breaking of one, two, three,... nucleon pairs as a function of excitation energy, while the canonical ensemble reveals the general average properties of this phase-like transition. The proper definition of thermally driven first- and second-order phase transitions in systems with few particles is a long standing problem, which will not be discussed in the present experimental work.
Micro-canonical ensemble
------------------------
The micro-canonical partition function is simply the multiplicity of nuclear states, which experimentally corresponds to the level density of accessible states. Thus, the experimental level density $\rho(E)$ is our starting point for the extraction of thermodynamic properties of nuclei.
The entropy is determined by $$S(E)=S_0 + \ln \rho(E),$$ where the Boltzmann constant for simplicity is set to unity ($k_B\equiv 1$), and the normalization constant $S_0$ can be adjusted to fulfill the condition of the third law of thermodynamics; $S\rightarrow 0$ when $T\rightarrow 0$, $T$ being the nuclear temperature $$T(E)=\left(\frac{\partial S}{\partial E}\right)^{-1}.
\label{eq:t}$$ The ground-state band of even-even nuclei is assumed to have $T=0$. Therefore, $S_0$ is determined so that the entropy of the ground-state band in $^{166}$Er is approximately zero.
Figures \[fig:s\] and \[fig:mu\] show the entropy and the temperature, respectively, deduced in the micro-canonical ensemble for $^{166}$Er and $^{167}$Er. The entropy curve plotted in a linear scale is essentially identical to the level-density curve in a logarithmic scale. The small bumps in the entropy curves of Fig. \[fig:s\] are enhanced through the differentiation performed in Eq. (\[eq:t\]) to obtain the temperatures of Fig. \[fig:mu\].
The specific heat can be determined from differentiating the temperature $$C_V(E) =\left(\frac{\partial T}{\partial E}\right)^{-1}$$ and is displayed in the lower panel of Fig. \[fig:mu\] for $^{166}$Er and $^{167}$Er. The double differentiation of the entropy has introduced strong fluctuations in the specific heat. Still it is distinct that the oscillations in the temperature give rise to successive positive and negative heat capacity. The spectacular feature of negative heat capacity is a direct consequence of the decrease in the micro-canonical temperature, and has recently been observed experimentally in the critical region of nuclear fragmentation in Au quasi-projectile sources formed in Au+Au collisions [@agost].
An anomalous decrease in the temperature means an unusual increase in the entropy, i.e. a more than normal opening of new domains of the phase space, e.g. opening of new degrees of freedom as more particle pairs are broken. The successive bumps of Figs. \[fig:s\] and \[fig:mu\] indicate that the transition from a paired to an unpaired phase is a gradual process of breaking up more and more nucleon pairs.
When we interpret the area of negative heat capacity, corresponding to the region where the temperature has a negative slope, as the location of the break up of a nucleon pair, a signal of the break up of the first pair can be seen around $E\sim 1.5$ MeV in $^{167}$Er in Fig. \[fig:mu\]. This value is close to twice the pairing-gap parameters $2\Delta_n=1.7$ MeV and $2\Delta_p=1.5$ MeV, which are the expected cost of breaking of a neutron and proton pair, respectively. The pairing-gap parameters are calculated from the empirical masses of a sequence of isotopes or isotones [@bm1] of $^{167}$Er. The first negative slope in the temperature as a function of excitation energy in $^{161}$Dy, $^{162}$Dy, $^{171}$Yb, and $^{172}$Yb [@andreasdubna] also coincide roughly with $2\Delta$. A similar argument is however less successful for $^{166}$Er. Deviations from $2\Delta$ in the localization of the break up of the first pair of particles can be due to the influence of several structural effects in the nuclei, such as e.g. the Fermi-level position in the Nilsson single-particle scheme, variation in the density of single-particle orbitals, and two quasi-particle couplings to collective degrees of freedom.
It has been shown [@gross3] that phase transitions of finite-size systems can be observed in the micro-canonical ensemble without invoking the thermodynamic limit. According to [@gross], a phase transition of first order can be recognized from the caloric curve in the region of co-existing phases, i.e. in the excitation energy region where the phase transition takes place. There, both the phase the system is leaving and the phase the system is attending are existing simultaneously. For this co-existency the system has to “pay” an amount of entropy which is eventually returned as the whole system is converted into the new phase [@gross]. The non-oriented area between the inverse temperature $T(E)^{-1}$ and the line $T_c^{-1}$ is twice this entropy. The critical temperature has to satisfy the condition that the oriented area between $T(E)^{-1}$ and $T_c^{-1}$ equals zero.
The process studied in the present work, namely the breaking of nucleon pairs and the general quenching of pair correlations, deviates from the process of multifragmentation [@gross3; @gross]. In the latter process, the geometrical surfaces of the fragments play a crucial role. In the depairing process, the long-range two-nucleon force has correlation lengths longer than the diameter of the nucleus. Thus, the phase-like situation of having a pair coupled or not, reveals no geometrical surface of interaction.
In order to qualitatively analyze our experimental findings along the procedure of Ref. [@gross], the excitation energy region around the outstanding bump in the temperature of $^{167}$Er is expanded, and the entropy, inverse temperature and heat capacity of $^{167}$Er are plotted in Fig. \[fig:gross\]. The qualitative similarity with the picture of Ref. [@gross] suggests that the general process of breaking a nucleon pair around $E\sim 2\Delta$ appears like a phase transition of first order.
Accepting this theoretical result, however, the similar experimental signals of breaking of further nucleon pairs indicate that the process of quenching of pairing correlations is a series of first order phase transitions. This interpretation might seem physically unattractive, and therefore we are anticipating further theoretical clarifications.
The convex shape of the entropy curve of Fig. \[fig:gross\] indicates a transition from the one quasi-particle regime to the three quasi-particle regime. The transition takes place between the two tangent points of the entropy and its enveloping curve (dashed line in the entropy plot). The transition region is marked with two dotted vertical lines. The critical temperature for breaking of the first nucleon pair in $^{167}$Er is derived to be $T_c=0.51(4)$ MeV (horizontal dashed line in the caloric curve of Fig. \[fig:gross\]).
The inter-phase entropy[^1] could in principle be expected to be equal to the envelope entropy, that is, an interpolation of the entropies before and after the pair-breaking process. But compared to this expectation, the entropy is seen to be lowered by $\sim 0.6$ units, corresponding to the sum of the primary loss and the final gain in entropy through the transition. It is unclear how this entropy-mixing contribution shall be interpreted in a small nuclear system.
The first law of thermodynamics states that supplied heat equals the change in internal energy plus the work performed by the system: $dQ=dE+dW$. Since the nucleus is an isolated system the work $dW$ is zero, and the increase in energy equals the supplied heat through the phase transition. In macroscopic systems, where the temperature is constant through a phase transition of first order, this increase in energy is the latent heat. We interpret the transition-region energy of approximately $1.8$ MeV (see Fig. \[fig:gross\]) as the latent heat, even though the temperature is not constant. We note that this value is also very close to the expected cost $2\Delta$, of breaking up a particle pair. The dashed lines in the specific heat are purely meant as guiding lines towards the theoretical prediction of poles between the successive positive and negative branches.
The derivative of the entropy-envelope curve in Fig. \[fig:gross\] equals the critical temperature. The somewhat stronger concave shape of the micro-canonical entropy of $^{167}$Er compared to $^{166}$Er (see Fig. \[fig:s\]) suggests that increasing temperature is necessary to break further nucleon pairs of $^{167}$Er, whereas the critical temperature for breaking of particle pairs is rather constant in $^{166}$Er. After the break up of the first particle pair in $^{167}$Er, the slope of the entropy becomes however more constant. A linear fit to the entropy above $E\sim 2$ MeV gives a critical temperature of $0.53(10)$ MeV for $^{166}$Er and $0.55(4)$ MeV for $^{167}$Er, indicating that the critical temperature for breaking of nucleon pairs in $^{167}$Er has a slight increase. This can be understood from the blocking effect of the unpaired nucleon, increasing the distance to the Fermi surface for low-lying orbitals with coupled pairs.
The critical temperature for breaking of the first particle pair in $^{167}$Er could be nicely determined from equal areas in the caloric plot. Unfortunately, the thermodynamic quantities of small systems are strongly fluctuating. Thus, the critical temperature for each back bend of the temperature, e.i. breaking of a particle pair, can generally be difficult to obtain by this method. The slope of the entropy-envelope curve can however always serve as a measure.
Canonical ensemble
------------------
The transformation from the micro-canonical to the canonical ensemble is performed by the canonical partition function $$Z(T)=\sum_{E=0}^{\infty}\omega (E)e^{-E/T}.
\label{eq:z}$$ The partition function is a Laplace transform of the multiplicity of states $\omega (E)=\Delta E\rho (E)$, where $\rho (E)$ is the level density of accessible states in the nuclear reaction at the discrete energy $E$ given in energy bins $\Delta E$.
The thermal average of the excitation energy in the canonical ensemble is $$\langle E(T)\rangle =Z^{-1}\sum_{E=0}^{\infty}E \omega (E)e^{-E/T}.
\label{eq:e}$$ By the Laplace transform in Eq. (\[eq:z\]) much of the information contained in the micro-canonical level density becomes smeared out. This smoothing effect can be quantified by the standard deviation for the thermal average of the energy $$\sigma_{\scriptscriptstyle E}= \sqrt{\langle E^2\rangle -\langle E\rangle ^2},$$ which e.g. gives $\sigma_E= 3$ MeV at $E=7$ MeV. Thus, fine structure in the thermodynamic observables in the micro-canonical ensemble will not be visible in the canonical ensemble. The lines in the upper panel of Fig. \[fig:mu\] display the smooth variance of the canonical temperature as a function of the thermal average of the excitation energy.
The mathematical justification of Eqs. (\[eq:z\]) and (\[eq:e\]) is that the summation is performed to infinity. However, for the typical temperatures studied here, $T<1$ MeV, the level density should be known up to $40$ MeV, which was found [@andreas] to be a sufficiently high upper limit for the summation. The experimental level density is only covering the excitation energy region from zero to about $S_n-1$ MeV. In the region of and above the neutron-separation energy $S_n$, the Fermi-gas model is believed to describe the nuclear properties. Therefore, the experimental level density is extrapolated to higher excitation energies with the Fermi-gas approximation of Ref. [@egidy], see the solid lines of Fig. \[fig:normrho\].
The heat capacity in the canonical ensemble is the derivative of the thermal average of the excitation energy $$C_V(T)=\frac{\partial <E>}{\partial T},$$ and it is shown for $^{166,167}$Er as a function of the thermal average of the excitation energy in Fig. \[fig:mu\] and as a function of temperature in Fig. \[fig:crit\]. The averaging done by the partition function can be seen to provide a smooth energy and temperature dependence of the heat capacity. The heat capacity of both nuclei shows an $S$ shape as a function of temperature. This feature is interpreted [@andreas] as a fingerprint of a phase transition from a state with strong pairing interaction to a state where the pairing correlations are quenched. Because of the smoothing performed by the canonical-partition function, discrete transitions between the different quasi-particle regimes, as observed within the micro-canonical ensemble, are hidden, and only the phase transition related to the quenching of the pair correlations as a whole can be seen.
The shape of the heat-capacity curve is related to the level-density. In general a constant-temperature level density gives a pronounced $S$ shape [@andreas], while a pure Fermi-gas level density provides a linear heat capacity $C_V = 2aT-3/2$; see the dashed lines of Fig. \[fig:crit\]. The more prominent $S$ shape in $^{166}$Er is due to a quicker rise in the heat capacity. This is the same phenomenon as observed in the micro-canonical ensemble, where the concave shape of the $^{167}$Er entropy was assigned to the blocking effect of the odd neutron. Shell model Monte Carlo calculations [@alha] on various Fe isotopes have shown that the pairing-phase transition is strongly correlated with the suppression of neutron pairs with increasing temperature. It is also observed that the reduction of neutron pairs is much stronger in the even- than in the odd-mass isotopes, giving rise to a more pronounced $S$ shape in the even nuclei. The same difference in the heat capacity is also observed experimentally between $^{161}$Dy and $^{162}$Dy, and $^{171}$Yb and $^{172}$Yb [@andreas].
In order to extract a critical temperature for the quenching of pair correlations from the canonical data, two different approaches has been tested. First, the local maximum of the heat capacity relative to the Fermi-gas approximation of Fig. \[fig:crit\] have been determined to be $0.47(10)$ MeV for $^{166}$Er and $0.59(10)$ MeV for $^{167}$Er, see the arrows of Fig. \[fig:crit\]. This method relies very much on the Fermi-gas extrapolation of the level density. The second approach depends on the assumption that the nuclear level density can be approximated by a constant-temperature expression at low excitation energies. Within the canonical ensemble this assumption provides the relations $$T^{-1}=\langle E(T)\rangle^{-1}+\tau^{-1},
\label{eq:invT}$$ $$C_V(T)=(1-T/\tau)^{-2},
\label{eq:cvt}$$ where $\tau$ is the constant-temperature parameter. By fitting $T^{-1}$ as a function of $\langle E(T)\rangle^{-1}$ to a straight line of slope $1$, $\tau$ can be determined. To minimize the dependency of the extracted critical temperature on the extrapolated level density, the fit is performed between $\langle E(T)\rangle\sim 0.5-2$ MeV. According to Eq. (\[eq:e\]) this corresponds to energies in the level density curve up to $\sim 6$ MeV. Then $\tau$ is identified with the critical temperature, since $C_V(T)$ has a pole at $T=\tau$. The critical temperature is determined to $T_c=0.44(10)$ MeV for $^{166}$Er and $T_c=0.52(9)$ MeV for $^{167}$Er. The dashed-dotted lines of Fig. \[fig:crit\] describe Eq. (\[eq:cvt\]) with asymptotes at $\tau=T_c$.
The different values derived for $T_c$ are quoted in Table \[tab:crit\]. The micro-canonical caloric-curve method (Fig. \[fig:gross\]) is only supposed to provide the critical temperature for breaking of the first particle pair, while the others should give the critical temperature for the total quenching of the pairing correlations. The canonical constant-temperature approach relies on experimental data, but it can be questionable whether Eqs. (\[eq:invT\]) and (\[eq:cvt\]) describes the nucleus well at low excitation energies. The method of deducing $T_c$ from the local maximum of the canonical heat capacity relative to the Fermi-gas approximation depends highly on the extrapolation of the level density. The most straight forward method is to obtain the temperature directly from the slope of the micro-canonical entropy. There seem to be a tendency for the critical temperature in $^{166}$Er to be slightly lower than in $^{167}$Er. All results are however equal within the approximated uncertainties, with a mean value of $0.5(1)$ MeV for both nuclei.
The results for $^{166}$Er agrees with recent finite-temperature Hartree Fock Bogoliubov calculations for $^{164}$Er [@egido], in which the critical-temperature estimate extracted is $0.7$ MeV. This value is somewhat higher than the present derived result for $^{166}$Er. This diversity is probably due to application of different interpretations of where exactly the phase transition takes place. Since the shapes of the heat-capacity curves of $^{164}$Er and $^{166}$Er are practically identical, we assume that the calculated critical temperature for quenching of pair correlations in $^{164}$Er coincides with the experimental findings for $^{166}$Er.
Relativistic Hartree-BCS calculations [@agrawal] also find the proton and neutron pairing gaps to vanish around $0.4-0.5$ MeV of temperature in $^{166}$Er and $^{170}$Er. The heat capacity shows two peaks corresponding to these events, characteristic of second order phase transitions from superfluid to normal phase [@agrawal]. Thus theoretically, the pair quenching is calculated to be a phase transition of second order. Provided that negative heat capacity is a sufficient signal, the observed breaking of nucleon pairs in the micro-canonical ensemble appears like first-order phase transitions. The process of breaking one nucleon pair does however not lead to a total quenching of the pair correlations, and may not be described by the same physics as the pairing-phase transition. It is not clear how a series of phase transitions should be interpreted physically.
Electromagnetic properties
==========================
The $\gamma$-strength function is governed by different multipolarities of electric and magnetic character. We assume however, that the $\gamma$ decay in the continuum is dominated by dipole transitions and will try to model the $\gamma$-strength function theoretically. It is commonly adopted that the E1 strength is determined by the giant electric dipole resonance (GEDR) at high $\gamma$ energies. More doubtful is the assumption that the tail of the GEDR governs the E1 strength at low $\gamma$ energies when the tail approaches zero [@voinov]. A model [@kad] taking into account the energy and temperature dependence of the GEDR width is often utilized to describe the experimental data [@kop]. The M1 strength function also plays an important role governing the $\gamma$ emission for low $\gamma$ energies. Experiments indicate the existence of an M1 giant resonance due to spin-flip excitations in the nucleus [@kop2].
In Fig. \[fig:sffit\], the experimental $\gamma$-strength function is fitted by a theoretical strength function taking into account both the giant electric dipole resonance and the spin-flip resonance. In addition, a weaker resonance at lower energies is needed in order to fit the experimental data. Because of the much lower strength compared to the GEDR, the resonance is denoted the pygmy resonance. To account for the E1 radiation, the model [@kad] $$f_{\mathrm{E1}}(E_\gamma)=\frac{1}{3\pi^2\hbar^2c^2} \frac{0.7\sigma_{\mathrm{E1}}\Gamma_{\mathrm{E1}}^2(E_\gamma^2+4\pi^2T^2)} {E_{\mathrm{E1}}(E_\gamma^2-E_{\mathrm{E1}}^2)^2}$$ is adopted. The values for the giant electric dipole resonance parameters $\sigma_{\rm E1}$, $\Gamma_{\rm E1}$ and $E_{\rm E1}$ are taken from [@AogN]. The temperature parameter $T$ is utilized as a constant fit parameter. The M1 radiation is described by $$f_{\mathrm{M1}}(E_\gamma)=\frac{1}{3\pi^2\hbar^2c^2} \frac{\sigma_{\mathrm{M1}}E_\gamma\Gamma_{\mathrm{M1}}^2} {(E_\gamma^2-E_{\mathrm{M1}}^2)^2+E_\gamma^2\Gamma_{\mathrm{M1}}^2},
\label{eq:M1}$$ where $\sigma_{\rm M1}$, $\Gamma_{\rm M1}$ and $E_{\rm M1}$ are the giant magnetic dipole resonance parameters, which are taken from [@tecdoc]. The pygmy resonance is here described with a similar Lorentzian function $f_{\rm py}$ as Eq. (\[eq:M1\]), where the pygmy-resonance strength $\sigma_{\rm py}$, width $\Gamma_{\rm py}$ and centroid $E_{\rm py}$ have been fitted in order to adjust the total theoretical strength function $$f=K(f_{{\mathrm{E}}1} + f_{{\mathrm{M}}1})+f_{\mathrm{py}}$$ to the experimental data. The resulting theoretical $\gamma$-strength functions are shown as solid lines in Fig. \[fig:sffit\]. The obtained values of the fitting parameters for the pygmy resonance, the normalization constant $K$ and the temperature $T$ are given in Table \[tab:fit\].
The low-energetic resonance has its centroid at $E_{\gamma}=2.98(8)$ MeV in $^{166}$Er and at $E_{\gamma}=3.24(7)$ MeV in $^{167}$Er. In $(n,\gamma)$ reactions a pygmy resonance is observed [@iga] in the energy region $2.5-3.5$ MeV for nuclei of mass 160-176. Both the centroid and the width of the resonance is found [@iga] to increase gradually with neutron number. In $^{166}$Er the centroid, width and strength of the pygmy resonance are somewhat lower than in $^{167}$Er, fitting very well into the systematics of [@iga]. The enhanced $\gamma$ strength around $3$ MeV in $^{161,162}$Dy and $^{171,172}$Yb [@voinov] also fit into this systematics, indicating that the $\sim 3$ MeV bump after ($^3$He,$\alpha$) and ($^3$He,$^3$He’) reactions probably is of the same origin as the pygmy resonance reported in [@iga].
The pygmy resonance has been explained by the enhancement of the E1 $\gamma$-strength function [@iga]. The possibility that it on the contrary is of M1 character can not be excluded. At an excitation energy around $3$ MeV, there is a concentration of orbital M1 strength in the weakly collective scissors mode [@rich]. The scissors mode was first observed in electron-scattering experiments [@riht], and is confirmed by the $(\gamma,\gamma ')$ reaction [@pietralla]. A systematic survey of centroids of observed M1 strength distributions in even-even nuclei in the $A=134-196$ mass region is given in Ref. [@pietralla], where the scissors mode of $^{166}$Er is found to be centered at $2.961(26)$ MeV, in perfect agreement with the pygmy resonance found here.
Figure \[fig:sffit\] also shows the predicted individual contributions from the giant electric dipole resonance $f_{E1}$, the giant magnetic dipole resonance $f_{M1}$ and the pygmy resonance $f_{\rm py}$ to the $\gamma$-strength function. The strength function is generally dominated by E1 radiation. The M1-strength function $f_{M1}$ is always $\sim 20$% lower than the E1-strength function $f_{E1}$. Moreover, the pygmy resonance contributes considerably around $3$ MeV.
The normalization constant $K$ is close to 1 for $^{167}$Er (Table \[tab:fit\]), showing that the adopted theoretical model reproduces the absolute values of the $\gamma$-strength functions very well. The deviation from $1$ can be explained by uncertain values of the GEDR parameters for the investigated nuclei and uncertainties in the experimental normalization of the $\gamma$-strength functions. Also the $K$ value for $^{166}$Er is reasonable, provided that the adopted value for the neutron-resonance radiative width is reliable.
Conclusion
==========
Levels in $^{166}$Er and $^{167}$Er in the excitation region up to the neutron separation energy were populated with the ($^3$He,$\alpha$$\gamma$) and ($^3$He,$^3$He’$\gamma$) reaction, respectively. The level density and $\gamma$-ray strength function of $^{166}$Er and $^{167}$Er are determined from their corresponding primary $\gamma$-ray spectra.
Thermodynamic observables are deduced from the level density and display signatures of phase-like transitions within the micro-canonical and the canonical ensemble, interpreted as the transition from a strongly pair-correlated phase to an uncorrelated phase. Micro-canonical thermodynamics give the possibility of investigating the successive breaking of nucleon pairs in detail, information which is hidden in the canonical approach. The canonical ensemble on the other hand, reveals the average properties of the pairing transition. In addition, the canonical ensemble provides an excellent opportunity to study the different mechanisms governing the thermodynamic properties of odd and even systems. The increase in the heat capacity with temperature is much steeper in $^{166}$Er than in $^{167}$Er, probably due to the tendency of the odd neutron to block higher quasiparticle excitations. Various estimates of the critical temperature for the pairing-phase transition is performed, giving the result $T_c\sim 0.5$ MeV.
The experimental $\gamma$-strength function is fitted by a theoretical strength function, assuming that the $\gamma$ decay in the continuum is governed by dipole transitions. The contribution of electric and magnetic dipole radiation to the $\gamma$-strength function is recognized. A bump is observed in the $\gamma$-strength function at $3.0$ and $3.2$ MeV in the $^{166}$Er and $^{167}$Er, respectively, and is probably of the same origin as the pygmy resonance found in the $(n,\gamma)$ reaction.
Financial support from the Norwegian Research Council (NFR) is acknowledged.
A.S. Iljinov, M.V. Mebel, N. Bianchi, E. De Sanctis, C. Guaraldo, V. Lucherini, V. Muccifora, E. Polli, A.R. Reolon, and P. Rossi, Nucl. Phys. **A543, 517 (1992), and references therein. R.B. Firestone, and V.S. Shirley, [*Table of Isotopes*]{}, $8^{\mathrm{th}}$ edition, Vol. II (New York: John Wiley & Sons, Inc., 1996). S.S. Dietrich and B.L. Berman, At. Data Nucl. Data Tables **38, 199 (1988). F. Bečvář, P. Cejnar, J. Honzátko, K. Konečný, I. Tomandl, and R.E. Chrien, Phys. Rev. C [**52**]{}, 1278 (1995). A. Schiller, L. Bergholt, M. Guttormsen, E. Melby, J. Rekstad, and S. Siem, Nucl. Instrum. Methods Phys. Res. A [**447**]{} 498 (2000). J. Kopecky and M. Uhl, Phys. Rev. C [**41**]{}, 1941 (1990). A. Voinov, M. Guttormsen, E. Melby, J. Rekstad, A. Schiller, and S. Siem, nucl-ex0009018. M. Igashira, H. Kitazawa, M. Shimizu, H. Komano, and N. Yamamuro, Nucl. Phys. **A457, 301 (1986). M. Guttormsen, A. Atac, G. L[ø]{}vh[ø]{}iden, S. Messelt, T. Rams[ø]{}y, J. Rekstad, T.F. Thorsteinsen, T.S. Tveter, and Z. Zelazny, Phys. Scr. **T32, 54 (1990). M. Guttormsen, T.S. Tveter, L. Bergholt, F. Ingebretsen, and J. Rekstad, Nucl. Instrum. Methods Phys. Res. A **374, 371 (1996). M. Guttormsen, T. Ramsøy, and J. Rekstad, Nucl. Instrum. Methods Phys. Res. A [**255**]{}, 518 (1987). L. Henden, L. Bergholt, M. Guttormsen, J. Rekstad, and T.S. Tveter, Nucl. Phys. **A589, 249 (1995). D.M. Brink, Ph.D. thesis, Oxford University (1955). P. Axel, Phys. Rev. [**126**]{}, 671 (1962). T.S. Tveter, L. Bergholt, M. Guttormsen, E. Melby, and J. Rekstad, Phys. Rev. Lett. **77, 2404 (1996). E. Melby, L. Bergholt, M. Guttormsen, M. Hjorth-Jensen, F. Ingebretsen, S. Messelt, J. Rekstad, A. Schiller, S. Siem, and S.W. [Ø]{}deg[å]{}rd, Phys. Rev. Lett. **83, 3150 (1999). A. Schiller, M. Guttormsen, E. Melby, J. Rekstad, and S. Siem, Phys. Rev. C **61, 044324 (2000). T. von Egidy, H.H. Schmidt, and A.N. Behkami, Nucl. Phys. [**A481**]{}, 189 (1988). M. Guttormsen, A. Bjerve, M. Hjorth-Jensen, E. Melby, J. Rekstad, A. Schiller, S. Siem, and A. Belić, Phys. Rev. C [**62**]{} 024306 (2000). A. Schiller, Doctorial Thesis, University of Oslo (2000). J.M. Blatt and V.F. Weisskopf, [*Theoretical Nuclear Physics*]{} (John Wiley & Sons, New York, 1952). Handbook for calculations of nuclear reaction data, IAEA, Vienna, IAEA-TECDOC-1024 (1998). M. D’Agostino [*et al.*]{}, Phys. Lett. B [**473**]{}, 219 (2000). A. Bohr and B.R. Mottelson, [*Nuclear Structure*]{}, Vol. I (Benjamin, New York, 1969) 169. A. Schiller, M. Guttormsen, M. Hjorth-Jensen, E. Melby, J. Rekstad, and S. Siem, nucl-ex/0007009. D.H.E. Gross and E.V. Votyakov, Eur. Phys. J. B [**15**]{}, 115 (2000). D.H.E. Gross, Phys. Rep. 279, 119 (1997), and references therein. A. Schiller, A. Bjerve, M. Guttormsen, M. Hjorth-Jensen, F. Ingebretsen, E. Melby, S. Messelt, J. Rekstad, S. Siem, and S.W. [Ø]{}deg[å]{}rd, nucl-ex/9909011. S. Liu and Y. Alhassid, nucl-th/0009006. J.L. Egido, L.M. Robledo, and V. Martin, Phys. Rev. Lett. [**85**]{}, 26 (2000). B.K. Agrawal, Tapas Sil, J.N. De, and S.K. Samaddar, Phys. Rev. C [**62**]{}, 044307 (2000). S.G. Kadmenski[ĭ]{}, V.P. Markushev, and V.I. Furman, Yad. Fiz. **37, 277 (1983) \[Sov. J. Nucl. Phys. **37, 165 (1983)\]. J. Kopecky and R.E. Chrien, Nucl. Phys. [**A468**]{}, 285 (1987). A. Richter, Nucl. Phys. **A507, 99c (1990). D. Bohle, A. Richter, W. Steffen, A.E.L. Dieperink, N. Lo Iudice, F. Palumbo, and O. Scholten, Phys. Lett. B **137, 27 (1984). N. Pietralla, P. von Brentano, R.-D. Herzberg, U. Kneissl, N. Lo Iudice, H. Maser, H.H. Pitz, and A. Zilges, Phys. Rev. C **58, 184 (1998), and references therein.****************************
-------------------------------- ------------- -------------
Extraction method $^{166}$Er $^{167}$Er
$T_c$ (MeV) $T_c$ (MeV)
micro-canonical caloric curve 0.51(4)
micro-canonical entropy 0.53(10) 0.55(4)
canonical heat capacity 0.47(10) 0.59(10)
canonical constant temperature 0.44(10) 0.51(9)
-------------------------------- ------------- -------------
: Different extracted values for the critical temperature.
\[tab:crit\]
------------ ------------------- ------------------------ ------------------------ --------- ---------
Nucleus $E_{\mathrm{py}}$ $\sigma_{\mathrm{py}}$ $\Gamma_{\mathrm{py}}$ $T$ K
(MeV) (mb) (MeV) (MeV)
$^{166}$Er 2.98(8) 0.30(4) 1.3(3) 0.31(5) 1.3(2)
$^{167}$Er 3.24(7) 0.43(4) 1.7(2) 0.36(2) 1.27(6)
------------ ------------------- ------------------------ ------------------------ --------- ---------
: Parameters obtained from the fitting of the $\gamma$-ray strength function.
\[tab:fit\]
![Gamma-ray spectra at $E=4$ and $6$ MeV of excitation energy for $^{166}$Er. Left panel: Raw $\gamma$-ray spectra, central panel: Unfolded $\gamma$-ray spectra, and right panel: Primary $\gamma$-ray spectra.[]{data-label="fig:adata"}](fig1.eps){width="15cm"}
![Same as Fig. 1 for $^{167}$Er.[]{data-label="fig:tdata"}](fig2.eps){width="17cm"}
![Normalization of the experimental level density (data points) of $^{166}$Er (left panel) and $^{167}$Er (right panel) between the arrows to known levels at low excitation energy (histograms) and to the Fermi-gas level density ($^{166}$Er) or to the level density calculated from neutron-resonance spacings ($^{167}$Er) at the neutron-separation energy (squares). The lines are the Fermi-gas approximations.[]{data-label="fig:normrho"}](fig3.eps){width="17cm"}
![ The normalized level density of $^{166}$Er (left) and $^{167}$Er (right) compared to $^{162}$Dy and $^{172}$Yb, and $^{161}$Dy and $^{171}$Yb, respectively. The Dy isotopes are multiplied by 0.1 and the Yb isotopes by 10 for better visualization. All nuclei are popolated through the ($^3$He,$\alpha$) reaction except for $^{167}$Er, which is populated through the ($^3$He,$^3$He’) reaction. []{data-label="fig:rhosmnl"}](fig4.eps){width="17cm"}
![The $\gamma$-energy dependent factor of $^{166,167}$Er. The lines are extrapolations needed to calculate the normalization integral of Eq. (\[eq:norm\]).[]{data-label="fig:f"}](fig5.eps){width="17cm"}
![The normalized $\gamma$-ray strength function of $^{166}$Er (left) compared to $^{162}$Dy and $^{172}$Yb, and $^{167}$Er (right) compared to $^{161}$Dy and $^{171}$Yb. The Dy isotopes are multiplied by 0.1 and the Yb isotopes by 10 for better visualization. []{data-label="fig:sfsmnl"}](fig6.eps){width="17cm"}
![The entropy as a function of excitation energy in $^{166}$Er and $^{167}$Er[]{data-label="fig:s"}](fig7.eps){width="17cm"}
![The temperature (upper panel) and heat capacity (lower panel) of $^{166}$Er (left) and $^{167}$Er (right) from the micro-canonical ensemble (data points) and the canonical ensemble (lines). []{data-label="fig:mu"}](fig8.eps){width="17cm"}
![Entropy (upper panel), inverse temperature (central panel) and specific heat (lower panel) as functions of excitation energy in $^{167}$Er. The vertical, dotted lines indicate the approximate region in which the phase-like transition takes place. See text.[]{data-label="fig:gross"}](fig9.eps){width="10cm"}
![The semi-experimental heat capacity of $^{166}$Er (left) and $^{167}$Er (right) as a function of temperature. The dashed lines are the Fermi-gas approximations $C_V=2aT-3/2$. The local maxima relative to the Fermi-gas expressions are marked with arrows, while the dashed-dotted lines describe the estimates of Eq. (\[eq:cvt\]), where $\tau$ is recognized as the critical temperature and marked with the vertival lines.[]{data-label="fig:crit"}](fig10.eps){width="17cm"}
![The experimental $\gamma$-ray strength function (data points) of $^{166}$Er (left) and $^{167}$Er (right). The solid line is the fit to the data by the theoretical model. The dashed lines are the respective contributions of the GEDR, the GMDR, and the pygmy resonance to the total theoretical strength function.[]{data-label="fig:sffit"}](fig11.eps){width="17cm"}
[^1]: As stated above, these phases are not physically separated.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The conformally flat families of initial data typically used in numerical relativity to represent boosted black holes are not those of a boosted slice of the Schwarzschild spacetime. If such data are used for each black hole in a collision, the emitted radiation will be partially due to the “relaxation” of the individual holes to “boosted Schwarzschild” form. We attempt to compute this radiation by treating the geometry for a single boosted conformally flat hole as a perturbation of a Schwarzschild black hole, which requires the use of second order perturbation theory. In this we attempt to mimic a previous calculation we did for the conformally flat initial data for spinning holes. We find that the boosted black hole case presents additional subtleties, and although one can evolve perturbatively and compute radiated energies, it is much less clear than in the spinning case how useful for the study of collisions are the radiation estimates for the “spurious energy” in each hole. In addition to this we draw some lessons on which frame of reference appears as more favorable for computing black hole collisions in the close limit approximation.'
address:
- |
1. Facultad de Matemática, Astronomía y Física, Universidad Nacional de Córdoba,\
Ciudad Universitaria, 5000 Córdoba, Argentina.
- ' 2. Natural Science Division, Southampton College of Long Island University, Southampton NY 11968'
- ' 3. Department of Physics and Astronomy, 202 Nicholson Hall, Louisiana State University, Baton Rouge LA 70803-4001'
author:
- 'Reinaldo J. Gleiser$^1$, Gaurav Khanna $^2$, Jorge Pullin$^3$'
date: May 16th 2002
title: Perturbative evolution of conformally flat initial data for a single boosted black hole
---
psfig
Introduction {#sec:intro}
============
There is significant interest in obtaining waveforms for the gravitational radiation produced in the collision of black holes. Progress is being made on this problem both using supercomputers [@grandlychallenged] and perturbative calculations [@PuKy]. One of the open issues is what families of initial data are appropriate to represent the collision of two black holes, especially when the latter are not far away from each other. The state of the art of numerical simulations suggests that for some time we may not be able to start simulations with the black holes at a sufficiently large separation, such that one can assume a simple linear superposition will work. This leaves open the issue of how much “spurious radiation” is one introducing in the various proposals for superpositions in the non-linear regime. Bowen and York [@BoYo] (and with a different set of boundary conditions more recently Brandt and Brügmann [@BrBr]) studied the problem of giving initial data for boosted and spinning holes in such a way that a superposition is possible. They assume the spatial metric is conformally flat and as a consequence one can superpose the extrinsic curvatures for the two holes and still solve the momentum constraint. One then proceeds to find a conformally flat spatial metric for the superposed holes by solving a nonlinear elliptic partial differential equation. The procedure achieves superposition at the price of assuming conformal flatness of the three metrics, which is not generically possible, and more importantly, is not possible in situations of interest. For instance a single spinning black hole, described by the Kerr solution, is not known to admit conformally flat spatial slices [@BoYo; @GaPr].
A similar situation develops for the case of a single boosted black hole. The initial data constructed by Bowen and York or Brandt and Brügmann do not correspond to those one would find on a boosted slice of the Schwarzschild spacetime. In this paper we will refer to these families as “conformal boosted black hole” Therefore if one evolves these families of data one should find that the black hole “settles down” to a Schwarzschild form through the emission of gravitational radiation. The original purpose of this paper was to study the emitted radiation by treating the conformal boosted hole as a perturbation of a Schwarzschild black hole. In a previous paper we had carried out a similar discussion for the case of a single spinning conformally flat hole [@GlNiPrPuboyo]. We will see that the boosted case is much more subtle than anticipated. We will be able to evolve the spacetime, but questions will remain about the usefulness of the results obtained, at least for the original purpose of gaining insight into the spurious radiation content of data for black hole collisions for interesting ranges of parameter values.
The organization of this paper is as follows. In the next section we will discuss perturbations of a boosted black hole. We recall that the $\ell=1$ even modes are pure gauge and therefore all the physics of interest takes place in the $\ell=2$ modes, which can be treated easily up to second order in perturbation theory. In section III we will discuss the perturbative evolution and the amount of radiation produced. We end with a discussion of the results and their implication for the choice of frame of reference one uses in perturbative evolutions of black hole collisions with linear momentum.
Conformal boosted black hole as a perturbation of Schwarzschild
===============================================================
Initial data in the conformal approach
--------------------------------------
The families of initial data that we will consider in this paper are obtained via the “conformal approach” to the initial value problem in general relativity. In it, one assumes the metric to be conformally flat $g_{ab} =\phi^4 \delta_{ab}$ and defines the conformal extrinsic curvature $\widehat{K}_{ab} = \phi^{2} K_{ab}$. In terms of these variables the initial value constraint equations (assuming maximal slicing ${\rm Tr} K =0$) read, $$\begin{aligned}
\nabla_a \widehat{K}^{ab} &=& 0\\
\nabla^2 \phi &=& -{1\over 8} {\widehat{K}^{ab} \widehat{K}_{ab}
\over \phi^{7}}
\label{Hamil}\end{aligned}$$ where all the derivatives are with respect to flat space.
One can construct [@BoYo] solutions to the first set of equations (momentum constraint) for a single black hole centered at $R=0$, with linear momentum $P_a$, $$\label{onehole}
\widehat{K}_{ab} = {3 \over 2 R^2} \left[ 2 P_{(a} n_{b)} -(\delta_{ab}
-n_a n_b)P^c n_c\right]\ . \label{boyok}$$ where $R$ is a spherical radial coordinate and $n_a$ a radial unit vector, and both are defined in the fiducial flat space that one obtains setting the conformal factor to unity.
Without loss of generality we may assume that $P_a$ points along the positive z-axis, and has magnitude $P$. If we write $\widehat{K}_{ab}$ in spherical coordinates, the only non vanishing components are $$\begin{aligned}
\label{Ksphe}
\widehat{K}_{RR} &=& {3 P \over R^2} \cos(\theta) \nonumber \\
\widehat{K}_{R\theta} &=& -{3 P \over 2 R} \sin(\theta) =
\widehat{K}_{\theta R}\nonumber \\
\widehat{K}_{\theta\theta} &=& -{3 P \over 2} \cos(\theta) \nonumber \\
\widehat{K}_{\phi \phi} &=& \sin(\theta)^2 \widehat{K}_{\theta\theta}
\label{extriseca}\end{aligned}$$
If we write the extrinsic curvature in terms of tensor spherical harmonics [@ReWh], we see that it consists of a pure $\ell=1$ even term. This is reasonable, since the presence of momentum in an initial slice configuration is determined by the presence of a “dipole” term asymptotically, that makes the ADM integral $$P_i = {1 \over 8 \pi } \int_\infty K_{ij} d^2 S_j$$ nonvanishing.
From (\[extriseca\]), we have, $$\label{Kcuadrado}
\widehat{K}_{ab}\widehat{K}^{ab} = {9 P^2 \over 2 R^4} \left( 1 + 2
\cos^2
\theta \right)$$
We now need to solve (\[Hamil\]), and this involves imposing boundary conditions on $\phi$. One possibility is that given by Bowen and York [@BoYo][^1]. In this case one chooses a certain constant $a$, solves (\[Hamil\]) for $R
\geq a$, and imposes $${\partial \phi \over \partial R} +{1 \over 2R} \phi =0
\;\; \mbox{for} \;\; R= a, \label{robin}$$ and $$\phi > 0
\;\; , \;\; \lim_{R \rightarrow \infty} \phi = 1. \label{bounds}$$
Equation (\[robin\]) implies that the inner boundary is an extremal surface, as was shown in reference [@BoYo] and leads to a well posed elliptic problem when used with equations (\[Hamil\],\[bounds\]). One can also use the boundary condition (\[robin\]) to generate initial data that obeys an isometry condition, provided one chooses an extrinsic curvature —different from the one we are choosing here— that satisfies the isometry condition. This was explored in great detail by Cook [@Cook] in his numerical work constructing solutions for the conformal factor for multiple black holes. The symmetrization procedure yields non-trivial results even for a single black hole as we are considering here [@CoYo]. We have chosen, for simplicity, not to symmetrize the extrinsic curvature. Experience has shown that symmetrization does not change significantly the amount of radiated energy in head-on black hole collisions [@GlNiPrPuboost2]. Therefore our choice should not crucially influence the central conclusions we are attempting to obtain about the conformally flat black hole solutions.
Even for the simple form of (\[Kcuadrado\]), in general we cannot solve (\[Hamil\]) exactly, and one has to resort to numerical methods. In the present analysis, however, we will be interested in solutions for “small” $P$. Since for $P=0$ the solution is $$\phi^{(0)} = 1 + {a \over R}$$ we may solve (\[Hamil\]) iteratively replacing $$\phi = \phi^{(0)}(R) + P^2 \phi^{(2)}(R,\theta) +P^4
\phi^{(4)}(R,\theta)
+ \cdots$$ in (\[Hamil\]), and imposing $${\partial \phi^{(i)} \over \partial R} +{1 \over 2R} \phi^{(i)} =0
\;\; \mbox{for} \;\; R= a, \;\; \;\;\; \lim_{R \rightarrow \infty}
\phi^{(i)} = 0. \;\; \mbox{for} \;\; i\neq 0,$$
The form of (\[Kcuadrado\]) further suggests that we expand $\phi^{(i)}$ in Legendre polynomials $P_{\ell}(\cos\theta)$, so that $\phi^{(2)}$ may be written as $$\label{eqphi}
\phi = \phi^{(0)}(R) + P^2 \left[\phi^{(2)}_0(R) P_0(\cos \theta)+
\phi^{(2)}_2(R) P_2(\cos \theta)\right] + {\cal{O}}(P^4).$$
Solving for the coefficients taking into account the boundary condition (\[robin\]) we get, $$\begin{aligned}
\phi^{(2)}_0(R) & = & {(R+a)^5-R^5 \over 32 a^2(R+a)^5} +{13 \over 512
a R} \nonumber \\
\phi^{(2)}_2(R) & = & { 75 R^6+291 a R^5 -650 a^2 R^4 -3400 a^3 R^3
-4800 a^4 R^2 -2925 a^5 R -669 a^6 \over 400 R^3 (R+a)^5} \nonumber \\
& & + {21 a\over 20 R^3} [\ln(2a)-\ln(R+a)] + {121 a \over 128 R^3}
\label{misnerbound}\end{aligned}$$ Notice that $\phi^{(2)}_2(R)$ falls off only as $R^{-2}$ for large $R$.
For completeness, we also present the solution one obtains if one chooses the “puncture” boundary condition considered by Brandt and Brügmann [@BrBr] recently, $$\begin{aligned}
\phi^{(2)}_0(R) &=&
{\frac {{M}^{4}+10\,{M}^{3}R+40\,{M}^{2}{R}^{2}+80
\,M{R}^{3}+80\,{R}^{4}}{8\,M\left (2\,R+M\right )^{
5}}} \label{brbrbrbr}\\
\phi^{(2)}_2(R) &=&{\frac {120\,{R}^{5}+768\,{R}^{4}M+1078\,{R}^{3}{M}
^{2}+658\,{R}^{2}{M}^{3}+189\,R{M}^{4}+21\,{M}^{5}}
{20\,{R}^{2}\left (2\,R+M\right )^{5}}}\\
&&-21{\frac {
\left (\ln (2\,R+M)-\ln (M)\right )M}{40\,
{R}^{3}}}\nonumber\end{aligned}$$ but we will only consider the solution (\[misnerbound\]) in the rest of this paper.
To obtain an initial data set appropriate for a perturbative evolution we proceed as follows. First, in an ADM type decomposition, we choose our shift functions $N_a=0$, so that we have,
$$\begin{aligned}
\label{ADMeqs}
ds^2 & = & g_{ab} dx^a dx^b - N^2 (dt)^2 \nonumber \\
{\partial g_{ab} \over \partial t} & = & -2NK_{ab}\end{aligned}$$
Next, we assume that on our initial slice, at $t=0$, the 3-metric $g_{ab}$ and extrinsic curvature $K_{ab}$ are given by the above conformal flatness construction. Namely, we have $g_{ab}=\phi^4 \delta_{ab}$ and $K_{ab}=\phi^{-2} \widehat{K}_{ab} $, with $\widehat{K}_{ab}$ given by (\[Ksphe\]), and $\phi$ given by (\[eqphi\]). (This, of course, ensures that our initial data satisfies the constraint equations). We next change from the conformal spherical radial coordinates $R$, to a “Schwarzschild” radial coordinates $r$, with $R=(\sqrt{r}+\sqrt{r-2M})^2/4$, where $M = 2 a$, and choose $N=\sqrt{1-2M/r}$. It can be checked that with these choices, for $P=0$ the extrinsic curvature vanishes and we recover the Schwarzschild metric in the usual Schwarzschild coordinates. This last expression for the metric in $(r,\theta,\phi)$ coordinates (or, rather, the initial data) has, therefore, the appropriate form for a perturbative treatment of the Regge - Wheeler type, as extended to second order in [@physrep].
The final form for the initial data that results from this construction has thefollowing multipolar components: to zeroth order in the momentum, we only have $\ell=0$ components; to first order in the linear momentum,only $\ell=1$ contributions; to order ${\cal O}(P^2)$, we have $\ell=0,2$ contributions. All contributions are even-parity and are analyzed in more detail in the following Sections.
Multipolar decomposition of the initial data : the $\ell=1$ contributions
-------------------------------------------------------------------------
The contribution to the three-metric of zeroth order in $P$ is just the Schwarzschild solution. The first apparently non-trivial contribution is given at ${\cal O}(P^1)$ and corresponds as we discussed in the previous subsection to an $\ell=1$ multipole. Let us analyze the $\ell=1$ perturbations of a spherically symmetric spacetime. In order to do this we use the traditional Regge–Wheeler [@ReWh] decomposition. One starts with a background metric written as, $$g^{(0)}_{\mu \nu} dx^{\mu} dx^{\nu} = -(1 -2M/r) \; dt^2 + (1 -2M
/r)^{-1} dr^2 + r^2 d\theta^2 + r^2 \sin^2 \theta d \varphi^2.$$
For axisymmetric perturbations, the general $\ell = 1$ even parity terms take the form [@ReWh] $$\begin{aligned}
h^{(1)}_{tt} & = & (1 -2 M/r ) H_0(t,r) \cos \theta \nonumber \\
h^{(1)}_{tr} & = & H_1(t,r) \cos \theta \nonumber \\
h^{(1)}_{rr} & = & (1 -2 M/r )^{-1} H_2(t,r) \cos \theta \nonumber \\
h^{(1)}_{t \theta} & = & - h_0(t,r) \sin \theta \nonumber \\
h^{(1)}_{r \theta} & = & - h_1(t,r) \sin \theta \nonumber \\
h^{(1)}_{\theta \theta} & = & r^2 K(t,r) \cos \theta \nonumber \\
h^{(1)}_{\phi \phi} & = & r^2 \sin^2 \theta K(t,r) \cos \theta \end{aligned}$$
The first order metric perturbation coefficients are not uniquely defined, but may be changed by “gauge transformations” of the form [@ReWh] $$\widetilde{h}^{(1)}_{\mu\nu} =
h^{(1)}_{\mu\nu} -g_{\mu \nu},_{\rho} \xi_{(1)}^{\rho} -
g_{\mu \rho} \xi_{(1)}^{\rho},_{\nu} - g_{\rho \nu}
\xi_{(1)}^{\rho},_{\mu}$$ where the gauge 4-vector $\xi_{(1)}^{\mu}$ is arbitrary, except for the requirement of axisymmetry.
In particular, the $\ell = 1$ even parity coefficients transform as $$\begin{aligned}
\label{gauge1}
\widetilde{H}_0(t,r) & = & H_0(t,r) +2 {\partial {\cal{M}}_0(t,r)
\over \partial t} + {2M \over r(r-2M)} {\cal{M}}_1(t,r) \nonumber \\
\widetilde{H}_1(t,r) & = & H_1(t,r) +{r-2M \over r}{\partial
{\cal{M}}_0(t,r) \over \partial r} - {r \over r-2M} {\partial
{\cal{M}}_1(t,r) \over \partial t} \nonumber \\
\widetilde{H}_2(t,r) & = & H_2(t,r) +{2M \over r(r-2M)}
{\cal{M}}_1(t,r) -2 {\partial {\cal{M}}_1(t,r) \over \partial
r}\nonumber \\
\widetilde{h}_0(t,r) & = & h_0(t,r) +{r-2M \over r} {\cal{M}}_0(t,r)
-r^2 {\partial {\cal{M}}_2(t,r) \over \partial t} \nonumber \\
\widetilde{h}_1(t,r) & = & h_1(t,r) -{r \over r-2M}{\cal{M}}_1(t,r)
-r^2 {\partial {\cal{M}}_2(t,r) \over \partial r}\nonumber \\
\widetilde{K}(t,r) & = & K(t,r) -{2 \over r} {\cal{M}}_1(t,r) +2
{\cal{M}}_2(t,r)\end{aligned}$$ where the functions ${\cal{M}}_i$ are arbitrary.
One can use this gauge freedom to go to a restricted gauge in which $h_0 = h_1 = K =0$. This gauge is not completely fixed. One still can perform gauge transformations of the form, $${\cal{M}}_0 = {r^3 \over r -2M} {\partial {\cal{M}}_2 \over \partial
t}
\;\;\;,\;\;\;
{\cal{M}}_1 = r {\cal{M}}_2 \;\;\;,\;\;\;
{\cal{M}}_2 = {f(t) \over r-2M}
\label{eqxi1}$$ where $f(t)$ is arbitrary.
An interesting result, is that in this gauge it is straightforward to find the [*general*]{} solution of the linearized Einstein equations for $\ell=1$ even parity perturbations. The result is, $$\begin{aligned}
H_1 & = & - {r \over (r-2M)^2} {d F_1 \over dt} \nonumber \\
H_0 & = & {1 \over 3(r-2M)^2} F_1 +{r^3 \over 3M(r-2M)^2} {d^2 F_1
\over dt^2} \nonumber \\
H_2 & = & {1 \over (r-2M)^2} F_1(t), \end{aligned}$$ where $F_1(t)$ is an arbitrary function. Remarkably, one can show that this solution is pure gauge. Choosing $f(t)=-F_1(t)/(6M)$ in (\[eqxi1\]) leads to vanishing gauge transformed quantities.
We therefore see that the $\ell=1$ perturbations are pure gauge. This result was first noticed by Zerilli [@zerilli]. This was to be expected in physical grounds since one could always imagine setting coordinates boosted in such a way that the black hole would not move. It has the further implication that we can compute the radiated energy by computing the $\ell=2$ perturbations in a gauge in which the $\ell=1$ perturbations vanish, and studying their evolution.
Multipolar decomposition of the initial data : the $\ell=2$ contributions
-------------------------------------------------------------------------
The relevant $\ell=2$ perturbations are of second order in our perturbation parameter, $P$. In principle, when one works out second order perturbations of a given metric, the evolution equations one gets have the general form of a linear operator (similar to the one that acts in first order giving rise to the Zerilli equation) acting on second order quantities, equal to a quadratic “source” term formed with the first order perturbations [@GlNiPrPucqg]. The source term is complicated and is delicate to handle numerically when evolving the perturbations. A place where this was explicitly done was for instance in the evolution of boosted black hole collisions to second order [@GlNiPrPuboost2]. The calculations are lengthy and complicated. Fortunately, in our case one can proceed in a different way. We have just shown that there is a gauge in which the first order perturbations (the $\ell=1$ ones) vanish. Therefore in that gauge one can write a second order Zerilli equation that is source-free. Moreover, the linear portion of the second order perturbative equation is exactly the same as the first order perturbative equation (the Zerilli equation), and this equation can be written in terms of quantities that are gauge invariant. We notice, however, that eliminating the first order $\ell=1$ terms through a first order gauge transformation introduces second-order changes in the metric that are not a pure second order gauge transformation, and must be taken into account. One must then be careful in making sure that the initial data for the $\ell=2$ perturbations used in the Zerilli equation corresponds precisely to that gauge. In other words, we need to carry out a first order gauge transformation on the initial data that provides a new initial data corresponding to a gauge where the $\ell=1$ perturbations vanish. Since all perturbations satisfy equations that are second order in time, this requires that the $\ell=1$ terms of the metric, and their first time derivative vanish on the initial slice. If we consider (\[gauge1\]), this requires that the gauge vector components ${\cal{M}}_i$ are such that both the left hand sides in (\[gauge1\]), and their first time derivatives, vanish when evaluated at $t=0$. This, in principle requires only the knowledge of the metric functions on the right hand side of (\[gauge1\]), and their first time derivative, at the fiducial time $t=0$. It turns out, however, that we also need second order time derivatives,
to implement the second order gauge transformation required to obtain the $\ell=2$ initial data. These second order time derivatives may be straighforwardly evaluated from the corresponding Einstein equations for the $\ell=1$ perturbations.
For the particular case in question, the $\ell=1$ initial data (and first time derivative) is determined by (\[Ksphe\]), and one can use this information to construct the space-time solution of the Einstein equations produced by the initial data as a Taylor expansion in $t$, up to the appropriate order. One gets, using the usual Regge-Wheeler notation [@ReWh] the following expansions for the $\ell=1$ components of the metric,
$$\begin{aligned}
H_2 &=& -{3\over 2} {(\sqrt{r}+\sqrt{r-2M})^2 \sqrt{r-2M}\over r^{7/2}} P t
+O(t^3)\\
K&=&{3\over 4} {(\sqrt{r}+\sqrt{r-2M})^2 \sqrt{r-2M}\over r^{7/2}} P t
+O(t^3)\\
h_1 &=&
-{3\over 4} {(\sqrt{r}+\sqrt{r-2M})^2 \over r^{2}} P t
+O(t^3)\end{aligned}$$
all the other components being $O(t^3)$.
The components of the gauge vector generating the first order gauge transformation that makes the initial $\ell=1$ data purely $O(t^3)$ are, $$\begin{aligned}
\xi_t &=& {\cal{M}}_0 \cos\theta\\
\xi_r &=& {\cal{M}}_1 \cos\theta\\
\xi_\theta &=& -{\cal{M}}_2 \sin\theta\\
\xi_t &=& 0,\end{aligned}$$ where, $$\begin{aligned}
{\cal{M}}_0 &=&
-{\frac {P\left (-3\,\sqrt {r-2\,M}rM-5\,{r}^{3/2}M+2\,{M}^{2}\sqrt {r}
+2\,{r}^{5/2}+2\,\sqrt {r-2\,M}{r}^{2}\right )}{4\,\sqrt {r}M\left (r-2\,
M\right )}}\\
{\cal{M}}_1 &=&
-{\frac {P\left (-6\,\sqrt {r-2\,M}rM+2\,{r}^{5/2}-8\,{r}^{3/2}M+8\,{M}^{2}
\sqrt {r}+3\,\sqrt {r-2\,M}{M}^{2}+2\,\sqrt {r-2\,M}{r}^{2}\right )t}{4\,
{r}^{5/2}M}}\\
{\cal{M}}_2 &=&
-{\frac {P\left (-3\,\sqrt {r-2\,M}rM-5\,{r}^{3/2}M+2\,{M}^{2}\sqrt {r}
+2\,{r}^{5/2}+2\,\sqrt {r-2\,M}{r}^{2}\right )t}{4\,{r}^{7/2}M}}.\end{aligned}$$
Performing a first order gauge transformation with this generator, one eliminates the first order $\ell=1$ component of the metric. Therefore the leading terms in the initial data become second order. The latter have two contributions, both of $\ell=2$ multipolar order. One contribution simply comes from the expansion to second order of the initial data generated via the conformal approach. The other contribution comes from the fact that the first order gauge transformation we just performed has second order pieces of the form, $$\begin{aligned}
\widetilde{h}^{(2)}_{\mu\nu} & = &
h^{(2)}_{\mu\nu} \nonumber \\
& &
-\widetilde{h}^{(1)}_{\mu
\nu},_{\rho} \xi_{(1)}^{\rho} - \widetilde{h}^{(1)}_{\mu \rho}
\xi_{(1)}^{\rho},_{\nu} - \widetilde{h}^{(1)}_{\rho \nu}
\xi_{(1)}^{\rho},_{\mu} \\
& & -
{1\over2}g_{\mu \nu},_{\sigma},_{\lambda} \xi_{(1)}^{\sigma} \xi_{(1)}^{\lambda}
- g_{\mu \lambda},_{\sigma} \xi_{(1)}^{\sigma}
\xi_{(1)}^{\lambda},_{\nu} - g_{\lambda \nu},_{\sigma}
\xi_{(1)}^{\sigma} \xi_{(1)}^{\lambda},_{\mu} - g_{\sigma \lambda}
\xi_{(1)}^{\sigma},_{\mu} \xi_{(1)}^{\lambda},_{\nu} \nonumber\end{aligned}$$
The contribution to the second order Zerilli function coming from the nonlinear terms in the gauge transformation can be found after a straightforward but tedious calculation. The relevant contributions to the second order $\ell=2$ metric at $t=0$, are, $$\begin{aligned}
{}^{(2)}K &=& {2 \over 3} {P^2 R^2 \over M^2 r^3} \nonumber\\
{}^{(2)}H_2 &=& {2 \over 3} {P^2R^2 (R-M)^2 \over M^2 r^3 (r-2M)}
\nonumber\\
{}^{(2)}h_1 &=& {1 \over 3} {P^2 R^{5/2} (R-M) \over M^2 r^{3/2}
(r-2M)} \nonumber\\
{}^{(2)}G &=& {1 \over 3} {P^2 R^2 \over M^2 r^3}\end{aligned}$$ where $R = (\sqrt{r}+\sqrt{r-2M})^2/4$.
Another contribution to this initial data comes from the second order corrections to the conformal factor, calculated above. These contributions to the second order $\ell=2$ metric all vanish, except for, $$\begin{aligned}
{}^{(2)}H_2 & = & 4 P^2 \left( 1 + {M \over 2R}
\right)^3 \phi^{(2)}_2(R) {R^2 \over r^2} \nonumber \\
{}^{(2)}K & = & {}^{(2)}H_2, \end{aligned}$$ and all first order time derivatives vanish.
The Zerilli function in an arbitrary gauge is given by [@moncrief],
$$\begin{aligned}
\psi(t,r) & = & \frac{r(r-2M)}{3(2 r +3M)}\left[
{\ }^{(2)}{H}_2
-r\frac{\partial {\ }^{(2)}{K}}{\partial r}
-\frac{r-3M}{r-2M}{\ }^{(2)}{K}
\right] \nonumber \\
& & +{r^2 \over 2 r +3M} \left[{\ }^{(2)}{K}
+(r-2M)
\left(
\frac{\partial {\ }^{(2)}{G}}{\partial r}
-{2 \over r^2}{\ }^{(2)}{h}_1
\right) \right] .\end{aligned}$$
Substituting the above components yields the following initial data for the Zerilli function, $$\psi(t=0,r) =
P^2\left[ {(3M+8R)R^2 \over 6M^2r(2r+3M)} + {4 r (2M+3R)
\over 3 (2r+3M)} \phi^{(2)}_2(R)- {(4r-8M) R^{1/2} r^{3/2} \over 3
(2r+3M)} \left({\partial \phi^{(2)}_2(R) \over \partial r}\right)
\right]$$
$$\left.{\partial \psi(t,r) \over \partial t}\right|_{t=0} =0.$$
For the conformal factor we computed in (\[misnerbound\]), this results in, $$\begin{aligned}
\psi(t=0,r) &=&
{P^2 R^6 \over M (M+2R)^4 (M^2+10MR+4R^2)} \left(
{128 \over 3} +56 {M \over R} +{6113 \over 25} {M^2 \over R^2} +
{32607 \over 100} {M^3 \over R^3} \right. \nonumber \\
& & \left. +{3882 \over 25} {M^4 \over R^4} -{235 \over 16} {M^5 \over R^5}
- {691 \over 16} {M^6 \over R^6} - {26679 \over 1600} {M^7 \over R^7} -
{2047 \over 800} {M^8 \over R^8} -{2327 \over 19200} {M^9 \over R^9} \right)
\nonumber \\
& & +{7 P^2 M (M+12R)(M+2R)^2(\ln(2M)-\ln(M+2R)) \over 40 R^3 (M^2+10MR+4R^2)}.\end{aligned}$$
Physical validity of the perturbative treatment
-----------------------------------------------
Inspecting the initial data for the Zerilli function one finds that it behaves in a non-conventional way. Here is where we note significant differences with the case of single spinning holes [@GlNiPrPuboyo].
The first unusual thing one notices is that the initial data goes asymptotically to a constant value for $R\rightarrow\infty$. This is different from the data for the “close limit” of two black holes (momentarily stationary or boosted) where the Zerilli function goes to zero at infinity. The root of this problem can be traced down to the falloff conditions of the extrinsic curvature. In all other cases in which the “close limit” approximation has been applied, the extrinsic curvature falls off as $1/r^3$ at infinity. For the case in this paper, it decreases as $1/r^2$ (otherwise the momentum would vanish). This, in particular, leads to falloff conditions in the gauge vectors we use to eliminate the $\ell=1$ even pieces of the extrinsic curvature. In turn, the falloff condition of the gauge vectors influences, via the quadratic terms in the second order gauge transformation, the behavior of the Zerilli function we obtain at the end of the process.
Does the appearance of a Zerilli function that does not vanish at infinity indicate something problematic in itself? We do not seem to see any difficulty in evolving the problem in this case. Our evolution code evolves a “slab” region consistent of the initial data and its domain of dependence, therefore we do not need to specify any boundary conditions. If one observes the behavior of the Zerilli function as a function of time for a fixed (finite) value of the radius, it starts having a constant value followed by a quasinormal ringing and a power law tail decrease towards zero. That is, the constant behavior at infinity translates itself in a certain behavior at the beginning of the waveform, it does not leave any visible effect after the ringdown and power law tail behavior. The radiated energy [*and all observable quantities at infinity*]{} (even to second order in perturbation theory, see section III.B of [@physrep]) are not determined by the Zerilli function itself but by its time derivative, for which the initial data indeed goes to zero for large values of the radius, and therefore no problem in the evaluation of physical quantities is present. This might be at first surprising, but in reality the Zerilli function deos noe play any physical role, it is its time derivative what does, and the latter does not have any unusual asymptotic behavior. To put it in different terms: one could carry out perturbation theory and work out all relevant physics entirely in terms of the time derivative of the Zerilli function (which also satisfies the Zerilli equation) and there one would not see any unusual asymptotic behavior.
Another aspect that could cause concern is the nature of the gauge transformation considered. The gauge transformation is well behaved in any finite value of $r_*$ (that is, any point of the exterior of the black hole excluding the horizon and spatial infinity). Since in order to compute the radiated energies and waveforms we will never need information from either the horizon nor infinity, the gauge transformation is well defined in all relevant points for our calculation.
Results and conclusions
=======================
We have numerically evolved the Zerilli equation with the initial data presented above and computed the energy and waveforms for the “relaxation” of a single boosted Bowen–York hole to a boosted Schwarzschild black hole. The results for the energy can be summarized by a simple formula which, to leading order in $P$ reads, $$\left({E_{\rm radiated} \over M_{ADM}}\right) = 4.1 \times 10^{-2}
\left({P\over M_{ADM}}\right)^{4}$$ where the prefactor was computed numerically [^2].
Figure \[fig1\] shows the radiated energy as a function of the momentum. We see that for values of the momentum close to $P/M_{ADM}\sim 0.4$ the total radiated energy by the “relaxation” of a single boosted Bowen–York black hole to a Schwarzschild black hole is similar to the total radiated energy in a close limit collision of Bowen–York black holes [@GlNiPrPuboost2] and figure \[fig2\]. One could therefore be tempted to say that for values of $P/M_{ADM} > 0.4$ one should stop using these families of initial data. A puzzling element is that we have already evolved these families of initial data in collision situations and compared with actual non-linear integrations of the Einstein equations [@boost1; @GlNiPrPuboost2] and [*we know*]{} that these families of initial data radiate less in collisions than the values we are predicting here for each individual hole, at least for close separations. The results for these collisions are recollected in figure \[fig2\].
What is going on? One has to keep these results in perspective, since it is easy to get carried away and believe that perturbation theory should work way beyond where it was meant to do so. In the calculation of the present paper we find that the radiated energy goes as $P^4$. This is good, since our perturbative parameter is $P$ and therefore this means that the corrections are small. In the case of colliding black holes the radiated energies contain terms that go as (see reference [@GlNiPrPuboost2]) $L^4$, $P^2 L^2$ and $P L^3$, where $L$ is the separation of the black hole centers in the conformally related flat space. If one simply increases the value of $P$ keeping $L$ constant, it is obvious that the contribution we consider in this paper will quickly dominate. However, one is clearly pushing things beyond the realm in which these calculations were meant to be reliable. One is essentially forcing [*a higher order term in perturbation theory*]{} to a regime in which it is larger than the lower order terms!
Therefore the conclusion of this paper has to be read in the following way: the terms coming from the “relaxation” of a conformally flat hole to a usual boosted slice of Schwarzschild are higher order in perturbation theory than the ones one obtains in a collision. Because of this fact, they grow fast with the perturbative parameter and perturbation theory breaks down early for the estimation of the involved energy in the calculations of the current paper. The breakdown occurs earlier for a single hole than for a collision of holes (at least computed in the center-of-momentum frame).
Connected with the latter point is another interesting insight gained from the analysis of this paper. It has to do with the choice of frame used to describe collisions of black holes in perturbation theory. The situation is illustrated in figure \[fig3\]. One would expect these two collisions to be physically equivalent. However, if one considers conformal black holes and expands in perturbation theory, the collision at the bottom will contain terms that behave exactly like those we discuss in the current paper, whereas the top collision does not. If we power count, for the bottom collision, the extrinsic curvature has leading terms that behave like $P$ and are $\ell=1$ and the conformal factor has terms that behave like $L^2$ plus terms that behave like $P^2$ at leading order. The energy, being quadratic in the fields, will generically contain terms that behave like $L^4$, $P^2
L^2$ and $P^4$, the latter being the terms we encountered in this paper. However, the top collision contains terms $L^4$, $P L^3$ and $P^2 L^2$, as discussed in reference [@GlNiPrPuboost2] (the extrinsic curvature goes as $PL$ and the conformal factor as $L^2$ at leading order). Therefore if one were to consider “close” black holes and were to consider the radiated energy as a function of $P$, as we have done above, one would encounter that the perturbative predictions —at the order considered— will differ significantly as soon as the $P^4$ terms start to grow. The moral from this paper insofar as the choice of the origin is: perturbation theory breaks down quickly as a function of momentum for situations with net linear momentum, it is best to analyze collisions set up in the center-of-momentum frame.
As expected, given the nature of the Zerilli equation, the waveforms that one obtains from the “relaxation” just behave like quasinormal ringing. Figure \[fig4\] shows the waveform of the decay. The form is that of a typical ringdown, and it therefore takes an amount of time of the order of the light-crossing time of the black hole size to decay.
Summarizing, as in the case of spinning Bowen–York black holes, one has extra radiation present in the initial data, that grows with the value of the momentum. In evolutions of binary black hole collisions, one can either wait long enough for this energy to be “flushed away” from the system, or restrict oneself to values of the momentum that are small enough that the extra energy is small respect to the total energy produced in the collision. The latter is the only option in the case of “close limit” collisions. Another conclusion is that perturbation theory breaks down quickly as a function of the momentum of the holes for single boosted holes (and collisions of black holes not computed in the center-of-momentum frame) and therefore cannot be used to reliably estimate the “energy content” of each hole in a regime that might be of interest for the momenta and energies relevant for black hole collisions.
Acknowledgments
===============
We wish to thank a anonymous referees for comments on an earlier version of this manuscript. This work was supported in part by grants of the National Science Foundation of the US INT-9512894, PHY-9423950, PHY-9800973, PHY-9407194, by funds of the University of Córdoba, the Pennsylvania State University and the Eberly Family Research Fund at Penn State. We also acknowledge support of CONICET and CONICOR (Argentina). JP also acknowledges support from the Alfred P. Sloan and John S. Guggenheim Foundations. JP was a visitor at ITP Santa Barbara during the part of the completion of this work. The authors wish to thank Richard Price and the University of Utah for hospitality. This work was supported in part by funds of the Horace Hearne Jr. Institute for Theoretical Physics.
See for instance Cook et al. Phys. Rev. Lett. [**80**]{} 2512 (1998), and Gómez et al. Phys. Rev. Lett. [**80**]{} 3915 (1998).
J. Pullin, Prog. Theor. Phys. Suppl. [**136**]{}, 107 (1999)
J. Bowen, J. York, Phys. Rev. [**D21**]{}, 2047 (1980).
A. Garat, R. Price, Phys. Rev. D [**61**]{}, 124011 (2000)
S. Brandt, B. Brügmann, Phys. Rev. Lett. [**78**]{}, 3606 (1997).
R. Gleiser, C. O. Nicasio, R. Price, J. Pullin, Phys. Rev. [**D57**]{}, 3401 (1998).
T. Regge, J. Wheeler, Phys. Rev. [**108**]{}, 1063 (1957).
D. Brill, R. Lindquist, Phys. Rev. [**131**]{}, 471 (1963).
G. Cook, Ph.D. thesis, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, 1990; G. B. Cook, M. W. Choptuik, M. R. Dubal , Phys. Rev. D [**47**]{}, 1471 (1993); G. B. Cook, Phys. Rev. D [**50**]{}, 5025 (1994).
R. Gleiser, C. Nicasio, R. Price, J. Pullin, Phys. Rept. [**325**]{}, 41 (2000).
G. Cook, J. York, Phys. Rev. [**D41**]{}, 1077 (1990).
R. Gleiser, O. Nicasio, R. Price, J. Pullin, Phys. Rev. [**D59**]{} 044024, (1999).
F. J. Zerilli, Phys. Rev. [**D2**]{} 2141 (1970).
R. Gleiser, O. Nicasio, R. Price, J. Pullin, Class. Quan. Grav. [**13**]{}, L117 (1996).
V. Moncrief, Ann. Phys. (NY) [**88**]{}, 323 (1974).
J. Baker, A. Abrahams, P. Anninos, R. Price, J. Pullin, E. Seidel, Phys. Rev. [**D55**]{}, 829 (1997).
[^1]: Alternatively, we may use a Brill–Lindquist type boundary condition [@BrLi] and as a result one obtains the “puncture” solutions of [@BrBr].
[^2]: The ADM mass of the slice depends on the value of the momentum. For these calculations we are using the zeroth order approximation to the ADM mass, which is constant. It is known from full numerical calculations that for $P/M_{ADM}<0.6$ the discrepancy between the zeroth order approximation and the full value is less than a $1\%$ (see figure 1 of reference [@CoYo]). This can also be seen from the perturbative calculation of the mass, which one can obtain from the conformal factor we computed, and yields $M_{ADM} = M+ {P^2
\over 8M} - {2287 P^4 \over 240000 M^3}$, for instance, for the “puncture” case corresponding to the conformal factor of equation (\[brbrbrbr\]).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The cosmological effects of the tachyon rolling down to its ground state are discussed by coupling a simple effective field theory for the tachyon field to Einstein gravity. As the tachyon rolls down to the minimum of its potential the universe expands. Depending upon initial conditions, the scale factor may or may not start off accelerating, but ultimately it ceases to do so and the final flat spacetime is either static in the rest frame of the tachyon (if $k=0$) or ( if $k=-1$ ) given by the Milne model.'
author:
- |
G W Gibbons\
D.A.M.T.P.,\
C.M.S.,\
Cambridge University,\
Wilberforce Road,\
Cambridge CB3 0WA,\
U.K.
title: Cosmological Evolution of the Rolling Tachyon
---
Introduction
============
The reconciliation of fundamental theories such as M/String theory with the basic facts of cosmology continues to present great challenges. The most straightforward approach is to pass to an effective field theory in which gravity is coupled to matter, for instance in a supergravity theory. Because the theories are formulated in higher dimensions, one must either construct a spontaneous compactification scenario or imagine a scheme in which the universe is sort of 3-brane [@Gibbons-Wiltshire]. It seems to be very difficult to construct models with $N=1$ supersymmetry on the 3-brane [@Gibbons-Lambert]. The trouble with compactifications is that they come with associated massless scalars and degenerate vacua: one must address the question of how these evolve with time [@Gibbons-Townsend]. One must also ensure that the resultant time-variability of coupling constants is compatible with observations. One possibility is to give the potentials a mass “by hand” such a way that Minkowski-space times a Calabi-Yau is an attractor at late times, [@Gibbons-Chapline] but this is [*ad hoc*]{} and ugly.
There is fairly good evidence from the BOOMERANG observations of the Cosmic Microwave Background that the scale factor of the universe underwent a period of acceleration (so-called Primordial Inflation) at early times and from Type Ia super-novae that it may also have been accelerating very recently if not today. (For a recent review see [@Straumann]). It is quite difficult to get accelerating universe out of pure supergravity theories [@Gibbons; @Maldacena-Nunez; @Hull; @Gibbons-Hull; @Townsend; @Gates] although with super-matter, providing one gauges a suitable axial current this is possible [@Freedman]. The problem is that the axial gauging gives rise to anomalies [@Freedman2]. It is possible that these anomalies can be cancelled in staring models with D-branes [@Kallosh].
In recent years there has been great progress, particularly due to Sen, in our understanding of the role of the tachyon in String Theory (see [@Sen1] for a recent account with references to earlier work) . The basic idea is that the usual open string vacuum is unstable but there exist a stable vacuum with zero energy density which is stable. There is evidence that this state is associated with the condensation of electric flux tubes of closed strings (see [@Sen1; @Sen]). These flux tubes described successfully using an effective Born-Infeld action (see [@Gibbons-Hori-Yi; @Sen1; @Sen3] and references therein). This success of effective action methods, together with the difficulties of other approaches described the encourages one to pursue this further and to attempt a description of the cosmology of tachyon rolling. Moreover not to take into account the effects gravity during the process is inconsistent, since it involves a spatially uniform distribution of energy. It is the purpose of this note to rectify this omission and initiate a study of tachyon cosmology .
The Rolling Tachyon
===================
The tachyon of string theory may described by effective field theory describing some sort of tachyon condensate which in flat space has a Lagrangian density = -V(T) , where $T$ is the tachyon field, $V(T)$ is the tachyon potential and $\eta_{\mu \nu} = {\rm diag}(-,+,+,\dots
)$ is the metric of Minkowski spacetime ( see [@Sen] for a discussion with references to earlier work) . The tachyon potential $V(T)$ has a positive maximum at the origin and has a minimum at $T=T_0$ where the potential vanishes. In [@Sen], $T_0$ is taken to lie at infinity. In Minkowski spacetime the rolling down of the towards its minimum value is described by a spatially homogeneous but time-dependent solution obtained from the Lagrangian density = -V . \[Lagrangian\] During rolling the Hamiltonian density = [V(T) ]{} \[Hamiltonian\] has a constant value $E$. Thus T = \[evolution\].As $T$ increases $V(T)$ decreases and $\dot T$ increases to attain its maximum value of $1$ in infinite time as $T$ tends to infinity. Note that as explained in [@Sen] the tachyon field behaves like a fluid of positive energy density = [V(T) ]{} \[density\] and negative pressure P= -V(T) . \[pressure\] Thus P = - V\^2 (T) and = w= -( 1- T\^2 ), and therefore, $-1 \le w \le 0$. Note that both the Weak Energy Condition, $\rho >0$ and Dominant Energy Condition, $\rho \ge |P|$ hold. However because + 3P = -[2 V(T) ]{} ( 1- [ 3 2]{} T\^2 ) the Strong Energy Condition fails to hold for small $|\dot
T|$ but does hold for large $|\dot T|$.
The discussion above has neglected the gravitational field generated by the tachyon condensate. To take it into account we use the Lagrangian density ( [ R 16 G]{} - V(T) ), where $g_{\mu \nu}$ is the metric and $R$ its scalar curvature. We shall work in $3+1$ spacetime dimensions and assume that the metric has Friedman-Lemaitre-Robertson-Walker form ds\^2 = -dt \^2 + a\^2(t) d \^2 \_k, where $a(t)$ is the scale factor and $d \Omega^2 _k $ is, locally at least, the metric on $S^3$ , ${\Bbb E } ^3$ or $ H^3$ according as $k=1,0,-1$ respectively. Note that in this model we have assumed that the cosmological constant $\Lambda$ vanishes in the tachyon ground state. The expressions (\[density\] ) and ( \[pressure\]) for the density and pressure remain valid and thus the Friedman and Raychaudhuri equations governing the evolution of the scale factor are + [ k a\^2 ]{} = [ 8G 3]{} [ V(T) ]{}, \[Friedman\] and
= [ 8G 3]{} [ V(T) ]{} ( 1- [ 3 2]{} T \^2 ) .\[Raychaudhuri\]
The Hamiltonian density of the tachyon field is no longer constant because the tachyon Lagrangian density is now explicitly time dependent. Equations (\[Lagrangian\]) and (\[Hamiltonian\]) must be replaced by = -a\^3 V . \[Lagrangian2\] and = a\^3 [V(T) ]{} \[Hamiltonian2\] .The equation = - [ t]{} \[timedependent\] is formally equivalent to the conservation of entropy of the fluid which reads = - [ a a]{} (+ P). Because + P =[ T\^2 ]{}, we have ( [ V ]{} ) = - [ 3 a a]{} ( [ V T\^2 ]{} ). Thus the evolution equation (\[evolution\]) remains valid but the quantity $E$ is no longer constant but rather decreases in time.
The course of cosmic evolution is now rather clear from these equations. The tachyon field rolls down hill with an accelerated motion and the universe expands. It still follows from (\[evolution\] ) that as $T$ increases $V(T)$ increases but since $E$ decreases, in principal $\dot T$ could decrease but in any case $\dot T$ remains positive and so $T$ increases monotonically to attain its maximum value of 1. From the Friedman equation (\[Friedman\]), it follows that if $k\le 0$, then $\dot
a $ will always be positive. This is because the Weak Energy Condition holds, $\rho
>0$. From the Raychaudhuri equation (\[Raychaudhuri\] ) one deduces that initially if $ |\dot T | < { 2 \over 3} $ the scale factor initially accelerates, $\ddot a >0 $ but eventually, once $\dot T$ exceeds $\sqrt { 2\over 3}$ the acceleration will cease and deceleration will set in. If the universe is flat, i.e. if $k=0$, then ultimately the scale factor will halt $a(t) \rightarrow {\rm constant}$. If the universe has hyperbolic sections, that is if $k=-1$, then ultimately the scale factor increases linearly with time, $a
\rightarrow t$. In both cases the final state of the universe is flat, the case $k=-1$ being the Milne model. In the case of spherical sections, $k=+1$ re-collapse will take place. The possibility of cosmic acceleration arises from the positive potential $V(T)$ and should be contrasted with the situation in pure supergravity theories for which the Strong Energy Condition holds and cosmic acceleration is not possible [@Gibbons; @Maldacena-Nunez], unless the internal space is non-compact [@Hull; @Gibbons-Hull]. The inclusion of supermatter matter may allow acceleration [@Townsend].
Despite the violation of the Strong Energy condition one sees from the Friedman equation (\[Friedman\]) that in the cases $k\le 0$ the positivity of energy precludes the avoidance of singularities in the past. If however $k=1$, it is conceivable that, for special initial conditions, the scale factor might pass through a finite sequence of minima and maxima or even that periodic or quasi-periodic solutions exist with an infinite sequence of maxima and minima.
Acknowlegement
==============
After submitting this paper to the archive I was told of an earlier paper of Stephon Alexander [@SA] of which I was unaware and which anticipated some of the ideas discussed here, in terms of $D$-$\bar D$ annihilation. The action adopted for the tachyon field is different from that used in this paper. Another relevant pre-cursor brought to my attention by Anupam Mazumdar and to which the same remarks apply is [@MPP]. I would like to thank them and also Andrew Chamblin, Neil Lambert, Mohammad Garousi, Thanu Padmanabhan, Nakoi Saskura and Arkady Tseytlin for helpful comments and pointng out typos and small inaccuracies in the wording.
[99]{} G W Gibbons and P K Townsend, Cosmological Evolution of Degenerate Vacua‘ [*Nucl Phys*]{} [**B282**]{} (1987) 610 G Chapline and G W Gibbons, Unification of Elementary Particle Physics and Cosmology in Ten dimensions [*Phys Letts* ]{} [**B135**]{}(1984)43 G W Gibbons and D Wiltshire,‘Spacetime as a Membrane in Higher Dimensions‘ [*Nucl Phys* ]{}[**B 287**]{}(1987) 717, [hep-th/0109093]{} G W Gibbons and N D Lambert, ‘Domain Walls and Solitons in Odd Dimensions ’ [*Phys Lett*]{} [**B488**]{} (2000) 90-96, [hep-th/0003197]{}
D Z Freedman, ‘Supergravity with axial-gauge invariance‘ [*Phys Rev*]{} [**D15**]{} (1977) 1173-1174 D Z Freeman, private communication (December 2001) C Herdeiro, S Hirano and R Kallosh, String Theory and Hybrid Inflation/Acceleration [*JHEP*]{} [**0102**]{} (2001) 027, [hep-th/0110271]{}
N Straumann, ‘On the Cosmological Constant Problems and Astronomical Evidence for a Homogeneous Energy Density with Negative Pressure, [astro-ph/020330]{} G W Gibbons, ‘Aspects of Supergravity Theories [*GIFT Seminar on Supersymmetry, Supergravity and Related Topics*]{} eds F. del Aguila, J A de Azcarraga and L E Ibanez, World Scientific(1984) J Maldacena and C Nunez,‘ Supergravity description of field theories on curved manifolds and a no go theorem’ [*Int J Mod Phys*]{} [**A16**]{} (2001) 822, [hep-th/0007018]{} S J Gates,‘ Is Stringy-Superysmmetry Quintessentially Challenged?’, [hep-th/0122112]{}
C M Hull, ‘De-Sitter Space in Supergravity and M-Theory’[*JHEP*]{} [**0109**]{} (2001) 027, [hep-th/0109213]{} G W Gibbons and C M Hull, ‘De-Sitter Space from Warped Supergravity Solutions‘, [hep-th/0111072]{} P K Townsend, ‘Quintessence from M-Theory‘ [*JHEP*]{} [**0111**]{} (2001) 042, [hep-th/0110072]{} A Sen, ‘Rolling Tachyon’, [hep-th/0203211]{} A Sen, ‘Tachyon Matter’, [hep-th/0203265]{} G W Gibbons, K Hori and P Yi, ‘ String fluid from unstable D-branes, [*Nucl Phys*]{} [**B596**]{} (2001) 136, [hep-th/0009061]{} A Sen, ‘Fundamamental Strings in open string theory at the tachyonic vacuum‘ [*J Math Phys*]{} [**42**]{} (2001) 2844, [ hep-th/0010240]{}
S Alexander, ‘Inflation from $D-{\bar D}$ Brane Annihilation, [*Phys Rev*]{} [**D65**]{} (2002) 023507 [hep-th/0105032]{}
A Mazumdar, S Panda and A Per/’ez-Lorenzan ‘Assisted inflation via tachyon condensation ’ [*Nucl Phys*]{} [**B614**]{} (2001) 101-116 [hep-ph/0107058]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We provide a brief review of the current status of resummed predictions for QCD final state observables such as event shapes and jet rates in a variety of different hard processes. Particular emphasis is given to more recent developments such as the study of non global observables, development of generalized resummation formulae and resummations for new types of event shape variables.'
address: |
Theory Division, CERN\
CH-1211, Geneva 23, Switzerland.
author:
- Mrinal Dasgupta
title: Status of resummed predictions for QCD final state observables
---
INTRODUCTION
============
All order resummed predictions have long been a common method to extend the predictive power of theoretical QCD estimates for several observables compared to that provided by fixed order perturbative computations alone. This applies in particular to observables that are measured close to their exclusive limits, such as event shape variables in $e^{+}e^{-}$ annihilation [@CTTW] where the highest statistics are typically found close to the two jet limit.
More generally, if a physical variable such as a typical event shape, defined for an arbitrary hard process, is constrained, by observation/measurement, to be near its Born value , it turns out to be sensitive to soft emissions. In fact if the deviation from the Born value of an observable (which we can take to be zero for the following discussion) is denoted by $V$, near the Born limit the cross-section for values up to $V$ ($V \ll 1$) is dominated by soft and/or collinear logarithms. In particular the $n^{\mathrm{th}}$ order perturbative estimate typically goes as : $$R^{(n)}(V,Q) \sim \left ( \alpha_s(Q) \ln^2 {V} \right)^{n}+\cdots$$ where the dots denote terms less singular in the $V \to 0$ limit and $Q$ is a scale relevant to the underlying hard process. Hence the predictive power of fixed order perturbation theory is spoilt in the small $V$ region, since the smallness of the expansion parameter $\alpha_s(Q)$ is compensated by the presence of large logarithms with upto two powers of $\ln 1/V \equiv L$ for each power of $\alpha_s$.
By now there are solid techniques in place to handle such logarithmic behaviour to all orders, which result in improved predictions valid over a much larger range of variable values than the fixed order computations alone, which become meaningless at very small $V$. These resummation techniques rely mainly on factorisation methods (broadly speaking), which in turn follow from QCD coherence properties and the observables dependence on final state emissions being factorisable, usually in some appropriate transform space [@CTTW]. Due to the latter property not all variables turn out to belong to the class of resummable observables–an example is jet rates defined with the JADE jet algorithm [@JADE].
While resummed predictions are often needed to improve perturbative accuracy, they also serve as a pathway to approach the non-perturbative domain one of the most challenging and certainly the least understood aspect of QCD. Specifically, to access large non-perturbative effects one has to enter the low $k_t$ domain, where $k_t$ is the transverse momentum of a typical gluon emission, leading to a measured value of the observable. Hence in the small $k_t \sim VQ$ region, aside from large logarithms one also has power corrections $\lambda^n/(VQ)^n$ where the $\lambda$ are non-perturbative coefficients. While for $VQ \sim \Lambda_{{\mathrm{QCD}}}$ the problem becomes highly non-perturbative, nevertheless for hard enough processes a wide range of values exists, $\Lambda_{\mathrm{QCD}} \ll VQ \ll Q$, where both perturbative resummations and non-perturbative power corrections are important effects. Exploring this region quantitatively is important as it provides valuable information on the onset of confinement effects.
In short resummations are a useful probe of QCD dynamics, which comes into its own in the infrared region, where parton multiplication is copious and hadronisation effects are important. In what follows we shall confine the discussion to observables that do admit phase space factorisation and hence are resummable. We shall begin by considering the anatomy of a typical resummed answer and indicate what is the state-of–the art with regard to computation of the various pieces we will mention. We shall then discuss some significant recent developments, which contributed to the improvement of resummed predictions and consequently reflect an improved understanding of QCD dynamics: in particular the discovery of non global logarithms and the advent of generalised approaches to resummation. We shall also present some recent comparisons of such improved predictions with data.
PROFILE OF A RESUMMED PREDICTION
================================
Consider an observable which has large logarithms in its perturbation expansion as below: $$R(V) = 1+\sum_{n=1} \alpha_s^{n} \left ( \sum_{m=0}^{2n} R_{nm} \ln^{m} \frac{1}{V}+\mathcal{O}(V) \right ).$$
One can naturally define leading logarithms (LL) as being the double logs $m=2n$ , next-to–leading logs (NLL) as those terms with $m=2n-1$ etc and in some cases this nomenclature is indeed employed [@burbglov]. Note however that in the regime where $\alpha_s L^2 \gg 1$, any truncation at ${\mathrm{N^{p} LL}}$ order, in this terminology, is no longer useful since one always has neglected terms $\alpha_s^n L^m$ with $2n-p >m>n$ which are larger than one. It should be pointed out that the state-of–the art resummation enables us to go to NNLL order in this notation.
However since variables that admit phase space factorisation typically exponentiate, it is possible to have a different (and in fact more commonly used) classification of logarithmic terms. Exponentiation means one can write [@CTTW] $$R(V) = (1+C_1\alpha_s+\cdots) \Sigma(\alpha_s, L)+ D(V,\alpha_s)$$
The first piece in brackets is simply a well behaved perturbative expansion (with no dependence on $V$) in $\alpha_s$. This piece treats for instance the mismatch between the full real and virtual emission and their soft limits, included in the form factor $\Sigma$.
All the singular dependence on $V$ is contained in the form factor $\Sigma$. It takes the form $$\Sigma = e^{\left [ L g_1(\alpha_s L) +g_2(\alpha_s L) +\alpha_s g_3(\alpha_s L) +\cdots \right ]}.$$
With this exponentiation it is now possible to define $g_1$ as a LL function (since it is accompanied by an extra logarithm $L$), $g_2$ as an NLL function (which is purely single logarithmic $\alpha_s^{n} L^n$), $g_3$ as NNLL due to the extra factor of $\alpha_s$ relative to the single logarithmic $g_2$ piece. In this notation the state of the art is NLL resummation in that resummed predictions aim to compute up to the $g_2$ piece. This prediction is then matched to fixed order calculations. The ‘remainder’ piece $D(V)$ vanishes in the $V \to 0$ limit .
Having clarified the basic notation and terminology we now turn our attention to some more recent developments. We begin the next section by providing a short discussion on non global observables.
NON GLOBAL EFFECTS
==================
Until very recently the resummed predictions available in the literature were all made employing an independent emission formalism. In other words the matrix element for multiple gluon production was approximated by a factorised product of single soft emission contributions: $$d \mathcal{P}_n \approx \prod_{i=1}^{n}
C_F \alpha_s(k_{ti}) d\eta \frac{dk_{ti}}{\pi k_{ti}}.
\label{eq:indem}$$ with $k_t$ the transverse momentum and $\eta$ the rapidity wrt the emitting jet axis.
The independent emission pattern (in one form or another) was used in conjunction with phase space factorisation to yield the form factor $\Sigma$. Independent emission essentially means one can assume each final state gluon to be emitted directly by the hard initiating parton as in an abelian theory. Subsequent branching of these gluons can in fact be neglected for several observables, apart from the contribution of such decays to the running of the coupling.
However the independent emission approximation (\[eq:indem\]) was found to be insufficient at single logarithmic level (ie at the level of the function $g_2$) for several observables [@dassalng1]. In some of these cases, such as energy flow away from jets [@dassalng2], single logarithms were in fact a leading effect due to absence of collinear enhancements, which meant that the function $g_1$ was absent. In general the observables for which independent emission breaks down at single-log level have one common feature responsible for this breakdown – they are sensitive to emissions only in a limited rapidity region. For this reason such observables and corresponding logarithms are refered to as [*[non global]{}*]{}, while the complementary set of observables is refered to as global.
The origin of the problem is simple to understand and briefly explained below.
  
Consider a measurement of say energy flow away from hard jets in a solid angle $\Omega$ as depicted above. Then by restricting, for example, the transverse energy in $\Omega$ such that $Q_\Omega \ll Q$ with $Q$ a hard scale, one is vetoing real radiation above scale $Q_\Omega$ into $\Omega$. Assuming an independent emission ansatz one needs to just veto direct emission from the hard jet lines as in diagram (a) above.
![Non global function $S(t)$ plotted against $t = \int_{Q_\Omega}^{Q} \frac{dk_t}{k_t} \alpha_s (k_t) $ for different definitions of $\Omega$.[]{data-label="fig:ng"}](Sfull.eps)
This yields a single logarithmic form factor $$\label{eq:primary}
\Sigma_{\mathrm{SL}} \sim \exp[- \alpha_s \ln Q/Q_\Omega ].$$ The above result is incorrect at single log level. This can be seen by expanding it to ${\mathcal{O}}(\alpha_s^2)$ and comparing with the logarithmic dependence of an exact fixed order estimate at ${\mathcal{O}}(\alpha_s^2)$. In fact an additional source of single logarithms is found starting from $\mathcal{O}(\alpha_s^2)$ which removes the discrepancy with the single logarithms in the fixed order estimate.
At order $\mathcal{O}(\alpha_s^2)$ this is just emission of a soft gluon with energy $\omega \sim Q_\Omega$ by a relatively harder gluon with energy $\omega'$ with $Q_\Omega \ll \omega' \ll Q$ outside $\Omega$ (see diagram (b)). This and analogous higher order contributions are missed by the independent emission approach which would be correct if both the gluon outside $\Omega$ and the softest gluon inside $\Omega$ were measured (vetoed). The effect of the softest gluon would cancel against virtual corrections at single logarithmic level. However since we are only measuring [*[inside]{}*]{} $\Omega$ the softest gluon emission is significant even in the presence of the harder emitter outside $\Omega$ and hence the gluon branching contribution appears.
The generalization of the above effect to $n^{\mathrm{th}}$ order is the coherent emission of a single softest gluon into $\Omega$ by an ensemble of $n-1$ gluons outside $\Omega$ (diagram (c) above. The $n-1$ gluon emitters are themselves ordered in energy, $\omega_0 \gg \omega_1 \gg \omega_{n-1}$ . The effect of such multiple wide-angle (non collinear-enhanced) soft emission is precisely a single logarithm, $\alpha_s^{n} \ln^{n} \frac{Q}{Q_\Omega}$. Resumming these terms needs a change of approach from the independent emission approximation. In Ref. [@dassalng1; @dassalng2], the resummation of single logarithms is carried out numerically by using the dipole evolution picture which captures the essential single logarithms in the large $N_c$ limit. The resulting function is plotted in Fig. \[fig:ng\]. Following this a non-linear evolution equation corresponding to the dipole evolution was derived in Ref. [@BMS] and its numerical solution yielded identical results to those obtained in Ref. [@dassalng2].
Subsequent to the discovery of non global logs, there was some effort made at understanding how to define observables in a way such that non global effects may be minimised, which would reduce any uncertainty to do with missing non global $1/N_c^2$ suppressed effects which cannot yet be resummed. The reduction or removal of the non global component is also useful for facilitating the phenomenology of multi jet event shapes and energy flows. This is because non global logs have thus far only been explicitly computed in the two jet case, although a similar structure is expected in the extension to multi-jet events and their calculation will also be along the same lines, employing dipole (non-linear) evolution.
In the above regard Appleby and Seymour considered the effect of jet clustering algorithms on non global logarithms. They found that if they defined rapidity gaps in terms of minijet energy flows (with the minijets being defined by a clustering alogorithm), rather than a sum over hadronic energies in the gap, the non global component was significantly reduced. This is essentially because the clustering algorithm has the effect of pulling soft hadrons (partons) out of the gap and clustering them with harder emissions outside the gap. Hence the non global contribution is not triggered except in specific geometrical configurations which survive the clustering, which reduces its numerical significance (see [@applesey] for details).
Berger, Kucs and Sterman introduced event-shape/energy flow correlations [@BKS] aimed at controlling non global effects in energy flow. This meant simultaneously restricting the energy $Q_\Omega$ in $\Omega$ alongside limiting the value of an event shape variable $V$, defined in terms of hadron momenta outside $\Omega$. Doing so amounts to restricting transverse momenta of soft gluons both outside the gap region, by controlling the event shape $k_t \leq VQ$, and inside the gap $k_t \leq Q_{\Omega}$ with $Q_{\Omega} <VQ \ll Q$. Non global or secondary logarithms now appear as $\alpha_s^n \ln^n \frac{VQ}{Q_\Omega}$ contributions and hence varying the values of $V$ and $Q_\Omega$ in tandem, allows one to control the significance the of the non-global contribution.
In the course of these studies the same authors also introduced new kinds of event shapes with an adjustable parameter $a$ that allows one to control the approach to the two jet limit [@BKS]. These variables have a parametric dependence on the transverse momenta $k_t$ and rapidity $\eta$ with respect to the thrust axis, of the form $V = \sum_i k_{ti} e^{-(1-a)|\eta_i|}$, where the sum extends over all final state partons. The resummation to NLL accutacy, for these observables can be found in Ref. [@BKS]. A special case, $a=0$, of the above class of variables is the much studied thrust variable.
Dokshitzer and Marchesini [@dokmarch] further extended the study of event shape/energy flow correlations by showing that the non global part of the answer, which involves logarithms of $QV/Q_\Omega$, factorises from the usual global resummation of the event shape $V$ (this time defined as a sum over all final state hadron contributions in and outside the gap).
Although some effort has been devoted to understanding and computing non global logarithms, the only inclusion of this effect in comparisons to data thus far has been in the case of DIS event shape variables. For details the reader is refered to Ref. [@dassaldis]. A recent comparison of a matched resummed prediction (including non global effects and power corrections) to H1 data is shown in Fig. 2, for the thrust defined wrt the thrust axis in the Breit current hemisphere of DIS.
THREE JET SHAPE VARIABLES
=========================
An interesting and fairly recent development is that of NLL resummed predictions for three jet event shapes in $e^{+}e^{-}$ , DIS (two final state +one incoming jet) and hadron-hadron collisions (two incoming jets and a final state (radiated) jet+vector boson). For an example see [@BDMZ].
![Resummed distribution for the thrust defined wrt the thrust axis in the current hemisphere of the DIS Breit frame compared to H1 data. From top to bottom represents increasing $Q$ values from 15 – 81 GeV.[]{data-label="fig:DIS"}](ladder-tautE.eps)
For all these processes, at Born level one finds that the three hard jets lie in a plane defined say by the thrust and thrust major axes. Small deviations from the Born level can be accesed by studying small values of the out of plane momentum $K_{\mathrm{out}}$. Compared to two jet event shapes such as thrust, the three jet variables offer some new challenges. From the point of view of the resummed prediction, one as to also take account of coherent, soft interjet radiation while in the global two jet case only radiation emitted collinearly (effectively incoherently) by each jet contributes to NLL accuracy. The interjet radiation gives a geometry dependence at NLL level, characteristic of three jet variables. Such variables would be very interesting to study experimentally especially from the viewpoint of hadronisation corrections. This is a particularly important intermediate step along the way to understanding power behaved hadronisation effects in multijet configurations, such as in dijet event shapes in hadroproduction ($2 \to 2$ processes).
GENERALISED FORMULAE AND AUTOMATED RESUMMATIONS
===============================================
Of late there has been significant effort and progress made in developing resummation formulae that are as general as currently possible and which reduce the need for treating different observables and different processeses on a laborious case-by–case basis. The idea is to write down master formulae which have general applicability and contain process and observable dependence in a few parameters which need to be computed on a case specific basis [@KCLOS; @BCMN; @NUMSUM2]. For example processes involving an arbitrary number of hard partons were treated in Ref. [@BCMN].
A generalised approach that aims at facilitating the computablity of resummed results , using numerical methods, and in fact automating the entire process of resummation was developed in [@NUMSUM2]. The idea roughly, is to test the observable’s dependence on soft and collinear emissions and determine numerically the parametric dependence on transverse momenta, rapidity and azimuth wrt the nearest leg. Once this dependence is obtained, the results are automatically inserted into a master formula valid for an arbitrary case (there however being some conditions such as globalness, that the observable has to satisfy for the master formula to be valid). Final results are numerically obtained by means of a computer program. This is a particularly powerful method since almost no analytical effort is needed to generate the final results and the entire procedure is automated. Its advantage is particularly manifest in cases where analytical calculations are too cumbersome, requiring the inversion of multiple Mellin and/or Fourier transforms or the observable depends in a complicated way on multiple soft emissions. A good example of this latter point is the Durham jet finding algorithm for which the analytic computation of the full NLL function $g_2$ was intractable due to effects to do with recombination of soft partons. The numerical resummation methods were on the other hand able to determine this function for the first time [@NUMSUM] .
Another example of the power of the automated resummation approach was provided by the computation, for the first time, to NLL accuracy of an event shape in hadron–hadron collisions (with a dijet final state) – the transverse thrust distribution [@NUMSUM2]. This is defined as follows: $$T_\perp = {\mathrm{max}}_{\vec{n}_{\perp}} \frac{\sum_i |\vec{p}_{\perp i}.\vec{n}_\perp|}{\sum_i p_{\perp i}}$$ where as usual the sum extends over all final state particles and $p_{\perp}$ is the transverse momentum wrt the beam direction while $\vec{n}$ is a unit vector found by maximising the sum in the numerator. The results for the different dijet production channels is illustrated in Fig. 2.
In conclusion it is worthwhile to note that there has been significant progress in the development of resummed computations in the last few years. The consequences of these developments are naturally of great value to experimental studies and future QCD phenomenology. Additionally they point towards unravelling of previously unexplored QCD dynamics and therefore to important progress on the theoretical front.
![Plot of the transverse thrust distribution in hadronic collisions at $\sqrt{s}=1.96$ GeV.. Fig. taken from Ref. [@NUMSUM2].[]{data-label="fig:transthrus"}](resdif-othrsub-num.eps)
ACKNOWLEDGEMENT {#acknowledgement .unnumbered}
===============
I would like to thank the authors of Ref. [@NUMSUM2] for the use of their figure for Fig. 3.
[9]{} S. Catani, L. Trentadue, G. Turnock and B.R. Webber Nucl. Phys B407 (1993) 3. N.Brown and W.J. Stirling, Phys. Lett. B 252 (1990) 657. S. Catani, Yu.L. Dokshitzer and B.R. Webber, Phys. Lett. B322 (1994) 263. M. Dasgupta and G.P. Salam, Phys. Lett. B512 (2001) 323. M.Dasgupta and G.P. Salam, JHEP 0203 (2002) 017. A. Banfi, G. Marchesini and G.E. Smye, JHEP 0208 (2002) 006. R.B. Appleby and M.H. Seymour, JHEP 0212 (2002) 063. C.F. Berger, T. Kucs and G. Sterman, Phys. Rev. D68 (2003) 014012. Yu.L. Dokshitzer and G. Marchesini, JHEP 0303 (2003) 040. M. Dasgupta and G.P. Salam, JHEP 0208,(2002) 032. A.Banfi, Yu.L. Dokshitzer, G. Marchesini and G. Zanderighi, Phys. Lett. B508 (2001). N. Kidonakis, G. Oderda and G. Sterman, Nucl. Phys B531 (1998) 365; E. Laenen, G. Oderda and G. Sterman, Phys. Lett. B438 (1998) 173. R. Bonciani, S. Catani, M.L. Mangano and P. Nason, hep-ph/0307035 (2003). A. Banfi, G.P. Salam and G. Zanderighi, preprint LPTHE-03-12, hep-ph/0304148. A. Banfi, G.P. Salam and G. Zanderighi, JHEP 0201, (2002), 18.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Matthew Amodio\
Yale University\
Rim Assouel\
Université de Montréal, Mila\
Victor Schmidt\
Université de Montréal, Mila\
Tristan Sylvain\
Université de Montréal, Mila\
Smita Krishnaswamy[^1]\
Yale University\
Yoshua Bengio\
Université de Montréal, Mila\
bibliography:
- 'references.bib'
title: 'Image-to-image Mapping with Many Domains by Sparse Attribute Transfer'
---
Introduction
============
Related Work
============
Model
=====
Experiments
===========
Discussion
==========
Broader Impact
==============
[^1]: Equal contribution.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Vehicle-to-grid increases the low utilization rate of privately owned electric vehicles by making their batteries available to the electricity grid. We formulate a robust optimization problem in continuous time that maximizes the expected profit from selling primary frequency regulation to the grid and guarantees that vehicle owners can meet their market commitments for all frequency deviation trajectories in an uncertainty set that encodes applicable EU legislation. Faithfully modeling the energy conversion losses during battery charging and discharging renders this optimization problem non-convex. By exploiting a total unimodularity property of the proposed uncertainty sets and an exact linear decision rule reformulation, we prove that this non-convex robust optimization problem in continuous time is equivalent to a tractable linear program. Through extensive numerical experiments based on real-world data we investigate how the value of vehicle-to-grid depends on the penalties for non-delivery of promised regulation power, the delivery guarantees, and the vehicle’s battery, charger, and yearly mileage.
**Keywords:** Vehicle-to-Grid, Frequency Regulation, Energy Storage, Energy Economics, Robust Optimization, Continuous-Time Linear Programming.
author:
- 'Dirk Lauinger[^1]'
- François Vuille
- Daniel Kuhn
bibliography:
- 'bibliography.bib'
title: |
Reliable Frequency Regulation through Vehicle-to-Grid:\
From EU Legislation to Robust Optimization
---
[^1]: Corresponding author: [`[email protected]`](mailto:[email protected])
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Based on a recently introduced by the author notion of [*parity*]{}, in the present paper we construct a sequence of invariants (indexed by natural numbers $m$) of long virtual knots, valued in certain simply-defined group ${\tilde G}_{m}$ (the Cayley graphs of these groups are represented by grids in the $(m+1)$-space); the conjugacy classes of elements of $G_{m}$ play the role of invariants of [*compact*]{} virtual knots. By construction, all invariants do not change under [*virtualization*]{}. Factoring the group algebra of the corresponding group by certain polynomial relations leads to finite order invariants of (long) knots which do not change under [*virtualization*]{}.'
author:
- 'V.O.Manturov[^1]'
title: 'Free Knots, Groups, and Finite-Type Invariants'
---
Introduction
============
Virtual knot theory was invented by Kauffman, [@KaV]; virtual knots correspond to knots in thickened surfaces $S_{g}\times
S^{1}$, considered up to isotopies and stabilizations.
The theory of finite type invariants of classical knots was invented independently by V.A.Vassiliev [@Vas] and M.N.Goussarov [@Gus], and it turned out that many well-known invariants are expressible in terms of finite-type invariants [@BL; @BN]; a breakthrough in Vassiliev’s theory of finite type invariants was marked by the celebrated paper by Kontsevich [@Kon], where the structure of the space of finite-type Vassiliev invariant was explicitely described. The Vassiliev knot invariants initiated the study of virtual knots by Goussarov, Polyak, and Viro [@GPV]. In the latter paper, the authors gave a definition of virtual knots equivalent to the original one due to Kauffman; this allowed one to classify invariants of finite type of [*classical and virtual knots*]{} in terms of combinatorial formulae, though their definition of finite type invariants of virtual knots was quite limited. In the present paper, we shall use a more general (and more natural) definition of finite-type (Vassiliev) invariants of virtual knots due to Kauffman [@KaV].
In the case of knots in thickened surfaces (without stabilization), an analogue of the Kontsevich theorem was obtained in [@AM]; however this solution of the classification problem (an explicit universal formula) does not look as elegant as in the classical case; in any case, this formula is quite far away from practical implementation.
Virtual knots are much more complicated objects than classical ones: all invariants of classical knots of order zero are constants, whence the space of finite type invariants of order zero is infinite-dimensional. The latter statement can be reformulated in the following manner: there are infinitely many types of virtual knots where two virtual knots represent the same type whenever one of them can be obtained from the other by a sequence of (equivalences and) crossing changes. These “types” (or “equivalence classes”) are called [*flat virtual knots*]{} (see [@HK]); representing themselves the dual space of the space of invariants of order $0$ they are important for investigation of invariants of finite type.
A thorough simplification for the notion of flat virtual knot is the notion of [*free knot*]{}. In [@Tu], V.G.Turaev (who first introduced free knots under the name of “homotopy classes of Gauss words”) conjectured that all free virtual knots were trivial; this conjecture was first disproved by the author in [@Ma], and later, in [@Gib].
The aim of the present work is to construct invariants of free knots valued in certain groups $G_{m}$ (depending on a certain natural parameter $m$), see [@MM; @MM2], and to extend this construction to some simply defined infinite series of finite type invariants for (long) virtual knots. The groups $G_{m}$ are defined by generators and relators, and have very simple Cayley graphs; the group $G_{1}$ is isomorphic to the infinite dihedral group.
I am very grateful to O.V.Manturov for fruitful consultations.
Basic Definitions
=================
Throughout the paper, by [*graph*]{} we mean a finite multi-graph (loops and multiple edges are allowed). Here a $4$-valent graph is called [*framed*]{} if for each vertex of it, the four emanating half-edges are split into two pairs; half-edges belonging to the same pair are called (formally) [*opposite*]{}. Assume a framed $4$-valent graph $K$ is given by its Gaussian diagram $C(K)$. We shall use the generic term “$4$-graph” to denote a topological object obtained from a four-valent graph by adding free circles (without vertices) as connected components.
For framed $4$-graphs one naturally defines unicursal components: if a graph has connected components homeomeorphic to the circle, they are treated as unicursal components; for the rest of the graph, unicursal components are defined as equivalence classes of edges generated by “local opposite” relation at vertices. This relation of unicursal components naturally yields the [*number of uncursal components*]{} which agrees with the number of components of a link diagram drawn on the plane. Certainly, a framed $4$-graph can be encoded by a chord diagram if and only if it can be has one unicursal component.
By a [*chord diagram*]{} we mean a pair $(S_{0}\sqcup \cdots \sqcup
S_{0}\subset S^{1})$ consisting of an oriented circle (called [*the core circle)*]{} and a set of $n$ unordered pairs of points (all $2n$ points are pairwise distinct). These pairs are called [*chords*]{}; points of the pairs are called [*ends of chords*]{}.
In figures, we shall depict chords by solid lines connecting pairs of points on the core circle. We say that a chord diagram is [*labeled*]{} if it is marked by a point on the core circle distinct from chord ends. We consider chord diagrams up to a natural equivalence (orientation-preseriving homeomorphism of core circle which respects the collection of chords).
Framed $4$-graphs with one unicursal component are in one-one correspondence with chord diagrams. The $4$-graph without vertices corresponds to the chord diagram without chords. Every framed $4$-valent graph can be represented as the image of the circle going transversely through all edges. The two preimages of every vertex correspond to a chord.
One can naturally consider [*oriented*]{} or [*non-oriented*]{} chord diagrams (framed $4$-graphs).
All statements about chord diagrams can be translated into the language of framed $4$-graphs and vice versa.
By a [*free knot*]{} [@Ma] is meant an equivalence class of chord diagrams by the equivalence relation generated by the following three elementary equivalences. The first Reidemeister move corresponds to an addition/removal of a solitary chord. The second Reidemeister move is an addition/removal of a couple of [*similar*]{} chords. Here chords $a$ and $b$ are called similar if their chord ends $a_{1},a_{2}$ and $b_{1},b_{2}$ can be numbered in a way such that $a_{1}$ and $b_{1}$ are adjacent and $a_{2}$ and $b_{2}$ are adjacent (here chord ends $x,y$ are [*adjacent*]{} if one of the two components of the complement $C\backslash\{x,y\}$ has no chord ends inside it).
The third Reidemeister move is depicted in Fig. \[sootvtt\].
![The third Reidemeister move and the correspondence between crossings[]{data-label="sootvtt"}](sootvtt.eps){width="200pt"}
Here the only changing part of the chord diagram consists of three segments with two chord ends on each. The chords belonging to the “stable” part of the chord diagram are not drawn not depicted, and the part of the circle containing only stable chord ends is depicted by dotted lines; in the remaining part, every chord “moves” each of its two ends from one position to the other.
All the three Reidemeister moves on chord diagram originate from the Reidemeister moves for virtual knots [@KaV], defined later in the present paper.
Every Reidemeister move transforms one fragment of the frame four-graph. When depicting such a fragment in a figure we show only the “changing part”, leaving the rest of the diagram outside. In the case of one unicursal component this corresponds to a rule for transforming a Gauss diagram; this transformation deals with some arcs of the Gauss diagrams. When depicting on the plane, we do not show “stable chords”; we use dotted line for depicting “stable parts” of the core circle.
By a [*long free knot*]{} we mean an equivalence class of based chord diagrams by the same Reidemeister move; here we only require that the marked point lies outside the transformation domain (equivalently, this marked point lies on a dotted part of the core circle), see Fig.\[lng\].
![A long free knot[]{data-label="lng"}](longfree.eps){width="200pt"}
The Groups $G_{m}$. Even and Odd Chords
=======================================
Fix a positive integer $m$. Consider the group $G_{m}=\langle
a'_{0},a''_{0},\dots,
a'_{m-1},a''_{m-1},a_{m}|(a_{m})^{2}=e,(a'_{i})^{2}=e,(a''_{i})^{2}=e,i=0,\dots,
m\!-\!1,
a'_{i}a'_{j}=a'_{j}a''_{i},i<j;a'_{i}a''_{j}=a''_{j}a''_{i},i<j;
a'_{i}a_{m}=a_{m}a''_{i} \rangle$.
With every framed chord diagram $D$ we associate an element $G_{m}$. For this sake, we associate with each chord its [*index and type*]{} in the following way.
Let $f$ be the operation on chord diagrams deleting all odd chords. It is proved in [@Ma] that this operation is well defined: if two chord diagrams $D,D'$ represent equivalent free knots, then so are $f(D),f(D')$.
We say that a chord diagram $D$ a chord has index $0$ if this chord [*odd*]{}; those [*even chords*]{} of $D$ which become odd in $f(D)$, are decreed to have index $1$, inductively, for $m-1$ we define those chords of $D$ which remain in $f^{m-1}(D)$ and become odd ones in $f^{m-1}(D)$, to have index $m-1$; all the remaining chords are said to have index $m$.
Now, for chords of index $m$, we do not define the type; for chords of index $k<m$ we define the type ([*the first type*]{} or [*the second type*]{}) depending on the fact whether the number of chords of the diagram $f^{k}(D)$, linked with the given chord, is [*even*]{} (in $f^{k}(D)$) or odd.
Let us take a chord diagram $D$ and let us start walking from the base point of the diagram $D$ along the orientation of the circle. Whenever we meet a chord end of index $i$ and type $a$, we write down the generator of the group $G$ with index $i$ and, if index is less than $m$, with the number of primes equal to $a$. If the index is equal to $m$ then we write down the generator $a_{m}$. When we make a full turn returning to the base point, we get a certain word $\gamma(D)$. Denote the corresponding element of the group $G$ by $[\gamma(D)]$, see Fig. \[primer\].
![Constructing $\gamma(D)$ from a chord diagram[]{data-label="primer"}](primer.eps){width="200pt"}
In [@MM; @MM2], the following statements are proved
1\. $[\gamma(D)]$ is an invariant of long free knots. 2. The conjugacy class of the element $[\gamma(D)]$ is an invariant of free knots.
The proof of this theorem follows from a straightforward check.
The Cayley graph of $G_{m}$
---------------------------
Elements of $G_{m}$ are in one-one correspondence with integer points in the Euclidean space ${\bf R}^{m+1}$ with last coordinate equal to zero or $1$. The unit of the group is represented by the origin of coordinates. With an element of the group, we associate a point in ${\bf R}^{m+1}$ defined by induction, as follows. Assume for some elements $g\in G_{m}$ represented by a word in generators, we have already defined the corresponding point in the Euclidean space; let us define those points corresponding to $g\alpha$, where $\alpha$ is the generator of $G_{m}$. The right multiplication by the generator with lower index $k$ ($a'_{k}$ or $a''_{k}$ or $a_{k}$ for $k=m$) correspond to the shift of the $(k+1)$-th coordinate. The direction of the shift is defined as follows: for $0\le k\le m-1$ if the sum of coordinates of the initial point (all except the first $k$) is even then the multiplication by $a'_{k}$ increases the first coordinate, whence the multiplication by $a''_{k}$ decreases it; if this sum of all coordinates except the first $k$ is odd, then $a'_{k}$ and $a''_{k}$ change their roles; the multiplication by $a_{m}$ changes the last $(m+1)$-st coordinate: from zero to one and from one to zero. It can be readily checked that this correspondence is well defined, i.e., the resulting element of the group $G_{m}$ does not depend on the way of representing an element of the group as a word in generators.
It can be easily checked that the last coordinated of the element of $G_{m}$ corresponding to a word coming from a chord diagram, is zero. The cojugacy class of the element of $G_{m}$ having coordinates $(x_{0},\dots, x_{m-1})\in \Z^{m}$ (with the last coordinate equal to zero) consists of elements with coordinates $(\pm x_{0},\dots, \pm x_{m-1})$ having the last coordinate equal to zero.
The Cayley graphs of $G_{1}$ and $G_{2}$ are given in Fig. \[g2\].
![The Cayley graphs for $G_1$ and $G_2$[]{data-label="g2"}](grids.eps){width="350pt"}
Virtual knots
-------------
In the present paper, virtual knots appear as a generalization of free knots with special decorations at vertices of the framed $4$-graph (resp., chords of the chord diagram). We shall give a definition in terms of planar diagrams and Reidemeister moves, the corresponding moves on chord diagram with decorations (called [*Gauss diagrams*]{}) can be written straightforwardly.
A [*virtual diagram*]{}, an example see in Fig. \[virt\], is a generic planar immersion of a four-valent graph, where every image of a vertex is marked as a classical crossing (one pair of opposite edges is said to form an overcrossing, and the other pair is said to form an undercrossing); when depicting on the plane, undercrossings are marked by a broken line; we also encircle the intersection points between images of edges and say that they from [*virtual crossings*]{}. We also admit the case when some components of the virtual diagram are not graphs, but circles without vertices. For this sake, we slightly change the notation and will say that a virtual diagram is a generic immersion of a $4$-graph in $\R^{2}$, where a $4$-graph means a disjoint sum of a $4$-valent graph with a collection of circles.
![A virtual diagram[]{data-label="virt"}](virtknot.eps){width="200pt"}
Unlike a framed $4$-graph, in a virtual diagram, we have got two new structures at classical crossings: some pair of opposite edges are said to form an [*overcrossing*]{}, the remaining pair forms an [*undercrossing*]{}; besides, the four edges incident to classical crossing obtain a [*counterclockwise ordering*]{}: being immersed in the plane, we know not only which edges are opposite, but also which non-opposite edge follows after the given one in the counterclockwise direction.
A [*virtual link*]{} is an equivalence class of virtual diagram modulo three classical Reidemeister moves and the detour move, see [@Mybook]. The Reidemeister moves are applied to fragments of the diagram containing only classical crossings. The detour move deals with an arc containing only virtual crossings and virtual self-crossings. This move replaces such an arc with another (possibliy, self-intersecting) arc having the same endpoints; all new crossings and self-crossings are marked as virtual ones.
For a virtual diagram, one naturally defines [*unicursal components*]{}. The immersion image of the $4$-graph is again a $4$-graph; the latter consists of a $4$-valent graph and a collection of circles disjoint from the rest of the graph. Every circle is treated as a unicursal component; other unicursal components are equivalence classes of edges of $4$-graph components, where the equivalence is generated by opposite edge relation. This agrees with the term “component” for a link diagram drawn on the plane. Certainly, the number of unicursal components remains unchanged under Reidemeister moves.
A [*virtual knot*]{} is a one-component virtual link. Analogously, one defines [*long virtual knots*]{}. In this case, to define [*long virtual diagrams*]{}, instead of framed $4$-graph, one has to consider the result of breaking a framed $4$-graph at some edge midpoint; this will result in two vertices of valency $1$. All vertices of valency $4$ inherit the framing (the half-edges incident to a four-valent vertex are split into two pairs of formally opposite edges); one may also say that the newborn vertices of valency $1$ are mapped to $\infty$ and $+\infty$ and the intersection of the image of the graph with the exterior of some large circle coincides with $Ox$, and the graph is oriented from $-\infty$ to $+\infty$ along its unicursal component.
A [*long virtual knot*]{} is an equivalence class of long virtual diagrams by the same Reidemeister moves. It is allowed to apply these Reidemeister moves only inside some prefixed large circle.
Virtual Knots and Gauss Diagrams
--------------------------------
With a virtual knot $K$ diagram one naturally associates (see [@GPV]) a [*Gauss diagram*]{} $\Gamma(K)$, with all chords endowed with arrows and signs; chords of the Gauss diagrams are in one-one corresondence with [*classical*]{} crossings of the diagram; the arrow is pointed from the preimage of the overcrossing arc to the preimage of the undercrossing arc; the sign of the crossing locally looking like ${\raisebox{-0.25\height}{\includegraphics[width=0.5cm]{skcrro.eps}}}$ is positive, and the sign of a crossing locally looking like ${\raisebox{-0.25\height}{\includegraphics[width=0.5cm]{skcrlo.eps}}}$ is negative. With a (long) virtual knot diagram $K$ one associates a (long) free knot whose chord diagram is obtained from $\Gamma(K)$ by forgetting arrows and signs. Note that the detour move does not change the Gauss diagram corresponding to a virtual diagram. By construction, Reidemeister moves on virtual knots generate the equivalence relations on free knots described above also called [*Reidemeister moves*]{}.
The Gauss diagram of classical and virtual trefoils are given below.
![The Right Trefoil and Its Gauss Diagram[]{data-label="Gaussdiagram"}](gd1.eps){width="200pt"}
![The Virtual Trefoil and Its Gauss Diagram[]{data-label="VGaussdiagram"}](vgd2.eps){width="200pt"}
By [*virtualization move*]{} for a classical crossing we mean the local transformation of a (virtual) diagram, which inverts the arrow direction for the arrow corresponding to this crossing, and preserves the sign of the crossings (in the level of Gauss diagrams). This map is well defined up to detour moves. In the language of virtual diagrams the virtualization move looks as shown in Fig. \[virtua\].
![The virtualization move[]{data-label="virtua"}](virtua.eps){width="120pt"}
In [@Chr], Chrisman proved that Goussarov-Poylak Viro combinatorial formulae for virtual knots do not yield invariants preserved by virtualization. The invariants of virtual knots constructed in the present paper are all invariant under virtualization.
Vassiliev (finite-type) invariants. The main result
===================================================
Let $h$ be an invariant of (long) virtual knots. We say that $h$ is a [*finite-type (Vassiliev)*]{} invariant of order at most $k$, if for every virtual knot $K$ with $k+1$ fixed classical crossings the following takes place $\sum_{s} (-1)^{\#(s)} h(K_{s})=0$, where the sum is taken over all possible $2^{k+1}$ knots $K_{s}$ coiniciding with $K$ everywhere except neighborhoods of given crossings, and at the fixed $k+1$ places, the crossings ${\raisebox{-0.25\height}{\includegraphics[width=0.5cm]{skcrro.eps}}}$ or ${\raisebox{-0.25\height}{\includegraphics[width=0.5cm]{skcrlo.eps}}}$ are chosen arbitrarily. Here the symbol $\#(s)$ stays for the number of crossings of type ${\raisebox{-0.25\height}{\includegraphics[width=0.5cm]{skcrlo.eps}}}$ among the chosen ones.
The other way of defining finite-type invariants uses [*singular knots*]{} and [*rigid vertices*]{}, for more details see, [@KaV]. A rigid vertex ${\raisebox{-0.25\height}{\includegraphics[width=0.5cm]{skcross.eps}}}$ is a formal linear combination ${\raisebox{-0.25\height}{\includegraphics[width=0.5cm]{skcross.eps}}}={\raisebox{-0.25\height}{\includegraphics[width=0.5cm]{skcrro.eps}}}-{\raisebox{-0.25\height}{\includegraphics[width=0.5cm]{skcrlo.eps}}}$. A [*singular knot*]{} of order $k$ is a formal linear combination of $2^{k}$ knots which appear as resolutions of $k$ formal singular crossings (rigid vertices). Every knot invariant naturally extends to linear combinations of knots, hence, to singular knots. An invariant is of finite type at most $k$ whenever its extension to singular knots of order $k+1$ vanishes.
Our next goal is to use a wider group than $G_{m}$ to construct invariants of [*virtual knots*]{} (with some over/undercrossing information). This goal can be achieved by extending the group $G_{m}$.
Consider the group ${\tilde G}_{m}=\langle a'_{0},a''_{0},\dots,
a'_{m-1},a''_{m-1},a_{m}|a_{m}^{2}=e,(a'_{i})^{2}=(a''_{i})^{2},
i=0,\dots, m\!-\!1, a'_{i}a'_{j}=a'_{j}a''_{i},
a'_{i}a''_{j}=a''_{j}a''_{i},
a'_{j}a'_{i}=a''_{i}a'_{j},a''_{j}a'_{i}=a''_{i}a''_{j}, i<j;
a'_{i}a_{m}=a_{m}a''_{i}, i=0,\dots, m-1 \rangle$.
Let us fix a natural number $m$, and write $G=G_{m}$. It is clear that the group $G$ defined above is obtained from ${\tilde G}_{m}$ by adding the following defining relations: $(a'_{i})^{2}=1,i=0,\dots, m-1$. Let ${\cal G}=\Q{\tilde G}_{m}$ be the group algebra of${\cal G}$, and let ${\cal G}_{k}$ denote the quotient algebra of ${\cal G}$ by the following relations $\prod_{j=1}^{k+1} ((a'_{n_{j}})^{2}-1)=0$, where $n_{j}$ stays for any arbitrary set of numbers from $0$ to $m-1$. It is clear that ${\cal G}_{0}=\Q G$.
From the relations for the group $G$, it easily follows that every element of ${\cal G}_{k}$ looking like $\alpha_{1}\cdot A_{1}\cdot
\alpha_{2} A_{2}\cdots \alpha_{k+1} A_{k+1}\cdot \alpha_{k+2}$ for arbitrary $\alpha$ and $A_{j}=(u-u^{-1})$, where $u$ stays for a generator of the group $G$, is [*equal to zero*]{}. Indeed, it suffices to note that the commutation relations for $u$ and $u^{-1}$ are similar, e.g. for $u=a'_{1}$ we have $a'_{i}u=a''_{1}$ and $a'_{i}u^{-1}=(a''_{1})^{-1}$ for $i>1$. So, $(u-u^{-1})$ in this case transforms into $a''_{1}-(a''_{1})^{-1}$. So, we can collect all expressions of type $(u-u^{-1})$ together, and get $0$.
Let $K$ be a long virtual knot, (see [@MaLong]). With it, we associate an element $\delta(K)\in {\tilde G}\subset {\cal G}$ as follows. Take a Gauss diagram $\Gamma(K)$ of $K$, and start writing a word in generators of the group ${\tilde G}$ and its inverses exactly in the way we were writing down the word $\gamma(D)$, with the only difference that instead of each generator $a'_{j}$ or $a''_{j}, j=0,\dots, m-1$ we shall write down either this generator or its inverse depending on whether the crossing in question is positive $({\raisebox{-0.25\height}{\includegraphics[width=0.5cm]{skcrro.eps}}})$ or negative $({\raisebox{-0.25\height}{\includegraphics[width=0.5cm]{skcrlo.eps}}})$. Denote the obtained word by $\delta(K)$. Note that the word $\delta(K)$ by definition does not depend on the direction of arrows in the Gauss diagram of $K$.
The main result of the present paper is the following
The element $\delta(K)$ of the group $\tilde G$ is an invariant of long virtual knots, which does not change under virtualization (change of arrow direction on the Gauss diagram).
The conjugacy class of the element $\delta(K)$ in $\tilde G$ is an invariant of (compact) virtual knot.
For every $k$, the map $K\to \delta(K)\in {\cal G}_{k}$ is a Vassiliev invariant of long virtual knots of order less than or equal to $k$.
The first statement of the theorem easily follows from a comparison of Reidemeister moves for virtual knots and the relations in the group ${\tilde G}$; by construction, the invariant $\delta$ does not depend on the arrow direction on the Gauss diagram: the latter statement means the invariance under virutalization (see [@FKM]). When passing from long virtual knots to compact virtual knots, we allow the marked point to go along the Gauss diagram (or, equivalently, we allow the infinity change); when moving the marked point through a crossing we conjugate the correspondent element of ${\tilde G}$, which yields the second statement of the theorem.
Now, consider the alternating difference of the values of $\delta$ for all long virtual knots corresponding to $k+1$ choices crossing types ${\raisebox{-0.25\height}{\includegraphics[width=0.5cm]{skcrro.eps}}}$ and ${\raisebox{-0.25\height}{\includegraphics[width=0.5cm]{skcrlo.eps}}}$ (or, equivalently, the value of $\delta$ on the singular knot of order $k+1$). By construction of $\delta$, each singular crossing will contribute a factor of one of the following types: $(a'_{i}-(a'_{i})^{-1})$ or $(a''_{i}-(a''_{i})^{-1})$ or $(a_{m}-(a_{m})^{-1})$; the latter factor is zero. Taking into account the factorization relation defining ${\tilde G}_{k}$, and the fact that the number of factors in our product is equal to $k+1$, we conclude that the desired alternating sum in ${\tilde G}_{k}$ is equal to zero.
The invariants described in the present paper are constructive. The group ${\tilde G}_{m}$ admits a simple description similar to that of the group $G$. The Cayley graph of the group ${\tilde G}_{m}$ is the integer grid in $\R^{2n+1}$; all coordinates of the vertices of this graph are arbitrary integer numbers, except the last one, which is equal to either zero or one. The right multiplication by $a_{m}$ changes the last coordinate, whence the right multiplication by $a'_{i}$ or $a''_{i}$ corresponds to a shift in positive direction along one of coordinates $x_{2i+1}$ or $x_{2i+2}$; here the choice which coordinate changes depends on the parity of sum of all coordinates from $2i+1$ to $m$.
Values of the Invariants and Further Discussion
===============================================
First we note that even for free knots and even for the group $G_{1}$, the values of the invariant $[\gamma]$ are very interesting. In Fig. \[exs\], we give two knots where the two coordinates of these values are equal to $16$ and $8$, respectively.
![Two non-slice free knots[]{data-label="exs"}](exs.eps){width="200pt"}
The knots above are not [*slice*]{} in the sense of paper [@Ma4]. In fact, the invariant $[\gamma]$ provides a sliceness obstruction for free knots, as shown in [@Ma4].
It is not a difficult exercise to show that the $L(K)$ is divisible by $4$ for every free knot. The divisibility of this invariant is conjectured by O.V.Manturov and will be proved elsewhere.
So, one can see that even the invariant of free knots and even for the case of the group $G_{1}$ is highly non-trivial. Certainly, so are invariants of long virtual knots.
In the present paper, we have constructed finite-type invariants of [*long*]{} virtual knots. For compact virtual knots, this can be done as well, but not in that elegant manner: one cannot further deal with specific elements of ${\tilde G}_{m}$, but should rather take conjugacy classes; the problem of extracting conjugacy classes from one another and taking alternating sum looks cumbersome. We shall touch on this problem in a separate publication.
[100]{}
J.E. Andersen and J. Mattes, Configuration space integrals and universal Vassiliev invariants over closed surfaces, arXiv:q-alg/9704019.
J.Birman, X-S.Lin (1993), Knot Polynomials and Vassiliev’s Invariants, [*Inventiones Mathematicae*]{}, [**111**]{}, P. 225-270.
D.Bar-Natan, On the Vassiliev Knot Invariants, (2005), [*Topology*]{}, [**34**]{}, pp. 423-475.
M.Chrisman, Twist Lattices and the Jones-Kauffman Polynomial for Long Virtual Knots. [*J. Knot Theory & Ramif.*]{}, to appear.
R.A.Fenn, L.H.Kauffman, V.O.Manturov (2005), Virtual Knots: Unsolved Problems, [*Fund. Math.*]{},[**188**]{}, pp. 293-323.
A.Gibson, Homotopy invariants of Gauss words, ArXiv: Math.GT/0902.0062
Goussarov M., Polyak M., and Viro O (2000), Finite type invariants of classical and virtual knots, [*Topology*]{}. [**39**]{}. P. 1045–1068.
M.N. Goussarov (1991), Novaya forma polinoma Jones’a-Conway’a dlya orientirovannyh zacepleniy (A new form of the Jones-Conway polynomial for oriented links), [*Zap. Nauchn. Seminarov LOMI.*]{} [**193**]{}, Geometry and Topology, 1. P. 4-9.
D.Hrencecin, L.H.Kauffman (2003), On filamentations and virtual knots, [*Topology Appl.*]{} [**134**]{}, pp. 23-52.
L.H. Kauffman (1999), Virtual Knot Theory, [*Eur. J. Combinatorics*]{} [**20**]{} (7), P. 662–690.
M.Kontsevich (1993), Vassiliev’s Knot Invariants, [*Adv. Sov. Math.*]{}, [**16**]{} (2), P. 137-150.
V.O.Manturov (2010), [*Parity in Knot Theory*]{}, Sbornik Mathematics., [**201**]{} (5), pp. 65-110.
V.O.Manturov (2005), [*On Long Virtual Knots*]{}, Doklady Mathematics, [**141**]{} (5), С. 195-198.
V.O.Manturov (2003), [*Knot Theory*]{}, Chapman and Hall/CRC., 416 pp.
V.O.Manturov (2010), [*Parity and Cobordisms of Free Knots*]{}, arXiv.Math/GT:1001.2827
O.V.Manturov, V.O.Manturov, Free Knots and Groups (2010), [*Journal of Knot Theory and Its Ramifications*]{}, [**19**]{}, (2)
O.V.Manturov, V.O.Manturov, Svobodnye Uzly i Gruppy (Free Knots and Groups), [*Doklady Mathematics*]{}, to appear.
V.G.Turaev,Topology of words, [*Proc. Lond. Math. Soc.*]{} (3) 95 (2007), no. 2, С. 360–412.
V.A.Vassiliev (1990), Cohomology of Knot Spaces, In: Theory of Singularities and Its Applications, [*Adv. Sov. Math.*]{}, [**1**]{} (23), P. 23-70
[^1]: vomanturov at yandex.ru
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
This paper provides the basis for new methods of inference for max-stable processes $\xi$ on general spaces that admit a certain incremental representation, which, in important cases, has a much simpler structure than the max-stable process itself. A corresponding peaks-over-threshold approach will incorporate all single events that are extreme in some sense and will therefore rely on a substantially larger amount of data in comparison to estimation procedures based on block maxima.\
Conditioning a process $\eta$ in the max-domain of attraction of $\xi$ on being *extremal*, several convergence results for the increments of $\eta$ are proved. In a similar way, the shape functions of mixed moving maxima (M3) processes can be extracted from suitably conditioned single events $\eta$. Connecting the two approaches, transformation formulae for processes that admit both an incremental and an M3 representation are identified.
bibliography:
- 'HREstimation.bib'
title: |
Representations of max-stable processes\
based on single extreme events
---
Introduction
============
The joint extremal behavior at multiple locations of some random process $\{\eta(t): t\in T\}$, $T$ an arbitrary index set, can be captured via its limiting *max-stable process*, assuming the latter exists and is non-trivial everywhere. Then, for independent copies $\eta_i$ of $\eta$, $i\in{\mathbb N}$, the functions $b_n: T \to {\mathbb R}$, $c_n : T\to (0,\infty)$ can be chosen such that the convergence $$\begin{aligned}
\label{MDA}
\xi(t) = \lim_{n\to\infty} c_n(t) \Big(\max_{i=1}^n \eta_i(t) - b_n(t)\Big),
\quad t\in T,\end{aligned}$$ holds in the sense of finite-dimensional distributions. The process $\xi$ is said to be *max-stable* and $\eta$ is in its max-domain of attraction (MDA). The theory of max-stable processes is mainly concerned with the dependence structure while the marginals are usually assumed to be known. Even for finite-dimensional max-stable distributions, the space of possible dependence structures is uncountably infinite-dimensional and parametric models are required to find a balance between flexibility and analytical tractability [@deh2006a; @res2008].
A general construction principle for max-stable processes was provided by [@deh1984; @smi1990]: Let $\sum_{i\in{\mathbb N}} \delta_{(U_i, S_i)}$ be a Poisson point process (PPP) on $(0,\infty)\times{\mathcal S}$ with intensity measure $u^{-2}\rd u\cdot \nu(\rd s)$, where $({\mathcal S},
\mathfrak S)$ is an arbitrary measurable space and $\nu$ a positive measure on ${\mathcal S}$. Further, let $f:{\mathcal S}\times T \to [0, \infty)$ be a non-negative function with $\int_{{\mathcal S}} f(s,t) \nu(\rd s) = 1$ for all $t\in T$. Then the process $$\begin{aligned}
\xi(t) = \max_{i\in{\mathbb N}} U_i f(S_i, t), \quad t\in T,\label{constr_max_stable}\end{aligned}$$ is max-stable and has standard Fréchet margins with distribution function $\exp(-1/x)$ for $x \geq 0$. In this paper, we restrict to two specific choices for $f$ and $({\mathcal S}, \mathfrak S, \nu)$ and consider processes that admit one of the resulting representations. First, let $\{W(t) : t\in T\}$ be a non-negative stochastic process with $\sE W(t) = 1$, $t\in T$, and $W(t_0) = 1$ a.s. for some point $t_0 \in T$. The latter condition means that $W(t)$ simply describes the multiplicative increment of $W$ w.r.t. the location $t_0$. For $({\mathcal S}, \mathfrak S, \nu)$ being the canonical probability space for the sample paths of $W$ and with $f(w,t)=w(t)$, $w\in{\mathcal S}$, $t\in T$, we refer to $$\begin{aligned}
\label{def_xi}
\xi(t) = \max_{i\in{\mathbb N}} U_i W_i(t), \quad t\in T,\end{aligned}$$ as the *incremental representation* of $\xi$, where $\{W_i\}_{i\in{\mathbb N}}$ are independent copies of $W$. Since $T$ is an arbitrary index set, the above definition covers multivariate extreme value distributions, i.e. $T=\{t_1,\dots,t_k\}$, as well as max-stable random fields, i.e. $T = {\mathbb R}^d$.\
For the second specification, let $\{F(t): \ t \in {\mathbb R}^d\}$ be a stochastic process with sample paths in the space $C({\mathbb R}^d)$ of non-negative continuous functions, such that $$\begin{aligned}
\label{assumption_integral}
\textstyle \sE \int_{{\mathbb R}^d} F(t) \rd t = 1.\end{aligned}$$ With $S_i = (T_i,F_i)$, $i\in{\mathbb N}$, in ${\mathcal S}= {\mathbb R}^d\times C({\mathbb R}^d)$, intensity measure $\nu(\rd t \times \rd g)=\rd t\sP_F(\rd g)$ and $f((t,g), s)=g(s-t)$, $(t,g)\in{\mathcal S}$, we obtain the class of *mixed moving maxima (M3) processes* $$\begin{aligned}
\xi(t) = \max_{i\in{\mathbb N}} U_i F_i(t- T_i), \quad t\in{\mathbb R}^d. \label{def_M3}\end{aligned}$$ These processes are max-stable and stationary on ${\mathbb R}^d$ (see for instance [@wan2010]). The function $F$ is called *shape function of $\xi$* and can also be deterministic (e.g., in case of the Smith process). In Smith’s “rainfall-storm” interpretation [@smi1990], $U_i$ and $T_i$ are the strength and center point of the $i$th storm, respectively, and $U_i F_i(t- T_i)$ represents the corresponding amount of rainfall at location $t$. In this case, $\xi(t)$ is the process of extremal precipitation.
When i.i.d. realizations $\eta_1, \ldots, \eta_n$ of $\eta$ in the MDA of a max-stable process $\xi$ are observed, a classical approach for parametric inference on $\xi$ is based on generating (approximate) realizations of $\xi$ out of the data $\eta_1, \ldots, \eta_n$ via componentwise block maxima and applying maximum likelihood (ML) estimation afterwards. A clear drawback of this method is that it ignores all information on large values that is contained in the order statistics below the within-block maximum. Further, ML estimation needs to evaluate the multivariate densities while for many max-stable models only the bivariate densities are known in closed form. Thus, composite likelihood approaches have been proposed [@pad2010; @dav2012].\
In univariate extreme-value theory, the second standard procedure estimates parameters by fitting a certain PPP to the *peaks-over-thresholds* (POT), i.e., to the empirical process of exceedances over a certain critical value [@lea1991; @emb1997]. Also in the multivariate framework we can expect to profit from using all extremal data via generalized POT methods instead of aggregated data. In contrast to the ML approach, in this paper, we assume that $\xi$ admits one of the two representations and and we aim at extracting realizations of the processes $W$ and $F$, respectively, from *single extreme events*. Here, the specification of a single extreme event will depend on the respective representation.\
In [@eng2012a], this concept is applied to derive estimators for the class of Brown-Resnick processes [@bro1977; @kab2009], which have the form by construction. With $a(n)$ being a sequence of positive numbers with $\lim_{n\to\infty} a(n) = \infty$, the convergence in distribution $$\begin{aligned}
\Bigg( \frac{\eta(t_1)}{\eta(t_0)}, \ldots,
\frac{\eta(t_k)}{\eta(t_0)}
\ \Bigg|\
\eta(t_0) > a(n) \Bigg)\cvgdist \bigl(
W(t_1),\dots, W(t_k) \bigr),
\label{cond_incr_conv}\end{aligned}$$ $t_0,t_1,\dots,t_k\in T$, $k\in {\mathbb N}$, is established for $\eta$ being in the MDA of a Brown-Resnick process and with $W$ being the corresponding log-Gaussian random field. A similar approach exists in the theory of homogeneous discrete-time Markov chains. For instance, [@seg2007] and [@ehl2011] investigate the behavior of a Markov chain $\{M(t): t\in {\mathbb Z}\}$ conditional on the event that $M(0)$ is large. The resulting extremal process is coined the tail chain and turns out to be Markovian again. In this paper, the convergence result is generalized in different aspects. Arbitrary non-negative processes $\{W(t) : t\in T\}$ with $\sE W(t) = 1$, $t\in T$, are considered, and convergence of the conditional increments of $\eta$ in the sense of finite-dimensional distributions as well as weak convergence in continuous function spaces is shown (Theorems \[theo\_cond\_increments\_general\] and \[theo\_cond\_increments\_cont\]). Moreover, in Section \[M3representation\], similar results are established for M3 processes by considering realizations of $\eta$ around their (local) maxima. Since one and the same max-stable process $\xi$ might admit both representations and we provide formulae for switching between them in Section \[sec:switching\]. Section \[sec:application\] gives an exemplary outlook on how our results can be applied for statistical inference.
Incremental representation {#examples_increment_representation}
==========================
Throughout this section, we suppose that $\{\xi(t): \ t\in T\}$, where $T$ is an arbitrary index set, is normalized to standard Fréchet margins and admits a representation $$\begin{aligned}
\label{def_xi2}
\xi(t) = \max_{i\in{\mathbb N}} U_i V_i(t), \quad t\in T,\end{aligned}$$ where $\sum_{i\in{\mathbb N}}\delta_{U_i}$ is a PPP on $(0,\infty)$ with intensity $u^{-2}du$, which we call *Fréchet point process* in the following. The $\{V_i\}_{i\in{\mathbb N}}$ are independent copies of a non-negative stochastic process $\{V(t): \ t\in T\}$ with $\sE V(t) =
1$, $t\in T$. Note that is slightly less restrictive than the representation in that we do not require that $V(t_0)=1$ a.s. for some $t_0\in T$. For any fixed $t_0\in T$, we have $$\begin{aligned}
\label{decomp_V}
\xi(t) \eqdist \max_{i\in{\mathbb N}} U_i
\left({\mathbf 1}_{P_i=0}V^{(1)}_i(t) + {\mathbf 1}_{P_i=1}V^{(2)}_i(t)\right),
\quad t\in T,\end{aligned}$$ where $\{P_i\}_{i\in{\mathbb N}}$ are i.i.d. Bernoulli variables with parameter $p=\sP(V(t_0) = 0)$ and the $V^{(1)}_i$ and $V^{(2)}_i$ are independent copies of the process $\{V(t): \ t\in T\}$, conditioned on the events $\{V(t_0) > 0\}$ and $\{V(t_0)= 0\}$, respectively.
Note that for $k\in{\mathbb N}$, $t_0, \ldots, t_k\in T$, the vector $\Xi =
(\xi(t_0),\dots,\xi(t_k))$ follows a $(k+1)$-variate extreme-value distribution and its distribution function $G$ can therefore be written as $$\begin{aligned}
\label{def_mu}
G(\mathbf{x}) = \exp( -\mu( [{\mathbf 0},\mathbf{x}]^C) ),
\quad \mathbf{x} \in {\mathbb R}^{k+1},\end{aligned}$$ where $\mu$ is a measure on $E = [0,\infty)^{k+1}\setminus\{{\mathbf 0}\}$, the so-called *exponent measure* of $G$ [@res2008 Prop. 5.8], and $[{\mathbf 0},\mathbf{x}]^C = E\setminus
[{\mathbf 0},\mathbf{x}]$.
The following convergence result provides the theoretical foundation for statistical inference based on the incremental process $V$.
\[theo\_cond\_increments\_general\] Let $\{\eta(t): \ t\in T\}$ be non-negative and in the MDA of some max-stable process $\xi$ that admits a representation and suppose that $\eta$ is normalized such that holds with $c_n(t) = 1/n$ and $b_n(t) = 0$ for $n\in{\mathbb N}$ and $t\in T$. Let $a(n)\to\infty$ as $n\to\infty$. For $k\in {\mathbb N}$ and $t_0,\dots,t_k\in T$ we have the convergence in distribution on ${\mathbb R}^{k+1}$ $$\begin{aligned}
\left(\frac{\eta(t_0)}{a(n)}, \frac{\eta(t_1)}{\eta(t_0)} ,\dots,
\frac{\eta(t_k)}{\eta(t_0)}
\ \Bigg|\
\eta(t_0) > a(n)\right) \cvgdist
\left(Z, \Delta\mathbf{\tilde{V}}^{(1)}\right),\quad n\to \infty,
\end{aligned}$$ where the distribution of $\Delta\mathbf{\tilde{V}}^{(1)}$ is given by $$\begin{aligned}
\sP(\Delta\mathbf{\tilde{V}}^{(1)}\in d \mathbf z) =
(1-p)\sP(\Delta\mathbf V^{(1)}\in d \mathbf z)
\sE\bigl( V^{(1)}(t_0) \big|
\Delta\mathbf V^{(1)}=\mathbf z \bigr), \quad \mathbf{z} \geq {\mathbf 0}.
\label{density_increment}
\end{aligned}$$ Here, $\Delta\mathbf V^{(1)}$ denotes the vector of increments $\left(\frac{V^{(1)}(t_1)}{V^{(1)}(t_0)}, \ldots,
\frac{V^{(1)}(t_k)}{V^{(1)}(t_0)}\right)$ with respect to $t_0$, and $Z$ is an independent Pareto variable.
Note that any process $\eta$ that satisfies the convergence in for a process $\xi$ with standard Fréchet margins can be normalized such that the norming functions in become $c_n(t) = 1/n$ and $b_n(t) = 0$, $n\in{\mathbb N}$, $t\in T$ [@res2008 Prop. 5.10].
For $\mathbf{X} = (\eta(t_0),\dots,\eta(t_k))$, which is in the MDA of the random vector $\Xi=(\xi(t_0),\dots,\xi(t_k))$, it follows from [@res2008 Prop. 5.17] that $$\begin{aligned}
\label{conv_resnick}
\lim_{m\to\infty} m \sP( \mathbf{X}/m \in B ) = \mu(B),\end{aligned}$$ for all elements $B$ of the Borel $\sigma$-algebra $\mathcal B(E)$ of $E$ bounded away from $\{{\mathbf 0}\}$ with $\mu(\partial B)=0$, where $\mu$ is defined by . For $s_0> 0$ and ${\mathbf s}=(s_1, \ldots,
s_k)\in [0, \infty)^{k}$, we consider the sets $A_{s_0}=(s_0,\infty)\times [0, \infty)^k$, $A=A_1$ and $B_{\mathbf{s}} = \{\mathbf{x} \in [0, \infty)^{k+1} :
(x^{(1)},\dots,x^{(k)}) \leq x^{(0)}\mathbf{s}\}$ for ${\mathbf s}$ satisfying $\sP( \Delta\tilde {\mathbf V}^{(1)}\in \partial
[{\mathbf 0},{\mathbf s}])=0$. Then $$\begin{aligned}
\left\{ \eta(t_0) > s_0 a(n),\,
\big( \eta(t_1) / \eta(t_0) ,\dots, \eta(t_k) / \eta(t_0) \big)
\leq \mathbf{s} \right\}
= \{ \mathbf{X} / a(n) \in B_{\mathbf{s}}\cap A_{s_0} \},\end{aligned}$$ since $B_{\mathbf{s}}$ is invariant under multiplication, i.e., $B_{\mathbf s}=cB_{\mathbf s}$ for any $c>0$. Thus, we obtain $$\begin{aligned}
\notag
\sP&\left( \eta(t_0) > s_0 a(n),
\, \left( \eta(t_1) / \eta(t_0) ,\dots,
\eta(t_k) / \eta(t_0) \right) \leq \mathbf{s} \,\Big|\,
\eta(t_0) > a(n) \right) \\
\notag&= \frac{
{a(n)} \sP( \mathbf{X} / a(n)
\in B_{\mathbf{s}} \cap A \cap A_{s_0} )}{
{a(n)} \sP( \mathbf{X} / a(n) \in A)} \\
\label{eq:01}
& \longrightarrow
\frac{\mu(B_{\mathbf{s}} \cap A \cap A_{s_0})}{\mu(A)},\quad (n\to\infty),\end{aligned}$$ where the convergence follows from , as long as $\mu\{ \partial (B_{\mathbf{s}} \cap A \cap A_{s_0})\} = 0$.\
Let $$\begin{aligned}
\label{def_xi3}
\xi^{(1)}(t) = \max_{i\in{\mathbb N}} U_i^{(1)} V^{(1)}_i(t),
\quad t\in T,
\end{aligned}$$ where $\sum_{i\in{\mathbb N}} \delta_{U_i^{(1)}}$ is a Poisson point process with intensity $(1-p)u^{-2}\rd u$ and let $\mu^{(1)}$ be the exponent measure of the associated max-stable random vector $(\xi^{(1)}(t_0), \ldots, \xi^{(1)}(t_k))$. Then the choice $A =
(1,\infty)\times [0,\infty)^k$ guarantees that $\mu(\cdot \cap A) =
\mu^{(1)}(\cdot \cap A)$. Comparing the construction of $\xi^{(1)}$ in with the definition of the exponent measure, we see that $\mu^{(1)}$ is the intensity measure of the Poisson point process $\sum_{i\in{\mathbb N}} \delta_{(U_i^{(1)}
V_i^{(1)}(t_0),\, \ldots,\, U_i^{(1)} V_i^{(1)}(t_k))}$ on $E$. Hence, $$\begin{aligned}
\mu(A) &= \int_0^\infty (1-p)u^{-2} \sP(u V^{(1)}(t_0) > 1) \rd u \notag\\
&= (1-p)\int_0^\infty u^{-2} \int_{[u^{-1}, \infty)}
\sP(V^{(1)}(t_0) \in \rd y) \rd u \notag\\
&= (1-p)\int_0^\infty y \sP(V^{(1)}(t_0) \in \rd y)
= (1-p)\sE V^{(1)}(t_0) = 1,
\end{aligned}$$ where the last equality follows from $\sE V^{(1)}(t_0) = \sE
V(t_0)/(1-p)$. Furthermore, for $s_0\geq 1$ and ${\mathbf s}\in[0,\infty)^k$ with $\sP(\Delta\mathbf{\tilde V}^{(1)} \in \partial [{\mathbf 0},{\mathbf s}])=0$, $$\begin{aligned}
&\mu(B_{\mathbf{s}} \cap A \cap A_{s_0}) / ((1-p)\mu(A)) \notag\\
&= \int_0^\infty u^{-2}
\sP\Bigl(u V^{(1)}(t_0) > s_0,\,
\big(u V^{(1)}(t_1),\dots, u V^{(1)}(t_k)\big) \leq
\mathbf{s} u V^{(1)}(t_0) \Bigr) \rd u\notag\\
&=
\int_0^\infty \int_{[s_0 u^{-1},\, \infty)} u^{-2}
\sP\Bigl(V^{(1)}(t_0)\in \rd y
\Big|\Delta\mathbf V^{(1)} \leq \mathbf{s} \Bigr)
\sP(\Delta\mathbf V^{(1)} \leq \mathbf{s} )\rd u
\notag\\
&=
\int_{[{\mathbf 0}, \mathbf s]} \int_{[0,\infty)} y s_0^{-1} \cdot
\sP\Bigl(V^{(1)}(t_0)\in \rd y \Big|
\Delta\mathbf V^{(1)}=\mathbf z \Bigr)
\sP(\Delta\mathbf V^{(1)}\in \rd{\mathbf z})
\notag\\
&=
s_0^{-1}\int_{[{\mathbf 0}, \mathbf s]} \sE\Bigl( V^{(1)}(t_0) \Big|
\Delta\mathbf V^{(1)}=\mathbf z \Bigr)
\sP(\Delta\mathbf V^{(1)}\in \rd{\mathbf z}).
\label{mu_expl}
\end{aligned}$$ Equation shows that the convergence in holds for all continuity points ${\mathbf s}\in [0, \infty)^{k}$ of the distribution function of $\Delta{\mathbf V}^{(1)}$. Since $s_0\geq 1$ was arbitrary, this concludes the proof.
1. If $V^{(1)}(t_0)$ is stochastically independent of the increments $\Delta\mathbf V^{(1)}$, we simply have $\sP(\Delta\mathbf{\tilde{V}}^{(1)}\in d \mathbf z) =
\sP(\Delta\mathbf{{V}}^{(1)}\in d \mathbf z)$.
2. If $p=\sP(V(t_0) = 0)=0$, the exponent measure $\mu$ of any finite-dimensional vector $\Xi=(\xi(t_0), \ldots, \xi(t_k))$, $t_0,
\ldots, t_k\in T$, $k\in{\mathbb N}$, satisfies the condition $\mu\left( \{0\}\times [0,\infty)^k \right)=0,$ and following Proposition \[calculateW\], the incremental representation of $\Xi$ according to is given by $\Xi = \max_{i\in{\mathbb N}}
U_i \cdot (1, \Delta\mathbf{\tilde{V}}_i)^\top$, where $\Delta\mathbf{\tilde{V}}_i$, $i\in{\mathbb N}$, are independent copies of $\Delta\mathbf{\tilde{V}}=\Delta\mathbf{\tilde{V}}^{(1)}$.
3. If $\xi$ admits a representation , we have $\sP(\Delta\mathbf{\tilde{V}}^{(1)}\in d \mathbf z) =
\sP(\Delta\mathbf{{V}}\in d \mathbf z)$, which shows that is indeed a special case of Theorem \[theo\_cond\_increments\_general\].
\[rem\_thres\] In the above theorem, the sequence $a(n)$ of thresholds is only assumed to converge to $\infty$, as $n\to\infty$, ensuring that $\{\eta(t_0) > a(n)\}$ becomes a rare event. For statistical applications $a(n)$ should also be chosen such that the number of exceedances $$\begin{aligned}
N(n) = \sum_{i=1}^n {\mathbf 1}\{ \eta_i(t_0) > a(n) \}
\end{aligned}$$ converges to $\infty$ almost surely, where $(\eta_i)_{i\in{\mathbb N}}$ is a sequence of independent copies of $\eta$. By the Poisson limit theorem, this is equivalent to the additional assumption that $\lim_{n\to\infty} a(n)/n = 0$, since in that case $n\sP(\eta(t_0) >
a(n)) = n / a(n) \to \infty$, as $n\to\infty$.
[@eng2012a] consider Hüsler-Reiss distributions [@hue1989; @kab2011] and obtain their limiting results by conditioning on certain extremal events $A\subset E$. They show that various choices of $A$ are sensible in the Hüsler-Reiss case, leading to different limiting distributions of the increments of $\eta$. In case $\xi$ is a Brown-Resnick process and $A =
(1,\infty)\times [0, \infty)^{k}$ the assertions of Theorem \[theo\_cond\_increments\_general\] and [@eng2012a Thm. 3.3] coincide.
A commonly used class of stationary yet non-ergodic max-stable processes on ${\mathbb R}^d$ is defined by $$\begin{aligned}
\label{schlather_model}
\xi(t) = \max_{i\in{\mathbb N}} U_i Y_i(t), \quad t\in{\mathbb R}^d,\end{aligned}$$ where $\sum_{i\in{\mathbb N}} \delta_{U_i}$ is a Fréchet point process, $Y_i(t)=\max(0, \tilde Y_i(t))$, $i \in {\mathbb N}$, and the $\tilde Y_i$ are i.i.d. stationary, centered Gaussian processes with $\sE(\max(0,
\tilde Y_i(t))) =1$ for all $t\in{\mathbb R}^d$ [@sch2002; @bla2011]. Note that in general, a $t_0\in{\mathbb R}^d$ s.t. $Y_i(t_0)=1$ a.s. does not exist, i.e., the process admits representation but not representation . In particular, for the extremal Gaussian process we have $p=\sP(V(t_0)=0)=1/2$ and the distribution of the increments in becomes $$\begin{aligned}
\sP(\Delta\mathbf{\tilde{V}}^{(1)} \! \in \rd \mathbf z)
&= \frac12 \sE\Bigl[ Y(t_0) \, \Big|\,
(Y(t_1)/Y(t_0), \ldots, Y(t_k)/Y(t_0)) = \mathbf z,
\, Y(t_0)>0\Bigr]\\
& \qquad \cdot\sP\Bigl( \bigl(Y(t_1)/Y(t_0), \ldots,
Y(t_k)/Y(t_0)\bigr)
\in \rd{\mathbf z}\, \Big|\, Y(t_0)>0 \Bigr).\end{aligned}$$
While the Hüsler-Reiss distribution is already given by the incremental representation , cf. [@kab2011], other distributions can be suitably rewritten, provided that the cumulative distribution function and hence the respective exponent measure $\mu$ is known.
\[calculateW\] Let $\Xi = (\xi(t_0),\dots,\xi(t_k))$ be a max-stable process on $T = \{t_0, \ldots, t_k \}$ with standard Fréchet margins and suppose that its exponent measure $\mu$ is concentrated on $(0, \infty) \times [0, \infty)^{k}$. Define a random vector ${\mathbf W}=(W^{(1)}, \ldots, W^{(k)})$ via its cumulative distribution function $$\begin{aligned}
\label{def_W}
\sP( {\mathbf W}\leq \mathbf{s}) = \mu(B_{\mathbf{s}} \cap A),
\quad \mathbf{s}\in [0,\infty)^{k},
\end{aligned}$$ where $A = (1,\infty)\times [0, \infty)^{k}$ and $B_{\mathbf s} =
\{{\mathbf x}\in [0,\infty)^{k+1}: \, (x^{(1)},\ldots,x^{(k)}) \leq
x^{(0)} {\mathbf s}\}$. Then, $\Xi$ allows for an incremental representation with ${\mathbf W}_i$, $i\in{\mathbb N}$, being independent copies of ${\mathbf W}$.
First, we note that indeed defines a valid cumulative distribution function. To this end, consider the measurable transformation $$\begin{aligned}
T: (0, \infty)\times [0, \infty)^{k}
\to (0, \infty)\times [0, \infty)^{k}, \ (x_0,\dots, x_k)
\mapsto \left(x_0, \frac{x_1}{x_0}, \dots, \frac{x_k}{x_0}\right).
\end{aligned}$$ Then, $ T(B_{\mathbf{s}} \cap A) = (1, \infty) \times [{\mathbf 0}, {\mathbf s}]$ and the measure $\mu^T(\cdot) = \mu(T^{-1}((1,\infty)\times \,\cdot\,))$ is a probability measure on $[0, \infty)^{k}$. Since $$\begin{aligned}
\mu(B_{\mathbf{s}} \cap A)
= \mu(T^{-1}((1,\infty)\times[{\mathbf 0}, {\mathbf s}]))
= \mu^T([{\mathbf 0}, {\mathbf s}]),
\end{aligned}$$ the random vector ${\mathbf W}$ is well-defined and has law $\mu^T$.
By definition of the exponent measure, we have $\Xi \eqdist \max_{i
\in {\mathbb N}} {\mathbf X}_i$, where $\Pi = \sum_{i\in{\mathbb N}} \delta_{{\mathbf X}_i}$ is a PPP on $E$ with intensity measure $\mu$. Then, the transformed point process $T\Pi = \sum_{i\in{\mathbb N}} \delta_{(X_i^{(0)},\,
X_i^{(1)}/X_i^{(0)},\, \ldots,\, X_i^{(k)} /X_i^{(0)})}$ has intensity measure $$\begin{aligned}
\tilde \mu((c,\infty) \times [{\mathbf 0},{\mathbf s}])
={}& \mu\left(T^{-1}\left( (c,\infty) \times [{\mathbf 0},{\mathbf s}] \right)
\right)\\
={} & \mu(B_{\mathbf{s}} \cap ((c,\infty) \times [0,\infty)^k)) {}
={} c^{-1} \mu(B_{\mathbf{s}} \cap A)
\end{aligned}$$ for any $c > 0$, $\mathbf{s} \in [0,\infty)^k$, where we use the fact that $\mu$, as an exponent measure, has the homogeneity property $c^{-1}\mu(\rd{\mathbf x})=\mu(\rd(c{\mathbf x}))$. Thus, $T\Pi$ has the same intensity as $\sum_{i\in{\mathbb N}} \delta_{(U_i, {\mathbf W}_i)}$, where $\sum_{i\in{\mathbb N}} \delta_{U_i}$ is a Fréchet point process and ${\mathbf W}_i$, $i \in {\mathbb N}$, are i.i.d. vectors with law $\sP({\mathbf W}\leq
\mathbf{s}) = \mu(B_{\mathbf{s}} \cap A)$. Hence, we have $$\begin{aligned}
\Xi\eqdist{}& \max_{i \in {\mathbb N}} T^{-1}\left(\big(X_i^{(0)}, X_i^{(1)} / X_i^{(0)},
\ldots, X_i^{(k)} / X_i^{(0)}\big)\right)\\
\eqdist{}& \max_{i \in {\mathbb N}} T^{-1}\left(\big(U_i,{\mathbf W}_i\big)\right) {}
={} \max_{i \in {\mathbb N}} U_i {\mathbf W}_i,
\end{aligned}$$ which completes the proof.
\[ex:symm\_log\] For $T=\{t_0,\dots,t_k\}$, the symmetric logistic distribution is given by $$\begin{aligned}
\sP(\xi(t_0) \leq x_0,\dots, \xi(t_k) \leq x_k) =
\exp\left[ - \left( x_0^{-q}+ \dots + x_k^{-q}\right)^{1/q} \right],
\label{eq:cdf_symm_log}
\end{aligned}$$ for $x_0,\dots,x_k>0$ and $q > 1$. Hence, the density of the exponent measure is $$\begin{aligned}
\mu(\rd x_0,\dots,\rd x_k) =
\left(\sum_{i=0}^k x_i^{-q}\right)^{1/q -(k+1)}
\left(\prod_{i=1}^k(iq-1)\right)
\prod_{i=0}^k x_i^{-q-1} \rd x_0\dots \rd x_k.
\end{aligned}$$ Applying Proposition \[calculateW\], the incremental process $W$ in the representation is given by $$\begin{aligned}
\sP(W(t_1) \leq s_1, \dots W(t_k) \leq s_k)
= \left(1 + \sum_{i=1}^k s_i^{-q}\right)^{1/q - 1}.
\end{aligned}$$
Continuous sample paths
-----------------------
In this subsection, we provide an analog result to Theorem \[theo\_cond\_increments\_general\], in which convergence in the sense of finite-dimensional distributions is replaced by weak convergence on function spaces. In the following, for a Borel set $U\subset{\mathbb R}^d$, we denote by $C(U)$ and $C^+(U)$ the space of non-negative and strictly positive continuous functions on $U$, respectively, equipped with the topology of uniform convergence on compact sets.
\[theo\_cond\_increments\_cont\] Let $K$ be a compact subset of ${\mathbb R}^d$ and $\{\eta(t): \ t\in K\}$ be a process with positive and continuous sample paths in the MDA of a max-stable process $\{\xi(t): \ t\in K\}$ as in in the sense of weak convergence on $C(K)$. In particular, suppose that $$\frac 1n \max_{i=1}^n \eta_i(\cdot) \cvgdist \xi(\cdot), \quad
n\to\infty.$$ Let $W$ be the incremental process from and $Z$ a Pareto random variable, independent of $W$. Then, for any sequence $a(n)$ of real numbers with $a(n) \to \infty$, we have the weak convergence on $(0,\infty)\times C(K)$ $$\begin{aligned}
\left(\frac{\eta(t_0)}{a(n)},
\frac{\eta(\cdot)}{\eta(t_0)} \ \Big|\ \eta(t_0) > a(n) \right)
\cvgdist (Z, W(\cdot)),
\end{aligned}$$ as $n$ tends to $\infty$.
\[weak\_conv\_Rd\] Analogously to [@whi1970 Thm. 5], weak convergence of a sequence of probability measures $P_n$, $n\in{\mathbb N}$, to some probability measure $P$ on $C({\mathbb R}^d)$ is equivalent to weak convergence of $P_n r_j^{-1}$ to $P r_j^{-1}$ on $C([-j, j]^d)$ for all $j\geq 1$, where $r_j :
C({\mathbb R}^d) \to C([-j,j]^d)$ denotes the restriction of a function to the cube $[-j, j]^d$. Hence the assertion of Theorem \[theo\_cond\_increments\_cont\] remains valid if the compact set $K$ is replaced by ${\mathbb R}^d$.
As the process $\xi$ is max-stable and $\eta\in\text{MDA}(\xi)$, similarly to the case of multivariate max-stable distributions (cf. Theorem \[theo\_cond\_increments\_general\]), we have that $$\begin{aligned}
\label{conv_dehaan}
\lim_{u \to \infty} u\sP(\eta / u \in B) = \mu(B)
\end{aligned}$$ for any Borel set $B \subset C(K)$ bounded away from $0^K$, i.e., $\inf\{\sup_{s\in K} f(s) : \ f\in B\} > 0$, and with $\mu(\partial
B) = 0$ [@deh2006a Cor. 9.3.2], where $\mu$ is the *exponent measure* of $\xi$, defined by $$\begin{aligned}
& \sP(\xi(s) \leq x_j, \ s \in K_j, \ j=1,\ldots,m) \nonumber \\
&={} \exp\left[-\mu\left(\left\{ f \in C(K): \ \textstyle\sup_{s \in K_j} f(s) > x_j \textrm{ for some } j \in \{1,\ldots,m\}
\right\}\right)\right]
\end{aligned}$$ for $x_j \geq 0$, $K_j \subset K$ compact. Thus, $\mu$ equals the intensity measure of the Poisson point process $\sum_{i \in {\mathbb N}}
\delta_{U_i W_i(\cdot)}$. For $z>0$ and $D\subset C(K)$ Borel, we consider the sets $$\begin{aligned}
A_{z} &= \{f \in C(K): \ f(t_0) > z\}\\
B_D &= \{f \in C(K) : f(\cdot)/f(t_0)\in D\}\end{aligned}$$ and $A=A_1$. Note that $B_D$ is invariant w.r.t. multiplication by any positive constant. Then, as $W(t_0) = 1$ a.s., we have $\mu(A_{z}) = \int_{z}^\infty u^{-2} \sd u = z^{-1}$ and for $s_0\geq 1$ and any Borel set $D \subset C(K)$ with $\sP(W \in \partial D) = 0$, by , we get $$\begin{aligned}
&\sP\left\{\eta(t_0) / a(n) > s_0,\ \eta(\cdot)/\eta(t_0) \in D \, \Big|\, \eta(t_0) > a(n) \right\}\\
&= \frac{a(n) \sP\bigl\{\eta(\cdot) / a(n) \in A_{s_0} \cap B_D \cap A\bigr\}}{a(n) \sP\bigl\{\eta(\cdot) / a(n) \in A\bigr\}}\\
&\stackrel{n \to \infty}{\longrightarrow}{} \frac{\mu(B_D \cap A_{s_0})}{\mu(A)}\\
&={} \int_{s_0}^\infty u^{-2}\sP\bigl\{u W(\cdot) \in B_D\bigr\} \sd u\\
&={} s_0^{-1} \sP\bigl\{W(\cdot) \in D\bigr\},
\end{aligned}$$ which is the joint distribution of $Z$ and $W(\cdot)$.
\[BRproc\] For $T={\mathbb R}^d$, $d\geq 1$, let $\{Y(t): \ t\in T\}$ be a centered Gaussian process with stationary increments, continuous sample paths and $Y(t_0) = 0$ for some $t_0\in{\mathbb R}^d$. Note that by [@adl2007 Thm. 1.4.1] it is sufficient for the continuity of $Y$ that there exist constants $C,\alpha,\delta > 0$, such that $$\begin{aligned}
\sE |Y(s) - Y(t)|^2 \leq \frac{C}{|\log \|s-t\| |^{1+\alpha}}
\end{aligned}$$ for all $s,t\in{\mathbb R}^d$ with $\|s-t\|<\delta$. Further let $\gamma(t) =
\sE(Y(t) - Y(0))^2$ and $\sigma^2(t) = \sE(Y(t))^2$, $t \in {\mathbb R}^d$, denote the variogram and the variance of $Y$, respectively. Then, with a Fréchet point process $\sum_{i\in{\mathbb N}} \delta_{U_i}$ and independent copies $Y_i$ of $Y$, $i\in{\mathbb N}$, the process $$\begin{aligned}
\label{BR_proc}
\xi(t) = \max_{i\in{\mathbb N}} U_i \exp\left(Y_i(t) - \sigma^2(t) / 2\right),
\quad t\in{\mathbb R}^d,
\end{aligned}$$ is stationary and its distribution only depends on the variogram $\gamma$. Comparing with the incremental representation , the distribution of the increments is given by the log-Gaussian random field $W(t) = \exp\left(Y(t) -
\sigma^2(t) / 2\right)$, $t\in{\mathbb R}^d$, and Theorem \[theo\_cond\_increments\_cont\] applies.
Mixed moving maxima representation {#M3representation}
==================================
A large and commonly used class of max-stable processes is the class of M3 processes . Let $$\begin{aligned}
\label{pi0}
\Pi_0 = \sum_{i\in{\mathbb N}} \delta_{(U_i.T_i,F_i)}
\end{aligned}$$ be the corresponding PPP on $(0,\infty)\times{\mathbb R}^d\times C({\mathbb R}^d)$ with intensity $u^{-2}\rd u \,\rd t \,\sP_F(\rd f)$. In the sequel, M3 processes are denoted by $$\begin{aligned}
M(t) = \max_{i\in{\mathbb N}} U_i F_i(t- T_i), \quad t\in{\mathbb R}^d.\end{aligned}$$ The marginal distributions of $M$ are given by $$\begin{aligned}
& \sP(M(t_0)\leq s_0, \ldots, M(t_k)\leq s_k) \notag\\
&= \sP\left[ \Pi_0
\left(\left\{(u,t,f):
\max_{l=0}^k u f(t_l-t)/s_l > 1\right\}\right) = 0\right] \notag\\
&= \exp\left(- \int_{C({\mathbb R}^d)} \int_{{\mathbb R}^d}
\max_{l=0}^k (f(t_l-t)/s_l)\, \rd t \, \sP_F(\rd f) \right),\label{M3_marginal}\end{aligned}$$ $t_0, \ldots, t_k\in{\mathbb R}^d$, $s_0, \ldots, s_k\geq 0$, $k\in{\mathbb N}$.
In Section \[examples\_increment\_representation\], we were interested in recovering the incremental process $W$ from processes in the MDA of a max-stable process with incremental representation. In case of M3 processes, the object of interest is clearly the distribution of the shape function $F$. Thus, in what follows, we provide the corresponding convergence results for processes $\eta$ in the MDA of an M3 process. We distinguish between processes on ${\mathbb R}^d$ with continuous sample paths and processes on a grid (${\mathbb Z}^d$). The main idea is to consider $\eta$ in the neighborhood of its own (local) maximum, conditional on this maximum being large.
Continuous Case
---------------
Let $\{\eta(t): \, t \in {\mathbb R}^d\}$ be strictly positive and in the MDA of a mixed moving maxima process $M$ in the sense of weak convergence in $C({\mathbb R}^d)$. We assume that $\eta$ is normalized such that the norming functions in are given by $c_n(t) = 1 / n$ and $b_n(t) = 0$, for any $n\in{\mathbb N}$ and $t\in{\mathbb R}^d$. Further suppose that the shape function $F$ of $M$ is sample-continuous and satisfies $$\begin{aligned}
\begin{split}
F({\mathbf}0) &= \lambda \quad a.s., \\
F(t) & \in [0,\lambda) \ \forall t \in {\mathbb R}^d \setminus \{{\mathbf}0\} \quad a.s.
\label{eq:Fmaxatorigin}
\end{split}\end{aligned}$$ for some $\lambda > 0$ and $$\int_{{\mathbb R}^d} \sE\left\{ \max_{t_0 \in K} F(t_0 - t) \right\} \sd t < \infty
\label{eq:sup-integrability-cont}$$ for any compact set $K \subset {\mathbb R}^d$. Under these assumptions, there is an analog result to Theorem \[theo\_cond\_increments\_cont\].
\[thm\_conv\_mmm\] Let ${Q}, K \subset {\mathbb R}^d$ be compact such that $\partial {Q}$ is a Lebesgue null set and let $$\tau_{Q}: \ C({Q}) \to {\mathbb R}^d, \ f \mapsto \inf\left( \operatorname*{arg\,max}_{t \in {Q}}
f(t) \right),$$ where [“inf”]{} is understood in the lexicographic sense. Then, under the above assumptions, for any Borel set $B \subset C(K)$ with $\sP(F / \lambda \in \partial B) = 0$, and any sequence $a(n)$ with $a(n) \to \infty$ as $n \to \infty$, we have $$\begin{aligned}
& \lim_{\substack{\{{\mathbf}0\} \in L \nearrow {\mathbb R}^d\\ {\rm compact}}}
\limsup_{n \to \infty} \sP\Big\{ \eta\big(\tau_{Q}(\eta|_{Q})+\cdot\big) \big/ \eta(\tau_{Q}(\eta|_{Q})) \in B \ \Big| \\[-1em]
& \hspace{2.5cm} \max_{t\in {Q}}\eta(t) = \max_{t \in {Q}\oplus L} \eta(t),
\ \max_{t\in {Q}}\eta(t) \geq a(n)\Big\} \hfill {}={} \hfill \sP\big\{F(\cdot) / \lambda \in B\big\},
\end{aligned}$$ where $\oplus$ denotes morphological dilation.
The same result holds true if we replace $\limsup_{n \to \infty}$ by $\liminf_{n \to \infty}$.
First, we consider a fixed compact set $L\subset{\mathbb R}^d$ large enough such that $K \cup \{{\bf 0}\} \subset L$ and define $$\begin{aligned}
A_L = \left\{ f \in C({Q}\oplus L): \ \max_{t\in {Q}}f(t) \geq 1, \ \max_{t\in {Q}}f(t) = \max_{t \in {Q}\oplus L} f(t)\right\}
\end{aligned}$$ and $$\begin{aligned}
C_B = \left\{f \in C({Q}\oplus L):
\ f\big(\tau_{Q}(f|_{Q}) + \,\cdot\,\big) \big/ f(\tau_{Q}(f|_{Q})) \in B\right\}
\end{aligned}$$ for any Borel set $B \subset C(K)$. Note that $C_B$ is invariant w.r.t. multiplication by any positive constant. Thus, we get $$\begin{aligned}
& \sP\Big\{\eta\big(\tau_{Q}(\eta|_{Q}) + \cdot\big)
\big/ \eta(\tau_{Q}(\eta|_{Q})) \in B \ \Big|\ \max_{t\in {Q}}\eta(t) = \max_{t \in {Q}\oplus L} \eta(t) \geq a(n)\Big\}
\nonumber \\
& ={} \sP\big\{\eta / a(n) \in C_B \,\big|\, \eta / a(n) \in A_L\big\}
\nonumber\\
& ={} \frac{a(n) \sP\big\{\eta/a(n) \in C_B,\,\eta/a(n)
\in A_L \big\}}{a(n) \sP\big\{\eta/a(n) \in A_L \big\}}.
\label{eq:expand-cont}
\end{aligned}$$
By [@deh2006 Cor. 9.3.2] and [@res2008 Prop. 3.12] we have $$\begin{aligned}
\limsup_{u \to \infty} u\sP(\eta / u \in C) \leq{}& \mu(C), \quad C \subset C({Q}\oplus L) \text{ closed},\\
\liminf_{u \to \infty} u\sP(\eta / u \in O) \geq{}& \mu(O), \quad O \subset C({Q}\oplus L) \text{ open},
\end{aligned}$$ where $C$ and $O$ are bounded away from $0^K$. Here, $\mu$ is the intensity measure of the PPP $\sum_{i \in {\mathbb N}} \delta_{U_i F_i(\,\cdot\, -
T_i)}$ restricted to $C({Q}\oplus L)$. Thus, by adding or removing the boundary, we see that all the limit points of Equation lie in the interval $$\label{eq:liminterval}
\left[ \frac{\mu(C_B \cap A_L) - \mu(\partial (C_B \cap A_L))}{\mu(A_L) + \mu(\partial A_L)}, \frac{\mu(C_B \cap A_L) + \mu(\partial (C_B \cap A_L))}{\mu(A_L) - \mu(\partial A_L)}\right].$$ We note that $A_L$ is closed and the set $$\begin{aligned}
A_L^* ={} & \bigg\{ f \in C({Q}\oplus L): \\[-.5em]
&\quad \, \tau_{Q}(f|_{Q}) \in {Q}^o, \ \max_{t\in {Q}} f(t) > \max\big\{1, f(t)\big\} \ \forall t \in {Q}\oplus L \setminus\{\tau_{Q}(f|_{Q})\} \bigg\}
\end{aligned}$$ is in the interior of $A_L$ (Lemma \[lem:AL\]). Hence, we can assess $$\begin{aligned}
\mu(\partial A_L) \leq{} & \quad \ \mu(\{f \in C({Q}\oplus L): \ \max_{t\in {Q}}f(t) = 1\}) \notag\\
& + \mu\bigg( \quad \bigg( \quad \{f \in C({Q}\oplus L): \ \tau_{Q}(f|_{Q}) \in \partial {Q}\} \notag\\
& \hspace{1.55cm} \cup \left\{f \in C({Q}\oplus L): \ \operatorname*{arg\,max}_{t \in {Q}\oplus L} f(t) \text{ is not unique}\right\}\bigg) \notag\\
& \qquad \cap \left\{f \in C({Q}\oplus L): \ \max_{t\in {Q}}f(t) = \max_{t \in {Q}\oplus L} f(t) \geq 1 \right\}\bigg) \notag \\
\leq{} & 0 + \int_{\partial {Q}} \int_{\lambda^{-1}}^\infty u^{-2} \sd u \sd t_0 \notag\\
& \phantom{0} + \int_{{\mathbb R}^d \setminus ({Q}\oplus L)} \int_{\lambda^{-1}}^\infty u^{-2} \sP\left\{u \max_{t_0 \in {Q}} F(t_0 -x) \geq 1\right\} \sd u \sd x \label{eq:partAL}.
\end{aligned}$$ Here, the equality $\mu(\{f \in C({Q}\oplus L): \ \max_{t\in {Q}}f(t) = 1\}) = 0$ holds as $\max_{t \in {Q}} M(t)$ is Fréchet distributed (cf. [@deh2006 Lemma 9.3.4]). Since $\partial {Q}$ is a Lebesgue null set, the second term on the right-hand side of also vanishes. Thus, $$\begin{aligned}
\mu(\partial A_L) \leq{} & \int_{{\mathbb R}^d \setminus ({Q}\oplus L)} \int_{\lambda^{-1}}^\infty u^{-2} \sP\left\{u \max_{t_0 \in {Q}} F(t_0 -x) \geq 1\right\} \sd u \sd x =: c(L) \label{eq:partAL2}.
\end{aligned}$$
Now, let $B \subset C(K)$ a be Borel set such that $\sP(F / \lambda \in \partial B) = 0$. For the set $C_B$, we obtain that the set $$\begin{aligned}
C_B^* ={} & \bigg\{ f \in C({Q}\oplus L): \ \operatorname*{arg\,max}_{f \in {Q}} f(t) \text{ is unique},\ \frac{f\big(\tau_{Q}(f|_{Q}) + \cdot\big)}{f(\tau_{Q}(f|_{Q}))} \in B^o \bigg\}
\end{aligned}$$ is in the interior of $C_B$ and that the closure of $C_B$ is a subset of $$\begin{aligned}
C_B^* \cup{} & \left\{f \in C({Q}\oplus L): \, \operatorname*{arg\,max}_{t \in {Q}} f(t) \text{ is not unique}\right\}\\
\cup{} & \left\{f \in C({Q}\oplus L): \, f\big(\tau_{Q}(f|_{Q}) + \cdot\big) \big/ f(\tau_{Q}(f|_{Q})) \in \partial B\right\}
\end{aligned}$$ (Lemma \[lem:interCB\] and Lemma \[lem:CB\]). Thus, by , we can assess $$\begin{aligned}
\mu(\partial (C_B \cap A_L)) \leq{} & \mu(\partial A_L) + \mu(\partial C_B \cap A_L) \notag\\
\leq{} & c(L) + \int_{{\mathbb R}^d \setminus ({Q}\oplus L)} \int_{\lambda^{-1}}^\infty u^{-2}
\sP\left\{u \max_{t_0 \in {Q}} F(t_0 -x) \geq 1\right\}\sd u \sd x \notag\\
& \hspace{0.7cm} + \int_{{Q}} \int_{\lambda^{-1}}^\infty u^{-2} \sP(F / \lambda \in \partial B) \sd u \sd t \quad
{}={} \quad 2 c(L). \label{eq:partCB}
\end{aligned}$$
Furthermore, we get $$\begin{aligned}
& \mu(C_B \cap A_L) \nonumber \\
={} & \int_{Q}\int_{\lambda^{-1}}^\infty u^{-2} \sP\Big\{F(\cdot) / \lambda \in B\Big\}
\sd u \sd t_0 \nonumber \\
& + \int_{{\mathbb R}^d \setminus({Q}\oplus L)} \int_{\lambda^{-1}}^\infty u^{-2}
\sP\bigg\{u \max_{t_0 \in {Q}} F(t_0 -x) \geq 1,\ \nonumber \\
& \hspace{2.5cm}
F\left(\Big(\tau_{Q}(F(\cdot-x)|_{Q})\Big)+\cdot-x\right) \Big/ \max_{t_0 \in {Q}} F(t_0-x) \in B,\nonumber \\
& \hspace{2.5cm} F(t-x) / \max_{t_0 \in {Q}} F(t_0-x) \leq 1
\ \forall t \in {Q}\oplus L \bigg\} \sd u \sd x. \label{eq:CB}
\end{aligned}$$ The second term in is positive and can be bounded from above by $c(L)$. Setting $B= C(K)$, $\mu(A_L)$ can be expressed in an analogous way. Now, we plug in the results of , and into to obtain that all the limit points of are in the interval $$\begin{aligned}
\left[\frac{\lambda \cdot |{Q}| \cdot \sP\big\{F(\cdot) / \lambda \in B\big\} - 2c(L)}{\lambda \cdot |{Q}| + 2c(L)},
\frac{\lambda \cdot |{Q}| \cdot \sP\big\{F(\cdot) / \lambda \in B\big\} + 3c(L)}{\lambda \cdot |{Q}| - c(L)} \right].
\end{aligned}$$
Finally, we note that $c(L)$ can be bounded from above by $$\int_{{\mathbb R}^d \setminus ({Q}\oplus L)} \sE \Big\{\max_{t_0 \in {Q}}
F(t_0-x)\Big\} \sd x,$$ which vanishes for $L \nearrow {\mathbb R}^d$ because of assumption . This yields the assertion of the theorem.
We conclude the treatment of the continuous case with an example of a process $\eta$ that allows for an application of Theorem \[thm\_conv\_mmm\]. As $\eta$ will be composed of a (locally) finite number of shape functions from the M3 construction in , $\eta$ may directly model rainfall data and has therefore the potential for various practical applications.
\[ex:mmm-mda\] Let $\{F(t): \ t \in {\mathbb R}^d\}$ be a random shape function as defined in . For $c, \epsilon >
0$ let $\Pi_{c, \epsilon} = \sum_{i\in{\mathbb N}} \delta_{(U_i.T_i,F_i)}$ be a PPP on $(0,\infty)\times{\mathbb R}^d\times C({\mathbb R}^d)$ with intensity $$\begin{aligned}
c{\mathbf 1}_{\{u\geq\epsilon\}}u^{-2}\rd u\,\rd t\,\sP_F(\rd f).\end{aligned}$$ and, for $\kappa > 0$, define a process $\tilde M=\tilde M_{c,\epsilon,\kappa}$ by $$\tilde M(\cdot) = \kappa \vee \max_{(u, t, f)\in \Pi_{c, \epsilon}}
u f(\,\cdot\, - t).$$ Then, the following statements hold.
1. $\tilde M$ is in the MDA of the M3 process $M$ associated to $F$ in the sense of finite-dimensional distributions.
2. If $F$ satisfies , then $\tilde M$ is in the MDA of $M$ in the sense of weak convergence on $C({\mathbb R}^d)$.
For a proof of this example, the reader is referred to Appendix \[sec:proof\_ex\_mmm-mda\].
Discrete Case
-------------
Theorem \[thm\_conv\_mmm\] allows for estimation of $F$ if the complete sample paths of $\eta$ are known, at least on a large set ${Q}\oplus L \subset {\mathbb R}^d$. For many applications, this assumption might be too restrictive. Therefore, we seek after a weaker assumption that only requires to know $\eta$ on a grid. This needs a modification of the underlying model leading to a discretized mixed moving maxima process.
Let $\{F(t): \ t \in {\mathbb Z}^d\}$ be a measurable stochastic process with values in $[0, \infty)$ and $$\begin{aligned}
\label{assumption_integral_discr}
\sum_{t \in {\mathbb Z}^d} \sE F(t) = 1.\end{aligned}$$ Further, let $\Pi_{0,\operatorname{discr}} = \sum_{i\in{\mathbb N}} \delta_{(U_i.T_i,F_i)}$ be a Poisson point process on $(0,\infty) \times {\mathbb Z}^d\times
[0,\infty)^{{\mathbb Z}^d}$ with intensity $u^{-2}\rd u\, \delta_{{\mathbb Z}^d}(\rd t)
\, \sP_F(\rd f)$. Then, the discrete mixed moving maxima process $M_{\operatorname{discr}}$ is defined by $$\begin{aligned}
M_{\operatorname{discr}}(t) = \max_{i\in{\mathbb N}} U_i F_i(t- T_i), \quad t\in{\mathbb Z}^d.
\label{def_M3_discr}\end{aligned}$$ The process $M_{\operatorname{discr}}$ is max-stable and stationary on ${\mathbb Z}^d$ and has standard Fréchet margins.
Let $\{\eta(t), \ t \in {\mathbb Z}^d\}$ be in the MDA of a discrete mixed moving maxima process $M_{\operatorname{discr}}$ in the sense of convergence of finite-dimensional distributions with norming functions $c_n(t) = 1/n$ and $b_n(t) = 0$ in , $n\in{\mathbb N}$ and $t\in{\mathbb Z}^d$. Furthermore, we assume that the shape function $F$ satisfies with ${\mathbb R}^d$ being replaced by ${\mathbb Z}^d$. Then, analogously to Theorem \[thm\_conv\_mmm\], the following convergence result can be shown.
Under the above assumptions, for any $k \in {\mathbb N}$, $k+1$ distinct points $t_0, \ldots, t_k \in {\mathbb Z}^d$, any Borel sets $B_1, \ldots, B_k
\subset [0,\infty)$ such that $$\sP\big\{(F(t_1)/\lambda,\ldots,F(t_k)/\lambda) \in
\partial(B_1\times\cdots\times B_k)\big\}=0,$$ and any sequence $a(n)$ with $a(n) \to \infty$ as $n \to \infty$, it holds $$\begin{aligned}
\lim_{\substack{\{{\mathbf}0\} \in L \nearrow {\mathbb Z}^d\\ {\rm compact}}}
\lim_{n \to \infty} &\sP\big\{ \eta(t_0+t_i) / \eta(t_0) \in B_i, \ i=1,\ldots,k \ \big| \\[-1em]
& \hspace{3.5cm} \eta(t_0) = \max_{t \in L} \eta(t_0+t),\ \eta(t_0) \geq a(n)\big\}\\[.5em]
= &\sP\big\{F(t_i) / \lambda \in B_i, \ i=1,\ldots,k\big\}.
\end{aligned}$$
Switching between the different representations {#sec:switching}
===============================================
In the previous sections we analyzed processes that admit the incremental representations or and, on the other hand, processes of M3 type as in . We show that under certain assumptions, we can switch from one representation to the other.
Incremental representation of mixed moving maxima processes
-----------------------------------------------------------
We distinguish between M3 processes with strictly positive shape functions, for which we can find an incremental representation , and general non-negative shape functions, for which only the weaker representation can be obtained.
### Mixed moving maxima processes with positive shape functions {#ex:M3}
Let $M$ be an M3 process on ${\mathbb R}^d$ as in with a shape function $F$ with $F(t) > 0$ for all $t \in {\mathbb R}^d$. Then $M$ admits a representation with $t_0 = 0$ and incremental process $W$ given by $$\begin{aligned}
\sP(W\in L)
= \int_{C^+({\mathbb R}^d)} \int_{{\mathbb R}^d} {\mathbf 1}_{\{f(\cdot - t)/f(-t)\in L\}} f(-t)
\sd t \, \sP_F(\rd f), \quad L\in\mathcal B(C^+({\mathbb R}^d)).\label{defWofM3}\end{aligned}$$
We consider the two Poisson point processes on $(0,\infty)\times C^+({\mathbb R}^d)$ $$\begin{aligned}
\Pi_1 = \sum_{i\in{\mathbb N}}\delta_{(U_i F_i(-T_i), F_i(\cdot - T_i)/F_i(-T_i))}, \label{auxPPP1}\end{aligned}$$ as a transformation of $\Pi_0$ in , and $$\begin{aligned}
\Pi_2 = \sum_{i\in{\mathbb N}}\delta_{(U'_i, W_i(\cdot))}, \label{auxPPP2}\end{aligned}$$ with $W_i$, $i\in{\mathbb N}$, being independent copies of $W$, and with $\sum_{i\in{\mathbb N}}\delta_{U'_i}$ being a Fréchet point process. Then the intensity measures of $\Pi_1$ and $\Pi_2$ satisfy $$\begin{aligned}
&\sE\Pi_1([z,\infty)\times L)\\
&=\int_{C^+({\mathbb R}^d)}\int_{{\mathbb R}^d}\int_0^\infty u^{-2} {\mathbf 1}_{\{u f(-t) \geq z\}}
{\mathbf 1}_{\{f(\cdot - t)/f(-t)\in L\}}
\, \rd u \, \rd t \, \sP_F(\rd f)\\
&= z^{-1} \int_{C^+({\mathbb R}^d)}\int_{{\mathbb R}^d} {\mathbf 1}_{\{f(\cdot - t)/f(-t)\in L\}} f(-t)
\, \rd t \, \sP_F(\rd f)\\
&=z^{-1} \sP(W\in L)\\
&=\sE\Pi_2([z,\infty)\times L), \end{aligned}$$ $L\in\mathcal B({C^+({\mathbb R}^d)})$, $z> 0$, and hence $\Pi_1\eqdist\Pi_2$. The assertion follows from the fact that $M$ is uniquely determined by $\Pi_1$ via the relation $M(t) = \max_{(v,g)\in\Pi_1} v g(t)$, $t\in{\mathbb R}^d$.
While the definition of $W$ in is rather implicit, in the following, we provide an explicit construction of the incremental process $W$, which can also be used for simulation. To this end, let $\sum_{i\in{\mathbb N}}\delta_{U_i''}$ be a Fréchet point process and let the distribution of $(S,G)\in C^+({\mathbb R}^d)\times{\mathbb R}^d$ be given by $$\begin{aligned}
\label{hat_distr}
&\sP\bigl((S, G)\in (B\times L)\bigr)\\ &=
\int_{C^+({\mathbb R}^d)}\int_{{\mathbb R}^d} {\mathbf 1}_{s\in B}{\mathbf 1}_{f\in L}
\frac{f(-s)}{\int f(r) \rd r} \,\rd s \left(\int f(r) \rd r\right)
\sP_F(\rd f) \notag\\
\notag &= \int_{C^+({\mathbb R}^d)}\int_{{\mathbb R}^d} {\mathbf 1}_{s\in B}{\mathbf 1}_{f\in L} f(-s) \, \rd s \, \sP_F(\rd f),\end{aligned}$$ $B\in\mathcal B^d$, $L\in\mathcal B({C^+({\mathbb R}^d)})$. In other words, $\sP_G(\rd f)= (\int f(r)\sd r)\,\sP_F(\rd f)$ and, conditional on $\{G=f\}$, the density function of the shift $S$ is proportional to $f(- \cdot)$. Putting $W(\cdot) = G(\cdot - S)/G(-S)$, equation is satisfied and with i.i.d. copies $W_i$, $i\in{\mathbb N}$, of $W$, we get that $\max_{i\in{\mathbb N}} U_i'' W_i(\cdot)$ is indeed an incremental representation of the mixed moving maxima process $M$.
\[mmm\_BRproc\] We consider the following two special cases of mixed moving maxima processes:
1. Let $\Sigma\in{\mathbb R}^{d\times d}$ be a positive definite matrix and let the shape function be given by $F(t) = (2\pi)^{-d/2} |\Sigma|^{-1/2}
\exp\left\{-\frac{1}{2}t^\top\Sigma^{-1} t\right\}$, $t\in{\mathbb R}^d$. Then, $M$ becomes the well-known Smith process. At the same time, by , $S\sim N(0,\Sigma)$ and $G \equiv F$. Thus $$\begin{aligned}
Y(t)&=\exp\left\{-\textstyle\frac{1}{2}(t-S)^\top\Sigma^{-1}(t-S) + \frac{1}{2}S^\top\Sigma^{-1}S\right\}\\
&= \exp\left\{-\textstyle\frac{1}{2}t^\top\Sigma^{-1}t + t^\top \Sigma^{-1} S\right\}.
\end{aligned}$$ Since $\sE(t^\top \Sigma^{-1} S)^2 = t^\top\Sigma^{-1}t $, $M$ is equivalent to the Brown-Resnick process in with variogram $\gamma(h) = h^\top \Sigma^{-1}h$.
2. For the one-dimensional Brown-Resnick process $\xi$ in with variogram $\gamma(h) =|h|$, i.e., $Y$ is the exponential of a standard Brownian motion with drift $-|t| / 2$, [@eng2011] recently showed that the M3 representation is given by $\{F(t): \, t\in{\mathbb R}\} = \{ Y(t) \mid Y(s)\leq 0 \ \forall s \in {\mathbb R}:
\, t\in{\mathbb R}\}$, i.e., the shape function is the exponential of a conditionally negative drifted Brownian motion. Having these two representations, it follows that the law of the conditional Brownian motion $F$, re-weighted by $\int F(t) \rd t$ and randomly shifted with density $F(-\cdot) / \int F(t) \rd t$, coincides with the law of $Y$.
### Mixed moving maxima processes with finitely supported shape functions {#ex:M32}
Let $M$ be an M3 process on ${\mathbb R}^d$ as in . In contrast to Section \[ex:M3\], where the shape functions are required to take positive values, here, we allow for arbitrary shape functions with values in $[0, \infty)$.
\[finite\_supp\] The M3 process $M$ as in allows for an incremental representation of the form , with incremental processes $V_i$ given by $$V_i(\cdot) = F_i(\cdot -R_i) / g(R_i).$$ Here $R_i$, $i\in{\mathbb N}$, are i.i.d. copies of a random vector $R$ with arbitrary density $g$ satisfying $g(t)>0$ for all $t\in{\mathbb R}^d$, and $F_i$, $i\in{\mathbb N}$, are i.i.d. copies of the random shape function $F$.
With $\sum_{i\in{\mathbb N}} \delta_{U_i}$ being a Fréchet point process, we consider the process $$\begin{aligned}
\tilde M(t) = \max_{i\in{\mathbb N}} U_i F_i(t-R_i) / g(R_i), \qquad t\in{\mathbb R}^d,\end{aligned}$$ which clearly is of the form . Then, $$\begin{aligned}
\sP&(\tilde M(t_0)\leq s_0, \ldots, \tilde M(t_k)\leq s_k) \\
&= \exp\left(- \int_{C({\mathbb R}^d)}\int_{{\mathbb R}^d}
\max_{l=0}^k (f(t_l-t)/(g(t)s_l)) g(t) \, \rd t \, \sP_F(\rd f)\right)\\
&= \exp\left(- \int_{C({\mathbb R}^d)}\int_{{\mathbb R}^d}
\max_{l=0}^k (f(t_l-t)/s_l)) \, \rd t \, \sP_F(\rd f)\right).$$ The right-hand side coincides with the marginal distribution of $M$, which is given by . This concludes the proof.
Decomposing $V$ as in with $t_0=0$, we obtain the equality in distribution $$\begin{aligned}
V^{(1)}(\cdot) \eqdist \bigl(F(\cdot -R) / g(R) \ \big|
-R\in \operatorname{supp}(F) \bigr).\end{aligned}$$ Applying Theorem \[theo\_cond\_increments\_general\] yields $$\begin{aligned}
&\sP\bigl(\Delta\mathbf{\tilde{V}}^{(1)}\in \rd{\mathbf z}\bigr)\notag\\
&= \sP\bigl( F(-R) / g(R) > 0\bigr)
\cdot\int_0^\infty y
\sP\big(V^{(1)}(0)\in \rd y,\ \Delta{\mathbf V}^{(1)}\in\rd{\mathbf z}\big)\notag\\
&= \int_{C({\mathbb R}^d)} \int_{-\operatorname{supp}(f)} g(s)\sd s \,\sP_F(\rd f)\notag\\
&\hspace{1cm}
\cdot\int_0^\infty y \int_{C({\mathbb R}^d)} \int_{-\operatorname{supp}(f)} {\mathbf 1}_{f(-t)/g(t)\in \rd y} {\mathbf 1}_{(f(t_l-t)/f(-t))_{l=1}^k \in \rd{\mathbf z}} \notag\\
&\hspace{6cm} \cdot g(t)
\left(\textstyle\int_{-\operatorname{supp}(f)} g(s)\rd s\right)^{-1} \sd t \,\sP_F(\rd f) \sd y\notag\\
&= \int_{C({\mathbb R}^d)} \int_{\operatorname{supp}(f)} g(-s)\sd s \sP_F(\rd f)\notag\\
&\hspace{1cm}
\cdot \int_{C({\mathbb R}^d)} \int_{\operatorname{supp}(f)} f(t) {\mathbf 1}_{(f(t_l+t)/f(t))_{l=1}^k \in \rd{\mathbf z}} \left(\textstyle\int_{\operatorname{supp}(f)} g(-s)\rd s\right)^{-1} \sd t \,\sP_F(\rd f).
\label{distr_gen_incr}\end{aligned}$$ If the shape function $F$ is deterministic, the right-hand side of simplifies to $ \int_{\operatorname{supp}(f)} f(t) {\mathbf 1}_{(f(t_l+t)/f(t))_{l=1}^k \in \rd{\mathbf z}} \sd t$, i.e., the asymptotic conditional increments of $\eta\in\text{MDA}(M)$ can be seen as a convolution of the shape function’s increments with a random shift, whose density is given by the shape function itself. Note in particular, that this distribution is independent of the choice of the density $g$ in Theorem \[finite\_supp\].
Section \[ex:M3\] considers the subclass of M3 processes with strictly positive shape functions and provides an incremental representation as in , which is nicely related to the conditional increments of $\eta$ due to the property $W(0)=1$. Section \[ex:M32\] applies to arbitrary M3 processes but only yields an incremental representation as in , for which the incremental process $V$ does not directly represent the conditional increments of $\eta$.
Mixed moving maxima representation of the incremental construction
------------------------------------------------------------------
\[theo:M3ofIncremental\] Let $\sum_{i \in {\mathbb N}} \delta_{U_i}$ be a Fréchet point process and let $W_i$, $i \in {\mathbb N}$, be independent copies of a non-negative, sample-continuous process $\{W(t),
\ t\in{\mathbb R}^d\}$, satisfying $$\begin{aligned}
\lim_{||t|| \to \infty} W(t) &= 0 && a.s.,\\
\sE W(t) &=1 &&\text{for all } t \in {\mathbb R}^d,\\
\text{and} \quad \sE\left\{\textstyle\max_{t \in K} W(t)\right\} & < \infty
&&\text{for any compact set $K \subset {\mathbb R}^d$.}
\end{aligned}$$ Furthermore, let $W$ be Brown-Resnick stationary, i.e., the process $\xi$, defined by $$\xi(t) = \max_{i \in {\mathbb N}} U_i W_i(t), \quad t \in {\mathbb R}^d,$$ is stationary with standard Fréchet margins. Then, the following assertions hold:
1. The random variables $$\begin{aligned}
\tau_i = \inf \left\{\operatorname*{arg\,sup}_{t \in {\mathbb R}^d} W_i(t)\right\} \quad \text{and}
\quad \gamma_i = \sup_{t \in {\mathbb R}^d} W_i(t)
\end{aligned}$$ are well-defined. Furthermore, $\sum_{i \in {\mathbb N}} \delta_{(U_i \gamma_i, \tau_i, W_i(\cdot + \tau_i)
/ \gamma_i)}$ is a Poisson point process on $(0,\infty) \times
{\mathbb R}^d \times C({\mathbb R}^d)$ with intensity measure $ \Psi(\rd u, \sd t,
\sd f) = c u^{-2} \sd u \, \sd t \, \sP_{\tilde F}(\rd f)$ for some $c >
0$ and some probability measure $\sP_{\tilde F}$.
2. $\xi$ has an M3 representation with $\sP_F(\rd f) = \sP_{\tilde F}(c\sd f)$ being the probability measure of the shape function $F$. The constant $c>0$ is given by $$\begin{aligned}
\label{eq:c}
c = \left(\int_{{\mathbb R}^d} \int_{C({\mathbb R}^d)} f(t) \, \sP_{\tilde F}(\rd f) \sd t\right)^{-1}
\end{aligned}$$ and the probability measure $\sP_{\tilde F}$ is defined by $$\sP_{\tilde F}(A) = \frac{\int_0^\infty y \sP(W(\cdot +\tau) / y \in A, \ \tau \in
K \mid \gamma=y) \, \sP_\gamma(\rd y)} {\int_0^\infty y \sP(\tau \in K
\mid \gamma=y) \, \sP_\gamma(\rd y)}$$ for any Borel set $A \subset C({\mathbb R}^d)$ and any compact set $K \subset {\mathbb R}^d$, where $\tau$ and $\gamma$ are defined as $\tau_i$ and $\gamma_i$, respectively, replacing $W_i$ by $W$, and $\sP_\gamma$ is the probability measure belonging to $\gamma$.
<!-- -->
1. Analogously to the proof of [@kab2009 Thm. 14].
2. From the first part it follows that $$\Phi_0 = \sum_{i \in {\mathbb N}} \delta_{(U_i \gamma_i / c,\, \tau_i,\,
c \cdot W_i(\cdot + \tau_i) / \gamma_i)}$$ is a PPP with intensity measure $\Psi_0(\rd u, \sd t, \sd f) = u^{-2} \sd u \times \sd t \times \, \sP_F(\rd f)$ where $\sP_F(\rd f) = \sP_{\tilde F}(c \sd f)$. Hence, $\Phi_0$ is of the same type as $\Pi_0$ from the beginning of Section \[M3representation\] and $$\xi(t) = \max_{(y,s,f) \in \Phi_0} y f(\cdot - s), \quad
t \in {\mathbb R}^d,$$ is a mixed moving maxima representation. The integrability condition follows from the fact that $\xi$ has standard Fréchet marginals. Thus, $$\int_{{\mathbb R}^d} \int_{C({\mathbb R}^d)} c f(t) \, \sP_{\tilde F}(\rd f) \sd t= 1,$$ which implies . In order to calculate $\sP_{\tilde F}$, let $A \in \mathcal B (C({\mathbb R}^d))$ and $K \in \mathcal{B}^d$ be compact. The first part of this Theorem implies that $$\Psi([1, \infty)\times K \times A) = c \cdot |K| \cdot \sP_{\tilde F}(A).$$ Therefore, $$\begin{aligned}
\label{eq:qlim}
\sP_{\tilde F}(A) = \frac{ \Psi([1, \infty) \times K \times A)}{
\Psi([1, \infty) \times K \times C({\mathbb R}^d))},
\end{aligned}$$ and both the enumerator and the denominator are finite. For the enumerator, we get $$\begin{aligned}
\Psi(&[1, \infty) \times K \times A)\\
={} & \int_0^\infty u^{-2} \int^{\infty}_{u^{-1}} \sP(W(\cdot + \tau) / \gamma \in A, \ \tau \in K \mid \gamma=y) \, \sP_\gamma(\rd y) \sd u\\
={} & \int_0^\infty \int^{\infty}_{y^{-1}} u^{-2} \sd u \cdot \sP(W(\cdot + \tau) / y \in A, \ \tau \in K \mid \gamma=y) \, \sP_\gamma(\rd y)\\
={} & \int_0^\infty y \sP(W(\cdot + \tau) / y \in A, \ \tau \in K \mid \gamma=y) \, \sP_\gamma(\rd y).
\end{aligned}$$ Thus, by , $$\begin{aligned}
\sP_{\tilde F}(A) = \frac{\int_0^\infty y \sP(W(\cdot + \tau) / y \in A, \ \tau \in K \mid \gamma=y) \, \sP_\gamma(\rd y)}{\int_0^\infty y \sP(\tau \in K \mid \gamma=y) \, \sP_\gamma(\rd y)},
\end{aligned}$$ which completes the proof.
Outlook: Statistical applications {#sec:application}
=================================
In univariate extreme value theory, a standard method for estimating the extreme value parameters fits all data exceeding a high threshold to a certain Poisson point process. This peaks-over-threshold approach has been generalized in [@roo2006] to the multivariate setting. Therein, generalized multivariate Pareto distributions are obtained as the max-limit of some multivariate random vector in the MDA of an extreme value distribution by conditioning on the event that at least one of the components is large. Conditioning on the same extremal events, the recent contribution [@fal2012] analyzes the asymptotic distribution of exceedance counts of stationary sequences.\
Here, we have suggested conditioning a stochastic process $\eta(t) : t\in
T\}$ in the MDA of a max-stable process $\{\xi(t) : t\in T\}$ such that it converges to the incremental processes $W$ in or the shape functions $F$ in . In this section we provide several examples how these theoretical results can be used for statistical inference. The approach is based on a multivariate peaks-over-threshold method for max-stable processes, though the definition of extreme events differs from that in [@roo2006; @fal2012].\
In the sequel, suppose that $\eta_1,\dots,\eta_n,$ $n\in{\mathbb N},$ are independent observations of the random process $\eta$, already normalized to standard Pareto margins.
Incremental representation {#incremental-representation}
--------------------------
For a max-stable process $\xi$ that admits an incremental representation $$\begin{aligned}
\label{def_xi_again}
\xi(t) = \max_{i\in{\mathbb N}} U_i W_i(t), \quad t\in T,\end{aligned}$$ as in , the statistical merit of the convergence results in Theorem \[theo\_cond\_increments\_general\] and Theorem \[theo\_cond\_increments\_cont\] is the “deconvolution” of $U$ and $W$ which allows to substitute estimation of $\xi$ by estimation of the process $W$. As only the single *extreme* events converge to $W$, we define the index set of extremal observations as $$\begin{aligned}
I_1(n) = \bigl\{ i\in \{1,\dots n\}: \ \eta_i(t_0) > a(n) \bigr\},\end{aligned}$$ for some fixed $t_0\in T$. The set $\{ \eta_i(\cdot) / \eta_i(t_0) :
\ i \in I_1(n) \}$ then represents a collection of independent random variables that approximately follow the distribution of $W$. Thus, once the representation in is known, both parametric and non-parametric estimation for the process $W$ is feasible. For statistical inference it is necessary that the number of extremal observations $|I_1(n)|$ converges to $\infty$, as $n\to\infty$. This is achieved by choosing the sequence of thresholds $a(n)$ according to Remark \[rem\_thres\].
The dependence parameter $q\geq 1$ of the symmetric logistic distribution can be estimated by perceiving the conditional increments of $\eta$ in the MDA as realizations of $W$ and maximizing the likelihood $$\begin{aligned}
&\sP\big(W(t_1) \in \rd s_1, \dots W(t_k) \in \rd s_k \,\big|\, q\big) \\
&= \left(1+\textstyle\sum_{i=1}^k s_i^{-q}\right)^{1/q -(k+1)}
\left(\textstyle\prod_{i=1}^k(iq-1)\right)
\textstyle\prod_{i=0}^k s_i^{-q-1} \ \rd s_1\dots \rd s_k.\end{aligned}$$
Recall that the Brown-Resnick processes in Example \[BRproc\] admit a representation with log-Gaussian incremental process $W(t) = \exp\left\{Y(t) - \sigma^2(t) / 2\right\}, t\in{\mathbb R}^d$. Hence, standard estimation procedures for Gaussian vectors or processes can be applied for statistical inference. [@eng2012a] explicitly construct several new estimators of the variogram $\gamma$ based on the incremental representation, which also covers Hüsler-Reiss distributions, and they provide some basic performance analyses.
Mixed moving maxima representation {#mixed-moving-maxima-representation}
----------------------------------
Similarly, in case of the mixed moving maxima representation $$\begin{aligned}
M(t) = \max_{i \in {\mathbb N}} U_i F_i(t - S_i), \quad t \in {\mathbb R}^d,\end{aligned}$$ the convergence results of Theorem \[thm\_conv\_mmm\] can be used to estimate $F$ (or $F_1 = F / \lambda$) on some compact domain $K$ instead of estimating $M$ directly. Here, the index set $T$ of the observed processes $\{\eta_i(t): \ t \in T\}$, $i=1,\ldots, n$, can be identified with ${Q}\oplus L$ from Theorem \[thm\_conv\_mmm\]. The set $L$ should be sufficiently large such that it is reasonable to assume that the components $\{U_i F_i(\cdot - S_i): \ S_i \notin {Q}\oplus L\}$ hardly affect the process $M$ on ${Q}\oplus K$ (that is, $\mu(C_B \cap A_L) / \mu(A_L) \approx \sP(F(\cdot) - \lambda \in B)$ in the proof of Theorem \[thm\_conv\_mmm\]). At the same time, a large set ${Q}$ leads to a rich set of usable observations $ \widetilde
F_1^{(i)} = \eta_i(\tau_{Q}(\eta_i)+\cdot) / \eta_i(\tau_{Q}(\eta_i)),
\ i \in I_2(n),$ where $$I_2(n) = \left\{ i \in \{1,\ldots,n\}: \ \max_{t\in {Q}}\eta_i(t) =
\max_{t \in {Q}\oplus L} \eta_i(t) \geq a(n)\right\}.$$ The resulting processes $\widetilde F_1^{(i)},\ i \in I_2(n),$ can be interpreted as independent samples from an approximation to $F_1$. This approach can be expected to be particularly promising in case of F having a simple distribution or even being deterministic.
Some examples of mixed moving maxima processes have already been analyzed for statistical inference by [@deh2006] who use normal, exponential and t densities as shape functions. More precisely, they consider M3 models with $$\begin{aligned}
F_1(t) &= \exp\left\{- \frac{\beta^2 t^2} 2\right\}, \quad &\lambda {}={}& \frac{\beta}{\sqrt{2\pi}}, \label{eq:mmm-ex-1}\\
F_1(t) &= \exp\left\{- \beta |t|\right\}, \quad &\lambda {}={}& \frac{\beta}{2}, \label{eq:mmm-ex-2}\\
\text{and} \quad F_1(t) &= \left( 1 + \frac{\beta^2 t^2}{\nu} \right)^{- \frac{\nu+1} 2}, \quad
&\lambda {}={}& \frac{\beta \Gamma\left( \frac{\nu+1} 2\right)}{\sqrt{\pi\nu} \Gamma\left(\frac \nu 2\right)}\, \quad \nu > 0, \label{eq:mmm-ex-3}\end{aligned}$$ all parametrized by $\beta > 0$. [@deh2006] introduce consistent and asymptotically normal estimators based on the interpretation of $\beta$ as a dependence parameter. From the samples $\widetilde F_1^{(i)}$, $i \in I_2(n)$, we get a new estimator $$\widehat F_1 = \frac 1 {|I_2(n)|} \sum_{i \in I_2} \widetilde F_1^{(i)}$$ for $F_1$. Applying this estimator, $\beta$ can be estimated by a least squares fit of – to $\widehat F_1$ at some locations $t_1, \ldots t_m \in K$. Note that in case of the normal model and the exponential model , the logarithm of the shape function $F_1$ depends linearly on $\beta^2$ and $\beta$, respectively, and $\log \widehat F_1$ can be fitted by ordinary least squares.
The mixed moving maxima representation can also be employed for estimation of Brown-Resnick processes although the distribution of $F$ is much more sophisticated than the one of $W$ in the incremental representation (cf. [@eng2011; @oes2012]). A relation between the shape function $F$ and the variogram $\gamma$ of the Brown-Resnick process can be obtained via the *extremal coefficient function* $\theta(\cdot)$. For a stationary, max-stable process $\xi$ with identically distributed marginals, [@sch2003] defined the extremal coefficient function $\theta$ via the relation $$\begin{aligned}
\sP(\xi(0)\leq u,\, \xi(h) \leq u) = \sP(\xi(0) \leq u)^{\theta(h)},
\quad h \in {\mathbb R}^d.\end{aligned}$$ For mixed moving maxima processes, we have $$\begin{aligned}
\label{eq:theta-mmm}
\theta(h)
=\sE\left. \int_{{\mathbb R}^d} \{ F(t) \vee F(t+h)\} \sd t\right.
=\frac{\sE\left. \int_{{\mathbb R}^d} \{ F_1(t) \vee F_1(t+h)\} \sd t\right.}{
\sE\left.\int_{{\mathbb R}^d} F_1(t) \sd t\right.}\end{aligned}$$ and, at the same time, for Brown-Resnick processes [@kab2009], $$\begin{aligned}
\label{eq:theta-br}
\theta(h) = 2\Phi\left(\sqrt{\gamma(h)}/ 2\right),\end{aligned}$$ where $\Phi$ is the standard Gaussian distribution function. Identifying with and plugging in the samples $\widetilde F_1^{(i)}$, $i \in I_2(n)$, we get the variogram estimator $$\begin{aligned}
\widehat \gamma(h) =
\left\{2 \Phi^{-1}\left(
\frac{\displaystyle \sum_{i \in I_2(n)} \int_{\widetilde K} \widetilde F_1^{(i)}(t)
\vee \widetilde F_1^{(i)}(t+h) \sd t}{2 \displaystyle \sum_{i \in I_2(n)} \int_{\widetilde K} \widetilde F_1^{(i)}(t) \sd t}
\right)\right\}^2,\end{aligned}$$ where $\widetilde K$ is a large set such that $\widetilde K, \widetilde K +h \subset K$.
Auxiliary Results for the Proof of Theorem \[thm\_conv\_mmm\]
=============================================================
\[lem:AL\] $A_L$ is closed. The set $A_L^*$ is in the interior of $A_L$.
The first assertion is obvious. For the second one, let $f^* \in
A_L^*$. Then, we have $f^*(\tau_{Q}(f^*|_{Q})) =: \alpha >
1$. Furthermore, there is $\delta > 0$ such that $B_\delta(\tau_{Q}(f^*|_{Q})) = \{t \in {\mathbb R}^d: \ ||t - \tau_{Q}(f|_{Q})||
< \delta \} \in {Q}^o$ and we have $$\beta := \sup_{t \in {Q}\oplus L \setminus B_\delta(\tau_{Q}(f^*|_{Q}))} f^*(t) - \max_{t\in {Q}}f^*(t) < 0.$$ Now, we choose $\varepsilon < \min \{ \frac {\alpha-1} 2, \frac {|\beta|} 2\}$ and show that $B_\varepsilon(f^*) = \{ f \in C({Q}\oplus L):\ ||f - f^*||_\infty < \varepsilon\} \subset A_L$. This holds, as for any $f \in B_\varepsilon(f^*)$, we have $$\begin{aligned}
& f(\tau_{Q}(f|_{Q})) \geq f(\tau_{Q}(f^*|_{Q})) > \alpha - \varepsilon > \frac {1+\alpha }2 > 1\\
\text{and} \quad & \max_{t\in {Q}}f(t) \leq \max_{t \in {Q}\oplus L} f(t) = \max\left\{\max_{t \in {Q}\oplus L \setminus {Q}^o} f(t), \ \max_{t\in {Q}}f(t)\right\}\\
& {}\leq{} \max\left\{ \beta + \alpha + \varepsilon, \ \max_{t\in {Q}}f(t)\right\} {}\leq{} \max\left\{ \alpha - \varepsilon, \ \max_{t\in {Q}}f(t)\right\} {}={} \max_{t\in {Q}}f(t),
\end{aligned}$$ which means equality.
\[lem:interCB\] The set $C_B^*$ is in the interior of $C_B$.
Let $f^* \in C_B^*$. Then, $t^* = \operatorname*{arg\,max}_{t \in {Q}} f^*(t)$ is well-defined and necessarily, as $f \geq 0$, $$\label{eq:alpha}
\alpha := f^*(t^*) \in (0,||f^*||_\infty].$$ Since $f^*(t^* + \cdot) / f^*(t^*) \in B^o$, there is some $\varepsilon > 0$ such that $$\label{eq:ball}
\left\{f \in C(K): \ \left|\left|f^*(t^* + \cdot) \Big/ f^*(t^*) - f\right|\right|_\infty < \varepsilon\right\} \subset B.$$ Furthermore, $f^*$ is uniformly continuous on the compact set ${Q}\oplus L$, i.e. there exists some $\delta > 0$ such that $$\label{eq:unifcont}
\sup_{s,t \in {Q}\oplus L, \ ||s-t|| < \delta} |f^*(s) - f^*(t)| < \frac \varepsilon 3 \alpha.$$ Then, as $\operatorname*{arg\,max}_{t \in {Q}} f^*(t)$ is unique, we have that $$\label{eq:beta}
\beta := \max_{t \in {Q}\setminus \{t \in {\mathbb R}^d: \ ||t-t^*|| < \delta\}} f^*(t) - f^*(t^*) \in [-\alpha,0).$$ Choose $\varepsilon^* < \min\left\{\frac {|\beta|} {2\alpha}, \frac \varepsilon 6 \frac \alpha {||f^*||_\infty}\right\}$. We will show that $B_{\varepsilon^*\alpha}(f^*) =\{f \in C({Q}\oplus L): \ ||f-f^*||_\infty < \varepsilon^*\alpha\} \subset C_B$. To this end, let $f_0 \in B_{\varepsilon^*\alpha}(f^*)$. Then, because of Equation and $\varepsilon^*\alpha < \frac {|\beta|} 2$, we have that $||t_0 - t^*|| \leq \delta$ for $t_0 = \tau_{Q}(f_0|_{Q})$. Therefore, $$\begin{aligned}
& \sup_{t \in K} \left| \frac{f^*(t^*+ \cdot)}{f^*(t^*)} - \frac{f_0(t_0 + \cdot)}{f_0(t_0)} \right| \notag\\
\leq{} & \sup_{t \in K} \left| \frac{f^*(t^*+\cdot)}{f^*(t^*)} - \frac{f^*(t_0+\cdot)}{f^*(t^*)}\right|
+ \sup_{t \in K} \left| \frac{f^*(t_0+\cdot)}{f^*(t^*)} - \frac{f^*(t_0+\cdot)}{f_0(t_0)}\right| \notag\\
& + \sup_{t \in K} \left| \frac{f^*(t_0+\cdot)}{f_0(t_0)} - \frac{f_0(t_0+\cdot)}{f_0(t_0)}\right| \qquad
\leq \qquad \frac \varepsilon 3 + \frac{\varepsilon^*}{1-\varepsilon^*} \frac{||f^*||_\infty}{\alpha} + \frac{\varepsilon^*}{1-\varepsilon^*}, \label{eq:dist}
\end{aligned}$$ where we used Equation and the fact that $f_0 \in B_{\varepsilon^*\alpha}(f^*)$. Equations and and the choice of $\varepsilon^*$ yield that $\varepsilon^*/(1-\varepsilon^*) \leq (\varepsilon^* ||f^*||_\infty)/((1-\varepsilon^*)\alpha) \leq 2\varepsilon^*||f^*||_\infty / \alpha < \varepsilon/3$, i.e. each summand on the right-hand side of is smaller than $\varepsilon/3$. Thus, $f_0(t_0 + \cdot) / f_0(t_0) \in \{f \in C(K): \ ||f^*(t^* + \cdot) / f^*(t^*) - f||_\infty < \varepsilon\} \subset B$ by Equation and $f_0 \in C_B$.
\[lem:CB\] The closure of $C_B$ is a subset of $$\begin{aligned}
B^* \cup{} & \left\{f \in C({Q}\oplus L): \, \operatorname*{arg\,max}_{t \in {Q}} f(t) \text{ is not unique} \right\}\\
\cup{} & \left\{f \in C({Q}\oplus L): \, f(\tau_{Q}(f|_{Q}) + \cdot) \Big/ f(\tau_{Q}(f|_{Q})) \in \partial B \right\}.
\end{aligned}$$
Let $\{f_n\} \subset C_B$ be a sequence converging uniformly to some $f^* \in C({Q}\oplus L)$. We have to verify that $f^*(\tau_{Q}(f^*|_{Q}) + \cdot) / f^*(\tau_{Q}(f^*|_{Q})) \in B \cup \partial B$ if $\operatorname*{arg\,max}_{t \in {Q}} f^*(t)$ is unique. Analogously to the proof of Lemma \[lem:interCB\] we can show that for any $\varepsilon_2 > 0$ there is some $\varepsilon_1 > 0$ such that $$\begin{aligned}
& ||f-f^*||_{\infty,{Q}\oplus L} < \varepsilon_1\\
{}\Longrightarrow{} \quad & \left|\left|\frac{f(\tau_{Q}(f|_{Q})+\cdot)}{f(\tau_{Q}(f|_{Q}))} - \frac{f^*(\tau_{Q}(f^*|_{Q})+\cdot)}{f^*(\tau_{Q}(f^*|_{Q}))}\right|\right|_{\infty,K} < \varepsilon_2.
\end{aligned}$$ Thus, $f_n(\tau_{Q}(f_n|_{Q}) + \cdot)/f_n(\tau_{Q}(f_n|_{Q}))$ converges to $f^*(\tau_{Q}(f^*|_{Q}) + \cdot)/f^*(\tau_{Q}(f^*|_{Q}))$ in $C(K)$. Hence, as $B \cup \partial B$ is closed, $f^*(\tau_{Q}(f^*|_{Q}) + \cdot)/f^*(\tau_{Q}(f^*|_{Q})) \in B \cup \partial B$.
Proof of Example \[ex:mmm-mda\] {#sec:proof_ex_mmm-mda}
===============================
Let $\tilde M_j$, $j\in{\mathbb N}$, be independent copies of the process $\tilde M$ and consider $$M_n(\cdot) = \frac{1}{cn} \max_{i=1}^n \tilde M_i(\cdot).$$ Further, suppose that $L \subset {\mathbb R}^d$ is an arbitrary compact set. Note that by Remark \[weak\_conv\_Rd\] it suffices to show weak convergence of $M_n \cvgdist M$, $n\to\infty$, on $C(L)$.
To prove the first assertion, note that, for $t_0, \ldots, t_k\in{\mathbb R}^d$, $s_0, \ldots, s_k\geq 0$, $k\in{\mathbb N}$, we have $$\begin{aligned}
&\sP(M_n(t_0)\leq s_0, \ldots, M_n(t_k)\leq s_k) \notag\\
&= \left[{\mathbf 1}_{\kappa \leq \min_{l=0}^k cns_l }\cdot \sP\left\{ \Pi_{c, \epsilon}
\left(\left\{(u,t,f):
\max_{l=0}^k u f(t_l-t)/(cns_l) > 1\right\}\right) = 0\right\}\right]^n \notag\\
&= {\mathbf 1}_{\kappa \leq \min_{l=0}^k cns_l }\cdot \exp\left(- n\int_{C({\mathbb R}^d)} \int_{{\mathbb R}^d}
\min\left\{\frac{1}{\epsilon}, \max_{l=0}^k \frac{f(t_l-t)}{cns_l}\right\}\, c\sd t \, \sP_F(\rd f) \right)\notag\\
&\longrightarrow
\exp\left(- \int_{C({\mathbb R}^d)} \int_{{\mathbb R}^d}
\max_{l=0}^k (f(t_l-t)/s_l) \sd t \, \sP_F(\rd f) \right), \label{proofMDAM3}\end{aligned}$$ as $n\to\infty$, where the convergence holds due to monotone convergence. The right-hand side of coincides with the marginal distribution of $M$ (cf. ).
For convergence of $M_n$ to $M$ in the sense of weak convergence in $C^+(L)$ endowed with the topology of uniform convergence, it remains to show that the sequence of restricted processes $\{M_n|_L : n\in {\mathbb N}\}$ is tight. To this end, by [@bil1999 Thm. 7.3], it suffices to verify that for any $\varepsilon > 0$, $\eta \in (0,1)$, there exist $\delta > 0$, $n_0 \in {\mathbb N}$ such that $$\sP\left\{ \sup_{||s-t|| < \delta} |M_n(s) - M_n(t)| \geq \varepsilon\right\} \leq \eta, \qquad n \geq n_0.$$ By Equation , we can choose $R > 0$ such that $$\label{eq:smallintegral}
\int_{{\mathbb R}^d \setminus (L \oplus B_R({\mathbf 0}))} \sE\left( \sup_{t \in L} F(t-s) \right) \sd s < \frac {\varepsilon \eta} 2,$$ where $B_R({\mathbf 0}) = \{ x \in {\mathbb R}^d: \ ||x|| \leq R\}$. Furthermore, implies that $\sE\left( \sup_{t \in K} F(s)\right) < \infty$ for any compact set $K \subset {\mathbb R}^d$. Therefore, as each realization of $F$ is uniformly continuous on $B_{R+d(L)}({\mathbf 0})$, where $d(L) = \sup_{s_1,s_2 \in L} ||s_1-s_2||$ denotes the diameter of $L$, dominated convergence yields $$\lim_{\delta \searrow 0} \sE\left( \sup_{s,t \in B_{R+d(L)}({\mathbf 0}), \ ||s-t|| < \delta} |F(s) - F(t)|\right) = 0.$$ In particular, we can choose $\delta > 0$ such that $$\label{eq:delta}
\sE\left( \sup_{s_1,s_2 \in B_{R+d(L)}({\mathbf 0}), \ ||s_1-s_2|| < \delta} |F(s_1) - F(s_2)|\right) < \frac {\varepsilon \eta}{2 |L \oplus B_R({\mathbf 0})|}.$$ Then, we get $$\begin{aligned}
& \sP\left\{ \sup_{||s_1-s_2|| < \delta, \ s_1,s_2 \in L} | M_n(s_1) - M_n(s_2) | \geq \varepsilon \right\} \\
\leq{} & n \sP\left\{ \sup_{||s_1-s_2|| < \delta, \ s_1,s_2 \in L} |\tilde M_n(s_1) - \tilde M_n(s_2) | \geq c n\varepsilon \right\} \\
\leq{}& n \bigg( \sP\bigg\{ \Pi\bigg(\bigg\{ (u,t,f): \ t \in L \oplus B_R({\mathbf 0}), \\
& \hspace{3cm} \sup_{\substack{s_1,s_2 \in B_{R+d(L)}({\mathbf 0}),\\ ||s_1-s_2|| < \delta} } |f(s_1)-f(s_2)| > \frac{cn\varepsilon}{u}
\bigg\}\bigg) > 0\bigg\}\\
& + \sP\bigg\{ \Pi\bigg(\bigg\{ (u,t,f): \ t \in {\mathbb R}^d \setminus (L \oplus B_R({\mathbf 0})), \sup_{s \in L} |f(s-t)| > \frac{cn\varepsilon}{u} \bigg\} \bigg) > 0\bigg\}\bigg) \displaybreak[0]\\
\leq{} & n\bigg( 1 - \exp\bigg\{ - \int_{L \oplus B_R({\mathbf 0})} \int_\epsilon^\infty u^{-2} \\
& \hspace{3.2cm} \cdot \sP\bigg( \sup_{\substack{s_1,s_2 \in B_{R+d(L)}({\mathbf 0}),\\ ||s_1-s_2|| < \delta} } |F(s_1)-F(s_2)| > \frac{c n\varepsilon}{u}\bigg) \sd u \, c \sd t \bigg\}\\
& + 1- \exp\bigg\{ - \int_{{\mathbb R}^d \setminus (L \oplus B_R({\mathbf 0}))} \int_\epsilon^\infty u^{-2}
\sP\bigg( \sup_{s \in L} |F(s-t)| > \frac{c n\varepsilon}{u}\bigg) \sd u \, c \sd t \bigg\} \bigg) \displaybreak[0]\\
\leq{} & n\bigg( 1 - \exp\bigg( - \frac{|L \oplus B_R({\mathbf 0})|}{n \varepsilon}
\sE\bigg\{ \sup_{\substack{s_1,s_2 \in B_{R+d(L)}({\mathbf 0}),\\ ||s_1-s_2|| < \delta} } |F(s_1)-F(s_2)| \bigg\} \bigg)\\
& + 1 - \exp\bigg(- \frac 1 {n\varepsilon} \int_{{\mathbb R}\setminus (L \oplus B_R({\mathbf 0}))} \sE\left\{ \sup_{s \in L} |F(s-t)| \right\} \sd t \bigg) \bigg)\\
\leq{} & n\left( 1 - \exp\left(- \frac \eta {2n}\right) + 1 - \exp\left(- \frac \eta {2n}\right)\right) \quad \leq \quad \eta,
\end{aligned}$$ where we used Equation and . Thus, the sequence of processes $\{M_n|_L : n\in {\mathbb N}\}$ is tight.
The authors are grateful to Zakhar Kabluchko for useful suggestions and hints. S. Engelke has been financially supported by Deutsche Telekom Stiftung. A. Malinowski has been financially supported the German Science Foundation (DFG), Research Training Group 1644 ‘Scaling problems in Statistics’. M. Oesting and M. Schlather have been financially supported by Volkswagen Stiftung within the project ‘WEX-MOP’.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'It has recently shown that in supersymmetric left-right models with automatic R-parity conservation, the theory below the $W_R$ scale is given by MSSM with massive neutrinos and a pair of doubly charged superfields with masses in the 100 GeV range ( with or without an extra pair of heavy Higgs doublets (M$\geq$ 10 TeV) depending on the model). In this paper we study the unification prospects for such theories and their phenomenological implications for collider experiments. We study two versions of the theory, one with supersymmetry breaking transmitted via the gauge and another where the same occurs via gravitational forces. We point out that looking at multi $\tau$ final states can considerably constrain the parameter space of the model.'
address: |
$^1$Institute of Theoretical Sciences, University of Oregon, Eugene, OR-97403, USA.\
$^2$ Department of Physics, University of Maryland, College Park, MD-20742, USA.
author:
- 'B. Dutta$^1$ and R. N. Mohapatra$^2$'
date: 'April, 1998'
title: |
Phenomenology of light remnant doubly charged Higgs fields\
in the supersymmetric left-right model
---
epsf.tex (\#1 width \#2)[=\#2 ]{}
Introduction
============
Supersymmetric left-right models (SUSYLR) where the $SU(2)_R$ gauge symmetry is broken by triplet Higgs fields with $B-L=2$ have many attractive features: 1) they imply automatic conservation of baryon and lepton number[@moh], a property which makes the standard model so attractive, but is not shared by the minimal supersymmetric standard model (MSSM); 2) they provide a natural solution to the strong and weak CP problems of the MSSM[@rasin]; 3) They yield a natural embedding of the see-saw mechanism for small neutrino masses[@gell] where the right handed triplet field ($\Delta^c$) that breaks the $SU(2)_R$ symmetry also gives heavy mass to the righthanded Majorana neutrino needed for implementing the see-saw mechanism.
In order to cancel anomalies as well as to maintain supersymmetry below the $SU(2)_R$ scale ($v_R$), one needs a pair of fields $\Delta^c\oplus
\overline{\Delta^c}$ with $B-L=\mp 2$ respectively. An important distinguishing characteristic of these Higgs multiplets is that they contain doubly charged Higgs bosons and Higgsinos in them that remain as physical fields subsequent to symmetry breakdown. It has recently been shown that the vacuum of the theory may or may not conserve R-parity[@kuchi]. If however we require the ground state to conserve R-parity , one must include higher dimensional operators[@rasin; @aulakh] or additional Higgs fields which break parity[@kuchi; @goran2]. In this case, [@goran2; @chacko] the model in its simple versions, always predicts that some of the doubly charged fields mentioned above are massless in the absence of the higher dimensional operators (HDO). This is independent of whether the hidden sector supersymmetry breaking scale is above or below the $W_R$ scale. In the presence of HDO’s, they acquire masses of order $\sim v^2_R/M_{Pl}$. Since the measurement of the Z-width at LEP and SLC implies that such particles must have a mass of at least 45 GeV, this puts a lower limit on the $W_R$ scale of about $10^{10}$ GeV or so. This result is interesting since $W_R$ masses in this range also lead to neutrino masses expected on the basis of current solar and atmospheric neutrino experiments. If we take the $W_R$ mass to be close to this lower limit (say of order $10^{10}-10^{11}$ GeV), it implies the masses of the doubly charged particles in the 100 GeV range. There are two bosonic and two fermionic particles of this type. The rest of the particle spectrum below the $W_R$ scale can be same as that of the minimal supersymmetric standard model (MSSM) with a massive neutrino or it can have an extra pair of Higgs doublets in the 10 TeV range depending on the structure of the model. It is the goal of this paper to explore the constraints on the parameters of the model and suggest tests in the $e^+e^-$ and $p\bar{p}$ collider.
Our main results are that for a large range of parameters in this model the stau (the superpartner of $\tau$) is the lightest of the sleptons due to the renormalization group running arising from the $\Delta^{++}\tau^-\tau^-$ coupling which is not very constrained from phenomenology. (Note that the $\Delta^{++}$ couplings to other leptons are severely bounded by the recent PSI results on muonium- antimuonium oscillation[@jungman]). As a result, tau lepton final states, generated from the production of the $\Delta^{\pm\pm}$ and its fermionic part, provide a crucial signature of this class of models. For instance we find that detection of final states of type $\tau^-\tau^-\tau^+\tau^+\gamma
\gamma$ plus missing energy or $\tau^-\tau^-\tau^+\tau^+$ with or without missing energy in both $p\bar{p}$ and $e^+e^-$ collsion will provide test of these models. Thus non-observation of such signals will significantly reduce the domain of allowed parameters of this model. We point out the difference between the multi $\tau$ signals of this model and the same type of signals that appear in the other models e. g. conventional GMSB type. We will show that the allowed region of parameter space is larger when we choose GMSB type of theories. We also discuss the unification prospects of this model in different SUSY breaking scenarios.
This paper is organized as follows: in section 2, we review the arguments leading to the existence of light doubly charged fields despite a high $W_R$ scale in the context of a simple model and discuss the low energy interactions of these fields; in section 3 we discuss gauge unification in these models; in section 4, We discuss its parameter space and experimental signals in gravity mediated SUSY breaking scenario; in section 5, we discuss the same for the gauge mediated SUSY breaking scenarios and in section 6 we present our conclusion.
Overview of the model
======================
In this section, we present a brief review of the arguments leading to the existence of the light doubly charged Higgs fields in the SUSYLR model. In order to give our arguments, we start by giving the basic features of the model, which is based on the gauge group $SU(2)_L\times SU(2)_R\times U(1)_{B-L}\times SU(3)_c$. In Table I, we give the particle content of the model. We will suppress the $SU(3)_c$ indices in what follows.\
------------------ ------------------------------ -------------------------------
Fields SU$(2)_L \, \times$ SU$(2)_R group transformation
\, \times$ U$(1)_{B-L}$
representation
Q (2,1,$+ {1 UQ
\over 3}$)
$Q^c$ (1,2,$- {1 \over 3}$) VQ$^c$
L (2,1,$- 1$) UL
$L^c$ (1,2,+ 1) V$L^c$
$\Phi_{1,2}$ (2,2,0) $U\phi V^{\dagger}$
$\Delta$ (3,1,+ 2) U$\Delta U^{\dagger}$
$\bar{\Delta}$ (3,1,$- 2$) $U\bar{\Delta}U^{\dagger}$
$\Delta^c$ (1,3,+ 2) $V\Delta^c V^{\dagger}$
$\bar{\Delta}^c$ (1,3,$- 2$) $ V\bar{\Delta}^cV^{\dagger}$
$S$ (1,1,0) $S$
------------------ ------------------------------ -------------------------------
\
Table 1:Field content of the SUSY LR model; we assume that $S$\
is odd under parity; $U$ and $V$ denote the $SU(2)_{L,R}$ transformations respectively.
The superpotential for this theory is given by (we have suppressed the generation index):
$$\begin{aligned}
W & = & {\bf h}^{(i)}_q Q^T \tau_2 \Phi_i \tau_2 Q^c + {\bf
h}^{(i)}_l L^T \tau_2 \Phi_i \tau_2 L^c
\nonumber\\
& + & i ( {\bf f} L^T \tau_2 \Delta L + {\bf f}_c {L^c}^T \tau_2 \Delta^c
L^c)
\nonumber\\
& + & M_{\Delta} [{\rm Tr} ( \Delta \bar{\Delta} ) +
{\rm Tr} ( \Delta^c \bar{\Delta}^c )] +\lambda S(\Delta\overline{\Delta}
-\Delta^c\overline{\Delta^c}) + \mu_S S^2 \nonumber\\
& + &
\mu_{ij} {\rm Tr} ( \tau_2 \Phi^T_i \tau_2 \Phi_j )+ W_{\it NR}
\label{eq:superpot}\end{aligned}$$
where $W_{\it NR}$ denotes non-renormalizable terms arising from higher scale physics such as grand unified theories or Planck scale effects.
$$\begin{aligned}
W_{\it NR}= A [Tr(\Delta^c \overline {\Delta}^c)]^2/2 + B Tr(\Delta^c
\Delta^c) Tr (\overline{\Delta}^c \overline{\Delta}^c)/2\end{aligned}$$
where A and B are of order 1/$M_{Planck}$.
We will work in the vacuum which conserves R-parity. The Higgs vevs then have the following pattern: $$\begin{aligned}
<\phi>=\left(\begin{array}{cc}
\kappa & 0\\ 0 & \kappa '\end{array}\right); <\Delta^c>=\left(\begin{array}{cc}
0 & v \\ 0 & 0 \end{array}\right)\end{aligned}$$ Similar pattern for $<\overline{\Delta^c}>$ is assumed.
Using Eq. 1, one can give a group theoretical argument for the existence of light doubly charged particles in the supersymmetric limit as follows. Let us first ignore the higher dimensional terms $A$ and $B$ as well as the leptonic couplings $f$. It is then clear that the superpotential has a complexified $U(3)$ symmetry (i.e. a $U(3)$ symmetry whose parameters are taken to be complex) that operates on the $\Delta^c$ and $\bar\Delta^c$ fields. This is due to the holomorphy of the superpotential. After one component of each of the above fields acquires vev the resulting symmetry is the complexified $U(2)$. This leaves 10 massless fields. Once we bring in the D-terms and switch on the gauge fields, six of these fields become massive as a consequence of the Higgs mechanism of supersymmetric theories. That leaves four massless fields in the absence of higher dimensional terms. These are the two complex doubly charged fields. Of the two non-renormalizable terms $A$ and $B$, only the A-term has the complexified $U(3)$ symmetry. Hence the supersymmetric contribution to the doubly charged particles will come only from the B-term. It is then clear that the masses of the doubly charged fields are of order $v^2_R/M_{Pl}$. Requiring that these masses satisfy the Z-width bound then implies that $v_R\geq
10^{10}-10^{11}$ GeV or so. In this paper we will assume that $v_R$ is at the lower bound value so that the doubly charged fields are accessibly to the existing collider experiments. Note incidentally that although the leptonic couplings do not respect the above mentioned symmetry, they are unimportant in determining the vacuum structure as long as R-parity is conserved and hence they do not effect the doubly charged field masses.
Let us now give an explicit calculation of the masses of the doubly charged fields in the supersymmetric limit using the superpotential in Eq.(1). Let us write down the F-terms for the $S$, $\Delta$ and $\Delta^c$ terms: $$\begin{aligned}
F_S=2\mu_S S +\lambda (\Delta
\overline{\Delta}-\Delta^c\overline{\Delta^c})
\nonumber \\ F_{\Delta} = (\lambda S + M_{\Delta} ) \overline{\Delta} \nonumber
\\ F_{\Delta^c}= (-\lambda S + M_{\Delta})\overline{\Delta^c} \end{aligned}$$ If the effective supersymmetry breaking scale is below the $W_R$ scale, then these $F$ terms must vanish. It is then clear that if we choose the $\Delta^c$ and $\overline{\Delta^c}$ vev’s (denoted by $v_R$ and $\bar{v}_R$) to be nonvanishing (and they are equal in the supersymmetric limit), then we must have $<S>=M_{\Delta}/\lambda$. This implies that the $\Delta$ and $\overline{\Delta}$ vev’s vanish and the masses of these fields are of order $2M_{\Delta}$. Thus the left triplet fields decouple from the low energy spectrum. It is then easy to see from the superpotential (in the absence of the A and B terms) that all the particles in the superfields $\Delta^c$ and $\overline{\Delta^c}$ are massless in the limit of exact supersymmetry. One linear combination of the neutral fields and another ofsingly charged fields disappear due to the Higgs mechanism. The remaining singly charged and neutral Higgs fields pick up mass of order of $v_R$ and disappear from the low energy spectrum.
The story of the doubly charged fields is however very different in this theory as has been shown in Ref.[@chacko]. Once supersymmetry breaking is turned on but the higher dimensional terms $A$ and $B$ are excluded from the analysis, these fields acquire negative mass-squares signalling the breakdown of electric charge. This problem is cured as soon as the $A$ and $B$ terms are included. The doubly charged fields (all of them) then acquire masses of order $v^2_R/M_{Pl}$ and the vacuum becomes charge conserving.
Let us now discuss the Higgs doublet spectrum of the model at low energies. At the $W_R$ scale, one generally takes two bi-doublet fields $\phi$’s to make the model realistic. In order to get the MSSM at low energies, one must decouple one pair of $H_{u}$ and $H_d$ from the low energy spectrum. This has been called doublet-doublet splitting problem in literature. In the model without HDO contributions, it is clear from the superpotential in Eq. (1) that doublet Higgsino matrix is symmetric: $$\begin{aligned}
M_H= \left(\begin{array}{cc}
\mu_{11} & \mu_{12}\\
\mu_{12} & \mu_{22} \end{array}\right)\end{aligned}$$ If we now do fine tuning to get one pair of $H_{u,d}$ at low energies, the $H_{u,d}$ appear as identical combinations of the doublets in $\phi_i$’s. As a result, at the MSSM level, we have proportionality of the $M_u$ and $M_d$ leading to vanishing CKM angles. This result holds even if we increase the number of bi-doublets arbitrarily and uses only the fact that bilinear mass matrix $\mu_{ij}$ is symmetric. This in fact raises the interesting possibility[@rabi] that all mixing angles in the quark and lepton sector may arise purely out of radiative corrections involving the soft breaking terms[@raby]. An advantage of this version of the model is that there are no new flavor changing effects other than those from the usual supersymmetric sources[@masiero].
On the other hand, one may choose the $\mu_{ij}$ parameters of this model to be of the order of electroweak scale so that the low energy model is not exactly the MSSM but rather the two Higgs pair extension of MSSM. The phenomenology of these models are very similar to the previous case except that there are new contributions to the flavor changing neutral current effects in this model similar to those in the nonsupersymmetric left-right models[@grimus] which puts a lower limit on the masses of the second pair of Higgs doublets to be in the 5-10 TeV range. As a result they will essentially decouple from the low energy spectrum.
Our results are independent of which of the above choices for the Higgs sector is made, except that unification discussion applies only to the second version.
Gauge unification
==================
The presence of the doubly charged fields at low energies distinguishes the gauge coupling evolution in this model from the MSSM and one might expect that one will lose the unification property. It however turns out that the gauge couplings do unify in this model, albeit at a lower scale as we see below, for the case which has two pairs of Higgs doublets at the weak scale. The couplings evolve according to their respective beta functions. As just mentioned, below the $v_R$ scale, we assume four Higgs doublet fields instead of the usual two of MSSM and we assume that one of the doublet pairs has mass of 10 TeV. They lead to a trivial modification of the beta function below $v_R$. The beta functions above the $v_R$ scale are given below for one loop : $$\begin{aligned}
b^{1223}_{i}=\left( \matrix{ 0\cr-6\cr -6\cr -9\cr} \right) +N_F
\left( \matrix{ 2\cr 2\cr 2\cr 2\cr} \right) +n_{\Phi} \left( \matrix{ 0\cr
1\cr 1\cr 0\cr} \right) +n_{\Delta} \left( \matrix{ 9\cr 4\cr 0\cr 0\cr}
\right) + n_{\Delta^c}
\left(
\matrix{ 9\cr 0\cr 4\cr 0\cr} \right) \ \ \, ,\end{aligned}$$ and in the following equation for the two-loop: $$\begin{aligned}
b^{1223}_{ij}&=&\left( \matrix{ 0&0&0&0\cr 0&-24&0&0\cr
0&0&-24&0\cr 0&0&0&-54\cr} \right) + N_F \left( \matrix{ 7/3&3&3&8/3\cr
1&14&0&8\cr 1&0&14&8\cr 1/3&3&3&68/3\cr}\right) + n_{\Phi} \left( \matrix{
0&0&{0}&0\cr 0&{7}&{3}&0\cr {0}&{3}&{7}&0\cr 0&0&0&0\cr}\right)+ n_{\Delta}
\left( \matrix{ 54&72&0&{0}\cr 24&{48}& {0}&0\cr {0}&{0}&{0}&0\cr 0&0&0&0 \cr}
\right)\\\nonumber &&+n_{\Delta^c} \left( \matrix{ 54&0& {72}&0\cr
0&{0}&{0}&0\cr {24}&{0}&{48}&0\cr 0&0&0&0\cr} \right) \, , \end{aligned}$$ where $i=U(1)_{B-L},SU(2)_L,SU(2)_R,SU(3)_C$ respectively in the matrices, $N_F$ is the number of fermion generations. $N_F=3$ always and $n_{\phi}$ is the number of bidoublets which we take to be 2. We also take $n_\Delta$(${\Delta}
+ \bar\Delta$) and $n_{\Delta^c}$($\Delta^c+\bar\Delta^c$) to be 1. Since the $\Delta ^{\pm\pm}$ leaks down to the weak scale, we need to include its contribution to the running of the gauge couplings in between the weak and the intermediate scale. Since this field $\Delta$ has only hypercharge quantum number under the SM representation, the hypercharge gauge coupling RGE gets an extra term of 24/5 in one loop and in the two loop, the hypercharge squared elements gets and additional factor $ 72\times 16/25$.
We see from figure 1a. that the gauge couplings unify at a scale $\sim 10^{11}$ GeV and the intermediate scale is $\sim 10^8$ GeV. We take $\alpha_c=0.118$, $\alpha=1/128.7$ and $\sin^2\theta_w=0.2321$ at the weak scale. The low unification scale implies that the proton decay constraint would rule out groups like SO(10). However the final unifying group can be $SU(3)\times SU(3)\times SU(3)$ or any group that conserves baryon number.
In the supergravity motivated models, the gauge unification is necessary in order to have gaugino mass unification. The masses of all the sparticles can then be determined in terms of the parameters e.g. universal scalar mass $m_0$ at the unification, universal gaugino mass $m_{1/2}$ and the triliniear coefficient in the potential A’s. We will assume the A’s to be 0. We choose to assume the universality of $m_0$ at the unification scale rather than at the Planck scale for two reasons: first is that we do not know the theory above the unification scale and the nature of the evolution of the parameters is obviously dependent on those details. The second reason is that the experimental signature we are interested in involves only the tau lepton and its partner and not the other super particles of the theory. So even if we assumed universality of scalar masses at the Planck scale which would necessarily imply some splitting between generations due to running between the Planck scale to the unification scale, our final conclusions will be uneffected by this.
If we use GMSB (gauge mediated supersymmetry breaking models) models where SUSY breaking is communicated to the observable sector by gauge mediation, the soft susy breaking scalar and the gaugino masses are generated at a scale $\sim 10^{5}$ GeV by gauge interactions. In this type of models gauge unification is not necessary in order to have unified gaugino mass, since they are generated in one loop. In our calculation for the GMSB cases we do not consider any gauge unification. However in figure 1b we show one example where we have the gauge unification with a messenger sector composed of: one copy of $D_R+\bar D_R$, $U_R+\bar U_R$ and $L_L+\bar L_L$. The couplings unify at $10^{10.5}$ GeV, the intermediate scale is at $\sim 10^{5.2}$ GeV and the $\alpha_R/\alpha_{B-L}$ is 1/2 at the intermediate scale.
(dmdeltafig1.epsf width 7 cm)
(dmdeltafig2.epsf width 7 cm)
Gravity mediated scenario
=========================
In the standard gravity mediated scenarios, one starts with a universal mass-square for all scalar components of the chiral superfields at the Planck scale and they are then extrapolated to the weak scale to determine their masses[@martin]. For those fields with large Yukawa couplings such as the Higgs doublet $H_u$, top squark, the weak scale value is significantly lower than the planck scale one. (In fact, for the $H_u$ turning negative gives rise to the celebrated phenomenon of radiative electroweak symmetry breaking). For other squarks, the gaugino mass contribution has the effect of increasing their value over the Planck scale value. On the other hand for the sleptons, the change between the weak scale and Planck scale value is not very significant since they neither have large Yukawa couplings nor do they have strong interactions.
Turning now to the SUSYLR model, as mentioned before, we will assume universality at the unification and in contrast with the MSSM, the effective theory below the right handed scale, contains the following new coupling in the superpotential: $$\begin{aligned}
W_{extra}=f_{i}{\Delta^c}^{\pm\pm}{\em l}_i{\em l}_i\end{aligned}$$ which will have a major impact on our spectrum at the weak scale.
In writing the above coupling, we have used the fact that experimental limits on lepton flavor changing processes such as $\mu\rightarrow 3e$ and $\tau\rightarrow 3e$ imply that $f_{ij}$ with $i\neq j$ are very small compared to the diagonal couplings $f_i$ and have therefore taken the liberty to simply drop the off-diagonal couplings. In what follows, we will denote $\Delta^c\equiv \Delta$. This interaction gives rise to the process $\mu^+e^-\rightarrow
\mu^-e^+$ with a strength $G_{M-\bar{M}}\simeq \frac{f_1 f_{2}}{4\sqrt{2}
M^2_{\Delta}}$. Recent PSI experiment[@jungman] has yielded an upper limit on $G_{M-\bar{M}}\leq 3\times G_F\times 10^{-3}$. For $M_{\Delta}=100 $ GeV, this implies that $f_1f_2\leq 1.2\times 10^{-3}$. Thus we expect each of the couplings to be less than $.1$ barring pathological situations where only one of the couplings bears the brunt of the constraint. On the other hand there is no such constraint on $f_{3}$ from experiments. So we will choose it to be of order 0$.5$.
The first implication of the relatively large $f_{3}$ is on the weak scale value for the mass of the $\Delta$-boson. As shown in [@chacko], at the $v_R$ scale, both the $\Delta$ and its fermionic partner $\tilde{\Delta}$ ($\Delta$-ino) have nearly the same mass. The bosonic component however runs faster than the fermionic one. As a result, at low energies we can expect that $M_{\Delta} < M_{\tilde{\Delta}}$. This has important implications for phenomenology.
A second consequence of the possible large $f_{3}$ is that the $\tilde{\tau^c}$ mass is drawn down to smaller values at lower scales. In order to calculate the mass spectrum we use the fermion masses, mixing angles, $f$’s and the magnitude of the gauge couplings as inputs at the weak scale. We use $m_t=175$ GeV and $m_b=4.5 GeV$. We then use the RGE’s shown in the appendix and in the reference [@barger] to determine the Yukawa and the gauge couplings at the GUT scale. Then we use the universal boundary conditions and use the RGE’s for the soft SUSY breaking masses and the couplings to determine the masses at the weak scale. We determine the parameter $\mu$ from the condition that electroweak symmetry is broken radiatively. We choose the sign of $\mu$ to be negative, since the other choice will give large $b\rightarrow s\gamma $ rate. The lighter stau mass will be much smaller than the other sleptons (even when $tan\beta$ is small) due to the presence of the new coupling $f_3$. In figure 3 we show the mass contours of the lighter stau (dotted line), the $\Delta^{\pm\pm}$ (solid line) and the lighter selectron(dot-dashed line) in the $m_0$ and $tan\beta$ plane for $m_{1/2}=180 GeV$. We choose $\tilde\Delta^{\pm\pm}$ mass $\sim 100$ GeV at the Weak scale (90 GeV at the intermediate scale). As $m_0$ increases the $\Delta^{\pm\pm}$ mass decreases due to the subtractive effect in the RGE originating from the larger soft SUSY breaking scalar mass. We can see from the figure that the bound on $\Delta^{\pm\pm}$ mass rules out the upper range of $m_0$. The lower bound on lighter stau ($\tilde\tau_1$) mass on the other hand rules out lower range of $m_0$.
The latest bound on lighter stau mass is about 57 GeV [@aleph1]. In this scenario $\tilde\Delta$ mass is too high to be pair produced at LEPII. We also observe (due to the smallness of $f_{1,2}$ compared to $f_3$) that the lighter selectron mass is much higher than the $\tilde\tau_1$ mass even when the $\tan\beta$ is small for the same $m_0$ and $m_{1/2}$ values. This will differentiate between the final states of chargino pair production in MSSM and in this model. In fig 4 we show the same mass contours for $\tilde\Delta^{\pm\pm}\sim 80$ GeV at the weak scale (70 GeV at the intermediate scale). Lower $\Delta^{\pm\pm}$ mass indicates lesser effect on the $\tilde\tau_1$ mass from the new interactions and consequently more parameter space for lower $m_0$, however low $\Delta$ mass rule out more parameter space from the higher $m_0$ range. In this scenario the $\tilde\Delta$s can be pair produced at LEPII. In Fig 5 we exhibit a scenario where $\tilde\Delta$ mass is $\sim 120$ GeV at the weak scale (110 GeV at the intermediate scale). The lightest neutralino ($\chi^0_1$)in all these scenarios are $\sim 49-57$ GeV and the lightest chargino mass is around 80-90 GeV. If we vary the $m_{1/2}$, e.g $m_{1/2}=150 GeV$, the $\chi^0_1$ becomes 33-43 GeV and the lightest chargino becomes 57-67 GeV. the scalar masses also get reduced by 10 -20 GeV. In Fig 6 we show a scenario where the third generation coupling is smaller $f_3\sim 0.25$. As one can expect, the effect of the new couplings in the stau mass is reduced. For low $tan\beta$, as in the conventional SUGRA mode, the $\tilde\tau_1$ and the lighter selectron mass ($\tilde e_R$) become very close. The $\Delta$ boson mass lower than 88 GeV appears for $m_0\rangle 270 $ GeV, where as in Fig. 3 the same mass contour appears at $m_0\rangle 140 $ GeV (In both the cases the $m_{1/2}$ at the GUT scale and $M_{\tilde\Delta}$ at the intermediate scale are same. From the analysis of the parameter space we surmise that most of the allowed parameter space for large $f_3$ can be searched in the present colliders.
(dmdeltafig3.epsf width 6 cm)
(dmdeltafig4.epsf width 6 cm)
(dmdeltafig5.epsf width 6 cm)
(dmdeltafig6.epsf width 6 cm)
The $\Delta$ will decay into a pair of $\tau$s (since the coupling to the other leptons are suppressed). The $\tau$’s have high $p_T$. We have 4 $\tau$s in the final state. The like charged $\tau$s originate from the same vertex.
The SM background for this process can come from the pair production of $Z^0$ and the subsequent decay of each $Z^0$ to a $\tau^{+}\tau^{-}$ pair (in all these cases oppositely charged $\tau$s come from the same vertex). This event will have no missing energy, but the rate is small due to the small branching ratio for $Z^0\rightarrow\tau^{+}\tau^{-}$ ($\sigma.B^2\sim 10^{-3} pb$). The $4\tau$ background can also come from the production of $Z\gamma^*$, $\gamma^*$ converting into $\tau^{+}\tau^{-}$ and the $Z$ decaying into $\tau^{+}\tau^{-}$ pair. Another source can be the production of two virtual photons along with $e^{+}e^{-}$ and each photon converting into $\tau^{+}\tau^{-}$. However, in both the above processes, the cross section is $\sim 10^{-3}$pb. Thus, the SM background is negligible small for the 4$\tau$ signal.
The $\tilde \Delta$ will primarily decay into $\tilde\tau_1$ and $\tau$ (almost 100$\%$). The $\tilde\tau_1$ will then decay into $\tau$ and $\chi^0_1$ (missing energy) (100$\%)$. The final state has 4 $\tau$’s plus missing energy. Two of the $\tau$s have high $p_T$ and these originate from the decay of lighter stau. This kind of 4 $\tau $ plus missing energy signal also originate from the $\chi^0_1$ pair production in the GMSB scenarios where lighter stau is the NLSP. But there is a subtle difference in the final state which we will discuss in the next section.
The chargino pair production can also give rise to $2\tau$ plus missing energy states. Since the staus are much lighter than the other sleptons, the chargino will primarily decay into $\tilde\tau_1$ and $\nu_{\tau}$ in the leptonic decay channel. On the other hand, the chargino decays into $e$’s and $\mu$’s in MSSM. The charginos can be pair produced at LEP II and Tevatron. The production of chargino and the second lightest neutralino (this crosssection is larger than the chargino pair production ) at the Tevatron will also give rise to lots of high $p_T$ taus in the final state. The second lightest neutralino ($\chi^0_2$) primarily decays into $\tau$ and $\tilde\tau_1$ and $\tilde\tau_1$ then decays into a tau and the lightest neutralino. Altogether, there can be 3$\tau$s along with missing energy. Since $\chi^0_2$ mass is much larger than the $\tilde\tau_1$ mass, all the three $\tau$s will have high $p_T$.
The $\Delta$ bosons and the $\Delta$-inos can be pair produced at the LEPII and at the Tevatron as well. The productions of $\Delta$’s (fermion and boson) at LEP II and $\Delta$ scalar at Tevatron have been considered in the references[@huitu]. We give the production cross-sections of the $\Delta $ bosons at the LEP II and at the Tevatron in Figures 7 and 8. The $\Delta$-ino is pair produced at the LEP II and at the Tevatron via Z, $\gamma$ exchange. Usually there is also a selectron mediated t-channel contribution in the case of $\Delta$-ino production at the $e^{+}e^{-}$ collider. However the contribution from this diagram in this model is negligible since the $\Delta$ coupling to the first generation leptons is very small. The production crosssections are larger than the scalar counterparts for the same mass. In Figure 9 we show the production of $\tilde\Delta$ at LEPII for $\sqrt s=182 $ GeV and 194 GeV. We can see that the $\Delta$-ino mass of 95 GeV gives rise to a cross section of 3 pb at LEPII. Thus, if the $\Delta$-inos are produced at LEPII the cross section will be quite large.
In Figure 10 we show the production cross-section at the Tevatron for $\sqrt s=2 TeV.$ The production crosssection is about 0.7 pb at $\sqrt s=2$TeV for $\Delta$-ino mass of 95 GeV and the cross section is about 0.6 pb at $\sqrt s=1.8$TeV for the same $\Delta$-ino mass. Hence with 110$pb^{-1}$ of already accumulated luminosity, the number of events are $\sim 66$. Since the final state are pure $\tau$ leptons, detection is difficult. But we will have some events left even after taking the $\tau$ detection efficiency to be small. We urge the experimentalists to look for the $\tau $ signals in the data which has been already accumulated and also in the data that will be generated in the future runs.
(dmdeltanew.epsf width 6 cm)
(dmdeltanew1.epsf width 7 cm)
(dmdeltafig7.epsf width 7 cm)
(dmdeltafig8.epsf width 7 cm)
Gauge mediated susy breaking scenario
=====================================
In the previous section we assumed that the scale at which supersymmetry is broken is higher than the $W_R$ scale. However this need not be the case and in particular there has recently been a lot of interest in theories where gauge interactions are the mediators of supersymmetry breaking at a relatively low scale[@dine]. Let us discuss the implications of this scenario for our model. The soft SUSY breaking terms are now generated explicitly only at the scale at which the messenger fields are integrated out, and are not explicitly present at the $W_R$ scale. Since they are generated by loop graphs involving the gauge bosons of the residual symmetries, their form will be such as to respect only the surviving gauge symmetries. We will show that this difference has consequences for phenomenology. Let us assume for simplicity that the messenger sector consists of a vectorlike isosinglet pair of fields (charge -1/3) $Q\oplus \bar{Q}$ and a vectorlike weak isodoublet pair $L\oplus \bar{L}$.
As shown in [@chacko], in this case, the scalar $\Delta$ mass has two contributions at the SUSY breaking scale of $100$ TeV or so: one coming from the two loop gauge contributions in the usual manner (for a review see [@martin]) and another coming from the higher dimensional terms. On the other hand, the $\tilde{\Delta}$ mass gets contribution only from the latter kind of terms. We therefore expect that in the gauge mediated scenario, the $\tilde{\Delta}$ will be lighter of the two particles.
The soft SUSY breaking gaugino and the scalar masses at the messenger scale $M$ are given by [@martin]
$\tilde M_i(M) = n\,g\left({\Lambda\over M}\right)\,
{\alpha_i(M)\over4\pi}\,\Lambda$
and
$\tilde m^2(M) = 2 \,n\, f\left({\Lambda\over M}\right)\,
\sum_{i=1}^3\, k_i \, C_i\,
\biggl({\alpha_i(M)\over 4\pi}\biggr)^2\,
\Lambda^2 $
where $\alpha_i\; (i=1-3)$ are the three SM gauge couplings and $k_i=1,1,3/5$ for SU(3), SU(2), and U(1), respectively. The $C_i$ are zero for gauge singlets, and 4/3, 3/4, and $(Y/2)^2$ for the fundamental representations of $SU(3)$ and $SU(2)$ and $U(1)_Y$ respectively (with $Y$ defined by $Q=I_3+Y/2)$. Here $n$ corresponds to $n((Q,L)+{\bar {(Q,L)}})$. $g(x)$ and $f(x)$ are messenger scale threshold functions with $x=\Lambda/M$.
We calculate the SUSY mass spectrum using the appropriate RGE equations [@barger] with the boundary conditions given by the equations above and we vary five free parameters $\Lambda$, $M/\Lambda$, $tan\beta$, $n$ and the sign of $\mu$ ($\mu$ is the coefficient of the bilinear Higgs term in the superpotential). We first run the Yukawa couplings (along with the three new couplings $f_{1,2,3}$) and gauge couplings from the weak scale upto the GMSB scale. At the GMSB scale we use the boundary conditions and then use the necessary RGEs for the soft SUSY breaking masses in order run down to the weak scale.
The CLEO constraint on the $b\rightarrow s\gamma $ rate restricts $\mu<0$ [@dwt]. In the absence of late inflation, cosmological constraints put an upper bound on the gravitino mass of about $10^4$ eV [@pp], which restricts $M/\Lambda=1.1-10^4$. In the figures we show the results for n=1, but we discuss the other values of n also. For reasons discussed before, we assume that $f_{1,2}$ are small ( $\sim 0.05$), but $f_3$ is larger. We show our result for $f_3\sim 0.5$ and $f_3\sim 0.25$. We also vary $M_{\tilde\Delta}$ between 70-120 GeV at the GMSB scale.
The gravitino is always the LSP. In the usual GMSB case the $\chi^0_1$ and the $\tilde\tau_1$ fight for the NLSP spot. In this model the $\tilde\Delta^{\pm\pm}$ also joins the race to become NLSP. The third generation lighter stau mass gets affected due to the presence of the additional large coupling $f_{3}$. Thus the $\tilde\tau_1$ is much smaller compared to the conventional GMSB case for the same parameter space. Consequently the lighter stau will be the NLSP for a wider region of parameter space compared to the lighter neutralino.
In figure 11a we have shown the mass contours of $\tilde\tau_1$ (solid line), $\tilde e_R$ (dotted line), $\chi^0_1$ (dash-dotted line) and the chargino masses (dashed line) for $M=1.1 \Lambda$, n=1 and $M_{\tilde\Delta}=90$ GeV at the GMSB scale (94 GeV at the weak scale). We also show the contour along which the lighter stau mass and the neutralino mass are same (thick solid line). The region above the contour has lighter stau as the NLSP. We see that only a small region for $tan\beta$ 3-15 and $\Lambda$ 40-60 TeV has $\chi^0_1$ as NLSP. When $\chi^0_1$ is the NLSP, it decays into a photon and a gravitino. If $\chi^0_1$ is pair produced at LEPII, the final state has $\gamma\gamma$ plus ${\rlap/E}_T$. The photons are hard and easy to detect. Already we have bound on the $\chi^0_1$ mass of around 80 GeV at LEP II [@aleph2] provided the selectrons are not too heavy. We can see that if we use $\chi^0_1$ mass bound as 80 GeV, the region left out in fig 11a where $\chi^0_1$ is the NLSP is very small. The region where $\tilde\tau_1$ is the NLSP, which is the dominant region, stau decays into a high $p_T$ $\tau$ and a gravitino (missing energy). So far there is not much bound in these regions, other than the $\tilde\tau_1$ has to be larger than 57 GeV. In the figures we have shown the stau mass contours of 45, 60, 80 and 100 GeV. The $\tilde\tau_1$ mass contours have large dependence on the $tan\beta$, the $\tilde\tau_1$ decreases with the increase in $\tan\beta$. The $\tilde e_R$, chargino and the $\chi^0_1$ mass contours do not have much $tan\beta $ dependence.
In figure 11b we show the same mass contours in the plane of $\Lambda$ and $M/\Lambda$ for $\tan\beta=10$. The contours along which $\tilde\tau_1$ mass is equal to the $\chi^0_1$ mass form an envelope. Within the envelope $\chi^0_1$ is NLSP. The $\tilde e_R$ masses increase as the $M/\Lambda$ ratio increase. The couplings $f_{1,2}$ are small to affect the $\tilde e_R$ or the $\tilde \mu_R$ masses (selectron and smuon masses are almost same). The lighter stau mass however decreases with the increase in the ratio of $M/\Lambda$ (subtractive effect from the soft scalar masses due to the presence of the new coupling overcome effects coming from the gaugino masses in the RGE). In the conventional GMSB model the stau mass increases with the increase in the $M/\Lambda$ ratio. With the improvement of the $\tilde\tau_1$ mass bound much more parameter space will be ruled out in the higher$M/\Lambda$ ratio. The $\chi^0_1$ and the chargino masses initially decrease with the increase in the ratio of $M/\Lambda$ due to the threshold corrections. The 80 GeV $\chi^0_1$ bound rules out much of the parameter space in the envelope where$\chi^0_1$ is the NLSP. The $M_{\tilde\Delta^{\pm\pm}}$ varies between 94-108 GeV at the weak scale. Unlike the SUGRA scenario, the $\Delta^{\pm\pm}$ mass is much larger than the fermionic part at the weak scale and its mass, in the full range shown in the figure, varies from 170 -570 GeV (due to the large soft SUSY breaking contribution at the GMSB scale). Thus, the allowed parameter space in GMSB scenarios are much more than the SUGRA scenarios.
In figures 12 and 13, we have shown the mass contours in the $\Lambda-M/\Lambda$ plane for $M_{\tilde\Delta^{\pm\pm}}=70$ and 110 GeV at the GMSB scale. The $tan\beta$ is chosen to be 10. When $M_{\tilde\Delta^{\pm\pm}}=70$ GeV, the envelope is larger, since $\tilde{\tau_1}$ mass has less subtractive contribution from the delta mass. On the other hand when $M_{\tilde\Delta^{\pm\pm}}=110$ GeV, the envelope shrinks. The envelope can also increase in the size if we have smaller coupling $f_3$. We show the effect of a smaller $f_3$ in figure 14. This figure looks more like the conventional GMSB model. We see that there is no parameter space where $\tilde{\tau_1}$ is NLSP. But if we increase $tan\beta$ we will hit the region where $\tilde{\tau_1}$ is the NLSP. The $\tilde{\tau_1}$ mass in this case, as expected, increases with the increases in the ratio of $M/\Lambda$. So far in all these figures we have used n=1. The effect of larger values of n can easily be surmised from the mass formula. As n increases gaugino masses increase proportionally, on the other hand the scalar masses increase as $\sqrt n$. Hence stau mass is the NLSP for a even wider region of parameter space. The lighter selectron’s mass becomes closer to the lighter stau mass and lower than the neutralino mass.
Let us now discuss the signals. At LEP II, the main production processes are the $\chi^0_1$ pair, the $\tilde\Delta^{\pm\pm}$ pair, the $\tilde e_R$ pair and the $\tilde{\tau_1}$ pair.
In the case of $\tilde\Delta^{\pm\pm}$ pair production, each $\tilde\Delta^{\pm\pm}$ wll decay primarily into a $\tilde{\tau_1}$ and a tau (both having same sign of charge). The other decay modes involving e.g. a electron and a $\tilde e_R$ and a muon and a smuon are suppressed primarily because of small $f_{1,2}$ couplings. In the parameter space when stau is the NLSP, $\tilde\Delta^{\pm\pm}$ decays into a $\tilde\tau_1$ and a $\tau$ and $\tilde\tau_1\rightarrow\tau\tilde G$. Thus, the final state has 4$\tau$ plus missing energy and out of the 4 $\tau$s, two have high $p_T$ (higher $p_T$ than the SUGRA case). These two high $p_T$ $\tau$s have opposite sign electric charges.
The $\chi^0_1$ pair production also gives rise to 4 $\tau$s plus missing energy (since each $\chi^0_1\rightarrow\tilde\tau_1\tau$), with two of these $\tau$s having high $p_T$ [@ddn]. However there is an essential difference between the signal in this case and in the previous case. Since the neutralino is a majorana particle, the two high $p_T$ $\tau$s can have same or opposite sign (with equal probability) for the electric charges and thus providing a way to discrminate between the two cases.
In the case of $\tilde e_R$ pair productions, each selectron will either decay into a electron and a $\chi^0_1$ or a $\tilde\Delta^{\pm\pm}$ and a positron (if the selectron mass is higher than the $\chi^0_1$ mass and the $\tilde\Delta^{\pm\pm}$ mass). Both $\chi^0_1$ and $\tilde\Delta^{\pm\pm}$ decay to $2\tau$ plus missing energy. The final states in either case will have $2e 4\tau$ plus missing energy with two of the $\tau$s having high $p_T$. Their relative sign will determine the decay channel of the selectron. If, however, the $\tilde e_R$ mass is lower than both the $\chi^0_1$ mass (large n case) and the $\tilde\Delta^{\pm\pm}$ mass, the selectron can decay into an electron and a gravitino or via offshell production of the $\chi^0_1$ or $\tilde\Delta^{\pm\pm}$. The $\chi^0_1$ and $\tilde\Delta^{\pm\pm}$ then convert into $2\tau$ plus missing energy. In the former case the final state of the selectron pair production has $2 e$ plus missing energy and in the later case the signal is $2e4\tau$ plus missing enrgy. Depending on the parameter space, these offshell decay modes can be comparable or greater than the onshell decay mode [@amK]. In the case of smuons the electrons in the final states will be replaced by the muons.
When $\chi^0_1$ is the NLSP, the $\tilde\Delta^{\pm\pm}$ will decay into a $\tilde\tau_1$ and a $\tau$ (100$\%$). The $\tilde\tau_1$ will decay into a $\chi^0_1$ and a $\tau$. The $\chi^0_1$ then decays into a photon and a gravitino. All the branching ratios are 100 $\%$. The final state from the $\tilde\Delta^{\pm\pm}$ pair production will have $2\gamma 4\tau$ plus missing energy-which is a spectacular signal and very hard to miss. This signal will appear along with the electrons or the muons in the case of selectron or smuon production, where the selectron decays as described above i.e. $\tilde
e_R\rightarrow e\tilde\Delta^{\pm\pm}$.
In the conventional GMSB models when $\chi^0_1$ is the NLSP case, one gets at most 2 leptons along with 2 $\gamma$ (through the slepton productions whose cross section is much smaller than the delta-ino production cross section) at the LEP II and at the Tevatron or 3 leptons plus 2 photons (chargino-second lightest neutralino production) at the Tevatron. Thus the signal $2\gamma 4\tau$plus missing energy will clearly distinguish this model from the ordinary GMSB models in the parameter space where $\chi^0_1$ is the NLSP. It may also happen that the $\tilde\Delta^{\pm\pm}$ mass is smaller than the $\tilde\tau_1$ or the $\chi^0_1$ mass. In that case the $\tilde\Delta^{\pm\pm}$ will decay into a $\tau$ and a virtual $\tilde\tau_1$ which will convert into a $\tau$ plus gravitino or if the $\chi^0_1$ mass is lower than the $\tilde\tau_1$ mass, then the $\tilde\tau_1$ will decay into a $\tau$ and a $\chi^0_1$. The $\chi^0_1$ will then convert into a photon and the gravitino. Thus the final states are same as discussed in the cases when $\chi^0_1$ is the NLSP or $\tilde\tau_1$ is the NLSP.
(dmdeltafig9ab.epsf width 6 cm)
(dmdeltafig10.epsf width 6 cm)
(dmdeltafig11.epsf width 6 cm)
(dmdeltafig12.epsf width 6 cm)
conclusion
==========
To summarize, we have studied the phenomenological implications and the collider signatures of the remnant doubly charged Higgs boson $\Delta^{\pm\pm}$ and its fermionic partner the Deltaino $\tilde\Delta^{\pm\pm}$. We find that existing limits on lepton flavor violation and muonium-anti-muonium transition can be used to conclude that the dominant coupling of these particles are diagonal and mostly to the third generation of leptons. They of course are forbidden from coupling to the quarks by electric charge conservation. The effect of the dominant third generation lepton coupling has the consequence that the $\tilde{\tau_1}$ mass is much smaller than the other slepton masses. This gives rise to mulitiple $\tau$ enriched signal at LEP II and Tevatron. The fermionic as well as the bosonic partner of the doubly charged Higgs bosons can be produced at the present colliders and the signals contain 4 $\tau$ (plus 2$\gamma$ in some scenarios of the GMSB version of the models) with and without missing energy. This signal is found to be detectable for reasonable range of mass values of the particles and could be used to test the supersymmetric left-right models of the type discussed here or to restrict the allowed parameter range of the model. Most of the allowed parameter space can be searched in the existing colliders. We therefore urge the experimentalists to analyze the tau events in the existing as well as in the future data.
[**Acknowledgement**]{}
Work of R. N. M. has been supported by the National Science Foundation grant No. PHY-9421385 and that of B. D. by DE-FG06-854ER 40224. We thank G. Altarelli, J. Gunion and G. Snow for encouragement and comments.
Appendix
========
The standard MSSM RGEs for the Yukawa couplings and the soft SUSY breaking terms will be modified due to the presence of the new couplings and the fields,. There will be new RGEs for the additional fields and the couplings. The new interaction involves the third generaion righthanded leptons and the new field $\Delta^{\pm\pm}$. We will keep only the $f_3$ coupling in the RGE’s (in our numerical calculation we use all of them). The new RGEs and the modified ones are listed below: $$\begin{aligned}
2{\cal D}\lambda_{\tau} &=&\lambda_{\tau}(-
\sum_i{C\tau_i(4\pi\alpha_i)}+
\lambda_{b}^2+4 \lambda_{\tau}^2+4 {f_3}^2),\end{aligned}$$ where $C\tau_i=9/5,3,0$ and ${\cal D}\equiv{{16\pi^2}\over 2}{d\over{dt}}$.
$$\begin{aligned}
2{\cal D}f_3 &=&f_3(-
\sum_i{Cf_{3i}(4\pi\alpha_i)} +10 {f_3}^2+4 \lambda_{\tau}^2),\end{aligned}$$
where $Cf_{3i}=36/5,0,0$.
$$\begin{aligned}
{\cal D}m_{\tau^c}^2 &=& -{48\over 5}\pi\alpha_1{\tilde M_1}^2
+ 2\lambda_{\tau}^2(m_{\tau_c}^2+m_{\tau}^2+M_{H1}^2+A_{\tau}^2)\\\nonumber &+&
4 {f_3}^2(m_{\Delta^{\pm\pm}}^2+M_{\tau^c}^2+{A_{3}}^2),\end{aligned}$$
$$\begin{aligned}
{\cal D}m_{\Delta^{\pm\pm}}^2 &=& -{192\over
5}\pi\alpha_1{\tilde M_1}^2+ 2 {f_3}^2(m_{\Delta^{\pm\pm}}^2+2
m_{\tau^c}^2+{A_{3}}^2),\end{aligned}$$
We do not include the effects of the couplings to the other generation $f_{1,2}$ since they are very small. $$\begin{aligned}
{\cal D}A_{\tau} &=&(\sum_i{C\tau_i4\pi\alpha_i M_i} +
\lambda_{b}^2 A_b+4 \lambda_{\tau}^2 A_{\tau}+4 {f_3}^2 A_3),\end{aligned}$$
$$\begin{aligned}
{\cal D}A_3 &=&( (4\pi\alpha_1){36\over 5} M_1 +10 {f_3}^2
A_3+4
\lambda_{\tau}^2 A_{\tau}),\end{aligned}$$
$$\begin{aligned}
2{\cal D}M_{\tilde\Delta^{\pm\pm}} &=&M_{\tilde\Delta^{\pm\pm}}
( {-(4\pi\alpha_1){48\over 5} M_1} +4 {f_3}^2).\end{aligned}$$
[\[001\]]{}
R. N. Mohapatra, Phys. Rev. [**D34**]{}, 3457 (1986);
A. Font, L. Ibanez and F. Quevedo, Phys. Lett. [**B228**]{}, 79 (1989);
S. Martin, Phys. Rev. [**D 46** ]{}, 2769 (1992).
R. N. Mohapatra and A. Rasin, Phys. Rev. Lett. [**76**]{}, 3490 (1996);
R. Kuchimanchi, Phys. Rev. Lett. [**76**]{}, 3486 (1996);
R. N. Mohapatra , A. Rasin and G. Senjanović, Phys. Rev. Lett. [**79**]{}, 4744 (1997).
M. Gell-Mann, P. Ramond and R. Slansky, in [*Supergravity*]{} ed. D. Freedman et al. (North Holland, 1979);
T. Yanagida, KEK Lectures (1979);
R. N. Mohapatra and G. Senjanović, Phys. Rev. Lett. [**44**]{}, 912 (1980).
R. Kuchimanchi and R. N. Mohapatra, Phys. Rev. [**D48**]{}, 4352 (1993); Phys. Rev. lett. [**75**]{}, 3989 (1995).
C. S. Aulakh, A. Melfo and G. Senjanović, hep-ph/9707258;
C. S. Aulakh, A. Melfo, A. Rasin and G. Senjanović, hep-ph/9712551.
Z. Chacko and R. N. Mohapatra, hep-ph/9712359.
C. S. Aulakh, K. Benakli and G. Senjanović, Phys. Rev. Lett. [**79**]{}, 2188 (1997).
K. Jungman, invited talk in PASCOS98 (1998).
M Dine, A Nelson, Phys. Rev. [**D 48**]{}, 1277 (1993);
M. Dine, A.E. Nelson, Y. Nir, and Y. Shirman, Phys. Rev. [**D53**]{} 2658 (1996);
A.E. Nelson, hep-ph/9511218; M. Dine, A.E. Nelson, and Y. Shirman, Phys. Rev. [**D51**]{} (1995) 1362.
K. S. Babu, B. Dutta and R. N. Mohapatra, in preparation.
Similar ideas in the context of MSSM and other models have been considered earlier by
T. Blazek, S. Raby and S. Pokorski, Phys. Rev. [**D 52**]{}, 4151 (1995);
C. Hamazoui and M. Pospelov, hep-ph/9803354.
F. Gabbiani, E. Gabrielli, A. Masiero and L. Sivestrini, hep-ph/9604387.
R. N. Mohapatra, G. Senjanović and M. Tran, Phys. Rev. [**D 28**]{}, 546 (1983);
G. Ecker, W. Grimus and H. Neufeld, Phys. Lett. [**B127**]{}, 365 (1983);
F. Gilman and M. Reno, Phys. Rev. [**D29**]{}, 937 (1983);
M. Pospelov, Phys. Rev. [**D56**]{}, 259 (1997);
G. Barenboim, J. Bernabeau, M. Raidal, Valencia preprint (1996).
S. Martin, hep-ph/9709356, to appear in [*Perspectives in Supersymmetry*]{}, ed. G. Kane (World Scientific, 1998)
V. Barger, M. Berger and P. Ohman, Phys. Rev. [**D**]{}, 4908 (1994).
The ALEPH collaboration; ALEPH-98-014, CONF 98-004.
M. Lusignoli and S. Petrarca, Phys. Lett. [**B 226**]{}, 397 (1989);
K. Huitu, J. Maalampi and M. Raidal, Nucl. Phys. [**B420**]{}, 449 (1994); Phys. Lett. [**B320**]{}, 60 (1994).
F. Cuyper and S. Davidson, hep-ph/9609487;
F. Cuyper and M. Raidal, hep-ph/9704224;
M. Raidal Phys. Rev. [**D57**]{}, 2013 (1998).
K. Huitu, J. Maalampi, A. Pietila and M. Raidal, hep-ph/9606311;
J. Gunion, C. Loomis and K. T. Pitts, hep-ph/9610237;
C. Picciotto, Phys. Rev. [**D 56**]{}, 1612 (1997);
S. Chakrabarti, D. Choudhury, R. M. Godbole and B. Mukhopadhyaya, hep-ph/9804297;
For earlier applications of the light doubly charged Higgs bosons of left-right symmetric models, see
M. L. Swartz, Phys. Rev. [**D 40**]{}, 1521 (1989);
P. Herczeg and R. N. Mohapatra, Phys. Rev. lett. [**69**]{}, 2475 (1992);
A. Halprin, Phys. Rev. lett. [**48**]{}, 1313 (1982).
J. Gunion, J. Grifols, B. Kayser, A. Mendez and F. Olness, Phys. Rev. [**D40**]{}, 1546 (1989).
S. Dimopoulos, S. Thomas and J.D. Wells, Nucl. Phys. [**B488**]{}, 39 (1997) ;
H. Baer, M. Brhlik, C.-H. Chen and X. Tata, Phys. Rev. [**D55**]{}, 4463 (1997);
N. G. Deshpande, B. Dutta and S. Oh, Phys. Rev. [**D 56**]{}, 519 (1997);
R. Rattazzi and Uri Sarid, Nucl. Phys. [**B501**]{}, 297 (1997).
H. Pagels, J. Primack, Phys. Rev. Lett. [**48**]{}, 223 1982.
The ALEPH collaboration; ALEPH-98-013, CONF 98-003;
C. Chang and G. Snow; UMD-PP-98-92.
D.A. Dicus, B. Dutta, S. Nandi, Phys. Rev. Lett.[**78**]{}, 3055 (1997);
F.M. Borzumati, hep-ph/9702307;
S. Ambrosanio, G. Kribs, S. Martin, Phys. Rev. [**D56**]{}, 1761 (1997).
S. Ambrosanio, Graham D. Kribs and Stephen P. Martin, hep-ph/9710217;
K. Cheung, D.A. Dicus, B. Dutta, S. Nandi,hep-ph/9711216 (to appear in PRD).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present a new approach for image reconstruction and weak lensing measurements with interferometers. Based on the shapelet formalism presented in Refregier (2001), object images are decomposed into orthonormal Hermite basis functions. The shapelet coefficients of a collection of sources are simultaneously fit on the $uv$ plane, the Fourier transform of the sky brightness distribution observed by interferometers. The resulting $\chi^2$-fit is linear in its parameters and can thus be performed efficiently by simple matrix multiplications. We show how the complex effects of bandwidth smearing, time averaging and non-coplanarity of the array can be easily and fully corrected for in our method. Optimal image reconstruction, co-addition, astrometry, and photometry can all be achieved using weighted sums of the derived coefficients. As an example we consider the observing conditions of the FIRST radio survey (Becker, White & Helfand 1995; White et al. 1997). We find that our method accurately recovers the shapes of simulated images even for the sparse $uv$ sampling of this snapshot survey. Using one of the FIRST pointings, we find our method compares well with CLEAN, the commonly used method for interferometric imaging. Our method has the advantage of being linear in the fit parameters, of fitting all sources simultaneously, and of providing the full covariance matrix of the coefficients, which allows us to quantify the errors and cross-talk in image shapes. It is therefore well-suited for quantitative shape measurements which require high-precision. In particular, we show how our method can be combined with the results of Refregier & Bacon (2001) to provide an accurate measurement of weak lensing from interferometric data.'
author:
- 'Tzu-Ching Chang$^{1}$ and Alexandre Refregier$^{2}$'
title: 'Shape Reconstruction and Weak Lensing Measurement with Interferometers: A Shapelet Approach'
---
Introduction
============
Interferometers are widely used for astronomical observations as they provide high angular resolutions and large collecting areas. Existing interferometers in the radio (e.g. the Very Large Array (VLA), the Berkeley-Illinois-Maryland Association (BIMA), etc.) and in the optical band (e.g., the Center for High Angular Resolution Astronomy (CHARA), the Palomar Testbed Interferometer, the Keck Interferometer, the Very Large Telescope (VLT) Interferometer, etc.) will soon be complemented by new facilities such as the Expanded Very Large Array (EVLA), the Square Kilometer Array (SKA), the Atacama Large Millimeter Array (ALMA), and the Low Frequency Array (LOFAR). Interferometers are now also being developed to produce maps of the Cosmic Microwave Background on small scales (e.g., the Cosmic Background Imager (CBI), the Array for Microwave Background Anisotropy (AMiBA), the Very Small Array (VSA), etc).
Interferometric arrays, however, do not provide a direct image of the observed sky, but instead measure its Fourier transform at a finite number of discrete samplings, or ‘$uv$’ points, corresponding to each antenna pair in the array. The image in real space must therefore be reconstructed from the $uv$ plane by inverse Fourier transform while deconvolving the effective beam arising from the finite sampling (see Thompson et al. 1986, Perley et al. 1989, and Taylor et al. 1999 for reviews). For this purpose several elaborate methods have been developed. For instance, the commonly used CLEANing algorithm implemented in the NRAO AIPS software package (Hogbom 1974; Schwarz 1978; Clark 1980; Cornwell 1983), relies on successive subtraction of real-space delta functions from the $uv$ plane. Another method is based on Maximum Entropy (e.g., Cornwell & Evans 1985) and consists of finding the simplest image consistent with the $uv$ data. These methods are well-tested and appropriate for various applications; however, the methods are non-linear and do not necessarily converge in a well-defined manner. Consequently, they are not well-suited for quantitative image shape measurements requiring high precision. In particular, weak gravitational lensing (see Mellier 1999; Bartelmann & Schneider 2000 for reviews) requires the statistical measurements of weak distortions in the shapes of background objects and thus cannot afford the instabilities and potential biases inherent in these methods. While interferometric surveys offer great promises for weak lensing (Kamionkowski et al. 1998; Refregier et al 1998; Schneider 1999), a different approach for shape measurements is therefore required to achieve the necessary accuracy and control of systematics.
In this paper, we present a new method for reconstructing images from interferometric observations. It is based on the formalism introduced by Refregier (2001, Paper I) and Refregier & Bacon (2001, Paper II), in which object shapes are decomposed into orthonormal shape components, or ‘shapelets’. The Hermite basis functions used in this approach have a number of remarkable properties which greatly facilitate the modeling of object shapes. In particular, they are invariant under Fourier transformation (up to a rescaling) and are thus a natural choice for interferometric imaging. We show how shapelet components can be directly fitted on the $uv$ plane to reconstruct an interferometric image. The fit is linear in the shapelet coefficients and can thus be performed by simple matrix multiplications. Since the shapelet components of all sources are fitted simultaneously, cross-talk between different sources (e.g., when the sidelobe from one source falls at the position of a second source) are avoided, or at least quantified. The method also provides the full covariance matrix of the shapelet coefficients, and is robust. We also show how the complex effects of bandwidth smearing, time averaging and non-coplanarity of the array can be easily and fully corrected for in our method. Our method is thus well-suited for applications requiring unbiased, high-precision measurements of object shapes. In particular, we show how the method can be combined with the results of Paper II to provide a clean measurement of weak gravitational lensing with interferometers. We test our methods using both observations from the FIRST radio survey (Becker et al. 1995; White et al. 1997) and numerical simulations corresponding to the observing conditions of that survey. We also show how our method can be implemented on parallel computers and discuss its performance in comparison with the CLEANing method.
Our paper is organized as follows. In §\[shapelets\], we first summarize the relevant features of the shapelet method. In §\[method\], we describe how shapelets can be applied to image reconstruction with interferometers. In §\[results\], we discuss tests of the method using both simulated and real FIRST observations. In §\[lensing\] we show how our method can be used for weak lensing applications. Our conclusions are summarized in §\[conclusion\].
Shapelet Method {#shapelets}
===============
We begin by summarizing the relevant components of the shapelet method described in Paper I. In this approach, the surface brightness $f({\mathbf
x})$ of an object is decomposed as $$f({\mathbf x}) = \sum_{{\mathbf n}} f_{{\mathbf n}} B_{{\mathbf n}}
({\mathbf x};\beta),$$ where $$B_{{\mathbf n}}({\mathbf x};\beta) \equiv
\frac{H_{n_{1}}(\beta^{-1} x_{1}) ~H_{n_{2}}(\beta^{-1} x_{2})
~e^{-\frac{|x|^{2}}{2 \beta^{2}}}}
{\left[ 2^{(n_{1}+n_{2})} \pi ~\beta^{2}
~n_{1}! ~n_{2}! \right]^{\frac{1}{2}}}$$ are the two-dimensional orthonormal Hermite basis functions of characteristic scale $\beta$, $H_{m}(\xi)$ is the Hermite polynomial of order m, ${\mathbf x}=(x_{1},x_{2})$ and ${\mathbf
n}=(n_{1},n_{2})$. The basis is complete and yields fast convergence in the expansion if $\beta$ and $x=0$ are, respectively, close to the size and location of the object. The basis functions can be thought of as perturbations around a two-dimensional Gaussian, and are thus natural bases for describing the shapes of most astronomical objects. They are also the eigenfunctions of the Quantum Harmonic Oscillator (QHO), allowing us to use the powerful formalism developed for that problem. A similar decomposition scheme using Laguerre basis functions has been independently proposed by Bernstein & Jarvis (2001).
The Hermite basis functions have remarkable mathematical properties. In particular, let us consider the Fourier transform of an object intensity, $\tilde{f}({\mathbf k}) = (2\pi)^{-\frac{1}{2}}
\int_{-\infty}^{\infty} d^{2}{x} f({\mathbf x}) e^{i{\mathbf k} \cdot
{\mathbf x}}$. It can be decomposed as $\tilde{f}({\mathbf k}) =
\sum_{{\mathbf n}} f_{{\mathbf n}} ~\widetilde{B}_{{\mathbf
n}}({\mathbf k};\beta)$, where $\widetilde{B}_{{\mathbf n}}({\mathbf
k};\beta)$ are the Fourier-transforms of the basis functions, which obey the dual property $$\label{eq:duality}
\widetilde{B}_{{\mathbf n}}({\mathbf k};\beta) = i^{(n_{1}+n_{2})}
B_{{\mathbf n}}({\mathbf k};\beta^{-1}).$$ From the orthonomality of the basis functions, the coefficients are given by $$f_{{\mathbf n}} = \int_{-\infty}^{\infty} ~d^{2}{k}~ \tilde{f}
({\mathbf k}) ~\widetilde{B}_{{\mathbf n}}({\mathbf k};\beta).$$ This invariance (up to a rescaling) under Fourier transformation (Eq. \[\[eq:duality\]\]) makes this basis set a natural choice for interferometric imaging.
Shapelet Reconstruction with Interferometers {#method}
============================================
In this section, we describe how shapelets can be applied to interferometric imaging. We first briefly discuss how images are mapped onto the $uv$ plane by interferometers. We also show how the $uv$ plane can be binned into cells to reduce computation time and memory requirements. We then describe how the shapelet coefficients can be directly fit onto the binned $uv$ plane using a linear $\chi^2$ procedure. Finally, we describe how the resulting shapelet coefficients can be optimally combined to reconstruct the image, to co-add several pointings, and to measure shape parameters.
Interferometric Observations {#interferometers}
----------------------------
An interferometer consists of an array of antennae whose output signals are correlated to measure a complex ‘visibility’ for each antenna pair (see Thompson et al. 1986, Perley et al. 1989, and Taylor et al. 1999 for reviews). Each visibility is then assigned a point on the ‘$uv$ plane’ corresponding to the two-dimensional spacings between the antennae. In practice, the visibilities are close to, but not exactly equal to a two-dimensional Fourier transform of the sky brightness. Within the conventions of Perley, Schwab & Bridle (1989) for the VLA, the visibility measured for the antenna pair $(i,j)$ at time $t$ and at frequency $\nu$ is indeed given by $$\label{eq:v_ij}
V_{ij}(\nu,t) = \int d^{2}l~ \frac{A({\mathbf l},\nu)
f({\mathbf l},\nu,t)}{\sqrt{1-|l|^2}} e^{-2\pi i [ ul + vm +w (\sqrt{1-|l|^2}-1)]},$$ where $f({\mathbf l},\nu,t)$ is the surface brightness of the sky at position ${\mathbf l}=(l,m)$ with respect to the phase center, and $A({\mathbf l},\nu)$ is the (frequency-dependent) primary beam. For the VLA, the primary beam power pattern can be well-approximated as the Bessel function $2J_{1}(z)/z$, where $z \simeq 3.234~r
\theta_{p}^{-1}, ~\theta_{p}=30'.83 \times \left(\frac{1.4 {\rm
GHz}}{\nu} \right)$, $\nu$ is the observation frequency and $r$ is the position offset from the phase center (Condon et al. 1998). The $u,v,w$ coordinates are given by $$\begin{aligned}
\left( \begin{array}{c} u \\ v \\ w \end{array} \right) & = &
\left( \begin{array}{ccc} \sin H_{0} & \cos H_{0} &
0 \\ - \sin \delta_{0} \cos H_{0} & \sin \delta_{0} \sin H_{0} & \cos
\delta_{0} \\ \cos \delta_{0} \cos H_{0} & - \cos \delta_{0} \sin
H_{0} & \sin \delta_{0} \\ \end{array} \right) \nonumber \\
& & \times \left( \begin{array}{c}
L_{x} \\ L_{y} \\ L_{z} \end{array} \right) \frac{1}{\lambda},\end{aligned}$$ where $\lambda=c \nu^{-1}$ is the wavelength of observation, $H_{0}$ and $\delta_{0}$ are the hour angle and declination of the phase center, and $(L_{x},L_{y},L_{z})$ are the coordinate differences for the two antennas. The latter are measured in a fixed-Earth coordinate system, for which the sky rotates about the $\hat{L}_{z}$ axis. Note that the $(u,v,w)$ positions of the visibilities define the synthesized beam pattern. Since the $(u,v,w)$ coordinates are entirely determined by the antenna positions, source coordinates, and time and frequency of the observations, the synthesized beam is precisely known for interforemeters.
Only in the absence of a primary beam ($A=1$), for observations at zenith ($w=0$), and for small displacements from the phase center ($l,m \ll 1$), does the visibility reduce to an exact Fourier Transform of the intensity. Furthermore, the visibilities are measured in practice by averaging over small time and frequency intervals. The resulting averaged visibility is given by $$\label{eq:v_ij_bar}
\overline{V}_{ij} = \int dt \int d\nu ~T(t) G(\nu) V_{ij}(t,\nu)$$ where $T(t)$ and $G(t)$ are the time and frequency window functions, respectively, and are normalized as $\int dt T(t) = \int d\nu G(\nu)
\equiv 1$. Because the time and frequency intervals are typically very small, this double integral can be evaluated by Taylor expanding $V_{ij}(t,\nu)$ about the central values $t_{0}$ and $\nu_{0}$ of the window functions. For square-hat window functions of width $\Delta t$ (exact) and $\Delta \nu$ (approximate), respectively, we obtain $$\begin{aligned}
\overline{V}_{ij} & \simeq & V_{ij}(t_{0},\nu_{0}) + \frac{1}{24} \left[
\frac{\partial^{2} V_{ij}(t_{0},\nu_{0})}{\partial t^{2}} (\Delta
t)^{2} \nonumber \right. \\
& & \left. + \frac{\partial^{2} V_{ij}(t_{0},\nu_{0})}{\partial \nu^{2}}
(\Delta \nu)^{2} \right] + \cdots\end{aligned}$$ When the telescope points to a fixed location on the sky, the hour angle of the phase center changes as $H_{0}(t)\propto \omega_{E}t$, where $\omega_{E}$ is the angular frequency of the Earth. On the other hand, the declination $\delta_{0}$ of the phase center remains constant.
The above expression for $\overline{V}_{ij}$ can thus be computed analytically, leaving the two-dimensional ${\mathbf l}$-integral to evaluate numerically. Note that this provides a direct and complete treatment of primary beam attenuation, time-averaging, bandwidth smearing and non-coplanarity of the array. These effects are difficult to correct for in the context of the standard CLEANing method.
Binning in the $uv$ plane
-------------------------
In practice, the number of visibilities per observation is large ($>
10^5$). Directly fitting the shape parameters to all $uv$ points would thus require prohibitively large computing time and memory. Instead, we use a binning scheme to reduce the effective number of $uv$ points without loosing information. In the $uv$ plane, we set a grid of size $\Delta u = \frac{1}{2} ~{\Delta l}^{-1}$ and average the visibilities inside each cell, where $\Delta l$ is one-half of the intended field of view, and the factor $\frac{1}{2}$ accounts for the Nyquist frequency. The choice of $\Delta u$ is designed both to minimize the number of cells and to avoid smearing at large angular scales, which would otherwise act like an effective primary beam attenuation. We thus calculate the average visibility in the $uv$ cell $c$ (of size $\Delta u$) as $$\label{eq:v_c}
\overline{V}_{c} = \frac{1}{N_{c}}\sum_{ij \in c} \overline{V}_{ij}$$ where $N_{c}$ is the number of visibilities in the cell. This is the data we will use to reconstruct the image.
Fitting for the shapelet coefficients
-------------------------------------
We now wish to model the intensity $f_{s}(l,m)$ of each source $s$ as a sum of shapelet basis functions $$f_{s}({\mathbf l}) = \sum_{\mathbf n} f_{{\mathbf n}s} B_{\mathbf
n}({\mathbf l}-{\mathbf l}_{s};\beta_{s}),$$ centered on the source centroid ${\mathbf l}_s=(l_{s},m_{s})$, and scale $\beta_{s}$. Our goal is to estimate the shapelet coefficients $f_{{\mathbf n}s}$ of the sources given the binned $uv$ data $\{\overline{V}_{c}\}$. (We will describe how the centroid and shapelet scales are chosen in practice in §\[sim\]). In principle, the full $uv$ plane provides complete shape information for the sources. However, due to the finite number and non-uniform spacings of the antennae, the $uv$ (Fourier) space is poorly sampled, thus hampering the decomposition. This prevents us from performing a simple linear decomposition as is done with optical images in real space (see Paper I). This problem can be largely resolved by making a linear fit to the $uv$ plane with the shapelet coefficients as the free parameters.
For this purpose, the first step is to compute the binned visibilities $\overline{V}_{c}^{{\mathbf n}s}$ corresponding to each shapelet basis functions $B_{{\mathbf n}s}({\mathbf l}-{\mathbf l}_{s};\beta_{s})$ for each source $s$. This can be done by first computing the time- and frequency-averaged visibility $\overline{V}_{ij}^{{\mathbf n}s}$ by setting $f({\mathbf l})=B_{{\mathbf n}s}({\mathbf l}-{\mathbf
l}_{s};\beta_{s})$ in Equations (\[eq:v\_ij\]) and (\[eq:v\_ij\_bar\]). To prevent potential biases introduced by the binning scheme, we evaluate the basis functions at every visibility point and then average them inside each cell to compute $\overline{V}_{c}^{{\mathbf n}s}$ just as in Equation (\[eq:v\_c\]). Note that this ensures that the systematic distortions induced by the primary beam, bandwidth smearing, time-averaging and non-coplanarity can all be fully corrected in our method.
The next step is to form and minimize $$\label{eq:chi2}
{\chi}^{2}=({\mathbf d} - {\mathbf M ~f})^{T}~{\mathbf
C}^{-1}~({\mathbf d} - {\mathbf M~f}),$$ where ${\mathbf d}=\{ \overline{V}_{c} \}$ is the data vector, ${\mathbf M}=\{ \overline{V}_{c}^{{\mathbf n}s} \}$ is the theory matrix, and ${\mathbf f}=\{ f_{{\mathbf n}s} \}$ is the parameter vector. The covariance error matrix $${\mathbf C} = {\rm cov}[{\mathbf d},{\mathbf d}] =
\left\langle ({\mathbf d} - \langle
{\mathbf d} \rangle)^{T}{(\mathbf d} - \langle {\mathbf d} \rangle)
\right\rangle$$ for the binned visibilities can be estimated in practice either from the distribution of the visibilities in each bin or from the error tables provided by the interferometric hardware.
Because the model is linear in the fitting parameters, the best-fit parameters $\widehat{\mathbf f}$ can be computed analytically as (e.g., Lupton 1993) $$\label{eq:f}
\widehat{\mathbf f}= ({\mathbf M}^{T}{\mathbf C}^{-1}{\mathbf
M})^{-1}{\mathbf M}^{T}{\mathbf C}^{-1}{\mathbf d}.$$ The covariance error matrix ${\mathbf W}={\rm cov}[\widehat{\mathbf
f},\widehat{\mathbf f}]$ of the best-fit parameters is given by $$\label{eq:cova}
{\mathbf W}= ( {\mathbf M}^{T} {\mathbf C}^{-1} {\mathbf M} )^{-1}$$ This provides us with an estimate for the shapelet coefficients for each source and for their full covariance matrix. Note that all sources are fitted simultaneously thus avoiding (or at least quantifying) potential cross-talk between different sources (e.g., when a sidelobe from one source falls at the position of a second source). The coefficient covariance matrix can also be used to determine degeneracies produce by the finite $uv$ sampling of the array.
Combining the Shapelet Coefficients {#weighting}
-----------------------------------
Now that we have derived estimates $\widehat{f}_{\mathbf n}$ for the shapelet coefficients $f_{\mathbf n}$ for each source in a pointing, we can combine them to construct an image and to compute useful quantities such as the fluxes, centroids and sizes of the sources.
We first consider the practical problem of co-adding several pointings to derive an optimal image of a source. Let $\widehat{f}_{{\mathbf
n}p}$ be the coefficients of a source derived from pointing $p$, and let $W_{{\mathbf nm}p}$ be the associated covariance error matrix (from Eq. (\[eq:cova\])). It is easy to see that the error in the co-added coefficients $\widehat{f}_{\mathbf n}$ will be minimized if they are given by the weighted sum $$\widehat{f}_{\mathbf n} = \frac{ \sum_{p} W_{{\mathbf
n}\mathbf{n}p}^{-1}\widehat{f}_{{\mathbf n}p}}{\sum_{p}
W_{{\mathbf n}{\mathbf n}p}^{-1}}.$$ The covariance error matrix $W_{{\mathbf n}\mathbf{m}} =
{\rm cov}[ \widehat{f}_{\mathbf n},\widehat{f}_{\mathbf m}]$ of the co-added coefficients are then given by $$W_{{\mathbf n}{\mathbf m}} = \frac{ \sum_{p}
W_{{\mathbf n}{\mathbf n}p}^{-1}
W_{{\mathbf n}{\mathbf m}p}
W_{{\mathbf m}{\mathbf m}p}^{-1}}
{ \left(\sum_{p} W_{{\mathbf n}{\mathbf n}p}^{-1} \right)
\left(\sum_{p} W_{{\mathbf m}{\mathbf m}p}^{-1} \right)}.$$
We can then find an optimal weighting to reconstruct the image of a source from the estimated coefficients $\widehat{f}_{\mathbf n}$. To do so we seek the reconstructed coefficients given by $$f^{r}_{\mathbf n} = \phi_{\mathbf n} \widehat{f}_{\mathbf n}.$$ The weights $\phi_{\mathbf n}$ are chosen so that the reconstructed image $f^{r}({\mathbf l}) = \sum_{\mathbf n} f^{r}_{\mathbf
n} B_{\mathbf n}({\mathbf l})$ is ‘as close as possible’ to the true image $f({\mathbf l})$, in the sense that the least-square difference $$\int d^{2}l~\left[ f^{r}({\mathbf l})-f({\mathbf l}) \right]^2 =
\sum_{\mathbf n} [ f^{r}_{\mathbf n} - f_{\mathbf n} ]^{2}$$ is minimized. It is easy to show that this will be the case when $$\phi_{\mathbf n} = \frac{ |f_{\mathbf n}|^2 }{ |f_{\mathbf n}|^2 +
W_{{\mathbf n}{\mathbf n}}} \approx \frac{ |\widehat{f}_{\mathbf n}|^2 -
W_{{\mathbf n}{\mathbf n}}}{|\widehat{f}_{\mathbf n}|^2},$$ where the right-hand side provides an approximation which can be directly derived from the data. This weighting amounts to Wiener filtering in Shapelet space, in analogy with that performed in Fourier Space (see, e.g., Press et al. 1987). Figures \[fig:sim\] and \[fig:data\] show several reconstructed images using this weighting scheme. Note that this produces an estimate for the [*deconvolved*]{} image of the source. For display purposes, it is sometimes useful to smooth the reconstructed image by a Gaussian kernel (the ’restoring beam’ in radio parlance). This can easily be done in shapelet space by multiplying the coefficients by the analytic smoothing matrix described in Paper I.
While Wiener filtering yields an optimal image reconstruction, it is [*not*]{} to be used to measure source parameters such as flux, centroid, size, etc. Instead, an unbiased estimator for shape parameters can be derived directly from the shapelet coefficients (see Paper I). For instance, an estimate for the flux $F\equiv \int d^{2}l f({\mathbf
l})$ of a source is given by $\widehat{F} =
{\mathbf A}^{T} \widehat{\mathbf f}$ where $$\label{eq:flux}
A_{n_{1}n_{2}} = \pi^{\frac{1}{2}} \beta
2^{\frac{1}{2}(2-n_{1}-n_{2})}
\left( \begin{array}{c} n_{1} \\ n_{1}/2 \end{array} \right)^{\frac{1}{2}}
\left( \begin{array}{c} n_{2} \\ n_{2}/2 \end{array}
\right)^{\frac{1}{2}},$$ if $n_{1}$ and $n_{2}$ are both even (and vanishes otherwise). The variance uncertainty in the flux is then simply $$\label{eq:error}
\sigma^{2}[\widehat{F}] = {\mathbf A}^{T} {\mathbf W} {\mathbf A},$$ which provides a robust estimate of the signal-to-noise SNR$=\widehat{F}/\sigma[\widehat{F}]$ of the source. Similar expressions can be used to compute the centroid and rms size of the source. This can be easily generalized to compute in addition the major and minor axes of the source and its position angle. Note that these expressions are, again, estimates for deconvolved quantities.
Test of the Method {#results}
==================
As an application, we consider the FIRST radio survey (Becker et al. 1995; White et al. 1997), being conducted with the VLA at 1.4 GHz in the B configuration. For this survey, the primary beam FWHM is $\sim$ 30$'$ and the angular resolution is $5''.4$ (FWHM). The survey currently contains about $7.2\times 10^5$ sources with a $5\sigma$ flux-density limit of 1.0 mJy over $A \simeq 8,000$ deg$^2$; the mean source redshift is $\langle z \rangle \sim 1$. Observing time has been allocated to extend its coverage to 9,000 deg$^2$. The survey is composed of 165-second ‘grid-pointings’ with a time-averaging interval $\Delta t = 5$ seconds. It was conducted in the spectral synthesis mode, with a channel bandwidth of $\Delta \nu =$ 3 MHz. Because this wide-field survey was performed in the snapshot mode, its $uv$ sampling is very sparse. This makes shape reconstruction particularly challenging for FIRST, providing a good test for our method.
As explained in §\[interferometers\], higher order effects such as bandwidth smearing and time-averaging produce small distortions in the reconstructed image shapes if they are left uncounted for. These must be carefully corrected for high-precision statistical measurements of object shapes such as those required in weak lensing surveys. The effects are, however, very small and not noticeable on an object-by-object basis. For the purpose of this test, we thus ignore these effects and instead focus on the dominant factor in shape reconstruction, the finite and discrete $uv$ sampling.
{width="6.in"}
Simulations {#sim}
-----------
As a first test, we generated simulated VLA data using the observational parameters of FIRST. A grid pointing was generated at zenith with 33 5-second time-averaging intervals and 14 3-MHz channels in the B configuration. Simulated sources were randomly distributed within 23$'$.5 of the phase center, the cutoff adopted for creating the final co-added FIRST maps (Becker et al. 1995); the number density, flux density and size distributions chosen for the sources were similar to sources in the FIRST catalog. After generating the visibilities, we added uncorrelated Gaussian noise to the real and imaginary component of each $uv$ data point, with an rms of $\sigma_v
= \sigma_n N_{\rm vis}^{0.5}$, where $N_{\rm vis}$ is the total number of visibilities. The real-space rms noise $\sigma_n$ was set to $0.3$ mJy beam$^{-1}$, which is somewhat higher than the typical FIRST noise level, $\sim$ 0.2 mJy beam$^{-1}$.
We then simultaneously fitted all 23 sources in the grid pointing directly in the $uv$ plane. We imposed the constraint that that source intensities are real (i.e., non-imaginary). Each source $s$ was modeled as a shapelet with scale $\beta_{s}$, maximum shapelet order $n_{{\rm max},s}$, and center position ${\mathbf l}_{s}$. In principle, it is possible to determine these parameters with a source detection algorithm which directly uses shapelets. One can, for instance, tile ground-state shapelets with different sizes in the $uv$ plane, and thus detect sources with different sizes. However, this is computationally expensive and, since the FIRST catalog is conveniently available, we have not implemented this algorithm.
Instead, good choices for these parameters were derived from the FIRST catalog, which lists basic shape parameters for each source, such as its centroid, flux density, major and minor axes, and position angle, all obtained from an elliptical Gaussian fit. The shapelet position ${\mathbf l}_{s}$ was simply set to the centroid position from the catalog. The choices for the shapelet scales $\beta_{s}$ and maximum shapelet orders $n_{{\rm max},s}$ were derived as follows. As described in Paper I, the Hermite basis functions have two natural scales: $\theta_{\rm max}$ corresponding to the overall extent of the basis functions, and $\theta_{\rm min}$ corresponding to the smallest-scale oscillations in the basis functions. These scales are related to the shapelet scale and maximum order by $\theta_{\rm max}
\sim \beta (n_{\rm max}+1)$ and $\theta_{\rm min} \sim \beta (n_{\rm
max}+1)^{-1}$. As $n_{\rm max}$ increases, the large-scale size of the shapelet grows, while its small-scale features become finer. The shapelet thus becomes more extended both in real and in Fourier space. We therefore choose $\theta_{\rm max}$ to be the rms major axis from the FIRST catalog, and $\theta_{\rm min}$ to correspond to the longest baseline of the VLA: $\sim$ 1$''$.8 (rms) in real space. This provides us with a choice for $\beta \simeq (\theta_{\rm
max}~\theta_{\rm min})^{0.5}$ and for $n_{\rm max} \simeq
\frac{\theta_{\rm max}}{\theta_{\rm min}}-1$ for each source.
Solving Equation (\[eq:chi2\]), we then obtain the shapelet coefficients and the covariance matrix using Equations (\[eq:f\]) and (\[eq:cova\]). The results are presented in Fig. \[fig:sim\], where the input images (before the addition of noise), inverse Fourier-transformed $uv$ data (‘dirty’ images), and shapelet-reconstructed images (with Weiner filtering, see §\[weighting\]) of three of the sources are shown. Each image shown is 32$''$ across and the resolution is about 5$''$.4 (FWHM). The poor $uv$ sampling of FIRST and the effect of noise are evident in the dirty images. For both resolved (left panels) and unresolved or marginally resolved (right panels) sources, the reconstructions agree with the inputs very well. The more complicated structure in the central panel is not fully recovered by shapelets. This is expected, since the small-scale structure of the source is not resolved and therefore can not be fully recovered in the reconstruction.
The comparison between the input and shapelet-reconstructed flux density for all sources in the grid pointing is shown in Figure \[fig:flux\]. The shapelet flux density is given by Equation (\[eq:flux\]) and its 1$\sigma$ error by Equation (\[eq:error\]). The source flux densities are well-recovered by the shapelets in an unbiased manner. Note the range of error bars at a given input flux is due to the range of source sizes. For instance, for an input flux density of about 2 mJy, the source with a relatively large error bar has a major axis of about $8''$ (FWHM), while those with small error bars are unresolved or barely resolved (major axis FWHM $< 5''$). In general, we find the shapelet reconstruction from the sparsely sampled and noisy simulated data to be in good agreement with the input (noise-free) image. Note that our method can be used to identify and discard spurious sources arising from sidelobes and other artifacts in the dirty image. Indeed, when we place an extra shapelet centered at a random positions in the field, the coefficients of that shapelet are consistent with zero.
0.1in {width="3.in"} 0.1in
Data
----
Next, we test our method by applying it to one of the FIRST grid pointings (14195+38531). For this purpose, we selected all sources within 23$'$.5 of the phase center from the FIRST catalog with a measured flux density limit (i.e., including the primary beam response) of 0.75 mJy[^1]. For each of the resulting 23 sources, we use the source major axis to estimate $\beta$ and $n_{\rm max}$ as described in the previous section. We then simultaneously fit all the sources for the shapelet coefficients directly in the $uv$ plane. Note that the shapelet coefficients obtained are deconvolved coefficients.
Figure \[fig:data\] shows the reconstruction of three representative sources in the bottom panels. Also shown for comparison are the images of the sources constructed using the standard AIPS CLEAN algorithm with a CLEANing limit of 0.5 mJy (central panel), along with the dirty images (top panel). Each panel is 32$''$ across and the FWHM of the FIRST resolution is 5$''$.4. The shapelet method does not involve image pixels in the modeling; one is therefore free to specify the pixel size when constructing the images. Here the dirty and CLEANed images have pixel sizes of 1$''$.8, while the shapelet images have pixel sizes of 1$''$ and thus show finer details. For demonstration, the shapelet reconstructions have been Weiner-filtered using the resulting covariance matrix. For a direct comparison, they have also been smoothed with a Gaussian kernel with a standard deviation of 2$''$.3, reproducing the ’restoring beam’ of the CLEANed image. We find that the shapelet reconstructions compare well with the CLEANed images.
In further tests, we have encountered cases where a bright source ($>
100$ mJy) lies in or near a grid pointing. We have found that the presence of the bright source does not affect the fit of the other sources in the grid in a noticeable way. Our method can thus well handle the dynamical range of the FIRST survey, which spans more than 3-orders of magnitude. For fainter sources ($< 1$ mJy; i.e., detection SNR $< 6$), the reconstructions are rather poor at times, in contrast to those for brighter sources (which are almost always well fitted). This is of course reasonable, given the larger impact of noise for faint sources.
In Figure \[fig:cova\] we display a portion of the covariance matrix for the shapelet coefficients for the nine sources in the pointing with the highest peak flux densities. The horizontal and vertical lines separate the nine sources. The diagonal line from the lower-left to the upper-right corner represents the variance of the shapelet coefficients. The block-diagonal boxes are the covariance matrix of the coefficients of the nine sources. The off-diagonal blocks quantify the cross-talk between sources. Note that the correlation between coefficients are roughly an order of magnitude smaller than the variance. Figure \[fig:cova\_s4\] shows the error in the shapelet coefficients (n1,n2) of the source shown in the left panels of Figure \[fig:data\]. (These errors are the diagonal segment of the 4$^{th}$ diagonal box counting from the lower left in Fig. \[fig:cova\]). In general, we find that higher-$n$ coefficients tend to be noisier. This is expected since convolution (or, equivalently, $uv$ sampling) suppresses the small scale information encoded by coefficients with large $n$ (see paper I). The covariance matrix thus provides us with useful information on the error in each coefficient, and quantifies cross-talk between coefficients both within and among sources.
{width="6in"}
Computation
-----------
Since the shapelet coefficients of all sources are simultaneously fit to a large number of visibilities, the computing memory required for the calculation is not negligible. We have implemented the method on the UK COSMOS SGI Origin 2000 supercomputer, which has 64 R10000 MIPS processors with a shared-memory structure. Numerically, the shapelet coefficients can be obtained by performing simple matrix operations as in Equation (\[eq:f\]), or by solving the linear least-squares problem, ${\mathbf M ~f} = {\mathbf d}$, using matrix factorization or singular value decomposition, and assuming that the data covariance matrix ${\mathbf C}$ is diagonal. Both methods can be efficiently parallelized. With our binning scheme, the run-time memory required for this particular FIRST grid pointing was about 700 MB, for 23 sources and a total of 177 shapelet parameters. The CPU time required was about 26 seconds with 10 processors or about 5 minutes in scalar mode. For other grid pointings with different numbers of sources, the computation time ranges between 20 and 60 seconds with 10 processors, with a run time memory between 0.5 to 1.5 GB.
Applications to Weak Lensing {#lensing}
============================
Weak gravitational lensing is now established as a powerful method for mapping the distribution of the total mass in the Universe (for reviews see Mellier 1999; Bartelmann & Schneider 2000). This technique is now routinely used to study the dark matter distribution of galaxy clusters and has recently been detected in the field (Wittman et al 2000; van Waerbeke et al 2000; Bacon, Refregier & Ellis 2000; Kaiser et al 2000; Maoli et al 2001; Rhodes, Refregier & Groth 2001; van Waerbeke et al 2001). All studies of weak lensing have been performed in the optical and IR bands, where the images are directly obtained in real space.
{width="3.in"} 0.1in
{width="3.in"} 0.2in
There are a number of reasons to try to extend these studies to interferometric images in the radio band. Firstly, the brightest radio sources are at high redshift, thereby increasing the strength of the lensing signal. Secondly, radio interferometers have a well-known and deterministic convolution beam, and thus do not suffer from the irreproducible effects of atmospheric seeing. Thirdly, existing surveys such as the FIRST radio survey (Becker et al. 1995; White el al. 1997) provide a sparsely sampled but very wide-area survey, which offers the unique opportunity to measure a weak lensing signal on large angular scales (Kamionkowski et al. 1998; Refregier et al. 1998; see also Schneider (1999) for the case of SKA). Finally, surveys at higher frequencies or in more extended antenna configurations could potentially yield very high angular resolution and are not limited by the irreducible effects of the seeing disk in ground-based optical surveys.
Because the distortions induced by lensing are only on the order of 1%, the shapes of background objects must be measured with high precision. In addition, systematic effects such as the convolution beam and instrumental distortions must be tightly controlled. For this purpose, a number of shear measurement methods have been developed. The original method of Kaiser, Squires & Broadhurst (1995) was found to be acceptable for current cluster and large-scale structure surveys (Bacon et al. 2000b; Erben et al. 2000), but are not sufficiently reliable for future high-precision surveys. Consequently, several other methods have been proposed (Kuijken 1999; Kaiser 2000; Rhodes, Refregier & Groth 2000, Berstein & Jarvis 2001).
Recently, Refregier & Bacon (2001, Paper II) developed a new method based on shapelets and demonstrated its simplicity and accuracy for ground-based surveys. It is thus straightforward to apply this method to interferometric measurements. Indeed, the shapelet coefficients $f_{\mathbf n}$ which we derive from the fit on the $uv$ plane (after co-adding if required) are already deconvolved from the effective beam and can thus be directly used to estimate the shear. This can be done using the estimators for the shear components $\gamma_{1}$ and $\gamma_{2}$ which are given by (see Paper II) $$\gamma_{i} = \frac{ f_{\mathbf n} - \langle f_{\mathbf n}
\rangle}{\sum_{\mathbf m}
\hat{S}_{i{\mathbf n}{\mathbf m}} \langle f_{\mathbf m} \rangle },$$ where the sum is over even (odd) shapelet coefficients for $\gamma_{1}$ ($\gamma_{2}$) and the brackets denote an average over an (unlensed) object ensemble. The matrix $\hat{S}_{i{\mathbf n}{\mathbf
m}}$ is the shear matrix, and can be expressed as simple combinations of ladder operators in the QHO formalism. These estimators for individual shapelet components are then optimally weighted and combined to provide a minimum-variance estimator for the shear. This permits us to achieve the highest possible sensitivity (while remaining linear in the surface brightness) by using all the available shape information of the lensed sources.
In Kamionkowski et al. (1998) and Refregier et al. (1998), it has been shown that the FIRST radio survey is a unique database for measuring weak lensing by large-scale structure on large angular scales. In a future paper, we will apply the method described here to this survey, search for the lensing signal, and, from its amplitude, derive constraints on cosmological parameters.
Conclusions {#conclusion}
===========
We have presented a new method for image reconstruction from interferometers. Our method is based on shapelet decomposition and is simple and robust. It consists of a linear fit of the shapelet coefficients directly in the $uv$ plane, and thus permits a full correction of systematic shape distortions caused by bandwidth smearing, time-averaging and non-coplanarity. Because the fit is linear in the shapelet coefficients it can be implemented as simple matrix multiplications. It provides the full covariance matrix of the shapelet coefficients which can then be used to estimate errors and cross-talk in the recovered shapes of sources. We have shown how source shapes from different pointings can be easily co-added using a weighted sum of the recovered shapelet coefficients. We have also described how the shapelet parameters could be combined to derive optimal image reconstruction, photometry, astrometry and pointing co-addition.
Our method can be efficiently implemented on parallel computers. We find that a fit to all the sources in a FIRST grid pointing takes about 1 minute on an Origin 2000 supercomputer with 10 processors (10 minutes in scalar mode). Because we are fitting all sources simultaneously, 0.5 to 1.5 GB of memory is required.
To test our methods, we considered the observing conditions of the FIRST radio survey (Becker et al. 1995; White et al. 1997) whose snapshot mode yields a sparse sampling in $uv$ space. Using numerical simulations tuned to reproduce the conditions of FIRST, we find that the sources are well-reconstructed with our method. We have also applied our method to a FIRST snapshot pointing and found that the shapes are well-recovered. The reconstruction of our method compares well with the CLEAN reconstruction, without suffering the potential biases inherent in the latter method. Our method is thus well-suited for applications requiring quantitative and high-precision shape measurements.
In particular, our method is ideal for the measurement of the small distortions induced by gravitational lensing in the shape of background sources by intervening structures. Such a measurement from CLEANed images may well not be practical since the systematic distortions induced by that method are very difficult to control. (One could perhaps imagine running numerical simulations to calibrate the shear estimator, but this would be both computationally expensive and rather uncertain). We have shown how our results can be combined with the shear measurement method described in Refregier & Bacon (2001) to derive a measurement of weak lensing with interferometers. This is facilitated both by the fact that our recovered shapelet coefficients are already deconvolved from the effective (dirty) beam, and as a consequence of the remarkable properties of shapelets under shears.
Our method therefore opens the possibility of high-precision measurements of weak lensing with interferometers. While to date all weak-lensing studies have been carried using optical data (and therefore in real space), an interferometric measurement of weak lensing in the radio band is very attractive (Kamionkowski et al. 1998; Refregier et al. 1998; Schneider 1999). Indeed, the lensing signal is expected to be larger because radio sources have a higher mean redshift. In addition, such a measurement would not suffer from the irreproducible effects of atmospheric seeing. Instead, the effective (dirty) beam is fully known for interferometers and the noise properties of the antennas are well-understood. As a result, the impact of systematic effects, the crucial limitation in the search for weak lensing, are expected to be lower with radio interferometers. In a future paper, we will describe our measurement of weak lensing by large-scale structure with the FIRST survey using the present method.
[99]{} Bacon, D., Refregier, A., Ellis, R., 2000, MNRAS, 318, 625 Bacon, D. Refregier, A., Clowe, D., & Ellis, R., 2000b, to appear in MNRAS, preprint astro-ph/0007023 Bartelmann, M., & Schneider, P., 2000, preprint astro-ph/0007023 Becker, R.H., White, R.L., Helfand, D.J., 1995, , 450, 559 Bernstein, G.M., & Jarvis, M., 2001, accepted by AJ, astro-ph/0107431 Clark, B.G., 1980, A&A, 89, 377 Condon,J.J., Cotton,W.D., Greisen, E.W., Yin, Q.F., Perley, R.A., Taylor, G.B., Broderick, J.J., 1998, , 115, 1695 Cornwell, T,J., 1983, A&A, 121, 281 Cornwell, T.J. & Evans, K.F., 1985, A&A, 143, 77 Erben T., van Waerbeke, L., Bertin, E., Mellier, Y., Schneider, P., 2001, A&A, 366, 717 Hogbom, J.,A., 1974, A&AS, 15, 417 Kaiser, N., 2000, ApJ, 537, 555 Kaiser, N., Wilson, G., Luppino, G. A., 2000, preprint astro-ph/0003338 Kamionkowski, M., Babul, A., Cress, C., Refregier, A., 1998, MNRAS, 301, 1064 Kuijken, K., 1999, A&A, 352, 355 Lupton, R., 1993, Statistics in Theory and Practice, Princeton University Press Maoli, R. et al, 2001, A&A, 368, 766 Mellier, Y., 1999, ARA&A, 37, 127 Narayan, R., & Bartelmann, M., 1999, in Formation of Structure in the Universe. Ed. by Dekel, A. and Ostriker, J.P., p.360 (preprint astro-ph/9606001) Perley, R.A., Schwab, F.R., & Bridle, A.H., 1989, Synthesis Imaging in Radio Astronomy, A.S.P.C.S. Vol. 6 Press, W.H., Teukolsky, S.A., Vetterling, W.T., Flannery, B.P., 1987, Numerical Recipes, Cambridge University Press Refregier et al. 1998, in Proc. of the XIVth IAP meeting, Wide-Field Surveys in Cosmology, held in Paris in May 1998, eds. Mellier, Y. & Colombi, S. (Paris: Frontieres), preprint astro-ph/9810025 Refregier, A., 2001, (Paper I) submitted to MNRAS, preprint astro-ph/0105178 Refregier, A. & Bacon, D.J., 2001, (Paper II) submitted to MNRAS, preprint astro-ph/0105179 Rhodes, J., Refregier, A., & Groth, E., 2000, ApJ, 536, 79 Rhodes, J., Refregier, A., & Groth, E., 2001, to appear in ApJL, preprint astro-ph/0101213 Schneider, P., 1999, in Perspectives on Radio Astronomy, Scientific Imperatives at cm and m wavelengths, Proceedings of a workshop in Amsterdam, April 1999, preprint astro-ph/9907146 Schwarz, U.J., 1978, A&A, 65, 345 Taylor, G.B., Carilli, C.L., & Perley, R.A., 1999, Synthesis Imaging in Radio Astronomy II, A.S.P.C.S. Vol. 180 Thompson, A.R., Moran, J. & Swenson, JR., G.W, 1986, Interferometry and Synthesis in Radio Astronomy (Wiley-Interscience) van Waerbeke, L. et al, 2000, A&A, 358, 30. van Waerbeke, L. et al, 2001, submitted to A&A, preprint astroph/0101511. White, R.L., Becker, R.H., Helfand, D.J., Gregg, M.D., 1997, , 475, 479 Wittman, D., Tyson, J. A., Kirkman, D., Dell’Antonio, I., Bernstein, G., 2000, Nature, 405, 143.
[^1]: As explained in Becker et al. (1995), a map flux density of 0.75 mJy corresponds to a source flux density of 1.0 mJy owning to “CLEAN bias” corrections.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the condition of black hole formation in five-dimensional space-time. We analytically solve the constraint equations of five-dimensional Einstein equations for momentarily static and conformally flat initial data of a spheroidal mass. We numerically search for an apparent horizon in various initial hypersurfaces and find both necessary and sufficient conditions for the horizon formation in terms of inequalities relating a geometric quantity and a mass defined in an appropriate manner. In the case of infinitely thin spheroid, our results suggest a possibility of naked singularity formation by the spindle gravitational collapse in five-dimensional space-time.'
address:
- ' Department of Mathematics and Physics, Graduate School of Science, Osaka City University, Osaka 558-8585, Japan '
- 'Department of Physics, Gakushuin University, Tokyo 171-8588, Japan '
author:
- 'Chul-Moon Yoo[^1], Ken-ichi Nakao[^2]'
- 'Daisuke Ida[^3]'
title: |
Hoop Conjecture in Five-dimensions\
-Violation of Cosmic Censorship-
---
Introduction {#sec:1}
============
In an attempt to unify fundamental forces including gravity, the possibility that the space-time dimensions of our universe is higher than four has been much discussed. Such higher-dimensional theories need mechanism to reduce the space-time dimensions down to four, for example via Kaluza-Klein type compactifications of extra dimensions, so as to be consistent with the observed world. The brane world scenario is another attractive idea of dimensional reduction. In this scenario, the standard model particles are confined to the boundary of a higher-dimensional space-time and only gravity can propagate in the extra dimensions. Models of the brane world scenario with large extra dimensions compared to the four-dimensional Planck scale ($\approx 1.6\times 10^{-33}$cm) have been considered in some recent works [@BRANE]. According to these models, the fundamental (namely, higher-dimensional) Planck scale may be set to rather low energy scale, even to $1$TeV, of which low energy effects just alter the short distance behaviour of classical gravitational interactions. The discrepancy in the gravitational interaction between the four and higher-dimensional theories arises only at the length scale below $0.1{\rm mm}$ so that it is consistent with the gravitational experiments [@experiments]. In such TeV gravity models, it is suggested that small black holes are produced in accelerators, such as the CERN Large Hadron Collider [@LHC] or in high energy cosmic ray events [@Feng:2001ib].
In order to understand physical phenomena caused by strong gravitational fields, the criterion for black hole formation is very crucial. In the case of four-dimensional Einstein gravity, such a criterion is well known as the hoop conjecture [@HC]. Hoop conjecture claims that the necessary and sufficient condition for black hole formation is given by the following; [*Black holes with horizons form when and only when a mass $M$ gets compacted into a region whose circumference in every direction is ${\cal C}\lesssim 4\pi G_4M$*]{}, where $G_4$ is the gravitational constant in four-dimensional theory of gravity. It is remarkable that no serious counterexample against hoop conjecture has been presented. However, at first glance, hoop conjecture is not valid in higher-dimensional Einstein gravity [@Nakao:2001pn]; there is black string solutions in five or higher-dimensions, which have infinitely long event horizons, while hoop conjecture claims that any length scale characterizing black hole should be less than the gravitational length scale determined by the Schwarzschild radius.
Recently, two of the present authors, DI and KN, proposed a higher-dimensional version of hoop conjecture [@IN]. Here we call it the hyperhoop conjecture in the sense that it is a possible generalization of the original hoop conjecture; [ *Black holes with horizons form when and only when a mass $M$ gets compacted into a region whose $(D-3)$-dimensional area $V_{D-3}$ in every direction is* ]{} $$V_{D-3}\lesssim G_D M, \label{eq:hyperhoop}$$ where $G_D$ is the gravitational constant in $D$-dimensional theory of gravity, and the ($D-3$)-dimensional area means the volume of $(D-3)$-dimensional closed submanifold of a spacelike hypersurface. Hereafter we call this $(D-3)$-dimensional closed submanifold [*the hyperhoop*]{}. The necessity of the condition (\[eq:hyperhoop\]) was confirmed in the case of momentarily static and conformally flat initial data sets of an axially symmetric line, disk and thin ring source for the five-dimensional Einstein equations [@IN] and for the system of point-particles [@Yoshino:2002tx]. Consistent results with the previous ones were obtained by Barrabés [*et al*]{} [@BFL]. They derived two inequalities for ($D-3$)-dimensional volume as the necessary and sufficient conditions for apparent horizon formation in the case of a ($D-2$)-dimensional convex thin shell collapsing with the speed of light in a $D$-dimensional space-time.
The purpose of the present paper is to study both the necessity and in particular [*sufficiency*]{} of the inequality (\[eq:hyperhoop\]) for the horizon formation in different situations from the case treated in Ref. [@BFL]. We consider the momentarily static and conformally flat four-dimensional initial hypersurfaces in which a four-dimensional homogeneous spheroid is put as a gravitational source. This procedure has been implemented by Nakamura [*et al.*]{} [@Nakamura:1988zq]. We apply their method to higher-dimensional case. Then, we analytically solve the constraint equations for five-dimensional Einstein equations. In order to investigate the validity of hyperhoop conjecture, we numerically search for an apparent horizon and calculate the ratio $V_2/G_5M$ for substantially various hyperhoops.
This paper is organized as follows. In Sec. \[sec:2\], assuming five-dimensional Einstein gravity, we derive the constraint equations for conformally flat initial hypersurfaces and then give analytic solutions of these equations for a homogeneous mass of a spheroidal shape. In Sec. \[sec:3\], we search for an apparent horizon in initial hypersurfaces with various shapes of a homogeneous spheroid including infinitely thin case by numerically solving a second order ordinary differential equation. This equation corresponds to the minimum volume condition for a three-dimensional closed submanifold of an initial hypersurface. The suggestion of the naked singularity formation is given in this section. In Sec. \[sec:4\], we define $V_2/G_5M$ in a reasonable manner and then give a procedure to select the hyperhoop with minimal value of $V_2/G_5M$. In Sec. \[sec:5\], we show numerical results and their implication to the necessary and sufficient condition for the horizon formation. Finally, Sec. \[sec:6\] is devoted to summary. In Appendix \[sct:new\], we derive analytic solutions for the Newtonian gravitational potential of an ellipsoid in arbitrary space dimension. In Appendix \[sct:nec\], the necessary condition of black hole formation based on Ref.[@IN] is derived.
In this paper, we adopt the unit of $c=1$. We basically follow the notations and sign conventions in Ref.[@Wald].
A Momentarily static spheroid in five-dimensional space-time {#sec:2}
============================================================
Let us consider an initial data set $\left(h_{ab}, K_{ab}\right)$ in a four-dimensional spacelike hypersurface $\Sigma$, where $h_{ab}$ is the induced metric in $\Sigma$ and $K_{ab}$ is the extrinsic curvature which represents how $\Sigma$ is embedded in the five-dimensional space-time. Denoting the unit normal vector to $\Sigma$ by $n^a$, $h_{ab}$ and $K_{ab}$ are, respectively, written as $$\begin{aligned}
h_{ab}&=&g_{ab}+n_a n_b, \\
K_{ab}&=&-h_a^c\nabla_c n_b,\end{aligned}$$ where $\nabla_c$ is the covariant derivative in the five-dimensional space-time.
The initial data set $(h_{ab},K_{ab})$ has to satisfy the Hamiltonian and momentum constraints given by $${\cal R}-K_{ab}K^{ab}+K^2=24\pi^2 G_5 \rho$$ and $$D_b\left(K^{ab}-h^{ab}K\right)=12\pi^2G_5 J^a,$$ where $\rho$ and $J^a$ are the energy density and energy flux for normal line observers to $\Sigma$, $D_a$ and ${\cal R}$ are the covariant derivative within $\Sigma$ and the scalar curvature of $h_{ab}$, and $G_5$ is the gravitational constant in five-dimensional theory of gravity. In this paper, we focus on momentarily static and conformally flat initial hypersurfaces: $$\begin{aligned}
K_{ab}&=&0\\
h_{ab}&=&f^2\delta_{ab},\end{aligned}$$ where $\delta_{ab}$ is the metric tensor of four-dimensional Euclidean space. We also require the axial symmetry in the sense that the metric on $\Sigma$ have the form $$dl^2=f^2(R,z)\left[dR^2
+R^2\left(d\vartheta^2+\sin^2\vartheta d\varphi^2\right)+dz^2\right],$$ where $0\leq R<+\infty$ and $-\infty<z<+\infty$ while $\vartheta$ and $\varphi$ are the round coordinates. Then the momentum constraint leads to zero flux condition $J^a=0$, and the Hamiltonian constraint becomes $${\partial^2 f \over \partial R^2}
+{2\over R}{\partial f \over\partial R}
+{\partial^2 f \over \partial z^2}
=-4\pi^2 G_5 f^3\rho.\label{eq:rap}$$ Here we note that the Hamiltonian constraint (\[eq:rap\]) is equivalent to the Poisson equation for axi-symmetric Newtonian gravitational potential. Let us consider the density profile respecting the axial symmetry given by $$f^3\rho=
\left\{\begin{array}{lll}
2M/\pi^2 a^3b & \mbox{for} &
R^2/a^2+z^2/b^2{\leq}1, \\
~~~~0 & \mbox{for} & \mbox{elsewhere},
\end{array}\right. \label{eq:rhon}$$ where $a$, $b$ and $M$ are constant parameters.
We consider the gravitational field of an isolated body, so that we assume the asymptotic condition given by $$f\rightarrow 1 ~~~{\rm for}~~r\rightarrow\infty,$$ where $$r=\sqrt{R^2+z^2}.$$ The regular solution is then obtained as $$\begin{aligned}
f&=&1-
\frac{4G_5M\left[b(2a+b)R^2+3a^2 z^2 -3a^2b(a+b)
\right] }{3a^3b{\left( a + b \right) }^2} \nonumber \\
&&~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
{\rm for}~~ \frac{R^2}{a^2}+\frac{z^2}{b^2}~{\leq}~1,
\\
&& \nonumber \\
f&=&1 - \frac{4G_5M}{3 e^4b^4}
\Biggl[
2R^2-6z^2+3e^2b^2+\sqrt{F^2-e^4b^4} \nonumber \\
&\times&
\biggl(
{2e^2b^2 R^2 \over (F-e^2b^2)^2}
-{2R^2-3z^2+3e^2b^2 \over F-e^2b^2}
+{3z^2 \over F+e^2b^2}
\biggr)
\Biggr]\nonumber \\
&&~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~{\rm for}~~\frac{R^2}{a^2}+\frac{z^2}{b^2}>1,%
\label{eq:outerf}\end{aligned}$$ where $e$ is the eccentricity defined by $$e=\sqrt{1-{a^2\over b^2}}$$ and $F$ is a function of $R$ and $z$ defined by $$\begin{aligned}
F&=&F(R,z;a,b)\nonumber \\
&=&R^2+z^2+\sqrt{4b^2R^2 - 4a^2\left( b^2 - z^2 \right)
+{\left( a^2 + b^2 - R^2 - z^2 \right)}^2}.\nonumber \\\end{aligned}$$ The detailed derivation of the above solution is given in Appendix \[sct:new\]. Newtonian gravitational potential of an ellipsoid in Euclidean space of arbitrary dimensions is shown there. Here we only investigate the prolate case $a<b$.
In the thin limit $a\rightarrow0$ with $M$ and $b$ fixed, two disconnected singularities appear at the poles $(R,z)=(0,\pm b)$ of the resultant “singular spheroid”. In order to see this, we evaluate the Kretschmann invariant $$I={\cal R}^{abcd}{\cal R}_{abcd},\label{eq:inv}$$ where ${\cal R}_{abcd}$ is the four-dimensional Riemann tensor of the spacelike hypersurface. Typical examples are shown in Figs. \[fig:inv\] and \[fig:invR0\]. The coordinate values and the Kretschmann invariant $I$ is normalized by $$r_{\rm s}:=\sqrt{G_5 M}.$$ It will be found in the next section that $r_{\rm s}$ is the coordinate radius of the apparent horizon in the case of a point source $a=b=0$.
![The logarithm to the base 10 of the Kretschmann invariant $I$ is plotted as function of the coordinates $R$ and $z$ in the case $a=0$ and $b=2r_{\rm s}$.[]{data-label="fig:inv"}](inv.eps)
![The logarithm to the base 10 of the Kretschmann invariant $I$ on the polar axis $(R=0)$ is plotted as function of $z$ in the case $a=0$ and $b=2r_{\rm s}$. The value of $I$ is diverge at only $(0,b)$. []{data-label="fig:invR0"}](invR0.eps)
Here it should be noted that the Kretschmann invariant $I$ is finite between these two singularities on the singular spheroid, $R=0$ and $|z|<b$. Further, we can see that the energy density $\rho$ is also finite there. The conformal factor on the surface of the spheroid is given by $$\begin{aligned}
f&=&f_{\rm sf}(z)\nonumber \\
&:=&1+{4G_5M \over 3ab^2(a+b)^2}\left[b^2(a+2b)-2(b-a)z^2\right].~~\end{aligned}$$ Therefore we find that the energy density at the surface becomes in the thin limit $a\rightarrow0$ as $$\rho=\rho_{\rm sf}(z):={2M\over \pi^2a^3 b f_{\rm sf}^{3}}
\longrightarrow {27b^8 \over 256\pi^2 G_5^3 M^2 (b^2-z^2)^3}.$$ Here note that the inequality $f(R,z)\geq f_{\rm sf}(z)$ is satisfied within the spheroid and hence $0\leq\rho(R,z)\leq \rho_{\rm sf}(z)$. Therefore $\rho$ is finite except for the poles $z=\pm b$ even in the thin limit $a\rightarrow0$. This fact means that the scalar polynomials of the five-dimensional Riemann tensor are also finite if the stress of matter fields is assumed to be reasonable. For example, assuming the dust as the matter and adopting Gaussian normal coordinate, Einstein equations leads to $${\partial K^a{}_b \over \partial t}={\cal R}^a{}_b
-4\pi^2 G_5\rho ~\delta^a{}_b$$ on the momentarily static initial hypersurface, where ${\cal R}^a{}_b$ is the four-dimensional Ricci tensor of this hypersurface. Here note that ${\cal R}^a{}_b$ is finite in the region of finite Kretschmann invariant $I$ since the metric of the spacelike hypersurface is positive definite. Therefore the finiteness of the energy density $\rho$ guarantees that the time derivative of the extrinsic curvature $K^a{}_b$ is finite in the region of finite Kretschmann invariant $I$. This means that the scalar polynomials of Riemann tensor of five-dimensional space-time are also everywhere finite except for the poles of the singular spheroid since these are expressed as polynomials of $\partial K^a{}_b/\partial t$, ${\cal R}^a{}_b$, $K^a{}_b$ and $D_cK^a{}_b$ in the Gaussian normal coordinate. Further, we can see that this singular spheroid except for the poles corresponds to spatial infinity. Consider a curve $z=\zeta(R)$ connecting two points $R=R_1$ and $R=R_2$ ($R_1<R_2$) and assume that both $R_1$ and $R_2$ are sufficiently small and $0<\zeta(R)<z_{\rm max}<b$. In this situation, the function $F$ is written as $$F(R,\zeta;0,b)=b^2+\frac{2b^2}{b^2-\zeta^2}R^2+\mathcal{O}(R^3).$$ Hence substituting this for Eq. (\[eq:outerf\]), we find $$f\left(R,\zeta\right)={8(b^2-\zeta^2)^{3/2}G_5 M \over 3b^4 R}+\mathcal{O}(R^0).$$ The proper length between $R=R_1$ and $R_2$ along the curve $z=\zeta(R)$ is bounded below as $$\begin{aligned}
&&\int_{R_1}^{R_2}f(R,\zeta)\sqrt{1+\left({d\zeta\over dR}\right)^2}~dR
\nonumber \\
&&\geq\int_{R_1}^{R_2}f(R,\zeta)dR
\simeq {8G_5 M\over 3b^4}
\int_{R_1}^{R_2}(b^2-\zeta^2)^{3/2}{dR\over R} \nonumber \\
&&\geq{8G_5 M\over 3b^4}(b^2-z_{\rm max}^2)^{3/2}\int_{R_1}^{R_2}{dR\over R}
={8G_5 M\over 3b^4}(b^2-z_{\rm max}^2)^{3/2}\ln{R_2 \over R_1}. \end{aligned}$$ We can see from the above equation that the proper length diverges in the limit of $R_1\rightarrow 0$ with $R_2$ fixed. Therefore each point on the singular spheroid except for the poles, $R=0$ and $|z|<b$, is spacelike infinity.
Apparent horizons {#sec:3}
=================
In a momentarily static initial hypersurface in five-dimensional asymptotically flat space-time, an apparent horizon is a three-dimensional closed marginal surface. Because of the axial symmetry of the initial hypersurface, the apparent horizon will also be axially symmetric and thus will be expressed by $r=r_{\rm m}(\xi)$ in the present case, where $$\xi=\arctan{R\over z}.$$ Then $r=r_{\rm m}(\xi)$ is a closed marginal surface only if $r_{\rm m}(\xi)$ satisfies $$\begin{aligned}
&&{\ddot r}_{\rm m}-{4 {\dot r}_{\rm m}{}^2\over r_{\rm m}}
-3r_{\rm m}+\frac{r_{\rm m}{}^2+{\dot r}_{\rm m}{}^2}{r_{\rm m}}
\nonumber\\&&
\times\left[
\frac{2{\dot r}_{\rm m}}{r_{\rm m}}\cot\xi
-{3\over f}\left({\dot r}_{\rm m}\sin\xi+r_{\rm m}\cos\xi\right)
\frac{\partial f}{\partial z}
+{3\over f}\left({\dot r}_{\rm m}\cos\xi-r_{\rm m}\sin\xi\right)
\frac{\partial f}{\partial R}\right]=0
\label{eq:joubi}\end{aligned}$$ with boundary conditions ${\dot r}_{\rm m}=0$ at $\xi=0$ and $\pi$, where a dot means the derivative with respect to $\xi$. The derivation of Eq. (\[eq:joubi\]) is shown in Appendix \[sec:margi\]. Since the present system has a reflection symmetry with respect to $z=0$, the apparent horizon should satisfy $${\dot r}_{\rm m}=0~~{\rm at}~~{\xi= \frac{\pi}{2}}.
\label{eq:BC}$$ In the case of a point source $a=b=0$, we can analytically solve Eq. (\[eq:joubi\]) and find $$r_{\rm m}=r_{\rm s}.$$
Replacing derivatives with respect to $\xi$ in Eq. (\[eq:joubi\]) by finite differences, we numerically search for solutions of this equation by relaxation method [@ref:SMMN]. If apparent horizons exist in the initial hypersurface, we can find solutions of Eq. (\[eq:joubi\]). Typical examples are shown in Fig. \[fig:aph\]. The coordinate values are normalized by $r_{\rm s}$.
![Apparent horizons for each shape of the four-dimensional spheroid are depicted in $(R,z)$-plane. The solid line shows the surface of the spheroid. The dashed line shows the apparent horizon if it is present. []{data-label="fig:aph"}](horizon.eps)
In the case of singular source $a=0$, we also find solutions of Eq. (\[eq:joubi\]) satisfying the boundary condition (\[eq:BC\]). In the case of $b\leq1.48 r_{\rm s}$, there is an apparent horizon enclosing whole the singular source. By contrast, for $b\geq 1.49r_{\rm s}$, the marginal surface covers only a central part of the singular source, and the space-time singularities at the poles $(R,z)=(0,\pm b)$ are not enclosed by the marginal surface (see Fig. \[fig:aphs\]). In this case, this marginal surface is not a closed three-surface and thus is not the apparent horizon since as mentioned in the previous section, $|z|<b$ on the polar axis $R=0$ is the spacelike infinity. This result implies that a long enough spindle singular source can produce naked singularities, which is quite different from the singular line source in Ref. [@IN].
![In the case of $a=0$, apparent horizons for each $b$ are depicted. []{data-label="fig:aphs"}](horizonsin.eps)
How to check hyperhoop conjecture {#sec:4}
=================================
We have presented the statement of hyperhoop conjecture in Sec. \[sec:1\]. Here we should note that in general, the mass $M$ in hyperhoop conjecture (and also in hoop conjecture) is not the total mass of the system but the mass encircled by the hyperhoop (the hoop). Therefore to check the sufficiency of this conjecture, we have to confirm that in an initial hypersurface without an apparent horizon, there is no hyperhoop satisfying $$V_{D-3} \lesssim G_D M_{\rm in}, \label{eq:hutou}$$ where $M_{\rm in}$ is not the total mass of the system but the one included in the region which is encircled by this hyperhoop.
Since we cannot try all the possible hyperhoops in $\Sigma$, we focus on the axi-symmetric hyperhoops which have the reflection symmetry with respect to $z=0$. Further, these hyperhoops are expressed in the form $r=r_{\rm h}(\xi)$ and $\vartheta=\pi/2$ since the spheroid is assumed to be prolate $a<b$.(In the oblate case, we should consider two-surface of $z=0$ and $R=$constant, where the hyperhoop is parameterized by two parameters $\xi$ and $\varphi$.) The two-dimensional area $V_2$ of the hyperhoop $r=r_{\rm h}(\xi)$ and $\vartheta=\pi/2$ is then given by $$V_2=\int_0^{2\pi}d\varphi\int^{\pi}_0 d\xi
f^2\sqrt{{\dot r}_{\rm h}{}^2
+r_{\rm h}{}^2}~r_{\rm h}\sin\xi=4\pi\int^{\pi/2}_0 d\xi
f^2\sqrt{{\dot r}_{\rm h}{}^2
+r_{\rm h}{}^2}~r_{\rm h}\sin\xi,$$ where we have taken account of the reflection symmetry in the second equality.
In the framework of general relativity, there is no unique definition of the mass in a quasilocal manner although the total mass is well defined for the isolated system. This is one of reasons why it is very difficult to formulate precisely the hoop and also hyperhoop conjectures. However, despite of this mathematical indefiniteness, the hoop and hyperhoop conjectures might be useful in understanding black hole formation processes on the physical ground. In this sense, all the reasonable definitions of quasilocal mass will be meaningful in the hoop and hyperhoop conjectures since these might give the results not so different from each other. Here we adopt the following simple definition for the mass $M_{\rm in}$ as $$M_{\rm in}=8\pi\int^{\pi/2}_0d\xi
\int^{r_{\rm h}\left(\xi\right)}_0 dr \rho f^3 r^3\sin^2\xi.
\label{eq:Min-definition}$$ Then we numerically calculate $V_2/G_5 M_{\rm in}$ for various hyperhoops in an initial hypersurface of a spheroid. Hereafter for notational simplicity, we introduce $$\Gamma:={V_2\over 16\pi G_5 M_{\rm in}}, \label{eq:gam}$$ where $16\pi G_5M_{\rm in}$ is the minimal value of $V_2$ in the case of a point source whose mass is $M_{\rm in}$.
Our first task is the selection of relevant hyperhoops from all the possible hyperhoops expressed by $r=r_h(\xi)$ and $\vartheta=\pi/2$ with the reflection symmetry with respect to $z=0$. A hyperhoop is represented as a continuous curve from a point on $z$-axis to a point on $R$-axis in the first quadrant of $(R,z)$-plane. Thus first, we select a point $z=z_0$ on $z$-axis and consider various hyperhoops which start from this point. Since the spheroid of the mass is assumed to be prolate, we expect that the significant hyperhoops are also prolate, and thus we restrict the curves within the region of $R^2+z^2\leq z_0^2$. The number of possible curves is still infinite. Therefore we impose further restrictions. Consider 100 points $\left(R,z\right)=\left(R^{(1)}_{j_i}(z_0),z_i(z_0)\right)$ $(i,j_i=1,2,..,10)$ within this spherical region; $z_i(z_0)$ is determined in the following manner $$z_{i}(z_0)={z_0\over10}(10-i),$$ and then $R_{j_i}^{(1)}(z_0)$ is given as $$R^{(1)}_{j_i}(z_0)={\sqrt{z_0{}^2-z_{i}{}^2(z_0)}\over 10}j_i.$$ We consider the curves composed of ten straight lines connected at $(R,z)=\left(R^{(1)}_{j_i}(z_0),z_i(z_0)\right)$; $i$ is in order from $1$ to $10$, and then for each $i$, $j_i$ is appropriately chosen among ten integers from 1 to 10.
By investigating several randomly chosen hyperhoops, we found that sharply bended one might have a value of $\Gamma$ larger than the ones not so sharply bended. Hence in the systematic numerical search, a line from the point of $(R,z)=\left(R^{(1)}_{j_i}(z_0),z_i(z_0)\right)$ to $(R,z)=\left(R^{(1)}_{j_{i+1}}(z_0),z_{i+1}(z_0)\right)$ is adopted as a constitutive one of the hyperhoop only if $j_{i+1}$ is equal to $j_i$ or $j_i\pm 1$ for $j_i>1$ and is equal to $j_i$ or $j_i+1$ for $j_i=1$. All the possible constitutive lines are depicted in Fig. \[fig:keiro1\]. This means that we consider only the hyperhoops each of which is a connected set of ten constitutive lines. Then we calculate $\Gamma$ for each hyperhoop and search for the minimal one which is specified by a set of ten integers $\{m_i\}$ in the manner of $j_i=m_i$. Further for several values of $z_0$, we carry out the same calculations as the above. We select the values of $z_0$ at even intervals $0.1r_s$ and search for the hyperhoop with the smallest value of $\Gamma$. Finally, we obtain the minimal one which is specified by a set of ten integers and one real number $\{m_i,q\}$, where $q$ is the value of $z_0$. The above hyperhoop $\{m_i,q\}$ might not be exactly minimal since the hyperhoops obtained by the above procedure are too restrictive. Thus we might find hyperhoops smaller than the one $\{m_i,q\}$ in the following refinement. We consider a neighbourhood $R^{(1)}_{m_{i}-1}(q_0)\leq R \leq R^{(1)}_{m_{i}+1}(q_0)$ of the hyperhoop $\{m_i,q_0\}$, where $q_0$ is a value which is equal or near to $q$. In this region, we put further grid points at $$R=R^{(2)}_{k_i}(q_0)
=R^{(1)}_{m_{i}-1}(q_0)
+{\sqrt{q_0^2-q_i^2(q_0)}\over 50}(k_i-1),~~
(k_i=1,2,\cdots,11),$$ where $$q_i(q_0)=\frac{q_0}{10}\times (10-i).$$ Then by the same procedure as in the previous search for the minimal hyperhoop, we will obtain the hyperhoop with $\Gamma$ smaller than the previous one (see Fig. \[fig:keiro2\]). Further for several value of $q_0$ in the vicinity of $q$, we carry out the same calculations as the above. We select the 11 values $$q-0.1r_s+0.02r_s(l-1),~~(l=1,2,\cdots,11)$$ as $q_0$ and search for the hyperhoop with the smallest value of $\Gamma$.
![ The connecting points (left figure) and the hyperhoops which we calculate for a value of $z_0$ (right figure) in first search.[]{data-label="fig:keiro1"}](keirof.eps)
![ The hyperhoop having the smallest value of $\Gamma$ in first search (left figure) and the hyperhoops which we calculate for a value of $q_0$ (right figure) in second search .[]{data-label="fig:keiro2"}](keirof2.eps)
Numerical Results {#sec:5}
=================
For various $a$ and $b$, numerical results of the minimal value of $\Gamma$ are listed in TABLE \[tab:re\]. Hereafter we denote the minimal value of $\Gamma$ by $\Gamma_{\rm min}$ which is a function of $a$ and $b$.
We see from TABLE \[tab:re\] that $\Gamma_{\rm min}$ is not smaller than unity in the case of no apparent horizon. This implies that the inequality (\[eq:hutou\]) is really a sufficient condition in the situations studied here.
$b \backslash a$ 0 0.01 0.1 0.3 0.5 0.7 0.9
------------------ ----------- ----------- ----------- ----------- ----------- ----------- -----------
1.0 1.05(Yes) 1.05(Yes) 1.05(Yes) 1.05(Yes) 1.04(Yes) 1.03(Yes) 1.01(Yes)
1.5 1.11(No) 1.11(Yes) 1.11(Yes) 1.10(Yes) 1.10(Yes) 1.09(No) 1.08(No)
2.0 1.17(No) 1.17(Yes) 1.17(Yes) 1.17(No) 1.16(No) 1.16(No) 1.17(No)
2.5 1.23(No) 1.23(Yes) 1.23(No) 1.23(No) 1.23(No) 1.24(No) 1.26(No)
3.0 1.29(No) 1.29(Yes) 1.29(No) 1.29(No) 1.30(No) 1.31(No) 1.35(No)
3.5 1.35(No) 1.34(No) 1.34(No) 1.34(No) 1.35(No) 1.38(No) 1.43(No)
4.0 1.40(No) 1.38(No) 1.38(No) 1.38(No) 1.40(No) 1.44(No) 1.51(No)
Next, let us study the necessity of hyperhoop conjecture (\[eq:hutou\]). By the investigation of the singular line source of a “constant” line energy density studied in Ref. [@IN], we obtain a necessary condition as follows $$V_2\lesssim\frac{\pi}{2}
\times16\pi G_5M.
\label{eq:nec}$$ The derivation of this quantity $\pi/2$ is shown in Appendix \[sct:nec\]. The counterexample for this condition is not found in TABLE \[tab:re\]. Here, it is again noted that a closed marginal surface is formed only when $b\leq1.48r_{\rm s}$, and thus our result of $a=0$ suggests the formation of naked singularities in the case $b\geq 1.49r_{\rm s}$. If it is true, the naked singularities might form by the gravitational collapse starting from the initial data of nonvanishing but sufficiently small $a$ and $b$ larger than $1.49r_{\rm s}$. If a naked singularity exists, the apparent horizon does not necessarily mean the existence of a black hole. Therefore, although the apparent horizon forms for $a\in [0.01,0.5]$ and $b=1.5r_{\rm s}$ (see TABLE \[tab:re\]), these results do not necessarily mean the black hole formation.
Finally, in the present situation, we find the following inequalities, $${\rm Necessary ~condition:}~~ V_2\lesssim\frac{\pi}{2}
16\pi G_5M,
%$$ $${\rm Sufficient ~condition:}~~V_2\lesssim16\pi G_5M. \label{eq:suf}$$ Our numerical results suggest that the necessary condition is not identical to the sufficient condition.
Summary {#sec:6}
=======
We have investigated the condition of apparent horizon formation in the case of momentarily static and conformally flat initial data of a spheroidal mass in the framework of five-dimensional Einstein gravity. All our results are consistent with the hyperhoop conjecture. Particularly, we confirmed the sufficiency of the inequality (\[eq:hyperhoop\]). (More precisely, inequality (\[eq:hutou\]) holds in the present case.)
We also consider the limit of infinitely prolate spheroid. The gravitational field of such spheroids are singular and have a nontrivial structure. The poles at the both ends of this singular spheroid are the space-time singularities since the Kretschmann invariant diverges at the poles of the spheroid. On the other hand, both the Kretschmann invariant and the energy density are finite elsewhere.
We find that the singular spheroid is spacelike infinity except for the poles. Furthermore, when the singular spheroid is sufficiently long, in an appropriate sense, no apparent horizon appears. This property can be regarded as being peculiar to nonuniform distribution of material energy, because a uniform line energy density is always enclosed by an apparent horizon in spite of its length [@IN].
One might wonder if the singular spheroid is a counterexample to the hyperhoop conjecture, since it is infinitely thin and hence there seems to be a hyperhoop satisfying the inequality (\[eq:hyperhoop\]). However as we have shown, the singular spheroid is not a counterexample to the hyperhoop conjecture. In order to understand its reason, we have to note two important features of the initial hypersurface studied here. First, the proper area $V_2$ of a hyperhoop tightly encircling the surface of the spheroid is not necessarily smaller than those encircling outside of the spheroid, since the conformal factor in the inner region takes the value larger than that in the outer region. Second, the proper area $V_2$ of a hyperhoop tightly encircling the spheroid does not necessarily become smaller when the coordinate size of the spheroid characterized by $a$ and $b$ becomes smaller, since the conformal factor in the spheroid of the smaller size becomes larger if the mass $M$ is fixed.
The difference between the present case and the previous work [@IN] is that the line energy density vanishes continuously at the poles in the present case, while it vanishes discontinuously at the poles in the previous case. In general, if an infinitely thin line object forms by the gravitational collapse, it might have a line energy density which continuously vanishes at the end of the matter distribution. Therefore, the naked singularity formation seems to be generic in the axi-symmetric gravitational collapse of highly elongated matter distribution in five-dimensional space-time, although we would need numerical simulation to have a definite evidence for the naked singularity formation [@Shapiro:1990]. This might strongly depend on the spacetime dimension [@Patil:2003yp] and this is also a future work.
Acknowledgements {#acknowledgements .unnumbered}
================
We are grateful to H. Ishihara and colleagues in the astrophysics and gravity group of Osaka City University for helpful discussion and criticism. This work is supported by the Grant-in-Aid for Scientific Research (No.16540264) from JSPS.
Newtonian potentials of a homogeneous ellipsoid in $D$-dimensional space-time {#sct:new}
=============================================================================
In this section, we extend Newtonian potentials of a homogeneous ellipsoid in four-dimensional space-time to $(n+1)$-dimensional space-time. The reader may refer to Ref. [@EFE] about the potentials of four-dimensional space-time.
We want to obtain the potentials of the homogeneous ellipsoid of which bounding ellipsoid is $$\sum^n_{i=1}\frac{x_i^2}{a_i^2}=1.\label{eq:source}$$ At the beginning of derivation, we define the [*homoeoid”*]{}.\
\
[*Definition*]{}. A $n$-dimensional homoeoid is a shell bounded by two similar concentric $n$-dimensional ellipsoids in which strata of equal density are also $n$-dimensional ellipsoids that are concentric with and similar to the bounding ellipsoids.\
\
Following theorem and corollary is derived in the same way as the case of four-dimensional space-time [@EFE].\
\
[*Theorem*]{}. The potential at internal point of a $n$-dimensional homoeoid is constant.\
\
[*Corollary*]{}. The equipotential surfaces external to a thin $n$-dimensional homoeoid are $n$-dimensional ellipsoids confocal to the homoeoid.\
\
We can obtain the potential of a thin $n$-dimensional homoeoid expressed as $$\sum^n_{i=1}\frac{x_i^2}{a_i^2}=1,~~(a_1<a_2<\cdots<a_n),\label{eq:ellip}$$ using $n$-dimensional ellipsoidal coordinates $(y_1,y_2,\cdots,y_n)$ which satisfy $$\sum^n_{i=1}\frac{x_i^2}{a_i^2-y_j}=1,~~(j=1,2,\cdots,n)\label{eq:renritu}$$ and $$y_1<y_2<\cdots<y_n.$$ We can solve Eq. (\[eq:renritu\]) for $x_i^2$ and $$x_i^2=\frac{\prod_{j=1}^{n}(a_i^2-y_j)}{\prod_{k\neq i}(a_i^2-a_k^2)}.
\label{eq:ri2}$$ Therefore $$\frac{\partial x_i}{\partial y_j}=\frac{x_i}{2(y_j-a_i^2)}. \label{eq:bib}$$ The metric is expressed as $$\sum_{i=1}^{n}dx_i^2=\frac{1}{4}\sum_{i=1}^{n}
\frac{\prod_{k\neq i}(y_i-y_k)}{\prod_{j=1}^{n}(a_j^2-y_i)}dy_i^2,
\label{eq:rsenso}$$ thus $$\det g=\left(\frac{1}{4}\right)^n\frac{\prod_{j=1}^{n}
\left[\prod_{k\neq j}(y_j-y_k)\right]}{\prod_{i,j}(a_i^2-y_j)}.
\label{eq:detg}$$
In order to obtain the potential $\Phi_N$ of a thin $n$-dimensional homoeoid, we only have to solve following equation because of the theorem and the corollary. $$\triangle\Phi_N(y_1)=0\label{eq:tri}$$ with $$\Phi_N \rightarrow
-\frac{M}{(n-2)r^{n-2}}~~~{\rm for}~~~r\rightarrow\infty, \label{eq:bcth}$$ where $M$ is constant which correspond to the mass of the thin $n$-dimensional homoeoid and $r$ is the length from the center of the homoeoid. It can be seen that $-y_1\sim r^2$ at infinity. Using ellipsoidal coordinates to Eq. (\[eq:tri\]), we obtain $$\frac{\partial}{\partial y_1}\left[\prod_{i=1}^n\sqrt{a_i^2-y_1}
\frac{\partial\Phi_N(y_1)}{\partial y_1}\right]=0. \label{eq:lap0p}$$ The solution of Eq. (\[eq:lap0p\]) with (\[eq:bcth\]) is $$\Phi_N(u)=-\frac{n-2}{2}M\int^{\infty}_{u}\frac{1}{\prod_{i=1}^n
\sqrt{a_i^2+u}}du, \label{eq:homopo}$$ where $u=-y_1$.
Integrating this potentials of thin homoeoid which is foliated in all region of ellipsoid in the same manner as four-dimensional space-time [@EFE], we can finally obtain the Newtonian potential of a $D$-dimensional homogeneous ellipsoid. The integration can be done as follows. We can express the thin homoeoids which is similar and concentric to the ellipsoid (\[eq:source\]) as $$\sum^n_{i=1}\frac{x_i^2}{a_i^2}=m^2,$$ where $m$ is constant and we assume $0\leq m \leq1$. Let us consider the homogeneous homoeoid bounded by two ellipsoids $\sum^n_{i=1}\frac{x_i^2}{a_i^2}=m^2$ and $\sum^n_{i=1}\frac{x_i^2}{a_i^2}=(m+dm)^2$, where $dm$ is small deviation. The mass of this homoeoid is $$\omega_na_1a_2\cdots a_n\rho m^{n-1}dm,\label{eq:homomass}$$ where $\omega_n$ and $\rho$ is $n$-dimensional solid angle and the density of the homoeoid respectively.
First, we derive the potential in outer region of homogeneous ellipsoid. Substituting (\[eq:homomass\]) into $M$ of (\[eq:homopo\]), we can obtain the potential of the homoeoidal element (\[eq:homomass\]) at $(x'_1,x'_2,\cdots,x'_n)$ $$-\frac{n-2}{2}\omega_na_1a_2\cdots a_n\rho m^{n-1}dm
\int^\infty_{\lambda(m^2)}\frac{du'}{\prod^n_{i=1}\sqrt{a_i^2m^2+u'}},
\label{eq:homopote}$$ where $\lambda$ is the largest root of $\sum^n_{i=1}\frac{{x'_i}^2}{m^2a_i^2
+\lambda}=1$. Integrating this equation about $0\leq m\leq 1$, we have $$\begin{aligned}
-\frac{n-2}{2}\omega_na_1a_2\cdots a_n\rho \int^1_0dmm^{n-1}
\int^\infty_{\lambda(m^2)}\frac{du'}{\prod^n_{i=1}\sqrt{a_i^2+u'}}
&&\nonumber \\
=-\frac{n-2}{4}\omega_na_1a_2\cdots a_n\rho \int^1_0dm^2
\int^\infty_{\mu(m^2)}\frac{du}{\prod^n_{i=1}\sqrt{a_i^2+u}},&&\nonumber\\\end{aligned}$$ where $u$ and $\mu$ defined by $u'=m^2u$ and $\lambda(m^2)=m^2\mu(m^2)$ respectively. Now, we can invert the order of integrations because $\mu(m^2)$ is the monotone decreasing function of $m^2$. Since $\mu \rightarrow \infty$ when $m \rightarrow 0$ and $\mu=\lambda$ when $m=1$, we can write $$\Phi_N=-\frac{n-2}{4}\omega_na_1\cdots a_n\rho
\int^\infty_{\lambda(1)}\frac{(1-m^2(u))}{\prod^n_{i=1}\sqrt{a_i^2+u}}du
\label{eq:sotofpote}$$ in outer region, where $$m^2(u)=\sum^n_{i=1}\frac{x_i^2}{a_i^2+u}.$$
Next, we derive the potential in inner region of homogeneous ellipsoid at the point $(x'_1,x'_2,\cdots,x'_n)$ which satisfy $\sum^n_{i=1}\frac{x'^2_i}{a_i^2}=m'^2$, where $m'$ is constant and $0\leq m'<1$. On the one hand, the potential of homogeneous ellipsoid bounded by $\sum^n_{i=1}\frac{x^2_i}{a_i^2m'^2}=1$ at $(x'_1,x'_2,\cdots,x'_n)$ is $$-\frac{n-2}{4}\omega_na_1\cdots a_n \rho
\int^\infty_0\frac{(m'^2-m^2(u))}{\prod^n_{i=1}\sqrt{a_i^2+u}}du,
\label{eq:elipote}$$ where we use (\[eq:sotofpote\]). On the other hand, the potential of homogeneous homoeoid bounded by $\sum^n_{i=1}\frac{x^2_i}{a_i^2}=m'^2$ and$\sum^n_{i=1}\frac{x^2_i}{a_i^2}=1$ at $(x'_1,x'_2,\cdots,x'_n)$ is obtained by integration of (\[eq:homopote\]) from $m=m'$ to $m=1$, and we have $$-\frac{n-2}{4}\omega_na_1a_2\cdots a_n \rho (1-m'^2)
\int^\infty_0\frac{du}{\prod^n_{i=1}\sqrt{a_i^2+u}},\label{eq:homo}$$ Adding together (\[eq:homo\]) and (\[eq:elipote\]), we can obtain the required potential $$\begin{aligned}
\Phi_N=-\frac{n-2}{4}\omega_na_1\cdots a_n\rho
\int^\infty_0\frac{(1-m^2(u))}{\prod^n_{i=1}\sqrt{a_i^2+u}}du~~~~\end{aligned}$$ in inner region.
If we have same radiuses to the directions corresponding to coordinates $x_i,x_{i+1},\cdots$ in (\[eq:source\]), we can introduce multipolar coordinates to these and calculate as same.
the necessary condition of black hole formation in five-dimensional space-time {#sct:nec}
==============================================================================
We consider the singular line source studied in Ref. [@IN]. The energy density is given by $$f^3\rho=\frac{1}{4\pi R^2}\frac{G_5M}{2b}\delta(R)\theta(b-|z|)
\label{eq:spindle}$$ where $\theta$ is the Heaviside’s step function and the “length” of this line source is given by $2b$. In this case, the solution of the Hamiltonian constraint (\[eq:rap\]) is given by $$f=1+\frac{G_5M}{2bR}\left(\arctan\frac{z+b}{R}-\arctan\frac{z-b}{R}\right).$$ As shown in Ref. [@IN], this line source is always covered by an apparent horizon.
In order to obtain the necessary condition of an apparent horizon formation, we have to calculate the values of $\Gamma_{\rm min}$ defined in Sec. \[sec:5\]. Here, we consider the hyperhoops which are expressed in the form $r=r_h(\xi)$ and $\vartheta=\pi/2$ with the same symmetry as discussed in Sec. \[sec:4\] (axi-symmetry and reflection symmetry with respect to $z=0$).
In this case, the hyperhoop which intersects the line source has infinite two-dimensional area $V_2$. In order to see this, consider a hyperhoop expressed as $\vartheta=\pi/2$ and $z=\eta(R)$, where $|\eta(0)|<b$ so that the hyperhoop intersects the line source. We focus on a segment $R_1<R<R_2$ and $z>0$ of this hyperhoop. We assume that $R_1$ and $R_2$ are sufficiently small and $0<\eta(R)<b$ on this segment. The conformal factor is approximately given by $$f\simeq \frac{\pi \left(2\eta-b\right)G_5M}{4bR\eta}$$ on this segment of the hyperhoop. Hence, the area of this segment is bounded below as $$\begin{aligned}
&&\int^{R_2}_{R_1}\int^{2\pi}_{0}f(R,\eta)^2R
\sqrt{1+\left(\frac{d\eta}{dR}\right)^2}d\varphi dR
\nonumber \\
&&\geq2\pi\int^{R_2}_{R_1}f(R,\eta)^2RdR\simeq\frac{\pi^3 G_5^2M^2}{8b^2}\int^{R_2}_{R_1}
\left(2-\frac{b}{\eta}\right)^2\frac{dR}{R}\nonumber \\
&&\geq\frac{\pi^3 G_5^2M^2}{8b^2}\int^{R_2}_{R_1}\frac{dR}{R}
=\frac{\pi^3 G_5^2M^2}{8b^2}\ln \frac{R_2}{R_1}.\end{aligned}$$ We can see from the above equation that the area of the hyperhoop diverges in the limit of $R_1\rightarrow0$ with $R_2$ fixed. Therefore we have to consider only the hyperhoop which entirely encircle the line source, and hence we find $$\Gamma_{\rm min}=\left(\frac{V_2}{16\pi G_5M_{\rm in}}\right)_{\rm min}
=\frac{\left(V_2\right)_{\rm min}}{16\pi G_5M},$$ where $\left(V_2\right)_{\rm min}$ is the area of the hyperhoop which entirely encircle the source and has the smallest area. In order to evaluate $\left(V_2\right)_{\rm min}$, we focus on the hyperhoop $r=r_{\rm a}(\xi)$ and $\vartheta=\pi /2$ which satisfy following minimum area condition $$\delta V_2=0, \label{eq:vari}$$ where $\delta V_2$ is the small variation of $V_2$ for slight deformation of the hyperhoop which keeps it on $\vartheta=\pi/2$ and the symmetry holds. Namely, Eq. (\[eq:vari\]) leads to the Euler-Lagrange equation for the Lagrangian $L=V_2\left(r_{\rm a},\dot{r_{\rm a}}\right)$ as $$\begin{aligned}
&&{\ddot r}_{\rm a}-{3 {\dot r}_{\rm a}{}^2\over r_{\rm a}}
-2r_{\rm a}
+\frac{r_{\rm a}{}^2+{\dot r}_{\rm a}{}^2}{r_{\rm a}}
\nonumber \\
&&\times\left[
\frac{{\dot r}_{\rm a}}{r_{\rm a}}\cot\xi
-{2\over f}\left({\dot r}_{\rm a}\sin\xi+r_{\rm a}\cos\xi\right)
\frac{\partial f}{\partial z}
+{2\over f}\left({\dot r}_{\rm a}\cos\xi-r_{\rm a}\sin\xi\right)
\frac{\partial f}{\partial R}\right]=0.\label{eq:deforma}\end{aligned}$$ We impose following boundary conditions so that every part of the hyperhoop locally satisfy Eq. (\[eq:deforma\]) $${\dot r}_{\rm a}=0~~~~~{\rm at}~~\xi=0,~\frac{\pi}{2}.\label{eq:bcforma}$$ Then $r=r_{\rm a}(\xi)$ is the hyperhoop of the minimum area if and only if $r_{\rm a}(\xi)$ satisfies Eq. (\[eq:deforma\]) with the boundary condition (\[eq:bcforma\]). We adopt the area of this hyperhoop as $(V_2)_{\rm min}$.
We numerically search for solutions of Eq. (\[eq:deforma\]) with (\[eq:bcforma\]). Accordingly, the solutions always can be found and the hyperhoop always encircle the source in spite of its length. The value of $\Gamma_{\rm min}$ is depicted in Fig. \[fig:deb\] as a function of $b$.
![ The value of $\Gamma_{\rm min}$ is plotted as a function of $b$. The dashed line bound this value above. This line is the corresponding quantity for the infinitely long spindle source which have line density $M/2b$. []{data-label="fig:deb"}](bdep.eps)
The value of $\Gamma_{\rm min}$ monotonically increases with $b$ but has a finite limit for $b\rightarrow \infty$, while an apparent horizon always covers this line source. Therefore it is necessary for apparent horizon formation that $\Gamma_{\rm min}$ is smaller than this asymptotic value. The asymptotic value will be obtained by evaluating the corresponding quantity of the infinitely long source case. Let us consider the infinitely long singular line source whose density profile is given by $$f^3\rho=\frac{1}{4\pi R^2}\frac{G_5M}{2b}\delta(R).$$ In this case, we can easily solve the Hamiltonian constraint (\[eq:rap\]) and obtain $$f=1+\frac{\pi G_5M}{2bR}.$$ The area $V_{\rm c}$ of the cylindrical two-surface $R=R_0$ and $\vartheta=\pi/2$ with coordinate length $2b$ is $$V_{\rm c}=2\pi Rf|_{R=R_0} \times 2bf|_{R=R_0},$$ where $2\pi Rf|_{R=R_0}$ is the proper length of circle around singular line source and $2bf|_{R=R_0}$ is the proper length of the cylinder measured along the $z$-direction. The minimal value of $V_{\rm c}$ is realized when $R_0=\pi G_5M/2b$, and in its value is equal to $8\pi^2 G_5M$. This minimal value might be almost equal to $\left(V_2\right)_{\rm min}$ of the singular line source (\[eq:spindle\]) if $b$ is much longer than $r_{\rm s}$. As a result, the asymptotic value of $\Gamma_{\rm min}$ for $b\rightarrow\infty$ with the mass $M$ fixed is evaluated as $\pi/2$.
The derivation of the equation for a marginal surface {#sec:margi}
=====================================================
In this section, we show the derivation of Eq.(\[eq:joubi\]). Here, we generalize Ref.[@ref:SMMN] to the $D$-dimensional case.
We denote the spacelike unit vector outward from and normal to the marginal surface by $s_a$, and the spacelike unit vectors spanning the marginal surface are denoted by $(e^A)_a$, where $A=1,..,D-2$. All these vectors are chosen to be orthogonal to each other and to the unit vector normal to the initial hypersurface $n_a$. Then the future directed outward null vector $l^a$ orthogonal to the marginal surface is written by $$l^a=n^a+s^a.$$ The expansion $\chi$ of this null vector is defined by $$\chi=\delta_{AB}(e^A)^a(e^B)^b\nabla_bl_a=(h^{ab}-s^as^b)(K_{ab}-D_bs_a)
,\label{eq:chi}$$ where $h_{ab}$ and $K_{ab}$ are the induced metric and the extrinsic curvature defined in Sec.\[sec:2\] respectively. The marginal surface is a closed $(D-2)$-dimensional spacelike submanifold such that the outward null vector orthogonal to the $(D-2)$-dimensional spacelike submanifold has vanishing expansion. Hence, the equation to define the marginal surface is given by $\chi=0$. In the momentarily static case, this equation reduces to $$\delta_{AB}(e^A)^aD_a(e^B)^bs_b=0.\label{eq:momestaj}$$
In the situation presented in this paper, coordinates of points on the marginal surface are represented as $$x^\mu=
\left(
r_{\rm m}(\xi)\cos\xi,r_{\rm m}(\xi)\sin\xi,\vartheta,\varphi
\right)$$ and following vectors are tangent to the marginal surface as $$\begin{aligned}
\frac{\partial x^\mu}{\partial\xi}
&=&\left(\dot{r}_{\rm m}\cos\xi-r\sin\xi,
\dot{r}_{\rm m}\sin\xi+r_{\rm m}\cos\xi,0,0\right),~~~~~\\
\frac{\partial x^\mu}{\partial\vartheta}&=&(0,0,1,0),\\
\frac{\partial x^\mu}{\partial\varphi}&=&(0,0,0,1).\end{aligned}$$ Hence, we can obtain the components of $s^a$ and $(e^A)^a$ as $$\begin{aligned}
s^\mu&=&\frac{1}{f\sqrt{r_{\rm m}^2+\dot{r}_{\rm m}^2}}
\left(\dot{r}_{\rm m}\sin\xi+r_{\rm m}\cos\xi,-\dot{r}_{\rm m}\cos\xi+r_{\rm m}\sin\xi,0,0\right),\label{eq:vectors4} \\
(e^1)^\mu&
=&\frac{1}{f\sqrt{r_{\rm m}^2+\dot{r}_{\rm m}^2}}
\left(\dot{r}_{\rm m}\cos\xi-r_{\rm m}\sin\xi,\dot{r}_{\rm m}\sin\xi+r_{\rm m}\cos\xi,0,0\right),\label{eq:vectors1}\\
(e^2)^\mu&=&\frac{1}{fR}(0,0,1,0), \label{eq:vectors2}\\
(e^3)^\mu&=&\frac{1}{fR\sin\vartheta}(0,0,0,1).\label{eq:vectors3}\end{aligned}$$ Substituting Eqs.(\[eq:vectors4\])$-$(\[eq:vectors3\]) into Eq.(\[eq:momestaj\]), we obtain Eq.(\[eq:joubi\]).
[99]{} N. Arkani-Hamed, S. Dimopoulos and G. R. Dvali, Phys. Lett. B [**429**]{}, 263 (1998) \[arXiv:hep-ph/9803315\] ; I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos and G. R. Dvali, Phys. Lett. B [**436**]{}, 257 (1998) \[arXiv:hep-ph/9804398\] ; L. Randall and R. Sundrum, Phys. Rev. Lett. [**83**]{}, 3370 (1999) \[arXiv:hep-ph/9905221\]. J. C. Long, H. W. Chan and J. C. Price Nucl. Phys. B [**539**]{}, 23 (1999) \[arXiv:hep-ph/9805217\]; C. D. Hoyle, D. J. Kapner, B. R. Heckel, E. G. Adelberger, J. H. Gundlach, U. Schmidt and H. E. Swanson, Phys. Rev. D [**70**]{}, 042004 (2004) \[arXiv:hep-ph/0405262\]. P. C. Argyres, S. Dimopoulos and J. March-Russell, Phys. Lett. B [**441**]{}, 96 (1998) \[arXiv:hep-th/9808138\] ; T. Banks and W. Fischler, arXiv:hep-th/9906038 ; S. Dimopoulos and G. Landsberg, Phys. Rev. Lett. [**87**]{}, 161602 (2001) \[arXiv:hep-ph/0106295\] ; S. B. Giddings and S. Thomas, Phys. Rev. D [**65**]{}, 056010 (2002) \[arXiv:hep-ph/0106219\] ; E. J. Ahn, M. Cavaglia and A. V. Olinto, Phys. Lett. B [**551**]{}, 1 (2003) \[arXiv:hep-th/0201042\]. J. L. Feng and A. D. Shapere, Phys. Rev. Lett. [**88**]{}, 021303 (2002) \[arXiv:hep-ph/0109106\] ; R. Emparan, M. Masip and R. Rattazzi, Phys. Rev. D [**65**]{}, 064023 (2002) \[arXiv:hep-ph/0109287\] ; L. Anchordoqui and H. Goldberg, Phys. Rev. D [**65**]{}, 047502 (2002) \[arXiv:hep-ph/0109242\]. K.S. Thorne, in [*Magic Without Magic*]{}, edited by J. Klauder (Freeman, San Francisco, 1972). K. Nakao, K. Nakamura and T. Mishima, Phys. Lett. B [**564**]{}, 143 (2003) \[arXiv:gr-qc/0112067\]. D. Ida and K. Nakao, Phys. Rev. D [**66**]{}, 064026 (2002) \[arXiv:gr-qc/0204082\]. H. Yoshino and Y. Nambu, Phys. Rev. D [**67**]{}, 024009 (2003) \[arXiv:gr-qc/0209003\]; H. Yoshino and Y. Nambu, Phys. Rev. D [**70**]{}, 084036 (2004) \[arXiv:gr-qc/0404109\]. C. Barrabes, V. P. Frolov and E. Lesigne, Phys. Rev. D [**69**]{}, 101501(R) (2004) \[arXiv:gr-qc/0402081\]. T. Nakamura, S. L. Shapiro and S. A. Teukolsky, Phys. Rev. D [**38**]{}, 2972 (1988). R.N. Wald, [*General Relativity*]{}, (University of Chicago Press, Chicago, 1984). M. Sasaki, K. Maeda, S. Miyama and T. Nakamura, Prog. Theor. Phys. [**63**]{}, 1051 (1980). S. L. Shapiro and S. A. Teukolsky, Phys. Rev. Lett. [**66**]{}, 994 (1991). K. D. Patil, Phys. Rev. D [**67**]{}, 024017 (2003) ; R. Goswami and P. S. Joshi, Phys. Rev. D [**69**]{}, 104002 (2004) \[arXiv:gr-qc/0405049\]; A. Mahajan, R. Goswami and P. S. Joshi, arXiv:gr-qc/0502080. S. Chandrasekhar, [*Ellipsoidal Figures of Equilibrium”*]{},(Yale University Press, New Haven, London,1969)
[^1]: E-mail:c\_m\[email protected]
[^2]: E-mail:[email protected]
[^3]: E-mail:[email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The generation of ultrashort pulses hinges on the careful management of dispersion. Traditionally, this has exclusively involved second-order dispersion, while higher-order dispersion was treated as a nuisance to be minimized. Here we show that high-order dispersion can be strategically leveraged to access an uncharted regime of ultrafast laser operation. In particular, we demonstrate a mode-locked laser, with an intra-cavity spectral pulse-shaper, that emits pure-quartic soliton pulses, which arise from the interaction of the fourth-order dispersion and the Kerr nonlinearity. Using phase-resolved measurements we demonstrate that the energy of these pulses is proportional to the third power of the inverse pulse duration. This implies a dramatic increase in the energy of ultrashort pulses compared to those emitted by soliton lasers to date. These results not only demonstrate a novel approach to ultrafast lasers, but more fundamentally, they clarify the use of higher-order dispersion for optical pulse control, opening up a plethora of possibilities in nonlinear optics and its applications.'
author:
- 'Antoine F. J. Runge$^{1,*}$'
- 'Darren D. Hudson$^{2}$'
- 'Kevin K. K. Tam$^{1}$'
- 'C. Martijn de Sterke$^{1,3}$'
- 'Andrea Blanco-Redondo$^{4}$'
bibliography:
- 'PQS\_bib.bib'
title: 'The pure-quartic soliton laser'
---
Introduction {#introduction .unnumbered}
============
Ultrafast lasers have been fundamental for the development of major photonic applications such as telecommunications [@Mollenauer_1991; @Mollenauer2_1991; @Haus_1996], supercontinuum [@Husakou_2001; @Dudley_2006], nonlinear imaging [@Xu_2013] and frequency comb generation [@Cundiff_2003]. Soliton effects, based on the balance of quadratic dispersion and nonlinearity, have allowed for the direct generation of optical pulses with duration below 10 fs [@Zhou_1994; @Jung_1997]. Soliton fibre lasers, which have relatively simple configurations can be built using off-the-shelf components [@Mollenauer_1984; @Kafka_1989; @Matsas_1992]. However, the energy of the pulses emitted by these lasers is limited by the soliton area theorem [@Zakharov_1972; @Hasegawa_1973], and by the appearance of resonant spectral sidebands, arising from periodic perturbations in the optical cavity [@Kelly_1992].
Pure-Quartic Solitons (PQSs), shape-maintaining pulses that arise from the balance of the Kerr nonlinearity and negative fourth-order dispersion (FOD), were recently discovered in dispersion-engineered photonic crystal waveguides [@Blanco_Redondo_2016]. They were observed in a spectral range where the second-order (quadratic) dispersion $\beta_2$ was positive, the third-order (cubic) dispersion $\beta_3$ was negligible and the fourth-order (quartic) dispersion $\beta_4$ was negative. Recent theoretical studies have unveiled that PQSs possess an advantageous energy scaling [@Lo_2018; @Tam_2019], which grants them the potential to achieve significantly higher energy than their conventional soliton counterpart for short pulse durations. This discovery has led to efforts to transition from planar silicon photonic crystals to other platforms [@Lo_2018; @TaheriMatsko_2019], including optical fibres where mature fibre laser technology can be utilized. Initial approaches involved carefully designed photonic crystal fibre (PCF) with multi-layer air holes that exhibit negligible $\beta_2$ and large, negative $\beta_4$ [@Lo_2018]. However, the fabrication tolerances required for achieving the necessary control of dispersion are very tight. Consequently, PQSs have until now not been observed experimentally in optical fibres.
Here, we report the first experimental demonstration of a mode-locked fibre laser emitting pure-quartic soliton pulses. We achieve the necessary laser cavity dispersion conditions by utilizing an intra-cavity spectral pulse shaper that simultaneously cancels the $\beta_2$ and $\beta_3$ of the otherwise conventional-fibre cavity, while imparting a strong, negative $\beta_4$. PQSs arise from this lumped quartic dispersion in much the same way that conventional solitons can arise in laser cavities with discrete segments of normal and anomalous dispersion fibre [@Knox_1994; @Turitsyn_2012].
Using spectral, temporal and phase-resolved measurements we demonstrate that the output pulses behave significantly differently than any other type of laser pulses: we find that the energy of the emitted pulses $E_{\rm PQS}$ is proportional to $\tau^{-3}$, consistent with theoretical predictions [@Tam_2019], and in contrast to conventional solitons for which $E\propto \tau^{-1}$ [@Agrawal_NFO]. We also find that the output spectra exhibit sidebands, which we associate with resonant dispersive waves. Though these sidebands also occur in conventional soliton lasers [@Kelly_1992], the measured sideband frequencies are characteristic for cavities with quartic dispersion, and obey a novel equation that we present here. These results constitute strong evidence for a novel operating regime for mode-locked lasers, opening the way for the generation of high-energy, ultrashort optical pulses, arising from SPM and higher-order cavity dispersion.
Results {#results .unnumbered}
=======
Our laser configuration is shown schematically in Fig. \[laser\](a). It is an erbium-doped fibre laser using nonlinear polarization evolution (NPE) for mode-locking, and incorporating an intracavity programmable optical pulse-shaper [@Schroder_2010; @Peng_2016]. The total cavity length is approximately $L=21.4~{\rm m}$, leading to a fundamental repetition rate of 9.3 MHz. Mode-locking is achieved by adjusting the pump power and the polarization controllers (PCs). An output coupler (OC) is located after the fibre polarizer and extracts 50% of the intracavity power. The pulse-shaper, based on a spatial light modulator (SLM), can produce an arbitrary phase mask and is used to adjust the net cavity dispersion as illustrated in Figs. \[laser\](b) and (c).
-0mm
-1mm
-3mm
-1mm
When no phase mask is applied, the laser operates in the conventional soliton regime, with the intrinsic anomalous GVD of the fibre balancing the SPM. The output pulses display the well-known hyperbolic secant spectrum centred around 1563 nm with a 3.72 nm bandwidth at $-3~{\rm dB}$, and strong spectral (“Kelly”) sidebands [@Kelly_1992] as seen in Fig. \[spectrum\](a) (solid blue curve). The truncated long wavelength edge is due to the finite spectral bandwidth of the pulse shaper. To gain insight in the temporal properties of the output pulses we also perform a set of temporal and phase-resolved measurements using a frequency-resolved electrical gating (FREG) setup [@Blanco_Redondo_2016; @Dorrer_2002]. The retrieved pulse of the laser operating in conventional soliton regime is shown in Fig. \[spectrum\](b) (blue solid line).
We then program the pulse-shaper to induce a phase profile that can be written as $$\phi = \exp\left(iL\sum{\beta_n\omega^n/n!}\right).
\label{phase_eq}$$ where $\beta_n$ is the $n^{th}$ dispersion order for $n=2,3,4$. For the results discussed below $\beta_2 = +21.4~{\rm ps^2/km}$, $\beta_3 = -0.12~{\rm ps^3/km}$, and we take initially $\beta_4 = -80~{\rm ps^4/km}$. The pulse shaper serves two aims; (i) to compensate the $2^{\rm nd}$ and $3^{\rm rd}$ order dispersion introduced by the optical fibres [@Hammani_2011], so that net cavity dispersion of these orders is negligible (Fig. \[laser\](b)); and (ii) the pulse shaper generates a large negative FOD, which dominates the propagation (Fig. \[laser\](c)). The applied FOD is several orders of magnitude larger than the intrinsic FOD of single mode fibres ($\approx10^{-3}$ ps$^4$/km) [@Ito_2016]. The output spectrum of the laser operating in this regime, shown in Fig. \[spectrum\](e), has a 3.16 nm bandwidth with a shape that differs significantly from that of conventional solitons, but is in very good agreement with the theoretically predicted spectral profile of the PQS (red dashed line) [@Tam_2019]. We note in particular the PQS’s distinct flatness of the spectral maximum. We also observe several strong, narrowly spaced spectral sidebands. The retrieved pulse corresponding to this spectrum is shown in Fig. \[spectrum\](f) (blue solid line). The recovered pulse has a duration of $\tau = 1.74$ ps (at full half width maximum). The retrieved temporal pulse shape is again in good agreement with the theoretically predicted PQS shape for the same pulse duration (red dashed line) [@Tam_2019]. From this measurement and the spectral bandwidth at $-3~{\rm dB}$ measured from the corresponding spectrum, we calculate a time-bandwidth product of $0.67$. This value is larger than the predicted value of $0.53$ for transform-limited PQSs [@Tam_2019], indicating that the pulses at the laser output have a nearly flat phase with a modest higher-order phase distortion, as suggested by the recovered phase (green solid line in Fig. \[spectrum\](f)). Finally, we note that the predicted temporal shape of the PQS exhibits periodic oscillations in the tails that are not observed in the retrieved experimental profile [@Tam_2019]. This is because the first lobe is expected to appear approximately $28$ dB below the pulse’s central maximum, which is below the continuous background arising from the dispersive spectral sidebands in our experiments.
We model the laser dynamics using an iterative cavity map [@Oktem_2010; @Woodward_2018], in which the propagation through every element is modeled using a generalized nonlinear Schrödinger equation with Kerr nonlinearity, and with dispersion up to the fourth order (see Methods). The simulated spectra and intensity profiles corresponding to the two operating regimes are shown in Fig. \[spectrum\](c), (d), (g) and (h), are in excellent agreement with the experimental results, and are consistent with the change in the pulse-shaping mechanism.
We can assess the quartic nature of the laser by analyzing the spectral positions of the dispersive waves in Fig. \[spectrum\](e). Similar to conventional solitons, these dispersive waves arise from perturbations when the soliton propagates through the cavity; the spectral positions of these peaks provide information about the cavity dispersion [@Dennis_1994]. The dispersive waves are generated every round trip and interfere constructively when $\beta_{PQS}-\beta_{\rm lin} =2m\pi/L$, where $\beta_{PQS,{\rm lin}}$ are the propagation constants of the soliton and the dispersive wave, respectively, and $m$ is an integer. When this condition is satisfied, then dispersive waves generated in consecutive passages through the cavity, interfere constructively, leading to narrow spectral peaks [@Kelly_1992].
For a linear wave propagating in a quartic dispersion cavity, $\beta_{\rm lin} = -|\beta_4|(\omega-\omega_0)^4/24$, whereas the PQS experiences a constant dispersion across its entire bandwidth of $\beta_{PQS} = K|\beta_4|/\tau^4$, where the constant $K=1.67$ [@Tam_2019]. Using the argument outlined in the previous paragraph, it is straightforward to show that the $m^{th}$-order spectral resonances $\omega_m$ satisfy $$\omega_m = \pm{1\over\tau} \left({48m\pi\tau^4\over |\beta_4|L}-24K\right)^{1/4}.
\label{sideband_eq}$$ This equation shows that the $4^{th}$ power of the sideband frequency offsets are equally spaced by $48\pi/(|\beta_4|L)$. In Fig. \[sidebands\](a) we show the optical spectrum of the PQS centred around $\omega_0$ and we mark the measured spectral positions of the sidebands for the low (circles) and high (diamonds) frequencies of the spectrum. Figure \[sidebands\](b) shows the $4^{th}$ power of these sidebands positions ($\Delta\omega^4$) versus the sideband order. This confirms that they follow a linear relation, consistent with Eq. \[sideband\_eq\]. The spacing determined from the measured spectral position of the sidebands is fixed at $0.90~{\rm ps^{-4}}$ and $0.867~{\rm ps^{-4}}$ for the low (circles) and high (diamonds) frequencies, respectively. These results are in excellent agreement with the theoretical value of $0.88~{\rm ps^{-4}}$ following from Eq. \[sideband\_eq\]. The small discrepancies between the calculated and theoretical values could be due to the limited resolution of the measurement of the position of the sidebands, or to residual, uncompensated quadratic and cubic dispersion.
While the results discussed thus far were taken at a fixed $\beta_4=-80~{\rm ps^4/km}$, in Figs \[sidebands\](c) and (d) we show, for completeness, similar measurements for seven different values of the quartic dispersion, namely $\beta_4=-20, -40, -60, -80,$ $
-90, -100, -110~{\rm ps^4/km}$. Fig. \[sidebands\](c) shows measured spectra for the shortest pulses that we generated at each of these $\beta_4$ values (see Fig. \[E\_vs\_tau\](b)). The similarity of these spectra implies that the pulse shaping mechanism is unchanged for all the $\beta_4$ values considered. Similarly, in Fig. \[sidebands\](d) we show an analysis of the positions of the side bands. Since for all cases the $\Delta\omega^4$ values lie on straight lines, we confirm that the dispersion remains quartic for all $\beta_4$ values. The presence of the last term in the brackets in Eq. \[sideband\_eq\] implies that the straight lines do not go through the origin, or even go through a common point.
We now investigate in more detail the results of FREG measurements as we vary $\beta_4$. In Fig. \[E\_vs\_tau\](a) we show spectrograms of the shortest pulses measured, for each $\beta_4$. The corresponding retrieved temporal intensity profiles are shown in Fig. \[E\_vs\_tau\](b). Similar to conventional soliton lasers, reducing the net-cavity dispersion allows for the generation of shorter optical pulses [@Tamura_1993; @Chen_1999; @Turitsyn_2012]. The vertical streaks at short and long wavelengths in Fig. \[E\_vs\_tau\](a) are the first sidebands on either side of the pulse spectrum, discussed earlier. As required for negative quartic dispersion, on the short wavelength side they precede the pulse, whereas on the long wavelength side they follow it.
Finally, we report the energy-width scaling of the emitted PQSs. While for conventional solitons the pulse energy is inversely proportional to the pulse duration, recent theoretical and numerical studies show that for a PQS the pulse energy is given by [@Tam_2019] $$E_{PQS} = \frac{2.87|\beta_4|}{\gamma\tau^3}.
\label{energy}$$ where $\gamma$ is the average cavity nonlinear parameter. At each value of the quartic dispersion we measure the spectral bandwidth of the output pulses for different pulse energies by adjusting the pump power, and we deduct the portion of the energy in the spectral sidebands by integrating the measured optical spectrum. The corresponding pulse durations are then calculated using the time-bandwidth product of $0.67$ determined above, which we find to be constant for the entire range of parameters that we consider. The results of the procedure described in the previous paragraph are summarized in Figs \[E\_vs\_tau\](c) and (d). The circles in Figure \[E\_vs\_tau\](c) show the measured pulse energies versus the pulse duration $\tau$, for different values of quartic dispersion. The results in Fig. \[E\_vs\_tau\](c) are in excellent agreement with Eq. \[energy\] once we account for the output coupling and the variations of the pulse parameters within the cavity (see Methods). To see this more clearly, we plot $E\propto\beta_4\tau^{-3}$ for each of the $\beta_4$ values. This shows that for fixed $\beta_4$, the energy $E\propto \tau^{-3}$ and that for fixed $\tau$, the energy $E\propto\beta_4$, consistent with Eq. \[energy\].
Figure \[E\_vs\_tau\](d) shows the same data as Fig. \[E\_vs\_tau\](c), but has $E^{-1/3}$ on the vertical axis. Plotted in this way, the data should form a fan of straight lines, with each rib, corresponding to a particular value of $\beta_4$, going through the origin. The consistency between the measured data and the prediction from Eq. \[energy\] is conclusive evidence of the unique scaling properties of the pulses emitted by our laser.
A second set of measurements (not shown here) is performed with a different output coupler, extracting 10% of the intracavity power, and we obtain similar results. We also perform a similar measurement (see Supplementary information) for the laser operating in the conventional soliton regime (as seen in Fig. \[spectrum\](a)) and find the output pulse energy follows the well-known relation $E\propto 1/\tau$ [@Agrawal_NFO]. These results confirm that PQS pulses follow a different energy-width scaling relation and that they could outperform conventional soliton for short pulse durations, as suggested by previous studies [@Tam_2019; @Blanco_Redondo_2017].
-3mm
-1mm
Discussion {#discussion .unnumbered}
==========
By harnessing the interaction between fourth-order dispersion with the nonlinearity in a fibre laser cavity we provide the first experimental realization of a pure-quartic soliton laser. We have provided experimental evidence of a novel pulse spectrum and temporal shape, energy-width scaling, and spectral sideband behaviour, with respect to any type of laser pulses produced to date.
The necessary FOD-dominating dispersion profile was achieved by incorporating a reconfigurable spectral pulse shaper inside the laser cavity. This innovation is crucial to leverage the effects of high-order dispersion in ultrafast lasers, without the need for challenging dispersion engineering of the waveguide or fibre.
The range of pulse energies achievable with our current setup is limited by the spectral and delay range of the spectral pulse-shaper. This limitation is in no way fundamental and it could be overcome by using a more broadband pulse-shaper with a larger dynamic range. In the future, fibre platforms with intrinsically high negative FOD [@Lo_2018] could substitute the pulse shaper. Alternatively, microresonator geometries with dominant negative FOD have been recently theoretically studied [@TaheriMatsko_2019], which highlights a potential route towards laser systems leveraging high-order dispersion in integrated platforms.
Our results could pave the way for a new class of simple and low-cost mode-locked fibre laser emitting ultrashort high-energy pulses. The simple laser cavity configuration presented in this work offers an exceedingly flexible testbed for the generation and the study of optical pulses arising from the interplay of Kerr nonlinearity and hybrid dispersion – combination of different orders of dispersion, including orders higher than fourth. This provides new degrees of freedom to optically control the shape and the energy scaling of optical pulses which could not only have impact on only ultrafast lasers, but also on other areas in which these features are crucial such as frequency combs, supercontinuum generation, and advanced modulation formats for communications.
Methods {#methods .unnumbered}
=======
**Numerical simulation model.** Numerical simulations are based on the nonlinear Schrödinger equation: $$\frac{\partial A}{\partial z}+i\hat{D}\left(i\frac{\partial}{\partial T}\right)A=\frac{g}{2}A+i\gamma A|A|^2.
\label{GNLSE}$$ Here, $A = A(z,T)$ is the slowly varying amplitude of the pulse envelope, $z$ is the propagation coordinate, $T$ is the pulse local time, and $\gamma$ is the nonlinear parameter given by $\gamma = n_2\omega_0/(cA_{eff})$, with $n_2$ is the nonlinear refractive index, $\omega_0$ is the central frequency , $c$ is the speed of light in vacuum, and $A_{eff}$ is the effective mode area. The dispersion operator is defined as: $$\hat{D}\left(i\frac{\partial}{\partial T}\right) = \sum_{k}\frac{\beta_k}{k!}\left(i\frac{\partial}{\partial T}\right)^k,
\label{Dop}$$ for $k = 2,3,4$. $\beta_k$ is the $k^{th}$ order of dispersion. The gain in the doped fibre section is calculated using: $$g = \frac{g_0}{1+E(z)/E_{sat}}.
\label{gain}$$ where $g_0 = 3.45$ is the small-signal gain (corresponding to 30 dB in power and non-zero only in the doped-fibre section), $E(z) = \int|A(z,T)|^2$ is the pulse energy and $E_{sat}$ is the saturation energy, which is adjusted to simulate changing the pump power. We multiply $g(z)$ with a Lorentzian profile of $50$ nm width to form the finite gain bandwidth $g(z,\omega_0)$. The saturable absorber is modelled by a transfer function that describes its transmittance $$T(\tau) = 1-\frac{q_0}{1+P(\tau)/P_0}.
\label{SA}$$ where $q_0$ is the unsaturated loss of the saturable absorber, $P(\tau) = |A(z,\tau)|^2$ is the instantaneous pulse power and $P_0$ is the saturation power. The spectral pulse-shaper is modelled by multiplying the electric field by a phase following the expression in Eq. \[phase\_eq\] in the spectral domain. The insertion losses ($\approx 5.6$ dB) of the spectral pulse-shaper are also taken into account in the simulations. Our numerical model is solved with a standard symmetric split-step Fourier method algorithm. The dispersion operator is applied in the frequency domain, whilst the nonlinear terms and gain are calculated in the time domain. Four our simulations we have used an initial field composed of Gaussian random noise multiplied by a sech shape in the time domain. The same stable solutions are reached for different initial noise fields.
The parameters used in our numerical simulations are the same as their experimental values. The erbium-doped fibre is $1.5$ m long, with a mode-field diameter (MFD) of $9.5$ $\mu$m, numerical aperture (NA) of 0.13 and $\gamma = 0.0016$ W$^{-1}$m$^{-1}$ at $1560$ nm. The rest of the cavity is built from SMF-28 fibre. SMF-28 has an MFD of $10.4$ $\mu$m, NA of 0.14, $\gamma = 0.0013$ W$^{-1}$m$^{-1}$ at $1560$ nm. The dispersion coefficient are, $\beta_2 = -21.4$ ps$^{2}$km$^{-1}$, $\beta_3 = 0.12$ ps$^{3}$km$^{-1}$ and $\beta_4 = -0.0022$ ps$^{4}$km$^{-1}$. For the simulated results shown in Fig. \[spectrum\], $q_0 = 0.7$ and $P_0 = 500 W$. The saturation energy was set at $E_{sat} = 90$ pJ and $E_{sat} = 100$ pJ, for the conventional and pure-quartic solitons, respectively.
**Mode-locking.** $980$ nm light from two laser diodes is delivered to the cavity through two $980/1550$ nm wavelength division multiplexers. An optical isolator is used to ensure unidirectional pulse propagation in the cavity. Nonlinear polarisation evolution is implemented using a set of two fibre polarisation controllers and a fibre polariser, which act as an artificial saturable absorber. The laser is self-starting and multi-pulsing at a pump power of $370$ mW after adjusting the polarisation controller. Single pulsing is then achieved by decreasing the pump power to $155$ mW.
**Phase-resolved characterization method.** The output pulses were input into the FREG apparatus. The pulses were split into to two branches by a 70/30 fibre-coupler. 30% of the output power was sent to a branch with a variable delay, before being detected by a fast photodiode and transferred to the electrical domain. This electrical signal drove a drove a Mach–Zender modulator that gated the optical pulses from the 70% branch of the fibre-coupler. Using an optical spectrum analyser (OSA), we measured the spectra as a function of the delay to generate a series of optical spectrograms. We then de-convolved the spectrograms with a blind deconvolution numerical algorithm (512x512 grid-retrieval errors < 0.005) to retrieve the pulse intensity and the phase in the temporal domain. We verified the validity of the retrieved pulses by taking their Fourier transform and checked these matched with the output spectra measured with the OSA.
Funding Information {#funding-information .unnumbered}
===================
Australian Research Project (ARC) Discovery Project (DP180102234); University of Sydney *Professor Harry Messel Research Fellowship*; Asian Office of Aerospace R&D (AOARD) grant (FA2386-19-1-4067).
Author contributions {#author-contributions .unnumbered}
====================
A. B.-R , C. M. d. S, and D. D. H. conceived the idea of the pure-quartic soliton laser. A. F. J. R , D. D. H, and A. B. -R. designed the experiment. A. F. J. R. performed the experiments and the numerical simulations. K. K. K. T and C. M. d. S. carried out the theoretical analysis. C. M. d. S. and A. B.-R supervised the overall project. All the authors contributed to the interpretation of data and wrote the manuscript.
References {#references .unnumbered}
==========
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The LHCb experiment at LHC is a single-armed spectrometer designed to pursue extensive, high precision studies of CP violation and rare phenomena in $b$ hadron decays. In this contribution, the trigger and reconstruction performance are summarised, and the expected performance for selected $b$ physics measurements is discussed.'
author:
- 'C.Lazzeroni. For the LHCb Collaboration'
bibliography:
- 'LHCb\_phys.bib'
title: 'Trigger, Reconstruction and Physics Performances in LHCb'
---
=1
[ address=[Cavendish Laboratory, JJ Thomson Avenue, Cambridge, CB3 0HE, United Kingdom]{} ]{}
Introduction
============
LHCb is one of the four major experiments that will take data at the LHC, due to start operation in 2007. The primary aims of LHCb are to perform precision tests of CP violation and to search for new physics in $b$ hadron decays. About $10^{12}$ $b \bar{b}$ pairs will be produced in LHCb per $10^7$ seconds (a nominal year) in $pp$ collisions with a luminosity of $2 \times 10^{32} \ cm^{-2} s^{-1}$. A large, high-purity sample of $b$ hadrons, decaying in a variety of channels, will be accumulated. LHCb will perform a detailed study of B meson mixing, precise measurements of the angles of the unitary triangle and investigations of rare decays in $b$ hadrons, looking for new physics in loop-induced processes.
The LHCb detector is optimised to reach these physics goals. Here a brief description of the reconstruction performance is given, and the sensitivities in typical channels for the study of $B_s$ mixing, CP violation and rare decays are summarised.
Trigger and Event Reconstruction performance
============================================
At the LHC energies, the production of $b$ and $\bar{b}$ quarks is predominantly along the beam axis, and highly correlated with each other, so that if one $b$ goes forward into the detector acceptance, the corresponding $\bar{b}$ products are also captured with high probability. This leads to high effienciency for a single arm spectrometer design, like LHCb, where only a factor $\sim$2 is lost compared to a twin-arm spectrometer. Moreover, owing to an extended acceptance in the forward direction (higher $\eta$), a factor 2 better $b\bar{b}$ cross section is expected at LHCb compared to the ATLAS and CMS experiments. At the LHCb luminosity, events are dominated by single proton-proton intereactions and the occupancy in the detector remains low.
The LHCb detector comprises a beam pipe, a Vertex Locator, a tracking system with a dipole magnet, two Ring Imaging Cherenkov detectors, an electromagnetic and a hadronic calorimeter and a muon system. To achieve its physics goals, efficient trigger, exclusive signal reconstruction, good proper time resolution and flavour tagging are essential for LHCb. For a detailed description of the detector, see [@VeloTDR; @ITTDR; @OTTDR; @RichTDR; @CaloTDR; @MuonTDR]. For an update on the installation and commisioning status, see [@Bolek].
Trigger
-------
The experiment will use various levels of trigger to reduce the 10 MHz rate of visible interactions to the 2 kHz that will be stored. The $b$-quark production cross section of $\sim 0.5$ mb at $\sqrt s \sim $14 TeV $pp$ interactions is only a small fraction of the total visible cross section of $\sim 100$ mb. The LHCb trigger fully exploits the $b$-decay topology, characterised by significant transverse momentum due to the high $b$-quark mass, and long lifetime yielding large impact parameter (IP) values. The first level trigger (Level 0, hardware) relies on high-$p_T$ $e,\gamma ,
\pi^0, \mu , h$ candidates, decreasing the rate to 1 MHz. The higher level trigger (HLT) is software based and makes use of the full detector information. Four independent trigger streams are defined, depending on the Level 0 output: $\mu$, $\mu + h$, $h$, and $ecal$ streams. The $p_T$ and IP discriminations are then used to reduce the output to 2 kHz. The overall trigger efficiency varies from 30% to 80% depending on the signal channels.
Reconstruction
--------------
Exclusive decay reconstruction requires good mass resolution, which is linked to excellent momentum resolution. For tracks with a momentum higher than 10 GeV/c the average efficiency is $\sim$95%, tuned to have a ghost rate of $\sim$4% for $p_t >
0.5$ GeV/c [@Jonesc]. The average track momentum resolution is $\delta p/p = 0.37$%.
The proper time of a $b$ hadron decay is determined from the distance between its production and decay vertex and from its momentum. Using a double Gaussian fit to the distribution of the vertex residuals, the resolution on the primary vertex position along the beam direction (z coordinate) is measured to be 44 $\mu$m, with 22% of the events being in the second Gaussian which is 124 $\mu$m wide (see Fig. \[vtx\] (left)). A resolution of 168 $\mu$m is obtained for the $B_s \rightarrow D_s K$ decay vertex, taken as an example, using a simple Gaussian fit (see Fig. \[vtx\] (centre)). Therefore, the proper time resolution is usually dominated by the resolution on the B decay vertex. The proper time resolution for the same decay is shown in Fig. \[vtx\] (right). Using a double Gaussian fit the core resolution is measured to be 33 $fs$. The second Gaussian accounts for 31% of the events and has a width of 67 $fs$.
Efficient particle identification in the momentum range 2-100 GeV/c is needed for flavour tagging and background rejection. The average kaon identification efficiency for momenta between 2 and 100 GeV/c is 83% with a $\pi$ mis-identification rate of 6%. Different optimisations can be performed, in order to improve the mis-identification rate at the expences of the efficiency. A detailed review of tracking performance and particle identification can be found in [@Jonesc].
Neutral particle reconstruction is performed using both resolved (separate) clusters and merged (overlapping) cluster shapes in the electromagnetic calorimeter. A mean efficiency of $\sim$53% is obtained for $B^0 \rightarrow \pi^+
\pi^- \pi^0$ events, as shown in Fig. \[neutral\] (left) as a function of the $\pi^0$ transverse momentum. The $\pi^0$ mass resolution is shown in Fig. \[neutral\] (centre) and (right) for resolved and merged $\pi^0$ respectively.
As an example of the mass resolutions that can be achieved, the reconstructed $D_s$ and $B_s$ mass distributions for $B_s \rightarrow D_s(KK\pi) K$ events are shown in Fig \[resolution\] (left, centre). The mass resolutions are $\sim 5.5$ MeV/c$^2$ for $D_s$ and $\sim 14$ MeV/c$^2$ for $B_s$. The second peak in the $B_s$ distribution corresponds to $B_s \rightarrow D_s(KK\pi) \pi$ where the $\pi$ has been mis-identified as a $K$.
The identification of the initial b-quark charge (flavour) of a reconstructed $b$ hadron decay is performed using opposite-side and same-side tagging algorithms. To determine the flavour of the accompanying B, opposite-side tagging uses the charge of: a) leptons from a semileptonic decay; b) a kaon from a $b \rightarrow c
\rightarrow s$ chain; c) the charge of all particles in a jet or at a vertex. Same-side tagging uses the charge of fragmentation particles which are correlated in phase space with the signal B meson to determine its flavour. Figure \[table\] (left) lists the tagging power ($\epsilon D^2 =
\epsilon (1 - 2 w)^2$), where $\epsilon$ is the tagging efficiency and $w$ is the wrong tagging fraction) of each tagging category and the combined tagging power for $B_d$ and $B_s$ mesons. The range of values depends on the signal channel considered.
For further details on reconstruction performances, see [@ReopTDR].
![Primary vertex resolution (left), secondary vertex resolution (centre), proper time resolution (right).[]{data-label="vtx"}](primvtx "fig:"){height=".24\textheight"} ![Primary vertex resolution (left), secondary vertex resolution (centre), proper time resolution (right).[]{data-label="vtx"}](secvtx "fig:"){height=".24\textheight"} ![Primary vertex resolution (left), secondary vertex resolution (centre), proper time resolution (right).[]{data-label="vtx"}](timeresolution "fig:"){height=".22\textheight"}
![Efficiency for $\pi^0$ reconstruction as a function of $p_T$ (left), $\pi^0$ mass resolution for resolved $\pi^0$s (centre), $\pi^0$ mass resolution for merged $\pi^0$s (right).[]{data-label="neutral"}](neuteffic "fig:"){height=".24\textheight"} ![Efficiency for $\pi^0$ reconstruction as a function of $p_T$ (left), $\pi^0$ mass resolution for resolved $\pi^0$s (centre), $\pi^0$ mass resolution for merged $\pi^0$s (right).[]{data-label="neutral"}](resolved "fig:"){height=".24\textheight"} ![Efficiency for $\pi^0$ reconstruction as a function of $p_T$ (left), $\pi^0$ mass resolution for resolved $\pi^0$s (centre), $\pi^0$ mass resolution for merged $\pi^0$s (right).[]{data-label="neutral"}](merged "fig:"){height=".24\textheight"}
![$D_s$ mass resolution (left), $B_S$ mass resolution (right). The arrow indicates the $3\sigma$ region.[]{data-label="resolution"}](dsmass "fig:"){height=".24\textheight"} ![$D_s$ mass resolution (left), $B_S$ mass resolution (right). The arrow indicates the $3\sigma$ region.[]{data-label="resolution"}](bsmass "fig:"){height=".24\textheight"}
Physics performance
===================
Selected baseline physics measurements are discussed below. All event samples were produced using the full detector simulation, realistic digitization and reconstruction with full pattern recognition and realistic trigger simulation. Toy Monte Carlo algorithms were used for sensitivity studies. The background was assumed to come from $b\bar{b}$ inclusive events, having similar topologies. All sensitivities are relative to one year of data taking or $10^7$ s, equivalent to $2 fb^{-1}$ accumulated statistics, unless otherwise specified. For the estimates of the background over signal ratio (B/S), the 90% CL is used throughout this paper.
$B_s \bar{B_s}$ mixing
----------------------
The LHCb precision for measuring $B_s \bar{B_s}$ oscillations has been estimated using the decay $B_s \rightarrow D_s \pi$. A sample of 80k events ($B/S < 0.3$) will be collected per year of running, with a proper time resolution of $\sim 40 \ fs$. A clear observation of the oscillation pattern can be performed, as seen in Fig.\[sec12\] (left). Given the value of $\Delta m_s$ recently measured [@CDF; @D0], LHCb will be able to measure it with much less than $2 fb^{-1}$ of data, and a high precision measurement is expected in 1 year: $\sigma_{stat} (\Delta m_s)\sim 0.01 \ ps^{-1}$.
The $B_s \rightarrow J/\Psi \phi$ decay measures simultaneously $\phi_s = -2 \chi$ and $\Delta \Gamma_s / \Gamma_s$, the oscillation phase and the decay width difference between the two CP eigenstates. The values for those quantities predicted by the Standard Model are respectively -0.04 and 0.1. This decay mode provides high statistics ($\sim$120k events/year, $B/S < 0.33$), but requires partial wave analysis since it contains both CP-even and CP-odd contributions. The sensitivities after 1 year of running are $\sigma(\phi_s) \sim 0.03$ and $\sigma(\Delta \Gamma_s / \Gamma_s) \sim 0.02$; the first is comparable to the Standard Model prediction, while the second is significantly better. Using the CP modes $B \rightarrow J/\Psi \eta, \eta_c \phi$, the sensitivity is improved to $\sigma(\phi_s) \sim 0.013$ after 5 years.
Measurement of $\alpha$ and $\gamma$
------------------------------------
LHCb will provide precise measurements of all the angles of the unitary triangle. Together with the excellent knowledge already obtained on the angle $\beta$ from experiments at $e^+e^-$ colliders, and the information available on the sides of the triangle, a measurement of the other two angles will over-constrain the unitary triangle and may allow detection of NP contributions to CP violation.
The sensitivity for the angle $\alpha$ has been studied using the decay $B_d \rightarrow \rho \pi$. A sample of 14k events will be reconstructed per year, with a selection based on multivariate analysis, and a ratio $B/S < 1$. Using a time-dependent Dalitz fit method [@alpha], the angle $\alpha$ can be determined with one year of data to a precision of $10^\circ$ (see Fig.\[sec12\] (right) for the Dalitz plot).
The angle $\gamma$ can be measured at LHCb using several methods.
The decay $B_s \rightarrow D_s K$, where only tree diagrams contribute, should provide a measurement of $\gamma -2 \chi$ free from new physics effects. The mixing phase $2 \chi$ will then be measured separately, as illustrated above. Here the $K/\pi$ separation significantly suppresses reflections from $B_s \rightarrow D_s \pi$. An annual yield of 5.4k events is expected, with a ratio $B/S < 1$. The measurement of $\gamma -2 \chi$ will be performed using the time-dependent rates of $B_s \rightarrow D_s^+ K^-$ and $B_s \rightarrow D_s^- K^+$, and their charge conjugates (see Fig \[dsk\]). For $\Delta m_s = 20 ps^{-1}$ this measurement gives a precision $\sigma(\gamma)$ of 14$^\circ $ with 2$fb^{-1}$.
$\gamma$ can be extracted from the interference of $B^\pm \rightarrow D^0 K^\pm$ and $B^\pm \rightarrow \bar{D^0} K^\pm$ when $D^0$ and $\bar{D^0}$ decay to a common final state. The decay amplitude can be parameterised as $A(B^- \rightarrow \bar{D^0} K^-)=
A_B r_B e^{i(\delta - \gamma)}$ where $A_B = A(B^- \rightarrow D^0 K^-)$, $r_B$ is the relative colour and CKM suppression between the two modes, and $\delta$ is the strong phase difference. Two types of decays are currently under study in LHCb: Cabibbo favoured self-conjugate decays (like $K_s \pi\pi$), the sensitivity of which is under study, and Cabibbo favoured/doubly Cabibbo suppressed modes (like $K\pi$,$K\pi\pi\pi$). The latter study is based on the ADS method [@ads; @ads1]: the relative rates of $B^+ \rightarrow D K^+$ and $B^- \rightarrow D K^-$ depend on the parameters $\gamma$, $r_B$, $\delta_B$, $r_D$, $\delta_D$. Taking $r_B=0.15$, $\sim$60k $B^\pm \rightarrow D^0 K^\pm$ and $\sim$2k $B^\pm \rightarrow \bar{D^0} K^\pm$ decays are expected in one year, and the sensitivity to $\gamma$ is $\sigma(\gamma)\sim 5^\circ$ considering the background ($\sim 3.9^\circ$ with no background, see Fig. \[dsk\] (right) for an example of the fit results). The sensitivity with lower values of $r_B$ is under study.
The decay $B_d \rightarrow D^0 K^{*0}$ also proceeds via a tree diagram, and has no sensitivity to new physics. The method [@glw] to extract $\gamma$ is based on the measurement of six time-integrated decay rates for $B_d \rightarrow D^0 K^{*0}$, $B_d \rightarrow \bar{D^0} K^{*0}$, $B_d \rightarrow D^0_{CP} K^{*0}$ and their CP conjugates. Samples of about 3.4k $B_d \rightarrow \bar{D^0}(K^- \pi^+)K^{*0}$, 0.5k $B_d \rightarrow D^0(K^+ \pi^-)K^{*0}$, and 0.6k $B_d \rightarrow D^0_{CP}(K^+ K^-)K^{*0}$ decays will be reconstructed per year (see Table \[table\] (right)). The precision reached is $\sigma(\gamma) \sim 8^\circ$ with 2$fb^{-1}$.
$\gamma$ can be measured from time-dependent asymmetries for $B_d \rightarrow \pi \pi$ and $B_s \rightarrow KK$ decays [@battaglia; @fleischer1; @fleischer2], where both penguin and tree diagrams contribute. This measurement is sensitive to new physics appearing in the penguin loops. Here again the $K/\pi$ separation is crucial because of numerous reflections from $B_{d,s}$ and $\Lambda_b$ decays. The asymmetries can be parameterised as $A_{CP} = A^{dir} cos(\Delta mt) +
A^{mix} sin(\Delta mt)$. The four observables are functions of the parameters: $\gamma$, mixing phases $\phi_d$ and $\phi_s$, ratios of penguin over tree contributions $P/T = d e^{i\theta}$. Measuring $\phi_d$ and $\phi_s$ independently and assuming U-spin symmetry ($d_{\pi\pi} = d_{KK}$, $\theta_{\pi\pi} = \theta_{KK}$), the relations above reduce to four measured quantities with three unknowns, which can be solved for $\gamma$. The annual yield of 26k $B_d \rightarrow \pi \pi$ ($B/S < 0.7$) and 37k $B_s \rightarrow KK$ ($B/S < 0.3$) events allows a precision of $\sigma(\gamma) \sim 5^\circ$. Since the system is overconstrained, the hypothesis of U-spin conservation can be also tested.
![Distribution of $B_s \rightarrow D_s \pi$ proper time (ps) after 1 year of data taking, for $\Delta m_s = 15, 25 \ ps^{-1}$ (left). Dalitz plot for the decay $B \rightarrow \pi^0 \pi^+ \pi^-$ (right).[]{data-label="sec12"}](oscillaz "fig:"){height=".24\textheight"} ![Distribution of $B_s \rightarrow D_s \pi$ proper time (ps) after 1 year of data taking, for $\Delta m_s = 15, 25 \ ps^{-1}$ (left). Dalitz plot for the decay $B \rightarrow \pi^0 \pi^+ \pi^-$ (right).[]{data-label="sec12"}](alpha "fig:"){height=".24\textheight"}
![Distribution of $B_s \rightarrow D_s K$ asymmetry versus proper time (ps) for $D_s^+ K^-$ and $D_s^- K^+$ (left). Example of fit results for $r_B$, $\delta_B$, $\delta_D$ and $\gamma$ in $B^\pm \rightarrow D^0 K^\pm$ using the ADS method with no background.[]{data-label="dsk"}](dsk_asym "fig:"){height=".24\textheight"} ![Distribution of $B_s \rightarrow D_s K$ asymmetry versus proper time (ps) for $D_s^+ K^-$ and $D_s^- K^+$ (left). Example of fit results for $r_B$, $\delta_B$, $\delta_D$ and $\gamma$ in $B^\pm \rightarrow D^0 K^\pm$ using the ADS method with no background.[]{data-label="dsk"}](fitads "fig:"){height=".24\textheight"}
Rare decays: $B \rightarrow K^* \mu \mu$
----------------------------------------
Rare loop-induced decays are sensitive to new physics in many Standard Model extensions. At LHCb, rare decays such as decay $B_d \rightarrow K^* \mu \mu$ will be studied. The expected annual yield is $\sim$4.5k, with a ratio $B/S < 0.2$, using the Standard Model prediction for the branching ratio of $\sim 10^{-6}$. Further improvements are expected in the signal selection to enhance the yield.
This channel is well suited to searches for new physics, since New Physics models make definite predictions for the shape of the forward-backward asymmetry of the $\mu^+$ in the $\mu\mu$ rest frame with respect to the $B$ direction, as a function of the $\mu\mu$ invariant mass [@rare] (see Fig. \[kstmumu\] (left)). In particular the zero crossing of the forward-backward asymmetry is predicted with small theoretical uncertainties. With a year of data taking, a clear observation of NP should be possible; with $10 \ fb^{-1}$ the position of the zero should be located with a precision of $\pm 0.53$ GeV$^2$ (see Fig \[kstmumu\] (right)).
![Distribution of the forward-backward asymmetry in $B_d \rightarrow K^* \mu \mu$, as a function of the $\mu\mu$ invariant mass, for different extensions of the Standard Model (left). Distribution of the forward-backward asymmetry as expected in LHCb with $10 fb^{-1}$ (right).[]{data-label="kstmumu"}](afbtheo "fig:"){height=".24\textheight"} ![Distribution of the forward-backward asymmetry in $B_d \rightarrow K^* \mu \mu$, as a function of the $\mu\mu$ invariant mass, for different extensions of the Standard Model (left). Distribution of the forward-backward asymmetry as expected in LHCb with $10 fb^{-1}$ (right).[]{data-label="kstmumu"}](afb "fig:"){height=".24\textheight"}
![Tagging power (left). Annual yield and $S/B$ ratio for the decays $B_d \rightarrow D^0 K ^{*0}$ (right).[]{data-label="table"}](tabletag "fig:"){height=".15\textheight"} ![Tagging power (left). Annual yield and $S/B$ ratio for the decays $B_d \rightarrow D^0 K ^{*0}$ (right).[]{data-label="table"}](glw "fig:"){height=".10\textheight"}
Conclusions
===========
Using a detailed simulation of the LHCb detector, we demonstrate that the LHCb experiment can efficiently reconstruct many different decay modes with a very good performance in the trigger, proper time resolution, mass resolution, and flavour tagging. This will allow LHCb to fully explore the $B_s$ mixing, to extract CKM parameters with various methods, and to perform studies of rare decays. The experiment will make precision tests of the Standard Model in order to constrain, and possibly discover, new physics.
The author would like to thank I. Belyaev, O. Deschamps, H. Dijkstra, R. Forty, C. R. Jones, T. Ruf, O. Schneider G. Wilkinson, and Y. Xie for their help in preparing these proceedings.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Nathanial P. Brown'
title: Quasitraces need not be traces
---
This paper has been withdrawn.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We review here the main advances made by using effective field theories (EFTs) in classical gravity, with notable focus on those unique to the EFTs of post-Newtonian (PN) gravity. We then proceed to overview the various prospects of using field theory to study the real-world gravitational wave (GW) data, as well as to ameliorate our fundamental understanding of gravity at all scales, by going from the EFTs of PN gravity to modern advances in scattering amplitudes, including computational techniques and intriguing duality relations between gauge and gravity theories.'
address: |
Institut de Physique Théorique, CEA & CNRS, Université Paris-Saclay\
91191 Gif-sur-Yvette, France\
[email protected]
author:
- Michele Levi
title: Field theory for gravity at all scales
---
Introduction
============
In recent years there has been a growing recognition in the all-encompassing relevance of the field theoretic framework beyond the quantum realm of particle physics, in which much of it has been conceived. Most notably, for long, in the absence of a viable quantum theory of gravity, the overlap between field theory and classical gravity has been largely deemed nonexistent. In particular, the conceptual framework of effective field theories (EFTs), which is intimately tied with the notion of renormalization, was widely regarded as uniquely ingrained in the domain of quantum field theories (QFTs). In order to dispel the latter fallacy, however, suffice it to acknowledge that ‘renormalization’ can in fact be taken as synonymous with ‘resolution of physical scales’. Once this realization is made, it becomes clear that the framework of EFTs is evidently a universal one, and that it can be applied on any wide range of physical scales. Moreover, the EFT framework is a solid and powerful one, providing an inherent setup strategy and a robust toolbox, geared towards yielding high precision predictions to a desirable accuracy.
Indeed, just over a decade ago, an EFT approach to handle gravitational waves (GWs), emitted from inspiralling compact binaries, which are analytically treated with post-Newtonian (PN) gravity [@Blanchet:2013haa], was put forward by Goldberger et al.[@Goldberger:2004jt; @Goldberger:2007hy; @Goldberger:2009qd]. This novel approach has resulted, in turn, further applications of EFTs in classical gravity. Notably, significant progress was demonstrated in the application of EFTs to higher dimensional gravity in the context of large extra dimensions[@Chu:2006ce; @Kol:2007rx; @Emparan:2009cs; @Gilmore:2009ea; @Emparan:2009at]. Further, also in the context of GWs, an investigation of an EFT treatment for the extreme mass ratio inspiral (EMRI) case of compact binaries, was initiated in Ref. . Moreover, within a decade the initial EFT approach to PN gravity has been able – in important sectors of the theory – to catch up and go beyond the spectacular state of the art[@Levi:2018nxp], already accomplished within the traditional framework of General Relativity (GR).
In what follows we highlight the main meaningful unique advances made in the EFT approach to PN gravity, yet more importantly, we point to the main prospects on a broader scope of using field theory to study gravity, stretching from delivering highly demanding accurate predictions for real-world GW data, to confronting our fundamental grasp of QFTs and gravity theories at all scales.
From Gravitational Waves To Gravity At All Scales
=================================================
Significant progress was made via the EFT approach to PN gravity, which is concerned with the analytical prediction of the inspiral portion of the GW signal. While this progress is evident and encouraging, it crucially serves to illustrate the great potential, still awaiting, in the use of field theory to study gravity, with many implications at various levels. First, there are clear anticipated developments still within the EFTs of PN gravity and the theoretical modeling of the GW signal. Further, even if still only in the context of GWs, modern advances in scattering amplitudes hold the promise to possibly enable an analytical treatment of the GW signal in its entirety, that is also in the strong field regime, beyond the inspiral phase. This may significantly improve upon the current phenomenological GW modeling, which relies heavily on the effective-one-body (EOB) formulation [@Buonanno:1998gg], by allowing for a smooth analytical model of the whole signal. Yet, more broadly, these advances from scattering amplitudes also importantly allow us to study gauge and gravity theories at a fundamental level, across the classical and quantum regimes.
Let us first review the main unique advancements accomplished within the EFTs of PN gravity so far. The most notable and extensive progress was realized in the treatment of the spinning sector, with several new higher order PN corrections. Next-to-leading order (NLO) effects in the conservative sector were first tackled in Refs. , following Refs. , and further resolved and extended to the current state of the art, at the fourth PN (4PN) order for rapidly rotating compact objects, in Refs. , following Refs. . Further, partial results at NLO in the radiative sector were presented in Ref. . Moreover, following Ref. , which also led to a new finite size correction of radiation reaction in classical electrodynamics[@Galley:2010es], and Refs. , a leading order effect of radiation reaction with spins was obtained in Ref. . All in all, beyond the specific new PN results obtained with spins, there has been an improvement in the understanding of classical spins in gravity.
Another remarkable finding in the field was the uncovering of classical renormalization group (RG) flows of the Wilson coefficients, which characterize the effective theories, both at the one-particle level[@Kol:2011vg], and at the level of the composite system, for the multipole moments of the binary[@Goldberger:2009qd; @Goldberger:2012kf]. These give rise to higher order PN logarithm corrections, which constitute unique predictions of the EFT framework. In addition, the 2PN order correction for a generic $n$-body problem was considered in Ref. , where automated computations were first advocated. Finally, the ‘EFTofPNG’ code for high precision Feynman calculations in PN gravity was created[@Levi:2017kzq], comprising the first comprehensive code in PN theory made public. The code consists of independent modules for easy adaptation and future development, where some prospects of building on this state of the art PN technology for various research purposes, were outlined in Ref. .
Within the methodology of the EFTs of PN gravity, there are several important directions, where development is required. The first is the solution of higher order PN equations of motion (EOMs) for accurate orbital dynamics. One interesting related way to find such closed form solutions, is the dynamical RG method, suggested in Ref. . Further, the treatment of non-conservative sectors with the EFTs of PN gravity has been rather limited, and so these various sectors should be tackled with a proper EFT formulation, and implementation for new PN results. Another objective to follow is the possible improvement or extension of the EFT formulation for spinning objects. For example, an alternative EFT formulation, which is ideally suited for the treatment of slowly rotating objects, based on a coset construction, was presented in Ref. . Further investigation is required in order to see whether these independent EFT formulations may possibly be integrated, and entail an even better understanding of classical spins in gravity. Next, we note the analysis of the mass- and spin-induced Wilson coefficients, and of the binary multipole coefficients, which also mostly remains to be approached via formal EFT matching. Finally, the continual public development of the ‘EFTofPNG’ code, or similar open codes, is required in order to keep up to date with the actual progress in PN theory. In that regard, it is important to note that classical theories of gravity, which modify GR in the IR, motivated by the cosmological constant problem and the dark matter puzzle[@Clifton:2011jh; @Deffayet:2013lga; @deRham:2014zqa; @Joyce:2014kja], can be incorporated in a rather straightforward manner into the ‘EFTofPNG’ code, in order to study their analytical predictions for the GW signal from the compact binary inspiral, see, e.g., Refs. .
Advances and prospects for gravity in field theory
--------------------------------------------------
Ultimately, we would like to invoke the correspondence – at any level – between gauge and gravity theories, in order to push beyond the state of the art in high precision computation for concrete analytical predictions of the real-world GW signal, as well as to directly ameliorate our fundamental understanding of the foundations of these theories, in particular of gravity. These objectives can be considerably facilitated by turning to modern field theory advances in the domain of scattering amplitudes, see, e.g., Ref. . In fact, the field of EFTs in PN gravity, on which we have focused hitherto, and of scattering amplitudes, which are both directly concerned with baldly tackling demanding high precision computation driven by experiment, share more profound parallels beyond the obvious robust technical tools they entail, see, e.g., in Ref. . Both of these fields push to the exposure of the universal commonalities across classical and quantum field theories, and drive us to confront our fundamental grasp of the underlying foundations of these theories. One crucial difference to note, though, between these fields, is that where in scattering amplitudes the computational outcome is directly related with the physical observables, the objects which are commonly computed in classical gravity are coordinate/gauge dependent quantities.
Let us then note, more specifically, some prospective avenues to deploy this broad correspondence of gauge theories and gravity. First, there is the standard working knowledge, which can be exchanged between the theories, e.g. multi-loop techniques in QFT, such as integration by parts (IBP), and high loop master integrals, along with other Feynman calculus and technology[@Smirnov:2012gma; @Levi:2017kzq; @Levi:2018stw]. This was nicely demonstrated, e.g., in Refs. , where an analytic evaluation of a four-loop master integral was provided for the first time from the classical gravity context. Furthermore, modern scattering amplitudes advances, such as the BCFW on-shell recursion relations[@Britto:2004ap], and generalized unitarity methods, which imply that tree level data encodes all multiplicity at the integrand level, were put forward to extract classical higher order loop results for gravitational scattering[@Bjerrum-Bohr:2013bxa; @Cachazo:2017jef; @Guevara:2017csg; @Bjerrum-Bohr:2018xdl]. Such a scattering treatment may also enable to analytically tackle the strong field regime of the GW signal, and hence smoothly model analytically the entire signal, as we noted above.
Yet, of particular interest are the novel intriguing color-kinematics or BCJ duality relations, and the related double copy correspondence [@Bern:2008qj; @Bern:2010ue], see, e.g., review in Ref. , which were discovered in the context of high loop computations of amplitudes in supersymmetry and supergravity theories. Such relations were already known from string theory to hold at tree level, but formulated in terms of the novel generic double copy correspondence, these relations have been successfully used to study UV divergences of supergravity theories. This recent new perspective on gravity, viewing gravitons as double copies of gluons, suggests that even the simplest gauge theories and GR are in fact frameworks, which are intimately connected. Though this correspondence was uncovered in the perturbative context, it was also found, remarkably, that particular classical *exact* solutions in GR, are in fact double copies of exact “single copy” counterparts in corresponding gauge theories. Such classical solutions include all vacuum stationary solutions, most notably, Kerr black holes, as well as their higher dimensional generalizations, see Ref. and references therein. Yet, the underlying origin of these perturbative relations, as well as their essential connection to the particular classical exact correspondence revealed, are yet to be uncovered.
In conjunction with the recent novel methods from scattering amplitudes, it is expected that this double copy correspondence can be used to advance the analytical calculations for the theoretical prediction of the GW signal, which may also help to shed more light on the nature and origin of this correspondence. Related with that end, an important time dependent case was studied in terms of the exact double copy of the Kerr-Schild form, of an arbitrarily accelerating point source [@Luna:2016due]. In this case the radiation current is double copied to the radiation stress-energy tensor in Fourier space, and both are related to the corresponding scattering amplitudes, which can be obtained from the amputated currents. Hence, the work in Ref. made a first explicit connection between the classical double copy in an exact form, to that in the perturbative scattering amplitudes context. At this point it should also be stressed, that from the scattering amplitudes context, it is already known that gluon amplitudes can double copy to arbitrary combinations of amplitudes for gravitons, and the additional unobserved dilaton and B fields, depending on the choice of the polarization states in the gauge theory amplitudes. This was indeed an ambiguous issue in Ref. , which built on Ref. , and that subsequently Ref. set out to address, with Ref. following up successfully at NLO.
Finally, we note the topic of soft graviton theorems for scattering amplitudes, which were recently demonstrated to be equivalent to gravitational memory effects, as part of a triple equivalence of the IR structure of gauge and gravity theories among soft theorems, asymptotic symmetry, and memory effects[@Strominger:2017zoo]. The latter may have observable signatures on the GW signal.
Acknowledgments {#acknowledgments .unnumbered}
===============
I am grateful to John Joseph Carrasco for his meaningful encouragement. I would like to thank Donato Bini for the warm hospitality throughout the MG15 meeting, and in particular on the session, where this review was presented. It is also a delightful pleasure to acknowledge Roy Kerr who graced us with his inspiring presence. My work is supported by the European Research Council under the European Union’s Horizon 2020 Framework Programme FP8/2014-2020 “preQFT” grant no. 639729, “Strategic Predictions for Quantum Field Theories” project.
[69]{} L. Blanchet, [*[Gravitational Radiation from Post-Newtonian Sources and Inspiralling Compact Binaries]{}*]{}, [*Living Rev.Rel.*]{} [**17**]{} (2014). W. D. Goldberger and I. Z. Rothstein, [*[An Effective field theory of gravity for extended objects]{}*]{}, [*Phys.Rev.*]{} [**D73**]{} (2006) 104029. W. D. Goldberger, [*[Les Houches lectures on effective field theories and gravitational radiation]{}*]{}, in [*[Les Houches Summer School - Session 86: Particle Physics and Cosmology: The Fabric of Spacetime Les Houches, France, July 31-August 25, 2006]{}*]{}, 2007. W. D. Goldberger and A. Ross, [*[Gravitational radiative corrections from effective field theory]{}*]{}, [*Phys.Rev.*]{} [**D81**]{} (2010) 124015. Y.-Z. Chu, W. D. Goldberger, and I. Z. Rothstein, [*[Asymptotics of d-dimensional Kaluza-Klein black holes: Beyond the Newtonian approximation]{}*]{}, [*JHEP*]{} [**03**]{} (2006) 013. B. Kol and M. Smolkin, [*[Classical Effective Field Theory and Caged Black Holes]{}*]{}, [*Phys. Rev.*]{} [**D77**]{} (2008) 064033. R. Emparan, T. Harmark, V. Niarchos, and N. A. Obers, [*[World-Volume Effective Theory for Higher-Dimensional Black Holes]{}*]{}, [*Phys. Rev. Lett.*]{} [**102**]{} (2009) 191301. J. B. Gilmore, A. Ross, and M. Smolkin, [*[Caged black hole thermodynamics: Charge, the extremal limit, and finite size effects]{}*]{}, [*JHEP*]{} [**09**]{} (2009) 104. R. Emparan, T. Harmark, V. Niarchos, and N. A. Obers, [*[Essentials of Blackfold Dynamics]{}*]{}, [*JHEP*]{} [**03**]{} (2010) 063. C. R. Galley and B. Hu, [*[Self-force on extreme mass ratio inspirals via curved spacetime effective field theory]{}*]{}, [*Phys.Rev.*]{} [**D79**]{} (2009) 064002. M. Levi, [arXiv:1807.01699]{} [*[Effective Field Theories of Post-Newtonian Gravity: A comprehensive review]{}*]{}, 2018.
A. Buonanno and T. Damour, [*[Effective one-body approach to general relativistic two-body dynamics]{}*]{}, [*Phys.Rev.*]{} [**D59**]{} (1999) 084006. R. A. Porto and I. Z. Rothstein, [*[The Hyperfine Einstein-Infeld-Hoffmann potential]{}*]{}, [*Phys.Rev.Lett.*]{} [**97**]{} (2006) 021101. R. A. Porto and I. Z. Rothstein, [*[Spin(1)Spin(2) Effects in the Motion of Inspiralling Compact Binaries at Third Order in the Post-Newtonian Expansion]{}*]{}, [*Phys.Rev.*]{} [**D78**]{} (2008) 044012, \[*Erratum-ibid.* D[**81**]{} (2010) 029904\].
R. A. Porto and I. Z. Rothstein, [*[Next to Leading Order Spin(1)Spin(1) Effects in the Motion of Inspiralling Compact Binaries]{}*]{}, [*Phys.Rev.*]{} [**D78**]{} (2008) 044013, \[*Erratum-ibid.* D[**81**]{} (2010) 029905\].
R. A. Porto, [*[Post-Newtonian corrections to the motion of spinning bodies in NRGR]{}*]{}, [*Phys.Rev.*]{} [**D73**]{} (2006) 104031. M. Levi, [*[Next to Leading Order gravitational Spin1-Spin2 coupling with Kaluza-Klein reduction]{}*]{}, [*Phys.Rev.*]{} [**D82**]{} (2010) 064029. M. Levi and J. Steinhoff, [*[Spinning gravitating objects in the effective field theory in the post-Newtonian scheme]{}*]{}, [*JHEP*]{} [**09**]{} (2015) 219. M. Levi, [*[Binary dynamics from spin1-spin2 coupling at fourth post-Newtonian order]{}*]{}, [*Phys.Rev.*]{} [**D85**]{} (2012) 064043. M. Levi and J. Steinhoff, [*[Equivalence of ADM Hamiltonian and Effective Field Theory approaches at next-to-next-to-leading order spin1-spin2 coupling of binary inspirals]{}*]{}, [*JCAP*]{} [**1412**]{} (2014) 003. M. Levi and J. Steinhoff, [*[Leading order finite size effects with spins for inspiralling compact binaries]{}*]{}, [*JHEP*]{} [**06**]{} (2015) 059. M. Levi and J. Steinhoff, [*[Next-to-next-to-leading order gravitational spin-squared potential via the effective field theory for spinning objects in the post-Newtonian scheme]{}*]{}, [*JCAP*]{} [**1601**]{} (2016) 008. M. Levi and J. Steinhoff, [arXiv:1607.04252]{} [*[Complete conservative dynamics for inspiralling compact binaries with spins at fourth post-Newtonian order]{}*]{}, 2016. M. Levi, [*[Next to Leading Order gravitational Spin-Orbit coupling in an Effective Field Theory approach]{}*]{}, [*Phys.Rev.*]{} [**D82**]{} (2010) 104004. R. A. Porto, A. Ross, and I. Z. Rothstein, [*[Spin induced multipole moments for the gravitational wave flux from binary inspirals to third Post-Newtonian order]{}*]{}, [*JCAP*]{} [**1103**]{} (2011) 009. C. R. Galley and M. Tiglio, [*[Radiation reaction and gravitational waves in the effective field theory approach]{}*]{}, [*Phys.Rev.*]{} [**D79**]{} (2009) 124027. C. R. Galley, A. K. Leibovich, and I. Z. Rothstein, [*[Finite size corrections to the radiation reaction force in classical electrodynamics]{}*]{}, [*Phys.Rev.Lett.*]{} [**105**]{} (2010) 094802. C. R. Galley, [*[Classical Mechanics of Nonconservative Systems]{}*]{}, [*Phys.Rev.Lett.*]{} [**110**]{} (2013) 174301. C. R. Galley, D. Tsang, and L. C. Stein, [arXiv:1412.3082]{} [*[The principle of stationary nonconservative action for classical mechanics and field theories]{}*]{}, 2014. N. T. Maia, C. R. Galley, A. K. Leibovich, and R. A. Porto, [*[Radiation reaction for spinning bodies in effective field theory II: Spin-spin effects]{}*]{}, [*Phys. Rev.*]{} [**D96**]{} (2017) 084065. B. Kol and M. Smolkin, [*[Black hole stereotyping: Induced gravito-static polarization]{}*]{}, [*JHEP*]{} [**02**]{} (2012) 010. W. D. Goldberger, A. Ross, and I. Z. Rothstein, [*[Black hole mass dynamics and renormalization group evolution]{}*]{}, [*Phys. Rev.*]{} [**D89**]{} (2014) 124033. Y.-Z. Chu, [*[The n-body problem in General Relativity up to the second post-Newtonian order from perturbative field theory]{}*]{}, [*Phys.Rev.*]{} [**D79**]{} (2009) 044031. M. Levi and J. Steinhoff, [*[EFTofPNG: A package for high precision computation with the Effective Field Theory of Post-Newtonian Gravity]{}*]{}, [*Class. Quant. Grav.*]{} [**34**]{} (2017) 244001. M. Levi, [arXiv:1811.12401]{} [*[A public framework for Feynman calculations and post-Newtonian gravity]{}*]{}, 2018.
C. R. Galley and I. Z. Rothstein, [*[Deriving analytic solutions for compact binary inspirals without recourse to adiabatic approximations]{}*]{}, [*Phys. Rev.*]{} [**D95**]{} (2017) 104054. L. V. Delacrétaz, S. Endlich, A. Monin, R. Penco, and F. Riva, [*[(Re-)Inventing the Relativistic Wheel: Gravity, Cosets, and Spinning Objects]{}*]{}, [*JHEP*]{} [**11**]{} (2014) 008. T. Clifton, P. G. Ferreira, A. Padilla, and C. Skordis, [*[Modified Gravity and Cosmology]{}*]{}, [*Phys. Rept.*]{} [**513**]{} (2012) 1–189. C. Deffayet and D. A. Steer, [*[A formal introduction to Horndeski and Galileon theories and their generalizations]{}*]{}, [*Class. Quant. Grav.*]{} [**30**]{} (2013) 214006. C. de Rham, [*[Massive Gravity]{}*]{}, [*Living Rev. Rel.*]{} [**17**]{} (2014) 7. A. Joyce, B. Jain, J. Khoury, and M. Trodden, [*[Beyond the Cosmological Standard Model]{}*]{}, [*Phys. Rept.*]{} [**568**]{} (2015) 1–98. Y.-Z. Chu and M. Trodden, [*[Retarded Green’s function of a Vainshtein system and Galileon waves]{}*]{}, [*Phys. Rev.*]{} [**D87**]{} (2013) 024011. C. de Rham, A. J. Tolley, and D. H. Wesley, [*[Vainshtein Mechanism in Binary Pulsars]{}*]{}, [*Phys. Rev.*]{} [**D87**]{} (2013) 044025. C. de Rham, A. Matas, and A. J. Tolley, [*[Galileon Radiation from Binary Systems]{}*]{}, [*Phys. Rev.*]{} [**D87**]{} (2013) 064024. H. Elvang and Y. t. Huang, [*[Scattering Amplitudes in Gauge Theory and Gravity]{}*]{} (Cambridge University Press, 2015).
V. A. Smirnov, [*[Analytic tools for Feynman integrals,]{}*]{} Springer Tracts Mod. Phys. [**250**]{}, 1 (2012).
S. Foffa, P. Mastrolia, R. Sturani, and C. Sturm, [*[Effective field theory approach to the gravitational two-body dynamics, at fourth post-Newtonian order and quintic in the Newton constant]{}*]{}, [*Phys. Rev.*]{} [**D95**]{} (2017) 104009. T. Damour and P. Jaranowski, [*[Four-loop static contribution to the gravitational interaction potential of two point masses]{}*]{}, [*Phys. Rev.*]{} [**D95**]{} (2017) 084005. R. Britto, F. Cachazo, and B. Feng, [*[New recursion relations for tree amplitudes of gluons]{}*]{}, [*Nucl. Phys.*]{} [**B715**]{} (2005) 499–522. N. E. J. Bjerrum-Bohr, J. F. Donoghue, and P. Vanhove, [*[On-shell Techniques and Universal Results in Quantum Gravity]{}*]{}, [*JHEP*]{} [**02**]{} (2014) 111. F. Cachazo and A. Guevara, [arXiv:1705.10262]{} [*[Leading Singularities and Classical Gravitational Scattering]{}*]{}, 2017. A. Guevara, [arXiv:1706.02314]{} [*[Holomorphic Classical Limit for Spin Effects in Gravitational and Electromagnetic Scattering]{}*]{}, 2017. N. E. J. Bjerrum-Bohr, P. H. Damgaard, G. Festuccia, L. Planté, and P. Vanhove, [*[General Relativity from Scattering Amplitudes]{}*]{}, [*Phys. Rev. Lett.*]{} [**121**]{} (2018) 171601. Z. Bern, J. J. M. Carrasco, and H. Johansson, [*[New Relations for Gauge-Theory Amplitudes]{}*]{}, [*Phys. Rev.*]{} [**D78**]{} (2008) 085011. Z. Bern, J. J. M. Carrasco, and H. Johansson, [*[Perturbative Quantum Gravity as a Double Copy of Gauge Theory]{}*]{}, [*Phys. Rev. Lett.*]{} [**105**]{} (2010) 061602. J. J. M. Carrasco, [*[Gauge and Gravity Amplitude Relations]{}*]{}, in [*[Proceedings, Theoretical Advanced Study Institute in Elementary Particle Physics: Journeys Through the Precision Frontier: Amplitudes for Colliders (TASI 2014): Boulder, Colorado, June 2-27, 2014]{}*]{}, pp. 477–557, WSP, 2015. A. Luna, R. Monteiro, I. Nicholson, D. O’Connell, and C. D. White, [*[The double copy: Bremsstrahlung and accelerating black holes]{}*]{}, [*JHEP*]{} [**06**]{} (2016) 023. W. D. Goldberger and A. K. Ridgway, [*[Radiation and the classical double copy for color charges]{}*]{}, [*Phys. Rev.*]{} [**D95**]{} (2017) 125010. A. Luna, I. Nicholson, D. O’Connell, and C. D. White, [*[Inelastic Black Hole Scattering from Charged Scalar Amplitudes]{}*]{}, [*JHEP*]{} [**03**]{} (2018) 044. C.-H. Shen, [*[Gravitational Radiation from Color-Kinematics Duality]{}*]{}, [*JHEP*]{} [**11**]{} (2018) 162. A. Strominger, [arXiv:1703.05448]{} [*[Lectures on the Infrared Structure of Gravity and Gauge Theory]{}*]{}, 2017.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We report on Expanded Very Large Array (EVLA) observations of the Type IIb supernova 2011dh, performed over the first 100 days of its evolution and spanning 1–40 GHz in frequency. The radio emission is well-described by the self-similar propagation of a spherical shockwave, generated as the supernova ejecta interact with the local circumstellar environment. Modeling this emission with a standard synchrotron self-absorption (SSA) model gives an average expansion velocity of $v \approx 0.1c$, supporting the classification of the progenitor as a compact star ($R_{\star} \approx 10^{11}\rm~cm$). We find that the circumstellar density is consistent with a $\rho \propto r^{-2}$ profile. We determine that the progenitor shed mass at a constant rate of $\approx 3 \times 10^{-5}~M_{\sun}\rm~yr^{-1}$, assuming a wind velocity of $1000\rm~km~s^{-1}$ (values appropriate for a Wolf-Rayet star), or $\approx 7 \times 10^{-7}~M_{\sun}\rm~yr^{-1}$ assuming $20\rm~km~s^{-1}$ (appropriate for a yellow supergiant \[YSG\] star). Both values of the mass-loss rate assume a converted fraction of kinetic to magnetic energy density of $\epsilon_B = 0.1$. Although optical imaging shows the presence of a YSG, the rapid optical evolution and fast expansion argue that the progenitor is a more compact star—perhaps a companion to the YSG. Furthermore, the excellent agreement of the radio properties of with the SSA model implies that any YSG companion is likely in a wide, non-interacting orbit.'
author:
- 'M. I. Krauss, A. M. Soderberg, L. Chomiuk, B. A. Zauderer, A. Brunthaler, M. F. Bietenholz, R. A. Chevalier, C. Fransson, M. Rupen'
title: EVLA Observations of the Radio Evolution of
---
Introduction
============
Type IIb supernovae (SNe IIb) were first identified as a distinct class of core-collapse events after detailed observations of the “canonical” Type IIb SN1993J revealed broad hydrogen [*and*]{} helium features [@filippenko97]. Recent studies have shown that this spectroscopic class shows a broad diversity in properties including H$\alpha$ strength, profile, and evolution (e.g., compare with SN2003bg; @matheson [@hamuy; @mazzali]). In several cases, YSG stars with extended radii, $R_*\sim 100~R_{\odot}$, have been identified at the explosion sites of SNe II (1993J, 2008cn, 2009kr, 2011dh; @93J [@maund11; @vandyk11; @eliasrosa09; @eliasrosa10]). Yet progenitor diagnostics from multi-wavelength studies indicate that some SNe IIb bear stronger similarity to hydrogen-poor Type Ibc supernovae commonly associated with compact progenitors, $R_*\approx R_{\odot}$ (hereafter SNe cIIb; @chevalier10). In particular, radio-derived estimates for the shockwave velocities are typically high, $v\gtrsim 0.1c$, and difficult to explain in the context of shock breakout from an extended object [@nakar10]. Furthermore, stellar evolution tracks place YSGs outside of the SN explosion phase space of the HR diagram (@meynet05, but see @georgy11). In addition, @chevalier10 find that all proposed SNe cIIb for which there is sufficient radio data show light curve variations indicative of density modulations in the explosion environment, consistent with wind variability from a compact progenitor. Finally, binary companions have been reported for two SNe IIb to date (1993J and 2001ig; @woosley94 [@ryder06; @maund07]). Similarly, the YSG may be a binary companion rather than the progenitor star.
[rlrccccccc]{} Image rms & & & 0.045 & 0.037 & 0.033 & 0.026 & 0.026 & 0.019\
June 17 & 55729.2 & 16.4 & & & & & 2.430$\pm$0.044 & 4.090$\pm$0.063\
June 21 & 55733.2 & 20.4 & $<0.13$ & $<0.12$ & 0.540$\pm$0.079 & 1.400$\pm$0.055 & 3.150$\pm$0.043 & 4.800$\pm$0.055\
June 26 & 55738.2 & 25.4 & 0.243$\pm$0.079 & 0.800$\pm$0.066 & 1.626$\pm$0.070 & 2.920$\pm$0.061 & 4.920$\pm$0.063 & 5.980$\pm$0.070\
July 6 & 55748.1 & 35.3 & 0.331$\pm$0.072 & 1.236$\pm$0.067 & 2.982$\pm$0.073 & 4.908$\pm$0.072 & 6.871$\pm$0.091 & 7.222$\pm$0.078\
July 16 & 55758.1 & 45.3 & 0.719$\pm$0.074 & 1.858$\pm$0.069 & 4.092$\pm$0.083 & 6.188$\pm$0.080 & 7.836$\pm$0.086 & 6.987$\pm$0.077\
July 29 & 55771.0 & 58.2 & 2.47$\pm$0.12 & 3.31$\pm$0.11 & 5.84$\pm$0.15 & 7.33$\pm$0.14 & 7.47$\pm$0.12 & 6.11$\pm$0.11\
Sept 1 & 55805.7 & 92.9 & 3.45$\pm$0.11 & 5.00$\pm$0.12 & 7.02$\pm$0.17 & 6.98$\pm$0.13 & 4.884$\pm$0.073 & 3.941$\pm$0.061\
\[srcTab1\]
[rlrccccccc]{} Image rms & & & 0.015 & 0.026 & 0.029 & 0.034 & 0.036 & 0.028 & 0.037\
June 17 & 55729.2 & 16.4 & 5.535$\pm$0.057 & 6.970$\pm$0.074 & 6.790$\pm$0.073 & 6.50$\pm$0.20 & 5.13$\pm$0.16 & 4.60$\pm$0.14 & 3.47$\pm$0.11\
June 21 & 55733.2 & 20.4 & 5.870$\pm$0.060 & 5.940$\pm$0.064 & 5.313$\pm$0.057 & 4.56$\pm$0.14 & 3.61$\pm$0.12 & 3.117$\pm$0.097 & 2.190$\pm$0.074\
June 26 & 55738.2 & 25.4 & 6.935$\pm$0.071 & 5.574$\pm$0.080 & 4.744$\pm$0.096 & 3.70$\pm$0.13 & 2.88$\pm$0.11 & 2.349$\pm$0.077 & 1.644$\pm$0.064\
July 6 & 55748.1 & 35.3 & 6.820$\pm$0.071 & 4.334$\pm$0.068 & 3.917$\pm$0.073 & 2.92$\pm$0.11 & 2.53$\pm$0.10 & 2.063$\pm$0.073 & 1.772$\pm$0.074\
July 16 & 55758.1 & 45.3 & 6.082$\pm$0.064 & 3.790$\pm$0.063 & 2.960$\pm$0.067 & 2.493$\pm$0.097 & 1.819$\pm$0.080 & 1.549$\pm$0.053 & 1.159$\pm$0.051\
July 29 & 55771.0 & 58.2 & 5.097$\pm$0.057 & 2.83$\pm$0.14 & 2.32$\pm$0.14 & 2.35$\pm$0.15 & 1.53$\pm$0.16 & 1.32$\pm$0.15 & $< 0.69$\
Sept 1 & 55805.7 & 92.9 & 2.891$\pm$0.043 & 1.627$\pm$0.064 & 1.321$\pm$0.072 & 1.28$\pm$0.15 & 0.71$\pm$0.10 & 0.60$\pm$0.11 & $< 0.42$\
\[srcTab2\]
Probing the distinguishing differences between SNe cIIb and extended Type IIb supernovae requires early discovery, since rapid follow-up observations are crucial for identifying unique characteristics. Since the advent of dedicated transient surveys and improvements in amateur astronomical equipment, such early discoveries are becoming more common. In June 2011, an optical transient was found in M51 by amateur astronomer Amédée Riou [@iaucriou]. Prompt spectroscopic follow-up indicated that it was a Type II supernova, and further spectroscopy revealed that the object most closely resembled a Type IIb [@arcavi11; @marion11]. A YSG was identified in pre-explosion HST imaging at the position of the SN [@maund11; @vandyk11], similar to the case of SN1993J. However, rapid optical follow-up observations from the Palomar Transient Factory [@law] pointed to a compact progenitor star, as evidenced by its short-lived ($\Delta t\approx 1$ day) cooling-envelope emission [@arcavi11]. In our recent paper (@soderberg11; hereafter Paper I) we reported on early radio, mm-band and X-ray emission. Based on modeling of the non-thermal emission, we found that the shockwave velocity was $v\sim 0.1c$, more typical of a SN Ibc (assumed to have a compact Wolf-Rayet progenitor) than the explosion of an extended supergiant. If did, in fact, have a YSG progenitor, it would be necessary to explain the high shockwave velocity and rapid optical evolution in the context of an extended star.
In this paper, we present our detailed Expanded Very Large Array (EVLA; @perley11) observations of spanning $\Delta t\approx 100$ days after explosion. This project capitalizes on the nearly continuous coverage now available from 1–40 GHz with the EVLA. We model the synchrotron emission over 7 epochs to derive the evolution of the shockwave radius and magnetic field as a function of time. We confirm the initial results of Paper I and find that the radio emission over the course of these observations evolves smoothly, with no evidence as yet for the circumstellar density variations seen in other compact SNe IIb.
Observations and Data Reduction {#observations}
===============================
We obtained multi-frequency monitoring observations with the EVLA beginning 17 days after explosion (taking $t_0 =$ 2011 May 31.8 UTC) and continuing through day 92 (Program 11A-277: PI Soderberg). These observations comprise seven epochs with roughly logarithmic spacing, matching the expected evolution of the supernova light curves. The second through final epochs covered 1.0–36.5 GHz, utilizing the L (1–2 GHz), S (2–4 GHz), C (4–8 GHz), X (8–8.8 GHz), Ku (12–18 GHz), K (18–26.5 GHz), and Ka-band (26.5–40 GHz) receivers (see @perley11 for a description of new observing capabilities with the EVLA). Within each observing band (except X-band),[^1] each of two basebands was tuned to a different frequency in order to maximize spectral coverage. We took data with the maximum available 1 GHz of bandwidth per baseband at all bands during the first five epochs, and 128 MHz for the final two. Each epoch was three hours in duration. All observations were performed while the EVLA was in its most extended A-configuration, giving the highest available spatial resolution. EVLA observations were not continued in the subsequent (most compact) D-configuration, due to concerns about confusion with other sources of radio emission in M51 at lower frequencies, to which the spectral peak of had shifted by this time.
Phase reference calibration was carried out using J1335+4542 (1.7 away; 8–40 GHz) or J1327+4326 (3.8 away; 1–8 GHz). At the highest frequencies ($>20$ GHz), rapid switching (2 minute cadence) between and J1335+4542 was done to enable optimal correction for atmospheric phase variations, and referenced pointing was used to correct for antenna pointing offsets. Each observation included data at all bands for the standard flux density calibration source 3C286. Processing was performed with NRAO’s Common Astronomy Software Applications (CASA; @mcmullin07) or Astronomical Image Processing System (AIPS; @greisen03), using the same procedure in each.
Bad data identified by the EVLA online system were deleted, as were pure zeros (sometimes generated by the correlator as a result of failure); further editing out of radio-frequency interference and poorly performing antennas was done by hand. Frequency-dependent atmospheric opacity was accounted for using the average of a seasonal model and observation-specific information from the weather station [@marvil10]. At frequencies higher than 5 GHz, where elevation-dependent antenna gain effects become important ($\gtrsim 1\%$ variance from zenith values), gain curve information was applied at relevant points in the calibration process.
To calibrate the data, a bandpass solution was derived using 3C286; applying this solution, we solved for the complex gains of the calibration sources. We scaled the amplitude gains of the phase calibrator according to the values derived for 3C286 using the “Perley-Butler 2010” flux density standard, and applied these solutions to .[^2] When there was sufficient signal, we performed phase-only self-calibration, and natural weighting was used during the imaging process. We fit an elliptical Gaussian model at each frequency to derive integrated flux density values. To estimate the uncertainties on our measured flux densities, we added (in quadrature) uncertainties from the Gaussian fitting with the rms noise of the images, as well as 1% systematic errors at low frequencies ($<20$ GHz) and 3% at high frequencies ($>20$ GHz). The integrated flux density values and associated 1$\sigma$ uncertainties, as well as mean rms image noises for each frequency, are reported in Tables \[srcTab1\] and \[srcTab2\].
Modeling and Results
====================
Radio emission from supernovae arises when the expanding ejecta interact with pre-existing circumstellar material, which for SN IIb is provided by the progenitor’s stellar wind. The interaction of the supernova blast wave with the circumstellar environment probes recent stellar mass-loss [@chevalier06]. Here, we model the radio emission from using the standard circumstellar interaction model (“model 1” in @chevalier96): as the expanding shock moves into the circumstellar medium, the magnetic field strength in the interaction zone increases via turbulence generated in the shocked region. Electrons that have been accelerated to relativistic energies interact with this enhanced magnetic field, producing synchrotron emission. This emission is subject to self-absorption, as further explored by @chevalier98; we employ this form of the synchrotron self-absorption (SSA) model to provide an analytic description of the observed radio spectra.
SSA Model Fits and Derived Parameters {#analysis}
-------------------------------------
For each given epoch, we fit the radio spectrum using the parameterization
$$S(\nu) = 1.582 \, S_{\nu_{\tau}} \left(\frac{\nu}{\nu_{\tau}}\right)^{5/2} \left\{1-\exp\left[-\left(\frac{\nu}{\nu_{\tau}}\right)^{-(p+4)/2}\right]\right\},$$
where $S_{\nu_{\tau}}$ is the flux density at $\nu_{\tau}$, the frequency at which the optical depth is unity, and $p$ the electron power-law index [@chevalier98]. We do not see any evidence of external free-free absorption, as was found for SN1993J [@weiler07]. Including free-free absorption in our fits did not reduce the resulting $\chi^2$ values; the fitting minimized its effect to an insignificant contribution. This is not surprising, since the spectra often show excess emission relative to the SSA model at low frequencies—the opposite of what is expected if free-free absorption were significant.
Allowing $p$ to vary did not significantly improve the resulting reduced $\chi^2$ values, so we froze this parameter to its average fitted value of $p = 2.8$ to limit the number of free parameters. For $p = 2.8$, the observed peak radio flux $S_{\nu_{\rm op}}$ occurs at $\nu_{\rm op} = 1.17\,\nu_{\tau}$.
Although the SSA model likely represents an over-simplification of the actual source geometry and physics, it provides reasonable fits to the observations (see Figure \[specPlot\]). Despite some systematic deviations, the model matches the peaks in the spectra well; large variations of the true average radii from the ones implied by the models therefore seem unlikely.
However, the formal errors described in §\[observations\] are small relative to deviations from the fits, resulting in large values of $\chi^2$ and unacceptably small uncertainties on the fitted values of $S_{\nu_{\tau}}$ and $\nu_{\tau}$. This is probably due to a combination of underestimated formal errors (the EVLA was still in its commissioning phase at the time of the observations), as well as unaccounted for physics that is beyond the scope of this paper. To better estimate the errors on the fitted parameters, we scaled the fitted uncertainties for each epoch so that the reduced $\chi^2$ values were unity, equivalent to increasing the errors by a factor of 3–7. These uncertainties were propagated throughout subsequent calculations. Values for $S_{\nu_{\tau}}$ and $\nu_{\tau}$, as well as the parameters derived below, are presented in Table \[fitResults\].
We find that the SSA-derived peak frequencies are systematically $\approx 10\%$ lower than the apparent peaks, but since this trend is consistent over the course of the observations, it will not affect the time-dependence of the derived parameters. We also note that for the final three epochs (days 45, 58, and 92), the high-frequency data trend below the model. This is likely due to the fact that for these epochs, there was insufficient flux to perform self-calibration above 20 GHz (K or Ka-bands), resulting in possible underestimation of the flux densities due to phase decoherence.
The fitted spectra can be used as observational tracers of the outer shock radius ($R_s$), strength of the magnetic field ($B_s$), and density of the progenitor’s wind ($\rho_{\rm wind}$), given a minimal set of assumptions [@chevalier98; @chevalier06]:
$$\begin{aligned}
R_s &=& 3.9 \times 10^{14} \, \alpha^{-1/19} \left( \frac{f}{0.5} \right)^{-1/19} \left( \frac{D}{\rm Mpc} \right)^{18/19} \nonumber \\
& & \left( \frac{S_{\nu_{\rm op}}}{\rm mJy} \right)^{9/19} \times \left( \frac{\nu_{\rm op}}{\rm 5~GHz} \right)^{-1} {\rm cm}, \\
B_s &=& 1.0 \, \alpha^{-4/19} \left( \frac{f}{0.5} \right)^{-4/19} \left( \frac{D}{\rm Mpc} \right)^{-4/19} \nonumber \\
& & \times \left( \frac{S_{\nu_{\rm op}}}{\rm mJy} \right)^{-2/19} \left( \frac{\nu_{\rm op}}{\rm 5~GHz} \right) {\rm G},\end{aligned}$$
and
$$\begin{aligned}
\rho_{\rm wind} &=& A r^{-2} {\rm~g~cm^{-3}},\end{aligned}$$
where the circumstellar density is parametrized as $A_{*} = A/(5 \times 10^{11}\rm~g~cm^{-1})$, and
$$\begin{aligned}
A_{*} &=& 0.82 \, \alpha^{-8/19} \left( \frac{\epsilon_B}{0.1} \right)^{-1} \left( \frac{f}{0.5} \right)^{-8/19} \left( \frac{D}{\rm Mpc} \right)^{-8/19} \nonumber \\
& & \times \left( \frac{S_{\nu_{\rm op}}}{\rm mJy} \right)^{-4/19} \left( \frac{\nu_{\rm op}}{\rm 5~GHz} \right)^2 \left( \frac{t}{10\rm~d} \right)^2.\end{aligned}$$
Here, $\alpha$ is the ratio of electron to magnetic energy densites ($u_e / u_B$), $f$ the filling fraction of emitting material, $D$ the distance, $t$ the age, and $\epsilon_B$ the converted fraction of kinetic to magnetic energy density ($\epsilon_B = u_B / \rho_{\rm wind}\,v_s$). We assume equipartition ($\alpha = 1$), and take $p=2.8$ (as fitted). In addition, we assume a filling factor $f = 0.5$ (approximately as was found for SN1993J; @bartel02), and a distance $D = 8.4\pm0.6$ Mpc [@feldmeier97; @vinko11]. The time evolution of the shock radius is consistent with $R_s \propto t^{0.9}$ (see also @bietenholz11), and the magnetic field strength with $B_s \propto t^{-1}$ (top and center panels of Figure \[RpBpPlot\]).
[lccccc]{} 16.4 & 7.03$\pm$0.25 & 10.01$\pm$0.29 & 3.2$\pm$0.4 & 1.21$\pm$0.06 & 3.2$\pm$0.3\
20.4 & 6.19$\pm$0.15 & 8.13$\pm$0.18 & 3.7$\pm$0.4 & 1.00$\pm$0.04 & 3.4$\pm$0.3\
25.4 & 6.52$\pm$0.22 & 6.18$\pm$0.21 & 5.0$\pm$0.6 & 0.76$\pm$0.04 & 3.0$\pm$0.3\
35.3 & 7.28$\pm$0.15 & 4.72$\pm$0.10 & 6.9$\pm$0.7 & 0.57$\pm$0.02 & 3.3$\pm$0.2\
45.3 & 7.36$\pm$0.19 & 3.91$\pm$0.11 & 8.4$\pm$0.9 & 0.47$\pm$0.02 & 3.8$\pm$0.3\
58.2 & 7.69$\pm$0.20 & 3.184$\pm$0.088 & 11$\pm$1 & 0.38$\pm$0.02 & 4.1$\pm$0.4\
92.9 & 6.44$\pm$0.21 & 2.235$\pm$0.076 & 14$\pm$2 & 0.27$\pm$0.01 & 5.3$\pm$0.6\
\[fitResults\]
The expectation of $\rho_{\rm wind} \propto r^{-2}$ may be questionable at large radii, but it is reasonable to approximate the immediate circumstellar environment by assuming a constant progenitor wind [@dwarkadas11]. These observations probe a region extending to $\sim 1000$ AU, corresponding to $\sim 5$ yr for a $1000~\rm km~s^{-1}$ wind — much shorter than necessary for substantial wind variability. Given these constraints, and taking $\epsilon_B = 0.1$, we find $A_{*} \approx 3.5$. There is no strong evidence for time variability, suggesting that our assumption of a constant progenitor wind was reasonable (bottom panel of Figure \[RpBpPlot\]).
In Paper I, joint radio and X-ray model fits pointed to deviations from equipartition, $\alpha \approx 30$ and $\epsilon_B \approx 0.01$, under the assumption that inverse Compton emission dominates the X-ray signal. With these values, our radius estimates are smaller by a factor of 0.8, magnetic field values smaller by a factor of 0.5, and $A_*$ larger by a factor of 2.
Physical Interpretation {#discussion}
-----------------------
The derived outer shock radii imply an average shock velocity of $dR_s/dt \approx 25,000$ km s$^{-1}$, or $\approx 0.1c$, in agreement with Paper I. As noted there, the shockwave is traveling a factor of $\sim1.5$ faster than material at the optical photosphere ($\approx 17,000$ km s$^{-1}$ at $\Delta t \approx 3$ days; @silverman11).
Assuming that the supernova ejecta and progenitor wind have power-law density profiles, the time evolution of the shock radius can be expressed as $R_s \propto t^m$. For the expected circumstellar density $\rho_{\rm wind} \propto r^{-2}$, $m = (n-3)/(n-2)$, where $n$ is the power-law index of the outer supernova density profile. We find $m = 0.87\pm0.07$, corresponding to $n = 9.7^{+12}_{-3.7}$, and reasonable for a fast blastwave from a compact progenitor (canonical value of $m = 0.9$; @chevalier92). This value of $m$ is also consistent with joint EVLA-VLBI fits reported in our companion paper [@bietenholz11].
The measured decrease in magnetic field strength agrees with the standard model for the hydrodynamic evolution of a self-similar shock. The magnetic field generation is thought to arise via turbulence in the shocked region, so it is proportional to the total post-shock energy density ($\propto t^{-2}$; @chevalier98); therefore, $B_s \propto t^{-1}$, as observed.
We find that the scaling factor for the circumstellar density is consistent with a constant value, $A_{*} \approx 3.5$, over the course of our observations. This implies a constant mass-loss rate of $\approx 3 \times 10^{-5}\rm~M_{\sun}~yr^{-1}$, assuming a wind velocity of $1000~\rm km~s^{-1}$. These values are in the expected range for a Wolf-Rayet progenitor [@crowther07], and agree with our analysis of the early-time radio data (Paper I). Since the implied mass-loss rate scales linearly with wind velocity ($\dot{M} \propto A \, v_{\rm wind}$), a wind velocity of $20\rm~km~s^{-1}$ gives a mass-loss rate of $\approx 7 \times 10^{-7}\rm~M_{\sun}~yr^{-1}$, reasonable for a YSG progenitor [@georgy11].
However, the high shock velocity, as well as the rapid cooling observed in early-time optical spectra, suggest a compact progenitor star for , and a Type cIIb classification (@arcavi11, Paper I). Furthermore, the fitted electron power-law index of $p=2.8$ is close to what is typically found for observations of SNe Ibc [$p \approx 3$; @chevalier06], which are presumed to have compact progenitors.
One further observational characteristic of Type cIIb SNe is that they often display late-time radio variability, as was seen in SN2001ig and SN2003bg [@ryder04; @soderberg06]. We do not yet see evidence of variability in , but note that our observations only cover the first $\sim 100$ days of evolution, around the time that variability was discovered in other SNe cIIb.
Conclusions
===========
The radio spectra of are well-characterized by an SSA model without any need for additional free-free absorption. This is in contrast with SN1993J, which required both, and also had a substantially higher derived mass-loss rate ($A_* \approx 500$, compared with $\approx 4$ for ; @fransson98). SN1993J was classified as a SN eIIb because of its extended progenitor [@chevalier10; @woosley94]: the observed differences in absorption and circumstellar density could be characteristic of these two classes of IIb SNe.
However, pre-supernova images of M51 show a YSG co-located with the explosion site, which was suggested as a potential progenitor or binary companion [@maund11; @vandyk11; @murphy11]. As we have mentioned, it seems unlikely that the YSG is in fact the progenitor. If it were, this would require a process to enable the escape of fast shockwaves from YSGs, either through steep ejecta density profiles or ejecta asymmetries.
If the YSG was a binary companion, then some interaction of the shock and the YSG might be expected. The observational agreement of the radio measurements with the standard model suggests that this did not occur within $\sim1000$ AU of the explosion site. In this scenario, then, the orbit must have been quite wide, with an orbital period 6000 yr, so any interaction between the YSG and the supernova progenitor would have been limited. For comparison, the binary progenitor of SN1993J likely had an orbital period of $\sim 2000$ d [@stancliffe09], allowing substantial mass transfer to occur and stripping the presupernova star of its H-rich envelope [@woosley94]. This would not have been possible for . Alternatively, if the blast wave were highly asymmetric, interaction with a more nearby companion could have been minimized. We consider this unlikely, since the radio data agree quite well with the standard, spherical ejecta model. Furthermore, the YSG phase is estimated to last only $\sim 3000$ yr [@drout09], making it improbable that the companion would happen to be a YSG at the time of the supernova.
Finally, it may be that the YSG is unrelated to the supernova, and is only coincidentally along the same line of sight. Future observations, including optical imaging of the field after SN 2011dh has faded, will help determine any association with the YSG and the true nature of the progenitor system.
[43]{} natexlab\#1[\#1]{}
, I. [et al.]{} 2011, , 742, L18
, N. [et al.]{} 2002, , 581, 404
, M. F., [Brunthaler]{}, A., [Soderberg]{}, A. M., [Krauss]{}, M., [Bartel]{}, N., [Chomiuk]{}, L., & [Rupen]{}, M. P. 2012, ArXiv e-prints, 1201.0771
, R. A. 1992, , 394, 599
, R. A. 1996, in Astronomical Society of the Pacific Conference Series, Vol. 93, Radio Emission from the Stars and the Sun, ed. [A. R. Taylor & J. M. Paredes]{}, 125
——. 1998, , 499, 810
, R. A., & [Fransson]{}, C. 2006, , 651, 381
, R. A., & [Soderberg]{}, A. M. 2010, , 711, L40
, P. A. 2007, , 45, 177
, M. R., [Massey]{}, P., [Meynet]{}, G., [Tokarz]{}, S., & [Caldwell]{}, N. 2009, , 703, 441
, V. V., & [Gruszko]{}, J. 2012, , 419, 1515
, N. [et al.]{} 2010, , 714, L254
——. 2009, , 706, 1174
, J. J., [Ciardullo]{}, R., & [Jacoby]{}, G. H. 1997, , 479, 231
, A. V. 1997, , 35, 309
, C., & [Bj[ö]{}rnsson]{}, C.-I. 1998, , 509, 861
, C. 2012, , 538, L8
, E. W. 2003, Information Handling in Astronomy - Historical Vistas, 285, 109
, T. [et al.]{} 2011, Central Bureau Electronic Telegrams, 2736, 1
, M. [et al.]{} 2009, , 703, 1612
, N. M. [et al.]{} 2009, , 121, 1395
2012, in preparation
, J. 2010, EVLA Memo 143: Improving the frequency resolution of the default atmospheric opacity model
, T., [Filippenko]{}, A. V., [Li]{}, W., [Leonard]{}, D. C., & [Shields]{}, J. C. 2001, , 121, 1648
, J. R. [et al.]{} 2011, , 739, L37
, J. R., & [Smartt]{}, S. J. 2009, Science, 324, 486
, J. R., [Wheeler]{}, J. C., [Patat]{}, F., [Wang]{}, L., [Baade]{}, D., & [H[ö]{}flich]{}, P. A. 2007, , 671, 1944
, P. A., [Deng]{}, J., [Hamuy]{}, M., & [Nomoto]{}, K. 2009, , 703, 1624
, J. P., [Waters]{}, B., [Schiebel]{}, D., [Young]{}, W., & [Golap]{}, K. 2007, in Astronomical Society of the Pacific Conference Series, Vol. 376, Astronomical Data Analysis Software and Systems XVI, ed. [R. A. Shaw, F. Hill, & D. J. Bell]{}, 127
, G., & [Maeder]{}, A. 2005, , 429, 581
, J. W., [Jennings]{}, Z. G., [Williams]{}, B., [Dalcanton]{}, J. J., & [Dolphin]{}, A. E. 2011, , 742, L4
, E., & [Sari]{}, R. 2010, , 725, 904
, R. A., [Chandler]{}, C. J., [Butler]{}, B. J., & [Wrobel]{}, J. M. 2011, , 739, L1
, S. D., [Murrowood]{}, C. E., & [Stathakis]{}, R. A. 2006, , 369, L32
, S. D., [Sadler]{}, E. M., [Subrahmanyan]{}, R., [Weiler]{}, K. W., [Panagia]{}, N., & [Stockdale]{}, C. 2004, , 349, 1093
, J. M., [Filippenko]{}, A. V., & [Cenko]{}, S. B. 2011, The Astronomer’s Telegram, 3398, 1
, A. M., [Chevalier]{}, R. A., [Kulkarni]{}, S. R., & [Frail]{}, D. A. 2006, , 651, 1005
, A. M. [et al.]{} 2011, ArXiv e-prints, 1107.1876
, R. J., & [Eldridge]{}, J. J. 2009, , 396, 1699
, S. D. [et al.]{} 2011, , 741, L28
, J. [et al.]{} 2011, ArXiv e-prints, 1111.0596
, K. W., [Williams]{}, C. L., [Panagia]{}, N., [Stockdale]{}, C. J., [Kelley]{}, M. T., [Sramek]{}, R. A., [Van Dyk]{}, S. D., & [Marcaide]{}, J. M. 2007, , 671, 1959
, S. E., [Eastman]{}, R. G., [Weaver]{}, T. A., & [Pinto]{}, P. A. 1994, , 429, 300
[^1]: Very few wideband X-band receivers were available at the time of our observations. The total bandwidth at X-band was 0.8 GHz and 256 MHz.
[^2]: Due to a problem with data acquisition, the flux density calibrator was not present at Ku-band for epoch 3. We characterized the variability of J1335+4542 over the other epochs and found it to be small (rms of approximately 1% at both 13.5 and 16.0 GHz). We fit its average spectrum and used this model to flux-calibrate the data for , adding the additional source of error from J1335+4542’s variability in quadrature with the image noise and fit errors to determine the total uncertainty on flux density.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
It is well known that for an arbitrary $n$-tuple of commuting contractions, $n\geq 3$, neither the existence of isometric dilation nor the von-Neumann inequality holds. In this paper we provide an explicit isometric dilation for a large class of $n$-tuple $(n\geq
3)$ of commuting contractions. The present class of tuples of operators is motivated by a polydisc version of commutant lifting theorem by Ball, Li, Timotin and Trent. The present class of operators is larger than the one considered in [@BDHS]. Also we prove a sharper von-Neumann inequality on an algebraic variety in the closure of the polydisc in $\mathbb{C}^n$.
address:
- 'Department of Mathematics, Indian Institute of Technology Bombay, Powai, Mumbai, 400076, India'
- 'Department of Mathematics, Indian Institute of Technology Bombay, Powai, Mumbai, 400076, India'
- 'Indian Statistical Institute, Statistics and Mathematics Unit, 8th Mile, Mysore Road, Bangalore, 560059, India'
author:
- Sibaprasad Barik
- 'B. Krishna Das'
- Jaydeb Sarkar
title: Isometric dilations and von Neumann inequality for finite rank commuting contractions
---
Introduction
============
This paper concerns contractive and isometric dilations, commutant lifting theorem and von Neumann inequality for commuting $n$-tuples of contractions, $n \geq 3$, on Hilbert spaces. We denote by ${\mathcal{T}}^n({\mathcal{H}})$ the set of all ordered $n$-tuples of commuting contractions on a Hilbert space ${\mathcal{H}}$: $${\mathcal{T}}^n({\mathcal{H}}) = \{(T_1, \ldots, T_n): T_i \in {\mathcal{B}}({\mathcal{H}}), \|T_i\| \leq
1, T_i T_j = T_j T_i, 1 \leq i, j \leq n\},$$ where ${\mathcal{B}}({\mathcal{H}})$ denotes the set of all bounded linear operators on ${\mathcal{H}}$.
The existence of isometric dilations, commutant lifting theorem and von Neumann inequality for a single contraction on a Hilbert space are useful tools for analyzing the structure of bounded linear operators on Hilbert spaces and related function theoretic problems like (Toeplitz) corona theorem, Nevanlinna-Pick interpolation theorem, Caratheodory-Fejer theorem and invariant subspace problem. On the other hand, Ando’s powerful isometric dilation theorem and the von Neumann inequality for elements in ${\mathcal{T}}^2({\mathcal{H}})$ provides deep insight into the complicated structure of commuting pairs of bounded linear operators and similar function theoretic problems in $\mathbb{C}^2$. However, neither the existence of isometric dilations nor the von Neumann inequality holds in general when $n >
2$. For instance, see the counterexamples by Varopoulous [@V] and Crabb and Davie [@CD1] (the von Neumann inequality does not hold in general for tuples in ${\mathcal{T}}^3({\mathcal{H}})$), and Parrott’s example [@Par] of a tuple $T \in {\mathcal{T}}^3({\mathcal{H}})$ such that $T$ does satisfy the von Neumann inequality but does not dilate to three commuting isometries. Also see Choi and Davidson [@CD], Drury [@Dr], Holbrook [@Ho1; @Ho2], Knese [@K1] and Kosiński [@LK] for relevant examples and results.
An intriguing question therefore is to characterize $n$ tuples $T$ in ${\mathcal{T}}^n({\mathcal{H}})$, $n \geq 3$, such that $T$ admits an isometric dilation (and satisfy the von Neumann inequality over a variety in $\overline{{\mathbb{D}}}^n$ or ${\mathbb{D}}^n$ in the sense of [@AM1], [@DS], [@DSS] and [@BDHS]). This has turned into one of the most challenging questions in multivariable operator theory and functions of several complex variables. However, the research in this direction seems unexplored except the work of Grinshpan, Kaliuzhnyi-Verbovetskyi, Vinnikov and Woerdeman [@VV] and the recent paper [@BDHS] (and also see Ball, Li, Timotin and Trent in [@BLTT]).
The purpose of this paper is to study a class of $n$-tuples of commuting contractions ${\mathcal{P}}_n({\mathcal{H}})$ $(n\geqslant3)$ (see subsection \[sub-pnh\]) admitting (isometric) dilations to the Hardy space over ${\mathbb{D}}^n$. The existence of isometric dilations for tuples in ${\mathcal{P}}_n({\mathcal{H}})$ is known by the commutant lifting theorem obtained by Ball, Li, Timotin and Trent in [@BLTT]. Here we first prove an explicit isometric dilation for tuples in ${\mathcal{P}}_n({\mathcal{H}})$. This is achieved via an explicit commutant lifting theorem which is similar to that of [@BLTT]. However, the core of our approach uses a technique of McCullough and Sultanic’s [@MS] commutant lifting on an annulus. The class of commuting contractions considered in this article is larger than the one considered in [@BDHS]. Also as an application, we obtain a version of sharper von Neumann inequality, in terms of algebraic varieties in $\bar{{\mathbb{D}}}^n$, for finite rank tuples in ${\mathcal{P}}_n({\mathcal{H}})$.
The rest of the paper is organized as follows. In the next section we set up notation and terminology. In Section 3, we consider the dilation of finite rank commuting operator tuples in the class ${\mathcal{P}}_n({\mathcal{H}})$. This is obtained by first proving a general commuting lifting theorem. A sharp von Neumann inequality on varieties is considered in Section 4.
Preliminaries
=============
In this section, we set up some of the needed background and terminology on $n$-tuples of commuting contractions. We follow the same notation as in [@BDHS]. We set $$\mathbb{Z}_+^n = \{\bm{k} = (k_1, \ldots, k_n) : k_i \in
\mathbb{Z}_+, i = 1, \ldots, n\},$$ and for each multi-index $\bm{k} \in \mathbb{Z}_+^n$, commuting tuple $T = (T_1, \ldots, T_n)$ on a Hilbert space ${\mathcal{H}}$, and ${\bm{z}}\in \mathbb{C}^n$ we denote $$T^{\bm{k}} = T_1^{k_1} \cdots T_n^{k_n},$$ and $${\bm{z}}^{\bm{k}} = z_1^{k_1} \cdots z_n^{k_n}.$$
Dilations {#ss-dilation}
---------
Let ${\mathcal{H}}$ and ${\mathcal{K}}$ be Hilbert spaces, and let $T \in {\mathcal{T}}^n({\mathcal{H}})$ and $V \in {\mathcal{T}}^n ({\mathcal{K}})$. Following usual convention, we say that $V$ is a *dilation* of $T$ if there exists an isometry $\Pi: {\mathcal{H}}{\rightarrow}{\mathcal{K}}$ such that $$\Pi T^{\bm{* k}} = V^{\bm{* k}} \Pi,$$ for all $\bm{k} \in \mathbb{Z}_+^n $. We also say that $T$ *dilates* to $V$. If, in addition, $V$ is a commuting tuple of isometries, then we say that $V$ is an *isometric dilation* of $T$.
If $V \in {\mathcal{T}}^n ({\mathcal{K}})$ is a dilation of $T \in {\mathcal{T}}^n({\mathcal{H}})$ with the isometry $\Pi : {\mathcal{H}}{\rightarrow}{\mathcal{K}}$ as above, and $${\mathcal{Q}}= \textit{ran~} \Pi,$$ then it follows that (cf. [@BDHS]) ${\mathcal{Q}}$ is a joint invariant subspace for $(V_1^*, \ldots, V_n^*)$, $(T_1, \ldots, T_n)$ on ${\mathcal{H}}$ and $(P_{{\mathcal{Q}}} V_1|_{{\mathcal{Q}}}, \ldots, P_{{\mathcal{Q}}} V_n|_{{\mathcal{Q}}})$ on ${\mathcal{Q}}$ are unitarily equivalent (via the unitary $\Pi : {\mathcal{H}}{\rightarrow}{\mathcal{Q}}$), and $$(P_{{\mathcal{Q}}} V|_{{\mathcal{Q}}})^{* \bm{k}} = V^{* \bm{k}}|_{{\mathcal{Q}}},$$ for all $\bm{k} \in {\mathbb{Z}_+}^n$. Here $P_{{\mathcal{Q}}}$ denotes the orthogonal projection of ${\mathcal{K}}$ onto ${\mathcal{Q}}$.
This therefore implies that the tuple $T$ has a power dilation (in the sense of Sz.-Nagy and Foias and Halmos ([@NF])) to the tuple $V$.
Szegö tuples
------------
The Hardy space $H^2({\mathbb{D}}^n)$ over ${\mathbb{D}}^n$ is the Hilbert space of all analytic functions $f = \sum_{\bm{k} \in {\mathbb{Z}_+}^n} a_{\bm{k}}
z^{\bm{k}}$ on ${\mathbb{D}}^n$ such that $$\|f\| = (\sum_{\bm{k} \in {\mathbb{Z}_+}^n} |a_{\bm{k}}|^2)^{\frac{1}{2}} < \infty.$$ Moreover, for a Hilbert space ${\mathcal{E}}$, the ${\mathcal{E}}$-valued Hardy space on ${\mathbb{D}}^n$ is denoted by $H^2_{{\mathcal{E}}}({\mathbb{D}}^n)$. We will also identify $H^2_{{\mathcal{E}}}({\mathbb{D}}^n)$ with $H^2({\mathbb{D}}^n) \otimes {\mathcal{E}}$ via the unitary map $\bm{z}^{\bm{k}} \eta \mapsto \bm{z}^{\bm{k}}\otimes \eta$ for all $\bm{k} \in {\mathbb{Z}_+}^n$ and $\eta \in {\mathcal{E}}$. It is well known that the ${\mathcal{B}}({\mathcal{E}})$-valued function $$({\bm{z}}, {\bm{w}}) \in {\mathbb{D}}^n \times {\mathbb{D}}^n {\rightarrow}\mathbb{S}_n({\bm{z}}, {\bm{w}}) I_{{\mathcal{E}}}.$$ is a reproducing kernel for $H^2_{{\mathcal{E}}}({\mathbb{D}}^n)$, where $$\mathbb{S}_n({\bm{z}},{\bm{w}})= \prod_{i=1}^n(1-z_i\bar{w_i})^{-1} \quad \quad
({\bm{z}}, {\bm{w}}\in {\mathbb{D}}^n),$$ is the Szegö kernel on ${\mathbb{D}}^n$. Let $(M_{z_1}, \ldots, M_{z_n})$ denote the $n$-tuple of multiplication operators on $H^2_{{\mathcal{E}}}({\mathbb{D}}^n)$ defined by $$(M_{z_i} f)({\bm{w}}) = w_i f({\bm{w}}),$$ for all $f \in H^2_{{\mathcal{E}}}({\mathbb{D}}^n)$, ${\bm{w}}\in {\mathbb{D}}^n$ and $i = 1, \ldots,
n$. Let $$\mathbb{S}_n({\mathcal{H}}) = \{ T \in {\mathcal{T}}_n({\mathcal{H}}) : \mathbb{S}_n^{-1}(T, T^*)
\geq 0\},$$ where $$\mathbb{S}_n^{-1}(T, T^*) = \sum_{\bm{k} \in \{0,1\}^n}
(-1)^{|\bm{k}|} T^{\bm{k}} T^{*\bm{k}}.$$ The elements of $\mathbb{S}_n({\mathcal{H}})$ are called *Szegö $n$-tuples* (cf. [@BDHS]).
The following theorem (see [@CV1], [@CV2] and [@MV]) says that a pure Szegö $n$-tuple admits an isometric dilation to a (canonical) vector-valued Hardy space over ${\mathbb{D}}^n$. Recall that an $n$-tuple $T \in {\mathcal{T}}^n({\mathcal{H}})$ is said to be *pure* if $$\|T^{*m}_i h\| {\rightarrow}0\quad\text{as}~ m{\rightarrow}\infty,$$ for all $h \in {\mathcal{H}}$ and $i = 1, \ldots, n$.
\[th-dc dil\] Let $T \in \mathbb{S}_n({\mathcal{H}})$ be a pure tuple. Suppose that ${D_{T}}=
\mathbb{S}_n^{-1}(T, T^*)^{1/2}$ and ${{\mathcal{D}}_{T}}=\overline{\text{ran}}~{D_{T}}$. Then $\Pi: {\mathcal{H}}\to H^2_{{\mathcal{D}}_T}({\mathbb{D}}^n)$ defined by $$(\Pi h)({\bm{z}}) = \sum_{\bm{k} \in {\mathbb{Z}_+}^n} {\bm{z}}^{\bm{k}} D_T
T^{*\bm{k}} h \quad \quad (h \in {\mathcal{H}}, {\bm{z}}\in {\mathbb{D}}^n),$$ is an isometry and $\Pi T_i^* = M_{z_i}^*\Pi$ for all $i = 1,
\ldots, n$.
The isometry $\Pi$ defined above is called the *canonical isometry*.
Commuting tuples in ${\mathcal{P}}_n({\mathcal{H}})$ {#sub-pnh}
----------------------------------------------------
This class of $n$-tuples of operators, $n \geq 3$, is motivated by the commutant lifting theorem on the polydisc by Ball, Li, Timotin and Trent [@BLTT]. We introduce first some notations. Let $A \in
{\mathcal{B}}({\mathcal{H}})$. The conjugacy operator corresponding to $A$ is the completely positive map $C_A : {\mathcal{B}}({\mathcal{H}}) {\rightarrow}{\mathcal{B}}({\mathcal{H}})$ defined by $$C_A(X) = A X A^* \quad \quad (X \in {\mathcal{B}}({\mathcal{H}})).$$ Also for $T \in {\mathcal{T}}^n({\mathcal{H}})$ and $1 \leq i \leq n$, define $$\hat{T}_{i} = (T_1,\dots,T_{i-1}, T_{i+1},\dots,T_{n})
\in {\mathcal{T}}^{(n-1)}({\mathcal{H}}),$$ the $(n-1)$-tuple obtained from $T$ by deleting $T_i$. An $n$-tuple $T \in
{\mathcal{T}}^n({\mathcal{H}})$ is said to be in ${\mathcal{P}}_n({\mathcal{H}})$ if ${\hat{T}}_n \in
\mathbb{S}_{n-1}({\mathcal{H}})$, ${\hat{T}}_n$ is a pure $(n-1)$-tuple, and if there exist $n-1$ positive operators $G_1, \ldots, G_{n-1}$ (depending on $T$) in ${\mathcal{B}}({\mathcal{H}})$ such that $$I-T_nT_n^* = G_1 + \dots+G_{n-1},$$ and $$\Big(\prod_{\stackrel{j=1}{j\ne i}}^{n-1}(I_{\mathcal{B}({\mathcal{H}})}-
C_{T_j})\Big)(G_i) \ge 0,$$ for all $i = 1,\ldots,n-1$. Note that the above product is induced by the composition of linear maps $\{I_{\mathcal{B}({\mathcal{H}})}-
C_{T_l}\}_{l=1}^n$. An easy calculation yields $$\Big(\prod_{\stackrel{j=1}{j\ne i}}^{n-1}(I_{\mathcal{B}({\mathcal{H}})}-
C_{T_j})\Big)(G_i) = \sum_{\bm{k} \in {\mathbb{Z}_+}^{n-2}} (-1)^{|\bm{k}|}
{\hat{T}}_{i,n}^{\bm{k}} G_i {\hat{T}}_{i,n}^{*\bm{k}},$$ where ${\hat{T}}_{i,n}=(T_1,\dots,T_{i-1}, T_{i+1},\dots,T_{n-1})
\in {\mathcal{T}}^{(n-2)}({\mathcal{H}})$ for $1\leqslant i\leqslant n-1$. In particular, if $n = 4$, then $$\prod_{\stackrel{j=1}{j\ne 3}}^{3}(I_{\mathcal{B}({\mathcal{H}})}-
C_{T_j})(G_3) = G_3 - T_1 G_3 T_1^* - T_2 G_3 T_2^* + T_1 T_2 G_3
T_1^* T_2^*.$$ As a concrete example, let us consider a pure pair of commuting contractions $(T_1,T_2) \in \mathbb{S}_2({\mathcal{H}})$ and set $T_3 = T_1^j
T_2^k$ for some $j, k \geq 1$. Then $T = (T_1, T_2, T_3) \in
{\mathcal{P}}_3({\mathcal{H}})$. Indeed, choose $$G_1 = I-T_1^j T_1^{*j},$$ and $$G_2 = T_1^j (I - T_2^k T_2^{*k}) T_1^{*j}.$$ This readily implies that $$I - T_3 T_3^* = G_1 + G_2.$$ On the other hand, by Szegö positivity of $(T_1, T_2)$, we have $$T_2(I-T_1T_1^*)T_2^*\le I-T_1T_1^*,$$ from which it follows that $$\begin{split}
T_2 G_1T_2^* & = T_2 (\sum _{i=0}^{j-1}T_1^i(I-T_1T_1^*)T_1^{*i})
T_2^*
\\
&=\sum_{i=0}^{j-1}T_1^{i}T_2(I-T_1T_1^*)T_2^*T_1^{* i}
\\
&\le \sum_{i=0}^{j-1}T_1^{i}(I-T_1T_1^*)T_1^{* i}
\\
&= I-T_1^jT_1^{*j},
\end{split}$$ that is $$G_1 - T_2 G_1 T_2^* \geq 0.$$ Similarly, $T_1(I-T_2T_2^*)T_1^*\le I- T_2T_2^*$ implies that $$G_2 - T_1 G_2 T_1^* \geq 0,$$ and hence the claim follows.
Transfer functions {#sub-trans}
------------------
Now we recall the definition and some basics of transfer functions (also known as functions in the Schur-Agler class) and refer the reader to [@AEM] and [@BLTT] for further details. Let ${\mathcal{H}}_i$, $i = 1, \ldots, n$, be Hilbert spaces, and let ${\bm{z}}\in
{\mathbb{D}}^n$. Let $${\mathcal{H}}= \mathop{\oplus}_{i=1}^n {\mathcal{H}}_i,$$ and let $E({\bm{z}})$ denote the diagonal operator $$E({\bm{z}}) = \mathop{\oplus}^n_{i=1} z_iI_{{\mathcal{H}}_i},$$ on ${\mathcal{H}}$. Let ${\mathcal{E}}$ and ${\mathcal{E}}_*$ be Hilbert spaces, and let $U:
{\mathcal{E}}\oplus {\mathcal{H}}\rightarrow {\mathcal{E}}_* \oplus {\mathcal{H}}$ be a unitary operator. Assume that $$U=\begin{bmatrix}A &B\\ C& D\end{bmatrix} : {\mathcal{E}}\oplus {\mathcal{H}}{\rightarrow}{\mathcal{E}}_* \oplus {\mathcal{H}}.$$ Then the *transfer function* $\tau_{U}$ corresponding to $U$ is defined by $$\tau_{U}({\bm{z}}) = A + B E({\bm{z}}) (I_{{\mathcal{H}}}- D E({\bm{z}}))^{-1}C,$$ for all ${\bm{z}}\in{\mathbb{D}}^n$. Since $\|D\| \leq 1$, and so $\|D E({\bm{z}}) \| <
1$ for all ${\bm{z}}\in {\mathbb{D}}^n$, it follows that $\tau_U$ is a $\mathcal{B}({\mathcal{E}}, {\mathcal{E}}_*)$-valued analytic function on ${\mathbb{D}}^n$. Moreover, a standard and well-known computation (cf. [@AEM], [@BLTT]) yields that $$\label{identity}
I - \tau_U({\bm{z}})^* \tau_{U}({\bm{z}}) = C^* (I_{{\mathcal{H}}} - E({\bm{z}})^* D^*)^{-1}
(I_{{\mathcal{H}}}- E({\bm{z}})^* E({\bm{z}})) (I_{{\mathcal{H}}} - D E({\bm{z}}))^{-1}C,$$ for all ${\bm{z}}\in {\mathbb{D}}^n$. We conclude that $\tau_U$ is a contractive multiplier, that is, $\tau_U \in H^\infty_{{\mathcal{B}}({\mathcal{E}},
{\mathcal{E}}_*)}({\mathbb{D}}^n)$ (the set of all bounded ${\mathcal{B}}({\mathcal{E}}, {\mathcal{E}}_*)$-valued analytic functions on ${\mathbb{D}}^n$) and $\|M_{\tau_U}\| \leq 1$, where $M_{\tau_U} : H^2_{{\mathcal{E}}}({\mathbb{D}}^n) {\rightarrow}H^2_{{\mathcal{E}}_*}({\mathbb{D}}^n)$ and $$M_{\tau_U} f = \tau_U f \quad \quad (f \in H^2_{{\mathcal{E}}}({\mathbb{D}}^n)).$$ Now let $$\mathcal{S}_n({\mathcal{E}}, {\mathcal{E}}_*) = \{ \Phi \in H^\infty_{{\mathcal{B}}({\mathcal{E}},
{\mathcal{E}}_*)}({\mathbb{D}}^n) : \|\Phi(T)\| \leq 1, T \in {\mathcal{T}}^n_1 ({\mathcal{H}})
~\text{and}~{\mathcal{H}}\text{~a Hilbert space}\},$$ where ${\mathcal{T}}^n_1({\mathcal{H}}) = \{T \in {\mathcal{T}}^n({\mathcal{H}}) : \|T_i \| < 1, i = 1,
\ldots, n\}$. For more details on $\mathcal{S}_n({\mathcal{E}}, {\mathcal{E}}_*)$ and the notations used above, we refer the reader to [@AEM] and [@BLTT]. In other words, ${\mathcal{S}}_n({\mathcal{E}}, {\mathcal{E}}_*)$ is the set of all bounded ${\mathcal{B}}({\mathcal{E}}, {\mathcal{E}}_*)$-valued analytic functions on ${\mathbb{D}}^n$ such that the von-Neumann inequality holds for all $n$-tuples of commuting strict contractions on Hilbert spaces. The celebrated realization theorem of Agler states the following: $\Phi \in
{\mathcal{S}}_n({\mathcal{E}}, {\mathcal{E}}_*)$ if and only if there exist Hilbert spaces $\{{\mathcal{H}}_i\}_{i=1}^n$ and a unitary operator $$U=\begin{bmatrix}A &B\\ C& D\end{bmatrix} : {\mathcal{E}}\oplus
(\mathop{\oplus}_{i=1}^n {\mathcal{H}}_i) {\rightarrow}{\mathcal{E}}_* \oplus
(\mathop{\oplus}_{i=1}^n {\mathcal{H}}_i),$$ such that $\Phi = \tau_U$.
We conclude this section by proving a useful lemma.
\[transfer\] Let ${\mathcal{H}}_1, \ldots, {\mathcal{H}}_n$ and ${\mathcal{E}}$ be finite dimensional Hilbert spaces, and let ${\mathcal{H}}=\oplus^n_{i=1}{\mathcal{H}}_i$. Suppose that $$U=\begin{bmatrix}A &B\\ C& D\end{bmatrix},$$ is a unitary matrix on ${\mathcal{E}}\oplus {\mathcal{H}}$ and let $$\tau_{U}({\bm{z}}) = A + B E({\bm{z}}) (I_{{\mathcal{H}}}-D E({\bm{z}}))^{-1}C \quad \quad ({\bm{z}}\in {\mathbb{D}}^n).$$ Then $\tau_{U}$ is unitary valued almost everywhere on $\mathbb{T}^n$. Moreover, if $A \in {\mathcal{B}}({\mathcal{E}})$ is completely non-unitary then for all ${\bm{z}}\in {\mathbb{D}}^n$, $\tau_{U}({\bm{z}})$ does not have unimodular eigenvalues.
First note that the determinat of $(I_{{\mathcal{H}}}-D E({\bm{z}}))$ is a non-zero rational function in $H^{\infty}({\mathbb{D}}^n)$. Therefore, $\tau_{U}$ cannot vanish on a set of positive measure. Hence $(I-D E({\bm{z}}))^{-1}$ exists for almost all ${\bm{z}}\in \mathbb{T}^n$, and therefore the proof of the first part now follows from .
For the second part, assume by contradiction that there exist ${\bm{z}}\in {\mathbb{D}}^n$, $\lambda \in \mathbb{T}$ and a non-zero vector $v \in
{\mathcal{E}}$ such that $$\tau_U({\bm{z}}) v = \lambda v.$$ Since $\tau_U({\bm{z}})$ is a contraction, it follows (in general) that $$\begin{split}
\|\tau_U({\bm{z}})^* v -\bar{\lambda} v\|^2 & = \|\tau_U({\bm{z}})^* v\|^2 +
\|v\|^2 - \lambda \langle \tau_U({\bm{z}})^* v, v\rangle - \bar{\lambda}
\langle v, \tau_U({\bm{z}})^* v \rangle
\\
& = \|\tau_U({\bm{z}})^* v\|^2 - \|v\|^2
\\
& \leq \|\tau_U({\bm{z}})\| \| v\|^2 - \|v\|^2
\\
& \leq 0,
\end{split}$$ that is $$\tau_U({\bm{z}})^* v = \bar{\lambda} v.$$ This yields $$(I_{{\mathcal{E}}} - \tau_U({\bm{z}})^* \tau_{U}({\bm{z}})) v = 0,$$ and consequently, the right side of implies that $$C v = 0.$$ We see immediately that $$\begin{split}
A v & = \tau_U({\bm{z}}) v
\\
& = \lambda v,
\end{split}$$ and so $$A^* v = \bar{\lambda} v.$$ This implies that $A$ has a non-trivial unitary part, a contradiction. Thus the lemma follows.
Dilations for tuples in ${\mathcal{P}}_n({\mathcal{H}})$
========================================================
\[2\]
The main purpose of this section is to provide an explicit dilation theory for finite rank tuples in ${\mathcal{P}}_n({\mathcal{H}})$. Our dilation theorem is a consequence of an explicit commutant lifting theorem. For this we adopt a technique from the commuatnt lifting theorem on annulus [@MS].
Let $T\in{\mathcal{T}}_n({\mathcal{H}})$ be such that ${\hat{T}}_n\in {\mathbb{S}}_{n-1}({\mathcal{H}})$ is a pure tuple. Also assume that there exist positive operators $F_1,\ldots,F_{n-1}$ in ${\mathcal{B}}({\mathcal{H}})$ and a unitary $$\label{unitary}
U =\begin{bmatrix}A &B\\C&D\end{bmatrix}:
{\mathcal{D}}_{{\hat{T}}_n}\oplus(\mathop{\oplus}_{i=1}^{n-1}{\mathcal{F}}_i)\oplus {\mathcal{K}}\to
{\mathcal{D}}_{{\hat{T}}_n}\oplus(\mathop{\oplus}_{i=1}^{n-1}{\mathcal{F}}_i)\oplus{\mathcal{K}}$$ such that $$\label{generating identity}
U(D_{{\hat{T}}_n}h,F_1T_1^*h,\ldots, F_{n-1}T_{n-1}^*h, 0_{{\mathcal{K}}})=(D_{{\hat{T}}_n}T_n^*h,F_1h,\ldots, F_{n-1}h,0_{{\mathcal{K}}}),$$ where ${\mathcal{F}}_i$ is the $\overline{\operatorname{Ran}} F_i$ for all $i=1,\dots,n-1$, and ${\mathcal{K}}$ is a Hilbert space. We set ${\mathcal{F}}:=(\oplus_{i=1}^{n-1}{\mathcal{F}}_i)\oplus{\mathcal{K}}$ and define $\imath,Y
\in {\mathcal{B}}({\mathcal{H}}, {\mathcal{F}})$ by $$\imath(h)=(F_1h,\ldots, F_{n-1}h,0_{{\mathcal{K}}}),$$ and $$Y(h)=(F_1T_1^*h,\ldots, F_{n-1}T_{n-1}^*h,0_{{\mathcal{K}}}),$$ for all $h\in{\mathcal{H}}$. Then by , we have $$\label{id1}
D_{{\hat{T}}_n}T_n^*=AD_{{\hat{T}}_n}+BY,$$ and $$\label{id2}
\imath=CD_{{\hat{T}}_n}+DY.$$
We also make use of the contractive multiplier $Z=M_{E}:H^2_{{\mathcal{F}}}({\mathbb{D}}^{n-1})\to H^2_{{\mathcal{F}}}({\mathbb{D}}^{n-1})$ where $$E({\bm{z}})=(\mathop{\oplus}_{i=1}^{n-2}z_iI_{{\mathcal{F}}_i})\oplus
z_{n-1}I_{({\mathcal{F}}_{n-1}\oplus {\mathcal{K}})},$$ for all ${\bm{z}}\in{\mathbb{D}}^{n-1}$. Then $Z^*$ has the following well-known action on kernel functions: $$\label{Z}
Z^*(k_{{\bm{z}}}\otimes\xi)=k_{{\bm{z}}}\otimes E({\bm{z}})^*\xi \quad (\xi\in {\mathcal{F}}, {\bm{z}}\in {\mathbb{D}}^{n-1} ),$$ where $k_{{\bm{z}}}$ is the kernel function at ${\bm{z}}$. Finally, we define $J:{\mathcal{H}}\to H^2_{{\mathcal{H}}}({\mathbb{D}}^{n-1})$ by $$\label{J}
(Jh)({\bm{z}})=\sum_{\bm{k} \in {\mathbb{Z}_+}^{n-1}} {{\bm{z}}}^{\bm{k}} \otimes{\hat{T}}_n^{*\bm{k}} h\quad (h\in{\mathcal{H}}).$$ Since ${\hat{T}}_n$ is pure then the above sum converges, and $J$ is a well defined bounded operator.
It is worth mentioning here that the dilation map $\Pi:{\mathcal{H}}\to H^2_{{\mathcal{D}}_{{\hat{T}}_n}}({\mathbb{D}}^{n-1})$ defined by $$\label{Pi}
(\Pi h)({\bm{z}})=\sum_{\bm{k} \in {\mathbb{Z}_+}^{n-1}} {{\bm{z}}}^{\bm{k}} \otimes D_{{\hat{T}}_n} {\hat{T}}_n^{*\bm{k}} h, \quad (h\in{\mathcal{H}})$$ for ${\hat{T}}_n$ and the map $J$ defined above are related by the following identity $$\label{id4}
(I\otimes D_{{\hat{T}}_n})J=\Pi.$$ Now we exhibit various relations among these operators $\imath, Y,Z, J$ and $\Pi$, and describe them through several lemmas below.
\[l1\] Let $\imath, Y, Z,J$ and $\Pi$ be as above. Then for any $\bm{p}\in {\mathbb{Z}_+}^{n-1}$, and $\xi\in{\mathcal{F}}$, $$\label{id5}
J^*(I\otimes\imath^*)Z({\bm{z}}^{\bm{p}}\otimes\xi)={\hat{T}}_n^{\bm{p}}Y^*\xi,$$ and $$\label{id6}
J^*(I\otimes\imath^*)({\bm{z}}^{\bm{p}}\otimes\xi)={\hat{T}}_n^{\bm{p}}\imath^*\xi.$$
For $h\in {\mathcal{H}}$ and $\xi=(\xi_1,\dots,\xi_{n-1})\in{\mathcal{F}}$, we have $$\begin{split}
\langle J^*(I\otimes\imath^*)Z({\bm{z}}^{\bm{p}}\otimes\xi), h \rangle
&=\sum_{\bm{k} \in {\mathbb{Z}_+}^{n-1}}\langle Z({\bm{z}}^{\bm{p}}\otimes\xi), {{\bm{z}}}^{\bm{k}} \otimes\imath({\hat{T}}_n^{*\bm{k}}h)\rangle\\
&=\sum_{\bm{k} \in {\mathbb{Z}_+}^{n-1}}\big\langle\sum_{i=1}^{n-1}z_i{\bm{z}}^{\bm{p}}\otimes \xi_i,
{{\bm{z}}}^{\bm{k}} \otimes (F_1{\hat{T}}_n^{*\bm{k}}h,\ldots,F_{n-1}{\hat{T}}_n^{*\bm{k}}h,0_{{\mathcal{K}}})\big\rangle\\
&= \sum_{i=1}^{n-2}\langle \xi_i, F_i T_i^* {\hat{T}}_n^{*\bm{p}}h\rangle+\langle\xi_{n-1},(F_{n-1} T_{n-1}^* {\hat{T}}_n^{*\bm{p}}h,0_{{\mathcal{K}}})\rangle\\
&=\langle\xi,(F_1T_1^*{\hat{T}}_n^{*\bm{p}}h,\ldots,F_{n-1}T_{n-1}^*{\hat{T}}_n^{*\bm{p}}h,0_{{\mathcal{K}}})\rangle\\
&=\langle\xi,Y{\hat{T}}_n^{*\bm{p}}h\rangle\\
&=\langle{\hat{T}}_n^{\bm{p}}Y^*\xi,h\rangle.
\end{split}$$ This proves the first identity and proof of the second identity is similar.
\[l2\] For any $m\in {\mathcal{D}}_{{\hat{T}}_n}$ and $\bm{p}\in{\mathbb{Z}_+}^{n-1}$, $$\Pi^*({\bm{z}}^{\bm{p}}\otimes m)={\hat{T}}_n^{\bm{p}}{\mathcal{D}}_{{\hat{T}}_n}^*m.$$
For $h\in {\mathcal{H}}$, $m\in {\mathcal{D}}_{{\hat{T}}_n}$ and $\bm{p}\in{\mathbb{Z}_+}^{n-1}$, note that $$\begin{split}
\langle\Pi^*({\bm{z}}^{\bm{p}}\otimes m),h\rangle& =\big\langle{\bm{z}}^{\bm{p}}
\otimes m,\sum_{\bm{k} \in {\mathbb{Z}_+}^{n-1}} {{\bm{z}}}^{\bm{k}} \otimes D_{{\hat{T}}_n} {\hat{T}}_n^{*\bm{k}} h\big\rangle\\
&=\langle m,D_{{\hat{T}}_n}{\hat{T}}_n^{*\bm{p}}h\rangle\\
&=\langle{\hat{T}}_n^{\bm{p}}D_{{\hat{T}}_n}^*m,h\rangle.
\end{split}$$ The proof follows.
\[l3\] Let $U:=\begin{bmatrix}A &B\\C&D\end{bmatrix}:
{\mathcal{D}}_{{\hat{T}}_n}\oplus{\mathcal{F}}\to
{\mathcal{D}}_{{\hat{T}}_n}\oplus{\mathcal{F}}$ be a unitary operator as in . Then $$J^*(I_{H^2({\mathbb{D}}^{n-1})}\otimes\imath^*)[I_{H^2({\mathbb{D}}^{n-1})\otimes{\mathcal{F}}}-Z(I_{H^2({\mathbb{D}}^{n-1})}\otimes D^*)]
=\Pi^*(I_{H^2({\mathbb{D}}^{n-1})}\otimes C^*).$$
For any $\xi\in{\mathcal{F}}$ and $\bm{p}\in{\mathbb{Z}_+}^{n-1}$, $$\begin{split}
J^*(I\otimes\imath^*)[I-Z(I\otimes D^*)]({\bm{z}}^{\bm{p}}\otimes\xi)
&=J^*(I\otimes\imath^*)[{\bm{z}}^{\bm{p}}\otimes\xi-Z({\bm{z}}^{\bm{p}}\otimes D^*\xi)]\\
&={\hat{T}}_n^{\bm{p}}[\imath^*-Y^*D^*]\xi\quad\quad [\textnormal{by (\ref{id5}) and (\ref{id6})}] \\
&={\hat{T}}_n^{\bm{p}}[D_{{\hat{T}}_n}^*C^*]\xi \quad\quad [\textnormal{by (\ref{id2})}] \\
&=\Pi^*({\bm{z}}^{\bm{p}}\otimes C^*\xi)\quad\quad [\textnormal{by lemma \ref{l2}}]\\
&=\Pi^*(I\otimes C^*)({\bm{z}}^{\bm{p}}\otimes\xi).
\end{split}$$ Since $\{{\bm{z}}^{\bm{p}}\otimes\xi : \bm{p}\in{\mathbb{Z}_+}^{n-1}, \xi\in {\mathcal{F}}\}$ is a total set in $H^2_{{\mathcal{F}}}({\mathbb{D}}^{n-1})$, the proof follows.
\[l4\] Let $U:=\begin{bmatrix}A &B\\C&D\end{bmatrix}$ be a unitary as in the previous lemma. Let $\Phi ({\bm{z}})=A^* + C^* E({\bm{z}})(I_{{\mathcal{F}}} -D^*E({\bm{z}})
)^{-1}B^*$, and let $\tilde{\Phi}({\bm{z}})=\Phi({\bm{z}})-A^*$ for all ${\bm{z}}\in{\mathbb{D}}^{n-1}$. Then for all $m\in{\mathcal{D}}_{{\hat{T}}_n}$ $$J^*(I\otimes\imath^*)Z(I\otimes B^*)(1\otimes m)=\Pi^*M_{\tilde{\Phi}}(1\otimes m).$$
Let $0<t_n<1$ be a sequence converging to 1, and let $X_n:=(1-t_n)[I-t_n(I\otimes D^*)Z]^{-1}$. We first claim that $X_n\to 0$ in the weak operator topology. To prove the claim we do the following computation. For the shake of simplicity of the computation we denote $(I\otimes D^*)Z$ by $S$. First note that $$\begin{split}
I-X_nX_n^*&=(I-t_nS)^{-1}[(I-t_nS)(I-t_nS^*)-(1-t_n)^2](I-t_nS^*)^{-1}\\
&=t_n(I-t_nS)^{-1}[-S-S^*-t_n(I-SS^*)](I-t_nS^*)^{-1}\\
&=t_n(I-t_nS)^{-1}[(I-S)(I-S^*)+(1-t_n)(I-SS^*)](I-t_nS)^{-1*}\geq 0.
\end{split}$$ So, $X_n$ is a contraction for all $n$. Now for given $f\in H^2_{{\mathcal{F}}}({\mathbb{D}}^{n-1})$ and $\xi\in{\mathcal{F}}$ $$\begin{split}
\langle X_nf,k_{{\bm{z}}}\otimes\xi\rangle
&=(1-t_n)\sum_j\langle f,(t_nZ^*(I\otimes D))^j(k_{{\bm{z}}}\otimes\xi)\rangle\\
&=(1-t_n)\sum_j\langle f, k_{{\bm{z}}}\otimes (t_nE({\bm{z}})^*D)^j\xi\rangle\\
&=(1-t_n)\langle f,k_{{\bm{z}}}\otimes(I-t_nE({\bm{z}})^*D)^{-1}\xi\rangle.
\end{split}$$ Since $\|E({\bm{z}})^*D\|<1$, the last term tends to $0$ as $t_n\to1$. This implies that $X_n\to0$ in the weak operator topology. It is also easy to see that $\tilde{\Phi}_n({\bm{z}}):=C^*E({\bm{z}})(I-t_nD^*E({\bm{z}}))^{-1}B^*\to\tilde{\Phi}({\bm{z}})$ as $t_n\to 1$ for all ${\bm{z}}\in{\mathbb{D}}^{n-1}$. Thus $M_{\tilde{\Phi}_{n}}\to M_{\tilde{\Phi}}$ in the weak operator topology.
Now for $m_1,m_2\in{\mathcal{D}}_{{\hat{T}}_n}$, $$\begin{split}
&\langle(I\otimes C^*)Z(I-t_n(I\otimes D^*)Z)^{-1}(I\otimes B^*)(1\otimes m_1),(k_{{\bm{z}}}\otimes m_2)\rangle\\
=&\langle (I-t_n(I\otimes D^*)Z)^{-1}(1\otimes B^*m_1),k_{{\bm{z}}}\otimes E({\bm{z}})^*Cm_2\rangle\\
=&\sum_{j} t_n^j\langle 1\otimes B^*m_1,k_{{\bm{z}}}\otimes(E({\bm{z}})^*D)^j E({\bm{z}})^*Cm_2\rangle\\
=&\sum_{j}t_n^j\langle E({\bm{z}})(D^*E({\bm{z}}))^jB^*m_1, Cm_2\rangle\\
=&\langle\tilde{\Phi}_n({\bm{z}})m_1,m_2\rangle\\
=&\langle M_{\tilde{\Phi}_n}(1\otimes m_1),k_{{\bm{z}}}\otimes m_2\rangle.
\end{split}$$ Thus for all $m\in {\mathcal{D}}_{{\hat{T}}_n}$, $$(I\otimes C^*)Z(I-t_n(I\otimes D^*)Z)^{-1}(I\otimes B^*)(1\otimes m)=M_{\tilde{\Phi}_n}(1\otimes m).$$ Finally, $$\begin{split}
&\Pi^*M_{\tilde{\Phi}_n}(1\otimes m)\\
&=\Pi^*(I\otimes C^*)Z(I-t_n(I\otimes D^*)Z)^{-1}(I\otimes B^*)(1\otimes m)\\
&=J^*(I\otimes \imath^*)[I-Z(I\otimes D^*)]Z(I-t_n(I\otimes D^*)Z)^{-1}(I\otimes B^*)(1\otimes m)\quad\quad [\textnormal{by lemma \ref{l3}}]\\
&=J^*(I\otimes \imath^*)[I-t_nZ(I\otimes D^*)+t_nZ(I\otimes D^*)-Z(I\otimes D^*)]\\
& \hspace{3in} \times Z(I-t_n(I\otimes D^*)Z)^{-1}(I\otimes B^*)(1\otimes m)\\
&=J^*(I\otimes\imath^*)Z(I\otimes B^*)(1\otimes m)-J^*(I\otimes\imath^*)Z(I\otimes D^*)ZX_n(I\otimes B^*)(1\otimes m)
\end{split}$$ Since $X_n\to 0$ and $M_{\tilde{\Phi}_{n}}\to M_{\tilde{\Phi}}$ in the weak operator topology as $n\to \infty$, we have $$J^*(I\otimes\imath^*)Z(I\otimes B^*)(1\otimes m)=\Pi^*M_{\tilde{\Phi}}(1\otimes m)\ (m\in{\mathcal{D}}_{{\hat{T}}_n} ).$$ This completes the proof.
The above lemmas lead us to the following explicit commutant lifting theorem which is very crucial in what follows.
\[key theorem\] Let $T\in {\mathcal{T}}_n({\mathcal{H}})$ be such that ${\hat{T}}_n\in{\mathbb{S}}_{n-1}({\mathcal{H}})$ is a pure tuple. Suppose that there exist positive operators $F_1,\ldots,F_{n-1}$ in ${\mathcal{B}}({\mathcal{H}})$ and a unitary $U:
{\mathcal{D}}_{{\hat{T}}_n}\oplus{\mathcal{F}}\to
{\mathcal{D}}_{{\hat{T}}_n}\oplus {\mathcal{F}}$ satisfying $$\label{generating identity 2}
U(D_{{\hat{T}}_n}h,F_1T_1^*h,\ldots, F_{n-1}T_{n-1}^*h,0_{{\mathcal{K}}})=(D_{{\hat{T}}_n}T_n^*h,F_1h,\ldots, F_{n-1}h,0_{{\mathcal{K}}})\
(h\in{\mathcal{H}}),$$ where ${\mathcal{F}}_i=\overline{\operatorname{Ran}} F_i$ $(i=1,\dots, n-1)$ and ${\mathcal{F}}:=(\oplus_{i=1}^{n-1}{\mathcal{F}}_i)\oplus{\mathcal{K}}$ for some Hilbert space ${\mathcal{K}}$. Then $$M_{\Phi}^*\Pi=\Pi T_n^*,$$ where $\Phi\in {\mathcal{S}}_{n-1}({\mathcal{D}}_{{\hat{T}}_n},{\mathcal{D}}_{{\hat{T}}_n})$ is the transfer function of $U^*$ and $\Pi:{\mathcal{H}}\to H^2_{{\mathcal{D}}_{{\hat{T}}_n}}({\mathbb{D}}^{n-1})$ is the dilation map for the tuple ${\hat{T}}_n$ as in Theorem \[th-dc dil\].
Let $
U=\begin{bmatrix}A &B\\C&D\end{bmatrix}
$ be the block decomposition of $U$ with respect to the decomposition $D_{{\hat{T}}_n}\oplus {\mathcal{F}}$. Let $\Phi$ be the transfer function of the unitary operator $U^*$, that is $$\Phi({\bm{z}})= A^* + C^* E({\bm{z}})
(I_{{\mathcal{F}}} - D^*E({\bm{z}}))^{-1}B^*,$$ for all ${\bm{z}}\in{\mathbb{D}}^{n-1}$. We set $\tilde{\Phi}:=\Phi-A^*$. Then for any $\bm{p}\in {\mathbb{Z}_+}^{n-1} $ and $m\in{\mathcal{D}}_{{\hat{T}}_n}$, $$\begin{split}
\Pi^*M_{\Phi}({\bm{z}}^{\bm{p}}\otimes m)
=&\Pi^*M_{{\bm{z}}^{\bm{p}}}M_{\Phi}(1\otimes m)\\
=&{\hat{T}}_n^{\bm{p}}\Pi^*M_{\Phi}(1\otimes m)\\
=&{\hat{T}}_n^{\bm{p}}\Pi^*[(1\otimes A^*m)+M_{\tilde{\Phi}}(1\otimes m)]\\
=&{\hat{T}}_n^{\bm{p}}[{\mathcal{D}}_{{\hat{T}}_n}^*A^*m+J^*(I\otimes\imath^*)Z(1\otimes B^*m)]\quad\quad[\textnormal{by lemma \ref{l2} and lemma \ref{l4}}]\\
=&{\hat{T}}_n^{\bm{p}}[{\mathcal{D}}_{{\hat{T}}_n}^*A^*+Y^*B^*]m\quad\quad[\textnormal{by (\ref{id5})}]\\
=&{\hat{T}}_n^{\bm{p}}T_n{\mathcal{D}}_{{\hat{T}}_n}^*m\quad\quad[\textnormal{by (\ref{id1})}]\\
=&T_n{\hat{T}}_n^{\bm{p}}{\mathcal{D}}_{{\hat{T}}_n}^*m\\
=&T_n\Pi^*({\bm{z}}^{\bm{p}}\otimes m)\quad\quad[\textnormal{by\hspace{1mm}lemma \ref{l2}}].
\end{split}$$ Since $\{{\bm{z}}^{\bm{p}}\otimes m :\bm{p}\in {\mathbb{Z}_+}^{n-1}, m\in{\mathcal{D}}_{{\hat{T}}_n}\}$ is a total set in $H^2_{{\mathcal{D}}_{{\hat{T}}_n}}({\mathbb{D}}^{n-1})$, $$\Pi^*M_{\Phi}=T_n\Pi^*.$$ This completes the proof of the theorem.
Let $T \in {\mathcal{P}}_n({\mathcal{H}})$. We say that $T$ is of *finite rank* if $$\dim {\mathcal{D}}_{{\hat{T}}_n} < \infty,$$ and $$\dim {\mathcal{F}}_i < \infty,$$ where $${\mathcal{F}}_i = \overline{\text{ran}}~ F_i,$$ and $$F_i^2 = \prod_{\stackrel{j=1}{j\ne i}}^{n-1}(I_{\mathcal{B}({\mathcal{H}})}-
C_{T_j})(G_i).$$ for all $i=1,\dots,n-1$ (see the definition of ${\mathcal{P}}_n({\mathcal{H}})$ in Subsection \[sub-pnh\]).
We are now ready to prove the dilation theorem for commuting tuples in ${\mathcal{P}}_n({\mathcal{H}})$.
\[dilation2\] If $T$ is an $n$-tuple of operators in ${\mathcal{P}}_n({\mathcal{H}})$, then there exist an isometry $\Pi:{\mathcal{H}}\to H^2_{{\mathcal{D}}_{{\hat{T}}_n}}({\mathbb{D}}^{n-1})$ and a contractive multiplier $\Phi\in {\mathcal{S}}_{n-1}({\mathcal{D}}_{{\hat{T}}_n}, {\mathcal{D}}_{{\hat{T}}_n})$ such that $$\Pi T_i^* = \left \{\begin{array}{ll}
M_{z_i}^* \Pi \;\; \quad \quad \text{if~} i = 1, \ldots, n-1,\\
M_{\Phi}^* \Pi \quad \quad \; \;\;\text{if~} i = n.
\end{array}
\right.$$ Moreover, if $T$ is of finite rank then $\Phi$ is an inner function in ${\mathcal{S}}_{n-1}({\mathcal{D}}_{{\hat{T}}_n}, {\mathcal{D}}_{{\hat{T}}_n})$. In particular, the finite rank $n$-tuple $T \in {\mathcal{P}}_n({\mathcal{H}})$ dilates to the $n$-tuple of commuting isometries $$(M_{z_1},\dots, M_{z_{n-1}},M_{\Phi}),$$ on $H^2_{D_{{\hat{T}}_n}}({\mathbb{D}}^{n-1})$.
According to the definition of ${\mathcal{P}}_n({\mathcal{H}})$, we know that ${\hat{T}}_n \in
\mathbb{S}_{n-1}({\mathcal{H}})$ is a pure tuple. Let $\Pi : {\mathcal{H}}{\rightarrow}H^2_{{\mathcal{D}}_{{\hat{T}}_{n}}}({\mathbb{D}}^{n-1})$ be the canonical isometry corresponding to ${\hat{T}}_{n}$ as in Theorem \[th-dc dil\]. Then $$\Pi T_i^*=M_{z_i}^*\Pi,$$ for all $i=1,\dots,n-1$. Clearly (see Subsection \[sub-pnh\]) $$\begin{split}
D_{{\hat{T}}_n}^2-T_nD_{{\hat{T}}_n}^2T_n^* & = \sum_{\bm{k} \in {\mathbb{Z}_+}^{n-1}}
(-1)^{|\bm{k}|} {\hat{T}}_{n}^{\bm{k}} (I - T_n T_n^*) {\hat{T}}_{n}^{*\bm{k}}
\\
& = \prod_{j=1}^{n-1}(I_{\mathcal{B}({\mathcal{H}})}- C_{T_j})(I-T_nT_n^*).
\end{split}$$ Now let $\{G_1, \ldots, G_{n-1}\}$ be the positive operators on ${\mathcal{H}}$ associated to $T$, that is $$I-T_nT_n^*= G_1+\dots+G_{n-1}.$$ Then $$D_{{\hat{T}}_n}^2-T_nD_{{\hat{T}}_n}^2T_n^* =
\sum_{i=1}^{n-1}\prod_{j=1}^{n-1}(I_{\mathcal{B}({\mathcal{H}})}-
C_{T_j})(G_i),$$ and so, in view of $$C_{T_l} C_{T_m} = C_{T_m} C_{T_l},$$ for all $l, m = 1, \ldots, n-1$, we have $$\begin{split}
D_{{\hat{T}}_n}^2 - T_nD_{{\hat{T}}_n}^2T_n^* & = \sum_{i=1}^{n-1}
\prod_{\stackrel{j=1}{j\ne i}}^{n-1} (I_{\mathcal{B}({\mathcal{H}})}-
C_{T_j}) (I_{\mathcal{B}({\mathcal{H}})}- C_{T_i})(G_i)
\\
&=\sum_{i=1}^{n-1}\Big(\prod_{\stackrel{j=1}{j\ne i}}^{n-1}
(I_{\mathcal{B}({\mathcal{H}})}- C_{T_j})
(G_i)-T_i\Big(\prod_{\stackrel{j=1}{j\ne
i}}^{n-1}(I_{\mathcal{B}({\mathcal{H}})}- C_{T_j})(G_i)\Big)T_i^*\Big)
\\
&=\sum_{i=1}^{n-1}(F_i^2-T_iF_i^2T_i^*).
\end{split}$$ as $$F_i^2 = \prod_{\stackrel{j=1}{j\ne i}}^{n-1}(I_{\mathcal{B}({\mathcal{H}})}-
C_{T_j})(G_i).$$ Hence $$D_{{\hat{T}}_n}^2 + \sum_{i=1}^{n-1} T_iF_i^2T_i^* = T_nD_{{\hat{T}}_n}^2T_n^* +
\sum_{i=1}^{n-1}F_i^2,$$ and so the map $$\label{un}
U: \{(D_{{\hat{T}}_n} h, F_1T_1^* h,\dots, F_{n-1} T_{n-1}^*h):
h\in{\mathcal{H}}\}\to \{(D_{{\hat{T}}_n} T_n^*h, F_1h,\dots, F_{n-1}h):h\in{\mathcal{H}}\}$$ defined by $$U (D_{{\hat{T}}_n} h, F_1T_1^*h,\dots, F_{n-1}T_{n-1}^*h) = (D_{{\hat{T}}_n}
T_n^*h, F_1 h,\dots, F_{n-1}h),$$ for all $h\in{\mathcal{H}}$, is an isometry. Now by adding an infinite dimensional Hilbert space ${\mathcal{K}}$, if necessary, we can find a unitary $U: {\mathcal{D}}_{{\hat{T}}_n}\oplus(\oplus_{i=1}^{n-1}{\mathcal{F}}_i)\oplus {\mathcal{K}}\to {\mathcal{D}}_{{\hat{T}}_n}\oplus(\oplus_{i=1}^{n-1}{\mathcal{F}}_i)\oplus {\mathcal{K}}$ such that for all $h\in{\mathcal{H}}$ $$U (D_{{\hat{T}}_n} h, F_1T_1^*h,\dots, F_{n-1}T_{n-1}^*h, 0_{{\mathcal{K}}}) = (D_{{\hat{T}}_n}
T_n^*h, F_1 h,\dots, F_{n-1}h, 0_{{\mathcal{K}}}),$$ where ${\mathcal{F}}_i=\overline{\operatorname{Ran}}F_i$ for all $i=1,\dots,n-1$. Thus the first part of the theorem follows from the theorem \[key theorem\].
In the case when $T$ is a finite rank operator tuple, since ${\mathcal{D}}_{{\hat{T}}_n}$ and ${\mathcal{F}}_i$, $i = 1, \ldots, n-1$, are all finite dimensional Hilbert spaces, the unitary $U$ in extends to a unitary on ${\mathcal{D}}_{{\hat{T}}_n}\oplus(\oplus_{i=1}^{n-1}{\mathcal{F}}_i)$, which we also denote by $U$. Then again applying Theorem \[key theorem\] with ${\mathcal{K}}=0$, we have $M_{\Phi}^*\Pi=\Pi T_n^*$, where $\Phi$ is the transfer function corresponding to $U^*$. Finally, it follows from Lemma \[transfer\] that $\Phi$ is a $\mathcal{B}(D_{{\hat{T}}_n})$-valued inner multiplier in this case. This completes the proof of the theorem.
\(i) It is worth mentioning that the converse of the above theorem is also true. Indeed, if ${\mathcal{Q}}$ is a joint invariant subspace for $(M_{z_1}^*, \ldots, M_{z_{n-1}}^*, M_{\Phi}^*)$ on $H^2_{{\mathcal{E}}}({\mathbb{D}}^{n-1})$ for some Hilbert space ${\mathcal{E}}$ and multiplier $\Phi \in {\mathcal{S}}_{n-1}({\mathcal{E}}, {\mathcal{E}})$, then using [@BLTT Theorem 5.1], it follows that $$(P_{{\mathcal{Q}}}M_{z_1}|_{{\mathcal{Q}}},\dots, P_{{\mathcal{Q}}}M_{z_{n-1}}|_{{\mathcal{Q}}},
P_{{\mathcal{Q}}}M_{\Phi}|_{{\mathcal{Q}}}) \in {\mathcal{P}}_n({\mathcal{Q}}).$$
\(ii) Let $T=(T_1,\dots,T_n)\in {\mathcal{T}}_n({\mathcal{H}})$ be such that ${\hat{T}}_n, \hat{T_1}\in \mathbb{S}_n({\mathcal{H}})$ and ${\hat{T}}_n$ is pure. Then $T\in{\mathcal{P}}_{n}({\mathcal{H}})$. Indeed, for $G_1=(I-T_nT_n^*)$ and $G_i=0$ for all $i=2,\dots,n-1$, we have $$\prod_{j=2}^{n-1}(I_{\mathcal{B}({\mathcal{H}})}-
C_{T_j})(G_1)=\mathsf{D}_{\hat{T_{1}}}^2\ge 0,$$ as $\hat{T_1}\in\mathbb{S}({\mathcal{H}})$. Thus the class of operator tuples ${\mathcal{P}}_n({\mathcal{H}})$ is larger than the class of operators considered in [@BDHS], and therefore the above theorem provides a different proof of the dilation results obtained in [@BDHS].
von-Neumann inequality for finite rank tuples in ${\mathcal{P}}_n({\mathcal{H}})$ {#3}
=================================================================================
In this section we consider the von-Neumann inequality of operator tuples in ${\mathcal{P}}_n({\mathcal{H}})$ and establish a sharp von-Neumann inequality over a variety in $\overline{{\mathbb{D}}}^n$ for the class of finite rank operator tuples in ${\mathcal{P}}_n({\mathcal{H}})$.
\[vN2\] Let $T\in{\mathcal{P}}_n({\mathcal{H}})$ be a finite rank operator tuple. Then there exists a variety $V$ in $\overline{{\mathbb{D}}}^n$ such that for any polynomial $P\in {\mathbb{C}}[z_1,\dots,z_n]$, $$\| P(T)\|\le \sup_{{\bm{z}}\in V}|P({\bm{z}})|.$$ Moreover, if $T_n$ is a pure contraction then one can take $V$ to be a variety in ${\mathbb{D}}^n$.
Let $(M_{z_1},\dots, M_{z_{n-1}}, M_{\Phi(z_1,\dots,z_{n-1})})$ on $H^2_{D_{{\hat{T}}_n}}({\mathbb{D}}^{n-1})$ be an isometric dilation of $T$ as in Theorem \[dilation2\], where $\Phi({\bm{z}})= A^*+
C^*E({\bm{z}})(I_{{\mathcal{F}}}-D^*E({\bm{z}}))^{-1}B^*$ is the transfer function associated to a unitary $$U^*=\begin{bmatrix} A^*& C^*\\ B^*& D^*\end{bmatrix}: D_{{\hat{T}}_n}\oplus
{\mathcal{F}}\to
D_{{\hat{T}}_n}\oplus{\mathcal{F}}$$ as constructed in the proof of Theorem \[dilation2\]. Let $A^*=\begin{bmatrix} W^*& 0\\0&
E^*\end{bmatrix}$ be the canonical decomposition of $A^*$ into the unitary part $W^*$ on ${\mathcal{H}}_0$ and the completely non-unitary part $E^*$ on ${\mathcal{H}}_1$, where ${\mathcal{H}}_0\oplus {\mathcal{H}}_1=D_{{\hat{T}}_n}$. With respect to the above decomposition of $A^*$, let $$\Phi({\bm{z}})=\begin{bmatrix} \Phi_0({\bm{z}})& 0\\0& \Phi_1({\bm{z}})\end{bmatrix}$$ be the decomposition of $\Phi$, where $\Phi_0({\bm{z}})\equiv W^*$ $({\bm{z}}\in{\mathbb{D}}^{n-1})$ and $\Phi_1({\bm{z}})=E^*+C^*E({\bm{z}})(I_{{\mathcal{F}}}-D^*E({\bm{z}}))^{-1}B^*$ $({\bm{z}}\in{\mathbb{D}}^{n-1})$ is a multiplier in $H^{\infty}_{\mathcal{B}({\mathcal{H}}_1)}({\mathbb{D}}^{n-1})$. We set $$V_0:=\{{\bm{z}}\in\bar{{\mathbb{D}}}^n: \det(z_n I_{{\mathcal{H}}_0}-\Phi_0(z_1,\dots,z_{n-1}))=0\}= {\mathbb{D}}^{n-1}\times \sigma(W^*)$$ and $$V_1:=\{{\bm{z}}\in{\mathbb{D}}^n: \det (z_nI_{{\mathcal{H}}_1}-\Phi_1(z_1,\dots,z_{n-1}))=0\}.$$ Now for any polynomial $P\in{\mathbb{C}}[z_1,\dots,z_n]$, we have $$\begin{split}
\|P(T)\| &\le \| P(M_{z_1},\dots,M_{z_{n-1}}, M_{\Phi(z_1,\dots,z_{n-1})})\|\\
&=\| M_{P(z_1,\dots,z_{n-1},\Phi(z_1,\dots,z_{n-1}))}\|\\
&=\sup_{\theta_1,\dots,\theta_{n-1}}\| P(e^{i\theta_1}I_{D_{{\hat{T}}_n}},\dots,e^{i\theta_{n-1}}I_{D_{{\hat{T}}_n}},\Phi(e^{i\theta_1},\dots,e^{i\theta_{n-1}}))\|.
\end{split}$$ Also note that for $j=0,1$, $$\begin{split}
&\sup_{\theta_1,\dots,\theta_{n-1}}\| P(e^{i\theta_1}I_{H_j},\dots,e^{i\theta_{n-1}}I_{H_j},\Phi_j(e^{i\theta_1},\dots,e^{i\theta_{n-1}}))\| \\
&= \sup_{\theta_1,\dots,\theta_{n-1}}\{|P(e^{i\theta_1},\dots,e^{i\theta_{n-1}},\lambda)|: \lambda\in \sigma(\Phi_j(e^{i\theta_1},\dots,e^{i\theta_{n-1}}))\}\\
& \le \|P\|_{\partial V_j}.
\end{split}$$ Since $\Phi(e^{i\theta_1},\dots,e^{i\theta_{n-1}})=
\Phi_0(e^{i\theta_1},\dots,e^{i\theta_{n-1}})\oplus
\Phi_1(e^{i\theta_1},\dots,e^{i\theta_{n-1}})$, we have by continuity and Lemma \[transfer\] that $$\|P(T)\|\le \sup_{{\bm{z}}\in V}|P({\bm{z}})|,$$ where $V= V_0\cup V_1$.
For the second part, we prove $V_0$ is empty by showing that ${\mathcal{H}}_0=\{0\}$. Let ${\mathcal{Q}}=\Pi({\mathcal{H}})$, where $\Pi$ is the dilation map for $T$. First, we prove that $${\mathcal{Q}}\subseteq H^2_{{\mathcal{H}}_1}({\mathbb{D}}^{n-1}).$$ To prove this let $g\in H^2_{{\mathcal{H}}_0}({\mathbb{D}}^{n-1})$. Then for any $m \in {\mathbb{Z}_+}$ and there exits $g_m =
M_{\Phi_0}^{*m} g\in H^2_{{\mathcal{H}}_0}({\mathbb{D}}^{n-1})$ such that $$g= M_{\Phi_0}^{ m} g_m.$$ Now, for $f\in{\mathcal{Q}}$ $$\begin{split}
\langle g,f\rangle & = \langle M_{\Phi_0}^{m}g_m, f\rangle
\\
&=\langle g_m, M_{\Phi_0}^{*m} f \rangle\\
&=\langle g_m, M_{\Phi}^{*m} f \rangle\\
& = \langle g_m , T_n^{* m} f \rangle.
\end{split}$$ Since $T_n$ is pure, $\langle g_m , T_n^{* m} f \rangle\to 0$ as $m\to\infty$. This implies that ${\mathcal{Q}}\subseteq H^2_{{\mathcal{H}}_1}({\mathbb{D}}^{n-1})$. By the minimality of the isometric dilation of ${\hat{T}}_n$ we have $$\bigvee_{\bm{k} \in {\mathbb{Z}_+}^{n-1}} M_{{\bm{z}}}^{\bm{k}}{\mathcal{Q}}=
H^2_{{\mathcal{D}}_{{\hat{T}}_n}}({\mathbb{D}}^{n-1}).$$ Also $H^2_{{\mathcal{H}}_1}({\mathbb{D}}^{n-1}) \subseteq
H^2_{{\mathbb{D}}_{{\hat{T}}_n}}({\mathbb{D}}^{n-1})$ is a joint reducing subspace for $(M_{z_1},
\ldots, M_{z_{n-1}})$. Hence $H^2_{{\mathcal{H}}_0}({\mathbb{D}}^{n-1})=\{0\}$ and therefore ${\mathcal{H}}_0=\{0\}$. The proof follows.
**Acknowledgement:** The research of the second named author is supported by DST-INSPIRE Faculty Fellowship No. DST/INSPIRE/04/2015/001094. The research of the third named author is supported in part by the Mathematical Research Impact Centric Support (MATRICS) grant, File No : MTR/2017/000522, by the Science and Engineering Research Board (SERB), Department of Science & Technology (DST), Government of India.
[99]{}
J. Agler and J. McCarthy, *Distinguished Varieties*, Acta. Math. 194 (2005), 133-153.
C. G. Ambrozie, M. Engliš ans V. Müller, [*Operator tuples and analytic models over general domains in ${\mathbb{C}}^n$*]{}, J. Operator Theory 47 (2002), 287â-€302.
T. Ando, [*On a pair of commutative contractions*]{}, Acta Sci. Math. (Szeged) 24 (1963), 88-–90.
W. Arveson, [*Subalgebras of $C^*$-algebras*]{}, Acta Math. 123 (1969), 141–-224.
W. Arveson, [*Subalgebras of $C^*$-algebras. II*]{}, Acta Math. 128 (1972), 271-–308.
J. A. Ball, T. T. Trent, [*Unitary colligations, reproducing kernel Hilbert spaces, and Nevanlinna–Pick interpolation in several variables*]{}, J. Funct. Anal. 157 (1998), 1-–61.
J. A. Ball, C. Sadosky, V. Vinnikov, [*Scattering systems with several evolutions and multidimensional input/state/output systems*]{}, Integral Equations Operator Theory 52 (2005), 323–-393.
J. A. Ball, W.S. Li, D. Timotin and T. T. Trent, [*A commutant lifting theorem on the polydisc*]{}, Indiana Univ. Math. J. 48 (1999), 653–675.
S. Barik, B. K. Das. K. J. Haria and J. Sarkar, [*Isometric dilations and von Neumann inequality for a class of tuples in the polydisc*]{}, arXiv:1710.07624.
C. A. Berger, L. A. Coburn and A. Lebow, [*Representation and index theory for $C^*$-algebras generated by commuting isometries*]{}, J. Funct. Anal. 27 (1978), no. 1, 51–99.
M.-D. Choi and K. R. Davidson, [*A $3 \times 3$ dilation counterexample*]{}, Bull. London Math. Soc. 45 (2013), 511–-519.
M. Cotlar and C. Sadosky, [*Transference of metrics induced by unitary couplings, a Sarason theorem for the bidimensional torus, and a Sz.-Nagy-Foias theorem for two pairs of dilations*]{}, J. Funct. Anal. 111 (1993), 473–-488.
M. Crabb and A. Davie, [*Von Neumann’s inequality for Hilbert space operators*]{}, Bull. London Math. Soc. 7 (1975), 49-–50.
R. E. Curto and F.-H. Vasilescu, [ *Standard operator models in the polydisc*]{}, Indiana Univ. Math. J. 42 (1993), 791-810.
R. E. Curto and F.-H. Vasilescu, [*Standard operator models in the polydisc*]{}, II, Indiana Univ. Math. J. 44 (1995), 727-746.
B. K. Das and J. Sarkar, [*Ando dilations, von Neumann inequality, and distinguished varieties*]{}, J. Funct. Anal. 272 (2017), 2114-2131.
B. K. Das, J. Sarkar and S. Sarkar, [*Factorizations of contractions*]{}, Adv. Math 322 (2017), 186-200.
L. de Branges, J. Rovnyak, [*Square Summable Power Series, Holt*]{}, Rinehart & Winston, New York, 1966.
S. Drury, [*Remarks on von Neumann’s inequality*]{}, Banach spaces, harmonic analysis, and probability theory, Storrs, CT, 1980/1981, Lecture Notes in Mathematics 995 (Springer, Berlin, 1983) 14–32.
A. Grinshpan, D.S. Kaliuzhnyi-Verbovetskyi, V. Vinnikov and H.J. Woerdeman, [*Classes of tuples of commuting contractions satisfying the multivariable von Neumann inequality*]{}, J. Funct. Anal. 256 (2009), 3035-3054.
J. A. Holbrook, [*Inequalities of von Neumann type for small matrices*]{}, Function spaces, Edwardsville, IL, 1990, Lecture Notes in Pure and Applied Mathematics 136 (Dekker, New York, 1992), 189-–193.
J. A. Holbrook, [*Schur norms and the multivariate von Neumann inequality*]{}, Recent advances in operator theory and related topics, Szeged, 1999, Operator Theory: Advances and Applications 127 (Birkhäuser, Basel, 2001), 375-–386.
G. Knese, [*Polynomials defining distinguished varieties*]{}, Trans. Amer. Math. Soc. 362 (2010), 5635–5655.
G. Knese, [*The von Neumann inequality for $3 \times 3$ matrices*]{}, Bull. Lond. Math. Soc. 48 (2016), 53-–57.
. Kosiński, [*Three-point Nevanlinna Pick problem in the polydisc*]{}, Proc. London Math. Soc. 111 (2015), 887–-910.
S. McCullough and S. Siltanic, [*Agler-Commutant Lifting on an Annulus*]{}, Integr. Equ. Oper. Theory 72 (2012), 449–-482.
V. Müller, [*Commutant lifting theorem for n -tuples of contractions*]{}, Acta Sci. Math. (Szeged) 59 (1994), 465-474.
V. Müller and F.-H. Vasilescu, [*Standard models for some commuting multioperators*]{}, Proc. Amer. Math. Soc. 117 (1993), 979-989.
B. Sz.-Nagy and C. Foias, [*Harmonic Analysis of Operators on Hilbert Space. North-Holland*]{}, Amsterdam-London, 1970.
B. Sz.-Nagy, [*Sur les contractions de l’espace de Hilbert*]{}, Acta Sci. Math. (Szeged) 15 (1953), 87–92.
S. Parrott, [ *Unitary dilations for commuting contractions*]{}, Pacific J. Math. 34 (1970), 481â-€“490.
N. Th. Varopoulos, [*On an inequality of von Neumann and an application of the metric theory of tensor products to operator theory*]{}, J. Funct. Anal. 16 (1974), 83-–100.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We consider two sets of issues in this paper. The first has to do with moduli stabilization, existence of “area codes" [@Giryavets] and the possibility of getting a non-supersymmetric dS minimum without the addition of $\overline{D3}$-branes as in KKLT for type II flux compactifications. The second has to do with the “Inverse Problem" [@VafaInverse] and “Fake Superpotentials" [@Ceresole+Dall''agata] for extremal (non)supersymmetric black holes in type II compactifications. We use (orientifold of) a “Swiss Cheese" Calabi-Yau [@SwissCheese] expressed as a degree-18 hypersurface in ${\bf WCP}^4[1,1,1,6,9]$ in the “large-volume-scenario" limit [@Balaetal2]. The main result of our paper is that we show that by including non-perturbative $\alpha^\prime$ and instanton corrections in the Kähler potential and superpotential [@Grimm], it may be possible to obtain a large-volume non-supersymmetric dS minimum [*without*]{} the addition of anti-D3 branes a la KKLT. The chosen Calabi-Yau has been of relevance also from the point of other studies of Kähler moduli stabilization via nonperturbative instanton contributions [@Denef+Douglas+Florea] and non-supersymmetric AdS vacua (and their subsequent dS-uplifts) using $(\alpha^\prime)^3$ corrections to the Kähler potential [@Balaetal2; @BBHL; @westphal; @Balaetal1].'
---
-2.8cm ‘@=11 ‘@=11
[**Moduli Stabilization, Large-Volume dS Minimum Without $\overline{D3}$-Branes, (Non-)Supersymmetric Black Hole Attractors and Two-Parameter Swiss Cheese Calabi-Yau’s**]{} 0.1in [ Aalok Misra$^{(a),(b)}$[^1] and Pramod Shukla$^{(a)}$ [^2]\
(a) Department of Physics, Indian Institute of Technology, Roorkee - 247 667, Uttaranchal, India\
(b) Physics Department, Theory Unit, CERN, CH1211, Geneva 23, Switzerland ]{} 0.5 true in
Introduction
============
Flux compactifications have been extensively studied from the point of view of moduli stabilization (See [@Grana] and references therein). Though, generically only the complex structure moduli get stabilized by turning on fluxes and one needs to consider non-perturbative moduli stabilization for the Kähler moduli[@KKLT]. In the context of type II compactifications, it is naturally interesting to look for examples wherein it may be possible to stabilize the complex structure moduli (and the axion-dilaton modulus) at different points of the moduli space that are finitely separated, for the [*same*]{} value of the fluxes. This phenomenon is referred to as “area codes" that leads to formation of domain walls. Another extremely important issue related to moduli stabilization is the problem of getting a non-supersymmetric de Sitter vacuum in string theory. The KKLT scenario which even though does precisely that, has the problem of addition of an uplift term to the the potential, corresponding to addition of $\overline{D3}$-branes, that can not be cast into an ${\cal N}=1$ SUGRA formalism. It would be interesting to be able to get a de Sitter vacuum without the addition of such $\overline{D3}$-branes. The Large Volume Scenarios’ study initiated in [@Balaetal2] provides a hope for the same. Further, there is a close connection between flux vacua and black-hole attractors. It has been shown that extremal black holes exhibit an interesting phenomenon - the attractor mechanism [@attractor]. In the same, the moduli are “attracted" to some fixed values determined by the charges of the black hole, independent of the asymptotic values of the moduli. Supersymmetric black holes at the attractor point, correspond to minimizing the central charge and the effective black hole potential, whereas nonsupersymmetric attractors [@nonsusybh1], at the attractor point, correspond to minimizing only the potential and not the central charge. The latter have recently been (re)discussed [@nonsusybh2] in the literature.
In this paper, we try to address all the issues of the previous paragraph by exploring different perturbative and non-perturbative (in $\alpha^\prime$ and instanton contributions) aspects of (non)supersymmetric flux vacua and black holes in the context of type II compactifications on (orientifold) of compact Calabi-Yau’s of a projective variety with multiple singular conifold loci in their moduli space. The compact Calabi-Yau we work with is of the “Swiss cheese" type. The paper is planned as follows. In section [**2**]{}, based on [@Candelasetal], we perform a detailed analysis of the periods of the Calabi-Yau three-fold considered in this paper, working out their forms in the symplectic basis for points away and close to the two singular conifold loci. The results of section [**2**]{} get used in the subsequent section ([**3**]{}). We then discuss, in section [**3**]{}, stabilization of the complex structure moduli including the axion-dilaton modulus by extremizing the flux superpotential for points near and close to the two conifold loci, arguing the existence of “area codes" and domain walls. In section [**4**]{}, we show that by the inclusion of non-perturbative $\alpha^\prime$-corrections to the Kähler potential that survive orientifolding and instanton contributions to the superpotential, one can, analogous to [@Balaetal2], get a large-volume non-supersymmetric dS vacuum [*without the addition of $\overline{D3}$-branes*]{}. We consider this to be the most significant result of this paper. In section [**5**]{}, we explicitly solve the “inverse problem" using the techniques of [@VafaInverse]. In section [**6**]{}, using the techniques of [@Ceresole+Dall'agata] we show the existence of multiple superpotentials (including therefore “fake superpotentials"). Section [**7**]{} has the conclusions.
The Moduli Space Scan and the Periods
=====================================
In this section, based on results in [@Candelasetal], we look at different regions in the moduli space of a two-parameter Calabi-Yau three fold of a projective variety expressed as a hypersurface in a weighted complex projective space, and write out the explicit expressions for the periods. The explicit expressions, though cumbersome, will be extremely useful when studying complex structure moduli stabilization and existence of “area codes" in section [**3**]{}, solving explicitly the “inverse problem" in section [**4**]{} and showing explicitly the existence of “fake superpotentials" in section [**5**]{} in the context of non-supersymmetric black hole attractors. More precisely, based on [@Candelasetal], we will consider the periods of the “Swiss cheese" [^3] Calabi-Yau three-fold obtained as a resolution of the degree-18 hypersurface in ${\bf WCP}^4[1,1,1,6,9]$: $$\label{eq:hypersurface}
x_1^{18} + x_2^{18} + x_3^{18} + x_4^3 + x_5^2 - 18\psi \prod_{i=1}^5x_i - 3\phi x_1^6x_2^6x_3^6 = 0.$$ Similar to the explanation given in [@Kachruetal], it is understood that only two complex structure moduli $\psi$ and $\phi$ are retained in (\[eq:hypersurface\]) which are invariant under the group $G$ of footnote 3, setting the other invariant complex structure moduli appearing at a higher order (due to invariance under $G$) at their values at the origin.
Defining $\rho\equiv (3^4.2)^{\frac{1}{3}}\psi$, the singular loci of (\[eq:hypersurface\]) are in ${\bf WCP}^2[3,1,1]$ with homogenous coordinates $[1,\rho^6,\phi]$ and are given as under:
1. ${\it Conifold\ Locus 1}: \{(\rho,\phi)|(\rho^6+\phi)^3=1\}$
2. ${\it Conifold\ Locus 2}: \{(\rho,\phi)|\phi^3=1\}$
3. ${\it Boundary}: (\rho,\phi)\rightarrow\infty$
4. ${\it Fixed\ point\ of\ quotienting}$: The fixed point $\rho=0$ of ${\cal A}^3$ where ${\cal A}:(\rho,\phi)\rightarrow(\alpha\rho,\alpha^6\phi)$, where $\alpha\equiv e^{\frac{2\pi i}{18}}
$[^4].
We will be considering the following sectors in the $(\rho,\phi)$ moduli space:
- $\underline{|\phi^3|>1, 0<arg\phi<\frac{2\pi}{3},{\rm large}\ \psi}$
The fundamental period $\varpi_0$, obtained by directly integrating the holomorphic three-form over the “fundamental cycle" (See [@Candelasetal]), is given by: $$\begin{aligned}
\label{eq:largephilargepsi1}
& & \varpi_0=\sum_{k=0}^\infty \frac{(6k)!}{k!(2k)!(3k)!}\left(\frac{-3}{18^6\psi^6}\right) U_k(\phi)\nonumber\\
& & =\sum_{k=0}^\infty\frac{(-)^k\Gamma(k+\frac{1}{6})\Gamma(k+\frac{5}{6})}{(k!)^2}\left(\frac{1}{\rho^{6k}}\right)
U_k(\phi),\end{aligned}$$ where $U_\nu(\phi)\equiv \phi^\nu\ _3F_2(-\frac{\nu}{3},\frac{1-\nu}{3},\frac{2-\nu}{3};1,1;\frac{1}{\phi^3})$; the other components of the period vector are given by: $\varpi_i=\varpi_0(\alpha^i\psi,\alpha^{6i}\phi)$ where $\alpha\equiv e^{\frac{2\pi i}{18}},\ i=1,2,3,4,5$.
- $\underline{|\phi^3|<1,\ {\rm large}\ \psi}$
The fundamental period is given by: $$\label{eq:smallphilargepsi11}
\varpi_0=\sum_{m=0}^\infty\sum_{n=0}^\infty\frac{(18n + 6m)!(-3\phi)^m}{(9n + 3m)!(6n + 2m)!(n!)^3m!(18\psi)^{18n + 6m}},$$ implying that around a suitable $\rho=\rho_0$ and $\phi=\phi_0$: $$\label{eq:smallphilargepsi2}
\left(\begin{array}{c}
\varpi_0\\
\varpi_1\\
\varpi_2\\
\varpi_3\\
\varpi_4\\
\varpi_5\end{array}\right)
=\left(\begin{array}{ccc}P_1 & P_2 & P_3\end{array}\right)\left(\begin{array}{c} 1\\(\phi - \phi_0)\\(\rho - \rho_0)
\end{array}\right),$$ where $P_{1,2,3}$ are given in appendix A.
- $\underline{|\frac{\rho^6}{\phi - \omega^{0,-1,-2}}|<1}$
$$\label{eq:awaycl11}
\varpi_{3a+\sigma}=\frac{1}{3\pi}\sum_{r=1,5}\alpha^{3ar}sin\left(\frac{\pi r}{3}\right)\xi^\sigma_r(\psi,\phi)[a=0,1;\sigma=0,1,2],$$
where $\xi_r^\sigma(\psi,\phi)=\sum_{k=0}^\infty\frac{(\Gamma(k+\frac{r}{6})^2}{k!\Gamma(k+\frac{r}{3})}\rho^{6k+r}
U_{-(k+\frac{r}{6})}^\sigma(\phi), U^\sigma_\nu(\phi)=\omega^{-\nu\sigma}U_\nu(\omega^\sigma\phi)\ \footnote{The three values of $\sigma$ correspond to the three solutions to $(1-\phi^3)U_\nu^{\prime\prime\prime}(\phi)
+3(\nu-1)\phi^2U^{\prime\prime}_\nu(\phi) - (3\nu^2-3\nu+1)\phi U^\prime_\nu(\phi)+\nu^3 U_\nu(\phi)=0$; the
Wronskian of the three solutions is given by: $\frac{-27i}{2\pi^3}e^{-i\pi\nu}sin^2(\pi\nu)(1-\phi^3)^{\nu-1}$ -
the solutions are hence linearly independent except when $\nu\in{\bf Z}$}, \omega\equiv e^{\frac{2\pi i}{3}}$; for small $\phi$, $$\label{eq:Usmallphi}
U_\nu(\phi)=\frac{3^{-1-\nu}}{\Gamma(-\nu)}\sum_{m=0}^\infty\frac{\Gamma(\frac{m-\nu}{3})(3\omega\phi)^m}{(\Gamma(1 - \frac{m-\nu}{3})^2m!}.$$ Expanding about a suitable $\phi=\phi_0$ and $\rho=\rho_0$, one can show: $$\label{eq:awaycl12}
\left(\begin{array}{c}
\varpi_0\\
\varpi_1\\
\varpi_2\\
\varpi_3\\
\varpi_4\\
\varpi_5\\
\end{array}\right)=\left(\begin{array}{ccc}
M_1 & M_2 & M_3
\end{array}\right)\left(\begin{array}{c}
1\\
(\rho - \rho_0)\\
(\phi - \phi_0)\\
\end{array}\right),$$ where $M_{1,2,3}$ are given in appendix A.
- $\underline{{\rm Near\ the\ conifold\ locus:}\ \rho^6 + \phi = 1}$
The periods are given by: $$\label{eq:confl11}
\varpi_i = C_i g(\rho,\phi) ln(\rho^6 + \phi - 1) + f_i(\rho,\phi),$$ where $C_i=(1,1,-2,1,0,0)$, $g(\rho,\phi)=\frac{i}{2\pi}(\varpi_1 - \varpi_0)\sim a(\rho^6 + \phi - 1)$ near $\rho^6 + \phi - 1\sim0$ where $a$ is a constant and $f_i$ are analytic in $\rho$ and $\psi$. The analytic functions near the conifold locus are given by: $$\label{eq:confl12}
f_{3a + \sigma}=\frac{1}{2\pi}\sum_{r=1,5}e^{\frac{i\pi ar}{3}}sin\left(\frac{\pi r}{3}\right) \xi_r^\sigma(\rho,\phi), a=0,1;\ \sigma=0,1,2.$$ Defining $x\equiv(\rho^6 + \phi - 1)$, one can show that: $$\label{eq:confl13}
\xi_r^\sigma=\sum_{k=0}^\infty\sum_{m=0}^\infty
\frac{3^{-1 + k + \frac{r}{6} + m}e^{\frac{2i\pi(\sigma + 1)}{3} + \frac{-i\pi(k + \frac{r}{6})}{3}}(-)^k(\Gamma(k + \frac{r}{6}))^2}{\Gamma(k + 1)\Gamma(k + \frac{r}{6})\Gamma(k + \frac{r}{6})(\Gamma(1 - \frac{m + k + \frac{r}{6}}{3}))^2m!}(x - \phi + 1)^{k + \frac{r}{6}}.$$ One can hence see that: $$\label{eq:confl14}
\left(\begin{array}{c}
f_0\\
f_1\\
f_2\\
f_3\\
f_4\\
f_5\end{array}\right)=\left(\begin{array}{ccc} N_1 & N_2 & N_3\end{array}\right)\left(\begin{array}{c}
1\\
x\\
\phi\end{array}\right),$$ where $N_{1,2,3}$ are given in appendix A.
- $\underline{{\rm Near}\ \phi^3=1,\ {\rm Large}\ \rho}$
From asymptotic analysis of the coefficients, one can argue: $$\begin{aligned}
\label{eq:confl21}
& & U_\nu(\phi)\sim-\frac{\sqrt{3}}{2\pi(\nu+1)}\left[(\phi-1)^{\nu+1} - 2\omega(\phi-\omega^{-1}) + \omega^2(\phi-\omega^{-2})\right],\nonumber\\
& & \equiv y^0_\nu - 2y^1_\nu + y^2_\nu,\end{aligned}$$ where $\omega\equiv e^{\frac{2i\pi}{3}}$. Defining $U^\sigma_\nu(\phi)=\sum_{\tau=0}^2\gamma_\nu^{\sigma,\tau}y_\nu^\tau(\phi)$, where $\gamma^{\sigma,\tau}_\nu=\left(\begin{array}{ccc}
1 & -2 & 1 \\
e^{-2i\pi\nu} & 1 & -2 \\
-2e^{-2i\pi\nu} & e^{-2i\pi\nu} & 1\\
\end{array}\right)$, one can show that $U_\nu,\left(\frac{\sum_{\sigma=0}^2U^\sigma_\nu(\phi)}{1-e^{-2i\pi\nu}}=\right)y^0_\nu(\phi) - y^1_\nu(\phi)\equiv V_\nu(\phi),\left(\frac{3V_\nu(\phi) - 2U_\nu(\phi) - U^1_\nu(\phi)}{1-e^{-2i\pi\nu}}=\right)y^0_\nu(\phi)\equiv W_\nu(\phi)$ are linearly independent even for $\nu\in{\bf Z}$[^5].
For small $\rho$, $$\label{eq:confl22}
\xi^\sigma_r=\int_\Gamma \frac{d\mu}{2i sin(\pi(\mu+\frac{r}{6}))}
\frac{(\Gamma(-\mu))^2}{\Gamma(-\mu+\frac{1}{6})\Gamma(-\mu+\frac{5}{6})}\rho^{-6\mu}U^\sigma_\mu(\phi),$$ where the contour $\Gamma$ goes around the Im$(\mu)<0$ axis. To deform the contour to a contour $\Gamma^\prime$ going around the Im$(\mu)>0$ axis, one sees that one can do so for $\sigma=0$ but not for $\sigma=1,2$. For the latter, one modifies $U^\sigma_\mu(\phi)$ by adding a function which does not contribute to the poles and has simple zeros at integers as follows: $$\label{eq:confl23}
U^\sigma_\mu(\phi)\rightarrow\tilde{U^\sigma_{\mu,r}(\phi)}\equiv U^\sigma_\mu(\phi) - e^{\frac{i\pi r}{6}}\frac{sin(\pi(\mu+\frac{r}{6}))}{sin(\pi\mu)}f^\sigma_\mu(\phi),$$ where $$\begin{aligned}
\label{eq:confl231}
& & f^0_\mu(\phi)=0,\nonumber\\
& & f^1_\mu(\phi)=-(1-e^{-2i\pi\nu})y^0_\mu(\phi),\nonumber\\
& & f^2_\mu(\phi)=(1-e^{-2i\pi\nu})V_\nu(\phi) + (1-e^{-2i\pi\nu})W_\mu(\phi).\end{aligned}$$ One can then deform the contour $\Gamma$ to the contour $\Gamma^\prime$ to evaluate the periods. This is done in appendix A.
Expanding about $\phi=\omega^{-1}$, and a large $\rho=\rho_0$, one gets the following periods: $$\label{eq:confl210}
\left(\begin{array}{c}
\varpi_0\\
\varpi_1\\
\varpi_2\\
\varpi_3\\
\varpi_4\\
\varpi_5\\
\end{array}\right)=\left(\begin{array}{c}
A^\prime_0 + B^\prime_{01}x + C^\prime_0(\rho - \rho_0)\\
A^\prime_1 + B^\prime_{11}x + C^\prime_1(\rho - \rho_0)\\
A^\prime_2 + B^\prime_{21}x + C^\prime_2(\rho - \rho_0)\\
A^\prime_3 + B^\prime_{31}x + B^\prime_{32}\ x\ ln x + C^\prime_3(\rho - \rho_0)\\
A^\prime_4 + B^\prime_{41}x + B^\prime_{42}\ x\ ln x + C^\prime_4(\rho - \rho_0)\\
A^\prime_5 + B^\prime_{51}x + B^\prime_{52}\ x\ ln x + C^\prime_5(\rho - \rho_0)\\
\end{array}\right),$$ where $x\equiv(\phi - \omega^{-1})$.
The equations (\[eq:confl24\]) and (\[eq:confl210\]) will get used to arrive at (\[eq:Wconfl11\]) and finally (\[eq:Wconfl161\]) and (\[eq:Wconfl17\]).
The Picard-Fuchs basis of periods evaluated above can be transformed to a symplectic basis as under (See [@Candelasetal]): $$\label{eq:PFbasis1}
\Pi=\left(\begin{array}{c}
F_0\\
F_1\\
F_2\\
X^0\\
X^1\\
X^2
\end{array}\right) = M \left(\begin{array}{c}
\varpi_0\\
\varpi_1\\
\varpi_2\\
\varpi_3\\
\varpi_4\\
\varpi_4
\end{array}\right),$$ where $$\label{eq:PFbasis2}
M=\left(\begin{array}{cccccc}
-1&1&0&0&0&0\\
1&3&3&2&1&0\\
0&1&1&1&0&0\\
1&0&0&0&0&0\\
-1&0&0&1&0&0\\
2&0&0&-2&1&1
\end{array}\right).$$ In the next section, we use information about the periods evaluated in this section, in looking for “area codes".
Extremization of the Superpotential and Existence of “Area Codes"
=================================================================
In this section, we argue the existence of area codes, i.e., points in the moduli space close to and away from the two singular conifold loci that are finitely separated where for the same large values (and hence not necessarily integral) of RR and NS-NS fluxes, one can extremize the (complex structure and axion-dilaton) superpotential (for different values of the complex structure and axion-dilaton moduli)[^6].
The axion-dilaton modulus $\tau$ gets stabilized (from $D_\tau W_{c.s.}=0$, $W_{c.s.}$ being the Gukov-Vafa-Witten complex structure superpotential $\int (F_3 - \tau H_3)\wedge\Omega=(2\pi)^2\alpha^\prime(f-\tau h)\cdot\Pi$, $F_3$ and $H_3$ being respectively the NS-NS and RR three-form field strengths, and are given by: $F_3=(2\pi)^2\alpha^\prime\sum_{a=1}^3(f_a\beta_a + f_{a+3}\alpha_a)$ and $H_3=(2\pi)^2\alpha^\prime\sum_{a=1}^3(h_a\beta_a+h_{a+3}\alpha_a)$; $\alpha_a,\beta^a$, $a=1,2,3$, form an integral cohomology basis) at a value given by: $$\label{eq:taufix1}
\tau=\frac{f^T.{\bar\Pi_0}}{h^T.{\bar\Pi_0}},$$ where $f$ and $h$ are the fluxes corresponding to the NS-NS and RR fluxes; it is understood that the complex structure moduli appearing in (\[eq:taufix1\]) are already fixed from $D_iW=0,\ i=1,2$.
- $\underline{{\rm Near\ the\ conifold\ locus:}\ \phi^3=1,\ {\rm Large}\ \psi}$
The period vector in the symplectic basis, is given by: $$\begin{aligned}
\label{eq:Wconfl11}
& & \Pi=\nonumber\\
& & \hskip - 2.8cm\left(\begin{array}{c}
-A^\prime_0 + A^\prime_1 + (C^\prime_1 - C^\prime_0) (\rho - \rho_0) + (B^\prime_{11} - B^\prime_{01}) x \\
(A^\prime_0 + A^\prime_4 + 3 A^\prime_1 + 3 A^\prime_2 + 2A^\prime_3) + (C^\prime_0 + C^\prime_4) (\rho - \rho_0) + (B^\prime_{01} + B^\prime_{41} + 3 B^\prime_{11} + 3B^\prime_{21} + 2 B^\prime_{31})x + (3 B^\prime_{42} + 2 B^\prime_{32}) x ln x \\
A^\prime_1 + A^\prime_2 + A^\prime_3 + (B^\prime_{11} + B^\prime_{21} + B^\prime_{31}) x + (C^\prime_1 + C^\prime_2 + C^\prime_3) (\rho - \rho_0) + B^\prime_{32} x ln x \\
A^\prime_0 + C^\prime_{01} x + C^\prime_0 (\rho - \rho_0) \\
- A^\prime_0 + C^\prime_{01} x + C^\prime_0 (\rho - \rho_0) \\
- A^\prime_0 + A^\prime_3 + (B^\prime_{31} - B^\prime_{01})x + (C^\prime_3 - C^\prime_0) (\rho - \rho_0) + B^\prime_{32} x ln x\\
2 A^\prime_0 - 2 A^\prime_3 + A^\prime_4 + A^\prime_5 + (2 B^\prime_{01} - 2 B^\prime_{31} + B^\prime_{41} + B^\prime_{51})x + (2C^\prime_0 - 2 C^\prime_3 + C^\prime_4 + C^\prime_5) (\rho - \rho_0) + (-2 B^\prime_{32} + B^\prime_{42} + B^\prime_{52}) x ln x
\end{array}\right)\nonumber\\
& & \equiv \left(\begin{array}{c} A_0 + B_{01} x + C_0 (\rho - \rho_0) \\ A_1 + B_{11} x + B_{12} x ln x + C_1 (\rho - \rho_0) \\ A_2 + B_{21} x + B_{22} x ln x + C_2 (\rho - \rho_0) \\ A_3 + B_{31} x + C_3 (\rho - \rho_0) \\ A_4 + B_{41} x + B_{42} x ln x + C_4 (\rho - \rho_0)\\ A_5 + B_{51} x + B_{52} x ln x + C_5 (\rho - \rho_0) \end{array} \right).\end{aligned}$$ The tree-level Kähler potential is given by: $$\label{eq:Wconfl12}
K = -ln\left(-i(\tau - {\bar\tau})\right) - ln\left(-i\Pi^\dagger\Sigma\Pi\right),$$ where the symplectic metric $\Sigma=\left(\begin{array}{cc} 0 & {\bf 1}_3\\ -{\bf 1}_3 & 0 \end{array}\right)$. Near $x=0$, one can evaluate $\partial_xK, \tau$ and $\partial_xW_{c.s.}$ - this is done in appendix B.
Using (\[eq:Wconfl11\]) - (\[eq:Wconfl12\]) and (\[eq:Wconfl13\])-(\[eq:Xisdefs\]), one gets the following (near $x=0,\rho - \rho_0=0$): $$\label{eq:Wconfl161}
D_xW_{c.s.}\approx ln x\Biggl({\cal A}_1 + {\cal B}_1x + {\cal C}_1 x ln x + {\cal D}_1 (\rho - \rho_0)
+ {\cal B}_1^\prime{\bar x} + {\cal C}_1^\prime{\bar x} ln {\bar x} + {\cal D}_1^\prime ({\bar\rho} - {\bar\rho_0})\Biggr)=0.$$ Similarly, $$\label{eq:Wconfl17}
D_{\rho - \rho_0}W_{c.s.}\approx {\cal A}_2 + {\cal B}_2 x + {\cal C}_2 x ln x + {\cal D}_2 (\rho - \rho_0)
+ {\cal B}_2^\prime{\bar x} + {\cal C}_2^\prime{\bar x} ln {\bar x} + {\cal D}_2^\prime ({\bar\rho} - {\bar\rho_0})=0.$$
- $\underline{{\rm Near}\ \rho^6 + \phi= 1}$
Near $y\equiv \rho^6 + \rho - 1 =0$ and a small $\rho=\rho_0^\prime$, one can follow a similar analysis as (\[eq:Wconfl11\] - (\[eq:Wconfl17\]) and arrive at similar equations: $$\begin{aligned}
\label{eq:Wconfl21}
& & D_yW_{c.s.}\approx ln y\Biggl({\cal A}_3 + {\cal B}_3y + {\cal C}_3 y ln y + {\cal D}_3 (\rho - \rho_0^\prime)
+ {\cal B}_3^\prime{\bar y} + {\cal C}_3^\prime{\bar y} ln {\bar y} + {\cal D}_3^\prime ({\bar\rho} - {\bar\rho_0^\prime})\Biggr)=0,\nonumber\\
& & D_{\rho - \rho_0^\prime}W_{c.s.}\approx {\cal A}_4 + {\cal B}_4 y + {\cal C}_4 y ln y + {\cal D}_4 (\rho - \rho_0^\prime)
+ {\cal B}_4^\prime{\bar y} + {\cal C}_4^\prime{\bar y} ln {\bar y} + {\cal D}_4^\prime ({\bar\rho} - {\bar\rho_0^\prime})=0.\end{aligned}$$
- $\underline{\rm Points\ away\ from\ both\ conifold\ loci}$
It can be shown, again following an analysis similar to the one carried out in (\[eq:Wconfl11\]) - (\[eq:Wconfl21\]), that one gets the following set of equations from extremization of the complex-structure moduli superpotential: $$\label{eq:Wawayconfl}
{\cal A}_i + {\cal B}_i\psi + {\cal C}_i\phi + {\cal B^\prime}_i{\bar\psi} + {\cal C^\prime}_i{\bar\phi} = 0,$$ where $i$ indexes the different regions in the moduli space away from the two conifold loci, as discussed in section [**2**]{} earlier.
Therefore, to summarize, $$\begin{aligned}
\label{eq:DW=0}
& & \underline{{\rm Near}\ \phi=\omega^{-1}}:\nonumber\\
& & {\cal A}_1 + {\cal B}_1 (\phi_1 - \omega^{-1}) + {\cal C}_1 (\phi_1 - \omega^{-1}) ln (\phi_1 - \omega^{-1}) + {\cal D}_1 (\rho_1 - \rho_0)\nonumber\\
& & + {\cal B}_1^\prime({\bar\phi_1} - \omega) + {\cal C}_1^\prime({\bar\phi} - \omega) ln ({\bar\phi} - \omega) + {\cal D}_1^\prime ({\bar\rho_1} - {\bar\rho_0})=0,\nonumber\\
& & {\cal A}_2 + {\cal B}_2 (\phi_1 - \omega^{-1}) + {\cal C}_2 (\phi_1 - \omega^{-1})
ln (\phi_1 - \omega^{-1}) + {\cal D}_2 (\rho_1 - \rho_0)\nonumber\\
& & + {\cal B}_2^\prime({\bar\phi_1} - \omega) + {\cal C}_2^\prime({\bar\phi_1} - \omega) ln ({\bar\phi_1} - \omega) + {\cal D}_2^\prime ({\bar\rho_1} - {\bar\rho_0})=0,\nonumber\\
& & \tau_1=\frac{\Xi[f_i;{\bar\phi_1} - \omega,\rho_1 - \rho_0]}{\sum_{i=0}^5h_i{\bar A_i}}\left[1 - \frac{\Xi[h_i;{\bar\phi_1} - \omega,\rho_1 - \rho_0]}{\sum_{j=0}^5h_i{\bar A_i}}\right];\nonumber\\
& & \underline{{\rm Near}\ \rho^6 + \phi - 1 = 0}:\nonumber\\
& & {\cal A}_3 + {\cal B}_3 (\rho_2^6+\phi-1) + {\cal C}_3 (\rho_2^6+\phi_2-1) ln (\rho_2^6+\phi_2-1) + {\cal D}_3 \phi_2
+ {\cal B}_3^\prime({\bar\rho_2^6}+{\bar\phi}-1)\nonumber\\
& & + {\cal C}_3^\prime({\bar\rho_2^6}+{\bar\phi}-1) ln ({\bar\rho_2^6}+{\bar\phi}-1) + {\cal D}_3^\prime{\bar\phi}=0,\nonumber\\
& & {\cal A}_4 + {\cal B}_4 (\rho_2^6+\phi_2-1) + {\cal C}_4 (\rho_2^6+\phi_2-1)
ln (\rho_2^6+\phi_2-1) + {\cal D}_4 \phi_2\nonumber\\
& & + {\cal B}_4^\prime({\bar\rho}^6-{\bar\phi}-1) + {\cal C}_4^\prime({\bar\rho}^6+{\bar\phi}-1) ln ({\bar\rho}^6+{\bar\phi}-1)+ {\cal D}_4^\prime {\bar\phi_2}=0,\nonumber\\
& & \tau_2=\frac{\Xi[f_i;{\bar\rho}^6+{\bar\phi}-1,\phi_2 ]}{\sum_{i=0}^5h_i{\bar A_i^\prime}}\left[1 - \frac{\Xi[h_i;{\bar\rho}^6+{\bar\phi}-1,\phi_2 ]}{\sum_{j=0}^5h_i{\bar A_i^\prime}}\right]\nonumber\\
& & \underline{|\phi^3|<1,\ {\rm Large}\ \psi}:\nonumber\\
& & {\cal A}_5 + {\cal B}_5(\phi_3 - \phi_0^{\prime\prime}) + {\cal C}_5(\rho_3 - \rho_0^{\prime\prime})
{\cal B}^\prime_5({\bar\phi}_3 - {\bar\phi}_0^{\prime\prime})
+ {\cal C}^\prime_5({\bar\rho}_3 - {\bar\rho}_0^{\prime\prime})= 0,\nonumber\\
& & {\cal A}_6 + {\cal B}_6(\phi_3 - \phi_0^{\prime\prime}) + {\cal C}_6(\rho_3 - \rho_0^{\prime\prime})
+ {\cal B}^\prime_6({\bar\phi}_3 - {\bar\phi}_0^{\prime\prime})
+ {\cal C}^\prime_6({\bar\rho}_3 - {\bar\rho}_0^{\prime\prime}) = 0,\nonumber\\
& & \tau_3=\frac{\tilde{\Xi}[f_i;{\bar\phi_3},\rho_3 ]}{\sum_{i=0}^5h_i{\bar A_i^{\prime\prime}}}\left[1 - \frac{\tilde{\Xi}[h_i;{\bar\phi_3},\rho_3]}{\sum_{j=0}^5h_i{\bar A_i^{\prime\prime}}}\right];\nonumber\\
& & \underline{\left|\frac{\rho^6}{\phi - \omega^{0,-1,-2}}\right|<1}:\nonumber\\
& & {\cal A}_7 + {\cal B}_7(\phi_4 - \phi_0^{\prime\prime\prime}) + {\cal C}_7(\rho_4 - \rho_0^{\prime\prime\prime})
+ {\cal B}^\prime_7({\bar\phi}_4 - {\bar\phi}_0^{\prime\prime\prime}) + {\cal C}^\prime_7({\bar\rho}_4 - {\bar\rho}_0^{\prime\prime\prime}) = 0,\nonumber\\
& & {\cal A}_7 + {\cal B}_7(\phi_4 - \phi_0^{\prime\prime\prime}) + {\cal C}_7(\rho_4 - \rho_0^{\prime\prime\prime})
+ + {\cal B}^\prime_7({\bar\phi}_4 - {\bar\phi}_0^{\prime\prime\prime})
+ {\cal C}^\prime_7({\bar\rho}_4 - {\bar\rho}_0^{\prime\prime\prime}) = 0,\nonumber\\
& & \tau_4=\frac{\tilde{\Xi}[f_i;{\bar\phi_4},\rho_4 ]}{\sum_{i=0}^5h_i{\bar A_i^{\prime\prime\prime}}}\left[1 - \frac{\tilde{\Xi}[h_i;{\bar\phi_4},\rho_4]}{\sum_{j=0}^5h_i{\bar A_i^{\prime\prime\prime}}}\right],\nonumber\\\end{aligned}$$ where on deleting the $ln$ terms in $\Xi$ one gets the form of $\tilde{\Xi}$ in (\[eq:DW=0\]). Given that the Euler characteristic of the elliptically-fibered Calabi-Yau four-fold to which, according to the Sen’s construction [@SenFIIBorient], the orientifold of the Calabi-Yau three-fold of (\[eq:hypersurface\]) corresponds to, will be very large[^7], and further assuming the absence of $D3$-branes, this would imply that one is allowed to take a large value of $f^T.\Sigma.h$, and hence the fluxes - therefore, similar to the philosophy of [@VafaInverse], we would disregard the integrality of fluxes. Without doing the numerics, we will now give a plausibility argument about the existence of solution to any one of the four sets of equations in (\[eq:DW=0\]). As one can drop $x$ as compared to $x ln x$ for $x\sim0$, the equations in (\[eq:DW=0\]) pair off either as:
- Near either of the two conifold loci: $$\begin{aligned}
\label{eq:DWsimp1}
& & A_i + (B_i cos \alpha_i + B_i^\prime sin \alpha_i) \epsilon_i ln \epsilon_i + C_i\beta_i + C_i^\prime {\bar\beta_i} = 0,\nonumber\\
& & \tilde{a}_i + (\tilde{b}_i cos \alpha_i + \tilde{b}_i^\prime sin \alpha_i) \epsilon_i ln \epsilon_i + \tilde{c}_i\beta_i + \tilde{c}_i^\prime {\bar\beta_i} = 0,\end{aligned}$$ or
- Away from both the conifold loci: $$\begin{aligned}
\label{eq:DWsimp2}
& & A_i + B_i \gamma_i + C_i\delta_i + B_i^\prime {\bar\gamma}_i + C_i^\prime {\bar\delta}_i = 0,\nonumber\\
& & \tilde{A}_i + \tilde{B}_i \gamma_i + \tilde{C}_i\delta_i + \tilde{B}_i^\prime {\bar\gamma}_i
+ \tilde{C}_i^\prime {\bar\delta}_i = 0,\end{aligned}$$ where $\epsilon_i,\alpha_i$ correspond to the magnitude and phase of the extremum values of either $\phi - \omega^{-1}$ or $\rho^6 + \phi - 1$, and $\gamma_i,\delta_i$ are different (functions of) extremum values of $\phi,\psi$ near and away, respectively, from the two conifold loci, and both sets are understood to be “close to zero" each.
From the point of view of practical calculations, let us rewrite, e.g., (\[eq:DWsimp1\]) as the equivalent four real equations: $$\begin{aligned}
\label{eq:DWsimp12}
& & {\cal A}_i + {\cal B}_i cos\alpha_i \epsilon_i ln\epsilon_i + {\cal B}_i^\prime sin\alpha_i \epsilon_i ln\epsilon_i + {\cal C}_i Re(\beta_i) + {\cal C}_i^\prime Im(\beta_i) = 0,\nonumber\\
& & \widetilde{{\cal A}_i} + \widetilde{{\cal B}_i} cos\alpha_i \epsilon_i ln\epsilon_i +
\widetilde{{\cal B}_i}^\prime sin\alpha_i \epsilon_i ln\epsilon_i + \widetilde{{\cal C}_i} Re(\beta_i) + \widetilde{\cal C}_i^\prime Im(\beta_i) = 0,\nonumber\\
& & {\nu}_i + {\chi}_i cos\alpha_i \epsilon_i ln\epsilon_i + {\chi}_i^\prime sin\alpha_i \epsilon_i ln\epsilon_i + {\vartheta}_i Re(\beta_i) + {\vartheta}^\prime Im(\beta_i) = 0,\nonumber\\
& & \widetilde{{\nu}_i} + \widetilde{{\chi}_i} cos\alpha_i \epsilon_i ln\epsilon_i +
\widetilde{{\chi}_i}^\prime sin\alpha_i \epsilon_i ln\epsilon_i + \widetilde{{\vartheta}_i} Re(\beta_i) + \widetilde{\vartheta}_i^\prime Im(\beta_i) = 0.
\end{aligned}$$ In (\[eq:DW=0\]), by “close to zero", what we would be admitting are, e.g., $\epsilon_i,|\beta_i|\sim e^{-5}\approx 7\times10^{-3}$ implying that $\epsilon_i ln\epsilon_i\approx10^{-2}$. Let us choose the moduli-independent constants in (\[eq:DWsimp12\]), after suitable rationalization, to be $7\times {\cal O}(1)$, the coefficients of the $\epsilon_i ln\epsilon_i$-terms to be $7\times10^2$ and the coefficients of $Re(\beta_i)$ and $Im(\beta_i)$ to be $\sim10^3$. On similar lines, for (\[eq:DWsimp2\]), we could take the moduli to be $\sim e^{-5}$ and the moduli-independent and moduli-dependent constants to be $7\times {\cal O}(1)$ and $\sim10^3$ respectively. Now, the constants appearing in (\[eq:DWsimp12\]) (and therefore (\[eq:DW=0\])) are cubic in the fluxes (more precisely, they are of the type $h^2f$ in obvious notations), which for (\[eq:hypersurface\]) would be $\sim10^3$ (See [@Denef+Douglas+Florea]). In other words, [*for the same choice of the NS-NS and RR fluxes*]{} - 12 in number - one gets 6 or 9 or 12 complex (inhomogenous \[in $\psi,\phi$\] algebraic/transcendetal) constraints (coming from (\[eq:DW=0\])) on the 6 or 9 or 12 extremum values of the complex structure moduli ($\phi_i,\psi_i,\tau_i;\ i=1,2,3,4$) finitely separated from each other in the moduli space. In principle, as long as one keeps $f^T.\Sigma.h$ fixed, one should be able to tune the fluxes $f_i,h_i;\ i=0,...,5$ to be able to solve these equations. Therefore, the expected estimates of the values of the constants and the moduli tuned by the algebraic-geometric inputs of the periods in the different regions of the moduli space as discussed in section [**2**]{}, are reasonable implying the possibility of existence of “area codes", and the interpolating domain walls [@CDGKL]. Of course, complete numerical calculations, which will be quite involved, will be needed to see explicitly everything working out.
Non-supersymmetric dS minimum via Non-perturbative $\alpha^\prime$- and Instanton Corrections
=============================================================================================
In this section, using the results of [@Grimm], we show that after inclusion of non-perturbative $\alpha^\prime$-corrections to the Kähler potential, in addition to the perturbative $\alpha^\prime$ corrections of [@BBHL], as well as the non-perturbative instanton contributions to the superpotential, it may be possible to obtain a large volume non-supersymmetric dS minimum (analogous to [@Balaetal2] for the non-supersymmetric AdS minimum) [*without the addition of $\overline{D3}$-branes*]{} - see also [@westphal].
Let us begin with a summary of the inclusion of perturbative $\alpha^\prime$-corrections to the Kähler potential in type IIB string theory compactified on Calabi-Yau three-folds with NS-NS and RR fluxes turned on, as discussed in [@BBHL]. The $(\alpha^\prime)^3-$ corrections contributing to the Kähler moduli space metric are contained in $$\label{eq:nonpert1}
\int d^{10}x\sqrt{g}e^{-2\phi}\left(R + (\partial\phi)^2 + (\alpha^\prime)^3\frac{\zeta(3) J_0}{3.2^{11}}
+ (\alpha^\prime)^3(\bigtriangledown^2\phi) Q\right),$$ where $$\begin{aligned}
\label{eq:nonpert2}
& & J_0\equiv t^{M_1N_1...M_4N_4}t_{M_1^\prime N_1^\prime....M_4^\prime N_4^\prime}
R^{M_1N_1}_{\ \ \ \ \ M_1^\prime N_1^\prime}...R^{M_4N_4}_{\ \ \ \ \ M_4^\prime N_4^\prime}\nonumber\\
& & + \frac{1}{4}\epsilon^{ABM_1N_1...M_4N_4}\epsilon_{ABM_1^\prime N_1^\prime...M_4^\prime N_4^\prime}
R^{M_1^\prime N_1^\prime}_{\ \ \ \ \ M_1 N_1}...R^{M_4^\prime N_4^\prime}_{\ \ \ \ \ M_4 N_4},\end{aligned}$$ the second term in (\[eq:nonpert2\]) being the ten-dimensional generalization of the eight-dimensional Euler density, and $$\begin{aligned}
\label{eq:nonpert3}
& & t^{IJKLMNPQ}\equiv-\frac{1}{2}\epsilon^{IJKLMNPQ} -\frac{1}{2}\biggl[(\delta^{IK}\delta^{JL}
-\delta^{IL}\delta^{JK})(\delta^{MP}\delta^{NQ}-\delta^{MQ}\delta^{NP})\nonumber\\
& & +(\delta^{KM}\delta^{LN}-\delta^{KN}\delta^{LM})(\delta^{PI}\delta^{QJ}-\delta^{PJ}\delta^{QI})
+(\delta^{IM}\delta^{JN}-\delta^{IN}\delta^{JM})(\delta^{KP}\delta^{LQ}-\delta^{KQ}\delta^{LP})
\biggr]\nonumber\\
& & +\frac{1}{2}\left[\delta^{JK}\delta^{KM}\delta^{NP}\delta^{QI}+\delta^{JM}\delta^{NK}\delta^{LP}\delta^{QI}
+\delta^{JM}\delta^{NP}\delta^{QK}\delta^{KI}\right]\nonumber\\
& & {\rm +\ 45\ terms\ obtained\ by\ antisymmetrization\ w.r.t.}\ (ij),(kl),(mn),(pq),\end{aligned}$$ and $$\label{eq:nonpert31}
Q\equiv\frac{1}{12(2\pi)^3}\biggl(R_{IJ}^{\ \ }R_{KL}^{\ \ MN}R_{MN}^{\ \ IJ}
-2 R_{I\ K}^{\ K\ L}R_{K\ L}^{\ M\ N}R_{M\ N}^{\ I\ J}\biggr).$$ The perturbative world-sheet corrections to the hypermultiplet moduli space of Calabi-Yau three-fold compactifications of type II theories are captured by the prepotential: $$\label{eq:nonpert4}
F(X)=\frac{i}{3}\kappa_{abc}\frac{X^aX^bX^c}{X^0} + (X^0)^2
\xi,$$ where the $(\alpha^\prime)^3$-corrections are contained in $\xi\equiv-(\alpha^\prime)^3\frac{\chi(CY_3)\zeta(3)}{2}$, $\kappa_{abc}$ being the classical $CY_3$ intersection numbers. Substituting (\[eq:nonpert4\]) in $K=-ln\left[X^i{\bar F}_i + {\bar X}^iF_i\right]$ gives: $$\label{eq:nonpert5}
K = - ln\left[-\frac{i}{6}(z^a - {\bar z}^a)(z^b - {\bar z}^b)(z^c - {\bar z}^c) + 4\xi\right].$$ Truncation of ${\cal N}=2$ to ${\cal N}=1$, implying reduction of the quaternionic geometry to Kähler geometry, corresponds to a Kähler metric which becomes manifest in Kähler coordinates: $T^a=\frac{1}{3}g^a+i\hat{V}^a,
\tau=l+ie^{-\phi_0}$, the hat denoting the Einstein’s frame in which, e.g., $\hat{V}_a=e^{\phi_0}\left(\frac{1}{6}\kappa_{abc}v^bv^c\right)$, $v^a$ being the Kähler moduli, and the Kähler potential is given by: $$\label{eq:nonpert6}
K = - ln\left(-(\tau-{\bar\tau})\right) - 2 ln\left(\hat{{\cal V}} + \frac{1}{2}\xi e^{{-3\phi_0}{2}}\right) -
ln\left(-i\int_{CY_3}\Omega\wedge{\bar\Omega}\right),$$ substituting which into the ${\cal N}=1$ potential $V = e^K\left(g^{i{\bar j}}D_iW{\bar D}_{\bar j}{\bar W} - 3 |W|^2\right)$ (one sums over all the moduli), one gets: $$\begin{aligned}
\label{eq:nonpert7}
& & V = e^K\Biggl[(G^{-1})^{\alpha{\bar\beta}}D_\alpha W D_{\bar\beta}{\bar W} + (G^{-1})^{\tau{\bar\tau}}D_\tau W
D_{\bar\tau}{\bar W}
- \frac{9\hat{\xi}\hat{{\cal V}}e^{-\phi_0}}{(\hat{\xi}-\hat{{\cal V}})(\hat{\xi} + 2\hat{{\cal V}})}(W{\bar D}_{\bar\tau}{\bar W}
+ {\bar W}D_\tau W)\nonumber\\
& & -3\hat{\xi}\frac{((\hat{\xi}) ^2 + 7\hat{\xi}\hat{{\cal V}} + (\hat{{\cal V}})^2)}{(\hat{\xi}-\hat{{\cal V}})
(\hat{\xi} + 2\hat{{\cal V}})^2}|W|^2\Biggr],\end{aligned}$$ the hats being indicative of the Einstein frame - in our subsequent discussion, we will drop the hats for notational convenience. The structure of the $\alpha^\prime$-corrected potential shows that the no-scale structure is no longer preserved due to explicit dependence of $V$ on $\hat{{\cal V}}$ and the $|W|^2$ term is not cancelled. In what follows, we will be setting $2\pi\alpha^\prime=1$.
The type IIB Calabi-Yau orientifolds containing O3/O7-planes considered involve modding out by $(-)^{F_L}\Omega\sigma$ where ${\cal N}=1$ supersymmetry requires $\sigma$ to be a holomorphic and isometric involution: $\sigma^*(J)=J,\ \sigma^*(\Omega)=-\Omega$. Writing the complexified Kähler form $-B_2+iJ=t^A\omega=-b^a\omega_a+iv^\alpha\omega_\alpha$ where $(\omega_a,\omega_\alpha)$ form canonical bases for ($H^2_-(CY_3,{\bf Z}), H^2_+(CY_3,{\bf Z})$), the $\pm$ subscript indicative of being odd under $\sigma$, one sees that in the large volume limit of $CY_3/\sigma$, contributions from large $t^\alpha=v^\alpha$ are exponentially suppressed, however the contributions from $t^a=-B_a$ are not. Note that it is understood that $a$ indexes the [**real**]{} subspace of [**real**]{} dimensionality $h^{1,1}_-=2$; the [**complexified**]{} Kähler moduli correspond to $H^{1,1}(CY_3)$ with [**complex**]{} dimensionality $h^{1,1}=2$ or equivalently real dimensionality equal to 4. So, even though $G^a=c^a-\tau b^a$ (for real $c^a$ and $b^a$ and complex $\tau$) is complex, the number of $G^a$’s is indexed by $a$ which runs over the real subspace $h^{1,1}_-(CY_3)$[^8] As shown in [@Grimm], based on the $R^4$-correction to the $D=10$ type IIB supergravity action [@Green+Gutperle] and the modular completion of ${\cal N}=2$ quaternionic geometry by summation over all $SL(2,{\bf Z})$ images of world sheet corrections as discussed in [@Llanaetal], the non-perturbative large-volume $\alpha^\prime$-corrections that survive the process of orientifolding of type IIB theories (to yield ${\cal N}=1$) to the Kähler potential is given by (in the Einstein’s frame): $$\begin{aligned}
\label{eq:nonpert8}
& & K = - ln\left(-i(\tau-{\bar\tau})\right) - 2 ln\Biggl[{\cal V} + \frac{\chi}{2}\sum_{m,n\in{\bf Z}^2/(0,0)}
\frac{({\bar\tau}-\tau)^{\frac{3}{2}}}{(2i)^{\frac{3}{2}}|m+n\tau|^3}\nonumber\\
& & - 4\sum_{\beta\in H_2^-(CY_3,{\bf Z})} n^0_\beta\sum_{m,n\in{\bf Z}^2/(0,0)}
\frac{({\bar\tau}-\tau)^{\frac{3}{2}}}{(2i)^{\frac{3}{2}}|m+n\tau|^3}cos\left((n+m\tau)k_a\frac{(G^a-{\bar G}^a)}{\tau - {\bar\tau}}
- mk_aG^a\right)\Biggr],\end{aligned}$$ where $n^0_\beta$ are the genus-0 Gopakumar-Vafa invariants for the curve $\beta$ and $k_a=\int_\beta\omega_a$, , and $G^a=c^a-\tau b^a$, the real RR two-form potential $C_2=C_a\omega^a$ and the real NS-NS two-form potential $B_2=B_a\omega^a$. As pointed out in [@Grimm], in (\[eq:nonpert8\]), one should probably sum over the orbits of the discrete subgroup to which the symmetry group $SL(2,{\bf Z})$ reduces. Its more natural to write out the Kähler potential and the superpotential in terms of the ${\cal N}=1$ coordinates $\tau, G^a$ and $T_\alpha$ where $$\label{eq:nonpert10}
T_\alpha = \frac{i}{2}e^{-\phi_0}\kappa_{\alpha\beta\gamma}v^\beta v^\gamma - (\tilde{\rho}_\alpha - \frac{1}{2}\kappa_{\alpha ab}c^a b^b)
-\frac{1}{2(\tau - {\bar\tau})}\kappa_{\alpha ab}G^a(G^b - {\bar G}^b),$$ where $\tilde{\rho}_\alpha$ being defined via $C_4$(the RR four-form potential)$=\tilde{\rho}_\alpha\tilde{\omega}_\alpha,
\tilde{\omega}_\alpha\in H^4_+(CY_3,{\bf Z})$.
Based on the action for the Euclidean $D3$-brane world volume (denoted by $\Sigma_4$) action\
$iT_{D3}\int_{\Sigma_4} e^{-\phi}\sqrt{g-B_2+F}+T_{D3}\int_{\Sigma_4}e^C\wedge e^{-B_2+F}$, the nonperturbative superpotential coming from a $D3$-brane wrapping a divisor $\Sigma\in H^4(CY_3/\sigma,{\bf Z})$ such that the unit arithmetic genus condition of Witten [@Witten] is satisfied, will be proportional to (See [@Grimm]) $$\label{eq:nonpertW1}
e^{\frac{1}{2}\int_\Sigma e^{-\phi}(-B_2+iJ)^2-i\int_\Sigma(C_4-C_2\wedge B_2+\frac{1}{2}C_0B_2^2}
= e^{iT_\alpha\int_\Sigma\tilde{\omega}_\alpha}\equiv e^{in_\Sigma^\alpha T_\alpha},$$ where $C_{0,2,4}$ are the RR potentials. The prefactor multiplying (\[eq:nonpertW1\]) is assumeto factorize into a function of the ${\cal N}=1$ coordinates $\tau,G^a$ and a function of the other moduli. Based on appropriate transformation properties of the superpotential under the shift symmetry and $\Gamma_S\subset SL(2,{\bf Z})$: $$\begin{aligned}
\label{eq:nonpert12}
& & (i)\ \tau\rightarrow\frac{a\tau + b}{c\tau + d},\nonumber\\
& & \left(\begin{array}{c} C_2\\B_2\end{array}\right)\rightarrow
\left(\begin{array}{cc} a & b \\ c & d \end{array}\right)
\left(\begin{array}{c} C_2\\B_2\end{array}\right),
\nonumber\\
& & G^a\rightarrow\frac{G^a}{(c\tau + d)},\nonumber\\
& & T^\alpha\rightarrow T_\alpha + \frac{c}{2}\frac{\kappa_{\alpha ab}G^a G^b}{(c\tau + d)};\nonumber\\
& & (ii)\ b^a\rightarrow b^a + 2\pi n^a,\nonumber\\
& & G^a\rightarrow G^a - 2\pi\tau n^a,\nonumber\\
& & T_\alpha\rightarrow T_\alpha - 2\pi\kappa_{\alpha ab}n^aG^b + 2\pi^2\tau\kappa_{\alpha ab}n^an^b,\end{aligned}$$ the non-perturbative instanton-corrected superpotential was shown in [@Grimm] to be: $$\label{eq:nonpert9}
W = \int_{CY_3}G_3\wedge\Omega + \sum_{n^\alpha}\frac{\theta_{n^\alpha}(\tau,G)}{f(\eta(\tau))}e^{in^\alpha T_\alpha},$$ where the theta function is given as: $$\label{eq:nonpert11}
\theta_{n^\alpha}(\tau,G)=\sum_{m_a}e^{\frac{i\tau m^2}{2}}e^{in^\alpha G^am_a}.$$ In (\[eq:nonpert11\]), $m^2=C^{ab}m_am_b, C_{ab}=-\kappa_{\alpha^\prime ab}$, $\alpha=\alpha^\prime$ corresponding to that $T_\alpha=T_{\alpha^\prime}$ (for simplicity) that is invariant under (\[eq:nonpert12\]).
Now, for (\[eq:hypersurface\]), as shown in [@Denef+Douglas+Florea], there are two divisors which when uplifted to an elliptically-fibered Calabi-Yau, have a unit arithmetic genus ([@Witten]): $\tau_1\equiv\partial_{t_1}{\cal V}=\frac{t^2_1}{2},\ \tau_2\equiv\partial_{t_2}{\cal V}=\frac{(t_1+6t_2)^2}{2}$. In (\[eq:nonpert10\]), $\rho_1=\tilde{\rho}_1-i\tau_1$ and $\rho_2=\tilde{\rho}_2-i\tau_2$.
To set the notations, the metric corresponding to the Kähler potential in (\[eq:nonpert8\]), will be given as: $$\label{eq:nonpert13}
G_{A{\bar B}}=\left(\begin{array}{cccc}
\partial_{\rho_1}{\bar\partial}_{{\bar\rho_1}}K & \partial_{\rho_1}{\bar\partial}_{{\bar\rho_2}}K
& \partial_{\rho_1}{\bar\partial}_{{\bar G}^1}K
& \partial_{\rho_1}{\bar\partial}_{{\bar G}^2}K\\
\partial_{\rho_2}{\bar\partial}_{{\bar\rho_1}}K & \partial_{\rho_2}{\bar\partial}_{{\bar\rho_2}}K
& \partial_{\rho_2}{\bar\partial}_{{\bar G}^1}K
& \partial_{\rho_2}{\bar\partial}_{{\bar G}^2}K\\
\partial_{G^1}{\bar\partial}_{{\bar\rho}_1}K & \partial_{G^1}{\bar\partial}_{{\bar\rho}_2}K &
\partial_{G^1}{\bar\partial}_{{\bar G}^1}K & \partial_{G^1}{\bar\partial}_{{\bar G}^2}K \\
\partial_{G^2}{\bar\partial}_{{\bar\rho}_1}K & \partial_{G^2}{\bar\partial}_{{\bar\rho}_2}K &
\partial_{G^2}{\bar\partial}_{{\bar G}^2}K & \partial_{G^2}{\bar\partial}_{{\bar G}^2}K
\end{array}\right),$$ where $A\equiv\rho^{1,2},G^{1,2}$. We have taken the involution to be such that $h^{1,1}_-=2$. From the Kähler potential given in (\[eq:nonpert8\]), one can show that the corresponding Kähler metric of (\[eq:nonpert13\]) is given by: $$\begin{aligned}
\label{eq:nonpert131}
& & G_{A{\bar B}} =\nonumber\\
& & \left(\begin{array}{cccc}
\frac{1}{4}\left(\frac{1}{6\sqrt{2}}\frac{1}{\sqrt{{\bar\rho}_1 - \rho_1}{\cal Y}}
+ \frac{1}{18}\frac{({\bar\rho}_1 - \rho_1)}{{\cal Y}^2}\right)
&\frac{1}{144}\left(\frac{\sqrt{({\bar\rho}_1 - \rho_1)({\bar\rho}_2 - \rho_2)}}{{\cal Y}^2}\right)
& \frac{-ie^{-\frac{3\phi_0}{2}}\sqrt{{\bar\rho}_1 - \rho_1}{\cal Z}(\tau)}{6\sqrt{2}{\cal Y}^2}
& \frac{-ie^{-\frac{3\phi_0}{2}}\sqrt{{\bar\rho}_1 - \rho_1}{\cal Z}(\tau)}{6\sqrt{2}{\cal Y}^2} \\
\frac{1}{144}\left(\frac{\sqrt{({\bar\rho}_1 - \rho_1)({\bar\rho}_2 - \rho_2)}}{{\cal Y}^2}\right)
& \frac{1}{4}\left(\frac{1}{6\sqrt{2}}\frac{\sqrt{{\bar\rho}_2 - \rho_2}}{{\cal Y}}
+ \frac{1}{18}\frac{\sqrt{{\bar\rho}_2 - \rho_2}}{{\cal Y}^2}\right)
& \frac{-ie^{-\frac{3\phi_0}{2}}\sqrt{{\bar\rho}_2 - \rho_2}{\cal Z}(\tau)}{6\sqrt{2}{\cal Y}^2}
& \frac{-ie^{-\frac{3\phi_0}{2}}\sqrt{{\bar\rho}_2 - \rho_2}{\cal Z}(\tau)}{6\sqrt{2}{\cal Y}^2} \\
\frac{ik_1e^{-\frac{3\phi_0}{2}}\sqrt{{\bar\rho}_1 - \rho_1}{\cal Z}({\bar\tau})}{6\sqrt{2}{\cal Y}^2}
& \frac{ik_1e^{-\frac{3\phi_0}{2}}\sqrt{{\bar\rho}_2 - \rho_2}{\cal Z}({\bar\tau})}{6\sqrt{2}{\cal Y}^2}
& k_1^2{\cal X}_1 & k_1k_2{\cal X}_1 \\
\frac{ik_2e^{-\frac{3\phi_0}{2}}\sqrt{{\bar\rho}_1 - \rho_1}{\cal Z}({\bar\tau})}{6\sqrt{2}{\cal Y}^2}
& \frac{ik_2e^{-\frac{3\phi_0}{2}}\sqrt{{\bar\rho}_2 - \rho_2}{\cal Z}({\bar\tau})}{6\sqrt{2}{\cal Y}^2}
& k_1k_2 {\cal X}_1 & k_2^2{\cal X}_1
\end{array}\right), \nonumber\\
& &\end{aligned}$$ where $$\begin{aligned}
\label{eq:nonpert14}
& & {\cal Z}(\tau)\equiv \sum_c\sum_{m,n}A_{n,m,n_{k^c}}(\tau) sin(nk.b + mk.c),\nonumber\\
&& {\cal Y}\equiv {\cal V}_E + \frac{\chi}{2}\sum_{m,n\in{\bf Z}^2/(0,0)}
\frac{(\tau - {\bar\tau})^{\frac{3}{2}}}{(2i)^{\frac{3}{2}}|m+n\tau|^3} \nonumber\\
& &
- 4\sum_{\beta\in H_2(CY_3,{\bf Z})}n^0_\beta\sum_{m,n\in{\bf Z}^2/(0,0)}
\frac{(\tau - {\bar\tau})^{\frac{3}{2}}}{(2i)^{\frac{3}{2}}|m+n\tau|^3}cos\left((n+m\tau)k_a\frac{(G^a-{\bar G}^a)}{\tau - {\bar\tau}}
- mk_aG^a\right),\nonumber\\
& & {\cal X}_1\equiv\frac{\sum_c\sum_{m,n\in{\bf Z}^2/(0,0)}e^{-\frac{3\phi_0}{2}}|n+m\tau|^3|
A_{n,m,n_{k^c}}(\tau)|^2cos(nk.b + mk.c)}{{\cal Y}}
\nonumber\\
& & + \frac{|\sum_c\sum_{m,n\in{\bf Z}^2/(0,0)}e^{-\frac{3\phi_0}{2}}|n+m\tau|^3A_{n,m,n_{k^c}}(\tau)sin(nk.b + mk.c)|^2}{{\cal Y}^2},
\nonumber\\
& & A_{n,m,n_{k^c}}(\tau)\equiv \frac{(n+m\tau)n_{k^c}}{|n+m\tau|^3}.\end{aligned}$$
The inverse metric is given as: $$\label{eq:nonpert15}
G^{-1}=\left(\begin{array}{cccc}
(G^{-1})^{\rho_1{\bar\rho_1}} & (G^{-1})^{\rho_1{\bar\rho_2}} & (G^{-1})^{\rho_1{\bar G^1}} & 0 \\
\overline{(G^{-1})^{\rho_1{\bar\rho_2}}} & (G^{-1})^{\rho_2{\bar\rho_2}} & (G^{-1})^{\rho_2{\bar G^1}} & 0 \\
\overline{(G^{-1})^{\rho_1{\bar G^1}}} & \overline{(G^{-1})^{\rho_2{\bar G^1}}} & \frac{1}{(k_1^2-k_2^2){\cal X}_1}
& \frac{k_2}{(k_1k_2^2-k_1^3){\cal X}_1} \\
0 & 0 & \frac{k_2}{(k_1k_2^2-k_1^3){\cal X}_1} & \frac{1}{(k_1^2-k_2^2){\cal X}_1}
\end{array}\right),$$ where the non-zero elements are given in appendix C.
Now, analogous to [@Balaetal2], we will work in the large volume limit: ${\cal V}\rightarrow\infty,
\tau_1\sim ln {\cal V},\ \tau_2\sim {\cal V}^{\frac{2}{3}}$. In this limit, the inverse metric (\[eq:nonpert15\]) simplifies to (we will not be careful about the magnitudes of the numerical factors in the following): $$\label{eq:nonpert16}
G^{-1}\sim\left(\begin{array}{cccc}
-{\cal V}\sqrt{ln {\cal V}} & {\cal V}^{\frac{2}{3}}ln {\cal V}
& \frac{-i{\cal X}ln {\cal V}}{{\cal X}_2}& 0 \\
{\cal V}^{\frac{2}{3}}ln {\cal V} & {\cal V}^{\frac{4}{3}} &
\frac{i{\cal X}{\cal V}^{\frac{2}{3}}}{k_1{\cal X}_2}& 0\\
\frac{i{\cal X} ln {\cal V}}{{\cal X}_2}
& \frac{-i{\cal X}{\cal V}^{\frac{2}{3}}}{k_1{\cal X}_2}
& \frac{1}{(k_1^2-k_2^2){\cal X}_1}& \frac{k_2}{(k_1k_2^2-k_1^3){\cal X}_1} \\
0 & 0 & \frac{k_2}{(k_1k_2^2-k_1^3){\cal X}_1} & \frac{1}{(k_1^2-k_2^2){\cal X}_1}
\end{array}\right),$$ where $${\cal X}_2\equiv \sum_c\sum_{m,n\in{\bf Z}^2/(0,0)}|n+m\tau|^3|
A_{n,m,n_{k^c}}(\tau)|^2cos(nk.b + mk.c).$$ Refer to [@Balaetal2] for discussion on the minus sign in the $\left(G^{-1}\right)^{\rho_1{\bar\rho}_1}$. Having extremized the superpotential w.r.t. the complex structure moduli and the axion-dilaton modulus, the ${\cal N}=1$ potential will be given by: $$\begin{aligned}
\label{eq:nonpert17}
& & V = e^K\Biggl[\sum_{A,B=\rho_\alpha,G^a}\Biggl\{(G^{-1})^{A{\bar B}}\partial_A W_{np}{\bar\partial}_{\bar B}{\bar W_{np}}
+ \left((G^{-1})^{A{\bar B}}(\partial_A K){\bar\partial}_{\bar B}{\bar W_{np}}W + c.c.\right)\Biggr\} \nonumber\\
& & + \left(\sum_{A,B=\rho_\alpha,G^a}(G^{-1})^{A{\bar B}}\partial_A K{\bar\partial}_{\bar B}K - 3\right)|W|^2 + \sum_{\alpha,{\bar\beta}\in{\rm c.s.}}(G^{-1})^{\alpha{\bar\beta}}\partial_{\alpha} K_{c.s.}{\bar\partial}_{\bar\beta}K_{c.s.}|W_{np}|^2
\Biggr],\end{aligned}$$ where the total superpotential $W$ is the sum of the complex structure moduli Gukov-Vafa-Witten superpotential $W_{c.s.}$ and the non-perturbative superpotential $W_{np}$ arising because of instantons (obtained by wrapping of $D3$-branes around the divisors with complexified volumes $\tau_1$ and $\tau_2$).
Now, using: $$\begin{aligned}
\label{eq:nonpert18}
& & \partial_{\rho_\alpha}W=\frac{\theta_{n^\alpha}(\tau,G^a)}{f(\eta(\tau))}e^{in^\alpha T_\alpha}in^\alpha
\left(-\frac{e^{-\phi}}{2}\right),\nonumber\\
& & \hskip -2cm
\partial_{G^a}W=\sum_{n^\alpha}\frac{e^{i\frac{\tau m^2}{2}}}{f(\eta(\tau))}e^{im_aG^an^\alpha}e^{in^\alpha T_\alpha}
\left(im_an^\alpha - in^\alpha\frac{\kappa_{\alpha ab}}{2|\tau-{\bar\tau}|^2}\left[{\bar\tau}(G^b-{\bar G}^b)
+({\bar\tau}G^b-\tau{\bar G}^b) + \frac{(2G^b - {\bar G}^b)}{(\tau-{\bar\tau})}\right]\right),\end{aligned}$$ in the large-volume limit, one forms tables 1, 2 and 3.
One therefore sees from table 1 that the dominant term in $(G^{-1})^{A{\bar B}}\partial_A W_{np}{\bar\partial}_{\bar B}{\bar W_{np}}$ is $(G^{-1})^{\rho_1{\bar\rho}_1}|\partial_{\rho_1}W_{np}|^2$, given by: $$\label{eq:nonpert181}
\frac{{\cal Y}\sqrt{ln {\cal V}}}{{\cal V}^{2n^1e^{-\phi}}}e^{-2\phi}(n^1)^2\left|\frac{\theta_{n^1}(\tau,G)}{f(\eta(\tau))}
\right|^2e^{-2n^1 Im(T_1)}.$$
From table 2 we see that the dominant term in $(G^{-1})^{A{\bar B}}(\partial_A K){\bar\partial}_{\bar B}{\bar W_{np}}W$ is $(G^{-1})^{\rho_1{\bar\rho}_1}\partial_{\rho_1}K{\bar\partial}_{\bar\rho_1}{\bar W}_{np}W+c.c.$, which gives: $$\label{eq:nonpert19}
\frac{W_{c.s.} ln {\cal V}}{{\cal V}^{n^1e^{-\phi}}}\left(\frac{\theta_{n^1}({\bar\tau},{\bar G})}{f(\eta({\bar\tau}))}
\right)e^{-in^1(-\tilde{\rho_1}+\frac{1}{2}\kappa_{1ab}
\frac{{\bar\tau}G^a-\tau{\bar G}^a}{({\bar\tau}-\tau)}\frac{(G^b-{\bar G}^b)}{({\bar\tau}-\tau)} -
\frac{1}{2}\kappa_{1ab}\frac{G^a(G^b-{\bar G}^b)}{(\tau-{\bar\tau})})}+c.c.$$
Note, from table 3, the dominant and the (sub) sub-dominant terms in $(G^{-1})^{A{\bar B}}\partial_A K{\bar\partial}_{\bar B}K|W|^2$, given respectively by $(G^{-1})^{\rho_2{\bar\rho}_2}|\partial_{\rho_2}K|^2|W|^2$ and $\left[(G^{-1})^{\rho_1{\bar\rho}_1}|\partial_{\rho_1}K|^2 + (G^{-1})^{\rho_1{\bar\rho}_2}\partial_{\rho_1}{\bar\partial}_{\bar\rho_2}K + c.c.\right]|W|^2$ are actually of the form: $\left[\frac{{\cal V}}{{\cal V} + \xi}\approx 1-\frac{\xi}{{\cal V}}+{\cal O}\left(\frac{1}{{\cal V}^2}\right)\right]|W|^2$ and $\left[\frac{(ln {\cal V})^{\frac{3}{2}}}{({\cal V}+\xi)} + \frac{(ln {\cal V})^{\frac{3}{2}}}{({\cal V}+\xi)^2}\right]|W|^2$ $\approx\biggl[\frac{(ln {\cal V})^{\frac{3}{2}}}{{\cal V}}-\xi\frac{(ln {\cal V})^{\frac{3}{2}}}{{\cal V}^2}
+ \frac{(ln {\cal V})^{\frac{3}{2}}}{{\cal V}^2} -\xi\frac{(ln {\cal V})^{\frac{3}{2}}}{{\cal V}^3}$ $+{\cal O}\left(\frac{1}{{\cal V}^3}\right)\biggr]|W|^2$ respectively and the $\xi$-independent terms together cancel the “-3" in (\[eq:nonpert16\]). This is just a rederivation of the last term in (\[eq:nonpert7\]). One notes that there are additional terms of ${\cal O}\left(\frac{1}{{\cal V}}\right)$ that one gets from $\Biggl[(G^{-1})^{G^1{\bar G^1}}|\partial_{G^1}K|^2 + (G^{-1})^{G^2{\bar G^2}}|\partial_{G^2}K|^2 +
(G^{-1})^{G^1{\bar G^2}}\partial_{G^1}K{\bar\partial}_{\bar G^2}K\Biggr]|W|^2$, which is given by: $$\label{eq:nonpert20}
\frac{|W|^2}{{\cal V}}\left(\frac{3k_2^2+k_1^2}{k_1^2-k_2^2}\right)
\frac{\left|\sum_c\sum_{n,m\in{\bf Z}^2/(0,0)}e^{-\frac{3\phi}{2}}A_{n,m,n_{k^c}}(\tau) sin(nk.b+mk.c)\right|^2}
{\sum_{c^\prime}\sum_{m^\prime,n^\prime\in{\bf Z}^2/(0,0)} e^{-\frac{3\phi}{2}}|n+m\tau|^3
|A_{n^\prime,m^\prime,n_{k^{c^{\prime}}}}(\tau)|^2 cos(n^\prime k.b+m^\prime k.c)},$$ which one sees [*can be either positive or negative*]{}.
To summarize, from (\[eq:nonpert18\]) - (\[eq:nonpert20\]), one gets the following potential: $$\begin{aligned}
\label{eq:nonpert21}
& & V\sim\frac{{\cal Y}\sqrt{ln {\cal V}}}{{\cal V}^{2n^1+2}}e^{-2\phi}(n^1)^2\frac{\left(\sum_{m^a}e^{-\frac{m^2}{2g_s} + \frac{m_ab^a n^1}{g_s} + \frac{n^1\kappa_{1ab}b^ab^b}{2g_s}}\right)^2}{\left|f(\eta(\tau))\right|^2}
\nonumber\\
& & +\frac{ln {\cal V}}{{\cal V}^{n^1+2}}\left(\frac{\theta_{n^1}({\bar\tau},{\bar G})}{f(\eta({\bar\tau}))}
\right)e^{-in^1(-\tilde{\rho_1}+\frac{1}{2}\kappa_{1ab}
\frac{{\bar\tau}G^a-\tau{\bar G}^a}{({\bar\tau}-\tau)}\frac{(G^b-{\bar G}^b)}{({\bar\tau}-\tau)} -
\frac{1}{2}\kappa_{1ab}\frac{G^a(G^b-{\bar G}^b)}{(\tau-{\bar\tau})})}+c.c.\nonumber\\
& & +
\frac{|W|^2}{{\cal V}^3}\left(\frac{3k_2^2+k_1^2}{k_1^2-k_2^2}\right)
\frac{\left|\sum_c\sum_{n,m\in{\bf Z}^2/(0,0)}e^{-\frac{3\phi}{2}}A_{n,m,n_{k^c}}(\tau) sin(nk.b+mk.c)\right|^2}
{\sum_{c^\prime}\sum_{m^\prime,n^\prime\in{\bf Z}^2/(0,0)} e^{-\frac{3\phi}{2}}|n+m\tau|^3
|A_{n^\prime,m^\prime,n_{k^{c^{\prime}}}}(\tau)|^2 cos(n^\prime k.b+m^\prime k.c)}+\frac{\xi|W|^2}{{\cal V}^3}.
\nonumber\\
& &\end{aligned}$$ On comparing (\[eq:nonpert21\]) with the analysis of [@Balaetal2], one sees that for generic values of the moduli $\rho_\alpha, G^a, k^{1,2}$ and ${\cal O}(1)\ W_{c.s.}$, [*and $n^1=1$*]{}, analogous to [@Balaetal2], the second term dominates; the third term is a new term. However, as in KKLT scenarios (See [@KKLT]), $W_{c.s.}<<1$; we would henceforth assume that the fluxes and complex structure moduli have been so fine tuned/fixed that $W\sim W_{n.p.}$. Further, from studies related to study of axionic slow roll inflation in Swiss Cheese models [@misra-shukla-II], it becomes necessary to take $n^1>2$. We assume that the fundamental-domain-valued $b^a$’s satisfy: $\frac{|b^a|}{\pi}<<1$[^9]. This implies that the first term in (\[eq:nonpert21\]) - $|\partial_{\rho^1}W_{np}|^2$ - a positive definite term and denoted henceforth by $V_I$, is the most dominant. Hence, if a minimum exists, it will be positive. To evaluate the extremum of the potential, one sees that: $$\begin{aligned}
\label{eq:extrV-c_b}
& & \partial_{c^a}V_I\nonumber\\
& & \hskip -0.3in\sim- 4\frac{\sqrt{ln {\cal V}}}{{\cal V}^{2n^1+2}}\sum_{\beta\in H_2^-(CY_3,{\bf Z})} n^0_\beta\sum_{m,n\in{\bf Z}^2/(0,0)}mk^a
\frac{({\bar\tau}-\tau)^{\frac{3}{2}}}{(2i)^{\frac{3}{2}}|m+n\tau|^3} sin(n k.b + mk.c)\frac{\left(\sum_{m^a}e^{-\frac{m^2}{2g_s} + \frac{m_ab^a n^1}{g_s} + \frac{n^1\kappa_{1ab}b^ab^b}{2g_s}}\right)^2}{\left|f(\eta(\tau))\right|^2}=0\nonumber\\
& & \Leftrightarrow nk.b + mk.c = N\pi;\nonumber\\
& & \hskip-0.23in\partial_{b^a}V_I|_{nk.b + mk.c = N\pi}\sim\frac{{\cal V}\sqrt{ln {\cal V}}}{{\cal V}^{2n^1+1}}\frac{e^{-\frac{m^2}{2g_s} + \frac{m_{a^\prime}b^{a^\prime} n^1}{g_s} + \frac{n^1\kappa_{1a^\prime b^\prime}b^{a^\prime}b^{b^\prime}}{2g_s}}\sum_{m^a}e^{-\frac{m^2}{2g_s} + \frac{m_ab^a n^1}{g_s} + \frac{n^1\kappa_{1ab}b^ab^b}{2g_s}}}{\left|f(\eta(\tau))\right|^2}\left(\frac{n^1m^a}{g_s} + \frac{n^1\kappa_{1ab}b^b}{g_s}\right)=0.\nonumber\\
& & \end{aligned}$$ Now, given the ${\cal O}(1)$ triple-intersection numbers and super sub-Planckian NS-NS axions, we see that potential $V_I$ gets automatically extremized for $D1$-instanton numbers $m^a>>1$. Note that if the NS-NS axions get stabilized as per $\frac{n^1m^a}{g_s} + \frac{n^1\kappa_{1ab}b^b}{g_s}=0$, satisfying $\partial_{b^a}V=0$, this would imply that the NS-NS axions get stabilized at a rational number, and in particular, a value which is not a rational multiple of $\pi$, the same being in conflict with the requirement $nk.b + mk.c = N \pi$. It turns out that the locus $nk.b + mk.c = N\pi$ for $|b^a|<<\pi$ and $|c^a|<<\pi$ corresponds to a flat saddle point with the NS-NS axions providing a flat direction - See [@misra-shukla-II].
Analogous to [@Balaetal2], for all directions in the moduli space with ${\cal O}(1)$ $W_{c.s.}$ and away from $D_iW_{cs}=D_\tau W=0=\partial_{c^a}V=\partial_{b^a}V=0$, the ${\cal O}(\frac{1}{{\cal V}^2})$ contribution of $\sum_{\alpha,{\bar\beta}\in{c.s.}}(G^{-1})^{\alpha{\bar\beta}}D_\alpha W_{cs}{\bar D}_{\bar\beta}{\bar W}_{cs}$ dominates over (\[eq:nonpert21\]), ensuring that that there must exist a minimum, and given the positive definiteness of the potential $V_I$, this will be a dS minimum. There has been no need to add any $\overline{D3}$-branes as in KKLT to generate a dS vacuum. Also, interestingly, one can show that the condition $nk.b + mk.c = N \pi$ gurantees that the slow roll parameters “$\epsilon$" and “$\eta$" are much smaller than one for slow roll inflation beginning from the saddle point and proceeding along an NS-NS axionic flat direction towards the nearest dS minimum (See [@misra-shukla-II]).
The “Inverse Problem" for Extremal Black Holes
==============================================
We now switch gears and address two issues in this and the subsequent sections, related to supersymmetric and non-supersymmetric black hole attractors[^10]. In this section, using the techniques discussed in [@VafaInverse], we explicitly solve the “inverse problem" for extremal black holes in type II compactifications on (the mirror of) (\[eq:hypersurface\]) - given a point in the moduli space, to find the charges $(p^I,q_I)$ that would satisfy $\partial_iV_{BH}=0$, $V_{BH}$ being the black-hole potential. In the next section, we address the issue of existence of “fake superpotentials" in the same context.
We will now summarize the “inverse problem" as discussed in [@VafaInverse]). Consider $D=4, N=2$ supergravity coupled to $n_V$ vector multiplets in the absence of higher derivative terms. The black-hole potential can be written as [@nonsusybh1]: $$\label{eq:BHinv1}
V_{BH} = -\frac{1}{2}(q_I - {\cal N}_{IK} p^K)\left((Im {\cal N})^{-1}\right)^{IJ}(q_J - {\bar{\cal N}}p^L),$$ where the $(n_V + 1)\times(n_V + 1)$ symmetric complex matrix, ${\cal N}_{IJ}$, the vector multiplet moduli space metric, is defined as: $$\label{eq:BHinv2}
{\cal N}_{IJ} \equiv {\bar F}_{IJ} + \frac{2i Im(F_{IK}) X^K Im (F_{IL}) X^L}{Im(F_{MN}) X^M X^N},$$ $X^I,F_J$ being the symplectic sections and $F_{IJ}\equiv\partial_IF_J=\partial_JF_I$. The black-hole potential (\[eq:BHinv1\]) can be rewritten (See [@VafaInverse]) as: $$\label{eq:BHinv3}
\tilde{V}_{BH} = \frac{1}{2}{\cal P}^I Im({\cal N}_{IJ}){\bar{\cal P}}^J - \frac{i}{2}{\cal P}^I(q_I - {\cal N}_{IJ}p^J)
+ \frac{i}{2}{\bar{\cal P}}^I(q_I - {\bar{\cal N}}_{IJ}p^J).$$ The variation of (\[eq:BHinv3\]) w.r.t. ${\cal P}^I$ gives: $$\label{eq:BHinv4}
{\cal P}^I=-i\left((Im {\cal N})^{-1})^{IJ}\right)(q_J - {\cal N}_{IJ}p^J),$$ which when substituted back into (\[eq:BHinv3\]), gives (\[eq:BHinv1\]). From (\[eq:BHinv4\]), one gets: $$\begin{aligned}
\label{eq:BHinv5}
& & p^I = Re({\cal P}^I),\nonumber\\
& & q_I = Re({\cal N}_{IJ}{\cal P}^J).\end{aligned}$$ Extremizing $\tilde{V}_{BH}$ gives: $$\label{eq:BHinv6}
{\cal P}^I{\bar{\cal P}}^J\partial_i Im({\cal N}_{IJ}) + i({\cal P}^I\partial_i {\cal N}_{IJ} - {\bar{\cal P}}^J\partial_i{\bar{\cal N}}_{IJ})p^J = 0,$$ which using (\[eq:BHinv5\]) yields: $$\label{eq:BHinv7}
\partial_i Im({\cal P}^I{\cal N}_{IJ}{\cal P}^J) = 0.$$ Similar to what was done in section [**3**]{}, one uses the semi-classical approximation and disregards the integrality of the electric and magnetic charges taking them to be large.
The inverse problem is not straight forward to define as all sets of charges $(p^I,q_I)$ which are related to each other by an $Sp(2n_V + 2,{\bf Z})$-transformation, correspond to the same point in the moduli space. This is because the $V_{BH})$ (and $\partial_iV_{BH}$) is (are) symplectic invariants. Further, $\partial_iV_{BH}=0$ give $2n_V$ real equations in $2n_V+2$ real variables $(p^I,q_I)$. To fix these two problems, one looks at critical values of $V_{BH}$ in a fixed gauge $W=w\in{\bf C}$. In other words, $$\label{eq:BHinv8}
W=\int_M\Omega\wedge H = q_I X^I - p^I F_I = X^I(q_I - {\cal N}_{IJ}p^J) = w,$$ which using (\[eq:BHinv5\]), gives: $$\label{eq:BHinv9}
X^I Im({\cal N}_{IJ}){\bar{\cal P}}^J = w.$$ Thus, the inverse problem boils down to solving: $$\begin{aligned}
\label{eq:BHinv10}
& & p^I = Re({\cal P}^I),\ q_I=Re({\cal N}_{IJ}{\cal P}^J);\nonumber\\
& & \partial_i({\cal P}^I{\cal N}_{IJ}{\cal P}^J)=0,\ X^I{\cal N}_{IJ}{\bar{\cal P}}^J = iw.\end{aligned}$$ One solves for ${\cal P}^I$s from the last two equations in (\[eq:BHinv10\]) and substitutes the result into the first two equations of (\[eq:BHinv10\]).
We will now solve the last two equations of (\[eq:BHinv10\]) for (\[eq:hypersurface\]). As an example, we work with points in the moduli space close to one of the two conifold loci: $\phi^3=1$. We need to work out the matrix $F_{IJ}$ so that one can work out the matrix ${\cal N}_{IJ}$. From the symmetry of $F_{IJ}$ w.r.t. $I$ and $J$, one sees that the constants appearing in (\[eq:confl210\]) must satisfy some constraints (which must be borne out by actual numerical computations). To summarize, near $x=0$ and using (\[eq:confl24\])-(\[eq:confl210\]): $$\begin{aligned}
\label{eq:BHinv11}
& & F_{01}=F_{10}\Leftrightarrow ln x \frac{B_{01}}{B_{31}} + \frac{C_1}{C_3}
= \frac{B_{01}}{B_{41} + B_{42} (ln x + 1)} + \frac{C_0}{C_4}\Rightarrow B_{12}=0,\ \frac{C_1}{C_3}=\frac{C_0}{C_4};\nonumber\\
& & F_{02}=F_{20}\Leftrightarrow ln x \frac{B_{22}}{B_{31}} + \frac{C_2}{C_3} = \frac{B_{01}}{B_{51} + B_{52} (ln x + 1)} + \frac{C_0}{C_5}\Rightarrow B_{22}=0,\ \frac{C_2}{C_3}=\frac{C_0}{C_5};\nonumber\\
& & F_{12}=F_{21}\Leftrightarrow \frac{B_{22}}{B_{42}} + \frac{C_2}{C_4} = \frac{B_{11}}{B_{51} + B_{52} (ln x + 1)} + \frac{C_1}{C_5}\Rightarrow\frac{C_2}{C_4}=\frac{C_1}{C_5}.\end{aligned}$$ In (\[eq:BHinv11\]), the constants $A_i, B_{ij},C_k$ are related to the constants $A_i, B_{ij}, C_k$ via matrix elements of $M$ of (\[eq:PFbasis2\]). Therefore, one gets the following form of $F_{IJ}$: $$\label{eq:BHinv12}
F_{IJ}=\left(\begin{array}{ccc} \frac{B_{01}}{C_3} + \frac{C_0}{C_3} & \frac{C_1}{C_3} & \frac{C_2}{C_3}\\
\frac{C_1}{C_3} & \frac{C_1}{C_4} & \frac{C_2}{C_4}\\
\frac{C_2}{C_3} & \frac{C_2}{C_4} & \frac{C_2}{C_5}
\end{array}\right)$$ Using (\[eq:BHinv12\]), one can evaluate $X^I Im(F_{IJ}) X^J$ - this is done in appendix D.
Using (\[eq:BHinv12\]), (\[eq:BHinv13\]) - (\[eq:BHinv14\]), one gets: $$\begin{aligned}
\label{eq:BHinv15}
& & {\cal N}=\nonumber\\
& & \hskip -2cm\left(\begin{array}{ccc}
a_{00}+b^{(1)}_{00}x + b^{(2)}_{00} x ln x + c_{00}(\rho-\rho_0) &
a_{01}+b^{(1)}_{01}x + b^{(2)}_{01} x ln x + c_{01}(\rho-\rho_0) &
a_{02}+b^{(1)}_{02}x + b^{(2)}_{02} x ln x + c_{02}(\rho-\rho_0) \\
a_{01}+b^{(1)}_{01}x + b^{(2)}_{01} x ln x + c_{01}(\rho-\rho_0) &
a_{11}+b^{(1)}_{11}x + b^{(2)}_{11} x ln x + c_{11}(\rho-\rho_0) &
a_{12}+b^{(1)}_{12}x + b^{(2)}_{12} x ln x + c_{12}(\rho-\rho_0) \\
a_{02}+b^{(1)}_{02}x + b^{(2)}_{02} x ln x + c_{02}(\rho-\rho_0) &
a_{12}+b^{(1)}_{12}x + b^{(2)}_{12} x ln x + c_{12}(\rho-\rho_0) &
a_{22}+b^{(1)}_{22}x + b^{(2)}_{22} x ln x + c_{22}(\rho-\rho_0)
\end{array}\right).\nonumber\\
& &\end{aligned}$$ The constants $a_{ij}, b^{(1),(2)}_{jk},c_{lm}$ are constrained by relations, e.g., $$F_I={\cal N}_{IJ}X^J,$$ which, e.g., for $I=0$ would imply: $$\begin{aligned}
\label{eq:BHinv16}
& & a_{00} A_3 + a_{01} A_4 + a_{02} A_5 = A_0\nonumber\\
& & a_{00} B_{31} + b^{(1)}_{00} A_3 + a_{01} B_{41} + A_4 b^{(1)}_{01} + a_{02} B_{51} + b^{(1)}_{02} A_5
= B_{01}\nonumber\\
& & b^{(2)}_{00} A_3 + a_{01} B_{42} + b^{(2)}_{01} A_4 + a_{02} B_{52} + A_5 b^{(2)}_{02} = 0\nonumber\\
& & a_{00} C_3 + c_{00} A_3 + a_{01} C_4 + c_{01} A_4 + a_{02} C_5 + c_{02} A_5 = C_0.\end{aligned}$$
So, substituting (\[eq:BHinv15\]) into the last two equations of (\[eq:BHinv10\]), one gets: $$\begin{aligned}
\label{eq:BHinv17}
& & \partial_x({\cal P}^I{\cal N}_{IJ}{\cal P}^J)=0\Rightarrow ln x\left[({\cal P}^0)^2b^{(2)}_{00} + ({\cal P}^1)^2b^{(2)}_{11} + ({\cal P}^2)^2b^{(2)}_{22} + 2{\cal P}^0{\cal P}^1b^{(2)}_{01} + 2{\cal P}^0{\cal P}^2b^{(2)}_{02} + 2{\cal P}^1{\cal P}^2b^{(2)}_{12}\right]=0;\nonumber\\
& & \partial_{\rho-\rho_0}({\cal P}^I{\cal N}_{IJ}{\cal P}^J)=0\Rightarrow({\cal P}^0)^2c^{(2)}_{00} + ({\cal P}^1)^2c_{11} + ({\cal P}^2)^2c_{22} + 2{\cal P}^0{\cal P}^1c_{01} + 2{\cal P}^0{\cal P}^2c_{02} + 2{\cal P}^1{\cal P}^2c_{12}=0,\end{aligned}$$ and ${\bar X}^I Im({\cal N}_{IJ}){\cal P}^J=-iw$ implies: $$\begin{aligned}
\label{eq:BHinv18}
& & {\bar A}_I(a_{IJ}-{\bar a}_{IJ}){\cal P}^J+{\bar x}[{\bar B}_{I1}(a_{IJ} - {\bar a}_{IJ}){\cal P}^J
- {\bar b^{(1)}}_{IJ}{\bar A}_I{\cal P}^J] + x[b^{(1)}_{IJ}{\bar A}_I{\cal P}^J]
+ x ln x[{\bar A}_I b^{(2)}_{IJ}{\cal P}^J] + (\rho-\rho_0)[{\bar A}_I c_{IJ}{\cal P}^J] \nonumber\\
& & + ({\bar\rho} - {\bar\rho_0})[{\bar C}_I(a_{IJ} - {\bar a}_{IJ}){\cal P}^J - {\bar c}_{IJ}A_I{\cal P}^J]
+ {\bar x} ln{\bar x}[B_{I2}a_{IJ}{\cal P}^J]=-2{\bar w}\nonumber\\
& & {\rm or}\nonumber\\
& & \sum_{I=0}^2\Upsilon^I(x,{\bar x}, x ln x, {\bar x} ln {\bar x};\rho-\rho_0,{\bar\rho}-{\bar\rho_0}){\cal P}^I={\bar w}.\end{aligned}$$ Using (\[eq:BHinv18\]), we eliminate ${\cal P}^2$ from (\[eq:BHinv17\]) to get: $$\begin{aligned}
\label{eq:BHinv19}
& & \alpha_1({\cal P}^0)^2 + \beta_1({\cal P}^1)^2 + \gamma_1{\cal P}^0{\cal P}^1 = \lambda_1,\nonumber\\
& & \alpha_2({\cal P}^0)^2 + \beta_2({\cal P}^1)^2 + \gamma_2{\cal P}^0{\cal P}^1 = \lambda_2.\end{aligned}$$ The equations (\[eq:BHinv19\]) can be solved and yield four solutions which are: $$\begin{aligned}
\label{eq:BHinv20}
& &
{\cal P}^0=
\frac{1}{2\,{\sqrt{2}}\,\Biggl( {\alpha_2}\,{\lambda_1} - {\alpha_1}\,{\lambda_2} \Biggr) }\Biggl( {\gamma_2}\,{\lambda_1} - {\gamma_1}\,{\lambda_2} +
\sqrt{Y} \Biggr)\sqrt{X} \,\nonumber\\
& & {{\cal P}^1}=-\frac{\sqrt{X}}{\sqrt{2}};\nonumber\\
& & {\cal P}^0=
-\frac{1}{2\,{\sqrt{2}}\,\Biggl( {\alpha_2}\,{\lambda_1} - {\alpha_1}\,{\lambda_2} \Biggr) }\Biggl( {\gamma_2}\,{\lambda_1} - {\gamma_1}\,{\lambda_2} +
\sqrt{Y} \Biggr)\sqrt{X} \,\nonumber\\
& & {{\cal P}^1}=\frac{\sqrt{X}}{\sqrt{2}};\nonumber\\
& & {\cal P}^0=
\frac{1}{2\,{\sqrt{2}}\,\Biggl( {\alpha_2}\,{\lambda_1} - {\alpha_1}\,{\lambda_2} \Biggr) }\Biggl( {\gamma_2}\,{\lambda_1} - {\gamma_1}\,{\lambda_2} -
\sqrt{Y} \Biggr)\sqrt{X} \,\nonumber\\
& & {{\cal P}^1}=-\frac{\sqrt{X}}{\sqrt{2}};\nonumber\\
& & {\cal P}^0=
-\frac{1}{2\,{\sqrt{2}}\,\Biggl( {\alpha_2}\,{\lambda_1} - {\alpha_1}\,{\lambda_2} \Biggr) }\Biggl( {\gamma_2}\,{\lambda_1} - {\gamma_1}\,{\lambda_2} -
\sqrt{Y} \Biggr)\sqrt{X} \,\nonumber\\
& & {{\cal P}^1}=\frac{\sqrt{X}}{\sqrt{2}}
\end{aligned}$$ where $$\begin{aligned}
\label{eq:BHinv21}
& & X\equiv\frac{1}{{{\alpha_2}}^2\,{{\beta_1}}^2 +
{\alpha_2}\,\Biggl[ -2\,{\alpha_1}\,{\beta_1}\,{\beta_2} +
{\gamma_1}\,\Biggl( {\beta_2}\,{\gamma_1} - {\beta_1}\,{\gamma_2} \Biggr) \Biggr] +
{\alpha_1}\,\Biggl[ {\alpha_1}\,{{\beta_2}}^2 +
{\gamma_2}\,\Biggl( -{\beta_2}\,{\gamma_1} + {\beta_1}\,{\gamma_2} \Biggr)
\Biggr] }\nonumber\\
& & \times\Biggl[2\,{{\alpha_2}}^2\,{\beta_1}\,{\lambda_1} +
{\alpha_1}\,\Biggl( {{\gamma_2}}^2\,{\lambda_1} + 2\,{\alpha_1}\,{\beta_2}\,{\lambda_2} -
{\gamma_2}\,\Biggl( {\gamma_1}\,{\lambda_2} +
{\sqrt{X_1}}
\Biggr) \Biggr)\Biggr],\nonumber\\
& & Y\equiv{{\gamma_2}}^2\,{{\lambda_1}}^2 - 2\,{\gamma_1}\,{\gamma_2}\,{\lambda_1}\,{\lambda_2} +
4\,{\alpha_2}\,{\lambda_1}\,\Biggl( -{\beta_2}\,{\lambda_1} +
{\beta_1}\,{\lambda_2} \Biggr) +
{\lambda_2}\,\Biggl( 4\,{\alpha_1}\,{\beta_2}\,{\lambda_1} -
4\,{\alpha_1}\,{\beta_1}\,{\lambda_2} + {{\gamma_1}}^2\,{\lambda_2} \Biggr), \end{aligned}$$ and $$\begin{aligned}
\label{eq:BHinv22}
& & X_1\equiv Y + {\alpha_2}\,
\Biggl[ -2\,{\alpha_1}\,\Biggl( {\beta_2}\,{\lambda_1} + {\beta_1}\,{\lambda_2} \Biggr) +
{\gamma_1}\,\Biggl( -{\gamma_2}\,{\lambda_1} + {\gamma_1}\,{\lambda_2}\nonumber\\
& & +
\sqrt{{{\gamma_2}}^2\,{{\lambda_1}}^2 -
2\,{\gamma_1}\,{\gamma_2}\,{\lambda_1}\,{\lambda_2} +
4\,{\alpha_2}\,{\lambda_1}\,
\Biggl( -{\beta_2}\,{\lambda_1} + {\beta_1}\,{\lambda_2} \Biggr) +
{\lambda_2}\,\Biggl( 4\,{\alpha_1}\,{\beta_2}\,{\lambda_1} -
4\,{\alpha_1}\,{\beta_1}\,{\lambda_2} + {{\gamma_1}}^2\,{\lambda_2} \Biggr)}\ \Biggr)\Biggr].\nonumber\\
& &
\end{aligned}$$ One can show that one does get ${\cal P}^I\sim X^I$ as one of the solutions - this corresponds to a supersymmetric black hole, and the other solutions correspond to non-supersymmetric black holes.
“Fake Superpotentials"
======================
In this section, using the results of [@Ceresole+Dall'agata], we show the existence of “fake superpotentials" corresponding to black-hole solutions for type II compactification on (\[eq:hypersurface\]).
As argued in [@Ceresole+Dall'agata], dS-curved domain wall solutions in gauged supergravity and non-extremal black hole solutions in Maxwell-Einstein theory have the same effective action. In the context of domain wall solutions, if there exists a ${\cal W}(z^i,{\bar z}^i)\in{\bf R}: V_{DW}(\equiv{\rm Domain\ Wall\ Potential})=-{\cal W}^2 + \frac{4}{3\gamma^2}g^{i{\bar j}}\partial_i{\cal W}\partial_{\bar j}{\cal W}$, $z^i$ being complex scalar fields, then the solution to the second-order equations for domain walls, can also be derived from the following first-order flow equations: $U^\prime=\pm e^U\gamma(r){\cal W};\ (z^i)^\prime = \mp e^U\frac{2}{\gamma^2}g^{i{\bar j}}\partial_{\bar j}{\cal W}$, where $\gamma\equiv\sqrt{1 + \frac{e^{-2U}\Lambda}{{\cal W}^2}}$.
Now, spherically symmetric, charged, static and asymptotically flat black hole solutions of Einstein-Maxwell theory coupled to complex scalar fields have the form: $dz^2 = - e^{2U(r)} dt^2 + e^{-2U(r)}\biggl[\frac{c^4}{sinh^4(cr)} dr^2 $ $+ \frac{c^2}{sinh^2(cr)}(d\theta^2 + sin^2\theta d\phi^2)\biggr]$, where the non-extremality parameter $c$ gets related to the positive cosmological constant $\Lambda>0$ for domain walls. For non-constant scalar fields, only for $c=0$ that corresponds to extremal black holes, one can write down first-order flow equations in terms of a ${\cal W}(z^i,{\bar z}^i)\in{\bf R}$: $U^\prime=\pm e^U{\cal W};\ (z^i)^\prime=\pm2e^U g^{i{\bar j}}\partial_{\bar j}{\cal W},$ and the potential $\tilde{V}_{BH}\equiv {\cal W}^2 + 4g^{i{\bar j}}\partial_i{\cal W}\partial_{\bar j}{\cal W}$ can be compared with the ${\cal N}=2$ supergravity black-hole potential $V_{BH}=|Z|^2+g^{i{\bar j}}D_iZD_{\bar j}{\bar Z}$ by identifying ${\cal W}\equiv|Z|$. For non-supersymmetric theories or supersymmetric theories where the black-hole constraint equation admits multiple solutions which may happen because several ${\cal W}$s may correspond to the same $\tilde{V}_{BH}$ of which only one choice of ${\cal W}$ would correspond to the true central charge, one hence talks about “fake superpotential" or “fake supersymmetry" - a ${\cal W}:\partial_i{\cal W}=0$ would correspond to a stable non-BPS black hole. Defining ${\cal V}\equiv e^{2U}V(z^i,{\bar z}^i),
{\cal\bf W}\equiv e^U{\cal W}(z^i,{\bar z}^i)$, one sees that ${\cal V}(x^A\equiv U,z^i,{\bar z}^i)=g^{AB}\partial_A{\bf W}(x)\partial_B{\bf W}(x)$, where $g_{UU}=1$ and $g_{Ui}=0$. This illustrates the fact that one gets the same potential ${\cal V}(x)$ for all vectors $\partial_A{\cal\bf W}$ with the same norm. In other words, ${\bf W}$ and $\tilde{\bf W}$ defined via: $\partial_A{\bf W}=R_A^{\ B}(z,{\bar z})\partial_B\tilde{\bf W}$ correspond to the same ${\cal V}$ provided: $R^TgR=g$.
For ${\cal N}=2$ supergravity, the black hole potential $V_{BH}=Q^T{\cal M}Q$ where $Q=(p^\Lambda,q_\Lambda)$ is an $Sp(2n_v+2,{\bf Z})$-valued vector ($n_V$ being the number of vector multiplets) and ${\cal M}\in Sp(2n_V+2)$ is given by: $$\label{eq:FakeW1}
{\cal M}=\left(\begin{array}{cc}
A&B\\
C&D
\end{array}\right),$$ where $$\begin{aligned}
\label{eq:FakeW2}
& & A\equiv Re {\cal N} (Im {\cal N})^{-1}\nonumber\\
& & B\equiv-Im {\cal N} - Re {\cal N} (Im {\cal N})^{-1} Re {\cal N} \nonumber\\
& & C\equiv (Im {\cal N})^{-1} \nonumber\\
& & D = -A^T = - (Im {\cal N}^{-1})^T (Re {\cal N})^T.\end{aligned}$$ Defining $M: {\cal M}={\cal I}M$ where $$\begin{aligned}
\label{eq:FakeW3}
& & M\equiv\left(\begin{array}{cc}
D&C\\
B&A
\end{array}\right),\nonumber\\
& & {\cal I}\equiv\left(\begin{array}{cc}
0&-{\bf 1}_{n_V+1}\\
{\bf 1}_{n_V+1}&0
\end{array}\right).\end{aligned}$$ The central charge $Z=e^{\frac{K}{2}}(q_\Lambda X^\Lambda - p^\Lambda F_\lambda)$, a symplectic invariant is expressed as a symplectic dot product of $Q$ and covariantly holomorphic sections: ${\cal V}\equiv e^{\frac{K}{2}}(X^\Lambda,F_\Lambda)=(L^\Lambda,M_\Lambda) (M_\Lambda={\cal N}_{\Lambda\Sigma}L^\Sigma)$, and hence can be written as $$\label{eq:FakeW4}
Z = Q^T{\cal I}{\cal V} = L^\Lambda q_\Lambda - M_\lambda p^\Lambda.$$ Now, the black-hole potential $V_{BH}=Q^T{\cal M}Q$ (being a symplectic invariant) is invariant under: $$\begin{aligned}
\label{eq:FakeW5}
& & Q\rightarrow SQ,\nonumber\\
& & S^T{\cal M}S={\cal M}.\end{aligned}$$ As $S$ is a symplectic matrix, $S^T{\cal I}={\cal I}S^{-1}$, which when substituted in (\[eq:FakeW5\]) yields: $$\label{eq:fakeW6}
[S,M]=0.$$ In other words, if there exists a constant symplectic matrix $S:[S,M]=0$, then there exists a fake superpotential $Q^TS^T{\cal I}{\cal V}$ whose critical points, if they exist, describe non-supersymmetric black holes.
We now construct an explicit form of $S$. For concreteness, we work at the point in the moduli space for (\[eq:hypersurface\]): $\phi^3=1$ and large $\psi$ near $x=0$ and $\rho=\rho_0$. Given the form of ${\cal N}_{IJ}$ in (\[eq:BHinv17\]), one sees that: $$\begin{aligned}
\label{eq:FakeW7}
& & {\cal N}^{-1}=\nonumber\\
& & \hskip -2cm\left(\begin{array}{ccc}
\tilde{a}_{00}+\tilde{b}^{(1)}_{00}x + \tilde{b}^{(2)}_{00} x ln x + \tilde{c}_{00}(\rho-\rho_0) &
\tilde{a}_{01}+\tilde{b}^{(1)}_{01}x + \tilde{b}^{(2)}_{01} x ln x + \tilde{c}_{01}(\rho-\rho_0) &
\tilde{a}_{02}+\tilde{b}^{(1)}_{02}x + \tilde{b}^{(2)}_{02} x ln x + \tilde{c}_{02}(\rho-\rho_0) \\
\tilde{a}_{01}+\tilde{b}^{(1)}_{01}x + \tilde{b}^{(2)}_{01} x ln x + \tilde{c}_{01}(\rho-\rho_0) &
\tilde{a}_{11}+\tilde{b}^{(1)}_{11}x + \tilde{b}^{(2)}_{11} x ln x + \tilde{c}_{11}(\rho-\rho_0) &
\tilde{a}_{12}+\tilde{b}^{(1)}_{12}x + \tilde{b}^{(2)}_{12} x ln x + \tilde{c}_{12}(\rho-\rho_0) \\
\tilde{a}_{02}+\tilde{b}^{(1)}_{02}x + \tilde{b}^{(2)}_{02} x ln x + \tilde{c}_{02}(\rho-\rho_0) &
\tilde{a}_{12}+\tilde{b}^{(1)}_{12}x + \tilde{b}^{(2)}_{12} x ln x + \tilde{c}_{12}(\rho-\rho_0) &
\tilde{a}_{22}+\tilde{b}^{(1)}_{22}x + \tilde{b}^{(2)}_{22} x ln x + \tilde{c}_{22}(\rho-\rho_0)
\end{array}\right),\nonumber\\
& &\end{aligned}$$ which as expected is symmetric (and hence so will $Re {\cal N}$ and $(Im {\cal N})^{-1}$). One can therefore write $$\label{eq:fakeW8}
M\equiv\left(\begin{array}{cc}
U & V \\
X & -U^T
\end{array}\right),$$ where $V^T=V,\ X^T=X$ and $U, V, X$ are $3\times 3$ matrices constructed from $Re {\cal N}$ and $(Im {\cal N})^{-1}$. Writing $$\label{eq:FakeW9}
S=\left(\begin{array}{cc}
{\cal A} & {\cal B} \\
{\cal C} & {\cal D}
\end{array}\right),$$ (${\cal A}, {\cal B}, {\cal C}, {\cal D}$ are $3\times3$ matrices) and given that $S\in Sp(6)$, implying: $$\label{eq:FakeW10}
\left(\begin{array}{cc}
{\cal A}^T & {\cal C}^T\\
{\cal B}^T & {\cal D}^T
\end{array}\right)\left(\begin{array}{cc}
0 & -{\bf 1}_3 \\
{\bf 1}_3 & 0
\end{array}\right)\left(\begin{array}{cc}
{\cal A} & {\cal B} \\
{\cal C} & {\cal D}
\end{array}\right)=\left(\begin{array}{cc}
0 & -{\bf 1}_3 \\
{\bf 1}_3 & 0
\end{array}\right),$$ which in turn implies the following matrix equations: $$\begin{aligned}
\label{eq:FakeW11}
& & -{\cal A}^T{\cal C} + {\cal C}^T{\cal A} = 0,\nonumber\\
& & -{\cal B}^T{\cal D} + {\cal D}^T{\cal B} = 0,\nonumber\\
& & -{\cal A}^T{\cal D} + {\cal C}^T{\cal B} = - {\bf 1}_3,\nonumber\\
& & -{\cal B}^T{\cal C} + {\cal D}^T{\cal A} = {\bf 1}_3.\end{aligned}$$ Now, $[S,M]=0$ implies: $$\label{eq:FakeW12}
\left(\begin{array}{cc}
{\cal A} U + {\cal B} X & {\cal A} V - {\cal B} U^T \\
{\cal C} U + {\cal D} X & {\cal C} V - {\cal D} U^T
\end{array}\right) = \left(\begin{array}{cc}
U {\cal A} + V {\cal C} & U {\cal B} + V {\cal D} \\
X {\cal A} - U^T {\cal C} & X {\cal B} - U^T {\cal D}
\end{array}\right).$$ The system of equations (\[eq:FakeW11\]) can be satisfied, e.g., by the following choice of ${\cal A}, {\cal B}, {\cal C}, {\cal D}$: $$\label{eq:FakeW13}
{\cal B} = {\cal C}=0;\ {\cal D} = ({\cal A}^{-1})^T.$$ To simplify matters further, let us assume that ${\cal A}\in O(3)$ implying that $({\cal A}^{-1})^T = {\cal A}$. Then (\[eq:FakeW12\]) would imply: $$\begin{aligned}
\label{eq:FakeW14}
& & [{\cal A},V] = 0,\nonumber\\
& & [{\cal A},X] = 0,\nonumber\\
& & [{\cal A}^{-1},U] = 0,\nonumber\\
& & [{\cal A},U] = 0.\end{aligned}$$ For points near the conifold locus $\phi=\omega^{-1},\rho=\rho_0$, using (\[eq:confl24\])-(\[eq:confl210\]) and (\[eq:BHinv12\]) and dropping the moduli-dependent terms in (\[eq:Wconfl11\]), one can show: $$\begin{aligned}
\label{eq:FakeW141}
& & \left(Im {\cal N}^{-1}\right)_{0I}\left(Re {\cal N}\right)_{IK}=0,\ K=1,2\nonumber\\
& & \left(Im {\cal N}^{-1}\right)_{0K}=0,\ K=1,2\nonumber\\
& & \left(Im {\cal N}\right)_{0K} + \left(Re {\cal N}\right)_{0I}\left(Im {\cal N}^{-1}\right)_{IJ}
\left(Re{\cal N}\right)_{JK} = 0,\ K=1,2.\end{aligned}$$ This is equivalent to saying that the first two and the last equations in (\[eq:FakeW14\]) can be satisfied by: $$\label{eq:FakeW15}
{\cal A} = \left(\begin{array}{ccc}
1&0&0\\
0&-1&0\\
0&0&-1
\end{array}\right).$$ The form of $A$ chosen in (\[eq:FakeW15\]) also satisfies the third equation in (\[eq:FakeW14\]) - similar solutions were also considered in [@Ceresole+Dall'agata]. Hence, $$\label{eq:FakeW16}
S = \left(\begin{array}{cccccc}
1&0&0&0&0&0\\
0&-1&0&0&0&0\\
0&0&-1&0&0&0\\
0&0&0&1&0&0\\
0&0&0&0&-1&0\\
0&0&0&0&0&-1
\end{array}\right).$$ We therefore see that the non-supersymmetric black-hole corresponding to the fake superpotential $Q^TS^T{\cal I}{\cal V}$, $S$ being given by (\[eq:FakeW16\]), corresponds to the change of sign of two of the three electric and magetic charges as compared to a supersymmetric black hole. The symmetry properties of the elements of ${\cal M}$ and hence $M$ may make it generically possible to find a constant $S$ like the one in (\[eq:FakeW16\]) for two-paramater Calabi-Yau compactifications.
Conclusion
==========
We looked at several aspects of complex structure moduli stabilization and inclusion, in the large volume limit, of perturbative and specially non-perturbative $\alpha^\prime$-corrections and instanton contributions in the Kähler potential and superpotential in the context of Kähler moduli, for a two-parameter “Swiss cheese" Calabi-Yau three-fold of a projective variety expressed as a (resolution of a) hypersurface in a complex weighted projective space, with mutliple conifold loci in its moduli space. As regards ${\cal N}=1$ type IIB compactifications on orientifold of the aforementioned Calabi-Yau, we argued the existence of (extended) “area codes" wherein for the same values of the RR and NS-NS fluxes, one is able to stabilize the complex structure and axion-dilaton moduli at points away from and close to the two singular conifold loci. It would be nice to explicitly work out the numerics and find the set of fluxes corresponding to the aforementioned area codes (whose existence we argued), as well as the flow of the moduli corresponding to the domain walls arising as a consequence of such area codes. Further, in the large volume limit of the orientifold, we show that with the inclusion of non-perturbative $\alpha^\prime$-corrections that survive the orientifolding alongwith the nonperturbative contributions from instantons, it is possible to get a non-supersymmetric dS minimum [*without the inclusion of anti-D3 branes*]{}. It would interesting to investigate the effect of string loop corrections in the context of orientifolds of compact Calabi-Yau of the type considered in this work (See [@BHP]). As regards supersymmetric and non-supersymmetric black-hole attractors in ${\cal N}=2$ type II compactifications on the same Calabi-Yau three-fold, we explicitly solve the “inverse problem" of determining the electric and magnetic charges of an extremal black hole given the extremum values of the moduli. In the same context, we also show explicitly the existence of “fake superpotentials" as a consequence of non-unique superpotentials for the same black-hole potential corresponding to reversal of signs of some of the electric and magnetic charges. There may be interesting connection between the existence of such fake superpotentials and works like [@EH][^11]
Acknowledgements {#acknowledgements .unnumbered}
================
PS is supported by a C.S.I.R., Government of India, (junior) research fellowship and the work of AM is partially supported by D.A.E. (Government of India) Young Scientist Award project grant. AM also thanks the CERN theory group for hosptiality and the theoretical high energy physics groups at McGill University (specially Keshav Dasgupta), University of Pennsylvania (specially Vijay Balasubramanian), Ohio State University (specially Samir Mathur) and Columbia University (specially D.Kabat), for hospitality where part of this work was done and also where preliminary versions of this work were presented. AM also thanks Samir Mathur and Jeremy Michelson for interesting discussions, and Rajesh Gopakumar and Ashoke Sen for interesting questions (that helped improve the arguments of section 4) during AM’s presentation of part of the material of this paper at ISM07, Harish-Chandra Research Institute, Allahabad.
Periods
=======
In this appendix, we fill in the details relevant to evaluation of periods in different portions of the complex structure moduli space of section ${\bf 2}$.
$\underline{|\phi^3|<1,\ {\rm large}\ \psi}$
The expressions for $P_{1,2,3}$ relevant to (\[eq:smallphilargepsi2\]) in section [**2**]{} are given as under:
$$\label{eq:smallphilargepsi3}
P_1\equiv\left(\begin{array}{c}
\sum_{m=0}^\infty\sum_{n=0}^\infty A_{m,n}\frac{\phi_0^m}{\rho_0^{18n + 6m}}\\
\sum_{m=0}^\infty\sum_{n=0}^\infty e^{\frac{i\pi(-35m + 128 n)}{9}}
A_{m,n}\frac{\phi_0^m}{\rho_0^{18n + 6m}}\\
\sum_{m=0}^\infty\sum_{n=0}^\infty e^{\frac{2i\pi(-35m + 128 n)}{9}}
A_{m,n}\frac{\phi_0^m}{\rho_0^{18n + 6m}}\\
\sum_{m=0}^\infty\sum_{n=0}^\infty e^{\frac{i\pi(-35m + 128 n)}{3}}
A_{m,n}\frac{\phi_0^m}{\rho_0^{18n + 6m}}\\
\sum_{m=0}^\infty\sum_{n=0}^\infty e^{\frac{4i\pi(-35m + 128 n)}{9}}
A_{m,n}\frac{\phi_0^m}{\rho_0^{18n + 6m}}\\
\sum_{m=0}^\infty\sum_{n=0}^\infty e^{\frac{5i\pi(-35m + 128 n)}{9}}
A_{m,n}\frac{\phi_0^m}{\rho_0^{18n + 6m}}\\
\end{array}\right),$$
$$\label{eq:smallphilargepsi4}
P_2\equiv\left(\begin{array}{c}
\sum_{m=0}^\infty\sum_{n=0}^\infty A_{m,n}\frac{m\phi_0^{m-1}}{\rho_0^{18n + 6m}}\\
\sum_{m=0}^\infty\sum_{n=0}^\infty A_{m,n}\frac{m\phi_0^{m-1}}{\rho_0^{18n + 6m}}
e^{\frac{i\pi(-35m + 128 n)}{9}}\\
\sum_{m=0}^\infty\sum_{n=0}^\infty A_{m,n}\frac{m\phi_0^{m-1}}{\rho_0^{18n + 6m}}
e^{2\frac{i\pi(-35m + 128 n)}{9}}\\
\sum_{m=0}^\infty\sum_{n=0}^\infty A_{m,n}\frac{m\phi_0^{m-1}}{\rho_0^{18n + 6m}}
e^{\frac{i\pi(-35m + 128 n)}{3}}\\
\sum_{m=0}^\infty\sum_{n=0}^\infty A_{m,n}\frac{m\phi_0^{m-1}}{\rho_0^{18n + 6m}}
e^{\frac{4i\pi(-35m + 128 n)}{9}}\\
\sum_{m=0}^\infty\sum_{n=0}^\infty A_{m,n}\frac{m\phi_0^{m-1}}{\rho_0^{18n + 6m}}
e^{\frac{5i\pi(-35m + 128 n)}{9}}\\
\end{array}\right)$$
and $$\label{eq:smallphilargepsi5}
P_3\equiv\left(\begin{array}{c}
-\sum_{m=0}^\infty\sum_{n=0}^\infty A_{m,n}\frac{(18n + 6m)\phi_0^m}{\rho_0^{18n + 6m + 1}}\\
-\sum_{m=0}^\infty\sum_{n=0}^\infty A_{m,n}\frac{(18n + 6m)\phi_0^m}{\rho_0^{18n + 6m + 1}}e^{\frac{i\pi(-35m + 128 n)}{9}}\\
-\sum_{m=0}^\infty\sum_{n=0}^\infty A_{m,n}\frac{(18n + 6m)\phi_0^m}{\rho_0^{18n + 6m + 1}}e^{\frac{2i\pi(-35m + 128 n)}{9}}\\
-\sum_{m=0}^\infty\sum_{n=0}^\infty A_{m,n}\frac{(18n + 6m)\phi_0^m}{\rho_0^{18n + 6m + 1}}e^{\frac{i\pi(-35m + 128 n)}{3}}\\
-\sum_{m=0}^\infty\sum_{n=0}^\infty A_{m,n}\frac{(18n + 6m)\phi_0^m}{\rho_0^{18n + 6m + 1}}e^{\frac{4i\pi(-35m + 128 n)}{9}}\\
-\sum_{m=0}^\infty\sum_{n=0}^\infty A_{m,n}\frac{(18n + 6m)\phi_0^m}{\rho_0^{18n + 6m + 1}}e^{\frac{5i\pi(-35m + 128 n)}{9}}\\
\end{array}\right).$$ The coefficients $A_{m,n}$ appearing in (\[eq:smallphilargepsi3\])-(\[eq:smallphilargepsi5\]) are given by: $$A_{m,n}\equiv \frac{(18n + 6m)!(-3\phi)^m(3^4.2)^{18n + 6m}}{(9n + 3m)!(6n + 2m)!(n!)^3m!18^{18n + 6m}}.$$ The equations (\[eq:smallphilargepsi3\])-(\[eq:smallphilargepsi5\]) will be used in obtaining (\[eq:Wawayconfl\]) and the third set of equations in (\[eq:DW=0\]).
$\underline{|\frac{\rho^6}{\phi - \omega^{0,-1,-2}}|<1}$
The expressions for $M_{1,2,3}$ relevant to (\[eq:awaycl12\]) in section are given as under:
$$\label{eq:awaycl13}
M_1\equiv\left(\begin{array}{c}
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}\rho_0^{6k+r}\phi_0^m
\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}\rho_0^{6k+r}\phi_0^m
e^{\frac{2i\pi(k+\frac{r}{6})}{3} + \frac{2i\pi m}{3}}
\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}\rho_0^{6k+r}\phi_0^m
e^{\frac{4i\pi(k+\frac{r}{6})}{3} + \frac{4i\pi m}{3}}
\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}\rho_0^{6k+r}\phi_0^m e^{\frac{i\pi r}{3}}
\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}\rho_0^{6k+r}\phi_0^m
e^{\frac{2i\pi(k+\frac{r}{6})}{3} + \frac{2i\pi m}{3} + \frac{i\pi r}{3}}
\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}\rho_0^{6k+r}\phi_0^m
e^{\frac{4i\pi(k+\frac{r}{6})}{3} + \frac{4i\pi m}{3} + \frac{i\pi r}{3}}
\end{array}\right),$$
$$\label{eq:awaycfl4}
M_2\equiv\left(\begin{array}{c}
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}(6k + r)\rho_0^{6k + r - 1}\phi_0^m
\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}(6k + r)\rho_0^{6k + r - 1}\phi_0^m
e^{\frac{2i\pi(k+\frac{r}{6})}{3} + \frac{2i\pi m}{3}}
\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}(6k + r)\rho_0^{6k + r - 1}\phi_0^m
e^{\frac{4i\pi(k+\frac{r}{6})}{3} + \frac{4i\pi m}{3}}
\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}(6k + r)\rho_0^{6k + r - 1}\phi_0^{m - 1} e^{\frac{i\pi r}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}(6k + r)\rho_0^{6k + r - 1}\phi_0^m e^{\frac{2i\pi(k+\frac{r}{6})}{3} + \frac{2i\pi m}{3} + \frac{i\pi r}{3}}
\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}(6k + r)\rho_0^{6k + r - 1}\phi_0^m
e^{\frac{4i\pi(k+\frac{r}{6})}{3} + \frac{4i\pi m}{3} + \frac{i\pi r}{3}}
\end{array}\right)$$
and $$\label{eq:awaycl15}
M_3\equiv\left(\begin{array}{c}
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}m\rho_0^{6k + r}\phi_0^{m - 1}\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}m\rho_0^{6k + r}\phi_0^{m - 1}
e^{\frac{2i\pi(k+\frac{r}{6})}{3} + \frac{2i\pi m}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}m\rho_0^{6k + r}\phi_0^{m - 1}
e^{\frac{4i\pi(k+\frac{r}{6})}{3} + \frac{4i\pi m}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}m
\rho_0^{6k + r}\phi_0^{m-1} e^{\frac{i\pi r}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}m
\rho_0^{6k + r}\phi_0^{m - 1} e^{\frac{2i\pi(k+\frac{r}{6})}{3} + \frac{2i\pi m}{3} + \frac{i\pi r}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty\sum_{m=0}^\infty A_{k,m,r}m\rho_0^{6k + r}\phi_0^{m - 1}
e^{\frac{4i\pi(k+\frac{r}{6})}{3} + \frac{4i\pi m}{3} + \frac{i\pi r}{3}}
\end{array}\right).$$ In equations (\[eq:awaycl13\])-(\[eq:awaycl15\]), the coefficients $A_{k,m,r}$ are given by: $$A_{k,m,r}\equiv e^{\frac{i\pi ar}{3}}sin\left(\frac{\pi r}{3}\right)\frac{(-)^k3^{-1 + k+\frac{r}{6} + m}e^{-\frac{i\pi(k + \frac{r}{6})}{3} + \frac{2im\pi}{3}}\Gamma(\frac{m + k + \frac{r}{6}}{3})(\Gamma(k+\frac{r}{6}))^2}{(\Gamma(1 - \frac{m + k + \frac{r}{6}}{3}))^2m!}.$$ The equations (\[eq:awaycl13\])-(\[eq:awaycl15\]) will be used in obtaining (\[eq:Wawayconfl\]) and the fourth set of equations in (\[eq:DW=0\]).
$\underline{{\rm Near\ the\ conifold\ locus:}\ \rho^6 + \phi = 1}$
The expressions for $N_{1,2,3}$ relevant for evaluation of (\[eq:confl14\]) in section [**2**]{}, are given as under:
$$\label{eq:confl15}
N_1\equiv\left(\begin{array}{c}
\sum_{r=1,5}\sum_{k=0}^\infty A_{k,0,r}e^{-\frac{i\pi(k + \frac{r}{6})}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty A_{k,0,r}e^{\frac{i\pi(k + \frac{r}{6})}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty A_{k,0,r}e^{i\pi(k + \frac{r}{6}}\\
\sum_{r=1,5}\sum_{k=0}^\infty A_{k,0,r}e^{\frac{i\pi(-k + 5\frac{r}{6})}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty A_{k,0,r}e^{-\frac{i\pi(k + 7\frac{r}{6})}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty A_{k,0,r}e^{-\frac{i\pi(3k + 3\frac{r}{6})}{3}}\end{array}\right),$$
$$\label{eq:confl16}
N_2\equiv\left(\begin{array}{c}
\sum_{r=1,5}\sum_{k=0}^\infty (k + \frac{r}{6}) A_{k,0,r}e^{-\frac{i\pi(k + \frac{r}{6})}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty (k + \frac{r}{6}) A_{k,0,r}e^{\frac{i\pi(k + \frac{r}{6})}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty (k + \frac{r}{6}) A_{k,0,r}e^{i\pi(k + \frac{r}{6}}\\
\sum_{r=1,5}\sum_{k=0}^\infty (k + \frac{r}{6}) A_{k,0,r}e^{\frac{i\pi(-k + 5\frac{r}{6})}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty (k + \frac{r}{6}) A_{k,0,r}e^{-\frac{i\pi(k + 7\frac{r}{6})}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty (k + \frac{r}{6}) A_{k,0,r}e^{-\frac{i\pi(3k + 3\frac{r}{6})}{3}}\end{array}\right)$$
and $$\label{eq:confl17}
N_3\equiv\left(\begin{array}{c}
\sum_{r=1,5}\sum_{k=0}^\infty (A_{k,1,r} e^{\frac{2i\pi}{3}} - A_{k,0,r}(k + \frac{r}{6}))
e^{-\frac{i\pi(k + \frac{r}{6})}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty (A_{k,1,r} e^{\frac{4i\pi}{3}} - A_{k,0,r}(k + \frac{r}{6}))
e^{\frac{i\pi(k + \frac{r}{6})}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty (A_{k,1,r} - A_{k,0,r}(k + \frac{r}{6}))
e^{i\pi(k + \frac{r}{6}}\\
\sum_{r=1,5}\sum_{k=0}^\infty (A_{k,1,r} e^{\frac{2i\pi}{3}} - A_{k,0,r}(k + \frac{r}{6}))
e^{\frac{i\pi(-k + 5\frac{r}{6})}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty (A_{k,1,r} e^{\frac{4i\pi}{3}} - A_{k,0,r}(k + \frac{r}{6}))
e^{-\frac{i\pi(k + 7\frac{r}{6})}{3}}\\
\sum_{r=1,5}\sum_{k=0}^\infty (A_{k,1,r} - A_{k,0,r}(k + \frac{r}{6}))
e^{-\frac{i\pi(3k + 3\frac{r}{6})}{3}}
\end{array}\right).$$ The coefficients $A_{k,m,r}$ figuring in (\[eq:confl15\])-(\[eq:confl17\]) are given by: $$A_{k,m,r}\equiv \frac{3^{-1 + k + \frac{r}{6} + m}e^{\frac{2i\pi(\sigma + 1)}{3} + \frac{-i\pi(k + \frac{r}{6})}{3}}(-)^k(\Gamma(k + \frac{r}{6}))^2}{\Gamma(k + 1)\Gamma(k + \frac{r}{6})\Gamma(k + \frac{r}{6})(\Gamma(1 - \frac{m + k + \frac{r}{6}}{3}))^2m!}sin\left(\frac{\pi r}{3}\right).$$
$\underline{{\rm Near}\ \phi^3=1,\ {\rm Large}\ \rho}$
The expressions for $\varpi_{0,...,5}$ relevant for evaluation of (\[eq:confl210\]) in section [**2**]{}, are given as under:
$$\begin{aligned}
\label{eq:confl24}
& &
(i) \varpi_0 \sim -\int_{\Gamma^\prime}\frac{d\mu}{4\pi^2i}\frac{\Gamma(-\mu)\Gamma(\mu+\frac{1}{6})}{\Gamma(1+\mu)}\rho^{-6\mu}
\frac{\sqrt{3}\sum_{\tau=0}^2\gamma^{0,\tau}_\mu\omega^\tau(\phi-\omega^{-\tau})^{\mu+1}}{2\pi(\mu+1)}\nonumber\\
& & \sim A_{0,0}\left[(\phi-1) - 2\omega(\phi - \omega^{-1}) + \omega^2(\phi - \omega^2)\right] + \frac{A_{0,1}}{\rho^6}\left[(\phi - 1)^2 - 2\omega(\phi - \omega^{-1})^2 + \omega^2(\phi - \omega^{-2})^2\right],
\nonumber\\
& &\end{aligned}$$
where $A_{0,0}=-\frac{\sqrt{3}}{16\pi^4i}\Gamma(\frac{1}{6})\Gamma(\frac{5}{6})$ and $A_{0,1}=\frac{\sqrt{3}}{16\pi^4i}\Gamma(\frac{7}{6})\Gamma(\frac{11}{6})$.
$$\begin{aligned}
\label{eq:confl25}
& & (ii) \varpi_1\sim
\int_{\Gamma^\prime}\frac{d\mu}{8\pi^3}(\Gamma(-\mu))^2\Gamma(\mu+\frac{1}{6})\Gamma(\mu+\frac{5}{6})
\rho^{-6\mu}\nonumber\\
& & \times\frac{\sqrt{3}}{2\pi(\mu+1)}\biggl[2i sin(\pi\mu)\left(e^{-2i\pi\mu}(\phi - 1)^{\mu+1} + \omega(\phi - \omega^{-1})^{\mu+1} - 2\omega^2(\phi - \omega^{-2})^{\mu+1} + (\phi - 1)^{\mu+1}\right)\nonumber\\
& & + e^{-3i\pi\mu}(\phi - 1)^{\mu + 1}\biggr]
\nonumber\\
& & \sim A_{0,1}\left[(\phi - 1)(A_0 + ln(\rho^{-6})) + i\pi((\phi - 1) + 2\omega(\phi - \omega^{-1} - 4\omega^2(\phi - \omega^{-2}) + (\phi - 1)ln(\phi - 1)\right]\nonumber\\
& & + \frac{A_{0,1}}{\rho^6}\left[(\phi - 1)^2(A_1 + ln(\rho^{-6}) + i\pi\left((\phi - 1) + 2\omega(\phi - \omega^{-1}) - 4\omega^2(\phi - \omega^{-2})\right) + (\phi - 1)^2ln(\phi - 1)\right],\nonumber\\
& &\end{aligned}$$
where $A_0\equiv -1 - 2\Psi(1) + \Psi(\frac{1}{6}) + \Psi(\frac{5}{6})$ and $A_1\equiv -\frac{1}{2} - 2\Psi(1) - 2 + \Psi(\frac{7}{6})+ \psi(\frac{11}{6})$.
$$\begin{aligned}
\label{eq:confl26}
& & (iii)\ \varpi_2\sim-\int_{\Gamma^\prime}\frac{d\mu}{8\pi^3}(\Gamma(-\mu))^2\Gamma(\mu+\frac{1}{6})\Gamma(\mu+\frac{5}{6})
\rho^{-6\mu}\nonumber\\
& & \times\frac{\sqrt{3}}{2\pi(\mu+1)}\biggl[2i sin(\pi\mu)\left(-2e^{-2i\pi\mu}(\phi - 1)^{\mu + 1} + e^{-2i\pi\mu}\omega(\phi - \omega^{-1})^{\mu+1} + \omega^2(\phi - \omega^{-2})^{\mu+1}\right)\nonumber\\
& & - \left((\phi - 1)^{\mu+1} - \omega(\phi - \omega^{-1})^{\mu+1}\right) - (\phi - 1)^{\mu+1}\biggr]\nonumber\\
& &
\sim A_{0,0}\biggl[\left(\omega(\phi - \omega^{-1}) - 2(\phi - 1)\right)(A_0 + ln(\rho^{-6})) + 2i\pi\left(-2(\phi - 1) + \omega(\phi - \omega^{-1}) + \omega^2(\phi - \omega^{-2})\right)\nonumber\\
& & - 2(\phi - 1)ln(\phi - 1)\biggr]
+\frac{A_{0,1}}{\rho^6}\biggl[\left(\omega(\phi - \omega^{-1}) - 2(\phi - 1)^2\right)(A_1 + ln(\rho^{-6}))\nonumber\\
& & -2i\pi\biggl(-2(\phi - 1)^2 + \omega(\phi - \omega^{-1})^2
+ \omega^2(\phi - \omega^{-2})^2\biggr) - 2(\phi - 1)^2 ln(\phi - 1)\biggr]\end{aligned}$$
$$\begin{aligned}
\label{eq:confl27}
& & (iv)\ \varpi_3\sim
\int_{\Gamma^\prime}\frac{d\mu}{8\pi^3}(\Gamma(-\mu))^2\Gamma(\mu+\frac{1}{6})\Gamma(\mu + \frac{5}{6})\rho^{-6\mu}e^{i\pi\mu}\nonumber\\
& & \times-\frac{\sqrt{3}}{2\pi(\mu+1)}\left[(\phi - 1)^{\mu+1} - 2\omega(\phi - \omega^{-1})^{\mu+1} + \omega^2(\phi - \omega^{-2})^{\mu+1}\right]\nonumber\\
& & \sim A_{0,0}\biggl[\left((\phi - 1) - 2\omega(\phi - \omega^{-1}) + \omega^2(\phi - \omega^{-2})\right)(A_0 + i\pi + ln(\rho^{-6}))\nonumber\\
& & +(\phi - 1)ln(\phi - 1) - 2\omega(\phi - \omega^{-1})ln(\phi - \omega^{-1}) + \omega^2(\phi - \omega^{-2})ln(\phi - \omega^{-2})\biggr] \nonumber\\
& & + \frac{A_{0,1}}{\rho^6}\biggl[\left((\phi - 1)^2
- 2\omega(\phi - \omega^{-1})^2 + \omega^2(\phi - \omega^{-2})^2\right)(A_1 + i\pi + ln(\rho^{-6}))\nonumber\\
& & +(\phi - 1)^2ln(\phi - 1) - 2\omega(\phi - \omega^{-1})^2ln(\phi - \omega^{-1}) + \omega^2(\phi - \omega^{-2})^2ln(\phi - \omega^{-2})\biggr]\end{aligned}$$
$$\begin{aligned}
\label{eq:confl28}
& & (v)\ \varpi_4\sim
\int_{\Gamma^\prime}\frac{d\mu}{8\pi^3}(\Gamma(-\mu))^2\Gamma(\mu+\frac{1}{6})\Gamma(\mu + \frac{5}{6})\rho^{-6\mu}e^{i\pi\mu}\nonumber\\
& & \times-\frac{\sqrt{3}}{2\pi(\mu+1)}\left[e^{-2i\pi\mu}(\phi - 1)^{\mu+1} + \omega\omega(\phi - \omega^{-1})^{\mu+1} -2\omega^2(\phi - \omega^{-2})^{\mu+1}\right]\nonumber\\
& & \sim A_{0,0}\biggl[\left((\phi - 1) - 2\omega(\phi - \omega^{-1}) + \omega^2(\phi - \omega^{-2})\right)(A_0 + i\pi + ln(\rho^{-6}))\nonumber\\
& & -2i\pi(\phi - 1) +(\phi - 1)ln(\phi - 1) + \omega(\phi - \omega^{-1})ln(\phi - \omega^{-1}) -2\omega^2(\phi - \omega^{-2})ln(\phi - \omega^{-2})\biggr] \nonumber\\
& & + \frac{A_{0,1}}{\rho^6}\biggl[\left((\phi - 1)^2
- 2\omega(\phi - \omega^{-1})^2 + \omega^2(\phi - \omega^{-2})^2\right)(A_1 + i\pi + ln(\rho^{-6}))\nonumber\\
& & -2i\pi(\phi - 1)^2 +(\phi - 1)^2ln(\phi - 1) + \omega(\phi - \omega^{-1})^2ln(\phi - \omega^{-1}) - 2 \omega^2(\phi - \omega^{-2})^2ln(\phi - \omega^{-2})\biggr]\nonumber\\
& &\end{aligned}$$
$$\begin{aligned}
\label{eq:confl29}
& & (vi)\ \varpi_5\sim
\int_{\Gamma^\prime}\frac{d\mu}{8\pi^3}(\Gamma(-\mu))^2\Gamma(\mu+\frac{1}{6})\Gamma(\mu + \frac{5}{6})\rho^{-6\mu}e^{i\pi\mu}\nonumber\\
& & \times-\frac{\sqrt{3}}{2\pi(\mu+1)}\left[-2e^{-2i\pi\mu}(\phi - 1)^{\mu+1} + e^{-2i\pi\mu}\omega\omega(\phi - \omega^{-1})^{\mu+1} + \omega^2(\phi - \omega^{-2})^{\mu+1}\right]\nonumber\\
& & \sim A_{0,0}\biggl[\left(-2(\phi - 1) + \omega(\phi - \omega^{-1}) + \omega^2(\phi - \omega^{-2})\right)(A_0 + i\pi + ln(\rho^{-6}))\nonumber\\
& & +4i\pi(\phi - 1) - 2(\phi - 1)ln(\phi - 1) -2i\pi\omega(\phi - \omega^{-1}) + \omega(\phi - \omega^{-1})ln(\phi - \omega^{-1}) + \omega^2(\phi - \omega^{-2})ln(\phi - \omega^{-2})\biggr] \nonumber\\
& & + \frac{A_{0,1}}{\rho^6}\biggl[\left(-2(\phi - 1)^2 +
\omega(\phi - \omega^{-1})^2 + \omega^2(\phi - \omega^{-2})^2\right)(A_1 + i\pi + ln(\rho^{-6}))\nonumber\\
& & 4i\pi(\phi - 1)^2 -2(\phi - 1)^2ln(\phi - 1) - 2i\pi\omega(\phi - \omega^{-1})^2 + \omega(\phi - \omega^{-1})^2ln(\phi - \omega^{-1})\nonumber\\
& & + \omega^2(\phi - \omega^{-2})^2ln(\phi - \omega^{-2})\biggr].\end{aligned}$$
Complex Structure Superpotential Extremization
==============================================
In this appendix, the details pertaining to evaluation of the covariant derivative of the complex structure superpotential in (\[eq:Wconfl161\]), are given.
One can see from (\[eq:Wconfl12\]): $$\label{eq:Wconfl13}
\hskip - 3cm
\partial_x K\sim \frac{- ln x\left({\bar A_1} B_{42} - {\bar B_{12}} {\bar A_4} + {\bar A_2}B_{52} -
{\bar A_5}B_{22}\right)}{{\cal K}},$$ where $$\begin{aligned}
\label{eq:Wconfl14}
& & {\cal K}\equiv2i\ {\rm Im}\Biggl[ ({\bar A_0} A_3 + {\bar A_1} A_4 + {\bar A}_2 A_5) + ({\bar B_{01}} A_3 + {\bar A_0}B_{31} + {\bar A_1}B_{41} + {\bar B_{11}} A_4 + {\bar A_2} B_{51} + {\bar B_{21}} A_5)x
\nonumber\\
& & + ({\bar A_1} B_{42} + {\bar B_{12}} A_4 + {\bar A_2} B_{52} + {\bar B_{22}} A_5)x ln x + ({\bar A_0} C_3 + {\bar C_0} A_3 + {\bar A_1} C_4 + {\bar C_1} A_4 + {\bar A_2} C_5 + {\bar C_2} A_5)(\rho - \rho_0)\Biggr].\nonumber\\
& &\end{aligned}$$ At the extremum values of the complex structure moduli ($x,\rho - \rho_0$), $$\begin{aligned}
\label{eq:Wconfl15}
& & \tau=\frac{f^T.{\bar\Pi}}{h^T.{\bar\Pi}}\approx\nonumber\\
& & \frac{1}{(\sum_{i=0}^5h_i{\bar A_i})}\Biggl[f_0\left({\bar A_0} + {\bar B_{01}}{\bar x} + {\bar C_0}({\bar\rho} - {\bar \rho_0})\right) + f_1\left({\bar A_1} + {\bar B_{11}}{\bar x} + {\bar B_{12}}{\bar x} ln{\bar x} + {\bar C_1}({\bar\rho} - {\bar \rho_0})\right)\nonumber\\
& & + f_2\left({\bar A_2} + {\bar B_{21}}{\bar x} + {\bar B_{22}}{\bar x} ln{\bar x} + {\bar C_2}({\bar\rho} - {\bar \rho_0})\right)+ f_3\left(({\bar A_3} + {\bar B_{31}}{\bar x} + {\bar C_3}({\bar\rho} - {\bar \rho_0})\right)\nonumber\\
& & + f_4\left({\bar A_4} + {\bar B_{41}}{\bar x} + {\bar B_{42}}{\bar x} ln{\bar x} + {\bar C_4}({\bar\rho} - {\bar \rho_0})\right) + f_5\left({\bar A_5} + {\bar B_{51}}{\bar x} + {\bar B_{52}}{\bar x} ln{\bar x} + {\bar C_1}({\bar\rho} - {\bar \rho_0})\right)\Biggr]\nonumber\\
& & \times\Biggl( 1 - \frac{1}{(\sum_{i=0}^5h_i{\bar A_i})}\Biggl[h_0({\bar B_{01}}{\bar x} + {\bar C_0}({\bar\rho} - {\bar \rho_0})) + h_1({\bar B_{11}}{\bar x} + {\bar B_{12}}{\bar x} ln{\bar x} + {\bar C_1}({\bar\rho} - {\bar \rho_0}))\nonumber\\
& & + h_2({\bar B_{21}}{\bar x} + {\bar B_{22}}{\bar x} ln{\bar x} + {\bar C_2}({\bar\rho} - {\bar \rho_0})) + h_3({\bar A_3} + {\bar B_{31}}{\bar x} + {\bar C_3}({\bar\rho} - {\bar \rho_0}))\nonumber\\
& & + h_4({\bar A_4} + {\bar B_{41}}{\bar x} + {\bar B_{42}}{\bar x} ln{\bar x} + {\bar C_4}({\bar\rho} - {\bar \rho_0})) + h_5({\bar A_5} + {\bar B_{51}}{\bar x} + {\bar B_{52}}{\bar x} ln{\bar x} + {\bar C_1}({\bar\rho} - {\bar \rho_0}))\Biggr).\nonumber\\
& &\end{aligned}$$ Hence, $$\begin{aligned}
\label{eq:Wconfl16}
& & \partial_x W_{c.s.} \approx ln x\Biggl[ {\bar B_{12}}\Biggl(f_1 - \frac{h_1\Xi[f_i;{\bar x},({\bar\rho} - {\bar \rho_0})]}{\sum_{i=0}^5h_i{\bar A_i}} + \frac{h_1(\sum_{j=0}^5f_i{\bar A_i})\Xi[h_i;{\bar x},({\bar\rho} - {\bar \rho_0})]}{(\sum_{i=0}^5h_i{\bar A_i})^2}\Biggr)\nonumber\\
& & + {\bar B_{22}}\Biggl(f_2 - \frac{h_2\Xi[f_i;{\bar x},({\bar\rho} - {\bar \rho_0})]}{\sum_{i=0}^5h_i{\bar A_i}} + \frac{h_2(\sum_{j=0}^5f_i{\bar A_i})\Xi[h_i;{\bar x},({\bar\rho} - {\bar \rho_0})]}{(\sum_{i=0}^5h_i{\bar A_i})^2}\Biggr)\nonumber\\
& & + {\bar B_{42}}\Biggl(f_4 - \frac{h_2\Xi[f_i;{\bar x},({\bar\rho} - {\bar \rho_0})]}{\sum_{i=0}^5h_i{\bar A_i}} + \frac{h_2(\sum_{j=0}^5f_i{\bar A_i})\Xi[h_i;{\bar x},({\bar\rho} - {\bar \rho_0})]}{(\sum_{i=0}^5h_i{\bar A_i})^2}\Biggr)\nonumber\\
& & + {\bar B_{52}}\Biggl(f_5 - \frac{h_5\Xi[f_i;{\bar x},({\bar\rho} - {\bar \rho_0})]}{\sum_{i=0}^5h_i{\bar A_i}} + \frac{h_5(\sum_{j=0}^5f_i{\bar A_i})\Xi[h_i;{\bar x},({\bar\rho} - {\bar \rho_0})]}{(\sum_{i=0}^5h_i{\bar A_i})^2}\Biggr)\Biggr],\end{aligned}$$ where $$\begin{aligned}
\label{eq:Xisdefs}
& &
\Xi[f_i; {\bar x},({\bar\rho} - {\bar \rho_0})]\equiv f_0({\bar A_0} + {\bar B_{01}}{\bar x} + {\bar C_0}({\bar\rho} - {\bar \rho_0})) + f_1({\bar A_1} + {\bar B_{11}}{\bar x} + {\bar B_{12}}{\bar x} ln{\bar x} + {\bar C_1}({\bar\rho} - {\bar \rho_0}))\nonumber\\
& & + f_2({\bar A_2} + {\bar B_{21}}{\bar x} + {\bar B_{22}}{\bar x} ln{\bar x} +
{\bar C_2}({\bar\rho}-{\bar \rho_0})
+ f_3({\bar A_3}+{\bar B_{31}}{\bar x}+{\bar C_3}({\bar\rho}-{\bar \rho_0}))\nonumber\\
& & + f_4({\bar A_4}+{\bar B_{41}}{\bar x}+{\bar B_{42}}{\bar x} ln{\bar x}+{\bar C_4}({\bar\rho}- {\bar \rho_0}))+f_5({\bar A_5}+{\bar B_{51}}{\bar x}+{\bar B_{52}}{\bar x} ln{\bar x}+{\bar C_1}({\bar\rho}-{\bar \rho_0})),\nonumber\\
& & \equiv f^T{\bar\Pi}({\bar x},{\bar\rho} - {\bar\rho_0})\nonumber\\
& &
\Xi[h_i;{\bar x},({\bar\rho} - {\bar \rho_0})]= h_0({\bar B_{01}}{\bar x} + {\bar C_0}({\bar\rho} - {\bar \rho_0})) + h_1({\bar A_1} + {\bar B_{11}}{\bar x} + {\bar B_{12}}{\bar x} ln{\bar x} + {\bar C_1}({\bar\rho} - {\bar \rho_0}))\nonumber\\
& & + h_2({\bar B_{21}}{\bar x} + {\bar B_{22}}{\bar x} ln{\bar x} + {\bar C_2}({\bar\rho} - {\bar \rho_0})) + h_3({\bar B_{31}}{\bar x} + {\bar C_3}({\bar\rho} - {\bar \rho_0}))\nonumber\\
& & + h_4({\bar B_{41}}{\bar x} + {\bar B_{42}}{\bar x} ln{\bar x} + {\bar C_4}({\bar\rho} - {\bar \rho_0})) + h_5({\bar A_5} + {\bar B_{51}}{\bar x} + {\bar B_{52}}{\bar x} ln{\bar x} + {\bar C_1}({\bar\rho} - {\bar \rho_0}))\nonumber\\
& & \equiv h^T\left[{\bar\Pi}({\bar x},{\bar\rho} - {\bar\rho_0}) - {\bar\Pi}(x=0,\rho=\rho_0)\right].\end{aligned}$$
Inverse Metric Components
=========================
The components of the inverse of the metric (\[eq:nonpert13\]), relevant to almost all equations in section [**4**]{} starting from (\[eq:nonpert15\]) are given as under:
$$\begin{aligned}
& & (G^{-1})^{\rho_1{\bar\rho_1}}=\nonumber\\
& & \frac{1}{\Delta}\Biggl[
144\,{\cal Y}^2\,{\sqrt{-{\rho_1} + {\bar\rho_1}}}\,
\biggl( 2\,{\rho_2}\,{\cal X}^2\,{\sqrt{-{\rho_2} + {\bar\rho_2}}} \nonumber\\
& & -
\left( 2\,{\cal X}^2 + e^{3\,\phi}\,{\cal X}_1\,{\cal Y}^2 \right) \,{\bar\rho_2}\,
{\sqrt{-{\rho_2} + {\bar\rho_2}}}
e^{3\,\phi}\, + {\cal X}_1\,{\cal Y}^2\,\left( 3\,{\sqrt{2}}\,{\cal Y} +
{\rho_2}\,{\sqrt{-{\rho_2} + {\bar\rho_2}}} \right) \biggr)\Biggr],\nonumber\\
& & (G^{-1})^{\rho_1{\bar\rho_2}}=\frac{1}{\Delta}\Biggl[
144\,{\cal Y}^2\,\left( -2\,{\cal X}^2 + e^{3\,\phi}\,{\cal X}_1\,{\cal Y}^2 \right) \,
\left( {\rho_1} - {\bar\rho_1} \right) \,
\left( {\rho_2} - {\bar\rho_2} \right)\Biggr],\nonumber\\
& & (G^{-1})^{\rho_1{\bar G^1}}=\frac{1}{\Delta}
24\,i \,e^{\frac{3\,\phi}{2}}\,{\cal X}\,{\cal Y}^2\,\left( {\rho_1} - {\bar\rho_1} \right) \,
\left( 3\,{\cal Y} + {\sqrt{2}}\,{\rho_2}\,{\sqrt{-{\rho_2} + {\bar\rho_2}}} -
{\sqrt{2}}\,{\bar\rho_2}\,{\sqrt{-{\rho_2} + {\bar\rho_2}}}
\right)\nonumber\\
& & (G^{-1})^{\rho_2{\bar\rho_2}}=\nonumber\\
& & \frac{1}{\Delta}
144\,{\cal Y}^2\,\biggl[ -2\,{\rho_1}\,{\cal X}^2\,{\sqrt{-{\rho_1} + {\bar\rho_1}}} +
\left( 2\,{\cal X}^2 + e^{3\,\phi}\,{\cal X}_1\,{\cal Y}^2 \right) \,{\bar\rho_1}\,
{\sqrt{-{\rho_1} + {\bar\rho_1}}} \nonumber\\
& & +
e^{3\,\phi}\,{\cal X}_1\,{\cal Y}^2\,\left( 3\,{\sqrt{2}}\,{\cal Y} -
{\rho_1}\,{\sqrt{-{\rho_1} + {\bar\rho_1}}} \right) \biggr] \,
{\sqrt{-{\rho_2} + {\bar\rho_2}}},\nonumber\\
& & (G^{-1})^{\rho_2{\bar G^1}}=\frac{1}{\Delta}\Biggl[
-24\,i \,e^{\frac{3\,\phi}{2}}\,{\cal X}\,{\cal Y}^2\,\left( 3\,{\cal Y} -
{\sqrt{2}}\,{\rho_1}\,{\sqrt{-{\rho_1} + {\bar\rho_1}}} +
{\sqrt{2}}\,{\bar\rho_1}\,{\sqrt{-{\rho_1} + {\bar\rho_1}}}
\right) \,\left( {\rho_2} - {\bar\rho_2} \right)\Biggr],\nonumber\\
& & (G^{-1})^{G^1{\bar G^1}}=\frac{1}{\Delta}\Biggl[
18\,e^{3\,\phi}\,{{k1}}^2\,{\cal X}_1\,{\cal Y}^4 -
6\,{\sqrt{2}}\,{{k2}}^2\,{\rho_1}\,{\cal X}^2\,{\cal Y}\,
{\sqrt{-{\rho_1} + {\bar\rho_1}}} -
3\,{\sqrt{2}}\,e^{3\,\phi}\,{{k1}}^2\,{\rho_1}\,{\cal X}_1\,{\cal Y}^3\,
{\sqrt{-{\rho_1} + {\bar\rho_1}}} \nonumber\\
& & +
6\,{\sqrt{2}}\,{{k_2}}^2\,{\rho_2}\,{\cal X}^2\,{\cal Y}\,
{\sqrt{-{\rho_2} + {\bar\rho_2}}} +
3\,{\sqrt{2}}\,e^{3\,\phi}\,{{k_1}}^2\,{\rho_2}\,{\cal X}_1\,{\cal Y}^3\,
{\sqrt{-{\rho_2} + {\bar\rho_2}}} -
8\,{{k_2}}^2\,{\rho_1}\,{\rho_2}\,{\cal X}^2\,
{\sqrt{-{\rho_1} + {\bar\rho_1}}}\,
{\sqrt{-{\rho_2} + {\bar\rho_2}}}\nonumber\\
& & -
\left( 3\,{\sqrt{2}}\,e^{3\,\phi}\,{{k_1}}^2\,{\cal X}_1\,{\cal Y}^3 +
2\,{{k_2}}^2\,{\cal X}^2\,\left( 3\,{\sqrt{2}}\,{\cal Y} -
4\,{\rho_1}\,{\sqrt{-{\rho_1} + {\bar\rho_1}}} \right) \right) \,
{\bar\rho_2}\,{\sqrt{-{\rho_2} + {\bar\rho_2}}}\nonumber\\
& & +
{\bar\rho_1}\,{\sqrt{-{\rho_1} + {\bar\rho_1}}}\,
\left( 3\,{\sqrt{2}}\,e^{3\,\phi}\,{{k_1}}^2\,{\cal X}_1\,{\cal Y}^3 -
8\,{{k_2}}^2\,{\cal X}^2\,{\bar\rho_2}\,
{\sqrt{-{\rho_2} + {\bar\rho_2}}} +
2\,{{k_2}}^2\,{\cal X}^2\,\left( 3\,{\sqrt{2}}\,{\cal Y} +
4\,{\rho_2}\,{\sqrt{-{\rho_2} + {\bar\rho_2}}} \right) \right)\Biggr],
\end{aligned}$$
with: $$\begin{aligned}
& & \Delta=
-18\,e^{3\,\phi}\,{\cal X}_1\,Y^4 + 6\,{\sqrt{2}}\,{\rho_1}\,{\cal X}^2\,{\cal Y}\,
{\sqrt{-{\rho_1} + {\bar\rho_1}}} +
3\,{\sqrt{2}}\,e^{3\,\phi}\,{\rho_1}\,{\cal X}_1\,{\cal Y}^3\,
{\sqrt{-{\rho_1} + {\bar\rho_1}}}\nonumber\\
& & -
6\,{\sqrt{2}}\,{\rho_2}\,{\cal X}^2\,{\cal Y}\,{\sqrt{-{\rho_2} + {\bar\rho_2}}} -
3\,{\sqrt{2}}\,e^{3\,\phi}\,{\rho_2}\,{\cal X}_1\,{\cal Y}^3\,
{\sqrt{-{\rho_2} + {\bar\rho_2}}} +
8\,{\rho_1}\,{\rho_2}\,X^2\,{\sqrt{-{\rho_1} + {\bar\rho_1}}}\,
{\sqrt{-{\rho_2} + {\bar\rho_2}}} \nonumber\\
& & +
\left( 3\,{\sqrt{2}}\,e^{3\,\phi}\,{\cal X}_1\,{\cal Y}^3 +
{\cal X}^2\,\left( 6\,{\sqrt{2}}\,{\cal Y} - 8\,{\rho_1}\,{\sqrt{-{\rho_1} + {\bar\rho_1}}}
\right) \right) \,{\bar\rho_2}\,
{\sqrt{-{\rho_2} + {\bar\rho_2}}} \nonumber\\
& & -
{\bar\rho_1}\,{\sqrt{-{\rho_1} + {\bar\rho_1}}}\,
\left( 3\,{\sqrt{2}}\,e^{3\,\phi}\,{\cal X}_1\,{\cal Y}^3 -
8\,{\cal X}^2\,{\bar\rho_2}\,{\sqrt{-{\rho_2} + {\bar\rho_2}}} +
{{\cal X}}^2\,\left( 6\,{\sqrt{2}}\,{\cal Y} + 8\,{\rho_2}\,
{\sqrt{-{\rho_2} + {\bar\rho_2}}} \right) \right);\nonumber\\
& & {\cal X}\equiv \sum_c\sum_{(n,m)\in{\bf Z}^2/(0,0)}A_{n,m,n_{k^c}}(\tau)sin(nk.b + mk.c).
\end{aligned}$$
Ingredients for Evaluation of ${\cal N}_{IJ}$
=============================================
In this appendix we fill in the details relevant to evaluation of $X^I Im(F_{IJ}) X^J$ to arrive at (\[eq:BHinv15\]).
First, using (\[eq:BHinv12\]), one arrives at; $$\begin{aligned}
\label{eq:BHinv13}
& & Im(F_{0I})X^I=-\frac{i}{2}\Biggl[\left(\frac{B_{01}}{B_3}+\frac{C_0}{C_3}-\frac{{\bar B_{01}}}{{\bar B_3}}-\frac{{\bar C_0}}{{\bar C_3}}\right)(A_3+B_{31}x+C_3(\rho-\rho_0)\nonumber\\
& & +\left(\frac{C_1}{C_3}-\frac{{\bar C_1}}{{\bar C_3}}\right)(A_4+B_{41}x+B_{42}x ln x+C_4(\rho-\rho_0))\Biggr] +\left(\frac{C_2}{C_3}-\frac{{\bar C_2}}{{\bar C_3}}\right)(A_5+B_{51}x+B_{52}x ln x+C_5(\rho-\rho_0))\Biggr];\nonumber\\
& & Im(F_{1I})X^I=-\frac{i}{2}\Biggl[\left(\frac{C_1}{C_3}-\frac{{\bar C_1}}{{\bar C_3}}\right)(A_3+B_{31}x+C_3(\rho-\rho_0)\nonumber\\
& & +\left(\frac{C_1}{C_4}-\frac{{\bar C_1}}{{\bar C_4}}\right)(A_4+B_{41}x+B_{42}x ln x+C_4(\rho-\rho_0))\Biggr] +\left(\frac{C_2}{C_4}-\frac{{\bar C_2}}{{\bar C_4}}\right)(A_5+B_{51}x+B_{52}x ln x+C_5(\rho-\rho_0))\Biggr];\nonumber\\
& & Im(F_{2I})X^I=-\frac{i}{2}\Biggl[\left(\frac{C_2}{C_3}-\frac{{\bar C_2}}{{\bar C_3}}\right)(A_3+B_{31}x+C_3(\rho-\rho_0)\nonumber\\
& & +\left(\frac{C_2}{C_4}-\frac{{\bar C_2}}{{\bar C_4}}\right)(A_4+B_{41}x+B_{42}x ln x+C_4(\rho-\rho_0))\Biggr] +\left(\frac{C_2}{C_5}-\frac{{\bar C_2}}{{\bar C_5}}\right)(A_5+B_{51}x+B_{52}x ln x+C_5(\rho-\rho_0))\Biggr].\nonumber\\
& &\end{aligned}$$ This hence yields $$\begin{aligned}
\label{eq:BHinv14}
& & X^I Im(F_{IJ} X^J = (X^0)^2 Im(F_{00}) + (X^1)^2 Im(F_{11}) + (X^2)^2 Im(F_{22})\nonumber\\
& & + 2x^0X1 Im(F_{01}) + 2X^0X^2 Im(F_{02}) + 2X^1X^2 Im(F_{12})\nonumber\\
& & \approx-\frac{i}{2}\Biggl[\Biggl(A_3^2 + 2A_3B_{31} x + 2A_3C_3 (\rho-\rho_0)\Biggr)\left(\frac{B_{01}}{B_3} + \frac{C_0}{C_3} - \frac{{\bar B_{01}}}{{\bar B_3}} - \frac{{\bar C_0}}{{\bar C_3}}\right)\nonumber\\
& & + \Biggl(A_4^2 + 2A_4B_{41} x + 2A_4B_{42} x ln x + 2A_4C_4 (\rho-\rho_0)\Biggr)\left(\frac{C_1}{C_4} - \frac{{\bar C_1}}{{\bar C_4}}\right)\nonumber\\
& & + \Biggl(A_5^2 + 2A_5B_{51} x + 2A_5B_{52} x ln x + 2A_5C_5 (\rho-\rho_0)\Biggr)
\left(\frac{C_2}{C_4} - \frac{{\bar C_2}}{{\bar C_5}}\right)\nonumber\\
& & + \Biggl(A_3A_4 + [A_3B_{41}+A_4B_{31}] x + A_3B_{42} x ln x + [A_3C_4 + A_4C_3](\rho - \rho_0)\Biggr)\left(\frac{C_1}{C_3} - \frac{{\bar C_1}}{{\bar C_3}}\right)\nonumber\\
& & + \Biggl(A_3A_5 + [A_3B_{51}+A_5B_{31}] x + A_3B_{52} x ln x + [A_3C_5 + A_5C_3](\rho - \rho_0)\Biggr)\left(\frac{C_2}{C_3} - \frac{{\bar C_2}}{{\bar C_3}}\right)\nonumber\\
& & + \Biggl(A_4A_5 + [A_4B_{51}+A_5B_{41}] x + A_4B_{52} x ln x + [A_4C_5 + A_5C_4](\rho - \rho_0)\Biggr)\left(\frac{C_2}{C_4} - \frac{{\bar C_2}}{{\bar C_4}}\right)\Biggr].\end{aligned}$$
[99]{} A. Giryavets, [*New attractors and area codes*]{}, JHEP [**0603**]{}, (2006) 020 \[arXiv:hep-th/0511215\]. K. Saraikin and C. Vafa, [*Non-supersymmetric Black Holes and Topological Strings*]{}, arXiv:hep-th/0703214. A. Ceresole and G. Dall’Agata, [*Flow equations for non-BPS extremal black holes*]{}, JHEP [**0703**]{} (2007) 110 \[arXiv:hep-th/0702088\]. J. P. Conlon, F. Quevedo and K. Suruliz, “Large-volume flux compactifications: Moduli spectrum and D3/D7 soft supersymmetry breaking,” JHEP [**0508**]{}, (2005) 007 \[arXiv:hep-th/0505076\]. V. Balasubramanian, P. Berglund, J. P. Conlon and F. Quevedo, [*Systematics of moduli stabilisation in Calabi-Yau flux compactifications*]{}, JHEP [**0503**]{}, (2005) 007 \[arXiv:hep-th/0502058\]. T. W. Grimm, [*Non-Perturbative Corrections and Modularity in N=1 Type IIB Compactifications*]{}, arXiv:0705.3253 \[hep-th\]. F. Denef, M. R. Douglas and B. Florea, [*Building a better racetrack*]{}, JHEP [**0406**]{}, (2004) 034 \[arXiv:hep-th/0404257\]. K. Becker, M. Becker, M. Haack and J. Louis, [*Supersymmetry breaking and alpha’-corrections to flux induced potentials*]{}, JHEP [**0206**]{}, (2002) 060 \[arXiv:hep-th/0204254\]. A. Westphal, [*de Sitter String Vacua from Kahler Uplifting*]{}, JHEP [**0703**]{}, (2007)102 \[arXiv:hep-th/0611332\]. V. Balasubramanian and P. Berglund, [*Stringy corrections to Kähler potentials, SUSY breaking, and the cosmological constant problem*]{}, JHEP [**0411**]{}, (2004) 085 \[arXiv:hep-th/0408054\]. M. Grana, [*Flux compactifications in string theory: A comprehensive review*]{}, Phys. Rept. [**423**]{}, (2006) 91 \[arXiv:hep-th/0509003\]. S. Kachru, R. Kallosh, A. Linde and S. P. Trivedi, [*De Sitter vacua in string theory*]{}, Phys. Rev. D [**68**]{} (2003) 046005 \[arXiv:hep-th/0301240\]. S. Ferrara, R. Kallosh and A. Strominger, [*N=2 extremal black holes*]{}, Phys. Rev. D [**52**]{}, (1995) 5412 \[arXiv:hep-th/9508072\]; A. Strominger, [*Macroscopic Entropy of $N=2$ Extremal Black Holes*]{}, Phys. Lett. B [**383**]{}, (1996) 39 \[arXiv:hep-th/9602111\]; H. Ooguri, A. Strominger and C. Vafa, [*Black hole attractors and the topological string*]{}, Phys.Rev. D [**70**]{}, (2004) 106007 \[arXiv:hep-th/0405146\]; L. Andrianopoli, R. D’Auria, S. Ferrara and M. Trigiante, [*Extremal black holes in supergravity*]{}, arXiv:hep-th/0611345; S. Ferrara and M. Gunaydin, [*Orbits and attractors for N = 2 Maxwell-Einstein supergravity theories in five dimensions*]{}, Nucl. Phys. B [**759**]{}, (2006) 1 \[arXiv:hep-th/0606108\]; M. Alishahiha and H. Ebrahim, [*New attractor, entropy function and black hole partition function*]{}, JHEP [**0611**]{}, (2006) 017 \[arXiv:hep-th/0605279\]; D. Astefanesei, K. Goldstein and S. Mahapatra, [*Moduli and (un)attractor black hole thermodynamics*]{}, arXiv:hep-th/0611140; D. Astefanesei, K. Goldstein, R. P. Jena, A. Sen and S. P. Trivedi, [*Rotating attractors*]{}, JHEP [**0610**]{}, (2006) 058 \[arXiv:hep-th/0606244\]. S. Ferrara, G. W. Gibbons and R. Kallosh, [*Black holes and critical points in moduli space*]{}, Nucl. Phys. B [**500**]{}, (1997) 75 \[arXiv:hep-th/9702103\]. K. Goldstein, N. Iizuka, R. P. Jena and S. P. Trivedi, [*Non-supersymmetric attractors*]{}, Phys. Rev. D [**72**]{}, (2005) 124021 \[arXiv:hep-th/0507096\]; P. K. Tripathy and S. P. Trivedi, [*Non-supersymmetric attractors in string theory*]{}, JHEP [**0603**]{}, (2006) 022 \[arXiv:hep-th/0511117\]; A. Sen, [*Black hole entropy function and the attractor mechanism in higher derivative gravity*]{}, JHEP [**0509**]{},(2005) 038 \[arXiv:hep-th/0506177\]; B. Sahoo and A. Sen, [*Higher derivative corrections to non-supersymmetric extremal black holes in N = 2 supergravity*]{},\[arXiv:hep-th/0603149\]; P. Kaura and A. Misra, [*On the existence of non-supersymmetric black hole attractors for two-parameter Calabi-Yau’s and attractor equations*]{}, Fortsch. Phys. [**54**]{},(2006) 1109 \[arXiv:hep-th/0607132\]; A. Dabholkar, A. Sen and S. P. Trivedi, [*Black hole microstates and attractor without supersymmetry*]{}, JHEP [**0701**]{}, (2007) 096 \[arXiv:hep-th/0611143\]; S. Ferrara and A. Marrani, [*On the Moduli Space of non-BPS Attractors for N=2 Symmetric Manifolds*]{}, arXiv:0706.1667 \[hep-th\]. P. Candelas, A. Font, S. H. Katz and D. R. Morrison, [*Mirror symmetry for two parameter models. 2*]{}, Nucl. Phys. B [**429**]{}, (1994) 626 \[arXiv:hep-th/9403187\]. A. Giryavets, S. Kachru, P. K. Tripathy and S. P. Trivedi, [*Flux compactifications on Calabi-Yau threefolds*]{}, JHEP [**0404**]{}, 003 (2004) \[arXiv:hep-th/0312104\]. T. Mohaupt, [*Special geometry, black holes and Euclidean supersymmetry*]{}, arXiv:hep-th/0703037. G. Curio and V. Spillner, [*On the modified KKLT procedure: A case study for the P(11169)(18) model*]{}, arXiv:hep-th/0606047. A. Sen, [*Orientifold limit of F-theory vacua*]{}, Phys. Rev. D [**55**]{}, (1997) 7345 \[arXiv:hep-th/9702165\]. A. Ceresole, G. Dall’Agata, A. Giryavets, R. Kallosh and A. Linde, [*Domain walls, near-BPS bubbles, and probabilities in the landscape*]{}, Phys. Rev. D [**74**]{}, (2006) 086010 \[arXiv:hep-th/0605266\]. M. B. Green and M. Gutperle, [*Effects of D-instantons*]{}, Nucl. Phys. B [**498**]{}, (1997) 195 \[arXiv:hep-th/9701093\]. D. Robles-Llana, M. Rocek, F. Saueressig, U. Theis and S. Vandoren, [*Nonperturbative corrections to 4D string theory effective actions from SL(2,Z) duality and supersymmetry*]{}, Phys. Rev. Lett. [**98**]{}, (2007) 211602 \[arXiv:hep-th/0612027\]. E. Witten, [*Non-Perturbative Superpotentials In String Theory*]{}, Nucl. Phys. B [**474**]{}, (1996) 343 \[arXiv:hep-th/9604030\]. M. Berg, M. Haack and E. Pajer, [*Jumping Through Loops: On Soft Terms from Large Volume Compactifications*]{}, arXiv:0704.0737 \[hep-th\]. R. Emparan, G. Horowitz, [*Microstates of a Neutral Black Hole in M Theory*]{}, hep-th/0607023. A. Misra and P. Shukla, [*Large Volume Axionic Swiss-Cheese Inflation*]{}, arXiv:0712.1260 \[hep-th\].
$(G^{-1})^{\rho_1{\bar\rho}_1}|\partial_{\rho_1}W_{np}|^2$ $\frac{{\cal V}\sqrt{ln {\cal V}}}{{\cal V}^{2n^1e^{-\phi}}}$
------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------
$(G^{-1})^{\rho_2{\bar\rho}_2}|\partial_{\rho_2}W_{np}|^2$ ${\cal V}^{\frac{4}{3}}e^{-2n^2e^{-\phi}{\cal V}^{\frac{2}{3}}}$
$(G^{-1})^{\rho_1{\bar\rho_2}}\partial_{\rho_1}W_{np}{\bar\partial}_{\bar\rho_2}{\bar W}_{np} + c.c.$ $\frac{{\cal V}^{\frac{2}{3}} ln {\cal V} e^{-n^2e^{-\phi}{\cal V}^{\frac{2}{3}}}}{{\cal V}^{n^1e^{-\phi}}}$
$(G^{-1})^{G^1{\bar G^1}}|\partial_{G^1}W|^2 + (G^{-1})^{G^2{\bar G^2}}|\partial_{G^2}W|^2 + $ ${\cal V}^{1-(2n^1\ or\ 2n^2\ or\ n^1+n^2)e^{-\phi}}$
$(G^{-1})^{G^1{\bar G^2}}\partial_{G^1}W_{np}{\bar\partial}_{\bar G^2}{\bar W}_{np} + c.c.$
$(G^{-1})^{\rho_1{\bar G^1}}\partial_{\rho_1}W_{np}{\bar\partial}_{\bar G^1}{\bar W}_{np} + c.c.$ $\frac{ln {\cal V} }{{\cal V}^{2n^1e^{-\phi}}}$
$(G^{-1})^{\rho_2{\bar G^1}}\partial_{\rho_2}W_{np}{\bar\partial}_{\bar G^1}{\bar W}_{np} + c.c.$ $\frac{{\cal V}^{\frac{2}{3}} e^{-n^1e^{-\phi}{\cal V}^{\frac{2}{3}}}}{{\cal V}^{n^1e^{-\phi}+\frac{1}{3}}}$
: $(G^{-1})^{A{\bar B}}\partial_A W_{np}{\bar\partial}_{\bar B}{\bar W}_{np}$
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
$(G^{-1})^{\rho_1{\bar\rho}_1}\partial_{\rho_1}K{\bar\partial}_{\bar\rho_1}{\bar W}_{np}+c.c.$ $\frac{ln {\cal V}}{{\cal V}^{n^1e^{-\phi}}}$
------------------------------------------------------------------------------------------------ -----------------------------------------------------------------------------------------------------
$(G^{-1})^{\rho_2{\bar\rho}_2}\partial_{\rho_2}K{\bar\partial}_{\bar\rho_1}{\bar W}_{np}+c.c.$ ${\cal V}^{\frac{2}{3}}e^{-n^2e^{-\phi}{\cal V}^{\frac{2}{3}}}$
$(G^{-1})^{\rho_1{\bar\rho}_2}\partial_{\rho_1}K{\bar\partial}_{\bar\rho_2}{\bar W}_{np}+c.c.$ $\frac{(ln {\cal V})^{\frac{3}{2}}e^{-n^2e^{-\phi}{\cal V}^{\frac{2}{3}}}}{{\cal V}^{\frac{1}{3}}}$
$(G^{-1})^{G^1{\bar G^1}}\partial_{G^1}K{\bar\partial}_{\bar G^1}{\bar W}_{np} + $\frac{1}{{\cal V}^{n^1e^{-\phi}}}$
(G^{-1})^{G^1{\bar G^1}}\partial_{G^1}K{\bar\partial}_{\bar G^1}{\bar W}_{np} +$
$(G^{-1})^{G^1{\bar G^2}}\partial_{G^1}K{\bar\partial}_{\bar G^2}{\bar W}_{np} +
(G^{-1})^{\rho_2{\bar G^1}}\partial_{\rho_2}K{\bar\partial}_{\bar G^1}{\bar W}_{np} + c.c.$
$(G^{-1})^{\rho_1{\bar G^1}}\partial_{\rho_1}K{\bar\partial}_{\bar G^1}{\bar W}_{np} + c.c.$ $\frac{(ln {\cal V})^{\frac{3}{2}}}{{\cal V}^{1+n^1e^{-\phi}}}$
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
: $(G^{-1})^{A{\bar B}}\partial_AK{\bar\partial}_{\bar B}{\bar W}_{np} + c.c.$
---------------------------------------------------------------------------------------------------------------------------------------------------
$(G^{-1})^{\rho_1{\bar\rho}_1}|\partial_{\rho_1}K|^2$ $\frac{(ln {\cal V})^{\frac{3}{2}}}{{\cal V}}$
------------------------------------------------------------------------------------------------ --------------------------------------------------
$(G^{-1})^{\rho_2{\bar\rho}_2}|\partial_{\rho_2}K|^2$ ${\cal O}(1)$
$(G^{-1})^{\rho_1{\bar\rho}_2}\partial_{\rho_1}{\bar\partial}_{\bar\rho_2}K + c.c.$ $\frac{(ln {\cal V})^{\frac{3}{2}}}{{\cal V}}$
$(G^{-1})^{G^1{\bar G^1}}|\partial_{G^1}K|^2 + (G^{-1})^{G^2{\bar G^2}}|\partial_{G^2}K|^2 + $ $\frac{1}{\cal V}$
$(G^{-1})^{G^1{\bar G^2}}\partial_{G^1}K{\bar\partial}_{\bar G^2}K
+ (G^{-1})^{\rho_2{\bar G^1}}\partial_{\rho_2}K{\bar\partial}_{\bar G^1}K + c.c.$
$(G^{-1})^{\rho_1{\bar G^1}}\partial_{\rho_1}K{\bar\partial}_{\bar G^1}K + c.c.$ $\frac{(ln {\cal V})^{\frac{3}{2}}}{{\cal V}^2}$
---------------------------------------------------------------------------------------------------------------------------------------------------
: $(G^{-1})^{A{\bar B}}\partial_AK{\bar\partial}_{\bar B}K$
[^1]: e-mail: [email protected]
[^2]: email: [email protected]
[^3]: The term “Swiss cheese" (See [@SwissCheese]) is used to denote those Calabi-Yau’s whose volume can be written as: ${\cal V}=(\tau^B + \sum_{i\neq B} a_i\tau^S_i)^{\frac{3}{2}} - (\sum_{j\neq B}b_j\tau^S_j)^{\frac{3}{2}} - ...$, where $\tau^B$ is the volume of the big divisor and $\tau^S_i$ are the volumes of the $h^{1,1}-1$ (corresponding to the (1,$h^{1,1}-1$)-signature of the Hessian) small divisors. The big divisor governs the size of the Swiss cheese and the small divisors control the size of the holes of the same Swiss cheese.
[^4]: This is induced by the group action: $(x_1,x_2,x_3,x_4,x_5;\psi,\phi)\rightarrow(\alpha^{A_1}x_1,\alpha^{A_2}x_2,\alpha^{A_3}x_3,
\alpha^{6A_4}x_4,\alpha^{9A_5}x_5;
\alpha^{-a}\psi,\alpha^{-6a}\phi)$ where $a=A_1 + A_2 + A_3 + 6A_4 + 9A_5$ and $(A_1,A_2,A_3,A_4,A_5)$ are related to the coefficients of the most general degree-18 polynomial in $(x_1,x_2,x_3,x_4,x_5)$ invariant under $G={\bf Z}_6\times{\bf Z}_{18}$ (${\bf Z}_6:(0,1,3,2,0,0); {\bf Z}_{18}:(1,-1,0,0,0)$). The mirror to $p=x_1^{18} + x_2^{18} + x_3^{18} + x_4^3 + x_5^2 =0$, according to the Greene-Plesser construction, is given by $\{p=0\}/G$. The $G-$invariant polynomial is given by: $A_0^\prime\prod_{i=1}^6x_i + A_1^\prime x_1^2 x_2^2 x_3^2 x_4^2
+ A_2^\prime x_1^3 x_2^3 x_3^3 x_5 + A_3^\prime x_1^4 x_2^4 x_3^4 x_4 + A_4^\prime x_1^6 x_2^6x_3^6
+ A_5^\prime x_1^{18} + A_6^\prime x_2^{18} + A_7^\prime x_3^{18} + A_8^\prime x_4^3 + A_9^\prime x_5^2$; some of the deformations can be redefined away by suitable automorphisms.
[^5]: The Wronskian of these three solutions is given by $\frac{27i}{(2\pi)^3}e^{i\pi\nu}(1-\phi^3)^{\nu-1}\neq0,\nu\in{\bf Z}$.
[^6]: For techniques in special geometry relevant to this work, see [@Mohaupt] for a recent review; see [@Curio+Spillner] for moduli-stablization calculations as well.
[^7]: See [@Denef+Douglas+Florea] - $\chi(CY_4)=6552$ where the $CY_4$ for the ${\bf WCP^4}[1,1,1,6,9]$-model, is the resolution of a Weierstrass over a three-fold B with $D_4$ and $E_6$ singularities along two sections, with the three-fold a ${\bf CP^1}$-fibration over ${\bf CP^2}$ with the two divisors contributing to the instanton superpotential a la Witten being sections thereof.
[^8]: To make the idea more explicit, the involution $\sigma$ under which the NS-NS two-form $B_2$ and the RR two-form $C_2$ are odd can be implemented as follows. Let $z_i, {\bar z}_i, i=1,2,3$ be the complex coordinates and the action of $\sigma$ be defined as: $z_1\leftrightarrow z_2, z_3\rightarrow z_3$; in terms of the $x_i$ figuring in the defining hypersurface in equation (1) on page 2, one could take $z_{1,2}=\frac{x_{1,2}^9}{x_5}$, etc. in the $x_5\neq0$ coordinate patch. One can construct the following bases $\omega^{(\pm)}$ of real two-forms of $H^2$ even/odd under the involution $\sigma$: $$\begin{aligned}
\label{eq:bases}
& & \omega^{(-)}=\{\sum(dz^1\wedge d{\bar z}^{\bar 2} - dz^2\wedge d{\bar z}^{\bar 1}),
i(dz^1\wedge d{\bar z}^{\bar 1} - dz^2\wedge d{\bar z}^{\bar 2})\}\equiv\{\omega^{(-)}_1,\omega^{(-)}_2\},\nonumber\\
& & \omega^{(+)}=\{\sum i(dz^1\wedge d{\bar z}^{\bar 2} + dz^2\wedge d{\bar z}^{\bar 1}),\sum i dz^1\wedge d{\bar z}^{\bar 1}\}\equiv\{\omega^{(+)}_1,\omega^{(+)}_2\}.\end{aligned}$$ This implies that $h^{1,1}_+(CY_3)=h^{1,1}_-(CY_3)=2$ - the two add up to give 4 which is the [**real**]{} dimensionality of $H^2(CY_3)$ for the given Swiss Cheese Calabi-Yau. As an example, let us write down $B_2\in{\bf R}$ as $$\begin{aligned}
\label{eq:Bform}
B_2 & = & B_{1{\bar 2}}dz^1\wedge d{\bar z}^{\bar 2} + B_{2{\bar 3}}dz^2\wedge d{\bar z}^{\bar 3} + B_{3{\bar 1}}dz^3\wedge d{\bar z}^{\bar 1} + B_{2{\bar 1}}dz^2\wedge d{\bar z}^{\bar 1} + B_{3{\bar 2}}dz^3\wedge d{\bar z}^{\bar 2} + B_{1{\bar 3}}dz^1\wedge d{\bar z}^{\bar 3}\nonumber\\
& & + B_{1{\bar 1}}dz^1\wedge d{\bar z}^{\bar 1}+ B_{2{\bar 2}}dz^2\wedge d{\bar z}^{\bar 2}+ B_{3{\bar 3}}dz^3\wedge d{\bar z}^{\bar 3}.\end{aligned}$$ Now, using (\[eq:bases\]), one sees that by assuming $B_{1{\bar 2}}=B_{2{\bar 3}}=B_{3{\bar 1}}=b^1$, and $B_{1{\bar 1}}=-B_{2{\bar 2}}= i b^2, B_{3{\bar 3}}=0$, one can write $B_2=b^1\omega^{(-)}_1 + b^2\omega^{(-)}_2\equiv\sum_{a=1}^{h^{1,1}_-=2}b^a\omega^{(-)}_a$.
[^9]: If one puts in appropriate powers of the Planck mass $M_p$, $\frac{|b^a|}{\pi}<<1$ is equivalent to $|b^a|<<M_p$, i.e., NS-NS axions are super sub-Planckian.
[^10]: See [@Mohaupt] for a nice review of special geometry relevant to sections [**5**]{} and [**6**]{}.
[^11]: AM thanks S.Mathur for bringing [@EH] to our attention.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Darran Nathan, Eva Rosdiana, Chua Beng Koon[^1]'
nocite: '[@*]'
title: DAB Content Annotation and Receiver Hardware Control with XML
---
[Murray and Balemi: Using the Document Class IEEEtran.cls]{}
The Eureka-147 Digital Audio Broadcasting (DAB) standard defines the ’dynamic labels’ data field for holding information about the transmission content. However, this information does not follow a well-defined structure since it is designed to carry text for direct output to displays, for human interpretation. This poses a problem when machine interpretation of DAB content information is desired. Extensible Markup Language (XML) was developed to allow for the well-defined, structured machine-to-machine exchange of data over computer networks. This article proposes a novel technique of machine-interpretable DAB content annotation and receiver hardware control, involving the utilisation of XML as metadata in the transmitted DAB frames.
Introduction
============
are various digital audio broadcast standards in existance. The European Eureka-147 DAB standard is described in this article. This standard offers advanced methods for source coding, channel coding, and modulation, with the DAB frames having fields defined for both data and audio. The main advantages of this standard over FM are CD-quality audio, a ’Single Frequency Network’ throughout the coverage area, data transmission capability, spectrum usage efficiency, and suitability for mobile reception.
However, one limitation of this standard is that it does not define structured data fields for annotation of the audio content. The ’dynamic-labels’ field is used for carrying transmission content information, such as artiste name and song title. This field does not impose a well-defined structure on the data it holds, since it is designed to carry text that will be directly displayed to the end-user. As such, it has limited machine interpretability.
Machine interpretability of received data will allow for powerful and flexible applications at the receiver; different receivers may choose to decode different parts of the bitstream, or effect different responses to the same received data, depending on the application.
Extensible Markup Language (XML) is a specification of the World Wide Web Consortium (W3C) that allows for the description (or markup) of data using a set of well-defined tags. The utilisation of XML gives data a structure that allows for machine interpretation of that data. Its main application thus far has been in the exchange of business information over the internet.
The need for structured annotation of DAB content, and the ability of XML to provide structured annotation of data, imply a synergistic combination of both technologies in a novel application of XML and an innovative technique of DAB transmission content description - the transmission of XML over DAB.
This article begins with an introduction to the ’Campus DAB’ initiative undertaken by NgeeAnn Polytechnic, which led to the innovation of ’XML over DAB’ as a base technology. Next, a description of how non-audio data is transmitted in DAB frames is given. It is shown how this leads to the problem of unstructured content annotation. The following section explains how XML is designed to give data a well-defined structure, and discusses the XML schema and messaging protocol developed for both annotation of DAB transmission content and control of the receiver hardware. Next, the necessary transmission frame header settings and extensions to the existing DAB standard to allow for transmission of XML are discussed. Finally, the architectural set-up and software deployment as developed in the working prototype system are analyzed.
The Campus DAB Initiative {#sectCampusDAB}
=========================
Traditionally, DAB has been used by radio broadcasters who wish to offer their listeners the clarity of digital audio and the transmission of data, such as weather forecasts and traffic information, alongside that audio. The DSP Technology Centre of NgeeAnn Polytechnic has taken DAB to the next level of application, under its ’Campus DAB’ initiative. This project will see DAB receivers deployed at various locations throughout the campus, each of which selectively decodes a particular subchannel of the DAB transmission, and performs an action on the decoded data depending on the receiver’s location and purpose. For example, a DAB receiver located at the canteen may be configured to decode the ’Campus Radio’ subchannel and output the audio through speakers, while another one located at a lift lobby may be used to decode and display announcement messages transmitted on another subchannel of the same DAB transmission ensemble. Yet another receiver located at a study area may decode e-learning content on a PC terminal.
The complication arises from the fact that different data meant for different receivers may be transmitted on the same subchannel; for example, an announcement message meant for the Mathematics Department should be decoded by a receiver located at that department, but not one located at the Engineering Department. This means that the DAB receivers need to be able to interpret the received data, and decide on a course of action depending on that data. This led to the development of the ’XML over DAB’ concept that has since been implemented and tested as a prototype system at the DSP Centre. The next sections describe DAB data transmission and this prototype implementation.
DAB Data Transmission {#sectDataTransmission}
=====================
Data is transmitted in the Programme Associated Data (PAD) field of each DAB audio frame, as shown in Figure \[figDABaudioFrame\]. A data object to be transmitted is split up into multiple segments, so that different segments can be sent out in the PAD fields of multiple DAB audio frames. At the receiver, these segments are extracted and combined to form the complete data object.
![A Single DAB Audio Frame[]{data-label="figDABaudioFrame"}](DABaudioFrame){width="40.00000%"}
The two main methods of data transmission in the PAD field are Dynamic Labels and Multimedia Object Transfer (MOT) objects [@bib301234].
Dynamic Labels hold audio descriptive information for presentation to the end-user, such as song title and artiste name. The data is sent as unstructured strings of ASCII characters, for direct display by the receiver. The absence of a well-defined structure is insignificant, since the Dynamic Labels Field has been defined to carry information for human interpretation. However, a problem arises when attempting machine interpretation of the Dynamic Label. In this case, the unstructured information poses a barrier to the ability of the machine to identify portions of the Dynamic Label with their corresponding intended use. For example, the dynamic label “Dancing Queen by ABBA” can be easily understood by a human being as describing a song called “Dancing Queen” sung by “ABBA”. To a machine, on the other hand, there is no way of identifying the purpose of one segment of text from another in an unstructured block. In this case, “ABBA” will have an equal probability of being a song title as “Queen” or “Dancing”. This illustrates the limited use of the Dynamic Labels Field in the machine-interpretable annotation of a DAB transmission.
The MOT protocol has been defined for the transmission of data objects, such as JPEG image files. The file to be transmitted is segmented into Data Groups, as shown in Figure \[figMOTobj\]. These Data Groups are then sent into a PAD encoder for transmission in the DAB audio frames. The MOT specification currently defines a limited, extensible set of MOT object types comprising of the ContentType and ContentSubType fields of the MOT Object header.
![Building the MOT Data Group[]{data-label="figMOTobj"}](MOTobj){width="45.00000%"}
This explanation of the DAB non-audio data transmission capability shows that only unstructured audio annotation information can be sent in the Dynamic Labels Field. MOT objects, however, offer an avenue for transmitting data objects that can be used for providing structured audio annotation information. How this can be done is described in Section \[sectXMLoverDAB\].
Using XML {#sectXML}
=========
The XML specification was created by the W3C to define a markup language that can be extended to suit the context of the data it describes. XML is designed to be used for the marking-up of data in a machine-to-machine information exchange environment. The mark-up is done with ’tags’ that designate the properties of the enclosed data. This gives the data a well-defined structure that allows for machine interpretation of that data. However, it is the extensibility of XML that gives this markup language its true power - by utilising this capability, different sets of tags can be defined for different uses of data.
The structured data annotation capability of XML, its inherent design for machine interpretability, and the extensibility of the language, imply that an application of XML can be defined for annotation of the DAB audio bitstream. This will give the transmitted data a well-defined structure that allows machines to interpret, understand and react to audio content.
To use XML in such an environment, two main areas have to be looked into: the development of an XML schema that serves this purpose (i.e., the extension of XML), and the utilisation of a suitable XML messaging protocol.
The XML Schema - DABml
----------------------
A schema declares the vocabulary (tags) of an XML application, as well as the usage of these tags. It was decided that the schema defined has to offer the ability to describe the transmitted bitstream content, and specify the reaction the machine should exhibit to particular content received.
As a result, the DABml schema has been designed with four main tags:
- $<$audioContent$>$ describes the content of the audio portion of the DAB transmission, such as the song title, artiste name and music genre;
- $<$dataContent$>$ describes the content of the non-audio data portion of the DAB transmission, such as image files, web pages, etc;
- $<$hardwareControl$>$ controls the DAB receiver hardware and the computer to which the receiver is attached. Such controls may include turning down the volume or recording a subchannel to the PC’s harddisk;
- $<$behaviours$>$ pulls together the descriptive $<$audioContent$>$ & $<$dataContent$>$ tags, and the reactive $<$hardwareControl$>$ tag, to define the response to particular content received.
The UML (Unified Markup Language) static structure model of parts of the DABml schema is given in Figure \[figSchemaStaticUMLmain\].
![UML Static Structure Model of the four main DABml tags[]{data-label="figSchemaStaticUMLmain"}](schemaStaticUMLmain){width="40.00000%"}
The XML Messaging Protocol
--------------------------
To use XML in the exchange of information between a sender and a receiver, a well-defined communications protocol has to be utilised. This XML messaging protocol lays down the rules of communications between parties, so that messages are received as intended by the sender.
There are multiple XML messaging protocol specifications in existance. SOAP is one such specification that is gaining great momentum due to its open standard, widespread industry support, and simple design. As such, SOAP was selected as the DABml messaging protocol.
SOAP messages are composed of an outer envelope that encloses a header and a body. The header contains metadata about the rest of the message, while the body holds the message payload. The example below shows how SOAP can be used to transport a DABml message which specifies the artiste name and song title as ’ABBA’ and ’Dancing Queen’ respectively.
<SOAP-ENV:Envelope xmlns:SOAP-ENV=
"http://schemas.xmlsoap.org/soap/envelope/"
SOAP-ENV:encodingStyle=
"http://schemas.xmlsoap.org/soap/encoding/">
<SOAP-ENV:Header>
</SOAP=ENV:Header>
<SOAP-ENV:Body>
<dabml:DAB xmlns:dabml=
"http://location/dabml/">
<audioContent>
<artiste>ABBA</artiste>
<songTitle>Dancing Queen</songTitle>
</audioContent>
</dabml:DAB>
</SOAP-ENV:Body>
</SOAP-ENV:Envelope>
Transmitting XML over DAB {#sectXMLoverDAB}
=========================
The XML message to be transmitted has to be inserted into the PAD field of DAB Audio Frames, of the particular subchannel concerned. To notify the receiver that such a message is available, information is inserted into the Fast Information Channel (FIC). This requirement is not fully satisfied by the DAB standard [@bib300401]. However, the standard has defined several fields as being ’Reserved for future definition’. By utilising these fields to extend the current state of the DAB standard, the requirements for transmission of XML over DAB can be satisfied.
Another requirement is indication of the interval during which an XML message is ’valid’. This period of validity can, for example, indicate when the information in the XML message applies to the audio content. This allows for earlier transmission of a message that will be effective only at a specified later time.
The requirements for transmitting XML over DAB, and how these requirements are met, are detailed below:
User Application Information
----------------------------
As shown in Figure \[figTransFrame\], the DAB transmission frame is divided into the FIC and the MSC (Main Service Channel). The FIC holds information about the MSC, which in turn holds all the subchannels. Information in the FIC is carried in FIGs (Fast Information Groups), each kind of which is identified by a type and extension in the format FIG TYPE/EXTENSION.
![Structure of the DAB Transmission Frame[]{data-label="figTransFrame"}](transFrame){width="30.00000%"}
FIG 0/13 (’User Application Type’) allows for signaling of the kind of data being carried in each subchannel. This is so that the user application will be able to use the correct decoder to extract the information from that subchannel. It is required that a User Application Type be defined for XML messages. The bit representation of ’00000000110’, defined in the DAB standard as ’Reserved for future definition’, has been utilised to represent ’MOT XML Message’ in this prototype implementation.
MOT ContentType / ContentSubType
--------------------------------
For the MOT decoder to decide where to channel a decoded MOT object, there must be some technique of signaling the type of data in the object. The MOT specification defines the ’ContentType’, ’ContentSubType’ and ’ContentName’ fields of the MOT object header core for this purpose, as shown in Figure \[figContentTypeSub\].
![MOT Object Structure[]{data-label="figContentTypeSub"}](contentTypeSub){width="40.00000%"}
To transmit XML, the MOT Header ’ContentType’ and ’ContentSubType’ will have to contain ’000000’ (’General Data’) and ’000000001’ (’MIME/HTTP’) respectively, so that the Header Extension can contain the ’ContentName’ string of ’TEXT/XML’. This ContentName will identify the presence of XML annotation to the receiver.
Object Validity
---------------
As required for the synchronized interpretation of XML messages, and as offered by the MOT specification, the validity of MOT objects is signaled by the parameter data field of the MOT object’s header extension. The parameters ’StartValidity’ and ’ExpireTime’ are coded in UTC format, and represent the start and end of the validity period respectively.
The System Architecture {#sectArch}
=======================
Since SOAP is used to communicate with the DAB receiver, a computing platform is needed to extract and interpret the SOAP message before sending lower-level instructions to the DAB hardware. This computing platform may take the form of a PC which is connected to the DAB receiver via the Universal Serial Bus (USB).
If this PC is connected to the network, the possibility arises for remote communications with the DAB receiver. In this case, that PC takes on the role of a server, while any other PC on the network plays the client.
Since SOAP has been designed for client-server communications over computer network, the client in this set-up can utilise the same SOAP protocol and XML schema as that transmitted over the air, to communicate with the DAB receiver. The origin of the message destined for the receiver is transparent to the part of the server application that interprets SOAP - whether the message arrives from a client over the network or from a broadcast over the air is of concern only to the higher layers.
Deployment of Resources
-----------------------
The set-up of the prototype system is illustrated in the UML Deployment Diagram of Figure \[figUMLdeployment\].
![UML Deployment Diagram of the set-up[]{data-label="figUMLdeployment"}](UMLdeployment){width="50.00000%"}
### DAB Receiver
- The DAB Receiver is connected to the Server via the USB. The Server controls and receives data from the receiver over this bus.
### Server
The server is the link between the DAB receiver and the ’outside world’. All DABml SOAP messages are interpreted and executed by the server. To do this, the following software are deployed:
- Server-side Application: This application runs various threads to extract SOAP from the subchannel bitstream, interpret the message, and perform necessary hardware control;
- SAX XML parser: It is used by the server-side application to parse the XML in the SOAP message, so that pertinent information can be extracted;
- ASP Server: This is the ’listener’ that receives messages from network-based clients and passes them to the server-side application;
- DAB Receiver Device Driver: The driver abstracts bus communications between the DAB receiver hardware and the server. The server-side application utilises the driver for all communications with the DAB receiver;
- SOAP Toolkit Library: A SOAP Toolkit is a library which exposes a set of APIs to allow for transparent encoding and decoding of SOAP messages. The server-side application uses this library to send and receive SOAP messages over the network.
### Client
The client communicates with the server using DABml SOAP messages. These messages may involve requests for DAB subchannel content information, or control of the hardware. The following are deployed at the client:
- Client Application: This software presents an interface to the end-user, which displays information about the DAB services and accepts input from the user. It acts as a ’frontend’ for communications with the DAB receiver, making it transparent to the end-user that the receiver is not connected to his PC, but is instead remotely accessed over the network;
- SOAP Toolkit Library: Just as for the case of the server, the client uses this set of APIs to encode / decode SOAP messages for communications over the network.
How It Works
------------
Figure \[figLayers\] shows the main layers of software and hardware involved in the prototype set-up developed.
![Main Hardware and Software Layers of the Set-Up[]{data-label="figLayers"}](Layers){width="50.00000%"}
The server-side application consists of three threads of execution, labelled on Figure \[figLayers\] as ’Thread 1’, ’Thread 2’ and ’Thread 3’.
### Thread 1 (Server-Side Application Main Thread) performs the following tasks:
- It is the first thread started when the application is loaded. Correspondingly, it has to start the other threads;
- SOAP messages received from either the DAB broadcast or the client are parsed using SAX, interpreted, and processed;
- If necessary, it builds up SOAP messages for reply to the client through the SOAP Toolkit Library;
- It monitors the set of ’behaviours’ defined, and triggers a reaction whenever a message of interest is received.
### Thread 2 (DAB Subchannel Bitstream SOAP Extractor) performs the following tasks:
- Extracts and builds up the MOT object to obtain the SOAP message;
- Signals the Server-Side Application Main Thread and passes it the new SOAP message decoded.
### Thread 3 (Hardware Control) performs the following tasks:
- All hardware access is done by this thread. PC resources are accessed via the Operating System (OS) API. The DAB receiver hardware is accessed via the device driver;
- It performs all necessary initialisations of the DAB hardware upon startup, such as tuning to a default ensemble and service;
- It engages in its own automatic monitoring and control of the DAB receiver, such as adjustment of the Automatic Frequency Control (AFC) to cater for variations in the received signal;
- It receives hardware-control messages from the Server-Side Application Main Thread, and performs the corresponding operations on the hardware.
- It saves decoded MOT objects from the DAB subchannel bitstream to the PC’s harddisk, the URLs of which are returned to the Server-Side Application Main Thread.
Figure \[figUMLseq\] shows a UML Sequence Diagram that illustrates an example of the operations that occur at the DAB receiver, the server, and the client. It also shows the calls and messages exchanged among them.
![UML Sequence Diagram of Control-Flow XML Utilisation Example[]{data-label="figUMLseq"}](UMLseq){width="50.00000%"}
In this scenario, the server initialises the DAB receiver and requests a default subchannel. This subchannel bitstream is sent to the server at regular intervals, upon which SOAP (if present) is extracted and interpreted. Some time later, a client connects to the server and requests information about the transmitted audio and data. This information can, for example, be used as service information for display to the end-user. The client then sends a new user-defined behaviour setting to the server, to which the server responds with a message indicating success in adding the new behaviour. Next, the server invokes behaviour reactions that operate on both the server itself (server-centric reaction, such as saving a bitstream to the harddisk) and the DAB receiver (device-centric reaction, such as turning up the volume). Some time later, the client sends the server a $<$hardwareControl$>$ message (such as switching of subchannels). The server correspondingly acts on the receiver, and relays the result of the operation back to the client.
This diagram also shows that all communications between the server and the client utilise SOAP, and all communications between the client and the DAB receiver have to go through the server.
The Prototype Implementation
----------------------------
The system architecture described in this section has been implemented as a prototype at the DSP Centre, to test its feasibility for deployment under the ’Campus DAB’ initiative (described in Section \[sectCampusDAB\]).
It is planned that the ’Campus DAB’ project will utilize PC-based DAB receivers, since PC resources such as harddisk space for data storage and the visual display monitor are desired at each deployment location.
The prototype has therefore been tested with both the client and server running on the same PC (to simulate a deployment location), as well as on separate PCs (to simulate a remote client being used to administer a deployment over the network).
The prototype implementation has been shown to be ideally suited for the ’Campus DAB’ initiative. In particular, the ability to remotely program the server at any deployment point from a client on any PC is a boon for administration of the Campus DAB network.
Conclusions
===========
This article has presented a novel technique of DAB content annotation and receiver hardware control involving the use of SOAP embedded in DAB transmission frames. The transmission of PAD in DAB has been described, and the problem of unstructured data exchange explained. The features of XML that allow for well-defined, structured machine-machine exchange of data has been illustrated, and the capabilities offered by a marriage of XML and DAB explained. An XML schema (DABml), which allows for both DAB content annotation and receiver hardware control, has been proposed and implemented. The client-server architecture set-up as demonstrated in the prototype test system has been discussed and shown to be suitable for campus-wide deployment under the Centre’s ’Campus DAB’ initiative.
Lessons Learned
---------------
One trade-off of this proposal is that XML is verbose text-based metadata that can significantly increase the overall size of data to be transmitted. This results in a lowering of the efficiency of the transmission.However, the fact that XML is just text brings about the advantage that it can be sent over the network without being a cause of concern for firewalls. This allows the same markup language and protocol to be used for messages sent either over the broadcast transmission or over the computer network to PCs beyond the firewall.
A mistake made in an early design and implementation was to require transfer of XML received over the air to the client for analysis, with the ’behaviours’ also stored at the clients. The rationale behind this idea was that the client-orientation would allow a single receiver to be shared among different clients, each of which picks out only the data it is interested in from the entire received DAB transmission channeled to each client.The problem encountered was that by performing subchannel selection and extraction at the client, followed by further MOT object extraction and decoding, and perhaps even decoding and output of the MPEG audio, the client PC processors were insufficiently powerful enough to handle all these tasks resulting in buffer overruns and loss of data. It was found that letting the DAB receiver hardware perform the subchannel selection and audio decoding, with the server receiving a particular subchannel for MOT data extraction / storage, and the client only involved in control and data request, resulted in an optimal distribution of processing-intensive tasks - although this meant each deployment point required a DAB receiver and server.
Future Implementation
---------------------
Given the many incompatible digital audio broadcast standards in existance, such as IBOC (USA), DAB (Europe) and DRM, DABml can be a common metadata language used across all these standards to describe the data content of the transmission regardless of the standard. This provides for an abstraction of the actual frame structure of the transmission, allowing a common higher-level application program concerned with data interpretation and display to be used across these standards.
This can even be taken one step further to the development of a ’broadcast markup language’ that will serve to describe content across all digital broadcast standards, such as DTV, DVB, XM Radio, Sirius, DAB and DRM.
Acknowledgments {#acknowledgments .unnumbered}
===============
Special thanks to Oliver Faust, Bernhard Sputh, and Lim Choo Min, of NgeeAnn Polytechnic. We gratefully acknowledge the funding support provided by the NgeeAnn Kongsi (Singapore).
[1]{}
ETSI, , Feb 1999.
ETSI, , April 2000.
Nathan D., Sputh B., Faust O. and Chua B.K., “Design and Features of an Intelligent PC-based DAB Receiver,” IEEE Transactions on Consumer Electronics, May 2002.
Mark Birbeck, et. al., , Wrox Press, 2000.
John Paul Mueller, , Que, 2002.
Grady Booch, James Rumbaugh, Ivar Jacobson, , Addison Wesley, 1999.
[^1]: The authors are with the DSP Technology Centre of NgeeAnn Polytechnic, Singapore. (e-mail: [email protected] \[Darran Nathan\], [email protected] \[Eva Rosdiana\], [email protected] \[Chua Beng Koon\]).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We have carried out two-dimensional simulations of the nonlinear evolution of unstable sheared magnetohydrodynamic flows. These calculations extend the earlier work of [@miura84] and consider periodic sections of flows containing aligned magnetic fields. Two equal density, compressible fluids are separated by a shear layer with a hyperbolic tangent velocity profile. We considered two cases: a strong magnetic field (Alfvén Mach number, $M_a = 2.5$) and a weak field ($M_a = 5$). Each flow rapidly evolves until it reaches a nearly steady condition, which is fundamentally different from the analogous gasdynamic state. Both MHD flows relax to a stable, laminar flow on timescales less than or of the order of 15 linear growth times, measured from saturation of the instability. That timescale is several orders of magnitude less than the nominal dissipation time for these simulated flows, so this condition represents an quasi-steady relaxed state analogous to the long-lived single vortex, known as “Kelvin’s Cat’s Eye”, formed in two-dimensional nearly ideal gasdynamic simulations of a vortex sheet.
The strong magnetic field case reaches saturation as magnetic tension in the displaced flow boundary becomes sufficient to stabilize it. That flow then relaxes in a straightforward way to the steady, laminar flow condition. The weak magnetic field case, on the other hand, begins development of the vortex expected for gasdynamics, but that vortex is destroyed by magnetic stresses that locally become strong. Magnetic topologies lead to reconnection and dynamical alignment between magnetic and velocity fields. Together these processes produce a sequence of intermittent vortices and subsequent relaxation to a nearly laminar flow condition in which the magnetic cross helicity is nearly maximized. Remaining irregularities show several interesting properties. A pair of magnetic flux tubes are formed that straddle the boundary between the oppositely moving fluids. Velocity and magnetic fluctuations within those features are closely aligned, representing Alfvén waves propagating locally downstream. The flux tubes surround a low density channel of hot gas that contains most of the excess entropy generated through the relaxation process.
author:
- 'Adam Frank , T. W. Jones , Dongsu Ryu'
- 'Joseph B. Gaalaas'
title: |
The MHD Kelvin-Helmholtz Instability:\
A Two-Dimensional Numerical Study
---
= cmmi10 scaled1 = cmmi10 = cmmi7 = cmmib10 scaled1 = cmmib10 = cmmib10 scaled 833 = cmbsy10 scaled1 = cmbsy10 = cmbsy10 scaled 833 1 = 1 = 1 = 9 = 9 = 9 = 10 = 10 = 10 =
Introduction
============
Sheared boundary flows are ubiquitous in astrophysical environments as diverse as the earth’s magnetopause and supersonic jets. The susceptibility of such boundaries to the Kelvin-Helmholtz (KH) instability is well- known (e.g., [@chan61; @birk91]). Development of the KH instability leads to momentum and energy transport, dissipation, mixing of fluids and, when magnetic fields are involved, the possibility of field amplification, reconnection and dynamical alignment, or self-organization processes. Although the nonlinear evolution of the KH instability has been studied in considerable detail for non-magnetized fluids through extensive analytical calculations and numerical simulations (e.g., [@corsher84] and references therein), much less progress has been made in understanding the nature of the KH instability in magnetohydrodynamics (MHD). That is both because the MHD KH instability is substantially more complex and because until recently adequate numerical MHD simulations were difficult. Yet it is clear that understanding the physics of the MHD KH instability would be very useful in a wide variety of contexts.
The basic MHD KH linear stability analysis was carried out a long time ago (e.g., [@chan61]), and it has since been applied to numerous astrophysical situations including those requiring cylindrical geometries (e.g., [@ferr81; @fied84]). It is clear from these studies that the influence a magnetic field has on the stability of an initial flow equilibrium depends on both the strength and orientation of the field. Obviously, an equilibrium exists only when the field initially is parallel to the shear layer. When the field is aligned also with the flow vectors, magnetic tension directly inhibits the instability, while if the field is perpendicular, its role in ideal MHD comes through pressure contributions modifying the characteristic magnetosonic wave speed. For equal density, incompressible, inviscid fluid layers separated discontinuously in a “vortex sheet” the linear growth rate is, $\Gamma = {1\over 2} |{\bmit k} \cdot {\bmit U_o}|[ 1~-~(2 c_a
\hat{\bmit k\cdot} \hat{\bmit B_o})^2/(\hat{\bmit k} \cdot{\bmit
U_o})^2]^{1/2}$, where $\bmit U_o$ is the velocity difference between the two layers, $c_a$ is the Alfvén speed, ${\bmit k}$ is the wave vector and $\hat{\bmit B_o}$ is the direction of the magnetic field ([@chan61]). From this, one also can see that a necessary condition for instability is $c_a < |(\hat {\bmit k} \cdot {\bmit
U_o})/(2 \hat {\bmit k} \cdot \hat{\bmit B_o})|$. The instability constraint on the Alfvén Mach number, $M_a$, is $M_a = U_o/c_a > 2$, when magnetic field, velocity and perturbation wave vectors line up. If the shear layer is of finite thickness, $\Gamma$ will generally fall below the value predicted above, with a maximum growth rate for wave numbers $k \sim
{1\over a}$, where $a$ measures the width of the transition layer. Similarly, the finite width of the shear layer stabilizes very short wavelength modes,${1\over k} << a$, preventing the divergence in the linear growth rate that occurs for a discontinuity. It appears that the same constraint on $c_a$ applies in the finite shear layer ([@miupr82], hereafter “MP”). Compressibility of the fluid also tends to reduce the growth of the instability. There has been much discussion in the literature about the existence or nonexistence of an upper limit of order unity to the sonic Mach number for KH unstable flows (e.g., [@lan44; @miles57; @blum70; @blumet75; @artma87; @miura90]). Especially when one considers superAlfvénic shear layers of finite thickness, it appears that there is no such limit, although the growth of the instability under highly supersonic conditions is generally much slower than the classical, incompressible result.
Even though these results tell us when to expect the MHD KH instability and when it might have time to develop fully, only nonlinear calculations can determine its consequences. In recent years a number of two-dimensional numerical studies have been published aiming at various aspects of the problem (e.g.,[@tajleb80; @miura84; @miura87; @miura92; @wangro84; @wu86; @belham88; @belmon89]), especially as it applies to the earth’s magnetosphere. However, the nonlinear physics of MHD KH unstable flows is very complex, and studies have generally been too limited by computational constraints (low numerical resolution and insufficient dynamical time coverage) to answer basic questions definitively about either the important physical processes involved or the eventual fluid state.
In the present paper we take some steps to help remedy that situation. Our intent is to address some basic issues identifying the character of fluid and magnetic field interactions once the MHD KH instability becomes nonlinear. Thus, we consider only the idealized, periodic KH instability, postponing for now a study of the evolution of KH unstable MHD flows in more general and realistic settings. In this way, we hope subsequently to understand better how those less controlled flows are formed. As our starting point we use the very nice work by [@miura84], who examined the evolution of some normal mode perturbations to an unstable MHD shear layer. That problem is both numerically well posed and leads to a straightforward evolution. The role of the magnetic field is most critical when it is aligned with the flow, so we shall focus on that configuration. This means that our simulations are strictly two-dimensional, leading to some important properties in the flows, especially as they apply to the evolution of the magnetic field. For example, for planar symmetry a regenerative dynamo action cannot develop to amplify the field ([*e.g.,*]{} [@zel57; @moff78]). On the other hand, both 2D and 3D MHD turbulent flows cascade energy to larger wavenumbers ([*e.g.,*]{} [@pouq78]), whereas 2D hydrodynamical turbulence produces an inverted energy cascade. A 2D simulation, however, enables us to achieve high numerical resolution that can be crucial to obtaining solutions in which the large scale motions and global statistical properties are not strongly influenced by numerical diffusion. Since the character of the flow evolution is rather different for flows in which the initial magnetic field strength is large ($c_a \sim U_o$) than when it is small ($c_a << U_o$) we consider both of those cases. Miura’s calculations were terminated shortly after the nonlinear saturation of the instability, because of concerns about numerical limitations and because the periodic version of the instability is idealized as already mentioned (cf. [@wu86]). On the other hand, understanding how the flow relaxes after saturation could be very useful in deciphering the fully nonlinear behavior of the instability in more general settings. So, we have extended our simulations until an approximate equilibrium is achieved. We note here that stable, equilibrium flows develop in these simulations on a much shorter timescale than that required to dissipate the kinetic energy in the initial configurations. Miura’s calculations were carried out on a rather coarse numerical grid ($100\times 100$ zones), and he made no effort to examine the convergence of his results. On the other hand, the numerical evolution of complex flows is often highly dependent on numerical diffusion that effectively limits the Reynolds numbers in the flow. [@maclet94] recently emphasized the special importance of this concern for MHD flows. Thus, we have carried out each of our simulations at two or more resolutions to provide some information about that issue. We note that [@maletal95] have independently concluded a similar numerical study using different methods. Their results are entirely consistent with ours and their discussion complements this one.
The plan of the paper is as follows. In §2 we describe our numerical methods and the initial conditions used. Section 3 contains a discussion of the characteristics of the several phases of the evolution in the instability, while §4 is a brief summary with conclusions.
Methods and Initial Conditions
==============================
MHD describes the behavior of the combined system of a conducting fluid and magnetic field in the limit that the displacement current and the separation between ions and electrons are neglected. So, the MHD equations represent coupling of the equations of fluid dynamics with the Maxwell’s equations of electrodynamics. When the effects of viscosity, electrical resistivity and thermal conductivity can be neglected on large scales one works with the ideal, compressible MHD equations, $${\partial\rho\over\partial t} + {\bmsy\nabla}\cdot
\left(\rho {\bmit u}\right) = 0,
\label{masscon}$$ $${\partial{\bmit u}\over\partial t} + {\bmit u}\cdot{\bmsy\nabla}
{\bmit u} +{1\over\rho}{\bmsy \nabla}p - {1\over\rho}
\left({\bmsy\nabla}\times{\bmit B}\right)\times{\bmit B} = 0,
\label{forceeq}$$ $${\partial p\over\partial t} + {\bmit u}\cdot{\bmsy\nabla}p
+ \gamma p{\bmsy\nabla}\cdot{\bmit u} = 0,
\label{energy}$$ $${\partial{\bmit B}\over\partial t} - {\bmsy\nabla}\times
\left({\bmit u}\times{\bmit B}\right) = 0,
\label{induct}$$ along with the constraint ${\bmsy\nabla}\cdot{\bmit B}=0$ imposed to account for the absence of magnetic monopoles (e.g., [@priest84]). The entropic gas equation of state is $p\propto \rho^{\gamma}$, where $\rho$ and $p$ are the density and gas pressure. Standard symbols are used for other common quantities. Here, we have chosen rationalized units for the magnetic field so that the magnetic pressure $p_b = {1\over 2}B^2$ and the Alfvén speed is simply $c_a =
B/\sqrt{\rho}$.
These equations were solved numerically using a multidimensional MHD code based on the explicit, finite difference “Total Variation Diminishing” or “TVD” scheme. That method is an MHD extension of the second-order finite-difference, upwinded, conservative gasdynamics scheme of Harten (1983), as described by [@ryuj95]. The multidimensional version of the code, along with a description of various one and two-dimensional flow tests is contained in [@ryujf95]. The code contains an fft-based routine that maintains the ${\bmsy\nabla}\cdot{\bmit B}=0$ condition at each time step within machine accuracy.
Numerical solution of equations \[masscon\] - \[induct\] on a discrete grid leads, of course, to diffusion of energy and momentum as well as to entropy generation. Of course, such effects are also present in nature and are important to defining the character of the flows. The existence of effective numerical resistivity is necessary, for example, to allow magnetic reconnection to occur. There is fairly good evidence that conservative monotonic schemes, as this one is, do a good job of approximately representing physical viscous and resistive dissipative processes that are expected to take place on scales smaller than the grid ([*e.g.,*]{} [@porwood94]). For the astrophysical environments being simulated the expected dissipative scales are likely very much smaller than those that can be modeled directly. One anticipates then that increased numerical resolution leads to solutions in which the large-scale flow patterns “converge” in a statistical sense over time scales of interest, or in more detail over limited time intervals.
Our calculations are carried out in the ($x - y$) plane with the initial background flow aligned with the $x$ direction. For those simulations presented here we have used a computational space that is square and extends over $x=[0,L_x]$ and $y=[0,L_y]$, with $L_x = L_y =
L$. In some related tests we applied different proportions for $L_x$ and $L_y$. The calculations are formally of “$2~+~{1\over 2}$” dimensions, since we can include the $u_z$ and $B_z$ components. In practice, however, $u_z = 0$ and $B_z = 0$ for the simulations we present here. We will consider the more general case elsewhere. Thus, the simulations are practically two-dimensional. Following MP, the computational domain is periodic in $x$, while the $y$ boundaries are reflecting; i.e., both normal velocity and magnetic field change sign across the top and bottom boundaries. In addition it is straightforward to show using equation (2.4) and Stokes theorem that under these conditions we expect ${{\partial \langle {\bmit B} \rangle } \over
{\partial t}} = 0$, where the brackets represent a spatial average over the computational domain . We have confirmed that this is exactly true in our simulations. The field is locally changing, of course, and the mean magnetic energy and pressure are variable, as well.
We consider an initial background flow of uniform density, $\rho = 1$ and gas pressure, $p = 0.6$, and an adiabatic index, $\gamma = {5\over
3}$, so that the sound speed, $c_s = \sqrt{{{\gamma p}\over {\rho}}}=
1.0$. The magnetic field, ${\bmit B_o} = B_o \hat {\bmit x}$, is also uniform, but its value is chosen to be either $B_o = 0.4$ or $B_o =
0.2$ in the two cases we have considered, as described below. The velocity in the background state is antisymmetric about $y = L/2$ according to the relation $${\bmit u_o} =u_o(y)\hat {\bmit x} = ~-~{U_o
\over 2} \tanh({y - {L\over 2} \over {a}})\hat {\bmit x},
\label{vprof}$$ with $U_o = 1$. This describes a smoothly varying flow within a shear layer of full width $2a$. Flow is to the left in the top half-plane and to the right below that. To this state we add a perturbation, $\delta (\rho, p, \bmit u, \bmit B)$. As in MP we define that perturbation to be a normal mode found from the linearized MHD equations, periodic in $x$ and evanescent in $y$, with period equal to the length of the computational box, $L$. Thus, each perturbed quantity can be expressed in a form, $$\delta f(x,y,t) = f(y)\exp(ik_x x + i\omega t),
\label{delf}$$ where each $f(y)$ is a complex function determined by numerical integration of the linearized MHD equations as outlined in MP. Physically, one requires the real part of each $\delta f(x,y,t)$, of course. Further, we have $$\label{kxeq}
k_x = {2 \pi \over L}$$ and $$\omega = \omega_r - i \Gamma.
\label{linrat}$$ From symmetry the real frequency, $\omega_r$, is zero in the computational frame, so that disturbances remain stationary as they evolve. The growth rate, $\Gamma$, can be computed by iteration on the solution for the $f's$, although we found it adequate to obtain $\Gamma$ from the published figures in MP. Other parameters that characterize the system are the Alfvénic Mach number, $M_a =
U_o/c_a$, and the sonic Mach number, $M_s = U_o/c_s$. It is common and convenient also to use the parameter, $\beta = {{p_g}\over {p_b}}
= {6\over 5} ({M_a\over M_s})^2$, to measure the relative importance of thermal and magnetic pressures. For the initial conditions given above we have $M_s = 1$ and either $M_a = 2.5$ ($\beta = 7.5$) or $M_a
= 5$ ($\beta = 30$). These properties are summarized in Table 1. To minimize the influence of the reflecting boundaries we followed Miura’s (1984) prescription to keep $L \ge 20 a$. For the simulations presented here, $L = 25 a$. With this scale boundary zone errors in the initial normal mode perturbation are insignificant. It is most convenient in using MP’s formulation to determine the various perturbed variables from a solution to $\delta p^*$, where $p^* = p + p_b$. Beginning with an initial value of $\delta p^*$ at one of the $y$ boundaries, one can integrate to the midplane and then extend the solution to the other boundary using appropriate even or odd symmetries for the flow variables around a pair of symmetry nodes along the line of flow symmetry. In the two cases presented, we used $\delta p^* (y = L) =
0.001 p^*_o$, ($M_a = 2.5$), and $\delta p^* (y = L) = 0.01 p^*_o$, ($M_a = 5$). The node points were $y = L/2$, $x = L/4, 3L/4$. From the initial conditions (background state + perturbations) we allowed the unstable flow to evolve until it appeared to approach an apparently relaxed state. In practice this meant running the simulations for up to 20 linear growth times; i.e., $\tau = 20~t/t_g$, where $t_g =
\Gamma^{-1}$.
To understand better how well our solutions are numerically converged in the sense described near the beginning of this section, we computed each case on the above domains using several different resolutions. Each case was computed on both $256\times 256$ and $512\times 512$ grids. In addition the $M_a = 5$ case was carried out on a $128\times 128$ grid. The highest resolution simulations required between 20 and 40 CPU hours on a Cray C90 computer. Setting $\lambda = {{2\pi}\over k_x} = L$ maximizes our ability to resolve structures small compared to the initial perturbed wavelength. On the other hand, one might be concerned that important structures should form on scales longer than the initial perturbation wavelength. That is not possible when $\lambda = L$, of course. To evaluate this concern we repeated the $256\times 256$, $M_a = 5$ simulation on a $x = [0,2L]$, $y = [0,L]$ domain using a $512\times
256$ grid. In other words we extended the computational box to include two perturbation wavelengths, but kept everything else unchanged. The results were indistinguishable from the simulation on the square box, so we conclude that no important information of this type has been lost by letting the perturbation wavelength match the box length. We should emphasize, of course, that the choice of periodic boundaries has itself restricted possible Fourier components, prohibiting the kind of large scale coalescent structures noted by [@wu86], for example. So, as pointed out by Wu, a study designed to predict the detailed structures in an unstable, convected shear layer at an arbitrary stage would need a different kind of symmetry.
We conducted one additional set of tests to evaluate a related issue; namely the influence of the location of the reflecting boundaries. Here we doubled the distance to the reflecting boundary, $y = [0,2L]$. In this case we concluded that the general properties of the flows were unchanged, especially within the central, strongly sheared region. Some structural details were slightly changed, but not sufficiently to alter any of the conclusions we will present below.
In summary, on a space periodic in one direction we computed the evolution of two uniform density, but compressible MHD KH unstable shear layers through their initial exponential growth, saturation and nonlinear relaxation phases. Both cases had in common an initial magnetic field aligned with the fluid flow and a sonic Mach number for the velocity spread, $M_s = 1$. The two cases were distinguished only by the Alfvénic Mach numbers; those being $M_a = 2.5$ and $M_a = 5$. Although these Alfvénic Mach numbers differ by just a factor 2, the flows are qualitatively very different.
Results
========
Overview: Weak Field and Strong Field Conditions
------------------------------------------------
Before discussing the two MHD shear layers we have studied it is useful to outline very briefly the well-known behavior in a two-dimensional gasdynamic, unstable shear layer (e.g., [@mas81]). For comparison we carried out a simulation equivalent to the MHD cases described here, except that the magnetic field was absent. In that case the perturbation, once it becomes nonlinear, quickly evolves into a single, large vortex that extends across the full length, $L$, of the grid. The vortex is flattened in $y$, so that it spans about $1/3$ of the grid in that direction. Since numerical viscosity is small, the fully formed, elongated vortex, sometimes called “Kelvin’s Cat’s Eye”, is stable and would spin almost indefinitely. It becomes the only identifiable structure in the flow. Considering the periodic nature of the computational space, there would in the gasdynamic situation be a periodic line of vortices separating the two oppositely moving and mostly undisturbed fluids. For nonperiodic flow, vortex coalescence would lead to merging of these structures (e.g., [@corsher84]). Eventually, over a timescale several orders of magnitude longer than we have considered, numerical viscosity would dissipate the vortex flow and the initial kinetic energy would be converted into thermal energy. That history, as we shall see, is rather different from what happens in either of the MHD flows we studied.
A sense of the evolutionary histories of MHD cases can be obtained by examining the transverse velocity, $u_y$. Initially, that is just the perturbation $\delta u_y$. Figs. 1a and 1b show the spatial RMS values, $\sqrt{\langle u_y^2\rangle}$, for both cases as functions of time, given in units of the predicted growth times; i.e, $\tau
= t\Gamma$. The different curves represent the various numerical resolutions used. Velocities are normalized by the initial value found for the perturbation in each run. Several key features are immediately evident. First, for both cases the amplitude, $\sqrt{\langle
u_y^2\rangle}$, initially increases exponentially as expected, with a growth rate close to the theoretical value (shown by the short dashed line in each panel). For the $M_a = 2.5$ case there is a startup error that disturbs the growth momentarily, but it then resumes the predicted rate very closely. The error, which we attribute to the way our code’s Riemann solver handled the initial conditions at one point along the midplane, is greater at the lower resolution, and it disturbed an even lower resolution run ($128\times 128$ grid) sufficiently that we elected to ignore those results. The second point to be made from Fig. 1 is that after only a few linear growth times, $\sqrt{\langle
u_y^2\rangle}$ saturates and then relaxes towards values that are below the initial perturbed levels. That “final”, “relaxed” state has some interesting properties, so we shall return shortly to discuss it in some detail. The relaxed states that develop are analogous to the long-lived stable vortex described for gasdynamics. Since that state cannot be the ultimate condition for these flows in the presence of finite dissipation, we will refer to it as a “quasi-steady” relaxed state. We also note that the time for an initial perturbation to reach saturation depends on the amplitude of the perturbation. However, the time elapsed from saturation, measured in units of $\tau$, ought to be characteristic of the properties of the unperturbed configuration, including the symmetry. This last statement comes from the observation that for a given background flow the characteristic global properties when the perturbation saturates are not very sensitive to the details of the perturbation itself. This is demonstrated for the present case by the good agreement between the saturation-phase flow properties seen in our simulations and those of [@maletal95]. The latter authors used a simple periodic velocity perturbation rather than a linear normal mode.
The third point from this figure is that the flow evolution is considerably more complex in the $M_a = 5$ case than in the $M_a =
2.5$ case. The difference in flows is even more obvious in the images of gas density and magnetic field lines displayed from our highest resolution runs in Figs. 2 and 3. To facilitate visualizing the extension of structures on the periodic space we doubled each image. For the $M_a = 2.5$ case flows remain relatively laminar throughout, and qualitatively resemble the initial, simply periodic form well past the stage when a linear description becomes invalid. The magnetic field lines as shown in Fig 2b suffer relatively modest bending or stretching. There is no indication of magnetic reconnection, nor that any suitable field topology for that process has formed. Once the oscillations reach their maximum amplitude, the flow seems simply to flatten out into a broadened shear layer. In marked contrast, the $M_a = 5$ case develops flows that become quite complex, with distinct, intermittent vortices. The large amplitude oscillations in $\sqrt{\langle u_y^2\rangle}$ after $\tau \approx 2$ are associated with the generation and decay of these vortex structures. The large “primary vortex” centered on $x = 3L/4$ at the $\tau = 2.4$ in Figs. 3 reaches its maximum development after $\tau =
2$ and corresponds to the first maximum in $\sqrt{\langle
u_y^2\rangle}$. Around $\tau = 6$ the magnetic field lines on the outside of that vortex, having been wrapped back on themselves, tear in a major reconnection event that matches with the second maximum in $\sqrt{\langle u_y^2\rangle}$. Fig. 3b shows the field configuration at $\tau = 5.6$, just before that event. The third maximum in $\sqrt{\langle u_y^2\rangle}$ corresponds to a similar reconnection event associated with the well-formed “secondary vortex” centered at $x = L/4$ and $\tau = 8$ in Fig. 3. Smaller peaks in $\sqrt{\langle
u_y^2\rangle}$ can also be seen at $\tau \approx 12, 15~{\rm and}~
18$. Each is associated with reconnection in a vortex structure within the flow.
There are clear differences between the descriptions just given for the MHD flows and that presented earlier for the analogous gasdynamic situation. Even when the magnetic field is initially weak its presence becomes crucial in determining the nonlinear evolution of the flow. The initially weak field is able to mediate dynamical and dissipative processes through local growth and decay. Not surprisingly, the role played early on by the magnetic field depends considerably on its initial strength. That is apparent in the initial, exponential growth rates. The growth rate for the $M_a = 5$ case ($\Gamma = 0.108 \times {{U_o}\over {2a}}$) is about 90% of the purely hydrodynamic rate ($\Gamma = 0.122\times {{U_o}\over {2a}}$ [@miupr82]), whereas the rate for the $M_a = 2.5$ case ($\Gamma =
0.053\times {{U_o}\over {2a}}$) is less than half the (same) hydrodynamic rate. From these various contrasts it is convenient and, we believe appropriate, to distinguish in our remaining discussion the two cases as “weak field” ($M_a = 5$) and “strong field” ($M_a =
2.5$), thus being generally representative of qualitatively different flows. [@maletal95] also saw a qualitative transition in properties between these two field strengths. We note for comparison that the incompressible hydrodynamical linear perturbation growth rate of a discontinuous shear layer is $\Gamma = {\pi}\times {{U_o}\over
{\lambda}}$. Here, $\lambda = 25 a$ ($\Gamma \rightarrow
{\pi\over{12.5}}\times {{U_o}\over {2a}}$), so a combination of compressibility and the finite spread in the shear layer has reduced the initial growth rate by more than a factor of two even before Maxwell stresses are introduced. The sonic Mach number is unity, after all, and in each case thermal energy is globally the dominant energy form throughout the simulations by about an order of magnitude.
As noted before, numerical resolution is an important issue in evaluating computations of this kind. We will comment on specific resolution-related questions as they come up, but we can make some preliminary observations here. It is clear from an examination of Fig. 1 and by comparing images of various physical quantities, that many qualitative behaviors are similarly captured in all of the calculations. In Fig. 1b, for example, the same major oscillations in $\sqrt{\langle u_y^2\rangle}$ occur for all three weak field simulations. By viewing animations of the three flow simulations one can see that the oscillations represent the same flow events as well. Likewise, there are no essential differences in the global characters of the two strong field simulations. At the same time it is obvious that the lower resolution runs are not converged quantitatively in important physical variables such as peak magnetic energy, or kinetic energy during the saturation phase. [@ryujf95] demonstrated with this code that under conditions similar to the strong field case here, isolated structures need to be resolved within about 30 zones before one can realistically neglect numerical diffusion; i.e., before the effective “Reynolds number” of the structures $> 10^3$. For conditions similar to the weak field case that requirement is somewhat less restrictive. There we can estimate that about 20 zones are needed. Thus, while our lowest resolution runs would contain as few as 6 “fully resolved” structures in each direction, that number climbs to around 25 or more in the highest resolution simulations.
We will now proceed to discuss the characteristics of three phases in the nonlinear evolution of the simulated flows. From the preceding outline we identify these as: a) saturation, b) relaxation and c) the quasi-steady, relaxed state. The initial, exponential growth phase is simply an extension to larger amplitude of the initial perturbation as discussed in MP, so we will not discuss it here. The quasi-steady relaxed state, is relatively simpler than any of the other nonlinear stages and comes closest to having conditions that are straightforward to predict. Consequently, we begin our analysis there.
The Quasi-steady, Relaxed State
-------------------------------
As mentioned in the previous summary, the simulated flows in both strong and weak field cases relax fairly quickly to an apparently steady condition. The fact that we have used periodic boundaries in our simulations and have conserved total energy in the domain of computation means that eventually the flow kinetic energy must be dissipated. But, that ultimately anticipated condition is not what is observed here, as emphasized by the comparison made with the gasdynamic case. The final density and magnetic field structures in our MHD simulations as shown at $\tau = 17.9$ for the strong field flow in Fig. 2 and at $\tau = 20$ for the weak field flow in Fig. 3 give a sense of the character of the relaxed states reached. The flows are nearly laminar, reflecting the fact that the mean transverse velocity has decayed to very small levels by this time (see Fig. 1). There are remaining visible striations in both the density and magnetic fields, however (see also Fig. 4). Within the magnetic field variations the magnetic and velocity fields are almost perfectly aligned, so that the normalized cross helicity, $$\langle H \rangle =
{{\int \hat {\bmit u} \cdot\hat {\bmit B}dxdy}\over{\int dxdy}},
\label{crosshel}$$ very closely approaches the possible extrema, $\mp 1$, in the top and bottom half planes respectively. Thus, the flows are highly organized by this time. Initially, the cross helicity is also large, but for the weak field flow it becomes much reduced during the nonlinear phases of flow evolution. In that case the maximized cross helicity at the end may be related to so-called “dynamical alignment”, first noted with respect to similar characteristics in the solar wind by [@dobro80], and subsequently examined theoretically and numerically in more general contexts of MHD turbulence (e.g., [@pouq86; @tingetal]). Outside of the small influences of the striations, the transverse velocity gradients, $\partial u_x/\partial y$, are almost constant within the broadened shear layer. We have examined animations of a number of variables to ascertain any remaining trends in the flow evolution. Although it is apparent that some of the feature details may be expected to relax further, it seems clear to us that the general properties of the flow patterns visible at these times will continue to exist as seen well beyond the end points of our simulations and that no new features are developing.
The properties of the magnetic field in the relaxed states are perhaps the easiest to understand. They come largely from the assumed symmetry in the problem and the laminar nature of the flow in the relaxed states. $\langle B_x \rangle$ and $\langle B_y \rangle$ are both exactly conserved during the calculation, so in the relaxed states we may expect fairly uniform fields. The magnetic energy in the flow may be reasonably expected to return to something resembling its original value. The manner of those trends can be made evident by using equation (\[induct\]) to derive a relation for the evolution in the mean magnetic energy density, $\langle E_b \rangle = {1\over 2}
\langle B^2 \rangle$. As mentioned in the introduction we depend on numerical dissipation to mimic the effects of viscosity and resistivity, although we cannot describe an analytic model for the two effects or accurately separate them. However, to understand heuristically their effects qualitatively we can add a term $\eta \nabla^2 \bmit B$ to equation (\[induct\]), where $\eta$ represents the effects of numerical resistivity. For a two-dimensional flow with our boundary conditions this leads to the equation $${{d \langle E_b\rangle}\over{dt}}={\partial \langle E_b\rangle \over
\partial t} + \langle ({\bmit u \cdot \nabla})E_b\rangle = -\eta
\langle{\bmit j}^2\rangle - \langle 2E_b{\bmit \nabla \cdot u}\rangle
+
\langle {\bmit B} \cdot [{\bmit B \cdot \nabla}{\bmit u}]\rangle,
\label{ebent}$$ where ${\bmit j} = {\bmit \nabla \times B}$ is the current density, and we assumed $\eta$ to be constant. The average is over the entire domain. The three right hand terms represent respectively, Joule dissipation or magnetic diffusion, compression of the field and field line stretching. Unless the relaxed field configuration is uniform there will ordinarily be a finite $\langle {\bmit j}^2 \rangle$, and hence dissipation. That term always will decrease the magnetic energy and can only be countered by enhancement through compression and stretching. But, in the laminar relaxed state those effects are fairly small, because of the absence of strong gradients in the velocity structure. Thus, we must expect that the final field will be relatively uniform and that the associated magnetic energy will resemble the initial value in order to satisfy the constant mean field values.
Fig. 5 shows the fractional change between the initial and final magnetic energies, $\Delta E_b/E_{bo}$, for both the strong and weak field cases as a function of the number of grid points, $N_x$, used in the various simulations. We see that for the strong field case the final and initial magnetic energies are, indeed, almost precisely the same for both numerical resolutions used. They agree to much better than 1%. For the weak field case, the two lower resolution simulations also relax to a state with magnetic energy equaling the initial value to better than 1%. Remarkably, however, the highest resolution simulation in this case ends with a magnetic energy greater than the initial value by about 7.7%. Fig. 6, which shows the time evolution of $\langle B^2_x\rangle$ and $\langle B^2_y\rangle$ for that simulation, also illustrates this fact. Although $\langle
B^2_y\rangle$ returns very close to its initial value, $\langle B^2_x\rangle$ does not evidently do so. We cannot rule out the possibility that $\langle B^2_x\rangle$ will slowly decay to its initial value, but we prefer another interpretation of this result; namely, that it represents the initial development of “two-dimensional flux tubes”. Fig. 4 shows that the final state in this flow contains a pair of broad striations in enhanced magnetic pressure positioned on opposite sides of the midplane, $y = L/2$. It is also apparent from Fig. 4 that the magnetic and gas pressure distributions are strongly anti-correlated at this time. Moreover, the maxima in magnetic pressure are also local minima in gas density. These features can, therefore, be described as slow mode rarefactions, similar to the “plasma depletion” zone described for an earlier stage of the flow by Miura. In addition, we already mentioned that the velocity and magnetic field vectors are almost exactly parallel in the lower half plane and anti-parallel in the upper half plane. Thus, the magnetic field lines and velocity streamlines are aligned and the flow is directed along the tubes. By watching animations of the evolution of the magnetic field structure one can see that the two flux tubes (stronger field regions) have survived as coherent features since the reconnection event associated with the breakup of the large vortex. The tubes can be identified relatively easily as clusters of field lines stretching obliquely above and below the vortex in the $\tau = 8$ panel of Fig. 3b. From that time they gradually settle down towards the midplane as activity in that region decays. We see no evidence of these tubes as coherent structures in our lower resolution simulations of this flow. Numerical diffusion presumably prevented them from forming, a fact that emphasizes the need for adequate numerical resolution in simulations of this type. There is also no indication of such structures in the strong field case, because there were no opportunities for dynamical self-reorganization through magnetic reconnection during evolution of that flow.
These final structures are also apparent in the Fourier power spectrum of the magnetic pressure, $p_b$. We constructed two-dimensional power spectra, $P^b_{x,y}$, at a number of times for both the magnetic field cases. The power spectra are not isotropic, as obvious from the flow structures seen in Figs. 2 and 3. Initially, $P^b_x$, the power spectrum along different $k_x$, has a single mode at $k_x = 1$ of course. $P^b_y$, the power spectrum along different $k_y$, has a fairly broad distribution at the beginning, reflecting the finite thickness of the shear layer. At the end of the simulations, $P^b_x$ has returned to have almost a monochromatic mode for the both cases. For the strong field case $P^b_y$ is also similar to its initial form, but with two notable changes. The dominant mode is now at $k_y = 3$ rather than at $k_y = 1$. In addition, it is spread over more modes to reflect an increased thickness of the shear layer, and there is some structure in the power spectrum coming from the features that can be seen in Fig. 2. In the weak field case $P^b_y$ has a narrower peak than that at the beginning with the dominant power at $k_y = 4$, but also with distinct peaks at $k_y = 8$ and 12. Some power extends to the highest modes in this case, as expected from thin stripes seen in Fig. 3.
As already mentioned, except for small deviations aligned with remaining magnetic features, the final velocity structure can be described as a simple shear layer (see Fig. 3c). During the evolution of the flow the initial shear layer spreads, of course, because momentum is transported between the bottom and top halves of the computational domain. Horizontal momentum, $\rho u_x$, is transported in the $y$ direction by way of Reynolds stresses, $\rho u_xu_y$, Maxwell stresses, $B_xB_y$, and also by numerical diffusion. In the same spirit as we had for equation (\[ebent\]), we can represent that last behavior qualitatively in terms of an effective viscous stress, $\sigma_{xy}$, (e.g., [@lanlif87]). Then, for example, using the symmetry and boundary properties of this problem the change in the mean $\rho u_x$ within the [*upper half*]{} of the domain is $${{\partial \langle \rho u_x\rangle}\over {\partial t}} = {2\over L^2}
\left\lbrack\int^{L}_{o} (\rho u_x u_y - B_x B_y + \sigma_{xy})dx
\Bigg|_{y = {L\over 2}}\right\rbrack
\label{momtran}$$ (see also [@miura84]). The shear layer should spread so long as $\partial \langle \rho u_x \rangle /\partial t > 0$, since initially $\langle \rho u_x\rangle < 0$ in this region. Once the flow becomes nearly laminar the Reynolds and Maxwell stress contributions to momentum flux across the midplane should become small, leaving only the viscous stresses as a source of transport. Although viscous stresses can never be entirely eliminated, they can be reduced to nominal levels as the numerical resolution is increased. For our highest resolution runs of both cases the momentum flux as measured by the right hand side of equation (\[momtran\]) is very small for $\tau > 10$. It is almost exactly zero in the strong field case, but oscillates around zero with decreasing amplitude in the weak field case.
As the disturbed flows become nearly laminar, shear layer spreading slows for reasons just given. We will address in §3.3 how that return to near laminar flow happens. But, one might reasonably ask why remaining fluctuations in $u_y$ do not lead to renewed growth and further spreading as that final state is approached. Qualitatively, at least, the answer seems straightforward. In particular, if we note that the $u_x$ structure of the final state is similar to that in the initial flow except for the width in $y$, then we know from the linear dispersion relation (e.g., MP) that there is a minimum unstable wavelength related to the thickness of the shear layer; i.e., for unstable perturbations, $k < {\phi\over {2a}}$, where $\phi \sim
1~-~2$, depending on $M_a~{\rm and}~M_s$. The fastest growing modes are concentrated at wave numbers about half this limit. On the other hand, our periodic boundary conditions limit Fourier components in the flow to $k\ge {2\pi\over L}$. Thus, we expect that once the shear layer has spread so that the nominal width, $a\sim |U_o/(\partial
u_x/\partial y)| > {{\phi L}\over{4\pi}}$, a quasi-laminar flow will be mostly stable to any further perturbation, independent of the magnetic field strength. For both cases we studied the final shear layer width as estimated from the slope of the central region is at least $\sim 0.5 L$. So, from the conditions just derived, we would not expect renewed growth. For the weak field case this estimated final value of $a$ from the observed flow is $\sim L$, which could be limited or modified by boundary effects. As further confirmation of our conclusion, then, we carried out a $256\times 256$ simulation using $L_x = L_y = 2L = 2\lambda$. Except for details in the flows near the top and bottom boundaries the properties of that flow were the same as for the $128\times 128$ simulation using $L_x = L_y = L =
\lambda$.
We previously noted that in the relaxed state the magnetic field lines and flow streamlines are highly correlated. That is evident in Fig. 3b and 3c. In fact it is apparent that within the “flux tubes” velocity and magnetic field fluctuations satisfy the relation $\delta
{\bmit u} \sim \mp \delta {\bmit B}$, where the minus sign corresponds to the fluctuations in the upper half plane and the plus sign corresponds to the lower half plane. That relation is consistent with the remaining fluctuations being carried in Alfvén waves that propagate locally downstream within the flux tubes and are linearly polarized so that $\delta u_z~=~\delta B_z~=~0$. This feature and the convergence and divergence of flow along the varying width of the tubes explains the good correlation of magnetic pressure and $u^2$ observed when the velocity is corrected for the mean shear structure, $\langle u_x(y)\rangle$.
In contrast to the magnetic energy, we should anticipate that the final total kinetic energy inside the computational domain, $E_k$, will be reduced in response to the relaxation of the flows. Since in the relaxed flow the velocity transition is spread out more between the top and bottom states, and the computation was carried out in the center-of-momentum frame, we can expect that the quantity $\langle
u^2_x\rangle$ is reduced over the domain by the spreading of the transition. In both the strong and weak field cases the kinetic energy in the relaxed state exceeds the magnetic energy. For the strong field case $E_k/E_b \approx 1.09$ at the end of the simulation. It is conceivable that an eventual fairly long term equipartition between magnetic and kinetic energy develops. However, kinetic energy in the weak field flow always significantly exceeds magnetic energy in our simulation. In the relaxed state $E_k/E_b \approx 2$. On a much longer time frame, it is not inconceivable that an equipartition between kinetic and magnetic energies develops in response to the dissipative decay of the flow mentioned in the next paragraph. On the other hand, it is also possible that the observed non-equipartition condition [*does*]{} represent the relaxed condition of the flow. [@tingetal] have pointed out that turbulent MHD flows dominated by dynamical alignment can trend to long-term states in which the kinetic energy remains dominant over the magnetic energy.
Since the magnetic energy is little changed at the end from the beginning and kinetic energy is reduced, it follows that the final states must possess increased thermal energy over the initial flows. Figs. 7 and 8, which display the evolution of the total kinetic, magnetic and thermal energy forms over time, verify this. For the strong field case the kinetic energy decreases by almost 24%, while that number is close to 60% for the weak field case. This leads to increases of thermal energy, $E_t = p_g/(\gamma - 1)$, of about 3.3% (7.8%) in the strong (weak) field case by the end of our simulation. Taking advantage of the symmetries and boundary properties in the simulations, it is straightforward to show that the mean thermal energy obeys the relation $${{\partial \langle E_{t}\rangle}\over{\partial t}} = - \langle \ p_g
{\bmit \nabla \cdot u}\rangle +\eta \langle {\bmit j}^2 \rangle +
\langle \sigma_{ik} {{\partial u_i}\over{\partial x_k}}\rangle,
\label{etherm}$$ where we have once again for heuristic purposes included effective resistive and viscous dissipation terms in addition to the consequences of adiabatic compression. Below we use equation (\[etherm\]) to estimate the degree to which our simulations are influenced by finite numerical resolution.
In these calculations the total mass is conserved inside the grid. Furthermore, the use of periodic boundaries means that any adiabatic work applied to mass leaving the domain is exactly replaced by mass with identical properties entering from the opposite boundary; i.e., thermodynamically, the system is closed. Thus, since the average compression must be zero over the fixed volume, the net adiabatic work done on the gas is expected to be small, and the first term in equation (\[etherm\]) is small. Consequently, ${{\partial\langle E_t\rangle}\over{\partial t}}$ must be due almost entirely to the two dissipative effects. We have verified numerically that $\langle \ p_g {\bmit \nabla \cdot u}\rangle$ is indeed small. There are no shocks formed in either of these cases, so only dissipation in smooth flows matters. Even though the dissipation in our computation is numerical, that does not necessarily mean the simulated flows are physically incorrect, of course, at least in a statistical sense. Pseudo-ideal fluid simulations of shocks and turbulence, for example, depend on the existence of a small amount of numerical dissipation to obtain physically meaningful behaviors. But, it is necessary that the numerical dissipation be small enough that the computed flows approach the intended physical flows (meaning those with very small viscosity and resistivity). To evaluate this situation we should ask how well the total dissipation in the simulated flows has converged. Examination of the time evolution of the thermal energies as shown in Fig. 7 and 8 provides that information. The final increases in thermal energies agree among the different simulations of the same flows to within 10% of the same value in the strong field case, and about 6% for the weak field. Therefore, we consider the final states moderately well converged in that respect. We will explore in the next subsection some details about how that dissipation takes place prior to the development of the relaxed states.
We emphasize again that, although the final conditions of our simulations are not changing very fast, they cannot represent the expected conditions if one extended the computations for an arbitrarily long time. In fact, dissipation cannot entirely vanish so long as there is any shear left within the system. In the limit that the flow is laminar the viscous term in equation (\[etherm\]) can be represented qualitatively in terms of an effective shear viscosity, $\mu$, and the mean-squared vorticity or enstrophy in the flow, $\langle W \rangle = \langle ({\bmit \nabla \times u})^2\rangle$, as $\langle \sigma_{ik} {{\partial u_i}\over{\partial x_k}}\rangle \sim
\mu \langle W \rangle$. From this and equation (\[etherm\]) the time for the flow kinetic energy to decay to thermal energy is $> R
a/U_o$, where $R$ is the effective viscous Reynolds number in the flow and $a$ is the nominal width of the shear layer. Using the properties of our code as determined by [@ryujf95] we can estimate that $R \gsim 10^5$ in our highest resolution runs. So, this decay time is several orders of magnitude longer than our simulation.
We noted earlier that the final magnetic field configuration in the weak field case showed signs of an organized structure; namely flux tubes containing magnetic field lines aligned with the flow streamlines. The final state of the weak field flow contains a second structure that warrants comment. Examination of Figs. 3a and 4 shows, in addition, the presence of a low density “channel” containing hot, high entropy gas concentrated along the midplane. That feature does not correspond to the flux tubes, but rather lies between the flux tubes. The origin of this second feature becomes obvious if one views animations of the evolution of several quantities, especially the density, magnetic field, electric current and entropy, $S =
\ln(P/\rho^{5\over 3})$. Most of the entropy generated in dissipative events that we will describe in the next subsection has been concentrated into this thin layer, which appears to be contained by the two magnetic flux tubes, since they lie just outside it. Gas within the central layer is about 25% hotter than the average over the domain. The gas pressure is moderately high within the layer, but there is a near equilibrium balance in the total pressure, $p~+~p_b$. Recalling that the flux tubes were traced to an early magnetic reconnection event separating the magnetic field in the outer and inner parts of the flow, we can recognize that a consequence of the evolution in the weak field KH unstable flow has been a reorganization of the flow into two distinct regions; one in which the energy dissipation necessary for relaxation takes place and one that surrounds that region. There is no evidence of such a development in the strong field flow, again presumably because the stronger initial field prevents the opportunity for the reorganization.
Saturation and Relaxation
-------------------------
As noted before, the history of a simulated two-dimensional ideal gasdynamic shear layer in a periodic flow is very simple; i.e., a single, long lived vortex forms. That history seems quite distinct from what happens in the MHD flows, where the formation of vortices is either intermittent or entirely absent. In the weak field case we computed the primary vortex begins development, but is disrupted before it is fully formed. Its destruction leads to the generation of a sequence of smaller vortices (each located at one of the symmetry or node points in the computational plane), but they, too, are quickly destroyed. In the strong field case the initial oscillation is stabilized before a vortex forms, and the flow then directly begins to approach a laminar shear layer. It might seem odd that our two MHD flows are so different, because their properties as given in Table 1 are the same except for a factor two difference in the strength of the magnetic field. In both MHD flows the Alfvén Mach number is greater than unity and the plasma $\beta$ parameter is substantially greater than one. However, it should be recalled that a parallel flow discontinuity is KH stable whenever the Alfvén Mach number, $M_a <
2$, only slightly smaller than the value for our strong field case. This criterion represents the condition that the restoring force from magnetic tension due to field line stretching along a perturbed boundary equals the “lift” force on the boundary. Thus, for our strong field flow the initial Maxwell restoring force is only slightly less than needed to stabilize the oscillation. By comparing the magnitude of the Maxwell stress, $|({\bmit j \times B})_y|$, and the gas pressure gradient, $|\partial p/\partial y|$, in the strong field flow we confirmed directly that the initial perturbation grows only until the Maxwell stresses [*along the displaced boundary*]{} exceed the gas pressure gradient that drives the instability. From that moment, which is simultaneous with the peak in $\langle\sqrt{u^2_y}\rangle$ in Fig. 1a, the two stresses seem quickly to achieve a close balance almost everywhere, while both transverse motions and excess magnetic energy steadily decay (see Figs. 1, 2, 3 and 7). Magnetic enhancements during the growth of the instability are concentrated into a close pair of magnetic flux tubes formed along the displaced flow boundary that have reduced plasma density and pressure. Although the total Maxwell stress does locally exceed the gas pressure gradient at saturation, the $\beta$ parameter is never less than about 2.5 (see Fig. 7), so an “equipartition” in the common sense never develops. The total magnetic energy is enhanced by about 8% at its peak in the higher resolution simulation for this case. That compares to about a 9% enhancement reported for this case by Miura, so the agreement is reasonably good.
The early development of the weak field flow is very different, of course, and resembles in most ways the behavior of the pure gasdynamical KH instability. Just as in the gasdynamic case the applied oscillation grows in amplitude and then begins to role up into a large vortex, as shown in Fig. 3. However, it is well known that the presence of even an initially weak background magnetic field can have a profound influence on turbulent or sheared flows (e.g., [@chan60; @biswel89; @catvain91; @balhal91; @noret92]), especially if it has a non-vanishing mean vector value. That property is soon apparent in this simulation. Instead of rolling up and continuing to spin, the vortex formed in our weak field flow becomes distorted in response to increases in magnetic field strength. Both magnetic pressure and tension contribute. That effect is apparent at $\tau = 5.6$ in Fig. 3. The maximum total magnetic energy during the simulation occurs near this time. That energy, which exceeds the original value by $\sim 140$% in the highest resolution simulation of this case (see Fig. 8), is concentrated into a thin flux tube feature. Our peak magnetic energy enhancement in this simulation agrees well with that found by [@maletal95] in their analogous calculation. The flux tube feature corresponds to the low density plasma channel that approximately traces the displaced initial shear layer being wrapped into the vortex at this time. Miura pointed to the almost straight segment of that feature, centered on $x = L/4$ and called by him the “plasma depletion zone”, as the place where most of the enhanced magnetic energy is concentrated. However, we find a much larger total magnetic energy enhancement (his was only 26%) and that by far the greater portion of the energy is held within the main vortex structure itself. In his computation, which had much lower resolution and, hence, greater diffusion and dissipation, the vortex was more strongly damped this time. We find, as well, a somewhat greater concentration of magnetic field within the plasma depletion zone, leading to a minimum $\beta = 0.55$, compared to the $\beta = 1.7$ that he quotes. We note that the plasma depletion zone seen at this time is [*not*]{} the same structure that leads to the two-dimensional flux tubes described for the final state. Those form separately and can be seen simultaneously with this feature at $\tau = 8$ in Fig. 3b.
About the time, $\tau \approx 4.5$, that the magnetic energy is reaching its maximum, the total kinetic energy is at a local minimum (Fig. 8), a development also apparent in the first local minimum in $\langle\sqrt{u^2_y}\rangle$ seen in Fig. 1b. Thus, it is clear that magnetic stresses have impacted on the gasdynamics in a substantial way. Immediately after these developments, Fig. 8 shows that the magnetic energy drops, while the kinetic energy rises. These further developments come from the fact that as the magnetic field is being drawn into the vortex, field lines from just outside the initial shear layer become folded back on themselves around the “tips” of the low density channel (see Fig. 3b). The magnetic energy maximum corresponds to the moment just before the field becomes unstable to a tearing mode instability near the “tips”. That leads to magnetic field reconnection and field annihilation. Blobs of heated, low density plasma are ejected from the “tips” into the outer regions of the flow and the magnetic field reorganizes itself. The ejected blobs contain magnetic flux islands; i.e., closed flux loops. Within the vortex the field lines snap back towards the vortex center (accelerating plasma with them as they do so), while the exterior field structure becomes isolated from further involvement with vortex behaviors within the central shear layer. The vortex collapses subsequently into the central shear layer.
As the original, primary vortex is destroyed in the manner just described, a strong, secondary vortex spins up, centered at $x = L/4$. That vortex also wraps and amplifies magnetic field within it, until at $\tau \approx 7.5$, field lines have been pulled completely around the outside of the vortex causing another reconnection event (see also Fig. 3). Once again, this event corresponds to a maximum in the total magnetic field energy. There is also a peak in the kinetic energy near this time, but this actually corresponds to the “ejection” of hot plasma from the primary vortex, at $\tau \approx
6$, when the magnetic energy is minimum. Magnetic energy and vorticity are concentrated around the perimeter of the vortex. That is also where reconnection begins. The reconnection event, as well as the one mentioned earlier, shows up clearly in images of such quantities as electric current, entropy and $\beta$. Where magnetic “$\times$”, or neutral points form, intense local current sheets develop and excess entropy is generated, along with expansions in response to the heating. Local maxima in $\beta$ also develop around the $\times$ points, of course. As with the earlier vortex-directed magnetic reconnection, the initial reconnection event here separates magnetic field within and outside the vortex. In this case the magnetic field is somewhat weaker than before and the vortex is stronger. The vortex continues to spin for a while and the magnetic islands formed end up inside the vortex. It is a magnetically isolated structure, with detached fields lines surrounding it. The vortex and the magnetic energy within it decay by about $\tau \approx 10.5$. Remnant magnetic islands are still visible at $\tau = 10.4$ in Fig. 3b. Weaker vortices briefly form and decay at later times, leaving noticeable signature peaks in $E_b$ around $\tau = 14 {\rm~ and~} \tau = 17$. Those features are anticorrelated with kinetic energy maxima at $\tau
= 15$ and $18$.
We commented earlier on the properties of the power spectrum of the magnetic pressure structure at the end of our simulations. To complete that discussion we briefly add a few comments about the power spectrum evolution during the saturation and relaxation phases for the weak field case. There is little change in the $P^b_x$ spectrum, except that as the secondary vortex strengthens it leads to a shift from dominance by the $k_x = 1$ mode to similar strength in both the $k_x = 1$ and $k_x = 2$ modes. For $P^b_y$ the evolution is more remarkable in that power spikes at a number of scales are evident. The number and strength of those spikes changes as the number and scales of the dominant structures change. For example, as the primary vortex forms there is a single broad peak, around $k_y = 3$, which breaks up into at least four comparable narrow spikes at such wave modes as $k_y = 2,~7,~9$ and 13, once that physical feature is destroyed. Those are replaced during the life of the secondary vortex by a pair of comparable spikes at $k_y = 1$ and 7. The structure at $\tau = 14.4$ (see Fig. 3) leads to four comparable sharp power spectrum spikes at $k_y = 1,~4,~6$ and 9.
Since the secondary vortex is strong and spins several times before decay it presents a good opportunity to study the interactions of the flow and the magnetic field within it. The equation for the evolution of vorticity in two-dimensional ideal MHD can be written as $${d\rho {\bmit \omega} \over dt} = ({\bmit \nabla \times (B \cdot \nabla ) B})
+ {\bmit \nabla}\rho {\bmit \times} ({d {\bmit u} \over dt}),
\label{vordif}$$ with ${\bmit\omega} = {\bmit\nabla\times u}$. The ${d{\bmit u}\over{dt}}$ term in equation (\[vordif\]) represents the total acceleration of a fluid element due to all forces (see equation (\[forceeq\])). For an entropic equation of state ${\bmit \nabla} \rho {\bmit\times \nabla}
p = 0$, so the only nonvanishing contributions (for ideal flows) to the last term would come from the Lorentz force; i.e., ${1\over\rho}{\bmit \nabla}\rho
{\bmit\times}\left({\bmsy\nabla}\times{\bmit B}\right)\times{\bmit B}$. On the other hand, these flows are only slightly compressible ($M_s
\sim 1$), so that $|{1\over\rho}{\bmit\nabla}\rho|<<1$, and this term ought to be small. Then with the aid of Stokes theorem we can relate the change in the total vorticity within an area to the line integral of the magnetic tension around the boundary, $${\int {d (\rho {\bmit \omega}) \over dt}\cdot{\bmit dA}} \approx
\oint ({\bmit B \cdot \nabla) B\cdot dl} = \oint {\bmit T \cdot dl}.
\label{inteq}$$ Fig. 9 displays the two sides of equation (\[inteq\]) computed from the flow properties during the life of the secondary vortex, where the area integral was carried out over the vortex and the line integral surrounded it. The vorticity is positive inside the area, so the signs indicate a reduction in the strength of the vortex. Clearly the two quantities are in good agreement, so that equation (\[inteq\]) properly represents the evolution of the vortex. Magnetic tension around the vortex peaks in magnitude at $\tau = 7.5$, in agreement with the magnetic energy peak noted earlier. Between $\tau = 7$ and $\tau = 9.5$ the value of $\rho w_z$ inside the vortex decreases from its maximum value by about 75%. The integrated effect of magnetic tension according to equation (\[inteq\]) accounts for this change to within about 4%.
Finally, we make a few comments about the evolution of the cross helicity in the weak field flow. The cross helicity, $\langle H
\rangle$, (defined in equation (\[crosshel\])) provides a way to express the self-organization of the flow. We already commented that at the end of our weak field simulation the cross helicity approaches very close to its possible extrema, $\pm 1$, in the bottom and top halves of the computational domain. The two signs just reflect differences in the initial velocity signs in the two regions of the grid. Since we began with a condition $\langle H \rangle = \pm
1$ in the appropriate halves, it may not be very surprising that the flows eventually return to that condition. In fact, if the flows were exactly laminar, this would necessarily be so. On the other hand, the flows still contain fluctuations at the end, and within those fluctuations the velocity and magnetic field structures are very well aligned. At intermediate times the cross helicity is only about half these extreme values, as can be seen in Fig. 10. So, we believe this is an indication of real self-organization. Evidence for dynamical alignment can be found by comparing Fig. 10 with Fig. 6, which follows evolution of the mean squared magnetic field components. The initial large drop in $|\langle H \rangle|$ relates to the growth of the initial, primary vortex. We have already pointed out that the valleys in $\langle B^2_x\rangle$ and $\langle B^2_y\rangle$ correspond to reconnection events associated with various flow vortices. Each large increase in $|\langle H \rangle|$ visible in Fig. 10 at, for example, $\tau = 7$, is generally simultaneous with one of those reconnection events, as well. Thus, the reconnection process serves to realign the flow and field vectors as the flow relaxes.
Summary and Conclusion
======================
We have carried out numerical simulations of the evolution of unstable sheared MHD flows in two-dimensions. Although we chose to conduct these experiments in a rather idealized setting, using a box that was periodic, we believe they have enabled us to discover several important insights about the nature of such flows that can be helpful in understanding more general situations. Our unperturbed initial conditions involved smooth shear layers of constant gas density separating flows with relative motions at sonic Mach number unity. There was a uniform magnetic field aligned with the flow.
We considered two cases. In the first the magnetic field strength was only slightly less than required to stabilize the perturbation. In the second case the magnetic field had a value 2.5 times smaller than that critical value. We followed each flow until it reached an apparently steady condition. For both flows the properties of that steady condition were fundamentally different from what happens in an analogous, two-dimensional gasdynamic unstable sheared flow. In the gasdynamic situation the flow evolves to include a single well-formed vortex that separates the two counter-flowing fluids. That vortex would spin for a very long time, until viscous interactions spread it and dissipated it. By contrast, in the MHD flows, the steady condition that develops is one of a broad, nearly [*laminar*]{} layer separating the two counter-flowing fluids. That difference is a consequence of the magnetic field aligned with the original shear layer. For the stronger magnetic field case this result is an obvious expectation, since magnetic tension stabilizes the instability before any vortices can develop and the flow relaxes in a straightforward manner to one that qualitatively resembles the initial conditions, except that the shear layer is broadened until it is stable against further perturbations. Because of the symmetries in our computations the mean vector magnetic field is a constant throughout each simulation, so the final magnetic field in this case is very little different from that at the beginning. The total magnetic energy, for example is the same as at the start to within 1%. By the end of our simulation there is a rough equipartition between magnetic and kinetic energies, although both are small compared to the thermal energy contained within the gas.
That the quasi-steady, relaxed flow state should also be laminar for the weaker magnetic field is perhaps not so immediately obvious, since the initial magnetic energy is almost an order of magnitude less than the kinetic energy and more than an order of magnitude less than the thermal energy. It never becomes globally more than 5% of the total energy and never reaches global equipartition with the kinetic energy. What happens is that as vortices do develop within the flow, magnetic stresses are built up that have sufficient strength to locally modify the dynamics. At the same time magnetic field wrapped into the vortices becomes unstable to tearing mode reconnection events that isolate some magnetic flux and lead to large scale reorganization of other magnetic structures. The isolated magnetic flux is subsequently annihilated. Together these developments produce a reorganized flow and magnetic field in which the magnetic and flow fields are almost perfectly aligned. The magnetic field is also approximately as it was at the beginning in this case, except for a pair of stronger flux structures, or flux tubes, and fluctuations within them that appear to correspond to Alfvén waves propagating in the same direction as the fluid. However, close examination reveals that the magnetic energy is actually significantly greater than it was at the beginning. That difference comes through the formation the flux tubes. They formed out of an early magnetic reconnection episode that magnetically isolated the shear layer from adjacent plasma. Those flux tubes surround a sheet of hot, low density gas that contains most of the excess entropy generated through the various reorganization events leading to relaxation. Thus, through the action of the magnetic field the flow has been reorganized into a thin, hot entropy layer surrounded by magnetically enhanced structures within a broad laminar shear layer. In this case the shear layer is much broader than for the stronger field case and actually substantially broader than the width of the single, “Cat’s Eye”, vortex that forms in the pure gasdynamic case. The last distinction results from the fact that magnetic reconnection following the initial generation of the vortex lead to a significant impulsive transfer of both energy and momentum away from the initial shear layer.
Even though our “weak field” case had a magnetic field initially dynamically fairly small, it was strong enough that before the primary vortex could become fully developed, magnetic stresses became [*locally*]{} important. It is reasonable, therefore, to ask what differences should be expected when the initial field is initially truly very weak. Some discussions of MHD turbulence in this very weak field regime (e.g., [@tingetal]) have argued that magnetic fields will never become dynamically important; i.e, the fields remain weak and passive. But, we suspect that the answer depends on the effective magnetic Reynolds of the flow, $R_m = U_oL/\eta$, comparing the rate at which magnetic field is rolled into a vortex to the rate at which it diffuses out of the vortex. The simulations in the [@tingetal] study, for example, had modest magnetic Reynolds numbers, $< 100$. On the other hand, as pointed out, for example, by [@catvain91] and [@cat94], when the magnetic Reynolds number is big and there is a large scale magnetic field, field line stretching can produce locally strong fields. In astrophysical flows we may often anticipate very large magnetic Reynolds numbers, so it is reasonable to expect in cases where the large scale field starts out being quite small that it might eventually play a role in determining the outcome (This point has been made quite clearly with regard to MHD turbulent transport ([@catvain91])). As applied to the present problem, this would mean that the flows might exist longer as pseudo-gasdynamic, but that eventually field lines would be stretched sufficiently that magnetic tension would become locally involved in evolution of the vortex. This process should also, of course, naturally create tearing mode unstable magnetic topologies (e.g., [@biswel89]), whose growth rates depend on the magnetic Reynolds number as well (e.g., [@mel86]). That would provide a limiting process to accompany the eventual feedback on dynamical stresses. So, an outcome similar to what we have observed in the two cases studied here is still reasonable for very weak fields when the magnetic Reynolds number is large enough. In practical problems, where flow conditions may be more complex or time variable on large scales, these characteristics would have to be balanced against time and length constraints imposed by those features, of course.
This work by AF, TWJ, and JBG was supported in part by the NSF through grants AST-9100486 and AST-9318959, by NASA through grant NAGW-2548 and by the University of Minnesota Supercomputer Institute. The work by DR was supported in part by the Basic Science Research Institute Program, Korean Ministry of Education 1994, Project No. BSRI-94-5408. We are grateful to Ibrahim Hallaj for very helpful assistance in analyzing the results. We also appreciate some very constructive comments on the manuscript by Akira Miura and Andrea Malagoli.
Artola, M. & Majda A. J. 1987, Physica D, 28, 253
Balbus, S. & Hawley, J. 1991, , 376, 214
Belle-Hamer, A. L., Fu, Z. F., & Lee, L. C. 1988, Geophys. Res. Lett., 15, 152 Belmont, G. & Chanteur, G. 1989, Physica Scripta, 40, 124
Biskamp, D. & Welter, H. 1989, Phys. Fluids B 1, 1964
Birkinshaw, M. 1991, in “Beams and Jets in Astrophysics", P. A. Hughes, ed., Cambridge University Press, Cambridge, p 278
Blumen, W. 1970, J. Fluid Mech., 40, 769
Blumen, W., Drazin, P. G., & Billings, D. F. 1975, J. Fluid Mech., 71, 305
Cattaneo, F. 1994, , 434, 200
Cattaneo, F. & Vainstein, S. I. 1991, 376, L21
Chandrasekhar, S. 1960, Proc. Nat. Acad. Sci., 46, 53
Chandrasekhar, S. 1961, “Hydrodynamic & Hydromagnetic Stability”, Oxford University Press, New York
Corcos, G. M. & Sherman, F. S., 1984, J. Fluid Mech., 139, 29
Dobrowolny, M., Mangeney, A. & Veltri, P. 1980, Phys. Rev. Lett., 45, 144
Ferrari, A., Trussoni, E. & Zaninetti, L. 1981 , 196, 1051
Fiedler, R. L. & Jones, T. W. 1984, , 283, 532
Landau, L. 1944, Akad. Nauk. S.S.S.R., Comptes Rendus (Doklady), 44, 139
Landau, L. D. & Lifshitz, E. M. 1987, “Fluid Mechanics”, Pergamon Press, Oxford
MacLow, M.-M., McKee, C. F., Klein, R. I., Stone, J. M., Norman, M. L. 1994, , 433, 757
Malagoli, A., Bodo, G. & Rosner R. 1995 (in press)
Maslowe, S. A. 1981, in “Hydrodynamic Instabilities and the Transition to Turbulence”, H. L. Swinney & J. P. Gollub, eds., Springer-Verlag, Berlin, 181
Melrose, D. B. 1986, “Instabilities in Space and Laboratory Plasmas”, Cambridge University Press, Cambridge
Miles, J. W. 1957, J. Acoust. Soc. Am., 29, 226
Miura, A. 1984, JGR, 89, 801
Miura, A. 1987, JGR, 92, 3195
Miura, A. 1990, Geophys. Res. Lett., 17, 749
Miura, A. 1992, JGR, 97, 10,655
Miura, A. & Pritchett, P. L. 1982, JGR, 87, 7431 (MP)
Moffatt, H. K. 1978, “Magnetic Field Generation in Electrically Conducting Fluids”, Cambridge University Press, Cambridge
Nordlund, Å, etal 1992, , 392, 647
Porter, D. H. & Woodward, P. R. 1994 , 93, 309
Pouquet, A 1978, J. Fluid Mech., 88, 1
Pouquet, A., Meneguzzi, M. & Frisch, U. 1986, Phys. Rev. A, 33, 4266
Priest, E. R. 1984, “Solar Magnetohydrodynamics”, D. Reidel Publ. Co., Dortrecht
Ryu, D. & Jones, T. W. 1995, , 442, 228
Ryu, D., Jones, T. W. & Frank, A. 1995, , Oct 20.
Ting, A. C., Matthaeus, W. H. & Montgomery, D. 1986, Phys. Fluids, 29, 3261
Tajima, T. & Leboeuf, J. N. 1980, Phys. Fluids 23, 884
Wang, Y.-M. & Robertson, J. A. 1984, Å, 139, 103
Wu, C. C. 1986, JGR, 91, 3042
Zeldovich, Ya. B. 1957, Sov. Physics JETP, 4, 460
[lrrrrrrr]{} ${\rm strong~field}$ & $1$ & $0.4$ & $L/25$ & $1$ & $2.5$ & $7.5$ & $0.053(U_o/2a)$ ${\rm weak~field}$ & $1$ & $0.2$ & $L/25$ & $1$ & $5$ & $30$ & $0.108(U_o/2a)$
[**FIGURE CAPTIONS**]{}
Fig. 1
: [The evolution of spatially averaged root mean square transverse velocities normalized by their initial values. Shown are the natural logarithm of the square root of the average $u_y^2$ throughout the evolution of the: (a) strong field case; (b) weak field case. The solid, dotted, and dashed-dotted lines correspond to the high ($512^2$), medium ($256^2$), and low ($128^2$) resolution simulations respectively.]{}
Fig. 2
: [Gas density, (a), and magnetic field lines, (b), at 6 times in the evolution of the high resolution strong field simulation. The gas density is shown as greyscale images where dark correspond to low values. To facilitate visualization of periodic structures across the boundaries each image has been doubled. Coordinate values, measured from the lower left are in units of $L$. The times shown are 3.0, 6.0, 9.0, 11.9, 14.9 and 17.9 in units of $\tau= t/\Gamma^{-1}$.]{}
Fig. 3
: [Gas density, (a), magnetic field lines, (b), and velocity field, (c), at 6 times in the evolution of the highest resolution weak field simulation. Presentation is similar to Fig. 2. The velocity field is shown as vectors. The times shown are 2.4, 5.6, 8.0, 10.4, 14.4 and 20.0 in units of $\tau= t/\Gamma^{-1}$.]{}
Fig. 4
: [Greyscale images (top to bottom) of the entropy, gas and magnetic pressures at the end of the weak field run (t = 20 $\tau$). Dark correspond to low values. To facilitate visualization on the periodic space each image has been tripled.]{}
Fig. 5
: [Fractional difference in magnetic energy between initial and final states in simulations as a function of resolution. Shown are ${{E_{bf} - E_{bo}} \over E_{bo}}$ where the subscripts “f” and “o” refer to the final and initial values of the magnetic energy. Note the units on the abscissa in the strong field case.]{}
Fig. 6
: [The evolution of the spatially averaged squared magnetic field components for the highest resolution weak field simulation. Shown are the values of $\langle (B_{k} - B_{ko})^2\rangle$, where $k=x,y$, are indices. They are normalized to the initial $x$ component squared, $(B_{xo})^2$. The solid, and dotted lines show the evolution of the $x$, and $y$ components respectively.]{}
Fig. 7
: [Evolution of energy components and $\beta$ parameter for the strong field field case simulations. Shown are spatially averaged values of the thermal $E_t$, kinetic $E_t$, and magnetic energies $E_b$ (normalized to the total energy $E_{Tot}$ = $E_t$ + $E_k$ + $E_b$). In addition we show the evolution of the minimum value on the computational grid of the plasma parameter $\beta = {p_g \over p_b}$. The solid, dotted, and dashed-dotted lines correspond to the high ($512^2$), and medium ($256^2$) resolution simulations respectively.]{}
Fig. 8
: [Same as Fig. 7 for the weak field case simulations. The dashed-dotted lines corresponds to the low resolution simulation.]{}
Fig. 9
: [Evolution of the secondary vortex. Shown are the value of the surface integral of vorticity and line integral of magnetic tension in equation (3.6). Note that the area integral was carried out over the vortex and the line integral surrounded it. The vorticity is positive inside the area, so the signs indicate a reduction in the strength of the vortex.]{}
Fig. 10
: [Evolution of the cross helicity for the high resolution weak field simulation. Shown are the values of $|\langle H\rangle |$ where $H = \int {\hat{\bmit u}} \cdot
{\hat {\bmit B}}dxdy$. The dashed (dotted) line corresponds to the top (bottom) half-planes.]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The exact reduced dynamics for the independent oscillator model in the RWA approximation at zero and finite temperatures is derived. It is shown that the information about the interaction and the environment is encapsulated into three time dependent coefficients of the master equation, one of which vanishes in the zero temperature case. In currently used optical cavities all the information about the field dynamics is contained into [*two*]{} (or three) experimentally accesible and physically meaningful real functions of time. From the phenomenological point of view it suffices then to carefully measure two ([*three*]{}) adequate observables in order to map the evolution of any initial condition, as shown with several examples: (generalized) coherent states, Fock states, Schrödinger cat states, and squeezed states.'
author:
- 'K. M. Fonseca Romero$^{(1)}$[^1] [ ]{} and M. C. Nemes$^{(2,3)}$[^2]'
title: Exact Description of Decoherence in Optical Cavities
---
[$^{(1)}$ Departamento de Física, Facultad de Ciencias,\
*Universidad Nacional, Ciudad Universitaria, Bogotá, Colombia*]{}
[$^{(2)}$ Departamento de Física–Matemática, Instituto de Física,\
*Universidade de São Paulo, C.P. 66318, 05315-970 São Paulo, S.P., Brazil*]{}
[$^{(3)}$ Departamento de Física, ICEX,\
*Universidade Federal de Minas Gerais, C.P. 702, 30161-970 Belo Horizonte, M.G., Brazil*]{}
Introduction
============
Measuring the time development of the entanglement process of a system coupled to its environment is a most remarkable achievement and a challenging goal. The reason for this is that the entanglement process is a unique and typical quantum feature. Several attemps, both on the theoretical as well as on the experimental side have been recently made[@theor; @theor2]. In the particular case of high Q optical cavities a direct measure of the decoherence process has been given and suggestions of experiments with essentially the same set up have been made on how to directly measure the Wigner function of the initially correlated field produced in the cavity[@Haroche; @Lutterbach]. We show that all necessary information to construct the time development of any Wigner function (or any system’s density operator) can be obtained by a precise measure of three quantities as function of time: the average photon number and two orthogonal field quadratures.
Recently the exact master equation for quantum Brownian motion in a general environment has been derived using both path integral techniques[@Hu] and the tracing of the evolution equation for the Wigner function[@Habib; @Halliwell]. We closely follow the later approach to derive the exact master equation for the oscillator independent model in the RWA at zero temperature and with a factorized initial condition. The hamiltonian of the model is $$\label{Eq:RWAHamiltonian}
H=\hbar\omega(a^\dagger a + 1/2) +
\hbar \sum_k \omega_k(a^\dagger_k a_k + 1/2) +
\hbar \sum_k c_k (a^\dagger a_k + a^\dagger_k a).$$ Here we present the solution of the initial value problem, which have been solved in the Heisenberg picture in Refs. [@Hope; @Savage]. The solution allows for an easy visualization in contrast to the model without the RWA approximation. Besides its intrinsic interest as an exactly soluble model, the hamiltonian (\[Eq:RWAHamiltonian\]) may be useful in treating leaking Bose-Einstein condensates [@Moy], in materials with modified dispersion relations[@Bay], or in any case of non-ohmic strength function, where the Born-Markov approximation is not adequate[@Moy].
We assume that at $t=0$ the total density operator is given by $$\label{Eq:InitialRho}
\rho(0) = \rho_S(0) \otimes
\prod_k \frac{e^{-\beta \hbar \omega_k a_k^\dagger a}}
{{\rm Tr }e^{-\beta \hbar \omega_k a_k^\dagger a}}
\stackrel{\beta \rightarrow \infty}{\longrightarrow}
\rho_S(0) \otimes \prod_k |0_k\rangle \langle0_k|,$$ where the subscript $S$ for system refers to the main oscillator. The bath, i.e. the set of oscillators labelled by $k$, is initially in thermal equilibrium at inverse temperature $\beta$. At zero temperature, the tensor product of the vacuum of the main oscillator with the vacuum of the set of oscillators is the ground state of (\[Eq:RWAHamiltonian\]).
The Exact Master Equation
=========================
It is well known that for quadratic hamiltonians the Wigner function satisfies the classical Liouville equation. To obtain the classical hamiltonian corresponding to (\[Eq:RWAHamiltonian\]), one uses the correspondence rule $a_\mu^{(\dagger)} \rightarrow \alpha_\mu^{(*)}$, where $$\alpha_\mu^{(*)} = \sqrt{\frac{m_\mu \omega_\mu}{2 \hbar}} q_\mu
+ (-) \frac{i}{\sqrt{2\hbar m_\mu \omega_\mu}}p_\mu,$$ and discards the zero energy contributions. We use $\mu=0,1,2,..$, $k=1,2,..$ and $a_0\equiv a$, $\omega_0 = \omega$. In general greek subindices denote non negative integers while latin sub indices denote positive integers. Using these conventions, the classical Liouville equation $$\label{Eq:Liouville}
\frac{\partial W(\alpha_\mu,\alpha_\mu^*,t)}{\partial t} =
\frac{1}{i\hbar} \sum_{\mu} \frac{\partial H}{\partial \alpha_\mu}
\frac{\partial W}{\partial \alpha_\mu^*} -
\frac{1}{i\hbar} \sum_{\mu} \frac{\partial H}{\partial \alpha_\mu^*}
\frac{\partial W}{\partial \alpha_\mu},$$ where $W(\alpha_\mu,\alpha_\mu^*,t)$ is the Wigner function in the quantum case and the probability density function in the classical case, can be written as $$\begin{aligned}
\nonumber
\frac{\partial W(\alpha_\mu,\alpha_\mu^*,t)}{\partial t} & = &
-i\omega \alpha^* \frac{\partial W}{\partial \alpha^*}
+i\omega \alpha \frac{\partial W}{\partial \alpha}
-i\sum_k \omega_k \alpha_k^* \frac{\partial W}{\partial \alpha_k^*}
+i\sum_k \omega_k \alpha_k \frac{\partial W}{\partial \alpha_k}\\
\label{Eq:WignerTotal} & &
-i\sum_k c_k \alpha_k^* \frac{\partial W}{\partial \alpha^*}
+i\sum_k c_k \alpha_k \frac{\partial W}{\partial \alpha}
-i\sum_k c_k \alpha^* \frac{\partial W}{\partial \alpha_k^*}
+i\sum_k c_k \alpha \frac{\partial W}{\partial \alpha_k}.\end{aligned}$$ The initial condition Eq. (\[Eq:InitialRho\]) in the language of Wigner functions is $$\label{W0}
W(\alpha_\mu,\alpha_\mu^*,t=0) =
W^0(\alpha_\mu,\alpha_\mu^*) =
W^0_S(\alpha,\alpha^*) W^0_B(\alpha_k,\alpha_k^*)=
W^0_S(\alpha,\alpha^*) \prod_k N_k
{\rm e}^{-2\tanh (\hbar\omega_k\beta)\alpha_k \alpha_k^*},$$ where the $N_k$ are normalization constants. Integrating Eq.(\[Eq:WignerTotal\]) over the bath variables we get $$\frac{\partial \widetilde{W}(\alpha,\alpha^*,t)}{\partial t} =
-i\omega \alpha^* \frac{\partial \widetilde{W}}{\partial \alpha^*}
+i\omega \alpha \frac{\partial \widetilde{W}}{\partial \alpha}
-i \frac{\partial G^*}{\partial \alpha^*}
+i \frac{\partial G}{\partial \alpha},$$ with $$\widetilde{W}(\alpha,\alpha^*,t) = \int \left(
\prod_k d\alpha_k d\alpha_k^* \right)
W(\alpha_\mu,\alpha_\mu^*,t),$$ and $$G(\alpha,\alpha^*,t) = \int \left(
\prod_k d\alpha_k d\alpha_k^* \right)
\sum_k c_k \alpha_k W(\alpha_\mu,\alpha_\mu^*).$$ As in [@Halliwell] it is easy to show that $G(\alpha,\alpha^*,t)$ can be written in terms of $\widetilde{W}$. We notice that for quadratic hamiltonians $W(\alpha_\mu,\alpha_\mu^*,t) = W^0 (\alpha_\mu(-t),
\alpha_\mu^*(-t))$, where $\alpha_\mu(t)$ is the solution of the classical equations of motion. If we define $\vec{\alpha}(t)=(\alpha_0(t), \alpha_1(t), \ldots)$ and denote its transpose by $\vec{\alpha}^T(t)$, we have $$\label{Eq:Solution}
\vec{\alpha}^T(t) =
U^\dagger \Delta(t) U \vec{\alpha}^T(0),
\hspace{.5cm} \vec{\alpha}^* (t) = (\vec{\alpha}(t))^*,$$ where $U$ and $\Delta$ are unitary, and $\Delta$ is diagonal. Taking the Fourier transform of $G$, and changing variables from $\{\alpha_\mu(-t),\alpha_\mu^*(-t)\}$ to $\{\alpha_\mu(0),
\alpha_\mu^*(0)\}$, with unit Jacobian, we obtain $$\begin{aligned}
\nonumber
G(\kappa,\kappa') =& & \int \prod_\mu d\alpha_\mu(0) d\alpha_\mu^*(0)
{\rm e}^{i\kappa\sum_\nu p_\nu(t) \alpha_\nu (0)}
{\rm e}^{i\kappa'\sum_\nu p_\nu^*(t) \alpha_\nu^* (0)}
\\ \label{Eq:Gdek}&&
\times \sum_\nu q_\nu (t) \alpha_\nu (0)
W_S^0(\alpha^{(*)}(0)) W_B^0(\alpha_k^{(*)}(0)),\end{aligned}$$ with $\{p_\nu,p_\nu^*,q_\nu \}$ time dependent parameters. From Eq.(\[Eq:Gdek\]) it is easy to see that the multiplication by $\alpha_0 = \alpha$ is equivalent to a derivation with respect to $k$, plus terms corresponding to multiplication by $\alpha_k$, up to time dependent coefficients. These last terms, as can be seen from (\[W0\]), correspond to derivations w.r.t. $\alpha_k^*$, which in turn, are equivalent to multiplication by $k'$, as shows a simple integration by parts. Taking the inverse Fourier transform we obtain a multiplication by $\alpha$ and a derivation w.r.t. $\alpha^*$. Thus, observing that the Fourier transform of $\widetilde{W} (\alpha,\alpha^*)$, $\widetilde{W} (\kappa,\kappa')$ is given by $$\begin{aligned}
\nonumber
\widetilde{W} (\kappa,\kappa') = &&
\int \prod_\mu d\alpha_\mu(0) d\alpha_\mu^*(0)
{\rm e}^{i\kappa\sum_\nu p_\nu(t) \alpha_\nu (0)}
{\rm e}^{i\kappa'\sum_\nu p_\nu^*(t) \alpha_\nu^* (0)}
\\ \nonumber&&
\times W_S^0(\alpha^{(*)}(0)) W_B^0(\alpha_k^{(*)}(0)),\end{aligned}$$ we obtain $$\begin{aligned}
\nonumber
i\frac{\partial G}{\partial \alpha} -
i\frac{\partial G^*}{\partial \alpha^*}
& = & iY \frac{\partial}{\partial \alpha}
\left( \alpha \widetilde{W} \right)
-iY^* \frac{\partial}{\partial \alpha^*}
\left( \alpha^* \widetilde{W} \right)\\ & &
+(iZ-iZ^*) \frac{\partial^2 \widetilde{W}}
{\partial\alpha\partial\alpha^*},\end{aligned}$$ with time dependent functions $Y, Z$. Therefore the Wigner equation can be written as $$\begin{aligned}
\nonumber
\frac{\partial \widetilde{W}(\alpha,\alpha^*,t)}{\partial t} & = &
-i(\omega+\delta)
\left( \alpha^* \frac{\partial \widetilde{W}}{\partial \alpha^*}
- \alpha \frac{\partial \widetilde{W}}{\partial \alpha}\right)
+2\lambda \widetilde{W}\\ \label{Eq:WignerEq} & &
+\lambda \left(\alpha^*\frac{\partial \widetilde{W}}{\partial \alpha^*}
+ \alpha\frac{\partial \widetilde{W}}{\partial \alpha}\right)
+\lambda' \frac{\partial^2 \widetilde{W}}
{\partial\alpha\partial\alpha^*},\end{aligned}$$ where we have set $iY = \lambda + i\delta$ and $iZ-iZ^* = \lambda'$. All of $\lambda, \delta$ and $\lambda'$ are real functions. By comparing the system of equations found from both (\[Eq:WignerTotal\]) and (\[Eq:WignerEq\]) we get $$\begin{aligned}
(\lambda+i\delta)\langle \alpha \rangle & = & i
\sum_k c_k \langle \alpha_k \rangle, \\
(\lambda+i\delta)\langle \alpha^2 \rangle & = & i
\sum_k c_k \langle \alpha \alpha_k \rangle, \\
-2\lambda' \langle \alpha \alpha^*\rangle & = &
i \sum_k c_k \langle \alpha \alpha_k^*\rangle
-i \sum_k c_k \langle \alpha^* \alpha_k \rangle.\end{aligned}$$ We know that the solution of the Heisenberg equations can be written as follows $$\begin{aligned}
\label{aoft}
a(t) & = & \eta(t) a(0) + \sum_k \gamma_k(t) a_k(0), \\ \label{akoft}
a_k(t) & = & \eta_k(t) a(t) + \sum_l \gamma_{kl} a_l(0).\end{aligned}$$ Using the above solution of the Heisenberg equations, and the fact that all of the first and second (symmetric) moments involving bath operators are zero, with the exception of $\langle \{a_k^\dagger, a_k\} \rangle =
2n_k+1=2\coth (\hbar\omega_k\beta/2)$, we obtain $$\begin{aligned}
\label{Eq:Sol1}
\lambda(t)+i\delta(t) & = & i \sum_k c_k \eta_k(t)\\
\label{Eq:Sol2}
\lambda'(t) & = & \sum_{kl} c_k (n_k(\beta)+ \frac{1}{2})
(i\gamma_l\gamma_{kl}^*-i\gamma_l^*\gamma_{kl}).\end{aligned}$$ For reasons that will be clear soon, we write the diffusion coefficient $\lambda'(t)$ as $\lambda(t)+\epsilon(t,\beta)$, with $\lim_{\beta
\rightarrow \infty}\epsilon(t,\beta)=0$, as shown in section \[DMEC\]. At zero temperature, since the tensor product of vacua is the ground state of (\[Eq:RWAHamiltonian\]), the corresponding (reduced) Wigner function $W_S(\alpha,\alpha^*)$ should be a stationary solution of the Wigner equation. When this condition is applied to Eq.(\[Eq:WignerEq\]), we obtain $\lambda = \lambda'$. If they were not equal it would imply the non existence of an exact master equation, as seems to be claimed in Ref. [@Moy]. However, this is not the case, as we show next.
It is not hard to show that the operator equation for the system’s reduced density operator equivalent to the Wigner equation (\[Eq:WignerEq\]) is $$\label{Eq:Master}
\frac{d\rho}{dt} =
\frac{1}{i\hbar}\left[\hbar(\omega+\delta)a^\dagger a,\rho \right]
+(\lambda+\epsilon)(2 a\bullet a^\dagger-a^\dagger a\bullet
-\bullet a^\dagger a)\rho +\epsilon(2 a^\dagger\bullet a-a
a^\dagger\bullet
-\bullet a a^\dagger)\rho=
{\cal L}(t)\rho(t),$$ where the usual dot superoperator convention has been used. The usual Born-Markov RWA master equation is of this form with constant coefficients[@Louisell]. Some results can be obtained at once from (\[Eq:Master\]): premultiplying by $a$ and taking the trace we get $$\frac{d}{dt}\langle a \rangle = \frac{d\alpha}{dt} =
(-i(\omega+\delta)-\lambda)\alpha,$$ which can be immediately solved to give $$\label{Eq:Alphat}
\alpha(t)=\exp(-i\Omega(t)-\Lambda(t))\alpha(0),$$ with $$\label{Eq:Defs}
\Omega(t)=\int_0^t d\tau(\omega+\delta)(\tau),
\hspace{0.5cm}
\Lambda(t) =\int_0^t d\tau \lambda(\tau).$$ Note that this result is [*independent*]{} of $\epsilon$, i.e., it does not depend on the temperature. Premultiplying (\[Eq:Master\]) by $a^\dagger a$, and taking the trace we get the following differential equation $$\label{eqforepsilon}
\frac{d}{dt}\langle a^\dagger a \rangle (t) =
-2 \lambda \langle a^\dagger a \rangle (t) + 2 \epsilon$$ with the solution $$\label{meanadaggera}
\langle a^\dagger a\rangle (t)= \exp(-2\Lambda(t))
\langle a^\dagger a\rangle (0)+{\cal N}(t)=
\langle a^\dagger a\rangle (t;\beta\rightarrow\infty)+
2\exp(-2\Lambda(t))\int_0^t d\tau \epsilon(\tau)\exp(2\Lambda(\tau)),$$ where it is evident that ${\cal N}(t)$ vanishes in the zero temperature limit. Contrary to the exact equations found in [@Hu; @Habib; @Halliwell] We thus have the following interpretations for the real functions that appear in the master equation: $\delta(t)$ is the instantaneous frequency shift, $\lambda(t)$ is the instantaneous energy rate of change at zero temperature and $\epsilon(t)$ is the instantaneous energy rate of change at finite temperature but with the system in the vacuum state. Moreover ${\cal N}(t)$, which is related to both $\epsilon(t)$ and $\delta(t)$ is the mean number of excitations when the initial state was the ground state.
The evolution superoperator and some initial states
===================================================
We can use Lie algebraic methods[@Borges] to find the evolution superoperator ${\cal U}$. Indeed, we can verify that the superoperators ${\cal M} = a^\dagger a\bullet$, ${\cal P} =
\bullet a^\dagger a$, ${\cal J} = a\bullet a^\dagger$ and ${\cal R} = a^\dagger\bullet a$ form an algebra, $$\label{Eq:Commutation}
[{\cal M},{\cal P}]=0, \quad [{\cal M},{\cal J}]
= - {\cal J} = [{\cal P},{\cal J}],\quad
[{\cal M},{\cal R}]
= - {\cal R} = [{\cal P},{\cal R}].$$ Thus, we can assume that ${\cal U}(t) = v {\rm e}^{w{\cal R}} {\rm e}^{x{\cal M}} {\rm e}^{y{\cal
P}}
{\rm e}^{z{\cal J}}$. Deriving this expression, using the formula $ \exp(xA) B \exp(-xA)
= B+x[A,B]+x^2 [B,[B,A]]/2! +...$ and the commutation relations (\[Eq:Commutation\]), comparing coefficients in the equation $d{\cal U}/dt$ = ${\cal L}(t) {\cal U}(t)$, and solving the resulting differential equations we obtain $$\begin{aligned}
\nonumber
v(t) & = & \frac{1}{1+{\cal N}(t)}, \quad
w(t) = \frac{{\cal N}(t)}{1+{\cal N}(t)}, \quad
x(t) = -i\Omega(t)-\Lambda(t)-\frac{1}{2}\ln (1+{\cal N}(t)) =
y^*(t), \\ \label{Ucoefs}
& & z(t) = 1-\frac{\exp(-2 \Lambda(t))}{1+{\cal N}(t)}.\end{aligned}$$
Let us suppose that $\rho$ satisfies the equation $d\rho/dt =
{\cal L} (X_i\bullet,\bullet X_i;t) \rho(t)$, where the $X_i$ are operators (notice that ${\cal L}$ is a general linear superoperator), and that $\rho$ can be written as $U\rho'U^{-1}$. Then, $\rho'$ satisfy the equation $d\rho'/dt = {\cal L}' \rho'(t)$. If, moreover, we choose $U=\exp(\sigma a^\dagger -\sigma^* a)=D(\sigma)$, the displacement operator, and ${\cal L}$ is that of the RWA, then $${\cal L}'(t) = {\cal L}(t) +
\left( (i\dot{\Omega}+\lambda)\sigma+\frac{d\sigma}{dt}\right)
(a^\dagger\bullet-\bullet a^\dagger) -
\left( (-i\dot{\Omega}+\lambda)\sigma^*+\frac{d\sigma^*}{dt}\right)
(a\bullet-\bullet a).$$ It is easy to see that if $$\label{sigmaoft}
\sigma(t)=\sigma(0)\exp(-i\Omega(t)-\Lambda(t)),
\hspace{0.5cm} \sigma^*(t) = (\sigma(t))^*,$$ then both $\rho$ and $\rho'$ satisfy the same equation. That is, we have shown that $D(\sigma(t))\rho(t)D^\dagger(\sigma(t))$ satisfies Eq. (\[Eq:Master\]) whenever $\rho(t)$ does the same. We remark that this result does [*not*]{} depend on the temperature of the bath.
We now turn to the evaluation of the density matrix evolved with the superoperator found above. We chose initial states relevant from the point of view of quantum optics. As a first initial state we choose the system’s ground state, $\rho(0)=\mid 0\rangle \langle 0\mid$. Since $$\begin{aligned}
{\rm e}^{x{\cal M}}\mid 0\rangle \langle 0\mid &= &{\rm e}^{y{\cal P}}
\mid 0\rangle \langle 0\mid = {\rm e}^{z{\cal J}}\mid 0\rangle \langle
0\mid = \mid 0\rangle \langle 0\mid,
\quad {\rm and} \\
{\rm e}^{x{\cal R}}\mid 0\rangle \langle 0\mid & = &
\sum_0^\infty \frac{x^n}{n!} (a^\dagger)^n
\mid 0\rangle \langle 0\mid a^n= \sum_0^\infty x^n
\mid n\rangle \langle n\mid,\end{aligned}$$ we have $$\rho(t) = {\cal U}(t) \mid 0\rangle \langle 0\mid =
\sum_0^\infty \frac{1}{1+{\cal N}}\left(1+1/{\cal N}\right)^n
\mid n\rangle \langle n\mid = \sum_0^\infty P_n(t)
\mid n\rangle \langle n\mid.$$ The above formula displays the so called decomposition in natural orbits[@Nemes] where the quantities $P_n$ can be directly interpreted as probabilities. We can write the evolved density matrix in the alternative form $\rho(t) = \exp((1+1/{\cal N})a^\dagger a)/
(1+{\cal N})$, which is the form of an instantaneous thermal density matrix, with ${\cal N}(t)=<a^\dagger a>(t)$. Had we chosen an initial thermal state, with mean number of excitations $\bar{n}(0)$, the density matrix would have remained a thermal state, but now $M(t) =
\bar{n}(0)\exp(-2\Lambda(t))+{\cal N}(t)$. If we use the instantaneous oscillator frequency $\omega'=\omega+\delta$, it is possible to define an instantaneous temperature through the relation $T(t)= \hbar(\omega+\delta)/(k_B \ln(1+1/{\cal N}))$, with $k_B$ the Boltzmann constant. Moreover, we have obtained a physical interpretation for the quantity ${\cal N}(t)$: it is the mean number of excitations of the main oscillator at time $t$ when it was initially prepared in its ground state.
To calculate the density matrix for an initial Fock state it is better to write the evolution superoperator in the form ${\cal U}(t) = v
\exp (w{\cal R}) \exp(z'{\cal J}) \exp(x{\cal M}) \exp(y{\cal P})$, where $w,x,y$ are given in (\[Ucoefs\]), and $z'(t) = (1+{\cal
N})\exp(2\Lambda)$. We use $$\begin{aligned}
\nonumber
e^{z{\cal J}} e^{x{\cal M}}e^{y{\cal P}} \mid m\rangle \langle m\mid & =
& \sum_{k=0}^{m} (e^{x+y})^{m-k}(ze^{x+y})^{k} \frac{m!}{(m-k)!k!}
\mid m-k\rangle \langle m-k\mid \\ & = &
\sum_{k=0}^{m} \frac{m!}{(m-k)!k!} (e^{x+y})^{k} (ze^{x+y})^{m-k}
\mid k\rangle \langle k\mid,
\quad and\\
e^{u{\cal R}} \mid m\rangle \langle m\mid & = &
\sum_{k=0}^{\infty} \frac{m! u^k}{(m-k)!k!}
\mid k\rangle \langle k\mid,\end{aligned}$$ to see that the density matrix at time $t$ is given by $\rho(t) =
\sum_{k=0}^\infty P_{m,s}(t) \mid s\rangle \langle s\mid$, with $$\label{probsFock}
P_{m,s}(t) = \frac{e^{-2m\Lambda}}{(1+{\cal N})^{m+1}}
\frac{m!}{s!} \sum_{k=0}^{{\rm min}(m,s)}
\frac{([1+{\cal N}]e^{2\Lambda}-1)^k}{(m-k)!(s-k)!}
\left( 1+\frac{1}{{\cal N}}\right)^{s-k}.$$ Since the former density matrix have been expressed in terms of natural orbits, the quantities $P_{m,s}(t) $ are readily interpreted as probabilities. The transformation property discussed above allows us to write the evolution of an initial generalized coherent state $\mid \sigma
m \rangle = D(\sigma) \mid m\rangle $, where $D$ is the displacement operator and $\mid n\rangle$ the n-th number state. We have ${\cal U}(t)\mid \sigma_0 m\rangle \langle \sigma_0 m\mid =
\sum_{k=0}^\infty P_{m,s}(t) \mid \sigma(t) s\rangle \langle \sigma(t)
s\mid$, with $ P_{m,s}(t)$ given by (\[probsFock\]) and $\sigma(t)$ by (\[sigmaoft\]).
One interesting point to be investigated is if there exists an asymptotic density operator. Provided that our environment is such that $\lim_{t\uparrow\infty}\Lambda(t)\rightarrow\infty$ and $\lim_{t\uparrow\infty} {\cal N}(t)= n_\infty$, the asymptotic evolution superoperator can be written as $$\lim_{t\uparrow\infty} {\cal U}(t) =
\frac{1}{1+n_\infty}
\exp (\frac{n_\infty}{1+n_\infty}{\cal R})
(\mid 0 \rangle \langle 0 \mid \bullet)
\exp({\cal J}) (\bullet \mid 0 \rangle \langle 0 \mid ),$$ which applied to a generic normalized initial density $\rho(0)$ gives $$\begin{aligned}
\nonumber
\rho_\infty & = &
\frac{1}{1+n_\infty}
\exp({\cal J}) \rho(0)\langle 0 \mid
\exp (\frac{n_\infty}{1+n_\infty}{\cal R}) \mid 0 \rangle \langle 0 \mid
\\ & = &
\frac{1}{1+n_\infty}\left({\rm Tr} \rho(0)\right)
\exp [(1+\frac{1}{n_\infty})a^\dagger a]\mid 0 \rangle \langle 0 \mid
= \frac{1}{1+n_\infty}
\exp [(1+\frac{1}{n_\infty})a^\dagger a]\mid 0 \rangle \langle 0 \mid.\end{aligned}$$ Thus, whenever the established conditions are met, the density operator approaches asymptotically to a thermal state with a mean number of excitations equal to that of the environment. The existence of a unique asymptotic density can not be taken for granted: in the model of decoherence without damping studied in references [@qdec] even when the coefficient of decoherence grows indefinitely with time, the asymptotic state depends on the initial state.
The normal order characteristic functional$C^{(n)} (\xi,\xi^\dagger,t)$ given by [@Louisell] $$C^{(n)} (\xi,\xi^\dagger,t) =
{\rm Tr } e^{i\xi a^\dagger} e^{i\xi^* a} \rho(t) =
{\rm Tr } e^{i\xi a^\dagger} e^{i\xi^* a} {\cal U}(t) \rho(0) =
{\rm Tr } {\cal U}^\dagger(t) e^{i\xi a^\dagger} e^{i\xi^* a} \rho(0),$$ where ${\cal U}(t)$ is the evolution superoperator and ${\cal
U}^\dagger(t)$ its adjoint, is the generating functional of the normally ordered moments. After a somewhat lengthy but straightforward calculation we obtain $$C^{(n)} (\xi,\xi^\dagger,t) =
e^{-2 {\cal N}\xi xi^\dagger}
{\rm Tr } e^{i\xi \exp(-\Lambda+i\Omega)a^\dagger}
e^{i\xi^* \exp(-\Lambda-i\Omega) a} \rho(0).$$ If we calculate $\langle a \rangle (t)$ and $\langle a\dagger a \rangle (t)$ using $$\langle a \rangle (t)=
\frac{\partial}{\partial \xi^*}C^{(n)}
(\xi,\xi^\dagger,t)\mid_{\xi=0=\xi^\dagger},
\quad
\langle a\dagger a \rangle (t)=
\frac{\partial^2}{\partial \xi\partial \xi^*}
C^{(n)} (\xi,\xi^\dagger,t)\mid_{\xi=0=\xi^\dagger},$$ we arrive at the same results as before. These can be compared to those obtained in reference [@Hope].
The Zero Temperature Limit
===========================
The zero temperature limit has its own special interest, both as an approximation at low temperatures, and as the relevant case for leaking Bose-Einstein condensates. In this case the evolution superoperator can be expressed as ${\cal U}(t) = {\rm e}^{\tilde{x}{\cal M}}
{\rm e}^{\tilde{y}{\cal P}} {\rm e}^{\tilde{z}{\cal J}}$ with $$\label{Utildecoefs}
\tilde{x}(t) = -i\Omega(t)-\Lambda(t)= \tilde{y}^*(t), \quad
\tilde{z}(t) = 1-{\exp(-2 \Lambda(t))}.$$
Since the vacuum is solution of the master equation, the transformation property discussed above indicates that $$\label{Eq:CState}
D(\sigma(t))| 0 \rangle\langle 0| D^\dagger (\sigma(t)) =
|\sigma(t)\rangle \langle \sigma(t)| = {\cal U}(t)
|\sigma_0\rangle \langle \sigma_0|$$ also solves the master equation. That means that initial coherent states evolve preserving their coherence, not matter what the details of the environment. A measurement of the norm and the phase of an initial coherent state is enough to determine the functions $\Omega$ and $\Lambda$, and hence, in principle, to determine the evolution of any other initial state.
The evolution of an initial Fock state is also easily calculated. $$\label{Eq:FState}
{\cal U}(t) | m \rangle\langle m|
= \sum_{k=0}^{m} p_{k,m}(t) | k\rangle\langle k| =
p_{k,m}(t) = \frac{m!}{k! (m-k)!} ({\rm e}^{-2\Lambda(t)})^k
(1-{\rm e}^{-2\Lambda(t)})^{m-k}| k\rangle\langle k| .$$ From this solution we can generate another solution: if we have an initial generalized coherent state, $| m;\sigma_0 \rangle =
D(\sigma_0) | m \rangle$, it evolves into a mixture of generalized coherent states, as given by $$\label{Eq:GCState}
{\cal U}(t) | m;\sigma_0 \rangle\langle m;\sigma_0| =
\sum_{k=0}^{m} p_{k,m}(t) | k;\sigma(t)\rangle\langle k;\sigma(t)|.$$
Applying the evolution operator ${\cal U}(t)$ to an initial density matrix element $| \sigma_0\rangle\langle {\sigma'}_0|$, one obtains $$\label{Eq:ABEvolution}
{\cal U}(t) | \sigma_0\rangle \langle {\sigma'}_0| =
\frac{\langle {\sigma'}_0|\sigma_0\rangle}
{\langle {\sigma'}(t)|\sigma(t)\rangle}
| \sigma(t)\rangle \langle \sigma'(t)|.$$ Notice that an heuristic argumentation also leads to (\[Eq:ABEvolution\]). Indeed, since a coherent state remains coherent, we expect ${\cal U}(t) | \sigma_0\rangle \langle {\sigma'}_0| =
N(t) | \sigma(t)\rangle \langle \sigma'(t)|$. As long as the exact dynamical equation for $\rho$ preserves the trace, $ \frac{d}{dt}{\rm Tr} \rho_S(t) = {\rm Tr} (\frac{d\rho_S(t)}{dt}) =
{\rm Tr} ({\cal L}(t)\rho_S(t)) = 0$, the normalization factor $N(t)$ cannot be other than that of Eq. (\[Eq:ABEvolution\]). The evolution of an initial even ($\rho_{\sigma_0e}$) or odd cat ($\rho_{\sigma_0o}$) state can be calculated from (\[Eq:ABEvolution\]) and (\[Eq:CState\]). Indeed, $$\rho_{\sigma_0e(o)} =
N_{e(o)}(\sigma_0)
(|\sigma_0\rangle, |-\sigma_0\rangle)
\left( \begin{array}{lr}
1 & (-) 1 \\ (-) 1 & 1
\end{array}\right)
\left( \begin{array}{c}
\langle\sigma_0| \\ \langle -\sigma_0|
\end{array} \right),$$ where $N_{e(o)}(\sigma_0) =
(1+(-)\langle -\sigma_0|\sigma_0\rangle)^{-1} /2$ is a normalization factor, evolves as follows $$\label{Eq:Cat}
{\cal U}\rho_{\sigma_0e(o)} =
N_{e(o)}(\sigma_0) (|\sigma_t\rangle, |-\sigma_t\rangle)
\left( \begin{array}{c}
1 \hspace{0.3cm} \frac{(-)\langle -\sigma_0|\sigma_0\rangle}
{\langle -\sigma(t)|\sigma(t)\rangle}\\
\frac{(-)\langle -\sigma_0|\sigma_0\rangle}
{\langle -\sigma(t)|\sigma(t)\rangle} \hspace{0.3cm} 1
\end{array} \right)
\left( \begin{array}{c} |\sigma_t\rangle \\
|-\sigma_t\rangle\end{array}\right).$$ We can rewrite Eq. (\[Eq:Cat\]) in a more convenient way, in terms of natural orbitals, as $${\cal U}(t)\rho_{\sigma_0e(o)} =
p_{e(o)}^{e(o)}(t) \rho_{\sigma(t)e(o)} +
p_{e(o)}^{o(e)}(t) \rho_{\sigma(t)o(e)},$$ with $$\begin{aligned}
\label{Eq:pdet}
p_{e(o)}^{e(o)}(t) & = & \frac{1}{2}
\frac{1+(-)\langle -\sigma(t)|\sigma(t)\rangle}
{1+(-) \langle -\sigma_0|\sigma_0\rangle}
\left( 1 + \frac{\langle -\sigma_0|\sigma_0\rangle}
{\langle -\sigma(t)|\sigma(t)\rangle}\right) \\ \label{Eq:qdet}
p_{e(o)}^{o(e)}(t) & = & \frac{1}{2}
\frac{1-(-)\langle -\sigma(t)|\sigma(t)\rangle}
{1+(-) \langle -\sigma_0|\sigma_0\rangle}
\left( 1 - \frac{\langle -\sigma_0|\sigma_0\rangle}
{\langle -\sigma(t)|\sigma(t)\rangle}\right).\end{aligned}$$ Observe that an initial cat state evolves as a mixture of even and odd cat states. Eq. (\[Eq:pdet\]) gives the probability of the cat state of the same parity as the initial state, and Eq. (\[Eq:qdet\]) the probability of the cat state of the other parity.
The evolution of an initial squeezed state is also easily computed. We notice that the density operator for the squeezed vacuum $\rho(\zeta)$, can be written as $$\rho(\zeta)= \lim_{\gamma\rightarrow\infty}
\rho(\zeta,\gamma) = \lim_{\gamma\rightarrow\infty}
(1-e^{-\gamma}) S(\zeta) e^{-\gamma a^\dagger a} S^\dagger(\zeta),$$ where $S(\zeta)=\exp((\zeta (a^\dagger)^2-\zeta^\dagger a^2)/4)$. Setting $\zeta = \xi \exp{i\phi}$, the following expressions for the second moments of $\rho(\zeta,\gamma)$ are found $$\langle (a^\dagger)^2 \rangle = \frac{e^{-i\phi}}{2} \sinh (\xi),
\hspace{0.5cm}
\langle a^2 \rangle = \frac{e^{i\phi}}{2} \sinh (\xi),
\hspace{0.5cm}
\langle \{ a,a^\dagger\} \rangle = \cosh (\xi).$$ The system of equations for the second moments, $$\begin{aligned}
\frac{d\langle a^2\rangle}{dt} = -2 \frac{d}{dt}(i\Omega+\Lambda)
\langle a^2 \rangle, \\
\frac{d\langle (a^\dagger)^2\rangle}{dt} =
2 \frac{d}{dt}(i\Omega-\Lambda) \langle (a^\dagger)^2 \rangle, \\
\frac{d\langle \{a,a^\dagger\} \rangle}{dt} = -2 \frac{d\Lambda}{dt}
\langle \{ a, a^\dagger\} \rangle + \frac{d\Lambda}{dt},\end{aligned}$$ is integrated to give $$\begin{aligned}
\langle (a^2)^{(\dagger)}\rangle(t) =
\exp(-(-)2i\Omega-2\Lambda)\langle (a^2)^{(\dagger)}\rangle(0),
\\
\langle \{ a, a^\dagger\} \rangle (t) =
\exp(-2\Lambda) \langle \{ a, a^\dagger\} \rangle (0)
+(1-\exp(-2\Lambda))/2.\end{aligned}$$ From these relationships we find ${\cal U}(t) \rho(\zeta) =
\rho(\zeta(t),\gamma(t))$, with $$\begin{aligned}
\label{zetaoft}
\zeta(t) = \xi(t) \exp(i\phi(t))
\hspace{0.5cm} \phi(t)=\phi_0-2\Omega\\ \label{xioft}
\xi(t) = {\rm ArcTanh}
(\frac{\sinh(\xi_0)}{\cosh(\xi_0)+\exp(2\Lambda)-1})\\
\label{gammaoft}
\gamma(t) = {\rm ArcCoth}
(\sqrt{e^{-4\Lambda}+2\cosh(\xi_0)e^{-2\Lambda}(1-e^{-2\Lambda})
+(1-e^{-2\Lambda})^2}).\end{aligned}$$ Now the evolution of a general initial state $\rho(\sigma,\zeta) =
D(\sigma)\rho(\zeta)D^\dagger(\sigma)$ is $$\begin{aligned}
{\cal U}(t) \rho(\sigma,\zeta)& = &
\rho(\sigma(t),\zeta(t),\gamma(t))=
D(\sigma(t))\rho(\zeta(t),\gamma(t))D^\dagger(\sigma(t)),
\quad {\rm or} \\
{\cal U}(t) \left( |\sigma_0,\zeta_0,n=0\rangle
\langle \sigma_0,\zeta_0,n=0|\right) & = &
D(\sigma(t)) S(\zeta(t))
(1-e^{-\gamma(t)}) e^{-\gamma(t)a^\dagger a}
S^\dagger(\zeta(t))D^{\dagger}(\sigma(t)),\end{aligned}$$ where $\sigma(t),\zeta(t)$ and $\gamma(t)$ are those of Eqs. (\[sigmaoft\]) and (\[zetaoft\]–\[gammaoft\]), and $$|\sigma_0,\zeta_0,n\rangle = D(\sigma(t)) S(\zeta(t)) |n\rangle.$$ The corresponding expansion in terms of natural orbits is $$\begin{aligned}
\nonumber
{\cal U}(t) \left( |\sigma_0,\zeta_0,0\rangle
\langle \sigma_0,\zeta_0, 0| \right) & = & \sum_{n=0}^\infty
p_n(t) |\sigma (t),\zeta(t),n\rangle
\langle \sigma (t),\zeta(t),n| \\
& = & \sum_{n=0}^\infty
e^{-n\gamma(t)} (1-e^{-\gamma(t)}) |\sigma (t),\zeta(t),n\rangle
\langle \sigma (t),\zeta(t),n|.\end{aligned}$$
Determination of the Master Equation Coefficients {#DMEC}
=================================================
Let’s return to Eq.(\[Eq:Solution\]). If we call $w_\mu$ the exact eigenfrequencies of (\[Eq:RWAHamiltonian\]), we can write (remenber that greek indices can assume the value 0, while latin indices do not) $$\label{Eq:Sol3}
a_\nu (t) = \sum_{\mu\sigma} U_{\mu\nu}^* U_{\mu\sigma}
{\rm e}^{-iw_\mu t} a_{\sigma}(0) =
\sum_{\sigma}Z_{\nu\sigma} a_{\sigma}(0).$$ For the sake of convenience we write $$\eta = Z_{00}, \hspace{0.3cm} \gamma_k = Z_{0k}, \hspace{0.3cm}
\Delta_k = Z_{k0}, \hspace{0.3cm} \Gamma_{kl} = Z_{kl}.$$ Using (\[Eq:Sol3\]) in Eqs. (\[aoft\]) and (\[akoft\]), we obtain the following expressions for $\eta_k$ and $\gamma_{kl}$, $$\eta_k = \frac{\Delta_k}{\eta} \hspace{.5cm}
\gamma_{kl} = \Gamma_{kl} -\frac{\Delta_k \gamma_l}{\eta}.$$ From (\[Eq:Sol1\]) and (\[Eq:Sol2\]) we obtain $$\label{lambdas}
\lambda = - \sum_k c_k {\rm Im}(\eta_k), \hspace{.5cm}
\lambda' = - \sum_k c_k (2 n_k(\beta)+1) {\rm Im}(\beta_k) = -\sum_k c_k
(2 n_k(\beta)+1) {\rm Im}(\sum_l \gamma_l \gamma_{kl}^*).$$ Since $U$ is unitary we have $$\begin{aligned}
\nonumber
\sum_l \gamma_l \Gamma_{kl}^* & = & \sum_{\lambda\mu\nu}
U_{\mu 0}^* U_{\mu \lambda} U_{\nu k} U_{\nu \lambda}^*
{\rm e}^{-i(w_\mu-w_\nu)t} - \sum_{\mu\nu} U_{\mu 0}^* U_{\mu 0}
U_{\nu k} U_{\nu 0}^* {\rm e}^{-i(w_\mu-w_\nu)t}
\\ \label{Eq:Sum1} & = & \sum_{\mu\nu}
U_{\mu 0}^* U_{\nu k} {\rm e}^{-i(w_\mu-w_\nu)t} \delta_{\mu\nu}
- \eta\Delta^*_k = \sum_{\mu} U_{\mu 0}^*
U_{\mu k} - \eta\Delta^*_k = - \eta\Delta^*_k,\end{aligned}$$ and $$\label{Eq:Sum2}
\sum_l \gamma_l \gamma_l^* = \sum_{\lambda\mu\nu}
U_{\mu 0}^* U_{\mu \lambda} U_{\nu 0} U_{\nu \lambda}^* {\rm
e}^{-i(w_\mu-w_\nu)t}-\sum_{\mu\nu}
U_{\mu 0}^* U_{\mu 0} U_{\nu 0} U_{\nu 0}^* {\rm
e}^{-i(w_\mu-w_\nu)t} = 1 - \eta \eta^*.$$ Using (\[Eq:Sum1\]) and (\[Eq:Sum2\]) we get $$%\nonumber
\beta_k =
\sum_l \gamma_l \gamma_{kl}^* = \sum_l \gamma_l \Gamma_{kl}^*-
\frac{\Delta_k^*}{\eta^*}\sum_l \gamma_l \gamma_l^*
-\eta\Delta_k^*
-\frac{\Delta_k^*}{\eta^*}(1-\eta\eta^*) = -\eta_k^*.$$ Thus we finally notice that $\lambda'(t)$ can be expressed as a sum $$\lambda'(t)=-\sum_k c_k (2n_k(\beta)+1){\rm Im}(\eta_k)=
\lambda(t)-2 \sum_k c_k n_k(\beta){\rm Im}(\eta_k)=
\lambda(t)+\epsilon(t;\beta).$$ Observe that $\lim_{\beta\rightarrow\infty} \epsilon(t;\beta)=0$. This proves that indeed we have the equality $\lambda(t) = \lambda'(t)$ at zero temeperature, as can be seen by looking at their definitions as given in Eq. (\[lambdas\]).
We have seen that $\delta,\lambda$ and $\epsilon$ (or $\Omega, \Lambda$ and ${\cal N}$) are all we need to characterize the effect of the bath on the main oscillator. The Heisenberg equations of motion, $$\begin{aligned}
\label{Eq:System1}
\dot{a} & = & -i\omega_0 a -i\sum_k c_k a_k,\\
\label{Eq:System2}
\dot{a_k} & = & -i\omega_k a_k -i c_k a,\end{aligned}$$ can be solved in a number of ways. For example, an implicit method gives $$a(t) = a(0) e^{-i\Omega(t)-\Lambda(t)} -i\sum_k c_k a_k(0)
\int_0^t d\tau e^{-i\omega_k (t-\tau)} e^{-i\Omega(\tau)-\Lambda(\tau)}.$$ where $\eta(t)$ satisfies the integrodifferential equation $$\label{Eq:Integrodif}
\dot{\eta}+i\omega \eta + \int_0^t d\tau
\sum_k c_k^2 {\rm e}^{i\omega_k(t-\tau)} \eta(\tau)=0,$$ subject to the initial condition $eta_0=1$. Using the $\eta_k(t)$ implicitly defined above in eq (\[Eq:Sol1\]), and taking into account the equation satidfied by $\eta$, we obtain $\lambda+i\delta=
-i\omega+d(\ln \eta)/dt$, or $\eta(t)=\exp(-\Lambda(t)-i\Omega(t)$. We also find $$\epsilon(t) = \frac{e^{-2\Lambda(t)}}{2}\frac{d}{dt}
\left( e^{2\Lambda(t)} \sum_k c_k^2 \mid
\int_0^t d\tau e^{-i\omega_k (t-\tau)} e^{-i\Omega(\tau)-\Lambda(\tau)}
\mid^2 n_k (\beta) \right).$$ Had we used normal modes wewould have arrived at the expressions $$\begin{aligned}
\eta(t) & = & \sum_{\nu}
\frac{e^{-iw_{\nu}t}}{1+\sum_k\frac{c_k^2}{(w_\nu-\omega_k)^2}}, \quad
{\rm and}\\
\epsilon(t) & = & {\rm Re}\left(\frac{2i}{\eta(t)}\sum_{k,\nu}
\frac{n(\beta)c_k^2}{w_\nu-\omega_k}
\frac{e^{-iw_{\nu}t}}{1+\sum_l\frac{c_l^2}{(w_\nu-\omega_l)^2}}
\right).\end{aligned}$$ We moreover would have noticed that, in order not to have inverted oscillators the following condition has to be fulfilled, $\omega> \sum_k
c_k^2/\omega_k$. Since (\[Eq:Integrodif\]) can be hard to solve, methods to obtain approximate solutions are welcome. For instance, we can expand in powers of $c_k$ to second order, $\alpha(t) = \exp
(k_0 + c k_1 + c^2 k_2)$, to obtain $$\label{Eq:PerturbativeSol}
-i\Omega(t)-\Lambda(t) = -i\omega_0 t -\sum_k c_k^2 \int_0^t dt_1
\int_0^t dt_2 {\rm e}^{-i(\omega_k-\omega_0) t_2}.$$ We expect (\[Eq:PerturbativeSol\]) to be valid for small $c_k$. It can be shown that this is the time dependent Born–Markov approximation, which is valid also for “strong” coupling and short times[@Duarte].
Experimental Characterization
=============================
We observe that the calculation of the mean energy, and of the entropy of the above examples of initial conditions only require the knowledge of $\Lambda(t)$. However, if for example one wants to measure the Wigner function of any field density matrix, one would need to know $\Omega(t)$ as well. To determine this function one takes advantage of the experimental setup to measure the Wigner function[@Lutterbach], in which a two level atom is prepared in the excited state $|e\rangle$, sent trough an array of two low–Q cavities R1 and R2 and one high–Q cavity C between them, and is detected eventually. The field in C is displaced by the operator $D(\alpha) = \exp (\alpha a^\dagger -\alpha^* a)$, by a microwave source connected to it. R1 and R2 behave as “rotation” operators in the Hilbert space of atomic states, $\mid e \rangle \rightarrow (\mid e
\rangle + e^{i\xi} \mid g \rangle) / \sqrt{2}$, and $\mid g \rangle
\rightarrow (-e^{-i\xi}\mid e \rangle + \mid g \rangle) / \sqrt{2}$, with $\xi=0$ in R1 and $\xi=\pi/2$ in R2. The dispersive atom–field interaction in C produces entanglement: The field component associated with $\mid e \rangle$ suffers the action of the operator $\exp (i\pi
(a^\dagger a +1)/2)$, while the one associated with $\mid e \rangle$ suffers the action of $\exp (-i\pi a^\dagger a /2)$. It was shown in Ref. [@Lutterbach] that for this experimental arrangement $$\Delta P = P_e-P_g = W(-\alpha,-\alpha^*, t) /2,$$ where $P_e$ ($P_g$) is the probability to detect the probe atom in the upper (lower) state $|e\rangle$ ($|g\rangle$), and $W(-\alpha,
-\alpha^*, t)$ is the value of the Wigner function corresponding to the field in thehigh–Q cavity at the time $t$ the probe atom exits this cavity. Notice that, for the sake of convenience the normalization of $W$ has been changed by a factor of $\pi$. The Wigner function of the coherent state (\[Eq:CState\]) is $$W(\alpha,\alpha^*,t) = 2 e^{-2(\sigma(t)-\alpha)(\sigma(t)-\alpha)^*}.$$ Let’s notice that $W(\alpha,\alpha^*,t)$ presents a maximum at $\alpha =
\sigma(t) = \sigma_0 \exp(-\Lambda(t)-i\Omega(t))$. Since $\Lambda$ can be determined from a photocounting experiment, we just need to measure $\Omega(t)$. We choose $\alpha=\alpha(t)=\sigma_0 \exp(-\Lambda(t))
\exp(-i\Phi(t))$, and adjust $\Phi(t)$ to maximize $\Delta P$. If $\Phi
(t) = \Omega(t) {\rm mod} (2\pi)$, then we obtain $\Delta P = 1$, otherwise, it will be smaller than one, $\Delta P = \exp (-8 \sigma_0^2
\exp(-2\Lambda) \sin^2((\Omega-\Phi)/2))$. Choosing different exit times $t$ we can map $\Omega(t)$. Notice that what we do, in fact, is find the right $\alpha(t)$ that brings the field at the cavity C to the vacuum. If we succeed in so doing the probe atom is rotated in R1, getting in a superposition of upper and lower states, flies free till R2 where the rotation is reversed, and goes to the detectors again in the upper state.
Concluding Remarks
==================
In conclusion, we have shown that, if the optical cavities can be modelled through the hamiltonian (\[Eq:RWAHamiltonian\]), then only [*three, experimentally measurable*]{} real functions are necessary to characterize their behavior: the mean photon number of an initial ground state, the instantaneous frequency and the rate of change of the number of excitations. Measuring thes quantities once will enable one to get the Wigner function for any initial condition. So, provided the adequacy of the RWA model has been tested the present work provides for an alternative way to construct the time evolution of Wigner’s functions.
Acknowledgements
================
This work was partly funded by FAPESP, CNPq and PRONEX (Brazil), and Colciencias, DINAIN (Colombia). K.M.F.R. gratefully acknowledges the Instituto de Física, Universidade de São Paulo, for their hospitality and PRONEX for partial support.
[99]{} M. Brune, S. Haroche, V. Lefevre, J. M. Raimond and N. Zagury, Phys. Rev. Lett. [**65**]{}, 976 (1990); M. Brune, S. Haroche, J. M. Raimond, L. Davidovich and N. Zagury, Phys. Rev. [**A45**]{}, 5193 (1992). V. Buzek, H. Moya-Cessa, P. L. Knight and S. J. D. Phoenix, Phys. Rev. [**A45**]{}, 8190 (1992). S. Haroche and J. P. Raimond, [*Coherence and Quantum Optics VII*]{}, J. H. Eberly, L. Mandel and E. Wolf, Eds., Plenum Press 1996, p. 83. L. Davidovich, M. Brune, J. M. Raymond and S. Haroche, Phys. Rev. [**A53**]{}, 1295 (1996). V. V. Dodonov, I. A. Malkin and V. I. Man’ko, Physica [**72**]{}, 597 (1974). R. L. de Matos Filho and W. Vogel, Phys. Rev. Lett. [**76**]{}, 608 (1996). S. R. Jefferts, C. Monroe, E. W. Bell and D. J. Wineland, Phys. Rev. A [**51**]{}, 3112 (1995). C. Monroe, D. M. Meekhof, B. E. King and D. J. Wineland, Science [**272**]{}, 1131 (1996).
[^1]: E–mail: [email protected]
[^2]: E–mail: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Disk self-gravity could play an important role in the dynamic evolution of interaction between disks and embedded protoplanets. We have developed a fast and accurate solver to calculate the disk potential and disk self-gravity forces for disk systems on a uniform polar grid. Our method follows closely the method given by Chan et al. (2006), in which an FFT in the azimuthal direction is performed and a direct integral approach in the frequency domain in the radial direction is implemented on a uniform polar grid. This method can be very effective for disks with vertical structures that depend only on the disk radius, achieving the same computational efficiency as for zero-thickness disks.
We describe how to parallelize the solver efficiently on distributed parallel computers. We propose a mode-cutoff procedure to reduce the parallel communication cost and achieve nearly linear scalability for a large number of processors. For comparison, we have also developed a particle-based fast tree-code to calculate the self-gravity of the disk system with vertical structure. The numerical results show that our direct integral method is at least two order of magnitudes faster than our optimized tree-code approach.
author:
- 'Shengtai Li, Matthew J. Buoni, and Hui Li'
title: 'A Fast Potential and Self-Gravity Solver for Non-Axisymmetric Disks'
---
Introduction
============
Well before extrasolar planets were discovered, [@GoTr79; @GoTr80] and [@LiPa86a; @LiPa86b] have speculated that tidal interactions between disks and embedded protoplanets would lead to planet migration. [@Ward97] suggested that two different types of migration could occur. [@NeBe03a; @NeBe03b] studied numerically the effects of disk self-gravity in two-dimensional simulations of planet-disk interactions. [@PiHu05a] reported by an analytical derivation that the disk gravity accelerates the planetary migration. Recently, simulations by [@BaMa08] confirmed that the self-gravity indeed accelerates the type I migration. They implemented a 2D self-gravity solver in their code using the fast Fourier transform (FFT) method of [@BiTr87], which requires a logarithmic radial spacing ($\log(r)$).
In Newtonian gravity, we can define the gravitational potential $\Psi$ associated with the mass density, $\rho$, by the volume integral $$\Psi({\bf x}) = -G\int\int\int\frac{\rho({\bf x'})}{|{\bf x}-{\bf
x'}|}d^3x'
\label{eq_pot1}$$ over all space, where $G$ is the gravitational constant. Rewriting Eq. (\[eq\_pot1\]) in differential form, we obtain Poisson’s equation $$\nabla^2\Psi = 4\pi G\rho,
\label{eq_pot2}$$ with $\Psi$ satisfying the boundary condition $\Psi(\infty)= 0$ at all times. The numerical “Poisson solvers” to Eq. (\[eq\_pot2\]) can be classified into two categories: difference methods and integral methods. The difference methods solve Eq. (\[eq\_pot2\]) directly by either finite-difference or finite-element methods with necessary boundary conditions. The known boundary conditions at infinity are usually not very useful if a finite domain is considered. User specified boundary conditions or Dirichlet boundary conditions obtained by direct summation are often required. A key advantage of difference methods is that, generally, they are relatively fast once initialized. However, they have low or limited accuracy, and they often rely on the integral method to provide the boundary conditions.
The integral methods are to integrate Eq. (\[eq\_pot1\]) directly. They have the advantage that the summation stops naturally at the domain boundaries. However an integral method has two difficulties in a practical implementation. One is that the integral has a point mass singularities (i.e. when ${\bf x'}\to {\bf x}$ in Eq. (\[eq\_pot1\])), which is often circumvented by introducing softening. The other difficulty is that they are computationally prohibitive for a large system. The computational cost can be significantly lowered if the FFT method can be used.
The gravitational field, $g=-\nabla \Psi$, is more convenient in many applications. It can be calculated via numerical derivatives, which generally have poor precision. To obtain high accuracy, we can integrate the field directly by $$g({\bf x}) = G\int\int\int\frac{\rho({\bf x'})({\bf x'}-{\bf
x})}{|{\bf x}- {\bf x'}|^3}d^3x'~~,
\label{eq_f1}$$ which shares the same two difficulties as calculating Eq. (\[eq\_pot1\]).
In this paper, we present a method for computing the disk self-gravity for quasi-2D disk models. We consider a disk with the cylindrical grid $(r,\phi,z)$. We assume that the scale height of the disk (semi-thickness) is only radius-dependent, i.e., $H= H(r)$, and it is quite small, $H(r)/r \ll 1$ (namely geometrically thin disks). We also assume that the vertical structure of the density can be described by some function $Z(r,z)$ that is independent of $\phi$ and time $t$, i.e., $$\rho(t,r,\phi,z) = \Sigma(t,r,\phi)Z(r,z).$$ In this paper, we are particularly interested in the gravitational potential and field force at the $z=0$ plane. Although our method is valid for any function of $Z(r,z)$, we assume that the density vertically has a Gaussian distribution. Under this assumption, $$Z(r,z) = \frac{1}{\sqrt{2\pi
H(r)^2}}\exp\left(-\frac{z^2}{2H(r)^2}\right).
\label{eq_z}$$ With the presence of the vertical structure (\[eq\_z\]), the migration rate of the protoplanet is reduced up to 50% (see [@Koller04]). We assume that the disk has a constant sound speed $c_s$. Then from the scale height $$\frac{H}{r} = \frac{c_s}{v_K}~~,$$ where $v_K$ is the Keplerian velocity $v_K(r) = \sqrt{GM_\star/r}$, we obtain $$H(r) = \frac{c_s r^{3/2}}{\sqrt{GM_\star}}~~.$$ Other profiles for the scale height are also easy to handle. For example, if the aspect ratio, $h=H/r$, is constant, then $H(r) =
rh$. The numerical verification in §\[sec:test\] actually uses a constant $H$ in the whole domain.
Since we are interested in the potential and field in the $z=0$ plane, the easiest method to obtain $\Psi$ is to integrate directly the equation $$\Psi(t,r,\phi) = \int_{r_{\min}}^{r_{\max}}\int_0^{2\pi}\Sigma(t,r',\phi')r'dr'd\phi'
\int_{-\infty}^{\infty}-\frac{GZ(r',z')}{\sqrt{r^2+{r'}^2-
2rr'cos(\phi-\phi')+{z'}^2}} dz'
\label{2d_eq1}$$ where $r_{\min}$ and $r_{\max}$ are the inner and outer radial boundaries, and $\Sigma$ is the vertically integrated density. As in [@Hure05], we have employed two different coordinate systems in the radial direction: $r$ as the field grid and $r'$ as the source grid. For simplicity, we set $G=1$ and $M_\star = 1$.
The rest of the paper is organized as follows. In §\[sec:non-axi\], we describe the method given by [@ChPO06] on how to solve Eq. (\[2d\_eq1\]) using a direct integral via a Green’s function method. Our method, though essentially follows the one given by [@ChPO06], differs in that we use a direct summation method on a uniform grid in the radial direction, which is the same grid used in our hydro code. In addition, we present two approaches to circumvent the singularity in Eq. (\[2d\_eq1\]) and two methods to calculate the force field. For comparison, we have also implemented a 2D tree-code with a simplified 3D treatment to calculate the self-gravity for disks with vertical structures. In §\[sec:test\] we present an efficient parallel implementation scheme on distributed memory computers. We describe a new algorithm to calculate the gravity force at an arbitrary point. We also present numerical test results to compare different approaches. A few concluding remarks are given in §\[sec:conc\].
Green’s Function Method \[sec:non-axi\]
=======================================
We present here a numerical method to compute integral (\[2d\_eq1\]). For zero-thickness non-axisymmetric disk, the classical FFT method [e.g. @BiTr87] based on polar grid, which has been implemented by [@BaMa08], is often used. However, as pointed out by [@HuPi05], the FFT method has a few drawbacks. First it requires a grid with a logarithmic radial spacing, which could be inconvenient to most hydro-solvers using uniform spacing. Secondly, to avoid the well-known alias issue, the FFT method requires to double the number of cells along the radial direction. The FFT calculation is thus done on a grid that has twice the extent of the hydrodynamic grid along the radial direction, which induces some complications in the calculation of the convolution kernels, as well as many communications between both grids. Furthermore, with the presence of vertical structure, it is impossible to apply the FFT method in 3D cylindrical coordinates because neither the potential nor the gravitational field can be represented as convolution products in $z$. There is no such coordinate transformation as described in [@BiTr87] to make the integral (\[2d\_eq1\]) become a convolution product in all coordinates.
The Green’s function method given by [@ChPO06] avoids the FFT in the radial direction. Instead, a pseudo-spectral method on a scaled cosine radial grid is used to achieve the high-order accuracy. Moreover, a known, time-independent vertical structure is easy to incorporate. In the following, we describe modifications to their method so that it can be applied directly to a uniform radial grid.
Modified method given by [@ChPO06]
----------------------------------
[@ChPO06] proposed a direct integral method to solve Eq. (\[2d\_eq1\]). For the sake of completeness, we recap the key steps in their method here. First, we introduce a softening $\varepsilon$ to Eq. (\[2d\_eq1\]) and denote $$\mathcal{G}(r,r',\phi-\phi') =
\int_{-\infty}^{\infty}-\frac{Z(r',z')}{\sqrt{r^2+{r'}^2 -
2rr'cos(\phi-\phi')+\varepsilon^2+{z'}^2}}dz',
\label{intGz}$$ where $\varepsilon$ can be zero or a radius-dependent parameter that will be discussed later in Section \[sec:ts\]. Function $\mathcal{G}(r,r',\phi-\phi')$ can be computed either analytically or by numerical quadrature. For the special case of $Z(r,z)$ defined by (\[eq\_z\]), $$\mathcal{G}(r,r',\phi-\phi') = -
\frac{e^{R^2/4}K_0(R^2/4)}{\sqrt{2\pi}H(r')}$$ where $R^2 = (r^2+{r'}^2-2rr'\cos(\phi-\phi')+\varepsilon^2)/H^2(r')$, and $K_0$ denotes the modified Bessel function of the second kind. Let $$I(r,r',\phi-\phi') = 2\pi r'\mathcal{G}(r,r',\phi-\phi').
\label{eq_I}$$ Then Eq. (\[2d\_eq1\]) becomes $$\Psi(t,r,\phi) =
\int_{r_{\min}}^{r_{\max}}\frac1{2\pi}\int_0^{2\pi}\Sigma(t,r',\phi')
I(r,r',\phi-\phi')
dr'd\phi'
\label{2d_eq1_2}$$ Note that both $\Sigma$ and $I$ are periodic functions with period $2\pi$, and they are in a natural convolution representation in Eq. (\[2d\_eq1\_2\]). Applying Fourier transform to (\[2d\_eq1\_2\]) with respect to (w.r.t.) $\phi$, and using the convolution theorem, we obtain $$\hat\Psi_m(t,r) =
\int_{r_{\min}}^{r_{\max}}\hat\Sigma_m(t,r')\hat{I}_m(r,r')dr', \qquad
m\in[-\infty,+\infty],
\label{2d_eq3}$$ where $\hat{f}_m$ represents the coefficients in the Fourier series expansion of $f$, which is given as $$\hat{f}_m = \frac1{2\pi}\int_0^{2\pi}f(\phi)e^{-im\phi}d\phi~~.$$ Note that our equation (\[2d\_eq3\]) is slightly different from the equation (39) in [@ChPO06], since we have put all the relevant coefficients in the representation of $I$ in Eq. (\[eq\_I\]).
The integral over $r$ in Eq. (\[2d\_eq3\]) was evaluated using Chebyshev spectral method on a “Chebyshev-roots grid” by [@ChPO06]. However, for most hydro codes that use a uniform grid, it is desirable to use the same uniform grid for both hydro and self-gravity solutions to avoid interpolation between the hydro solver and self-gravity solver (which is inconvenient and introduces interpolation error).
We propose to integrate (\[2d\_eq3\]) directly with numerical quadrature using the available discrete values of $\hat\Sigma_m$ and $\hat{I}_m$ on a uniform grid. Note that $I(r,r',\phi-\phi')$ has an analytic expression and is not changed with time as long as the vertical structure remains the same. As proposed by [@ChPO06], we can pre-compute its Fourier transform, $\hat{I}_m(r,r')$, to speed up the algorithm. The whole algorithm can be summarized as follows:
1. For each pair $(r,r')$, we pre-compute $\hat{I}_m(r,r')$. Take the Fourier transform of $I(r,r',\phi-\phi')$, which is defined in Eq. (\[eq\_I\]), w.r.t. $\phi-\phi'$, $$\hat{I}_m(r,r') =
\mbox{FFT}(I(r,r',\phi-\phi')), \qquad m=1,2,...,N_\phi,$$ where $N_\phi$ is the number of cells in $\phi$-direction. Since $I(r,r,\phi-\phi')$ is a real and even function w.r.t. $\phi-\phi'$, only the discrete cosine transform is needed and $N_\phi/2$ modes need to be stored.
2. Take the Fourier transform of $\Sigma(r',\phi')$ w.r.t. $\phi'$ to obtain $\hat\Sigma_m(r')$, $m=1,2,...,N_\phi$.
3. Calculate Eq. (\[2d\_eq3\]) by numerical quadrature in the radial direction to obtain $\hat\Psi_m(r)$. Either the midpoint or trapezoidal rule can be used, depending on the grid point distribution of the source grid.
4. Take the inverse Fourier transform of $\hat\Psi(r)$ w.r.t. $\phi$ to obtain $\Psi(r,\phi)$.
We should emphasize that the pre-computing of $\hat{I}_m(r,r')$ plays a major role in the efficiency of our self-gravity solver. This step can be calculated once for all during a long time simulation of the disk-planet interaction system. We find that it costs at least fifty times more than the rest of steps. One cannot afford to calculate it in every time step.
The overall computational cost is $O(N_rN_\phi\log N_\phi+ N_\phi
N_r^2)$, where $N_r$ is the number of cells in the radial direction. Again, these steps are essentially the same as in [@ChPO06], except that the integration is done on a uniform polar grid using direct summation.
Treatment of singularity\[sec:ts\]
----------------------------------
When $\varepsilon=0$, the integrand in Eq. (\[intGz\]) contains a singularity. Note that the density splitting method of [@PiHu05b] cannot be used in our case because the density profile is unknown during the pre-computing of $\hat{I}_m(r,r')$.
We have tested two approaches to solve the singularity issue. One approach is to use a nonzero softening, $\varepsilon(r')$. We propose a function that scales roughly linearly with $\Delta r$, i.e., $\varepsilon(r') = \alpha(r')\Delta r$. The idea is that the mass distribution from the small spheres at each cell center should approximate a smooth mass distribution instead of a sum of $\delta$-functions. We optimize $\alpha(r')$ by minimizing the relative error in $\Psi(r,\phi)$ for sharply peaked Gaussian mass distributions located at $r'$ ranging from 0.6 to 1.8 (scaled value for the disk-planet problem) with $c_s = 0.05$. We find a piecewise linear function for $\alpha(r)$: $$\alpha(r) = \left\{\begin{array}{ll}
0.17 + \frac{0.06}{0.6}(r-r_{\min}), & r < 1.0 \\
0.23 + \frac{0.03}{0.2}(r-1.0), & 1.0\le r <1.2 \\
0.26 + \frac{0.04}{0.3}(r-1.2), & 1.2\le r <1.5 \\
0.30 + \frac{0.03}{0.2}(r-1.5), & 1.5\le r <1.7 \\
0.33 + \frac{0.04}{0.3}(r-1.7), & 1.7\le r
\end{array}\right.
\label{softn}$$ Our experiments show that $\alpha(r)$ is fairly insensitive to different resolutions in the radial direction. [@BaMa08], however, have argued that the softening must scale with $r$. This might be because the logarithmic grid in the radial direction is used in their FFT implementation. For the uniform grid along the radial direction, we find that even a constant $\alpha$, e.g., $\alpha(r)=0.23$, yields good results for the density peak near $r=1$.
We remark that our choice of the softening should not be mixed with the softening that is used for calculating the gravity from the planet. It has been pointed out by [@NeBe03a] that the softening used between the disk and planet should be at least 0.75 times the physical size of the grid zone. [@BaMa08] suggested using $\varepsilon(r) = 0.3H(r)$ for both disk self-gravity and the gravity between disk cells and planet. We find that this choice is too big and produces large error for disk self-gravity in our implementation. Therefore we use two different softenings in our simulations: for the disk self-gravity $\varepsilon =
\alpha(r)\Delta r$, and for the gravity between disk and planet $\varepsilon = 0.1H(r_p)$ ($r_p$ is the planet position in radial direction). The coefficient 0.1, which is still much smaller than the values generally used in the literature, is chosen partially due to the presence of vertical structure, and it matches very well with the [@Tanaka02] theory on the torque evaluation.
A question arises that the different softening choices may introduce inconsistency between the disk and planet. This concern can be alleviated by the fact that the gravity force for disk cells near the planet are dominated by the planet gravity. Therefore this inconsistency does not have much impact on the dynamics of the disk.
To eliminate the dependence on the softening, we have implemented the other approach, which is to use different grids for $r$ and $r'$ so that $r'_i$ is never equal to $r_j$. This approach has been used by [@ChPO06] for their spectral method. In our case, the integral of Eq. (\[2d\_eq3\]) is performed on a fixed uniform grid of $r'$. Because $r'$ is the cell-centered grid, we define $r$ as the node centered grid, i.e., $$r_0 = r'_1 - 0.5\Delta r, \qquad r_i = r'_i + 0.5\Delta r,
i=1,2,...,N.
\label{shiftG}$$ Note that the $r$ grid has one more points than the $r'$ grid. After calculating $\mathcal{G}(r,r',\phi-\phi')$ for $r$ at the node-center, we can obtain the $\mathcal{G}(r,r',\phi-\phi')$ for $r$ at the cell-center by interpolation, and then calculate the potential $\Psi(r,\phi)$ at the cell-center. We can also calculate directly the potential $\Psi(r,\phi)$ with $r$ at the node center for use of the field calculation.
Field Calculation
-----------------
We have implemented two approaches to calculate the field components. The first approach is the difference method: we calculate the potential first and then use the finite-difference method to calculate the field.
If the potential is calculated at the cell-center $(r_i,\phi_j)$, we use the central-difference to obtain the field: $$\begin{aligned}
\Psi_r(r_i,\phi_j) &=&
\frac{\Psi(r_{i+1},\phi_j)-\Psi(r_{i-1},\phi_j)}{2\Delta r} \label{fd1}\\
\Psi_\phi(r_i,\phi_j) &=&
\frac{\Psi(r_i,\phi_{j+1})-\Psi(r_i,\phi_{j-1})}{2r_i\Delta \phi} \label{fd2}~~,\end{aligned}$$ where the notation $(\cdot)_\phi = \partial(cdot)/(r\partial\phi)$. Note that we need values of $\Psi(r_0,\phi)$ and $\Psi(r_{N+1},\phi)$ for the $\Psi_r$ component at the boundary cells located at $r_1$ and $r_N$; otherwise, one-side finite-difference, which is not very accurate, should be used.
If the potential is calculated at the edge-center $(r_{i+\frac12},\phi_j)$, the field can be calculated as $$\begin{aligned}
\Psi_r(r_i,\phi_j) &=&
\frac{\Psi(r_{i+\frac12},\phi_j)-\Psi(r_{i-\frac12},\phi_j)}{\Delta
r}\label{fd3} \\
\Psi_\phi(r_i,\phi_j) &=&
\frac{\Psi(r_{i+\frac12},\phi_{j+1})+\Psi(r_{i-\frac12},\phi_{j+1})-
\Psi(r_{i+\frac12},\phi_{j-1}) - \Psi(r_{i-\frac12},\phi_{j-1})
}{4r_i\Delta \phi}~. \label{fd4}\end{aligned}$$
High order methods that use wide stencils are also possible. In fact, we can obtain the derivative $\Psi_\phi$ with only one extra FFT in the above algorithm of calculating the potential. After obtaining $\hat\Psi_m(r)$, we take the inverse transformation of $-im\hat\Psi_m(r)$ to get the $\Psi_\phi$.
The second approach for the field calculation is the integral method: we directly convolute the field Green’s function with the mass distribution. The differentiation in $\Psi_r$ and $\Psi_\phi$ can be applied directly to the Green function, which yields $$\begin{aligned}
\Psi_r(r,\phi) &=&
\int_{r_{\min}}^{r_{\max}}\int_0^{2\pi}\Sigma(t,r',\phi')r'
\mathcal{G}_r(r,r',\phi-\phi')dr'd\phi'
\label{Gr1}\\
\Psi_\phi(r,\phi) &=&
\int_{r_{\min}}^{r_{\max}}\int_0^{2\pi}\Sigma(t,r',\phi')r'
\mathcal{G}_\phi(r,r',\phi-\phi')dr'd\phi' \label{Gphi1}\end{aligned}$$
Since we have two field components, $\Psi_r$ and $\Psi_\phi$, the integral method takes twice the computation time of the difference method. But, we expect that the integral method might have higher accuracy than the finite difference approximation. However, if the mode cut-off approach, which will be described in Section \[sec:para\], is used, it is possible that the integral method may lose this advantage. In fact, we observed in our numerical tests that the integral method of (\[Gr1\]) and (\[Gphi1\]) with the mode cut-off gave larger errors and more than double the computation and communication times of the finite-difference method.
Tree-code force calculation\[sec:tc\]
-------------------------------------
Yet another approach we explored to solve for the field involved using a hierarchical tree-code. In this approach, we treat each cell as a particle, and hence an effective force law between two points, $(r,\phi)$ and $(r',\phi')$, in the disk that accounts for the vertical structure is required. Just as before, we assume $\rho({\bf x}) = \Sigma(r,\phi)Z(r,z)$. And since symmetry arguments require $Z(r,z)$ to be an even function of $z$, the z-component of the force must be zero at the $z=0$ plane. Therefore, the force is directed in the plane of the disk and it has a magnitude factor given by $$F(d) = \int_{-\infty}^{\infty}\int_{-\infty}^{\infty}
\frac{Z(r,z)Z(r',z')d}{\big(d^2 +
(z-z')^2\big)^{3/2}}dzdz'\label{eq_tc1}$$ where $d = \big(r^2+{r'}^2 - 2rr'cos(\phi-\phi')\big)^{1/2}$.
If we further assume $Z(r,z)$ is given by Eq. (\[eq\_z\]), then we can rewrite Eq. (\[eq\_tc1\]) as $$F(d;H,H') = \frac{C_{3D}(d;H,H')}{d^2} \label{eq_tc2}$$ where $$C_{3D}(d;H,H') = \frac{d^3}{2 \pi H
H'}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty} \frac{e^{-z^2/2
H^2} e^{-z'^2/2 H'^2}}{\big(d^2 + (z-z')^2\big)^{3/2}}dzdz'
\label{eq_tc3}$$ is called the 3D correction factor with respect to the 2D force without the vertical structure. As $H\to 0$, (\[eq\_tc3\]) approaches to the 3D factor defined by [@Koller04].
We note that this force law is a line-line force (giving the net interaction between the mass distributed along two vertical lines at $(r,\phi)$ and $(r',\phi')$). This is different from what is done in the Green’s function method, which uses a point-line interaction. By setting $H=0$ in Eq. (\[eq\_tc2\]), we can recover the interaction force law of the Green’s function method. However, if we wished to include line-line interactions in the Green’s function method, we would have to abandon the more efficient potential calculation of Eq. (\[intGz\]) and instead use Eqs. (\[Gr1\]) and (\[Gphi1\]) with $\mathcal{G}_r$ and $\mathcal{G}_{\phi}$ defined appropriately.
Re-scaling the integration in Eq. (\[eq\_tc3\]) and introducing the non-dimensional variables $d' = \frac{d}{\sqrt{H^2 + H'^2}}$ and $\alpha = \frac{H'}{H}$ gives $$C_{3D}(d';\alpha) = \frac{d'^3}{2
\pi}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty} \frac{e^{-z^2/2}
e^{-z'^2/2}}{\bigg(d'^2 + \frac{(z-z'\alpha)^2}{1 +
\alpha^2}\bigg)^{3/2}}dzdz' \label{eq_tc4}$$ Finally, we perform an orthogonal coordinate transformation using the variables $$\begin{aligned}
\zeta = \frac{1}{\sqrt{1+\alpha^2}}(z - z'\alpha) \nonumber
\\
\zeta' = \frac{1}{\sqrt{1+\alpha^2}}(\alpha z + z'). \nonumber\end{aligned}$$ This allows us to simplify Eq. (\[eq\_tc4\]) to $$C_{3D}(d') = \frac{d'^3}{2
\pi}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}
\frac{e^{-\zeta^2/2} e^{-\zeta'^2/2}}{\big(d'^2 +
\zeta^2\big)^{3/2}}d\zeta d\zeta' \label{eq_tc5}$$ revealing that it is actually independent of $\alpha$. In fact, similar to Eq. (\[intGz\]) we can express Eq. (\[eq\_tc5\]) as $$C_{3D}(d') = -\frac{d'^3}{2 \sqrt{2 \pi}}e^{d'^2/4}\bigg(K_0(d'^2/4)
+ K'_0(d'^2/4)\bigg) \label{eq_tc6}$$ where $K_0$ and $K'_0$ denote the modified Bessel function of the second kind and its derivative.
The tree-code requires the calculation of $C_{3D}(d')$ and its first and second derivatives over a wide range of $d'$ (ranging from $10^{-2}$ to $10^3$). Calculating Eq. (\[eq\_tc6\]) on the fly is prohibitively expensive and should be avoided. Another option is to pre-compute Eq. (\[eq\_tc6\]) and its derivatives at a discrete set of $d'$ values. Then when $C_{3D}(d')$ is needed, we interpolate the pre-computed data. This approach can provide a significant speedup, but because the data is stored in main memory it is still too slow for our purposes.
![\[fig\_tc1\] 3D correction factor (Eq. (\[eq\_tc6\])) and relative error plotted vs. non-dimensional distance ($d'$) using model Eq. (\[eq\_tc7\]).](f1a.eps "fig:"){width="45.00000%"} ![\[fig\_tc1\] 3D correction factor (Eq. (\[eq\_tc6\])) and relative error plotted vs. non-dimensional distance ($d'$) using model Eq. (\[eq\_tc7\]).](f1b.eps "fig:"){width="45.00000%"}
![\[fig\_tc2\] 3D Force factor (Eq. (\[eq\_tc2\])) and relative error plotted vs. non-dimensional distance ($d'$) using model Eq. (\[eq\_tc7\]).](f2a.eps "fig:"){width="45.00000%"} ![\[fig\_tc2\] 3D Force factor (Eq. (\[eq\_tc2\])) and relative error plotted vs. non-dimensional distance ($d'$) using model Eq. (\[eq\_tc7\]).](f2b.eps "fig:"){width="45.00000%"}
Another approach we explored involves approximating Eq. (\[eq\_tc6\]) with a model function that is accurate yet inexpensive enough to compute on the fly. Plotting $C_{3D}(d')$ reveals its form (Fig. \[fig\_tc1\]) and suggests a model function of the form $$C_{3D}^{(\mathrm{model})}(d') = \frac{(d'/d'_{1/2})^p +
(d'/d'_{1/2})^{2p}}{2 + (d'/d'_{1/2})^p + (d'/d'_{1/2})^{2p}}.
\label{eq_tc7}$$ We find optimal parameter values to be $d'_{1/2} = 0.8252$ and $p =
0.957$. The relative error introduced by using this model is plotted with $C_{3D}$ and $F$ in Fig. \[fig\_tc1\] and \[fig\_tc2\]. Since generally $d' = d/H(r) > \varepsilon/H(r)$ with $H'=0$, the model is acceptable for distance or softening larger than $0.005H(r)$. This model requires the calculation of one power function, a few additions and multiplications and one division. For our application, we found this to be the most efficient way to include 3D structure using a tree-code.
Our tree-code is implemented in a distributed memory parallel architecture using MPI, just like our Green’s function FFT algorithm. There are many papers detailing parallel tree-codes [@Dubin96; @MiDo02]. So we simply highlight the important points of our algorithm here.
1. We begin by pre-computing the tree structure of our grid, which does not change with time. Each processor retains a copy of this [*quad-tree*]{} to be used throughout the simulation.
2. All processors compute partial sums (using local particles) of the moments for all nodes in the tree.
3. These partial sums are added and broadcast back to all the processors.
4. A complete copy of the density profile is distributed to all the processors. This is required for load balancing and computing the direct summation part of the force (step 5(c)).
5. Forces are computed at every grid cell. The disk is partitioned azimuthally (the symmetry providing ideal load balancing), and processors are assigned equal-sized sectors of particles (grid cells). Each particle traverses the [*quad-tree*]{} to include interactions with all other particles.
1. If a particle-node interaction meets the accuracy criterion, then the particle interacts with that node via the moments computed in step 3.
2. If the accuracy criterion is not met (the node is too large or the particle-node separation too small), then we drop one level in the tree and check all the rejected node’s sub-nodes for particle-node interactions.
3. Last, if we reach a leaf-node (a node with no sub-nodes) and the accuracy criterion is not met we perform direct summation over all the leaf-node’s particles.
Numerical Implementation and Experiments\[sec:test\]
====================================================
Description of the test problem
-------------------------------
As in [@ChPO06], we consider the density function to be a Gaussian sphere given by $$\rho(r,\phi,z) =
\frac{M}{(2\pi\sigma^2)^{3/2}}\exp\left(-\frac{r^2+z^2}{2\sigma^2}\right),$$ where $M$ is the normalized total mass, $\sigma$ is a parameter that controls the width of the mass distribution and it has similar role as the scale height in Eq. (\[eq\_z\]). The potential on the $z=0$ plane has been given by [@ChPO06] $$\psi(r,\phi) = -\frac1r\mbox{erf}\left(\frac{r}{\sqrt{2}\sigma}\right)~,
\label{eq_psi}$$ where $\mbox{erf}(x)$ is the error function $$\mbox{erf}(x) = \frac{2}{\sqrt{\pi}}\int_0^xe^{-t^2}dt ~~.$$ For a collection of Gaussian spheres centered on the $z=0$ plane with the same $\sigma$, the $z$-dependent vertical structure can be factored out, i.e., $$\rho(r,\phi,z) = \sum_i\rho_{(r_i,\phi_i)}(r,\phi,z) =
Z(r,z)\sum_i\Sigma_{(r_i,\phi_i)}(r,\phi),$$ where $(r_i,\phi_i)$ is the center of each sphere, $Z(r,z)$ gives the vertical structure $$Z(r,z) = \frac{1}{\sqrt{2\pi
\sigma^2}}\exp\left(-\frac{z^2}{2\sigma^2}\right),
\label{eq_z2}$$ $\Sigma_{(r_i,\phi_i)}$ is the surface density $$\Sigma_{(r_i,\phi_i)} =
\frac1{2\pi\sigma^2}\exp\left(-\frac{R_i^2}{2\sigma^2}\right),$$ and $R_i(r,\phi) = \sqrt{r^2 +{r_i}^2 - 2rr'\cos(\phi-\phi_i)}$ is the distance between $(r,\phi)$ and $(r_i,\phi_i)$. Notice that Eq. (\[eq\_z2\]) is the same as Eq. (\[eq\_z\]) with a constant scale height $H(r)=\sigma$.
We use three Gaussian spheres located at $(r,\phi)=(1,0)$, $(0.9,3\pi/4)$, and $(1,-\pi/2)$ with a total mass of 2,1/2, and 1 respectively, i.e., $$\Sigma_{\mbox{ana}}(r,\phi) = 2\Sigma_{(1,0)}(r,\phi) +
\frac12\Sigma_{(0.9,3\pi/4)}(r,\phi) + \Sigma_{(1,-\pi/2)}(r,\phi),
\label{den_exact}$$ where $(*)_{\mbox{ana}}$ represents the analytic solution. The exact potential will be $$\Psi_{\mbox{ana}}(r,\phi) = 2\psi(R_{(1,0)}) + \frac12\psi(R_{(0.9,3\pi/4)}) +
\psi(R_{(1,-\pi/2)})
\label{pot_exact}$$ where $\psi$ is defined by Eq. (\[eq\_psi\]). The exact field $\Psi_r$ and $\Psi_\phi$ can also be calculated accordingly.
The tests are done in the computational domain $[0.4,2.0]\times[0,2\pi]$ with grid $N_r\times N_\phi$. The normalized total mass for the disk is $M=0.002M_\star$. We choose $N_\phi=4N_r$ so that the cells near $r=1$ are nearly square. Without specification, we always use $N_r=800$, which reaches the convergence in both the torque calculation on planet and the azimuthal averaged potential vorticity distribution on the disk in our disk-planet interaction simulations [@lietal05]. Since the surface density (\[den\_exact\]) approaches to zero outside the computational domain, the exact potential (\[pot\_exact\]) is still valid for our truncated domain. For convenience, we use the following notations for comparison: $N_p$ denotes the number of processors, $E_{\max}$ denotes the maximum error over the whole domain, $RE_{\max}$ denotes the maximum relative error, and $p$ is the convergence order in $E_{\max}$. Since the force is a vector that has two component, we use the following formula to calculate $E_{\max}$ $$E_{\max} = \sqrt{|f^r - f^r_{\mbox{ana}}|^2 + |f^\phi - f^\phi_{\mbox{ana}}|^2}$$ where $f^r$ and $f^\phi$ are the force components in $r$- and $\phi$-direction respectively. The local relative error for a variable $u$ is defined as $RE_{\max}(u) =
{E_{\max}}/{|u_{\mbox{ana}}|}$. For the force, $|u_{\mbox{ana}}|=\sqrt{(f^r_{\mbox{ana}})^2 + (f^\phi_{\mbox{ana}})^2}$. The global relative error for a variable $u$ is defined as $$RE(u) =
\frac{\sum_{j=1}^{N_\phi}\sum_{i=i}^{N_r}|u_{i,j}-u_{i,j,\mbox{ana}}|}
{\sum_{j=1}^{N_\phi}\sum_{i=i}^{N_r}|u_{i,j,\mbox{ana}}|}
\label{eq_err_re}$$
All of the computation are performed on a parallel Linux cluster at the Los Alamos National Lab. Each node of the cluster is dual-core AMD Opteron(tm) processor with 2.8G HZ and 2GB local memory.
Parallel implementation and comparison\[sec:para\]
--------------------------------------------------
Our hydro simulation for the interaction of disk and proto-planet problem is performed on a high resolution grid, e.g., $N_r\times
N_\phi=800\times3200$ grid. The whole domain is split into annular regions for parallel computation. Each annular region has the same number of cells in the radial direction.
The Fourier transform and its inverse can be parallelized without much difficulty because they require no communication between different processors. However, the numerical quadrature of Eq. (\[2d\_eq3\]) is done in the radial direction, where each processor holds only part of the information about the density distribution in the whole domain.
We have tested two approaches to parallelize the numerical quadrature. In the first approach, we start by computing a partial quadrature including only source terms from the local density distribution, i.e. each processor computes a partial sum of Eq. (\[2d\_eq3\]) for all $r$ and $m$. We then apply a global communication to obtain the complete summation.
In the second approach, we begin by performing a global communication over the whole domain so that every processor has a complete copy of the density profile for $r'\in[r_{\min},r_{\max}]$. Then the complete numerical quadrature (\[2d\_eq3\]) can be performed to obtain $\hat\Psi_m(r)$ for the local portion of $r$ grid.
We remark that in the pre-computing stage, no matter which approach is used, each processor can compute its own portion of the Fourier transform $\hat{I}_m(r,r')$ independently and store it for the later use. It does not need to have a copy of whole profile for all $r$ and $r'$.
In the following we will first propose a strategy to reduce the parallel communication cost, and then compare the above two approaches in calculating potential $\Psi(r,\phi)$. For simplicity, we denote the first approach as approach I, and the second as approach II.
### Mode cut-off approach to reduce the communication cost
The communication cost in both approaches is proportional to both the number of processors and the amount of data being communicated (total grid size). Consequently, we expect that the communication cost will dominate the computation cost as the number of processors is increased, eventually becoming a bottleneck for a large number of processors. Note that the communication is done in the frequency domain, where the high modes decay exponentially for a smooth function. Thus we can truncate the Fourier modes above a cutoff parameter ($M_{cut}$) without a significant loss of accuracy. If $M_{cut}$ is far smaller than $N_\phi$ (the total number of mode), the communicate cost can be reduced to a small of fraction of the cost before the cutoff.
For both approaches, we can calculate $M_{cut}$ based on the Fourier transformation of the density profile. For a specific radius $r_i$, we define the energy contained in $\hat\Sigma(r_i)$ as $\sum_{m=0}^{N_\phi}|\hat\Sigma_m(r_i)|^2$. Each processor then computes an $m_{cut}(r)$ based on satisfying some preset fraction $\epsilon_{cut}$ (e.g., $\epsilon_{cut}=10^{-4}$) of the total energy for modes above $m_{cut}(r)$. Next, each processor finds its maximum $m_{cut}$ over its range of $r$. Last, all the processors compare their $m_{cut}$ values to obtain the global maximum $M_{cut}$ that is used to truncate the Fourier series. Note that $M_{cut}$ calculated in this way does not vary with the number of processors.
For approach I, the $m_{cut}(r)$ can also be evaluated based on $\hat\Psi(r)$ instead of $\hat\Sigma(r)$, because the communication is done after the partial summation. To be more accurate in the field calculation, we instead can evaluate $m_{cut}(r)$ based on the energy defined by $\sum_{m=0}^{N_\phi}|m\hat\Psi_m(r_i)|^2$. The extra factor of $m$ is included so that we obtain the energy of the force, which is the derivative of $\Psi$. Notice, however, that if we compute $m_{cut}$ and $M_{cut}$ as described above, $M_{cut}$ will vary with the number of the processors. This is because each processor computes only a partial sum of $\hat\Psi_m(r)$. To produce a relatively constant $M_{cut}$, after each processor finds its maximum $m_{cut}$, all the processors take an energy-weighted average of their $m_{cut}$ to obtain the global $M_{cut}$. It is observed that the energy-weighting average gives an $M_{cut}$ that remains nearly constant while the number of processors varies between 1 to $N_p$. For verification reason, we can require that $M_{cut}$ be independent of the number of processors. However, that requires a costly global communication between different processors. Nonetheless, we find that $M_{cut}$ will only increase slightly with the number of processors, which means that increasing the number of processors will not degrade the numerical accuracy. We define a similar preset fraction $\epsilon_{cut}$ to the previous density mode cut-off approach. After extensive experiments, we observe that $\epsilon_{cut}\approx 10^{-7}$ in $\hat\Psi(r)$ cut-off produces similar results to $\epsilon_{cut} = 10^{-4}$ in $\hat\Sigma(r)$ cut-off.
For density distributions in disk-planet simulations, the energy of the zero mode ($m=0$) often dominates that of all other modes combined. Thus, when calculating the energy using $\hat\Sigma$ or $\hat\Psi$ , it is convenient to exclude the zero mode from the total energy. Notice that if the energy is defined as $|m\hat\Psi_m(r)|^2$, the zero-mode is naturally excluded. To remove the numerical noise, we multiply the energy of the zero mode by a small number (e.g., $\epsilon$) and add it to the total energy. In our disk-planet simulations, the initial density distribution is axisymmetric, and hence zero mode is enough. With the time evolution of the surface density, $M_{cut}$ begins to increase quickly until it reaches a certain number, which is insensitive to different resolutions and different initial power-law density profiles. However $M_{cut}$ does vary a lot with the planet mass and sound speed. For an example, we obtain the following data for simulations of an isothermal disk with different planet mass ($M_p$) and sound speed ($c_s$), $$M_{cut}=\left\{\begin{array}{cl}
190, & \mbox{ for } M_p=3\times10^{-5}M_\star, c_s = 0.05 \\
297, & \mbox{ for } M_p=3\times10^{-4}M_\star, c_s = 0.05 \\
298, & \mbox{ for } M_p=3\times10^{-5}M_\star, c_s = 0.035
\end{array}\right.$$ Note that $M_{cut}$ is much smaller than the total number of modes in $\phi$-direction, $N_\phi=3200$.
### Impact of the cutoff modes and cut-off thresholds\[sec:test\_cut-off\]
Here, we study how the number of cut-off modes, $M_{cut}$, impacts the computational efficiency and accuracy. We test the problem with parallel computation using 100 processors. We will show the results only for approach II. Similar results have also been obtained for approach I.
Fig. \[cpu1\] shows the variation in the computation and communication time with different numbers of cut-off modes. The total cost displayed in term of CPU time includes the computation of one FFT of the density field, the global communication of the density spectrum up-to the cut-off mode $M_{cut}$ right after the FFT, a quadrature calculation in the radial direction. Fig. \[cpu1\] shows that the communication time increases at a slower rate with $M_{cut}$ than the computation time .
![\[cpu1\]The computation and communication cost for evaluating $\hat\Psi(r,\phi)$ (left) and the error variation (right) as a function of $M_{cut}$. ](f3a.eps "fig:"){width="45.00000%"} ![\[cpu1\]The computation and communication cost for evaluating $\hat\Psi(r,\phi)$ (left) and the error variation (right) as a function of $M_{cut}$. ](f3b.eps "fig:"){width="45.00000%"}
![\[cpu2\]The computation and communication cost for evaluating $\hat\Psi(r,\phi)$ (left) and the error variation (right) as a function of energy cut-off thresholds $\epsilon_{cut}$. The cut-off is based on the energy of $\hat\Sigma$. ](f4a.eps "fig:"){width="45.00000%"} ![\[cpu2\]The computation and communication cost for evaluating $\hat\Psi(r,\phi)$ (left) and the error variation (right) as a function of energy cut-off thresholds $\epsilon_{cut}$. The cut-off is based on the energy of $\hat\Sigma$. ](f4b.eps "fig:"){width="45.00000%"}
The energy cut-off approach depends on the cut-off threshold $\epsilon_{cut}$. We also vary the size of $\epsilon_{cut}$ to see how it impacts the cut-off mode. Fig. \[cpu2\] shows the simulation results for different $\epsilon_{cut}$. Figs \[cpu1\] and \[cpu2\] show that our cut-off threshold, $\epsilon_{cut}=10^{-4}$, which corresponds to $M_{cut}=153$, appears to be the optimal number to balance accuracy and efficiency for this problem. This energy cut-off threshold is insensitive to grids with different resolutions.
### Parallelization via domain decomposition of the source grid: approach I
The approach I is implemented via domain decomposition of the source grid $r'$. In terms of the communication cost, it involves a global reduction of $N_rM_{cut}$ data from every processor and a distribution of $N_rM_{cut}/N_p$ data to each processor, where $N_p$ is the number of processors. Here we use a fixed mode cut-off number $M_{cut}=153$.
Fig. \[cpu3\] shows how the communication and computation cost vary with the number processors. Note that the communication cost remains relatively constant no matter how many number of the processors we use. This is a surprise, because we expect that the communication cost is proportional to the length of the communicated data, which is proportional to the number of processors. We remark that one might obtain a different performance for a different MPI implementation or a different parallel cluster. In year 2007, we observed using the same code that the communication cost increases linearly with the number of processors. In year 2008, our parallel cluster has been upgraded with a new parallel software and a new inter-connection. The data shown in Fig. \[cpu3\] is calculated on the new cluster.
![\[cpu3\]The communication and total cost for evaluating $\hat\Psi(r,\phi)$ varying with the number of processors in MPI. $M_{cut} = 153$. ](f5.eps){width="80.00000%"}
### Parallelization via domain decomposition of the field grid: approach II
The approach II described above is implemented via domain decomposition of the field grid $r$. It involves a global gathering of $N_rM_{cut}/N_p$ data from every processor and a broadcast of $N_rM_{cut}$ data to each processor. While the amount of communicated data is the same for both approaches, approach I also involves a global summation operation resulting in a total communication cost of about 10% more than approach II with the same value $M_{cut}$. Again we show some results using only a fixed cut-off mode number $M_{cut} = 153$.
Fig.\[cpu3\] (approach II) shows the performance comparison with the first approach. Similar to the results of approach I described in the last subsection, the communication cost remains nearly constant. It is clear that approach II (this approach) is faster than the approach I. Therefore we will use approach II as our choice of the methods in the tests hereafter.
Fig. \[cpu4\] shows the parallel efficiency for different numbers of processors ($N_p=2^n,0\le n\le7$) and grids with different resolutions. The parallel efficiency is defined by $$E(N_p) = \frac{T_{seq}(1)}{N_pT(N_p)}$$ where $T(N_p)$ is the run time of the parallel algorithm, and $T_{seq}(1)$ is the runtime of the sequential algorithm using one processor. For the fine resolution grid $1024\times4096$, we replace $T_{seq}(1)$ with $4T(4)$ due to the memory limitation of the processors. It is clear that for a smaller number of processors and a larger grid, the parallel efficiency is better. Fig. \[cpu4\] shows that the parallel efficiency can be larger than 1 at certain stages. The reason is not clear to us.
![\[cpu4\]The parallel efficiency for grids with different resolutions and different number of processors. $M_{cut} = 153$ is used. ](f6.eps){width="80.00000%"}
Comparison of different approaches for potential calculation\[sec:test1\]
-------------------------------------------------------------------------
We have tested two approaches for computing the potential and field: one is with the softening Eq. (\[softn\]) and the other is without softening but with a shifted grid defined by Eq. (\[shiftG\]). Note that the field is calculated differently using different central-difference methods. The field force with softening is calculated using central-difference Eqs. (\[fd1\]) and (\[fd2\]). The field force without softening is calculated using central-difference Eqs. (\[fd3\]) and (\[fd4\]).
To minimize the impact of the cut-off mode, we use a fixed cut-off number $M_{cut}=153$, which produces nearly the same results as without cut-off (see §\[sec:test\_cut-off\]). Table \[tab1\] shows the numerical errors for both approaches. The convergence order is calculated based on the maximum absolute error. The results show nearly second-order convergence as the mesh is refined. This is in agreement with the accuracy of our method, because both the quadrature rule to calculate the potential and the finite-difference method to calculate the force are of second-order accuracy.
Based on global relative errors ($RE$ in Table \[tab1\]) for both potential and field, we see that the potential method using the shifted grid without softening is more accurate than with softening. Yet the difference is relatively small.
------- ------------ ---------------- ------ ------------- ------------- ------
Grid $E_{\max}$ $RE$ $p$ $E_{\max}$ $RE$ $p$
$N_r$ ($10^{-2}$)(1) ($10^{-2}$) ($10^{-5}$)
128 2.6514 3.1388 3.2631 7.2648
256 0.7922 0.9205 1.74 1.7595 3.0652 0.89
512 0.2317 0.2983 1.77 0.6925 1.4737 1.35
1024 0.0725 0.1298 1.68 0.2485 0.9461 1.49
128 3.3427 2.4634 5.5060 5.1819
256 0.8539 0.6580 1.97 1.4317 1.8465 1.94
512 0.2106 0.2105 2.02 0.3775 0.9995 1.92
1024 0.0592 0.1036 1.83 0.1115 0.7909 1.76
------- ------------ ---------------- ------ ------------- ------------- ------
: \[tab1\] Numerical results for the potential calculation with and without softening Eq. (\[softn\]). The top half is for with softening and the bottom half is for shifted grid without softening.
Tests of point force calculation
--------------------------------
In the previous subsection, we have tested different approaches for disk self-gravity computed at the cell centers. In this subsection, we describe and compare methods to calculate the gravitational force of the disk on the planet at any point $(r_p,\phi_p)$. We assume a unit planet mass at that point.
We investigated the force calculations in three different ways. In the first approach, we performed a direct summation of $\mathcal{G}_r$ and $\mathcal{G}_\phi$ from the point $(r_p,\phi_p)$ to each of the cell centers \[Eqs. (\[Gr1\]) and (\[Gphi1\]) with $(r,\phi) =
(r_p,\phi_p)$\]. This is similar to a particle method and is widely used in calculating the disk force exerted on the planet in the disk-planet simulations. In the second approach, we applied bilinear interpolation to the force computed at the cell centers. In the third approach, we applied a finite difference formula directly to $\Psi$ at the grid cells closest to the point location. This approach is very similar to bilinear interpolation of the force, only it uses a smaller grid stencil, which gives better accuracy.
To calculate the force at a specific point and account for the fact that the point can be at an arbitrary location within a cell, we take a cell that contains point $(r_p,\phi_p)=(0.99,0.0)$, which is close to the center of one of the first Gaussian sphere, and compute the force for $11\times11$ equidistributed points within the cell. Since the third approach is similar to the second approach and verified to be more accurate, we only compare the first and third approaches.
We use the softening defined by (\[softn\]) in the direct summation approach to avoid singularities. Fig. \[pforce1\] shows the absolute error and relative error throughout the grid cell. We find that the maximum error and maximum relative error are 6.12 and 16.6%, respectively. The top two plots of Fig. \[pforce1\] show that, although softening (\[softn\]) works very well for the disk self-gravity calculated at the cell-center and point force at the node- and edge-centers, it gives large error for the point force at other locations. The relative error range is \[0.0152%,16.6%\], which means the direct summation with this softening is very sensitive to the location of the point. We have tried two other different softenings. The middle two plots of Fig. \[pforce1\] show the results with softening $\varepsilon = \min(\Delta r,r\Delta\phi)$, which approximates one grid spacing. Both the maximum error and maximum relative error are much reduced. Also the range of the relative error becomes \[0.63%,0.88%\], which means the direction summation with this softening is relatively insensitive to the location of the point. The bottom two plots of Fig. \[pforce1\] show the results with softening $\varepsilon = 0.3H(r)$, which has been adopted by [@BaMa08] and corresponds to seven grid spacings in our $800\times3200$ grid layout. Both the maximum error and relative error become large everywhere. The error range is \[6.04,7.40\], and the relative error range is \[16.5%,16.6%\].
![\[pforce1\]The accuracy of point force calculation by direct summation approach with different softenings: softening (\[softn\]) (top two), $\varepsilon = \min(\Delta r,r\Delta\phi)$ (middle two), and $\varepsilon = 0.3H(r)$ (bottom two). The left plots show the magnitude of the errors and the right plots show the relative errors, both throughout one grid cell. ](f7a.eps "fig:"){width="45.00000%"} ![\[pforce1\]The accuracy of point force calculation by direct summation approach with different softenings: softening (\[softn\]) (top two), $\varepsilon = \min(\Delta r,r\Delta\phi)$ (middle two), and $\varepsilon = 0.3H(r)$ (bottom two). The left plots show the magnitude of the errors and the right plots show the relative errors, both throughout one grid cell. ](f7b.eps "fig:"){width="45.00000%"} ![\[pforce1\]The accuracy of point force calculation by direct summation approach with different softenings: softening (\[softn\]) (top two), $\varepsilon = \min(\Delta r,r\Delta\phi)$ (middle two), and $\varepsilon = 0.3H(r)$ (bottom two). The left plots show the magnitude of the errors and the right plots show the relative errors, both throughout one grid cell. ](f7c.eps "fig:"){width="45.00000%"} ![\[pforce1\]The accuracy of point force calculation by direct summation approach with different softenings: softening (\[softn\]) (top two), $\varepsilon = \min(\Delta r,r\Delta\phi)$ (middle two), and $\varepsilon = 0.3H(r)$ (bottom two). The left plots show the magnitude of the errors and the right plots show the relative errors, both throughout one grid cell. ](f7d.eps "fig:"){width="45.00000%"} ![\[pforce1\]The accuracy of point force calculation by direct summation approach with different softenings: softening (\[softn\]) (top two), $\varepsilon = \min(\Delta r,r\Delta\phi)$ (middle two), and $\varepsilon = 0.3H(r)$ (bottom two). The left plots show the magnitude of the errors and the right plots show the relative errors, both throughout one grid cell. ](f7e.eps "fig:"){width="45.00000%"} ![\[pforce1\]The accuracy of point force calculation by direct summation approach with different softenings: softening (\[softn\]) (top two), $\varepsilon = \min(\Delta r,r\Delta\phi)$ (middle two), and $\varepsilon = 0.3H(r)$ (bottom two). The left plots show the magnitude of the errors and the right plots show the relative errors, both throughout one grid cell. ](f7f.eps "fig:"){width="45.00000%"}
The third approach using finite difference on $\Psi$ achieves much high accuracy, with maximum error and maximum relative error of 0.0315 and 0.078%. Fig. \[pforce2\] shows the absolute error and relative error throughout the grid cell. The relative error range is \[0.0264%,0.0778%\], which means that this approach is relatively insensitive to the location of the point.
![\[pforce2\]The accuracy of point force calculation finite-difference of the potential. The left plot show the magnitude of the errors and the right plot show the relative errors, both throughout one grid cell. ](f8a.eps "fig:"){width="45.00000%"} ![\[pforce2\]The accuracy of point force calculation finite-difference of the potential. The left plot show the magnitude of the errors and the right plot show the relative errors, both throughout one grid cell. ](f8b.eps "fig:"){width="45.00000%"}
We remark that in the actual disk-planet interaction simulations, the accuracy of the force on the planet also depends on how accurately the disk density around the planet is resolved. Since the disk within one Roche lobe is usually not well resolved, the third approach in the point force calculation, though by itself is very accurate, may not always give the most “accurate” force calculation if the disk density is poorly determined. In fact, we find that the direct summation approach with a relatively large softening, e.g. $\varepsilon=0.1H(r)$, gives results that are in good agreement with the linear theory results by [@Tanaka02].
Tests of the tree-code force calculation
----------------------------------------
Here, we test the performance of our tree-code method described in Section \[sec:tc\]. First, we test the accuracy with the same error and convergence measures used in section \[sec:test1\]. For simplicity, we use a fixed softening $\varepsilon = \alpha(1)\Delta r
= 0.23\Delta r$ throughout the grid. The approximate model (\[eq\_tc7\]) is used to calculate 3D correction factor. To obtain the field in the $z=0$ plane, we set $H=0$ in Eq. (\[eq\_tc3\]) resulting in an interaction force consistent with the Green’s function method. Also, we use the same density function with known analytical force field described in Section \[sec:test\]. Second, we test the parallel efficiency by varying the number of processors for a highly resolved grid (800x3200).
Grid $F_{max}$ $E_{\max}$ $RE$($10^{-2}$) $p$ total CPU Time (s)
---------------- ----------- ------------ ----------------- ------ -------------------- -- --
100 x 400 171.68 5.617 1.5216 0.1078
200 x 800 171.72 2.324 0.8452 1.27 0.6844
400 x 1600 171.73 1.390 0.6291 0.74 3.7167
800 x 3200 171.74 1.114 0.5463 0.32 16.543
800 x 3200$^*$ 171.74 0.08912 0.0126 0.02093
: \[tc\_fixed\_NP\] Accuracy and efficiency data for tree-code force calculation on $N_P$=100 processors.
[($^*$The last line is for Green function method, where the CPU time does not include the pre-computing time, which takes 1.001 second.)]{}
------- ---------------- ---------------- ---------------------
$N_p$ total time (s) comp. time (s) Parallel efficiency
10 182.32 180.97 0.99
20 92.78 91.83 0.98
40 46.92 46.06 0.96
50 37.71 36.85 0.96
80 24.07 23.05 0.94
100 20.07 18.94 0.90
160 13.80 12.25 0.82
200 11.28 9.49 0.80
------- ---------------- ---------------- ---------------------
: \[tc\_fixed\_grid\] Parallel efficiency data for tree-code force calculation on 800x3200 fixed grid.
From the results given in Table \[tc\_fixed\_NP\] we draw the following conclusions. First, this method does not appear to converge to the exact solution. This is actually expected since the model force function (Eq. (\[eq\_tc7\])) with our chosen softening does not become more accurate as the grid is refined. We observe that the maximum error asymptotically approaches to $1.0$, which is much larger than the error using the Green’s function method for highly resolved grids. Next, we see that the time complexity scales approximately with $O(N_rN_\phi \log(N_rN_\phi))$, which is consistent with the theoretical prediction. Finally, we should remark that the softening has a large impact on the accuracy of the solutions for the tree-code. If $\varepsilon = 0.3H(r)$ is used, the maximum error and global relative error become 20.57 and 3.045% respectively.
Parallel efficiency results can be seen in Table \[tc\_fixed\_grid\]. Although we find our tree-code scales well with the number of processors, the actual value of the CPU time is more than two orders of magnitude greater than our Green’s function method! We have experimented with tuning tree code parameters and other memory optimizations and have found that while it may be possible to gain a factor of $2-4$ in speedup over the results of Table \[tc\_fixed\_grid\], our tree-code method always performs [*much*]{} slower than our Green’s function method.
Conclusion\[sec:conc\]
======================
In this paper, we have presented a fast and accurate solver to calculate the potential and self-gravity forces for the disk systems. This method is implemented on a polar grid, and FFT can be used in the azimuthal direction. The pre-calculation of the Green’s function and its FFT play a major role in the algorithm to reduce the computational cost. We think it could be the main reason why the Green’s function method is much faster than the particle tree-code method. We also presented an efficient method in implementing the solver on parallel computers. We find the computational cost for the self-gravity solver to be comparable to that of the hydro solver for large, highly resolved grids run on a distributed memory parallel architecture. We also developed a 2D tree-code solver, which uses a relatively inexpensive model force to accurately account for the vertical structure of the disk.
Finally, we notice that if the disk vertical structure varies with time, our pre-calculation must be done every time step making our solver inefficient.
We have applied our self-gravity solver to simulations of disk-planet interaction system. Compared with the simulations without self-gravity, the total computation time is increased by only 30% for a parallel computation with 100 processors. We have also confirmed that the 2D self-gravity indeed accelerates the planet migration. These results will be reported elsewhere.
[**Acknowledgment:**]{} We would like to thank Dr. C.-K. Chan for helpful discussion during his stay at Los Alamos. We also thanks the referee for many useful comments. This research was performed under the auspices of the Department of Energy. It was supported by the Laboratory Directed Research and Development (LDRD) Program at Los Alamos. It is also available as Los Alamos National Laboratory Report, .
[19]{} natexlab\#1[\#1]{}
Baruteau, C., & Masset, F. 2008, ApJ, 678, 483
Binney, J., & Tremaine, S. 1987, Galactic Dynamics (Princeton, NJ: Princeton Univ. Press)
Chan, C. K., Psaltis, D., & Özel, F. 2006, ApJ, 645, 506
Dubinski, J. 1996, New Astronomy, 1, 133
Goldreich, P., & Tremaine, S. 1979, ApJ, 233, 857
—. 1980, ApJ, 241, 425
Huré, J. M. 2005, A&A, 434, 17
Huré, J. M., & Pierens, A. 2005, ApJ, 624, 289
Koller, J. 2004, PhD thesis, Rice University, Houston, TX
Li, H., Li, S., Wendroff, B., Koller, J., & Liska, R. 2005, ApJ, 624, 1003
Lin, D. N. C., & Papaloizou, J. 1986, ApJ, 307, 395
—. 1986, ApJ, 309, 846
Miocchi, P., & Capuzzo-Dolcetta, R. 2002, A&A, 382, 758
Nelson, A. F., & Benz, W. 2003, ApJ, 589, 556
—. 2003, ApJ, 589, 578
Pierens, A., & Huré, J. M. 2005, A&A, 433, L37
—. 2005, A&A, 434, 1
Tanaka, H. Takeuchi, T., & Ward, W. R. 2002, ApJ, 565, 1257
Ward, W. R. 1997, ApJ Letter, 482, L211
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Mattia Lupetti$^a$, Julia Hengster$^b$, Thorsten Uphues$^b$, and Armin Scrinzi$^a$'
title: Supplementary Material
---
Parameterization of the plasmon
===============================
The variance of the photoscopic spectrogram according to our theory is given by: $$\label{eq:var}
S(\tau) = \Delta p^2 + \frac{1}{\pi}\int_{-\infty}^{\infty} dx_i\,
g_\text{x}(x_i)\,\mathbf{a}_y^2(x_i, \tau).$$ In order to compute the spatial integral in Eq. (\[eq:var\]), we consider a case where the bright and dark modes can both be excited at frequencies $\omega_b$ and $\omega_d$, respectively. “Bright” and “dark” refer to the coupling properties of the modes: the bright mode couples efficiently with incident radiation, the dark poorly. The two frequencies are well separated. Each mode consists of two counter-propagating plasmons. In addition, we admit a term describing the ringing of localized modes excited in the focus of the NIR pulse (see, e.g., \[1\]). The contributions in each mode $m=b,d$ are $$P_m^{(\pm)}=e^{i\varphi_{\pm m}}e^{-\frac{\varphi_{\pm m}^{2}}{2\omega_m^{2}T_m^{2}}}$$ with the phase of the propagating plasmon $$\varphi_{\pm m} = \pm k_m x - \omega (t-t_m)$$ and $$P_m^{(0)}= \cos(\omega_m (t-t_m))\,e^{-x^2/2w_\text{nir}^2},$$ for the localized excitation.
Buildup and decay are assumed to obey a simple rate equation where a Gaussian-shaped buildup of width $\sigma$ is depleted by decay at a constant rate $\tau$: $$\dot{f}(t)= e^{-\frac{t^2}{2\sigma^2}} - \frac{1}{2\tau}f(t).$$ The resulting time-distribution is
$$\begin{aligned}
\label{eq:conv}
f(t,\sigma,\tau) &= \int_0^t \, e^{-\frac{t'^2}{2\sigma^2}}e^{-\frac{t-t'}{2\tau}}\,dt'\nonumber\\
&= e^{\frac{\sigma^2-4 \tau t}{8 \tau ^2}} \left[1-\text{erf}\left(\frac{\sigma^2-2 \tau t}{2 \sqrt{2} \tau \sigma}\right)\right].\end{aligned}$$
With this, the complete field is parametrized by $$\begin{aligned}
\label{eq:tot-fie}\nonumber
\mathsf{a}_y(x,t)=&\sum_{m=b,d}f(t\!-\!t_m,\sigma_m,\tau_m)\nonumber\\
&\times\left\{ \text{a}_m\left[ P_m^{(+)}\!-\! P_m^{(-)} \right]+\text{c}_mP_m^{(0)}\right\}.\end{aligned}$$ In practice, we find that the bright mode decays so fast that its propagation can be neglected in the spectrogram variance and we set a$_b\equiv0$.
We define the buildup time of each mode $\xi_m = \sigma_m \sqrt{\ln(2)}$ as the half-width half-maximum of the Gaussian function in Eq. (\[eq:conv\]), which allows for a direct comparison with the NIR pulse FWHM duration.
The dark mode plasmon duration $T$ has a measurable effect only during generation, when counter-propagating SPPs have not separated yet and form a standing wave. As this process is superposed by the bright mode, it cannot be reliably retrieved from the fit. However, $T$ is only weakly correlated with the dynamical parameters $\xi_m,\tau_m$ and $\omega_m$. Table \[tab:fit-res-fdtd\] shows the dynamical parameters for variations of $T$ over $[10, 20]$ fs (FWHM).
----------------- --------- --------- --------- --------- --------- --------- --------- ---------
$T $ $6$ $7$ $8$ $9$ $10$ $11$ $12$ Var
$T_\text{FWHM}$ $ 9.99$ $11.66$ $13.32$ $14.99$ $16.65$ $18.32$ $19.98$
$\xi_b$ $1.933$ $1.964$ $1.987$ $2.004$ $2.016$ $2.025$ $2.031$ $5$ %
$\tau_b$ $3.285$ $3.137$ $3.031$ $2.964$ $2.924$ $2.903$ $2.896$ $13$ %
$\xi_d$ $5.941$ $5.649$ $5.430$ $5.286$ $5.202$ $5.160$ $5.149$ $15$ %
$\tau_d$ $34.18$ $34.41$ $34.57$ $34.63$ $34.62$ $34.57$ $34.47$ $< 1$ %
$\omega_b$ $1.613$ $1.615$ $1.616$ $1.617$ $1.617$ $1.616$ $1.615$ $< 1$ %
$\omega_d$ $1.645$ $1.645$ $1.645$ $1.645$ $1.645$ $1.645$ $1.645$ $< 1$ %
----------------- --------- --------- --------- --------- --------- --------- --------- ---------
: \[tab:fit-res-fdtd\] Buildup-, life-time, and frequency of the bright and dark modes as obtained by fitting Eq. (\[eq:var\]) with the parameterization (\[eq:tot-fie\]), for a range of plasmon durations $T_m$. Times in fs, frequencies in eV, $T_{\text{FWHM}}=2\sqrt{\ln2}T$.
------------------------------------------------------------------------
\
\
[\[1\]]{} F.C. Garcia-Vidal [*et al.*]{}, J. Lightwave Technology [**17**]{}, 2191 (1999).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The performances of machine translation (MT) systems are usually evaluated by the metric BLEU when the golden references are provided. However, in the case of model inference or production deployment, golden references are usually expensively available, such as human annotation with bilingual expertise. In order to address the issue of translation quality estimation (QE) without reference, we propose a general framework for automatic evaluation of the translation output for the QE task in the Conference on Statistical Machine Translation (WMT). We first build a conditional target language model with a novel bidirectional transformer, named ***neural bilingual expert*** model, which is pre-trained on large parallel corpora for feature extraction. For QE inference, the bilingual expert model can simultaneously produce the joint latent representation between the source and the translation, and real-valued measurements of possible erroneous tokens based on the prior knowledge learned from parallel data. Subsequently, the features will further be fed into a simple Bi-LSTM predictive model for quality estimation. The experimental results show that our approach achieves the state-of-the-art performance in most public available datasets of WMT 2017/2018 QE task.'
author:
- |
Kai Fan[^1], Jiayi Wang$^*$, Bo Li$^*$, Fengming Zhou, Boxing Chen, Luo Si\
[{k.fan,joanne.wjy,shiji.lb,zfm104435,boxing.cbx,luo.si}@alibaba-inc.com]({k.fan,joanne.wjy,shiji.lb,zfm104435,boxing.cbx,luo.si}@alibaba-inc.com)\
Alibaba Group Inc.\
bibliography:
- 'aaai.bib'
title: '“Bilingual Expert" Can Find Translation Errors'
---
Introduction
============
The neural machine translation (NMT) in a sequence-to-sequence fashion, empowering an end-to-end learning approach for automatic translation system, has accomplished great success to potentially overcome many of the weaknesses of conventional phrase-based translation, and claimed being close to human parity for certain language pairs [@wu2016google; @hassan2018achieving]. However, current MT systems are still not perfect to meet the real-world applications without human post-editing (a popular example is the Chinese to English translation test at Google online system [^2], which translated “
[UTF8]{}[gbsn]{}苹果比谷歌厉害
" to “Apple is worse than Google", where the correct translation should be “Apple is better than Google".). Apparently, additional error correction is needed for even such a simple translation output. A possible solution to take advantage of the existing MT technologies is to collaborate with human translators within a computer-assisted translation (CAT) [@barrachina2009statistical]. In such cases, translation quality estimation (QE) plays a critical role in CAT to reduce human efforts, thereby increasing productivity [@specia2011exploiting]. Either the global sentence quality score or the fine-grained word “OK/BAD" tags can guide the CAT as an evidence to indicate whether a machine translation output requires further manual post-editing, or even which particular token needs special correction.
One traditional direction for translation quality estimation is to formulate the sentence level score or word level tags prediction as a constraint regression or sequence labeling problem respectively [@bojar2017findings]. The classical baseline model is to use the QuEst++ [@specia-paetzold-scarton:2015:ACL-IJCNLP-2015-System-Demonstrations] with two modules: rule based feature extractor and scikit-learn [^3] SVM algorithms. Similarly, the recent predictor-estimator model [@kim2017predictor] is a recurrent neural network (RNN) based feature extractor and quality estimation model, ranking first place at WMT 2017 QE.
Another promising direction is to build a multi-task learning model to incorporate quality estimation task with automatic post-editing (APE) together [@hokamp2017ensembling; @tan2017neural; @chatterjee2018combining], achieving the goal of CAT eventually. In this paper, we will first adopt the traditional single task framework to describe our model. In the experimental section, we also propose an extension to support multi-task learning for QE and APE simultaneously.
However, the final prediction model for scoring or tagging is not the main contribution in our work. Since there are many publicly available bilingual corpora, we can readily build a conditional language model as a robust feature extractor. The high level joint latent representation of the source and the target in a parallel pair can hopefully capture the alignment or semantic information. In contrast, when a source and a low-quality machine translation are fed into the pre-trained language model, the distribution of latent features is very likely to be different from the one that grammatically correct target has. Intuitively, people can learn the foreign language from reading the correct translation to their native language. Gradually, they may acquire the ability to be aware of the abnormality, even when errors appear in a sentence have never seen before. Additionally, we design 4-dimensional token mis-matching features from the pre-trained model, measuring the difference between what the bilingual expert model will predict and the actual token of machine translation output.
Particularly, we use the recent proposed self-attention mechanism and transformer neural networks [@vaswani2017attention] to build the conditional language model – *neural bilingual expert*. The model consists of the traditional transformer encoder for the source sentence and a novel bidirectional transformer decoder for the target sentence. It will be pre-trained on the large parallel corpus, and then produce high level features for the downstream quality estimation task. Constructing pre-trained word embedding on our designed language models has shown great improvement in many downstream NLP tasks. Both ELMO [@Peters:2018] and the OpenAI’s transformer decoder trained for monolingual language model [@radfordimproving] are good illustrations. Bidirectional attention mechanism was mainly proposed to achieve success in machine reading comprehension, such BiDAF [@seo2016bidirectional] and [@shen2018bi]. However, all of them are used for monolingual training without involving other conditional language.
The conditional language model can play the role of automatic post-editing as well. Since shifts were not annotated as word order errors (but rather as deletions and insertions) to avoid introducing noise in the annotation, missing tokens in the machine translations, as indicated by the TER tool [@snover2006study], are annotated as follows: after each token in the sentence and at sentence start, a gap tag is placed. In this situation, we can use the same network structure of conditional language model to enable the gap prediction (insertions) for missing token of translation output conditional on the source sentence. Using the deletion operation in word level tagging (by adding class “D" rather than “OK/BAD"), we are literally trying to predict post-editing.
This paper makes the following main contributions: i) we propose a novel approach with bidirectional transformer for building a conditional language model and pre-train it on available large bilingual corpora, which can further be used as automatic post-editing model. ii) we address the importance of the 4-dimensional mis-matching features, and in the experiments, with only these features, our approach can still achieve comparable results with No. 1 system in WMT 2017 QE task. iii) we develop a differentiable word-level quality estimation model to support data preprocessing with byte-pair-encoding (BPE) tokenization, bridging the gap between words and BPE tokens. iv) extensive experiments on real-world datasets (e.g., IT and pharmacy domain corpora) demonstrate our method is effective and achieve the state-of-the-art performance in most tasks.
Background
==========
#### Quality Estimation for Machine Translation
Given the bilingual corpus, from the statistical view we can formulate the machine translation system as $p(\mathbf{t}|\mathbf{s})=p(\mathbf{t}|\mathbf{z})p(\mathbf{z}|\mathbf{s})$, where $\mathbf{s}$ represents the tokens sequence of source sentence, $\mathbf{t}$ for target sentence, and $\mathbf{z}$ is the latent variable to represent the encoded source sentence. Therefore, $p(\mathbf{z}|\mathbf{s})$ and $p(\mathbf{t}|\mathbf{z})$ can be practically considered as the encoder and decoder. In the quality estimation task of machine translation, the machine translation system is agnostic and the training dataset is given in the format of triplet $(\mathbf{s}, \mathbf{m}, \mathbf{t})$, where $\mathbf{m}$ is the translation output from the unknown machine translation system with the input $\mathbf{s}$, and $\mathbf{t}$ represents the human post-edited sentence based on $\mathbf{s}$ and $\mathbf{m}$. Notice we abuse using notation $\mathbf{t}$ to refer both golden reference and human post-edited sentence.
In general, the quality of $\mathbf{m}$ can be evaluated either in the global sentence level or the fine-grained word level. The sentence level score is calculated by the percentage of edits needed to fix for $\mathbf{m}$, denoted as HTER. The word level evaluation is framed as the sequential binary classification problem to distinguish between ‘OK’ and ‘BAD’ for each token in translation output. Particularly, the binary word-level labels are generated by using the alignments provided by the TER tool [@snover2006study] between $\mathbf{m}$ and $\mathbf{t}$. Notice the sentence HTER and word labels can also deterministically be calculated by the TER tool when $\mathbf{m}$ and $\mathbf{t}$ are both present. However, in inference only the source sentence $\mathbf{s}$ and machine translation $\mathbf{m}$ are available, thus essentially requiring an automatic method for quality estimation of machine translation output at run-time, without relying on any reference.
We can assume that the training data contains the tuple $(\mathbf{s}, \mathbf{m}, \mathbf{t}, h, \mathbf{y})$, where $h$ is a scalar to represent HTER, and $\mathbf{y}$ is a binary vector to indicate the ‘OK/BAD’ labels of machine translation output. Considering the inference scenario, our task is to learn a regression model $p(h|\mathbf{s}, \mathbf{m})$ and a sequence labeling model $p(\mathbf{y}|\mathbf{s}, \mathbf{m})$.
{width="\textwidth"}
Methodology
===========
Bilingual Expert Model
----------------------
In this section, we will first highlight how to train a neural bilingual expert model with a parallel corpus including $(\mathbf{s},\mathbf{t})$ pairs. By default of QE task, the machine translation system $p(\mathbf{t}|\mathbf{z})p(\mathbf{z}|\mathbf{s})$ is unknown, but in representation learning we are usually interested in the latent variable $\mathbf{z}$, whose posterior may contain the deep semantic information between the source and the target languages, and be beneficial to many downstream tasks [@hill2016learning]. According to the Bayes rule, we can write the posterior distribution of the latent variable as, $$p(\mathbf{z}|\mathbf{t}, \mathbf{s}) = \frac{p(\mathbf{t}|\mathbf{z})p(\mathbf{z}|\mathbf{s})}{p(\mathbf{t}|\mathbf{s})}$$ where the integral $p(\mathbf{t}|\mathbf{s})=\int p(\mathbf{t}|\mathbf{z})p(\mathbf{z}|\mathbf{s}) \mathrm{d}\mathbf{z}$ is usually intractable. Instead of exact inference, we propose a variational distribution $q(\mathbf{z}|\mathbf{t}, \mathbf{s})$ to approximate true posterior by minimizing exclusive Kullback-Leibler (KL) divergence. $$\min D_{KL}(q(\mathbf{z}|\mathbf{t}, \mathbf{s})\|p(\mathbf{z}|\mathbf{t}, \mathbf{s}))$$ Rather than optimizing the objective function above, we can equivalently maximize the following one, $$\label{eq:elbo}
\max \mathbb{E}_{q(\mathbf{z}|\mathbf{t}, \mathbf{s})}[p(\mathbf{t}|\mathbf{z})] - D_{KL}(q(\mathbf{z}|\mathbf{t}, \mathbf{s})\|p(\mathbf{z}|\mathbf{s})) .$$ A nice property of the new objective is that it is unnecessary to parameterize or estimate the implicit machine translation model $p(\mathbf{t}|\mathbf{s})$. The first expectation term in (\[eq:elbo\]) can be readily considered as a conditional auto-encoder system if we use one sample Monte Carlo integration during optimization, and the second KL term can be analytically computed if we practically set the prior $p(\mathbf{z}|\mathbf{s})$ as standard Gaussian distribution, playing as a model regularization for latent variables. Furthermore, if we omit the conditional information $\mathbf{s}$, the objective exactly reduces to amortized variational inference or variational auto-encoders (VAE) framework [@kingma2013auto]. In analogous to most VAE models, the expected log-likelihood is commonly approximated by a practical surrogated term, $$\mathbb{E}_{q(\mathbf{z}|\mathbf{t}, \mathbf{s})}[p(\mathbf{t}|\mathbf{z})] \approx p(\mathbf{t}|\tilde{\mathbf{z}}), ~~\tilde{\mathbf{z}}\sim q(\mathbf{z}|\mathbf{t}, \mathbf{s})$$ Next, we will show the details of constructing the other two probability distributions appeared in (\[eq:elbo\]) with self-attention based transformer neural networks.
Bidirectional Transformer
-------------------------
Transformer [@vaswani2017attention] is based solely on attention mechanisms, dispensing with recurrence and convolution, becoming the state-of-the-art NMT model in most machine translation competitions. @vaswani2017attention claims that self-attention mechanism has several advantages: first, its gating or multiplication enables crisp error propagation; second, it can replace sequence-aligned recurrence entirely; third, from the implementation perspective, it is trivial to be parallelized during training. When we design the bidirectional transformer, we are trying to keep the three properties remained in our model.
The overall model architecture of bidirectional transformer is illustrated in the right block of Figure \[fig:bi\_trans\]. There are three modules in total, self-attention encoder for the source sentence, forward and backward self-attention encoders for target sentence, and the reconstructor for the target sentence, where the first two modules represent the proposed posterior approximation $q(\mathbf{z}|\mathbf{s},\mathbf{t})$ and the third reconstruction process corresponds to $p(\mathbf{t}|\mathbf{z})$. To make the inference efficient, we explicitly assume the conditional independence with the following factorization, $$\begin{aligned}
p(\mathbf{t}|\mathbf{z}) &= \prod_{k}p(t_k|\overrightarrow{\mathbf{z}_k},\overleftarrow{\mathbf{z}_k}) \label{eq:p_t} \\
q(\mathbf{z}|\mathbf{s},\mathbf{t}) &= \prod_{k} q(\overrightarrow{\mathbf{z}_k}|\mathbf{s},\mathbf{t}_{<k})q(\overleftarrow{\mathbf{z}_k}|\mathbf{s},\mathbf{t}_{>k}) \label{eq:q_z}\end{aligned}$$ where the bidirectional latent variable $\mathbf{z}$ includes all $\{\overrightarrow{\mathbf{z}_k},\overleftarrow{\mathbf{z}_k}\}$. Note that our factorization is different from ELMO [@Peters:2018], where they use a finer grained form $\prod_{k}p(t_k|\overrightarrow{\mathbf{z}_k})p(t_k|\overleftarrow{\mathbf{z}_k})$ but with the shared parameters between forward and backward reconstruction $p(t_k|\cdot)$.
Latent variables $\overrightarrow{\mathbf{z}_k},\overleftarrow{\mathbf{z}_k}$ are sampled from $q(\overrightarrow{\mathbf{z}_k}|\mathbf{s},\mathbf{t}_{<k})$ and $q(\overleftarrow{\mathbf{z}_k}|\mathbf{s},\mathbf{t}_{>k})$ respectively, assuming to follow the Gaussian distribution, e.g., $q(\cdot|\cdot) \sim \mathcal{N}(\mu(\mathbf{s},\mathbf{t}), \sigma^2\mathbf{I})$. Meanwhile, the mean $\mu(\mathbf{s},\mathbf{t})$ is learned in an amortized way, i.e., every single pair $\mathbf{s},\mathbf{t}$ will generate their own mean via the shared neural network model. By fixing $\sigma$ as a hyper-parameter, we can efficiently implement the stochastic layer as the deterministic one via dropout training with additive Gaussian noise [@srivastava2014dropout]. The stochastic layer can increase the uncertainty of the latent representation, potentially preventing overfitting. In practice, a small $\sigma$ is recommended. Notice that we didn’t follow the NMT parlance to call our bidirectional self-attention transformer as “decoder", since it is not actually a generative model during inference.
Model Derived Features
----------------------
Once the bilingual expert model has been fully trained on large parallel corpora, we can reasonably assume the model will predict higher likelihood for the correct target token, given the source and other context of the target, if only very few tokens are incorrect. Therefore, we will use the prior knowledge learned by bilingual expert to extract the features for subsequent translation error prediction. Basically, we will first design the sequential (token-wise) model derived features based upon the pre-trained model with $(\mathbf{s},\mathbf{m})$ pair as input. The latent representation $\mathbf{z}_k = \text{Concat}(\overrightarrow{\mathbf{z}_k}, \overleftarrow{\mathbf{z}_k})$ should naturally be the high level features. As we discussed previously, the entire latent variable $\mathbf{z}$ should generally summarize the information of the source and the target. In Equation (\[eq:q\_z\]), the distribution of $\mathbf{z}_k$ is deliberately defined to contain the information from the source and the context around the $k$-th token in the target. We see this by observing the computational graph in the right panel of Figure \[fig:bi\_trans\], e.g., the token “den" of target is desired to predict, but only the information of the source and all the other tokens in the target will be propagated to the final layer for prediction. It will be reasonably beneficial to our manually extracted mis-matching features introduced later.
In ELMO [@Peters:2018], the token embedding is also used as one linear component to compute the final feature. However, in our case that translation output is fed into the model, it is not guaranteed that every single token is correct. Therefore, we design a different token embedding feature following the rationale of subtle information flow within latent variable $\mathbf{z}_k$. In fact, we use the embedding concatenation of two neighbor tokens $\text{Concat}(\mathbf{e}_{t_{k-1}}, \mathbf{e}_{t_{k+1}})$. Since the possibly erroneous translation may mislead the model in the downstream quality estimation task, we did not extract any information from current token $t_k$. More importantly, the correct syntax representation of the token which is supposed to be translated should come from the source sentence, which has been encoded into $\mathbf{z}$ via joint attention.
Mis-matching Features
---------------------
Besides the proposed model derived features that are exactly the nodes within the computational graph of the bidirectional transformer, we intuitively found another type of crucial features that can directly measure how the prior knowledge from the well-trained bilingual expert model is different from the translation. To make it concrete, $p(t_k|\cdot)$ follows the categorical distribution with the number of classes equal to the vocabulary size. Since we pre-train the bilingual expert model on parallel corpara, the objective (\[eq:elbo\]) is theoretically to maximize the likelihood of each $p(t_k|\cdot)$, which achieves its maximum when $t_k$ is ground truth. Intuitively, we should have $p(m_k|\cdot) \leq p(t_k|\cdot)$ for optimal model if $m_k \neq t_k$, illustrated in top-left block of Figure \[fig:bi\_trans\]. Following this intuition, we propose the mis-matching features.
Suppose $\mathbf{l}_k$ is the logits vector before applying the softmax operation, i.e. $p(t_k|\cdot) \sim \text{Categorical}(softmax(\mathbf{l}_k))$, thus we can define the 4-dimensional mis-matching features as the following vector, $$\mathbf{f}_k^{mm} = (\mathbf{l}_{k, m_k}, \mathbf{l}_{k, i^k_{\max}}, \mathbf{l}_{k, m_k}-\mathbf{l}_{k, i^k_{\max}}, \mathbb{I}_{m_k\neq i^k_{\max}})$$ where $m_k$ represents the vocabulary id of the $k$-th token in translation output, $i^k_{\max} = \arg\max_i \mathbf{l}_k$ is the id that the bilingual expert predicts, and $\mathbb{I}$ is indicator function. Therefore, these four values will directly reflect the differences or errors. Apparently, if the machine translation coincides with the bilingual expert prediction, the first 2 elements of $\mathbf{f}_k^{mm}$ should be identical and the last two elements, representing soft and hard differences, should be both 0. We empirically found the quality estimation model can achieve comparable result even with the mis-matching features alone.
QE training data $(\mathbf{s}, \mathbf{m}, \mathbf{t}, h, \mathbf{y})_{1:M}$, QE inference data $(\mathbf{s},\mathbf{m})$, and parallel corpus $(\mathbf{s},\mathbf{t})_{1:N}$. Combine the parallel corpus with 10 copies of QE training parallel corpus $C = (\mathbf{s}_n,\mathbf{t}_n)_{n=1}^N \bigcup 10 \times (\mathbf{s}_m, \mathbf{t}_m)_{m=1}^M$ Pre-train bilingual expert model via the bidirectional transformer on the combined corpus $C$. Extract features $\mathbf{f}_k = \text{Concat}(\overrightarrow{\mathbf{z}_k}, \overleftarrow{\mathbf{z}_k}, \mathbf{e}_{t_{k-1}}, \mathbf{e}_{t_{k+1}}, \mathbf{f}_k^{mm})$ for QE training data $(\mathbf{s}, \mathbf{m})$. Train Bi-LSTM model via objectives (\[eq:sent\_loss\])(\[eq:word\_loss\]). Predict $h, \mathbf{y}$ for QE inference data
\[alg:double\_bi\]
--------------------- -------------------------- ------------------ ------------------- --------------------------- --------------------- -------------------------- ------------------ ------------------- --------------------------- ---------------------
Method **Pearson’s** $\uparrow$ MAE $\downarrow$ RMSE $\downarrow$ **Spearman’s** $\uparrow$ DeltaAvg $\uparrow$ **Pearson’s** $\uparrow$ MAE $\downarrow$ RMSE $\downarrow$ **Spearman’s** $\uparrow$ DeltaAvg $\uparrow$
Baseline 0.3970 0.1360 0.1750 0.4250 0.0745 0.4410 0.1280 0.1750 0.4500 0.0681
Unbabel 0.6410 0.1280 0.1690 0.6520 0.1136 0.6260 0.1210 0.1790 0.6100 0.9740
POSTECH Single 0.6599 0.1057 0.1450 0.6914 0.1188 0.6985 0.0952 0.1461 0.6408 0.1039
Ours Single (MD+MM) **0.6837** 0.1001 0.1441 **0.7091** 0.1200 **0.7099** 0.0927 0.1394 **0.6424** 0.1018
w/o MM 0.6763 0.1015 0.1466 0.7009 0.1182 0.7063 0.0947 0.1410 0.6212 0.1005
w/o MD 0.6408 0.1074 0.1478 0.6630 0.1101 0.6726 0.1089 0.1545 0.6334 0.0961
POSTECH Ensemble 0.6954 0.1019 0.1371 0.7253 0.1232 0.7280 0.0911 0.1332 0.6542 0.1064
Ours Ensemble **0.7159** 0.0965 0.1384 **0.7402** 0.1247 **0.7338** 0.0882 0.1333 **0.6700** 0.1050
--------------------- -------------------------- ------------------ ------------------- --------------------------- --------------------- -------------------------- ------------------ ------------------- --------------------------- ---------------------
Bi-LSTM Quality Estimation
--------------------------
To this end, we have the model derived and manually designed sequential features, each time stamp of which is corresponding to a fixed size vector. Our quality estimation task is built upon the bidirectional LSTM [@graves2005framewise] model, being widely used for sequence classification or sequence tagging problems. In sequence tagging, @huang2015bidirectional proposed a variant of Bi-LSTM with one Conditional Random Field (CRF) layer (Bi-LSTM-CRF). We empirically found that the extra CRF layer did not show any significant improvement over vanilla Bi-LSTM, which we simply adopted. Another natural question is whether the traditional encoder self-attention or our proposed forward/backward self-attention can be an alternative to the Bi-LSTM. We empirically found the results with self-attention module become even worse, and we suspect the scarcity of labelled quality estimation data, being incomparable to the sufficient parallel corpus, is the main reason.
We concatenate all sequential features along the depth direction to obtain a single vector, denoted as $\{\mathbf{f}_k\}_{k=1}^T$, where $T$ is the number of tokens in $\mathbf{m}$. Therefore, the sentence level score HTER prediction can be formulated as a regression problem (\[eq:sent\_loss\]), and the word error prediction is a sequence labeling problem (\[eq:word\_loss\]), $$\begin{aligned}
\overrightarrow{\mathbf{h}_{1:T}}, \overleftarrow{\mathbf{h}_{1:T}} &= \text{Bi-LSTM}(\{\mathbf{f}_k\}_{k=1}^T) \\
\arg\min & \left\|h - \text{sigmoid}\left(\mathbf{w}^\top [\overrightarrow{\mathbf{h}_T}, \overleftarrow{\mathbf{h}_T}]\right) \right\|_2^2 \label{eq:sent_loss} \\
\arg\min & \sum_{k=1}^T \text{XENT}(y_k, \mathbf{W}[\overrightarrow{\mathbf{h}_k}, \overleftarrow{\mathbf{h}_k}]) \label{eq:word_loss}\end{aligned}$$ where $\mathbf{w}$ is a vector, $\mathbf{W}$ is a matrix, $y_k$ is the error label for the $k$-th token of translation output, and XENT is the cross entropy loss (with logits). Notice HTER $h$ is a real value within interval $[0,1]$, we apply a squash function “sigmoid" for rescaling in the regression model. Since the HTER is a global score for the entire sentence, we use the hidden states of the last time stamp in the forward/backward LSTMs as the regression signals. Actually, we can train the two losses together in a multi-task setting. In summary, we describe the outline of our proposed approach in Algorithm \[alg:double\_bi\].
Experiments
===========
Setting Description
-------------------
The data resources that we used for training the neural Bilingual Expert model are mainly from WMT[^4]: (i) parallel corpora released for the WMT17/18 News Machine Translation Task, (ii) UFAL Medical Corpus and Khresmoi development data released for the WMT17/18 Biomedical Translation Task, (iii) src-pe pairs for the WMT17/18 QE Task. To ensure the quality of the corpora, we filtered the source and target sentence with length $\leq$70 and the length ratio between 1/3 to 3, thus resulting roughly 9 million (2017) and 25 million (2018) parallel sentences pairs for both English$\leftrightarrow$German directions. We mainly tried word tokenization for the corpus in the WMT17 QE task, where the word tokenization naturally fits the word level QE task. For WMT18, we applied byte-pair-encoding (BPE) [@sennrich2016neural] tokenization to reduce the number of unknown tokens. However, there exists the discrepancy between word token tagging prediction and BPE tokenization, and we will present how to bridge the gap in the next section. We also test our model on the CWMT 2018 Chinese English sentence QE task[^5]. Since the two languages are unrelated, we tokenize them separately.
The number of layers in the bidirectional transformer for each module is 2, and the number of hidden units for feedforward sub-layer is 512. We use the 8-head self-attention in practice, since the single one is just a weighted average of previous layers. The bilingual expert model is trained on 8 Nvidia P-100 GPUs for about 3 days until convergence. For translation QE model, we use only one layer Bi-LSTM, and it is trained on a single GPU.
We evaluate our algorithm on the testing data of WMT 2017/2018, and development data of CWMT 2018. Notice for the QE task of WMT 2017, it is forbidden to use any data from 2018, since the training data of 2018 includes some testing data of 2017. The same setting applies to all following experiments. For fair comparison, we tuned all the hyper-parameters of our model on the development data, and reported the corresponding results for the testing data.
Sentence Level Scoring And Ranking
----------------------------------
--------------- -------------------------- ------------------ ------------------- ---------------------------
**Pearson’s** $\uparrow$ MAE $\downarrow$ RMSE $\downarrow$ **Spearman’s** $\uparrow$
Method
Baseline 0.3653 0.1402 0.1772 0.3809
UNQE 0.7000 0.0962 0.1382 0.7244
Ours Ensemble **0.7308** 0.0953 0.1383 **0.7470**
Method
Baseline 0.3323 0.1508 0.1928 0.3247
UNQE **0.7667** 0.0945 0.1315 0.7261
Ours Ensemble 0.7631 0.0962 0.1328 **0.7318**
--------------- -------------------------- ------------------ ------------------- ---------------------------
: Results of sentence level QE on WMT 2018[]{data-label="tab:sent2018"}
System Used Bi-Corpus Ch->En En->Ch
---------------------------- ------------------ ----------- -----------
CWMT 1st ranked (Ensemble) CWMT 8m + 8m BT 0.465 0.405
Our Model 1 (Single) WMT 25m + 25m BT 0.612 0.620
Our Model 2 (Single) CWMT 8m 0.564 0.588
: Pearson’s coefficient of CWMT 2018 QE[]{data-label="tab:sent2018_cwmt"}
----------------------- -------- -------- --------------
F1-BAD F1-OK **F1-Multi**
Method
Baseline 0.407 0.886 0.361
DCU 0.614 0.910 0.559
Unbabel 0.625 0.906 0.566
POSTECH Ensemble 0.628 0.904 0.568
Ours Single (MM + MD) 0.6410 0.9083 **0.5826**
Method
Baseline 0.365 0.939 0.342
POSTECH Single 0.552 0.936 0.516
Unbabel 0.562 0.941 0.529
POSTECH Ensemble 0.569 0.940 0.535
Ours Single (MM + MD) 0.5816 0.9470 **0.5507**
Method
Baseline 0.4115 0.8821 0.3630
Conv64 0.4768 0.8166 0.3894
SHEF-PT 0.5080 0.8460 0.4298
Ours Ensemble 0.6616 0.9168 **0.6066**
Method
Baseline 0.1973 0.9184 0.1812
Conv64 0.3573 0.8520 0.3044
SHEF-PT 0.3353 0.8691 0.2914
Ours Ensemble 0.4750 0.9152 **0.4347**
Method
Baseline 0.4850 0.9015 0.4373
Conv64 0.4948 0.8474 0.4193
SHEF-PT 0.4853 0.8741 0.4242
Ours Ensemble 0.6475 0.9162 **0.5932**
----------------------- -------- -------- --------------
: Results of word level QE on WMT 2017/2018[]{data-label="tab:word201718"}
The sentence level results of WMT 2017 are listed in Table \[tab:sent2017\]. We mainly compared our single model with the two algorithms [@kim2017predictor; @martins2017unbabel], ranking top 3 in the WMT 2017 finalist. **Unbabel** is combination of a feature-rich sequential linear model with a neural network. **POSTECH** is a predictor-estimator model with all Bi-GRU modules. **Baseline** is the official provided system. The primary metrics of sentence level task are Pearson’s correlation and Spearman’s rank correlation of the entire testing data. Alternatively, mean average error (MAE), root mean squared error (RMSE), or the average of delta values (DeltaAvg) can also measure the performance of overall predictions, but not be a ranking reference in the QE task. For both single and ensemble model comparisons, our algorithm can outperform all other systems for the two primary metrics. The ranking results are generated by the predicted HTER scores. In addition, we also analyze the importance of model derived features (MD) and mis-matching features (MM) the ablation study. With 4-dimensional mis-matching features alone, the model can still achieve comparable or better performance than the second single system last year. It demonstrates that the low dimensional features can provide a strong prediction signal as well.
We also report the result on unrelated language pair, Chinese and English, as shown in Table \[tab:sent2018\_cwmt\], where BT means back-translation. Our single model without back-translation has outperformed the best system in the competition.
Word Level For Word Tagging
---------------------------
The metric of word level is evaluated in terms of classification performance via the multiplication of F1-scores for the ‘OK’ and ‘BAD’ classes against the true labels. For the binary classification, we tuned the threshold of the classifier on the development data and applied to the test data. The overall results are shown at Table \[tab:word201718\]. The baseline is provided by the offical WMT organizers, and the system is trained by CRFSuite toolkit with passive-aggressive algorithm [@CRFsuite]. We also compared the top 3 algorithms in WMT17 QE task, POSTECH [@kim2017predictor], Unbabel [@martins2017unbabel], and DCU [@martins2017pushing]. **DCU** is a stacked neural model by exploiting synergies between the related tasks of word-level quality estimation and automatic post-editing. In the primary metric F1-Multi, our algorithm of the single model outperforms all other models, including the best ensemble system in WMT17. In WMT18 word level QE task, our approach exceeds all other algorithms with significant better numbers.
The higher value of single F1-OK or F1-BAD cannot reflect the robustness of the algorithm, since it may result in lower F1 of another metric. Though we presented the F1-OK and F1-BAD, it is not a valid metric to QE task. However, by comparing them, we can conclude that all algorithms tend to classify the word tag as OK in general, since the true labels are very imbalanced. This is the reason why we use the threshold tuning strategy to finalize our classifier.
Word Level For Gap Tagging
--------------------------
Method F1-BAD F1-OK **F1-Multi**
--------------- -------- -------- --------------
UAlacante SBI 0.1997 0.9444 0.1886
SHEF-bRNN 0.2710 0.9552 0.2589
SHEF-PT 0.2937 0.9618 0.2824
Ours Ensemble 0.5109 0.9783 **0.4999**
----- ------------------------------------------------------------------------------------------------------------------
MT sie im bedienfeld “ profile ” des dialogfelds “ preflight ” auf die schaltfläche “ auswählen . ”
APE klicken sie im bedienfeld “ profile ” des dialogfelds “ preflight ” auf die schaltfläche “ profile auswählen . ”
PE klicken sie im bedienfeld “ profile ” des dialogfelds “ preflight ” auf die schaltfläche “ profile auswählen . ”
MT das von komplexen symbolen und große textblöcke kann viel zeit in anspruch nehmen .
APE das von komplexen symbolen und großen textblöcke kann viel zeit in anspruch nehmen .
PE das aufteilen von komplexen symbolen und großen textblöcke kann viel zeit in anspruch nehmen .
MT sie müssen nicht auf den ersten punkt , um das polygon zu schließen .
APE sie müssen nicht auf den ersten punkt , um das polygon zu schließen .
PE sie müssen nicht auf den ersten punkt klicken , um das polygon zu schließen .
MT sie können bis zu vier zeichen .
APE sie können bis zu vier zeichen .
PE sie können bis zu vier zeichen eingeben .
MT die standardmaßeinheit in illustrator ( ein punkt entspricht .3528 ) .
APE die standardmaßeinheit in illustrator ist punkt ( ein punkt entspricht .3528 ) .
PE die standardmaßeinheit in illustrator ist punkt ( ein punkt entspricht .3528 millimetern ) .
----- ------------------------------------------------------------------------------------------------------------------
The gap level error prediction is important to machine translation system as well. Missing tokens in the machine translation, as indicated by the TER tool, are annotated as follows: after each token in the sentence and at the sentence start, a gap tag is placed. Note that the number of gap tags for each translation sentence is $T+1$, including the predictions before the first token and after last one. Therefore, we can directly build the gap prediction model by modifying (\[eq:word\_loss\]) as, $$\label{eq:gap_loss}
\arg\min \sum_{k=0}^{T} \text{XENT}(g_k, \mathbf{W}[\overrightarrow{\mathbf{h}_k}, \overleftarrow{\mathbf{h}_k},\overrightarrow{\mathbf{h}_{k+1}}, \overleftarrow{\mathbf{h}_{k+1}}])$$ where $g_k$ is the gap tag between the $k$th and $k$+1st tokens. We can train the neural bilingual expert model for gap prediction to extract more representative features for the downstream task. Basically, we have the following factorization model $p(\mathbf{t},\mathbf{t}^g|\mathbf{z})=p(\mathbf{t}|\mathbf{z})p(\mathbf{t}^g|\mathbf{z})$ and $q(\mathbf{z}|\mathbf{s},\mathbf{m})$, where $p(\mathbf{t}|\mathbf{z})$ is identical as previously discussed model, gap token prediction distribution $p(\mathbf{t}^g|\mathbf{z})=\prod_k p(t_k^g|\overrightarrow{\mathbf{z}_k}, \overleftarrow{\mathbf{z}_k},\overrightarrow{\mathbf{z}_{k+1}}, \overleftarrow{\mathbf{z}_{k+1}})$ and $q$ becomes conditional on $\mathbf{m}$. Note that we need to define a “$<$blank$>$" token for gap prediction, meaning that nothing needs to be inserted. Therefore, it also results in a side product – automatic post-editing. If we label the human post-edited translations by the insertion or deletion operations to machine translations (which could be done by using TER tool), we can train the model to predict such operations on the target side, achieving a better APE system eventually. We leave this as the future work.
As we discussed in the introduction, most computer assisted translation scenarios use the quality estimation model as the an activator of APE, a guidance to APE corrections, or a selector of final translation output [@chatterjee2018combining]. Though QE can play the role of a helper function for APE, they are fundamentally considered as two separated tasks. In our proposed model, after we pre-trained the neural bilingual model for gap prediction, we can subsequently feed the model derived and mis-matching features to the Bi-LSTM model for gap quality estimation. We propose a direction to unify the quality estimation and automatic post-editing. First, we demonstrate the performance of our result for gap quality estimation in the left-side of Table \[tab:apegap\]. We also show several examples of APE results by our pre-trained model in the right-side of Table \[tab:apegap\].
Extending to BPE Tokenization
-----------------------------
In many NMT systems, using BPE or subword units gives an effective way to deal with rare words. Especially in German, there are a bunch of compound words, which are simply a combination of two or more words that function as a single unit of meaning, e.g. “handschuh" means glove in German, which is literally the “hand shoe". BPE tokenization gives a good balance between the flexibility of single characters and the efficiency of full words for decoding, and also sidesteps the need for special treatment of unknown words.
For sentence level HTER prediction, there is no harm or conflict to use BPE, since the regression signals only care about the hidden states of the last time stamps. However, for word level labeling, the length of sequential features $L_b$ with BPE tokenization is different from the number of word tokens $L_w$. We propose to average the features of all subword units belonging to one single word token, similar to average pooling along the time axis with dynamic sizes. To make the computational graph differentiable, the BPE segmentation information needs to be stored into a $L_w \times L_b$ sparse matrix $S$, where $S_{ij}\neq 0$ if $j$-th subword unit belongs to $i$-th word (see Fig \[fig:BPE\](a) for an example). The averaged features can be computed by matrix multiplication.
We compared the performance of the word and BPE tokenization on both sentence and word levels, and results are plotted as histograms in Fig \[fig:BPE\](b,c). Similar to NMT systems, the finer grained BPE tokenization can improve the QE performance in most tasks. In the sentence level, BPE model got a lower Pearson’s $r$ for en-de NMT QE task, which is very likely due to the small data size ($<$14000). In the word level, if we did not tune the threshold by using the default 0.5, the BPE model can always be better. After threshold tuning, the BPE model may have less improvement (we tune the threshold on development data and evaluate on it as well, since we did not have the ground truth of the testing data).
Actually, the two models can be jointly trained during the stage of quality estimation, no matter the preprocessing is word or BPE tokenization. Even for BPE tokenization, we can do back-propagation to update the “bilingual expert" model when we are training Bi-LSTM, if appropriate column and row paddings are added to the segmentation matrix. We will also leave this as another future work.
Conclusion
==========
In this paper, we present a novel approach to solve the quality estimation problem for machine translation systems. We first introduce the neural “bilingual expert" model as the prior knowledge model. Then, we use a simple Bi-LSTM as the quality estimation model with the extracted model derived and manually designed mis-matching features. In the end, we test our algorithm on the public available WMT 17/18 QE competition dataset and yield better performance than other algorithms in most downstream tasks.
[^1]: indicates equal contribution.
[^2]:
[^3]: <http://scikit-learn.org/>
[^4]: <http://www.statmt.org/wmt18/>
[^5]: <http://nlp.nju.edu.cn/cwmt2018/guidelines.html>
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
In this paper we study the problem of maintaining the strongly connected components of a graph in the presence of failures. In particular, we show that given a directed graph $G=(V,E)$ with $n=|V|$ and $m=|E|$, and an integer value $k\geq 1$, there is an algorithm that computes in $O(2^{k}n\log^2 n)$ time for any set $F$ of size at most $k$ the strongly connected components of the graph $G\setminus F$. The running time of our algorithm is almost optimal since the time for outputting the SCCs of $G\setminus F$ is at least $\Omega(n)$. The algorithm uses a data structure that is computed in a preprocessing phase in polynomial time and is of size $O(2^{k} n^2)$.
Our result is obtained using a new observation on the relation between strongly connected components (SCCs) and reachability. More specifically, one of the main building blocks in our result is a restricted variant of the problem in which we only compute strongly connected components that intersect a certain path. Restricting our attention to a path allows us to implicitly compute reachability between the path vertices and the rest of the graph in time that depends logarithmically rather than linearly in the size of the path. This new observation alone, however, is not enough, since we need to find an efficient way to represent the strongly connected components using paths. For this purpose we use a mixture of old and classical techniques such as the heavy path decomposition of Sleator and Tarjan [@ST:83] and the classical Depth-First-Search algorithm. Although, these are by now standard techniques, we are not aware of any usage of them in the context of dynamic maintenance of SCCs. Therefore, we expect that our new insights and mixture of new and old techniques will be of independent interest.
author:
- |
Surender Baswana\
Department of CSE,\
I.I.T. Kanpur, India\
[`[email protected]`]{}
- |
Keerti Choudhary\
Department of CSE,\
I.I.T. Kanpur, India\
[`[email protected]`]{}
- |
Liam Roditty\
Department of Comp. Sc.\
Bar Ilan University, Israel.\
[`[email protected]`]{}
bibliography:
- 'ref.bib'
title: |
An efficient strongly connected components algorithm\
in the fault tolerant model
---
Introduction
============
Computing the strongly connected components (SCCs) of a directed graph $G=(V,E)$, where $n=|V|$ and $m=|E|$, is one of the most fundamental problems in computer science. There are several classical algorithms for computing the SCCs in $O(m+n)$ time that are taught in any standard undergrad algorithms course [@Cormen].
In this paper we study the following natural variant of the problem in dynamic graphs. What is the fastest algorithm to compute the SCCs of $G\setminus F$, where $F$ is any set of edges or vertices. The algorithm can use a polynomial size data structure computed in polynomial time for $G$ during a preprocessing phase.
The main result of this paper is:
\[T-main\] There is an algorithm that computes the SCCs of $G\setminus F$, for any set $F$ of $k$ edges or vertices, in $O(2^{k}n\log^2 n)$ time. The algorithm uses a data structure of size $O(2^{k} n^2)$ computed in $O(2^kn^2m)$ time for $G$ during a preprocessing phase.
Since the time for outputting the SCCs of $G\setminus F$ is at least $\Omega(n)$, the running time of our algorithm is optimal (up to a polylogarithmic factor) for any fixed value of $k$.
This dynamic model is usually called the fault tolerant model and its most important parameter is the time that it takes to compute the output in the presence of faults. It is an important theoretical model as it can be viewed as a restriction of the deletion only (decremental) model in which edges (or vertices) are deleted one after another and queries are answered between deletions. The fault tolerant model is especially useful in cases where the worst case update time in the more general decremental model is high.
There is wide literature on the problem of decremental SCCs. Recently, in a major breakthrough, Henzinger, Krinninger and Nanongkai [@HenzingerKN14] presented a randomized algorithm with $O(mn^{0.9+o(1)})$ total update time and broke the barrier of $\Omega(mn)$ for the problem. Even more recently, Chechik et al. [@ChETAL] obtained an improved total running time of $O(m\sqrt{n \log n})$.
However, these algorithms and in fact all the previous algorithms have an $\Omega(m)$ worst case update time for a single edge deletion. This is not a coincidence. Recent developments in conditional lower bounds by Abboud and V. Williams [@AbboudW14] and by Henzinger, Krinninger, Nanongkai and Saranurak [@HenzingerKNS15] showed that unless a major breakthrough happens, the worst case update time of a single operation in any algorithm for decremental SCCs is $\Omega(m)$. Therefore, in order to obtain further theoretical understanding on the problem of decremental SCCs, and in particular on the worst case update time it is only natural to focus on the restricted dynamic model of fault tolerant.
In the recent decade several different researchers used the fault tolerant model to study the worst case update time per operation for dynamic connectivity in undirected graphs. Pǎtraşcu and Thorup [@PaTh07] presented connectivity algorithms that support edge deletions in this model. Their result was improved by the recent polylogarithmic worst case update time algorithm of Kapron, King and Mountjoy [@KapronKM13]. Duan and Pettie[@DuPe10; @DuPe17] used this model to obtain connectivity algorithms that support vertex deletions.
In directed graphs, very recently, Georgiadis, Italiano and Parotsidis [@GeItPa17] considered the problem of SCCs but only for a single edge or a single vertex failure, that is $|F|=1$. They showed that it is possible to compute the SCCs of $G\setminus \{e\}$ for any $e\in E$ (or of $G\setminus \{v\}$ for any $v \in V$) in $O(n)$ time using a data structure of size $O(n)$ that was computed for $G$ in a preprocessing phase in $O(m+n)$ time. Our result is the first generalized result for any constant size $F$. This comes with the price of an extra $O(\log^2 n)$ factor in the running time, a slower preprocessing time and a larger data structure. In [@GeItPa17], Georgiadis, Italiano and Parotsidis also considered the problem of answering strong connectivity queries after one failure. They show construction of an $O(n)$ size oracle that can answer in constant time whether any two given vertices of the graph are strongly connected after failure of a single edge or a single vertex.
In a recent result [@OUR-STOC] we considered the problem of finding a sparse subgraph that preserves single source reachability. More specifically, given a directed graph $G=(V,E)$ and a vertex $s\in V$, a subgraph $H$ of $G$ is said to be a $k$-Fault Tolerant Reachability Subgraph (${\textsc{$k$-FTRS}}$) for $G$ if for any set $F$ of at most $k$ edges (or vertices), a vertex $v\in V$ is reachable from $s$ in $G\setminus F$ if and only if $v$ is reachable from $s$ in $H\setminus F$. In [@OUR-STOC] we proved that there exists a ${\textsc{$k$-FTRS}}$ for $s$ with at most $2^k n$ edges.
Using the $k$-FTRS structure, it is relatively straightforward to obtain a data structure that, for any pair of vertices $u,v \in V$ and any set $F$ of size $k$, answers in $O(2^{k} n)$ time queries of the form: $$\text{``Are $u$ and $v$ in the same SCC of $G\setminus F$?''}$$
The data structure consists of a ${\textsc{$k$-FTRS}}$ for every $v\in V$. It is easy to see that $u$ and $v$ are in the same SCC of $G\setminus F$ if and only if $v$ is reachable from $u$ in ${\textsc{$k$-FTRS}}(u)\setminus F$ and $u$ is reachable from $v$ in ${\textsc{$k$-FTRS}}(v)\setminus F$. So the query can be answered by checking, using graph traversals, whether $v$ is reachable from $u$ in ${\textsc{$k$-FTRS}}(u)\setminus F$ and whether $u$ is reachable from $v$ in ${\textsc{$k$-FTRS}}(v)\setminus F$. The cost of these two graph traversals is $O(2^k n )$. The size of the data structure is $O(2^k n^2)$.
This problem, however, is much easier since the vertices in the query reveal which two ${\textsc{$k$-FTRS}}$ we need to scan. In the challenge that we address in this paper *all* the SCCs of $G\setminus F$, for an arbitrary set $F$, have to be computed. However, using the same data structure as before, it is not really clear a-priori which of the ${\textsc{$k$-FTRS}}$ we need to scan.
We note that our algorithm uses the ${\textsc{$k$-FTRS}}$ which seems to be an essential tool but is far from being a sufficient one and more involved ideas are required. As an example to such a relation between a new result and an old tool one can take the deterministic algorithm of [Ł]{}cki [@Lacki11] for decremental SCCs in which the classical algorithm of Italiano [@Italiano88] for decremental reachability trees in directed acyclic graphs is used. The main contribution of [Ł]{}cki [@Lacki11] is a new graph decomposition that made it possible to use Italiano’s algorithm [@Italiano88] efficiently.
An overview of our result {#subsection:overview}
-------------------------
We obtain our $O(2^k n \log^2 n)$-time algorithm using several new ideas. One of the main building blocks is surprisingly the following restricted variant of the problem.
To solve this restricted version, we implicitly solve the problem of reachability from $x$ (and to $x$) in $G\setminus F$, for each $x\in P$. Though it is trivial to do so in time $O(2^kn|P|)$ using ${\textsc{$k$-FTRS}}$ of each vertex on $P$, our goal is to preform this computation in $O(2^k n \log n)$ time, that is, in running time that is *independent* of the length of $P$ (up to a logarithmic factor). For this we use a careful insight into the structure of reachability between $P$ and $V$. Specifically, if $v\in V$ is reachable from $x\in P$, then $v$ is also reachable from any predecessor of $x$ on $P$, and if $v$ is not reachable from $x$, then it cannot be reachable from any successor of $x$ as well. Let $w$ be any vertex on $P$, and let $A$ be the set of vertices reachable from $w$ in $G\setminus F$. Then we can split $P$ at $w$ to obtain two paths: $P_1$ and $P_2$. We already know that all vertices in $P_1$ have a path to $A$, so for $P_1$ we only need to focus on set $V\setminus A$. Also the set of vertices reachable from any vertex on $P_2$ must be a subset of $A$, so for $P_2$ we only need to focus on set $A$. This suggests a divide-and-conquer approach which along with some more insight into the structure of ${\textsc{$k$-FTRS}}$ helps us to design an efficient algorithm for computing all the SCCs that intersect $P$.
In order to use the above result to compute all the SCCs of $G\setminus F$, we need a clever partitioning of $G$ into a set of vertex disjoint paths. A Depth-First-Search (DFS) tree plays a crucial role here as follows. Let $P$ be any path from root to a leaf node in a DFS tree $T$. If we compute the SCCs intersecting $P$ and remove them, then the remaining SCCs must be contained in subtrees hanging from path $P$. So to compute the remaining SCCs we do not need to work on the entire graph. Instead, we need to work on each subtree. In order to pursue this approach efficiently, we need to select path $P$ in such a manner that the subtrees hanging from $P$ are of small size. The heavy path decomposition of Sleator and Tarjan [@ST:83] helps to achieve this objective.[^1]
Our algorithm and data structure can be extended to support insertions as well. More specifically, we can report the SCCs of a graph that is updated by insertions and deletions of $k$ edges in the same running time.
Related work
------------
The problem of maintaining the SCCs of a graph was studied in the decremental model. In this model the goal is to maintain the SCCs of a graph whose edges are being deleted by an adversary. The main parameters in this model are the worst case update time per an edge deletion and the total update from the first edge deletion until the last. Frigioni [et al.]{}[@FMNZ01] presented an algorithm that has an *expected* total update time of $O(mn)$ if all the deleted edges are chosen at random. Roditty and Zwick [@RoZw08] presented a Las-Vegas algorithm with an *expected* total update time of $O(mn)$ and *expected* worst case update time per a single edge deletion of $O(m)$. [Ł]{}cki [@Lacki11] presented a deterministic algorithm with a total update time of $O(mn)$, and thus solved the open problem posed by Roditty and Zwick in [@RoZw08]. However, the worst case update time per a single edge deletion of his algorithm is $O(mn)$. Roditty [@Roditty13] improved the worst case update time of a single edge deletion to $O(m\log n)$. Recently, in a major breakthrough, Henzinger, Krinninger and Nanongkai [@HenzingerKN14] presented a randomized algorithm with $O(mn^{0.9+o(1)})$ total update time. Very recently, Chechik et al. [@ChETAL] obtained a total update time of $O(m\sqrt{n \log n})$. Note that all the previous works on decremental SCC are with $\Omega(m)$ worst case update time. Whereas, our result directly implies $O(n \log^2 n)$ worst case update time as long as the total deletion length is constant.
Most of the previous work in the fault tolerant model is on variants of the shortest path problem. Demetrescu, Thorup, Chowdhury and Ramachandran [@DT:08] designed an $O(n^2 \log n)$ size data structure that can report the distance from $u$ to $v$ avoiding $x$ for any $u,v,x\in V$ in $O(1)$ time. Bernstein and Karger [@BK:9] improved the preprocessing time of [@DT:08] to $O(mn ~{\mbox{polylog}}~ n)$. Duan and Pettie [@DP:09] designed such a data structure for two vertex faults of size $O(n^2 \log n)$. Weimann and Yuster [@WY:13] considered the question of optimizing the preprocessing time using Fast Matrix Multiplication (FMM) for graphs with integer weights from the range $[-M,M]$. Grandoni and Vassilevska Williams [@GW:12] improved the result of [@WY:13] based on a novel algorithm for computing all the replacement paths from a given source vertex in the same running time as solving APSP in directed graphs.
For the problem of single source shortest paths Parter and Peleg [@ParterP:13] showed that there is a subgraph with $O(n^{3/2})$ edges that supports one fault. They also showed a matching lower bound. Recently, Parter [@Parter15] extended this result to two faults with $O(n^{5/3})$ edges for undirected graphs. She also showed a lower bound of $\Omega(n^{5/3})$.
Baswana and Khanna [@BaswanaK10-stacs] showed that there is a subgraph with $O(n \log n)$ edges that preserves the distances from $s$ up to a multiplicative stretch of $3$ upon failure of any single vertex. For the case of edge failures, sparse fault tolerant subgraphs exist for general $k$. Bil[ò]{} et al. [@Bilo16] showed that we can compute a subgraph with $O(kn)$ edges that preserves distances from $s$ up to a multiplicative stretch of $(2k +1)$ upon failure of any $k$ edges. They also showed that we can compute a data structure of $O(kn \log^2 n)$ size that is able to report the $(2k +1)$-stretched distance from $s$ in $O(k^2 \log^2 n)$ time.
The questions of finding graph spanners, approximate distance oracles and compact routing schemes in the fault tolerant model were studied in [@DK:11; @CR:12; @C:13; @ChechikCFK17].
Organization of the paper
-------------------------
We describe notations, terminologies, some basic properties of DFS, heavy-path decomposition, and ${\textsc{$k$-FTRS}}$ in Section 2. In Section 3, we describe the fault tolerant algorithm for computing the strongly connected components intersecting any path. We present our main algorithm for handling $k$ failures in Section 4. In Section 5, we show how to extend our algorithm and data structure to also support insertions.
Preliminaries
=============
Let $G=(V,E)$ denote the input directed graph on $n=|V|$ vertices and $m=|E|$ edges. We assume that $G$ is strongly connected, since if it is not the case, then we may apply our result to each strongly connected component of $G$. We first introduce some notations that will be used throughout the paper.
- $T$: A DFS tree of $G$.
- $T(v)$: The subtree of $T$ rooted at a vertex $v$.
- $Path(a,b)$: The tree path from $a$ to $b$ in $T$. Here $a$ is assumed to be an ancestor of $b$.
- $depth(Path(a,b))$: The depth of vertex $a$ in $T$.
- $G^R$: The graph obtained by reversing all the edges in graph $G$.
- $H(A)$: The subgraph of a graph $H$ induced by the vertices of subset $A$.
- $H\setminus F$: The graph obtained by deleting the edges in set $F$ from graph $H$.
- ${\textsc{In-Edges}}(v,H)$: The set of all incoming edges to $v$ in graph $H$.
- $P[a,b]$: The subpath of path $P$ from vertex $a$ to vertex $b$, assuming $a$ and $b$ are in $P$ and $a$ precedes $b$.
- $P{$\normalfont::$}Q$ : The path formed by concatenating paths $P$ and $Q$ in $G$. Here it is assumed that the last vertex of $P$ is the same as the first vertex of $Q$.
Our algorithm for computing SCCs in a fault tolerant environment crucially uses the concept of a $k$-fault tolerant reachability subgraph ($k$-FTRS) which is a sparse subgraph that preserves reachability from a given source vertex even after the failure of at most $k$ edges in $G$. A $k$-FTRS is formally defined as follows.
Let $s\in V$ be any designated source. A subgraph $H$ of $G$ is said to be a $k$-Fault Tolerant Reachability Subgraph (${\textsc{$k$-FTRS}}$) of $G$ with respect to $s$ if for any subset $F\subseteq E$ of $k$ edges, a vertex $v\in V$ is reachable from $s$ in $G\setminus F$ if and only if $v$ is reachable from $s$ in $H\setminus F$. \[definition:k-FTRS\]
In [@OUR-STOC], we present the following result for the construction of a ${\textsc{$k$-FTRS}}$ for any $k\ge 1$.
There exists an $O(2^kmn)$ time algorithm that for any given integer $k\geq 1$, and any given directed graph $G$ on $n$ vertices, $m$ edges and a designated source vertex $s$, computes a ${\textsc{$k$-FTRS}}$ for $G$ with at most $2^k n$ edges. Moreover, the in-degree of each vertex in this ${\textsc{$k$-FTRS}}$ is bounded by $2^k$. \[theorem:kftrs\]
Our algorithm will require the knowledge of the vertices reachable from a vertex $v$ as well as the vertices that can reach $v$. So we define a ${\textsc{$k$-FTRS}}$ of both the graphs - $G$ and $G^R$ with respect to any source vertex $v$ as follows.
- ${\cal G}^{}(v)$: The ${\textsc{$k$-FTRS}}$ of graph $G$ with $v$ as source obtained by Theorem \[theorem:kftrs\].
- ${\cal G}^{R}(v)$: The ${\textsc{$k$-FTRS}}$ of graph $G^R$ with $v$ as source obtained by Theorem \[theorem:kftrs\].
The following lemma states that the subgraph of a ${\textsc{$k$-FTRS}}$ induced by $A\subset V$ can serve as a ${\textsc{$k$-FTRS}}$ for the subgraph $G(A)$ given that $A$ satisfies certain properties.
Let $s$ be any designated source and $H$ be a ${\textsc{$k$-FTRS}}$ of $G$ with respect to $s$. Let $A$ be a subset of $V$ containing $s$ such that every path from $s$ to any vertex in $A$ is contained in $G(A)$. Then $H(A)$ is a ${\textsc{$k$-FTRS}}$ of $G(A)$ with respect to $s$. \[lemma:kftrs-set\]
Let $F$ be any set of at most $k$ failing edges, and $v$ be any vertex reachable from $s$ in $G(A)\setminus F$. Since $v$ is reachable from $s$ in $G\setminus F$ and $H$ is a ${\textsc{$k$-FTRS}}$ of $G$, so $v$ must be reachable from $s$ in $H \setminus F$ as well. Let $P$ be any path from $s$ to $v$ in $H\setminus F$. Then (i) all edges of $P$ are present in $H$ and (ii) none of the edges of $F$ appear on $P$. Since it is already given that every path from $s$ to any vertex in $A$ is contained in $G(A)$, therefore, $P$ must be present in $G(A)$. So every vertex of $P$ belongs to $A$. This fact combined with the inferences (i) and (ii) imply that $P$ must be present in $H(A)\setminus F$. Hence $H(A)$ is ${\textsc{$k$-FTRS}}$ of $G(A)$ with respect to $s$.
The next lemma is an adaptation of Lemma 10 from Tarjan’s classical paper on Depth First Search [@Tarjan72] to our needs.
Let $T$ be a DFS tree of $G$. Let $a,b\in V$ be two vertices without any ancestor-descendant relationship in $T$, and assume that $a$ is visited before $b$ in the DFS traversal of $G$ corresponding to tree $T$. Every path from $a$ to $b$ in $G$ must pass through a common ancestor of $a,b$ in $T$. \[lemma:DFS-property\]
Let us assume on the contrary that there exists a path $P$ from $a$ to $b$ in $G$ that does not pass through any common ancestor of $a$, $b$ in $T$. Let $z$ be the LCA of $a,b$ in $T$, and $w$ be the child of $z$ lying on $Path(z,a)$ in $T$. See Figure \[figure:dfs\]. Let $A$ be the set of vertices which are either visited before $w$ in $T$ or lie in the subtree $T(w)$, and $B$ be the set of vertices visited after $w$ in $T$. Thus $a$ belongs to set $A$, and $b$ belongs to set $B$. Let $x$ be the last vertex in $P$ that lies in set $A$, and $y$ be the successor of $x$ on path $P$. Since none of vertices of $P$ is a common ancestor of $a$ and $b$, therefore, the edge $(x,y)$ must belong to set $A\times B$. So the following relationship must hold true- $\textsc{Finish-Time}(x)\leq\textsc{Finish-Time}(w)<\textsc{Visit-Time}(y)$. But such a relationship is not possible since all the out-neighbors of $x$ must be visited before the DFS traversal finishes for vertex $x$. Hence we get a contradiction.
![ Depiction of vertices $a,b,z,w$ and sets $A$ (shown in orange) and $B$ (shown in purple).[]{data-label="figure:dfs"}](dfs.pdf)
A heavy path decomposition
--------------------------
The heavy path decomposition of a tree was designed by Sleator and Tarjan [@ST:83] in the context of dynamic trees. This decomposition has been used in a variety of applications since then. Given any rooted tree $T$, this decomposition splits $T$ into a set ${\cal P}$ of vertex disjoint paths with the property that any path from the root to a leaf node in $T$ can be expressed as a concatenation of at most $\log n$ sub-paths of paths in ${\cal P}$. This decomposition is carried out as follows. Starting from the root, we follow the path downward such that once we are at a node, say $v$, the next node traversed is the child of $v$ in $T$ whose subtree is of maximum size, where the size of a subtree is the number of nodes it contains. We terminate upon reaching a leaf node. Let $P$ be the path obtained in this manner. If we remove $P$ from $T$, we are left with a collection of subtrees each of size at most $n/2$. Each of these trees hang from $P$ through an edge in $T$. We carry out the decomposition of these trees recursively. The following lemma is immediate from the construction of a heavy path decomposition.
For any vertex $v\in V$, the number of paths in $\cal P$ which start from either $v$ or an ancestor of $v$ in $T$ is at most $\log n$. \[lemma:heavy-path\]
We now introduce the notion of ancestor path.
A path $Path(a_1,b_1)\in {\cal P}$ is said to be an ancestor path of $Path(a_2,b_2)\in {\cal P}$, if $a_1$ is an ancestor of $a_2$ in $T$.
In this paper, we describe the algorithm for computing SCCs of graph $G$ after any $k$ edge failures. Vertex failures can be handled by simply splitting a vertex $v$ into an edge $(v_{in},v_{out})$, where the incoming and outgoing edges of $v$ are directed to $v_{in}$ and from $v_{out}$, respectively.
Computation of SCCs intersecting a given path
=============================================
Let $F$ be a set of at most $k$ failing edges, and $X=(x_1,x_2,\ldots,x_t)$ be any path in $G$ from $x_1$ to $x_t$ which is intact in $G\setminus F$. In this section, we present an algorithm that outputs in $O(2^k n \log n)$ time the SCCs of $G\setminus F$ that intersect $X$.
For each $v\in V$, let $X^{\textsc{\scriptsize{in}}}(v)$ be the vertex of $X$ of minimum index (if exists) that is reachable from $v$ in $G\setminus F$. Similarly, let $X^{\textsc{\scriptsize{out}}}(v)$ be the vertex of $X$ of maximum index (if exists) that has a path to $v$ in $G\setminus F$. (See Figure \[figure:in\_out\]).
![ Depiction of $X^{\textsc{\scriptsize{in}}}(v)$ and $X^{\textsc{\scriptsize{out}}}(v)$ for a vertex $v$ whose SCC intersects $X$.[]{data-label="figure:in_out"}](in_out.pdf)
We start by proving certain conditions that must hold for a vertex if its SCC in $G\setminus F$ intersects $X$.
For any vertex $w\in V$, the SCC that contains $w$ in $G\setminus F$ intersects $X$ if and only if the following two conditions are satisfied.
\(i) Both $X^{\textsc{\scriptsize{in}}}(w)$ and $X^{\textsc{\scriptsize{out}}}(w)$ are defined, and
\(ii) Either $X^{\textsc{\scriptsize{in}}}(w)=X^{\textsc{\scriptsize{out}}}(w)$, or $X^{\textsc{\scriptsize{in}}}(w)$ appears before $X^{\textsc{\scriptsize{out}}}(w)$ on $X$. \[lemma:SCC-intersect-X\]
Consider any vertex $w\in V$. Let $S$ be the SCC in $G\setminus F$ that contains $w$ and assume $S$ intersects $X$. Let $w_1$ and $w_2$ be the first and last vertices of $X$, respectively, that are in $S$. Since $w$ and $w_1$ are in $S$ there is a path from $w$ to $w_1$ in $G\setminus F$. Moreover, $w$ cannot reach a vertex that precedes $w_1$ in $X$ since such a vertex will be in $S$ as well and it will contradict the definition of $w_1$. Therefore, $w_1=X^{\textsc{\scriptsize{in}}}(w)$. Similarly we can prove that $w_2=X^{\textsc{\scriptsize{out}}}(w)$. Since $w_1$ and $w_2$ are defined to be the first and last vertices from $S$ on $X$, respectively, it follows that either $w_1=w_2$, or $w_1$ precedes $w_2$ on $X$. Hence conditions (i) and (ii) are satisfied.
Now assume that conditions (i) and (ii) are true. The definition of $X^{\textsc{\scriptsize{in}}}(\cdot)$ and $X^{\textsc{\scriptsize{out}}}(\cdot)$ implies that there is a path from $X^{\textsc{\scriptsize{out}}}(w)$ to $w$, and a path from $w$ to $X^{\textsc{\scriptsize{in}}}(w)$. Also, condition (ii) implies that there is a path from $X^{\textsc{\scriptsize{in}}}(w)$ to $X^{\textsc{\scriptsize{out}}}(w)$. Thus $w$, $X^{\textsc{\scriptsize{in}}}(w)$ and $X^{\textsc{\scriptsize{out}}}(w)$ are in the same SCC and it intersects $X$.
The following lemma states the condition under which any two vertices lie in the same SCC, given that their SCCs intersect $X$.
Let $a,b$ be any two vertices in $V$ whose SCCs intersect $X$. Then $a$ and $b$ lie in the same SCC if and only if $X^{\textsc{\scriptsize{in}}}(a)=X^{\textsc{\scriptsize{in}}}(b)$ and $X^{\textsc{\scriptsize{out}}}(a)=X^{\textsc{\scriptsize{out}}}(b)$.
In the proof of Lemma \[lemma:SCC-intersect-X\], we show that if SCC of $w$ intersects $X$, then $X^{\textsc{\scriptsize{in}}}(w)$ and $X^{\textsc{\scriptsize{out}}}(w)$ are precisely the first and last vertices on $X$ that lie in the SCC of $w$. Since SCCs forms a partition of $V$, vertices $a$ and $b$ will lie in the same SCC if and only if $X^{\textsc{\scriptsize{in}}}(a)=X^{\textsc{\scriptsize{in}}}(b)$ and $X^{\textsc{\scriptsize{out}}}(a)=X^{\textsc{\scriptsize{out}}}(b)$.
It follows from the above two lemmas that in order to compute the SCCs in $G\setminus F$ that intersect with $X$, it suffices to compute $X^{\textsc{\scriptsize{in}}}(\cdot)$ and $X^{\textsc{\scriptsize{out}}}(\cdot)$ for all vertices in $V$. It suffices to focus on computation of $X^{\textsc{\scriptsize{out}}}(\cdot)$ for all the vertices of $V$, since $X^{\textsc{\scriptsize{in}}}(\cdot)$ can be computed in an analogous manner by just looking at graph $G^R$. One trivial approach to achieve this goal is to compute the set $V_i$ consisting of all vertices reachable from each $x_i$ by performing a BFS or DFS traversal of graph ${\cal G}(x_i)\setminus F$. Using this straightforward approach it takes $O(2^knt)$ time to complete the task of computing $X^{\textsc{\scriptsize{out}}}(v)$ for every $v\in V$, while our target is to do so in $O(2^kn \log n)$ time.
Observe the nested structure underlying $V_i$’s, that is, $V_1\supseteq V_2\supseteq \cdots \supseteq V_t$. Consider any vertex $x_\ell, 1<\ell <t$. The nested structure implies for every $v\in V_\ell$ that $X^{\textsc{\scriptsize{out}}}(v)$ must be on the portion $(x_\ell,\ldots,x_t)$ of $X$. Similarly, it implies for every $v\in V_1\setminus V_\ell$ that $X^{\textsc{\scriptsize{out}}}(v)$ must be on the portion $(x_1,\ldots,x_{\ell-1})$ of $X$. This suggests a divide and conquer approach to efficiently compute $X^{\textsc{\scriptsize{out}}}(\cdot)$. We first compute the sets $V_1$ and $V_t$ in $O(2^k n)$ time each. For each $v\in V\setminus V_1$, we assign NULL to $X^{\textsc{\scriptsize{out}}}(v)$ as it is not reachable from any vertex on $X$; and for each $v\in V_t$ we set $X^{\textsc{\scriptsize{out}}}(v)$ to $x_t$. For vertices in set $V_1\setminus V_t$, $X^{\textsc{\scriptsize{out}}}(\cdot)$ is computed by calling the function Binary-Search($1,t-1,V_1\setminus V_t$). See Algorithm \[Binary-Search\].
In order to explain the function Binary-Search, we first state an assertion that holds true for each recursive call of the function Binary-Search. We prove this assertion in the next subsection.
Assertion 1:
: If Binary-Search($i,j,A$) is called, then $A$ is precisely the set of those vertices $v\in V$ whose $X^{\textsc{\scriptsize{out}}}(v)$ lies on the path $(x_i,x_{i+1},\ldots,x_j)$.
We now explain the execution of function Binary-Search($i,j,A$). If $i=j$, then we assign $x_i$ to $X^{{\textsc{\scriptsize{out}}}}(v)$ for each $v\in A$ as justified by Assertion 1. Let us consider the case when $i\neq j$. In this case we first compute the index $mid = \lceil(i+j)/2\rceil$. Next we compute the set $B$ consisting of all the vertices in $A$ that are reachable from $x_{mid}$. This set is computed using the function Reach($x_{mid},A$) which is explained later in Subsection \[function\_Reach\]. As follows from Assertion 1, $X^{\textsc{\scriptsize{out}}}(v)$ for each vertex $v\in A$ must belong to path $(x_i,\ldots,x_j)$. Thus, $X^{{\textsc{\scriptsize{out}}}}(v)$ for all $v\in B$ must lie on path $(x_{mid},\ldots,x_j)$, and $X^{{\textsc{\scriptsize{out}}}}(v)$ for all $v\in A\setminus B$ must lie on path $(x_i,\ldots,x_{mid\text{-}1})$. So for computing $X^{\textsc{\scriptsize{out}}}(\cdot)$ for vertices in $A\setminus B$ and $B$, we invoke the functions Binary-Search($i,mid\text{-}1,A\text{$\setminus$} B$) and Binary-Search($mid,j,B$), respectively.
Proof of correctness of algorithm
---------------------------------
In this section we prove that Assertion 1 holds for each call of the Binary-Search function. We also show how this assertion implies that $X^{\textsc{\scriptsize{out}}}(v)$ is correctly computed for every $v\in V$.
Let us first see how Assertion 1 implies the correctness of our algorithm. It follows from the description of the algorithm that for each $i, (1\leq i\leq t-1)$, the function Binary-Search($i,i,A$) is invoked for some $A\subseteq V$. Assertion 1 implies that $A$ must be the set of all those vertices $v\in V$ such that $X^{\textsc{\scriptsize{out}}}(v)=x_i$. As can be seen, the algorithm in this case correctly sets $X^{\textsc{\scriptsize{out}}}(v)$ to $x_i$ for each $v\in A$.
We now show that Assertion 1 holds true in each call of the function Binary-Search. It is easy to see that Assertion 1 holds true for the first call Binary-Search($1,t-1,V_1\setminus V_t$). Consider any intermediate recursive call Binary-Search($i,j,A$), where $i\neq j$. It suffices to show that if Assertion 1 holds true for this call, then it also holds true for the two recursive calls that it invokes. Thus let us assume $A$ is the set of those vertices $v\in V$ whose $X^{\textsc{\scriptsize{out}}}(v)$ lies on the path $(x_i,x_{i+1},\ldots,x_j)$. Recall that we compute index $mid$ lying between $i$ and $j$, and find the set $B$ consisting of all those vertices in $A$ that are reachable from $x_{mid}$. From the nested structure of the sets $V_i,V_{i+1},\ldots, V_j$, it follows that $X^{{\textsc{\scriptsize{out}}}}(v)$ for all $v\in B$ must lie on path $(x_{mid},\ldots,x_j)$, and $X^{{\textsc{\scriptsize{out}}}}(v)$ for all $v\in A\setminus B$ must lie on path $(x_i,\ldots,x_{mid\text{-}1})$. That is, $B$ is precisely the set of those vertices whose $X^{\textsc{\scriptsize{out}}}(v)$ lies on the path $(x_{mid},\ldots,x_j)$, and $A\setminus B$ is precisely the set of those vertices whose $X^{\textsc{\scriptsize{out}}}(v)$ lies on the path $(x_i,\ldots,x_{mid\text{-}1})$. Thus Assertion 1 holds true for the recursive calls Binary-Search($i,mid\text{-}1,A\text{$\setminus$} B$) and Binary-Search($mid,j,B$) as well.
Implementation of function Reach {#function_Reach}
--------------------------------
The main challenge left now is to find an efficient implementation of the function Reach which has to compute the vertices of its input set $A$ that are reachable from a given vertex $x \in X$ in $G\setminus F$. The function Reach can be easily implemented by a standard graph traversal initiated from $x$ in the graph ${\cal G}(x)\setminus F$ (recall that ${\cal G}(x)$ is a ${\textsc{$k$-FTRS}}$ of $x$ in $G$). This, however, will take $O(2^k n)$ time which is not good enough for our purpose, as the total running time of Binary-Search in this case will become $O(|X|2^k n)$. Our aim is to implement the function Reach in $O(2^k|A|)$ time. In general, for an arbitrary set $A$ this might not be possible. This is because $A$ might contain a vertex that is reachable from $x$ via a single path whose vertices are not in $A$, therefore, the algorithm must explore edges incident to vertices that are not in $A$ as well. However, the following lemma, that exploits Assertion 1, suggests that in our case as the call to Reach is done while running the function Binary-Search we can restrict ourselves to the set $A$ only.
If $(i,j,A)$ is called and $\ell\in[i,j]$, then for each path $P$ from $x_\ell$ to a vertex $z\in A$ in graph in $G\setminus F$, all the vertices of $P$ must be in the set $A$. \[lemma:set\_A\]
Assertion 1 implies that $A$ is precisely the set of those vertices in $V$ which are reachable from $x_i$ but not reachable from $x_{j+1}$ in $G\setminus F$. Consider any vertex $y\in P$. Observe that $y$ is reachable from $x_i$ by the path $X[x_i,x_\ell]$::$P[x_\ell,y]$. Moreover, $y$ is not reachable from $x_{j+1}$, because otherwise $z$ will also be reachable from $x_{j+1}$, which is not possible since $z\in A$. Thus vertex $y$ lies in the set $A$.
Lemma \[lemma:set\_A\] and Lemma \[lemma:kftrs-set\] imply that in order to find the vertices in $A$ that are reachable from $x_{mid}$, it suffices to do traversal from $x_{mid}$ in the graph $G_A$, the induced subgraph of $A$ in ${\cal G}(x)\setminus F$, that has $O(2^k |A|)$ edges. Therefore, based on the above discussion, Algorithm \[Reach\] given below, is an implementation of function Reach that takes $O(2^k |A|)$ time.
$H \gets {\cal G}(x_{mid})\setminus F$ $G_A \gets (A,\emptyset)$ $B \gets$ Vertices reachable from $x_{mid}$ obtained by a BFS or DFS traversal of graph $G_A$ Return $B$
The following lemma gives the analysis of running time of Binary-Search($1,t-1,V_1\setminus V_t$).
The total running time of $(1,t-1,V_1\setminus V_t)$ is $O(2^{k} n \log n)$. \[lemma:time-binary-search\]
The time complexity of Binary-Search($1,t-1,V_1\setminus V_t$) is dominated by the total time taken by all invocation of function Reach. Let us consider the recursion tree associated with Binary-Search($1,t-1,V_1\setminus V_t$). It can be seen that this tree will be of height $O(\log n)$. In each call of the Binary-Search, the input set $A$ is partitioned into two disjoint sets. As a result, the input sets associated with all recursive calls at any level $j$ in the recursion tree form a disjoint partition of $V_1\setminus V_t$. Since the time taken by Reach is $O(2^{k}|A|)$, so the total time taken by all invocations of Reach at any level $j$ is $O(2^{k}|V_1\setminus V_t|)$. As there are at most $\log n$ levels in the recursion tree, the total time taken by Binary-Search($1,t-1,V_1\setminus V_t$) is $O(2^{k} n \log n)$.
We conclude with the following theorem.
\[theorem:specific\_case\] Let $F$ be any set of at most $k$ failed edges, and $X=\{x_1,x_2,\ldots,x_t\}$ be any path in $G\setminus F$. If we have prestored the graphs ${\cal G}(x)$ and ${\cal G}^R(x)$ for each $x\in X$, then we can compute all the SCCs of $G\setminus F$ which intersect with $X$ in $O(2^k n\log n)$ time.
Main Algorithm {#section:main-algo}
==============
In the previous section we showed that given any path $P$, we can compute all the SCCs intersecting $P$ efficiently, if $P$ is intact in $G\setminus F$. In the case that $P$ contains $\ell$ failed edges from $F$ then $P$ is decomposed into $\ell+1$ paths, and we can apply Theorem \[theorem:specific\_case\] to each of these paths separately to get the following theorem:
Let $P$ be any given path in $G$. Then there exists an $O(2^k n|P|)$ size data structure that for any arbitrary set $F$ of at most $k$ edges computes the SCCs of $G\setminus F$ that intersect the path $P$ in $O((\ell+1) 2^k n \log n)$ time, where $\ell~$ ($\ell\leq k$) is the number of edges in $F$ that lie on $P$. \[theorem:split-path\]
Now in order to use Theorem \[theorem:split-path\] to design a fault tolerant algorithm for SCCs, we need to find a family of paths, say $\cal P$, such that for any $F$, each SCC of $G\setminus F$ intersects at least one path in $\cal P$. As described in the Subsection \[subsection:overview\], a heavy path decomposition of DFS tree $T$ serves as a good choice for $\cal P$. Choosing $T$ as a DFS tree helps us because of the following reason: let $P$ be any root-to-leaf path, and suppose we have already computed the SCCs in $G\setminus F$ intersecting $P$. Then each of the remaining SCCs must be contained in some subtree hanging from path $P$. The following lemma formally states this fact.
Let $F$ be any set of failed edges, and $Path(a,b)$ be any path in $\cal P$. Let $S$ be any SCC in $G\setminus F$ that intersects $Path(a,b)$ but does not intersect any path that is an ancestor path of $Path(a,b)$ in $\cal P$. Then all the vertices of $S$ must lie in the subtree $T(a)$. \[lemma:subtree\]
Consider a vertex $u$ on $Path(a,b)$ whose SCC $S_u$ in $G\setminus F$ is not completely contained in the subtree $T(a)$. We show that $S_u$ must contain an ancestor of $a$ in $T$, thereby proving that it intersects an ancestor-path of $Path(a,b)$ in $\cal P$. Let $v$ be any vertex in $S_u$ that is not in the subtree $T(a)$. Let $P_{u,v}$ and $P_{v,u}$ be paths from $u$ to $v$ and from $v$ to $u$, respectively, in $G\setminus F$. From Lemma \[lemma:DFS-property\] it follows that either $P_{u,v}$ or $P_{v,u}$ must pass through a common ancestor of $u$ and $v$ in $T$. Let this ancestor be $z$. Notice also that since $P_{u,v}$ and $P_{v,u}$ form a cycle all their vertices are in $S_u$. Therefore, $u$ and $z$ are in the same SCC in $G\setminus F$. Moreover, since $v\notin T(a)$ and $u\in T(a)$, their common ancestor $z$ in $T$ is an ancestor of $a$. Since $z\in S_u$ and it is an ancestor of $a$ in $T$, the lemma follows.
Lemma \[lemma:subtree\] suggests that if we process the paths from ${\cal P}$ in the non-decreasing order of their depths, then in order to compute the SCCs intersecting a path $Path(a,b)\in \cal P$, it suffices to focus on the subgraph induced by the vertices in $T(a)$ only. This is because the SCCs intersecting $Path(a,b)$ that do not completely lie in $T(a)$ would have already been computed during the processing of some ancestor path of $Path(a,b)$.
We preprocess the graph $G$ as follows. We first compute a heavy path decomposition $\cal P$ of DFS tree $T$. Next for each path $Path(a,b)\in \cal P$, we use Theorem \[theorem:split-path\] to construct the data structure for path $Path(a,b)$ and the subgraph of $G$ induced by vertices in $T(a)$. We use the notation ${\cal D}_{a,b}$ to denote this data structure. Our algorithm for reporting SCCs in $G\setminus F$ will use the collection of these data structures associated with the paths in $\cal P$ as follows.
Let $\cal C$ denote the collection of SCCs in $G\setminus F$ initialized to $\emptyset$. We process the paths from ${\cal P}$ in non-decreasing order of their depths. Let $P(a,b)$ be any path in ${\cal P}$ and let $A$ be the set of vertices belonging to $T(a)$. We use the data structure ${\cal D}_{a,b}$ to compute SCCs of $G(A)\setminus F$ intersecting $P(a,b)$. Let these be $S_1,\ldots,S_t$. Note that some of these SCCs might be a part of some bigger SCC computed earlier. We can detect it by keeping a set $W$ of all vertices for which we have computed their SCCs. So if $S_i\subseteq W$, then we can discard $S_i$, else we add $S_i$ to collection $\cal C$. Algorithm \[main-algo\] gives the complete pseudocode of this algorithm.
$\cal C \gets \emptyset$ $W \gets \emptyset$ ${\cal P}\gets$ A heavy-path decomposition of $T$, where paths are sorted in the non-decreasing order of their depths Return $~\cal C$
Note that, in the above explanation, we only used the fact that $T$ is a DFS tree, and $\cal P$ could have been any path decomposition of $T$. We now show how the fact that $\cal P$ is a heavy-path decomposition is crucial for the efficiency of our algorithm. Consider any vertex $v\in T$. The number of times $v$ is processed in Algorithm \[main-algo\] is equal to the number of paths in $\cal P$ that start from either $v$ or an ancestor of $v$. For this number to be small for each $v$, we choose $\cal P$ to be a heavy path decomposition of $T$. On applying Theorem \[theorem:split-path\], this immediately gives that the total time taken by Algorithm \[main-algo\] is $O(k2^kn\log^2n)$. In the next subsection, we do a more careful analysis and show that this bound can be improved to $O(2^kn\log^2n)$.
Analysis of time complexity of Algorithm \[main-algo\] {#section:improved-time-complexity}
------------------------------------------------------
For any path $Path(a,b)\in \cal P$ and any set $F$ of failing edges, let $\ell(a,b)$ denote the number of edges of $F$ that lie on $Path(a,b)$. It follows from Theorem \[theorem:split-path\] that the time spent in processing $Path(a,b)$ by Algorithm \[main-algo\] is $O\big((\ell(a,b)+1)\times 2^k|T(a)| \times \log n\big)$. Hence the time complexity of Algorithm \[main-algo\] is of the order of
$$\sum_{Path(a,b)\in{\cal P}} (\ell(a,b)+1)\times 2^k|T(a)| \times \log n$$
In order to calculate this we define a notation $\alpha(v,Path(a,b))$ as $\ell(a,b)+1$ if $v\in T(a)$, and $0$ otherwise, for each $v\in V$ and $Path(a,b)\in \cal P$. So the time complexity of Algorithm \[main-algo\] becomes
$$~2^k \log n \times \Big(\sum_{Path(a,b)\in{\cal P}} (\ell(a,b)+1)\times |T(a)|\Big)$$ $$=2^k \log n \times \Big(\sum_{Path(a,b)\in{\cal P}}~ \sum_{v\in V} \alpha(v,Path(a,b)) \Big)$$ $$=2^k \log n \times \Big(\sum_{v\in V} ~\sum_{Path(a,b)\in{\cal P}} \alpha(v,Path(a,b)) \Big)$$
Observe that for any vertex $v$ and $Path(a,b)\in \cal P$, $\alpha(v,Path(a,b))$ is equal to $\ell(a,b)+1$ if $a$ is either $v$ or an ancestor of $v$, otherwise it is zero. Consider any vertex $v\in V$. We now show that $\sum_{Path(a,b)\in{\cal P}} \alpha(v,Path(a,b))$ is at most $k+\log n$. Let $P_v$ denote the set of those paths in $\cal P$ which starts from either $v$ or an ancestor of $v$. Then $\sum_{Path(a,b)\in{\cal P}} \alpha(v,Path(a,b)) = \sum_{Path(a,b)\in{P_v}} \ell(a,b)+1$. Note that $\sum_{Path(a,b)\in{P_v}} \ell(a,b)$ is at most $k$, and Lemma \[lemma:heavy-path\] implies that the number of paths in $P_v$ is at most $\log n$. This shows that $\sum_{Path(a,b)\in{\cal P}} \alpha(v,Path(a,b))$ is at most $k+\log n$ which is $O(\log n)$, since $k\leq \log n$.
Hence the time complexity of Algorithm \[main-algo\] becomes $O(2^k n\log^2 n)$. We thus conclude with the following theorem.
For any $n$-vertex directed graph $G$, there exists an $O(2^k n^2)$ size data structure that, given any set $F$ of at most $k$ failing edges, can report all the SCCs of $G\setminus F$ in $O(2^k n\log^2 n)$ time. \[theorem:main-theorem\]
Extension to handle insertion as well as deletion of edges {#section:updates}
==========================================================
In this section we extend our algorithm to incorporate insertion as well as deletion of edges. That is, we describe an algorithm for reporting SCCs of a directed graph $G$ when there are at most $k$ edge insertions and at most $k$ edge deletions.
Let $\cal D$ denote the $O(2^k n^2)$ size data structure, described in Section \[section:main-algo\], for handling $k$ failures. In addition to $\cal D$, we store the two ${\textsc{$k$-FTRS}}$: ${\cal G}^{}(v)$ and ${\cal G}^{R}(v)$ for each vertex $v$ in $G$. Thus the space used remains the same, i.e. $O(2^k n^2)$. Now let $U=(X,Y)$ be the ordered pair of $k$ updates, with $X$ being the set of failing edges and $Y$ being the set of newly inserted edges. Also let $|X|\le k$ and $|Y|\le k$.
$\cal C \gets$ SCCs of graph $G\setminus X$ computed using data structure $\cal D$ $S \gets$ Subset of $V$ consisting of endpoints of edges in $Y$ $H \gets \bigcup_{v\in S} \big({\cal G}^{}(v) + {\cal G}^{R}(v) + Y\big)$ Compute SCCs of graph $H\setminus X$ using any standard static algorithm
Our first step is to compute the collection $\cal C$, consisting of SCCs of graph $G\setminus X$. This can be easily done in $O(2^k n \log^2 n)$ time using the data structure $\cal D$. Now on addition of set $Y$, some of the SCCs in $\cal C$ may get merged into bigger SCCs. Let $S$ be the subset of $V$ consisting of endpoints of edges in $Y$. Note that if the SCC of a vertex gets altered on addition of $Y$, then its new SCC must contain at least one edge from $Y$, and thus also a vertex from set $S$. Therefore, in order to compute SCCs of $G+U$, it suffices to recompute only the SCCs of vertices lying in the set $S$.
Let $H$ be a graph consisting of edge set $Y$, and the ${\textsc{$k$-FTRS}}$ ${\cal G}^{}(v)$ and ${\cal G}^{R}(v)$, for each $v\in S$. Then $SCC_{H\setminus X}(v)=SCC_{G+U}(v)$, for each $v\in S$. \[lemma:k-updates\]
Consider a vertex $v\in S$. Since $H\setminus X\subseteq G+U$, $SCC_{H\setminus X}(v)\subseteq SCC_{G+U}(v)$. We show that $SCC_{H\setminus X}(v)$ is indeed equal to $SCC_{G+U}(v)$.
Let $w$ be any vertex reachable from $v$ in $G+U$, by a path, say $P$. Our aim is to show that $w$ is reachable from $v$ in $H\setminus X$ as well. Notice that we can write $P$ as $(P_1{$\normalfont::$}e_1 {$\normalfont::$}P_2 {$\normalfont::$}e_2 \cdots e_{\ell-1} {$\normalfont::$}P_\ell)$, where $e_1,\ldots,e_{\ell-1}$ are edges in $Y\cap P$ and $P_1,\ldots,P_\ell$ are segments of $P$ obtained after removal of edges of set $Y$. Thus $P_1,\ldots,P_\ell$ lie in $G\setminus X$. For $i=1$ to $\ell$, let $a_i$ and $b_i$ be respectively the first and last vertices of path $P_i$. Since $a_1=v$ and $a_2,\ldots,a_\ell\in S$, the ${\textsc{$k$-FTRS}}$ of all the vertices $a_1$ to $a_\ell$ is contained in $H$. Thus for $i=1$ to $\ell$, vertex $b_i$ must be reachable from $a_i$ by some path, say $Q_i$, in graph $H\setminus X$. Hence $Q = (Q_1{$\normalfont::$}e_1 {$\normalfont::$}Q_2 \cdots e_{\ell-1} {$\normalfont::$}Q_\ell)$ is a path from $a_1=v$ to $b_\ell=w$ in graph $H\setminus X$.
In a similar manner we can show that if a vertex $w'$ has a path to $v$ in graph $G+U$, then $w'$ will also have path to $v$ in graph $H\setminus X$. Thus $SCC_{H\setminus X}(v)$ must be equal to $SCC_{G+U}(v)$.
So we compute the auxiliary graph $H$ as described in Lemma \[lemma:k-updates\]. Note that $H$ contains only $O(k 2^k n)$ edges. Next we compute the SCCs of graph $H\setminus X$ using any standard algorithm [@Cormen] that runs in time which is linear in terms of the number of edges and vertices. This algorithm will take $O(2^k n \log n)$ time, since $k$ is at most $\log n$. Finally, for each $v\in S$, we check if the $SCC_{H\setminus X}(v)$ has broken into smaller SCCs in $\cal C$, if so, then we merge all of them into a single SCC. We can accomplish this entire task in a total $O(nk)$ time only. This completes the description of our algorithm. For the pseudocode see Algorithm \[algo:k-updates\].
We conclude with the following theorem.
For any $n$-vertex directed graph $G$, there exists an $O(2^k n^2)$ size data structure that, given any set $U$ of at most $k$ edge insertions and at most $k$ edge deletions, can report the SCCs of graph $G+U$ in $O(2^k n\log^2 n)$ time. \[theorem:k-updates\]
[^1]: We note that the heavy path decomposition was also used in the fault tolerant model in STACS’10 paper of [@BaswanaK10-stacs], but in a completely different way and for a different problem.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, we propose a generalized notion of a distance function, called a $g$-metric. The $g$-metric with degree $n$ is a distance of $n+1$ points, generalizing the ordinary distance between two points and $G$-metric between three points. Indeed, it is shown that the $g$-metric with degree 1 (resp. degree 2) is equivalent to the ordinary metric (resp. the $G$-metric). Fundamental properties and several examples for the $g$-metric are also given. Moreover, topological properties on the $g$-metric space including the convergence of sequences and the continuity of mappings on the $g$-metric space are studied. Finally, we generalize some well-known fixed point theorems including Banach contraction mapping principle and Ćirić fixed point theorem in the $g$-metric space.'
address:
- 'School of Information Science and Technology, ShanghaiTech University, Pudong district, Shanghai 200031, China'
- 'Department of Mathematics, Chungbuk National University, Cheongju 361-763, Republic of Korea'
- 'Department of Mathematics, University of Denver, Denver, CO 80208, USA'
author:
- Hayoung Choi
- Sejong Kim
- Seung Yeop Yang
title: 'Structure for $g$-Metric Spaces and Related Fixed Point Theorems '
---
generalized metric, g-metric, G-metric, distance function, fixed point
Introduction and preliminaries
==============================
A distance is a measurement how far apart each pair elements of a given set are. The distance function in mathematics and many other scientific fields is a crucial concept. For instance, the distance function is used to quantify a dissimilarity (or equivalently similarity) between two objects in some sense. However, due to massive and complicated data sets today, the definition of a distance function is required to be generalized.
Numerous ways to generalize the notion of a distance function have been studied [@AKOH15; @Khamsi2015]. Among them, we consider the $G$-metric space, which allows us to establish many topological properties. Moreover, a variety of the fixed point theorems on the $G$-metric space have extensively been studied. In this paper we give a generalized notion of a distance function between $n+1$ points, called a $g$-metric. It coincides with the ordinary distance between two points and with the $G$-metric between three points. Furthermore, we establish topological notions and properties on the $g$-metric space including the convergence of sequences and continuity of mappings. From these topology on the $g$-metric space we generalize some well-known fixed point theorems such as the Banach contraction mapping principle, weak contraction mapping principle, and Ćirić fixed point theorem.
Let ${\mathbb{N}}$ (resp. ${\mathbb{R}}$) be the set of all nonnegative integers (resp. all real numbers), and let ${\mathbb{R}}$ be the set of all real numbers. We denote as ${\mathbb{R}}_{+}$ the set of all nonnegative real numbers. For a finite set $A$, we denote the number of distinct elements of $A$ by $n(A)$.
The definition of a distance function was proposed by M. Fréchet [@Frechet06] in 1906.
\[D:distance\] Let $\Omega$ be a nonempty set. A function $d: \Omega \times \Omega \longrightarrow {\mathbb{R}}_{+}$ is called a *distance function* or *metric* on $\Omega$ if it satisfies the following conditions:
- (identity) $d(x,y)=0$ if $x=y$,
- (non-negativity) $d(x,y)>0$ if $x \neq y$,
- (symmetry) $d(x,y) = d(y,x)$ for all $x, y \in \Omega$,
- (triangle inequality) $d(x,y) \leq d(x,z) + d(z,y)$ for all $x, y, z \in \Omega$.
The pair $(\Omega, d)$ is called a *metric space*.
The first attempt to generalize the ordinary distance function to a distance of three points was introduced by Gahler [@Gahler63; @Gahler66] in 1963.
\[D:Gahler\] Let $\Omega$ be a nonempty set. A function $d:\Omega\times \Omega \times \Omega \longrightarrow {\mathbb{R}}_+$ is called a *2-metric* on $\Omega$ if it satisfies the following properties:
- For all $x,y \in \Omega$ with $x \neq y$, there is $z\in \Omega$ such that $d(x,y,z) \neq 0$,
- $d(x,y,z)=0$ if $x=y,$ $y=z,$ or $z=x,$
- $d(x,y,z) = d(x,z,y) = \cdots$ (symmetry in all three variables),
- $d(x,y,z) \leq d(x,y,w) + d(x,w,z) + d(w,y,z)$ for all $x, y, z, w \in \Omega$.
The pair $(\Omega, d)$ is called a *2-metric space*.
\[ex:2-metric\] We give a few examples of a $2$-metric.
- Let $d(x,y,z)$ be the area of the triangle with vertices at $x,y,z\in {\mathbb{R}}^2$. Then $d$ is a 2-metric on ${\mathbb{R}}^{2}$ [@MS06].
- For a given metric $\delta$ on a set $\Omega$ with $n(\Omega) \geq 3$, define $d: \Omega \times \Omega \times \Omega \longrightarrow {\mathbb{R}}_+$ given by $$d(x,y,z) = \min\{ \delta(x,y), \delta(y,z), \delta(z,x) \},$$ for all $x,y,z\in \Omega$. Then $(\Omega, d)$ is a 2-metric space. The proof is left to the reader.
- Let $\Omega$ be a set with $n(\Omega) \geq 3$. Define $d: \Omega \times \Omega \times \Omega \longrightarrow {\mathbb{R}}_+$ given by $$d(x,y,z) =
\begin{cases}
0, & \hbox{if $x=y,$ $y=z,$ or $z=x$;}\\
1, & \hbox{otherwise.}
\end{cases}$$ Then, it is easy to check that $d$ is a 2-metric on $\Omega$.
It was shown that a 2-metric is not a generalization of the usual notion of a metric [@HCW88]. Dhage in 1992 introduced a new class of generalized metrics called D-metrics [@dhage92].
Let $\Omega$ be a nonempty set. A function $D: \Omega \times \Omega \times \Omega \longrightarrow {\mathbb{R}}_+$ is called a *D-metric* on $\Omega$ if it satisfies the following conditions:
- $D(x,y,z)=0$ if and only if $x=y=z$,
- $D(x,y,z) = D(x,z,y) =\cdots $ (symmetry in all three variables),
- $D(x,y,z) \leq D(x,y,w) + D(x,w,z) + D(w,y,z)$ for all $x,y,z,w\in \Omega,$
- $D(x, y, y) \leq D(x, z, z) + D(z, y, y)$ for all $x, y, z \in \Omega$.
The pair $(\Omega,D)$ is called a *D-metric space*.
\[example:D-metric\] For any given metric space $(\Omega, \delta)$ the following are $D$-metrics [@dhage92].
- $\displaystyle D(x, y, z) = \frac{1}{3} (\delta(x, y) + \delta(y, z) + \delta(x, z)).$
- $D(x, y, z) = \max\{ \delta(x, y), \delta(y, z), \delta(x, z) \}.$
Topological structures and fixed points in a $D$-metric space have been studied. However, several errors for fundamental topological properties in a $D$-metric space were found [@MS03; @NRS05]. Due to these considerations, Mustafa and Sims [@MS06] proposed a more appropriate notion of a generalized metric space. For more information, see [@AKOH15] and references therein.
Let $\Omega$ be a nonempty set. A function $G:\Omega\times \Omega \times \Omega \longrightarrow {\mathbb{R}}_+$ is called a *G-metric* on $\Omega$ if it satisfies the following conditions:
- $G(x,y,z)=0$ if $x=y=z$,
- $G(x,x,y) > 0$ for all $x,y \in \Omega$ with $x \neq y$,
- $G(x, x, y) \leq G(x, y, z)$ for all $x, y, z \in \Omega$ with $y \neq z$,
- $G(x,y,z) = G(x,z,y) = \cdots $ (symmetry in all three variables $x,y,z$),
- $G(x,y,z) \leq G(x,w,w) + G(w,y,z)$ for all $x,y,z,w \in \Omega$.
The pair $(\Omega,G)$ is called a *G-metric space*. A $G$-metric space $(\Omega,G)$ is said to be *symmetric* if
- $G(x,y,y) = G(x,x,y)$ for all $x,y\in \Omega$.
The following are G-metrics.
- Let $d(x,y,z)$ be the perimeter of the triangle with vertices at $x,y,z$ on ${\mathbb{R}}^2$. Then $({\mathbb{R}}^2,d)$ is a $G$-metric space. More generally, for a given normed space $(\Omega, \| \cdot \|)$, define $$d(x,y,z)= c(\|x-y \|+ \|y-z \| + \|z-x \|)$$ for $x,y,z \in \Omega$ with a fixed $c>0$. Then by Theorem \[thm:relation\] (2) and Example \[Ex:average g-metric\] it follows that $d$ is a $G$-metric.
- Let $\Omega = \{ x,y \}$ and let $G(x,x,x)=G(y,y,y)=0$, $G(x,x,y)=1,G(x,y,y)=2$ and extend $G$ to all of $\Omega\times \Omega \times \Omega$ by symmetry in the variables. Then $G$ is a $G$-metric which is not symmetric [@MS06].
- By Theorem \[thm:relation\] (2) and Example \[Ex:max g-metric\], it is clear that Example \[example:D-metric\] (2) is a $G$-metric.
Theory of a $g$-metric
======================
Now we propose a new definition of a generalized metric for $n$ points instead of two or three points in a given set. For a set $\Omega$, we denote $\displaystyle \Omega^n := \prod_{i=1}^n \Omega.$
\[D:g-metric space\] Let $\Omega$ be a nonempty set. A function $g:\Omega^{n+1} \longrightarrow {\mathbb{R}}_+$ is called a *$g$-metric with order $n$* on $\Omega$ if it satisfies the following conditions:
- (positive definiteness) $g(x_0, \ldots, x_{n})=0$ if and only if $x_0 = \cdots = x_{n}$,
- (permutation invariancy) $g(x_0, \ldots, x_{n}) = g(x_{\sigma(0)}, \ldots, x_{\sigma(n)})$ for any permutation $\sigma$ on $\{ 0, 1, \ldots, n \}$,
- (monotonicity) $g(x_0, \ldots, x_{n}) \leq g(y_0, \ldots, y_{n})$ for all $(x_0, \ldots, x_{n}), (y_0, \ldots, y_{n}) \in \Omega^{n+1}$ with $\{ x_i : i = 0, \ldots, n \} \subsetneq \{ y_i : i = 0, \ldots, n \}$,
- (triangle inequality) for all $x_0, \ldots, x_s, y_0, \ldots, y_t, w \in \Omega$ with $s + t + 1 = n$ $$g(x_0, \ldots, x_s, y_0, \ldots, y_t) \leq g(x_0, \ldots, x_s, w, \ldots, w) + g(y_0, \ldots, y_t, w, \ldots, w).$$
The pair $(\Omega, g)$ is called a *$g$-metric space*.
A $g$-metric on $\Omega$ is called *multiplicity-independent* if the following holds $$g(x_0, \ldots, x_{n}) = g(y_0, \ldots, y_{n})$$ for all $(x_0, \ldots, x_{n}), (y_0, \ldots, y_{n}) \in \Omega^{n+1}$ with $\{ x_i : i=0, \ldots, n \} = \{y_i : i=0, \ldots, n \}$.
Note that for a given multiplicity-independent $g$-metric with order 2, it holds that $g(x,y,y)=g(x,x,y)$. For a given multiplicity-independent $g$-metric with order 3, it holds that $g(x,y,y,y) = g(x,x,y,y) = g(x,x,x,y)$ and $g(x,x,y,z) = g(x,y,y,z) = g(x,y,z,z)$. This is why we call it multiplicity-independent rather than symmetric which was proposed in the $G$-metric.
If we allow equality under the condition of monotonicity in Definition \[D:g-metric space\], i.e., “$g(x_0, \ldots, x_{n}) \leq g(y_0, \ldots, y_{n})$ for all $(x_0, \ldots, x_{n}), (y_0, \ldots, y_{n}) \in \Omega^{n+1}$ with $\{ x_i : i = 0, \ldots, n \} \subseteq \{ y_i : i = 0, \ldots, n \}$", then every $g$-metric becomes multiplicity-independent.
Let us explain why the condition $(g4)$ can be considered as a generalization of the triangle inequality. Recall that the triangle inequality condition for a distance function $d$ is $d(x,y) \leq d(x,z) + d(z,y)$ for all $x,y,z$.
at (0,-0.2) (x3) ; at (1,-0.2) (x4) ; at (2,-0.2) (x5) ; at (0.5,1) (x2) ; at (1.5,1) (x6) ; at (1,2) (x1) ; at (x4) (E2) ; at (x2) (E3) ; at (x6) (E1) ;
The point $w$ is required to measure approximately the distance between $x$ and $y$ with the distances between $x$ and $w$ and between $w$ and $y$. Note that one cannot measure the distance between $x$ and $y$ by the distances $d(x,w_1)$ and $d(y,w_2)$ with $w_1 \neq w_2$. Consider $d(x,y)$ as a dissimilarity between $x$ and $y$. Clearly, if $x=y$, then the dissimilarity is 0, vice versa. Also, the dissimilarity between $x$ and $y$ is same as the dissimilarity between $y$ and $x$. If $x$ (resp. $y$) and $z$ (resp. $z$) are sufficiently similar, then by the triangle inequality $x$ and $y$ must be sufficiently similar.
In the similar way, one can generalize the definition of triangle inequality for the $g$-metric. Specifically, one can see from the definition of triangle inequality for the $g$-metric that if both $g(x_0, \ldots, x_s, w, \ldots, w)$ and $g(y_0, \ldots, y_t, w, \ldots, w)$ are sufficiently small, then $g(x_0, \ldots, x_s, y_0, \ldots, y_t)$ must be sufficiently small. That is, the higher similarities two data sets $\{ x_0, \ldots, x_s, w \}$ and $\{ y_0, \ldots, y_t, w \}$ have, the higher similarity data set $ \{ x_0, \ldots, x_s, y_0, \ldots, y_t \} $ does. Note that $w$ is a necessary point to combine information about similarity for each data set.
The following theorem shows us that $g$-metrics generalize the notions of ordinary metric and $G$-metric.
\[thm:relation\] Let $\Omega$ be a given nonempty set. The following are true.
- $d$ is a $g$-metric with order 1 on $\Omega$ if and only if $d$ is a metric on $\Omega$.
- $d$ is a (resp. multiplicity-independent) $g$-metric with order 2 on $\Omega$ if and only if $d$ is a (resp. symmetric) $G$-metric on $\Omega$.
<!-- -->
- By the definition of $g$-metric, $d$ is a $g$-metric with order 1 if and only if it satisfies the following conditions.
- $d(x,y)=0$ if and only if $x=y$,
- $d(x,y) = d(y,x)$,
- $0 = d(x,x) \leq d(x,y)$ if $x \neq y$,
- $d(x,z) \leq d(x,y) + d(y,z)$ for all $x,y,z \in \Omega$.
Clearly, these are equivalent to the axioms for a distance function.
- By the definition of $g$-metric, $d$ is a $g$-metric with order 2 if and only if it satisfies the following conditions.
- $d(x,y,z)=0$ if and only if $x=y=z$,
- $d(x,y,z) = d(x,z,y) = \cdots $ (symmetry in all three variables),
- $d(x,x,y) \leq d(x,y,z)$ for all $x,y,z \in \Omega $ with $y \neq z$,
- $d(x,y,z) \leq d(x,w,w) + d(y,z,w) $ for all $x,y,z,w \in \Omega$.
These are exactly same as the conditions of $G$-metric. Thus, the $g$-metric with order 2 and the $G$-metric are identical. Moreover, one can see easily the equivalence between the multiplicity-independence of a $g$-metric and the symmetry of a $G$-metric.
Remark that since a $g$-metric with order 2 on a nonempty set $\Omega$ is a $G$-metric, any $g$-metrics with order 2 satisfy all properties of the $G$-metric as shown in [@MS06]. Moreover, if $d$ is a $g$-metric with order 2 on $\Omega$, then $d$ is also a $D$-metric.
Next, we show that an explicit form of conditions for a $g$-metric with order 3.
$d$ is a $g$-metric with order 3 on $\Omega$ if and only if the following conditions hold:
- $g(x,y,p,q) = 0$ if and only if $x=y=p=q$.
- $g(x,y,p,q) = g(y,x,p,q) = g(p,y,x,q) =g(q,y,p,x) = g(x,p,y,q) = g(x,y,q,p) $.
- $g(x,y,y,y) \leq g(x,x,y,p)$,\
$g(x,y,y,y) \leq g(x,y,y,p)$,\
$g(x,y,y,y) \leq g(x,y,p,p)$,\
$g(x,x,y,y) \leq g(x,x,y,p)$,\
$g(x,x,y,y) \leq g(x,y,y,p)$,\
$g(x,x,y,y) \leq g(x,y,p,p)$,\
$g(x,y,p,p) \leq g(x,y,p,q)$\
for all distinct $x,y,p,q\in \Omega$.
- $g(x,y,p,q) \leq g(x,w,w,w) + g(y,p,q,w)$,\
$g(x,y,p,q) \leq g(x,y,w,w) + g(p,q,w,w)$ for all $x,y,p,q,w\in \Omega$.
It is easy to show that the condition (1) (resp. (2)) is equivalent to $(g1)$ (resp. $(g2)$). For $(g4)$ since $s+t=2$, there are three possibilities: (i) $s=0$ and $t=2$, (ii) $s=1$ and $t=1$, and (iii) $s=2$ and $t=0$. Since $x,y,p,q,w$ are arbitrary, (i) and (iii) are equivalent. Thus we have two different inequalities given in (4). For $(g3)$ let $X=\{ x_0, x_1,x_2,x_3 \}$ and $Y=\{y_0, y_1,y_2,y_3 \}$ with $X \subsetneq Y$. There are six possibilities: (i) $n(X)=1$ and $n(Y)=2$; (ii) $n(X)=1$ and $n(Y)=3$; (iii) $n(X)=1$ and $n(Y)=4$; (iv) $n(X)=2$ and $n(Y)=3$; (v) $n(X)=2$ and $n(Y)=4$; (vi) $n(X)=3$ and $n(Y)=4$. Note that $(1)$ implies the condition $(g3)$ for the cases (i), (ii), and (iii). Also the condition $(g3)$ for (iv) and (vi) implies the condition $(g3)$ for (v). Thus the conditions $(g1),(g2),(g3)$, and $(g4)$ are equivalent the conditions $(1),(2),(4)$, and (iv) and (vi). For (iv), let $X=\{x,y\}$ and $Y=\{x,y,p\}$. Then, we have for all distinct $x,y,p\in \Omega$, $$\begin{aligned}
g(x,y,y,y) \leq g(x,x,y,p),\quad & g(x,y,y,y) \leq g(x,y,y,p),\\
g(x,y,y,y) \leq g(x,y,p,p),\quad & g(x,x,y,y) \leq g(x,x,y,p),\\
g(x,x,y,y) \leq g(x,y,y,p),\quad & g(x,x,y,y) \leq g(x,y,p,p).\end{aligned}$$
For (vi), let $X=\{x,y,p\}$ and $Y=\{x,y,p,q\}$. Then, we have $g(x,y,p,p) \leq g(x,y,p,q)$ for all distinct $x,y,p,q\in \Omega$.
A new $g$-metric can be constructed from given $g$-metrics.
\[lemma:new\_g-metric\] Let $(\Omega, g)$ and $(\Omega, \tilde{g})$ be $g$-metric spaces. Then the following functions, denoted by $d$, are $g$-metrics on $\Omega$.
- $d(x_0, x_1, \ldots, x_{n}) = g(x_0, x_1, \ldots, x_{n}) + \tilde{g}(x_0, x_1, \ldots, x_{n})$.
- $d(x_0, x_1, \ldots, x_{n}) = \psi(g(x_0, x_1, \ldots, x_{n}))$ where $\psi $ is a function on $[0,\infty)$ satisfies
- $\psi $ is increasing on $[0,\infty)$;
- $\psi(0)=0 $;
- $\psi (x+y) \leq \psi (x) + \psi (y)$ for all $x,y\in [0,\infty)$.
<!-- -->
- It is easy to check that $$\begin{aligned}
d(x_0, x_1, \ldots, x_{n}) = 0
&\Longleftrightarrow g(x_0, x_1, \ldots, x_{n})=0 \text{ and } \tilde{g}(x_0, x_1, \ldots, x_{n})=0\\
&\Longleftrightarrow x_0 = x_1= \cdots = x_{n}.\end{aligned}$$ So, $d$ holds the condition $(g1)$. It is clear that $d$ holds the condition $(g2)$. Let $\{x_i: i=0,\ldots, n\} \subsetneq \{y_i: i=0,\ldots, n\}$. Since $g$ and $\tilde{g}$ are $g$-metrics, it follows that $$\begin{aligned}
d(x_0, x_1, \ldots, x_{n})
&= g(x_0, x_1, \ldots, x_{n}) + \tilde{g}(x_0, x_1, \ldots, x_{n})\\
&\leq g(y_0, y_1, \ldots, y_{n}) + \tilde{g}(y_0, y_1, \ldots, y_{n}) \\
&=d(y_0, y_1, \ldots, y_{n}).\end{aligned}$$ Thus $d$ satisfies the condition $(g3)$. Let $x_0, \ldots, x_s, y_0, \ldots, y_t, w \in \Omega$ with $s + t + 1 = n$. Then it follows that $$\begin{aligned}
d(x_0, \ldots, x_s, y_0, \ldots, y_t)
&= g(x_0, \ldots, x_s, y_0, \ldots, y_t) + \tilde{g}(x_0, \ldots, x_s, y_0, \ldots, y_t)\\
&\leq g(x_0, \ldots, x_s, w, \ldots, w) + g(y_0, \ldots, y_t, w, \ldots, w) \\
&\quad+ \tilde{g}(x_0, \ldots, x_s, w, \ldots, w) + \tilde{g}(y_0, \ldots, y_t, w, \ldots, w)\\
& \leq d(x_0, \ldots, x_s, w, \ldots, w) + d(y_0, \ldots, y_t, w, \ldots, w).\end{aligned}$$ Thus it satisfies the condition $(g4)$. Therefore, $d$ is a $g$-metric.
- Since $\psi(x)=0$ if and only if $x=0$, it holds the condition $(g1)$. It is clear that $d$ holds the condition $(g2)$. Let $\{x_i: i=0,\ldots, n\} \subsetneq \{y_i: i=0,\ldots, n\}$. Since $g$ is a $g$-metric, it follows that $g(x_0, x_1, \ldots, x_{n}) \leq g(y_0, y_1, \ldots, y_{n})$. Since $\psi$ is increasing on $[0,\infty)$, $$\begin{aligned}
d(x_0, x_1, \ldots, x_{n})
&= \psi(g(x_0, x_1, \ldots, x_{n}))\\
&\leq \psi(g(y_0, y_1, \ldots, y_{n})) \\
&=d(y_0, y_1, \ldots, y_{n}).\end{aligned}$$ Thus such $d$ satisfies the condition $(g3)$. Let $x_0, \ldots, x_s, y_0, \ldots, y_t, w \in \Omega$ with $s + t + 1 = n$. Then it follows that $$\begin{aligned}
d(x_0, \ldots, x_s, y_0, \ldots, y_t)
&= \psi(g(x_0, \ldots, x_s, y_0, \ldots, y_t)) \\
& \leq \psi(g(x_0, \ldots, x_s, w, \ldots, w) + g(y_0, \ldots, y_t, w, \ldots, w)) \\
& \leq \psi(g(x_0, \ldots, x_s, w, \ldots, w)) + \psi(g(y_0, \ldots, y_t, w, \ldots, w))\\
& =d(x_0, \ldots, x_s, w, \ldots, w) + d(y_0, \ldots, y_t, w, \ldots, w).\end{aligned}$$ Thus it satisfies the condition $(g4)$. Therefore, $d$ is a $g$-metric.
\[ex:newold:g-metric\] The following functions, denoted by $\psi$, satisfy the conditions in Lemma \[lemma:new\_g-metric\] (2). Thus, each $\psi \circ g$ is a $g$-metrics for any $g$-metric $g$.
- $(\psi \circ g)(x_0, \ldots, x_n) =kg(x_0, \ldots, x_n)$ where $\psi(x)=kx$ with a fixed $k>0$.
- $(\psi \circ g)(x_0, \ldots, x_n) = \dfrac{g(x_0, \ldots, x_n)}{1+g(x_0, \ldots, x_n)}$ where $\psi(x)=\dfrac{x}{1+x}$.
- $(\psi \circ g)(x_0, \ldots, x_n) = \sqrt{g(x_0, \ldots, x_n)}$ where $\psi(x)=\sqrt{x}$. Furthermore, it is true for $\psi(x)=x^{1/p}$ with a fixed $p \geq1$.
- $(\psi \circ g)(x_0, \ldots, x_n) =\log{(g(x_0, \ldots, x_n)+1)}$ where $\psi(x)=\log{(x+1)}$.
- $(\psi \circ g)(x_0, \ldots, x_n) =\min\{ k, g(x_0, \ldots, x_n) \}$ where $\psi(x)=\min\{ k, x \}$ with a fixed $k>0$.
We give several interesting examples of $g$-metric on a variety of settings in the following.
(Discrete $g$-metric) For a nonempty set $\Omega$, define $d: \Omega^{n+1} \rightarrow {\mathbb{R}}_+$ by $$d(x_0, \ldots, x_n)=
\begin{cases}
0 \quad \text{if } x_{0} = \cdots = x_{n}, \\
1 \quad \text{otherwise }
\end{cases}$$ for all $x_0, \ldots, x_n \in \Omega$. Then $d$ is a $g$-metric on $\Omega$.
It is trivial that $d$ satisfies the conditions $(g1)$ and $(g2)$.
- Let $x_0, \ldots,~ x_n,~y_0, \ldots,~ y_n \in \Omega$ such that $\{x_0, \ldots, x_n \} \subsetneq \{y_0, \ldots, y_n \}$. If $n(\{ x_0, \ldots, x_n \}) = 1$, then $x_{0} = \cdots = x_{n}$, and so\
$g(x_0, \ldots, x_n) = 0 < 1 = g(y_0, \ldots, y_n)$. If $n(\{ x_0, \ldots, x_n \}) > 1$, then $g(x_0, \ldots, x_n) = 1 = g(y_0, \ldots, y_n)$.
- Let $x_0, \ldots,~ x_s,~y_0, \ldots,~y_t,~w \in \Omega$ with $s + t + 1 = n$. If $d(x_0, \ldots, x_s, w, \ldots, w) = 1$ or $d(y_0, \ldots, y_t, w, \ldots, w) = 1$, then $$g(x_0, \ldots, x_s, y_0, \ldots, y_t) \leq 1 \leq g(x_0, \ldots, x_s, w, \ldots, w) + g(y_0, \ldots, y_t, w, \ldots, w).$$ If $d(x_0, \ldots, x_s, w, \ldots, w) = 0$ and $d(y_0, \ldots, y_t, w, \ldots, w) = 0$, then\
$x_i = y_j = w$ for all $i = 0, 1, \dots, s$ and $j = 0, 1, \dots, t$.\
So $g(x_0, \ldots, x_s, y_0, \ldots, y_t) = 0$. Thus, $$g(x_0, \ldots, x_s, y_0, \ldots, y_t) = 0 = g(x_0, \ldots, x_s, w, \ldots, w) + g(y_0, \ldots, y_t, w, \ldots, w).$$
(Diameter $g$-metric) Define $d: {\mathbb{R}}_+^{n+1} \longrightarrow {\mathbb{R}}_+$ by $$d(x_0, \ldots, x_n) = \max_{0 \leq i \leq n}{x_{i}} - \min_{0 \leq j \leq n}{x_{j}}$$ for all $x_0, \ldots, x_n \in {\mathbb{R}}_+$. Then $d$ is a $g$-metric on ${\mathbb{R}}_+$.
It is easy to check that $d$ holds the conditions $(g1)$, $(g2)$, and $(g3)$. Let us show that such $d$ holds the condition $(g4)$. Let $x_0, \ldots, x_s, y_0, \ldots, y_t, w \in {\mathbb{R}}_+$ with $s+t+1=n$. Set $M_x = \max\{ x_0, \ldots, x_s \}$, $m_x = \min\{ x_0, \ldots, x_s \}$, $M_y = \max\{ y_0, \ldots, y_t \}$, and $m_y = \min\{ y_0, \ldots, y_t \}$. Without loss of generality, we assume that $M_x \leq M_y$. Then there are three different cases: (i) $m_x \leq M_x \leq m_y \leq M_y$; (ii) $m_x \leq m_y \leq M_x \leq M_y$; (iii) $m_y \leq m_x \leq M_x \leq M_y$. For notational simplicity we denote $A= d(x_0, \ldots, x_s,y_0,\ldots,y_t)$, $B=d(x_0, \ldots, x_s,w,\ldots,w)$, $C=d(y_0, \ldots, y_t,w\ldots,w)$. For the case (i), clearly $A=M_y - m_x$, and there are five different possibilities with respect to $w$ as follows: $$\begin{aligned}
m_x \leq M_x \leq m_y \leq M_y \leq w & \Longrightarrow A \leq w- m_x = B \leq B+C; \\
m_x \leq M_x \leq m_y \leq w \leq M_y & \Longrightarrow A = M_y - m_y + m_y - m_x \leq B+C; \\
m_x \leq M_x \leq w \leq m_y \leq M_y & \Longrightarrow A= M_y - w + w - m_x = B +C; \\
m_x \leq w \leq M_x \leq m_y \leq M_y & \Longrightarrow A = M_y - w + w - m_x \leq B+C;\\
w \leq m_x \leq M_x \leq m_y \leq M_y & \Longrightarrow A \leq M_y- w = C \leq B+C.\end{aligned}$$ For the case (ii), $A = M_y - m_x $. There are five different possibilities with respect to the value of $w$ as follows: $$\begin{aligned}
m_x \leq m_y \leq M_x \leq M_y \leq w & \Longrightarrow A \leq w- m_x = B \leq B+C; \\
m_x \leq m_y \leq M_x \leq w \leq M_y & \Longrightarrow A = M_y - m_y + m_y - m_x \leq B +C; \\
m_x \leq m_y \leq w \leq M_x \leq M_y & \Longrightarrow A = M_y - w + w - m_x \leq B+C;\\
m_x \leq w \leq m_y \leq M_x \leq M_y & \Longrightarrow A = M_y - w + w - m_x \leq B+C;\\
w \leq m_x \leq m_y \leq M_x \leq M_y & \Longrightarrow A \leq M_y- w = C \leq B+C .\end{aligned}$$ For the case (iii), when $M_y \leq w$ or $w \leq m_y$, it holds that $d(x_0, \ldots, x_s,y_0,\ldots,y_t) \leq d(y_0, \ldots, y_t,w\ldots,w)$. Otherwise, $d(x_0, \ldots, x_s,y_0,\ldots,y_t) = d(y_0, \ldots, y_t,w\ldots,w)$. Therefore, $d$ satisfies the condition $(g4)$.
For a nonempty normed space $(\Omega, \| \cdot \|)$ we define $d: \Omega^{n+1} \longrightarrow {\mathbb{R}}_+$ by $$d(x_0, \ldots, x_n) = \max_{0 \leq i \leq n}{\| x_{i} \|} - \min_{0 \leq j \leq n}{\| x_{j} \|}$$ for all $x_0, \ldots, x_n \in \Omega$. Then it is not a $g$-metric on $\Omega$. In fact, it holds $(g2)$, $(g3)$, and $(g4)$, but does not hold $(g1)$ in general. Indeed, there possibly exist $x_0, x_1,\ldots, x_n\in \Omega$ such that $\| x_0\| = \| x_1\| = \cdots = \| x_n\|$ although $x_i \neq x_j$ for some $i\neq j$.
\[Ex:average g-metric\] (Average $g$-metric) For a given metric space $(\Omega, \delta)$, define $d: \Omega^{n+1} \longrightarrow {\mathbb{R}}_+$ by $$d(x_0, \ldots, x_n) = \frac{1}{(n+1)^2} \sum_{i,j=0}^{n} \delta (x_{i}, x_{j})$$ for all $x_0, \ldots, x_n \in \Omega$. Then $d$ is a $g$-metric on $\Omega$.
By Example \[ex:newold:g-metric\] (1), it is enough to show that $\displaystyle d(x_0, \ldots, x_n) = \sum_{i<j} \delta (x_{i}, x_{j})$ is a $g$-metric on $\Omega$. Clearly, $d$ holds the conditions $(g2)$ and $(g3)$.
- Since $\delta$ is a metric on $\Omega$, it follows that $d(x_0, \ldots, x_n) = 0$ if $x_0 = \cdots = x_n$. Conversely, if $d(x_0, \ldots, x_n) = 0$, then $\delta(x_{i}, x_{j}) = 0$ for all $i,j = 0, \dots, n$. So $x_0 = \cdots = x_n$.
- Let $x_0, \ldots, x_s, y_0, \ldots, y_t, w \in \Omega$ with $s + t + 1 = n$. Since $\delta$ is a metric on $\Omega$, it holds from the triangle inequality that $\delta(x_{i}, y_{j}) \leq \delta(x_{i}, w) + \delta(y_{j}, w)$ for all $i = 0, \dots, s$ and $j = 0, \dots, t$. Then it follows that $$\begin{aligned}
\sum_{i,j} \delta(x_{i}, y_{j}) \leq \sum_{i} \delta(x_{i}, w) + \sum_{j} \delta(y_{j}, w).\end{aligned}$$ Adding $\displaystyle \sum_{i<j} \delta(x_{i}, x_{j})$ and $\displaystyle \sum_{i<j} \delta(y_{i}, y_{j})$ on both sides, we have $$\begin{aligned}
d(x_0, \ldots, x_s, y_0, \ldots, y_t) \leq d(x_0, \ldots, x_s, w, \ldots, w) + d(y_0, \ldots, y_t, w, \ldots, w).\end{aligned}$$
\[Ex:max g-metric\] (Max $g$-metric) For a given metric space $(\Omega, \delta)$, define $d: \Omega^{n+1} \longrightarrow {\mathbb{R}}_+$ by $$d(x_0, \ldots, x_n) = \max_{0 \leq i,j \leq n} \delta(x_{i}, x_{j})$$ for all $x_0, \ldots, x_n \in \Omega$. Then $d$ is a $g$-metric on $\Omega$.
Obviously, $d$ satisfies $(g1)$, $(g2)$, and $(g3)$. Let $x_0, \ldots,~ x_s,~y_0, \ldots,~y_t,~w \in \Omega$ with $s + t + 1 = n$. Let $a$ and $b$ be distinct elements in $\Omega$ such that $$d(x_{0}, \dots, x_{s}, y_{0}, \dots, y_{t}) = \delta(a,b).$$ Then there are three different possibilities: (i) $a, b \in \{ x_0, \ldots, x_s \}$; (ii) $a, b \in \{ y_0, \ldots, y_t \}$; (iii) $a \in \{ x_0, \ldots, x_s \}$, $b \in \{ y_0, \ldots, y_t \}$. For (i) and (ii), it is clear that $d$ holds $(g4)$. For (iii), since $\delta(x_{i}, y_{j}) \leq \delta(x_{i}, w) + \delta(x_{j}, w)$ for all $i = 0, 1, \dots, s$ and $j = 0, 1, \dots, t$, it follows that $\displaystyle \max_{i,j} \delta(x_{i}, y_{j}) \leq \max_{i} \delta(x_{i}, w) + \max_{j} \delta(y_{j}, w)$. Thus, $$\begin{split}
d(x_0, \ldots, x_s, y_0, \ldots, y_t) & = \delta(a,b) = \max_{i,j} \delta(x_{i}, y_{j}) \\
& \leq \max_{i} \delta(x_{i}, w) + \max_{j} \delta(y_{j}, w) \\
& \leq d(x_0, \ldots, x_s, w, \ldots, w) + d(y_0, \ldots, y_t, w, \ldots, w).
\end{split}$$
In Example \[Ex:average g-metric\], on a given metric space $(\Omega, \delta)$ $$d(x_0, \ldots, x_n) = \sum_{i,j=0}^{n} \delta (x_{i}, x_{j})$$ is a $g$-metric by Example \[ex:newold:g-metric\] (1). Then this $g$-metric and the max $g$-metric in Example \[Ex:max g-metric\] can be considered as $$\begin{split}
d(x_0, \ldots, x_n) & = \sum_{i,j=0}^{n} \delta (x_{i}, x_{j}) = || M ||_{1}, \\
d(x_0, \ldots, x_n) & = \max_{0 \leq i,j \leq n} \delta(x_{i}, x_{j}) = || M ||_{\infty},
\end{split}$$ where $M = [m_{ij}]_{0\leq i,j \leq n}$ is the $(n+1) \times (n+1)$ matrix whose entries are $m_{ij} = \delta(x_{i}, x_{j})$. Here, $|| \cdot ||_{1}$ and $|| \cdot ||_{\infty}$ are $\ell_{1}$ and $\ell_{\infty}$ matrix norms, respectively. So it is a natural question whether or not $|| M ||_{p}$ for $1 < p < \infty$ is a $g$-metric on the metric space $(\Omega, \delta)$.
\[Ex:shortest\_path\] (Shortest path $g$-metric) Let $(\Omega, \delta)$ be a nonempty metric space and let $d: \Omega^{3} \to \mathbb{R}_{+}$ be a map defined by $$d(x,y,z) = \min\{ \delta(x,y) + \delta(y,z), \delta(x,z) + \delta(z,y), \delta(y,x) + \delta(x,z) \}.$$ Then $d$ is a $g$-metric with order $2$.
By Theorem \[thm:relation\] (2), it is enough to show that $d$ is a $G$-metric.
- Since $\delta$ is a metric on $\Omega$, $\delta(x,x) = 0$ for all $x \in \Omega$. So $d(x,x,x) = 0$.
- $d(x,x,y) = \min\{ \delta(x,y), 2\delta(x,y), \delta(x,y) \} = \delta(x,y) > 0$ for all $x \neq y$.
- $d(x,x,y) = \delta(x,y) \leq d(x,y,z)$ for all $x,y,z \in \Omega$.
- $d(x,y,z) = d(y,z,x) = d(z,x,y) = d(x,z,y) = d(z, y, x) = d(y,x,z)$.
- Without loss of generality, assume that $d(w,y,z) = \delta(w,y) + \delta(y,z)$. Then it holds that $$d(x,w,w) + d(w,y,z)
= \delta(x,w) + \delta(w,y) + \delta(y,z)
\geq \delta(x,y) + \delta(y,z)
\geq d(x,y,z).$$ The first inequality follows from the triangle inequality for the metric $\delta$, and the second inequality follows from the definition of $d(x,y,z)$.
Thus, $d$ is a $G$-metic. Therefore, $d$ is a $g$-metric with order 2.
- All $g$-metrics listed above are multiplicity-independent.
- For a given metric space $(\Omega, \delta)$ we can generalize the $g$-metric with order $2$ in Example \[Ex:shortest\_path\] as a map $d: \Omega^{n+1} \to \mathbb{R}_{+}$ defined by $$d(x_0, \ldots, x_n) = \min_{ \pi \in \mathcal{S}} \sum_{i=0}^{n-1} \delta( x_{\pi(i)}, x_{\pi(i+1)} )$$ for all $x_0, \ldots, x_n \in \Omega$. Here, $\mathcal{S}$ denotes the set of all permutations on $\{ 0, 1, \ldots, n \}$. That is, $d(x_0, \ldots, x_n)$ is the length of the shortest path connecting $x_0, \ldots, x_n$. Finding the shortest path is very important problem in operations research and theoretical computer science, which is also known as the traveling salesman problem[@PR91; @Verblunsky1951]. In Example \[Ex:shortest\_path\] we showed that the shortest path $g$-metric is a $g$-metric with order $2$, but it is an open problem that $d$ is a $g$-metric for any $n \geq 3$.
- Let $\Omega$ be a nonempty subset of $\mathbb{R}^{n},$ i.e., $\Omega$ can be considered as an $n$-dimensional data set. Define $d:\Omega^{n+1} \longrightarrow \mathbb{R}_{+}$ by $d(x_{0},\ldots,x_{n}) $ is the diameter of the smallest closed ball, $B$, such that $\{x_{0},\ldots,x_{k}\} \subseteq B$. This is called the smallest enclosing circle problem, which was introduced by Sylvester[@sylvester1857]. For more information, see [@DRAGER2007929; @Mordukhovich2013; @Welzl91]. It is an open problem that $d$ is a $g$-metric for any $n \geq 3$.
\[thm:g-metric:basic-properties\] Let $g$ be a $g$-metric with order $n$ on a nonempty set $\Omega$. The following are true:
- $g(\underbrace{x, \ldots, x}_\text{$s$ times},y, \ldots, y) \leq g( \underbrace{x, \ldots, x}_\text{$s$ times}, w, \ldots, w) + g(\underbrace{w, \ldots, w}_\text{$s$ times},y, \ldots, y)$,
- $g(x, y, \ldots, y) \leq g(x, w, \ldots, w) + g(w, y, \ldots, y)$, and\
$ g(x, y, \ldots, y) \leq g(x, w, \ldots, w) + g(y, w, \ldots, w) $ if $g$ is multiplicity-independent,
- $g(\underbrace{x, \ldots, x}_\text{$s$ times},w,\ldots, w) \leq s g(x, w, \ldots, w)$ and\
$g(\underbrace{x, \ldots, x}_\text{$s$ times},w,\ldots, w) \leq (n+1-s) g(w, x, \ldots, x)$,
- $ \displaystyle{g(x_0, x_1, \ldots, x_n) \leq \sum_{i=0}^{n} g(x_{i}, w, \ldots, w)}$,
- $\big| g(y, x_1, \ldots, x_n) - g(w, x_1, \ldots, x_n) \big| \leq \max\{ g(y, w, \ldots, w), g(w, y, \ldots, y) \}$,
- $\big| g(\underbrace{x, \ldots, x}_\text{$s$ times},w,\ldots,w) - g(\underbrace{x, \ldots, x}_\text{$\tilde{s}$ times},w,\ldots,w) \big| \leq \big| s-\tilde{s} \big| g(x, w, \ldots, w) $.
- $g(x, w, \ldots, w) \leq (1+(s-1)(n+1-s))g(\underbrace{x, \ldots, x}_\text{$s$ times},w,\ldots, w) $,
\(1) and (2) follow from the condition $(g4)$. Note that for a multiplicity-independent $g$-metric $g$, it is true that $g(y, w, \ldots, w) = g(w, y, \ldots, y)$.
- By the condition $(g4)$, it follows that $$\begin{aligned}
g(\underbrace{x, \ldots, x}_\text{$s$ times},w,\ldots, w)
&\leq g(\underbrace{x, \ldots, x}_\text{$s-1$ times}, w, w) + g(x, w, \ldots, w) \\
&\leq g(\underbrace{x, \ldots, x}_\text{$s-2$ times}, w, w, w) + g(x, w, \ldots, w) + g(x, w, \ldots, w) \\
& \quad \vdots \\
&\leq g(x, w, \ldots, w) + g(x, w, \ldots, w) + \cdots + g(x, w, \ldots, w) \\
&\leq s g(x, w, \ldots, w).\end{aligned}$$
- By the condition $(g2)$ and $(g4)$, it follows that $$\begin{aligned}
g(x_0, x_1, \ldots, x_n)
&\leq g(x_0, w, \ldots, w) + g(x_1, x_2, \ldots, x_n, w) \\
&\leq g(x_0, w, \ldots, w) + g(x_1, w, \ldots, w) + g(x_2, \ldots, x_n, w, w) \\
& \quad \vdots \\
&\leq \sum_{i=0}^{n} g(x_{i}, w, \ldots, w).\end{aligned}$$
- By the condition $(g4)$, it follows that $ g(y, x_1, \ldots, x_n) \leq g(w, x_1, \ldots, x_n) + g(y, w, \ldots, w)$. Then $$g(y, x_1, \ldots, x_n) - g(w, x_1, \ldots, x_n) \leq g(y, w, \ldots, w).$$ Similarly, we have $$g(w, x_1, \ldots, x_n) - g(y, x_1, \ldots, x_n) \leq g(w, y, \ldots, y).$$
- By (3), it is trivial.
- By Theorem \[thm:g-metric:basic-properties\] (3), we have $$\begin{aligned}
g(x, w, \ldots, w)
& \leq g(x,x,w,\ldots,w) + g(w,x,\ldots,x)\\
& \leq g(x,x,x,w,\ldots,w) + g(w,x,\ldots,x)+g(w,x,\ldots,x)\\
& \quad \vdots \\
&\leq g(\underbrace{x,\ldots,x}_\text{$s$ times}, w, \ldots, w) + (s-1)g(w,x,\ldots,x) \\
&\leq g(\underbrace{x,\ldots,x}_\text{$s$ times}, w, \ldots, w) + (s-1)(n+1-s)g(\underbrace{x,\ldots,x}_\text{$s$ times}, w, \ldots, w)\\
&= (1+(s-1)(n+1-s))g(\underbrace{x,\ldots,x}_\text{$s$ times}, w, \ldots, w).\end{aligned}$$
For a given $g$-metric, we can construct a distance function.
For any $g$-metric space $(\Omega,g)$, the following are distance functions:
- $d(x,y)= g(\underbrace{x, \ldots, x}_\text{$s$ times},y, \ldots, y) + g(\underbrace{y, \ldots, y}_\text{$s$ times}x, \ldots, x)$,
- $d(x,y) = g(x,y,\ldots, y) + g(x,x,y,\ldots, y) + \cdots + g(x,x,\ldots, x,y)$,
- $d(x,y) = \max\{ g(x_0, x_1, \ldots, x_n) : x_{i} \in \{ x,y \},~ 0 \leq i \leq n \}$.
It is easy to show that each function $d$ holds the conditions (1), (2), and (3) in Definition \[D:distance\]. We show that each function holds the triangle inequality (4) in Definition \[D:distance\].
- By the condition $(g4)$, it follows that $$\begin{aligned}
d(x,z) + d(z,y)
&= g(\underbrace{x, \ldots, x}_\text{$s$ times},z, \ldots, z) + g(\underbrace{z, \ldots, z}_\text{$s$ times},x, \ldots, x) \\
&\quad + g(\underbrace{z, \ldots, z}_\text{$s$ times},y, \ldots, y) + g(\underbrace{y, \ldots, y}_\text{$s$ times},z, \ldots, z)\\
& \geq
g(\underbrace{x, \ldots, x}_\text{$s$ times},y, \ldots, y) + g(\underbrace{y, \ldots, y}_\text{$s$ times},x, \ldots, x) \\
&= d(x,y).\end{aligned}$$
- By the condition $(g4)$, it follows that $$\begin{aligned}
d(x,z) + d(z,y)
&= g(x,z,\ldots, z)+g(x,x,z,\ldots, z) + \cdots + g(x,x,\ldots, x,z)\\
&\quad +g(z,y,\ldots, y)+g(z,z,y,\ldots, y) + \cdots + g(z,z,\ldots, z,y)\\
&\geq g(x,y,\ldots, y)+g(x,x,y,\ldots, y) + \cdots + g(x,x,\ldots, x,y)\\
&= d(x,y).\end{aligned}$$
- Let $x, y \in \Omega$. Since $d(x,y)=0$ for $x=y$, the function $d$ holds the triangle inequality. If $x\neq y$, then there exists $\alpha_0, \alpha_1, \ldots, \alpha_n \in \{ x,y \}$ such that $d(x,y)=g(\alpha_0, \alpha_1, \ldots, \alpha_n)$ and $$g(\alpha_0, \alpha_1, \ldots, \alpha_n) \geq g(\beta_0, \beta_1, \ldots, \beta_n)$$ for all $\beta_0, \beta_1, \ldots, \beta_n \in \{ x,y \} $. Let $A = \{i: \alpha_i =x \}$. Clearly, $1\leq n(A) \leq n$. Take $d= n(A)$. Without loss of generality, we assume that $\alpha_0 = \cdots = \alpha_d = x$ and $\alpha_{d+1} = \cdots = \alpha_{n} = y$. That is, $d(x,y) = g(\underbrace{x,x,\dots,x}_\text{$d$ times}, y,y, \ldots,y)$. Then by the condition $(g4)$ it follows that $$\begin{aligned}
d(x,z) + d(z,y)
& \geq g(\underbrace{x,x,\dots,x}_\text{$d$ times},z,z,\ldots,z) + g(\underbrace{z,z,\ldots, z}_\text{$d$ times}, y,y,\ldots,y)\\
& \geq g(\underbrace{x,x,\dots,x}_\text{$d$ times},y,y,\ldots,y) = d(x,y).
\end{aligned}$$
Topology on a $g$-metric space
==============================
For a given metric space $(\Omega,d)$, we denote the ball centered at $x_0$ with radius $r$ by $B_{d}(x_0,r).$ We define a ball on a $g$-metric space.
Let $(\Omega,g)$ be a $g$-metric space. For $x_0 \in \Omega$ and $r>0$, the $g$-ball centered at $x_0$ with radius $r$ is $$B_g(x_0,r) = \{y\in \Omega: g(x_0, y, y, \ldots, y)<r \}.$$
Let $(\Omega,g)$ be a $g$-metric space. Then the following hold.
- If $g(x_0,x_1,x_2,\ldots,x_n)<r$ and $n(\{x_0,x_1,x_2,\ldots,x_n\}) \geq 3$, then $x_i \in B_g(x_0,r)$ for all $i=0,\ldots,n$.
- If $g$ is multiplicity-independent and $g(x_0,x_1,x_2,\ldots,x_n)<r,$ then $x_i \in B_g(x_0,r)$ for all $i=0,\ldots,n$.
- Let $y \in B_g(x_1,r_1) \cap B_g(x_2,r_2).$ Then there exists $\delta > 0$ such that $B_g(y, \delta) \subseteq B_g(x_1,r_1) \cap B_g(x_2,r_2).$
Suppose that $g(x_0,x_1,x_2,\ldots,x_n)<r$. Set $X=\{ x_0,x_1,x_2,\ldots,x_n \}$.
- Since $n(X) \geq 3$, clearly $\{ x_0, x_i, x_i, \ldots, x_i \} \subsetneq X$ for each $i\in {\mathbb{N}}$. By monotonicity condition for the $g$-metric, it follows that $g(x_0, x_i, \ldots, x_i) \leq g(x_0,x_1,\ldots,x_n) <r$. So $x_i \in B_g(x_0,r)$ for all $i\in {\mathbb{N}}$.
- It suffices to show that it holds for $n(X) = 2$. Since a $g$-metric is multiplicity-independent, $g(x_0, x_i, \ldots, x_i) \leq g(x_0,x_1,\ldots,x_n) <r$.
- Since $y \in B_g(x_1,r_1) \cap B_g(x_2,r_2),$ it holds that $g(x_i, y, \ldots, y)<r_i$ for $i=1,2$. We take $\delta = \min\{r_i - g(x_i, y, \ldots, y):i=1,2\}.$ Then for every $z \in B_g(y, \delta),$ by Theorem \[thm:g-metric:basic-properties\] (2) we have $g(x_i, z, \ldots, z) \leq g(x_i, y, \ldots, y) + g(y, z, \ldots, z) < g(x_i, y, \ldots, y) + \delta < r_i$ for each $i=1,2.$ Therefore, $B_g(y, \delta) \subseteq B_g(x_1,r_1) \cap B_g(x_2,r_2).$
Due to the preceding proposition, the collection of all $g$-balls, $\mathcal{B}=\{B_g(x,r) : x\in \Omega, r>0 \}$ forms a basis for a topology on $\Omega.$ We call the topology generated by $\mathcal{B}$ the *$g$-metric topology* on $\Omega.$
Let $(\Omega,g)$ be a $g$-metric space and let $d(x,y)=g(x,y,\ldots,y) + g(y,x,\ldots,x)$. Then $$B_g \Big(x_0,\frac{r}{n+1} \Big) \subseteq B_{d}(x_0,r) \subseteq B_g(x_0,r).$$
Recall that $y\in B_g(x_0,r) \Longleftrightarrow g(x_0,y,y,\ldots,y) < r$.
\(i) Let $\displaystyle{x \in B_g\Big(x_0,\frac{r}{n+1} \Big)} $. Then $\displaystyle{g(x_0,x,x,\ldots,x) < \frac{r}{n+1}}$. It follows that $$\begin{aligned}
d(x_0,x)
&= g(x_0,x,x,\ldots,x) + g(x, x_0,x_0, \ldots, x_0)\\
&\leq g(x_0,x,x,\ldots,x) + ng(x_0, x,x, \ldots, x)\\
&\leq (n+1)g(x_0,x,x,\ldots,x) < r.\end{aligned}$$ So, $x \in B_{d}(x_0,r)$.
\(ii) Let $x \in B_{d}(x_0,r) $. Then $d(x_0,x) = g(x_0,x,x,\ldots,x) + g(x,x_0,x_0,\ldots,x_0)< r$. Since $g(x_0,x,x,\ldots,x) \leq n g(x,x_0,x_0,\ldots,x_0) $, it follows that $$\frac{n+1}{n} g(x_0,x,x,\ldots,x) \leq g(x_0,x,x,\ldots,x) + g(x,x_0,x_0,\ldots,x_0) < r.$$ Thus, $g(x_0,x,x,\ldots,x) < r,$ i.e., $x \in B_g(x_0,r)$ as desired.
Thus, every $g$-metric space is topologically equivalent to a metric space arising from the metric $d$. This makes it possible to transport many concepts and results from metric spaces into the $g$-metric setting.
Convergence and continuity in $g$-metric spaces
===============================================
/in [(2.01,2)/x,(1.9,1.4)/x\_k]{} circle (0.05) node\[above right\] [$\PointLabel$]{}; (1.3,2.7) –(2.01,2) node\[near start,below\][$\varepsilon$]{}; (2,2) circle (1cm); plot \[smooth cycle, tension=0.8\] coordinates [(4.4,1.6) (5,1.4) (5.8,1.7) (6.7773,1.4421)(6.7905,2.3074) (5.9752,2.7) (5.4,2.7) (4.6,2.2) ]{}; /in [(5.01,1.7)/x\_[i\_1]{},(5.21,2)/x\_[i\_2]{},(5.51,2.2)/x\_[i\_3]{},(5.81,2.3)/x\_[i\_4]{},(6.27,2)/x\_[i\_n]{},(6.7,1.7)/x]{} circle (0.05) node\[above right\] [$\PointLabel$]{}; (5.94,2.2) circle (0.01); (6.04,2.15) circle (0.01); (6.14,2.1) circle (0.01);
\[def:convergence\] Let $(\Omega,g)$ be a $g$-metric space. Let $x\in \Omega$ be a point and $\{x_k\} \subseteq \Omega$ be a sequence.
- $\{x_k\}$ *$g$-converges* to $x$, denoted by $\{x_k\} \overset{g}{\longrightarrow} x$, if for all $\varepsilon >0$ there exists $N\in {\mathbb{N}}$ such that $$i_1,\ldots, i_n \geq N \Longrightarrow g(x,x_{i_1},\ldots, x_{i_{n}})< \varepsilon.$$ For such a case, $\{x_k\}$ is said to be *$g$-convergent* in $\Omega$ and $x$ is called the *$g$-limit* of $\{x_k\}$.
- $\{x_k\}$ is said to be *$g$-Cauchy* if for all $\varepsilon >0$ there exists $N\in {\mathbb{N}}$ such that $$i_0,i_1,\ldots, i_n \geq N \Longrightarrow g(x_{i_0},x_{i_1},\ldots, x_{i_{n}})< \varepsilon.$$
- $(\Omega,g)$ is *complete* if every $g$-Cauchy sequence in $(\Omega,g)$ is $g$-convergent in $(\Omega,g)$.
The following are true.
- The limit of a $g$-convergent sequence in a $g$-metric space is unique.
- Every convergent sequence in a $g$-metric space is a $g$-Cauchy sequence.
<!-- -->
- Let $(\Omega,g)$ be a $g$-metric space and let $\{x_k\} \subseteq \Omega$ be a $g$-convergent sequence. Suppose that $x,y \in \Omega$ are the $g$-limits of $\{x_k\}$. By Definition \[def:convergence\] (1), there exists $N_1,N_2\in {\mathbb{N}}$ such that $$\begin{aligned}
g(x,x_{i_1},\ldots, x_{i_{n}}) &< \frac{\varepsilon}{n+1} \quad \text{for all }i_1,\ldots, i_n \geq N_1,\\
g(y,x_{i_1},\ldots, x_{i_{n}}) &< \frac{\varepsilon}{n+1} \quad \text{for all }i_1,\ldots, i_n \geq N_2.
\end{aligned}$$ Set $N=\max\{N_1,N_2\}$. If $m \geq N$, then by the condition $(g4)$ and Theorem \[thm:g-metric:basic-properties\] (3), it follows that $$\begin{aligned}
g(x,y,y,\ldots,y) &\leq g(x,x_m, x_m, \ldots, x_m) + g(x_m, y, y, \ldots, y)\\
&\leq g(x,x_m, x_m, \ldots, x_m) + ng(y, x_m, x_m, \ldots, x_m)\\
&< \frac{\varepsilon}{n+1} + \frac{n\varepsilon}{n+1} = \varepsilon.
\end{aligned}$$ Since $\varepsilon$ is arbitrary, $g(x,y,y,\ldots,y)=0$. Thus, $x=y$ by the condition $(g1)$.
- Let $(\Omega,g)$ be a $g$-metric space and let $\{x_k\} \subseteq \Omega$ be a convergent sequence with the $g$-limit $x$. By Definition \[def:convergence\] (1), there exists $N\in {\mathbb{N}}$ such that $$\begin{aligned}
g(x,x_{i_1},\ldots, x_{i_{n}}) &< \frac{\varepsilon}{n+1} \quad \text{for all }i_1,\ldots, i_n \geq N.
\end{aligned}$$ By Theorem \[thm:g-metric:basic-properties\] (4) and the monotonicity condition for the $g$-metric, it follows that $$\begin{aligned}
g(x_{i_0},x_{i_1},\ldots, x_{i_{n}}) \leq \sum_{k=0}^n g(x_{i_k},x,x,\ldots, x) < \sum_{k=0}^n \frac{\varepsilon}{n+1} = \varepsilon.
\end{aligned}$$ Thus, $\{x_k\}$ is a $g$-Cauchy sequence in $(\Omega, g)$.
\[lemma:g-convergent\] Let $(\Omega,g)$ be a $g$-metric space. Let $\{x_k\} \subseteq \Omega$ be a sequence and $x\in \Omega$. The following are equivalent.
- $\{x_k\} \overset{g}{\longrightarrow} x$.
- For a given $\varepsilon>0 $, there exists $N \in {\mathbb{N}}$ such that $x_k \in B_g(x,\varepsilon)$ for all $k \geq N$.
- $\displaystyle{\lim_{k_1,\ldots, k_s \rightarrow \infty} g(\underbrace{x_{k_1}, \ldots, x_{k_s}}_\text{$s$ times},x,\ldots,x) =0}$ for a fixed $1 \leq s \leq n$. That is, for all $\varepsilon>0$, there exists $N \in {\mathbb{N}}$ such that $k_1,\ldots, k_s \geq N$ $\Longrightarrow$ $g(x_{k_1},\ldots, x_{k_s},x,\ldots,x) < \varepsilon$.
($(1) \Longleftrightarrow (2)$) It is clear by the definition of $g$-convergence.
($(2) \Longrightarrow (3)$) Assume that for a given $\varepsilon > 0,$ there exists $N \in {\mathbb{N}}$ such that $k \geq N$ implies $\displaystyle{x_{k} \in B_{g} \Big( x, \frac{\varepsilon}{s} \Big)},$ i.e., $\displaystyle{g(x, x_{k}, \ldots, x_{k}) < \frac{\varepsilon}{s}}$. If $k_1,\ldots, k_s \geq N$, then by Theorem \[thm:g-metric:basic-properties\] (4), we have that $\displaystyle{g(x_{k_1},\ldots, x_{k_s},x,\ldots,x) \leq \sum_{j=1}^s g(x,x_{k_j},\ldots, x_{k_j}) < \varepsilon }.$
($(3) \Longrightarrow (2)$) Let $\varepsilon>0$. Assume that there exists $N \in {\mathbb{N}}$ such that $$k_1,\ldots, k_s \geq N \Longrightarrow g(k_1,\ldots, k_s,x,\ldots,x) < \frac{\varepsilon}{(1+(s-1)(n+1-s))}.$$ If $k \geq N$, then by Theorem \[thm:g-metric:basic-properties\] (7) it follows that $$\begin{aligned}
g(x,x_k,\ldots, x_k)
\leq (1+(s-1)(n+1-s)) g(\underbrace{x_k,\ldots, x_k}_\text{$s$ times},x,\ldots,x) < \varepsilon.\end{aligned}$$
\[lemma:g-cauchy\] Let $(\Omega,g)$ be a $g$-metric space. Let $\{x_k\} \subseteq \Omega$ be a sequence. The following are equivalent.
- $\{x_k\}$ is $g$-Cauchy.
- $g(x_k, x_{k+1}, x_{k+1},\ldots, x_{k+1}) \longrightarrow 0$ as $k \longrightarrow \infty.$
- $\displaystyle{\lim_{k,\ell \rightarrow \infty} g(\underbrace{x_k,\ldots,x_k}_\text{$s$ times},x_\ell,\ldots,x_\ell) =0}$ for a fixed $1 \leq s \leq n $.
($(1) \Longrightarrow (2)$) It is trivial by Definition \[def:convergence\] (2).
($(2) \Longrightarrow (3)$) Without loss of generality, we can assume $k < \ell.$ Let $\varepsilon > 0$ be given. Then for each $m=0, \ldots, \ell-k-1$ there exists $N_{m} \in {\mathbb{N}}$ such that $\displaystyle{g(x_{k+m},x_{k+m+1}, \ldots, x_{k+m+1}) < \frac{\varepsilon}{n(\ell-k)} }$. Let $N=\text{max}\{N_{0},\ldots, N_{\ell-k-1}\}.$ Then by Theorem \[thm:g-metric:basic-properties\] (3),(4), and the conditions $(g4)$, we have that $$\begin{aligned}
g(\underbrace{x_k,\ldots,x_k}_\text{$s$ times}, & x_\ell,\ldots,x_\ell)
\leq s g(x_{k},x_{\ell},\ldots,x_{\ell}) \\
&\leq s \big(g(x_{k},x_{k+1},\ldots,x_{k+1}) + g(x_{k+1},x_{\ell},\ldots,x_{\ell}) \big) \\
&\leq s \big(g(x_{k},x_{k+1},\ldots,x_{k+1}) + g(x_{k+1},x_{k+2},\ldots,x_{k+2})+ g(x_{k+2},x_{\ell},\ldots,x_{\ell}) \big) \\
& \quad \vdots \\
&\leq s\sum\limits_{i=k}^{\ell-1} g(x_{i},x_{i+1},\ldots,x_{i+1})< \varepsilon,
\end{aligned}$$ for all $k \geq N$. If $k,\ell \geq N$, then $g(\underbrace{x_k,\ldots,x_k}_\text{$s$ times}x_\ell,\ldots,x_\ell)< \varepsilon $.
($(3) \Longrightarrow (1)$) Let $\varepsilon > 0$ be given. Assume that there exists $N \in{\mathbb{N}}$ such that $$k,\ell \geq N ~ \Longrightarrow ~ g(\underbrace{x_{k},\ldots,x_{k}}_\text{$s$ times},x_{\ell},\ldots,x_{\ell}) < \frac{\varepsilon}{n(1+(s+1)(n+1-s))}.$$ If $i_0,i_1,\ldots,i_n \geq N$, then by Theorem \[thm:g-metric:basic-properties\] (4),(7) it follows that $$\begin{aligned}
g(x_{i_0},x_{i_1},\ldots, x_{i_{n}})
&\leq \sum_{k=0}^n g(x_{i_k},x_{i_0},\ldots, x_{i_{0}}) \\
&\leq \sum_{k=0}^n (1+(s+1)(n+1-s)) g(\underbrace{x_{i_k},\ldots,x_{i_k}}_\text{$s$ times},x_{i_0},\ldots,x_{i_0})
< \varepsilon.
\end{aligned}$$
Let $(\Omega,g)$ be a $g$-metric space, and let $\varepsilon>0$ be given.
- A set $A\subseteq \Omega$ is called an *$\varepsilon,g$-net* of $(\Omega,g)$ if for each $x\in \Omega$, there exists $a \in A$ such that $x \in B_g(a,\varepsilon)$. If the set $A$ is finite then $A$ is called a *finite $\varepsilon,g$-net* of $(\Omega,g)$.
- A $g$-metric space $(\Omega,g)$ is called *totally $g$-bounded* if for every $\varepsilon>0$ there exists a finite $\varepsilon,g$-net.
- A $g$-metric space $(\Omega,g)$ is called *$g$-compact* if it is complete and totally $g$-bounded.
Let $(\Omega_{1},g_{1})$ and $(\Omega_{2},g_{2})$ be $g$-metric spaces.
- A mapping $T: \Omega_{1} \longrightarrow \Omega_{2}$ is said to be $g$-*continuous at a point* $x \in \Omega_{1}$ provided that for each open ball $B_{g_{2}}(T(x), \varepsilon),$ there exists an open ball $B_{g_{1}}(x, \delta)$ such that $T(B_{g_{1}}(x, \delta)) \subseteq B_{g_{2}}(T(x), \varepsilon).$
- $T: \Omega_{1} \longrightarrow \Omega_{2}$ is said to be $g$-*continuous* if it is continuous at every point of $\Omega_{1}.$
- $T: \Omega_{1} \longrightarrow \Omega_{2}$ is called a $g$-*homeomorphism* if $T$ is bijective, and $T$ and $T^{-1}$ are $g$-continuous. In this case, the spaces $\Omega_{1}$ and $\Omega_{2}$ are said to be $g$-*homeomorphic*.
- A property $P$ of $g$-metric spaces is called a $g$-*topological invariant* if $P$ satisfies the condition:\
If a space $\Omega_{1}$ has the property $P$ and if $\Omega_{1}$ and $\Omega_{2}$ are $g$-homeomorphic, then $\Omega_{2}$ also has the property $P.$
Let $(\Omega_{1},g_{1})$ and $(\Omega_{2},g_{2})$ be $g$-metric spaces, and let $T: \Omega_{1} \longrightarrow \Omega_{2}$ be a mapping. Then the following are equivalent.
- T is $g$-continuous.
- For each point $x \in \Omega_{1}$ and for each sequence $\{x_{k}\}$ in $\Omega_{1}$ $g$-converging to $x,$ $\{T(x_{k})\}$ $g$-converges to $T(x).$
($(1) \Longrightarrow (2)$) Let $x \in \Omega_{1},$ and let $\{x_{k}\}$ be a sequence in $\Omega_{1}$ $g$-converging to $x.$ Since $T: \Omega_{1} \longrightarrow \Omega_{2}$ is $g$-continuous, for a given $\varepsilon >0$ there exists $\delta >0$ such that $T(B_{g_{1}}(x, \delta)) \subseteq B_{g_{2}}(T(x), \varepsilon n^{-2}).$ Since $\{x_{k}\} \overset{g}{\longrightarrow} x,$ there is $N \in \mathbb{N}$ such that $g(x, x_{i_{1}}, \ldots, x_{i_{n}}) < \delta$ for all $i_{1}, \ldots, i_{n} \geq N.$ Thus $g(x, x_{i_{k}}, \ldots, x_{i_{k}}) < \delta$ for each $k= 1, \ldots, n$. Then the $g$-continuity of $T$ gives rise to the inequality $$g(T(x), T(x_{i_{k}}), \ldots, T(x_{i_{k}})) < \frac{\varepsilon}{n^{2}}$$ for each $k\in {\mathbb{N}}.$ By Theorem \[thm:g-metric:basic-properties\] (3) and (4) we have $$\begin{aligned}
g(T(x), T(x_{i_{1}}), \ldots, T(x_{i_{n}}))
& \leq \sum\limits_{k=1}^{n} g(T(x_{i_{k}}), T(x), \ldots, T(x))\\
& \leq \sum\limits_{k=1}^{n} n g(T(x), T(x_{i_{k}}), \ldots, T(x_{i_{k}})) < \varepsilon.\end{aligned}$$ Therefore, $\{T(x_{k})\}$ $g$-converges to $T(x).$
($(2) \Longrightarrow (1)$) Suppose that $T$ is not $g$-continuous, i.e. there exists $x \in \Omega_{1}$ such that $T$ is not $g$-continuous at $x.$ Then there exists $\varepsilon > 0$ such that for each $\delta > 0$ there is $y \in \Omega_{1}$ with $g(x, y, \ldots, y) < \delta$ but $g(T(x), T(y), \ldots, T(y)) \geq \varepsilon.$ Then for each $k \in {\mathbb{N}}$ we can take $x_{k} \in \Omega_{1}$ such that $g(x, x_{k}, \ldots, x_{k}) < \frac{1}{k}$ but $g(T(x), T(x_{k}), \ldots, T(x_{k})) \geq \varepsilon.$ Hence, $\{x_{k}\}$ $g$-converges to $x$ but $\{T(x_{k})\}$ does not $g$-converges to $T(x),$ which contradicts to (2).
A $g$-metric space $(\Omega,g)$ is said to *have the fixed point property* if every $g$-continuous mapping $T: \Omega \longrightarrow \Omega$ has a fixed point.
The fixed point property is a $g$-topological invariant.
Let $(\Omega_{1},g_1)$ and $(\Omega_{2},g_2)$ be $g$-metric spaces, and let $h:\Omega_{1} \longrightarrow \Omega_{2}$ be a $g$-homeomorphism. Suppose that $\Omega_{1}$ has the fixed point property.\
Let $\widetilde{T}:\Omega_{2} \longrightarrow \Omega_{2}$ be a $g$-continuous function. We consider the function $T:\Omega_{1} \longrightarrow \Omega_{1}$ given by $T(x) = (h^{-1} \circ \widetilde{T} \circ h)(x).$ Since $\Omega_{1}$ has the fixed point property and $T$ is $g$-continuous, there exists a fixed point $x \in \Omega_{1}$ under $T,$ i.e. $T(x)=x.$ Denote $h(x)$ by $y.$ Then we have $$\widetilde{T}(y) = \widetilde{T}(h(x)) = (h \circ h^{-1} \circ \widetilde{T} \circ h)(x) = h(T(x)) = h(x) = y,$$ implying that $y$ is a fixed point under $\widetilde{T}.$ Therefore, $\Omega_{2}$ has the fixed point property.
\[lemma:g-continuous\] If $(\Omega,g)$ is a $g$-metric space, then the function $g$ is jointly continuous in all $n+1$ variables, i.e., if for each $i=0,1,\ldots, n$, $\{x_i^{(k)}\}_{k\in{\mathbb{N}}}$ is a sequence in $\Omega$ such that $\{x_i^{(k)}\} \overset{g}{\longrightarrow} x_i$, then $\{ g(x_0^{(k)},x_1^{(k)},\ldots, x_n^{(k)}) \}{\longrightarrow} \{ g(x_0,x_1,\ldots, x_n) \}$ as $k \longrightarrow \infty$.
Assume that $\{x_{i}^{(k)}\} \overset{g}{\longrightarrow} x_{i}$ as $k \longrightarrow \infty$ for each $i=0, \ldots, n$. For a given $\varepsilon > 0,$ there exists $N_{i} \in {\mathbb{N}}$ such that $\displaystyle{g(x_{i}^{(k)},x_{i}, \ldots, x_{i}) < \frac{\varepsilon}{n+1}}$ if $k \geq N_{i}$ by Lemma \[lemma:g-convergent\] (3). We let $N=\text{max}\{N_{0}, N_{1}, \ldots, N_{n}\}.$ Then by the conditions $(g2),(g4)$, if $k \geq N$, then $$\begin{aligned}
g(x_0^{(k)},x_1^{(k)},\ldots, x_n^{(k)})
&\leq g(x_0^{(k)},x_0,\ldots, x_0) + g(x_0,x_1^{(k)},\ldots, x_n^{(k)}) \\
&\leq g(x_0^{(k)},x_0,\ldots, x_0) + g(x_1,x_1^{(k)},x_1,\ldots, x_1) + g(x_0,x_1, x_2^{(k)}, \ldots, x_n^{(k)}) \\
& \quad \vdots \\
&\leq \sum\limits_{i=0}^{n} g(x_i^{(k)},x_i,\ldots, x_i) + g(x_0,x_1, \ldots, x_n)\\
&< \varepsilon + g(x_0,x_1, \ldots, x_n).\end{aligned}$$ In a similar way, we have $g(x_0,x_1, \ldots, x_n) < \varepsilon + g(x_0^{(k)},x_1^{(k)},\ldots, x_n^{(k)}).$\
Therefore, $\big|g(x_0^{(k)},x_1^{(k)},\ldots, x_n^{(k)})-g(x_0,x_1, \ldots, x_n) \big| < \varepsilon$ as desired.
Fixed point theorems
====================
Fixed point theorems on a $G$-metric space have extensively been studied: see [@AKOH15] and references therein. The interested reader can also refer to [@AN20131486; @GABA17; @GABA18; @Khamsi2015]. In this section we generalize several fixed point theorems on the $g$-metric space under the topology established in Section 3 and Section 4.
The following result can be considered as a generalization of the Banach contractive mapping principle with respect to a $g$-metric space.
\[T:Banach\] Let $(\Omega,g)$ be a complete $g$-metric space and let $T:\Omega \longrightarrow \Omega$ be a mapping such that there exists $\lambda \in [0, 1)$ satisfying $$\label{eq:contractive1}
g(T(x_0), T(x_1), \ldots, T(x_n)) \leq \lambda g(x_0, x_1, \ldots, x_n) \quad \text{for all } x_0, \ldots, x_n \in \Omega.$$ Then $T$ has a unique fixed point in $\Omega$.
Let $y_0$ be an arbitrary point in $\Omega$. Set $y_{k+1} = T(y_k)$ for all $k \in {\mathbb{N}}$.
[**[(Existence of a fixed point)]{}**]{} If $y_{m+1}=y_m$ for some $m\in {\mathbb{N}}$, then $y_m$ is a fixed point of $T$. We assume that $y_{k+1} \neq y_k$ for all $ k \in {\mathbb{N}}$. Then, by the condition it follows that $$\label{eq:contractive-a}
g(y_{k+1},y_{k+2},y_{k+2},\ldots, y_{k+2}) \leq \lambda g(y_{k},y_{k+1},y_{k+1},\ldots, y_{k+1}) \quad \text{for all } k\in {\mathbb{N}}.$$ So, by induction we have $g(y_{k},y_{k+1},y_{k+1},\ldots, y_{k+1}) \leq \lambda^k g(y_{0},y_{1},y_{1},\ldots, y_{1}) $, implying $$g(y_{k},y_{k+1},y_{k+1},\ldots, y_{k+1}) \longrightarrow 0 \quad \text{as } k\longrightarrow \infty.$$ Thus, $\{ y_k \}$ is a $g$-Cauchy sequence in $(\Omega,g)$ by Lemma \[lemma:g-cauchy\]. Since $(\Omega,g)$ is complete, there exists $y\in \Omega$ such that $\{y_k\} \overset{g}{\longrightarrow} y$. It follows that $$\label{eq:contractive-b}
g(y_{k+1},T(y),T(y),\ldots,T(y))
\leq \lambda g(y_k, y, y, \ldots, y).$$ As $k \longrightarrow \infty$, by Lemma \[lemma:g-continuous\] $$g(y,T(y),T(y),\ldots,T(y))
\leq \lambda g(y, y, y, \ldots, y)=0.$$ Therefore, $T(y) = y$ by the positive definiteness for the $g$-metric.
[**[(Uniquness of a fixed point)]{}**]{} Suppose that $y, \tilde{y}$ are distinct fixed points. Then $$\begin{aligned}
g(\tilde{y},y,y,\ldots,y)
& = g(T(\tilde{y}),T(y),T(y),\ldots,T(y)) \\
& \leq \lambda g(\tilde{y},y,y,\ldots,y) \\
& <g(\tilde{y},y,y,\ldots,y),\end{aligned}$$ which is a contradiction. Thus, $y = \tilde{y}$.
In fact, a weaker condition than the contractivity condition can lead to the same conclusion as follows.
\[T:Banach2\] Let $(\Omega,g)$ be a complete $g$-metric space and let $T:\Omega \longrightarrow \Omega$ be a mapping such that there exists $\lambda \in [0, 1)$ satisfying either
- $g(T(x_0), x_1, \ldots, x_n) \leq \lambda g(x_0, x_1, \ldots, x_n)$ for all $x_0, \ldots, x_n \in \Omega$
- - $g(T(x), T(y), \ldots, T(y)) \leq \lambda g(x, y, \ldots, y)$ for all $x, y \in \Omega$.
Then $T$ has a unique fixed point in $\Omega$.
The proof of Theorem \[T:Banach2\] is the same as the proof of Theorem \[T:Banach\]. Note that if the condition $(i)$ holds, then it follows that $$g(T(x_0), T(x_1), \ldots, T(x_n)) \leq \lambda^{n+1} g(x_0, x_1, \ldots, x_n) \quad \text{for all } x_0, \ldots, x_n \in \Omega,$$ which is the contractive condition. Also, it is noted that the condition $(ii)$ implies the inequalities and .
We weaken the contractive conditions based on the notion of weak $\phi$-contractions. The following are some families of control functions which are involved in establishing fixed point results. For more information about these families, see [@AKOH15]. $$\begin{aligned}
\mathcal{F}_{alt} & = \{ \phi:[0,\infty) \longrightarrow [0,\infty): \phi \text{ is continuous, non-decreasing, and } \phi^{-1}(\{0\})=\{0 \} \}, \\
\mathcal{F}'_{alt} & = \{ \phi:[0,\infty) \longrightarrow [0,\infty): \phi \text{ is lower semi-continuous and } \phi^{-1}(\{0\})=\{0 \} \}, \\
\mathcal{F}_{A} & = \{ \phi:[0,\infty) \longrightarrow [0,\infty): \phi \text{ is non-decreasing and } \phi^{-1}(\{0\})=\{0 \} \}.\end{aligned}$$
[@AKOH15 Lemma 2.3.3]\[lemma:FA\] If $\phi \in \mathcal{F}_A$ and $\{t_k\}_{k\in {\mathbb{N}}} \subseteq [0,\infty)$ is a sequence such that $\phi(t_k) \longrightarrow 0$, then $t_k \longrightarrow 0$.
[@AKOH15 Lemma 2.3.6]\[lemma:control3\] Let $\psi \in \mathcal{F}_{alt},~ \phi \in \mathcal{F}'_{alt}$ and let $\{t_k \}_{k\in {\mathbb{N}}} \subseteq [0,\infty)$ be a sequence such that $$\psi(t_{k+1}) \leq \psi(t_k) - \phi(t_k) \quad \text{for all } k \in {\mathbb{N}}.$$ Then $t_k \longrightarrow 0$.
We show some fixed point results in $g$-metric spaces with weaker contractivity conditions involving the families of control functions.
Let $(\Omega,g)$ be a complete $g$-metric space and let $T:\Omega \longrightarrow \Omega$ be a self-mapping. Assume that there exist two function $\psi \in \mathcal{F}_{alt}$ and $\phi \in \mathcal{F}'_{alt}$ such that $$\label{eq:contractive-control1}
\psi(g(T(x), T(y), \ldots, T(y))) \leq \psi(g(x, y, \ldots, y)) - \phi(g(x, y, \ldots, y))$$ for all $x,y \in \Omega$. Then $T$ has a unique fixed point in $\Omega$. Furthermore, $T$ is a Picard operator.
Let $x_0$ be an arbitrary point in $\Omega$. Set $x_{k+1} = T(x_k)$ for all $k \in {\mathbb{N}}$.
[**[(Existence of a fixed point)]{}**]{} If $x_{m+1}=x_m$ for some $m\in {\mathbb{N}}$, then $x_m$ is a fixed point of $T$. We assume that $x_{k+1} \neq x_k$ for all $ k \in {\mathbb{N}}$.
Then, by the condition it follows that for all $k\in {\mathbb{N}}$, $$\begin{aligned}
& \quad \psi(g(x_ {k+1}, x_ {k+2}, \ldots, x_ {k+2}))\\
&= \psi(g(T(x_k), T(x_{k+1}), \ldots, T(x_{k+1})) \\
& \leq \psi(g(T(x_k), T(x_{k+1}), \ldots, T(x_{k+1})) - \phi(g(T(x_k), T(x_{k+1}), \ldots, T(x_{k+1})).\end{aligned}$$ By Lemma \[lemma:control3\], it follows that $$\lim_{k\rightarrow \infty} g(x_ {k+1}, x_ {k+2}, \ldots, x_ {k+2}) = 0.$$ By Lemma \[lemma:g-cauchy\], $\{x_k\}_{k\in {\mathbb{N}}}$ is a $g$-Cauchy sequence. Since $(\Omega,g)$ is complete, there exists $x \in \Omega$ such that $\{x_k\} \overset{g}{\longrightarrow} x$. It follows that $$\begin{aligned}
\psi(g(x_{k+1},T(x),T(x),\ldots,T(x)))
&= \psi(g(T(x_k), T(x), T(x), \ldots, T(x))) \\
&\leq \psi(g(x_k, x, \ldots, x)) - \phi(g(x_k, x, \ldots, x))\\
& \leq \psi(g(x_k, x, \ldots, x)).\end{aligned}$$ Taking the limit as $k \longrightarrow \infty$ on the both sides, by the continuity of $g$ and $\psi$, we have $$\begin{aligned}
\lim_{k \rightarrow \infty} \psi(g(x_{k+1},T(x),T(x),\ldots,T(x)))
& \leq \lim_{k \rightarrow \infty} \psi(g(x_k, x, \ldots, x)) \\
& = \psi(g(x, x, \ldots, x))=0.\end{aligned}$$ By Lemma \[lemma:FA\] and the continuity of $g$, we have $g(x,T(x),T(x),\ldots,T(x))=0$. Therefore, $T(x) = x$ by the condition $(g1)$.
[**[(Uniquness of a fixed point)]{}**]{} Suppose that $x, \tilde{x}$ are distinct fixed points. Then $g(\tilde{x}, x, \ldots, x)>0$. Since $\phi \in \mathcal{F}'_{alt}$, it holds that $\phi (g(\tilde{x}, x, \ldots, x))>0$. By the contractivity condition, we have that $$\begin{aligned}
\psi(g(\tilde{x},x,x,\ldots,x))
& = \psi(g(T(\tilde{x}),T(x),T(x),\ldots,T(x))) \\
& \leq \psi(g(\tilde{x}, x, \ldots, x)) - \phi(g(\tilde{x}, x, \ldots, x))\\
& < \psi(g(\tilde{x},x,x,\ldots,x)),\end{aligned}$$ which is a contradiction. Thus, $x = \tilde{x}$.
Let $(\Omega, g)$ be a $g$-metric space. A mapping $T: \Omega \to \Omega$ is said to be *weak $g$-contractive* if $$g(T(x_0), T(x_1), \ldots, T(x_n)) < g(x_0, x_1, \ldots, x_n)$$ for which any two of $x_0, \ldots, x_n \in \Omega$ are distinct.
\[prop:continuity\] Let $(\Omega, g)$ be a $g$-metric space. Suppose that $T: \Omega \longrightarrow \Omega$ is a weak $g$-contractive function. Then the function $f: \Omega \longrightarrow \mathbb{R}_{+}$ given by $f(x) = g(x, T(x), \ldots, T(x))$ is continuous.
Let $x \in \Omega.$ We need to show that for any $\varepsilon > 0$ there exists $\delta > 0$ such that $\big| f(B_{g}(x, \delta)) - g(x, T(x), \ldots, T(x)) \big| < \varepsilon$. We let $\displaystyle{\delta = \frac{\varepsilon}{n+1}}$. For $y \in B_{g}(x, \delta),$ we first assume that $g(y, T(y), \ldots, T(y)) \leq g(x, T(x), \ldots, T(x)).$ Then
&|g(x, T(x), …, T(x))-g(y, T(y), …, T(y))|\
&= g(x, T(x), …, T(x))- g(y, T(y), …, T(y))\
& g(x, y, …, y) + g(y, T(x), …, T(x))- g(y, T(y), …, T(y))\
& g(x, y, …, y) + g(T(y), T(x), …, T(x))\
& g(x, y, …, y) + g(y, x, …, x)\
& g(x, y, …, y) + n g(x, y, …, y)\
&< (1+n)=
In a similar way, it can be proved that $|g(x, T(x), \ldots, T(x))-g(y, T(y), \ldots, T(y))| < \varepsilon$ holds when $g(y, T(y), \ldots, T(y)) \geq g(x, T(x), \ldots, T(x)).$ Hence, $f$ is continuous.
Let $T$ be a weak $g$-contractive mapping on a $g$-compact $g$-metric space $(\Omega, g)$. Then $T$ has a unique fixed point.
By Proposition \[prop:continuity\] the function $f: \Omega \ni x \mapsto g(x, T(x), \dots, T(x)) \in \mathbb{R}_{+}$ is continuous. Since $\Omega$ is $g$-compact, the continuous function $f$ attains its minimum at some $\bar{x} \in \Omega$. If $\bar{x} \neq T(\bar{x})$, then $$\begin{split}
g(\bar{x}, T(\bar{x}), \ldots, T(\bar{x})) &= \min_{x \in \Omega} g(x, T(x), \dots, T(x)) \\
& \leq g(T(\bar{x}), T(T(\bar{x})), \dots, T(T(\bar{x}))) \\
& < g(\bar{x}, T(\bar{x}), \ldots, T(\bar{x})),
\end{split}$$ which is a contradiction. So $\bar{x}$ is a fixed point of $T$. The uniqueness argument follows exactly same as in the proof of Theorem \[T:Banach\].
We next generalize Ćirić fixed point theorem [@Cir74] in a $g$-metric space.\
Let $(\Omega,g)$ be a $g$-metric space and $T: \Omega \longrightarrow \Omega$ a map. For each $x \in \Omega$, we denote $$O(x, N) = \{x,~T(x),~ T^{2}(x), \ldots ,~T^{N}(x)\} \text{~and~} O(x, \infty) = \{x,~T(x),~T^{2}(x),\ldots\} ,$$ where $T^{k+1}=T \circ T^k$ for all $k\in {\mathbb{N}}$ and $T^{0}$ is the identity mapping on $\Omega$.
- A $g$-metric space $\Omega$ is said to be *$T$-orbitally $g$-complete* if every $g$-Cauchy sequence contained in $O(x, \infty)$ for some $x \in \Omega$ is $g$-convergent in $\Omega.$
- A mapping $T: \Omega \longrightarrow \Omega$ is called a *$g$-quasi-contraction* if there exists $\lambda \in [0, 1)$ such that for all $x_0,\ldots,x_{n} \in \Omega.$ $$\begin{aligned}
g(T(x_{0}),\ldots,T(x_{n}))
\leq &~ \frac{\lambda}{n} \max \Big[\{g(x_{0},\ldots,x_{n})\} \\
& \cup \{g(x_{i},T(x_{j}),\ldots,T(x_{j})) : i,j = 0,\ldots,n\} \Big].\end{aligned}$$
For $A \subseteq \Omega,$ we denote $\hbox{sup}\{g(a_{0}, \ldots , a_{n}):a_{0},\ldots,a_{n} \in A\}$ by $s(A).$
\[lemma:g-quasi-contraction\] Suppose that $T: \Omega \longrightarrow \Omega$ is a $g$-quasi-contraction on a $g$-metric space $(\Omega,g).$ Then for each $x \in \Omega$ the following inequalities hold.
- $\displaystyle{ g(T^{k_{0}}(x),\ldots,T^{k_{n}}(x)) \leq \frac{\lambda}{n} s(O(x, N)) }$ for all $k_{0},\ldots,k_{n} \in \{1,\ldots,N\}$.
- $\displaystyle{ s(O(x, \infty)) \leq \frac{n}{1-\lambda}g(x,T(x),\ldots,T(x)) }$.
\(1) Let $x \in \Omega.$ Since $\{T^{k_{0}}(x),T^{k_{0}-1}(x),\ldots,T^{k_{n}}(x),T^{k_{n}-1}(x)\}$ is a subset of $O(x, N)$ and the mapping $T$ is a $g$-quasi-contraction, there exists $\lambda \in [0, 1)$ such that $$\begin{aligned}
g(T^{k_{0}}(x),\ldots,T^{k_{n}}(x))
&= g(TT^{k_{0}-1}(x),\ldots,TT^{k_{n}-1}(x))\\
& \leq \frac{\lambda}{n} \hbox{max}\Big[\{g(T^{k_{0}-1}(x),\ldots,T^{k_{n}-1}(x))\}\\
&\quad \cup \{g(T^{k_{i}-1}(x),T^{k_{j}}(x),\ldots,T^{k_{j}}(x)) : i,j = 0,\ldots,n\}\Big]\\
&\leq \frac{\lambda}{n} s(O(x, N)).\end{aligned}$$
\(2) Let $x \in \Omega.$ Since the sequence $\{s(O(x, N))\}_{N\in {\mathbb{N}}}$ is monotonically increasing, $s(O(x, \infty))=\hbox{sup}\{s(O(x, N)):N \in \mathbb{N}\}.$ For a fixed positive integer $N_{0},$ the statement (1) implies that there exist $k_{1},k_{2},\ldots,k_{n} \in \{0,1,\ldots,N_{0}\}$ such that\
$g(x,T^{k_{1}}(x), \ldots , T^{k_{n}}(x))=s(O(x, N_{0})).$ Without loss of generality, we can assume that $k_{1} \leq k_{2} \leq \cdots \leq k_{n}.$ If $k_{n}=0$ (i.e. $k_{i}=0$ for all $i$), then $s(O(x, N_{0}))=g(x,x, \ldots , x)=0.$ Suppose that there exists $1 \leq j \leq n $ such that $k_{j} \neq 0$ and $k_{j-1} = 0.$ Then by Theorem \[thm:g-metric:basic-properties\] (4) and the statement (1) we have $$\begin{aligned}
g(x,T^{k_{1}}(x), \ldots , & T^{k_{n}}(x))
\leq g(x,T(x), \ldots, T(x)) + \sum\limits_{i=1}^{n}g(T^{k_{i}}(x),T(x), \ldots, T(x))\\
& = jg(x,T(x), \ldots, T(x)) + \sum\limits_{i=j}^{n}g(T^{k_{i}}(x),T(x), \ldots, T(x))\\
& \leq jg(x,T(x), \ldots, T(x)) + (n-j+1) \frac{\lambda}{n} s(O(x, N_{0}))\\
& = jg(x,T(x), \ldots, T(x)) + (n-j+1) \frac{\lambda}{n} g(x,T^{k_{1}}(x), \ldots , T^{k_{n}}(x))\\\end{aligned}$$ Thus, it follows that $$\begin{aligned}
s(O(x, N_{0}))
&=g(x,T^{k_{1}}(x), \ldots , T^{k_{n}}(x)) \\
&\leq \dfrac{j}{1-\frac{n-j+1}{n}\lambda}g(x,T(x), \ldots, T(x)) \\
& \leq \dfrac{n}{1-\lambda}g(x,T(x), \ldots, T(x)).\end{aligned}$$ Since $N_{0}$ is arbitrary, $s(O(x, \infty)) \leq \dfrac{n}{1-\lambda}g(x,T(x),\ldots,T(x)).$
Let $\Omega$ be a $g$-metric space. Suppose that $\Omega$ is $T$-orbitally $g$-complete and $T: \Omega \longrightarrow \Omega$ is a $g$-quasi-contraction. Then the following are true.
- $T$ has a unique fixed point $y$ in $\Omega.$
- $\{T^{N}(x)\} \overset{g}{\longrightarrow} y$ as $N \longrightarrow \infty$.
- $g(T^{N}(x), y, \ldots, y) \leq \dfrac{\lambda^N}{n^{N-1}(1-\lambda)}g(x,T(x),\ldots,T(x)).$
<!-- -->
- Let $x \in \Omega.$ Since $T$ is a $g$-quasi-contraction, by Lemma \[lemma:g-quasi-contraction\] (1) it follows that $$\begin{aligned}
g(T^{k_{0}}(x), \ldots, T^{k_{n}}(x))
& = g(TT^{k_{0}-1}(x), T^{k_{1}-k_{0}+1}T^{k_{0}-1}(x), \ldots, T^{k_{n}-k_{0}+1}T^{k_{0}-1}(x))\\
& \leq \frac{\lambda}{n} s(O(T^{k_{0}-1}(x), k_{n}-k_{0}+1))\end{aligned}$$ for positive integers $k_{0}, k_{1}, \ldots, k_{n}$ with $k_{0} < k_{1} < \cdots < k_{n}$. By Lemma \[lemma:g-quasi-contraction\] (1), there exist $\ell_{1}, \ldots, \ell_{n} \in \{0, \ldots, k_{n}-k_{0}+1\}$ (without loss of generality, we assume that $\ell_{1} \leq \cdots \leq \ell_{n}$) such that $$s(O(T^{k_{0}-1}(x), k_{n}-k_{0}+1)) = g(T^{k_{0}-1}(x), T^{l_{1}}T^{k_{0}-1}(x), \ldots, T^{l_{n}}T^{k_{0}-1}(x)).$$ Then by Lemma \[lemma:g-quasi-contraction\] (1), we have $$\begin{aligned}
g(T^{k_{0}-1}(x), &T^{\ell_{1}}T^{k_{0}-1}(x), \ldots, T^{\ell_{n}}T^{k_{0}-1}(x))\\
& =g(TT^{k_{0}-2}(x), T^{\ell_{1}+1}T^{k_{0}-2}(x), \ldots, T^{\ell_{n}+1}T^{k_{0}-2}(x)) \\
& \leq \frac{\lambda}{n} s(O(T^{k_{0}-2}(x), \ell_{n}+1)) \leq \frac{\lambda}{n} s(O(T^{k_{0}-2}(x), k_{n}-k_{0}+2)).\end{aligned}$$ By repeating process, we eventually obtain the following inequalities: $$\begin{aligned}
g(T^{k_{0}}(x), \ldots, T^{k_{n}}(x))
&\leq \frac{\lambda}{n} s(O(T^{k_{0}-1}(x), k_{n}-k_{0}+1))\\
& \leq \bigg(\frac{\lambda}{n} \bigg)^{2} s(O(T^{k_{0}-2}(x), k_{n}-k_{0}+2))\\
& \quad \vdots\\
& \leq \bigg(\frac{\lambda}{n} \bigg)^{k_{0}} s(O(x, k_{n})).\end{aligned}$$ Then it follows from Lemma \[lemma:g-quasi-contraction\] (2) that $$\label{upperbound}
g(T^{k_{0}}(x), \ldots, T^{k_{n}}(x)) \leq \bigg(\frac{\lambda}{n} \bigg)^{k_{0}}\frac{n}{1-\lambda}g(x,T(x),\ldots,T(x)).$$ The sequence of iterates $\{T^{N}(x)\}$ is $g$-Cauchy because $\displaystyle{\Big(\frac{\lambda}{n} \Big)^{k_{0}}}$ tends to $0$ as $k_{0} \longrightarrow \infty.$ Therefore, since $\Omega$ is $T$-orbitally $g$-complete, $\{T^{N}(x)\}$ has the $g$-limit $y$ in $\Omega.$
- (**Existence of a fixed point**) We shall show that the $g$-limit $y$ is a fixed point under $T.$ Let us consider the following inequalities: $$\begin{aligned}
g(y, T(y), \ldots, T(y))
& \leq g(y, T^{N+1}(y), \ldots, T^{N+1}(y)) + g(TT^{N}(y), T(y), \ldots, T(y)) \\
& \leq g(y, T^{N+1}(y), \ldots, T^{N+1}(y)) + \frac{\lambda}{n} \hbox{max}\Big\{g(T^{N}(y), y, \ldots, y),\\
& \qquad g(T^{N}(y), T^{N+1}(y), \ldots, T^{N+1}(y)),~ g(y, T(y), \ldots, T(y)),\\
& \qquad g(T^{N}(y), T(y), \ldots, T(y)),~ g(y, T^{N+1}(y), \ldots, T^{N+1}(y)) \Big\} \\
& \leq g(y, T^{N+1}(y), \ldots, T^{N+1}(y)) + \frac{\lambda}{n} \Big(g(T^{N}(y), y, \ldots, y)\\
& \quad +g(T^{N}(y), T^{N+1}(y), \ldots, T^{N+1}(y))~ + g(y, T(y), \ldots, T(y)) \\
& \quad +g(y, T^{N+1}(y), \ldots, T^{N+1}(y)) \Big) \quad \text{(by Theorem \ref{thm:g-metric:basic-properties} (2))}.\end{aligned}$$ Then for every positive integer $N,$ we have $$\begin{aligned}
g(y, T(y), \ldots, T(y))
& \leq \frac{\lambda}{n-\lambda} \Big[ g(T^{N}(y), y, \ldots, y) + g(T^{N}(y), T^{N+1}(y), \ldots, T^{N+1}(y))\\
& \qquad + \Big(\frac{n}{\lambda}+1 \Big)g(y, T^{N+1}(y), \ldots, T^{N+1}(y)) \Big].\end{aligned}$$ Note that for any $x \in \Omega,$ $\{T^{N}(x)\} \overset{g}{\longrightarrow} y.$ Thus $g(y, T(y), \ldots, T(y))=0,$ i.e. $T(y)=y.$ Therefore, $y$ is a fixed point of $T.$
(**Uniqueness of a fixed point**) Suppose that $y$ and $\widetilde{y}$ are fixed points under $T,$ i.e. $T(y)=y$ and $T(\widetilde{y})=\widetilde{y}.$ The $g$-quasi-contractivity of $T$ gives rise to the following inequality: $$\begin{aligned}
g(\widetilde{y}, y, \ldots, y)
&= g(T(\widetilde{y}), T(y), \ldots, T(y))\\
& \leq \frac{\lambda}{n} \max\Big\{g(\widetilde{y}, y, \ldots, y), g(\widetilde{y}, T(\widetilde{y}), \ldots, T(\widetilde{y})), g(y, T(y), \ldots, T(y)),\\
& \qquad \qquad \quad g(\widetilde{y}, T(y), \ldots, T(y)), g(y, T(\widetilde{y}), \ldots, T(\widetilde{y})) \Big\} \\
& \leq \frac{\lambda}{n} \max \Big\{g(\widetilde{y}, y, \ldots, y), g(y, \widetilde{y}, \ldots, \widetilde{y}) \Big\} \\
& \leq \frac{\lambda}{n} \hbox{max}\Big\{g(\widetilde{y}, y, \ldots, y), n g(\widetilde{y}, y, \ldots, y) \Big\} \quad \text{(by Theorem \ref{thm:g-metric:basic-properties} (3))} \\
& = \lambda g(\widetilde{y}, y, \ldots, y).\end{aligned}$$ Since $0 \leq \lambda <1$, it holds that $g(\widetilde{y}, y, \ldots, y)=0.$ Therefore, $y = \widetilde{y}$ as desired.
- Taking the limit as $k_{1} \longrightarrow \infty$ on the both side of , we can obtain the inequality $$g(T^{k_0}(x), y, \ldots, y) \leq \bigg(\frac{\lambda}{n}\bigg)^{k_0} \bigg( \frac{n}{1-\lambda} \bigg) g(x,T(x),\ldots,T(x)).$$
Acknowledgments {#acknowledgments .unnumbered}
===============
The work of S. Kim was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MIST) (No. NRF-2018R1C1B6001394). This work of H. Choi was partially supported by Shanghai Sailing Program under Grant 16YF1407700, and National Nature Science Foundation of China (NSFC) under Grant No. 61601290.
References {#references .unnumbered}
==========
[99]{}
R. P. Agarwal, E. Karapinar, D. O’Regan, and A. F. Roldan-Lopez-de-Hierro. *Fixed Point Theory in Metric Type Spaces*. Springer International Publishing, 2015.
T. V. An, N. V. Dung, and V. T. L. Hang. A new approach to fixed point theorems on [G]{}-metric spaces. *Topol. Appl.*, 160(12):1486–1493, 2013.
Lj. B. Ćirić. A Generalization of [B]{}anach’s Contraction Principle. *Proc. Amer. Math. Soc.*, 45(2):267–273, 1974.
B. C. Dhage. Generalized metric spaces and mapping with fixed points. *Bull. Calcutta Math. Soc.*, 84:329–336, 1992.
L. D. Drager, J. M. Lee, and C. F. Martin. On the geometry of the smallest circle enclosing a finite set of points. *J. Franklin Inst.*, 344(7):929–940, 2007.
M. Fréchet. Sur quelques points du calcul fonctionnel. *Rendiconti del Circolo Matematico di Palermo*, 22(1):1–74, 1906.
Y. U. Gaba. Fixed point theorems in [G]{}-metric spaces. *J. Math. Anal. Appl.*, 455(1):528–537, 2017.
Y. U. Gaba. Fixed points of rational type contractions in [G]{}-metric spaces. *Cogent Mathematics $\&$ Statistics*, 5(1):1444904, 2018.
S. Gahler. 2-metrische [R]{}aume und ihre topologische [S]{}trukture. *Math. Nachr*, 26:115–148, 1963.
S. Gahler. Zur geometric 2-metrische [R]{}aume. *Reevue Roumaine de Math. Pures et Appl.*, XI:"664–669, 1966.
K. S. Ha, Y. J. Cho, and A. White. Strictly convex and strictly 2-convex 2-normed spaces. *Math. Japonica*, 33(3):375–384, 1988.
M. A. Khamsi, Generalized metric spaces: A survey. *Journal of Fixed Point Theory and Applications*, 17(3):455–475, 2015.
B. Mordukhovich, N. M. Nam, and C. Villalobos. The smallest enclosing ball problem and the smallest intersecting ball problem: existence and uniqueness of solutions. *Optim. Lett.*, 7(5):839–853, 2013.
Z. Mustafa and B. Sims. Some remarks concerning [D]{}-Metric Spaces. In *Proceedings of the International Conference on Fixed Point Theorey and Applications, Valencia (Spain)*, pages 189–198, 2003.
Z. Mustafa and B. Sims, A new approach to generalized metric spaces. *J. Nonlinear Convex Anal.*, 7(2):289–297, 2006.
S. V. R. Naidu, K. P. R. Rao, and N. Srinivasa Rao. On the concepts of balls in a [D]{}-metric space. *Int. J. Math. Math. Sci.*, 2005(1):133–141, 2005.
M. Padberg and G. Rinaldi. A Branch-and-Cut Algorithm for the Resolution of Large-Scale Symmetric Traveling Salesman Problems. *SIAM Review*, 33(1):60–100, 1991.
J. J. Sylvester. A question in the geometry of situation. *Q. J. Pure Appl. Math.*, 79(1), 1857.
S. Verblunsky. On the Shortest Path Through a Number of Points. *Proc. Am. Math. Soc.*, 2(6):904–913, 1951.
E. Welzl. Smallest enclosing disks (balls and ellipsoids). In *New Results and New Trends in Computer Science*, pages 359–370, Berlin, Heidelberg, 1991. Springer Berlin Heidelberg.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The early-type spiral NGC 4698 is known to host a nuclear disc of gas and stars which is rotating perpendicularly with respect to the galaxy main disc. In addition, the bulge and main disc are characterised by a remarkable geometrical decoupling. Indeed they appear elongated orthogonally to each other. In this work the complex structure of the galaxy is investigated by a detailed photometric decomposition of optical and near-infrared images. The intrinsic shape of the bulge was constrained from its apparent ellipticity, its twist angle with respect to the major axis of the main disc, and the inclination of the main disc. The bulge is actually elongated perpendicular to the main disc and it is equally likely to be triaxial or axisymmetric. The central surface brightness, scalelength, inclination, and position angle of the nuclear disc were derived by assuming it is infinitesimally thin and exponential. Its size, orientation, and location do not depend on the observed passband. These findings support a scenario in which the nuclear disc is the end result of the acquisition of external gas by the pre-existing triaxial bulge on the principal plane perpendicular to its shortest axis and perpendicular to the galaxy main disc. The subsequent star formation either occurred homogeneously all over the extension of the nuclear disc or through an inside-out process that ended more than 5 Gyr ago.'
author:
- 'E. M. Corsini,$^{1,2,}$[^1] J. Méndez-Abreu,$^{3,4}$ N. Pastorello,$^{1,5}$ E. Dalla Bontà,$^{1,2}$'
- |
L. Morelli,$^{1,2}$ A. Beifiori,$^{6}$ A. Pizzella$^{1,2}$ and F. Bertola$^{1,2}$\
\
$^{1}$Dipartimento di Fisica e Astronomia ‘G. Galilei’, Università di Padova, vicolo dell’Osservatorio 3, I-35122 Padova, Italy\
$^{2}$INAF–Osservatorio Astronomico di Padova, vicolo dell’Osservatorio 5, I-35122 Padova, Italy\
$^{3}$Instituto Astrofísico de Canarias, Calle Vía Láctea s/n, E-38200 La Laguna, Spain\
$^{4}$Departamento de Astrofísica, Universidad de La Laguna, Calle Astrofísico Francisco Sánchez, E-38205 La Laguna, Spain\
$^{5}$ Centre for Astrophysics and Supercomputing, Swinburne University of Technology, Hawthorn, VIC 3122, Australia\
$^{6}$Max-Planck-Institut für extraterrestrische Physik, Giessenbachstra[ß]{}e, D-85748 Garching bei München, Germany.
date: 'Accepted 2012 March 20. Received 2012 March 19; in original form 2011 December 23'
title: 'Polar bulges and polar nuclear discs: the case of NGC 4698'
---
\[firstpage\]
galaxies: bulges – galaxies: formation – galaxies: individual: NGC 4698 – galaxies: photometry – galaxies: spiral – galaxies: structure.
Introduction
============
Several mechanisms have been proposed for bulge assembly. Bulges formed either before the disc by hierarchical merging, or at the same time in a monolithic collapse, or after the disc as a results of secular evolution. Furthermore, there is evidence that bulges can experience accretion events via infall of external material or satellite galaxies. None of these scenarios alone is able to reproduce the observed properties of all the bulges along the Hubble sequence since bulges are a class of diverse and heterogeneous objects. The large bulges of lenticulars and early-type spirals are similar to low-luminosity elliptical galaxies, whereas the small bulges of late-type spirals are reminiscent of disc components (see @Kormendy2004 and references therein).
The study of the intrinsic shape of bulges is a key constraint on their formation. Although the kinematics of many bulges is well described by dynamical models of oblate ellipsoids which are flattened by rotation with little or no anisotropy [e.g., @Pignatelli2001; @Cappellari2006], the twisting of the bulge isophotes [@Lindblad1956; @Zaritsky1986], the presence of non-circular gas motions [e.g., @Coccato2004; @FalconBarroso2006], and the misalignment between the major axes of the bulge and disc [@Bertola1991; @MendezAbreu2008 hereafter MA+08] observed in several galaxies cannot be explained if the bulge and disc are both axisymmetric. These features are interpreted as the signature of bulge triaxiality. Perfect axisymmetry is ruled out when the intrinsic shape of bulges is determined by statistical analyses based on their observed ellipticity (@Fathi2003, MA+08). A large fraction of bulges are characterised by an elliptical equatorial cross-section, and most of them are flattened along the polar axis they share with the surrounding disc. As a consequence, triaxial bulges elongated perpendicularly with respect to the disc are expected to be very rare [@MendezAbreu2010 hereafter MA+10]. To date NGC 4698 [@Bertola1999], NGC 4672 [@Sarzi2000], and UGC 10043 [@Matthews2004] are the only spiral galaxies known to host a prominent bulge sticking out from the plane of the disc.
We decided to revisit the case of NGC 4698 since such a rare galaxy represents an excellent test case to derive the intrinsic shape of the bulge by applying the method of MA+10. NGC 4698 is an Sab(s) spiral [@RC3 hereafter RC3] in the Virgo Cluster. Its total $B$-band magnitude is $B_T=11.46$ mag (RC3), which corresponds to $M_B = -19.69$ mag assuming a distance of 17 Mpc [@Freedman1994]. The inner region of the large elliptical-like bulge is elongated perpendicular to the major axis of the main disc, i.e. it is a polar bulge; this is also true for the outer parts of the bulge if a parametric photometric decomposition is adopted [@Bertola1999]. At the same time, the stellar [@Bertola1999; @Corsini1999] and ionised-gas [@Bertola2000] components are characterised by an inner velocity gradient and a central zero-velocity plateau along the minor and major axes of the disc, respectively. This kinematically-decoupled core is rotating perpendicularly with respect to the galaxy main disc. It corresponds to a nuclear stellar disc [hereafter NSD; @Pizzella2002]. The geometrical and kinematical orthogonal decoupling of NGC 4698 can hardly be explained without invoking the acquisition of external material from the galaxy outskirts [see @Bertola2000].
In this Letter, we improve the previous results by quantitatively constraining the intrinsic shape of the bulge of NGC 4698 in Section \[sec:bulge\]. The analysis of a near-infrared (NIR) image from the United Kingdom Infrared Telescope (UKIRT) allows us to map the mass distribution and minimise the contamination from dust. We get a more reliable estimate of the galaxy’s structural parameters with respect to @Bertola1999 since they folded the eastern side of their optical image around the galaxy major axis to deal with the strongest dust lanes of the main disc. In addition, they assumed the galaxy surface brightness to be the sum of a de Vaucouleurs bulge and an exponential disc [*with*]{} orthogonal major axes. Here, no [*a priori*]{} choice is made about the orientation of the bulge. The structure of the NSD detected by @Pizzella2002 is analysed in Section \[sec:nsd\] in greater detail. The photometric decomposition of multi-band optical images of the galaxy nucleus obtained with the [*Hubble Space Telescope (HST)*]{} allows us to investigate the formation process of the NSD by studying its stellar populations. Our findings are discussed in Section \[sec:conclusions\].
The bulge {#sec:bulge}
=========
United Kingdom Infrared Telescope imaging
-----------------------------------------
A $K-$band image of NGC 4698 is available in the first data release of the Large Area Survey of UKIRT Infrared Deep Sky Survey (UKIDSS, @Lawrence2007). It was obtained on 23 January 2007 with a total exposure time of 40 s. The 3.8-m UKIRT telescope is operated in Mauna Kea Observatory (Hawaii, USA). It mounted the Wide Field Camera (WFCAM) with four Rockwell Hawaii-II devices. Each of them consists of $2048\times2048$ HgCdTe detectors of $18 \times 18$ $\mu$m$^2$ and covers a field of view of $13.7 \times 13.7$ arcmin$^2$ with an image scale of $0.4$ arcsec pixel$^{-1}$. The gain and read-out noise were 4.5 $e^-$ count$^{-1}$ and 25 $e^-$ rms, respectively. The image was reduced, cleaned for cosmic rays, sky subtracted, and flux calibrated with the WFCAM pipeline [@Hambly2008]. A two-dimensional fit with a circular Gaussian to the field stars in the resulting image yielded a $\rm FWHM = 0.76$ arcsec.
The photometric decomposition of the WFCAM image of NGC 4698 was performed using the Galaxy Surface Photometry Two-Dimensional Decomposition ([GASP2D]{}) algorithm by MA+08, which yields the structural parameters for a Sérsic bulge and an exponential main disc.
The fitting algorithm relies on a $\chi^2$ minimisation of the intensities in counts, for which we must adopt initial trial parameters that are as close as possible to their final values. The trial values of the effective surface brightness, effective radius, shape parameter, position angle, and axial ratio of the bulge ($I_{\rm
e}$, $r_{\rm e}$, $n$, PA$_{\rm b}$, and $q_{\rm b}$) and of the central surface brightness, scalelength, position angle, and axial ratio of the main disc ($I_0$, $h$, PA$_{\rm d}$, and $q_{\rm d}$) were estimated from the analysis of the ellipse-averaged radial profiles of surface brightness $\mu_K$, ellipticity $\epsilon$, and position angle PA. The latter were measured using the [IRAF]{} task [ELLIPSE]{} and analysed by following the prescriptions of MA+08. Starting from these initial trial parameters the different photometric models of the surface brightness were fitted iteratively to the galaxy image. Each image pixel was weighted according to the variance of its total observed photon counts due to the contribution of both galaxy and sky, and determined assuming photon noise limitation and taking the detector read-out noise into account. Seeing effects were also taken into account by convolving the model image with a circular Gaussian point spread function (PSF) with a FWHM matching the observed one. The convolution was performed as a product in Fourier domain before the least-squares minimisation. Only the image pixels with an intensity larger than 0.5 times the sky standard deviation were included in the fit. Foreground stars were masked and excluded from the fit. The result of the photometric decomposition is shown in Fig. \[fig:UKIDSS\]. The residual image shows the tightly-wound spiral arms aligned with the disc major axis. Nevertheless, the fit was satisfactory. The model surface brightness is consistent with the data within the error bars. The differences of ellipticity and position angle between the observed and model isophotes are smaller than 0.1 and $3^\circ$, respectively. The PA is scattered in the radial range between about 5 and 20 arcsec where the isophotes are nearly round.
The errors on the fitted parameters were estimated through Monte Carlo simulations on images of artificial galaxies. A set of disc galaxies with a Sérsic bulge and an exponential main disc with $7 < K_T < 8$ mag to bracket the magnitude of NGC 4698 ($K_T = 7.54$ mag, @Kassin2006) was generated. The artificial galaxies were assumed to be observed at the distance of the Virgo Cluster taking into account resolution effects. The parameters of the artificial galaxies were randomly chosen in the ranges observed for nearby S0/a-Sb galaxies by @Mollenhoff2001. Finally, a background level and photon noise were added to the simulated images in order to mimic the instrumental setup and signal-to-noise of the WFCAM image. The relative errors in the fitted parameters of the artificial galaxies were estimated by comparing the input and output values and were assumed to be normally distributed. The standard deviation was adopted as the $1\sigma$ error in the relevant parameter for the bulge-disc decomposition. The best-fitting values and their $3\sigma$ errors are $\mu_{\rm e} =
17.07 \pm 0.64$ mag arcsec$^{-2}$, $r_{\rm e} = 11.5 \pm 6.0$ arcsec, $n = 3.46 \pm 0.42$, PA$_{\rm b} = 72\fdg8 \pm 0\fdg9$, and $q_{\rm b}
= 0.88 \pm 0.03$ for the bulge, and $\mu_0 = 17.06 \pm 0.72$ mag arcsec$^{-2}$, $h = 35.4 \pm 18.0$ arcsec, PA$_{\rm d} = 165\fdg7 \pm
6\fdg6$, and $q_{\rm d} = 0.47 \pm 0.06$ for the main disc.
Systematic errors given by a wrong estimation of the PSF FWHM and sky level are the most significant contributors to the error budget, since the spiral arms do not affect the result and no further component is observed. We estimated a $3\sigma$ error of 2 and 1 per cent for the sky level and PSF FWHM, respectively. We analysed the artificial galaxies by adopting the correct sky level and a PSF FWHM that was 2 per cent larger (or smaller) than the actual one or the correct PSF FWHM and a sky level that was 1 per cent larger (or smaller) than the actual one. To derive the intrinsic shape of the bulge we are interested in $q_{\rm b}$, $q_{\rm d}$, PA$_{\rm b}$, and PA$_{\rm
d}$. Their $3\sigma$ errors including systematics are smaller than 15 per cent.
Intrinsic shape of the bulge
----------------------------
The method developed by MA+10 was applied to derive the intrinsic shape of the bulge of NGC 4698. It is based upon the geometrical relationships between the projected and intrinsic shapes of the bulge and its surrounding disc. The bulge is assumed to be a triaxial ellipsoid with semi-axes of length $A$ and $B$ in the equatorial plane and $C$ along the polar axis. The bulge shares the same centre and polar axis as its main disc, which is circular and lies on the equatorial plane of the bulge. The intrinsic shape of the bulge is recovered from the bulge ellipticity parameter $e = (1-q_{\rm
b}^2)/(1+q_{\rm b}^2) = 0.13$, the twist angle $\rm \delta =
180^\circ-|PA_{\rm b} -PA_{\rm d}| = 87\fdg1$ between the major axes of the bulge and main disc (i.e., the line of nodes), and the main disc inclination $\theta = \arccos{q_{\rm d}} = 61\fdg7$.
The relation between the intrinsic and projected parameters depends only on the spatial position of the bulge, i.e., on the angle $0 \le
\phi \le 90^\circ$ measured on the bulge equatorial plane between the principal axis corresponding to $A$ and the line of nodes. The equatorial ellipticity $Z=B^2/A^2$ and intrinsic flattening $F=C^2/A^2$ of the bulge are given by:
$$\begin{aligned}
\lefteqn{ \frac{\sin{\left(2\phi_C\right)}\,\sin^2{\theta}}{\cos^2{\theta}} F =
- \sin{\phi_B}\cos{\left(2\phi_C - \phi_B\right)}
\left(1+Z\right)^2 \nonumber } \\
&+ \sin{\left(2\phi_C-\phi_B\right)}
\sqrt{\left(1-Z\right)^2 - \sin^2{\phi_B}\left(1+Z\right)^2},
\label{eqn:vartie}\end{aligned}$$
where:
$$\begin{aligned}
\phi_B & = & \arctan{\frac{e\,\sin{2\delta}}{\cos{\theta}\,\left( 1+e\,\cos{2\delta}\right)}} \;= \;1\fdg8, \\
\phi_C & = & \frac{1}{2} \arctan{\frac{2\,e\,\sin{2\delta}\,\cos{\theta}}{e\,\cos{2\delta}\,\left(1+\cos^2{\theta}\right)-\sin^2{\theta}}} \;= \;89\fdg6.
\label{eqn:phi}\end{aligned}$$
Since the axial ratios $B/A$ and $C/A$ are both functions of the same variable $\phi$, their probabilities are identical, i.e., for a given value of $B/A$ with probability $P(B/A)$, the corresponding value of $C/A$ obtained by Eq. \[eqn:vartie\] has a probability $P(C/A)=P(B/A)$. This allows to obtain the range of possible values of $B/A$ and $C/A$ for the bulge of NGC 4698 and to constrain its most probable intrinsic shape by adopting the probabilities $P(Z)$ and $P(F)$ derived by following MA+10. Fig. \[fig:shape\] shows the distribution of $B/A$ and $C/A$ calculated via Monte Carlo simulations. We randomly chose 5000 geometric configurations assuming a Gaussian distribution of $q_{\rm
b}$, $q_{\rm d}$, PA$_{\rm b}$, and PA$_{\rm d}$ taking into account their $3\sigma$ errors. For each geometric configuration we derived 1000 values of $B/A$ and $C/A$ according to their probability distribution functions (see MA+10). Therefore, the resulting median values of the axial ratios and the $3\sigma$ confidence intervals from their cumulative distribution are $B/A=0.95^{+0.05}_{-0.91}$ and $C/A=1.60^{+0.18}_{-0.12}$, respectively. The bulge of NGC 4698 is elongated orthogonal to the main disc and the probability that it has an equatorial section ($B/A<0.95$) is 50 per cent.
The nuclear stellar disc {#sec:nsd}
========================
Hubble Space Telescope imaging
------------------------------
The Wide Field Planetary Camera 2 (WFPC2) images of NGC 4698 obtained with the filters [*F450W*]{} (Prop. Id. 9042, P.I. S. J. Smartt), [*F606W*]{} (Prop. Id. 6359, P.I. M. Stiavelli), and [*F814W*]{} (Prop. Id. 9042, P.I. S. J. Smartt) were retrieved from the [ *HST*]{} Science Data Archive. The [*F606W*]{} images were taken by centring the galaxy nucleus on the Planetary Camera (PC), whereas the others were taken with the Wide Field Camera (WFC). The PC and WFC detectors are Loral CCDs with $800
\times 800$ pixels and a pixel size of $15 \times 15$ $\mu$m$^2$. The PC image scale of $0.046$ arcsec pixel$^{-1}$ yields a field of view of about $36\times 36$ arcsec$^2$. Each WFC detector covers $72\times
72$ arcsec$^2$ with $0.091$ arcsec pixel$^{-1}$. To help in identifying and correcting cosmic ray events, different exposures were taken with each filter. The total exposure times was 460 s for the [*F450W*]{} and [*F814W*]{} filters and 600 s for the [*F606W*]{} filter. The telescope was always guided in fine lock, giving a typical rms tracking error per exposure of $0.005$ arcsec. The images were reduced using the [CalWFPC]{} reduction pipeline in [IRAF]{} [@McMaster2008]. Subsequent analysis including alignment and combination, rejection of cosmic rays, and sky subtraction was performed using [IRAF]{} standard tasks. The flux calibration to the Vega magnitude system was performed following @Holtzman1995. The flux calibration to the Vega magnitude system in the [*HST*]{} passbands was performed following @Holtzman1995.
Following @Pizzella2002 the unsharp-masked image of each WFPC2 frame was built to first gauge the structure and extent of the NSD in all the available images. This procedure enhanced any surface-brightness fluctuation and non-circular structure extending over a spatial region comparable to the standard deviation of the smoothing Gaussian. The galaxy nucleus clearly reveals the presence of a highly elongated structure in all the images. This is the NSD and its location, orientation, and size are the same in all the observed passbands. This nuclear structure is associated with a central increase in ellipticity, $\epsilon$, and fourth cosine Fourier coefficient, $a_4$, as measured by performing an isophotal analysis using [ELLIPSE]{} and shown in Fig. \[fig:HST\].
The photometric parameters of the NSD were derived in the different passbands using the method by @Scorza1995 [hereafter SB95] as implemented by @Morelli2004 although with a different treatment of the PSF. The photometric decomposition was performed independently for each band-pass. The SB95 method is based on the assumption that the isophotal disciness is the result of the superimposition of a host spheroidal component and an inclined exponential disc. Both the components are assumed to have perfectly elliptical isophotes with constant but different ellipticity. The method consists of the iterative subtraction of an infinitesimally thin disc model characterised by an exponential surface-brightness profile with central surface brightness $I_{0, {\rm NSD}}$, scalelength $h_{\rm NSD}$, axial ratio $q_{\rm
NSD}$, and position angle PA$_{\rm NSD}$. The NSD parameters are adjusted until the departures from perfect ellipses are minimised (i.e., $a_{4}\approx0$). To properly derive the photometric parameters of the NSD, it is important to account for the [*HST*]{} PSF. For each nuclear disc model, the disc-free image of the galaxy was obtained from the galaxy image by subtracting the nuclear disc model after convolving with the [*HST*]{} PSF. The adopted PSF model was calculated with the TINYTIM package taking into account the instrumental setup and position of the NSD on the given image [@Krist1999]. The best-fitting values of the NSD parameters and their $3\sigma$ errors were derived as in @Morelli2004 and are listed in Tab. \[tab:disc\_parameters\]. The NSD inclination is calculated as $i_{\rm NSD} =\arccos{q_{\rm NSD}}$. The NSD is elongated like the galaxy bulge as is evident from its position angle. The comparison between the isophotal parameters of NGC 4698 measured before and after the subtraction of the best-fitting model of the NSD are shown in Fig. \[fig:HST\].
@Pizzella2002 had already analysed the [*F606W*]{} image with the SB95 method. They deconvolved the galaxy image from the effects of the PSF using the Richardson-Lucy method. We decided to repeat the photometric decomposition since both an homogeneous photometric decomposition and an estimate of the errors on the fitted parameters are required to properly compare the structural parameters of the NSD in the different passbands. Moreover, this allowed us to verify that the NSD parameters obtained by subtracting the PSF-convolved disc model from the galaxy image are in agreement within the errors with those obtained by subtracting the unconvolved disc model from a deconvolved galaxy image. To this aim the conversion to Johnson $V$ band has to be taken into account for a proper comparison of $I_{0,{\rm NSD}}$ and was calculated with [IRAF]{} task [SYNPHOT]{} ($V-{\it F606W} = 0.27$ for a spiral galaxy).
------------- ------------------------- ----------------------- ---------------------- ---------- --
[*F450W*]{} $17.39^{+0.24}_{-0.04}$ $50.4^{+1.8}_{-6.7}$ $79.4^{+3.3}_{-6.4}$ $70\pm2$
[*F606W*]{} $16.97^{+0.49}_{-0.12}$ $45.2^{+4.1}_{-6.0}$ $73.4^{+2.2}_{-4.7}$ $71\pm2$
[*F814W*]{} $15.63^{+0.45}_{-0.29}$ $50.4^{+9.8}_{-11.1}$ $75.5^{+9.0}_{-5.2}$ $68\pm2$
------------- ------------------------- ----------------------- ---------------------- ---------- --
: Photometric parameters of the nuclear stellar disc.
\[tab:disc\_parameters\]
Formation of the nuclear stellar disc
-------------------------------------
The structural parameters of the NSD (i.e., scalelength, inclination, and position angle) are constant within the errors in all the available WFPC2 images. Therefore, the location, orientation, and size of the NSD do not depend on the observed passband. This also implies the absence of colour gradients in the NSD, an important constraint on the star formation process. The mean values of the NSD parameters are $\langle h_{\rm NSD} \rangle
= 48.7$ pc, $\langle i_{\rm NSD} \rangle = 76\fdg1$, and $\rm \langle
PA_{\rm NSD} \rangle = 69\fdg7$. The size and luminosity are consistent with those of the other NSDs detected so far [see @Ledo2010 for a census]. The NSD is oriented as is the bulge, and its major axis is perpendicular to that of the main disc, i.e. it is a polar NSD.
A linear combination of single-age stellar population synthesis models was used by @Sarzi2005 to interpret an [*HST*]{} spectrum measured within the central 0.13 arcsec (11 pc) of NGC 4698. The continuum spectral energy distribution between 3000-5700 Å was best fitted by a very old stellar population (10 Gyr) with no evidence for stars younger than 5 Gyr in the galaxy nucleus, where the NSD light contamination is large ($\sim20$ per cent, Fig. \[fig:HST\]). The light fraction from stars younger than 1 Gyr rises to 2 per cent if supersolar-metallicity models are considered. This implies that the centre of the NSD is made mostly of old stars. Models for chemical and spectro-photometric evolution of galaxy discs predict a strong evolution with age of their colour profiles as direct consequence of an inside-out formation [@Boissier1999; @Prantzos2000]. In this scenario colour gradients are produced early on in the disc centre and propagate outwards. The maximum colour gradient measured for the NSD ($|d(B-R)/dr| < 0.05$ mag kpc$^{-1}$) falls short of the model predictions and it is not consistent with a picture wherein the nuclear disc assembled inside out in the last 5-10 Gyr.
Discussion and conclusions {#sec:conclusions}
==========================
The structure of the nucleus and bulge of the Virgo spiral galaxy NGC 4698 was investigated through a detailed analysis of [*HST*]{} optical and UKIRT near-infrared images, respectively. The galaxy is known to host a nuclear disc of gas and stars which is rotating perpendicularly with respect to the galaxy main disc. Moreover, the bulge and main disc appear on the sky elongated perpendicular to each other [@Bertola1999; @Bertola2000; @Pizzella2002].
The equatorial ellipticity and intrinsic flattening of the polar bulge were obtained following MA+10 from the apparent ellipticity of the bulge, twist angle between the bulge and main disc, and main disc inclination measured in the UKIDSS near-infrared image by adopting the photometric decomposition method by MA+08. The bulge of NGC 4698 is remarkably elongated in a perpendicular direction with respect to the disc plane ($C/A=1.60^{+0.18}_{-0.12}$). Although the consistency with axisymmetry was recognised ($B/A=0.95^{+0.05}_{-0.91}$), still the probability that it is significantly triaxial (i.e., with an equatorial elliptical section with $B/A<0.95$) is 50 per cent. The central surface brightness, scalelength, inclination, and position angle of the polar NSD in all the available [*HST*]{} images were measured by assuming it is an infinitesimally thin exponential disc and applying the photometric decomposition method of SB95 as implemented by @Morelli2004. The size, orientation, and location of the polar NSD do not depend on the observed passband, as already observed for the few other NSDs for which a detailed multi-band photometric analysis was performed (@Krajnovic2004, @Morelli2010).
The combination of these new results about the complex structure of NGC 4698 gives us the opportunity to gain further insight on the formation of NSDs with the goal of motivating numerical modelling to test and refine this scenario. The kinematical decoupling between two components of a galaxy suggests the occurrence of an accretion event or merging [@BertolaCorsini1999]. Therefore, it is straightforward to explain the existence of the orthogonally-rotating dynamically-cold nuclear disc in NGC 4698 as the end result of the acquisition of external gas by the pre-existing galaxy. Gas dissipation is indeed a necessary ingredient since purely stellar dynamical mergers can not form a nuclear disc [@Hartmann2011]. In NGC 4698 the accreted gas settled on the principal plane perpendicular to the shortest axis of the triaxial bulge (i.e., perpendicular to the galaxy main disc) and formed stars. The major axis of the NSD is elongated in the same direction as the bulge. The link between the angular momentum transport from galactic to nuclear scales and formation of NSDs has been recently investigated by @Hopkins2010. The stellar population in the centre of NGC 4698 is very old [@Sarzi2005]. The absence of colour gradients in the NSD is explained if either the star formation homogeneously occurred all over the extension of the disc or the NSD assembled through an inside-out process that ended more than 5 Gyr ago.
Acknowledgments {#acknowledgments .unnumbered}
===============
We are grateful to Victor P. Debattista for valuable comments. This work is supported by Padua University (grant 60A02-1283/10) and Italian Space Agency (contract ASI-INAF I/009/10/0). LM acknowledges support from Padua University (grant CPS0204). JMA is partially funded by the Spanish MICINN (Consolider-Ingenio 2010 Program grant CSD2006-00070 and grants AYA2007-67965-C03-01 and AYA2010-21887-C04-04).
Bertola F., Corsini E. M., 1999, in Barnes J. E., Sanders, D. B., eds, IAU Symp. 186, Galaxy Interactions at Low and High Redshift, Kuwler, Dordrecht, p. 149
Bertola F., Corsini E. M., 2000, in Combes F., Mamon G. A., Charmandaris V., eds, ASP Conf. Ser. 197, Dynamics of Galaxies: from the Early Universe to the Present, Astron. Soc. Pac., San Francisco, p. 115
Bertola F., Vietri M., Zeilinger W. W., 1991, ApJ, 374, L13
Bertola F., Corsini E. M., Vega Beltr[á]{}n J. C., Pizzella A., Sarzi M., Cappellari M., Funes J. G., 1999, ApJ, 519, L127
Boissier S., Prantzos N., 1999, MNRAS, 307, 857
Cappellari M., et al., 2006, MNRAS, 366, 1126
Coccato L., Corsini E. M., Pizzella A., Morelli L., Funes J. G., Bertola F., 2004, A&A, 416, 507
Corsini E. M., et al., 1999, A&A, 342, 671
de Vaucouleurs G., de Vaucouleurs A., Corwin H. G., Jr., Buta R. J., Paturel G., Fouque P., 1991, Third Reference Catalogue of Bright Galaxies, Springer-Verlag, Berlin (RC3)
Falc[ó]{}n-Barroso J., et al., 2006, MNRAS, 369, 529
Fathi K., Peletier R. F., 2003, A&A, 407, 61
Freedman W. L., et al., 1994, Nature, 371, 757
Hambly N. C., et al., 2008, MNRAS, 384, 637
Hartmann M., Debattista V. P., Seth A., Cappellari M., Quinn T. R., 2011, MNRAS, 418, 2697
Holtzman J. A., et al., 1995, PASP, 107, 156
Hopkins P. F., Quataert E., 2010, MNRAS, 405, L41
Kassin S. A., de Jong R. S., Pogge R. W., 2006, ApJS, 162, 80
Kormendy J., Kennicutt R. C., Jr., 2004, ARA&A, 42, 603
Krajnovi[ć]{} D., Jaffe W., 2004, A&A, 428, 877
Krist J., Hook R., 1999, STIS Instrument Handbook, Version 4.0, STScI, Baltimore
Lawrence A., et al., 2007, MNRAS, 379, 1599
Ledo H. R., Sarzi M., Dotti M., Khochfar S., Morelli L., 2010, MNRAS, 407, 969
Lindblad B., 1956, Stockholms Observatoriums Annaler, 19, 7
Matthews L. D., de Grijs R., 2004, AJ, 128, 137
McMaster M., et al. 2008, WFPC2 Instrument Handbook, Version 10.0, STScI, Baltimore
M[é]{}ndez-Abreu J., Aguerri J. A. L., Corsini E. M., Simonneau E., 2008, A&A, 478, 353 (MA+08)
M[é]{}ndez-Abreu J., Simonneau E., Aguerri J. A. L., Corsini E. M., 2010, A&A, 521, A71 (MA+10)
M[ö]{}llenhoff C., Heidt J., 2001, A&A, 368, 16
Morelli L., et al., 2004, MNRAS, 354, 753
Morelli L., Cesetti M., Corsini E. M., Pizzella A., Dalla Bont[à]{} E., Sarzi M., Bertola F., 2010, A&A, 518, A32
Pignatelli E., et al., 2001, MNRAS, 323, 188
Pizzella A., Corsini E. M., Morelli L., Sarzi M., Scarlata C., Stiavelli M., Bertola F., 2002, ApJ, 573, 131
Prantzos N., Boissier S., 2000, MNRAS, 313, 338
Sarzi M., Corsini E. M., Pizzella A., Vega Beltr[á]{}n J. C., Cappellari M., Funes J. G., Bertola F., 2000, A&A, 360, 439
Sarzi M., Rix H.-W., Shields J. C., Ho L. C., Barth A. J., Rudnick G., Filippenko A. V., Sargent W. L. W., 2005, ApJ, 628, 169
Scorza C., Bender R., 1995, A&A, 293, 20 (SB95)
Zaritsky D., Lo K. Y., 1986, ApJ, 303, 66
\[lastpage\]
[^1]: E-mail: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In our paper we consider the notion of determinant of Clifford algebra elements. We present some new formulas for determinant of Clifford algebra elements for the cases of dimension $4$ and $5$. Also we consider the notion of trace of Clifford algebra elements. We use the generalization of the Pauli’s theorem for 2 sets of elements that satisfy the main anticommutation conditions of Clifford algebra.'
author:
- |
D. S. Shirokov\
\
Steklov Mathematical Institute\
Gubkin St.8, 119991 Moscow, Russia\
\
email: [email protected]
title: 'Concepts of trace, determinant and inverse of Clifford algebra elements'
---
Keywords: Clifford algebra, determinant, trace, inverse
MSC classes: 15A66
Introduction
============
The notion of determinant of Clifford algebra elements was considered in [@MarcMart]. In our work we present some new formulas for determinant of Clifford algebra elements for the cases of dimension $n=4$ and $5$. Also we consider the notion of trace of Clifford algebra elements. We use the generalization of the Pauli’s theorem for 2 sets of elements that satisfy the main anticommutation conditions of Clifford algebra.
After writing this paper author found the article [@new] on the subject that is close to the subject of this paper. In particular, the article [@new] contains the formulas that are similar to the formulas for the determinant in this paper. However, note that for the first time most of these formulas ($n=1, 2, 3$) were introduced in [@MarcMart].
Complex Clifford algebras
=========================
Let $p$ and $q$ be nonnegative integers such that $p+q=n\geq 1$. We consider complex Clifford algebra $\cl(p,q)$. The construction of Clifford algebra $\cl(p,q)$ is discussed in details in [@Marchuk:Shirokov].
Generators $e^1, e^2, \ldots, e^n$ satisfy the following conditions $$e^a e^b +e^b e^a = 2\eta^{ab} e,$$ where $\eta=||\eta^{ab}||$ is the diagonal matrix whose diagonal contains $p$ elements equal to $+1$ and $q$ elements equal to $-1$.
The elements $$e^{a_1}\ldots e^{a_k}=e^{a_1\ldots a_k},\qquad 1\leq a_1<\ldots a_k\leq n,\quad k=1, 2, \ldots n$$ together with the identity element $e$ form a basis of Clifford algebra $\cl(p,q)$. The number of basis elements equals to $2^n$.
Any Clifford algebra element $U\in\cl(p,q)$ can be written in the following form $$\begin{aligned}
U=ue+u_a e^a+\sum_{a_1<a_2}u_{a_1 a_2}e^{a_1 a_2}+\ldots+u_{1\ldots n}e^{1\ldots n},\label{U:decomp}\end{aligned}$$ where $u, u_a, u_{a_1 a_2},\ldots u_{1\ldots n}$ are complex constants.
We denote the vector subspaces spanned by the elements $e^{a_1 \ldots a_k}$ enumerated by the ordered multi-indices of length $k$ by $\cl_k(p,q)$. The elements of the subspace $\cl_k(p,q)$ are denoted by ${\stackrel}{k}{U}$ and called elements of rank $k$. We have $$\begin{aligned}
\cl(p,q)=\oplus_{k=0}^{n}\cl_k(p,q).\label{ranks}\end{aligned}$$
Clifford algebra $\cl(p,q)$ is a superalgebra, so we have even and odd subspaces: $$\begin{aligned}
\cl(p,q)=\cl_{\Even}(p,q)\oplus\cl_{\Odd}(p,q),\label{Evenness}\end{aligned}$$ where $$\cl_{\Even}(p,q)=\cl_0(p,q)\oplus\cl_2(p,q)\oplus\cl_4(p,q)\oplus\ldots,$$ $$\cl_{\Odd}(p,q)=\cl_1(p,q)\oplus\cl_3(p,q)\oplus\cl_5(p,q)\oplus\ldots$$
Operations of conjugation
=========================
Let denote complex conjugation of matrix by ${\overleftarrow{A}}$, transpose matrix by $A^T$, Hermitian conjugate matrix (composition of these 2 operations) by $A^\dagger$.
Now let define some operations on Clifford algebra elements.
[**Complex conjugation.**]{} Operation of complex conjugation $U\to\bar{U}$ acts in the following way $$\begin{aligned}
\bar U={\overleftarrow{u}}e+{\overleftarrow{u_a}} e^a+\sum_{a_1<a_2}{\overleftarrow{u_{a_1 a_2}}}e^{a_1
a_2}+\sum_{a_1<a_2<a_3}{\overleftarrow{u_{a_1 a_2 a_3}}}e^{a_1 a_2
a_3}+\ldots\label{U:decomp1}\end{aligned}$$ We have $$\bar e^a=e^a,\quad a=1, \ldots, n,\qquad \overline{\overline U}=U,\qquad (\overline{U V})=\bar U \bar V,\qquad (\overline{U+V})=\bar U+\bar V,$$ $$(\overline{\lambda U})={\overleftarrow{\lambda}} \bar U,\qquad \forall U,V\in \cl(p,q),\qquad \lambda\in\C.$$
[**Reverse.**]{} Let define operation reverse for $U\in\cl(p,q)$ in the following way $$\begin{aligned}
U^\sim &=& \sum_{k=0}^n(-1)^{\frac{k(k-1)}{2}}{\stackrel}{k}{U}.\end{aligned}$$ We have $$(e^a)^\sim=e^a,\quad a=1,\ldots,n,\qquad U^{\sim\sim}=U,\qquad (U V)^\sim=V^\sim U^\sim,$$ $$(U+V)^\sim=U^\sim+V^\sim,\qquad (\lambda U)^\sim=\lambda U^\sim.$$
[**Pseudo-Hermitian conjugation.**]{} Let define Pseudo-Hermitian conjugation as composition of reverse and complex conjugation: $$U^\ddagger=\bar U^\sim.$$ We have $$(e^a)^\ddagger=e^a,\qquad a=1,\ldots,n,\qquad U^{\ddagger\ddagger}=U,\qquad (U V)^\ddagger=V^\ddagger U^\ddagger,$$ $$(U+V)^\ddagger=U^\ddagger+V^\ddagger, \qquad (\lambda U)^\ddagger={\overleftarrow{\lambda}} U^\ddagger.$$
[**Grade involution.**]{} Let define operation of grade involution $U\to U^\curlywedge$ in the following way $$U^\curlywedge=\sum_{k=0}^n(-1)^k{\stackrel}{k}{U}.$$ We have $$(e^a)^\curlywedge=-e^a,\quad a=1,\ldots,n,\qquad U^{\curlywedge\curlywedge}=U,\qquad (U V)^\curlywedge=U^\curlywedge V^\curlywedge,$$ $$(U+V)^\curlywedge=U^\curlywedge+V^\curlywedge,\qquad (\lambda U)^\curlywedge=\lambda U^\curlywedge.$$
[**Clifford conjugation.**]{} Let define Clifford conjugation as composition of grade involution and reverse $U\to
U^{\curlywedge\sim}$: $$U^{\curlywedge\sim}=\sum_{k=0}^n(-1)^{\frac{k(k+1)}{2}}{\stackrel}{k}{U}.$$ We have $$(e^a)^{\curlywedge\sim}=-e^a,\quad a=1,\ldots,n,\qquad U^{\curlywedge\sim\curlywedge\sim}=U,\qquad (U V)^{\curlywedge\sim}=V^{\curlywedge\sim}U^{\curlywedge\sim},$$ $$(U+V)^{\curlywedge\sim}=U^{\curlywedge\sim}+V^{\curlywedge\sim},\qquad (\lambda U)^{\curlywedge\sim}=\lambda U^{\curlywedge\sim}.$$
[**Hermitian conjugation.**]{} In [@Marchuk:Shirokov] we consider operation of Hermitian conjugation. We have the following formulas for these operation: $$\begin{aligned}
U^\dagger&=&(e^{1\ldots p})^{-1}U^\ddagger e^{1 \ldots p},\qquad \mbox{if $p$ - odd},\nonumber\\
U^\dagger&=&(e^{1\ldots p})^{-1}U^{\ddagger\curlywedge} e^{1 \ldots p},\qquad \mbox{if $p$ - even},\\
U^\dagger&=&(e^{p+1\ldots n})^{-1}U^\ddagger e^{p+1 \ldots n},\qquad \mbox{if $q$ - even},\nonumber\\
U^\dagger&=&(e^{p+1\ldots n})^{-1}U^{\ddagger\curlywedge} e^{p+1 \ldots n},\qquad \mbox{if $q$ - odd},\nonumber\end{aligned}$$ We have $$(e^a)^\dagger=(e^a)^{-1},\qquad a=1,\ldots,n,\qquad U^{\dagger\dagger}=U,\qquad (U V)^\dagger=V^\dagger U^\dagger,$$ $$(U+V)^\ddagger=U^\dagger+V^\dagger, \qquad (\lambda U)^\dagger={\overleftarrow{\lambda}} U^\dagger.$$
Matrix representations of Clifford algebra elements, recurrent method.
======================================================================
Complex Clifford algebras $\cl(p,q)$ of dimension $n$ and different signatures $(p,q), p+q=n$ are isomorphic. Clifford algebras $\cl(p,q)$ are isomorphic to the matrix algebras of complex matrices. In the case of even $n$ these matrices are of order $2^{\frac{n}{2}}$. In the case of odd $n$ these matrices are block diagonal of order $2^{\frac{n+1}{2}}$ with 2 blocks of order $2^{\frac{n-1}{2}}$.
Consider the following matrix representations of Clifford algebra elements.
Identity element $e$ of Clifford algebra $\cl(p,q)$ maps to identity matrix of corresponding order: $e \to {\bf 1}$.
For $\cl(1,0)$ element $e^1$ maps to the following matrix $$e^1\to\left( \begin{array}{ll}
1 & 0 \\
0 & -1 \end{array}\right).$$ For $\cl(2,0)$ we have $$e^1\to\left( \begin{array}{ll}
1 & 0 \\
0 & -1 \end{array}\right)
,\quad
e^2\to\left( \begin{array}{ll}
0 & 1 \\
1 & 0 \end{array}\right).$$ Further, suppose we have a matrix representation for $\cl(2k,0)$, $n=2k$: $$e^1, \ldots, e^n \to \gamma^1, \ldots, \gamma^n.$$
Then, for Clifford algebra $\cl(2k+1,0)$ we have $$e^a\to\left( \begin{array}{ll}
\gamma^a & 0 \\
0 & -\gamma^a \end{array}\right)
,\quad a=1, \ldots, n,\qquad
e^{n+1}\to\left( \begin{array}{ll}
i^k \gamma^1\ldots \gamma^n & 0 \\
0 & -i^k \gamma^1\ldots \gamma^n \end{array}\right).$$
For Clifford algebra $\cl(2k+2,0)$ we have the same matrices for $e^a, a=1, \ldots, n+1$ as in the previous case $n=2k+1$ and for $e^{n+2}$ we have $$e^{n+2}\to\left( \begin{array}{ll}
0 & {\bf1} \\
{\bf1} & 0 \end{array}\right).$$
So, we have matrix representation for all Clifford algebras $\cl(n,0)$. In the cases of other signatures elements $e^a, a>p$ maps to the same matrices as in signature $(n,0)$ but with multiplication by imaginary unit $i$.
For example, we have the following matrix representations for Clifford algebras $\cl(3,0)$, $\cl(4,0)$ and $\cl(1,3)$.
$\cl(3,0)$:
: $$e^1 \to \gamma^1=\left( \begin{array}{llll}
1 & 0 & 0 & 0\\
0 & -1 & 0 & 0\\
0 & 0 & -1 & 0\\
0 & 0 & 0 & 1\end{array}\right),\quad
e^2 \to \gamma^2=\left( \begin{array}{llll}
0 & 1 & 0 & 0\\
1 & 0 & 0 & 0\\
0 & 0 & 0 & -1\\
0 & 0 & -1 & 0\end{array}\right),$$ $$e^3 \to \gamma^3=\left( \begin{array}{llll}
0 & i & 0 & 0\\
-i & 0 & 0 & 0\\
0 & 0 & 0 & -i\\
0 & 0 & i & 0\end{array}\right).$$
$\cl(4,0)$:
: $$e^1 \to \gamma^1,\quad e^2 \to \gamma^2,\quad e^3 \to \gamma^3,\quad e^4 \to \gamma^4=\left( \begin{array}{llll}
0 & 0 & 1 & 0\\
0 & 0 & 0 & 1\\
1 & 0 & 0 & 0\\
0 & 1 & 0 & 0\end{array}\right).$$
$\cl(1,3)$:
: $$e^1 \to \gamma^1,\quad e^2 \to i\gamma^2,\quad e^3 \to i\gamma^3,\quad e^4 \to i\gamma^4.$$
Operation of trace of Clifford algebra elements
===============================================
Consider complex Clifford algebra $\cl(p,q)$ and introduce the operation of [*trace*]{} of Clifford algebra element $U\in\cl(p,q)$ as the following operation of projection onto subspace $\cl_0(p,q)$: $$\Tr(U)=\langle U\rangle_0|_{e\to 1}.\label{tracedef}$$ For arbitrary element $U\in\cl(p,q)$ in the form (\[U:decomp\]) we have $$\Tr(ue+u_a e^a +\ldots)=u.$$
. \[theoremTraceProp\] Operation trace (\[tracedef\]) of Clifford algebra element $U\in\cl(p,q)$ has the following properties:
- linearity: $$\Tr(U+V)=\Tr(U)+\Tr(V),\qquad \Tr(\alpha U)=\alpha \Tr(U)$$ $$\forall U, V\in\cl(p,q),\qquad \forall \alpha\in\C,$$
- cyclic recurrence: $$\Tr(UV)=\Tr(VU),\qquad \Tr(UVW)=\Tr(VWU)=\Tr(WUV)$$ $$\forall U, V, W\in\cl(p,q),$$ but, in general: $$\Tr(UVW)\neq\Tr(UWV).$$
- invariance under similarity transformation: $$\Tr(U^{-1}VU)=\Tr(V) \qquad \forall V\in\cl(p,q),\, U\in\cl^{\times}(p,q),$$ where $\cl^{\times}(p,q)$ is the set of all invertible Clifford algebra elements.
- invariance under conjugations: $$\Tr(U)=\Tr(U^\curlywedge)=\Tr(U^\sim)={\overleftarrow{\Tr(\overline{U})}}={\overleftarrow{\Tr(U^\ddagger)}}={\overleftarrow{\Tr(U^\dagger)}}.$$
. Linearity follows from the definition (\[tracedef\]).
We have $$\Tr(UV)=\Tr(VU)$$ because $\Tr([{\stackrel}{k}{U},{\stackrel}{l}{V}])=0$ for $k,l=0,\ldots n$ (see [@Marchuk:Shirokov]). Then we obtain cyclic recurrence for 3 elements. We obtain invariance under similarity transformation as a simple consequence of cyclic recurrence. Last properties follow from properties of conjugations. $\blacksquare$
There is a relation between operation trace $\Tr$ of Clifford algebra element $U\in \cl(p,q)$ and operation trace $\tr$ of quadratic matrix. To obtain this relation, at first, we will prove the following statement.
. \[lemmaTrace\] Consider recurrent matrix representation of Clifford algebra $\cl(p,q)$ (see above). For this representation $U \to \underline U$ we have $$\tr(\underline U)=2^{[\frac{n+1}{2}]} \Tr(U),\qquad \tr(\underline U^\curlywedge)=\tr(\underline U).$$
. Coefficient $2^{[\frac{n+1}{2}]}$ equals to the order of corresponding matrices. It is not difficult to see that trace of almost all matrices that correspond to basis elements equals to zero $$\tr(\underline e^A)=0,\qquad \mbox{where $A$ - any multi-index except empty}.$$ The only exception is identity element $e$, which corresponds to the identity matrix. In this case we have $\tr(\underline e)=2^{[\frac{n+1}{2}]}$. Further we use linearity of trace and obtain $$\tr(\underline U)=2^{[\frac{n+1}{2}]} u=2^{[\frac{n+1}{2}]}\Tr(U).$$ The second property is a simple consequence of the first property, because $$\tr(\underline U^\curlywedge)=2^{[\frac{n+1}{2}]} \Tr(U^\curlywedge)=2^{[\frac{n+1}{2}]} \Tr(U)=\tr(\underline U).$$ $\blacksquare$
. \[theoremTraceDef\] Consider complex Clifford algebra $\cl(p,q)$ and operation trace $\Tr$. Then $$\begin{aligned}
\Tr(U)=\frac{1}{2^{[\frac{n+1}{2}]}}\tr(\gamma(U)),\label{tracedef2}\end{aligned}$$ where $\gamma(U)$ - any matrix representation of Clifford algebra $\cl(p,q)$ of minimal dimension. Moreover, this definition of trace (\[tracedef2\]) is equivalent to the definition (\[tracedef\]). New definition is well-defined because it doesn’t depend on the choice of matrix representation.
. This property proved in the previous statement for the recurrent matrix representation. Let we have besides recurrent matrix representation $$\underline U=U|_{e^a\to\gamma^a}$$ another matrix representation $$\underline{\underline U}=U|_{e^a\to\beta^a}.$$ Then, by Pauli’s theorem in Clifford algebra of even dimension $n$ there exists matrix $T$ such that $$\beta^a=T^{-1}\gamma^a T,\qquad a=1, \ldots, n.$$ Then, we have $$\underline{\underline U}= T^{-1} \underline UT$$ and $$\tr(\underline{\underline U})= \tr(T^{-1}\underline U T)=\tr(\underline U).$$ In the case of odd $n$ we can have also another case (by Pauli’s theorem), when two sets of matrices relate in the following way $$\beta^a=-T^{-1}\gamma^a T,\qquad a=1, \ldots, n.$$ In this case we have $$\underline{\underline{U}}= T^{-1}\underline{U^\curlywedge}T.$$ From $\tr(\underline{U^\curlywedge})=\tr(\underline U)$ (see Lemma \[lemmaTrace\]) we obtain $$\tr(\underline{\underline{U}})= \tr(T^{-1}\underline{U^\curlywedge}T)=\tr(\underline{U^\curlywedge})=\tr(\underline U).$$ $\blacksquare$
Determinant of Clifford algebra elements
========================================
[*Determinant of Clifford algebra element $U\in\cl(p,q)$*]{} is a complex number $$\begin{aligned}
\Det U=\det(\underline U),\label{det}\end{aligned}$$ which is a determinant of any matrix representation $\underline U$ of minimal dimension.
Now we want to show that this definition is well-defined. Let prove the following Lemma.
. \[lemmaDetRek\] Consider the recurrent matrix representation (see above) of Clifford algebra $\cl(p,q)$. For this representation $U \to \underline U$ we have $$\det(\underline U^\curlywedge)=\det(\underline U).$$
. In the case of Clifford algebra of even dimension $n$ we have $$U^\curlywedge=(e^{1\ldots n})^{-1}Ue^{1\ldots n}.$$ So, we obtain $$\det(\underline{U^\curlywedge})=\det(\underline{(e^{1\ldots n})^{-1}U e^{1\ldots n}})=\det(\underline{(e^{1\ldots n})^{-1}})\det(\underline{U})\det(\underline{e^{1\ldots n}})=\det(\underline{U}).$$
In the case of Clifford algebra of odd dimension generators maps to the block diagonal matrices and blocks are identical up to the sign: $$e^a\to\left( \begin{array}{ll}
\gamma^a & 0 \\
0 & -\gamma^a \end{array}\right).$$ Then for elements of the rank 2 we obtain $$e^{ab}\to\left( \begin{array}{ll}
\gamma^a\gamma^b & 0 \\
0 & \gamma^a\gamma^b \end{array}\right).$$ It is not difficult to see that even part $U_{\Even}$ of arbitrary element $U=U_{\Even}+U_{\Odd}$ maps to the matrix with identical blocks, and odd part $U_{\Odd}$ of the element $U$ maps to the matrix with the blocks differing in sign: $$U_{\Even}\to\left( \begin{array}{ll}
A & 0 \\
0 & A \end{array}\right),\qquad
U_{\Odd}\to\left( \begin{array}{ll}
B & 0 \\
0 &-B \end{array}\right).$$ Then we have $$U\to\left( \begin{array}{ll}
A+B & 0 \\
0 & A-B \end{array}\right),\qquad
U^\curlywedge\to\left( \begin{array}{ll}
A-B & 0 \\
0 & A+B \end{array}\right)$$ and $$\det(U)=(A-B)(A+B)=\det(U^\curlywedge).$$ $\blacksquare$
. \[predlDet\] Definition (\[det\]) is well-defined, i.e. it doesn’t depend on the matrix representation.
. Consider the recurrent matrix representation $$\underline U=U|_{e^a\to\gamma^a}.$$ The statement for this representation proved in the previous lemma. Let we have another matrix representation $$\underline{\underline U}=U|_{e^a\to\beta^a}.$$ Then, by Pauli’s theorem in Clifford algebra of even dimension $n$ there exists a matrix $T$ such that $$\beta^a=T^{-1}\gamma^a T,\qquad a=1, \ldots, n.$$ Then we have $$\underline{\underline U}= T^{-1} \underline UT$$ and obtain $$\det(\underline{\underline U})= \det(T^{-1}\underline U T)=\det(T^{-1})\det(\underline U)\det(T)=\det(\underline U).$$
In the case of odd $n$, by Pauli’s theorem we also have another case, where 2 sets of matrices relate in the following way $$\beta^a=-T^{-1}\gamma^a T,\qquad a=1, \ldots, n.$$ In this case we have $$\underline{\underline{U}}= T^{-1}\underline{U^\curlywedge}T.$$ From $\det(\underline{U^\curlywedge})=\det(\underline U)$ (see Lemma \[lemmaDetRek\]) we obtain $$\det(\underline{\underline{U}})= \det(T^{-1}\underline{U^\curlywedge}T)=\det(\underline{U^\curlywedge})=\det(\underline U).$$ $\blacksquare$
Let formulate some properties of operation determinant of Clifford algebra element.
. \[theoremDetProp\] Operation determinant (\[det\]) of Clifford algebra element $U\in\cl(p,q)$ has the following properties
- $$\Det(UV)=\Det(U)\Det(V),\qquad \Det(\alpha U)=\alpha^{2^{[\frac{n+1}{2}]}} \Det(U)\label{2pr}$$
$$\forall U, V\in\cl(p,q),\qquad \forall \alpha\in\C.$$
- Arbitrary element $U\in \cl(p,q)$ is invertible if and only if $\Det U\neq 0$.
- For any invertible element $U\in\cl(p,q)$ $$\Det (U^{-1}) = (\Det U)^{-1}.\label{3pr}$$
- Invariance under similarity transformation: $$\Det(U^{-1}VU)=\Det(V) \qquad \forall V\in\cl(p,q),\, U\in\cl^{\times}(p,q),$$ where $\cl^{\times}(p,q)$ is set of all invertible Clifford algebra elements.
- Invariance under conjugations: $$\Det(U)=\Det(U^\curlywedge)=\Det(U^\sim)={\overleftarrow{\Det(\overline{U})}}={\overleftarrow{\Det(U^\ddagger)}}={\overleftarrow{\Det(U^\dagger)}}.$$
. The first 4 properties are simple and follow from the definition of determinant (\[det\]).
In lemma \[lemmaDetRek\] we have the property $\Det(U)=\Det(U^\curlywedge)$ for the recurrent matrix representation. But it is also valid for other matrix representations because of independence on the choice of representations (see Theorem \[predlDet\]).
It is known that operation $\sim$ relates to the operation of matrix transpose (as similarity transformation) and also we have $\det(U)=\det(U^T)$. Analogously we can consider another operations of conjugation. $\blacksquare$
Definition (\[det\]) of determinant of Clifford algebra element $U\in\cl(p,q)$ is connected with its matrix representation. We have shown that this definition doesn’t depend on matrix representation. So, determinant is a function of complex coefficients $u_{a_1 \ldots a_k}$ located before basis elements $e^{a_1\ldots a_k}$ in (\[U:decomp\]). In the cases of small dimensions $n\leq 5$ we give expressions for determinant of Clifford algebra elements that doesn’t relate to the matrix representation.
Now we need also 2 another operations of conjugations $\bigtriangledown$, $\bigtriangleup$:
$$({\stackrel}{0}{U}+{\stackrel}{1}{U}+{\stackrel}{2}{U}+{\stackrel}{3}{U}+{\stackrel}{4}{U})^\bigtriangledown = {\stackrel}{0}{U}+{\stackrel}{1}{U}+{\stackrel}{2}{U}+{\stackrel}{3}{U}-{\stackrel}{4}{U},\qquad n=4,$$ $$({\stackrel}{0}{U}+{\stackrel}{1}{U}+{\stackrel}{2}{U}+{\stackrel}{3}{U}+{\stackrel}{4}{U}+{\stackrel}{5}{U})^\bigtriangledown = {\stackrel}{0}{U}+{\stackrel}{1}{U}+{\stackrel}{2}{U}+{\stackrel}{3}{U}-{\stackrel}{4}{U}-{\stackrel}{5}{U},\qquad n=5,$$ $$({\stackrel}{0}{U}+{\stackrel}{1}{U}+{\stackrel}{2}{U}+{\stackrel}{3}{U}+{\stackrel}{4}{U}+{\stackrel}{5}{U})^\bigtriangleup = {\stackrel}{0}{U}+{\stackrel}{1}{U}+{\stackrel}{2}{U}+{\stackrel}{3}{U}+{\stackrel}{4}{U}-{\stackrel}{5}{U},\qquad n=5.$$
. \[theoremDet\] We have the following formulas for the determinant of Clifford algebra element $U\in\cl(p,q)$: $$\begin{aligned}
\Det\,U=\left\lbrace
\begin{array}{ll}
U, & n=0;\\
U U^\curlywedge, & n=1;\\
U U^{\sim\curlywedge}, & n=2;\\
U U^\sim U^\curlywedge U^{\sim\curlywedge}= U U^{\sim\curlywedge} U^\curlywedge U^\sim, & n=3;\\
U U^\sim (U^\curlywedge U^{\sim\curlywedge})^\bigtriangledown= U U^{\sim\curlywedge} (U^\curlywedge U^{\sim})^\bigtriangledown, & n=4;\\
U U^\sim (U^\curlywedge U^{\sim\curlywedge})^\bigtriangledown (U U^\sim (U^\curlywedge U^{\sim\curlywedge})^\bigtriangledown)^\bigtriangleup, & n=5.
\end{array}
\right.\end{aligned}$$
Note, that these expressions are Clifford algebra elements of the rank $0$. In this case we identify them with the constants: $ue\equiv u$.
. The proof is by direct calculation. $\blacksquare$.
Note, that properties (\[2pr\]) and (\[3pr\]) for small dimensions also can be proved with the formulas from Theorem \[theoremDet\]. For example, in the case $n=3$ we have $$\Det(UV)=UV (UV)^{\sim\curlywedge} (UV)^\curlywedge (UV)^\sim =UVV^{\sim\curlywedge} U^{\sim\curlywedge}U^\curlywedge V^\curlywedge V^\sim U^\sim =$$ $$=UU^{\sim\curlywedge} U^\curlywedge U^\sim V V^{\sim\curlywedge} V^\curlywedge V^\sim =\Det(U)\Det(V)$$ We used the fact that $VV^{\sim\curlywedge}$ and $V^\curlywedge V^\sim=(VV^{\sim\curlywedge})^\sim$ are in Clifford algebra center $\cl_0(p,q)\oplus\cl_3(p,q)$ and commute with all elements.
Theorem \[theoremDet\] give us explicit formulas for inverse in $\cl(p,q)$. We have the following theorem.
. \[theoremObrat\] Let $U$ be invertible element of Clifford algebra $\cl(p,q)$. Then we have the following expressions for $U^{-1}$: $$(U)^{-1}=\left\lbrace
\begin{array}{ll}
\frac{e}{U}, & n=0;\\ \\
\frac{U^\curlywedge}{U U^\curlywedge}, & n=1;\\ \\
\frac{U^{\sim\curlywedge}}{U U^{\sim\curlywedge}}, & n=2;\\ \\
\frac{U^\sim U^\curlywedge U^{\sim\curlywedge}}{U U^\sim U^\curlywedge U^{\sim\curlywedge}}=\frac{U^{\sim\curlywedge} U^\curlywedge U^\sim}{U U^{\sim\curlywedge} U^\curlywedge U^\sim}, & n=3;\\ \\
\frac{U^\sim (U^\curlywedge U^{\sim\curlywedge})^\bigtriangledown}{U U^\sim (U^\curlywedge U^{\sim\curlywedge})^\bigtriangledown}=\frac{U^{\sim\curlywedge} (U^\curlywedge U^{\sim})^\bigtriangledown}{ U U^{\sim\curlywedge} (U^\curlywedge U^{\sim})^\bigtriangledown}, & n=4;\\ \\
\frac{U^\sim (U^\curlywedge U^{\sim\curlywedge})^\bigtriangledown (U U^\sim (U^\curlywedge U^{\sim\curlywedge})^\bigtriangledown)^\bigtriangleup}{U U^\sim (U^\curlywedge U^{\sim\curlywedge})^\bigtriangledown (U U^\sim (U^\curlywedge U^{\sim\curlywedge})^\bigtriangledown)^\bigtriangleup}, & n=5.
\end{array}
\right.$$
Note, that in denominators we have Clifford algebra elements of the rank $0$. We identify them with the constants: $ue\equiv u$.
. Statement follows from Theorem \[theoremDet\]. $\blacksquare$
Note, that formulas for determinant in Theorem \[theoremDet\] are not unique. For example, in the case of $n=4$ we can use the following formulas, but only for even and odd Clifford algebra elements $U\in\cl_{\Even}(p,q)\cup\cl_{\Odd}(p,q)$.
Consider operation $+$, that acts on the even elements such that it changes the sign before the basis elements that anticommutes with $e^1$. For example, elements $e, e^{23}, e^{24}, e^{34}$ maps under $+$ into themselves, and elements $e^{12}, e^{13}, e^{14}, e^{1234}$ change the sign.
. \[theoremDet2\] Let $U\in\cl_{\Even}(p,q)$, $n=p+q=4$. Then $$\Det U= U U^\sim U^{\sim +} U^+.$$ Let $U\in\cl_{\Odd}(p,q)$, $n=p+q=4$. Then $$\Det U= U U^\sim U^{\sim} U.$$
. The proof is by direct calculation. $\blacksquare$
[99]{} Lounesto P., [*Clifford Algebras and Spinors*]{}, Cambridge Univ. Press (1997, 2001). Marchuk N.G., Shirokov D.S., [*Unitary spaces on Clifford algebras*]{}, Advances in Applied Clifford Algebras, Volume 18, Number 2 (2008). Marchuk N.G., Martynova S.E., [*Notions of determinant, spectrum and Hermitian conjugation of Clifford algebra elements*]{}, arXiv:math-ph/0307043 (2008). P. Dadbeh, [*Inverse and determinant in 0 to 5 dimensional Clifford algebra*]{}, arXiv:1104.0067v1 \[math-ph\] (2011).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, a stabilized extended finite element method is proposed for Stokes interface problems on unfitted triangulation elements which do not require the interface align with the triangulation. The velocity solution and pressure solution on each side of the interface are separately expanded in the standard nonconforming piecewise linear polynomials and the piecewise constant polynomials, respectively. Harmonic weighted fluxes and arithmetic fluxes are used across the interface and cut edges (segment of the edges cut by the interface), respectively. Extra stabilization terms involving velocity and pressure are added to ensure the stable inf-sup condition. It is proved that the convergence orders of error estimates are optimal. Moreover, the errors are robust with respect to the viscosity. Results of numerical experiments are presented to verify the theoretical analysis.'
author:
- 'Xiaoxiao He[^1].'
- 'Fei Song[^2]'
- 'Weibing Deng[^3]'
bibliography:
- 'ref.bib'
title: 'Stabilized nonconforming Nitsche’s extended finite element method for Stokes interface problems'
---
Stokes interface problems, NXFEM, nonconforming finite element
65N12, 65N15, 65N30
Introduction
============
A variety of phenomena with discontinuities exist in the real world. For example, because of the different physical parameters, the velocity has kinks and the pressure is discontinuous for the multiphase flow. Therefore, simulating such phenomena must treat the discontinuities carefully. Standard finite element methods can perform well when the interface coincides with mesh lines, known as the body-fitted meshes. Optimal convergence orders can be obtained for body-fitted meshes where every element is contained in one sub-region (see [@bs08; @ciarlet78]).
However, it is expensive to generate a good body-fitted meshes for the complicated interface or time-dependent interface problems. Therefore, varieties of numerical methods have been proposed to handle these difficulties. Those methods do allow that the interface is not aligned with the mesh. One way is the immersed finite element methods based on Cartesian meshes where the finite element basis functions are locally modified for elements cut by the interface to satisfy the jump conditions across the interface exactly or approximately. We can see [@hl99; @llw03; @gll08; @Kwak2009An; @gl10; @Lin2015Nonconforming; @Lin2015Partially] for elliptic interface problems and [@Lin2013; @Adjerid2015] for Stokes interface problems.
The other way is the extended finite element methods (XFEMs) based on unfitted meshes, which are mainly designed to solve the problems with discontinuities, kinks and singularities within elements. For XFEMs, extra basis functions are added for elements intersected by the interface so that the discontinuities can be captured, and the jump conditions are enforced by a variant of Nitsche’s approach. This method was first proposed by Hansbo and Hansbo in [@hh02] to solve the elliptic interface problems. Then a large number of related methods have been developed, such as [@m09; @Wu2010An; @Barrau2012A; @Capatina2014NONCONFORMING; @Wang2016High; @Burman2016Robust; @hwx2017; @Capatina2017Extension] for elliptic interface problems, [@Hansbo2014A; @Cattaneo2015Stabilized; @Wang2015A] for Stokes interface problems and [@Massing2018A] for Oseen problems.
From now on, we will refer to the Nitsche-XFEM schemes. In this paper we focus on the following two-phase Stokes problem of two fluids with different kinematic viscosities on a bounded polygonal domain $\Omega \in \mathbb{R}^2$. The whole domain is crossed by an interface $\Gamma$ which is assumed to have at least $C^2$-smooth and is divided into two open sets $\Omega_1$ and $\Omega_2$. Denote by $[v]=v|_{\Omega_1}-v|_{\Omega_2}$ the jump across the interface $\Gamma$. Then we study the problem as follows: Find a velocity $\mathbf{u}$ and a pressure $p$ such that $$\label{eP}\left\{
\begin{aligned}
& - {\nabla}\cdot\big(\mu(x) {\nabla}\mathbf{u}\big)+\nabla p = \mathbf{f},\qquad &\text{ in }{\Omega}_1\cup{\Omega}_2,\\
& \nabla \cdot \mathbf{u} =0,\qquad &\text{ in }{\Omega}_1\cup{\Omega}_2,\\
& [\mathbf{u}]=0,\ [p\mathbf{n}-\mu \nabla \mathbf{u}\cdot \mathbf{n}]=\sigma \kappa \mathbf{n}, \qquad &\text{ on } {\Gamma}, \\
& \mathbf{u} = 0,\qquad &\text{ on } {\partial}{\Omega},
\end{aligned}\right.$$ where $\mathbf{f}\in [L^2(\Omega)]^2$ and $\mu$ is a piecewise constant viscosity, namely $\mu|_{\Omega_i}=\mu_i >0$. $\sigma$ is the surface tension coefficient, $\kappa$ is the curvature of the interface , and $\mathbf{n}$ is the unit normal vector on $\Gamma$ pointing from $\Omega_1$ to $\Omega_2$.
It is well known that mixed finite elements are a typical choice of approximation spaces for the discrete formulation of a saddle point problem without interface. It would be natural to expect that same finite element spaces would be adequate to solve the interface problem using the Nitsche-XFEM formulation. Since the computational meshes of the XFEMs do not fit the interface, the approximation of the pressure may be unstable near the interface even though for the inf-sup stable finite elements (see [@Cattaneo2015Stabilized]). That is to say, XFEM broke the stability condition for mixed problems. Therefore, extra pressure stabilization approaches in the elements cut by the interface are used to ensure the inf-sup condition. The Nitsche-XFEM with $P_1-bubble/P_1$ couple functions was proposed by Laura et al. in [@Cattaneo2015Stabilized] which used the symmetric pressure stabilization operator based on Brezzi-Pitkaranta stability technique on the cut region to ensure the stability. In [@Hansbo2014A], a Nitsche-XFEM based on $P_1$-iso-$P_2/P_1$ elements to solve Stokes interface problems was proposed. In the method, extra stabilization terms for normal-derivative jumps over some element faces with respect to not only pressure but also velocity are also added. Recently, a Nitsche formulation for Stokes interface problems based on $P_1/P_1$ elements was developed in [@Wang2015A], where extra penalty terms that contained the difference between the solution and an $L^2$ projection of the solution for velocity and pressure on a patch of elements intersected by the interface were added to ensure the stability. This extra penalty terms are called ghost penalty which was proposed by Burman in [@Burman2010Ghost]. We remark that stability technique are also used to solve Stokes problem with interfaces in the context of fictitious domain method in [@Massing2014A] where pressure was stabilized by the extra penalty terms for the jumps in the normal velocity and pressure gradients near the interface.
In this paper, we will propose an accurate and stable extended finite element method for Stokes interface problems based on $nonconforming-P_1/P_0$ shape functions using unfitted meshes. Harmonic weight fluxes and arithmetic average fluxes are used on the interface and cut edges (the local segment of edges cut by the interface) respectively. Moreover, stabilization terms involving the jumps in the normal pressure on the edges and velocity gradients in the vicinity of the interface are added in our method. Optimal error estimates in energy and $L^2$ norms for velocity and in $L^2$ norm for pressure are obtained. Moreover, the errors do not depend on the jump of different viscosities. Numerical examples support our theoretical analysis.
The rest of this paper is organized as follows. In Section \[method\], we describe the Nitsche’s extended finite element method with nonconforming elements. In Section \[prepare\], we list some preliminary lemmas. The stable inf-sup condition and error analysis are given in Section \[estimation\]. Numerical tests are presented in Section \[test\]. Finally, we make a conclusion in Section \[conclude\].
Throughout the paper, $C_A,C_{A_1},C_{A_2},C_{A_3},C_{b_1},C_{b_2},C_p,C_{B_1},C_{B_2},C_{B_3},\cdots$ are used to denote the generic positive constants which are independent of $h$, the penalty parameters, and the jump of the coefficient $\mu$. We also use the shorthand notation $A\lesssim B$ and $B\gtrsim A$ for the inequality $A\leq C B$ and $B\geq CA$. $A\eqsim B$ is for the statement $A\lesssim B$ and $B\lesssim A$. Moreover, denote by $H^s(\Omega_1\cup\Omega_2)=:\{v: v|_{\Omega_i}\in H^s(\Omega_i),i=1,2\}$ the piecewise $H^s$ space on $\Omega_1\cup \Omega_2$ and by $||v||_{s,\Omega_1\cup\Omega_2}$ and $|v|_{s,\Omega_1\cup\Omega_2}$ its norm and semi-norm.
Finite element formulation {#method}
==========================
Let ${\left\{{\mathcal{T}}_{h}\right\}}$ be a family of conforming, quasi-uniform, and regular triangulations of the domain $\Omega$ independent of the location of the interface $\Gamma$ where the mesh should be fine enough to ensure the interface is well resolved. Define $h_K$ as diam$(K)$ and $h:=\max_{K\in {\mathcal{T}}_h}h_K$ for any $K\in {{\mathcal{T}}}_h$. Note that any element $K\in {\mathcal{T}}_h$ is considered as closed. Let us introduce the set of cut elements $G_h^\Gamma :=\{K\in {\mathcal{T}}_h: K\cap \Gamma \neq \varnothing\}$ and denote $\Gamma_K=K\cap\Gamma$ for $K\in G_h^\Gamma$. Denote ${\mathcal{T}}_{h,i}:=\{K\in{\mathcal{T}}_{h}: K\cap\Omega_i\neq \varnothing\}$. Then we define the elements restricted and extended sub-domains $\Omega_{h,i}^{+}$ and $\Omega_{h,i}^{-}$ $$\Omega_{h,i}^{+}:= \bigcup_{K\in {\mathcal{T}}_{h,i}}K,\ \Omega_{h,i}^{-}:=\bigcup_{K\in {\mathcal{T}}_{h,i}\setminus G^\Gamma_h}K.$$ Let ${\mathcal{F}}_{h,i}$, ${\mathcal{F}}_h^{nc,i}$ and ${\mathcal{F}}^{cut,i}_h$ denote the set of all the edges of ${\mathcal{T}}_{h,i}$, the set of uncut edges of ${\mathcal{T}}_{h,i}$ and the set of cut segments contained in $\Omega_i$ respectively. ${\mathcal{F}}_h^{nc,i}$ and ${\mathcal{F}}^{cut,i}_h$ are given by $${\mathcal{F}}^{nc,i}:=\{e\in {\mathcal{F}}_{h,i}: e=\partial K_l\cap\partial K_r, K_l,K_r \in\mathcal{T}_{h,i}, \text{and} \ e \subset \Omega_i\},$$ and $${\mathcal{F}}^{cut,i}_h:=\{\widetilde{e}= e\cap \Omega_i: e=\partial K_l \cap \partial K_r, K_l,K_r \in G^{\Gamma}_h\}.$$
Finally, the set of all the edges of $G^{\Gamma}_h$ restricted to the interior of $\Omega^{+}_{h,i}$ is considered ${\mathcal{F}}^{\Gamma,i}_h:=\{e=\partial K_l\cap \partial K_r: K_l,K_r\in \mathcal{T}_{h,i}, K_l\ \text{or}\ K_r \in G^{\Gamma}_h\}$.
We make the following assumptions (see [@Hansbo2014A]):
- It is assumed that the interface intersects the edge of each triangle at most two points and each (open) edge at most once, or that the interface coincides with one edge of the element.
- We assume that for each $K\in G^{\Gamma}_h$ there exists one $K'\subset \Omega_i,i=1,2$ such that $K'$ shares an edge or a vertex with $K$. That is to say, if $z\in \Omega_i$ is a vertex of $K$ and $\triangle_z$ denotes the patch of elements associated to $z$, i.e. $\triangle_z=\bigcup\{K: K\in {{\mathcal{T}}}^i_h, z\in \partial K\}$, then there exists an element $K'\subset\Omega_i$ such that $K'\in \triangle_z$.
- It is assumed that the mesh coincides with the outer boundary $\partial\Omega$.
Assumption $A_1$ and $A_2$ make the interface is well resolved by the mesh with an enough small mesh. In this paper, we assume $0< h\leq h_0$ for some enough small constant $h_0$ depends on the curvature of the interface (see [@hwx2017]).
Now we assume that the velocity space is $$V_i :=[\{v\in L^2(\Omega^{+}_{h,i}): v\in H^2(K)\ \forall K \in \mathcal{T}_{h,i}\}]^2, i=1,2,$$ and $$Q_i :=\{p\in L^2(\Omega^{+}_{h,i}): p\in H^1({K})\ \forall K \in \mathcal{T}_{h,i}\}, i=1,2.$$ Further, we define the weak velocity space by $$V:= \{\mathbf{v}: \mathbf{v}|_{\Omega_i}\in V_i|_{\Omega_i},i=1,2, \mathbf{v}|_{\partial\Omega}=0\},$$ and the weak pressure space by $$Q=\{p: p|_{\Omega_i}\in {Q_i}|_{\Omega_i}, p\in L^2_{\mu}(\Omega_1\cup \Omega_2)\},$$ where $L^2_{\mu}(\Omega_1\cup \Omega_2)=\{q\in L^2(\Omega):(\mu^{-1}q,1)_{\Omega_1\cup\Omega_2}=0\}$. We now introduce the couple of inf-sup stable spaces on the extended sub-domain $\Omega^{+}_{h,i}$, $$\begin{aligned}
V^i_h:=\Big[\{ &v\in L^2(\Omega^{+}_{h,i}): v|_K\in S_h(K) \ if \ K\in{{\mathcal{T}}}_{h,i};\\
& \ if \ e=\partial K_l\cap\partial K_r,\ K_l,K_r\in {{\mathcal{T}}}_{h,i}, \ then \ \int_e [v]ds=0;\\
& \ if \ e=\partial K \cap \partial \Omega,\ K\in \mathcal{T}_{h,i}, \ then \ \int_e vds =0\}\Big]^2, i=1,2,
\end{aligned}$$ with $S_h(K):=\text{span}\{\phi_l: \phi_l \in P_1(K), \frac{1}{|e_m|}\int_{e_m}\phi_l ds=\delta_{lm}, e_m\in \partial K,l,m=1,2,3\},$ and $$Q_{h,i}:=\{p\in L^2(\Omega^{+}_{h,i}): p|_{K}\in P_0(K), \forall k\in {\mathcal{T}}_{h,i}\}.$$ Then we denote a couple of finite element spaces. Let $V_h$ be the extended velocity space of nonconforming piecewise linear polynomials defined on ${\mathcal{T}}_h$ as follows: $$V_h:=\{\mathbf{v}|_{\Omega_i}\in {V_{h,i}}|_{\Omega_i}, i=1,2\},$$ and $Q_h$ be the extended pressure space of piecewise constant functions defined on ${\mathcal{T}}_h$ as follows: $$Q_h:=\{p|_{\Omega_i}\in {Q_{h,i}}|_{\Omega_i},i=1,2, p\in L^2_{\mu}(\Omega_1\cup \Omega_2)\}.$$ The above extended finite element spaces double the degrees of freedom in the elements which are cut by the interface. Clearly, $V_h\nsubseteq V$ and $Q_h\subseteq Q$.
Recalling the definition of ${\mathcal{F}}^{cut,i}$, for each edge $\widetilde{e}\in{\mathcal{F}}^{cut,i}_h$, there exist two cut elements $K_l, K_r\in G^{\Gamma}_h$ and $K^i_j=K_j\cap \Omega_i, j=l,r$ such that $\widetilde{e}=K^i_l\cap K^i_r$. Define jumps of $\mathbf{v}\in V+V_h$ and $p\in Q$, and jump of the flux of $\mathbf{v}$ by $[\mathbf{v}]=\mathbf{v}|_{K^i_l}-\mathbf{v}|_{K^i_r}$, $[p]=p|_{K^i_l}-p|_{K^i_r}$ and $[\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}]=\nabla \mathbf{v}|_{K^i_l}\cdot \mathbf{n}_{\widetilde{e}}-\nabla \mathbf{v}|_{K^i_r}\cdot \mathbf{n}_{\widetilde{e}}$, respectively, provided that $\mathbf{n}_{\widetilde{e}}$ is a unit normal vector to the edge $\widetilde{e}$ pointing from $K_l^i$ to $K_r^i$. Similarly, for $e\in {\mathcal{F}}^{nc,i}_h$, we can also define the jumps of $\mathbf{v}\in V+V_h$ and $p\in Q$ on $e$ and a unit normal vector to the edge $e$ by $\mathbf{n}_e$. In particular, we note that $[\mathbf{v}]=\mathbf{v}|_K$ for $e\in {\mathcal{F}}^{nc,i}_h$ and $e=\partial K\cap\partial\Omega$ with $K\in {{\mathcal{T}}}_{h,i}$. Further, we define jump $[\nabla \mathbf{v}]=\nabla \mathbf{v}|_{K_l}-\nabla \mathbf{v}|_{K_r}$ for $\mathbf{v}\in V+V_h$ on each edge $e\in{\mathcal{F}}^{\Gamma,i}_h$.
For any $\mathbf{v}\in V+V_h$ and weights $w_i,i=1,2$, we define the averages $\{\mathbf{v\}}_{w}$ and $\{\mathbf{v}\}^{w}$ on the interface $\Gamma$ as follows: $$\{\mathbf{v}\}_{w}=w_1\mathbf{v}_1|_{\Gamma}+w_2\mathbf{v}_2|_{\Gamma}, \ \{\mathbf{v}\}^{w}=w_2\mathbf{v}_1|_{\Gamma}+w_1\mathbf{v}_2|_{\Gamma},$$ where $\mathbf{v}_i=\mathbf{v}|_{\Omega_i},i=1,2$. Similarly, for any $p\in Q$ and weights $w_i,i=1,2$, we define the averages $\{p\}_{w}$ and $\{p\}^{w}$ on the interface $\Gamma$ as follows: $$\{p\}_{w}=w_1p_1|_{\Gamma}+w_2p_2|_{\Gamma}, \ \{p\}^{w}=w_2p_1|_{\Gamma}+w_1p_2|_{\Gamma},$$ where $p_i=p|_{\Omega_i},i=1,2$. In this paper, we use the so-called “harmonic weights" as adopted by [@Cai2011Discontinuous; @hwx2017], $$w_1=\frac{\mu_2}{\mu_1+\mu_2},\ w_2=\frac{\mu_1}{\mu_1+\mu_2}.$$ Likewise, we denote the arithmetic weights $\{\mathbf{v}\}_k$, $\{p\}_k$ on the cut edges $\widetilde{e}\in {\mathcal{F}}^{cut,i}_h$ by $$\{\mathbf{v}\}_{k}=\frac{1}{2}\mathbf{v}_l|_{\widetilde{e}}+\frac{1}{2}\mathbf{v}_r|_{\widetilde{e}},\ \{p\}_{k}=\frac{1}{2}p_l|_{\widetilde{e}}+\frac{1}{2}p_r|_{\widetilde{e}},$$ where $\mathbf{v}_j=\mathbf{v}|_{K^i_j}$, $p_j=p|_{K^i_j},j=l,r$ provided $\widetilde{e}=\partial K^i_l\cap \partial K^i_r$, $K^i_j=K_j\cap \Omega_i$ for $K_l,K_r\in G^{\Gamma}_h$.
We propose the following Nitsche method: find $(\mathbf{u}_h,p_h)\in V_h\times Q_h$ such that $$\label{numer_sol}
B_h[(\mathbf{u}_h,p_h),(\mathbf{v}_h,q_h)]=L(\mathbf{v}_h), \forall (\mathbf{v}_h,q_h)\in V_h\times Q_h.$$ Here $B_h[(\cdot),(\cdot)]$ is a bilinear form defined by $$B_h[(\mathbf{u}_h,p_h),(\mathbf{v}_h,q_h)]=A_h(\mathbf{u}_h,\mathbf{v}_h)+b_h(p_h,\mathbf{v}_h)-b_h(q_h,\mathbf{u}_h)+J_p(p_h,q_h),$$ where $A_h(\mathbf{u}_h,\mathbf{v}_h)=a_h(\mathbf{u}_h,\mathbf{v}_h)+J_\mathbf{u}(\mathbf{u}_h,\mathbf{v}_h)$, $a_h(\cdot,\cdot)$, $J_{\mathbf{u}}(\cdot,\cdot)$ are the bilinear forms on $(V+V_h)\times (V+V_h)$ defined by $$\begin{aligned}
a_h(\mathbf{u},\mathbf{v})&=\sum^2_{i=1}\sum_{K\in {{\mathcal{T}}}_{h,i}}\int_{K\cap\Omega_i}\mu_i\nabla \mathbf{u}\cdot \nabla \mathbf{v}-\sum_{K\in G^{\Gamma}_h}\int_{{\Gamma}_K} \Big({\left\{\mu{\nabla}\mathbf{u}\cdot{\mathbf{n}}\right\}}_w{\left[\mathbf{v}\right]}\\
\\
&\quad +{\left[\mathbf{u}\right]}{\left\{\mu{\nabla}\mathbf{v}\cdot{\mathbf{n}}\right\}}_w\Big)\label{eah}+ \sum_{K\in G^{\Gamma}_h} \int_{{\Gamma}_K}\frac{{\gamma}_0\{\mu\}_w}{h} {\left[\mathbf{u}\right]}{\left[\mathbf{v}\right]}\\
&\quad+\sum^2_{i=1}\sum_{\widetilde{e}\in {\mathcal{F}}^{cut,i}_h}\Big(\int_e(-{\left\{\mu_i{\nabla}\mathbf{u}\cdot{\mathbf{n}}_{\widetilde{e}}\right\}}_k{\left[\mathbf{v}\right]}-{\left\{\mu_i{\nabla}\mathbf{v}\cdot{\mathbf{n}}_{\widetilde{e}}\right\}}_k{\left[\mathbf{u}\right]})\\
\\
&\quad +{\gamma}_i |e|^{-1}\mu_i\int_e [\mathbf{u}][\mathbf{v}]\Big),
\end{aligned}$$ and $$J_\mathbf{u}(\mathbf{u},\mathbf{v})=\sum^2_{i=1}\left(\sum_{e\in {\mathcal{F}}^{\Gamma,i}_h}|e|\mu_i\int_e[\nabla \mathbf{u}][\nabla \mathbf{v}]+\sum_{\widetilde{e}\in{\mathcal{F}}^{cut,i}_h }\int_{\widetilde{e}} |\widetilde{e}|\mu_i[\nabla \mathbf{u}\cdot \mathbf{n}_{\widetilde{e}}][\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}]\right),$$ $b_h(\cdot,\cdot)$ is defined in $Q\times(V+V_h)$ by $$\begin{aligned}
b_h(p,\mathbf{v})&=-\sum^2_{i=1}\sum_{K\in {{\mathcal{T}}}_{h,i}}\int_{K\cap\Omega_i}p\nabla\cdot \mathbf{v}\\
&\quad+\sum_{\widetilde{e}\in{\mathcal{F}}^{cut,i}_h }\int_{\widetilde{e}} \{p\}_k[\mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}]+ \sum_{K\in G^{\Gamma}_h}\int_{\Gamma_K}\{p\}_w[\mathbf{v}\cdot \mathbf{n}],
\end{aligned}$$ $J_p(\cdot,\cdot)$ is defined in $Q\times Q$ by $$J_p(p,q)=\sum^2_{i=1}\left(\sum_{e\in {\mathcal{F}}^{\Gamma,i}_{h}}|e|\int_e \mu_i^{-1}[p][q]+\sum_{\widetilde{e}\in {\mathcal{F}}^{cut,i}_{h}}|\widetilde{e}|\int_{\widetilde{e}} \mu_i^{-1}[p][q]\right),$$ and $L_h(\cdot)$ is a linear form defined by $$L_h(\mathbf{v})=(f,\mathbf{v})_{\Omega}+\sum_{K\in G^{\Gamma}_h}\int_{\Gamma_K}\sigma\kappa\{\mathbf{v}\cdot \mathbf{n}\}^w,$$ where $\gamma_0$, $\gamma_1$ and $\gamma_2$ are positive parameters to be chosen in Lemma \[ah-coer\].
The stabilization terms $J_\mathbf{u}$, $J_p$ appeared in the method are all consistent. The term $J_\mathbf{u}(\mathbf{u}_h,\mathbf{v}_h)$ is added to ensure the coercivity of $A_h(\cdot,\cdot)$ and the term $J_p(p_h,q_h)$ is used to prove the inf-sup stability of the method.
For any $\mathbf{u}\in {[H^2(\Omega_1\cup\Omega_2)\cap H_0^1(\Omega)]}^2$ and $p\in H^1(\Omega_1\cup\Omega_2)\cap L^2_{\mu}(\Omega)$, it is easy to see that the following equality holds, $$\label{orth}
\begin{aligned}
&B_h[(\mathbf{u}-\mathbf{u}_h,p_h),(\mathbf{v}_h,q_h)]\\
&=\sum^2_{i=1}\sum_{e\in {\mathcal{F}}_{h}^{nc,i}}\left(\int_e\mu_i \nabla \mathbf{u}\cdot \mathbf{n}_e[\mathbf{v}_h]+\int_e p[\mathbf{v}_h\cdot \mathbf{n}_e]\right),\ \forall (v_h,q_h)\in V_h\times Q_h.
\end{aligned}$$ Now we introduce the norms. For $\mathbf{v}\in V+V_h$, we define $$\label{norm1}
\begin{aligned}
|||\mathbf{v}|||^2:=&\sum^2_{i=1}\sum_{K\in {{\mathcal{T}}}_{h,i}}||\sqrt{\mu_i}\nabla \mathbf{v}||^2_{0,K\cap\Omega_i}+\frac{\{\mu\}_w}{h}\sum_{K\in G^{\Gamma}_h}||[\mathbf{v}]||^2_{0,\Gamma_K}\\
&+\sum^2_{i=1}\sum_{\widetilde{e}\in {\mathcal{F}}^{cut,i}_h} |\widetilde{e}|^{-1}\mu_i||[\mathbf{v}]||^2_{0,\widetilde{e}}+J_\mathbf{u}(\mathbf{v},\mathbf{v}),
\end{aligned}$$ and $$\label{norm2}
\begin{aligned}
|||\mathbf{v}|||^2_V:=&|||\mathbf{v}|||^2+\frac{h}{\{\mu\}_w}\sum_{K\in G^{\Gamma}_h}||\{\mu\nabla \mathbf{v}\cdot \mathbf{n}\}_w||_{0,\Gamma_K}^2\\
&+\sum^2_{i=1}\sum_{\widetilde{e}\in {\mathcal{F}}^{cut,i}_h}|\widetilde{e}|\mu_i||\{\nabla \mathbf{v}\cdot \mathbf{n}\}_k||^2_{0,\widetilde{e}}.
\end{aligned}$$ For $(\mathbf{v},p)\in (V+V_h)\times Q$, we define $$|||(\mathbf{v},p)|||^2:=|||\mathbf{v}|||^2+||\mu^{-1/2}p||^2_{0,\Omega}+J_p(p,p),$$ and $$\begin{aligned}
|||(\mathbf{v},p)|||_V^2:=&|||\mathbf{v}|||^2_V+||\mu^{-1/2}p||^2_{0,\Omega}+J_p(p,p)\\
&+\frac{h}{\{\mu\}_w}\sum_{K\in G^{\Gamma}_h}||\{p\}_w||^2_{0,\Gamma_K}+\sum^2_{i=1}\sum_{\widetilde{e}\in {\mathcal{F}}^{cut,i}_h}|\widetilde{e}|\mu_i^{-1}||\{p\}_k||^2_{0,\widetilde{e}}.
\end{aligned}$$
Preliminary {#prepare}
===========
In this section, we will give some preliminaries for the later error analysis. Firstly, we give the following Lemma, where Lemma \[sta2\] is proved in [@Guzm2015A].
\[sta2\] If $\widetilde{e}\in {\mathcal{F}}^{cut,i}_h$, that is to say, $\widetilde{e}=\partial K^i_l\cap\partial K^i_r$, where $K_l,\ K_r\in G^{\Gamma}_h$ and $K^i_j =K_j\cap \Omega_i, j=l,r$, then there exists a constant $\theta >0$ such that $$|\widetilde{e}|^2\leq \theta \max_{j=l,r}|K_j^i|.$$ The constant $\theta$ depends on the $C^2$-norm of the parametrization of $\Gamma$ and the shape regularity of $K_l$ and $K_r$.
Recalling the definition of interface segment $\Gamma_K$ in Section \[method\], we have the following trace inequality for the interface segment (see [@Wu2010An]).
\[tr\_intf\] For each $K\in G^\Gamma_h$ and $v\in H^1(K)$, it holds $${\left\Vertv\right\Vert}_{0,\Gamma_K}\lesssim h^{-1/2}_K{\left\Vertv\right\Vert}_{0,K}+{\left\Vertv\right\Vert}^{1/2}_{0,K}{\left\Vert\nabla v\right\Vert}^{1/2}_{0,K}.$$ Further, if $v\in P_1(K)$, then $${\left\Vertv\right\Vert}_{0,\Gamma_K}\lesssim h^{-1/2}_K{\left\Vertv\right\Vert}_{0,K}.$$
In order to estimate the error of our method, we need the following trace inequality for the cut segments contained in $\Omega_i$.
\[tr\_cut\][ Suppose that $v\in H^2(K)$ for $K\in G_h^\Gamma$. If $e\in \partial K$ and $\widetilde{e }\in {\mathcal{F}}^{cut,i}_h$ such that $\widetilde{e}\subseteq e$, then we have]{} $$\frac{1}{|\widetilde{e}|}{\left\Vertv\right\Vert}^2_{0,\widetilde{e}}\leq C\left(\frac{1}{h^2_K}{\left\Vertv\right\Vert}^2_{0,K}+{\left\Vert\nabla v\right\Vert}^2_{0,K}+h^2_K|\nabla v|^2_{1,K}\right).$$
Since [$\widetilde{e}\subseteq e$]{}, from the Lemma 3 of [@Dryja1994Domain], we have $$\frac{1}{|\widetilde{e}|}{\left\Vertv\right\Vert}^2_{L^2(\widetilde{e})}\leq C\left(\frac{1}{|e|}{\left\Vertv\right\Vert}^2_{L^2(e)}+|e|{\left\Vert\nabla v\right\Vert}^2_{L^2(e)}\right),$$ which combines with the trace inequality yields the result.
Then we give the properties of $A_h(\cdot,\cdot)$ and $b_h(\cdot,\cdot)$.
\[ah-coer\] The bilinear discrete form $A_h(\cdot,\cdot)$ is coercive on $V_h$ provided $\gamma_i,i=0,1,2$ are chosen large enough. That is, $$A_h(\mathbf{v},\mathbf{v})\geq \frac{1}{2}|||\mathbf{v}|||^2,\ \forall \mathbf{v}\in V_h.$$
Denote $\mathbf{v}_i=\mathbf{v}|_{\Omega_i}$. From the definition of $A_h(\cdot,\cdot)$, we have $$\begin{aligned}
A_h(\mathbf{v},\mathbf{v})=&\sum^2_{i=1}\sum_{K\in \mathcal{T}_{h,i}}{\left\Vert\sqrt{\mu_i}\nabla \mathbf{v}\right\Vert}^2_{0,K\cap\Omega_i}-2\sum_{K\in G^{\Gamma}_h}\int_{\Gamma_K} \{\mu\nabla \mathbf{v}\cdot\mathbf{n}\}_w[\mathbf{v}]\\
&+\frac{\gamma_0\{\mu\}_w}{h}\sum_{K\in G^{\Gamma}_h}{\left\Vert[\mathbf{v}]\right\Vert}^2_{0,\Gamma_K}-2\sum^2_{i=1}\sum_{\widetilde{e}\in {\mathcal{F}}_h^{cut,i}}\int_{\widetilde{e}}\{{\mu}_i\nabla \mathbf{v}\cdot \mathbf{n_{\widetilde{e}}}\}_k[\mathbf{v}]\\
&+\sum^2_{i=1}\sum_{\widetilde{e}\in {\mathcal{F}}^{cut,i}}\gamma_i\mu_i|\widetilde{e}|^{-1}{\left\Vert[\mathbf{v}]\right\Vert}^2_{0,\widetilde{e}}+J_{\mathbf{u}}(\mathbf{v},\mathbf{v}).
\end{aligned}$$ We mainly focus on the two nonsymmetric terms. For the first term, applying the Cauchy-Schwarz and Young inequalities, we have $$\begin{aligned}
2\sum_{K\in G^{\Gamma}_h}&\int_{\Gamma_K} \{\mu\nabla \mathbf{v}\cdot\mathbf{n}\}_w[\mathbf{v}]\\
&\leq \sum_{K\in G^{\Gamma}_h}\frac{h}{\{\mu\}_w\epsilon}||\{\mu\nabla \mathbf{v}\cdot\mathbf{n}\}_w||^2_{0,\Gamma_K}+\sum_{K\in G^{\Gamma}_h}\frac{\{\mu\}_w\epsilon}{h}||[\mathbf{v}]||^2_{0,\Gamma_K},
\end{aligned}$$ where $\epsilon$ is a positive number to be defined later. Further, using the fact that $\frac{\mu^2_iw^2_i}{\{\mu\}_w} \leq \frac{\mu_i}{2}$, we have $$\begin{aligned}
\frac{h}{\{\mu\}_w}||\{\mu\nabla \mathbf{v}\cdot\mathbf{n}\}_w||^2_{0,\Gamma_K}&\leq \sum^2_{i=1}2h\frac{\mu^2_iw^2_i}{\{\mu\}_w}||\nabla \mathbf{v}_i\cdot\mathbf{n}||^2_{0,\Gamma_K}\\
&\leq h\sum^2_{i=1}\mu_i||\nabla \mathbf{v}_i\cdot\mathbf{n}||^2_{0,\Gamma_K}.
\end{aligned}$$ Since the interface $\Gamma$ is $C^2$-smooth, it is easy to obtain $|\Gamma_K|\eqsim h_K$. Moreover, according to Assumption $A_2$, for $K\in G^{\Gamma}_h$, let $z\in \Omega_i$ be a vertex of $K$ and $\triangle_z$ denote the patch of elements associated to $z$, i.e. $\triangle_z=\bigcup\{K: K\in {{\mathcal{T}}}_{h,i}, z\in \partial K\}$, then there exist an element $K'\subset\Omega_i$ such that $K'\in \triangle_z$. We assume that $$e_1\in \partial K, e_1=\partial K\cap \partial K_2, e_2=\partial K_2\cap \partial K_3,\cdots, e_{m-1}=\partial K_{m-1}\cap \partial K', e_m\in \partial K'.$$ Since the triangulation is conforming, quasi-uniform and regular, $|e_1|^2 \eqsim |K|$, $|e_i|^2\eqsim |K_i|, i=2,3,\ldots,m-1$, $|e_m|^2\eqsim |K'|$ and $h\eqsim |e_i|, i=1,2,\ldots,m$. Then using the fact that $\nabla \mathbf{v}_i$ is a constant vector, we have $$\label{ah-coer-1}
\begin{aligned}
h||\nabla \mathbf{v}_i\cdot\mathbf{n}||^2_{0,\Gamma_K}&\lesssim |e_1|\ ||\nabla {\mathbf{v}_i}|_{K\cap\Omega_i}||^2_{0,e_1}\\
&\leq |e_1|\left(||\nabla {\mathbf{v}_i}|_{K_2\cap\Omega_i}||^2_{0,e_1}+||[\nabla \mathbf{v}]||^2_{0,e_1}\right)\\
&\lesssim |e_2|\ ||\nabla {\mathbf{v}_i}|_{K_2\cap\Omega_i}||^2_{0,e_2}+|e_1|\ ||[\nabla \mathbf{v}]||^2_{0,e_1}\\
&\leq |e_2|(||\nabla {\mathbf{v}_i}|_{K_3\cap\Omega_i}||^2_{0,e_2}+||[\nabla \mathbf{v}]||^2_{0,e_2})+|e_1|\ ||[\nabla \mathbf{v}]||^2_{0,e_1}\\
&\lesssim |e_3|\ ||\nabla {\mathbf{v}_i}|_{K_3\cap\Omega_i}||^2_{0,e_3}+\sum^2_{i=1}|e_i|\ ||[\nabla \mathbf{v}]||^2_{0,e_i}\\
&\vdots\\
&\lesssim |e_m|\ ||\nabla {\mathbf{v}_i}|_{K'}||^2_{0,e_m}+\sum^{m-1}_{i=1}|e_i|\ ||[\nabla \mathbf{v}]||^2_{0,e_i}\\
&\lesssim ||\nabla {\mathbf{v}_i}||^2_{0,K'}+\sum^{m-1}_{i=1}|e_i|\ ||[\nabla \mathbf{v}]||^2_{0,e_i}.
\end{aligned}$$ Hence $$\label{ah-coer-2}
\begin{aligned}
&\sum_{K\in G^{\Gamma}_h}\frac{h}{\{\mu\}_w}||\{\mu\nabla \mathbf{v}\cdot\mathbf{n}\}_w||^2_{0,\Gamma_K}\\
&\qquad\quad\leq \sum^2_{i=1}C_{i0} \left(\sum_{K\in {{\mathcal{T}}}_{h,i}\setminus G^{\Gamma}_h}\mu_i||\nabla \mathbf{v}||^2_{0,K}+\sum_{e\in {{\mathcal{F}}}^{\Gamma,i}_h}\mu_i|e|\ ||[\nabla \mathbf{v}]||^2_{0,e}\right).
\end{aligned}$$ Next, we estimate $$2\sum^2_{i=1}\sum_{\widetilde{e}\in {\mathcal{F}}_h^{cut,i}}\int_{\widetilde{e}}\{{\mu}_i\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}\}_k[\mathbf{v}].$$ For any $\widetilde{e}\in {\mathcal{F}}_h^{cut,i}$, let $K_l, K_r\in G^{\Gamma}_h$ and $K_l^i=K_l\cap \Omega_i$, $K_r^i=K_r\cap \Omega_i$ satisfy $\widetilde{e}=\partial K_l^i\cap \partial K_r^i$. Without loss of generality, we assume $|K_l^i|=\max_{j=l,r}|K_j^i|$. According to Lemma \[sta2\], $$\label{cute_cell}
|\widetilde{e}|^2\leq \theta |K_l^i|.$$ It is easy to see $$\label{ce_1}
\sum_{\widetilde{e}\in {\mathcal{F}}_h^{cut,i}}\int_{\widetilde{e}}\{{\mu}_i\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}\}_k[\mathbf{v}]= \sum_{\widetilde{e}\in {\mathcal{F}}_h^{cut,i}}\int_{\widetilde{e}}\mu_i(\nabla \mathbf{v}|_{K_l}\cdot \mathbf{n}_{\widetilde{e}}\pm \frac{1}{2}[\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}])[\mathbf{v}].$$ Using the fact that $\nabla \mathbf{v}$ is a constant vector, from , we have $$\label{ce_2}
\begin{aligned}
\int_{\widetilde{e}}\mu_i\nabla \mathbf{v}|_{K_l}\cdot \mathbf{n}_{\widetilde{e}}[\mathbf{v}]&\leq \sqrt{\theta}{\left\Vert\sqrt{\mu_i}\nabla \mathbf{v}\right\Vert}_{0,K_l^i}\sqrt{\frac{\mu_i}{|\widetilde{e}|}}{\left\Vert[\mathbf{v}]\right\Vert}_{0,\widetilde{e}}\\
&\leq \frac{\theta}{2\epsilon}{\left\Vert\sqrt{\mu_i}\nabla \mathbf{v}\right\Vert}^2_{0,K_l^i}+\frac{\epsilon}{2|\widetilde{e}|}\mu_i{\left\Vert[\mathbf{v}]\right\Vert}^2_{0,\widetilde{e}}.
\end{aligned}$$ Further, using the Cauchy-Schwarz inequality, we infer that $$\label{ce_3}
\int_{\widetilde{e}}\mu_i[\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}][\mathbf{v}]\leq \frac{1}{2}\mu_i|\widetilde{e}|\ {\left\Vert[\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}]\right\Vert}^2_{0,\widetilde{e}}+\frac{1}{2}\mu_i|\widetilde{e}|^{-1}{\left\Vert[\mathbf{v}]\right\Vert}^2_{0,\widetilde{e}}.$$ Then, collecting the above two estimates (\[ce\_2\]) and (\[ce\_3\]) , we obtain $$\begin{aligned}
&2\int_{\widetilde{e}}\mu_i(\nabla \mathbf{v}|_{K_l}\cdot \mathbf{n}_{\widetilde{e}}\pm \frac{1}{2}[\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}])[\mathbf{v}]\\
&\qquad\leq \frac{\theta}{\epsilon}{\left\Vert\sqrt{\mu_i}\nabla \mathbf{v}\right\Vert}^2_{0,K_l^i}+\frac{1}{2}\mu_i|\widetilde{e}|\ {\left\Vert[\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}]\right\Vert}^2_{0,\widetilde{e}}+\frac{\epsilon+1/2}{|\widetilde{e}|}\mu_i{\left\Vert[\mathbf{v}]\right\Vert}^2_{0,\widetilde{e}}.
\end{aligned}$$ Finally, $$\begin{aligned}
&A_h(\mathbf{v},\mathbf{v})\geq \sum^2_{i=1}\left(1-\frac{C_{i0}+\theta}{\epsilon}\right)\sum_{K\in {\mathcal{T}}_{h,i}}||\sqrt{\mu_i}\nabla \mathbf{v}||^2_{0,K\cap \Omega_i}\\
&\quad+\sum^2_{i=1}\sum_{e\in {{\mathcal{F}}}^{\Gamma,i}_h}\left(1-\frac{C_{i0}}{\epsilon}\right)\mu_i|e|\ ||[\nabla \mathbf{v}]||^2_{0,e}+(\gamma_0-\epsilon)\frac{\{\mu\}}{h}\sum_{K\in G^{\Gamma}_h}||[\mathbf{v}]||^2_{0,\Gamma_K}\\
&\quad+\frac{1}{2}\sum^2_{i=1}\sum_{\widetilde{e}\in {{\mathcal{F}}}^{cut,i}_h}\mu_i|\widetilde{e}|\ ||[\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}]||^2_{0,\widetilde{e}}+\sum^2_{i=1}\sum_{\widetilde{e}\in {{\mathcal{F}}}^{cut,i}_h}\frac{\gamma_i-\epsilon-1/2}{|\widetilde{e}|}\mu_i\ ||[\mathbf{v}]||^2_{0,\widetilde{e}}.
\end{aligned}$$ Taking $\epsilon=2~(\max_{i=1,2}\{C_{i0}\}+\theta)$, we can conclude the result by choosing $$\gamma_0>{\varepsilon}+1/2, \gamma_i>{\varepsilon}+1, i=1,2.$$ This completes the proof.
Applying the Cauchy-Schwarz inequality and the above process of Lemma \[ah-coer\], we obtain the following Lemma.
\[continu\_ah\] There exist three positive constants $ C_{A_1}$, $ C_{A_2}$ and $ C_{A_3}$ such that $$A_h(\mathbf{u},\mathbf{v})\leq C_{A_1}|||\mathbf{u}|||_V |||\mathbf{v}|||_V, \ \forall \mathbf{u},\mathbf{v}\in V,$$ $$A_h(\mathbf{u},\mathbf{v})\leq C_{A_2}|||\mathbf{u}|||_V |||\mathbf{v}|||, \ \forall \mathbf{u}\in V,\ \mathbf{v}\in V_h,$$ and $$|||\mathbf{v}|||_V\leq C_{A_3} |||\mathbf{v}|||, \ \forall \mathbf{v}\in V_h.$$
The first inequality can be obtained directively by using Cauchy-Schwarz inequality. Then we only need to prove the third inequality which infers the second one. From , we have $$\sum_{K\in G^{\Gamma}_h}\frac{h}{\{\mu\}_w}||\{\mu\nabla \mathbf{v}\cdot\mathbf{n}\}_w||^2_{0,\Gamma_K}\lesssim |||\mathbf{v}|||^2.$$ Next we bound $$\sum^2_{i=1}\sum_{\widetilde{e}\in {\mathcal{F}}^{cut,i}_h}\mu_i|\widetilde{e}|{\left\Vert\{\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}\}_k\right\Vert}^2_{L^2(\widetilde{e})}.$$ For any $\widetilde{e}\in {\mathcal{F}}_h^{cut,i}$, let $K_l, K_r\in G^{\Gamma}_h$ and $K_l^i=K_l\cap \Omega_i$, $K_r^i=K_r\cap \Omega_i$ satisfy $\widetilde{e}=\partial K_l^i\cap \partial K_r^i$, and assume $|K_l^i|=\max_{j=l,r}|K_j^i|$. According to Lemma \[sta2\], $|\widetilde{e}|^2\leq \theta |K_l^i|$ holds. Then $$\begin{aligned}
\mu_i|\widetilde{e}|{\left\Vert\{\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}\}_k\right\Vert}^2_{L^2(\widetilde{e})}&\leq \mu_i|\widetilde{e}|\sum_{j=l,r}||\nabla \mathbf{v}|_{K^i_j}\cdot \mathbf{n}_{\widetilde{e}}||^2_{L^2(\widetilde{e})}\\
&\leq \mu_i|\widetilde{e}|\left(||\nabla \mathbf{v}|_{K^i_l}\cdot \mathbf{n}_{\widetilde{e}}||^2_{L^2(\widetilde{e})}+||[\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}]||^2_{L^2(\widetilde{e})}\right)\\
&\lesssim ||\sqrt{\mu_i}\nabla \mathbf{v}||^2_{L^2(K^i_l)}+\mu_i|\widetilde{e}|\ ||[\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}]||^2_{L^2(\widetilde{e})}.
\end{aligned}$$ Hence, $$\sum^2_{i=1}\sum_{\widetilde{e}\in {\mathcal{F}}^{cut,i}_h}\mu_i|\widetilde{e}|{\left\Vert\{\nabla \mathbf{v}\cdot \mathbf{n}_{\widetilde{e}}\}_k\right\Vert}^2_{L^2(\widetilde{e})}\lesssim |||\mathbf{v}|||^2.$$ Recalling the definitions of $|||\mathbf{v}|||_V$ and $|||\mathbf{v}|||$, we complete the lemma.
\[continu\_bh\] There exist three positive constants $C_{b_1}$, $C_{b_2}$ and $C_{p}$ such that, for any $\mathbf{v}\in V+V_h, p\in Q$, the following inequality holds $$\label{bh_1}
\begin{aligned}
b_h(p,\mathbf{v})\leq& C_{b_1}|||\mathbf{v}|||_V\Big(\sum^2_{i=1}\sum_{K\in {{\mathcal{T}}}_{h,i}}||\mu^{-1/2}p||_{0,K\cap\Omega_i}\\
&+\left(\frac{h}{\{\mu\}_w}\right)^{\frac{1}{2}}\sum_{K\in G^{\Gamma}_h}||\{p\}_w||_{0,\Gamma_K}\quad+\sum^2_{i=1}\sum_{\widetilde{e}\in {\mathcal{F}}^{cut,i}_h}\left(\frac{|\widetilde{e}|}{\mu_i}\right)^{\frac{1}{2}}||\{p\}_k||_{0,\widetilde{e}}\Big).
\end{aligned}$$ Additionally, for any $\mathbf{v}\in V+V_h, p\in Q_h$, then $$\label{bh_2}
\begin{aligned}
&b_h(p,\mathbf{v})\leq C_{b_2}|||\mathbf{v}|||_V\left(\sum^2_{i=1}\sum_{K\in {{\mathcal{T}}}_{h,i}}||\mu^{-1/2}p||^2_{0,K\cap\Omega_i}+J_p(p,p)\right)^{\frac{1}{2}}.
\end{aligned}$$ Furthermore, for any $ p\in Q_h$, we have $$\label{bh_3}
\mu^{-1}_i\sum_{K\in {{\mathcal{T}}}_{h,i}}||p||^2_{0,K\cap\Omega_i}\leq C_p\left(\mu^{-1}_i\sum_{K\in {{\mathcal{T}}}_{h,i}\setminus G^{\Gamma}_h}||p||^2_{0,K}+J_p(p,p)\right).$$
It is easy to obtain the first inequality by using Cauchy-Schwarz inequality directly. We focus on the proof of and . Using $\frac{hw^2_i}{\{\mu\}_w}\leq \frac{h}{2\mu_i}$, we have $$\frac{h}{\{\mu\}_w}\sum_{K\in G^{\Gamma}_h}||\{p\}_w||^2_{0,\Gamma_K}\lesssim \sum_{i=1}^2\frac{h}{\mu_i}\sum_{K\in G^{\Gamma}_h}||p|_{\Omega_i}||^2_{0,\Gamma_K}.$$ For $K\in G^{\Gamma}_h$, similar to , we have $$\label{bh_proof0}
\begin{aligned}
h\sum_{K\in G^{\Gamma}_h}||p|_{\Omega_i}||^2_{0,\Gamma_K}&\lesssim \sum_{K\in {{\mathcal{T}}}_{h,i}\setminus G^{\Gamma}_h}||p||^2_{0,K}+\sum_{e\in {{\mathcal{F}}}^{\Gamma,i}_h}|e|\ ||[p]||^2_{0,e}.
\end{aligned}$$ Hence $$\label{bh_proof1}
\begin{split}
&\frac{h}{\{\mu\}_w}\sum_{K\in G^{\Gamma}_h}||\{p\}_w||^2_{0,\Gamma_K}\\
&\qquad\lesssim \sum^2_{i=1}\mu^{-1}_i\left(\sum_{K\in {{\mathcal{T}}}_{h,i}}||p||^2_{0,K\cap\Omega_i}+\sum_{e\in {\mathcal{F}}^{\Gamma,i}_h}|e|\ ||[p]||^2_{0,e}\right).
\end{split}$$ For any $\widetilde{e}\in {\mathcal{F}}^{cut,i}_h$, there exist two elements $K_l,K_r\in G^{\Gamma}_h$ so that $e=\partial K^i_l\cap K^i_r$ where $K^i_j=K_j\cap \Omega_i, j=l,r$. Assume $|K^i_l|=\max_{j=l,r}|K^i_j|$. According to Lemma \[sta2\], we have $|\widetilde{e}|^2\lesssim |K^i_l|$. Applying the fact that $p\in Q_{h,i}$ is a piecewise constant polynomial, we obtain $$\label{bh_proof2}
\begin{aligned}
\frac{|\widetilde{e}|}{\mu_i}||\{p\}_k||^2_{0,\widetilde{e}}&\lesssim \mu^{-1}_i|\widetilde{e}|\sum_{j=l,r}||p|_{K^i_j}||^2_{0,\widetilde{e}}\\
& \lesssim\mu^{-1}_i|\widetilde{e}|\left(||p|_{K^i_l}||^2_{0,\widetilde{e}}+||[p]||^2_{0,\widetilde{e}}\right)\\
&\lesssim \mu^{-1}_i\left(||p||^2_{0,K^i_l}+|\widetilde{e}|\ ||[p]||^2_{0,\widetilde{e}}\right).
\end{aligned}$$ We can prove by combining with and .
For any $q\in Q_h$, it is clear that there exist $q_{i}\in Q_{h,i}$ such that ${q}|_{\Omega_i}={q_i}|_{\Omega_i}, i=1,2$. Noticing that $$\label{bh_proof4}
\begin{aligned}
\mu^{-1}_i\sum_{K\in {{\mathcal{T}}}_{h,i}}&||p||^2_{0,K\cap\Omega_i}\leq \mu^{-1}_i\left(\sum_{K\in {{\mathcal{T}}}_{h,i}\setminus G^{\Gamma}_h}||p||^2_{0,K}+\sum_{K\in G^{\Gamma}_h}||p_i||^2_{0,K}\right),
\end{aligned}$$ and for $e\in\partial K, K\in G^{\Gamma}_h$ $$||p_i||^2_{0,K}\lesssim |e|\ ||p_i||^2_{0,e}.$$ Similar to , the following inequality holds $$\sum_{K\in G^{\Gamma}_h}||p_i||^2_{0,K}\lesssim \sum_{K\in {{\mathcal{T}}}_{h,i}\setminus G^{\Gamma}_h}||p||^2_{0,K}+\sum_{e\in {{\mathcal{F}}}^{\Gamma,i}_h}|e|\ ||[p]||^2_{0,e},$$ which combines yields .
Combining with Lemma \[continu\_ah\], Lemma \[continu\_bh\] and Cauchy-Schwarz inequality, we can easily obtain the following Lemma.
\[con\_bilinear\] There exist three constants $C_{B_1}$ $C_{B_2}$ and $C_{B_3}$ such that $$B_h[(\mathbf{u},p),(\mathbf{v},q)]\leq C_{B_1}|||(\mathbf{u},p)|||_V|||(\mathbf{v},q)|||_V, \ \forall (\mathbf{u},p),(\mathbf{v},q) \in (V+V_h)\times Q,$$ $$B_h[(\mathbf{u},p),(\mathbf{v},q)]\leq C_{B_2}|||(\mathbf{u},p)|||_V|||(\mathbf{v},q)|||, \ \forall (\mathbf{u},p)\in (V+V_h, Q), (\mathbf{v},q)\in V_h\times Q_h,$$ and $$|||(\mathbf{v},q)|||_V\leq C_{B_3}|||(\mathbf{v},q)|||, \ \forall (\mathbf{v},q)\in V_h\times Q_h.$$
Error analysis {#estimation}
==============
In this section, we will give the priori error estimates.
We will first prove the stability of the scheme.
Suppose that $\gamma_i, i=0,1,2$ are large enough. Let $(\mathbf{u}_h,p_h)\in V_h\times Q_h$, then $$\sup_{0\neq(\mathbf{v}_h,q_h)\in V_h\times Q_h}\frac{B_h[(\mathbf{u}_h,p_h),(\mathbf{v}_h,q_h)]}{|||(\mathbf{v}_h,q_h)|||}\geq C_s|||(\mathbf{u}_h,p_h)|||.$$
First,we start by choosing $(\mathbf{v_h},q_h)=(\mathbf{u_h},p_h)$ to obtain $$B_h[(\mathbf{u}_h,p_h),(\mathbf{u}_h,p_h)]=A_h(\mathbf{u}_h,\mathbf{u}_h)+J_p(p_h,p_h).$$ Using the Lemma \[ah-coer\], we get $$\label{infsup_1}
B_h[(\mathbf{u}_h,p_h),(\mathbf{u}_h,p_h)]\geq \frac{1}{2}|||\mathbf{u_h}|||^2+J_p(p_h,p_h).$$ Next, for $p_h\in Q_h$, we let $\overline{p_{h,i}}=\frac{1}{|\Omega_{h,i}^{-}|}\int_{\Omega_{h,i}^{-}}p_h$, then $\mu_i^{-1}(p_h-\overline{p_{h,i}})\in L^2_0({\Omega_{h,i}^{-}})$. By the surjectivity of the divergence operator from $[H_0^1(\Omega^{-}_{h,i})]^2$ to $L^2_0(\Omega^{-}_{h,i})$, there exists $\mathbf{v}_{p_{h,i}}\in [H^1_0(\Omega^{-}_{h,i})]$ such that $$-\nabla\cdot \mathbf{v}_{p_{h,i}} = \mu_i^{-1}(p_h-\overline{p_{h,i}}), \ \text{ in } \Omega^{-}_{h,i},$$ and $$\label{infsup_01}
||\mathbf{v}_{p_{h,i}}||_{1,\Omega^{-}_{h,i}}\lesssim\mu_i^{-1}||p_h-\overline{p_{h,i}}||_{0,\Omega^{-}_{h,i}}\leq C_{i1}\mu_i^{-1}||p_h||_{0,\Omega^{-}_{h,i}}.$$
Taking $(\mathbf{v}_h,q_h)=(I_h\mathbf{v}_{p_h},0)$, where $(I_h\mathbf{v}_{p_h})|_{\Omega_i}=(\Pi_h{(E_i{\mathbf{v}_{p_{h,i}}})|_{\Omega^{+}_{h,i}}})|_{\Omega_i}$, $E_i$ is a extension by zero of $\mathbf{v}_{p_{h,i}}$ from $H^1_0(\Omega^{-}_{h,i})$ to $H^1_0(\Omega)$, $\Pi_{h}$ is the standard Crouzein-Raviart interpolation operator to $V^i_{h}$ which satisfies $\int_{e}I_{h}\mathbf{v}_{p_{h}}=0$ for $e\in {\mathcal{F}}_{h,i}\cap{\partial \Omega^{-}_{h,i}}$ and $I_{h}\mathbf{v}_{p_{h}}=0$ on $G_h^{\Gamma}$ and $$\label{infsup_02}
\sum_{K\in {{\mathcal{T}}}_{h,i}\setminus G^{\Gamma}_h}||\nabla I_h\mathbf{v}_{p_h}||_{0,K}\leq C_{i2}||\mathbf{v}_{p_{h,i}}||_{1,\Omega^{-}_{h,i}}.$$ Then, from the definition of $B_h$, we have $$\label{infsup_2}
B_h[(\mathbf{u}_h,p_h),(I_h \mathbf{v}_{p_h},0)]=A_h(\mathbf{u}_h,I_h \mathbf{v}_{p_h})+b_h(p_h,I_h \mathbf{v}_{p_h}).$$ Using the Continuity of $A_h(\cdot,\cdot)$, and , we infer that $$\label{infsup_3}
\begin{aligned}
A_h(\mathbf{u}_h,I_h \mathbf{v}_{p_h})&\geq -C_A|||\mathbf{u}_h|||\ |||I_h\mathbf{v}_{p_h}|||\\
&\geq -C_A|||\mathbf{u}_h|||\ \sum^2_{i=1}\sum_{K\in {{\mathcal{T}}}_{h,i}\setminus G^{\Gamma}_h}||\sqrt{\mu_i}\nabla I_h\mathbf{v}_{p_h}||_{0,K}\\
&\geq -C_A|||\mathbf{u}_h|||\ \sum^2_{i=1}{\mu_i}^{1/2}C_{i2}||\mathbf{v}_{p_{h,i}}||_{1,\Omega^{-}_{h,i}}\\
&\geq -C_A|||\mathbf{u}_h|||\ \left(\sum^2_{i=1}C_{i1}C_{i2}||\mu^{-1/2}_ip_h||_{0,\Omega^{-}_{h,i}}\right),
\end{aligned}$$ where $C_A=C_{A_1}C^2_{A_3}$. Using the fact that $p_h\in Q_h$ and $I_h\mathbf{v}_{p_h}\in V_h$ we have $$\label{infsup_4}
\begin{aligned}
b_h(p_h,I_h\mathbf{v}_{p_h})&=-\sum^2_{i=1}\sum_{K\in {{\mathcal{T}}}_{h,i}\setminus G^{\Gamma}_h}p_h \int_{K} \nabla\cdot (I_h\mathbf{v}_{p_h})\\
&=-\sum^2_{i=1}\sum_{K\in {{\mathcal{T}}}_{h,i}\setminus G^{\Gamma}_h}p_h \int_{\partial K}I_h\mathbf{v}_{p_h} \cdot \mathbf{n}_K\\
&=-\sum^2_{i=1}\sum_{K\in {{\mathcal{T}}}_{h,i}\setminus G^{\Gamma}_h}p_h \int_{\partial K}\mathbf{v}_{p_h} \cdot \mathbf{n}_K\\
&=-\sum^2_{i=1}\sum_{K\in {{\mathcal{T}}}_{h,i}\setminus G^{\Gamma}_h}\int_{K}p_h\nabla\cdot \mathbf{v}_{p_h}\\
&=b_h(p_h,\mathbf{v}_{p_h})=\sum^2_{i=1}||\mu^{-1/2}_ip_h||^2_{0,\Omega^{-}_{h,i}},
\end{aligned}$$ where $\mathbf{n}_K$ is the outer unit normal vector to $\partial K$. Combining with and , yields $$\label{infsup_5}
\begin{aligned}
&B_h[(\mathbf{u}_h,p_h),(I_h \mathbf{v}_{p_h},0)]\\
&\qquad\geq \sum^2_{i=1}||\mu^{-1/2}_ip_h||^2_{0,\Omega^{-}_{h,i}}-C_AC_{max}|||\mathbf{u}_h|||\left(\sum^2_{i=1}||\mu^{-1/2}_ip_h||_{0,\Omega^{-}_{h,i}}\right),
\end{aligned}$$ where $C_{max}=\max_{i=1,2}\{C_{i1}C_{i2}\}$.
Finally, let $(\mathbf{v}_h,q_h)=(\mathbf{u}_h+\eta I_h\mathbf{v}_{p_h},p_h)$, we get $$\begin{aligned}
&B_h[(\mathbf{u}_h,p_h),(\mathbf{u}_h+\eta I_h \mathbf{v}_{p_h},p_h)]\\
&=B_h[(\mathbf{u}_h,p_h),(\mathbf{u}_h,p_h)]+\eta B_h[(\mathbf{u}_h,p_h),(I_h\mathbf{v}_{p_h},0)]\\
&\geq \frac{1}{2}|||\mathbf{u}_h|||^2+J_p(p_h,p_h)-\frac{\epsilon}{2}|||\mathbf{u}_h|||^2\\
&\quad-\frac{C^2_AC^2_{max}\eta^2}{2\epsilon}\sum^2_{i=1}||\mu^{-1/2}_ip_h||^2_{0,\Omega^{-}_{h,i}}+\eta\sum^2_{i=1}||\mu^{-1/2}_ip_h||^2_{0,\Omega^{-}_{h,i}}\\
&\geq\left( \frac{1}{2}-\frac{\epsilon}{2}\right)|||\mathbf{u}_h|||^2+J_p(p_h,p_h)+
\left(\eta-\frac{C^2_AC^2_{max}\eta^2}{2\epsilon}\right)\sum^2_{i=1}||\mu^{-1/2}_ip_h||^2_{0,\Omega^{-}_{h,i}}\\
&\geq \frac{1}{4}|||\mathbf{u}_h|||^2+\frac{1}{2}J_p(p_h,p_h)+\min\{\frac{1}{2},\eta-C^2_AC^2_{max}\eta^2\}C^{-1}_p\sum^2_{i=1}||\mu^{-1/2}_ip_h||^2_{0,\Omega_{i}}\\
&\geq C_s|||(\mathbf{u}_h,p_h)|||^2.
\end{aligned}$$ The above inequality holds by choosing $C_s=\min\{\frac{1}{4}, \frac{1}{2C_p}, (\eta-C^2_AC^2_{max}\eta^2)C^{-1}_p\}$ and $\epsilon$ and $\eta$ such that $\epsilon=\frac{1}{2}$ and $0<\eta<\frac{1}{C_A^2C_{max}^2}$. Combining with $|||(\mathbf{u}_h+I_h \mathbf{v}_{p_h},p_h)|||\lesssim |||(\mathbf{u}_h,p_h)|||$, we prove this theorem.
To obtain the priori error estimates, we need the interplation operators and their approximation error. To show these, we need construct the extension operator $E^2_i: [H^2(\Omega_i)\cap H^1_0(\Omega)]^2\rightarrow [H^2(\Omega)\cap H^1_0(\Omega)]^2$, and $E^1_i: H^1(\Omega_i)\rightarrow H^1(\Omega)\cap L^2_{\mu}(\Omega),i=1,2$ such that $(E^2_i\mathbf{v},E^1_iq)|_{\Omega_i}=(\mathbf{v},q)$, $$||E^2_i\mathbf{v}||_{s,\Omega}\leq C||\mathbf{v}||_{s,\Omega_i}, s=0,1,2,$$ where $\mathbf{v}\in [H^s(\Omega_i)\cap H^1_0(\Omega)]^2$ for $i=1,2$, and $$||E^1_iq||_{r,\Omega}\leq C||q||_{r,\Omega_i}, \ \forall q\in H^r(\Omega_i), r=0,1.$$
For any piecewise $H^2$ function $\mathbf{v}\in [H^2(\Omega_1\cup\Omega_2)]^2$ and any piecewise $H^1$ function $q\in H^1(\Omega_1\cup\Omega_2)$, let $\mathbf{v}_i={E^2_i(\mathbf{v}|_{\Omega_i})}|_{\Omega^{+}_{h,i}}$, $q_i={E^1_i(q|_{\Omega_i})}|_{\Omega^{+}_{h,i}}$ be the extension of the restriction of $\mathbf{v}$ and $q$ on $\Omega_i$ to $\Omega^{+}_{h,i}, i=1,2$, respectively. Let $\Pi^1_h$ be the standard Crouzeix-Raviart interpolant and $\Pi^0_h$ be the standard $L^2-$ projection operator onto piecewise constants space. We define interpolations $I_h$ on $V_h$ and $R_h$ on $Q_h$ by $$\label{inter_ope}
((I_h\mathbf{v})|_{\Omega_i},(R_hq)|_{\Omega_i}):=((I_h{\mathbf{v}}_i)|_{\Omega_i},(R_hq_i)|_{\Omega_i}), i=1,2,$$ where $I_h{\mathbf{v}}_i=\Pi^1_h\mathbf{v}_i$ and $R_hq_i=\Pi^0_hq_i$. Combined with the interpolation error and property of the extension operator, the following theorem is valid.
\[appr\_err\] Suppose that $\mathbf{v}\in [H^2(\Omega_1\cup\Omega_2)\cap H^1_0(\Omega)]^2$ and $q\in H^1(\Omega_1\cup\Omega_2)\cap L^2_{\mu}(\Omega)$ and $(I_h\mathbf{v},R_hq)$ be a pair of interpolant operators defined as in . Then $$|||(\mathbf{v}-I_h\mathbf{v},q-R_hq)|||_V \lesssim h\sum^2_{i=1}\left(\mu^{1/2}_i|\mathbf{v}|_{2,\Omega_i}+\mu^{-1/2}_i|q|_{1,\Omega_i}\right).$$
Denote by ${\mathbf{w}}_i = \mathbf{v}_i-I_h\mathbf{v}_i$, $\zeta_i =q_i-R_hq_i, i=1,2$ and $\mathbf{w}=\mathbf{v}-I_h\mathbf{v}$, $\zeta=q-R_hq$. Clearly, $\mathbf{w}|_{\Omega_i}={{\mathbf{w}}_i}|_{\Omega_i}$, $q|_{\Omega_i}={q_i}|_{\Omega_i}, i=1,2$. From the standard finite element interpolation theory in [@Ern2004Theory], for $j=0,1,2$, $$||{\mathbf{w}}_i||_{j,K}\lesssim h^{s-j}|\mathbf{v}_i|_{s,K},j\leq s\leq 2,$$ and for $l=0,1$ $$||\zeta_i||_{l,K}\lesssim h^{m-l}|q_i|_{m,K}, l\leq m\leq 1.$$ Further, collecting the property of extension operator, we have $$\sum_{K\in {{\mathcal{T}}}_{h,i}}||{\mathbf{w}}_i||^2_{j,K}\lesssim h^{4-2j}|\mathbf{v}_i|_{2,\Omega^{+}_{h,i}}\lesssim h^{4-2j}|\mathbf{v}|^2_{2,\Omega_i},\ j=0,1,2,$$ and $$\sum_{K\in {{\mathcal{T}}}_{h,i}}||\zeta_i||_{l,K}\lesssim h^{2-2l}|q_i|_{1,\Omega^{+}_{h,i}}\lesssim h^{2-2l}|q|_{1,\Omega_i}, l=0,1.$$ Next we estimate each term of $|||(\mathbf{w},\zeta)|||_V$. Clearly $$\begin{aligned}
&\sum_{K\in {{\mathcal{T}}}_{h,i}}||\mu_i^{1/2}\nabla \mathbf{w}||^2_{0,K\cap\Omega_i}\lesssim \mu_ih^2|v|^2_{2,\Omega_i},\\
&\sum_{K\in {{\mathcal{T}}}_{h,i}}||\mu_i^{-1/2}\zeta_i||^2_{0,K\cap\Omega_i}\lesssim \mu_i^{-1}h^2|q|^2_{1,\Omega_i}.
\end{aligned}$$ Further, using the fact $\{\mu\}_w\leq 2\mu_i, i=1,2$ and Lemma \[tr\_intf\], the following estimate holds $$\begin{aligned}
\frac{\{\mu\}_w}{h}\sum_{K\in G^{\Gamma}_h}||[\mathbf{w}]||^2_{0,\Gamma_K}&\leq \sum^2_{i=1}\frac{\mu_i}{h}\sum_{K\in G^{\Gamma}_h}||\mathbf{w}_i||^2_{0,\Gamma_K}\\
&\lesssim\sum^2_{i=1}\frac{\mu_i}{h}\sum_{K\in G^{\Gamma}_h}\left(h^{-1}_K||{\mathbf{w}}_i||^2_{0,K}+||{\mathbf{w}}_i||_{0,K}||\nabla {\mathbf{w}}_i||_{0,K}\right)\\
&\lesssim \sum_{i=1}^2\mu_i h^2|\mathbf{v}|^2_{2,\Omega_i}.
\end{aligned}$$ From $\frac{\mu_i^2w^2_i}{\{\mu\}_w}\leq \frac{\mu_i}{2},i=1,2$ and Lemma \[tr\_intf\], $$\begin{aligned}
\frac{h}{\{\mu\}_w}\sum_{K\in G^{\Gamma}_h}||\{\mu \nabla \mathbf{w} \cdot \mathbf{n}\}||^2_{0,\Gamma_K}&\leq \sum^2_{i=1}\sum_{K\in G^{\Gamma}_h}\mu_i h||\nabla \mathbf{w}_i \cdot \mathbf{n}||^2_{0,\Gamma_K}\\
&\lesssim \sum^2_{i=1}\sum_{K\in G^{\Gamma}_h}\mu_i h \left(h^{-1}_K||\nabla \mathbf{w}_i||^2_{0,K}+|\nabla \mathbf{w}_i|_{1,K}||\nabla \mathbf{w}_i||_{0,K}\right)\\
&\lesssim \sum_{i=1}^2\mu_i h^2|\mathbf{v}|^2_{2,\Omega_i}.
\end{aligned}$$ Similarly, from $\frac{w^2_i}{\{\mu\}_w}\leq \frac{1}{2\mu_i},i=1,2$, we have $$\frac{h}{\{\mu\}_w}\sum_{K\in G^{\Gamma}_h}||\{\zeta\}||^2_{0,\Gamma_K}\leq \sum^2_{i=1}\sum_{K\in G^{\Gamma}_h}\frac{ h}{\mu_i}||\zeta_i||^2_{0,\Gamma_K}\lesssim \sum_{i=1}^2\mu_i^{-1} h^2|q|^2_{1,\Omega_i}.$$ Using the triangle inequality and trace inequality, we obtain $$\sum_{e\in {{\mathcal{F}}}^{\Gamma,i}_h}\mu_i |e|\ ||[\nabla \mathbf{w}]||^2_{0,e}\lesssim\mu_i h^2|\mathbf{v}|^2_{2,\Omega_i},\ \sum_{e\in {{\mathcal{F}}}^{\Gamma,i}_h}\mu^{-1}_i |e|\ ||[\zeta]||^2_{0,e}\lesssim \mu^{-1}_i h^2|q|^2_{1,\Omega_i}.$$ For any $\widetilde{e}\in {{\mathcal{F}}}^{cut,i}_h$, we assume that $e=\partial K_l\cap \partial K_r, K_l,K_r\in {{\mathcal{T}}}_{h,i}$ and $\widetilde{e}\subseteq e$. Using Lemma \[tr\_cut\], we have $$\begin{aligned}
|\widetilde{e}|^{-1}||[\mathbf{w}_i]||^2_{0,\widetilde{e}}&\leq |\widetilde{e}|^{-1}\sum_{j=l,r}||\mathbf{w}_i|_{K_j}||^2_{0,\widetilde{e}}\\
&\lesssim \sum_{j=l,r}\left(\frac{1}{h^2_{K_j}}||\mathbf{w}_i||^2_{0,K_j}+||\nabla \mathbf{w}_i||^2_{0,K_j}+h^2_{K_j}|\nabla \mathbf{w}_i|^2_{1,K_j}\right)\\
&\lesssim \sum_{j=l,r}h^2|\mathbf{v}_i|^2_{2,K_j}.
\end{aligned}$$ Thus $$\sum_{e\in {{\mathcal{F}}}^{cut,i}_h}\mu_i|e|^{-1}||[\mathbf{w}_i]||^2_{0,\widetilde{e}}\lesssim \mu_ih^2|\mathbf{v}|^2_{2,\Omega_i}.$$ Applying the triangle and standard trace inequalities again, we have $$\begin{aligned}
\sum_{\widetilde{e}\in {{\mathcal{F}}}^{cut,i}}\mu_i|\widetilde{e}|\ ||[\nabla \mathbf{w}_i\cdot \mathbf{n}_{\widetilde{e}}]||^2_{\widetilde{e}}&\leq \sum_{e\in {{\mathcal{F}}}^{cut,i}}\mu_i|e|\ ||[\nabla \mathbf{w}_i\cdot \mathbf{n}_{e}]||^2_{e}\\
&\lesssim \sum_{\widetilde{e}\in {{\mathcal{F}}}^{cut,i}}\mu_i|e|\sum_{j=l,r}\left(|e|^{-1}||\nabla \mathbf{w}_i||^2_{0,K_j}+|e|\ |\nabla \mathbf{w}_i|^2_{1,K_j} \right)\\
&\lesssim \mu_i h^2 |\mathbf{v}|^2_{2,\Omega_i}.
\end{aligned}$$ Similarly, $$\sum_{\widetilde{e}\in {{\mathcal{F}}}^{cut,i}}\mu_i|\widetilde{e}|\ ||\{\nabla \mathbf{w}_i\cdot \mathbf{n}_{\widetilde{e}}\}_k||^2_{0,\widetilde{e}}\lesssim \mu_i h^2 |\mathbf{v}|^2_{2,\Omega_i}, \ \sum_{\widetilde{e}\in {{\mathcal{F}}}^{cut,i}}\mu^{-1}_i|\widetilde{e}|\ ||[\zeta_i]||^2_{\widetilde{e}}\lesssim \mu_i^{-1}h^2|q|^2_{1,\Omega_i},$$ and $$\sum_{\widetilde{e}\in {{\mathcal{F}}}^{cut,i}}\mu^{-1}_i|\widetilde{e}|\||\{\zeta\}_k||^2_{0,\widetilde{e}}\lesssim \mu_i^{-1}h^2|q|^2_{1,\Omega_i}.$$ So far, we complete the proof.
\[energy\_err\] Let $(\mathbf{u},p)$ be the weak solution of and $(\mathbf{u}_h,p_h)$ be the solution of the finite element formulation respectively. Assume that the solution $(\mathbf{u},p)\in [H^2(\Omega_1\cup\Omega_2)\cap H^1_0(\Omega)]^2\times H^1(\Omega_1\cup\Omega_2)\cap L^2_{\mu}(\Omega_1\cup\Omega_2)$ and the assumptions (A1)–(A3) hold. Suppose that $\gamma_i, i=0,1,2$ are large enough. Then the following error estimate holds $$|||(\mathbf{u}-\mathbf{u}_h,p-p_h)|||\lesssim h\sum^2_{i=1}\left(\mu^{1/2}_i|\mathbf{u}|_{2,\Omega_i}+\mu^{-1/2}_i|p|_{1,\Omega_i}\right).$$
Using the triangulation inequality, we have $$\label{errorterm_1}
|||(\mathbf{u}-\mathbf{u}_h,p-p_h)|||\leq |||(\mathbf{u}-I_h\mathbf{u},p-R_hp)|||+|||(I_h\mathbf{u}-\mathbf{u}_h,R_hp-p_h)|||.$$ For the second term, we use the inf-sup condition, Lemma \[con\_bilinear\] and then get $$\label{errorterm_2}
\begin{aligned}
&|||(I_h\mathbf{u}-\mathbf{u}_h,R_hp-p_h)|||\lesssim\sup_{0\neq(\mathbf{v}_h,q_h)\in V_h\times Q_h}\frac{B_h[(I_h\mathbf{u}-\mathbf{u}_h,R_hp-p_h),(\mathbf{v}_h,q_h)]}{|||(\mathbf{v}_h,q_h)|||}\\
&= \sup_{0\neq(\mathbf{v}_h,q_h)\in V_h\times Q_h}\frac{B_h[(\mathbf{u}-\mathbf{u}_h,p-p_h),(\mathbf{v}_h,q_h)]+B_h[(I_h\mathbf{u}-\mathbf{u},R_hp-p),(\mathbf{v}_h,q_h)]}{|||(\mathbf{v}_h,q_h)|||}\\
&\lesssim \sup_{0\neq(\mathbf{v}_h,q_h)\in V_h\times Q_h}\frac{B_h[(\mathbf{u}-\mathbf{u}_h,p-p_h),(\mathbf{v}_h,q_h)]}{|||(\mathbf{v}_h,q_h)|||}+|||(\mathbf{u}-I_h\mathbf{u},p-R_hp)|||_V.
\end{aligned}$$ From the equality , we get $$\label{inter_orth}
\begin{aligned}
&B_h[(\mathbf{u}-\mathbf{u}_h,p-p_h),(\mathbf{v}_h,q_h)]=\sum^2_{i=1}\sum_{e\in {\mathcal{F}}_{h}^{nc,i}}\left(\int_e\mu_i \nabla \mathbf{u}\cdot \mathbf{n}_e[\mathbf{v}_h]+\int_e p[\mathbf{v}_h\cdot \mathbf{n}_e]\right).
\end{aligned}$$ Let $\overline{v}=\frac{1}{|e|}\int_e v$. From the Poincar$\acute{e}$ inequality, we have $$||v-\overline{v}||_{0,e}\lesssim |e|\ ||\nabla v||_{0,e}.$$ For $e\in {{\mathcal{F}}}^{nc,i}_h$ is the non-cut edge which is totally contained in $\Omega_i$, the related two elements $K_l,K_r\in {{\mathcal{T}}}_{h,i}$ with $e=\partial K_l \cap\partial K_r$ can be classified three cases. One case is that $K_l,K_r$ are also contained in $\Omega_i$, we have $$\begin{aligned}
&\int_e\mu_i \nabla \mathbf{u}\cdot \mathbf{n}_e[\mathbf{v}_h]=\int_e\mu_i \{\nabla\mathbf{u}\cdot\mathbf{n}_e-\nabla I_h\mathbf{u}\cdot \mathbf{n}_e\}_k[\mathbf{v}_h-\overline{\mathbf{v_h}}]\\
&\leq \mu_i||\{\nabla\mathbf{u}\cdot\mathbf{n}_e-\nabla I_h\mathbf{u}\cdot \mathbf{n}_e\}_k||_{0,e}\sum_{j=l,r}|e|\ ||{\nabla \mathbf{v_h}}|_{K_j}||_{0,e}\\
&\lesssim \mu_i\sum_{j=l,r}\left(|e|^{-1/2}||\nabla (\mathbf{u}-I_h\mathbf{u})||_{0,K_j}+|e|^{1/2}|\nabla (\mathbf{u}-I_h\mathbf{u})|_{1,K_j}\right)\sum_{j=l,r}|e|\ ||{\nabla \mathbf{v_h}}|_{K_j}||_{0,e}\\
&\lesssim \sqrt{\mu_i}h|\mathbf{u}|_{2,K_l\cup K_r}\sum_{j=l,r}|e|^{1/2}\ ||\sqrt{\mu_i}\nabla \mathbf{v_h}|_{K_j}||_{0,e}\\
&\leq \mu_i^{1/2}h|\mathbf{u}|_{2,K_l\cup K_r}\sum_{j=l,r}||\sqrt{\mu_i}\nabla \mathbf{v_h}||_{0,K_j},
\end{aligned}$$ where we have used that $\nabla I_h\mathbf{v_h}$ is a constant matrix and $\int_e[\mathbf{v_h}]=0$. Similarly, for $\int_e p[\mathbf{v}_h\cdot \mathbf{n}_e]$ we have $$\begin{aligned}
\int_e p[\mathbf{v}_h\cdot \mathbf{n}_e]&=\int_e\{p-R_hp\}_k[\mathbf{v}_h\cdot \mathbf{n}_e-\overline{\mathbf{v_h}\cdot \mathbf{n}_e}]\\
&\lesssim\mu_i^{-1/2}h|p|_{1,K_l\cup K_r}\sum_{j=l,r}||\sqrt{\mu_i}\nabla \mathbf{v_h}||_{0,K_j}.
\end{aligned}$$ For the second case, we assume that $K_l$ is totally contained in $\Omega_i$ and $K_r\in G^{\Gamma}_h$. $$\begin{aligned}
\int_e\mu_i \nabla \mathbf{u}\cdot \mathbf{n}_e[\mathbf{v}_h]&=\int_e\mu_i \left(\nabla\mathbf{u}\cdot\mathbf{n}_e-\nabla {I_h\mathbf{u}}|_{K_l}\cdot \mathbf{n}_e\right)[\mathbf{v}_h-\overline{\mathbf{v_h}}]\\
&\leq\mu_i||\nabla\mathbf{u}\cdot\mathbf{n}_e-\nabla {I_h\mathbf{u}}|_{K_l}\cdot \mathbf{n}_e||_{0,e}||[\mathbf{v}_h-\overline{\mathbf{v_h}}]||_{0,e}\\
&\lesssim \mu_ih|\mathbf{u}|_{2,K_l}\sum_{j=l,r}|e|^{1/2}||{\nabla \mathbf{v_h}}|_{K_j}||_{0,e}\\
&\leq \mu_ih|\mathbf{u}|_{2,K_l}|e|^{1/2}\left(||{\nabla \mathbf{v_h}}|_{K_l}||_{0,e}+||{\nabla \mathbf{v_h}}|_{K_l}\pm[\nabla\mathbf{v_h}]||_{0,e}\right)\\
&\lesssim \sqrt{\mu_i}h|\mathbf{u}|_{2,K_l}\left(||\sqrt{\mu_i}{\nabla \mathbf{v_h}}||_{0,K_l}+|e|^{1/2}\sqrt{\mu_i}||[\nabla\mathbf{v_h}]||_{0,e}\right).
\end{aligned}$$ Likewise, for $\int_e p[\mathbf{v}_h\cdot \mathbf{n}_e]$ we have $$\begin{aligned}
\int_e p[\mathbf{v}_h\cdot \mathbf{n}_e]&=\int_e(p-R_hp)|_{K_l}[\mathbf{v}_h\cdot \mathbf{n}_e-\overline{\mathbf{v_h}\cdot \mathbf{n}_e}]\\
&\lesssim\mu_i^{-1/2}h|p|_{1,K_l}\left(||\sqrt{\mu_i}{\nabla \mathbf{v_h}}||_{0,K_l}+|e|^{1/2}\sqrt{\mu_i}||[\nabla\mathbf{v_h}]||_{0,e}\right).
\end{aligned}$$ Last case is that $e\in {\mathcal{F}}^{nc,i}_h$ and $e\in \partial K\cap\partial\Omega$ for $K\in {\mathcal{T}}_{h,i}$. Similarly, we have $$\begin{aligned}
\int_e\mu_i \nabla \mathbf{u}\cdot \mathbf{n}_e[\mathbf{v}_h]&=\int_e\mu_i \left(\nabla\mathbf{u}\cdot\mathbf{n}_e-\nabla {I_h\mathbf{u}}|_{K}\cdot \mathbf{n}_e\right)(\mathbf{v}_h-\overline{\mathbf{v_h}})\\
&\lesssim \mu_i^{1/2}h|\mathbf{u}|_{2,K}||\sqrt{\mu_i}\nabla\mathbf{v_h}||_{0,K},
\end{aligned}$$ and $$\begin{aligned}
\int_e p[\mathbf{v}_h\cdot \mathbf{n}_e]&=\int_e(p-R_hp)|_{K}(\mathbf{v}_h\cdot \mathbf{n}_e-\overline{\mathbf{v_h}\cdot \mathbf{n}_e})\\
&\lesssim\mu_i^{-1/2}h|p|_{1,K}||\sqrt{\mu_i}\nabla\mathbf{v_h}||_{0,K},
\end{aligned}$$ where we have used $\int_e \mathbf{v}_h=0$ for $\mathbf{v}_h\in V_h$.
Hence, is estimated by $$\label{errorterm_3}
B_h[(\mathbf{u}-\mathbf{u}_h,p-p_h),(\mathbf{v}_h,q_h)]\lesssim h\sum^2_{i=1}\left(\mu^{1/2}_i|\mathbf{u}|_{2,\Omega_i}+\mu^{-1/2}_i|p|_{1,\Omega_i}\right)|||\mathbf{v_h}|||.$$ The theorem follows by combining , , and Theorem \[appr\_err\].
Using the Aubin-Nitsche duality argument we prove the following $L^2$-estimate. Consider the dual adjoint problem. Let $\mathbf{z}$ and $r$ be the solution of the problem $$\label{arg_ep}\left\{
\begin{aligned}
& - {\nabla}\cdot\big(\mu(x) {\nabla}\mathbf{z}\big)-\nabla r = \mathbf{u}-\mathbf{u}_h,\qquad &\text{ in }{\Omega}_1\cup{\Omega}_2,\\
& \nabla \cdot \mathbf{z} =0,\qquad &\text{ in }{\Omega}_1\cup{\Omega}_2,\\
& [\mathbf{z}]=0,\ [r\mathbf{n}-\mu \nabla \mathbf{z}\cdot \mathbf{n}]=0, \qquad &\text{ on } {\Gamma}, \\
& \mathbf{z} = 0,\qquad &\text{ on } {\partial}{\Omega}.
\end{aligned}\right.$$ We assume that the solution of the adjoint problem satisfies the following regularity $$\mu_i|\mathbf{z}|_{2,\Omega_i}+|r|_{1,\Omega_i}\lesssim ||\mathbf{u}-\mathbf{u}_h||_{0,\Omega}.$$
\[l2\] Under the same assumptions of Theorem \[energy\_err\], there holds $$||\mathbf{u}-\mathbf{u}_h||_{0,\Omega}\lesssim \mu^{-1/2}_{min}h^2\sum^2_{i=1}\left(\mu^{1/2}_i|\mathbf{u}|_{2,\Omega_i}+\mu^{-1/2}_i|p|_{1,\Omega_i}\right),$$ where $\mu_{min}=\min_{i=1,2}\{\mu_i\}$.
Multiply the equation by $\mathbf{u}$, integrating on each sub-domain and using integration by parts, we have $$\begin{aligned}
(\mathbf{u}-\mathbf{u}_h,\mathbf{u})&=\sum^2_{i=1}\int_{\Omega_i}\left(\mu_i\nabla \mathbf{z}\cdot \nabla \mathbf{u}+ r\nabla\cdot \mathbf{u}\right)-\int_{\Gamma}\left(\mu\nabla \mathbf{z}\cdot \mathbf{n} \mathbf{u}+r \mathbf{n}\cdot \mathbf{u}\right)\\
&=A_h(\mathbf{z},\mathbf{u})-b_h(r,\mathbf{u}).
\end{aligned}$$ Further, let $(\mathbf{z}_h,r_h)\in V_h\times Q_h$ be the solution of the finite element method approximation of $(\mathbf{z},r)$ which satisfies $$\label{l2_1}
A_h(\mathbf{z}_h,\mathbf{v}_h)-b_h(r_h,\mathbf{v}_h)=(\mathbf{u}-\mathbf{u}_h,\mathbf{v}_h), \ \forall \mathbf{v}_h\in V_h.$$ It is easy to obtain $$\label{l2_2}
\begin{aligned}
&A_h(\mathbf{z},\mathbf{v}_h)-b_h(r,\mathbf{v}_h)=(\mathbf{u}-\mathbf{u}_h,\mathbf{v}_h)\\
&\qquad+\sum^2_{i=1}\sum_{e\in {\mathcal{F}}_{h}^{nc,i}}\left(\int_e\mu_i \nabla \mathbf{u}\cdot \mathbf{n}_e[\mathbf{v}_h]+\int_e p[\mathbf{v}_h\cdot \mathbf{n}_e]\right), \ \forall \mathbf{v}_h\in V_h.
\end{aligned}$$ Thus, $$\begin{aligned}
||\mathbf{u}-\mathbf{u}_h||^2_{L^2(\Omega)}&=(\mathbf{u}-\mathbf{u}_h,\mathbf{u})-(\mathbf{u}-\mathbf{u}_h,\mathbf{u}_h)\\
&=A_h(\mathbf{z},\mathbf{u})-b_h(r,\mathbf{u})-A_h(\mathbf{z}_h,\mathbf{u}_h)+b_h(r_h,\mathbf{u}_h)\\
&=A_h(\mathbf{z}-\mathbf{z}_h,\mathbf{u})+A_h(\mathbf{z}_h,\mathbf{u}-\mathbf{u}_h)-b_h(r,\mathbf{u})+b_h(r_h,\mathbf{u}_h)\\
&=A_h(\mathbf{z}-\mathbf{z}_h,\mathbf{u}-I_h\mathbf{u})-b_h(r-r_h,\mathbf{u}-I_h\mathbf{u})\\
&\quad+A_h(\mathbf{z}-\mathbf{z}_h,I_h\mathbf{u})-b_h(r-r_h,I_h\mathbf{u})\\
&\quad+A_h(\mathbf{u}-\mathbf{u}_h,\mathbf{z}_h)-b_h(r_h,\mathbf{u}-\mathbf{u}_h):=I_1+I_2+I_3,
\end{aligned}$$ where $I_1$ stands for the first two terms, $I_2$ is the third and fourth terms and $I_3$ is the last two terms. Using the continuities of $A_h(\cdot,\cdot)$ and $b_h(\cdot,\cdot)$, we get $$\begin{aligned}
I_1&\lesssim |||\mathbf{z}-\mathbf{z}_h|||_V|||\mathbf{u}-I_h\mathbf{u}|||_V\\
&\lesssim |||(\mathbf{z}-\mathbf{z}_h,r-r_h)|||_V|||\mathbf{u}-I_h\mathbf{u}|||_V\\
&\lesssim \left(|||(\mathbf{z}-I_h\mathbf{z},r-R_hr)|||_V+|||(I_h\mathbf{z}-\mathbf{z}_h,R_hr-r_h)|||\right)|||\mathbf{u}-I_h\mathbf{u}|||_V.
\end{aligned}$$ From the proof for Theorem \[energy\_err\], we can see $$\label{pf_1}
|||(I_h\mathbf{u}-\mathbf{u}_h,R_hp-p_h)|||\lesssim h\sum^2_{i=1}\left(\mu^{1/2}_i|\mathbf{u}|_{2,\Omega_i}+\mu^{-1/2}_i|p|_{1,\Omega_i}\right).$$ Likewise, $|||(\mathbf{z}_h-I_h\mathbf{z},r_h-R_hr)|||$ have the similar estimate $$\label{pf_2}
|||(I_h\mathbf{z}-\mathbf{z}_h,R_hr-r_h)|||\lesssim h\sum^2_{i=1}\left(\mu^{1/2}_i|\mathbf{z}|_{2,\Omega_i}+\mu^{-1/2}_i|r|_{1,\Omega_i}\right).$$ Further, from , Theorem \[appr\_err\] and the regularity, $I_1$ can be estimated by $$\begin{aligned}
I_1&\lesssim h\sum^2_{i=1}\left(\mu^{1/2}_i|\mathbf{z}|_{2,\Omega_i}+\mu^{-1/2}_i|r|_{1,\Omega_i}\right)|||\mathbf{u}-I_h\mathbf{u}|||_V\\
&\lesssim \mu^{-1/2}_{min}h^2\left(\sum^2_{i=1}\mu^{1/2}_i|\mathbf{u}|_{2,\Omega_i}\right)||\mathbf{u}-\mathbf{u}_h||^2_{L^2(\Omega)}.
\end{aligned}$$ Since $$I_2 = A_h(\mathbf{z},I_h\mathbf{u})-b_h(r,I_h\mathbf{u})-A_h(\mathbf{z}_h,I_h\mathbf{u})+b_h(r_h,I_h\mathbf{u}),$$ we can get the following estimate by combining and $$\begin{aligned}
I_2&=\sum^2_{i=1}\sum_{e\in {\mathcal{F}}_{h}^{nc,i}}\left(\int_e\mu_i \nabla \mathbf{z}\cdot \mathbf{n}_e[I_h\mathbf{u}]+\int_e r[I_h\mathbf{u}\cdot \mathbf{n}_e]\right)\\
&=\sum^2_{i=1}\sum_{e\in {\mathcal{F}}_{h}^{nc,i}}\left(\int_e\mu_i \{\nabla (\mathbf{z}-I_h\mathbf{z})\cdot \mathbf{n}_e\}_k[I_h\mathbf{u}-\mathbf{u}]+\int_e \{r-R_hr\}_k[(I_h\mathbf{u}-\mathbf{u})\cdot \mathbf{n}_e]\right)\\
&\lesssim h^2\sum^2_{i=1}\left(\mu^{1/2}_i|\mathbf{z}|_{2,\Omega_i}+\mu^{-1/2}_i|r|_{1,\Omega_i}\right)\mu^{1/2}_i|\mathbf{u}|_{2,\Omega_i}\\
&\lesssim \mu^{-1/2}_{min}h^2\left(\sum^2_{i=1}\mu^{1/2}_i|\mathbf{u}|_{2,\Omega_i}\right)||\mathbf{u}-\mathbf{u}_h||_{L^2(\Omega)},
\end{aligned}$$ where we have used the trace inequality and standard Crouzeix-Raviart interpolation error estimate and $L^2$-project operator. From the definition of $B_h[(\cdot),(\cdot)]$ and , we have $$\label{I3_term2}
\begin{aligned}
&A_h(\mathbf{u}-\mathbf{u}_h,\mathbf{v}_h)+b_h(p-p_h,\mathbf{v}_h)-b_h(q_h,\mathbf{u}-\mathbf{u}_h)+J_p(p-p_h,q_h)\\
&=\sum^2_{i=1}\sum_{e\in {\mathcal{F}}_{h}^{nc,i}}\left(\int_e\mu_i \nabla \mathbf{u}\cdot \mathbf{n}_e[\mathbf{v}_h]+\int_e p[\mathbf{v}_h\cdot \mathbf{n}_e]\right), \ \forall (v_h,q_h)\in V_h\times Q_h,
\end{aligned}$$ and $$\label{I3_term1}
\begin{aligned}
I_3&=A_h(\mathbf{u}-\mathbf{u}_h,\mathbf{z}_h-I_h\mathbf{z})-b_h(r_h-R_hr,\mathbf{u}-\mathbf{u}_h)\\
&\quad+A_h(\mathbf{u}-\mathbf{u}_h,I_h\mathbf{z})-b_h(R_hr,\mathbf{u}-\mathbf{u}_h)\\
&\leq B_h[(\mathbf{u}-\mathbf{u}_h,p-p_h),(\mathbf{z}_h-I_h\mathbf{z},r_h-R_hr)]\\
&\quad+A_h(\mathbf{u}-\mathbf{u}_h,I_h\mathbf{z})-b_h(R_hr,\mathbf{u}-\mathbf{u}_h).
\end{aligned}$$ For the first term of the right-hand side in , using Lemma \[con\_bilinear\], we have $$\label{l2_term1}
\begin{aligned}
&B_h[(\mathbf{u}-\mathbf{u}_h,p-p_h),(\mathbf{z}_h-I_h\mathbf{z},r_h-R_hr)]\\
&\lesssim|||(\mathbf{u}-\mathbf{u}_h,p-p_h)|||_V|||(\mathbf{z}_h-I_h\mathbf{z},r_h-R_hr)|||\\
&\lesssim\left(|||(\mathbf{u}-I_h\mathbf{u},p-R_hp)|||_V+|||(I_h\mathbf{u}-\mathbf{u}_h,R_hp-p_h)|||\right)|||(\mathbf{z}_h-I_h\mathbf{z},r_h-R_hr)|||.
\end{aligned}$$ Together with Theorem \[appr\_err\], , and regularity, can be estimated by $$\label{I3_term3}
\begin{aligned}
&B_h[(\mathbf{u}-\mathbf{u}_h,p-p_h),(\mathbf{z}_h-I_h\mathbf{z},r_h-R_hr)]\\
&\lesssim \mu^{-1/2}_{min}h^2\sum^2_{i=1}\left(\mu^{1/2}_i|\mathbf{u}|_{2,\Omega_i}+\mu^{-1/2}_i|p|_{1,\Omega_i}\right)||\mathbf{u}-\mathbf{u}_h||_{0,\Omega}.
\end{aligned}$$ Now, we bound the remaining terms of the right-hand side in . From , we have $$\label{I3_term4}
\begin{aligned}
&A_h(\mathbf{u}-\mathbf{u}_h,I_h\mathbf{z})-b_h(R_hr,\mathbf{u}-\mathbf{u}_h)\\
&=\sum^2_{i=1}\sum_{e\in {\mathcal{F}}_{h}^{nc,i}}\left(\int_e\mu_i \nabla \mathbf{u}\cdot \mathbf{n}_e[I_h\mathbf{z}]+\int_e p[I_h\mathbf{z}\cdot \mathbf{n}_e]\right)\\
&\quad-b_h(p-p_h,I_h\mathbf{z})-J_p(p-p_h,R_hr)\\
&=\sum^2_{i=1}\sum_{e\in {\mathcal{F}}_{h}^{nc,i}}\left(\int_e\mu_i \{\nabla \mathbf{u}\cdot \mathbf{n}_e-\nabla I_h\mathbf{u}\cdot\mathbf{n}_e\}_k[I_h\mathbf{z}-\mathbf{z}]+\int_e \{p-I_hp\}_k[(I_h\mathbf{z}-\mathbf{z})\cdot \mathbf{n}_e]\right)\\
&\quad+b_h(p-p_h,\mathbf{z}-I_h\mathbf{z})+J_p(p-p_h,r-R_hr).
\end{aligned}$$ Using trace inequality and interpolation estimate, we have $$\label{I3_term41}
\begin{aligned}
&\sum^2_{i=1}\sum_{e\in {\mathcal{F}}_{h}^{nc,i}}\left(\int_e\mu_i \{\nabla \mathbf{u}\cdot \mathbf{n}_e-\nabla I_h\mathbf{u}\cdot\mathbf{n}_e\}_k[I_h\mathbf{z}-\mathbf{z}]+\int_e \{p-I_hp\}_k[(I_h\mathbf{z}-\mathbf{z})\cdot \mathbf{n}_e]\right)\\
&\lesssim h^2\sum^2_{i=1}\mu^{1/2}_i\left(\mu^{1/2}_i|\mathbf{u}|_{2,\Omega_i}+\mu^{-1/2}_i|p|_{1,\Omega_i}\right)|\mathbf{z}|_{2,\Omega_i}.
\end{aligned}$$ Applying Lemma \[continu\_bh\] and Cauchy-Schwarz inequality, we obtain $$\label{I3_term42}
\begin{aligned}
b_h(p-p_h,\mathbf{z}&-I_h\mathbf{z})+J_p(p-p_h,r-R_hr)\\
&\lesssim|||(\mathbf{u}-\mathbf{u}_h,p-p_h)|||_V|||(\mathbf{z}-I_h\mathbf{z},r-R_hr)|||_V\\
&\lesssim h\sum^2_{i=1}\left(\mu^{1/2}_i|\mathbf{z}|_{2,\Omega_i}+\mu^{-1/2}_i|r|_{1,\Omega_i}\right)|||(\mathbf{u}-\mathbf{u}_h,p-p_h)|||_V\\
&\lesssim h^2\mu_{min}^{-1/2}\sum^2_{i=1}\left(\mu^{1/2}_i|\mathbf{u}|_{2,\Omega_i}+\mu^{-1/2}_i|p|_{1,\Omega_i}\right)||\mathbf{u}-\mathbf{u}_h||_{0,\Omega}.
\end{aligned}$$ Combined with and , we can estimate by $$\label{I3_term5}
\begin{aligned}
A_h(\mathbf{u}&-\mathbf{u}_h,I_h\mathbf{z})-b_h(R_hr,\mathbf{u}-\mathbf{u}_h)\\
&\lesssim h^2\mu_{min}^{-1/2}\sum^2_{i=1}\left(\mu^{1/2}_i|\mathbf{u}|_{2,\Omega_i}+\mu^{-1/2}_i|p|_{1,\Omega_i}\right)||\mathbf{u}-\mathbf{u}_h||_{0,\Omega}.
\end{aligned}$$ Thus, the result is proved.
Numerical examples {#test}
==================
In the above section, we have shown that the proposed nonconforming extended finite element method is of optimal convergence order. In this section we investigate results for numerical experiments in two dimension space for the Stokes interface problem. We present the convergence rate of $H^1$, $L^2$ errors for velocity and $L^2$ error for pressure from two examples. Let $|\cdot|_{1,h,\Omega}$ be the piecewise $H^1$ semi norm. Then we denote the errors as follows: $$e^0_{h,\mathbf{u}}:=\frac{||\mathbf{u}-\mathbf{u}_h||_{0,\Omega}}{||\mathbf{u}||_{0,\Omega}}, e^0_{h,p}=\frac{||\mu^{-1/2}(p-p_h)||_{0,\Omega}}{||\mu^{-1/2}p||_{0,\Omega}},e^1_{h,\mathbf{u}}:=\frac{|\mu^{1/2} (\mathbf{u}-\mathbf{u}_h)|_{1,h,\Omega}}{|\mu^{1/2}\mathbf{u}|_{1,h,\Omega}}.$$
Example 1: a continuous problem
-------------------------------
We consider a continuous problem presented in [@Becker2009A]. The computational domain is $\Omega=[-1,1]\times[-1,1]$, the interface is a circle centered in $(0,0)$ with radius $0.5$ and $\mu=1$. The Dirichlet boundary conditions on $\partial\Omega$ are chosen such that the exact solution satisfies $\mathbf{u}=(20xy^3,5x^4-5y^5)$ and $p=60x^2y-20y^3$.
$h$ $e^1_{\mathbf{u}}$ rate $e^0_{\mathbf{u}}$ rate $e^0_{p}$ rate
-------- -------------------- -------- -------------------- -------- ----------- --------
$1/4$ 0.5040 0.2726 0.5597
$1/8$ 0.2816 0.8389 0.0920 1.5671 0.3237 0.7900
$1/16$ 0.1458 0.9497 0.0262 1.8121 0.1439 1.1696
$1/32$ 0.0737 0.9843 0.0066 1.9890 0.0615 1.2264
$1/64$ 0.0372 0.9864 0.0016 2.0444 0.0300 1.0356
: Errors for a continuous problem with $\mu=1$.[]{data-label="ex1"}
The errors and their convergence order for the velocity in $L^2$ and $H^1$ norms and the pressure in $L^2$ norm are give in Table \[ex1\]. We can see that the convergence orders of the errors are optimal. These results support our theoretical results.
Example 2: an interface problem
-------------------------------
We now consider a problem where the pressure is continuous and the velocity field is discontinuous on the interface due to different fluid viscosities. Let $\Omega=[-1,1]\times[-1,1]$, the interface is a circle centered in $(0,0)$ with the radius $0.5$. The interface separates domain $\Omega$ into two regions $\Omega_1=\{(x,y): x^2+y^2>0.25\}$ and $\Omega_2=\{(x,y): x^2+y^2<0.25\}$. The Dirichlet boundary conditions on $\partial\Omega$ are chosen such that the exact solution of the Stokes equations is given by $$\mathbf{u}=\begin{cases}
\mathbf{u}=(\frac{y(x^2+y^2-0.25)}{\mu_1},\frac{-x(x^2+y^2-0.25}{\mu_1})^{T} &\text{$(x,y) \in \Omega_1$},\\
\mathbf{u}=(\frac{y(x^2+y^2-0.25)}{\mu_2},\frac{-x(x^2+y^2-0.25}{\mu_2})^{T} &\text{$(x,y) \in \Omega_2$},
\end{cases}$$ and $$p=4(y^2-x^2),$$ then the right hand side $\mathbf{f}=(-8x-8y,8x+8y)^T$ and the jump conditions $[\mathbf{u}]=0$, $[p\mathbf{n}-\mu\nabla\mathbf{u}\cdot\mathbf{n}]=0$ on the interface. The viscosity is taken by $\mu_1=1000$ and $\mu_2=1$.
$h$ $e^1_{\mathbf{u}}$ rate $e^0_{\mathbf{u}}$ rate $e^0_{p}$ rate
-------- -------------------- -------- -------------------- -------- ----------- --------
$1/4$ 0.5115 0.2754 0.5438
$1/8$ 0.2850 0.8438 0.0913 1.5928 0.2976 0.8697
$1/16$ 0.1463 0.9620 0.0253 1.8515 0.1503 0.9855
$1/32$ 0.0738 0.9872 0.0063 2.0057 0.0641 1.2294
$1/64$ 0.0373 0.9844 0.0016 1.9773 0.0302 1.0858
: Errors for an interface problem with $\mu_1=1000$ and $\mu_2=1$.[]{data-label="ex2"}
Numerical results are shown in Table \[ex2\], and we can see that the convergence orders are optimal for the velocity in $H^1$ norm, $L^2$ norm and the pressure in $L^2$ norm, which conform the theoretical analysis.
$\mu_1$ $\mu_2$ $e^1_{\mathbf{u}}$ $e^0_{\mathbf{u}}$ $e^0_{p}$
--------- --------- -------------------- -------------------- -----------
$1E+01$ $1$ 0.0738 0.0063 0.0598
$1E+02$ $1$ 0.0737 0.0066 0.0612
$1E+03$ $1$ 0.0737 0.0066 0.0615
$1E+04$ $1$ 0.0737 0.0066 0.0615
$1E+05$ $1$ 0.0737 0.0066 0.0615
: Errors for an interface problem with $(\mu_1,\mu_2)=(10,1),(10^2,1),\cdots, (10^5,1)$ and fixed mesh $h=1/32$.[]{data-label="ex3"}
Finally, the numerical test is designed to confirm the influence of the jump of the different viscosities on the errors. We fix the mesh size $h=1/32$. Table \[ex3\] list the errors for velocity and pressure with $(\mu_1,\mu_2)=(10,1),(10^2,1),\cdots, (10^5,1)$. It indicates that the errors converge as $\frac{\mu_{max}}{\mu_{min}}\rightarrow \infty$, which means that they are all independent of the jump of the viscosity.
Conclusions {#conclude}
===========
We have introduced a finite element method which gives a way to accurately solve the Stokes interface problems with different viscosities. The method allows for discontinuities across the interface, namely, the interface can be intersected by the mesh. Harmonic weighted averages and arithmetic averages are used. Furthermore, the extra stabilization terms involves both velocity and pressure are added such that the inf-sup condition holds for nonconforming-$P_1/P_0$ elements. It is proven that the convergence orders of errors are optimal. Moreover, errors do not depend on the jump of the viscosities. Numerical results for both the continuous problem and interface problem in two dimensions have been given to support our theoretical results.
[^1]: Department of Mathematics, Nanjing University, Nanjing, 210093, People’s Republic of China. ([[email protected]]{})
[^2]: Department of Applied Mathematics, Nanjing Forestry University, Nanjing 210037, People’s Republic of China. ([[email protected]]{}). The work of this author was partially supported the Natural Science Foundation of the Jiangsu Higher Institutions of China (18KJB110015) and by the GXL2018024.
[^3]: Department of Mathematics, Nanjing University, Jiangsu, 210093, People’s Republic of China. ([[email protected]]{}). The work of this author was partially supported by the the NSF of China grant 10971096, and by the Project Funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In his famous Princeton Notes, Thurston introduced the so-called gluing equations defining the deformation variety. Later, Kashaev defined a non-commutative ring from H-triangulations of 3-manifolds and observed that for trefoil and figure-eight knot complements the abelianization of this ring is isomorphic to the ring of regular functions on the deformation variety, [@Kashaev-definition_delta_groupoid; @def_anneau; @Kashaev-Delta-groupoids_and_ideal_triangulations]. In this paper, we prove that this is true for any knot complement in a homology sphere. We also analyse some examples on other manifolds.'
author:
- 'Xavier Morvan[^1]'
bibliography:
- 'A\_non\_commutative\_generalisation.bib'
title: 'A non-commutative generalisation of Thurston’s gluing equations'
---
Introduction
============
In his famous Princeton notes [@Princeton_notes], Thurston introduced the following gluing equations. Starting from an ideal triangulation $X$ of a cusped manifold, $M\setminus K$ assign to each edge of each tetraedron a shape parameter $z\in\mathbb{C}$ so that:
- if $z_1, z_2, z_3$ are three shape parameters counter-clockwisely ordered around a vertex of a tetrahedron, then $z_1z_2z_3=-1$ and $z_2-z_1z_2=1$;
- for each edge $E$ of $X$, let $z_1,\dots, z_n$ be the shape parameters of all the edges which are identified to $E$, then $\prod_{i=1}^n z_i=1$.
This defines a set of polynomial equations, hence an affine variety in $(\mathbb{C}\setminus\{0,1\})^{6n}$, where $n$ is the number of tetrahedra in the triangulation. This affine variety is called *deformation variety*. We denote by [$R_I$]{} its associated ring of regular functions. The deformation variety may be empty for some triangulations. In that case, [$R_I$]{} is defined to be the ring with one element.
Since then, the gluing equations have been extensively studied, see for example [@luo2012solving; @solutions_for_gluing_eq; @NEUMANN1985307; @Petronio; @Segerman_Generalisation; @MR2866925].
Later, Kashaev introduced $\Delta$-groupoids [@Kashaev-definition_delta_groupoid] and $B'$-rings associated to ideal triangulations of manifolds and computed them for the trefoil and the figure eight knot complements [@Kashaev-Delta-groupoids_and_ideal_triangulations].
Then, in [@def_anneau], Kashaev introduced the ring [$\tilde{R}_H$]{}, the abelianisation of which is studied in this article. This ring is associated to a particular 1-vertex H-triangulation of a pair ($M^3,K$) (where $M^3$ is a connected, oriented, closed 3-manifold and $K$ is a knot) to which naturally corresponds an ideal triangulation of the complement of $K$ in $M^3$. This will be precised in section \[Preliminaries\]. In the case of the trefoil knot, [$\tilde{R}_H$]{} is abelian and isomorphic to [$R_I$]{}. In the case of the figure-eight knot, [$\tilde{R}_H$]{} is not abelian, but its abelianization is isomorphic to [$R_I$]{}.
We prove in the following that for any knot embedded in a homology 3-sphere there exists an isomorphism between the abelianization of the ring [$\tilde{R}_H$]{} defined from particular H-triangulations and the ring of regular functions on the deformation variety of a corresponding ideal triangulation. This isomorphism is explicitly constructed in section \[morphism\] and the theorem proved in section \[proof\]. In section \[examples\] we give an example of a knot embedded in the 3-sphere and examples of knots embedded in other manifolds.
The author is sincerely grateful to Rinat Kashaev for very valuable discussions and usuful comments on this paper.
Preliminaries {#Preliminaries}
=============
Triangulations
--------------
Let $M$ be a connected, oriented, closed 3-manifold. A triangulation $\mathbf{\Delta}$ of $M$ is defined to consist of a pairwise disjoint union of oriented euclidian tetrahedra $\mathbf{\Delta^3}=\bigsqcup_{i=1}^n\Delta^{3}_i$, together with a collection $\mathbf{\Phi}$ of orientation-reversing affine isomorphisms pairing the faces of the tetrahedra in $\mathbf{\Delta^3}$, so that $M$ is homeomorphic to the identified space $\mathbf{\Delta^3} / \mathbf{\Phi}$. In the following, for $k\in\mathbb{N}$, ${\ensuremath{\mathbf{\Delta^{k}}}}$ will denote the set of cells of dimension $k$ of the triangulation $\mathbf{\Delta}$.
An H-triangulation is a pair $(\mathbf{\Delta},K)$, where $\mathbf{\Delta}$ is a triangulation of a 3-manifold $M$ and $K$ is a hamiltonian subcomplex of the 1-skeleton of $\mathbf{\Delta}$. In the following, we will consider only *particular* H-triangulations: we will assume that $\mathbf{\Delta}$ has only one vertex, and the hamiltonian subcomplex is given by a single edge [$K$]{}, which is contained in a single face, [$F^K$]{} of $\mathbf{\Delta}$, obtained from a single tetrahedron [$T^K$]{} of $\mathbf{\Delta}$, glued to itself as a closed book by $\mathbf{\Phi}$ along its two faces. The edge [$K$]{} represents a knot embedded in $M$. Such a H-triangulation of $(M,{\ensuremath{K}})$ will be denoted [$\mathcal{T}_H$]{}. In order to stress the “H-nature” of [$\mathcal{T}_H$]{}, the set of $k$ dimensional cells of [$\mathcal{T}_H$]{} will be denoted [$\mathbf{\Delta^{k,H}}$]{} and the set of face pairings, [$\mathbf{\Phi^H}$]{}. We will denote ${\ensuremath{p_H}}:{\ensuremath{\mathbf{\Delta^{3,H}}}}\ \rightarrow\ {\ensuremath{\mathbf{\Delta^{3,H}}}}/ {\ensuremath{\mathbf{\Phi^H}}}$ the identification projection. [$K$]{}, [$F^K$]{}, [$T^K$]{} will respectively be called *distinguished edge*, *distinguished face* and *distinguished tetrahedron* (see figure \[example of distinguished tetrahedron\]).
(-1.22,-7.66) rectangle (8.24,1.71); (3.5,1.58)– (3.5,-3.61); (3.5,-1.17) – (3.3,-1.01); (3.5,-1.17) – (3.7,-1.01); (8,-6.21)– (3.5,-3.61); (5.48,-4.75) – (5.71,-4.66); (5.48,-4.75) – (5.51,-5.01); (5.75,-4.91) – (5.99,-4.81); (5.75,-4.91) – (5.79,-5.17); (-1,-6.21)– (3.5,-3.61); (1.52,-4.75) – (1.49,-5.01); (1.52,-4.75) – (1.29,-4.66); (1.25,-4.91) – (1.21,-5.17); (1.25,-4.91) – (1.01,-4.81); (-1,-6.21)– (8,-6.21); (3.66,-6.21) – (3.5,-6.41); (3.66,-6.21) – (3.5,-6.01); (3.34,-6.21) – (3.18,-6.41); (3.34,-6.21) – (3.18,-6.01); (3.97,-6.21) – (3.82,-6.41); (3.97,-6.21) – (3.82,-6.01); (3.05,-6.51) node\[anchor=north west\] [$A$]{}; (3.02,-4.73) node\[anchor=north west\] [$B$]{}; (4.08,-2.84) node\[anchor=north west\] [${\ensuremath{F^K}}$]{}; (1.28,-2.84) node\[anchor=north west\] [${\ensuremath{F^K}}$]{}; (-1,-6.21)– (3.5,1.56); (1.33,-2.19) – (1.43,-2.43); (1.33,-2.19) – (1.07,-2.22); (8,-6.21)– (3.5,1.56); (5.67,-2.19) – (5.93,-2.22); (5.67,-2.19) – (5.57,-2.43);
For any knot $K$ in $\mathbb{S}^3$, there exists a particular H-triangulation of $(\mathbb{S}^3,K)$.
There exist algorithms for one vertex H-triangulations of couples $(\mathbb{S}^3,K)$. See, for example, [@Kashaev2015], section 4. A slight modification of this algorithm allows one to get a particular H-triangulation: at the last stage of the decomposition of the 3-cell onto several tetrahedra, one has to extract a distinguished tetrahedron, and the previous steps of the algorithm always make this possible.
From [$\mathcal{T}_H$]{}, one gets a cell decomposition of another manifold by removing an open neighbourhood of the vertex. This is equivalent to truncating the tetrahedra, so that the new 3-cells are bounded by *triangular* and *hexagonal* faces. The triangular faces are bounded by *short edges*, and hexagonal faces are bounded by short and *long edges* which are remnant of the edges of [$\mathcal{T}_H$]{}. The obtained manifold is the complement in $M$ of an open ball. Such a cell decomposition will be denoted [$\overline{\mathcal{T}_H}$]{} and the set of $k$ dimensional cells in [$\overline{\mathcal{T}_H}$]{} will be denoted ${\ensuremath{\overline{\mathbf{\Delta^{k,H}}}}}$. Figure \[truncated distinguished tetrahedron\] is an example of a truncated tetrahedron. We will denote ${\ensuremath{\mathcal{H}}}\subset {\ensuremath{\overline{\mathbf{\Delta^{2,H}}}}} / {\ensuremath{\mathbf{\Phi^H}}}$ and $\mathcal{\tilde{H}}\subset {\ensuremath{\overline{\mathbf{\Delta^{2,H}}}}}$ the sets of non distinguished hexagonal faces, ${\ensuremath{\mathcal{T}}}$ the set of triangular faces disjoint from the distinguished edge, $\mathcal{L}\subset {\ensuremath{\overline{\mathbf{\Delta^{1,H}}}}} / {\ensuremath{\mathbf{\Phi^H}}}$ and $\mathcal{\tilde{L}}\subset {\ensuremath{\overline{\mathbf{\Delta^{1,H}}}}}$ the sets of non distinguished long edges, $\mathcal{S}\subset {\ensuremath{\overline{\mathbf{\Delta^{1,H}}}}} / {\ensuremath{\mathbf{\Phi^H}}}$, and $\mathcal{\tilde{S}}\subset {\ensuremath{\overline{\mathbf{\Delta^{1,H}}}}}$ the sets of short edges disjoint from the knot.
An ideal triangulation of a connected, oriented, cusped manifold $N$, is defined to consist of a pairwise disjoint union of oriented euclidian tetrahedra ${\ensuremath{\mathbf{\Delta^{3,I}}}}=\bigsqcup_{i=1}^n\Delta^{3,I}_i$, together with a set ${\ensuremath{\mathbf{\Phi^I}}}$ of orientation-reversing affine isomorphisms pairing the faces of the tetrahedra in ${\ensuremath{\mathbf{\Delta^{3,I}}}}$ so that $N={\ensuremath{(\mathbf{\Delta^{3,I}}\setminus\mathbf{\Delta^{0,I}})/\mathbf{\Phi^I}}}$. Such a triangulation will be denoted [$\mathcal{T}_I$]{}. We will denote ${\ensuremath{p_I}}: {\ensuremath{\mathbf{\Delta^{3,I}}}}\setminus \mathbf{\Delta^{0,I}} \rightarrow {\ensuremath{(\mathbf{\Delta^{3,I}}\setminus\mathbf{\Delta^{0,I}})/\mathbf{\Phi^I}}}$ the identification projection. In the following, we will consider a knot $K$ embedded in a 3-manifold $M$ and ideal triangulations of $N=M\setminus K$, such that ${\ensuremath{\mathbf{\Delta^{3,I}}}}/ {\ensuremath{\mathbf{\Phi^I}}}$ is a pseudomanifold having one singular point, its only vertex, corresponding to $K$. The tetrahedra of [$\mathcal{T}_I$]{} can be seen as hyperbolic ideal tetrahedra with vertices at infinity.
Let us remark that from a particular H-triangulation of a pair $(M,K)$, one can get an ideal triangulation of $M\setminus K$ by collapsing the distinguished edge to a point in such a way that the distinguished face [$F^K$]{} is collapsed to an edge while the two other faces of the distinguished tetrahedron (not bounded by distinguished edge) are identified with each other. For example, in figure \[example of distinguished tetrahedron\], the edges with simple and double arrows are identified with each other; face ${\ensuremath{F^K}}$ collapses and faces $B$ and $A$ are identified with each other. Then, each cell of [$\mathcal{T}_H$]{} different from the distinguished ones have a canonical corresponding cell in [$\mathcal{T}_I$]{}.
One can find more information on triangulations of 3-manifolds in [@BenPetr; @livre_Matveev].
From an ideal triangulation of a knot complement in a 3-manifold, one can construct a ring [$R_I$]{} as in the introduction.
Definition of [$R_H$]{} {#section def de Rh}
-----------------------
According to [@def_anneau], the ring [$\tilde{R}_H$]{} is defined from ${\ensuremath{\overline{\mathcal{T}_H}}}$ with oriented short edges by the following presentation. The set of generators is given by associating to each oriented short edge $e\in\mathcal{S}$, disjoint with the distinguished edge, a pair of generators $(u_e,v_e)$; and the set of relations:
- if $\bar{e}$ is the edge $e$ with opposite orientation, $u_{\bar{e}}=u_e^{-1}$ and $v_{\bar{e}}=-u_e^{-1}v_e$;
- *hexagonal face relation*: if $\check{e}$ is the unique oriented short edge such that it belongs to the same hexagonal face as $e$, and the terminal points of $e$ and $\check{e}$ form the boundary of a long edge, then $u_{\check{e}}=v_e$ and $v_{\check{e}}=u_e$;
- *triangular face relations*: if $e_1, e_2, e_3$ are cyclically oriented short edges constituting the boundary of a triangular face, then $u_{e_1}u_{e_2}u_{e_3}=1$ and $u_{e_1}u_{e_2}v_{e_3}+u_{e_1}v_{e_2}+v_{e_1}=0$.
One can easily check that there exists a representation of the knot group onto $GL(2,{\ensuremath{\tilde{R}_H}})$ by associating the matrix $\begin{pmatrix}
u_e&v_e\\0&1
\end{pmatrix}$ to each oriented short edge disjoint from the distinguished edge and the matrix $\begin{pmatrix}
0&1\\1&0
\end{pmatrix}$ to each (oriented) long edge different from the distinguished one.
Note that the ring [$\tilde{R}_H$]{} is not necessarily commutative. We will denote by [$R_H$]{} the abelianization of [$\tilde{R}_H$]{}.
\[relations for the distinguished tetrahedra\] Let the truncated distinguished tetrahedron be labelled as in figure \[truncated distinguished tetrahedron\], then $u_l=u_m$, $u_p=v_m^{-1}v_l$, $v_p=0$.
Let us write the two triangular face relations where all generators with indices $i_1, i_2, j_1, j_2$ are expressed in terms of $(u_l,v_l)$ and $(u_m,v_m)$ through the use of the hexagonal faces relations.
For the left hand triangle: $$\begin{aligned}
1&=v_mu_pv_l^{-1} \label{eq1} \\
0&=-v_mu_pv_l^{-1}u_l+v_mv_p+u_m \label{eq2}\end{aligned}$$ and the right hand triangle: $$\begin{aligned}
1&=u_m^{-1}v_mu_pv_l^{-1}u_l \label{eq3} \\
0&=-u_m^{-1}v_mu_p(v_l^{-1}u_l)u_l^{-1}+u_m^{-1}v_mv_p+u_m^{-1} \label{eq4}\end{aligned}$$ Equations (\[eq1\]) and (\[eq3\]) imply that $u_m=u_l$. Then (\[eq2\]) implies that $v_p=0$ because $v_m$ is invertible.
(-1.24,-6.4) rectangle (8.2,1.61); (3.5,0.06)– (3,0.93); (0,-6)– (0.5,-5.13); (0.3,-5.49) – (0.35,-5.63); (0.3,-5.49) – (0.15,-5.51); (0.5,-5.13)– (-0.5,-5.13); (-0.09,-5.13) – (0,-5.02); (-0.09,-5.13) – (0,-5.25); (7,-6)– (6.5,-5.13); (6.7,-5.49) – (6.85,-5.51); (6.7,-5.49) – (6.65,-5.63); (6.5,-5.13)– (7.5,-5.13); (7.09,-5.13) – (7,-5.25); (7.09,-5.13) – (7,-5.02); (3.5,-2.82)– (4,-3.69); (3,0.93)– (3.5,0.06); (3.5,-2.82)– (3,-3.69); (0.5,-5.13)– (-0.5,-5.13); (-0.5,-5.13)– (3,0.93); (1.3,-2.02) – (1.35,-2.16); (1.3,-2.02) – (1.15,-2.04); (3.5,0.06)– (4,0.93); (4,0.93)– (7.5,-5.13); (7.5,-5.13)– (6.5,-5.13); (6.5,-5.13)– (4,-3.69); (5.17,-4.37) – (5.31,-4.31); (5.17,-4.37) – (5.19,-4.52); (7,-6)– (6.5,-5.13); (4,-3.69)– (3,-3.69); (3.41,-3.69) – (3.5,-3.57); (3.41,-3.69) – (3.5,-3.81); (0.5,-5.13)– (0,-6); (7,-6)– (7.5,-5.13); (7.3,-5.49) – (7.35,-5.63); (7.3,-5.49) – (7.15,-5.51); (7.5,-5.13)– (4,0.93); (5.7,-2.02) – (5.85,-2.04); (5.7,-2.02) – (5.65,-2.16); (4,0.93)– (3,0.93); (3.41,0.93) – (3.5,1.05); (3.41,0.93) – (3.5,0.81); (3,0.93)– (-0.5,-5.13); (0.5,-5.13)– (3,-3.69); (1.83,-4.37) – (1.81,-4.52); (1.83,-4.37) – (1.69,-4.31); (7,-6)– (0,-6); (3.41,-6) – (3.5,-5.88); (3.41,-6) – (3.5,-6.12); (3.5,-2.82)– (3.49,0.08); (3.5,-1.28) – (3.61,-1.37); (3.5,-1.28) – (3.38,-1.37); (0.02,-5.97)– (-0.5,-5.13); (-0.29,-5.48) – (-0.14,-5.49); (-0.29,-5.48) – (-0.34,-5.62);
(0.84,-5.71) node [$j_2$]{}; (0.14,-4.76) node [$p$]{}; (6.29,-5.65) node [$i_2$]{}; (6.83,-4.8) node [$p$]{}; (1.78,-2.23) node [$a_1$]{}; (5.26,-4.8) node [$a_2$]{}; (3.5,-4.06) node [$m$]{}; (7.81,-5.77) node [$i_1$]{}; (6.25,-2.06) node [$a_1$]{}; (3.51,1.35) node [$l$]{}; (2.12,-4.73) node [$a_2$]{}; (3.5,-5.62) node [$c$]{}; (-0.61,-5.71) node [$j_1$]{};
A ring homomorphism from [$R_I$]{} to [$R_H$]{} {#morphism}
===============================================
We define a ring homomorphism $f:{\ensuremath{R_I}}\rightarrow{\ensuremath{R_H}}$ as follows. Let $E$ be an edge of a tetrahedron of the ideal triangulation with shape parameter $z$. Let us still call $E$ the corresponding long edge in $\mathcal{\tilde{L}}$. We choose a boundary point of $E$ in [$\overline{\mathbf{\Delta^{0,H}}}$]{}, and let $e_i$ and $e_j$ be the short edges in $\mathcal{\tilde{S}}$ sharing this point at their origins and such that $E,e_i,e_j$ are clockwisely ordered. Let $(u_i,v_i)$ and $(u_j,v_j)$ be the couples of generators in [$R_H$]{} assigned to $e_i$ and $e_j$. We define, $f(z)=v_iv_j^{-1}$ (see figure \[definition of f\]).
\[autre def de f\] Keeping the notation of figure \[definition of f\], $f$ can be equivalently defined as $f(z)=u_pu_q^{-1}$.
(-0.87,-6.75) rectangle (7.97,1.54); (3.5,0.06)– (3,0.93); (3.2,0.57) – (3.35,0.55); (3.2,0.57) – (3.15,0.44); (-0.5,-5.13)– (0,-6); (0,-6)– (0.5,-5.13); (0.5,-5.13)– (-0.5,-5.13); (-0.09,-5.13) – (0,-5.02); (-0.09,-5.13) – (0,-5.25); (7.5,-5.13)– (7,-6); (7,-6)– (6.5,-5.13); (6.5,-5.13)– (7.5,-5.13); (7.09,-5.13) – (7,-5.25); (7.09,-5.13) – (7,-5.02); (3.5,-2.82)– (4,-3.69); (3.8,-3.34) – (3.65,-3.32); (3.8,-3.34) – (3.85,-3.2); (3,0.93)– (3.5,0.06); (3.5,0.06)– (3.5,-2.82); (3.5,-2.82)– (3,-3.69); (3.2,-3.34) – (3.15,-3.2); (3.2,-3.34) – (3.35,-3.32); (3,-3.69)– (0.5,-5.13); (0.5,-5.13)– (-0.5,-5.13); (-0.5,-5.13)– (3,0.93); (3.5,0.06)– (4,0.93); (3.8,0.57) – (3.85,0.44); (3.8,0.57) – (3.65,0.55); (4,0.93)– (7.5,-5.13); (7.5,-5.13)– (6.5,-5.13); (6.5,-5.13)– (4,-3.69); (3.5,-2.82)– (3.5,0.06); (7,-6)– (6.5,-5.13); (6.5,-5.13)– (4,-3.69); (4,-3.69)– (3,-3.69); (3.41,-3.69) – (3.5,-3.57); (3.41,-3.69) – (3.5,-3.81); (3,-3.69)– (0.5,-5.13); (0.5,-5.13)– (0,-6); (0,-6)– (7,-6); (7,-6)– (7.5,-5.13); (7.5,-5.13)– (4,0.93); (4,0.93)– (3,0.93); (3.41,0.93) – (3.5,1.05); (3.41,0.93) – (3.5,0.81); (3,0.93)– (-0.5,-5.13); (-0.5,-5.13)– (0,-6); (0,-6)– (7,-6);
(2.95,0.21) node [i]{}; (0.14,-4.76) node [p]{}; (6.83,-4.8) node [q]{}; (4.28,-3.09) node [n]{}; (3.13,-1.23) node [z]{}; (2.88,-3.1) node [k]{}; (4.04,0.21) node [j]{}; (3.5,-4.06) node [m]{}; (3.51,1.35) node [l]{};
Let us check that $f$ is a well defined ring homomorphism.
First, the definition of $f$ does not depend on the choice of the boundary point of E: if $e_k=\overline{\check{\bar{e_i}}}$ and $e_n=\overline{\check{\bar{e_j}}}$, then $v_i=v_k^{-1}$ and $v_j=v_n^{-1}$ so that $v_iv_j^{-1}=v_nv_k^{-1}$.
Let $E_1, E_2, E_3$ be three long edges counter-clockwisely ordered around a vertex of a tetrahedron in [$\mathbf{\Delta^{3,I}}$]{}, with shape parameters $z_1,z_2,z_3$, and let $e_i, e_j, e_k$ be three short edges bounding a triangular face obtained after truncation along the corresponding vertex and such that $E_1, e_i, e_j$ and $E_2, \overline{e_j}, e_k$ are clockwisely ordered. Then, the triangular face relations are $u_ju_ku_i^{-1}=1$ and $-v_i+u_j v_k+v_j=0$.
So, $$f(z_1)f(z_2)f(z_3)=(v_iv_j^{-1})(-u_j^{-1}v_jv_k^{-1})(u_k^{-1}v_kv_i^{-1}u_i)=-u_j^{-1}u_k^{-1}u_i=-1$$
and $$f(z_2)-f(z_1)f(z_2)=-u_j^{-1}v_jv_k^{-1}+v_iv_j^{-1}u_j^{-1}v_jv_k^{-1}=u_j^{-1}(-v_j+v_i)v_k^{-1}=1$$
Let us now check that the gluing equations are respected. If the ideal triangulation of the knot complement contains an edge with only one preimage by [$p_I$]{}, then both [$R_I$]{} and [$R_H$]{} are the ring with one element and all conditions are trivially satisfied.
Before considering the general case, let us first remark the following. Let two triangles in ${\ensuremath{\mathcal{T}}}$ be glued together along an oriented short edge $e_j$ with initial point corresponding to long edges with shape parameters $z_1$ and $z_2$. Let $e_i,e_j,e_k$ be short edges clockwisely ordered as in figure \[glued triangles\]. Then, $f(z_1)f(z_2)=(v_iv_j^{-1})(v_jv_k^{-1})=v_iv_k^{-1}$.
Let us now consider separately few different cases. First, if an edge $E$ in the ideal triangulation has no corresponding one in a distinguished tetrahedron of the H-triangulation, then the product of the $f(z_i)$ around $E$ telescopically reduces to 1.
Second, $E$ has a corresponding edge opposite to the distinguished edge in the distinguished tetrahedron. Then, after telescopication, keeping notations of lemma \[relations for the distinguished tetrahedra\], the product of $f(z_i)$’s around $E$ reduces to $v_{\check{m}}v_{\check{l}}^{-1}$ which is equal to $1$ due to lemma \[relations for the distinguished tetrahedra\].
Third case, $E$ has a corresponding edge in the distinguished tetrahedron in the boundary of the distinguished face. Again, and after telescopication, the product of $f(z_i)$’s around $E$ reduces to a product of parameters $v_i$ which is equal to 1. In order to see this, it is easier to use the equivalent definition of $f$ : $f(z)=u_pu_q^{-1}$ as in remark \[autre def de f\]. Then, by using the notation of figure 2, the edge $a_2$ will contribute $u_{i_2}u_{j_2}^{-1}$ and the edge $a_1$ will contribute $u_{i_1}^{-1}u_{j_1}$ (both $a_k$ appear twice in the distinguished tetrahedron). The product of those two contributions is 1 due to hexagonal face relations of figure \[truncated distinguished tetrahedron\].
(-2.12,-1.42) rectangle (1.8,2.86); (0.86,0)– (-1.24,1.46); (-0.26,0.78) – (-0.12,0.83); (-0.26,0.78) – (-0.26,0.63); (-1.24,1.46)– (-1.45,-1.09); (-1.45,-1.09)– (0.86,0); (0.86,0)– (-1.45,-1.09); (-0.39,-0.59) – (-0.35,-0.42); (-0.39,-0.59) – (-0.24,-0.67); (0.86,0)– (1.07,2.55); (0.98,1.38) – (1.1,1.26); (0.98,1.38) – (0.83,1.29); (1.07,2.55)– (-1.24,1.46);
(0.86,0) circle (1.5pt); (0.42,0.1) node [$z_1$]{}; (-0.12,1.12) node [j]{}; (-0.44,-0.2) node [i]{}; (0.86,0) circle (2.5pt); (0.68,0.42) node [$z_2$]{}; (1.3,1.42) node [k]{};
\[theorem\] If $H_1(M,\mathbb{Z})=0$, then $f$ is a ring isomorphism.
According to Poincaré duality and Universal coefficients theorem, if $H_1(M,\mathbb{Z})=0$, then $H_2(M,\mathbb{Z})=0$.
Proof of theorem \[theorem\] {#proof}
============================
First of all, let us clear up the particular cases where the ideal triangulation of a knot complement contains an edge with only one preimage by ${\ensuremath{p_I}}$. Such an edge must be in a tetrahedron closed as a book around it. Then, the shape parameter of that edge must be equal to $1$, which means that the deformation variety is empty. As a consequence, the ring ${\ensuremath{R_I}}$ is the ring with one element. Besides, according to lemma \[relations for the distinguished tetrahedra\], in that case, there is an invertible element in [$R_H$]{} equal to 0. So [$R_H$]{} is also the ring with one element and the isomorphism between [$R_I$]{} and [$R_H$]{} is trivial.
For the general case, let us first remark that the image of an element of [$R_I$]{} by $f$ is a priori a product of an even number of elements in [$R_H$]{}. Then, in order to prove the surjectivity of $f$, the main difficulty is to write an equality in [$R_H$]{} with one generating element on the left hand side and an even number of generating elements on the right hand side. For that, we will use the triviality of $H_1(M,\mathbb{Z})$.
From now on, we will say that an oriented short edge $e\in\mathcal{\tilde{S}}$ is *parallel* (resp. *antiparallel*) to an oriented long edge $E$ of $\mathcal{\tilde{L}}$ if there exists $E'$ of ${\ensuremath{p_H}}^{-1}({\ensuremath{p_H}}(E))$ such that $e$ and $E'$ are disjointly contained in the boundary of one and the same hexagonal face $F$, and if the orientations of $\bar{e}$ (resp. $e$) and $E'$ are induced from the orientation of $F$. We will also say that two oriented short edges $e$ and $e'$ are *parallel* (resp. *antiparallel*) if there exists a long edge $E$ of $\mathcal{\tilde{L}}$ such that $e$ and $e'$ are both parallel or antiparallel (resp. one is parallel and another one is antiparallel) to $E$. We will also say that an edge $e\in\mathcal{S}$ is parallel (resp. antiparallel) to an edge $E\in\mathcal{L}$ if there exists a parallel (resp. antiparallel) preimage of $e$ and $E$ by ${\ensuremath{p_H}}$. For instance, in figure \[truncated distinguished tetrahedron\], $m$, $c$ and $l$ are parallel.
\[Lemme\] If $e$ and $e'$ are parallel short edges in $\mathcal{S}$, then there exists $m\in{\ensuremath{R_I}}$ such that $f(m)=u_eu_{e'}^{-1}$.
We will prove that there exists $m\in{\ensuremath{R_I}}$ such that $f(m)=v_{\check{e}}v_{\check{e'}}^{-1}$(see subsection \[section def de Rh\] for the definition of $\check{e}$ and $\check{e'}$), which is equivalent to the lemma \[Lemme\].
Let $E\in\mathcal{L}$ be parallel to $e$ and $e'$. Then, among the triangles containing the edge $E$ in [$\overline{\mathcal{T}_H}$]{} there are some which also contain the edge $\check{e}$ or $\check{e'}$. On the disk composed by the triangular faces containing a given end point of $E$, choose a path from $\check{e}$ to $\check{e'}$. Let $e_1,\ldots,e_n$ be short edges met by this path and oriented so that they all have the same starting point (here $\check{e}=e_1$ and $\check{e'}=e_n$). Let $\tilde{E}_1,\ldots,\tilde{E}_{n-1}\in{\ensuremath{p_H}}^{-1}(E)$ be long edges intersecting $e_1,\ldots,e_n$ in their starting point, and let $z_1,\ldots,z_{n-1}$ be the corresponding parameters in the ideal triangulation (see figure \[image pour lemme\]). Let $m=\prod_{i=1}^{n-1} z_i\in{\ensuremath{R_I}}$. According to the gluing equations in ${\ensuremath{R_I}}$, $m$ does not depend on the choice of the path from $\check{e}$ to $\check{e'}$. Besides, $\prod_{i=1}^{n-1} f(z_i)=\prod_{i=1}^{n-1} v_iv_{i+1}^{-1}=v_1v_n^{-1}=v_{\check{e}}v_{\check{e'}}^{-1}$ (here, $(u_i,v_i)$ is the couple of generators in ${\ensuremath{R_H}}$ associated to the short edge $e_i$).
It will be usefull in the following to denote $m_{e\rightarrow e'}$ for such a word.
(-1.7,-4.1) rectangle (8.52,5.42); (0,0) – (178.01:1.8) arc (178.01:214.01:1.8) – cycle; (0,0) – (142.01:1.8) arc (142.01:178.01:1.8) – cycle; (0,0) – (-1.99:1.8) arc (-1.99:142.01:1.8) – cycle; (0,0) – (-37.99:1.8) arc (-37.99:-1.99:1.8) – cycle; (3.31,0.63)– (-0.26,-1.78); (1.44,-0.63) – (1.45,-0.46); (1.44,-0.63) – (1.6,-0.69); (3.31,0.63)– (-0.99,0.78); (1.05,0.71) – (1.16,0.84); (1.05,0.71) – (1.15,0.57); (3.31,0.63)– (-0.08,3.29); (1.53,2.02) – (1.7,2.06); (1.53,2.02) – (1.53,1.85); (3.31,0.63)– (6.71,-2.02); (5.1,-0.76) – (4.93,-0.8); (5.1,-0.76) – (5.1,-0.59); (3.31,0.63)– (7.62,0.48); (5.57,0.55) – (5.46,0.42); (5.57,0.55) – (5.47,0.69); plot\[domain=-0.03:2.48,variable=\]([1\*4.31\*cos(r)+0\*4.31\*sin(r)]{},[0\*4.31\*cos(r)+1\*4.31\*sin(r)]{}); (-1.99:1.8) arc (-1.99:142.01:1.8); plot\[domain=3.74:5.62,variable=\]([1\*4.31\*cos(r)+0\*4.31\*sin(r)]{},[0\*4.31\*cos(r)+1\*4.31\*sin(r)]{}); (-0.26,-1.78)– (-0.99,0.78); (-0.99,0.78)– (-0.08,3.29); (7.62,0.48)– (6.71,-2.02);
(1.48,-1.26) node [$e_1$]{}; (0.9,0.28) node [$e_2$]{}; (1.28,1.74) node [$e_3$]{}; (5.44,-1.34) node [$e_n$]{}; (6.34,0.16) node [$e_{n-1}$]{}; (2.58,0.3) node [$z_1$]{}; (2.58,0.86) node [$z_2$]{}; (4.28,0.28) node [$z_{n-1}$]{};
\[corollary\] Let $e\in\mathcal{S}$ and $m\in{\ensuremath{R_I}}$ be such that $f(m)=u_e$. Then, for any short edge $e'$ parallel to $e$, there exists $m'\in{\ensuremath{R_I}}$ such that $f(m')=u_{e'}$.
Being a cell complex decomposition of $M$, [$\mathcal{T}_H$]{} induces a presentation of the first homology group with 1-cells as generators, and 2-cells as relations. Let us denote by [$K$]{} the distinguished edge corresponding to the knot in $M$. Let us also denote by $E_i^H$ for $1\leqslant i \leqslant n$ all the other edges in ${\ensuremath{\mathbf{\Delta^{1,H}}}} / {\ensuremath{\mathbf{\Phi^H}}}$ and $E_i^{j,H}$ the edges of ${\ensuremath{\mathbf{\Delta^{1,H}}}}$ such that ${\ensuremath{p_H}}(E_i^{j,H})=E_i^H$. Let us denote by ${\ensuremath{\Delta_{l}^H}}(M)$ the group of oriented $l$-chains (the free abelian group with basis the oriented open $l$-simplices), with boundary maps: ${\ensuremath{\partial_{l}^H}}:{\ensuremath{\Delta_{l}^H}}(M)\rightarrow{\ensuremath{\Delta_{l-1}^H}}(M)$. Let us denote by [$F^K$]{} the distinguished face of ${\ensuremath{\mathbf{\Delta^{2,H}}}} / {\ensuremath{\mathbf{\Phi^H}}}$, and $F_i^H$ for $1\leqslant i \leqslant m$ the other faces. Then ${\ensuremath{\partial_{2}^H}}({\ensuremath{F^K}})=\epsilon_{{\ensuremath{K}}}{\ensuremath{K}}+\epsilon_j E_j^H+\epsilon_{j'} E_{j'}^H$ ($\epsilon_i=\pm 1$). Besides, according to the particular H-triangulations considered here, ${\ensuremath{F^K}}$ is the only face of [$\mathbf{\Delta^{2,H}}$]{} / [$\mathbf{\Phi^H}$]{} whose boundary contains ${\ensuremath{K}}$. Thus, one can consider another presentation of the first homology group with generators the edges in ${\ensuremath{\mathbf{\Delta^{1,H}}}} / {\ensuremath{\mathbf{\Phi^H}}}$ without the distinguished one, and relations corresponding to non-distinguished faces. On what follows, we assume that $K$ and [$F^K$]{} are removed from the corresponding sets of cells.
Let us assume from now on that $H_1(M,\mathbb{Z})=0$ and define some useful notations. Let [$A$]{} be an edge in ${\ensuremath{\mathbf{\Delta^{1,H}}}} / {\ensuremath{\mathbf{\Phi^H}}}$. Then, there exists a map ${\ensuremath{\eta_{{\ensuremath{A}}}}}: {\ensuremath{\mathbf{\Delta^{2,H}}}} / {\ensuremath{\mathbf{\Phi^H}}}\rightarrow \mathbb{Z}$ such that $${\ensuremath{A}}=\sum_{F\in{\ensuremath{\mathbf{\Delta^{2,H}}}}/ {\ensuremath{\mathbf{\Phi^H}}}}{\ensuremath{\eta_{{\ensuremath{A}}}}}(F){\ensuremath{\partial_{2}^H}}(F)\in {\ensuremath{\Delta_{1}^H}}(M).$$ It is a tautological equality in a free abelian group.
For any $F\in\mathcal{\tilde{H}}$ let $$\mathcal{\tilde{E}}_F=\{E\in\mathcal{\tilde{L}} \mid E \text{ is on the boundary of }F\}$$ and let us define a set $$X=\{(F,E) \mid F\in\mathcal{\tilde{H}}, E \in\mathcal{\tilde{E}}_F \}$$ and a map $r:X\rightarrow\mathcal{\tilde{S}}$ in the following way. $E$ is on the boundary of two faces in $\mathcal{\tilde{H}}$, $F$ and $F'$, so $r(F,E)$ is the short edge on the boundary of $F'$ disjoint from $E$ and parallel to it (see figure \[def de r\]).
(-0.98,-6.56) rectangle (8.11,1.41); (3.5,0.06)– (3,0.93); (0,-6)– (0.5,-5.13); (0.5,-5.13)– (-0.5,-5.13); (-0.07,-5.13) – (0,-5.04); (-0.07,-5.13) – (0,-5.23); (3.5,-2.82)– (3,-3.69); (3.5,0.06)– (4,0.93); (4,0.93)– (7.5,-5.13); (7.5,-5.13)– (6.5,-5.13); (6.5,-5.13)– (4,-3.69); (4,-3.69)– (3.5,-2.82); (7,-6)– (6.5,-5.13); (6.5,-5.13)– (4,-3.69); (4,-3.69)– (3,-3.69); (3,-3.69)– (0.5,-5.13); (0.5,-5.13)– (0,-6); (0,-6)– (7,-6); (7,-6)– (7.5,-5.13); (7.5,-5.13)– (4,0.93); (4,0.93)– (3,0.93); (3,0.93)– (-0.5,-5.13); (-0.5,-5.13)– (0,-6); (0,-6)– (7,-6); (3.5,-2.82)– (3.5,0.06); (3.5,-1.31) – (3.59,-1.38); (3.5,-1.31) – (3.41,-1.38);
(0.12,-4.7) node [$e$]{}; (4.7,-2.53) node [$F$]{}; (3,-1.48) node [$E$]{};
We fix a section $\sigma: {\ensuremath{\mathcal{H}}}\rightarrow \mathcal{\tilde{H}}$ with the image disjoint with the distinguished tetraedron. Denote $\tilde{{\ensuremath{\mathcal{F}}}}=\sigma({\ensuremath{\mathcal{H}}})$ and reinterpret ${\ensuremath{\eta_{{\ensuremath{A}}}}}$ as a map from $\tilde{{\ensuremath{\mathcal{F}}}}$ to $\mathbb{Z}$.
We define
$$U_{{\ensuremath{A}}}=\prod_{F\in\tilde{{\ensuremath{\mathcal{F}}}}}\prod_{E\in\mathcal{\tilde{E}}_F}(u_{r(F,E)})^{{\ensuremath{\eta_{{\ensuremath{A}}}}}(F)}$$ Let us also define a map $s:\mathcal{L}\rightarrow\mathcal{S}$ which associates arbitrarily, to each long edge, a parallel short edge, and $$U_{{\ensuremath{A}}}^s=\prod_{F\in\tilde{{\ensuremath{\mathcal{F}}}}}\prod_{E\in\mathcal{\tilde{E}}_F}(u_{s\circ{\ensuremath{p_H}}(E)})^{{\ensuremath{\eta_{{\ensuremath{A}}}}}(F)}$$ and remark that $U_{{\ensuremath{A}}}^s=u_{s(A)}$; it is a tautological equality in the free abelian group generated by the symbols $u_i$ associated to short edges. Then, for each ${\ensuremath{A}}\in{\ensuremath{\mathbf{\Delta^{1,H}}}} / {\ensuremath{\mathbf{\Phi^H}}}$, one can write the equality $$u_{s({\ensuremath{A}})}=\dfrac{U_{{\ensuremath{A}}}^s}{U_A}=\prod_{F\in\tilde{{\ensuremath{\mathcal{F}}}}}\prod_{E\in\mathcal{\tilde{E}}_F}\left(\frac{u_{s\circ{\ensuremath{p_H}}(E)}}{u_{r(F,E)}}\right)^{{\ensuremath{\eta_{{\ensuremath{A}}}}}(F)}.$$ By definitions of $r$ and $s$ and thanks to lemma \[Lemme\], there exist words $m_{F,E}\in{\ensuremath{R_I}}$ such that $\frac{u_{s\circ{\ensuremath{p_H}}(E)}}{u_{r(F,E)}}=f(m_{F,E})$ and thus there exists a word $m_{s({\ensuremath{A}})}$ in [$R_I$]{} such that $f(m_{s({\ensuremath{A}})})=u_{s({\ensuremath{A}})}$.
According to corollary \[corollary\], for any short edge $e_i$ parallel to $s(A)$ and with the generators of [$R_H$]{} $(u_i,v_i)$, there exists a word $m_i\in{\ensuremath{R_H}}$ such that $f(m_i)=u_i$. Besides, for any short edge $e_j$, there exists a short edge $e_i$ such that $v_j=u_i$ or $v_j=u_i^{-1}$. Thus $f$ is surjective, and this almost allows one to define a ring homomorphism $g:{\ensuremath{R_H}}\rightarrow {\ensuremath{R_I}}$: keeping the previous notation, $g(u_i)=g(v_j)=m_i$.
One still has \[fin enum\] things to check:
1. The images of the $u_i$ do not depend on the choice of ${\ensuremath{\eta_{{\ensuremath{A}}}}}$:
Let ${\ensuremath{\eta_{{\ensuremath{A}}}^{(1)}}}$ and ${\ensuremath{\eta_{{\ensuremath{A}}}^{(2)}}}$ be two maps such that $${\ensuremath{A}}=\sum_F{\ensuremath{\eta_{{\ensuremath{A}}}^{(1)}}}(F)F=\sum_F{\ensuremath{\eta_{{\ensuremath{A}}}^{(2)}}}(F)F\in {\ensuremath{\Delta_{1}^H}}(M).$$ As $H_2(M,\mathbb{Z})=0$, $\phi=\sum_F{\ensuremath{\eta_{{\ensuremath{A}}}^{(1)}}}(F)F-\sum_F{\ensuremath{\eta_{{\ensuremath{A}}}^{(2)}}}(F)F={\ensuremath{\partial_{3}^H}}(\tau)$, where $\tau$ is a linear combination of tetrahedra in [$\mathcal{T}_H$]{}. If $\tau$ contains the distinguished tetrahedron, then there is an ambiguity in the word $m_{s({\ensuremath{A}})}$ given by a power of the product of all shape parameters associated to the edge opposite to the distinguished edge in the distinguished tetraedron, which is $1$ thanks to the gluing equations. Let us now assume that $\tau$ contains only tetrahedra different from the distinguished one. Let us denote by $m_{s({\ensuremath{A}})}^{(1)}$ and $m_{s({\ensuremath{A}})}^{(2)}$ the corresponding words for ${\ensuremath{\eta_{{\ensuremath{A}}}}}{(1)}$ and ${\ensuremath{\eta_{{\ensuremath{A}}}}}{(2)}$. Then $\frac{m_{s({\ensuremath{A}})}^{(1)}}{m_{s({\ensuremath{A}})}^{(2)}}$ is the product of all shape parameters of all tetrahedra contained in $\tau$, so it is equal to $1$.
2. The definition of $U_{{\ensuremath{A}}}$ does not depend on the choice of $\tilde{{\ensuremath{\mathcal{F}}}}$ *i.e.* the section $\sigma:{\ensuremath{\mathcal{H}}}\rightarrow\tilde{{\ensuremath{\mathcal{H}}}}$:
Let $F\in{\ensuremath{\mathcal{H}}}$ and $\{F^1, F^2\}={\ensuremath{p_H}}^{-1}(F)$. If one of the $F^i$’s is on the boundary of the distinguished tetrahedron, then there is no problem thanks to the definition of $\sigma$. Else, let $E^i_k, 1\leqslant k \leqslant 3, 1\leqslant i \leqslant 2$ be the long edges in the boundary of $F^i$, with the corresponding shape parameters $z^i_k$. Let $e^i_k=r(F^i, E^i_k)$. It is then easy to see that for each $k$, $f(z^1_kz^2_k)=u^1_k(u^2_k)^{-1}$, then the ambiguity of $m_{s(A)}$ is given by a power of $\prod_{k=1}^3z^1_kz^2_k=1\in{\ensuremath{R_I}}$.
3. The definition of $m_i$ does not depend on the choice of $s$:
Let $A\in\mathcal{L}$, $e_1, e_2\in\mathcal{S}$ with parameters $(u_1,v_1)$, $(u_2, v_2)$, and $s_1, s_2:\mathcal{L}\rightarrow\mathcal{S}$ such that $s_1(A)=e_1$, $s_2(A)=e_2$ and for any $E\in\mathcal{L}, E\neq A$, $s_1(E)=s_2(E)$. Then, from the definition of $g$, one can write $g_{s_1}(u_1)=m_{s_1(A)}$, $g_{s_2}(u_2)=m_{s_2(A)}$ and $g_{s_2}(u_1)=m_{e_1\rightarrow e_2}m_{s_2(A)}$.
The sum of the powers of $u_i$ in the definition of $U_A^{s_i}$ is 1, so $U_A^{s_1}=\frac{u_1}{u_2}U_A^{s_2}$ and $$\begin{aligned}
g_{s_1}(u_1)&=m_{s_1(A)}\\&=m_{e_1\rightarrow e_2}m_{s_2(A)}\\&=g_{s_2}(u_1).\end{aligned}$$ The same reasonning works for $g_{s_1}(u_2)=g_{s_2}(u_2)$.
For $E\neq A$, the sum of the powers of $u_i$ in the definition of $U_E^{s_i}$ is 0, so changing $s$ has no influence on the definition of $g(u)$ for the short edges parallel with E.
4. Let us now prove that $g$ respects the relations in [$R_H$]{}.
The hexagonal face relations are preserved by definition of the image of $v_i$.
Until now, orientation of long edges, hence the sign, was implied in the notation, and short edges were considered parallel with long edges. Reversing orientation of a short edge naturally inverses the image of its generator $u$ by $g$.
Let us check now, that reversing orientation preserves the relation for the parameter $v$.
Let $e_1, e_2, e_3$ be three short edges on the boundary of a face $F\in\mathcal{\tilde{H}}$, respectively parallel to long edges $E_1, E_2, E_3$ (also on the boundary of $F$) and oriented such that the starting point of $e_1$ and $e_3$ bound $E_2$, and the end point of $e_1$ and $e_2$ bound $E_3$. Then, by definition of $g$, $g(v_2)=g(u_1)$. Let us reverse orientation of $e_2$ and let us call it $\bar{e_2}$, then $g(v_{\bar{2}})=g(u_3)$, and one has to check that $g(u_3)=-g(u_2)^{-1}g(u_1)$. Let us do this from the definition of $g(u_3)$ and let us assume without loss of generality, that $e_3=s(E_3)$. Let us write ${\ensuremath{\partial_{2}^H}}(F)=E_2+E_3-E_1\in{\ensuremath{\Delta_{1}^H}}(M)$, so $$\begin{aligned}
E_3&={\ensuremath{\partial_{2}^H}}(F)+E_1-E_2\\&={\ensuremath{\partial_{2}^H}}\left(F+\sum_{\phi}{\ensuremath{\eta_{E_1}}}(\phi) \phi-\sum_{\phi}{\ensuremath{\eta_{E_2}}}(\phi) \phi\right).\end{aligned}$$
Let us assume, as a first case, that ${\ensuremath{p_H}}(E_1), {\ensuremath{p_H}}(E_2), {\ensuremath{p_H}}(E_3)$ are three different edges and that $e_1=s(E_1)$ and $ e_2=s(E_2)$. Let us denote $u_i'=r(F,E_i)$. Then, one has the following tautological equality: $$u_3=\frac{u_3 u_1' u_2}{u_3' u_1 u_2'}\frac{U_1^s}{U_1}\frac{U_2}{U_2^s}.$$ $\frac{u_3 u_1' u_2}{u_3' u_1 u_2'}$ has as preimage in ${\ensuremath{R_I}}$ a product of three different shape parameters of a tetrahedron (which is equal to -1), $\frac{U_1^s}
{U_1}$ has as preimage in [$R_I$]{} a word $m_1$ from lemma \[Lemme\] such that $f(m_1)=u_1$ and $\frac{U_2^s}{U_2}$ has as preimage in [$R_I$]{} a word $m_2$ from lemma \[Lemme\] such that $f(m_2)=u_2$. So, by construction of $g$, $g(u_3)=-g(u_1)g(u_2)^{-1}$. Let us now assume that another edge of the boundary of $F$ has projection equal to $E_3$ in ${\ensuremath{\mathbf{\Delta^{1,H}}}} / {\ensuremath{\mathbf{\Phi^H}}}$, for example $E_1$, with $s(E_1)=s(E_3)=e_3$. Then, one has the following tautological equality: $u_3=\frac{u_3 u_1' u_2}{u_3' u_3 u_2'}\frac{U_1^s}{U_1}\frac{U_2}{U_2^s}$, but now, $\frac{U_1^s}{U_1}=u_3$ and $\frac{u_3 u_1' u_2}{u_3' u_3 u_2'}=\frac{u_3 u_1' u_2}{u_3' u_1 u_2'}\frac{u_1}{u_3}$ which has preimage in [$R_I$]{} the product of three different shape parameters of a tetrahedron and the word of corollary \[corollary\] going from $e_3$ to $e_1$.
Let us end with the triangular face relations. Let $e_i, e_j, e_k$ be short edges bounding a triangular face in ${\ensuremath{\overline{\mathcal{T}_H}}}$, with generators $(u_i,v_i),(u_j,v_j),(u_k,v_k)$. Let $F$ be the opposite hexagonal face, with long edges $E_i,E_j,E_k$ on its boundary. Let us assume they are oriented so that ${\ensuremath{\partial_{2}^H}}(F)=E_i+E_j-E_k$, and one has to prove that $g(u_i)g(u_j)g(u_k)^{ -1}=1$: the reasoning is similar to the previous one. One also has to prove that $$-g(v_k)+g(u_i)g(v_j)+g(v_i)=0,$$ or equivalently, $$g(u_i)g(v_j)g(v_k)^{-1}+g(v_i)g(v_k)^{-1}=1,$$ so, by definition of $g$, $$g(u_i)g(u_{\check{j}})g(u_{\check{k}})^{-1}+g(u_{\check{i}})g(u_{\check{k}})^{-1}=1.$$ Let us call $e_l$, the short edge in the same tetrahedron parallel with $e_i$, then it is equivalent to prove that $$g(u_i)g(u_l)^{-1}g(u_l)g(u_{\check{j}})g(u_{\check{k}})^{-1}+g(u_{\check{i}})g(u_{\check{k}})^{-1}=1.$$ The left part of the equality is equal, by definition of $g$ and thanks to the preceding relation, to $-z_kz_j+z_j=1$. \[fin enum\]
It is now a straightforward computation to check that $f\circ g =id_{{\ensuremath{R_I}}}$ and $g\circ f=id_{{\ensuremath{R_H}}}$.
We actually proved that for any edge $E$ of ${\ensuremath{\Delta_{1}^H}}(M)$ of which the projection onto $H_1(M)$ is trivial, the parameters $u_i$ associated to short edges parallel to $E$ are in the image of $f$.
Examples
========
Figure-eight knot
-----------------
Let $K$ be the figure-eight knot embedded in $\mathbb{S}^3$. The cell complex of figure \[figure eight knot\] is a truncated particular H-triangulation corresponding to the well known ideal triangulation of $\mathbb{S}^3\setminus K$, see \[3\]. The distinguished edge is dotted. The edges labelled with double and triple arrows are identified when collapsing that edge onto the vertex before truncation. The shape parameters $z_i$ and $w_i$ are actually associated to the corresponding edges of the ideal triangulation.
Thurston’s gluing equations are $z_1^2z_2w_1w_3^2=1$ and $z_2z_3^2w_1w_2^2=1$, or, equivalently, $z_1(1-z_1)w_3(1-w_3)=1$.
Keeping the previous notations, one has, for example, $f(z_1)=v_2v_{11}^{-1}=u_{7}u_{16}^{-1}$. According to theorem \[theorem\], $f$ is an isomorphism. Let us see, for example, what the preimage of $u_1$ is. $e_1$ is parallel to the edge labelled with a simple arrow. That edge is tautologically equal to ${\ensuremath{\partial_{2}^H}}(B)\in{\ensuremath{\Delta_{1}^H}}(\mathbb{S}^3)$. Let us consider the only triangular face opposite to the hexagonal face corresponding to $B$ which is not in the distinguished tetrahedron. Let us assume that the image by $s$ of the long edge labelled with a double arrow is $e_{15}$, and $e_1$ for the long one labelled with a simple arrow. Then, one has the following equality : $u_1=\frac{u_1 u_{15} u_{15}^{-1}}{u_{14} u_{15} u_{16}^{-1}}$. Besides, following the proof of lemma \[Lemme\], one has $f(w_2z_2w_2z_3w_1)=\frac{u_1}{u_{14}}$ and, by definition of $f$ and thanks to the hexagonal face relations in [$R_H$]{}, $f(w_3)=\frac{u_{16}}{u_{15}}$. Then, using the gluing equations in [$R_I$]{}, we find that $f(z_2z_3w_1w_{2}^{2}w_3)=u_1$.
(-0.8,-7.48) rectangle (39.02,4.33); (0.6,-6)– (9.4,-6); (5.25,-6) – (5,-6.32); (5.25,-6) – (5,-5.68); (4.2,-3)– (5.77,-2.98); (5.24,-2.99) – (4.99,-3.31); (5.24,-2.99) – (4.99,-2.67); (4.2,3.01)– (5.8,3.01); (5.25,3.01) – (5,2.69); (5.25,3.01) – (5,3.32); (18.23,-2.98)– (15.4,-4.61); (16.6,-3.92) – (16.65,-3.52); (16.6,-3.92) – (16.97,-4.07); (19.8,-3)– (22.6,-4.61); (21.41,-3.93) – (21.04,-4.08); (21.41,-3.93) – (21.36,-3.53); (22.6,-4.61)– (23.4,-6); (23.12,-5.52) – (22.72,-5.47); (23.12,-5.52) – (23.28,-5.15); (19,1.62)– (18.2,3.01); (18.48,2.53) – (18.88,2.47); (18.48,2.53) – (18.32,2.15); (15.4,-4.61)– (14.6,-6); (14.88,-5.52) – (14.72,-5.15); (14.88,-5.52) – (15.28,-5.47); (19,1.62)– (19.8,3.01); (19.52,2.53) – (19.68,2.15); (19.52,2.53) – (19.12,2.47); (33,-1.62)– (33,1.62); (33,0.25) – (33.32,0); (33,0.25) – (32.68,0); (29.4,-4.61)– (27.8,-4.61); (28.35,-4.61) – (28.6,-4.3); (28.35,-4.61) – (28.6,-4.93); (36.6,-4.61)– (38.2,-4.61); (37.65,-4.61) – (37.4,-4.93); (37.65,-4.61) – (37.4,-4.3); (27.8,-4.61)– (32.2,3.01); (30.12,-0.59) – (30.28,-0.96); (30.12,-0.59) – (29.72,-0.64); (32.2,-3)– (33,-1.62); (32.73,-2.09) – (32.88,-2.46); (32.73,-2.09) – (32.33,-2.15); (37.4,-6)– (38.2,-4.61); (37.92,-5.09) – (38.08,-5.47); (37.92,-5.09) – (37.52,-5.15); (33.77,-2.98)– (36.6,-4.61); (35.4,-3.92) – (35.03,-4.07); (35.4,-3.92) – (35.35,-3.52); (29.4,-4.61)– (28.6,-6); (28.88,-5.52) – (28.72,-5.15); (28.88,-5.52) – (29.28,-5.47); (33,1.62)– (33.8,3.01); (33.52,2.53) – (33.68,2.15); (33.52,2.53) – (33.12,2.47); (1.4,-4.61)– (4.2,-3); (3.23,-3.56) – (3.17,-3.96); (3.23,-3.56) – (2.86,-3.41); (2.8,-3.81) – (2.75,-4.2); (2.8,-3.81) – (2.43,-3.65); (8.6,-4.61)– (5.77,-2.98); (6.76,-3.55) – (7.13,-3.4); (6.76,-3.55) – (6.81,-3.95); (7.19,-3.8) – (7.56,-3.65); (7.19,-3.8) – (7.24,-4.2); (13.8,-4.61)– (18.2,3.01); (16.25,-0.38) – (16.4,-0.75); (16.25,-0.38) – (15.85,-0.43); (16,-0.8) – (16.15,-1.18); (16,-0.8) – (15.6,-0.86); (19,-1.62)– (19,1.62); (19,0.5) – (19.32,0.25); (19,0.5) – (18.68,0.25); (19,0) – (19.32,-0.24); (19,0) – (18.68,-0.24); (29.4,-4.61)– (32.2,-3); (31.23,-3.56) – (31.17,-3.96); (31.23,-3.56) – (30.86,-3.41); (30.8,-3.81) – (30.75,-4.2); (30.8,-3.81) – (30.43,-3.65); (9.4,-6)– (8.6,-4.61); (8.88,-5.09) – (9.28,-5.15); (8.88,-5.09) – (8.72,-5.47); (4.2,3.01)– (5,1.62); (0.6,-6)– (1.4,-4.61); (1.12,-5.09) – (1.28,-5.47); (1.12,-5.09) – (0.72,-5.15); (5.8,3.01)– (5,1.62); (15.4,-4.61)– (13.8,-4.61); (14.35,-4.61) – (14.6,-4.3); (14.35,-4.61) – (14.6,-4.93); (22.6,-4.61)– (24.2,-4.61); (23.65,-4.61) – (23.4,-4.93); (23.65,-4.61) – (23.4,-4.3); (18.23,-2.98)– (19,-1.62); (18.73,-2.08) – (18.89,-2.46); (18.73,-2.08) – (18.34,-2.14); (23.4,-6)– (24.2,-4.61); (23.92,-5.09) – (24.08,-5.47); (23.92,-5.09) – (23.52,-5.15); (37.4,-6)– (36.6,-4.61); (36.88,-5.09) – (37.28,-5.15); (36.88,-5.09) – (36.72,-5.47); (32.2,3.01)– (33,1.62); (32.72,2.1) – (32.32,2.15); (32.72,2.1) – (32.88,2.47); (-0.2,-4.61)– (4.2,3.01); (2.12,-0.59) – (2.28,-0.96); (2.12,-0.59) – (1.72,-0.64); (1.88,-1.02) – (2.03,-1.39); (1.88,-1.02) – (1.48,-1.07); (2.37,-0.16) – (2.52,-0.53); (2.37,-0.16) – (1.97,-0.22); (4.2,-3)– (5,-1.62); (9.4,-6)– (10.2,-4.61); (9.92,-5.09) – (10.08,-5.47); (9.92,-5.09) – (9.52,-5.15); (10.2,-4.61)– (5.8,3.01); (7.88,-0.59) – (8.28,-0.64); (7.88,-0.59) – (7.72,-0.96); (8.12,-1.02) – (8.52,-1.07); (8.12,-1.02) – (7.97,-1.39); (7.63,-0.16) – (8.03,-0.22); (7.63,-0.16) – (7.48,-0.53); (5.77,-2.98)– (5,-1.62); (0.6,-6)– (-0.2,-4.61); (0.08,-5.09) – (0.48,-5.15); (0.08,-5.09) – (-0.08,-5.47); (14.6,-6)– (23.4,-6); (19.25,-6) – (19,-6.32); (19.25,-6) – (19,-5.68); (18.75,-6) – (18.51,-6.32); (18.75,-6) – (18.51,-5.68); (19.74,-6) – (19.49,-6.32); (19.74,-6) – (19.49,-5.68); (18.23,-2.98)– (19.8,-3); (19.26,-2.99) – (19.01,-3.31); (19.26,-2.99) – (19.01,-2.67); (18.2,3.01)– (19.8,3.01); (19.25,3.01) – (19,2.69); (19.25,3.01) – (19,3.32); (28.6,-6)– (37.4,-6); (33.25,-6) – (33,-6.32); (33.25,-6) – (33,-5.68); (32.75,-6) – (32.51,-6.32); (32.75,-6) – (32.51,-5.68); (33.74,-6) – (33.49,-6.32); (33.74,-6) – (33.49,-5.68); (32.2,-3)– (33.77,-2.98); (33.24,-2.99) – (32.99,-3.31); (33.24,-2.99) – (32.99,-2.67); (32.2,3.01)– (33.8,3.01); (33.25,3.01) – (33,2.69); (33.25,3.01) – (33,3.32); (33.8,3.01)– (38.2,-4.61); (36.12,-1.02) – (35.72,-0.96); (36.12,-1.02) – (36.28,-0.64); (35.88,-0.59) – (35.48,-0.53); (35.88,-0.59) – (36.03,-0.22); (36.37,-1.45) – (35.97,-1.39); (36.37,-1.45) – (36.52,-1.07); (33,-1.62)– (33.77,-2.98); (33.51,-2.51) – (33.11,-2.46); (33.51,-2.51) – (33.66,-2.14); (27.8,-4.61)– (28.6,-6); (28.32,-5.52) – (27.92,-5.47); (28.32,-5.52) – (28.48,-5.15); (19.8,3.01)– (24.2,-4.61); (22.12,-1.02) – (21.72,-0.96); (22.12,-1.02) – (22.28,-0.64); (19,-1.62)– (19.8,-3); (19.52,-2.52) – (19.12,-2.46); (19.52,-2.52) – (19.67,-2.15); (13.8,-4.61)– (14.6,-6); (14.32,-5.52) – (13.92,-5.47); (14.32,-5.52) – (14.48,-5.15); (5,-1.62)– (5,1.62); (5,0.25) – (5.32,0); (5,0.25) – (4.68,0); (1.4,-4.61)– (-0.2,-4.61); (0.35,-4.61) – (0.6,-4.3); (0.35,-4.61) – (0.6,-4.93); (8.6,-4.61)– (10.2,-4.61); (9.65,-4.61) – (9.4,-4.93); (9.65,-4.61) – (9.4,-4.3); (19.16,0.57) node\[anchor=north west\] [$z_1$]{}; (22.17,0.05) node\[anchor=north west\] [$z_3$]{}; (14.03,0.28) node\[anchor=north west\] [$z_2$]{}; (33.14,0.47) node\[anchor=north west\] [$w_1$]{}; (36.43,-0.23) node\[anchor=north west\] [$w_3$]{}; (27.73,0.52) node\[anchor=north west\] [$w_2$]{}; (2.11,-1.32) node\[anchor=north west\] [A]{}; (6.78,-1.55) node\[anchor=north west\] [A]{}; (4.47,-4.19) node\[anchor=north west\] [B]{}; (7.81,2.36) node\[anchor=north west\] [C]{}; (15.91,-1.5) node\[anchor=north west\] [B]{}; (35.92,2.36) node\[anchor=north west\] [C]{}; (18.78,-4.09) node\[anchor=north west\] [D]{}; (34.93,-1.69) node\[anchor=north west\] [D]{}; (20.95,-1.46) node\[anchor=north west\] [E]{}; (29.84,-1.6) node\[anchor=north west\] [E]{}; (22.26,2.4) node\[anchor=north west\] [F]{}; (32.53,-4.14) node\[anchor=north west\] [F]{};
(5.08,-3.62) node [2]{}; (5.03,3.77) node [1]{}; (22.31,-5.41) node [14]{}; (18.17,1.79) node [2]{}; (15.58,-5.22) node [9]{}; (19.91,1.79) node [11]{}; (28.95,-3.86) node [12]{}; (37.14,-3.91) node [14]{}; (31.96,-2.02) node [11]{}; (38.37,-5.51) node [1]{}; (29.66,-5.36) node [8]{}; (33.8,1.7) node [9]{}; (8.42,-5.36) node [7]{}; (1.69,-5.32) node [4]{}; (14.87,-3.81) node [7]{}; (22.92,-3.86) node [16]{}; (18.07,-1.93) node [4]{}; (24.67,-5.32) node [15]{}; (36.06,-5.27) node [15]{}; (32.15,1.79) node [16]{}; (10.45,-5.46) node [6]{}; (-0.29,-5.65) node [5]{}; (18.92,-3.72) node [13]{}; (18.97,3.82) node [10]{}; (33.05,-3.67) node [10]{}; (33.19,3.77) node [6]{}; (33.99,-1.98) node [13]{}; (27.63,-5.55) node [5]{}; (20.05,-1.83) node [12]{}; (13.7,-5.55) node [8]{}; (0.84,-3.86) node [3]{}; (9.27,-3.91) node [3]{};
A knot in $\mathbb{S}^2\times\mathbb{S}^1$
------------------------------------------
Figure \[essedeuxcroixessun\] represents a truncated particular H-triangulation of a knot in $\mathbb{S}^2\times\mathbb{S}^1$. The knot is represented by the dotted edge. Here, $M=\mathbb{S}^2\times\mathbb{S}^1$, so $H_1(M,\mathbb{Z})\simeq H_2(M,\mathbb{Z})\simeq \mathbb{Z}$.
The shape parameters $z$ and $w$ are associated to the corresponding edges in [$\mathcal{T}_I$]{}. The Thurston’s gluing equation is $zw=1$. After computation, the presentation of [$R_H$]{} reduces to $\mathbb{Z}\langle u_1,u_{\bar{1}},u_2 \rvert u_1u_{\bar{1}}=1,u_2^3=1\rangle$
$f$ is neither surjective, nor injective. Indeed, $u_1\notin\text{Im}(f)$, and $f(w)=u_2$, so $f(w^3)=1$.
(-6,2.46) rectangle (12.47,7.38); (-2.75,6.9)– (-3.75,6.9); (-3.38,6.9) – (-3.25,7.07); (-3.38,6.9) – (-3.25,6.74); (-3.25,6.04)– (-3.75,6.9); (-3.56,6.58) – (-3.35,6.55); (-3.56,6.58) – (-3.64,6.39); (-3.25,6.04)– (-2.75,6.9); (-2.93,6.58) – (-2.85,6.39); (-2.93,6.58) – (-3.14,6.55); (-5.62,3.65)– (-5.12,2.78); (-5.31,3.1) – (-5.51,3.13); (-5.31,3.1) – (-5.23,3.3); (-5.62,3.65)– (-4.62,3.65); (-4.99,3.65) – (-5.12,3.48); (-4.99,3.65) – (-5.12,3.81); (-4.62,3.65)– (-5.12,2.78); (-4.93,3.1) – (-5.01,3.3); (-4.93,3.1) – (-4.73,3.13); (-0.86,3.65)– (-1.36,2.79); (-1.18,3.11) – (-1.26,3.3); (-1.18,3.11) – (-0.97,3.14); (-1.86,3.65)– (-1.36,2.79); (-1.55,3.11) – (-1.76,3.14); (-1.55,3.11) – (-1.47,3.3); (-3.24,5.03)– (-3.75,4.15); (-3.56,4.48) – (-3.64,4.67); (-3.56,4.48) – (-3.35,4.51); (-2.74,4.16)– (-3.75,4.15); (-3.37,4.16) – (-3.24,4.32); (-3.37,4.16) – (-3.24,3.99); (-3.24,5.03)– (-2.74,4.16); (-2.93,4.48) – (-3.14,4.51); (-2.93,4.48) – (-2.85,4.67); (-3.75,6.9)– (-5.62,3.65); (-4.75,5.16) – (-4.83,5.36); (-4.75,5.16) – (-4.54,5.19); (-2.75,6.9)– (-0.86,3.65); (-1.74,5.17) – (-1.95,5.19); (-1.74,5.17) – (-1.66,5.36); (-3.25,6.04)– (-3.24,5.03); (-3.24,5.4) – (-3.41,5.53); (-3.24,5.4) – (-3.08,5.53); (-1.36,2.79)– (-5.12,2.78); (-3.37,2.78) – (-3.24,2.95); (-3.37,2.78) – (-3.24,2.62); (-3.11,2.78) – (-2.98,2.95); (-3.11,2.78) – (-2.98,2.62); (-3.63,2.78) – (-3.5,2.95); (-3.63,2.78) – (-3.5,2.62); (3.75,6.95)– (2.75,6.96); (3.12,6.95) – (3.25,7.12); (3.12,6.95) – (3.25,6.79); (1.32,2.78)– (0.82,3.64); (1.01,3.32) – (1.22,3.29); (1.01,3.32) – (0.93,3.13); (5.15,2.76)– (5.65,3.63); (5.47,3.31) – (5.55,3.11); (5.47,3.31) – (5.26,3.28); (7.83,2.79)– (7.33,3.66); (7.51,3.34) – (7.72,3.31); (7.51,3.34) – (7.43,3.14); (11.66,2.8)– (12.16,3.67); (11.97,3.35) – (12.05,3.15); (11.97,3.35) – (11.76,3.32); (10.23,6.98)– (9.23,6.98); (9.6,6.98) – (9.73,7.15); (9.6,6.98) – (9.73,6.81); (1.82,3.64)– (1.32,2.78); (1.51,3.1) – (1.43,3.29); (1.51,3.1) – (1.72,3.13); (1.82,3.64)– (0.82,3.64); (1.2,3.64) – (1.33,3.81); (1.2,3.64) – (1.32,3.48); (4.65,3.63)– (5.15,2.76); (4.97,3.09) – (4.76,3.12); (4.97,3.09) – (5.05,3.28); (4.65,3.63)– (5.65,3.63); (5.28,3.63) – (5.15,3.46); (5.28,3.63) – (5.16,3.8); (3.74,4.16)– (3.24,5.03); (3.43,4.71) – (3.64,4.68); (3.43,4.71) – (3.35,4.51); (2.74,4.17)– (3.24,5.03); (3.06,4.71) – (3.14,4.52); (3.06,4.71) – (2.85,4.68); (3.74,4.16)– (2.74,4.17); (3.11,4.16) – (3.24,4.33); (3.11,4.16) – (3.24,4); (3.25,6.09)– (2.75,6.96); (2.93,6.63) – (3.14,6.6); (2.93,6.63) – (2.85,6.44); (3.25,6.09)– (3.75,6.95); (3.56,6.63) – (3.64,6.44); (3.56,6.63) – (3.35,6.6); (9.24,4.19)– (9.74,5.06); (9.55,4.74) – (9.63,4.54); (9.55,4.74) – (9.34,4.71); (10.24,4.19)– (9.74,5.06); (9.92,4.74) – (10.13,4.71); (9.92,4.74) – (9.84,4.54); (10.24,4.19)– (9.24,4.19); (9.61,4.19) – (9.74,4.36); (9.61,4.19) – (9.74,4.02); (7.33,3.66)– (8.33,3.66); (7.96,3.66) – (7.83,3.49); (7.96,3.66) – (7.83,3.82); (7.83,2.79)– (8.33,3.66); (8.14,3.34) – (8.22,3.14); (8.14,3.34) – (7.93,3.31); (11.66,2.8)– (11.16,3.67); (11.34,3.35) – (11.55,3.32); (11.34,3.35) – (11.26,3.15); (11.16,3.67)– (12.16,3.67); (11.79,3.67) – (11.66,3.5); (11.79,3.67) – (11.66,3.83); (9.23,6.98)– (9.73,6.11); (9.55,6.43) – (9.34,6.46); (9.55,6.43) – (9.63,6.63); (10.23,6.98)– (9.73,6.11); (9.92,6.43) – (9.84,6.63); (9.92,6.43) – (10.13,6.46); (0.82,3.64)– (2.75,6.96); (1.85,5.41) – (1.93,5.22); (1.85,5.41) – (1.64,5.38); (5.65,3.63)– (3.75,6.95); (4.64,5.4) – (4.85,5.37); (4.64,5.4) – (4.56,5.21); (2.74,4.17)– (1.82,3.64); (2.17,3.84) – (2.2,4.05); (2.17,3.84) – (2.37,3.76); (3.74,4.16)– (4.65,3.63); (4.31,3.83) – (4.12,3.75); (4.31,3.83) – (4.28,4.04); (3.24,5.03)– (3.25,6.09); (3.25,5.82) – (3.41,5.69); (3.25,5.82) – (3.08,5.69); (3.25,5.56) – (3.41,5.43); (3.25,5.56) – (3.08,5.43); (5.15,2.76)– (1.32,2.78); (3.11,2.77) – (3.24,2.94); (3.11,2.77) – (3.24,2.6); (3.37,2.77) – (3.5,2.94); (3.37,2.77) – (3.5,2.6); (2.85,2.77) – (2.98,2.94); (2.85,2.77) – (2.98,2.6); (11.66,2.8)– (7.83,2.79); (9.61,2.8) – (9.74,2.96); (9.61,2.8) – (9.74,2.63); (9.87,2.8) – (10,2.96); (9.87,2.8) – (10,2.63); (9.35,2.79) – (9.48,2.96); (9.35,2.79) – (9.48,2.63); (7.33,3.66)– (9.23,6.98); (8.41,5.54) – (8.49,5.35); (8.41,5.54) – (8.2,5.51); (8.28,5.32) – (8.36,5.12); (8.28,5.32) – (8.07,5.29); (12.16,3.67)– (10.23,6.98); (11.06,5.55) – (11.27,5.52); (11.06,5.55) – (10.98,5.35); (11.19,5.32) – (11.4,5.3); (11.19,5.32) – (11.11,5.13); (9.74,5.06)– (9.73,6.11); (9.73,5.72) – (9.9,5.59); (9.73,5.72) – (9.57,5.59); (8.33,3.66)– (9.24,4.19); (8.89,3.99) – (8.87,3.78); (8.89,3.99) – (8.7,4.07); (11.16,3.67)– (10.24,4.19); (10.59,3.99) – (10.78,4.07); (10.59,3.99) – (10.62,3.78); (-0.86,3.65)– (-1.86,3.65); (-1.49,3.65) – (-1.36,3.82); (-1.49,3.65) – (-1.36,3.49); (-4.62,3.65)– (-3.75,4.15); (-3.96,4.03) – (-3.99,3.82); (-3.96,4.03) – (-4.15,4.11); (-4.18,3.9) – (-4.21,3.69); (-4.18,3.9) – (-4.38,3.98); (-1.86,3.65)– (-2.74,4.16); (-2.53,4.03) – (-2.33,4.11); (-2.53,4.03) – (-2.5,3.82); (-2.3,3.9) – (-2.11,3.98); (-2.3,3.9) – (-2.27,3.69);
(-4.44,4.76) node [A]{}; (-2.09,4.71) node [A]{}; (-3.28,3.42) node [E]{}; (-1.87,6.14) node [B]{}; (2.09,4.73) node [C]{}; (4.56,4.63) node [D]{}; (3.15,3.45) node [B]{}; (8.49,4.73) node [C]{}; (10.92,4.73) node [D]{}; (4.68,6.04) node [F]{}; (9.65,3.42) node [F]{}; (11.21,6.07) node [E]{}; (3.3,7.18) node [$e_2$]{}; (2.73,6.39) node [$e_1$]{}; (9.9,3.87) node [$e_2$]{}; (9.23,6.41) node [$e_1$]{}; (3.57,5.33) node [$z$]{}; (10,5.38) node [$w$]{};
Knot in $L(3,1)$
----------------
(-6.36,-3.13) rectangle (13.05,8.16); (-2.75,6.9)– (-3.75,6.9); (-3.38,6.9) – (-3.25,7.08); (-3.38,6.9) – (-3.25,6.73); (-3.25,6.04)– (-3.75,6.9); (-3.56,6.59) – (-3.34,6.56); (-3.56,6.59) – (-3.65,6.38); (-3.25,6.04)– (-2.75,6.9); (-2.93,6.59) – (-2.84,6.38); (-2.93,6.59) – (-3.15,6.56); (-5.62,3.65)– (-5.12,2.78); (-5.3,3.1) – (-5.52,3.13); (-5.3,3.1) – (-5.22,3.3); (-5.62,3.65)– (-4.62,3.65); (-4.99,3.65) – (-5.12,3.47); (-4.99,3.65) – (-5.12,3.82); (-4.62,3.65)– (-5.12,2.78); (-4.94,3.1) – (-5.02,3.3); (-4.94,3.1) – (-4.72,3.13); (-0.86,3.65)– (-1.36,2.79); (-1.18,3.1) – (-1.26,3.31); (-1.18,3.1) – (-0.96,3.13); (-1.86,3.65)– (-1.36,2.79); (-1.55,3.1) – (-1.76,3.13); (-1.55,3.1) – (-1.46,3.31); (-3.24,5.03)– (-3.75,4.15); (-3.56,4.47) – (-3.65,4.68); (-3.56,4.47) – (-3.34,4.5); (-2.74,4.16)– (-3.75,4.15); (-3.38,4.16) – (-3.24,4.33); (-3.38,4.16) – (-3.24,3.98); (-3.24,5.03)– (-2.74,4.16); (-2.92,4.47) – (-3.14,4.5); (-2.92,4.47) – (-2.84,4.68); (-3.25,6.04)– (-3.24,5.03); (-3.24,5.4) – (-3.42,5.53); (-3.24,5.4) – (-3.07,5.53); (-1.36,2.79)– (-5.12,2.78); (-3.38,2.78) – (-3.24,2.96); (-3.38,2.78) – (-3.24,2.61); (3.75,6.95)– (2.75,6.96); (3.12,6.95) – (3.25,7.13); (3.12,6.95) – (3.25,6.78); (1.32,2.78)– (0.82,3.64); (1.01,3.33) – (1.22,3.3); (1.01,3.33) – (0.92,3.12); (5.15,2.76)– (5.65,3.63); (5.47,3.31) – (5.55,3.11); (5.47,3.31) – (5.25,3.28); (7.83,2.79)– (7.33,3.66); (7.51,3.34) – (7.73,3.31); (7.51,3.34) – (7.43,3.14); (11.66,2.8)– (12.16,3.67); (11.97,3.35) – (12.06,3.15); (11.97,3.35) – (11.76,3.32); (10.23,6.98)– (9.23,6.98); (9.6,6.98) – (9.73,7.15); (9.6,6.98) – (9.73,6.8); (1.82,3.64)– (0.82,3.64); (1.19,3.64) – (1.33,3.82); (1.19,3.64) – (1.32,3.47); (4.65,3.63)– (5.15,2.76); (4.97,3.08) – (4.75,3.11); (4.97,3.08) – (5.05,3.28); (4.65,3.63)– (5.65,3.63); (5.29,3.63) – (5.15,3.46); (5.29,3.63) – (5.16,3.8); (3.74,4.16)– (3.24,5.03); (3.43,4.72) – (3.65,4.68); (3.43,4.72) – (3.34,4.51); (2.74,4.17)– (3.24,5.03); (3.06,4.72) – (3.14,4.51); (3.06,4.72) – (2.84,4.69); (3.74,4.16)– (2.74,4.17); (3.11,4.16) – (3.24,4.34); (3.11,4.16) – (3.24,3.99); (3.25,6.09)– (2.75,6.96); (2.93,6.64) – (3.15,6.61); (2.93,6.64) – (2.85,6.43); (9.24,4.19)– (9.74,5.06); (9.55,4.74) – (9.64,4.54); (9.55,4.74) – (9.34,4.71); (10.24,4.19)– (9.74,5.06); (9.92,4.74) – (10.14,4.71); (9.92,4.74) – (9.84,4.54); (10.24,4.19)– (9.24,4.19); (9.6,4.19) – (9.74,4.36); (9.6,4.19) – (9.74,4.01); (7.33,3.66)– (8.33,3.66); (7.96,3.66) – (7.83,3.48); (7.96,3.66) – (7.83,3.83); (7.83,2.79)– (8.33,3.66); (8.14,3.34) – (8.23,3.14); (8.14,3.34) – (7.93,3.31); (11.66,2.8)– (11.16,3.67); (11.34,3.35) – (11.56,3.32); (11.34,3.35) – (11.26,3.15); (11.16,3.67)– (12.16,3.67); (11.79,3.67) – (11.66,3.49); (11.79,3.67) – (11.66,3.84); (9.23,6.98)– (9.73,6.11); (9.55,6.43) – (9.33,6.46); (9.55,6.43) – (9.63,6.63); (10.23,6.98)– (9.73,6.11); (9.92,6.43) – (9.83,6.63); (9.92,6.43) – (10.13,6.46); (0.82,3.64)– (2.75,6.96); (1.86,5.42) – (1.94,5.21); (1.86,5.42) – (1.64,5.39); (5.65,3.63)– (3.75,6.95); (4.64,5.41) – (4.85,5.38); (4.64,5.41) – (4.55,5.2); (2.74,4.17)– (1.82,3.64); (2.16,3.84) – (2.2,4.05); (2.16,3.84) – (2.37,3.75); (5.15,2.76)– (1.32,2.78); (3.1,2.77) – (3.24,2.94); (3.1,2.77) – (3.24,2.6); (11.66,2.8)– (7.83,2.79); (9.61,2.8) – (9.74,2.97); (9.61,2.8) – (9.74,2.62); (8.33,3.66)– (9.24,4.19); (8.9,3.99) – (8.87,3.77); (8.9,3.99) – (8.69,4.07); (11.16,3.67)– (10.24,4.19); (10.58,4) – (10.79,4.08); (10.58,4) – (10.61,3.78); (-4.62,3.65)– (-3.75,4.15); (-4.07,3.97) – (-4.1,3.75); (-4.07,3.97) – (-4.27,4.05); (0.51,2.01)– (-0.49,2.01); (-0.13,2.01) – (0.01,2.19); (-0.13,2.01) – (0.01,1.84); (0.01,1.15)– (-0.49,2.01); (-0.31,1.7) – (-0.09,1.67); (-0.31,1.7) – (-0.39,1.49); (0.01,1.15)– (0.51,2.01); (0.33,1.7) – (0.41,1.49); (0.33,1.7) – (0.11,1.67); (-2.37,-1.24)– (-1.86,-2.11); (-2.05,-1.79) – (-2.27,-1.76); (-2.05,-1.79) – (-1.96,-1.59); (-2.37,-1.24)– (-1.37,-1.24); (-1.73,-1.24) – (-1.87,-1.42); (-1.73,-1.24) – (-1.87,-1.07); (-1.37,-1.24)– (-1.86,-2.11); (-1.68,-1.79) – (-1.77,-1.59); (-1.68,-1.79) – (-1.46,-1.76); (2.39,-1.24)– (1.89,-2.11); (2.07,-1.79) – (1.99,-1.59); (2.07,-1.79) – (2.29,-1.76); (1.39,-1.24)– (1.89,-2.11); (1.71,-1.79) – (1.49,-1.76); (1.71,-1.79) – (1.79,-1.59); (0.01,0.13)– (-0.49,-0.74); (-0.31,-0.42) – (-0.39,-0.21); (-0.31,-0.42) – (-0.09,-0.39); (0.51,-0.74)– (-0.49,-0.74); (-0.12,-0.74) – (0.01,-0.56); (-0.12,-0.74) – (0.01,-0.91); (0.01,0.13)– (0.51,-0.74); (0.33,-0.42) – (0.11,-0.39); (0.33,-0.42) – (0.41,-0.21); (0.51,2.01)– (2.39,-1.24); (1.52,0.27) – (1.3,0.3); (1.52,0.27) – (1.6,0.47); (2.39,-1.24)– (1.39,-1.24); (1.76,-1.24) – (1.89,-1.07); (1.76,-1.24) – (1.89,-1.41); (7.33,1.92)– (6.33,1.92); (6.69,1.92) – (6.83,2.1); (6.69,1.92) – (6.83,1.75); (6.83,1.06)– (7.33,1.92); (7.15,1.61) – (7.23,1.4); (7.15,1.61) – (6.93,1.58); (4.45,-1.33)– (4.96,-2.2); (4.77,-1.88) – (4.55,-1.85); (4.77,-1.88) – (4.86,-1.68); (4.45,-1.33)– (5.45,-1.33); (5.09,-1.33) – (4.95,-1.51); (5.09,-1.33) – (4.95,-1.16); (5.45,-1.33)– (4.96,-2.2); (5.14,-1.88) – (5.05,-1.68); (5.14,-1.88) – (5.36,-1.85); (9.21,-1.33)– (8.71,-2.2); (8.89,-1.88) – (8.81,-1.68); (8.89,-1.88) – (9.11,-1.85); (6.83,0.04)– (6.33,-0.83); (6.51,-0.51) – (6.43,-0.3); (6.51,-0.51) – (6.73,-0.48); (7.33,-0.83)– (6.33,-0.83); (6.7,-0.83) – (6.83,-0.65); (6.7,-0.83) – (6.83,-1); (6.83,0.04)– (7.33,-0.83); (7.15,-0.51) – (6.93,-0.48); (7.15,-0.51) – (7.23,-0.3); (6.33,1.92)– (4.45,-1.33); (5.32,0.18) – (5.24,0.38); (5.32,0.18) – (5.54,0.21); (7.33,1.92)– (9.21,-1.33); (8.34,0.18) – (8.12,0.21); (8.34,0.18) – (8.42,0.38); (6.83,1.06)– (6.83,0.04); (6.83,0.41) – (6.66,0.55); (6.83,0.41) – (7,0.55); (9.21,-1.33)– (8.21,-1.33); (8.58,-1.33) – (8.71,-1.16); (8.58,-1.33) – (8.71,-1.5); (5.45,-1.33)– (6.33,-0.83); (6.01,-1.01) – (5.98,-1.23); (6.01,-1.01) – (5.81,-0.93); (-1.36,2.79)– (-5.12,2.78); (-5.62,3.65)– (-3.75,6.9); (-4.62,5.39) – (-4.53,5.19); (-4.62,5.39) – (-4.83,5.36); (-0.86,3.65)– (-2.75,6.9); (-1.87,5.4) – (-1.65,5.37); (-1.87,5.4) – (-1.96,5.19); (4.65,3.63)– (3.74,4.16); (4.08,3.97) – (4.29,4.05); (4.08,3.97) – (4.11,3.75); (-0.86,3.65)– (-1.86,3.64); (-1.5,3.64) – (-1.36,3.82); (-1.5,3.64) – (-1.36,3.47); (-1.86,3.64)– (-2.74,4.16); (-2.42,3.97) – (-2.21,4.05); (-2.42,3.97) – (-2.39,3.75); (9.23,6.98)– (7.33,3.66); (8.21,5.2) – (8.13,5.4); (8.21,5.2) – (8.43,5.23); (12.16,3.67)– (10.23,6.98); (11.13,5.44) – (11.34,5.41); (11.13,5.44) – (11.04,5.24); (-1.86,-2.11)– (1.89,-2.11); (0.15,-2.11) – (0.01,-2.28); (0.15,-2.11) – (0.01,-1.93); (-2.37,-1.24)– (-0.49,2.01); (-1.36,0.5) – (-1.28,0.3); (-1.36,0.5) – (-1.58,0.47); (0.01,0.13)– (0.01,1.15); (0.01,0.78) – (0.18,0.64); (0.01,0.78) – (-0.16,0.64); (-0.49,-0.74)– (-1.37,-1.24); (-1.04,-1.06) – (-1.01,-0.84); (-1.04,-1.06) – (-0.84,-1.14); (0.51,-0.74)– (1.39,-1.24); (1.07,-1.06) – (0.87,-1.14); (1.07,-1.06) – (1.04,-0.84); (4.96,-2.2)– (8.71,-2.2); (6.97,-2.2) – (6.83,-2.37); (6.97,-2.2) – (6.83,-2.02); (7.33,-0.83)– (8.21,-1.33); (7.89,-1.15) – (7.69,-1.23); (7.89,-1.15) – (7.86,-0.93); (9.74,5.06)– (9.73,6.11); (9.73,5.72) – (9.91,5.59); (9.73,5.72) – (9.56,5.59); (3.1,5.6) node\[anchor=north west\] [$z$]{}; (9.55,5.6) node\[anchor=north west\] [$w$]{}; (-0.09,0.84) node\[anchor=north west\] [$t$]{}; (6.69,0.66) node\[anchor=north west\] [$u$]{}; (1.32,2.78)– (1.82,3.64); (1.64,3.33) – (1.72,3.12); (1.64,3.33) – (1.42,3.3); (3.75,6.95)– (3.25,6.09); (3.43,6.4) – (3.35,6.61); (3.43,6.4) – (3.65,6.43); (3.25,6.09)– (3.24,5.03); (3.25,5.43) – (3.07,5.56); (3.25,5.43) – (3.42,5.56); (6.33,1.92)– (6.83,1.06); (6.65,1.37) – (6.43,1.4); (6.65,1.37) – (6.73,1.58); (8.71,-2.2)– (8.21,-1.33); (8.39,-1.65) – (8.61,-1.68); (8.39,-1.65) – (8.31,-1.85);
(-4.09,5.01) node [A]{}; (-2.41,5.01) node [A]{}; (-3.29,3.42) node [B]{}; (-1.82,6.2) node [C]{}; (2.25,5) node [F]{}; (4.19,5) node [E]{}; (3.15,3.44) node [H]{}; (8.88,5) node [F]{}; (10.55,5) node [E]{}; (4.78,6.12) node [D]{}; (9.68,3.42) node [C]{}; (11.28,6.15) node [H]{}; (-3.27,2.49) node [c]{}; (7.36,3.08) node [$e_1$]{}; (2.59,4.55) node [$e_2$]{}; (8.03,3.83) node [$e_2$]{}; (1.48,5.61) node [a]{}; (4.96,5.56) node [a]{}; (2.12,4.27) node [a]{}; (3.23,2.44) node [c]{}; (9.7,2.54) node [c]{}; (8.62,4.27) node [a]{}; (10.84,4.24) node [a]{}; (-4.35,4.24) node [b]{}; (1.32,-1.92) node [$e_3$]{}; (1.79,0.63) node [b]{}; (5.09,0.55) node [c]{}; (8.52,0.63) node [a]{}; (6.58,0.63) node [e]{}; (5.66,-0.71) node [c]{}; (-5.07,5.53) node [a]{}; (-1.51,5.53) node [a]{}; (4.39,4.24) node [d]{}; (-2.03,4.19) node [b]{}; (7.95,5.61) node [a]{}; (11.48,5.66) node [d]{}; (0.11,-2.44) node [b]{}; (-1.69,0.68) node [c]{}; (-0.25,0.79) node [e]{}; (-1.13,-0.66) node [c]{}; (1.17,-0.66) node [a]{}; (6.84,-2.51) node [a]{}; (8.05,-0.81) node [b]{}; (-0.02,-1.41) node [G]{}; (-0.97,0.09) node [I]{}; (0.91,0.04) node [J]{}; (1.4,1.61) node [B]{}; (5.73,-0.17) node [I]{}; (7.9,-0.14) node [J]{}; (8.31,1.64) node [D]{}; (6.79,-1.46) node [G]{}; (9.52,5.84) node [a]{}; (2.07,3.03) node [$e_1$]{}; (3.08,5.7) node [a]{}; (8.1,-2) node [$e_3$]{};
The figure \[knot in ltroisun\] represents a truncated particular H-triangulation of a knot in $L(3,1)$. The knot is represented by the dotted edge. The shape parameters $z,t,u,w$ are actually associated to edges of the corresponding ideal triangulation. Let us remind that $H_1(L(3,1),\mathbb{Z})=\mathbb{Z}/3\mathbb{Z}$ and $H_2(L(3,1),\mathbb{Z})=0$.
The gluing equations are $t=u^{-1}$ and $z=w^{-1}$ and the presentation of the ring [$R_H$]{} reduces to $\mathbb{Z}\langle u_1,u_{\bar{1}},u_2,u_3, u_{\bar{3}} \mid u_1u_{\bar{1}}=1, u_3u_{\bar{3}}=1, u_2^3=-1 \rangle$. $f$ is not surjective, because $\text{Im}(f)=\mathbb{Z}\langle u_1u_2^{2}, u_{\bar{1}}u_2,u_3, u_{\bar{3}} \mid u_1u_{\bar{1}}=1, u_3u_{\bar{3}}=1, u_2^3=-1 \rangle$.
References
==========
<span style="font-variant:small-caps;">Section de Mathématiques, Université de Genève, 2-4 rue du Lièvre, Case postale 64, 1211 Genève 4, Suisse.</span>
*E-mail address*: `[email protected]`
[^1]: The work is supported by Swiss National Science Foundation, subsides SNF 200020\_149226.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present results on the full one-loop electroweak radiative corrections to the squark decay partial widths into charginos and neutralinos. We show the renormalization framework, and present numerical results for the third squark family. The corrections can reach values of $\sim 10\%$, which are comparable to the radiative corrections from the strong sector of the model. Therefore they should be taken into account for the precise extraction of the SUSY parameters at future colliders.'
---
[**Full electroweak one-loop radiative corrections to squark decays in the MSSM**]{}
[J. Guasch$^{a}$, W. Hollik$^{a}$, J. Solà$^{b}$]{}
[*$^a$ Institut für Theoretische Physik, Universität Karlsruhe, D-76128 Karlsruhe, Germany\
$^b$ Grup de F[í]{}sica Te[ò]{}rica and Institut de F[í]{}sica d’Altes Energies, Universitat Aut[ò]{}noma de Barcelona, E-08193 Bellaterra (Barcelona), Catalonia, Spain*]{}
The Standard Model (SM) of the strong and electroweak interactions is the present paradigm of particle physics. Its validity has been tested to a level better than one per mil in the particle accelerators [@PDB2000]. Nevertheless, there are arguments against the SM being the fundamental model of particle interactions [@Carena:2000yx], giving rise to the investigation of competing alternative or extended models, which can be tested at high-energy colliders, such as the Large Hadron Collider (LHC) [@ATLASCMS], or a $e^+e^-$ Linear Collider (LC) [@Miller:1998fh]. One of the most promising possibilities for physics beyond the SM is the incorporation of Supersymmetry (SUSY), which leads to a renormalizable field theory with precisely calculable predictions to be tested in future experiments. The simplest supersymmetric extension of the SM is the Minimal Supersymmetric Standard Model (MSSM) [@MSSM]. If the masses of the extra non-standard particles are very large as compared to the SM electroweak scale, the effects of these particles decouple from the SM low-energy effective Lagrangian [@Dobado:1997up]. This means that if the extra particles are too heavy we could not discern between the SM and the MSSM by just looking at the low-energy end of the spectrum, since the only trace of the MSSM would be a light Higgs boson ($M_{h^0}{\mbox{ \raisebox{-4pt}{${\stackrel{\textstyle <}{\sim}}$} }}135{\mbox{ GeV}}$) [@Higgs2L], whose properties would not differ from the SM one. Nevertheless, when some of the extra particle masses are of the order of the electroweak scale, the next generation of colliders will be able to produce such kind of particles and investigate their properties. In this case non-decoupling effects appear [@Djouadi:1997wt; @GHS2]. While the LHC will be able to produce new particles with masses up to $2.5{\mbox{ TeV}}$, the LC will be able to make precision measurements of their properties, provided they are not too heavy. For an adequate analysis, precise theoretical predictions are required, going beyond the Born approximation for SUSY processes.
Up to now the major effort on the computation of SUSY radiative corrections has been put into the computation of virtual SUSY effects in observables that involve only SM external particles, or into the calculation of loop effects in the extended Higgs sector of the MSSM[^1]. But for the case of direct production of SUSY particles, one also needs a detailed knowledge of the higher-order effects for the processes with these SUSY particles in the external states. A number of studies have already addressed this issue, for production as well as for decay processes. For squark and gluino production in hadron collisions, the NLO QCD corrections are available [@Beenakker:1997ch]; for squark-pair production in $e^+e^-$ collisions, the NLO QCD are also known, together with the Yukawa corrections [@Eberl:1996wa]. Concerning the subsequent squark decays into charginos/neutralinos, the QCD corrections were presented in [@Kraml:1996kz; @Djouadi:1997wt][^2], whereas the Yukawa corrections were given in [@Guasch:1998as]. In this last work large corrections were found. They were derived, however, in the *higgsino* approximation for the chargino; hence, a full computation is required to consolidate the significance of the loop effects.
In this letter we present, for the first time, a complete one-loop computation of the electroweak radiative corrections to the partial decay widths of squarks into quarks and charginos/neutralinos, $$\Gamma(\tilde{q} \to q' \chi)\,\,,
\label{eq:gammadef}$$ with $\tilde{q}={\ensuremath{\tilde b}},\,{\ensuremath{\tilde t}}$, $q=b,\,t$, $\chi={\ensuremath{\chi^0}}_{\alpha},\,\chi^\pm_i$. We present the basic structure of the corrections, and illustrate their main features and their significance in representative numerical examples. The fine details, as well as a comprehensive analysis, will be presented elsewhere [@GHS2]. Although the explicit results, for definiteness, are displayed for the third squark generation, our analytic results are valid for all kinds of squarks.
In processes with exclusively SM particles in external states, it is possible to divide the 1-loop contributions into SM-like and non-SM-like subclasses. This separate treatment is often used in the literature, and it is useful since it allows to make the computation in small steps, checking each sector individually. As a distinctive feature of the radiative corrections to processes with supersymmetric particles in the external legs, this separability is lost. In this kind of processes the ultraviolet (UV) divergences of diagrams with virtual SM particles cancel the UV divergences of diagrams with non-SM particles. Any partial computation would yield UV-divergent and thus meaningless results. For this reason we have to compute all the SM and non-SM loop contributions at once, with the proper counter terms involving the renormalization of almost the full MSSM.
We have used an on-shell renormalization scheme, which can be obtained by extending the on-shell scheme of [@Bohm:1986rj] to the MSSM. For the SM sector, the gauge-boson and fermion masses are treated as input parameters; the electron charge is defined in the Thomson limit, and the weak mixing angle is given in terms of ${\ensuremath{\sin \theta_W}}^2\equiv {s_W}^2\equiv1-{\ensuremath{M^2_W}}/{\ensuremath{M^2_Z}}$. The SM sector and its renormalization is described in [@Bohm:1986rj] and will not be repeated here. On the other hand, different conventions for the SUSY sector exist; therefore, a few comments on our procedure are in order. Concerning the Higgs sector, one mass has to be specified as an input quantity (for which we take the mass of the $H^\pm$), and a definition for [$\tan\beta$]{}, the ratio between the vacuum expectation value of the two Higgs-boson doublets, $v_2/v_1$, is needed. Following [@Dabelstein:1995hb], an indirect definition of [$\tan\beta$]{} is given by the requirement[^3] $\delta v_2/v_2={\delta v_1}/{v_1}$. For the sfermion sector, we use the input and the renormalization conditions as described in [@Guasch:1998as], which fixes the squark masses and the squark-mixing angles $\theta_{{\ensuremath{\tilde q}}}$.
Finally, as a new ingredient, we have to specify the chargino–neutralino sector. The tree-level mass matrices are well known, but we list them in order to settle the conventions: $$\begin{aligned}
{\cal M}&=&\pmatrix{
M&\sqrt{2} M_W \sin{\beta} \cr
\sqrt{2} M_W \cos{\beta}&\mu
}\,\,,\nonumber\\
{\cal M}^0&=&
\pmatrix{
M^\prime&0&M_Z\cos{\beta}{s_W}&-M_Z\sin{\beta}{s_W}\cr
0&M&-M_Z\cos{\beta}{c_W}&M_Z\sin{\beta}{c_W}\cr
M_Z\cos{\beta}{s_W}&-M_Z\cos{\beta}{c_W}&0&-\mu \cr
-M_Z\sin{\beta}{s_W}&M_Z\sin{\beta}{c_W}&-\mu&0
}\,\,\,.
\label{eq:mmassacplus}\label{eq:mmassaneut}\end{aligned}$$ $M$ and $M'$ are the $SU(2)_L$ and $U(1)_Y$ soft-SUSY-breaking gaugino masses. These mass matrices are diagonalized by unitary matrices $U,V,N$ via $$\begin{array}{lcccl}
U^* {\cal M} V^\dagger&=&{\cal M}_D&=&{\rm diag}\left(M_1,M_2\right)\,\,(0<M_1<M_2)\,\,,\\
N^*{\cal M}^0 N^\dagger &=&{\cal M}^0_D&=&
{\rm diag}\left(M_1^0,M_2^0,M_3^0,M_4^0\right)\,\,(0<M_1^0<M_2^0<M_3^0<M_4^0)\,\,.
\end{array}\label{eq:defUVN}$$ This sector contains six particle masses, but only three free parameters are available for an independent renormalization. As a consequence, we are not allowed to impose on-shell conditions for all the particle masses. For the independent input parameters, we choose: the masses of the two charginos and the mass of the lightest neutralino. By introducing counterterms for all the independent parameters in (\[eq:mmassaneut\]), we are able to relate the counterterms of the fundamental parameters to the mass-counterterms $\delta M_i$ of the charginos. Similarly to [@Pierce:1994gj] we find: $$\begin{aligned}
M_1 \,\delta M_1 + M_2 \,\delta M_2 &=& M \,\delta M + \mu \,\delta\mu+ \delta
{\ensuremath{M^2_W}}, \nonumber\\
M_1 M_2
\left(M_1 \,\delta M_2+M_2\,\delta M_1\right)&=&\left(M \mu -{\ensuremath{M^2_W}}\sin(2\beta)\right)\bigg[M \,\delta \mu + \mu \,\delta M\nonumber\\
&&-{\ensuremath{M^2_W}}\,\delta\sin(2\beta)-\sin(2\beta) \,\delta{\ensuremath{M^2_W}}\bigg] .
\label{eq:defcountcha} \end{aligned}$$ The mass counterterms $\delta M_i$ are fixed using the on-shell scheme relation, in the convention of [@Bohm:1986rj] (but with opposite sign for $\Sigma$), $$\delta M_i=-\frac{1}{2}\left(\Sigma_L^i(M_i^2)+\Sigma_R^i(M_i^2)\right)-\Sigma_S^i(M_i^2)\,\,,
\label{eq:defdeltami}$$ where $\Sigma_{\{L,R,S\}}^i$ denote the one-loop unrenormalized left-, right-handed and scalar components of the self-energy for the $i$th-chargino. $\delta M'$ is determined from the lightest neutralino mass, inverting the relation $$\label{eq:defcountneut}
N^*_{1 \alpha} \delta{\cal M}^0_{\alpha\beta} N_{ 1\beta}^{*} = \delta M_1^0\,\,,$$ where the neutralino-mass counterterm $\delta M_1^0$ is fixed by the on-shell condition for $\chi^0_1$, in analogy to (\[eq:defdeltami\]). It is a non-trivial check that with the counterterms determined in eqs. (\[eq:defcountcha\]) and (\[eq:defcountneut\]), the one-loop masses for the residual neutralinos, computed as the pole masses, are UV-finite. The one-loop on-shell neutralino masses read $$M_\alpha^{0 \ {\rm os}}=M_\alpha^0+N^*_{\alpha\beta} \delta{\cal M}^0_{\beta\gamma} N_{\alpha\gamma}^{*}
+\frac{1}{2}\left(\Sigma_L^\alpha(M_\alpha^{02})+\Sigma_L^\alpha(M_\alpha^{02})\right)+\Sigma_S^\alpha(M_\alpha^{02})\,\, ,
\label{eq:defmass1l}$$ where now the parameters of eq. (\[eq:mmassaneut\]) and the masses and mixing matrices computed in (\[eq:defUVN\]) have to be regarded as *renormalized* quantities.
The choice of the lightest neutralino to fix the counterterm $\delta M'$ in (\[eq:defcountneut\]) is only efficient if it has a substantial *bino* component. If $M'\gg(|\mu|,M)$ then $|N_{11}|\ll1$, and the extraction of $\delta M'$ from (\[eq:defcountneut\]) would amplify the radiative corrections artificially. In this case it would be better to extract $\delta M'$ from the $\alpha$th neutralino, such that $|N_{1\alpha}|$ is large. This is, however, not relevant for the scenarios which are discussed in this letter. Notice also that our renormalization procedure makes use of positive-definite mass eigenvalues for charginos and neutralinos, which require the introduction of some purely-imaginary non-zero elements in the $N$-matrix (\[eq:defUVN\]). Had we chosen a real $N$-matrix, with some negative eigenvalues, the various renormalization conditions would be plagued with the explicit sign of the corresponding eigenvalue (see e.g. [@Guasch:1997dk]).
At one-loop, also mixing self-energies between the different neutralinos and charginos are generated, which we write as follows: $$-i \hat\Sigma^{\alpha\beta}(k^2)=-i \left(\hat \Sigma^{\alpha\beta}_L(k^2)
{\rlap/ k} {\ensuremath{P_L}}+ \hat \Sigma^{\alpha\beta}_R(k^2) {\rlap/ k} {\ensuremath{P_R}}+\hat\Sigma^{\alpha\beta}_{SL}(k^2) {\ensuremath{P_L}}+ \hat\Sigma^{\alpha\beta}_{SR}(k^2) {\ensuremath{P_R}}\right)\,\,,\,\,\alpha\neq\beta\,\,,
\label{eq:mixingneut}$$ with $\hat\Sigma$ denoting the renormalized two-point functions, and with the chirality projectors $P_{\{L,R\}}=1\mp\gamma_5$. For the neutralinos, the renormalized self-energies (\[eq:mixingneut\]) are related to the unrenormalized ones according to $$\hat\Sigma^{\alpha\beta}_{\{L,R\}}=\Sigma^{\alpha\beta}_{\{L,R\}}\,\,,\,\,
\hat\Sigma_{SL}^{\alpha\beta}=\Sigma_{SL}^{\alpha\beta}{
-N_{\alpha\gamma}\delta{\cal M}^{0*}_{\gamma\lambda} N_{\beta\lambda}}\,\,,\,\,
\hat\Sigma_{SR}^{\alpha\beta}=\Sigma_{SR}^{\alpha\beta}{
-N^*_{\alpha\gamma} \delta{\cal M}^{0}_{\gamma\lambda} N_{\beta\lambda}^{*}}\,\,,
\label{eq:mixinneutren}$$ and analogous expressions hold for the charginos.
As far as vertex renormalization is concerned, the vertex counterterms are already determined by the renormalization procedure described above. In addition to the parameter renormalization, we have introduced also a field-renormalization constant for each left- and right-handed chargino and neutralino field. As an explicit example, we list the renormalized bottom-sbottom-neutralino vertex. The tree-level interaction Lagrangian reads [@Coarasa:1996qa][^4] $$\begin{array}{l}
{\cal L}_{{\ensuremath{\chi^0}}b\tilde{b}}=
+\displaystyle{\frac{g}{\sqrt{2}}}
\sum_{a=1,2}\sum_{\alpha=1,\ldots ,4}
{\ensuremath{\tilde{b}_a}}^*{\ensuremath{\bar{\Psi}_\alpha^0}}\left({\ensuremath{A_{+a\alpha}^{(b)}}}{\ensuremath{P_L}}+{\ensuremath{A_{-a\alpha}^{(b)}}}{\ensuremath{P_R}}\right)\,b
+\mbox{\rm h.c.}\ ,\\ {\ensuremath{A_{+a\alpha}^{(b)}}}= {\ensuremath{R_{1a}^{(b)}}}\left({\ensuremath{N_{\alpha2}}}^*-\frac{1}{3}{t_W}{\ensuremath{N_{\alpha1}}}^*\right)
-\sqrt{2}{\ensuremath{\lambda_b}}{\ensuremath{R_{2a}^{(b)}}}{\ensuremath{N_{\alpha3}}}^*\, ,\ {\ensuremath{A_{-a\alpha}^{(b)}}}= -\sqrt{2}{\ensuremath{\lambda_b}}{\ensuremath{R_{1a}^{(b)}}}{\ensuremath{N_{\alpha3}}}-\frac{2}{3}{t_W}{\ensuremath{R_{2a}^{(b)}}}{\ensuremath{N_{\alpha1}}}\,\,, \label{eq:bsbntree}
\end{array}$$ where the bottom-quark Yukawa coupling is ${\ensuremath{\lambda_b}}={\ensuremath{m_b}}/(\sqrt{2}{\ensuremath{M_W}}\cos\beta)$. Introducing the one-loop counterterms analogously to [@Guasch:1998as] we obtain the following counterterm Lagrangian [@Guasch:1998as; @Coarasa:1996qa] $$\begin{aligned}
\delta {\cal L}_{{\ensuremath{\chi^0}}_\alpha b{\ensuremath{\tilde b}}_a}&\equiv& \displaystyle{\frac{g}{\sqrt{2}}}
{\ensuremath{\tilde{b}_a}}^*{\ensuremath{\bar{\Psi}_\alpha^0}}\left(\delta{\ensuremath{\Lambda_{+a\alpha}^{(b)}}}{\ensuremath{P_L}}+\delta{\ensuremath{\Lambda_{-a\alpha}^{(b)}}}{\ensuremath{P_R}}\right)\,b
+\mbox{\rm h.c.}\nonumber\\
&=& \frac{1}{2}\left\{
\left[{\frac{\delta\alpha}{\alpha}}+\frac{{c_W}^2}{{s_W}^2} \left({\frac{\delta {\ensuremath{M^2_W}}}{{\ensuremath{M^2_W}}}}-{\frac{\delta {\ensuremath{M^2_Z}}}{{\ensuremath{M^2_Z}}}}\right)\right]
+ {\delta Z}_{a}\right\} {\cal L}_{{\ensuremath{\chi^0}}_\alpha b{\ensuremath{\tilde b}}_a}
+{\cal L}_{{\ensuremath{\chi^0}}_\alpha
b{\ensuremath{\tilde b}}_c}{\delta Z}_{ca}\nonumber\\
&&+\Bigg\{\displaystyle{\frac{g}{\sqrt{2}}}{\ensuremath{\tilde{b}_a}}^*{\ensuremath{\bar{\Psi}_\alpha^0}}\left[{\ensuremath{A_{+a\alpha}^{(b)}}}\frac{1}{2} \left({\delta Z_R^0}{}_\alpha+{\delta Z_L^b}\right) {\ensuremath{P_L}}+{\ensuremath{A_{-a\alpha}^{(b)}}}\frac{1}{2} \left({\delta Z_L^0}{}_\alpha+{\delta Z_R^b}\right){\ensuremath{P_R}}\right]\,b\nonumber\\
&&+\displaystyle{\frac{g}{\sqrt{2}}}{\ensuremath{\tilde{b}_a}}^*{\ensuremath{\bar{\Psi}_\alpha^0}}\left(\delta{\ensuremath{A_{+a\alpha}^{(b)}}}{\ensuremath{P_L}}+\delta {\ensuremath{A_{-a\alpha}^{(b)}}}{\ensuremath{P_R}}\right)\,b
+\mbox{ h.c.}\Bigg\}\,\,,\,\,(c\neq a)\,\,,\nonumber \\
\delta {\ensuremath{A_{+a\alpha}^{(b)}}}&=& -\sqrt{2} \left( {\delta R^{(b)}}_{2a} {\ensuremath{\lambda_b}}N_{\alpha 3}
+ \delta {\ensuremath{\lambda_b}}R^{(b)}_{2a} N_{\alpha 3} \right)+
{\delta R^{(b)}}_{1a} \left(N_{\alpha 2}
-\frac{1}{3} {t_W}N_{\alpha 1} \right) - \frac{1}{3} \delta {t_W}R^{(b)}_{1a}
N_{\alpha 1}
\,\,,\nonumber\\
\delta {\ensuremath{A_{-a\alpha}^{(b)}}}&=&-\sqrt{2} \left({\delta R^{(b)}}_{1a} {\ensuremath{\lambda_b}}N^*_{\alpha3}
+ \delta {\ensuremath{\lambda_b}}R^{(b)}_{1a} N^*_{\alpha3} \right)
-\frac{2}{3} {t_W}N^*_{\alpha1} \left(
{\frac{\delta {t_W}}{{t_W}}}R^{(b)}_{2a}+{\delta R^{(b)}}_{2a}
\right)\,\,,\nonumber \\
\frac{\delta {\ensuremath{\lambda_b}}}{{\ensuremath{\lambda_b}}}&=&{\frac{\delta {\ensuremath{m_b}}}{{\ensuremath{m_b}}}}-\frac{1}{2}
{\frac{\delta {\ensuremath{M^2_W}}}{{\ensuremath{M^2_W}}}}-{\frac{\delta \cos \beta}{\cos\beta}}\,\,,\,\,
\delta{t_W}=\frac{1}{c_W^2}\left(\frac{\delta
M_Z^2}{M_Z^2}-\frac{\delta M_W^2}{M_W^2} \right) ,
\label{eq:counterl}\end{aligned}$$ where $\delta\alpha$, $\delta M_{W,Z}^2$ are the charge and mass counterterms for the MSSM, as given in [@Garcia:1994sb]. The renormalized one-loop part of the amplitude for the decay ${\ensuremath{\tilde b}}_a \rightarrow b \chi^0_\alpha$ can then be written as $$-i T_{a\alpha}^{\rm loop}=-i
\frac{g}{\sqrt{2}}\bar{u}_b \left[C_{+\alpha\beta}{\ensuremath{P_R}}+C_{-\alpha\beta}{\ensuremath{P_L}}\right] v_{{\ensuremath{\chi^0}}_\alpha}\,\,;\,\, C_{\pm\alpha\beta}=\delta{\ensuremath{\Lambda_{\pm a\alpha}^{(b)*}}}+{\ensuremath{\Lambda_{\pm a\alpha}^{(b)}}}{}^{\Sigma}
+{\ensuremath{\Lambda_{\pm a\alpha}^{(b)}}}{}^{\rm 1PI}\,\,.
\label{eq:renormamp}$$ Besides the counterterms $\delta\Lambda$ from eq. (\[eq:counterl\]), it contains the one-loop contribution $\Lambda^{\rm 1PI}$ to the one-particle-irreducible three-point vertex function and the quantity $\Lambda^{\Sigma}$ representing the contribution of the neutralino-mixing self-energies (\[eq:mixingneut\]), $${\ensuremath{\Lambda_{\pm a\alpha}^{(b)}}}{}^{\Sigma}=\sum_\beta {\ensuremath{A_{\pm a\beta}^{(b)*}}}\frac{
M_{\alpha}^0 \hat\Sigma_{SL}^{\beta\alpha}
+M_{\beta}^0 \hat\Sigma_{SR}^{\beta\alpha}
+M_{\alpha}^0 M_{\beta}^0 \hat\Sigma_L^{\beta\alpha}
+M_{\beta}^0{}^2 \hat\Sigma_R^{\beta\alpha}
}{M_{\alpha}^0{}^2-M_{\beta}^0{}^2}\,\, .
\label{eq:mixingeffect}$$ Due to the presence of photon loops, the amplitude (\[eq:renormamp\]) is infra-red divergent; hence, bremsstrahlung of real photons has to be added to cancel this divergence. We therefore include in our results the radiative partial decay width $\Gamma({\ensuremath{\tilde b}}_a \to b{\ensuremath{\chi^0}}_\alpha \gamma)$, including both the soft and the hard photon part. So finally, the complete one-loop electroweak correction is given by $$\begin{array}{rcl}
\delta^{a\alpha}_0&=&
\frac{\Gamma({\ensuremath{\tilde b}}_a\to b{\ensuremath{\chi^0}}_\alpha)}{\Gamma^0({\ensuremath{\tilde b}}_a\to b{\ensuremath{\chi^0}}_\alpha)}-1
= \delta^{a\alpha}_{0 {\rm virt}}+\frac{\Gamma({\ensuremath{\tilde b}}_a \to b{\ensuremath{\chi^0}}_\alpha
\gamma)}{\Gamma^0({\ensuremath{\tilde b}}_a\to b{\ensuremath{\chi^0}}_\alpha)}\,\,,\\[0.5cm]
\delta^{a\alpha}_{0 {\rm virt}}&=&
2{\rm Re}\left( \frac{ ({\ensuremath{m_{\tilde{b}_a}^2}}-M_\alpha^{02}-{\ensuremath{m_b^2}}) ({\ensuremath{A_{+a\alpha}^{(b)}}}{\ensuremath{C_{+a\alpha}^{(b)}}}+ {\ensuremath{A_{-a\alpha}^{(b)}}}{\ensuremath{C_{-a\alpha}^{(b)}}}) - 2 M_\alpha^0 {\ensuremath{m_b}}({\ensuremath{A_{+a\alpha}^{(b)}}}{\ensuremath{C_{-a\alpha}^{(b)}}}+{\ensuremath{A_{-a\alpha}^{(b)}}}{\ensuremath{C_{+a\alpha}^{(b)}}})}{({\ensuremath{m_{\tilde{b}_a}^2}}-M_\alpha^{02}-{\ensuremath{m_b^2}}) (|{\ensuremath{A_{+a\alpha}^{(b)}}}|^2+|{\ensuremath{A_{-a\alpha}^{(b)}}}|^2) -4 M_\alpha^0 {\ensuremath{m_b}}{\rm Re}({\ensuremath{A_{+a\alpha}^{(b)}}}{\ensuremath{A_{-a\alpha}^{(b)*}}})}\right)\,\,,
\label{eq:deltadef}
\end{array}$$ for the neutralino decay channels, and by corresponding expressions $\delta^{ai}_+$ for the chargino channels, as well as for the top-squark decays.
The loop computation itself is rather tedious, since there is a huge number of diagrams to compute. This is better done by means of automatized tools. The computation of the loop diagrams has been done by using the Computer Algebra Systems *FeynArts* and *FormCalc* [@Kublbeck:1990xc; @Hahn:1998yk]. We have produced a set of Computer Algebra programs that compute the one-loop diagrams (and the bremsstrahlung corrections), which are then plugged into a *Fortran* code for the numerical evaluation with the help of the one-loop routines *LoopTools* [@Hahn:1998yk]. A number of checks has been made on the results. The UV and infra-red finiteness of the result, relying on the relations between the different sectors of the model, is a non-trivial check. We also have recovered results already available in the literature; for instance, we used our set of programs to reproduce the strong corrections of [@Djouadi:1997wt], and, using the *higgsino* approximation, we could also reproduce the results of [@Guasch:1998as]. Moreover we also checked that, when using the $\overline{\mbox{MS}}$-scheme, the one-loop corrections to neutralino and chargino masses reproduce those of [@Pierce:1994gj].
Although we consider the chargino and neutralino masses as input parameters, in our numerical study we treat them in a slightly different way. We choose a set of renormalized input parameters $(M,M',\mu)$, and apply (\[eq:mmassaneut\]), (\[eq:defUVN\]) to obtain the one-loop renormalized masses. Of course, if SUSY would be discovered the procedure will be the other way around, that is, the MSSM parameters will be computed from the various observables measured, for example, from the chargino production cross-section and asymmetries at the LC [@Kneur:1999gy]. For a consistent treatment, the one-loop expressions for these observables will have to be used [@Blank:2000uc].
[cc]{}\
\
&\
(b) & (c)
As for the numerical analysis, we use a set of parameters relevant for the next generation of colliders. The squark masses have been chosen in the range of $\sim 300{\mbox{ GeV}}$. If the squarks have a mass around this scale, they will be produced at significant rates not only at the LHC, but also at a LC with $800{\mbox{ GeV}}$ center-of-mass energy. As for the gaugino sector, we make use of the GUT relation between the gaugino soft-SUSY-breaking mass parameters, $M'=5/3 \tan^2\theta_W M$. The input parameters for the SUSY-electroweak sector are chosen to be $${\ensuremath{\tan\beta}}= 4\,\,,\,\,
\mu = -100{\mbox{ GeV}}\,\,,\,\,
M = 150{\mbox{ GeV}}\,\,,\,\,
{\ensuremath{M_{H^\pm}}}=120{\mbox{ GeV}}\,\, .$$ For the quark-squark sector we take $$\begin{aligned}
&&{\ensuremath{m_b}}=5 {\mbox{ GeV}}\,\,,\,\,
m_{{\ensuremath{\tilde b}}_1}=300{\mbox{ GeV}}\,\,,\,\,m_{{\ensuremath{\tilde b}}_2}=m_{{\ensuremath{\tilde b}}_1}+5{\mbox{ GeV}}\,\,,\,\,\theta_{{\ensuremath{\tilde b}}}=0.3\,\,,\nonumber\\
&&{\ensuremath{m_t}}=175 {\mbox{ GeV}}\,\,,\,\, m_{{\ensuremath{\tilde t}}_1}= 300{\mbox{ GeV}}\,\,, \,\,
\theta_{{\ensuremath{\tilde t}}}=0.6\,\,,\end{aligned}$$ the rest of squarks are given a mass of $\sim 1 {\mbox{ TeV}}$ with a mixing angle $\theta_{\tilde q}=\pi/4$. All over our numerical results we apply the (approximate) condition for the non-existence of colour-breaking minima, demanding that the (computed) values of the trilinear soft-SUSY-breaking couplings $A_q$ do not exceed $3 m_{\tilde q}$ [@Frere:1983ag].
In Fig. \[fig:sbottom\] we present the results on the lightest-sbottom decay. Fig. \[fig:sbottom\]a presents the tree-level prediction for the various branching ratios ($BR$) as a function of the lightest-sbottom mass. We see that, aside from the third neutralino channel, all the channels have an appreciable branching ratio whenever they are possible. The opening of the bosonic decay channel (${\ensuremath{\tilde b}}_1\to {\ensuremath{\tilde t}}_1 W^-$) is clearly visible at $m_{{\ensuremath{\tilde b}}_1}=m_{{\ensuremath{\tilde t}}_1}+{\ensuremath{M_W}}=380{\mbox{ GeV}}$. The charged Higgs boson channel (${\ensuremath{\tilde b}}_1\to {\ensuremath{\tilde t}}_1 H^-$) opens at $m_{{\ensuremath{\tilde b}}_1}=m_{{\ensuremath{\tilde t}}_1}+{\ensuremath{M_{H^\pm}}}=420{\mbox{ GeV}}$, but its partial width is much smaller (for the chosen value of the parameters), and hence we can not visualize its effect in Fig. \[fig:sbottom\]a. Whenever the bosonic decay channels are open they amount to a large fraction of the branching ratio [@Bartl:1998xk], and they suffer from large radiative corrections [@Bartl:1997pb]. In Fig. \[fig:sbottom\]b and c we present the one-loop electroweak radiative corrections to the chargino and neutralino channels, respectively. The lightest-neutralino channel is specially important, in that it is always open when we require the lightest neutralino to be the Lightest Supersymmetric Particle. When making a numerical scan over the MSSM parameters the corrections show a rich structure, owing to the large number of thresholds, pseudo-thresholds, etc. For example, the variation of the higgsino-mass parameter $\mu$ does not only change the chargino–neutralino masses and mixing angles, but also the value of the soft-SUSY-breaking trilinear squark couplings, and the mass of the heaviest top-squark. We see in Fig. \[fig:sbottom\]a the opening of the ${\ensuremath{\tilde b}}_1\to t {\ensuremath{\chi^-}}_2$ channel at $m_{{\ensuremath{\tilde b}}_1}=365{\mbox{ GeV}}$. This threshold is accompanied by a corresponding divergence of the one-loop corrections to the lightest-chargino channel ($\delta^{11}_{+}$ in Fig. \[fig:sbottom\]b), and also in the various neutralino channels (Fig. \[fig:sbottom\]c). Of course, the (divergent) corrections near the threshold do not have a physical meaning, since perturbation theory breaks down. In the *light-sbottom* region ($m_{{\ensuremath{\tilde b}}_1}{\mbox{ \raisebox{-4pt}{${\stackrel{\textstyle <}{\sim}}$} }}350 {\mbox{ GeV}}$) the branching ratio is neutralino-dominated. In this region the corrections amount to a 5-10% positive corrections for all of the channels. After the opening of the ${\ensuremath{\chi^-}}_2$ channel the picture changes a little. The third neutralino can get corrections up to 30%, its $BR$, however, is smaller that 1%; these large corrections are thus of little interest. The rest of the channels continue with moderate corrections of the order of 5-10%; a special address to the heaviest-neutralino channel (${\ensuremath{\chi^0}}_4$) with a 15% correction and a non-negligible $BR\simeq1-8\%$ is in order. At this point, however, we do not know yet the net effect on the corrected branching ratios since the electroweak corrections are similar in all decay channels. For quantitative statements also the QCD corrections [@Kraml:1996kz; @Djouadi:1997wt] have to be included.
[cc]{}\
\
&\
(b) & (c)
The corresponding tree-level branching ratios and radiative corrections for the heaviest bottom-squark are displayed in Fig. \[fig:sbottom2\]. They show a similar pattern to that of the lightest sbottom, but with more relevance for the chargino channels. We note that in this case negative radiative corrections are attained for the lightest neutralino (Fig. \[fig:sbottom2\]c). The maximum radiative corrections to the neutralino channel are 15% (for the heaviest neutralino), which has a non-negligible branching ratio all over the allowed sbottom-mass range. Notice that the very different corrections to the chargino channels ($\sim 17\%$ for ${\ensuremath{\chi^-}}_1$ and $\sim 4\%$ for ${\ensuremath{\chi^-}}_2$) at large sbottom masses will translate into corresponding corrections to the branching ratios.
----- -----
(b) (c)
----- -----
We now turn our attention to the lightest-stop decay channels. In Fig. \[fig:stop\] we present the branching ratios and the corrections to all possible decays as a function of the lightest top-squark mass. The chargino decay channels are important whenever one of the charginos is lighter than the top-squark. The top-squark is expected to be one of the lightest squarks of the model (due to large squark mixing), it could be even lighter than the top quark. In this latter case, if the chargino channel is closed, it would decay through Flavour Changing Neutral Currents, ${\ensuremath{\tilde t}}_1\to c{\ensuremath{\chi^0}}_1$ [@Hikasa:1987db]. These two decay channels have been used for a experimental simulation to extract with high precision the top-squark parameters at the LC [@Berggren:1999ss]. For the chosen set of parameters the neutralino channels have never a $BR$ larger that $20\%$. When the lightest top-squark is lighter than $\sim200{\mbox{ GeV}}$ the only possible decay channel has corrections to the width up to $20\%$ (Fig. \[fig:stop\]b). These are, however, of little practical interest, since the $BR$ is 100% in any case. Above this mass, both chargino channels receive corrections in the 5-10% range. The neutralino channels obtain corrections in the 5-15% range, but, as in the case of the sbottom decays, the largest corrections occur in the channels with smallest branching ratios. Again, several threshold structures are visible in the plot, accompanied by the corresponding divergence of the corrections.
In summary, we have presented the set of full electroweak corrections to squark decays into quarks and charginos/neutralinos. These corrections can be sizeable, and therefore they have to be taken into account for the extraction of the MSSM parameters from experiment. A sample of the numerical results has been presented. The final impact on the branching ratios will also depend on the strong corrections [@Kraml:1996kz; @Djouadi:1997wt]. A combination of the two sets of corrections, as well as a comprehensive analysis will be presented elsewhere [@GHS2].
**Acknowledgments:**\
The calculations have been done using the QCM cluster of the DFG Forschergruppe “Quantenfeldtheorie, Computeralgebra und Monte-Carlo Simulation”. We are thankful to T. Hahn for his help regarding the Computer Algebra system, and to A. Vicini and D. St[ö]{}ckinger for useful discussions. The collaboration is part of the network “Physics at Colliders” of the European Union under contract HPRN-CT-2000-00149. The work of J.G. is supported by the European Union under contract No. HPMF-CT-1999-00150. The work of J. S. has been supported in part by CICYT under project No. AEN99-0766.
[10]{}
D. E. Groom [*et al.*]{}, [*Eur. Phys. J.*]{} [**C15**]{} (2000) 1; D. Strom, talk at the *5th International Symposium on Radiative Corrections (RADCOR 2000)*, Carmel, US, September 11-15, 2000; K. Hagiwara, talk at the *5th International Symposium on Radiative Corrections (RADCOR 2000)*, Carmel, US, September 11-15, 2000.
See e.g. M. Carena [*et al.*]{}, *Report of the Tevatron Higgs working group*, [[hep-ph/0010338]{}](http://xxx.lanl.gov/abs/hep-ph/0010338). Atlas Collaboration, *Atlas Technical Design Report* CERN/LHCC/99-14/15; CMS Collaboration, *CMS Technical Proposal* CERN/LHCC/94-38; F. Gianotti, proceedings of the *IVth International Symposium on Radiative Corrections (RADCOR 98)*, p. 270, World Scientific 1999, ed. J. Sol[à]{}.
D. J. Miller, proceedings of the *IVth International Symposium on Radiative Corrections (RADCOR 98)*, p. 289, World Scientific 1999, ed. J. Sol[à]{}, [[hep-ex/9901039]{}](http://xxx.lanl.gov/abs/hep-ex/9901039). H. P. Nilles, [*Phys. Rept.*]{} [**110**]{} (1984) 1; H. E. Haber, G. L. Kane, [*Phys. Rept.*]{} [**117**]{} (1985) 75; A. B. Lahanas, D. V. Nanopoulos, [*Phys. Rept.*]{} [**145**]{} (1987) 1; S. Ferrara, ed., [*Supersymmetry*]{}, vol. 1-2. North Holland/World Scientific, Singapore, 1987.
A. Dobado, M. J. Herrero, S. Pe[ñ]{}aranda, [*Eur. Phys. J.*]{} [**C7**]{} (1999) 313, [[hep-ph/9710313]{}](http://xxx.lanl.gov/abs/hep-ph/9710313); *ibid.* [**C12**]{} (2000) 673, [[hep-ph/9903211]{}](http://xxx.lanl.gov/abs/hep-ph/9903211); *ibid.* [**C17**]{} (2000) 487, [[hep-ph/0002134]{}](http://xxx.lanl.gov/abs/hep-ph/0002134).
M. Carena, M. Quir[ó]{}s, C. E. M. Wagner, [*Nucl. Phys.*]{} [**B461**]{} (1996) 407, [[hep-ph/9508343]{}](http://xxx.lanl.gov/abs/hep-ph/9508343); H. E. Haber, R. Hempfling, A. H. Hoang, [*Z. Phys.*]{} [**C75**]{} (1997) 539, [[hep-ph/9609331]{}](http://xxx.lanl.gov/abs/hep-ph/9609331); S. Heinemeyer, W. Hollik, G. Weiglein, [*Phys. Rev.*]{} [**D58**]{} (1998) 091701, [[ hep-ph/9803277]{}](http://xxx.lanl.gov/abs/hep-ph/9803277); [*Phys. Lett.*]{} [**B440**]{} (1998) 296, [[hep-ph/9807423]{}](http://xxx.lanl.gov/abs/hep-ph/9807423); [*Eur. Phys. J.*]{} [**C9**]{} (1999) 343, [[hep-ph/9812472]{}](http://xxx.lanl.gov/abs/hep-ph/9812472); J. R. Espinosa, R. Zhang, *JHEP* [**0003**]{} (2000) 026, [[hep-ph/9912236]{}](http://xxx.lanl.gov/abs/hep-ph/9912236); *Nucl. Phys. * [**B586**]{} (2000) 3, [[hep-ph/0003246]{}](http://xxx.lanl.gov/abs/hep-ph/0003246); M. Carena *et al.*, [*Nucl. Phys.*]{} [**B580**]{} (2000) 29, [[hep-ph/0001002]{}](http://xxx.lanl.gov/abs/hep-ph/0001002). A. Djouadi, W. Hollik, C. J[ü]{}nger, [*Phys. Rev.*]{} [**D55**]{} (1997) 6975, [[ hep-ph/9609419]{}](http://xxx.lanl.gov/abs/hep-ph/9609419). J. Guasch, W. Hollik, J. Sol[à]{}, preprint KA-TP, UAB-FT in preparation.
J. Sol[à]{}, ed., [*Quantum Effects in the MSSM*]{}, World Scientific, 1998.
W. Beenakker, R. Hopker, M. Spira, P. M. Zerwas, [*Nucl. Phys.*]{} [**B492**]{} (1997) 51, [[hep-ph/9610490]{}](http://xxx.lanl.gov/abs/hep-ph/9610490); W. Beenakker *et al.*, [*Nucl. Phys.*]{} [**B515**]{} (1998) 3, [[hep-ph/9710451]{}](http://xxx.lanl.gov/abs/hep-ph/9710451); T. Plehn, [*Phys. Lett.*]{} [**B488**]{} (2000) 359, [[hep-ph/0006182]{}](http://xxx.lanl.gov/abs/hep-ph/0006182). H. Eberl, A. Bartl, W. Majerotto, [*Nucl. Phys.*]{} [**B472**]{} (1996) 481, [[hep-ph/9603206]{}](http://xxx.lanl.gov/abs/hep-ph/9603206); H. Eberl, S. Kraml, W. Majerotto, [*JHEP*]{} [**05**]{} (1999) 016, [[hep-ph/9903413]{}](http://xxx.lanl.gov/abs/hep-ph/9903413). S. Kraml *et al.*, [*Phys. Lett.*]{} [**B386**]{} (1996) 175, [[hep-ph/9605412]{}](http://xxx.lanl.gov/abs/hep-ph/9605412). W. Beenakker, R. Hopker, T. Plehn, P. M. Zerwas, [*Z. Phys.*]{} [**C75**]{} (1997) 349, [[hep-ph/9610313]{}](http://xxx.lanl.gov/abs/hep-ph/9610313). J. Guasch, W. Hollik, J. Sol[à]{}, [*Phys. Lett.*]{} [**B437**]{} (1998) 88, [[hep-ph/9802329]{}](http://xxx.lanl.gov/abs/hep-ph/9802329).
M. B[ö]{}hm, H. Spiesberger, W. Hollik, [*Fortsch. Phys.*]{} [**34**]{} (1986) 687; W. Hollik, [*Fortschr. Phys.*]{} [**38**]{} (1990) 165. P. Chankowski, S. Pokorski, J. Rosiek, *Nucl. Phys. * [**B423**]{} (1994) 437 [[hep-ph/9303309]{}](http://xxx.lanl.gov/abs/hep-ph/9303309); A. Dabelstein, [*Z. Phys.*]{} [**C67**]{} (1995) 495, [[hep-ph/9409375]{}](http://xxx.lanl.gov/abs/hep-ph/9409375). J. A. Coarasa *et al.*, [*Eur. Phys. J.*]{} [**C2**]{} (1998) 373, [[hep-ph/9607485]{}](http://xxx.lanl.gov/abs/hep-ph/9607485). D. Pierce, A. Papadopoulos, [*Phys. Rev.*]{} [**D50**]{} (1994) 565, [[hep-ph/9312248]{}](http://xxx.lanl.gov/abs/hep-ph/9312248); [*Nucl. Phys.*]{} [**B430**]{} (1994) 278, [[hep-ph/9403240]{}](http://xxx.lanl.gov/abs/hep-ph/9403240). J. Guasch, J. Sol[à]{}, [*Z. Phys.*]{} [**C74**]{} (1997) 337, [[hep-ph/9603441]{}](http://xxx.lanl.gov/abs/hep-ph/9603441). D. Garcia, J. Sol[à]{}, *Mod. Phys. Lett. * [**A9**]{} (1994) 211; P. H. Chankowski *et al.*, *Nucl. Phys. * [**B417**]{} (1994) 101. J. K[ü]{}blbeck, M. B[ö]{}hm, A. Denner, [*Comput. Phys. Commun.*]{} [**60**]{} (1990) 165; T. Hahn, [[hep-ph/0012260]{}](http://xxx.lanl.gov/abs/hep-ph/0012260). T. Hahn, M. P[é]{}rez-Victoria, [*Comput. Phys. Commun.*]{} [**118**]{} (1999) 153, [[hep-ph/9807565]{}](http://xxx.lanl.gov/abs/hep-ph/9807565); T. Hahn, *FeynArts*, *FormCalc* and *LoopTools* user’s guides, available from [`http://www.feynarts.de`](http://www.feynarts.de).
J. L. Kneur, G. Moultaka, *Phys. Rev. * [**D59**]{} (1999) 015005, [[hep-ph/9807336]{}](http://xxx.lanl.gov/abs/hep-ph/9807336); *Phys. Rev. * [**D61**]{} (2000) 095003, [[hep-ph/9907360]{}](http://xxx.lanl.gov/abs/hep-ph/9907360); S. Y. Choi *et al.*, [*Eur. Phys. J.*]{} [**C14**]{} (2000) 535, [[hep-ph/0002033]{}](http://xxx.lanl.gov/abs/hep-ph/0002033). M. A. Diaz, S. F. King, D. A. Ross, *Nucl. Phys. *[**B529**]{} (1998) 23 [[hep-ph/9711307]{}](http://xxx.lanl.gov/abs/hep-ph/9711307); S. Kiyoura, M. M. Nojiri, D. M. Pierce, Y. Yamada, *Phys. Rev. * **D58** (1998) 075002 [[hep-ph/9803210]{}](http://xxx.lanl.gov/abs/hep-ph/9803210); T. Blank, W. Hollik, [[hep-ph/0011092]{}](http://xxx.lanl.gov/abs/hep-ph/0011092).
J. M. Fr[é]{}re, D. R. T. Jones, S. Raby, [*Nucl. Phys.*]{} [**B222**]{} (1983) 11; M. Claudson, L. J. Hall, I. Hinchliffe, [*Nucl. Phys.*]{} [**B228**]{} (1983) 501; C. Kounnas, A. B. Lahanas, D. V. Nanopoulos, M. Quir[ó]{}s, [*Nucl. Phys.*]{} [**B236**]{} (1984) 438; J. F. Gunion, H. E. Haber, M. Sher, [*Nucl. Phys.*]{} [**B306**]{} (1988) 1.
A. Bartl *et al.*, [*Phys. Lett.*]{} [**B435**]{} (1998) 118, [[hep-ph/9804265]{}](http://xxx.lanl.gov/abs/hep-ph/9804265). A. Bartl *et al.*, [*Phys. Lett.*]{} [**B419**]{} (1998) 243, [[hep-ph/9710286]{}](http://xxx.lanl.gov/abs/hep-ph/9710286); [*Phys. Rev.*]{} [**D59**]{} (1999) 115007; H. Eberl *et al.*, [*Phys. Rev.*]{} [**D62**]{} (2000) 055006, [[hep-ph/9912463]{}](http://xxx.lanl.gov/abs/hep-ph/9912463). K. Hikasa, M. Kobayashi, [*Phys. Rev.*]{} [**D36**]{} (1987) 724. M. Berggren, R. Ker[ä]{}nen, H. Nowak, A. Sopczak, proceedings of *International Workshop on Linear Colliders (LCWS 99)*, Sitges, Barcelona, Spain, 28 Apr - 5 May 1999, Universitat Aut[ò]{}noma de Barcelona 2000, eds. E. Fern[á]{}ndez, A. Pacheco, [[hep-ph/9911345]{}](http://xxx.lanl.gov/abs/hep-ph/9911345).
[^1]: See e.g. [@IWQEMSSM] and references therein.
[^2]: The gluino decay channel, which can be overwhelming for $m_{\tilde
q}>m_q+m_{\tilde g}$, was studied in [@Beenakker:1996de]. Here we will assume $m_{\tilde g}>m_{\{\tilde t_a,\tilde b_a\}}$.
[^3]: For possible other definitions see e.g. [@Coarasa:1996qa].
[^4]: Note that our convention for the neutralino mass-matrices (\[eq:mmassaneut\]) is different from that of [@Coarasa:1996qa].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Pulsar spin frequencies and their time evolution are an important source of information on compact stars and their internal composition. Oscillations of the star can reduce the rotational energy via the emission of gravitational waves. In particular unstable oscillation modes, like r-modes, are relevant since their amplitude becomes large and can lead to a fast spin-down of young stars if they are saturated by a non-linear saturation mechanism. We present a novel mechanism based on the pronounced large-amplitude enhancement of the bulk viscosity of dense matter. We show that the enhanced damping due to non-linear bulk viscosity can saturate r-modes of neutron stars at amplitudes appropriate for an efficient spin-down.'
author:
- Mark Alford
- Simin Mahmoodifar
- Kai Schwenzer
bibliography:
- 'cs.bib'
title: 'Non-linear viscous saturation of r-modes'
---
[ address=[Department of Physics, Washington University in St. Louis, Missouri, 63130, USA]{} ]{}
[ address=[Department of Physics, Washington University in St. Louis, Missouri, 63130, USA]{} ]{}
[ address=[Department of Physics, Washington University in St. Louis, Missouri, 63130, USA]{} ]{}
Compact stars are so dense that they could contain deconfined quark matter. To reveal their decomposition requires to connect microscopic properties to macroscopic observables. R-modes are oscillations modes of rotating stars that are unstable under the emission of gravitational waves [@Papaloizou:1978zz; @Andersson:1997xt; @Andersson:2000mf; @Lindblom:1999yk]. In case of neutron stars they could spin-down young stars within a short time interval or limit the rotational rate of old stars that are spun up by accretion. However, since viscous damping cannot stop low amplitude r-modes in certain instability regions at high frequency they have to be saturated by some non-linear damping mechanism. We propose the strong suprathermal enhancement of the bulk viscosity at large amplitudes [@Madsen:1992sx; @Alford:2010gw] as a viable saturation mechanism and in contrast to previous studies [@Reisenegger:2003pd] find that it can stop the growth of r-modes of neutron stars at amplitudes that are large enough for an efficient spin-down but small enough that viscous damping could dominate competing saturation mechanisms [@Bondarescu:2008qx; @Lin:2004wx].
The bulk viscosity of dense matter provides a measure for the energy dissipation in a compression and rarefaction cycle. It is maximal when the external oscillation frequency matches the time scale of the microscopic interactions that cause the dissipation, whereby the dominant interactions in the case of star oscillations are slow weak processes. The external density fluctuation $\Delta n$ induces a corresponding oscillation of the difference of chemical potentials $\mu_\Delta$ which would vanish in weak equilibrium. Similarly, the rate of weak interactions vanishes generally for fully degenerate matter in equilibrium and becomes finite either due to finite temperature effects or due to such deviations of the chemical potentials from their equilibrium value. The net rate for the considered weak process $\Gamma^{(\leftrightarrow)}$ takes the general form $$\Gamma^{(\leftrightarrow)} = -\tilde{\Gamma} T^{2N}\mu_\Delta \left(1+\sum_{j=1}^{N}\chi_j\left(\frac{\mu_\Delta^{2}}{T^{2}}\right)^{j}\right)\,.
\label{eq:gamma-parametrization}$$ where the coefficients $\chi_j$ of the terms that are non-linear in the oscillating chemical potential difference $\mu_\Delta$ and their number $N$ are determined by the respective weak process. In this work we study neutron matter with an APR equation of state [@Akmal:1998cf] at densities that are not high enough to allow direct Urca reactions, so that only the modified Urca process involving a bystander nucleon $n+n \rightarrow n+p+e^- +\bar\nu_e\, , \,\cdots $ is kinematically allowed. For this process $N=3$, corresponding to a strong non-linear dependence on $T$ and $\mu_\Delta=\mu_n-\mu_p-\mu_e$.
The bulk viscosity features three distinct characteristic regions [@Madsen:1992sx; @Alford:2010gw]. In the [*subthermal*]{} regime at low amplitudes, where $\mu_\Delta \ll T$, $\mu_\Delta$ is linear in $\Delta n$ and the viscosity is independent of the amplitude. The [*suprathermal*]{} regime $\mu_\Delta>T$ is divided into a part where $\mu_\Delta$ is still linear in $\Delta n$ but the viscosity strongly rises and another where the rise of $\mu_\Delta$ becomes weaker due to non-linear saturation effects and the viscosity decreases again. Previously mainly the subthermal limit $\mu_\Delta \ll T$ has been studied, but since the r-mode rises exponentially it eventually reaches the suprathermal regime that will be studied in the following.
![\[fig:all-viscosity\] [*Left panel:*]{} The amplitude of the chemical potential oscillation $\mu_\Delta/T$ as a function of the driving density oscillation amplitude $\Delta n/\bar{n}$ for different forms of dense matter. [*Right panel:*]{} Comparison of the bulk viscosity of the different forms of matter as a function of the density oscillation amplitude. In both plots, the frequency is $\omega=8.4$ kHz, corresponding to an r-mode in a millisecond pulsar, and characteristic values $T=10^9$ K of the temperature and $\bar n = 2n_0$ for the density are chosen. The solid curves show the result for interacting matter, the dashed curved for non-interacting matter, whereas the dotted lines denote the analytic approximation eq. (\[eq:madsen-approximation\]); for more details see [@Alford:2010gw].](chem-pot-amplitude)
![\[fig:all-viscosity\] [*Left panel:*]{} The amplitude of the chemical potential oscillation $\mu_\Delta/T$ as a function of the driving density oscillation amplitude $\Delta n/\bar{n}$ for different forms of dense matter. [*Right panel:*]{} Comparison of the bulk viscosity of the different forms of matter as a function of the density oscillation amplitude. In both plots, the frequency is $\omega=8.4$ kHz, corresponding to an r-mode in a millisecond pulsar, and characteristic values $T=10^9$ K of the temperature and $\bar n = 2n_0$ for the density are chosen. The solid curves show the result for interacting matter, the dashed curved for non-interacting matter, whereas the dotted lines denote the analytic approximation eq. (\[eq:madsen-approximation\]); for more details see [@Alford:2010gw].](viscosity-all-10E9nt)
At sufficiently low temperature and large frequency the bulk viscosity allows a general analytic approximation that covers both the subthermal and the part of the suprathermal regime where $\mu_\Delta$ is linear in $\Delta n/\bar{n}$ [@Madsen:1992sx; @Alford:2010gw; @Reisenegger:2003pd] $$\label{eq:madsen-approximation}
\zeta^\sim = \frac{C^{2}\tilde{\Gamma}T^{2N}}{\omega^{2}}\left(1+\sum_{j=1}^{N}\frac{\left(2j+1\right)!! \chi_{j}}{2^{j}\left(j+1\right)!} \left(\frac{C}{T} \frac{\Delta n}{\bar{n}}\right)^{2j}\right)$$ where $C$ is a susceptibility describing the strongly interacting state of matter which in the case of neutron star matter takes the form $$C\equiv\bar{n}\left.\frac{\partial\mu_\Delta}{\partial n}\right|_x = 4\!\left(1\!-\!2x\right)\!\left(\! n\!\frac{\partial S}{\partial n}\!-\!\frac{S}{3}\!\right)$$ in terms of the symmetry energy $S$, the baryon density $n$ and the proton fraction $x$.
The validity of the approximation eq. (\[eq:madsen-approximation\]) can be seen from fig. \[fig:all-viscosity\], where on the left panel the amplitude of the chemical potential oscillation $\mu_\Delta/T$ is plotted against the corresponding density amplitude $\Delta n/\bar{n}$. For the different forms of dense matter shown there the suprathermal regime is reached for amplitudes $\Delta n/\bar{n}=O(0.01)$. In particular, in the case of hadronic matter with modified Urca interactions the linear regime, where $\mu_\Delta\sim\Delta n/\bar{n}$, extends at large frequencies basically over the entire physical range of density amplitudes $\Delta n/\bar{n}<1$. Correspondingly, the approximation eq. (\[fig:all-viscosity\]) shown by the dotted curve on the right panel of fig. \[fig:all-viscosity\] reproduces the full result shown by the solid curve favorably in this case.
![\[fig:saturation-amplitude\] [*Left panel:*]{} Time scales of the fundamental $m=2$ r-mode of a $1.4\,M_\odot$ neutron star as a function of temperature. The dotted curve represents the gravitational time scale, the dashed curve the shear viscosity damping time scale and the solid curves the bulk viscosity time scales at various amplitudes. [*Right panel:*]{} Temperature dependence of the “static” saturation amplitude. The solid, dashed, and dotted curve are given for the maximum Kepler frequency $\Omega_K\approx 6$ kHz and lower values.](damping-times-APR.pdf)
![\[fig:saturation-amplitude\] [*Left panel:*]{} Time scales of the fundamental $m=2$ r-mode of a $1.4\,M_\odot$ neutron star as a function of temperature. The dotted curve represents the gravitational time scale, the dashed curve the shear viscosity damping time scale and the solid curves the bulk viscosity time scales at various amplitudes. [*Right panel:*]{} Temperature dependence of the “static” saturation amplitude. The solid, dashed, and dotted curve are given for the maximum Kepler frequency $\Omega_K\approx 6$ kHz and lower values.](hadronic-sat-amp.pdf)
We consider in the following an exemplary $1.4\,M_\odot$ neutron star model obtained as a solution of the TOV equations [@Tolman:1939jz] using the APR equation of state [@Akmal:1998cf]. The analysis of the bulk viscosity damping time requires in principle an expression for the r-mode to next to leading order in a slow rotation expansion [@Lindblom:1999yk]. Following [@Lindblom:1998wf; @Jaikumar:2008kh] we restrict ourselves to the leading order expression for the fundamental $m=2$ r-mode with density fluctuation $$\label{eq:r-mode-profile}
\left| \frac{\Delta n}{n} \right|\approx\sqrt{\frac{4\pi5!}{7}}\alpha R^{2}\Omega^{2} \frac{\partial\rho}{\partial p}\left(\left(\frac{r}{R}\right)^{3}+\delta\Phi_{0}\right)Y_{3}^{2}\left(\theta,\phi\right)$$ where $\alpha$ is a dimensionless amplitude parameter and $\delta\Phi_0$ denotes the deviation of the gravitational potential from the equlilibrium. Eq. (\[eq:r-mode-profile\]) has a very strong radial dependence both due to the explicit cubic factor as well due to the pronounced r-dependence of the pressure derivative of the energy density $\partial\rho/\partial p$ in a neutron star. The characteristic time scales of the gravitational instability $\tau_G$ and of the damping by shear and bulk viscosity $\tau_S$ and $\tau_B$ are defined via their energy dissipation $$\frac{1}{\tau_{i}}\equiv-\frac{1}{2E}\left(\frac{dE}{dt}\right)_{i}$$ The stability criterium for the r-mode is $1/\tau_G+1/\tau_S+1/\tau_V\geq0$ where the equality defines the boundary of the instability region. At small amplitudes in the subthermal regime of the bulk viscosity there is at high frequency an instability region where the r-mode is unstable and grows exponentially [@Andersson:2000mf; @Lindblom:1998wf; @Jaikumar:2008kh]. In the general case the damping time due to bulk viscosity reads$$\tau_B=\frac{2^{3}}{3^{5}5!\pi\alpha^{2}\tilde{J} MR^{2}}\int d^3x \left|\frac{\Delta n}{\bar{n}}\right|^{2}\zeta\left(\left|\frac{\Delta n}{\bar{n}}\right|^{2}\right)$$ where $\tilde J$ is a constant and the viscosity eq. (\[eq:madsen-approximation\]) depends on the density fluctuation induced by the r-mode. In the integral we neglect the contribution of the crust of the star since the bulk viscosity has not been computed there, yet.
The temperature dependence of the damping times is shown in fig. \[fig:saturation-amplitude\] where the bulk viscosity time scale is given at various amplitudes. As can be seen due to the strong increase of the bulk viscosity in the suprathermal regime its damping time decreases strongly with amplitude and eventually undercuts the gravitational time scale. Therefore the enhanced damping can effectively stop the r-mode growth and saturates the amplitude at finite values that are at fixed temperature and frequency determined by the above stability criterium. The result for this “static” saturation amplitude as a function of temperature is shown in fig. \[fig:saturation-amplitude\] for different frequencies. It features a plateau for temperatures inside the instability region and decreases with frequency. The size of the saturation amplitudes should lead to a fast spin down of young stars [@Lindblom:1998wf]. Actually, during the r-mode growth the star cools and starts spinning down so that the above amplitudes do not have to be reached. Due to the continuous increase at the right boundary of the instability region the r-mode should automatically saturate at an amplitude that is sufficient for spinning down the star and that could be lower than those of competing mechanisms [@Bondarescu:2008qx; @Lin:2004wx].
We thank Nils Andersson and Andreas Reisenegger for helpful discussions. This research was supported in part by the Offices of Nuclear Physics and High Energy Physics of the U.S. Department of Energy under contracts \#DE-FG02-91ER40628, \#DE-FG02-05ER41375.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Adam Kardos and Paolo Nason\
INFN, Sezione di Milano Bicocca, Piazza della Scienza 3, 20126 Milan, Italy\
E-mail:
- |
Carlo Oleari\
Università di Milano-Bicocca and INFN, Sezione di Milano-Bicocca\
Piazza della Scienza 3, 20126 Milan, Italy\
E-mail:
title: '[Three-jet production in POWHEG]{}'
---
Introduction
============
The production of hadronic jets is an ever present phenomenon in hadronic collisions. Jets are the manifestation of the production of coloured particles with large transverse momentum, and in hadronic collisions this phenomenon is very frequent, due to the relatively large size of the strong coupling constant, and to the presence of coloured incoming partons.
Electroweak processes with associated production of QCD jets are an ever present background both to Standard Model studies and to searches of new physics. It is therefore mandatory to understand these phenomena to our best. For this reason, basic QCD jet-production processes can constitute a framework where we can test our ability to simulate jet phenomena.
We stress that QCD jet production is more difficult to understand and simulate with respect to associated jet-production phenomena. In fact, in the latter case, our initial process already involves the production of a massive object, with a relatively well defined mass. This sets the relevant scale and momentum fractions for the parton distribution functions (pdfs), and the associated jet production probes the values of pdfs around this point. On the other hand, in the basic QCD jet-production processes, these scales and momentum fractions are instead determined by the jet system, that is not as well known. Small errors in the determination of the jet energy induce larger uncertainties due to the steep fall of the luminosity as a function of the mass of the produced system. Furthermore, at relatively low total transverse energies, we are approaching the high-energy regime, that is not usually dealt rigorously by current Monte Carlo implementations.[^1] Thus, a reasonable understanding of basic QCD jet simulation can increase our confidence that we can also model associated jet-production phenomena in a reliable way.
A NLO-accurate generator for dijet production that can be interfaced to parton-shower generators (i.e. a NLO+PS generator), using the [[POWHEG]{}]{} method [@Nason:2004rx; @Frixione:2007vw], was implemented in ref. [@Alioli:2010xa] (the [[dijet]{}]{} generator from now on), within the [[POWHEG BOX]{}]{} framework [@Alioli:2010xd].
In the present work, we implement a NLO+PS generator, built in the [[POWHEG BOX]{}]{} framework, for the production of three jets (the [[trijet]{}]{} generator from now on). Basically, we include the $2\to 3$ parton scattering processes and all the QCD corrections to them, that include, besides the virtual corrections, all the $2\to 4$ parton scattering processes at leading order. We neglect parton masses throughout. The [[trijet]{}]{} implementation is carried out within the [[POWHEG BOX V2]{}]{} framework.[^2]
The NLO virtual matrix elements for three-jet production were computed for the first time in refs. [@Bern:1993mq; @Bern:1994fz; @Kunszt:1994nq]. Compact expressions for the real contributions are also available from refs. [@Gunion:1985bp; @Gunion:1986zh; @Gunion:1986zg; @Kuijf:1991kn]. We have used these results as coded in the [[c++]{}]{} program [[NLOJET++]{}]{} [@Nagy:2003tz]. The other missing ingredients, needed to set up a [[POWHEG BOX]{}]{} generator, are the colour- and spin-correlated Born amplitudes. These are easily obtained using the [[[MadGraph4]{}]{}]{} [@Alwall:2007st] interface to the [[POWHEG BOX]{}]{} developed in ref. [@Campbell:2012am]. The Born phase space, due to the complex singularity structure of the Born amplitude, has been built using a multi-channel technique.
The Born process in the [[trijet]{}]{} generator has several singular kinematic regions, associated to pairs of final-state partons becoming collinear, or one parton acquiring small transverse momentum. A further, overall, singular configuration is the one where all partons have small transverse momentum. We find that the [[MiNLO]{}]{} [@Hamilton:2012np] procedure for setting the scales and assigning Sudakov form factors is particularly helpful here, since it tames the divergences in all kinematic regions but the overall one. As we will discuss in the following, we also find that, when using [[MiNLO]{}]{}, two-jet inclusive observables are fairly well described, so that we do not need to worry about the impact of configurations close to the limit where one jet becomes unresolved, and furthermore we only need to deal with the problems of the overall singularity of the Born-level cross section, since the others are regulated by [[MiNLO]{}]{}.
The paper is organized as follows. In section \[sec:technical\] we describe the construction of the Born phase space, and the multi-channel technique that we have used in order to probe adequately all the singular regions, and other technical details about the [[trijet]{}]{} implementation. In section \[sec:checks\] we discuss the checks that we have carried out in order to validate our cross-section formulae. In section \[sec:comparisons\] we compare the output of our generator at different levels. After the discussion of the common settings for the comparisons in section \[sec:settings\], we compare among each other the NLO, the Les Houches Event (LHE) level and the shower results in section \[sec:NLOvsLH\]. The LHE level is the stage where [[POWHEG]{}]{} has already generated the hardest radiation, but no other radiation has been added by the subsequent shower programs. The purpose of this comparison is to determine how the final result is built up. In section \[sec:NLOvsNLOMiNLO\] we compare the NLO and the [[MiNLO]{}]{}-improved NLO results, in order to show at what level the [[MiNLO]{}]{} procedure differs from the NLO results obtained with a standard scale choice. In section \[sec:NLOvsLHMiNLO\] we compare among each other the NLO, LHE and shower results, when [[MiNLO]{}]{} is turned on. Since [[MiNLO]{}]{} regulates the divergences related to the third jet becoming soft or collinear, but not those related to the whole event having small total transverse energy, we discuss how to enforce some physicality requests on the small transverse-energy region in section \[sec:smallHT\]. In section \[sec:trijet-dijet-comparison\], we compare the [[MiNLO]{}]{} [[trijet]{}]{} results with the [[dijet]{}]{} ones, when considering quantities inclusive in the third jet. We give our conclusions in section \[sec:conclusions\]. A short discussion on the choice and setting of the dynamical scales in the [[POWHEG BOX]{}]{} is presented in appendix \[app:btlscalereal\].
Technical details {#sec:technical}
=================
In this section we discuss a few technical details of the [[trijet]{}]{} implementation: the multi-channel Born phase space, the generation cuts and the production of weighted events, with and without [[MiNLO]{}]{}.
Phase-space generation with multi-channel technique {#sec:multichannel}
---------------------------------------------------
The Born cross section for the production of three partons has several sharp peaks, in correspondence to the singular regions where soft and collinear singularities are approached. A standard way to integrate a many-peak function is by using a multi-channel technique. In the following we illustrate how we implemented it on trijet production.
We label the particles as follows: with 1 and 2 we indicate the two incoming partons, using the label 0 when we refer to both, and with 3, 4, 5 the final-state ones. Momentum conservation in the center-of-mass frame at the Born level is then given by $$p_1 + p_2 = p_3 + p_4 + p_5\,.$$ The Born cross section has 9 singular regions, according to the 9 possible choices of emitter-emitted couples. We label them with two indexes: the first index identifies the emitting particle and the second the emitted one:
- 6 final-state regions: $\{35,53,45,54,34,43\}$, where, for example, 35 is the singular region associated with parton 5 being emitted by parton 3.
- 3 initial-state regions: $\{03,04,05\}$, where we treat as one the singular region associated with either of the incoming partons.
In order to perform an efficient importance sampling in each singular region, we introduce a function of the kinematic variables that approaches 1 only in one singular region and goes to zero fast enough in all the others. To do so, we define the following quantities $$S_{ij} = \frac{1}{d_{ij}}\,,$$ where, for final-state partons, $$d_{ij}= 2\, p_i\cdot p_j \,\frac{E_i E_j}{E_i^2+E_j^2}= 2\,\frac{E_i^2
E_j^2}{E_i^2+E_j^2}\( 1-\cos\theta_{ij}\) \,, \qquad i,j\ge 3\,,$$ with $E_i=p_i^0$, and $\theta_{ij}$ the angle between parton $i$ and parton $j$, in the center-of-mass frame, and $$d_{0j} = E_j^2 \(1-y^2\), \qquad y= 1- \frac{p_1\cdot p_j}{E_1 E_j}=\cos
\theta_{1j}\,,
\qquad j\ge 3\,,$$ where $\theta_{1j}$ is the angle between the direction of the first incoming beam and the outgoing parton $j$. It is clear from their definition that when two final-state partons become collinear or when one parton becomes soft, the corresponding $S_{ij}$ diverges. The same can be said about $S_{0j}$ when the $j$th parton becomes soft or collinear with respect to the incoming beam.
Defining $$S= S_{03}+S_{04}+S_{05}+S_{35} +S_{53}+S_{45}+S_{54}+S_{34}+S_{43}\,,$$ we can then write the following identity $$\begin{aligned}
\label{eq:ident}
1 &=& \frac{S_{03}}{S} + \frac{S_{04}}{S} + \frac{S_{05}}{S}
+\frac{S_{35}}{S} \,\frac{E_5}{E_3+E_5}+\frac{S_{53}}{S} \,\frac{E_3}{E_3+E_5}+
\frac{S_{45}}{S} \,\frac{E_5}{E_4+E_5}+\frac{S_{54}}{S} \,\frac{E_4}{E_4+E_5}
\nonumber\\
&&{}+\frac{S_{34}}{S} \,\frac{E_4}{E_3+E_4} +\frac{S_{43}}{S} \,\frac{E_3}{E_3+E_4}\,,
\\[2mm]
\label{eq:ident_short}
&\equiv& \sum_j \tilde{S}_{0j} + \sum_{ij} \tilde{S}_{ij}\,, \end{aligned}$$ where we have introduced a self-explanatory notation in eq. (\[eq:ident\_short\]). Each term on the right-hand-side of eqs. (\[eq:ident\]) or (\[eq:ident\_short\]) approaches 1 only in one particular singular region. For example, the terms $\tilde{S}_{0j}$ approach 1 when the $j$th parton is either soft or collinear to any of the two incoming beams, and go to 0 when other singular regions are approached. Similarly, terms of the form $\tilde{S}_{ij}$ approach 1 when the $j$th parton is either soft or collinear to the final-state parton $i$, and go to 0 when other singular regions are approached.
We insert then 1 written as eq. (\[eq:ident\_short\]) in the formula of the invariant phase-space element $${\mathrm{d}}\Phi_B = {\mathrm{d}}\Phi_B \lg \sum_j \tilde{S}_{0j} + \sum_{ij}
\tilde{S}_{ij} \rg = \sum_j \tilde{S}_{0j} \, {\mathrm{d}}\Phi_B + \sum_{ij}
\tilde{S}_{ij} \, {\mathrm{d}}\Phi_B \,.$$ We can now choose the best parametrization of the kinematic variables (i.e. the momenta $p_i$) in terms of the Monte Carlo integration variables, in order to do an importance sampling for each of the terms of the sum. Each ${\mathrm{d}}\Phi_B$ in the sum has then a different parametrization in terms of the Monte Carlo integration variables, so that each of them can be seen to depend on the summation indexes. We will indicate this by adding the subscript $ij$ to each phase-space element volume $$\label{eq:multichannel}
{\mathrm{d}}\Phi_B = \sum_{kj} \tilde{S}_{kj} \, \({\mathrm{d}}\Phi_B\)_{kj}\,.$$ In the [[trijet]{}]{} generator, each phase-space volume $\({\mathrm{d}}\Phi_B\)_{kj}$ is computed using a replica of the automatic machinery (see refs. [@Frixione:2007vw; @Alioli:2010xd]) for the generation of the real phase space, starting from the Born phase space for dijet production, i.e. starting from the $2\to 2$ phase space and attaching an extra parton, with different importance sampling according to the singular region where it has been adapted to.
When the subroutine for the computation of the Born phase space is invoked, one extra random number is used to choose, with equal probability, one of the 9 different parametrization of eq. (\[eq:multichannel\]). The returned Jacobian is then the product of $\({\mathrm{d}}\Phi_B\)_{kj}$ and of the corresponding suppression function $\tilde{S}_{kj}$, multiplied by 9, in order to compensate for the 1/9 factor introduced by choosing to evaluate randomly only one single term of the sum.
The real phase space for trijet production is then built as usual by the [[POWHEG BOX]{}]{} automatic machinery on top of the Born kinematics.
Generation cuts or weighted events
----------------------------------
As already stated, the Born cross section for three-jet production has several singular regions, associated to a pair of final-state partons becoming collinear among each other, or to a final-state parton becoming collinear to an initial-state parton, or becoming soft. Because of these singularities, an unweighted generator would end up generating all events in these singular regions. This problem is usually handled by requiring that the final-state partons satisfy some generation cuts, such as to avoid the singular configurations. In this case, of course, one should make sure that the contributions arising from the neglected regions of phase space do not end up affecting observables of interest. In the case at hand, one may require that the three partons form well separated jets. If final-state observables do require at least three jets, it is unlikely that the neglected regions would contribute to them. However, one should always check that the results are independent upon these cuts.
An alternative method is to generate weighted events. One chooses a weight function that diverges when approaching the singular regions. This is done as follows. One introduces a function of the Born phase-space kinematics, $F(\Phi_B)$, that vanishes in the singular regions, such that the following integral of the differential cross section $\sigma$ $$\int \frac{{\mathrm{d}}\sigma}{{\mathrm{d}}\Phi_B} F(\Phi_B) \;{\mathrm{d}}\Phi_B$$ is finite. One then generates the phase-space points with a probability proportional to $({\mathrm{d}}\sigma/{\mathrm{d}}\Phi_B) F(\Phi_B)$, assigning to each point a weight $1/F(\Phi_B)$. In this way one should not worry about the independence of the result upon generation cuts. Observables that do depend upon the singular regions will typically receive rare contributions with large weights, yielding large errors. This method was used in the [[POWHEG BOX]{}]{} since ref. [@Alioli:2010qp], where the function $F$ was dubbed “Born suppression factor”. In the [[POWHEG BOX]{}]{} framework the factor $F$ was applied to all cross-section contributions (i.e. not only the Born term), and, in the case of real terms and collinear remnants, it was computed as a function of the underlying-Born kinematics. Here we stress that, in spite of the name, no phase-space regions are really suppressed. In fact, the effect of the $F$ factor at the generation level is exactly compensated by the $1/F$ weight of the event.
In the present case, we have considered the following $F$ function $$\begin{aligned}
\label{eq:F}
F&=&F_1\times F_2\,, \\
F_1 &=& \exp\left[-S_1^p \times \left(\frac{1}{q_1^p}+\frac{1}{q_2^p}+\frac{1}{q_3^p}+
\frac{1}{q_{12}^p}+\frac{1}{q_{23}^p}+\frac{1}{q_{13}^p}\right)\right], \\
\label{eq:F2}
F_2 &=& \frac{\frac{1}{S_2^p}}{\left(\frac{1}{S_2}+\frac{1}{{H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}^2}\right)^p}\,,\end{aligned}$$ where $S_1$ and $S_2$ are suitably chosen scales, $q_1$, $q_2$ and $q_3$ are the square of the transverse momenta of the three final-state particles with respect to the beam axis, and $q_{ij}$ is the relative transverse momentum squared of particle $i$ and $j$, defined in the partonic center-of-mass frame as $$q_{ij}=p_i\cdot p_j \, \frac{E_i E_j}{E_i^2+E_j^2}\,.$$ Furthermore $${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}= \sqrt{q_1}+\sqrt{q_2}+\sqrt{q_3}\,.$$ The role of $F_2$ is to handle the singular region associated to all partons having small transverse momentum. It also plays the role of increasing the importance sampling in the region of large transverse-momentum jets, a feature that is needed in order to properly cover a large range of transverse energy.
MiNLO
-----
If we apply the [[MiNLO]{}]{} procedure to the [[trijet]{}]{} generator, the factor $F_1$ discussed above is no longer needed. All singular regions, except for the overall one, are regulated by the [[MiNLO]{}]{} Sudakov form factors. The [[MiNLO]{}]{} form factor is exactly as described in ref. [@Hamilton:2012np], with the only freedom of choosing the scale of the basic process,[^3] that in the case of ref. [@Hamilton:2012np] (dealing with Higgs boson production in association with jets) was taken equal to the Higgs boson virtuality. In the present case, we have chosen the scale of the basic process to be equal to the sum of the transverse momenta of the two final pseudoparticles (after the [[MiNLO]{}]{} clustering has taken place).
Checks of the code {#sec:checks}
==================
We have performed several checks on our code. We have generated subroutines for the virtual corrections using [[NJET]{}]{} [@Badger:2012pg], [[GoSam]{}]{} [@Cullen:2011ac; @Mastrolia:2010nb; @Mastrolia:2012du; @Binoth:2008uq; @Heinrich:2010ax; @Guillet:2013msa; @vanHameren:2010cp] and [[HELAC-NLO]{}]{} [@Bevilacqua:2011xh] and compared them to the virtual contributions obtained with the routines in our program. We have found that the [[NJET]{}]{}, [[GoSam]{}]{} and [[HELAC-NLO]{}]{} results were in agreement among each other for all subprocesses. We also found agreement with the [[NLOJET++]{}]{} routines, except for the $ q\bar{q} g g g$ amplitude, and all its crossings, where there was a problem in the colour sum. After fixing this, we found perfect agreement.
The Born contribution, together with the colour- and spin-correlated Born amplitude, were generated using the [[[MadGraph4]{}]{}]{} [[POWHEG BOX]{}]{} interface. This allowed also the generation of the real contribution according to [[[MadGraph4]{}]{}]{}, that was thus checked against the (much faster) one that we have implemented in our code.
As a further check, the full NLO calculation was compared with the one of ref. [@Badger:2012pf] by first comparing the plots displayed there with the results of our code, that we ran using the same parton distribution functions and the same scales. We found agreement within statistical errors. In order to carry out this comparison we needed to run the [[POWHEG BOX]{}]{} with the same dynamical scale choice of ref. [@Badger:2012pf]. Although the default scale choice in the [[POWHEG BOX]{}]{} is computed as a function of the underlying-Born kinematics, it is possible to set it up in such a way that any scale choice can be implemented. See appendix \[app:btlscalereal\] for more details.
Comparing [[trijet]{}]{} results at various levels {#sec:comparisons}
==================================================
The [[trijet]{}]{} generator can be used to compute three-jet observables at the NLO level, at the [[POWHEG]{}]{} Les Houches Event level (i.e. with the hardest emission generated according to the shower technique implemented in [[POWHEG]{}]{}), and after the full shower. The [[MiNLO]{}]{} feature can be turned on already at the NLO stage. In the present section we compare the output of our generator at these levels.
General settings for the forthcoming comparisons {#sec:settings}
------------------------------------------------
We consider jet production at the 8 TeV LHC. We use the [CT10nlo]{} pdf set [@Lai:2010vv]. We remind the reader that any of the modern pdf sets can be used [@Martin:2009iq; @Ball:2012cx], and our choice is only a matter of definiteness. We consider the shower output at the parton level, without the inclusion of hadronization effects and multiple interactions, since, at this stage, we are not comparing our result with jet data.[^4] We interface our generator to [[PYTHIA]{}]{} (version 6.4.25), [[Pythia8]{}]{} (version 8.183) and [[HERWIG]{}]{} (version 6.510). When using [[PYTHIA]{}]{} we use the Perugia 0 tune [@Skands:2010ak] ([ pytune(320)]{}). We use [[Pythia8]{}]{} and [[HERWIG]{}]{} with their default tunes.
In the [[POWHEG]{}]{} settings we have included the [doublefsr 1]{} option, and the modification of the [scalup]{} prescription obtained by setting [changescalup 1]{}. Both these features are illustrated in ref. [@Nason:2013uba].
We encountered severe stability problems when using [[HERWIG]{}]{}, showing up as spikes in our final histograms. We investigated them, and found that they were related to events having very small transverse energy (of the order of few GeV) at the Les Houches level, and developing very high transverse momentum jets (above 50 GeV) after shower. The cause of these problems was photon emission from quarks, that apparently does not comply with the [ scalup]{} veto in [[HERWIG]{}]{}. These problems disappeared completely by setting [vpcut=1D30]{}, that switches off photon radiation. Thus, all our [[HERWIG]{}]{} results were obtained with this setting. We verified that it has no visible effect on our results, but for the disappearance of the spikes.
The use of [[PYTHIA]{}]{} with the [[trijet]{}]{} generator in combination with [[MiNLO]{}]{} requires particular care. In fact, in this case, [[PYTHIA]{}]{} is unable to shower a sensible fraction of very small ${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}$ events. These events are not going to contribute to physically interesting distributions. Thus, they should be treated as events that did not pass the cuts, and should be counted when dividing the total weight entering a histogram bin by the total number of events. It turns out that, when [[PYTHIA]{}]{} finds such events, it silently discards them and loads a new one. In the analysis setup of the [[POWHEG BOX]{}]{}, the total number of events is usually computed as the number of times that the analysis routine is called, and, because of the aforementioned behaviour of [[PYTHIA]{}]{}, the fraction of discarded events is thus not counted. We coded a workaround for this problem in our analysis routines. A user adopting different analysis frameworks must make sure not to incur this problem.
We adopt the following values for the parameters entering the $F$ function in eqs. (\[eq:F\]) and (\[eq:F2\]): $$S_1=\left(50\;{\rm GeV}\right)^2\,,\quad\quad S_2=\left(800\;{\rm
GeV}\right)^2\,,\quad\quad p=2\,.$$ In addition, in order to avoid uninteresting regions with very small transverse momenta that may cause numerical problems, we reject all Born configurations such that $\min\{q_i,q_{ij}\}<\left(0.3\;{\rm GeV}\right)^2$. The value of the factorization and renormalization scale is taken equal to ${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}/2$, computed on the partonic configuration of the underlying-Born kinematics.
Finally, jets are reconstructed using the anti-${k_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm T}}$ algorithm [@Cacciari:2008gp] as implemented in the [[Fastjet]{}]{} package [@Cacciari:2011ma; @Cacciari:2005hq], with jet radius $R$.
The results we will show in the next sections have been obtained by generating 2.4 M events in two runs, with and without [[MiNLO]{}]{}. The runs have been performed on a 48 core machine, and they took roughly 37 and 100 hours, respectively.
NLO, LHE and shower-level comparisons {#sec:NLOvsLH}
-------------------------------------
We begin by showing in fig. \[fig:NLO-LH-pt3\] the comparison of the fixed-NLO and the LHE-level results for the transverse-momentum distribution of the third jet.
The [[PYTHIA]{}]{} and [[HERWIG]{}]{} showered output, compared to the NLO result, are displayed in figs. \[fig:NLO-PY-pt3\] and \[fig:NLO-HW-pt3\].
We first remark that the small transverse-momentum suppression of the LHE, [[PYTHIA]{}]{} and [[HERWIG]{}]{} showered results is simply due to the fact that, because of our $F$ function, events with transverse momentum smaller than 50 GeV are very rarely generated.
We observe that when $R=1$ all results are in better agreement. Differences arise for smaller values of $R$, and can be understood as follows. First of all, it can be easily checked that the LHE level result has very mild dependence upon $R$. This is due to the fact that the splitting of the third parton into two collinear partons has a very strong Sudakov suppression. In fact, in this case the [[POWHEG]{}]{} Sudakov form factor is the product of the form factors for vetoing harder final-state splittings of all final-state partons, times the form factor for vetoing harder initial-state radiation. Because of this suppression, partons are relatively well separated, and a small $R$ dependence is observed.[^5] When completing the shower, further splitting processes can take place at a suitable rate, and the $R$ dependence is reinstated. Notice that no Sudakov suppression for radiation is included in the fixed-order calculation, yielding a visible $R$ dependence. It is clear, however, that in the [[trijet]{}]{} generator the $R$ dependence will mostly arise at the shower stage. This is a desirable feature. In fact we do not expect the NLO result to be reliable in this region, since, among other things, it lacks the Sudakov form factor and the appropriate scale choice in the coupling constant. Because of this, the shower algorithm will acquire the responsibility to reliably describe the $R$ dependence.
In figs. \[fig:NLO-PY-pt3\] and \[fig:NLO-HW-pt3\] we observe a disturbing difference between [[PYTHIA]{}]{} and [[HERWIG]{}]{}. In the latter, the $R$ dependence is much stronger. Our expectation is that the shower result should be determined by two elements: on one side, the introduction of the correct Sudakov form factor for the splitting process (that will tend to reduce collinear splitting and thus increase the third jet cross section at smaller $R$), and multiple emissions, that will tend to increase collinear splitting processes, and thus reduce the jet cross section at smaller $R$. The net effect is an increase of the shower cross section at $R=0.5$ (with respect to the $R=1$ value) that is around 10% in [[PYTHIA]{}]{}, but is more of the order of 20% in [[HERWIG]{}]{}. Furthermore, the [[HERWIG]{}]{} result shows an increasing discrepancy with the fixed order result at large transverse momenta of the third jet in the $R=1$ case. We have no good understanding of why this is the case. On the other hand, the [[Pythia8]{}]{} result is compatible with the [[PYTHIA]{}]{} one.
When going from a fixed-NLO result to an LHE one, the most striking differences are usually observed in the spectrum of the extra parton emitted in the real process, that, in our case, corresponds to the fourth jet. We thus compare the transverse-momentum distribution of the fourth jet, with a cut on the third jet, ${p_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm T}}^{j_3}\ge 100$ GeV, computed at the NLO level, with the LHE level result (fig. \[fig:fourthNLO-LH\]), and with the [[PYTHIA]{}]{} showered one (fig. \[fig:fourthNLO-PY\]). The 100 GeV cut on the third jet is imposed for the following reason. If no cuts on the remaining jets are imposed, as ${p_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm T}}^{j_4}$ increases, also ${p_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm T}}^{j_1}$, ${p_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm T}}^{j_2}$ and ${p_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm T}}^{j_3}$ must increase, and we are thus probing an overall property of the cross section. With the cut on the third jet, by studying the ${p_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm T}}^{j_4}$ spectrum below the third-jet cut, we are studying the soft-collinear radiation dynamics from the three-jet Born configuration. We thus expect, for example, that the fixed-order result (that for this quantity has only leading order accuracy) diverges at very small transverse momenta, in contrast with the LHE result, where the soft-collinear region for the emission of the fourth jet is strongly Sudakov suppressed. We also remark that, in this case, the $R$ dependence of the NLO result is not at all reliable, since no further partons are emitted in this framework beyond the fourth.
Comparison between the NLO results with a standard choice of scales and NLO+[[MiNLO]{}]{} {#sec:NLOvsNLOMiNLO}
-----------------------------------------------------------------------------------------
Before turning to the discussion of the [[trijet]{}]{} results when [[MiNLO]{}]{} is active, we perform a comparison of the bare NLO calculation with or without [[MiNLO]{}]{}, whose purpose is to show that only minor differences are seen in the region where the three jets are resolved. Here we report in fig. \[fig:NLOminlo-NLO\] such comparison for the transverse-momentum distribution of the third jet. Our standard scale choice, when [[MiNLO]{}]{} is not used, is to set ${\mu_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm R}}$ and ${\mu_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm F}}$ equal to ${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}/2$, where ${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}$ is computed on the kinematics of the underlying-Born configuration. We see that the two results are fairly compatible, except in the very small transverse-momentum region. Here, the NLO with the standard scale choice grows much faster in magnitude, and becomes large and negative in the first bin. With [[MiNLO]{}]{}, the small transverse-momentum region is better behaved, as expected. However, we remind the reader that, in fig. \[fig:NLOminlo-NLO\], this region is divergent also with [[MiNLO]{}]{}, since it is dominated by the production of low transverse-momentum jets.
NLO, LHE and shower-level comparisons with [[MiNLO]{}]{} {#sec:NLOvsLHMiNLO}
--------------------------------------------------------
It is interesting to present the results for the transverse momentum of the third jet, when the [[MiNLO]{}]{} feature is turned on. Also in this case, for $R=1$ we see good agreement among the NLO, LHE and the showered results. We illustrate these results in figs. \[fig:NLO-LH-minlo-pt3\], \[fig:NLO-PY-minlo-pt3\] and \[fig:NLO-HW-minlo-pt3\]. It turns out, however, that for $R=0.5$ the NLO result is below the LHE one by a larger amount with respect to the no-[[MiNLO]{}]{} case, which leads to a slightly better agreement of the showered and NLO results.
The low transverse-energy region {#sec:smallHT}
--------------------------------
When showering a [[POWHEG]{}]{}-generated LHE configuration, we usually assume that the jet structure of the event is only marginally affected by the shower. In jet production, in particular, we assume that the configurations having very small transverse energy (i.e. ${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}$) at the Les Houches level should not contribute significantly to events that pass the jet cuts. It turns out, however, that such configurations have diverging cross section, and thus one may worry that the very small probability that the shower has for building up relatively hard jets starting from LHE configurations with small ${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}$, may end up being amplified by an unphysically large cross section.
A similar problem arises when we consider associated jets in a hard phenomenon. In this case, however, the [[MiNLO]{}]{} procedure ensures that the cross section for the Les Houches event is physically well behaved. On the other hand, in the case of jet production, the Sudakov resummation is not enough to guarantee a physical behaviour at small transverse energies.
In order to study this potential problem, we have taken a very simple approach. We have determined the cross section for events with transverse energy above a given cut at the Les Houches level. We have then found that for ${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}>10$ GeV such cross section is about 60 mb, going down to around 30 mb for ${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}>20$ GeV. Without imposing this cut, the cross section reaches 1000 mb. The diverging behaviour is, in fact, limited by the tiny cut-off that we impose upon the kinematics to avoid divergences. Since the inelastic cross section at the 7 TeV LHC is around 70 mb [@Antchev:2013haa], it seems reasonable to impose a cut on the transverse energy of our events at the Les Houches level, in the range between 10 and 20 GeV.
We have found that, for the shower Monte Carlo programs that we have considered, and with our settings, the Les Houches level cut has visible impact only on events with very small transverse momenta. For example, we find sensible differences in the distribution of the transverse momentum of the third jet only for ${p_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm T}}^{j_3}\lesssim 5$ GeV, for the cut ${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}>10$ GeV, and for ${p_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm T}}^{j_3}\lesssim 10$ GeV for the cut ${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}>20$ GeV. This is reasonable to expect, since the ${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}$ of the event is at least three times the transverse momentum of the third jet. In spite of this, we have preferred to maintain the Les Houches level ${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}>20$ GeV cut as our standard, since it seems, in all cases, unreasonable to have events with a cross section that becomes of the same order of the total inelastic cross section.
Comparison of the [[trijet]{}]{}+[[MiNLO]{}]{} results with the [[dijet]{}]{} results for inclusive quantities {#sec:trijet-dijet-comparison}
--------------------------------------------------------------------------------------------------------------
When using [[MiNLO]{}]{}, the [[trijet]{}]{} generator becomes predictive also for inclusive quantities, i.e. for observables that do not necessarily require the presence of a third jet. It is hard to quantify theoretically the accuracy of such predictions. We have shown that, for inclusive quantities, the [[VJ]{}]{} generators (i.e. generators for Higgs or $W/Z$ boson production in association with a jet) improved with the [[MiNLO]{}]{} procedure yield an accuracy that is better then LO, but not quite at the NLO level, unless one makes a careful tuning of the procedure [@Hamilton:2012rf]. In the case of jets, the argument of ref. [@Hamilton:2012rf] cannot be applied as is, since the soft-singularity structure of the two-parton production process is quite involved. Rather than trying to understand theoretically what is the accuracy of the [[trijet]{}]{}+[[MiNLO]{}]{} generator for inclusive quantities, here we simply compare its inclusive distributions to those obtained with the NLO-accurate [[dijet]{}]{} generator.
We begin by showing in fig. \[fig:trijet-dijet-PY8-j3\] the transverse-momentum distribution of the third jet, ${p_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm T}}^{j_3}$, in events where ${p_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm T}}^{j_2}> 20$ GeV. The aim of the figure is to compare the Sudakov effects affecting the production of the third jet, introduced by the [[POWHEG]{}]{} machinery, in the case of the [[dijet]{}]{} generator, with those introduced in the [[trijet]{}]{} generator by the [[MiNLO]{}]{} procedure. We see that, in both cases, the small transverse-momentum region is properly regulated by the Sudakov form factor. We remind the reader that, as far as the large transverse-momentum region is concerned, the [[trijet]{}]{} generator has NLO accuracy for this observable, while the [[dijet]{}]{} one is only LO accurate. The slightly strange features at very low transverse momentum, that we observe with the [[Pythia8]{}]{} shower, are shared for the same observable by the [[PYTHIA]{}]{} result, while, in the [[HERWIG]{}]{} case, we see a smoother behaviour, as shown in figs. \[fig:trijet-dijet-PY-j3\] and \[fig:trijet-dijet-HW-j3\].
We now turn to the comparison of the [[trijet]{}]{}+[[MiNLO]{}]{} and [[dijet]{}]{} generators for inclusive-jet distributions. We first compare the rapidity distribution of an inclusive jet in fig. \[fig:trijet-dijet-PY8\_y\]. No interesting differences are seen between the [[trijet]{}]{}+[[MiNLO]{}]{} and [[dijet]{}]{} curves, other than the obvious difference in normalization, due to the cut in transverse momentum, that could also be evinced from the transverse-momentum distribution. We show this distribution only for the case of [[Pythia8]{}]{} shower, since we obtain similar results when using [[HERWIG]{}]{} or [[PYTHIA]{}]{}.
Another interesting distribution is the transverse-momentum of an inclusive jet, displayed in figs. \[fig:trijet-dijet-PY8\], \[fig:trijet-dijet-PY\] and \[fig:trijet-dijet-HW\] for [[Pythia8]{}]{}, [[PYTHIA]{}]{} and [[HERWIG]{}]{} respectively. We notice the remarkable agreement between the [[trijet]{}]{}+[[MiNLO]{}]{} and [[dijet]{}]{} generators.
In conclusion, we find that the use of [[MiNLO]{}]{} considerably improves the [[trijet]{}]{} generator also in the region where one jet becomes unresolved. Observe that, while the [[dijet]{}]{} generator is NLO accurate for these distributions, the [[trijet]{}]{}+[[MiNLO]{}]{} generator is at most LO accurate. Thus, we do not advocate the use of the [[trijet]{}]{}+[[MiNLO]{}]{} generator for one or two jet inclusive distributions. However, the fact that also these regions are treated consistently gives us confidence that we should not be afraid to make predictions for three-jet observables also in the region where one jet is relatively close to being soft, or relatively close to a collinear configuration. We thus recommend that our generator is used with the [[MiNLO]{}]{} feature turned on.
Scale-variation bands {#sec:scalevar}
---------------------
In this section we show the scale-variation bands for some key distributions. The purpose of this section is threesome: to show that scale uncertainties can be easily computed, to give an idea of the uncertainty involved, and to show that the remarkable agreement of the [[trijet]{}]{}+[[MiNLO]{}]{} and [[dijet]{}]{} results for inclusive quantities is not accidental. A more thorough uncertainty study will be carried out in a forecoming publication where comparisons with available data will be considered.
We illustrate in figs. \[fig:LH-scalevar-j3-ptzoom\], \[fig:LH-scalevar-j-y\] and \[fig:LH-scalevar-j-pt\] the scale-variation bands for the transverse-momentum distribution of the third hardest jet, and the rapidity distribution and transverse momentum of the inclusive jet, for [[trijet]{}]{}+[[MiNLO]{}]{} at the LHE level. We compare the [[trijet]{}]{}+[[MiNLO]{}]{} scale variation with the same scale variation in [[dijet]{}]{} production.
The scale variation can be performed in a fast way, without regenerating the event sample, using the reweighting tool in the [[POWHEG BOX V2]{}]{}, that allows for a very fast re-evaluation of the weight associated to each event.[^6] The scale-variation band is obtained by taking the envelope of the 7 differential cross sections computed by multiplying the reference factorization and renormalization scales by the factors ${K_{\scriptscriptstyle \rm F}}$ and ${K_{\scriptscriptstyle \rm R}}$, respectively, where $$({K_{\scriptscriptstyle \rm R}},{K_{\scriptscriptstyle \rm F}})=(0.5,0.5), (0.5,1), (1,0.5), (1,1), (2,1), (1,2), (2,2).$$ In all the three distributions we are showing, we get a comparable scale-band size in the [[trijet]{}]{}+[[MiNLO]{}]{} and [[dijet]{}]{} results, of the order of 20%, and there is a very good degree of overlapping for the inclusive-jet rapidity and transverse momentum.
Conclusions {#sec:conclusions}
===========
In this paper, we have presented an implementation of a NLO plus parton-shower generator for three-jet production, built in the framework of the [[POWHEG BOX V2]{}]{}.
We have compared key kinematic distributions at different levels: NLO, Les Houches event level, and after the shower performed by [[PYTHIA]{}]{}, [[Pythia8]{}]{} and [[HERWIG6]{}]{}. We found very good agreement between the NLO and the [[PYTHIA]{}]{} and [[Pythia8]{}]{} results, for variables that are correctly described by a fixed-order calculation. Slightly worse agreement is found between the NLO and the [[HERWIG]{}]{}-showered results.
We have also applied the recently-proposed [[MiNLO]{}]{} procedure, for the scale assignment in the NLO calculation, to our generator. We have found that [[MiNLO]{}]{} considerably improves the [[trijet]{}]{} generator also in the regions where one jet becomes unresolved. The fact that also these regions are treated consistently gives us confidence that we can make predictions for three-jet observables also in the region where one jet is relatively close to being soft, or relatively close to a collinear configuration.
We have seen that the [[trijet]{}]{}+[[MiNLO]{}]{} and [[dijet]{}]{} results display remarkable consistency among each other. On the other hand, we have also evidence that the kind of shower generator that is used for the final shower has a non-negligible impact on the result, especially for relatively-small jet cone sizes.
The code can be downloaded following the instructions in the [[POWHEG BOX]{}]{} web site <http://powhegbox.mib.infn.it>.
Acknowledgments {#acknowledgments .unnumbered}
===============
We thank Simon Badger for helping us in the use of [[NJET]{}]{}, and Gionata Luisoni and Francesco Tramontano for their help in running the [[GoSam]{}]{} package. A.K. is grateful to Zoltán Nagy and Zoltán Trócsányi for several useful discussions and help with the `NLOJET++` code. We kindly acknowledge (A.K. in particular) the financial support provided by the LHCPhenoNet Training Network. A.K. acknowledges the Hungarian Scientific Research Fund grant K-101482 and is thankful to the Aspen Center for Physics for warm hospitality where part of this work was carried out. The computations shown in this paper were partially performed on the HPC facility of the University of Debrecen (NIIF Institute, Hungary).
Scale options in the [[POWHEG BOX]{}]{} {#app:btlscalereal}
=======================================
In the [[POWHEG BOX]{}]{}, the factorization and renormalization scales are usually set as a function of the underlying-Born kinematics. Since the [[POWHEG BOX]{}]{} can also be used as a parton level, fixed-order generator (by setting [ testplots 1]{} in the [powheg.input]{} file), it is convenient, at times, to remove this restriction (for example, in order to compare the fixed order NLO output to other codes). This is done as follows. If one sets the variable [btlscalereal 1]{} in the [powheg.input]{} file, the internal flag [ flg\_btildepart]{} is used to distinguish the Born, virtual and subtraction-term contributions from the real one. When [flg\_btildepart]{} equals [’b’]{}, the program is computing the Born or virtual. When it is set to [’r’]{} it is computing the real contribution. The user can then modify the [set\_fac\_ren\_scales]{} subroutine, so that, on the basis of the value of [flg\_btildepart]{}, the program uses the Born or real kinematics to compute the scales.
Another ambiguity in the scale choice has to do with the computation of the subtraction terms. It is acceptable to use for them the same scales used for the real contributions. On the other hand, it is also acceptable to use for them the scales of the corresponding underlying-Born configuration. In order to implement also this option, one sets the variable [btlscalect 1]{} in the [powheg.input]{} file. If this variable is set, the [ set\_fac\_ren\_scales]{} subroutine is called with [flg\_btildepart]{} equals to [’c’]{} when the subtraction terms are computed.
For the comparison with the results of ref. [@Badger:2012pf], we set ${\mu_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm R}}={\mu_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}\rm F}}={H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}/2$. ${H_{{\mathchoice{\displaystyle}{\scriptstyle}{\scriptscriptstyle}{\scriptscriptstyle}}T}}$ is computed using the Born kinematics, for the Born, the virtual and subtraction terms, and using the real-contribution kinematics for the real terms. A code that implements this choice of scales has the form:
subroutine set_fac_ren_scales(muf,mur)
...
include 'pwhg_kn.h'
include 'pwhg_flg.h'
...
if ((flg_btildepart.eq.'b').or.(flg_btildepart.eq.'c')) then
pt1 = sqrt(kn_cmpborn(1,3)**2+kn_cmpborn(2,3)**2)
pt2 = sqrt(kn_cmpborn(1,4)**2+kn_cmpborn(2,4)**2)
pt3 = sqrt(kn_cmpborn(1,5)**2+kn_cmpborn(2,5)**2)
Ht = pt1 + pt2 + pt3
elseif ((flg_btildepart.eq.'r')) then
pt1 = sqrt(kn_cmpreal(1,3)**2+kn_cmpreal(2,3)**2)
pt2 = sqrt(kn_cmpreal(1,4)**2+kn_cmpreal(2,4)**2)
pt3 = sqrt(kn_cmpreal(1,5)**2+kn_cmpreal(2,5)**2)
pt4 = sqrt(kn_cmpreal(1,6)**2+kn_cmpreal(2,6)**2)
Ht = pt1 + pt2 + pt3 + pt4
endif
muf=Ht/2
mur=Ht/2
where [kn\_cmpborn]{} and [kn\_cmpreal]{} are the arrays of the Born and real center-of-mass momenta, defined in the [[POWHEG BOX]{}]{} [pwhg\_kn.h]{} include file, [flg\_btildepart]{} is declared in the [pwhg\_flg.h]{} file, and [pt1]{} …[pt4]{} are local variables denoting the transverse momenta of the final-state partons.
[10]{}
J. R. Andersen and J. M. Smillie, [*[Multiple Jets at the LHC with High Energy Jets]{}*]{}, [*JHEP*]{} [**1106**]{} (2011) 010, \[[[arXiv:1101.5394]{}](http://xxx.lanl.gov/abs/1101.5394)\].
P. Nason, [*[A new method for combining NLO QCD with shower Monte Carlo algorithms]{}*]{}, [*JHEP*]{} [**11**]{} (2004) 040, \[[[hep-ph/0409146]{}](http://xxx.lanl.gov/abs/hep-ph/0409146)\].
S. Frixione, P. Nason, and C. Oleari, [*[Matching NLO QCD computations with Parton Shower simulations: the POWHEG method]{}*]{}, [*JHEP*]{} [**11**]{} (2007) 070, \[[[arXiv:0709.2092]{}](http://xxx.lanl.gov/abs/0709.2092)\].
S. Alioli, K. Hamilton, P. Nason, C. Oleari, and E. Re, [*[Jet pair production in POWHEG]{}*]{}, [*JHEP*]{} [**1104**]{} (2011) 081, \[[[arXiv:1012.3380]{}](http://xxx.lanl.gov/abs/1012.3380)\].
S. Alioli, P. Nason, C. Oleari, and E. Re, [*[A general framework for implementing NLO calculations in shower Monte Carlo programs: the POWHEG BOX]{}*]{}, [*JHEP*]{} [**06**]{} (2010) 043, \[[[arXiv:1002.2581]{}](http://xxx.lanl.gov/abs/1002.2581)\].
Z. Bern, L. J. Dixon, and D. A. Kosower, [*[One loop corrections to five gluon amplitudes]{}*]{}, [*Phys.Rev.Lett.*]{} [**70**]{} (1993) 2677–2680, \[[[hep-ph/9302280]{}](http://xxx.lanl.gov/abs/hep-ph/9302280)\].
Z. Bern, L. J. Dixon, and D. A. Kosower, [*[One loop corrections to two quark three gluon amplitudes]{}*]{}, [*Nucl.Phys.*]{} [**B437**]{} (1995) 259–304, \[[[hep-ph/9409393]{}](http://xxx.lanl.gov/abs/hep-ph/9409393)\].
Z. Kunszt, A. Signer, and Z. Trocsanyi, [*[One loop radiative corrections to the helicity amplitudes of QCD processes involving four quarks and one gluon]{}*]{}, [*Phys.Lett.*]{} [**B336**]{} (1994) 529–536, \[[[hep-ph/9405386]{}](http://xxx.lanl.gov/abs/hep-ph/9405386)\].
J. Gunion and Z. Kunszt, [*[Four jet processes: gluon-gluon scattering to nonidentical quark-antiquark pairs]{}*]{}, [*Phys.Lett.*]{} [**B159**]{} (1985) 167.
J. F. Gunion and Z. Kunszt, [*[Six Quark Subprocesses in [QCD]{}]{}*]{}, [ *Phys.Lett.*]{} [**B176**]{} (1986) 163.
J. F. Gunion and Z. Kunszt, [*[Addendum Concerning the Four Quark Two Gluon Subprocess]{}*]{}, [*Phys.Lett.*]{} [**B176**]{} (1986) 477.
J. G. M. Kuijf, [*[Multiparton production at hadron colliders]{}*]{}, .
Z. Nagy, [*[Next-to-leading order calculation of three jet observables in hadron hadron collision]{}*]{}, [*Phys.Rev.*]{} [**D68**]{} (2003) 094002, \[[[hep-ph/0307268]{}](http://xxx.lanl.gov/abs/hep-ph/0307268)\].
J. Alwall [*et. al.*]{}, [*[MadGraph/MadEvent v4: The New Web Generation]{}*]{}, [*JHEP*]{} [**09**]{} (2007) 028, \[[[arXiv:0706.2334]{}](http://xxx.lanl.gov/abs/0706.2334)\].
J. M. Campbell, R. K. Ellis, R. Frederix, P. Nason, C. Oleari, [*et. al.*]{}, [*[NLO Higgs Boson Production Plus One and Two Jets Using the POWHEG BOX, MadGraph4 and MCFM]{}*]{}, [*JHEP*]{} [**1207**]{} (2012) 092, \[[[arXiv:1202.5475]{}](http://xxx.lanl.gov/abs/1202.5475)\].
K. Hamilton, P. Nason, and G. Zanderighi, [*[MINLO: Multi-Scale Improved NLO]{}*]{}, [*JHEP*]{} [**1210**]{} (2012) 155, \[[[arXiv:1206.3572]{}](http://xxx.lanl.gov/abs/1206.3572)\].
S. Alioli, P. Nason, C. Oleari, and E. Re, [*[Vector boson plus one jet production in POWHEG]{}*]{}, [*JHEP*]{} [**01**]{} (2011) 095, \[[[arXiv:1009.5594]{}](http://xxx.lanl.gov/abs/1009.5594)\].
S. Badger, B. Biedermann, P. Uwer, and V. Yundin, [*[Numerical evaluation of virtual corrections to multi-jet production in massless QCD]{}*]{}, [ *Comput.Phys.Commun.*]{} [**184**]{} (2013) 1981–1998, \[[[arXiv:1209.0100]{}](http://xxx.lanl.gov/abs/1209.0100)\].
G. Cullen, N. Greiner, G. Heinrich, G. Luisoni, P. Mastrolia, [*et. al.*]{}, [*[Automated One-Loop Calculations with GoSam]{}*]{}, [*Eur.Phys.J.*]{} [ **C72**]{} (2012) 1889, \[[[ arXiv:1111.2034]{}](http://xxx.lanl.gov/abs/1111.2034)\].
P. Mastrolia, G. Ossola, T. Reiter, and F. Tramontano, [*[Scattering AMplitudes from Unitarity-based Reduction Algorithm at the Integrand-level]{}*]{}, [*JHEP*]{} [**1008**]{} (2010) 080, \[[[arXiv:1006.0710]{}](http://xxx.lanl.gov/abs/1006.0710)\].
P. Mastrolia, E. Mirabella, G. Ossola, T. Peraro, and H. van Deurzen, [*[The Integrand Reduction of One- and Two-Loop Scattering Amplitudes]{}*]{}, [*PoS*]{} [**LL2012**]{} (2012) 028, \[[[ arXiv:1209.5678]{}](http://xxx.lanl.gov/abs/1209.5678)\].
T. Binoth, J. P. Guillet, G. Heinrich, E. Pilon, and T. Reiter, [*[Golem95: a numerical program to calculate one-loop tensor integrals with up to six external legs]{}*]{}, [*Comput. Phys. Commun.*]{} [**180**]{} (2009) 2317–2330, \[[[arXiv:0810.0992]{}](http://xxx.lanl.gov/abs/0810.0992)\].
G. Heinrich, G. Ossola, T. Reiter, and F. Tramontano, [*[Tensorial Reconstruction at the Integrand Level]{}*]{}, [*JHEP*]{} [**1010**]{} (2010) 105, \[[[arXiv:1008.2441]{}](http://xxx.lanl.gov/abs/1008.2441)\].
J. P. Guillet, G. Heinrich, and J. von Soden-Fraunhofen, [*[Tools for NLO automation: extension of the golem95C integral library]{}*]{}, [[arXiv:1312.3887]{}](http://xxx.lanl.gov/abs/1312.3887).
A. van Hameren, [*[OneLOop: For the evaluation of one-loop scalar functions]{}*]{}, [*Comput.Phys.Commun.*]{} [**182**]{} (2011) 2427–2438, \[[[arXiv:1007.4716]{}](http://xxx.lanl.gov/abs/1007.4716)\].
G. Bevilacqua, M. Czakon, M. Garzelli, A. van Hameren, A. Kardos, [ *et. al.*]{}, [*[HELAC-NLO]{}*]{}, [*Comput.Phys.Commun.*]{} [**184**]{} (2013) 986–997, \[[[arXiv:1110.1499]{}](http://xxx.lanl.gov/abs/1110.1499)\].
S. Badger, B. Biedermann, P. Uwer, and V. Yundin, [*[NLO QCD corrections to multi-jet production at the LHC with a centre-of-mass energy of $\sqrt{s}=8$ TeV]{}*]{}, [*Phys.Lett.*]{} [**B718**]{} (2013) 965–978, \[[[arXiv:1209.0098]{}](http://xxx.lanl.gov/abs/1209.0098)\].
H.-L. Lai, M. Guzzi, J. Huston, Z. Li, P. M. Nadolsky, [*et. al.*]{}, [*[New parton distributions for collider physics]{}*]{}, [*Phys.Rev.*]{} [**D82**]{} (2010) 074024, \[[[ arXiv:1007.2241]{}](http://xxx.lanl.gov/abs/1007.2241)\].
A. D. Martin, W. J. Stirling, R. S. Thorne, and G. Watt, [*[Parton distributions for the LHC]{}*]{}, [*Eur. Phys. J.*]{} [**C63**]{} (2009) 189–285, \[[[arXiv:0901.0002]{}](http://xxx.lanl.gov/abs/0901.0002)\].
R. D. Ball, V. Bertone, S. Carrazza, C. S. Deans, L. Del Debbio, [*et. al.*]{}, [*[Parton distributions with LHC data]{}*]{}, [*Nucl.Phys.*]{} [**B867**]{} (2013) 244–289, \[[[ arXiv:1207.1303]{}](http://xxx.lanl.gov/abs/1207.1303)\].
P. Z. Skands, [*[Tuning Monte Carlo Generators: The Perugia Tunes]{}*]{}, [ *Phys.Rev.*]{} [**D82**]{} (2010) 074018, \[[[arXiv:1005.3457]{}](http://xxx.lanl.gov/abs/1005.3457)\].
P. Nason and C. Oleari, [*[Generation cuts and Born suppression in POWHEG]{}*]{}, [[arXiv:1303.3922]{}](http://xxx.lanl.gov/abs/1303.3922).
M. Cacciari, G. P. Salam, and G. Soyez, [*[The anti-$k_T$ jet clustering algorithm]{}*]{}, [*JHEP*]{} [**04**]{} (2008) 063, \[[[arXiv:0802.1189]{}](http://xxx.lanl.gov/abs/0802.1189)\].
M. Cacciari, G. P. Salam, and G. Soyez, [*[FastJet User Manual]{}*]{}, [ *Eur.Phys.J.*]{} [**C72**]{} (2012) 1896, \[[[arXiv:1111.6097]{}](http://xxx.lanl.gov/abs/1111.6097)\].
M. Cacciari and G. P. Salam, [*[Dispelling the $N^{3}$ myth for the $k_t$ jet-finder]{}*]{}, [*Phys.Lett.*]{} [**B641**]{} (2006) 57–61, \[[[hep-ph/0512210]{}](http://xxx.lanl.gov/abs/hep-ph/0512210)\].
Collaboration, G. Antchev [*et. al.*]{}, [*[Measurement of proton-proton inelastic scattering cross-section at $\sqrt{S} = 7$ TeV]{}*]{}, [*Europhys.Lett.*]{} [**101**]{} (2013) 21003.
K. Hamilton, P. Nason, C. Oleari, and G. Zanderighi, [*[Merging H/W/Z + 0 and 1 jet at NLO with no merging scale: a path to parton shower + NNLO matching]{}*]{}, [*JHEP*]{} [**1305**]{} (2013) 082, \[[[arXiv:1212.4504]{}](http://xxx.lanl.gov/abs/1212.4504)\].
[^1]: For a shower implementation focused upon the high-energy limit, see [@Andersen:2011hs].
[^2]: The [[POWHEG BOX V2]{}]{} framework is an enhanced version of the original [[POWHEG BOX]{}]{} package. A paper describing the new features of the [[POWHEG BOX V2]{}]{} is in preparation.
[^3]: In the [[MiNLO]{}]{} framework, by basic process we mean the process before any branching has occurred, i.e. $H$ production in $H$ + jets, and dijet production in the present case.
[^4]: A comparison with available data is in progress.
[^5]: This feature of the LHE level events was already noticed and discussed in ref. [@Alioli:2010xa].
[^6]: This feature is documented in the file [ README.LesHouchesReweighting]{} in the [Docs]{} directory of [[POWHEG BOX V2]{}]{}.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The mean-field Hubbard model is used to investigate the formation of the antiferromagnetic phase in hexagonal graphene rings with inner zigzag edges. The outer edge of the ring was taken to be either zigzag or armchair, and we found that both types of structures can have a larger antiferromagnetic interaction as compared with hexagonal dots. This difference could be partially ascribed to the larger number of zigzag edges per unit area in rings than in dots. Furthermore, edge states localized on the inner ring edge are found to hybridize differently than the edge states of dots, which results in important differences in the magnetism of graphene rings and dots. The largest staggered magnetization is found when the outer edge has a zigzag shape. However, narrow rings with armchair outer edge are found to have larger staggered magnetization than zigzag hexagons. The edge defects are shown to have the least effect on magnetization when the outer ring edge is armchair shaped.'
author:
- 'M. Grujić'
- 'M. Tadić'
- 'F. M. Peeters'
title: 'Antiferromagnetism in hexagonal graphene structures: rings vs dots'
---
Recent tremendous progress in graphene research is driven by its remarkable properties, e.g. high crystalline quality, high electron mobility, lack of a band gap, and a minimal possible thickness, to name a few.[@nov04] The mentioned properties are advantageous for various applications of graphene, such as piezoelectric devices, [@ong12] supercapacitors, [@el-kady12] photodetectors, [@photod12] and field-effect transistors. [@lin10; @brit12] Furthermore, it has been predicted that graphene structures could exhibit [*magnetic ordering*]{} which is potentially advantageous for spintronic applications. [@oleg10; @son06] This effect is essentially related to either the global or local imbalance of sublattice atoms in bipartite lattices. An imbalance might give rise to zero energy states in the electron spectrum. These states are localized near the zigzag edges or vacancies, and along with the repulsive electron-electron ([*e*]{}-[*e*]{}) interaction could eventually lead to a spin polarization of the ground state of the system. [@oleg10] Furthermore, the spins on the same sublattice are found to exhibit ferromagnetic coupling along the graphene edges, whereas the spins on different sublattices along the graphene edges couple antiferromagnetically.
In theory, magnetic ordering has been demonstrated for graphene flakes, [@rossier07] nanoribbons, [@rossier08] and vacancies in bulk graphene. [@palac08] On the other hand, experimental reports on magnetism in graphene structures are rare and conflicting. They range from the detection of ferromagnetic or antiferromagnetic ordering [@matte09; @enoki09; @joly10] to measurements of defect-induced paramagnetism. [@ney11; @nair12] Magnetic ordering was even found to be preserved at room temperature [@wangliang09; @hong12]. The essential cause of magnetism in graphene is the existence of a peak in the density of nonbonding edge states near the Fermi energy. However, due to the high reactivity of these states, magnetism might be strongly suppressed. [@kunst11] Several theoretical studies offered explanations for a diversity of phenomena related to magnetic ordering and its suppression, which might occur by means of nonmagnetic edge passivation, edge reconstruction, or vanishing of spin correlations with increasing temperature. [@kunst11; @yazyev08] Hence, in order to experimentally detect magnetic ordering graphene samples should be kept under rigorously controlled conditions. Yet, various applications of this effect have been proposed. They involve half-metallicity with electrically controlled spin propagation, [@son06] defect induced spin filtering, [@wimmer08] and spin logic devices. [@wang09; @ezawa10]
In this report we employ the mean-field Hubbard model to study the formation of local magnetic moments in hexagonal graphene rings. Our aim is to explore how magnetic ordering is affected by the ring size and the edge type. In order to identify different hexagonal rings, we introduce the following notation which might be visually aiding. We assume that the type of the inner ring edge is zigzag, and that $N$ unit hexagons are adjacent to this boundary. The outer ring edge is assumed to be comprised of either $M$ dimers if it is of armchair type, or $M$ unit hexagons if it is of zigzag type. Therefore, the ring is denoted by $M:N$. As an example, consider the ring shown in Fig. 1, which is assumed to be formed out of the hexagonal dot with armchair edge, which contains seven dimers at each side of the hexagon, as shown in Fig. 1(a). The ring is formed when the carbon atoms around the center of the dot are removed, as depicted in Figs. 1(a) and (b). Potentially these exotic structures could be manufactured via substitutional doping of boron nitride nanostructures with carbon.[@wei11] Because the edge of the removed dot has four unit hexagons at each side, the ring is denoted as $7_{AC}:4_{ZG}$. The distributions of the magnetic moments in the graphene rings will be compared with the magnetic moment distributions in the hexagonal graphene dots. Those dots are assumed to have zigzag edges, and are labeled by $N_{ZG}$, where $N$ has the same meaning as the symbol $M$ for the rings.
Magnetic ordering of a graphene structure is governed by Lieb’s theorem. [@lieb89] It states that the total ground-state spin of a bipartite lattice with repulsive [*e*]{}-[*e*]{} interaction as described by the Hubbard model equals half of the difference of the sublattice sites. For symmetrical structures, this rule is related to the arrangement of the carbon atoms with respect to lines of reflection symmetry in the graphene plane: if there is a symmetry line which does not intersect any of the carbon atoms the total ground state spin is zero; otherwise there exists a finite magnetic moment. All the hexagonal rings analyzed here possess such a symmetry, thus their total magnetization equals zero, unlike triangular rings which display a ferrimagnetic phase. [@potasz11] However, Lieb’s theorem does not dictate the distribution of the local magnetic moments or the lack of zero energy states. Furthermore, the number of zero-energy states in the analyzed ring is an integer multiple of six, which is a consequence of the $C_{6v}$ symmetry of the ring. In the $9_{AC}:10_{ZG}$ ring, six zero energy states are found, which agrees with graph theory, and which is a topological property related to the nonperfect matching of the $p_z$ orbitals. [@oleg10]
The Hubbard Hamiltonian $$H=H_0+H_I,$$ is employed to compute the distribution of magnetic moments. $H_0$ is the noninteracting part, which represents the nearest neighbor tight-binding Hamiltonian, and is given by $$H_0=-t\displaystyle\sum\limits_{<i,j>,\sigma}c_{i\sigma}^\dagger c_{j\sigma},
\label{h0}$$ where $c_{j\sigma}$ and $c_{j\sigma}^\dagger$ are the annihilation and creation operators, respectively, and $t$ denotes the hopping integral. The interacting part $H_I$ describes the e-e interaction $$H_I=U\displaystyle\sum\limits_i\left(n_{i\uparrow}\langle
n_{i\downarrow}\rangle+n_{i\downarrow}\langle
n_{i\uparrow}\rangle-\langle n_{i\uparrow}\rangle\langle
n_{i\downarrow}\rangle\right), \label{hi}$$ where $n_{i\sigma}=c_{i\sigma}^\dagger c_{i\sigma}$ is the number operator, and $U$ denotes the on-site Coulomb repulsion energy for each pair of electrons with the opposite spins orbiting the same atom.[@futnota1] Equation (\[hi\]) is obtained within the mean-field approximation, which assumes that the spin-up (spin-down) electrons interact with the average density of spin-down (spin-up) electrons on a particular atomic site.
In our calculations, we take $t=2.7$ eV and $U=1.2t$.[@oleg10] We note that there is no consensus on the actual value of the strength of the Coulomb interaction to be used in the Hubbard model in graphene. Recent density functional theory (DFT) calculations came up with a value closer to $U=3.4t$.[@wehling11] However, having in mind that the mean-field approximation can overestimate the tendency for magnetic order for large $U$,[@feldner10] we chose the more conservative value of $U=1.2t$. The solution is then obtained by means of a self-consistent procedure which starts from an initial distribution of the spins, and ends when the maximum change of the electron density over the atomic sites drops below $10^{-5}$. When the self-consistent spin densities are determined, the magnetic moment per site $m_i$ is computed as $$m_i=\langle s_i^z\rangle=\left(\langle
n_{i\uparrow}\rangle-\langle n_{i\downarrow}\rangle\right)/2.$$
For the antiferromagnetic order parameter we take the staggered magnetization $$\mu_s^z=\frac{1}{N}\sum_i(-1)^i\langle s_i^z\rangle,$$ where $(-1)^i$ symbolizes that we sum up the contributions from opposite sublattices with opposite signs. This is the appropriate order parameter for antiferromagnetism when examining spin polarization occurring in bipartite lattices. The larger $\mu_s^z$ the stronger is, the antiferromagnetic phase. In addition to $\mu_s^z$, the shift in the electron and the hole energy spectra which arises from the magnetic order is quantified as $\Delta
E=\left(E^{HOS}+E^{LUS}\right)/2$, where $E^{HOS}$ and $E^{LUS}$ are the highest occupied and lowest unoccupied states in the ground state at half filling, respectively. Note that in the nonmagnetic state we have $\Delta E=0$. We will explore how the maximum magnetic moment $m_{max}$ varies with the ring width.
{width="8.6cm"}
The distribution of the local magnetic moments in the $9_{AC}:N_{ZG}$ rings for several values of $N$ is shown in Fig. 1(c). The symmetry of each hexagonal ring is $C_{6v}$, whereas the symmetry of the magnetic moment distribution is $IC_{6v}$, i.e. the magnetic moments alter sign when rotated over $\pi/3$ rad. Therefore, it suffices to display the distribution of magnetic moments in sectors of $\pi/3$ rad, as done in Fig. 1(c), which combines the sectors of different $N$. Orientation and color of triangles denotes the orientation of the magnetic moments, and the absolute value of $m_i$ is depicted by color intensity.
{width="8.6cm"}
It is evident in Fig. 1(c) that both the appearance of staggered magnetization and the total magnetic moment situated on the inner edge of the ring depend on the ring width. Furthermore, we observe a phase change from nonmagnetic order for $N\leq 3$ to antiferromagnetic order for $N\geq 4$, which is similar to previous calculations for zigzag hexagonal graphene dots.[@rossier07; @rossier08] No magnetic ordering for zigzag segments shorter than three unit cells is found because of the close proximity of the opposite sublattice imbalance on the adjacent sides of the ring inner edge. When this edge is short, the edge states on the different sides of the inner ring boundary are subject to strong hybridization, and therefore their energy is lifted from the Dirac point. Hence, spontaneous spin polarization does not occur, which is similar to the case of nanoribbons. [@dressel96]
{width="6.8182cm"}
For $N\geq 4$, the spatial spin symmetry is broken due to the [*e*]{}-[*e*]{} interaction. When the ring width decreases, the maximum magnetic moment, which is located near the middle of the zigzag edge segment increases. Furthermore, nonzero magnetization is build up on the outer ring edge, and it increases when the ring width decreases. However, as a consequence of the increasing influence of the outer edge with decreasing ring width, the difference between the distributions of the magnetic moments on the two edges is not large for $N=10$ and $N=11$. Similarly, the staggered magnetization increases when the ring width decreases.
Figure 2 shows how the density of states (DOS) of the $9_{AC}:N_{ZG}$ rings (the cases depicted in Fig. 1) varies with $N$. The density of states for the noninteracting (interacting) case is displayed by the black (purple) lines. In order to align the interacting and noninteracting spectra for easier comparison we subtracted $\Delta E$ for each interacting spectrum. Note that the density of states is spin degenerate, which is in accordance with Lieb’s theorem. For $N=3$, magnetic order is not present, therefore the energy dependence of the density of states for the interacting and noninteracting systems coincide \[see Fig. 2 (a)\]. The interacting and noninteracting electron case exhibit a small difference in the energy dependence of the DOS for rings with $N=4$ and $N=5$, which is shown in Figs. 2(b) and 2(c). As could be inferred from Fig. 1(c), the magnetization along the inner ring edge is rather small for these values of $N$. For larger $N$, the discrepancy between the DOS’s for the interacting and non-interacting systems becomes larger, as demonstrated by Figs. 2(d)-(f) for $N=9$, $10$, and $11$. In all these cases, appreciable DOS for the noninteracting system is found around zero energy. Such a configuration becomes unstable in the presence of [*e*]{}-[*e*]{} interactions, which results in the appearance of an interaction gap.
{width="8.6cm"}
In order to demonstrate how the shape of the outer boundary affects the distribution of the magnetic moments in the ring, we show in Fig. 3 the magnetization in the $13_{ZG}:N_{ZG}$ rings. It is apparent that the shape of the outer edge has a large effect on the localization of the magnetic moments on this boundary (compare Figs. 1(c) and 3). It is clear that in the case of zigzag outer ring edge, the magnetization propagates much further into the ring.
Fig. 4 displays how $\mu_s^z$, $m_{max}$, and $\Delta E$ vary with the length of the side of the inner ring edge expressed by the number $N$. Along with the rings whose magnetic moment distributions were shown in Fig. 1(c) and Fig. 3, the case of a hexagonal graphene dot having zigzag edge, is also displayed in Fig. 4. Both the staggered magnetization $\mu_s^z$ and the energy shift $\Delta E$ increase with $N$, i.e. with the size of the inner ring, except for the extremely narrow $M_{ZG}:N_{ZG}$ rings. Interestingly, the staggered magnetization in the hexagonal quantum dots does not exceed 0.02, whereas for the $13_{ZG}:N_{ZG}$ ring it can reach almost up to $0.05$. The nearly twofold enlargement of the staggered magnetization could be accounted for by the double number of zigzag edges in the $M_{ZG}:N_{ZG}$ ring as compared to the $N_{ZG}$ graphene dot. On the other hand, most $9_{AC}:N_{ZG}$ rings exhibit larger staggered magnetization and all show larger maximum magnetic moment than the hexagonal graphene dot. As a matter of fact, in hexagonal graphene dots the zero-energy orbitals which are localized along the adjacent zigzag sides of the edge are oriented toward each other, whereas inner zigzag edges in rings face away from each other. Hence, hybridization between the states of the two edges is larger in the former case than in the latter case. This is why $9_{AC}:N_{ZG}$ rings turn magnetic for shorter lengths of zigzag edges than hexagonal dots (four versus seven, respectively). The decrease of $m_{max}$ with $N$ for $13_{ZG}:N_{ZG}$ is due to the more effective hybridization between the quasi-zero-energy states localized on the inner and outer edges of the ring when the ring width decreases. Hence, the electron energy shifts from the band of zero energy states, and therefore magnetic ordering decays, which is manifested by a smaller $m_{max}$ in the $13_{ZG}:11_{ZG}$ ring than in the $11_{ZG}$ dot. The shapes of the $\Delta E(N)$ curves shown in Fig. 4(c) resemble the $\mu_s^z(N)$ and $m_{max}(N)$ curves in Figs. 4(a) and 4(b).
{width="8.6cm"}
In order to elucidate the difference between magnetic ordering in rings and dots, one may also analyze how the local density of states (LDOS) depends on the geometry of the structure. More specifically, the spatial distribution of the states close to zero energy determines how the magnetic moments evolve when the dimensions of the structures varies. In order to enhance the contribution of the low-energy states, we will compute the weighted LDOS (WLDOS): [@palac08] $$W_{i}=\sum_j e^{-\beta E_j^2}\vert\phi_{ji}\vert^2.$$ Here, $i$ indexes the lattice sites, $j$ labels the eigenstates, $\beta$ is the damping coefficient chosen as $1/\sqrt\beta=0.1$ eV, whereas $\phi_{ji}$ is the value of the probability amplitude of the $j$-th state at the site $i$. Such defined WLDOS assumes that the contribution of the states with $\vert E_j\vert > 0.1$ eV is negligible. The plots of the WLDOS in Fig. 5(a) illustrate how the edge states form when the inner $7_{ZG}$ hexagonal dot is cleaved out of the outer $7_{AC}$ hexagonal dot. The inner dot is separated from the ring by severing the bonds one by one between the dot and the ring. The number of severed bonds between the dot and the ring is explicitly shown in Fig. 5(a), and the dot edge is depicted by the blue line. The local sublattice imbalance accumulates quickly with the number of severed bonds, but no edge states emerge when the number of cut bonds is less than four. The edge states, which are depicted by red contours around the edge, are initially distributed evenly between the ring and the dot, but they extend more to the ring when the number of cut bonds increases.
To explore this finding in more detail, we show in Fig. 5(b) how the total WLDOS (full purple circles), which is the sum over the atomic sites in the dot (full blue circles) and the ring (empty red circles), varies with the number of severed bonds. Also, the DOS at zero energy is shown by the black triangles in Fig. 5(b). Notice that the variation of the WLDOS has a similar shape for each side of the ring’s inner edge, and that the WLDOS displays step-like features. These steps arise because the imbalance between the two local sublattices, found at the ring and dot sides of the newly formed edge, are maximized when the formation of each side of the rings inner edge is completed. The next side of the rings inner edge contains the opposite sublattice imbalance, and therefore the states on this side hybridize with the states on the previous side, which leads to a decrease of WLDOS. [@palac08] Note that after the first edge has been cut the ring and the dot WLDOSs start deviating from each other more strongly. This is because the hybridization in the dot is stronger, as the edge states on adjacent segments hybridize inward and towards each other. In the ring part the edge states face away from each other and hybridize radially outward, hence the hybridization is weaker. This is why the WLDOS in the former case experiences a decline with the beginning of each new edge segment, while in the latter case the WLDOS keeps growing. The gradual increase of WLDOS for both cases near the end of each segment is related to the accumulation of the local sublattice imbalances. This pattern reappears with each new zigzag segment, with the exception of the last bond, which after being cut results in the separation of the two structures. By the end, the WLDOS in the ring is much larger than WLDOS in the dot, which accounts for the fact that the rings exhibit a larger maximum magnetic moment and staggered magnetization than the dots. Figure 5(c) shows a stacked plot of DOSs for each resulting structure. Plots are stacked from the bottom up, with each subsequent line corresponding to a structure with one more bond cut. DOSs for structures depicted in Fig. 5(a) are colored red. It shows that only features near zero energy evolve in a similar fashion as do the WLDOSs during the separation of the ring and the dot. This justifies the damping of states higher than 0.1 eV in calculating the WLDOS, as they are not artifacts of the edge forming between the ring and the dot.
{width="8.6cm"}
Finally, we examine the influence of the edge deformations on somewhat larger structures; namely, the $25_{ZG}:20_{ZG}$ and $16_{AC}:20_{ZG}$ rings and the $20_{ZG}$ dot. Larger structures are considered here because they can be deformed in a larger number of ways than smaller structures analyzed in the rest of the paper. Defects are induced by randomly deforming the polygons which outline the perfect structure as shown in Fig. 6 (a). The amplitude of this deformation is itself a randomly selected number out of a specific range and the final structure is made up of all atoms that are enclosed by the deformed outline,[@futnota2] which is shown in Fig. 6(b). In order to quantify the amount of defects, the defect ratio $f_{def}$ is defined as a fraction of the total number of the defects, which is a sum of the missing and the surplus sites, and the number of the sites in the original structure. The magnetic moment distributions in the $25_{ZG}:20_{ZG}$ ring for a few values of $f_{def}$ are shown in Fig. 6 (c). Also, variation of the staggered magnetization with the defect fraction for the $16_{AC}:20_{ZG}$ and $25_{ZG}:20_{ZG}$ rings and the $20_{ZG}$ dot is displayed in Fig. 6 (d). For the $25_{ZG}:20_{ZG}$ ring and the $20_{ZG}$ hexagon, $\mu_s^z$ decreases with defect fraction. This is expected, having in mind that the defects can only impair the conditions for magnetism in zigzag edges. On the other hand, for the $16_{AC}:20_{ZG}$ ring, small random defects are more likely to make the larger outer edge magnetic than to make the smaller inner edge nonmagnetic. This explains the initial rise in $\mu_s^z$ for $f_{def}$ up to 0.02.
In conclusion, we predict an antiferromagnetic phase in hexagonal graphene rings with zigzag inner edge within the mean-field Hubbard model. The distribution of magnetic moments is found to strongly depend on the type of outer edge, and larger antiferromagnetic order is found in rings than in hexagonal dots. Peculiar hybridization between the states of adjacent sides of the inner ring edge is found to lead to an increase of magnetization of rings with respect to dots. Also, the staggered magnetization and the maximum magnetic moment are found to be strongly influenced by the size and the shape of the rings. For wide rings, the maximum magnetic moment is largest when both the inner and outer edges are zigzag. But, as a consequence of the hybridization between the states of the two edges, the maximum magnetic moment in a ring with armchair outer edge exceeds the one for the zigzag outer edge when the ring width decreases. The staggered magnetization in both the hexagonal dots and the rings with zigzag outer edge is found to decrease faster than in the rings with armchair outer edge when the number of the edge defects increases.
This work was supported by the EuroGRAPHENE programme of the ESF (project CONGRAN), the Serbian Ministry of Education, Science, and Technological Development, and the Flemish Science Foundation (FWO-Vl).
[00]{} K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, and A. A. Firsov, Science [**306**]{}, 666 (2004). M. T. Ong and E. J. Reed, ACS Nano [**6**]{}, 1387 (2012). M. F. El-Kady, V. Strong, S. Dubin, and R. B. Kaner, Science [**335**]{}, 1326 (2012). T. J. Echtermeyer, L. Britnell, P. K. Jasnos, A. Lombardo, R. V. Gorbachev, A. N. Grigorenko, A. K. Geim, A. C. Ferrari, and K. S. Novoselov, Nat. Commun. [**2**]{}, 458 (2011). Y. M. Lin, C. Dimitrakopoulos, K. A. Jenkins, D. B. Farmer, H. Y. Chiu, A. Grill, and Ph. Avouris, Science [**327**]{}, 662 (2010). L. Britnell, R. V. Gorbachev, R. Jalil, B. D. Belle, F. Schedin, A. Mishchenko, T. Georgiou, M. I. Katsnelson, L. Eaves, S. V. Morozov, N. M. R. Peres, J. Leist, A. K. Geim, K. S. Novoselov, and L. A. Ponomarenko, Science [**335**]{}, 947 (2012). O. V. Yazuev, Rep. Prog. Phys. [**73**]{}, 056501 (2010). Y.-W. Son, M. L. Cohen, and S. G. Louie, Nature [**444**]{}, 347 (2006). J. Fernandez-Rossier and J. J. Palacios, Phys. Rev. Lett. [**99**]{}, 177204 (2007). J. Fernandez-Rossier, Phys. Rev. B [**77**]{}, 075430 (2008). J. J. Palacios, J. Fernandez-Rossier, and L. Brey, Phys. Rev. B [**77**]{}, 195428 (2008). H. S. S. Ramakrishna Matte, K. S. Subrahmanyam, and C. N. R. Rao, J. Phys. Chem. C [**113**]{}, 9982 (2009). T. Enoki and K. Takai, Solid State Commun. [**149**]{}, 1144 (2009). V. L. Joseph Joly, M. Kiguchi, Si-Jia Hao, K. Takai, T. Enoki, R. Sumii, K. Amemiya, H. Muramatsu, T. Hayashi, Y. A. Kim, M. Endo, J. Campos-Delgado, F. López-Urías, A. Botello-Méndez, H. Terrones, M. Terrones, and M. S. Dresselhaus, Phys. Rev. B [**81**]{}, 245428 (2010). A. Ney, P. Papakonstantinou, A. Kumar, N. -G. Shang, and N. Peng, Appl. Phys. Lett. [**99**]{}, 102504 (2011). R. R. Nair, M. Sepioni, I-Ling Tsai, O. Lehtinen, J. Keinonen, A. V. Krasheninnikov, T. Thomson, A. K. Geim, and I. V. Grigorieva, Nat. Phys. [**8**]{}, 199 (2012). Y. Wang, Y. Huang, Y. Song, X. Zhang, Y. Ma, J. Liang, and Y. Chen, Nano. Lett. [**9**]{}, 220 (2009). J. Hong, E. Bekyarova, P. Likang, W. A. de Heer, R. C. Haddon, and S. Khizroev, Sci. Rep. [**2**]{}, 624, DOI:10.1038/srep00624 (2012). J. Kunstmann, C. Özdoğan, A. Quandt, and H. Fehske, Phys. Rev. B [**83**]{}, 045414 (2011). O. V. Yazyev and M. I. Katsnelson, Phys. Rev. Lett. [**100**]{}, 047209 (2008). M. Wimmer, I. Adagideli, S. Berber, D. Toma´nek, and K. Richter, Phys. Rev. Lett. [**100**]{}, 177207 (2008). W. L. Wang, O. V. Yazyev, S. Meng, and E. Kaxiras, Phys. Rev. Lett. [**102**]{}, 157201 (2009). M. Ezawa, Physica E [**42**]{}, 703 (2010). X. Wei, M. Wang, Y. Bando and D. Golberg, ACS Nano [**5**]{}, 29162922 (2011). E. H. Lieb, Phys. Rev. Lett. [**62**]{}, 1201 (1989). P. Potasz, A. D. Guclu, O. Voznyy, J. A. Folk, and P. Hawrylak, Phys. Rev. B [**83**]{}, 174441 (2011). We also include the constant part $U/4-U/2\sum_{i,\sigma}n_{i,\sigma}$ in the Hamiltonian. T. O. Wehling, E. Şaşioğlu, C. Friedrich, A. I. Lichtenstein, M. I. Katsnelson, and S. Blügel, Phys. Rev. Lett. [**106**]{}, 236805 (2011). H. Feldner, Zi Yang Meng, Andreas Honecker, Daniel Cabra, Stefan Wessel and Fakher F. Assaad, Phys. Rev. B [**81**]{}, 115416 (2010). K. Nakada, M. Fujita, G. Dresselhaus, and M. S. Dresselhaus, Phys. Rev. B [**54**]{}, 17954 (1996). We remove all the dangling atoms which might appear after the edge is deformed. Also, we ensure that all the deformed structures have equal numbers of the A and B sublattice sites, thus our calculations always regarded the antiferromagnetic limit.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We define two related invariants for a $d$-dimensional local ring $(R,\mathfrak{m},{k})$ called syzygy and differential symmetric signature by looking at the maximal free splitting of reflexive symmetric powers of two modules: the top dimensional syzygy module $\operatorname{Syz}_R^{d}({k})$ of the residue field and the module of Kähler differentials $\Omega_{R/{k}}$ of $R$ over ${k}$. We compute these invariants for two-dimensional ADE singularities obtaining $1/|G|$, where $|G|$ is the order of the acting group, and for cones over elliptic curves obtaining $0$ for the differential symmetric signature. These values coincide with the F-signature of such rings in positive characteristic.'
address:
- '[Holger Brenner, Institut für Mathematik, Universität Osnabrück, Albrechtstrasse 28a, 49076 Osnabrück, Germany]{}'
- '[Alessio Caminata, Institut für Mathematik, Universität Osnabrück, Albrechtstrasse 28a, 49076 Osnabrück, Germany]{}'
author:
- Holger Brenner and Alessio Caminata
title: The symmetric signature
---
[^1]
Introduction {#introduction .unnumbered}
============
Let $(R,\mathfrak{m},{k})$ be a Noetherian local ring of prime characteristic $p$ and dimension $d$ which is F-finite, and with algebraically closed residue field ${k}$. For every $e\in\mathbb{N}$, let $q=p^e$, and let $^e\!R$ be the $R$-module which is equal to $R$ as abelian group and has scalar multiplication twisted via the $e$-th Frobenius homomorphism, that is $r\circ s:=r^{q}s$ for $r\in R$ and $s\in\ ^e\!R$. We decompose it as $$^e\!R=R^{a_q}\oplus M_q,$$ where the module $M_q$ has no free direct summands. The number $a_q$ is also called the *free rank* of $^e\!R$ and denoted by $\operatorname{frk}_R(^e\!R)$, while if $R$ is a domain, one has that $q^{d}=\operatorname{rank}_R(^e\!R)$, the usual rank of an $R$-module. Huneke and Leuschke [@HL02] defined the *F-signature* of $R$ as the limit $$s(R):=\lim_{e\rightarrow+\infty}\frac{a_q}{q^{d}}.$$
They proved that the limit defining the F-signature exists assuming that $R$ is Gorenstein and in other cases. Watanabe and Yoshida [@WY04] introduced the *minimal relative Hilbert-Kunz multiplicity*. Then, Yao [@Yao06] proved it actually coincides with the F-signature. Finally, Tucker [@Tuc12] proved that the F-signature exists for every reduced F-finite Noetherian local ring.
The F-signature is a real number between $0$ and $1$ and provides delicate information about the singularities of $R$. Two principal results in this direction are the fact that $s(R)=1$ if and only if the ring is regular [@WY00], and that $s(R) > 0$ if and only if $R$ is strongly F-regular [@AL03].
The main motivation for this paper is given by the following question.
Does there exist an invariant analogous to the F-signature which is independent of the characteristic and can be used to study singularities also in characteristic zero?
A possible natural attempt to define a characteristic zero version of the F-signature is the following. If $R$ is a reduced $\mathbb{Z}$-algebra such that $\mathrm{Spec} \,R\rightarrow\mathrm{Spec} \,\mathbb{Z}$ is dominant, then for every prime number $p$ we consider its reduction mod $p$, $R_p := R \otimes_{\mathbb{Z}} (\mathbb{Z}/p\mathbb{Z})$, and compute the F-signature $s(R_p)$. One may ask whether the limit $$\lim_{p\rightarrow+\infty} s(R_p)$$ exists, and use this limit to define a characteristic $0$ version of the F-signature.
Unfortunately, it is difficult to find an appropriate meaning and compute the previous limit, or even determine whether it exists at all. If $p_1$ and $p_2$ are distinct prime numbers, the two Frobenius homomorphisms are difficult to compare. If $R$ is a domain then the rank of $^e\!R_{p_1}$ and the rank of $^e\!R_{p_2}$ are different, the first one being $p_1^e$.
In this paper we define a new invariant called *symmetric signature* which works in any characteristic. As inspiration for our construction, we considered two important properties of the modules $^e\!R$ which we would like to keep. The first fact is a result of Kunz [@Kun69] which states that $^e\!R$ is a free module (for all or for some $e\in\mathbb{N}$) if and only if $R$ is a regular ring. Next, if the ring $R$ is Cohen-Macaulay then $^e\!R$ is a maximal Cohen-Macaulay (MCM for short) module.
Let $(R,\mathfrak{m},{k})$ be a Noetherian local ${k}$-algebra of dimension $d$. We look for an $R$-module with the property that it is free if and only if $R$ is regular. We have at least two modules with this property: the *top-dimensional syzygy module of the residue field* $\operatorname{Syz}^{d}_R({k})$, and the *module of (Kähler) differentials* $\Omega_{R/{k}}$ of $R$ over ${k}$ or its dual.
If $R$ is Cohen-Macaulay, then the module $\operatorname{Syz}^{d}_R({k})$ is MCM by the depth lemma, on the other hand $\Omega_{R/{k}}$ is not even reflexive in general. For this reason we prefer to work with its reflexive hull, $\Omega_{R/{k}}^{**}$, which is called *module of Zariski differentials of $R$ over ${k}$*. The sign $(-)^*$ stands for the functor $\operatorname{Hom}_R(-,R)$.
We want to study an asymptotic behaviour, so we construct two classes of modules (one for $\operatorname{Syz}^{d}_R({k})$, and one for $\Omega_{R/{k}}^{**}$) indexed by a natural number $q$. To do this we introduce the functor of $q$-th reflexive symmetric powers $\left(\operatorname{Sym}^q_R(-)\right)^{**}$, that is ordinary $q$-th symmetric powers of $R$-modules composed with the reflexive hull.
The choice of studying symmetric powers is motivated by an unpublished work of Brenner and Fischbacher-Weitz [@BF09]. Symmetric powers of cotangent bundles on projective varieties have been studied also by other authors such as Bogomolov and De Oliveira[@BDO08], and Sakai [@Sak78]. Instead, the reflexive hull is motivated by the fact that we want to stay inside the category of reflexive modules, which coincides with the category of MCM $R$-modules if $R$ is a normal two-dimensional domain.
More precisely, we procede as follows. We consider $\operatorname{Syz}^{d}_R({k})$ (for $\Omega_{R/{k}}^{**}$ the situation is analogous) and we apply to it the $q$-th reflexive symmetric powers functor for a natural number $q$. We obtain a reflexive $R$-module $$\mathcal{S}^q:=\left(\operatorname{Sym}_R^q\big(\operatorname{Syz}^{d}_R({k})\big)\right)^{**}$$
We decompose the module $\mathcal{S}^q$ as $$\mathcal{S}^q= R^{a_q} \oplus M_q,$$ with the module $M_q$ containing no free direct summands, so that $a_q=\operatorname{frk}_R\mathcal{S}^q$ the free rank of $\mathcal{S}^q$. For ease of notation we fix also $b_q:=\operatorname{rank}_R\mathcal{S}^q$, and we introduce the following (Definition \[defsymmetricsignature\]).
The real number $$s_{\sigma}(R) := \lim_{N\rightarrow+\infty} \frac{\sum_{q=0}^Na_q}{\sum_{q=0}^Nb_q}$$ is called *(syzygy) symmetric signature* of $R$, provided the limit exists.
Replacing the module $\operatorname{Syz}^{d}_R({k})$ with $\Omega_{R/{k}}^{**}$ in the previous construction we define also the *differential symmetric signature* of $R$, which we denote by $s_{d\sigma}(R)$.
In order to understand the behaviour of these new invariants in comparison with the F-signature we concentrate on two important classes of examples: two-dimensional Kleinian singularities, and coordinate rings of plane elliptic curves.
Two-dimensional Kleinian or ADE singularities are the rings of invariants $R=S^G$ of a power series ring in two variables $S={k}\llbracket u,v\rrbracket$ under the linear action of certain finite subgroups $G$ of ${\mathrm{SL}(2,{k})}$, called Klein groups. The field ${k}$ is algebraically closed and such that its characteristic does not divide the order of the acting group $G$. Watanabe and Yoshida [@WY04] proved that if the characteristic of ${k}$ is a prime number then the F-signature of $R$ is $s(R)=\frac{1}{|G|}$. We prove that the same holds for the symmetric and the differential symmetric signature in any characteristic (Corollary \[corollaryKleinian\]), that is $$s_{\sigma}(R)=s_{d\sigma}(R)=\frac{1}{|G|}.$$
Our main tool to prove this result is the so-called *Auslander correspondence* (Theorem \[Auslandertheorem\]). This states that for a Kleinian singularitiy $R=S^G$ there is a functorial one to one correspondence between irreducible ${k}$-representations of $G$ and indecomposable maximal Cohen-Macaulay $R$-modules. The module corresponding to a ${k}$-representation $V$ is given by $\mathcal{A}(V):=(S\otimes_{{k}}V)^G$.
We prove that this construction commutes with $q$-th reflexive symmetric powers (Theorem \[symcommuteswithauslandertheorem\]), that is we have $$\mathcal{A}\left(\operatorname{Sym}_{{k}}^q(V)\right)\cong\left(\operatorname{Sym}_R^q(\mathcal{A}(V))\right)^{**}.$$ Here, $\operatorname{Sym}_R^q(-)$ denotes the $q$-th symmetric power of an $R$-module, and $\operatorname{Sym}_{{k}}^q(V)$ is the $q$-th symmetric power of the representation $V$.
This result is the key to translate the problem of computing the symmetric signature into a problem in representation theory of finite groups. In fact, it turns out that the free rank of some MCM $R$-module $M=\mathcal{A}(V)$ is equal to the multiplicity of the trivial representation into $V$, and the latter can be more easily computed using tools from representation theory, such as characters.
The symmetric signature of two-dimensional cyclic quotient singularities ${k}\llbracket u,v\rrbracket^G$, with $G$ cyclic, is also $1/|G|$ and is computed by Katthän and the second author in the paper [@CK16].
Let $R$ be the coordinate ring of a plane elliptic curve over an algebraically closed field ${k}$. Then $R$ is not strongly F-regular, therefore its F-signature is $s(R)=0$. We prove that the differential symmetric signature of $R$ is $s_{d\sigma}(R)=0$ (Theorem \[Theocotangentelliptic\]), provided that $\operatorname{char}{k}$ is not $2$ or $3$. For the syzygy symmetric signature we give an upper bound, that is $s_{\sigma}(R)\leq\frac{1}{2}$ (Corollary \[corsleq12elliptic\]), provided that the limit exists.
The methods we use in this situation are geometric. We use the correspondence between graded MCM $R$-modules and vector bundles over the smooth projective curve $Y=\mathrm{Proj}\,R$, and we translate the problem of computing $s_{\sigma}(R)$ and $s_{d\sigma}(R)$ into an analogous problem in the category $\mathrm{VB}(Y)$ of vector bundles over $Y$. The main advantage of this approach is that over an elliptic curve $Y$ the indecomposable vector bundles have been classified and described by Atiyah [@Ati57].
The structure of this paper is the following.
In Section \[sectionFandsymmetric\] we give a short review of the F-signature and its properties. Then, we define the symmetric and differential symmetric signatures, together with some other variants: a generalized version for modules (Definition \[defgeneralizedsymmetric\]), and one for graded rings (Definition \[defgradedsymmetric\]). Some easy properties and consequences of the symmetric signature are also stated in this section.
In Section \[sectionAuslandersymmetric\] we review the theory of the Auslander correspondence and we prove that the Auslander functor commutes with reflexive symmetric powers (Theorem \[symcommuteswithauslandertheorem\]).
Section \[sectionKleinian\] is dedicated to the computation of the symmetric signatures of two-dimensional Kleinian singularities, and Section \[sectionelliptic\] to the computation of the symmetric signature for coordinate rings of plane elliptic curves.
F-signature and Symmetric signature {#sectionFandsymmetric}
===================================
F-signature {#subsectionFsignature}
-----------
We recall some general facts about rings of prime characteristic, and the definition and basic properties of the F-signature. Let $R$ be a commutative ring containing a field of prime characteristic $p$. With the letter $e$ we always denote a natural number, and with $q=p^e$ a power of the characteristic. The *Frobenius homomorphism* is the ring homomorphism $F:R\rightarrow R$, $F(r)=r^p$, we often consider also its iterates $F^e:R\rightarrow R$, $F^e(r)=r^{q}$. For any finitely generated $R$-module $M$, we denote by $^e\! M$ the $R$-module $M$, whose multiplicative structure is pulled back via $F^e$. The scalar multiplication on $^e\!M$ is given by $r\cdot m:=r^qm$, for any $r\in R$, $m\in \ \! ^e\!M$.
We say that $R$ is *F-finite* if $^1\!R$ is a finitely generated $R$-module, and we say that $R$ is *F-split* if the Frobenius map $F:R\rightarrow\ ^1\!R$ splits. If $(R,\mathfrak{m},{k})$ is a complete Noetherian local ring, then $R$ is F-finite if and only if $[{k}^{1/p}:{k}]$ is finite. We will always assume that our rings are F-finite, and that ${k}$ is perfect, that is $[{k}^{1/p}:{k}]=1$. The latter is not a strong restriction, but it is done simply to avoid that numbers like $[{k}^{1/p}:{k}]$ appear in the definitions and in the results.
\[eRisMCM\] If $R$ is F-finite and Cohen-Macaulay, then $ ^e\!R$ is a maximal Cohen-Macaulay module for every $e\in\mathbb{N}$. Actually, if $x_1,\dots,x_n$ is a maximal $R$-regular sequence, then it is also an $ ^e\!R$-regular sequence for every $e\in\mathbb{N}$.
Let $(R,\mathfrak{m},{k})$ be a $d$-dimensional Noetherian reduced local ring of prime characteristic $p$, which is F-finite and such that ${k}$ is perfect. We recall the definition of *free rank* of an $R$-module $M$ $$\operatorname{frk}_R(M):=\max\{n: \ \exists \text{ a split surjection } \varphi:M\twoheadrightarrow R^n\}.$$ Since the cancellation property holds for finitely generated modules over a local ring (cf. [@LW12 Corollary 1.16]), $\operatorname{frk}_R(M)$ is the unique integer such that we have a decomposition $$M\cong R^{\operatorname{frk}_R(M)}\oplus N,$$ and the module $N$ contains no free $R$-direct summands.
The *F-signature* of $R$, denoted by $s(R)$ is the limit $$s(R):=\lim_{e\rightarrow+\infty}\frac{\operatorname{frk}_R(^e\!R)}{q^d}.$$
Tucker [@Tuc12] proved that the F-signature exists for every reduced F-finite Noetherian local ring.
If $R$ is a domain of dimension $d$, then $\operatorname{rank}_R(^e\!R)=q^d$. So we can write the limit defining the F-signature as $$s(R)=\lim_{e\rightarrow+\infty}\frac{\operatorname{frk}_R (^e\!R)}{\operatorname{rank}_R(^e\!R)}.$$
The F-signature is always a real number in the interval $[0,1]$. Using a result of Watanabe and Yoshida [@WY00], Huneke and Leuschke [@HL02] proved that if $R$ is Cohen-Macaulay then the extreme value $1$ is obtained if and only if the ring is regular. Then, Yao [@Yao06] removed the Cohen-Macaulay assumption.
\[WatYoshTheorem1\] Let $(R,\mathfrak{m},{k})$ be a reduced F-finite local ring of prime characteristic such that ${k}$ is infinite and perfect. Then $s(R)=1$ if and only if $R$ is regular.
Also the value $s(R)=0$ has a special meaning in terms of the singularities of the ring, it is equivalent to the ring being not strongly F-regular, an important notion in tight closure theory.
\[AbeLeusTheorem\] Let $(R,\mathfrak{m},{k})$ be a reduced excellent F-finite local ring of prime characteristic such that ${k}$ is perfect. Then $s(R)>0$ if and only if $R$ is strongly F-regular.
If the ring $R$ has dimension $0$ or $1$, then it is normal if and only if it is regular if and only if it is strongly F-regular. So by Theorem \[WatYoshTheorem1\] and Theorem \[AbeLeusTheorem\] we have only two possibilities for the F-signature of $R$. We have $s(R)=1$ if $R$ is regular, and $s(R)=0$ otherwise.
Theorem \[WatYoshTheorem1\] and Theorem \[AbeLeusTheorem\] justify the statement that *F-signature measures singularities*. Roughly speaking, the closer the F-signature to $1$ is, the nicer the singularity. An important example is the F-signature of quotient singularities computed by Watanabe and Yoshida.
\[WatYoshTheorem2\] Let $R={k}\llbracket x_1,\dots x_n\rrbracket^G$ be a quotient singularity over a field ${k}$ of prime characteristic $p$. Assume that the acting group $G\subseteq{\mathrm{GL}(n,{k})}$ is a small finite group such that $(p,|G|)=1$. Then the F-signature of $R$ is $$s(R)=\frac{1}{|G|}.$$
We conclude with the notion of generalized F-signature, introduced by Hashimoto and Nakajima [@HN15]. Let $\mathcal{C}$ be a full subcategory of ${\mathrm{mod}(R)}$, the category of finitely generated $R$-modules, such that $\mathcal{C}$ has the Krull-Remak-Schmidt property (KRS property for short) and $^e\!R\in\mathcal{C}$ for every $e\in\mathbb{N}$. For example, if $R$ is complete we may choose $\mathcal{C}={\mathrm{mod}(R)}$. Let $M$ be an indecomposable object in $\mathcal{C}$, for each $e\in\mathbb{N}$ and $q=p^e$ we can consider the multiplicity $a_{q}$ of $M$ inside $^e\!R$. In other words, we can write $^e\!R=M^{a_q}\oplus N_q$, with the module $N_q$ containing no copies of $M$ as direct summands.
The *generalized F-signature of $R$ with respect to $M$* is $$s(R,M):=\lim_{e\rightarrow+\infty}\frac{a_q}{q^{d}},$$ provided the limit exists.
If the ring $R$ has finite F-representation type, then the generalized F-signature exists. This is a consequence of [@SVB96 Proposition 3.3.1] and of [@Yao05 Theorem 3.11]. Moreover, Seibert [@Sei97] proved that in this situation the F-signature and the Hilbert-Kunz multiplicity of $R$ are rational numbers.
\[YusukeTheorem\] Let $R={k}\llbracket x_1,\dots x_n\rrbracket^G$ be a quotient singularity over an algebraically closed field ${k}$ of prime characteristic $p$. Assume that the acting group $G\subseteq{\mathrm{GL}(n,{k})}$ is a small finite group such that that $(p,|G|)=1$. Let $V_t$ be an irreducible ${k}$-representation of $G$ and let $M_t=\mathcal{A}(V_t)=(S\otimes_{{k}}V_t)^G$ be the corresponding $R$-module via Auslander functor. Then the generalized F-signature of $R$ with respect to $M_t$ is $$s(R,M_t)=\frac{\operatorname{rank}_R M_t}{|G|}.$$
The definition of the Auslander functor $\mathcal{A}$ and the relation between ${k}$-representations of $G$ and reflexive modules over the invariant ring $R$ will be reviewed in Section \[sectionAuslandersymmetric\].
Symmetric signature {#subsectionsymmetricsignature}
-------------------
For this section, let $(R,\mathfrak{m},{k})$ be a local Noetherian ${k}$-domain of dimension $d$ (of any characteristic) and let $q$ be a natural number. We consider the top-dimensional syzygy module $\operatorname{Syz}^d_R({k})$ of the residue field, coming from a minimal free resolution of ${k}$, and the module of (Kähler) differentials of $R$ over ${k}$, denoted by $\Omega_{R/{k}}$.
It is an easy consequence of the Auslander-Buchsbaum-Serre theorem, and of Auslander-Buchsbaum formula that $\operatorname{Syz}^d_R({k})$ is $R$-free if and only if $R$ is regular. Similarly, under some mild conditions we have that $\Omega_{R/{k}}$ is $R$-free if and only if $R$ is regular (see e.g. [@Har77 Chapter II, Theorem 8.8]). Moreover from the depth lemma follows that if $R$ is Cohen-Macaulay then $\operatorname{Syz}^d_R({k})$ is maximal Cohen-Macaulay, and in particular reflexive. On the other hand $\Omega_{R/{k}}$ is not even reflexive in general (see e.g. [@Her78a]), so it is convenient to work with its reflexive hull $\Omega_{R/{k}}^{**}$, which is called *module of Zariski (or regular) differentials of $R$ over ${k}$*.
Let $q$ be a natural number, we define the following two classes of $R$-modules. $$\begin{split}
\mathcal{S}^q:&=\left(\operatorname{Sym}_R^q\big(\operatorname{Syz}^d_R({k})\big)\right)^{**}, \\
\mathcal{C}^q:&=\left(\operatorname{Sym}^q_R\big(\Omega_{R/{k}}^{**}\big)\right)^{**}=\left(\operatorname{Sym}^q_R\big(\Omega_{R/{k}}\big)\right)^{**}.
\end{split}$$
\[defsymmetricsignature\] The real numbers $$s_{\sigma}(R):=\lim_{N\rightarrow+\infty}\frac{\sum_{q=0}^N\operatorname{frk}_R\mathcal{S}^q}{\sum_{q=0}^N\operatorname{rank}_R\mathcal{S}^q},$$ and $$s_{d\sigma}(R):=\lim_{N\rightarrow+\infty}\frac{\sum_{q=0}^N\operatorname{frk}_R\mathcal{C}^q}{\sum_{q=0}^N\operatorname{rank}_R\mathcal{C}^q}$$ are called *(syzygy) symmetric signature* of $R$, and *differential symmetric signature* of $R$ respectively, provided the limits exist.
In the rest of the paper we will study both the symmetric signature and the differential symmetric signature. For the clarity of exposition we will focus sometimes on one definition and we will say what applies also to the other.
We don’t know whether the limit defining the symmetric signature always exists. To obtain an invariant which is for sure well defined and exists, one may replace the limits in Definition \[defsymmetricsignature\] with limits inferior.
Moreover one may ask why should we consider the limits of Definition \[defsymmetricsignature\], insted of the simpler limit $$\label{symmetricsimplerlimit}
\lim_{q\rightarrow+\infty} \frac{\operatorname{frk}_R\mathcal{S}^q}{\operatorname{rank}_R\mathcal{S}^q}.$$ The main reason is that the limit does not exist even in simple cases, as we see in Example \[exampleAnlimitnotexists\]. However, when the simpler limit exists, then also the symmetric signature exists and they coincide.
Let $(a_n)_{n\in\mathbb{N}}$, $(b_n)_{n\in\mathbb{N}}$ be sequences of real numbers such that $b_n>0$ for all $n\in\mathbb{N}$. Assume that the infinite series $\sum_{n\in\mathbb{N}}b_n$ diverges. If the limit $\lim_{n\rightarrow+\infty}\frac{a_n}{b_n}$ exists, then also the limit $$\lim_{n\rightarrow+\infty}\frac{\sum_{k=0}^na_k}{\sum_{k=0}^nb_k}$$ exists and the two limits coincide.
The modules $\mathcal{S}_q$ and $\mathcal{C}_q$ are reflexive for every $q\geq0$. If $R$ is Cohen-Macaulay of dimension $\leq2$, then $\mathcal{S}_q$ and $\mathcal{C}_q$ are also MCM.
\[exampleregularsymmetricis1\] If $R$ is a regular ring of dimension $d$, then $\operatorname{Syz}^d_R({k})$ and $\Omega_{R/{k}}^{**}$ are free modules. Then $\mathcal{S}^q$ and $\mathcal{C}^q$ are also free, therefore $\operatorname{frk}_R\mathcal{S}^q=\operatorname{rank}_R\mathcal{S}^q$ and $\operatorname{frk}_R\mathcal{C}^q=\operatorname{rank}_R\mathcal{C}^q$for every $q$. It follows that the symmetric signatures are $s_{\sigma}(R)=1$ and $s_{d\sigma}(R)=1$. In particular, this happens if $R$ is of dimension $0$, since it is forced to be a field, being a domain.
If $(R,\mathfrak{m},{k})$ has dimension $1$, then the first syzygy of the residue field $\operatorname{Syz}_R^1({k})$ is just the maximal ideal $\mathfrak{m}$. Since $\mathfrak{m}$ is an ideal it has rank $1$, therefore also all its symmetric powers have rank $1$, and the same holds for the reflexive hull. In other words, we have $\operatorname{rank}_R\left(\operatorname{Sym}^q_R(\operatorname{Syz}^1_R({k})\right)^{**}=1$ for all $q\in\mathbb{N}$. Therefore, for each $q$ we have only two possibilities: either $\mathcal{S}^q\cong R$ and $\operatorname{frk}_R\mathcal{S}^q=\operatorname{rank}_R\mathcal{S}^q=1$ or $\mathcal{S}^q$ is not free and $\operatorname{frk}_R\mathcal{S}^q=0$.
Now assume in addition that the category ${\mathrm{Ref}(R)}$ of finitely generated reflexive modules has the KRS property. We fix an indecomposable object $M$ in ${\mathrm{Ref}(R)}$, then for every $q\in\mathbb{N}$ we have a unique decomposition $$\mathcal{S}^q=M^{a_q}\oplus N_q,$$ where the module $N_q$ contains no copy of $M$ as direct summand. The natural number $a_q$ is called *the multiplicity of $M$ in $\mathcal{S}^q$*.
\[defgeneralizedsymmetric\] The number $$s_{\sigma}(R,M):=\lim_{N\rightarrow+\infty}\frac{\sum_{q=0}^Na_q}{\sum_{q=0}^N\operatorname{rank}_R\mathcal{S}^q}$$ is called *generalized symmetric signature of $R$ with respect to $M$*, provided the limit exists.
We can define the symmetric signature also for $\mathbb{N}$-graded rings.
Let $R$ be a standard graded Noetherian ${k}$-domain of dimension $d$. We work in the category ${\mathrm{mod}_{\mathbb{Z}}(R)}$ of finitely generated graded $R$-modules. In this category we consider the graded module $\operatorname{Syz}^d_R({k})$ coming from a minimal graded free resolution of ${k}$ as $R$-module, and the differential module $\Omega_{R/{k}}$, which is also naturally graded. Their $q$-th reflexive symmetric powers $\mathcal{S}^q$ and $\mathcal{C}^q$ are again graded modules. We also introduce a graded version of the free rank. For any finitely generated graded $R$-module $M$, we define the *graded free rank* of $M$ as $$\begin{split}
\operatorname{frk}_R^{\mathrm{gr}}(M):=\max\{n: \ \exists &\text{ a homogeneous of degree $0$ split surjection }\varphi: M\twoheadrightarrow F, \\ &\text{ with } F \text{ free graded $R$-module of rank } n\}.
\end{split}$$
The main reason to introduce this definition is that we want work in the Krull-Schmidt category ${\mathrm{mod}_{\mathbb{Z}}(R)}$ of finitely generated graded $R$-modules, whose maps are $R$-linear homomorphisms of degree $0$. Thus, the previous definition is more natural in this setting.
\[defgradedsymmetric\] The *graded (syzygy) symmetric signature* of $R$ is defined by the limit $$s_{\sigma}(R):=\lim_{N\rightarrow+\infty}\frac{\sum_{q=0}^N\operatorname{frk}^{\mathrm{gr}}_R\mathcal{S}^q}{\sum_{q=0}^N\operatorname{rank}_R\mathcal{S}^q},$$ provided it exists.
Replacing the module $\mathcal{S}^q$ with $\mathcal{C}^q$ in the definition above, we define also the graded version of the differential symmetric signature.
Notice that in the definition of $s_{\sigma}(R)$ for graded rings we consider the graded free rank instead of the free rank as in the local setting. Therefore some differences may occur. For example, one should remind that all direct summands of the form $R(a)$ for some integer $a$ contribute to the graded free rank part of the module.
Auslander correspondence and reflexive symmetric powers {#sectionAuslandersymmetric}
=======================================================
We fix the setting for this section. Let $S={k}\llbracket u,v\rrbracket$ be a power series ring in two variables over an algebraically closed field ${k}$. Let $G\subseteq{\mathrm{GL}(2,{k})}$ be a finite group which is small, i.e. it contains no pseudo-reflections, and such that the characteristic of ${k}$ and the order $|G|$ of $G$ are coprime. The group $G$ acts on $S$ via linear changes of variables, and we denote by $R:=S^G$ the invariant ring under this action. The invariant ring $R$ is called *quotient singularity* and has the following properties: it is a Noetherian local domain of dimension $2$, normal, Cohen-Macaulay, complete, and an isolated singularity (see e.g. [@BD08 Theorem 4.1]). Moreover, Watanabe [@Wat74a; @Wat74b] proved that $R$ is Gorenstein if and only if $G\subseteq{\mathrm{SL}(2,{k})}$. In this case $R$ is called *special quotient singularity*. If $G$ is a cyclic group $R$ is called *cyclic quotient singularity*. The *Kleinian singularities*, or *ADE singularities* are special quotient singularities, which actually coincides with them if the base field has characteristic zero.
In Section \[sectionKleinian\] we will compute the symmetric signature of Kleinian singularities. In this section we will review the theory of Auslander correspondence between linear ${k}$-representations of $G$ and MCM $R$-modules, and we will prove that the Auslander functor commutes with reflexive symmetric powers. For a more detailed survey of these facts the reader may consult the books of Yoshino [@Yos90], and of Leuschke and Wiegand [@LW12], or the second author’s Ph.D. thesis [@Cam16].
The Auslander correspondence
----------------------------
We denote by ${k}[G]$ the *group ring* of $G$ over ${k}$, that is the ${k}$-vector space with basis $\{e_g: g\in G\}$, and with multiplication given on the basis elements by group multiplication $e_ge_h=e_{gh}$ and extended linearly to arbitrary elements. The category ${\mathrm{mod}({k}[G])}$ of finitely generated left modules over the group ring can be identified with the category of linear ${k}$-representations of $G$. For this reason we will use indifferently the terminology of representation theory or the terminology of module theory over ${k}[G]$.
The *skew group ring* of $G$ and $S$ is denoted by $S*G$ and it is a free $S$-module on the elements of $G$ $S*G=\displaystyle\bigoplus_{g\in G}S g$, with a multiplicative structure given by $(s g)(t h):=sg(t)\cdot gh$ for all $s,t\in S$, $g,h\in G$.
The group ring and the skew group ring are in general non-commutative, so by ${k}[G]$-module and $S*G$-module we will always mean left module. We point out that an $S*G$-module is just an $S$-module with a compatible action of $G$: $g(sm)=g(s)g(m)$ for all $g\in G$, $s\in S$ and $m\in M$. The ring $S$ is clearly an $S*G$-module, but not every $S$-module has a natural $S*G$-module structure.
There is a natural functor $\mathcal{F}$ between the category ${\mathrm{mod}({k}[G])}$ and the category ${\mathrm{Proj}(S*G)}$ of finitely generated projective $S*G$-modules, namely $\mathcal{F}(V):=S\otimes_{{k}}V$ for every object $V$ in ${\mathrm{mod}({k}[G])}$. The functor $\mathcal{F}$ has a right adjoint given by $\mathcal{F}'(P):=P\otimes_{S}{k}$. The functors $\mathcal{F}$ and $\mathcal{F}'$ are an adjoint pair [@Yos90 Lemma 10.1], but in general they are not an equivalence of categories.
We remark that an $S*G$-module is $S*G$-projective if and only if it is $S$-projective, hence $S$-free, since $S$ is local.
We define two other functors $\mathcal{G}:{\mathrm{Proj}(S*G)}\rightarrow{\mathrm{Ref}(R)}$, $\mathcal{G}(M)=M^G$, and $\mathcal{G}':{\mathrm{Ref}(R)}\rightarrow{\mathrm{Proj}(S*G)}$, $\mathcal{G}'(N)=(S\otimes_RN)^{**}$, where $(-)^{*}:=\operatorname{Hom}_S(-,S)$, and ${\mathrm{Ref}(R)}$ denotes the category of finitely generated reflexive $R$-modules. Since $R$ has dimension $2$ the category ${\mathrm{Ref}(R)}$ coincides with the category ${\mathrm{MCM}(R)}$ of maximal Cohen-Macaulay $R$-modules, and also with the category $\mathrm{Add}_R(S)$ of $R$-direct summands of $S$ by a result of Herzog [@Her78b], that is ${\mathrm{Ref}(R)}={\mathrm{MCM}(R)}=\mathrm{Add}_R(S)$.
Auslander [@Aus86b] proved that the functors $\mathcal{G}$ and $\mathcal{G}'$ are an equivalence of categories. We will give a geometric proof of this fact, based on the following ring version of the Speiser’s Lemma of Galois theory due to Auslander and Goldman [@AG60], and Chase, Harrison, and Rosenberg [@CHR65].
\[speiserlemma\] Let $A$ be a commutative domain, $G$ a finite group of ring automorphisms of $A$ and denote by $B=A^G$ the invariant ring. Then the following facts are equivalent.
$A$ is a separable $B$-algebra.
For every $g\neq \mathrm{id}_G$ in $G$, and every maximal ideal $\mathfrak{p}$ in $A$, there exists $a\in A$ such that $a-g(a)\not\in\mathfrak{p}$.
The map $\varphi:A\otimes_BM^G\rightarrow M$ given by $\varphi(a\otimes m)=am$ is an isomorphism for every $A*G$-module $M$.
\[M=SxMGlemmadim2\] Let $S={k}\llbracket u,v\rrbracket$, let $G\subseteq{\mathrm{GL}(2,{k})}$ be a finite small group such that $(\operatorname{char}{k},|G|)=1$, and let $R=S^G$. Let $M$ be a projective $S*G$-module and consider the canonical map $$\delta: S\otimes_R M^G\rightarrow M,$$ given by $\delta(s\otimes m)=sm$. Then the following facts hold.
$\delta$ induces an isomorphism of coherent sheaves ${\ThisStyle{ \setbox0=\hbox{$\SavedStyleS\otimes_R M^G$} \stackengine{-.1\LMpt}{$\SavedStyleS\otimes_R M^G$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\rightarrow\widetilde{M}$ on the punctured spectrum $U'=\mathrm{Spec} \,S\setminus\{\mathrm{m}\}$, where $\mathfrak{m}=(u,v)$.
$\delta$ induces an isomorphism of $S*G$-modules $$(S\otimes_R M^G)^{**}\cong M.$$
The ring $S$ is a Cohen-Macaulay isolated singularity, so coherent sheaves associated to MCM $S$-modules are locally free on the punctured spectrum. Moreover we recall that projective $S$-modules are MCM in this case. Therefore it follows that property *1)* implies property *2)*.
We prove *1)*. Let $f_1,\dots,f_{\mu}$ be elements in $R$ which generate $\mathfrak{m}$ up to radical, that is $\sqrt{(f_1,\dots,f_{\mu})}=\mathfrak{m}$ as ideals in $S$. These elements exist, since the ring extension $R\hookrightarrow S$ is finite. Thus, we have an open covering $U'=\bigcup_{i}D(f_i)$.
We claim that for every $f=f_i$ the induced map $$\delta_f: S_f\otimes_{R_f}M_f^G\rightarrow M_f$$ is an isomorphism. We prove the claim, then the Lemma will follow from it. In fact, because we have a global homomorphism, we get a sheaf homomorphism defined on $U'$ which is locally an isomorphism, so it is forced to be an isomorphism on $U'$.
To prove the claim, let $f$ be one of the $f_i$’s. We check that condition *2)* of Lemma \[speiserlemma\] is true for $A=S_f$ and $B=R_f$. Let $\mathfrak{p}$ be a maximal ideal of $A$. If $a-g(a)\in\mathfrak{p}$ for every $a\in A$ and every $g\in G$, then we have $a\in\mathfrak{p}$ if and only if $g(a)\in\mathfrak{p}$, which is equivalent to say that $\mathfrak{p}$ is a fix point for the action of $G$ on $A$. On the other hand, since $G$ is small, its action on $U'$ is fixpoint-free, so we get a contradiction. Therefore $\delta_f$ is an isomorphism by Lemma \[speiserlemma\] above.
\[RandSGmodules\] The functors $\mathcal{G}:{\mathrm{Proj}(S*G)}\rightarrow{\mathrm{Ref}(R)}$ and $\mathcal{G}':{\mathrm{Ref}(R)}\rightarrow{\mathrm{Proj}(S*G)}$ give an equivalence of categories $${\mathrm{Proj}(S*G)}\cong{\mathrm{Ref}(R)}.$$
See [@Yos90 Proposition 10.9].
We compose the functors $\mathcal{F}$ and $\mathcal{G}$ and we obtain a functor $\mathcal{A}:{\mathrm{mod}({k}[G])}\rightarrow{\mathrm{Ref}(R)}$, $\mathcal{A}(V):=(S\otimes_{{k}}V)^G$ called *Auslander functor*. The Auslander functor has a right adjoint given by $\mathcal{A}'(N):=(S\otimes_RN)^{**}\otimes_{S}{k}$, for every reflexive $R$-module $N$. These functors give a one-one correspondence between ${k}$-representations of $G$ and maximal Cohen-Macaulay modules over $R$ called *Auslander correspondence*. We collect the properties of the functor $\mathcal{A}$ and of the Auslander correspondence in the following theorem and corollary.
\[Auslandertheorem\] The functors $\mathcal{A}:{\mathrm{mod}({k}[G])}\rightarrow{\mathrm{Ref}(R)}$ and $\mathcal{A}':{\mathrm{Ref}(R)}\rightarrow{\mathrm{mod}({k}[G])}$ have the following properties.
$\mathcal{A}(V)\cong\mathcal{A}(W)$ if and only if $V\cong W$.
$\mathcal{A}(V)$ is indecomposable in ${\mathrm{Ref}(R)}$ if and only $V$ is an irreducible representation.
$\mathcal{A}(\operatorname{Hom}_{{k}}(V,W))\cong\operatorname{Hom}_R(\mathcal{A}(V),\mathcal{A}(W))$ for every $V,W\in{\mathrm{mod}({k}[G])}$.
$\mathcal{A}(V\otimes_{{k}}W)\cong\left(\mathcal{A}(V)\otimes_R\mathcal{A}(W)\right)^{**}$ for every $V,W\in{\mathrm{mod}({k}[G])}$.
If $V_0$ is the trivial representation then $\mathcal{A}(V_0)=R$.
$\operatorname{rank}_R\mathcal{A}(V)=\dim_{{k}}V$ for every ${k}[G]$-module $V$.
\[corollarydecomposition\] Let $V_1,\dots,V_r$ be a complete set of non-isomorphic irreducible ${k}$-representations of $G$, and fix $N_i=\mathcal{A}(V_i)=(S\otimes_{{k}}V_i)^G$. Then $N_1,\dots,N_r$ is a complete set of non-isomorphic indecomposable MCM $R$-modules. Moreover let $V$ be a ${k}$-representation which decomposes as $$V=V_1^{n_1}\oplus\cdots\oplus V_r^{n_r},$$ for some natural numbers $n_i$. Then the MCM $R$-module $N:=\mathcal{A}(V)=(S\otimes_{{k}}V)^G$ decomposes as $$N=N_1^{n_1}\oplus\cdots\oplus N_r^{n_r}.$$
\[examplecyclicgroup3\] Let $C_n=<g>$ be the cyclic group of order $n$, with irreducible representations $(V_t,\rho_t)$ over an algebraically closed field ${k}$ given by $\rho_t(g)=\xi^t$, for a fixed primitive $n$-th root of unity $\xi\in{k}$. Assume that $\operatorname{char}{k}$ does not divide $|G|$. We can embed $G$ into ${\mathrm{GL}(2,{k})}$ via the representation $V_1\oplus V_a$, where $a$ is a natural number such that $(a,n)=1$, otherwise the representation is not faithful. In other words, we consider the cyclic group generated by $$\begin{pmatrix}
\xi&0\\0&\xi^a
\end{pmatrix}.$$ This group acts linearly on $S={k}\llbracket u,v\rrbracket$ and the invariant subring $R$ is generated by monomials $u^iv^j$ such that $i+aj\cong0\mod n$.
For each irreducible representation $V_t$, we have an indecomposable MCM $R$-module $M_t=(S\otimes_{{k}}V_t)^G$. A straightforward computation shows that this is given by $$M_t=R\left(u^iv^j: \ i+aj\cong-t\mod n\right).$$
Auslander functor and reflexive symmetric powers
------------------------------------------------
We want to prove that the Auslander functor commutes with $q$-th reflexive symmetric powers $\operatorname{Sym}^q(-)^{**}$, that is $$\mathcal{A}(\operatorname{Sym}_{{k}}^q(V))\cong(\operatorname{Sym}_R^q(\mathcal{A}(V)))^{**}$$ for every ${k}[G]$-module $V$. Since the Auslander functor is the composition of the functors $\mathcal{F}$ and $\mathcal{G}$ we will split the proof of this fact in two propositions.
Observe that in the categories ${\mathrm{mod}({k}[G])}$ and ${\mathrm{Proj}(S*G)}$ the reflexive symmetric powers coincide with the usual symmetric powers. In fact, every finitely generated ${k}[G]$-module $V$ is reflexive, so it is canonically isomorphic to its double dual $V^{**}$. Since $S$ is regular, for every finitely generated projective $S*G$-module $N$ the symmetric powers $\operatorname{Sym}^q_S(N)$ are $S$-free, hence reflexive.
\[symcommuteswithauslanderprop1\] For every ${k}[G]$-module $V$ we have an isomorphism of $S*G$-modules $$\operatorname{Sym}_S^q(S\otimes_{{k}}V)\cong S\otimes_{{k}}\operatorname{Sym}^q_{{k}}(V).$$
From [@Nor84 (6.5.1)] we have an isomorphism of $S$-modules $\psi:\operatorname{Sym}_S^q(S\otimes_{{k}}V)\rightarrow S\otimes_{{k}}\operatorname{Sym}^q_{{k}}(V)$ given by $\psi((a_1\otimes v_1)\circ\dots\circ (a_q\otimes v_q))=(a_1\cdots a_q)\otimes(v_1\circ\dots\circ v_q)$. The only thing that we have to check is that $\psi$ is compatible with the action of $G$, and this is shown by the following commutative diagram $$\begin{tikzcd}
&(a_1\otimes v_1)\circ\dots\circ (a_q\otimes v_q) \arrow{r}{\psi}\arrow{d}{g}
&(a_1\cdots a_q)\otimes(v_1\circ\dots\circ v_q) \arrow{d}{g}\\
&\left(g(a_1)\otimes g(v_1)\right)\circ\dots\circ\left(g(a_q)\otimes g(v_q)\right) \arrow{r}{\psi}
&\begin{split}&g(a_1\dots a_q)\otimes g(v_1\circ\dots\circ v_q)= \\ &\left(g(a_1)\cdots g(a_q)\right)\otimes \left(g(v_1)\circ\dots\circ g(v_q)\right).\end{split}
\end{tikzcd}$$
We will use often the following well-known lemma. We write it here for ease of reference.
\[reflexivehullpuncturedlemma\] Let $(R,\mathfrak{m},{k})$ be a normal two-dimensional local domain, and let $U=\mathrm{Spec} \,R\setminus\{\mathfrak{m}\}$ be the punctured spectrum of $R$. For a torsion-free finitely generated $R$-module $M$ we have an isomorphism $$M^{**}\cong\Gamma(U,\widetilde{M}).$$
\[identificationremark\] Let $S={k}\llbracket u,v\rrbracket$ and $R=S^G$ be as in the rest of the section. Consider the following commutative diagram $$\begin{CD}
U'@>>>\mathrm{Spec} \,S \\
@V\pi VV @V\pi VV \\
U@>>>\mathrm{Spec} \,R
\end{CD}$$ where the map $\pi$ is induced by the inclusion $R\hookrightarrow S$, $U$ is the punctured spectrum of $R$, and $U'=\pi^{-1}(U)$ is the pull-back of $U$ to $\mathrm{Spec} \,S$, which is actually the punctured spectrum of $S$. For every $R$-module $M$ this gives us the following identification of sheaves on $U'$ $$\label{identification}
\pi^{*}(\widetilde{M}|_{U})\cong\left.{\ThisStyle{ \setbox0=\hbox{$\SavedStyleS\otimes_RM$} \stackengine{-.1\LMpt}{$\SavedStyleS\otimes_RM$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\right|_{U'}.$$
\[symcommuteswithauslanderprop2\] Let $N$ be a projective $S*G$-module, then we have an isomorphism of $R$-modules $$(\operatorname{Sym}_S^q(N))^G\cong\left(\operatorname{Sym}_R^q(N^G)\right)^{**}.$$
Let $\mathcal{G}'(-)=(S\otimes_R-)^{**}$ be the right adjoint of the $G$-invariants functor $\mathcal{G}(-)=(-)^G$. Since $\mathcal{G}$ and $\mathcal{G}'$ are an equivalence of categories (Theorem \[RandSGmodules\]), it is enough to show that $$\mathcal{G}'\left((\operatorname{Sym}_S^q(N))^G\right)\cong\mathcal{G}'\left(\left(\operatorname{Sym}_R^q(N^G)\right)^{**}\right),$$ that is $$\label{aussymequat1}
\left(S\otimes_R\left(\operatorname{Sym}^q_S(N)\right)^G\right)^{**}\cong\left(S\otimes_R\left(\operatorname{Sym}_R^q(N^G)\right)^{**}\right)^{**},$$ as $S*G$-modules.
Notice that the two double duals on the right hand side of are different: the first one is the double dual in ${\mathrm{mod}(R)}$ and the second is the double dual in ${\mathrm{mod}(S)}$.
In order to prove , we consider the left hand side and the right hand side separately. From the second part of Lemma \[M=SxMGlemmadim2\], the left hand side of is $\left(S\otimes_R\left(\operatorname{Sym}^q_S(N)\right)^G\right)^{**}\cong\operatorname{Sym}^q_S(N)$.
For the $S$-module on the right hand side of we use Lemma \[reflexivehullpuncturedlemma\], and we interpret it as the evaluation of the sheaf $$\left.{\ThisStyle{ \setbox0=\hbox{$\SavedStyleS\otimes_R\left(\operatorname{Sym}_R^q(N^G)\right)^{**}$} \stackengine{-.1\LMpt}{$\SavedStyleS\otimes_R\left(\operatorname{Sym}_R^q(N^G)\right)^{**}$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\right|_{U'}$$ on the punctured spectrum $U'$ of $S$. From the commutative diagram of Remark \[identificationremark\] we get the isomorphism $$\left.{\ThisStyle{ \setbox0=\hbox{$\SavedStyleS\otimes_R\left(\operatorname{Sym}^q_R(N^G)\right)^{**}$} \stackengine{-.1\LMpt}{$\SavedStyleS\otimes_R\left(\operatorname{Sym}^q_R(N^G)\right)^{**}$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\right|_{U'}\cong\pi^{*}\left(\left.{\ThisStyle{ \setbox0=\hbox{$\SavedStyle\operatorname{Sym}_R^q(N^G)^{**}$} \stackengine{-.1\LMpt}{$\SavedStyle\operatorname{Sym}_R^q(N^G)^{**}$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\right|_{U}\right)=\pi^{*}\left(\left.{\ThisStyle{ \setbox0=\hbox{$\SavedStyle\operatorname{Sym}_R^q(N^G)$} \stackengine{-.1\LMpt}{$\SavedStyle\operatorname{Sym}_R^q(N^G)$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\right|_{U}\right),$$ where we can remove the double dual over $U$, thanks to Lemma \[reflexivehullpuncturedlemma\]. Since taking symmetric powers commute with sheafification and with the restriction map of sheaves we get $$\pi^{*}\left(\left.{\ThisStyle{ \setbox0=\hbox{$\SavedStyle\operatorname{Sym}_R^q(N^G)$} \stackengine{-.1\LMpt}{$\SavedStyle\operatorname{Sym}_R^q(N^G)$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\right|_{U}\right)\cong \pi^{*}\left(\left.\left(\operatorname{Sym}^q_{X}(\widetilde{N^G})\right)\right|_{U}\right)\cong \pi^{*}\left(\operatorname{Sym}^q_{X}\big(\widetilde{N^G}|_{U}\big)\right),$$ where the second and the third symmetric powers are sheaf symmetric powers taken over $X=\mathrm{Spec} \,R$. We set $Y=\mathrm{Spec} \,S$. Since sheaf symmetric powers commute with pullback we have $$\pi^{*}\left(\operatorname{Sym}^q_{X}\big(\widetilde{N^G}|_{U}\big)\right)\cong\operatorname{Sym}_{Y}^q\left(\pi^{*}\big(\widetilde{N^G}|_{U}\big)\right).$$ We apply again and Lemma \[reflexivehullpuncturedlemma\] to obtain $$\operatorname{Sym}_{Y}^q\left(\pi^{*}\big(\widetilde{N^G}|_{U}\big)\right)\cong\operatorname{Sym}_{Y}^q\left(\left.{\ThisStyle{ \setbox0=\hbox{$\SavedStyleS\otimes_RN^G$} \stackengine{-.1\LMpt}{$\SavedStyleS\otimes_RN^G$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\right|_{U'}\right)=\operatorname{Sym}_{Y}^q\left(\left.{\ThisStyle{ \setbox0=\hbox{$\SavedStyle(S\otimes_RN^G)^{**}$} \stackengine{-.1\LMpt}{$\SavedStyle(S\otimes_RN^G)^{**}$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\right|_{U'}\right).$$ Since taking symmetric powers commutes with sheafification and with the restriction map of sheaves we get $$\operatorname{Sym}_{Y}^q\left(\left.{\ThisStyle{ \setbox0=\hbox{$\SavedStyle(S\otimes_RN^G)^{**}$} \stackengine{-.1\LMpt}{$\SavedStyle(S\otimes_RN^G)^{**}$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\right|_{U'}\right)\cong \left.{\ThisStyle{ \setbox0=\hbox{$\SavedStyle\operatorname{Sym}_S^q\left(\left(S\otimes_RN^G\right)^{**}\right)$} \stackengine{-.1\LMpt}{$\SavedStyle\operatorname{Sym}_S^q\left(\left(S\otimes_RN^G\right)^{**}\right)$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\right|_{U'}\cong\left.{\ThisStyle{ \setbox0=\hbox{$\SavedStyle\operatorname{Sym}_S^q\left(N\right)$} \stackengine{-.1\LMpt}{$\SavedStyle\operatorname{Sym}_S^q\left(N\right)$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\right|_{U'},$$ where the last isomorphism follows from Lemma \[M=SxMGlemmadim2\]. Taking global sections on $U'$ we obtain that the right hand side of is also isomorphic to $\operatorname{Sym}^q_S(N)$. Therefore we have an isomorphism of $S$-modules as in . This is actually an isomorphism of $S*G$-modules. In fact, the natural map of $R$-modules $\operatorname{Sym}^q_R(N^G)\rightarrow\left(\operatorname{Sym}_S^q(N)\right)^G$ induces a natural map $S\otimes_R\operatorname{Sym}^q_R(N^G)\rightarrow S\otimes_R\left(\operatorname{Sym}_S^q(N)\right)^G$, which is $G$-compatible since the action is just on $S$.
From Proposition \[symcommuteswithauslanderprop1\] and Proposition \[symcommuteswithauslanderprop2\] we immediately obtain the following.
\[symcommuteswithauslandertheorem\] Let $V$ be a ${k}[G]$-module, and let $M=(S\otimes_{{k}}V)^G$ be the corresponding MCM $R$-module via Auslander functor. Then we have $$\operatorname{Sym}_R^q(M)^{**}\cong \left(S\otimes_{{k}}\operatorname{Sym}_{{k}}^q(V)\right)^G.$$ In other words $\operatorname{Sym}_R^{q}(\mathcal{A}(V))^{**}\cong\mathcal{A}(\operatorname{Sym}^{q}_{{k}}(V))$.
Symmetric signature of Kleinian singularities {#sectionKleinian}
=============================================
In this section we will compute the generalized symmetric signature (and differential symmetric signature) of two-dimensional Kleinian singularities. These are invariant rings $R:={k}\llbracket u,v\rrbracket^G$, where ${k}$ is an algebraically closed field, and $G$ is one of the following finite small subgroups of ${\mathrm{SL}(2,{k})}$: the cyclic group $C_n$ of order $n$, the binary dihedral group $BD_n$ of order $4n$, the binary tetrahedral group $BT$ of order $24$, the binary octahedral group $BO$ of order $48$, and the binary icosahedral group $BI$ of order $120$. We always assume that the characteristic of ${k}$ does not divide the order of the group $G$. It is an old result of Klein that if ${k}$ has characteristic zero, then every finite subgroup of ${\mathrm{SL}(2,{k})}$ is isomorphic to one of these groups. For this reason these groups are also called *Klein groups*.
The Kleinian singularities are hypersurface rings, that is they are isomorphic to a quotient ring ${k}\llbracket x,y,z\rrbracket/(f)$ for some polynomial $f$, according to the following table.
\[table\]
singularity name $G$ $|G|$ $f$
------------------ -------------------- ------- --------------------
$A_{n-1}$ cyclic $n$ $y^{n}-xz$
$D_{n+2}$ binary dihedral $4n$ $x^2+y^{n+1}+yz^2$
$E_6$ binary tetrahedral $24$ $x^2+y^3+z^4$
$E_7$ binary octahedral $48$ $x^2+y^3+yz^3$
$E_8$ binary icosahedral $120$ $x^2+y^3+z^5$
The fundamental module of Kleinian singularities
------------------------------------------------
Let $(R,\mathfrak{m},{k})$ be a complete two-dimensional normal non-regular domain with canonical module $K_R$. In this setting Auslander [@Aus86b] proved that in the category ${\mathrm{Ref}(R)}$ there exists a unique non-split short exact sequence of the form $$0\rightarrow K_R\rightarrow E\rightarrow \mathfrak{m}\rightarrow 0,$$ which is called the *fundamental sequence* of $R$.
The module $E$ appearing in the middle term is also unique up to isomorphism and is called the *fundamental module* or *Auslander module* of $R$ (cf. [@Yos90 Chapter 11]). It is a reflexive (hence MCM) module of rank $2$. Moreover we have an isomorphism of $R$-modules $\left(\bigwedge^2 E\right)^{**}\cong K_R$.
The following example (cf. [@Yos90 Example 11.8]) clarifies the name fundamental module.
\[fundamentalmodulerepresentation\] Let $V$ be a ${k}$-vector space of dimension $2$ with basis $u$, $v$ and let $G\subseteq{\mathrm{GL}(2,{k})}$ be a finite subgroup. Then, $G$ acts on the power series ring $S={k}\llbracket u,v\rrbracket$ and we consider the invariant ring $R=S^G$, which is a two-dimensional quotient singularity. The fundamental module of $R$ is the image via Auslander functor of the fundamental representation $V$ of $G$, that is $$E=(S\otimes_{{k}}V)^G.$$
\[yoshinokawamoto1\] Let $R={k}\llbracket x,y,z\rrbracket/(f)$ be a complete two-dimensional normal non-regular domain with canonical module, with $f\in{k}\llbracket x,y,z\rrbracket$. Then the fundamental module $E$ is isomorphic to the third syzygy of ${k}$, i.e. $$E\cong\mathrm{Syz}^3_R({k}).$$ Moreover, the following facts are equivalent.
The fundamental module $E$ is decomposable.
$R$ is a cyclic quotient singularity.
Observe that the Kleinian singularities satisfy the condition of the previous Theorem \[yoshinokawamoto1\]. Now we prove that also the second syzygy of the residue field is isomorphic to the fundamental module.
\[Theoremsecondsyzygyfield1\] Let $(R,\mathfrak{m},{k})$ be a two-dimensional Kleinian singularity. Then the second syzygy of the residue field ${k}$ is isomorphic to the fundamental module, that is $$\operatorname{Syz}_R^2({k})\cong E.$$
From Example \[fundamentalmodulerepresentation\], it is clear that Theorem \[Theoremsecondsyzygyfield1\] is equivalent to the following statement.
\[Theoremsecondsyzygyfield2\] Let $(R,\mathfrak{m}_R,{k})$ be a two-dimensional Kleinian singularity, and let $V_1$ be the two-dimensional fundamental representation of the acting group $G$ of $R$. Then, the second syzygy of the residue field ${k}$ is isomorphic to the image of $V_1$ via Auslander functor, that is $$\operatorname{Syz}_R^2({k})\cong\mathcal{A}(V_1)$$
Theorem \[Theoremsecondsyzygyfield2\] can be proved by case considerations. We illustrate the general strategy and we apply it to the singularity $A_{n-1}$ in Example \[examplecyclicsyzygy\]. The reader interested to the computations for $D_{n+2}$, $E_6$, $E_7$ and $E_8$ may consult [@Cam16].
Let $S={k}\llbracket u,v\rrbracket$ be the powers series ring over an algebraically closed field ${k}$, and let $G$ be one of the Klein subgroups of ${\mathrm{SL}(2,{k})}$ acting on $S$ through a faithful representation $V_1$. Assume that $\operatorname{char}{k}$ and $|G|$ are coprime. Then we have $R=S^G$, and we denote by $\mathfrak{m}_R$ the maximal ideal of $R$. Let $p_1, p_2, p_3$ be a minimal system of generators of $\mathfrak{m}_R$ as ideal in $S$.
We consider the following short exact sequence of $S$-modules, which is the beginning of an $S$-free resolution of $ S/\mathfrak{m}_RS$ $$\label{syzygysequenceKlenian}
0\rightarrow \operatorname{Syz}^1_S(p_1,p_2,p_3)\rightarrow S^3 \xrightarrow{p_1,p_2,p_3} S\rightarrow S/\mathfrak{m}_RS\rightarrow0.$$ The group $G$ acts linearly on $S$ through the fundamental representation $V_1$, and this action extends naturally to $S^3$, and its submodule $\operatorname{Syz}^1_S(p_1,p_2,p_3)$. In other words, the sequence is an exact sequence of $S*G$-modules. We apply the functor $\mathcal{G}(N)=N^G$, which is exact, to the sequence and we obtain an exact sequence $$0\rightarrow M\rightarrow R^3 \rightarrow R\rightarrow R/\mathfrak{m}_R\rightarrow0.$$
Since $p_1,p_2,p_3$ are a system of generators for the maximal ideal $\mathfrak{m}_R$ of $R$, the last non-zero module on the right is the residue field. It follows that the module $M$ appearing on the left of the last sequence is just the second syzygy of ${k}$, that is $M=\operatorname{Syz}_R^2({k})$. In other words we have that $$\operatorname{Syz}^1_S(p_1,p_2,p_3)^G=\operatorname{Syz}_R^2({k}).$$ So, to understand which $R$-module $\operatorname{Syz}_R^2({k})$ is, we need to understand which is the action of $G$ on $\operatorname{Syz}^1_S(p_1,p_2,p_3)$, that is its $S*G$-module structure.
We know that as $S$-module $\operatorname{Syz}_S^1(p_1,p_2,p_3)\cong S^2$ and it is therefore generated by two elements $s_1(u,v),s_2(u,v)\in S^3$. We need to keep track of the action of $G$ through this isomorphism. The action of $G$ on $\operatorname{Syz}^1_S(p_1,p_2,p_3)$ is inherited by the action on $S$, which is linear and given by matrices $M=M_g$ in ${\mathrm{SL}(2,{k})}$.
In order to understand how these matrices act on the generators $s_1(u,v)$ and $s_2(u,v)$ we procede as follows. For each matrix $M$ we apply the linear transformation $(u,v)\mapsto M (u,v)^T $ to $s_1(u,v)$ and $s_2(u,v)$. We obtain two elements $s'_1(u,v)$ and $s'_2(u,v)$ in $S^3$ which belong to $\operatorname{Syz}^1_S(p_1,p_2,p_3)$. Therefore we can write them as linear combination of $s_1$ and $s_2$ $$\begin{pmatrix} s'_1\\
s'_2
\end{pmatrix}
= N \begin{pmatrix} s_1 \\ s_2
\end{pmatrix}$$ for some matrix $N$. In this way we obtain a collection of matrices $N=N_g$ for $g\in G$, which gives us the representation corresponding to $\operatorname{Syz}^1_S(p_1,p_2,p_3)$. Then one has to check that this is actually isomorphic to the fundamental representation $V_1$.
\[examplecyclicsyzygy\] Let $\xi$ be a primitive $n$-th root of unity in ${k}$, and consider the cyclic group $C_n$ generated by $$A=\begin{pmatrix}
\xi & 0 \\ 0 & \xi^{-1}
\end{pmatrix}.$$ The maximal ideal $\mathfrak{m}_R$ of the invariant ring $R={k}\llbracket u,v\rrbracket^{C_n}$ is generated by polynomials $p_1=u^n$, $p_2=v^n$, $p_3=uv$. Their syzygy module $\operatorname{Syz}^1_S(p_1,p_2,p_3)$ is generated by $$s_1=\begin{pmatrix}
0\\-u\\v^{n-1}
\end{pmatrix} \ \ \text{ and } \ \
s_2=\begin{pmatrix}
-v \\ 0 \\ u^{n-1}
\end{pmatrix}.$$ We apply the linear transformation given by $A$: $u\mapsto\xi u$, $v\mapsto \xi^{-1}v$, to $s_1$ and $s_2$ $$\begin{split}
s_1&\mapsto s'_1=\begin{pmatrix}
0\\-\xi u\\(\xi^{-1})^{n-1} v^{n-1}
\end{pmatrix}=\xi\begin{pmatrix}
0\\-u\\v^{n-1}
\end{pmatrix}=\xi s_1;\\
s_2&\mapsto s'_2=\begin{pmatrix}
-\xi^{-1}v \\ 0 \\ \xi^{n-1}u^{n-1}
\end{pmatrix}=\xi^{-1}\begin{pmatrix}
-v \\ 0 \\ u^{n-1}
\end{pmatrix}=\xi^{-1}s_2.
\end{split}$$ Thus, the representation corresponding to $\operatorname{Syz}^1_S(p_1,p_2,p_3)$ is exactly the fundamental representation.
The referee suggested to us an alternative proof of Theorem \[Theoremsecondsyzygyfield2\] which works for the singularities $D_{n+2}$, $E_6$, $E_7$ and $E_8$. We sketch it in the following remark.
\[remarksecondsyzygyfield2\] We use the same notation as in the proof of Theorem \[Theoremsecondsyzygyfield2\]. First, we observe that by Auslander correspondence the isomorphism $\operatorname{Syz}_R^2({k})\cong\mathcal{A}(V_1)$ is equivalent to say that the top of $\mathrm{Ext}^2_S(S/\mathfrak{m}_RS,S)$ is $V_1^*$. By the equivariant local duality (cf. [@HO10 (5.5)]), this is equivalent to say that the socle of $S/\mathfrak{m}_RS$ is $V_1$. Now let $x,y,z\in S$ be a minimal system of generators of $\mathfrak{m}_R$ as in the table at page , and consider the quotient $\Lambda=S/(y,z)$. Since $y$ and $z$ are invariants, $\Lambda$ inherits a natural $G$-action. By the relation $f$ in the table, $\Lambda$ is Artinian and hence it is a complete intersection. Moreover, the degree relation $\deg x= \deg y+\deg z-1$ holds and implies that $x\cdot u=x\cdot v=0$, because $\deg(x\cdot u)=\deg(x\cdot u)=\deg y+ \deg z$. So $x$ must sit in the socle of $\Lambda$, which is one-dimensional and is therefore equal to $\Lambda_{\deg x}$, the degree $\deg x$ component of $\Lambda$. Since $x$ is invariant, the socle of $\Lambda$ is the trivial representation. The natural multiplication $\Lambda_i\times\Lambda_{\deg x-i}\rightarrow\Lambda_{\deg x}\cong{k}$ forces $\operatorname{Hom}_{{k}}(\Lambda,{k})\cong\Lambda$ as $S*G$-modules, and moreover we have isomorphisms in each degree, i.e. $\Lambda_i\cong\operatorname{Hom}_{{k}}(\Lambda,{k})_{\deg x-i}$. We quotient by $x$, and we have that the socle of $\Lambda/x\Lambda\cong S/\mathfrak{m}_RS$, which is two-dimensional, must agree with $\Lambda_{\deg x -1}$, which is also two-dimensional. So we obtain $\Lambda_{\deg x-1}\cong \operatorname{Hom}_{{k}}(\Lambda,{k})_{1}\cong \operatorname{Hom}_{{k}}(V_1,{k})\cong V_1$, since the fundamental representation of the Klein groups is self-dual, and we are done.
For Kleinian singularities, also the module of Zariski differentials $\Omega_{R/{k}}^{**}$ is isomorphic to the fundamental module.
\[Zariskiisfundamental\] Let $(R,\mathfrak{m},{k})$ be a two-dimensional quotient singularity over an algebraically closed field ${k}$, and assume that the characteristic of ${k}$ does not divide the order of the acting group. Then the module of Zariski differentials $\Omega_{R/{k}}^{**}$ of $R$ over ${k}$ is isomorphic to the fundamental module $E$.
The decomposition of $\operatorname{Sym}^q(V)$
----------------------------------------------
Let $G$ be a finite group of order $n$ and let ${k}$ be an algebraically closed field such that $\operatorname{char}{k}$ does not divide $n$. Let $\mu_n({k})$ be the group of $n$-th roots of unity in ${k}$ and let $\mu_n(\mathbb{C})$ be the group of complex $n$-th roots of unity. Both $\mu_n({k})$ and $\mu_n(\mathbb{C})$ are cyclic groups of order $n$, so we can fix an isomorphism $ \phi: \mu_n({k})\rightarrow\mu_n(\mathbb{C})$, which we name a *lift*. In the same way, we say that a complex root of unity $z\in\mu_n(\mathbb{C})$ is a lift of $a\in\mu_n({k})$ if $z=\phi(a)$.
Let $(V,\rho)$ be a ${k}$-representation of $G$ of dimension $r\geq1$ and let $g$ be an element of $G$. Then the matrix $\rho(g)$ is diagonalizable in ${k}$ and its eigenvalues are elements of $\mu_n({k})$, since the order of $g$ divides $n$. Let $\lambda_1,\dots,\lambda_r$ be these eigenvalues, counted with multiplicity. The *Brauer character* or simply the *character* of $(V,\rho)$ is the function $\chi:G\rightarrow\mathbb{C}$ given by $$\chi_V(g)=\phi(\lambda_1)+\cdots+\phi(\lambda_r).$$
This definition depend on the choice of the isomorphism $\phi$. Since there are in general many choices for $\phi$, we have a certain degree of arbitrariness. However once chosen $\phi$, it will never be changed and sometimes we will simply say that we lift the eigenvalues to $\mathbb{C}$, meaning that the isomorphism $\phi$ is fixed.
For a proof of the following classical results of character theory the reader may consult the books of Feit [@Fei82 Chapter IV], and of Fulton and Harris [@FH91 Chapter 2].
\[propertiesofcharacterprop\] Let $V$ and $W$ be ${k}$-representations of $G$ with characters $\chi_V$ and $\chi_W$. Then the following facts hold:
1. $\chi_V(e)=\dim_{{k}}V$, where $e$ is the identity of $G$;
2. $\chi_V(g^{-1})=\overline{\chi_V(g)}$, the complex conjugate of $\chi_V(g)$, for every $g\in G$;
3. $\chi_V(hgh^{-1})=\chi_V(g)$ for every $g,h\in G$;
4. $\chi_{V\oplus W}(g)=\chi_V(g)+\chi_W(g)$ for every $g\in G$;
5. $\chi_{V\otimes W}(g)=\chi_V(g)\cdot\chi_W(g)$ for every $g\in G$.
We define an Hermitian inner product on the set of characters of $G$ by $$\langle\varphi,\psi\rangle:=\frac{1}{|G|}\sum_{g\in G}\overline{\varphi(g)}\psi(g),$$ for every $\varphi, \psi$ characters, where $\overline{\varphi(g)}$ denotes the complex conjugation. This bilinear form satisfies $$\langle \chi,\psi\rangle=\frac{1}{|G|}\sum_{g\in G}\chi(g)\overline{\psi(g)}=\langle \psi,\chi\rangle,$$ for every characters $\chi$, $\psi$.
The characters of the irreducible representations of $G$ are orthonormal with respect to this inner product.
\[orthonormalrelationstheorem\] Let $G$ be a finite group and let ${k}$ be an algebraically closed field such that $\operatorname{char}{k}$ does not divide $|G|$. Then the following facts hold.
1. A representation $V$ over ${k}$ is irreducible if and only if $\langle \chi_V,\chi_V\rangle=1$.
2. If $\chi$ and $\psi$ are the characters of two non-isomorphic irreducible ${k}$-representations then $\langle \chi,\psi\rangle=0$.
\[orthonormalrelationscorollary\] Let ${k}$ and $G$ be as in Theorem \[orthonormalrelationstheorem\]. Let $V$ be a ${k}$-representation of $G$ and let $$V=V_1^{n_1}\oplus\cdots\oplus V_r^{n_r}$$ be its decomposition into irreducible representations $V_i$. If $\chi_V$ is the character of $V$ and $\chi_{V_i}$ is the character of $V_i$ then $$n_i=\langle \chi_V,\chi_{V_i}\rangle.$$ In particular, two representations are isomorphic if and only if they have the same character.
Thanks to Corollary \[orthonormalrelationscorollary\], characters are an important tool to understand the decomposition of a representation into irreducible representations. We will use them to obtain the decomposition of the $q$-th symmetric powers $\operatorname{Sym}^q_{{k}}(V)$ of a two-dimensional faithful representation $V$ of a Klein group. We begin with an elementary statement.
\[sumofrootofunitybounded\] Let $\lambda\neq1$ be a root of unity in $\mathbb{C}$, and let $f:\mathbb{N}\rightarrow\mathbb{C}$ be the function $$f(N):= \sum_{q=0}^N\sum_{t=0}^q\lambda^{2t-q}.$$ Then $f(N)=o(N^2)$, that is $$\lim_{N\rightarrow+\infty}\frac{|f(N)|}{N^2}=0.$$
We have $$\begin{split}
0\leq\frac{|f(N)|}{N^2}&\leq\frac{2\left|\sum_{q=0}^N\sum_{t=0}^q\lambda^{2t-q}\right|}{N^2}=\frac{\left|\left(\sum_{q=0}^{N+1}\lambda^q\right)\left(\sum_{q=0}^{N+1}\lambda^{-q}\right)-\sum_{t=0}^{N+1}\lambda^{2t-N-1}\right|}{N^2}\\
&\leq\frac{4/|1-\lambda|^2+N+2}{N^2},
\end{split}$$ which goes to $0$ for $N\rightarrow+\infty$.
Let $(V,\rho)$ be a two-dimensional representation of a finite group $G$ over an algebraically closed field ${k}$. Since ${k}$ is algebraically closed, for every element $g\in G$ the matrix $\rho(g)$ can be diagonalized $$\rho(g)=\begin{pmatrix}
\lambda & 0\\
0 & \mu
\end{pmatrix},$$ with $\lambda,\mu\in {k}$. The representation $\operatorname{Sym}^q(V)$ evaluated at the element $g$ is given by the matrix $$\begin{pmatrix}
\lambda^q & 0 & \cdots & \cdots & 0 \\
0 & \lambda^{q-1}\mu & 0 & \cdots & 0 \\
\vdots & & \ddots & & \vdots \\
& \cdots & \cdots & \lambda\mu^{q-1} & 0 \\
0 & \cdots & \cdots & 0 & \mu^{q}
\end{pmatrix}.$$ In other words, if $\lambda$ and $\mu$ are the eigenvalues of $V$ at the element $g$, then the eigenvalues of $\operatorname{Sym}^q(V)$ at $g$ are $\{\lambda^{t}\mu^{q-t}: \ t=0,\dots,q\}$.
\[characterofSymqV\] Let ${k}$ be an algebraically closed field and let $G$ be finite group such that $\operatorname{char}{k}$ and $|G|$ are coprime. Let $(V,\rho)$ be a two-dimensional representation of $G$ whose image is contained in ${\mathrm{SL}(2,{k})}$, then the (Brauer) character of the representation $\operatorname{Sym}^q(V)$ is given by $$\chi_{\operatorname{Sym}^q(V)}(g)=\sum_{t=0}^q\lambda_g^{2t-q},$$ where $\lambda_g$ is the lift to $\mathbb{C}$ of an eigenvalue of the matrix $\rho(g)$.
Let $g\in G$ and let $\lambda_g$ and $\mu_g$ be the lift to $\mathbb{C}$ of the eigenvalues of $\rho(g)$. Since $\rho(g)\in{\mathrm{SL}(2,{k})}$, we have $\mu_g=\lambda_g^{-1}$. Then the lift of the eigenvalues of $g$ in the representation $\operatorname{Sym}^q(V)$ are $$\{\lambda_g^t\cdot\mu_g^{q-t}: \ t=0,\dots, q\}=\{\lambda_g^{2t-q}: \ t=0,\dots,q\},$$ so the formula for the character of $\operatorname{Sym}^q(V)$ follows immediately.
The character of the symmetric representation $\operatorname{Sym}^q(V)$ can be computed also using *Molien’s formula* (cf. [@Stu08 Theorem 2.2.1]) $$\label{moliensformula}
\sum_{q=0}^{+\infty}\chi_{\operatorname{Sym}^q(V)}(g)t^q=\frac{1}{\det(I-t\rho(g))}.$$ Here $t$ is an indeterminate, and the determinant on the right hand side is of a matrix with entries conveniently lifted to the polynomial ring $\mathbb{C}[t]$. Expanding the rational function on the right, we obtain a formal power series which is equal to the formal power series on the left, and we can use this equality to compute the character $\chi_{\operatorname{Sym}^q(V)}$. Notice that Molien’s formula holds also if $\dim_{{k}}V>2$.
\[theoremkleinrepresentation\] Let ${k}$ be an algebraically closed field and let $G$ be a Klein subgroup of ${\mathrm{SL}(2,{k})}$, such that $\operatorname{char}{k}$ does not divide $|G|$. Let $(V,\rho)$ be a two-dimensional faithful ${k}$-representation of $G$ with image contained in ${\mathrm{SL}(2,{k})}$ and let $(V_i,\rho_i)$ be an irreducible ${k}$-representation of $G$. We denote by $\alpha_{i,q}(V)$ the multiplicity of $V_i$ in $\operatorname{Sym}^q(V)$ and by $\beta_{q}(V)=\dim_{{k}}\operatorname{Sym}^q(V)$. Then $$\lim_{N\rightarrow+\infty}\frac{\sum_{q=0}^N\alpha_{i,q}(V)}{\sum_{q=0}^N\beta_{q}(V)}=\frac{\dim_{{k}}V_i}{|G|}.$$
Since $\dim_{{k}}V=2$, the dimension of $\operatorname{Sym}^q(V)$ is $q+1$. So the denominator in the limit above is $\sum_{q=0}^N\beta_{q}(V)=\frac12(N+1)(N+2)$. From Corollary \[orthonormalrelationscorollary\] and Lemma \[characterofSymqV\] we have $$\begin{split}
|G|\alpha_{i,q}(V)&=\langle \chi_{\operatorname{Sym}^q(V)},\chi_{V_i}\rangle=\sum_{g\in G}\chi_{\operatorname{Sym}^q(V)}(g)\cdot\overline{\chi_{V_i}(g)}\\
&=\sum_{g\in G}\overline{\chi_{V_i}(g)}\left(\sum_{t=0}^q\lambda_g^{2t-q}\right),
\end{split}$$ where $\lambda_g$ is the lift to $\mathbb{C}$ of an eigenvalue of $\rho(g)$. The order $m_g$ of the root of unity $\lambda_g$ coincides with the order of $\rho(g)$ in ${\mathrm{SL}(2,{k})}$. Since $V$ is faithful, this coincides also with the order of the group element $g$ in $G$. In particular, $\lambda_g=1$ if and only if $g$ is the identity $I$ of $G$. Thus, we can write the previous sum as $$\overline{\chi_{V_i}(I)}\left(\sum_{t=0}^q1\right)+\sum_{\substack{g\in G\\ g\neq I}}\overline{\chi_{V_i}(g)}\left(\sum_{t=0}^q\lambda_g^{2t-q}\right)= \dim_{{k}}(V_i)(q+1)+\sum_{\substack{g\in G\\ g\neq I}}\overline{\chi_{V_i}(g)}\left(\sum_{t=0}^q\lambda_g^{2t-q}\right).$$ We sum for $q$ running from $0$ to a fixed natural number $N$ and we obtain $$\begin{split}
|G|\sum_{q=0}^N\alpha_{i,q}(V)=\frac12\dim_{{k}}(V_i)(N+1)(N+2)+o(N^2),
\end{split}$$ thanks to Lemma \[sumofrootofunitybounded\]. Therefore, the numerator of the limit is $$\sum_{q=0}^N\alpha_{i,q}(V)=\frac{(N+1)(N+2)\dim_{{k}}V_i}{2|G|}+o(N^2)$$ so the limit is equal to $\frac{\dim_{{k}}V_i}{|G|}$ as desired.
The multiplicity $\alpha_{i,q}(V)$ can be computed also using Molien’s formula , as showed by Springer in [@Spr87]. One obtains $$\sum_{q=0}^{+\infty} \alpha_{i,q}(V) t^q=\frac{1}{|G|}\sum_{g\in G}\frac{\chi_{V_i}(g^{-1})}{\det(I-t\rho(g))}.$$
Symmetric signature of Kleinian singularities {#symmetric-signature-of-kleinian-singularities}
---------------------------------------------
\[theoremssKleinian\] Let ${k}$ be an algebraically closed field and let $G$ be a Klein subgroup of ${\mathrm{SL}(2,{k})}$ such that $\operatorname{char}{k}$ does not divide $|G|$. Let $R$ be the corresponding Kleinian singularity and let $M_i$ be an indecomposable MCM $R$-module. Then the generalized symmetric signature and the generalized differential symmetric signature of $R$ with respect to $M_i$ are $$s_{\sigma}(R,M_i)=s_{d\sigma}(R,M_i)=\frac{\operatorname{rank}_RM_i}{|G|}.$$
Let $E$ be the fundamental module of $R$. By Theorem \[Theoremsecondsyzygyfield1\], and by Theorem \[Zariskiisfundamental\] we have $$E\cong\operatorname{Syz}^2_R({k})\cong\Omega_{R/{k}}^{**}.$$ Therefore the two symmetric signatures, differential and syzygy, coincide for Kleinian singularities.
Let $V_i$ be the irreducible ${k}$-representation of $G$ such that $M_i=\mathcal{A}(V_i)$, and let $V$ be the fundamental representation of $G$, that is the ${k}$-representation such that $\mathcal{A}(V)=E$. Let $\alpha_{i,q}(E)$ be the multiplicity of $M_i$ into $\operatorname{Sym}_R^q(E)^{**}$ and let $\beta_q(E)=\operatorname{rank}_R\operatorname{Sym}_R^q(E)^{**}$. From Theorem \[symcommuteswithauslandertheorem\] we have $$\mathcal{A}'\big(\operatorname{Sym}^q_R(E)^{**}\big)=\operatorname{Sym}^q(V).$$ It follows that $\beta_q(E)=\dim_{{k}}\operatorname{Sym}^q(V)$ and $\alpha_{i,q}(E)$ equals the multiplicity of the representation $V_i$ in $\operatorname{Sym}^q(V)$. Since $V$ is the fundamental representation of a Klein group, it is faithful and its image is contained in ${\mathrm{SL}(2,{k})}$, so we can apply Theorem \[theoremkleinrepresentation\] to conclude the proof.
\[corollaryKleinian\] Let ${k}$ be an algebraically closed field and let $G$ be a Klein subgroup of ${\mathrm{SL}(2,{k})}$ such that $\operatorname{char}{k}$ does not divide $|G|$. The symmetric signature and the differential symmetric signature of the corresponding Kleinian singularity $R$ are $$s_{\sigma}(R)=s_{d\sigma}(R)=\frac{1}{|G|}.$$
\[exampleAnlimitnotexists\] Consider the $A_{n-1}$-singularity $R={k}\llbracket u,v\rrbracket^{C_n}$ over an algebraically closed field ${k}$, and let $\mathcal{S}^q=\left(\operatorname{Sym}_R^q(\operatorname{Syz}_{R}^2({k}))\right)^{**}$ as in the definition of symmetric signature. From the proofs of Theorem \[theoremkleinrepresentation\] and Theorem \[theoremssKleinian\], we have that $\operatorname{rank}_R\mathcal{S}^q=q+1$, and $$\operatorname{frk}_R\mathcal{S}^q=\frac{q+1}{n}(1+(-1)^q)+O(1)=\begin{cases}
\frac{2(q+1)}{n}+O(1)\ &\text{ for }q \ \text{even}\\
O(1)\ &\text{ for }q \ \text{odd}.
\end{cases}$$ This shows that the limit $$\lim_{q\rightarrow+\infty}\frac{\operatorname{frk}_R\mathcal{S}^q}{\operatorname{rank}_R\mathcal{S}^q}$$ does not exist. Specifically, in this setting the syzygy module splits as $\operatorname{Syz}_R^2({k})\cong M_1\oplus M_{n-1}$ (with notation as in Example \[examplecyclicgroup3\]). Therefore we have $$\mathcal{S}^q\cong \bigoplus_{t=0}^q M_1^{\otimes t}\otimes M_{n-1}^{\otimes q-t}\cong\bigoplus_{t=0}^qM_1^{\otimes 2t-q}.$$
Symmetric signature of cones over elliptic curves {#sectionelliptic}
=================================================
In this section we prove that the differential symmetric signature of the coordinate ring $R$ of a plane elliptic curve over an algebraically closed field of characteristic different from $2$ and $3$ is zero. Our intention is motivated by the following remark, which shows that the F-signature of such rings (over a field of positive characteristic) is zero.
We consider $R={k}[x,y,z]/(f)$, with ${k}$ an algebraically closed field of positive characteristic, and $f$ a homogeneous non-singular polynomial of degree $3$. In other words, $R$ is the coordinate ring of a plane elliptic curve over ${k}$. The ring $R$ is not strongly F-regular, and in particular its F-signature $s(R)$ is $0$ by the result of Aberbach and Leuschke (Theorem \[AbeLeusTheorem\]). To see this, assume for simplicity that $f$ is in Weierstrass normal form, that is $f=y^2z-x^3-axz^2-bz^3$ for some $a,b\in{k}$. This is always possible if the characteristic of ${k}$ is different from $2$ and $3$. Then the ideal $I=(x,z)$ is not tightly closed in $R$, since $y^2\not\in I$, but $y^2\in I^{*}$. Therefore $R$ is not strongly F-regular and $s(R)=0$.
The methods we use in this situation are different from those of previous sections, and are of geometric nature. We will use the correspondence between graded MCM $R$-modules and vector bundles over the smooth projective curve $Y=\mathrm{Proj}\,R$ to translate the problem into geometric language. We will not make a general assumption on the characteristic of the base field ${k}$ in this section, but we will restrict to characteristic $\neq 2,3$ only when needed. The letter $q$ is used to denote a positive integer.
Generalities on vector bundles
------------------------------
We recall some preliminary facts concerning vector bundles over curves, and in particular over elliptic curves. For further details and the proofs of these facts, we refer to the books of Hartshorne [@Har77], Le Potier [@LeP97], and Mukai [@Muk03 Chapter 10].
Let $Y$ be a smooth projective curve over an algebraically closed field ${k}$. We denote by $\mathrm{VB}(Y)$ the category whose objects are vector bundles over $Y$, and whose maps are morphisms between them. The category $\mathrm{VB}(Y)$ is a Krull-Schmidt category (cf. [@Ati56 Theorem 3]). Moreover there is an equivalence of categories between the category of vector bundles over $Y$ and the category of locally free sheaves on $Y$ which respects the rank (cf. [@Har77 Ex. II 5.18]). Having this identification in mind, we will use indifferently the words vector bundle and locally-free sheaf. For example, we will use the notation for the structure sheaf $\mathcal{O}_Y$ of $Y$ also to denote the trivial bundle of rank one corresponding to it.
Let $\mathcal{S}$ be a locally free sheaf of rank $r$ over $X$. The degree of $\mathcal{S}$ is defined as the degree of the corresponding determinant line bundle $\mathrm{deg}\mathcal{S}=\deg\bigwedge^r\mathcal{S}$. The slope of $\mathcal{S}$ is $\mu(\mathcal{S})=\deg\mathcal{S}/r$. The degree is additive on short exact sequences and moreover $\mu(\mathcal{S}\otimes\mathcal{T})=\mu(\mathcal{S})+\mu(\mathcal{T})$.
The sheaf $\mathcal{S}$ is called *semistable* if for every locally free subsheaf $\mathcal{T}\subseteq\mathcal{S}$ the inequality $\mu(\mathcal{T})\leq\mu(\mathcal{S})$ holds. If the strict inequality $\mu(\mathcal{T})< \mu(\mathcal{S})$ holds for every proper subsheaf $\mathcal{T}\subset\mathcal{S}$, then $\mathcal{S}$ is called *stable*. Stable vector bundles are indecomposable, but the converse does not hold in general.
The usual operations on vector spaces like direct sum, tensor product, symmetric and wedge powers extend naturally to vector bundles and correspond to the same constructions in the category of locally free sheaves. The *dual* of a vector bundle $E$ is the bundle $E^{\vee}:=\operatorname{Hom}(E,\mathcal{O}_Y)$.
Since every vector bundle $E$ is canonically isomorphic to its double dual $E^{\vee\vee}$, reflexive symmetric powers $\operatorname{Sym}^q(-)^{\vee\vee}$ coincide with ordinary symmetric powers $\operatorname{Sym}^q(-)$ in the category $\mathrm{VB}(Y)$.
The following are standard properties of symmetric powers of vector bundles (see also [@Laz04 Corollary 6.4.14]).
\[propertiessymvb\] Let $Y$ be a smooth projective curve, let $E$ be a vector bundle and $\mathcal{L}$ a line bundle over $Y$. Then the following facts hold.
$\operatorname{Sym}^q(E\otimes\mathcal{L})\cong \operatorname{Sym}^q(E)\otimes\mathcal{L}^{\otimes q}$.
$\mu(\operatorname{Sym}^q(E))=q\cdot\mu(E)$.
If the field ${k}$ has characteristic $0$ and $E$ is semistable, then $\operatorname{Sym}^q(E)$ is also semistable.
Let $R$ be a normal standard-graded domain of dimension $2$ over an algebraically closed field ${k}$, that is $R_0={k}$ and $R$ is generated by finitely many elements of degree $1$. The normal assumption on $R$ implies that $R_{\mathfrak{p}}$ is a regular ring for every prime ideal $\mathfrak{p}\neq R_{+}$. It follows that the projective variety $Y=\mathrm{Proj}\,R$ is a smooth projective curve over ${k}$. For every graded module $M$ we denote by $\widetilde{M}$ the corresponding coherent sheaf on $Y$. The sheaves $\widetilde{R(n)}$ are invertible (cf. [@Har77 Proposition II.5.12]) and denoted by $\mathcal{O}_Y(n)$. Moreover, every MCM graded $R$-module $M$ is locally free on the punctured spectrum, so the associated coherent sheaf $\widetilde{M}$ is in fact a vector bundle over $Y$. If we denote by ${\mathrm{MCM}_{\mathbb{Z}}(R)}$ the category of finitely generated graded MCM $R$-modules, then we have a functor $$\widetilde{ }:{\mathrm{MCM}_{\mathbb{Z}}(R)}\rightarrow\mathrm{VB}(Y),$$ whose properties we collect in the following proposition.
\[propertiessheafification\] Let $R$ be a normal standard graded domain of dimension $2$ over a field ${k}$, let $Y=\mathrm{Proj}\,R$, and let $M$ and $N$ be finitely generated graded MCM $R$-modules. Then the following facts hold.
$\widetilde{M\oplus N}\cong \widetilde{M}\oplus\widetilde{N}$;
${\ThisStyle{ \setbox0=\hbox{$\SavedStyle(M\otimes_R N)^{**}$} \stackengine{-.1\LMpt}{$\SavedStyle(M\otimes_R N)^{**}$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\cong \widetilde{M}\otimes_{\mathcal{O}_Y}\widetilde{N}$;
${\ThisStyle{ \setbox0=\hbox{$\SavedStyle\operatorname{Sym}_R^q(M)^{**}$} \stackengine{-.1\LMpt}{$\SavedStyle\operatorname{Sym}_R^q(M)^{**}$}{ \stretchto{\scaleto{\SavedStyle\mkern.2mu\sim}{.5467\wd0}}{.7\ht0} }{O}{c}{F}{T}{S}}}\cong\operatorname{Sym}_Y^q(\widetilde{M})$.
Now we give an appropriate definition of free rank for the category of vector bundles over $Y$.
\[freerankVB\] Let $E$ be a vector bundle over $Y$ with a fixed very ample invertible sheaf $\mathcal{O}_Y(1)$. We define the *free rank* of $E$ as $$\begin{split}
\operatorname{frk}_{\mathcal{O}_Y(1)}(E):=\max\big\{n: \ \exists &\text{ a split surjection }\varphi: E\twoheadrightarrow F, \ \text{ with } F=\bigoplus_{i=1}^n\mathcal{O}_Y(-d_i) \\
&\text{ a splitting vector bundle of rank } n\big\}.
\end{split}$$
Since the category $\mathrm{VB}(Y)$ has the KRS property, to compute the free rank of a bundle $E$, one should count how many copies of twisted structure sheaves $\mathcal{O}_Y(-d_i)$ appear in the decomposition of $E$ into indecomposable bundles. For this reason, the free rank of vector bundles $\operatorname{frk}_{\mathcal{O}_Y(1)}$ agrees with the graded free rank $\operatorname{frk}_R^{\mathrm{gr}}$ and not with the free rank, as explained in the following corollary.
\[rankfreerankcorollary\] Let $R$ be a normal standard graded domain of dimension $2$ over a field ${k}$, let $Y=\mathrm{Proj}\,R$, and let $M$ be a finitely generated graded MCM $R$-module. Then the following facts hold.
$\operatorname{rank}_RM=\operatorname{rank}_{\mathcal{O}_Y}\widetilde{M}$.
$\operatorname{frk}_R^{\textrm{gr}}M=\operatorname{frk}_{\mathcal{O}_Y(1)}\widetilde{M}$.
\[remarksyzygybundle\] Let $I$ be an $R_+$-primary homogeneous ideal with homogeneous generators $f_1,\dots,f_n$ of degrees $d_i$. The following presentation of $R/I$ $$\bigoplus_{i=1}^nR(-d_i)\xrightarrow{f_1,\dots,f_n}R\rightarrow R/I\rightarrow0$$ induces a short exact sequence of sheaves on $Y$ $$0\rightarrow\mathcal{S}\rightarrow\bigoplus_{i=1}^n\mathcal{O}_Y(-d_i)\rightarrow\mathcal{O}_Y\rightarrow0,$$ where $R/I$ corresponds to $0$ because $I$ is $R_+$-primary. Since $\mathcal{O}_Y$ and $\bigoplus_{i=1}^n\mathcal{O}_Y(-d_i)$ are locally free, the sheaf $\mathcal{S}$ is also locally free. The sheaf $\mathcal{S}$ is the kernel of the sheaf morphism $f_1,\dots, f_n$ and is denoted by $\operatorname{Syz}(f_1,\dots,f_n)$ and called *syzygy bundle*. In fact, it is nothing but the vector bundle corresponding to the MCM graded $R$-module $\operatorname{Syz}_R^2(R/I)=\operatorname{Syz}_R^1(f_1,\dots,f_n)$.
Syzygies of the irrelevant maximal ideal $R_+$ play a special role, they correspond to the restriction of the cotangent bundle $\Omega_{\mathbb{P}^n}$ of $\mathbb{P}^n$ to the curve. In fact, it follows from the Euler sequence on $\mathbb{P}^n$ [@Har77 Theorem 8.13] $$0\rightarrow \Omega_{\mathbb{P}^n}\rightarrow \bigoplus_{i=0}^n\mathcal{O}_{\mathbb{P}^n}(-1)\rightarrow\mathcal{O}_{\mathbb{P}^n}\rightarrow0,$$ that $\Omega_{\mathbb{P}^n}$ is a syzygy bundle. So if $x_0,\dots,x_n$ is a system of generators for $\mathcal{O}_{\mathbb{P}^n}(1)$, we have the isomorphism $$\operatorname{Syz}(x_0,\dots,x_n)\cong\Omega_{\mathbb{P}^n}.$$ If we restrict these bundles to the curve $Y$, we obtain isomorphic bundles. In fact, we have just proved the following proposition.
\[equivsyzcotang\] Let $Y$ be a smooth projective curve with an embedding in $\mathbb{P}^n$ given by $\varphi:Y\rightarrow\mathbb{P}^n$. Let $x_0,\dots,x_n$ be a system of generators of $\mathcal{O}_{\mathbb{P}^n}(1)$ and denote by $\Omega_{\mathbb{P}^n}$ the cotangent bundle of $\mathbb{P}^n$ and by $\Omega_{\mathbb{P}^n}|_{Y}$ its restriction to $Y$. Then, we have the following isomorphism of vector bundles on $Y$ $$\operatorname{Syz}(x_0,\dots,x_n)|_Y\cong\Omega_{\mathbb{P}^n}|_{Y}.$$
It is a classical result, usually ascribed to Grothendieck, that the only indecomposable vector bundles over a curve of genus zero are of the form $\mathcal{O}_Y(a)$ for some integer $a$. Therefore the structure of the category $\mathrm{VB}(Y)$ is quite easy for such a curve. The next interesting case concerns a curve of genus $1$, an elliptic curve. This case was treated by Atiyah [@Ati57], who was able to give a description of all indecomposable vector bundles of fixed rank and degree. Moreover he gave explicit formulas for the multiplicative structure of the monoid $\mathrm{VB}(Y)$ with the tensor product operation if the base field has characteristic zero.
In the following theorem we collect some results of Atiyah that we will need later on.
\[Atiyahfundamentalbundle\] Let $Y$ be an elliptic curve over an algebraically closed field ${k}$ with structure sheaf $\mathcal{O}_Y$. For every positive integer $r$ there exists a unique (up to isomorphism) indecomposable vector bundle $F_r$ of rank $r$ and degree $0$ with $\Gamma(Y,F_r)\neq0$. Moreover, the following facts hold.
$F_r$ has only one section up to scalar multiplication, i.e. $\Gamma(Y,F_r)\cong{k}$.
$F_r\cong F_r^{\vee}$.
There is a non-split exact sequence $$0\rightarrow \mathcal{O}_Y\rightarrow F_r\rightarrow F_{r-1}\rightarrow0.$$
$\operatorname{Sym}^{q}(F_2)\cong F_{q+1}$.
If $E$ is an indecomposable vector bundle of rank $r$ and degree $0$ then $E\cong L\otimes F_r$, where $L$ is a line bundle of degree zero, unique up to isomorphism and such that $L\cong\det E$.
We call the bundle $F_r$ *Atiyah bundle of rank* $r$. Clearly we have that $F_1=\mathcal{O}_Y$, the structure sheaf is the unique line bundle of degree $0$ with non-zero sections. The bundle $F_2$ is given by the unique non-trivial extension
$$0\rightarrow \mathcal{O}_Y\rightarrow F_2\rightarrow \mathcal{O}_Y \rightarrow 0.$$
In other words, $F_2$ is the non-zero element of $\mathrm{Ext}^1(\mathcal{O}_Y,\mathcal{O}_Y)$.
The following two results are well known to experts. However, since we are unable to locate a proof in the literature, for the sake of completeness we add a proof here.
\[lemmaSymqFrsemistable\] Let $Y$ be an elliptic curve over an algebraically closed field ${k}$ with structure sheaf $\mathcal{O}_Y$. Then for every positive integers $r$ and $q$ the vector bundle $\operatorname{Sym}^q(F_r)$ is semistable.
Indecomposable bundles over an elliptic curve are semistable (cf. [@Tu93 Lemma 29]), so if $\operatorname{char}{k}=0$ the result follows from Proposition \[propertiessymvb\].
Now assume that $\operatorname{char}{k}=p>0$. The Atiyah bundle $F_r$ is defined inductively by a non-split short exact sequence $0\rightarrow \mathcal{O}_Y\rightarrow F_r\rightarrow F_{r-1}\rightarrow0$, hence by a non-zero cohomology class $c\in H^1(Y,F_{r-1}^{\vee})\cong H^1(Y,F_{r-1})$. By [@Bre05 Lemma 2.4] the $p^e$-multiplication $[p^e]:Y\rightarrow Y$ on the elliptic curve kills this cohomology class for a suitable choice of $e$, that is $[p^e]^*(c)=0$. Therefore the pullback via $[p^e]$ of the previous sequence splits, and inductively this shows that $[p^e]^*(F_r)$ is free. Hence its symmetric powers are also free and in particular semistable. Then the semistability of $[p^e]^*(\operatorname{Sym}^q(F_r))\cong\operatorname{Sym}^q([p^e]^*(F_r))$ implies that $\operatorname{Sym}^q(F_r)$ is semistable.
\[lemmasemistableelliptic\] Let $Y$ be an elliptic curve over an algebraically closed field ${k}$, and let $E$ be a semistable vector bundle, then $\operatorname{Sym}^q(E)$ is semistable.
If $\operatorname{char}{k}=0$, then the result follows from Proposition \[propertiessymvb\]. So we assume that $\operatorname{char}{k}=p>0$, and we denote by $r$ the rank of $E$. By a result of Oda [@Oda71] (see also [@Tu93]) the bundle $E$ is strongly semistable, so by [@Miy87 Proposition 5.1] it follows that $[n]^*E$ is semistable for every integer $n$, where $[n]:Y\rightarrow Y$ is the standard multiplication on the elliptic curve.
We consider the semistable bundle $[r]^*E$. Its degree is a multiple of $r$, so by tensoring with an appropriate line bundle $L$, we obtain a semistable vector bundle $E'=[r]^*E\otimes L$ of degree $0$. It follows that the indecomposable direct summands of $E'$ must have degree $0$ too, so they are of the form $F_s\otimes \mathcal{L}$, with $\mathcal{L}$ line bundle. With the help of a further multiplication map as in the proof of Lemma \[lemmaSymqFrsemistable\] we may achieve that $[n]^*(E)$ is a direct sum of line bundles of degree $0$. Hence its symmetric powers are semistable and this descends to the symmetric powers of $E$.
Symmetric signature of cones over elliptic curves {#symmetric-signature-of-cones-over-elliptic-curves}
-------------------------------------------------
Let $f$ be a homogeneous non-singular polynomial $f$ of degree $3$ in ${k}[x,y,z]$, and let $R={k}[x,y,z]/(f)$. Then $R$ is a normal standard graded ${k}$-domain of dimension $2$ and the projective curve $Y=\mathrm{Proj}\,R$ is a plane elliptic curve over ${k}$.
To compute the symmetric signature of $R$ one should consider reflexive symmetric powers of the second syzygy of the residue field, $\operatorname{Syz}^2_R({k})$. Thanks to Remark \[remarksyzygybundle\] and Proposition \[propertiessheafification\] this is equivalent to consider the syzygy bundle $\operatorname{Syz}(x,y,z)$ over $Y$ and taking ordinary symmetric powers in the category $\mathrm{VB}(Y)$. We recall that by Corollary \[rankfreerankcorollary\] we have $\operatorname{rank}_RM=\operatorname{rank}_{\mathcal{O}_Y}\widetilde{M}$ and $\operatorname{frk}^{\mathrm{gr}}_RM=\operatorname{frk}_{\mathcal{O}_Y(1)}\widetilde{M}$ for every MCM graded $R$-module. Therefore, we have that $s_{\sigma}(R)$ exists if and only if the limit $$\lim_{N\rightarrow+\infty}\frac{\sum_{q=0}^N\operatorname{frk}_{\mathcal{O}_Y(1)}\operatorname{Sym}^q(\operatorname{Syz}(x,y,z))} {\sum_{q=0}^N\operatorname{rank}_{\mathcal{O}_Y}\operatorname{Sym}^q(\operatorname{Syz}(x,y,z))}$$ exists, and in this case they coincide.
The same reasoning applies to the differential symmetric signature. In this case, we should consider the sheaf associated to the module of Zariski differentials $\Omega_{R/{k}}^{**}$ on $Y$ and then take ordinary symmetric powers in the category $\mathrm{VB}(Y)$. In other words, we consider the vector bundles $$\operatorname{Sym}^q_Y\left( \widetilde{\Omega_{R/{k}}}\right).$$ Notice that we can forget about the double dual inside $\operatorname{Sym}_Y^q(-)$, thanks to Proposition \[propertiessheafification\]. The differential symmetric signature of the coordinate ring $R$ exists if and only if the limit $$\label{cotangentellipticlimit}
\lim_{N\rightarrow+\infty}\frac{\sum_{q=0}^N\operatorname{frk}_{\mathcal{O}_Y(1)}\operatorname{Sym}^q_Y\left( \widetilde{\Omega_{R/{k}}}\right)} {\sum_{q=0}^N\operatorname{rank}_{\mathcal{O}_Y}\operatorname{Sym}^q_Y\left( \widetilde{\Omega_{R/{k}}}\right)}$$ exists, and in this case they coincide.
\[propSyziscotangent\] Let $Y$ be a plane elliptic curve over a field ${k}$ of characteristic $\neq2,3$ with coordinate ring ${k}[x,y,z]/(f)$, where $f$ is a homogeneous polynomial of degree $3$. Then we have $$\operatorname{Syz}\left(\frac{\partial f}{\partial x},\frac{\partial f}{\partial y},\frac{\partial f}{\partial z}\right)(3)\cong F_2.$$
It is enough to show that $\operatorname{Syz}\left(\frac{\partial f}{\partial x},\frac{\partial f}{\partial y},\frac{\partial f}{\partial z}\right)(3)$ corresponds to a non-zero element of $\mathrm{Ext}^1(\mathcal{O}_Y,\mathcal{O}_Y)$ with global sections, then the uniqueness of $F_2$ will imply the desired isomorphism.
We consider the following injective map of sheaves, $\varphi:\mathcal{O}_Y\rightarrow\mathcal{O}_Y(1)^{\oplus3}$, given by $\varphi(1)=(x,y,z)$. From the Euler formula $$\frac{\partial f}{\partial x}x+\frac{\partial f}{\partial y}y+\frac{\partial f}{\partial z}z=3f,$$ which vanishes on $Y$, we obtain that the image of $\varphi$ is contained in the syzygy bundle $\operatorname{Syz}\left(\frac{\partial f}{\partial x},\frac{\partial f}{\partial y},\frac{\partial f}{\partial z}\right)(3)$. Actually, since this image does not vanish anywhere, it defines a subbundle and hence also a quotient bundle, which is by rank and degree reasons the structure sheaf. In other words, we have the following short exact sequence of vector bundles $$\label{sequenceF2}
0\rightarrow\mathcal{O}_Y\xrightarrow{\varphi}\operatorname{Syz}\left(\frac{\partial f}{\partial x},\frac{\partial f}{\partial y},\frac{\partial f}{\partial z}\right)(3)\rightarrow\mathcal{O}_Y\rightarrow0,$$
It remains to prove that the sequence is non-split, equivalently $(x,y,z)$ is the unique non-zero global section. This can be done easily by explicit computations, for example assuming that $f$ is in Weierstrass normal form.
\[corolOmega=F2\] Let $\operatorname{char}{k}\neq2,3$, and let $\widetilde{\Omega_{R/{k}}}$ be the sheaf version of the cotangent module $\Omega_{R/{k}}$ of the cone $R$ of $Y$, then we have an isomorphism of vector bundles $$\widetilde{\Omega_{R/{k}}}(-1)\cong F_2.$$
We recall that the cotangent module $\Omega_{R/{k}}$ is the graded $R$-module $$\Omega_{R/{k}}=<\mathrm{d}x,\mathrm{d}y,\mathrm{d}z>/\left(\frac{\partial f}{\partial x}\mathrm{d}x+\frac{\partial f}{\partial y }\mathrm{d}y+\frac{\partial f}{\partial z}\mathrm{d}z\right).$$ In other words, $\Omega_{R/{k}}$ can be defined by the following short exact sequence of graded $R$-modules $$\label{cotangentsequence}
0\rightarrow R(-2)\xrightarrow{\psi} R^{\oplus3}\xrightarrow{\varphi}\Omega_{R/{k}}\rightarrow0,$$ where $\psi(1)=\left(\frac{\partial f}{\partial x},\frac{\partial f}{\partial y},\frac{\partial f}{\partial z}\right)$, and $\varphi$ sends the canonical basis to $\mathrm{d}x,\mathrm{d}y,\mathrm{d}z$.
Sequence induces a short exact sequence of locally free sheaves on $Y$ $$0\rightarrow\mathcal{O}_Y(-2)\rightarrow\mathcal{O}_Y^{\oplus3}\rightarrow\widetilde{\Omega_{R/{k}}}\rightarrow0.$$ We dualize this sequence and we get $$0\rightarrow \widetilde{\Omega_{R/{k}}}^{\vee}\rightarrow\mathcal{O}_Y^{\oplus3}\rightarrow\mathcal{O}_Y(2)\rightarrow0,$$ where the last map is given by $\psi$. We obtain that $$\widetilde{\Omega_{R/{k}}}^{\vee}\cong\operatorname{Syz}\left(\frac{\partial f}{\partial x},\frac{\partial f}{\partial y},\frac{\partial f}{\partial z}\right)(2).$$ So by the previous Proposition \[propSyziscotangent\] we have $\widetilde{\Omega_{R/{k}}}^{\vee}(1)\cong F_2$ on $Y$, but the Atiyah bundle $F_2$ is self-dual, so we have $\widetilde{\Omega_{R/{k}}}(-1)\cong F_2^{\vee}\cong F_2$.
We can compute the differential symmetric signature of the cone $R$.
\[Theocotangentelliptic\] Let $Y$ be a plane elliptic curve over an algebraically closed field ${k}$ of characteristic $\neq2,3$ with coordinate ring $R$. Then the differential symmetric signature of $R$ is $s_{d\sigma}(R)=0$.
By the previous observations we should compute the limit . From Corollary \[corolOmega=F2\], we have $\widetilde{\Omega_{R/{k}}}\cong F_2\otimes\mathcal{O}_Y(1)$. Therefore from part *4)* of Theorem \[Atiyahfundamentalbundle\] and from Proposition \[propertiessymvb\] we obtain $$\operatorname{Sym}^q_Y\left( \widetilde{\Omega_{R/{k}}}\right)\cong\operatorname{Sym}_Y^q(F_2)\otimes\mathcal{O}_Y(q)\cong F_{q+1}\otimes\mathcal{O}_Y(q)$$ for all $q\geq1$. So the module $\operatorname{Sym}^q_Y\left( \widetilde{\Omega_{R/{k}}}\right)$ is indecomposable, and in particular it has free rank $0$ and the claim follows.
\[propSyzstablebundle\] The vector bundle $\operatorname{Syz}(x,y,z)$ is stable of rank $2$ and degree $-9$. Moreover $\det\operatorname{Syz}(x,y,z)=\mathcal{O}_Y(-3)$.
The syzygy bundle fits into a short exact sequence $$\label{syzygysequenceelliptic}
0\rightarrow \operatorname{Syz}(x,y,z)\rightarrow \bigoplus_{i=1}^3\mathcal{O}_Y(-1)\xrightarrow{x,y,z}\mathcal{O}_Y\rightarrow0.$$ So from the additivity of rank and degree, we immediately get that $\operatorname{Syz}(x,y,z)$ has rank $2$ and degree $$\deg\operatorname{Syz}(x,y,z)=\left(\deg\bigoplus_{i=1}^3\mathcal{O}_Y(-1)-\deg\mathcal{O}_Y\right)\deg Y= -3\deg Y=-9.$$
For the indecomposable property, we have that $\operatorname{Syz}(x,y,z)$ equals the restriction of the cotangent bundle $\Omega_{\mathbb{P}^2}$ of $\mathbb{P}^2$ to $Y$ by Proposition \[equivsyzcotang\]. This bundle is stable, and in particular indecomposable, by the result of Brenner and Hein [@BH06 Theorem 1.3]. Finally, taking determinants of sequence yields $\det\operatorname{Syz}(x,y,z)=\mathcal{O}_Y(-3)$.
\[remrankSymSyz\] Since $\operatorname{Syz}(x,y,z)$ has rank $2$, we have that $\operatorname{rank}_{\mathcal{O}_Y}\operatorname{Sym}^q(\operatorname{Syz}(x,y,z))=q+1$.
The difficult part is to determine the free rank of $\operatorname{Sym}^q(\operatorname{Syz}(x,y,z))$.
\[corqodd0freerank\] Let $q$ be odd, then the bundle $\operatorname{Sym}^q(\operatorname{Syz}(x,y,z))$ contains no direct summand of rank one. In particular, $\operatorname{frk}_{\mathcal{O}_Y(1)}\operatorname{Sym}^q(\operatorname{Syz}(x,y,z))=0$.
From Proposition \[propertiessymvb\], Proposition \[propSyzstablebundle\] and Lemma \[lemmasemistableelliptic\] we know that $\operatorname{Sym}^q(\operatorname{Syz}(x,y,z))$ is semistable of slope $-\frac{9}{2}q$. Let $\mathcal{L}$ be a direct summand of rank one. Since $\operatorname{Sym}^q(\operatorname{Syz}(x,y,z))$ is semistable, we have $\mu(\mathcal{L})=-\frac{9}{2}q$, but $\mu(\mathcal{L})=\deg\mathcal{L}$ must be an integer. Since $q$ is odd we get a contradiction.
\[corsleq12elliptic\] If the symmetric signature of $R$ exists, then $s_{\sigma}(R)\leq\frac{1}{2}$.
For every $q\in\mathbb{N}$, let $a_q=\operatorname{frk}_{\mathcal{O}_Y(1)}\operatorname{Sym}^q(\operatorname{Syz}(x,y,z))$, and let $b_q=\operatorname{rank}_{\mathcal{O}_Y}\operatorname{Sym}^q(\operatorname{Syz}(x,y,z))$. From the previous Corollary \[corqodd0freerank\] and Remark \[remrankSymSyz\], we have $a_q=0$ for $q$ odd, and $b_q=q+1$ for all $q$.
For every even $q$ we have the following inequalities $$a_q+a_{q+1}=a_{q}\leq b_q=\frac{1}{2}(b_q+b_{q})\leq\frac{1}{2}(b_q+b_{q+1}).$$ It follows that $$\frac{\sum_{q=0}^Na_q}{\sum_{q=0}^Nb_q}\leq\frac{\frac{1}{2}\sum_{q=0}^Nb_q}{\sum_{q=0}^Nb_q}+\phi(N)=\frac{1}{2}+\phi(N),$$ where $\phi(N)=\frac{b_N}{\sum_{q=0}^Nb_q}$ if $N$ is even, and $0$ otherwise. In both cases $\phi(N)\rightarrow 0$ if $N\rightarrow+\infty$, so the claim is proved.
Is it true that also $s_{\sigma}(R)=0$?
Acknowledgements {#acknowledgements .unnumbered}
================
We would like to thank Winfried Bruns, Ragnar-Olaf Buchweitz, Helena Fischbacher-Weitz, Lukas Katthän, and Yusuke Nakajima for their interest and many valuable comments. We thank the referee for showing us how to simplify the proof of Theorem \[theoremkleinrepresentation\] and pointing out Remark \[remarksecondsyzygyfield2\].
[90]{} , *The F-signature and strong F-regularity*, Math. Res. Lett., vol. 10, no. 1, pp. 51–56, 2003.
, *On the Krull-Schmidt theorem with application to sheaves*, Bull. Soc. Math. France, vol. 84, pp. 307–317, 1956.
, *Vector bundles over an elliptic curve*, Proc. London Math. Soc. (3), vol. 7, pp. 414–452, 1957.
, *Rational singularities and almost split sequences*, Trans. Amer. Math. Soc., vol. 293, no. 2, pp. 511–531, 1986.
, *The Brauer group of a commutative ring*, Trans. Amer. Math. Soc., vol. 97, pp. 367–409, 1997.
, *Symmetric tensors and geometry of $\mathbb{P}^N$ subvarieties*, Geom. Funct. Anal. vol. 18, no. 3, 637–656, 2008.
, *Tight closure and plus closure for cones over elliptic curves*, Nagoya Math. J., vol. 177, pp. 31–45, 2005.
, *Symmetric powers of vector bundles and Hilbert-Kunz theory*, unpublished paper, 2009.
, *Restriction of the cotangent bundle to elliptic curves and Hilbert-Kunz functions*, Manuscripta Math., vol. 119, no. 1, pp. 17–36, 2006.
, *Maximal Cohen-Macaulay modules over surface singularities*, Trends in representation theory of algebras and related topics, pp. 101–166, EMS Ser. Congr. Rep., Eur. Math. Soc., Zürich, 2008.
, *The symmetric signature*, Ph.D. Thesis, Universität Osnabrück, 2016.
, *The symmetric signature of cyclic quotient singularities*, preprint arXiv:1603.06427.
, *Galois theory and Galois cohomology of commutative rings*, Mem. Amer. Math. Soc., no. 52, pp. 15–33, 1965.
, *The representation theory of finite groups*, North-Holland Mathematical Library, vol. 25, 1982.
, *Representation theory. A first course*, Graduate Texts in Mathematics, vol. 129, Springer, 1991.
, *Algebraic geometry*, Graduate Texts in Mathematics, vol. 52, Springer, 1977.
, *Generalized F-signature of invariant subrings*, J. Algebra, vol. 443, pp. 142–152, 2015.
, *Equivariant Matlis and the local duality*, J. Algebra, vol. 324, pp. 1447–1470, 2010.
, *Ein Cohen-Macaulay-Kriterium mit Anwendungen auf Konormalenmodul und den Differentialmodul*, Math. Z., vol. 163, no. 2, pp. 149–162, 1978.
, *Ringe mit nur endlich vielen Isomorphieklassen von maximalen, unzerlegbaren Cohen-Macaulay-Moduln*, Math. Ann., vol. 233, pp. 21–34, 1978.
, *Two theorems about maximal Cohen-Macaulay modules*, Math. Ann., vol. 324, no. 2, pp. 391–404, 2002.
, *Characterizations of regular local rings for characteristic p*, Amer. J. Math., vol. 91, pp. 772–784, 1969.
, *Positivity in Algebraic Geometry II, Positivity for Vector Bundles, and Multiplier Ideals*, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge / A Series of Modern Surveys in Mathematics, vol. 49, Springer, 2004.
, *Lectures on vector bundles*, Cambridge Studies in Advanced Mathematics, vol. 54, Cambridge University Press, 1997.
, *Cohen-Macaulay representations*, Mathematical Surveys and Monographs, vol. 181, American Mathematical Society, 2012.
, *Almost split sequences and Zariski differentials*, Trans. Amer. Math. Soc., vol. 319, pp. 285–307, 1990.
, *The Chern class and Kodaira dimension of a minimal variety*, Algebraic Geometry, Sendai 1985, Adv. Stud. Pure Math., vol. 10, pp. 449–476, 1987.
, *An introduction to invariants and moduli*, Cambridge Studies in Advanced Mathematics, vol. 81, Cambridge University Press, 2003.
, *Multilinear Algebra*, Cambridge University Press, 1984.
, *Vector bundles on a elliptic curve*, Nagoya Math. J., vol. 43, pp. 41–72, 1971.
, *Symmetric powers and the cotangent bundle and classification of algebraic varieties*, Lecture Notes in Math., vol. 732, pp. 545–563, 1978.
, *The Hilbert-Kunz function of rings of finite Cohen-Macaulay type*, Arch. Math. (Basel), vol. 69, no. 4, pp. 286–296, 1997.
, *Simplicity of rings of differential operators in prime characteristic*, Proc. London Math. Soc. (3), vol. 75, no. 1, pp. 32–62, 1997.
, *Poincaré Series of Binary Polyhedral Groups and McKay’s Correspondence*, Math. Ann., vol. 278, pp. 99–116, 1987.
, *Algorithms in invariant theory. Second edition*, Texts and Monographs in Symbolic Computation, Springer, Vienna, 2008.
, *Semistable bundles over an elliptic curve*, Adv. Math., vol. 98, no. 1, pp. 1–26, 1993.
, *F-signature exists*, Invent. Math., vol. 190, no. 3, pp. 743–765, 2012.
, *Certain invariant subrings are Gorenstein. I*, Osaka J. Math., vol. 11, pp. 1–8, 1974.
, *Certain invariant subrings are Gorenstein. II*, Osaka J. Math., vol. 11, pp. 379–388, 1974.
, *Hilbert-Kunz multiplicity and an inequality between multiplicity and colength*, J. Algebra, vol. 230, no. 1, pp. 295–317, 2000.
, *Minimal relative Hilbert-Kunz multiplicity*, Illinois J. Math., vol. 48, pp. 273–294, 2004.
, *Modules with finite F-representation type*, J. London Math. Soc. (2), vol. 72, pp. 53–72, 2005.
, *Observations on the F-signature of local rings of characteristic p*, J. Algebra, vol. 299, no. 1, pp. 198–218, 2006.
, *Cohen-Macaulay modules over Cohen-Macaulay rings*, London Mathematical Society Lecture Note Series, vol. 146, Cambridge University Press, Cambridge, 1990.
, *The fundamental module of a normal local domain of dimension $2$*, Trans. Amer. Math. Soc., vol. 309, no. 1, pp. 425–431, 1988.
[^1]: *Mathematics Subject Classification (2010)*: 13A35, 13A50.\
*Keywords*: F-signature, ADE singularities, free resolutions, Kähler differentials
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose a novel system for unsupervised skeleton-based action recognition. Given inputs of body keypoints sequences obtained during various movements, our system associates the sequences with actions. Our system is based on an encoder-decoder recurrent neural network, where the encoder learns a separable feature representation within its hidden states formed by training the model to perform prediction task. We show that according to such unsupervised training the decoder and the encoder self-organize their hidden states into a feature space which clusters similar movements into the same cluster and distinct movements into distant clusters. Current state-of-the-art methods for action recognition are strongly supervised, i.e., rely on providing labels for training. Unsupervised methods have been proposed, however, they require camera and depth inputs (RGB+D) at each time step. In contrast, our system is fully *unsupervised*, does not require labels of actions at any stage, and can operate with *body keypoints* input only. Furthermore, the method can perform on various dimensions of body keypoints (2D or 3D) and include additional cues describing movements. We evaluate our system on three extensive action recognition benchmarks with different number of actions and examples. Our results outperform prior unsupervised skeleton-based methods, unsupervised RGB+D based methods on cross-view tests and while being unsupervised have similar performance to supervised skeleton-based action recognition.'
author:
- Kun Su
- Xiulong Liu
- Eli Shlizerman
bibliography:
- 'egbib.bib'
title: 'PREDICT & CLUSTER: Unsupervised Skeleton Based Action Recognition'
---
Introduction
============
Robust action recognition, especially human action recognition, is a fundamental capability in ubiquitous computer vision and artificial intelligence systems. While recent methods have shown remarkable success rates in recognizing basic actions in videos, current methods rely on strong supervision with a large number of training examples accompanied with action labels. Collection and annotation of large scale datasets is implausible for various types of actions and applications. Furthermore, annotation is a challenging problem by itself, since it is often up to the interpretation of the annotator to assign a meaningful label for a given sequence. This is particularly the case in situations where it is unclear what is the ground truth label, e.g., annotation of animal movements. Indeed, annotations challenges are common in different contextual information on movement, such as video (RGB), depth (+D) and keypoints tracked over time. Compared to RGB+D data, keypoints include much less information and can be challenging to work with. However, on the other hand, focusing keypoints can often isolate the actions from other information and provide more robust unique features for actions.
For human action recognition, time-series of body joints (skeleton) tracked over time are indeed known as effective descriptors of actions. Here we focus on $3D$ skeleton time sequences and propose an unsupervised system to learn features and assign actions to classes according to them. We call our system PREDICT & CLUSTER (P&C) since it is based on training an encoder-decoder type network to both predict and cluster skeleton sequences such that the network learns an effective hidden feature representation of actions. Indeed, an intuitive replacement of a classification supervised task by a non-classification unsupervised task is to attempt to continue (predict) or reproduce (re-generate) the given sequence such that it leads the hidden states to capture key features of the actions. In the encoder-decoder architecture, the prediction task is typically implemented as follows: given an action sequence as the encoder input, the decoder continues or generates the encoder input sequence. Since inputs are sequences, both the decoder and the encoder are recurrent neural networks (RNN) containing cells with hidden variables for each time sample in a sequence. The final hidden state of the encoder is typically being utilized to represent the action feature. While the encoder contains the final action feature, since the gradient during training flows back from the decoder to the encoder, it turns out that the decoder training strategies significantly determine the effectiveness of the representation. Specifically, there are two types of decoder training strategies proposed for such prediction/re-generation task [@srivastava2015unsupervised]. The first strategy is a conditional strategy, where the output of the previous time-step of the decoder is used as input to the current time-step. With such strategy the output of the decoder is expected to be continuous. In contrast, the unconditional strategy assigns a zero input into each time-step of the decoder. Previous work showed that unconditional training of the decoder is expected to have better prediction performance since it effectively weakens the decoder and thus forces the encoder to learn a more informative representation.
In our system, we extend such strategies to enhance the encoder representation. This results in enhanced clustering and organization of actions in the feature space. In particular, we propose two decoder training strategies, Fixed Weights and Fixed States to further penalize the decoder. The implementation of these strategies guides the encoder to further learn the feature representation of the sequences that it processes. In fact, in both strategies, the decoder is a ‘weak decoder’, i.e., the decoder is effectively not being optimized and it serves the role of propagating the gradient to the encoder to further optimize its final state. Combining these two strategies together, we find that the network can learn a robust representation and our results show that this strategy can achieve significantly enhanced performance than unsupervised approaches trained without them. We demonstrate the effectiveness and the generality of our proposed methodology by evaluating our system on three extensive skeleton-based and RGB+D action recognition datasets. Specifically, we show that our P&C unsupervised system achieves high accuracy performance and outperforms prior methods.
{width="90.00000%"}
Related Work
============
The objective of action recognition is to assign a class label to a sequence of frames with context information on the action performed, Fig. \[fig:teaser\]. Numerous approaches have been introduced particularly for human movement action recognition. Such approaches use video frames (RGB), and/or depth (RGB+D) and/or skeleton data, i.e. tracking of body joints (keypoints). Performing exclusive skeleton-based action recognition is especially advantageous since requires much less data, which is relatively easy to acquire and therefore has the potential to be performed in real-time. Furthermore, in contrast to videos and depth, including various contexts such as background, skeleton data can be used to understand the exclusive features of the actions. Indeed, in recent years, various supervised and unsupervised approaches have been introduced for human skeleton-based action recognition. Most of skeleton-based approaches have been supervised approaches where an annotated set of actions and labels should be provided for training them. In an unsupervised setup, the problem of action recognition is much more challenging. Only a few unsupervised skeleton-based approaches have been proposed and several unsupervised approaches have been developed to use more information such as video frames and depth, i.e. unsupervised RGB+D. We review these prior approaches below and compare our results with them.
For *supervised* skeleton-based action recognition, prior to deep learning methods, classical approaches were proposed to map the actions from Lie group to its Lie algebra and to perform classification using a combination of dynamic time warping, Fourier temporal pyramid representation and linear SVM (e.g. LARP [@vemulapalli2014human]). Deep learning approaches have been developed to classify skeleton data as well, in particular, RNN based models that are designed to work with sequences. For example, Du et al. [@du2015hierarchical] used hierarchical RNN (HBRNN-L) for action classification and Shahroudy et al. [@shahroudy2016ntu] proposed part-aware LSTM (P-LSTM) as a baseline for the large scale skeleton action recognition NTU RGB+D dataset. Since skeleton data is noisy, largely due to variance in camera views, Zhang et al. [@zhang2019view] proposed a view-adaptive RNN (VA-RNN) which learns a transformation from original skeleton data to a general pose. CNN based approaches have been also proposed for supervised skeleton based recognition by constructing a representation of body joints that can be processed by CNN. In particular, Du et al. [@du2015skeleton] represented a skeleton sequence as a matrix by concatenating the joint coordinates in each instant and arranging those vector representations in a chronological order and transforming the matrix into an image on which CNN is trained for classification. In addition, Liu et al.[@liu2017enhanced] used an enhanced skeleton visualization method in conjunction with CNN classification for view invariant human action recognition. Recently, graph convolution networks (GNN) gained popularity in skeleton-based action recognition approaches. Yan et al. [@yan2018spatial] introduced Spatial Temporal Graph Convolutional Networks (ST-GCN), which were shown to be capable to learn both the spatial and temporal patterns from skeleton data. A recent extension of such an approach by Shi et al. [@shi2019skeleton],[@shi2019two] showed that directed GNN can be used to encode the skeleton representation and also showed that two-stream GNN can learn the graph in an adaptive manner.
While recent supervised approaches show robust performance on action recognition, the *unsupervised* setup is advantageous since it does not require labeling of sequences and may not need re-training when additional actions, not included in the training set, are introduced. Unsupervised methods typically aim to obtain an effective feature representations by predicting future frames of input action sequences or by re-generating the sequences. Unsupervised approaches were mostly proposed for videos of actions or videos with additional information such as depth or optical flow. Specifically, Srivastava et al. [@srivastava2015unsupervised] proposed a recurrent-based sequence to sequence (Seq2Seq) model as an autoencoder to learn the representation of a video. Such an approach is at the core of our method for body joints input data. However, as we show, the approach will not be able to achieve efficient performance without particular training strategies that we develop to weaken the decoder and strengthen the encoder. Subsequently, Luo et al. [@luo2017unsupervised] developed a convolutional LSTM to use depth and optical flow information such that the network encodes depth input and uses the decoder to predict the optical flow of future frames. Furthermore, Li et al. [@li2018unsupervised] proposed to employ a generative adversarial network (GAN) with a camera-view discriminator to assist the encoder in learning better representations.
As in unsupervised RGB+D approaches, skeleton-based approaches utilize the task of human motion prediction as the underlying task to learn action feature representation. For such a task, RNN-based Seq2Seq models [@martinez2017human] were shown to achieve improved accuracy in comparison to non-Seq2Seq based RNN models such as ERD [@fragkiadaki2015recurrent] and S-RNN [@jain2016structural]. Recently, networks incorporating GANs have achieved improved performance on this task by utilizing the predictor network being RNN Seq2Seq and the discriminator [@gui2018adversarial].
*Unsupervised* approaches for *skeleton-based action recognition* are scarce since obtaining effective feature representations from coordinate positions of body joints is challenging. In particular, based on successful human motion prediction network configurations, Zheng et al. [@zheng2018unsupervised] (LongT GAN) proposed a GAN encoder-decoder such that the decoder attempts to re-generate the input sequence and the discriminator is used to discriminate whether the re-generation is accurate. The feature representation used for action recognition is taken from the final state of the encoder hidden representation. During training, the masked ground truth input is provided to the decoder. The method was tested on motion-capture databases, e.g., CMU Mocap, HDM05[@cg-2007-2] and Berkeley MHAD[@ofli2013berkeley]. Such datasets were captured by physical sensors (markers) and thus are much cleaner than marker-less data collected by depth cameras and do not test for multi-view variance which significantly affects action recognition performance. Our baseline network architecture is similar to the structure in Zheng et al. [@zheng2018unsupervised] since we use an encoder and decoder and we also use the final state of the encoder as a features representation of the action sequences. However, as we show, it is required to develop extended training strategies for the system to be applicable to larger scale multi-view and multi-subject datasets. Specifically, instead of using the masked ground truth as an input into the decoder, we propose methods to improve learning of the encoder and to weaken the decoder.
![Pre-processing of body keypoints sequences according to view-invariant transformation.[]{data-label="fig:preprocess"}](Fig2_v1_new.png){width="\linewidth"}
Methods {#sec:methods}
=======
**Pre-processing of body keypoints**: Body keypoints data is a sequence $X^V_T$ of $T$ frames captured from a particular view, where each frame represents $N=3$D coordinates of $J$ joint keypoints $$\begin{aligned}
X_T^V = \{ x_1, x_2, \dots, x_T\}, x_t\in \mathbf{R}^{J\times N}.\end{aligned}$$ Action sequences are captured from different views by depth camera such as Microsoft Kinect and $3$D human joint positions are extracted from a single depth image by a real-time human skeleton tracking framework[@shotton2011real]. We align the action sequences by implementing a view-invariant transformation which transforms the keypoints coordinates from original coordinate system into a view-invariant coordinate system $X^V \to X$. The transformed skeleton joint coordinates are given by $$\begin{aligned}
x^j_t = R^{-1}(x^j_t - d_R), \forall j \in J, \forall t\in T,\end{aligned}$$ where $x^j_t \in \mathbf{R}^{3\times 1}$ are the coordinates of the $j$-th joint of the $t$-th frame, $R$ is the rotation matrix and $d_R$ is the origin of rotation. These are computed according to $$\begin{aligned}
R &= \left[ \frac{v_1}{\Vert{v_1}\Vert} \left| \frac{\hat{v}_2}{\Vert{\hat{v}_2}\Vert} \right| \frac{v_1 \times \hat{v}_2}{\Vert{v_1 \times \hat{v}_2}\Vert} \right],d_R = x^{\text{root}}_{t=0},\end{aligned}$$ where $v_1 = x^{\text{spine}}_{t=0} - x^{\text{root}}_{t=0}$ is the vector perpendicular to the ground, $v_2=x^{\text{hip left}}_{t=0} - x^{\text{hip right}}_{t=0}$ is the difference vector between left and right hips joints in the initial frame of each sequence and $\hat{v}_2 =\frac{v_2-\text{Proj}_{v_1}(v_2)}{\Vert v_2-\text{Proj}_{v_1}(v_2)\Vert}$. $\text{Proj}_{v_1}(v_2)$ and $v_1 \times \hat{v}_2$ denotes the vector projection of $v_2$ onto $v_1$ and the cross product of $v_1$ and $\hat{v}_2$, respectively. $x^{\text{Root}}_{t=0}$ is the coordinate of the root joint in the initial frame [@lee2017ensemble] (see Fig. \[fig:preprocess\]). Since actions can be of different lengths we down-sample each action sequence to be at most a fixed length $T_{max}$ and pad with zeros if the sequence length is smaller than that.\
**Self-organization of hidden states clustering:**
![Encoder states trajectories visualized by projection to 3 PCA space. Each color represents one type of action(blue: donning, red: sit down,green:carry, black: stand up). The cross symbol denotes the final state. Left: before training; Right: after training.[]{data-label="fig:encoderstates"}](traj_new.png){width="\linewidth"}
A key property that we utilize in our system is the recent observation that propagation of input sequences through RNN self-organizes them into clusters within the hidden states of the network, i.e., clusters represent features in an embedding of the hidden states[@farrell2019recurrent]. Such strategy is a promising unsupervised method for multi-dimensional sequence clustering such as body keypoints sequences [@suclustering]. As we show, self-organization is inherent to any RNN architecture and even holds for random RNN which are initialized with random weights and kept fixed, i.e., no training is performed. Indeed, when we input sequences of body keypoints of different actions into random RNN, the features in the hidden state space turn out to be effective filters. While such strategy is promising, the recognition accuracy outcome appears to be non-optimal (Table \[tab:Perf\_Table\] P&C Rand). We therefore implement an encoder-decoder system, which we call PREDICT & CLUSTER (P&C), where the encoder propagates input sequences and passes the last hidden state to the decoder. The decoder is used to regenerate the encoder input sequences. Furthermore, we utilize the random network setup (which does not require training) to choose the optimal hyper-parameters for the network to be trained. We describe the components of P&C below.\
**Motion prediction**: At the core of our unsupervised method is an encoder-decoder RNN (Seq2Seq). Such network models were shown to be effective at prediction of future evolution of multi-dimensional time-series of features including skeleton temporal data of various actions [@martinez2017human],[@gui2018adversarial]. In these applications the typical flow in the network is uni-directional. The *encoder* processes an initial sequence of activity and passes the last state to the *decoder* which in turn, based on this state generates the evolution forward. We extend such network structure for our method (see system overview in Fig. \[fig:sysoverview\]).
We propose a *bi-directional* flow such that the network can capture better long-term dependencies in the action sequences. Specifically, the encoder is a multi-layered bi-directional Gated Recurrent Unit (GRU) which input is a whole sequence of body keypoints corresponding to an action. We denote the forward and backward directions hidden states of the last layer of encoder at time $t$ as $ \overrightarrow{E_t}$ and $\overleftarrow{E_t}$ respectively, and the final state of the encoder as their concatenation $E_T = \{\overrightarrow{E_T} , \overleftarrow{E_T} \}$. The decoder is a uni-directional GRU with hidden states at time $t$ denoted as $D_t$. The final state of the encoder is fed into the decoder as its initial state, i.e., $D_0 = E_T$. In such a setup, the decoder generates a sequence based on $E_T$ initialization. In a typical prediction task, the generated sequence will be compared with forward evolution of the same sequence (prediction loss). In our system, since our goal is to perform action recognition, the decoder is required to re-generate the whole input sequence (re-generation loss). Specifically, for the decoder outputs $\hat{X} = \{\hat{x}_1, \hat{x}_2, \dots, \hat{x}_T\}$ the re-generation loss function is the error between $X$ and $\hat{X}$. In particular, we use mean square error (MSE) $L=\frac{1}{T}\sum^{T}_{t=1}(x_t-\hat{x}_t)^2$ or mean absolute error (MAE) $L=\frac{1}{T}\sum^{T}_{t=1} \left|x_t-\hat{x}_t\right|$ as plausible losses.\
**Hyper-parameter search**: As in any deep learning system, hyper-parameters significantly impact network performance and require tuning for optimal regime. We utilize the self-organization feature of random initialized RNN to propagate the sequences through the network and use network performance prior to training as an optimization for hyper-parameter tuning. Specifically, we evaluate the capacity of the encoder by propagating the skeleton sequence through the encoder and evaluate the performance of recognition on the final encoder state. We observe that this efficient hyper-parameter search significantly reduces total training time when an optimal network amenable for training is being selected.\
**Training**: With optimal hyper-parameter encoder being set, training is performed on the outputs of the decoder to predict (re-generate) the encoder’s input action sequence. Training for prediction is typically performed according to one of the two approaches: (i) *unconditional* training in which zeros are being fed into the decoder at each time step or (ii) *conditional* in which an initial input is fed into the first time-step of the decoder and subsequent time-steps use the predicted output of the previous time step as their input [@srivastava2015unsupervised]. Based on these training strategies, we propose two decoder configurations *(i) Fixed Weights decoder (FW)* or *(ii) Fixed States decoder (FS)* to weaken the decoder, i.e. to force it to perform the re-generation based upon the information provided by the hidden representation of the encoder and thus improve the encoder’s clustering performance, see Fig.\[fig:encoderstates\].\
*1.Fixed Weights decoder* (**FW**): The input into the decoder is unconditional in this configuration. The decoder is not expected to learn a useful information for prediction and it exclusively relies on the state passed by the encoder. The weights of the decoder can thereby be assigned as random and the decoder is used as a recurrent propagator of the sequences. In training for the re-generation loss such configuration is expected to *force the encoder to learn the latent features* and represent them with the final state passed to the decoder. This intuitive method turns out to be computationally efficient since only the encoder is being trained and our results indicate favorable performance in conjunction with KNN action classification.\
*2.Fixed States decoder* (**FS**): The external input into the decoder is conditional in this configuration ( external input into each time-step is the output of the previous time-step), however the internal input, typically the hidden state from previous step, is replaced by the final state of the encoder $E_T$. Namely, in RNN cell $$\begin{aligned}
&h_t = \sigma (W_x x_t + W_h g_{t}+b_h), g_{t}=h_{t-1} \to E_T, \\
&y_t = \sigma (W_y h_t+b_y),\\
&x_{t+1} = y_t,\end{aligned}$$ with $x_t$ the external input, $y_t$ the output and $h_t$ the hidden state at time-step $t$, $h_{t-1}$ terms are replaced by $E_T$. In addition, we also add residual connection between external input and output, which has been shown useful in human motion prediction as well [@martinez2017human]. The final output and next input will be $\hat{y}_t = y_t + x_t$ and $\hat{x}_{t+1}=\hat{y}_t$, respectively. The configuration forces the network to rely on $E_T$, instead of the hidden state at previous time-step and eliminates vanishing of the gradient since during back-propagation at each time-step there is a defined gradient back to the final encoder state.\
![Feature-level autoencoder and KNN Classifier[]{data-label="fig:classify"}](Fig4_v1.png){width="0.9\linewidth"}
{width="0.95\linewidth"}
**Feature level auto-encoder**: After training the prediction network we extract the final encoder state $E_T$ as the feature vector associated with each action sequence. Since the feature vector is high-dimensional, we use a feature-level auto-encoder that learns the core low dimensional components of the high-dimensional feature so it can be utilized for classification (Fig. \[fig:classify\]). Specifically, we implement the auto-encoder, denoted as $f$ to be of an encoder-decoder architecture with parameters $\theta$ such that $$\hat{E_T} = f_{\theta}(E_T)\approx E_T.$$ The encoder and the decoder are multi-layer FC networks with non-linear $\tanh$ activation and we implement the following loss $l_{aec}=\left| E_T-\hat{E_T} \right|$.\
**K-nearest neighbors classifier**: For evaluation of our method on action recognition task we use a K-nearest neighbors (KNN) classifier on the middle layer of the auto-encoder feature vector. Specifically, we apply the KNN classifier (with $k=1$) on the features of the trained network on all sequences in the training set to assign classes. We then use cosine similarity as the distance metric to perform recognition, i.e., place each tested sequence in a class. Notably, KNN classifier does not require to learn extra weights to action placement.
Experimental Results and Datasets
=================================
**Implementation details**: To train the network, all body keypoints sequences are pre-processed according to the view-invariant transformation and down-sampled to have at most $50$ frames (Fig. \[fig:preprocess\]). The coordinates are also normalized to the range of $[-1,1]$. Using the hyper-parameter search, employing random RNN propagation discussed above, we set the following architecture: *Encoder*: $3$-Layer Bi-GRU with $N=1024$ units in each layer. *Decoder*: 1-Layer Uni-GRU with $N=2048$ units such that it is compatible with the dimensions of the encoder final state $E_T$.All GRUs are initialized with a random uniform distribution. *Feature-level auto-encoder:* $6$ FC Layers with the following dimensions: input feature vector(dim$=2048$) $\rightarrow$ FC($1024$) $\rightarrow$ FC($512$) $\rightarrow$ FC($256$)$\rightarrow$ FC($512$)$\rightarrow$FC($1024$) $\rightarrow$FC($2048$). All FCs use $\tanh$ activation except the last layer which is linear. The middle layer of auto-encoder outputs a vector feature of $256$ elements which is used as the final feature. We use Adam optimizer and learning rate starting from $0.0001$ and $0.95$ decay rate at every $1000$ iterations. The gradients are clipped if the norm is greater than $25$ to avoid gradient explosion. It takes $0.7$sec per training iteration and $0.1$sec to forward propagate with batch size of $64$ on one Nvidia Titan X GPU. Please see additional details of architecture choices in the supplementary material.\
**Datasets**: We use three different data-sets for training, evaluation and comparison of our P&C system with related approaches. The three data-sets include various number of classes, types of actions, body keypoints captured from different views and on different subjects. In these datasets, the body keypoints are captured by depth cameras and also include additional data, e.g., videos (RGB) and depth (+D). Various types of action recognition approaches have been applied to these datasets, e.g., supervised skeleton approaches and unsupervised RGB+D approaches. We list these types of approaches and their performance on the tests in the datasets in Table \[tab:Perf\_Table\]. Notably, as far as we know, our work is the **first fully unsupervised skeleton based approach applied on these extensive action recognition tests**.
The datasets that we have applied our P&C system to are *(i) NW-UCLA*, *(ii) UWA3D*, and *(iii) NTU RGB+D*. The datasets include $3$D body keypoints of $10, 30, 60$ action classes respectively. We briefly describe them below. **North-Western UCLA (NW-UCLA)** dataset [@wang2014cross] is captured by Kinect v$1$ and contains $1494$ videos of $10$ actions. These actions are performed by $10$ subjects repeated $1$ to $6$ times. There are three views of each action and for each subject $20$ joints are being recorded. We follow [@liu2017enhanced] and [@wang2014cross] to use the first two views (V$1$,V$2$) for training and last views (V$3$) to test cross view action recognition. **UWA3D Multiview Activity II (UWA3D)** dataset [@rahmani2014hopc] contains $30$ human actions performed $4$ times by $10$ subjects. $15$ joints are being recorded and each action is observed from four views: frontal, left and right sides, and top. The dataset is challenging due to many views and the resulting self-occlusions from considering only part of them. In addition, there is a high similarity among actions, e.g., the two actions "drinking” and “phone answering” have many keypoints being near identical and not moving and there are subtle differences in the moving keypoints such as the location of the hand. **NTU RGB+D** dataset [@shahroudy2016ntu] is a large scale dataset for $3$D human activity analysis. This dataset consists of $56,880$ video samples, captured from $40$ different human subjects, using Microsoft Kinect v2. NTU RGB+D($60$) contains $60$ action classes. We use the $3$D skeleton data for our experiments such that each time sample contains $25$ joints. We test our P&C method on both cross-view and cross-subject protocols.
{width="0.8\linewidth"}
![t-SNE visualization of learned features on NW-UCLA dataset.[]{data-label="fig:ucla_tsne"}](UCLA_tSNE_final.png){width="0.7\linewidth"}
Evaluation and Comparison
=========================
**Evaluation**: In all experiments, we use the K-nearest neighbors classifier with $k=1$ to compute the action recognition accuracy and evaluate the performance of our P&C method. We test different variants of P&C architectures (combinations of components described in Section \[sec:methods\]) and report a subset of these in the paper: *baseline* random initialized encoder with no training (**P&C-Rand**), full system with FS decoder and feature-level auto-encoder (**P&C-FS-AEC**) and full system with FW decoder and feature-level auto-encoder (**P&C-FW-AEC**). We report the rest of the combinations and their results in the Supplementary material.
Fig. \[fig:all\_acc\] shows the optimization of the regeneration loss (red) and the resulting accuracy (blue) during training for each dataset. We include plots of additional P&C configurations in the Supplementary material. The initial accuracy appears to be substantial and this is attributed to the hyper-parameter search being performed on random initialized networks prior to training that we describe in Section \[sec:methods\]. Indeed, we find that using appropriate initialization, the encoder, without any training, effectively directs similar action sequences to similar final states. Training enhances this performance further in both P&C FW and P&C FS configurations. Over multiple training iterations both P&C FW and P&C FS achieve higher accuracy than no FW and no FS in all datasets. While the convergence of the loss curve indicates improvement on the accuracy, the value of the loss does not necessarily indicate a better accuracy as can be observed from loss and accuracy curves of training on UWA3D and NTU-RGB+D (Fig. \[fig:all\_acc\] middle, right). We show the confusion matrices for the three considered datasets in Fig. \[fig:all\_cm\]. In NW-UCLA (with least classes) we show the elements of the 10x10 matrix. Our method achieves high-accuracy ($>83\%$) on average and there are three actions (pick up with two hands, drop trash, sit down) for which it recognizes with nearly $100\%$ accuracy. We also show in Fig. \[fig:ucla\_tsne\] a t-SNE visualization of the learned features for NW-UCLA test. Even in this $2$D embedding it is clearly evident that the features for each class are well separated. As more action classes are considered, the recognition becomes a more difficult task and also depends on amount of training data. For example, while NTU RGB+D has more classes than UWA3D, the recognition accuracy on NTU RGB+D is smoother and results with better performance since it has $40$ times more data than UWA3D. Our results show that our method is compatible with varying data sizes and number of classes.\
**Comparison**: We compare the performance of our P&C method with prior related supervised and unsupervised methods applied to (left-to-right): NW-UCLA, UWA3D, NTU RGB+D datasets, see Table \[tab:Perf\_Table\]. In particular, we compare action recognition accuracy with approaches based on supervised skeleton data (blue), unsupervised RGB+D data (purple) and unsupervised skeleton data (red). For comparison with unsupervised skeleton methods, we implement and reproduce the LongTerm GAN model (LongT GAN) as introduced in [@zheng2018unsupervised] and list its performance.
For NW-UCLA, P&C outperforms previous unsupervised methods (both RGB+D and skeleton based). Our method even outperforms the first three supervised methods listed in Table \[tab:Perf\_Table\]-left. UWA3D is considered a challenging test for many deep learning approaches since the number of sequences is small, while it includes a large number of classes ($30$). Indeed, action recognition performance of many supervised skeleton approaches is low ($<50\%$). For such datasets, it appears that the unsupervised approach could be more favorable, i.e., even P&C Rand reaches performance of $\approx 50\%$. LongT GAN achieves slightly higher performance than P&C Rand, however, not as high as P&C FS/FW-AEC which perform with $\approx 60\%$. Only a single supervised skeleton method, VA-RNN-Aug, is able to perform better than our unsupervised approach, see Table \[tab:Perf\_Table\]-middle. On the large scale NTU-RGB+D dataset, our method performs extremely well on the cross-view test. It outperforms prior unsupervised methods (both RGB+D and skeleton based) and on-par with ST-LSTM (second best supervised skeleton method), see Table \[tab:Perf\_Table\]-right. On the cross-subject test we obtain performance that is higher (including P&C Rand) than the prior unsupervised skeleton approach, however, our accuracy does not outperform unsupervised RGB+D approaches. We believe that the reason stems from skeleton based approaches not performing well in general on cross-subject tests since additional aspects such as subjects parameters, e.g., skeleton geometry and invariant normalization from subject to subject, need to be taken into account.
In summary, for all three datasets, we used a single architecture and it was able to outperform the prior unsupervised skeleton method, LongT-GAN[@zheng2018unsupervised], most supervised skeleton methods and unsupervised RGB+D methods on cross view tests and some supervised skeleton and unsupervised RGB+D on large scale cross subject test.
Conclusion
==========
We presented a novel unsupervised model for human skeleton-based action recognition. Our system reaches enhanced performance compared to prior approaches due to novel training strategies which weaken the decoder and training of the encoder. As a result the network learns more separable representations. Experimental results demonstrate that our unsupervised model can effectively learn distinctive action features on three benchmark datasets and outperform prior unsupervised methods.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Following the work of Maldacena and Zhiboedov, we study the implementation of the Coleman-Mandula theorem in the free $O(N)$/Higher Spin correspondence. In the bi-local framework we first define an $S$-matrix for scattering of collective dipoles. Its evaluation in the case of free UV fixed point theory leads to the result $S=1$ stated in the title. We also present an appropriate field transformation that is seen to transform away all the non-linear $1/N$ interactions of this theory. A change of boundary conditions and/or external potentials results in a nontrivial $S$-matrix.'
---
[BROWN-HET-1629]{}\
[WITS-CTP-096 ]{}
\
[Robert de Mello Koch$^a$, Antal Jevicki$^b$, Kewang Jin$^c$,\
João P. Rodrigues$^a$, and Qibin Ye$^b$]{}
[*$^a$National Institute for Theoretical Physics,\
School of Physics and Centre for Theoretical Physics,\
University of the Witwatersrand, Wits 2050, South Africa\
,* ]{}
[*$^b$Department of Physics, Brown University, Providence, RI 02912, USA\
,* ]{}
[*$^c$Institut für Theoretische Physik, ETH-Zürich, CH-8093 Zürich, Switzerland\
*]{}
Introduction
============
The duality involving $O(N)$ vector models and Vasiliev’s Higher Spin Gravity in Anti de Sitter space represent some of the simplest examples of AdS/CFT and as such is a topic of vigorous recent studies [@Klebanov:2002ja; @Sezgin:2002rt; @Das:2003vw; @Giombi:2009wh; @Giombi:2010vg; @Koch:2010cy; @Douglas:2010rc; @Jevicki:2011ss; @Shenker:2011zf; @Maldacena:2011jn; @Bekaert:2012ux; @Vasiliev:2012vf; @Maldacena:2012sf]. Equally interesting is the correspondence between 2d minimal CFT’s and 3d Chern-Simons Higher Spin Gravity [@Henneaux:2010xg; @Campoleoni:2010zq; @Gaberdiel:2010pz; @Gaberdiel:2011zw; @Chang:2011mz; @Gary:2012ms; @Gaberdiel:2012ku]. These large $N$ dualities involve quantum field theories that have been thought to be understood for some time and a relatively novel version of HS Gravity built on a single Regge trajectory. These theories feature many properties that have been unreachable in String Theory, in particular the structure and explicit form of the higher spin gauge symmetry group. They also offer a potentially solvable framework for studies of black hole formation and de Sitter theory itself [@Anninos:2011ui; @Ng:2012xp].
In the case of three dimensional $O(N)$ vector field theory, one has two conformally invariant fixed points, the UV and the IR one. The HS duals are given by the same Vasiliev theory [@Vasiliev:1995dn; @Vasiliev:2003ev; @Bekaert:2005vh; @Metsaev:2005ar; @Metsaev:2007rn; @Polyakov:2010sk; @Iazeolla:2011cb] but with different boundary conditions on the scalar field [@Klebanov:2002ja]. This provides a simple relationship between a (HS) theory dual to a free $N$-component scalar field (UV) and the nontrivial dual corresponding to the IR CFT. The correspondence provided by the free $O(N)$ scalar field theory is then of central interest. This theory is characterized by an infinite sequence of conserved currents which are themselves boundary duals of higher spin fields and whose correlation functions represent a point of comparison [@Giombi:2009wh; @Giombi:2010vg] between the two descriptions.
Since conserved currents imply the existence of an infinite sequence of conserved charges and higher symmetries, one is faced with the question regarding the implementation (and implication) of the Coleman-Mandula theorem. This question was raised and addressed in the recent work of Maldacena and Zhiboedov [@Maldacena:2011jn; @Maldacena:2012sf] who were able to show the existence of conserved currents (charges) implies that the correlation functions are built in terms of free fields. This demonstrates the simplicity of the corresponding Vasiliev theory. One still, however, has the question regarding the triviality of the theory in the bulk. In standard field theories this question is addressed (and answered) through the $S$-matrix. The Coleman-Mandula theorem in particular would imply $S=1$ for theories with higher symmetries. Due to the equivalence theorem (under field transformations) this means that there exists a field redefinition which linearizes the field equations. In the AdS/CFT framework one sometimes thinks of the correlators as taking the role analogous to an $S$-matrix. A proposal along this line offered by Mack [@Mack:2009mi; @Mack:2009gy] has been nicely implemented in recent works [@Penedones:2010ue; @Fitzpatrick:2011ia; @Paulos:2011ie; @Nandan:2011wc]. If this analogy is taken at face value, one has the puzzling fact that this $S$-matrix is non-trivial, even for the correspondence based on free theory.
We have in [@Das:2003vw; @Koch:2010cy] formulated a constructive approach to bulk AdS duality and HS Gravity in terms of bi-locals. It leads to a nonlinear, interacting theory (with $1/N$ as the coupling constant) which was seen to posses all the properties of the dual AdS theory. This theory reproduces arbitrary-point correlation functions and provides a construction of HS theory (in various gauges) based on CFT [@Jevicki:2011ss]. The construction also, as we will explain, offers a framework for defining and calculating an $S$-matrix and addressing the implementation of the Coleman-Mandula theorem in the nonlinear bulk framework. The construction is based on the two-particle collective dipole and its interactions in the large $N$ limit. It has been known since the early works that nontrivial collective phenomena can appear “even” for free theories (for example excitations near the large $N$ fermion surface). In the present case we are led to consider the $S$-matrix for collective dipoles: the corresponding LSZ reduction formula is easily stated as a limit of bi-local correlators. Its evaluation will produce the result $S=1$ as claimed in the title.[^1]
$S=1$ implies triviality, namely that interactions can be removed by a nonlinear transformation of fields (by this we mean the $1/N$ interactions which equal $G_N$ interactions in Vasiliev’s theory). We demonstrate this for the nonlinear dipole representation, where we establish a construction of a nonlinear field transformation that linearizes the bi-local field theory. One might (naively) wonder if the linearization is (not) just a change from bulk HS fields to the $N$-component CFT fields. One should remember however that the large $N$ duality is Quantum $\leftrightarrow$ Classical, namely that for recovering the classical AdS theory quantum effects in CFT need to be taken into account. Consequently the issue of a transformation between the two boils down to the construction of the (elusive) master field construction at large $N$. Some knowledge on this exists [@Gopakumar:1994iq; @Douglas:1994kw].
The content of this paper is as follows. In section \[sec:CFT\] we summarize the basics of the $O(N)$ model and the associated (nonlinear) bi-local field construction of [@Koch:2010cy; @Jevicki:2011ss]. In section \[sec:CM\], after a summary of the Maldacena-Zhiboedov result, we discuss the differences between “Boundary $S$-matrix” and “Collective $S$-matrix” that we propose. In particular we give an LSZ formula for the $S$-matrix and evaluate the associated three- and four-point amplitudes using the cubic and quartic vertices of the $1/N$ theory demonstrating the result $S=1$. In section \[sec:Field\] we present a construction of a nonlinear bi-local field transformation that linearizes the theory. Conclusions are given in section \[sec:Con\].
Bi-local representation of $\mathbf{O(N)}$ CFT$_\mathbf{3}$ {#sec:CFT}
===========================================================
In the $O(N)$/Higher Spin duality one starts with a $N$-component scalar field theory $$\mathcal{L}= \frac{1}{2}\partial_\mu \phi^a \partial^\mu \phi^a+\frac{g}{4}(\phi\cdot\phi)^2 \ , \qquad a=1,\cdots,N$$ where $\phi^a=\phi^a(t,\vec{x})=\phi^a(x^+,x^-,x^\perp)$. This theory features two critical points with conformal symmetry: the UV fixed point at zero coupling ($g=0$) and the nontrivial IR fixed point at nonzero value of the constant coupling constant. The latter can be evaluated in the large $N$ limit and serves as the classic example of critical phenomena in 3d.
For the correspondence with higher spin fields, a central role is played by the sequence of traceless and symmetric higher spin currents $$J_{\mu_1 \mu_2 \cdots \mu_s}=\sum_{k=0}^s (-1)^k \begin{pmatrix} s-1/2 \cr k \end{pmatrix}
\begin{pmatrix} s-1/2 \cr s-k \end{pmatrix} \partial_{\mu_1} \cdots \partial_{\mu_k} \phi^a \;
\partial_{\mu_{k+1}} \cdots \partial_{\mu_s} \phi^a -\text{traces}$$ which are exactly conserved in the free case. These operators can be summarized in the semi bi-local form by the generating function $$\mathcal{O}(x,\epsilon)=\phi^a(x-\epsilon)\sum_{n=0}^\infty \frac{1}{(2n)!}
\left(2\epsilon^2 \overleftarrow{\partial_x} \cdot \overrightarrow{\partial_x}
-4(\epsilon \cdot \overleftarrow{\partial_x})(\epsilon \cdot \overrightarrow{\partial_x})\right)^n
\phi^a(x+\epsilon)$$ where $\epsilon^2=0$ (in order to satisfy the traceless condition). As a result, $\epsilon$ represents a cone with a two dimensional coordinate and altogether $\mathcal{O}(x,\epsilon)$ is a five dimensional semi bi-local field. The currents that it generates represent boundary duals of AdS$_4$ higher spin fields $$J_{\mu_1 \mu_2 \cdots \mu_s}(x) \leftrightarrow \mathcal{H}_{\hat{\mu}_1 \hat{\mu}_2 \cdots \hat{\mu}_s}(x,z\rightarrow 0)$$ where $ds^2=\frac{dx^2+dz^2}{z^2}$ is the AdS$_4$ metric. In the AdS/CFT correspondence, correlation functions of currents are to match up with the boundary transition amplitude (sometimes referred to as the boundary $S$-matrix) of the higher dimensional AdS theory. A successful demonstration of this was accomplished in the three-point case by Giombi and Yin [@Giombi:2009wh; @Giombi:2010vg] who were able to match the two critical points of the vector model with two versions of Vasiliev’s Higher Spin Gravity in AdS$_4$. The trivial and nontrivial fixed points are seen as conjectured by Klebanov and Polyakov [@Klebanov:2002ja] to correspond to different boundary conditions involving the lowest spin ($s=0$) field.
A constructive approach for this AdS$_4$/CFT$_3$ correspondence was given in [@Das:2003vw], based on the notion of collective fields. These are given by bi-local invariants of the $O(N)$ field theory $$\Phi(x,y)\equiv \phi(x)\cdot\phi(y)=\sum_{a=1}^N\phi^a(x)\cdot\phi^a(y)$$ and close under the Schwinger-Dyson equations (in the large $N$ limit). These operators represent a more general set than the conformal fields $\mathcal{O}(x,\epsilon)$ since there is no restriction to a cone. The collective action evaluates the complete $O(N)$ invariant partition function $$Z=\int [d\phi^a(x)] e^{-S[\phi]}=\int \prod_{x,y} [d\Phi(x,y)] \mu(\Phi) e^{-S_c[\Phi]}$$ where the measure is given by $\mu(\Phi)=(\det \Phi)^{V_x V_p}$ with $V_x=L^3$ the volume of space and $V_p=\Lambda^3$ the volume of momentum space where $\Lambda$ is the momentum cutoff. Explicitly one has the collective action $$S_c[\Phi]=\text{Tr}\left[-(\partial_x^2+\partial_y^2)\Phi(x,y)+V \right]+\frac{N}{2}\text{Tr} \, \ln \Phi(x,y)$$ where the trace is defined as ${\rm Tr} \, B=\int d^3 x \, B(x,x)$. This collective action is nonlinear, with $1/N$ appearing as the expansion parameter. Through the identification of $1/N $ with $G_N$ (the coupling constant of higher spin gravity), this collective field representation provides a bulk description of the dual AdS theory. One also has a natural (star) product defined as $(\Psi \star \Phi)(x,y)=\int dz \, \Psi(x,z)\Phi(z,y)$.
The perturbation expansion is defined in this (bi-local) space. The nonlinear equation of motion specified by $S_c$ gives the background in the expansion: $\Phi=\Phi_0+\frac{1}{\sqrt{N}}\eta$. Expanding about the background gives rise to an infinite number of interaction vertices [@deMelloKoch:1996mj] $$\begin{aligned}
S_c=S[\Phi_0]+\text{Tr}[\Phi_0^{-1} \eta \Phi_0^{-1} \eta]+\frac{g}{4}\eta^2+\sum_{n \ge 3} N^{1-n/2} \, \text{Tr}B^n \ , \quad \text{with} \quad B=\Phi_0^{-1} \eta \ .\end{aligned}$$ The nonlinearities built into $S_c$ are precisely such that all invariant correlators: $\langle \phi(x_1) \cdot \phi(y_1) \cdots \phi(x_n) \cdot \phi(y_n)\rangle$ are now reproduced through the Witten (Feynman) diagrams with $1/N$ vertices (the four-point example is shown in Figure \[fourdiagram\]). We stress that this nonlinear structure is there for both the free and the interacting fixed point.
![Illustration of the four-point collective field diagrams.[]{data-label="fourdiagram"}](fourptcoll.eps){width="55.00000%"}
This bi-local theory is expected to represent a (covariant) gauge fixing of Vasiliev’s gauge invariant theory. An attempt at a gauge invariant formalism is given in [@Douglas:2010rc]. A large number of degrees of freedom are removed in fixing a gauge in any gauge theory and this simplification happens in Higher Spin Gravity too. A one-to-one relationship between bi-local and AdS higher spin fields can be demonstrated in a physical (single-time) picture. The existence of such a gauge and the discussion of the collective dipole underlying the collective construction is given in [@Jevicki:2011ss]. The single-time formulation now involves the equal time bi-local $$\Psi(t,\vec{x},\vec{y})=\sum_a \phi^a(t,\vec{x})\phi^a(t,\vec{y}) \label{bilocal}$$ and its conjugate momenta: $\Pi(\vec{x},\vec{y})=-i\frac{\delta}{\delta\Psi(\vec{x},\vec{y})}$ with a Hamiltonian of the form $$\begin{aligned}
H=2\text{Tr}(\Pi \Psi \Pi)+\frac{1}{2}\int d\vec{x} [-\nabla_{\vec{x}}^2\Psi(\tilde{x},\tilde{y})\vert_{\tilde{x}=\tilde{y}}]+\frac{N^2}{8}\text{Tr} \, \Psi^{-1} \ ,\end{aligned}$$ where we have set the coupling constant $g=0$. This Hamiltonian again has a natural $1/N$ expansion, after a background shift $$\Psi=\Psi_0+\frac{1}{\sqrt{N}} \eta \ , \qquad \Pi=\sqrt{N} \pi \ ,$$ one gets a quadratic Hamiltonian $$H^{(2)}=2 {\rm Tr}( \pi \Psi_0 \pi )+\frac{1}{8} {\rm Tr} (\Psi_0^{-1} \eta \Psi_0^{-1} \eta \Psi_0^{-1}) \ .$$ Fourier transform the background field as well as the fluctuations $$\begin{aligned}
\Psi^0_{\vec{x}\vec{y}} &=& \int d\vec{k} e^{i \vec{k} \cdot (\vec{x}-\vec{y})} \psi^0_{\vec{k}} \ , \qquad \psi^0_{\vec{k}}=\frac{1}{2 \sqrt{\vec{k}^2}} \ , \\
\eta_{\vec{x}\vec{y}} &=& \int d \vec{k}_1 d \vec{k}_2 e^{-i \vec{k}_1 \cdot \vec{x}+i \vec{k}_2 \cdot \vec{y}} \eta_{\vec{k}_1 \vec{k}_2} \ , \\
\pi_{\vec{x}\vec{y}} &=& \int d \vec{k}_1 d \vec{k}_2 e^{+i \vec{k}_1 \cdot \vec{x}-i \vec{k}_2 \cdot \vec{y}} \pi_{\vec{k}_1 \vec{k}_2} \ ,\end{aligned}$$ one finds the quadratic Hamiltonian in momentum space $$H^{(2)}=\frac{1}{2} \int d \vec{k}_1 d \vec{k}_2 \; \pi_{\vec{k}_1 \vec{k}_2} \pi_{\vec{k}_1 \vec{k}_2}+\frac{1}{8} \int d \vec{k}_1 d \vec{k}_2 \; \eta_{\vec{k}_1 \vec{k}_2}
\left( \psi_{\vec{k}_1}^{0\,\,-1}+\psi_{\vec{k}_2}^{0\,\,-1} \right)^2\eta_{\vec{k}_1 \vec{k}_2}$$ representing the (singlet) spectrum $\omega_{\vec{k}_1 \vec{k}_2}=\sqrt{\vec{k}_1^2}+\sqrt{\vec{k}_2^2}$ of the $O(N)$ theory. A sequence of $1/N$ vertices representing interactions can be found similarly and the cubic and quartic interactions are given explicitly as $$\begin{aligned}
H^{(3)}&=&\frac{2}{\sqrt{N}}\text{Tr}(\pi \eta \pi)-\frac{1}{8\sqrt{N}}\text{Tr} \Psi_0^{-1} \eta \Psi_0^{-1} \eta \Psi_0^{-1} \eta \Psi_0^{-1} \ , \label{cubic} \\
H^{(4)}&=&\frac{1}{8N}\text{Tr} \Psi_0^{-1} \eta \Psi_0^{-1} \eta \Psi_0^{-1} \eta \Psi_0^{-1} \eta \Psi_0^{-1} \ .
\label{quartic}\end{aligned}$$ We note that the form of these vertices is the same for both the free (UV) and the interacting (IR) conformal theories (the only difference is induced by the different background shifts in these two cases).
One also has a null-plane version of this construction which would correspond to light-cone gauge higher spin theory. This was used in [@Koch:2010cy] to demonstrate the one-to-one map between the two descriptions: the null-plane bi-locals $\Psi(x^+;x_1^-,x_2^-;x_1,x_2)$ and the higher spin fields $\mathcal{H}(x^+;x^-,x,z;\theta)$ in AdS$_4$, where $\theta$ is higher spin coordinate generating a sequence of higher spins. Both fields have same number of dimensions $1+2+2=1+3+1$, the same representation of the conformal group, and the same number of degrees of freedom.
The explicit canonical transformation was given in [@Koch:2010cy] as $$\begin{aligned}
x^-&=&\frac{x_1^-p_1^++x_2^-p_2^+}{p_1^++p_2^+} \ , \\
x&=&\frac{x_1 p_1^++x_2 p_2^+}{p_1^++p_2^+} \ , \\
z&=&\frac{\sqrt{p_1^+ p_2^+}}{p_1^++p_2^+}(x_1-x_2) \ , \\
\theta&=&2\arctan \sqrt{p_2^+ / p_1^+} \ ,\end{aligned}$$ where $p_i^+$ are the conjugate momenta of $x_i^-$. The map going from the bi-local field to the higher spin field is given by an integral transformation $$\begin{aligned}
\text{$\mathcal{H}$}(x^-,x,z,\theta)&=&\int dp^+dp^xdp^z \, e^{i(x^-p^++xp^x+zp^z)}\int dp_1^+dp_1dp_2^+dp_2\cr
&&\delta(p_1^++p_2^+-p^+)\delta(p_1+p_2-p^x)\delta(p_1\sqrt{p_2^+/p_1^+}-p_2\sqrt{p_1^+/p_2^+}-p^z)\cr
&&\delta(2\arctan \sqrt{p_2^+/p_1^+}-\theta)\tilde{\Psi}(p_1^+,p_2^+,p_1,p_2) \ ,\end{aligned}$$ where $\tilde{\Psi}(p_1^+,p_2^+,p_1,p_2)$ is the Fourier transform of the bi-local field $\Psi(x_1^-,x_2^-,x_1,x_2)$.
It was shown in [@Koch:2010cy] that under this transformation all the generators of collective field theory map into the generators of light-cone gauge Higher Spin Gravity in the form given by Metsaev [@Metsaev:1999ui]. In particular the quadratic bi-local Hamiltonian $$P^{-(2)}=\int dx_1^- dx_1 dx_2^- dx_2 \, \Psi^\dagger \left(-\frac{\nabla_1^2}{2\partial_{x_1^-}}-\frac{\nabla_2^2}{2\partial_{x_2^-}} \right) \Psi$$ takes an AdS$_4$ form $$P^{-(2)}=\int dx^- dx dz d\theta \; \mathcal{H}^\dagger \left(-\frac{\partial_x^2+\partial_z^2}{2\partial_{x^-}} \right) \mathcal{H} \ .$$ This establishes at the quadratic level the bi-local representation is identical to the local AdS$_4$ higher spin representation. One should note that the $1/N$ vertices do not become local in AdS spacetime. Actually the light-cone gauge fixing of Vasiliev’s theory has not been established yet at the nonlinear level, so one would expect a nonlocal form.
Another important check regarding the identification of the “extra” AdS coordinate $z$ can be seen by taking the $z \rightarrow 0$ limit. Evaluating the bi-local field at $z=0$ gives the following “boundary” form $$\text{$\mathcal{H}$}(x^+,x^-,x,\theta)=\int dp_1^+dp_2^+e^{ix^-(p_1^++p_2^+)}\delta(\theta-2\tan ^{-1}\sqrt{p_2^+/p_1^+})\tilde{\Psi}(p_1^+,p_2^+;x,x) \ .
\label{agreecurrent}$$ Expanding the kernel in the above transformation into Fourier series, for a fixed even spin $s$, one has the binomial expansion $$\left(\sqrt{p_1^+}-i\sqrt{p_2^+}\right)^{2s}=\frac{(-1)^k s! \; \Gamma(s+\frac{1}{2})\Gamma(\frac{1}{2})}{k!(s-k)! \; \Gamma(s-k+\frac{1}{2})\Gamma(k+\frac{1}{2})} (p_1^+)^k \, (p_2^+)^{s-k} \ .$$ This is to be compared with conformal operators of a fixed spin $s$ which are explicitly given in [@Makeenko:1980bh; @Braun:2003rp] by $$\text{$\mathcal{O}$}^s=\sum_{k=0}^s\frac{(-1)^k \; \Gamma(s+\frac{1}{2})\Gamma(s+\frac{1}{2})}{k!(s-k)! \; \Gamma(s-k+\frac{1}{2})\Gamma(k+\frac{1}{2})}(\partial_+)^k \phi \, (\partial_+)^{s-k} \phi \ ,$$ which agree with up to an overall normalization constant.
As a result, in the bi-local picture one has a clear definition of the boundary $z=0$ and the notion of boundary amplitudes (boundary $S$-matrix). Due to the construction through collective field theory, one is guaranteed to reproduce the boundary correlators in full agreement with the $O(N)$ model which are now reproduced in terms of Witten diagrams through higher $n$-point vertices as always in AdS duals. The bulk/bi-local theory is nonlinear with nonlinearities governed by $1/N=G_N$. All this provides a nontrivial check of the collective picture and the proposal that bi-local fields provide a bulk representation of AdS$_4$ higher spin fields.
Coleman-Mandula Theorem in AdS$_\mathbf{4}$/CFT$_\mathbf{3}$ {#sec:CM}
============================================================
The simplest case of the correspondence involves the UV fixed point CFT of noninteracting $N$-component bosonic or fermionic fields. These theories are characterized by the existence of an infinite sequence of higher spin currents that are conserved. Consequently one has a higher symmetry and an infinite sequence of generators $$Q^s=\int d\vec{x}J_{0\mu_1\mu_2\cdots \mu_s} \ .$$
In such a theory, the Coleman-Mandula theorem would imply that the $S$-matrix should be 1. The relevance and implications of the Coleman-Mandula theorem in AdS$_4$/CFT$_3$ was recently addressed by Maldacena and Zhiboedov [@Maldacena:2011jn; @Maldacena:2012sf]. They work in the light cone and make extensive use of only one particular charge $$Q^s=\int dx^-dxJ_{---\cdots -} \ .$$
In general there is a question regarding the existence of an $S$-matrix in CFT (and also AdS spacetime gravity). Maldacena and Zhiboedov considered the implication of the theorem on correlation functions; they demonstrated that the existence of the sequence of currents implies that the correlators are given by free fields, and it is in this sense that the theory can be categorized as simple, i.e. represented in terms of free fields.
The recovered correlators $C_n=\langle \mathcal{O}_1 \mathcal{O}_2 \cdots \mathcal{O}_n \rangle$ even though expressible in terms of free $N$-component fields, are nonzero for all $n$. They are, as we have described above, associated with a nonlinear bulk theory, with nonlinearities governed by $1/N=G_N$. The question then concerns the fate of these nonlinearities characterizing the AdS$_4$ HS theory.
Boundary correlators are sometimes described in the literature as a “boundary $S$-matrix” of the AdS theory. In fact Mack [@Mack:2009mi; @Mack:2009gy] has put forward arguments whereby CFT correlation functions themselves posses a structure equivalent to an $S$-matrix. He argued that they can be in general written in an integral form (the Mellin representation) $$G(x_1 x_2\cdots x_3)=\int \prod ds_{ij} M(\{s_{ij}\}) \prod_{i<j} \Gamma(s_{ij})(x_{ij})^{-2 s_{ij}}$$ which then implies various properties (crossing, duality, etc.) in support of their $S$-matrix interpretation. This interpretation was strengthened by the AdS calculation [@Penedones:2010ue]. Nevertheless, this “boundary $S$-matrix” lacks some of the key features of a genuine scattering matrix.
Based on the collective construction we would like to put forward (and evaluate) another more direct $S$-matrix which we will base on the physical picture of (collective) dipoles that underlie the CFT$_3$/Higher Spin Holography. In this picture we identify an appropriate on-shell relation and define the $S$-matrix through a Lehman-Symanzik-Zimmermann reduction formula. As such it will be given as a limit of general bi-local correlation functions.
An example
----------
Before proceeding with the details, we describe an analogous example that features a simple (free) theory duality: the old $d=1$ Matrix Model / $2d$ non-critical string theory correspondence [@Das:1990kaa]. One has the matrix theory $$L=\frac{1}{2}\text{Tr}\left(\dot{M}^2(t)+M^2(t)\right)$$ and a Hamiltonian corresponding to $N^2$ decoupled harmonic oscillators $$H=-\frac{1}{2}\sum_{\alpha=1}^{N^2}\left(\frac{\partial^2}{\partial M_\alpha^2}-M_\alpha^2\right) .$$ In this model one also had an infinite sequence of higher charges: $Q_s=\text{Tr}[(P^2-M^2)^s]$ and an infinite $\mathcal{W}_{\infty}$ symmetry. In the basic matrix theory representation, there is clearly no scattering and no visible $S$-matrix. A spacetime interpretation of the model (and an $S$-matrix) is found through the collective (Fermi-Droplet) representation $$M_{ij}(t)\rightarrow \phi(x,t)=\text{Tr} \, \delta(x-M_{ij}(t)) \ .$$ The large $N$ collective Hamiltonian derived in [@Das:1990kaa] is given by $$H_c=\int dx \left( \frac{1}{2} \partial_x \Pi(x) \phi(x) \partial_x \Pi(x)+\frac{\pi^2}{6}\phi^3-\frac{x^2}{2}\phi \right)$$ where $\phi(x)$ and $\Pi(x)$ obey the canonical commutation relations $[\phi(x),\Pi(y)]=i\delta(x-y)$. This collective Hamiltonian correctly reproduces all the correlators $\langle \mathcal{O}_{n_1} \mathcal{O}_{n_2} \cdots \mathcal{O}_{n_k} \rangle$ for the most general invariant operators $\mathcal{O}_n=\text{Tr}(M^n)=\int dx \; x^n \phi \ .$
Small fluctuations of this (collective) theory $\phi=\phi_0+\partial_x \psi$, $\Pi=-\partial_x^{-1} \dot{\psi}$ features a 2d massless boson [@Demeterfi:1991cw] $$H^{(2)}=\int d\sigma \left( \frac{1}{2}\dot{\psi}^2(t,\sigma)+\frac{\pi^2}{2}{\psi^\prime}^2(t,\sigma)\right) \ ,$$ where the prime is the derivative with respect to the Liouville coordinate defined by $$\sigma=\frac{1}{\pi} \int_0^x \frac{d y}{\phi_0(y)} \ .$$ Consequently one is led to consider the scattering of collective massless bosons [@Gross:1991qp] with an on-shell condition: $K_\mu=(E,K)$ and $E^2-K^2=0$. Evaluation of the corresponding scattering amplitudes gives the $S$-matrix. For the three-point scattering amplitude, one has $$\begin{aligned}
S_3(E_1,E_2,E_3) &=& 2\pi\delta(E_1+E_2+E_3) \left[ \prod_{i=1}^3 (E_i-K_i)- \prod_{i=1}^3 (E_i+K_i) \right] \cr
&=& 2\pi\delta(E_1+E_2+E_3) \left[ \prod_{i=1}^3 \bigl( E_i-\vert E_i\vert \bigr)- \prod_{i=1}^3 \bigl(E_i+\vert E_i\vert \bigr) \right]\end{aligned}$$ where we have used $K_i=\vert E_i\vert$ (corresponding to Liouville as time). For the scattering of incoming (outgoing) particles, we have $E_1,E_2>0$, $E_3<0$, so that $$\begin{aligned}
S_3(+,+,-)=0 \ .\end{aligned}$$ In the same way one can show $S_{n \ge 4}=0$. A change of boundary conditions (in particular Dirichlet), gives however a non-trivial result $S_n \neq 0$ which was then compared with the string scattering amplitudes.
Evaluation of the three- and four-point amplitudes
--------------------------------------------------
Let us now return to the bi-local theory and consider therefore the $S$-matrix for scattering of “collective dipoles”. In a time-like gauge (single-time), one has the on-shell relation: $E^2-(\vert \vec{k}_1\vert+\vert \vec{k}_2\vert)^2=0$, and the $S$-matrix can be defined by the LSZ-type reduction formula $$\begin{aligned}
S=\lim \prod_{i} (E_i^2-(\vert \vec{k}_i\vert+\vert \vec{k}_{i'}\vert )^2) \langle \tilde{\Psi}(E_1,\vec{k}_1,\vec{k}_{1'})\tilde{\Psi}(E_2,\vec{k}_2,\vec{k}_{2'})\cdots \rangle
\label{Sdefintion}\end{aligned}$$ where the $\tilde{\Psi}$ operators denote energy-momentum transforms of the bi-local fields . The limit implies the on-shell specification for the energies of the dipoles. In the light-cone gauge, (\[Sdefintion\]) would correspond to $$\begin{aligned}
\lim \prod_{i} (P_i^--\frac{p_i^2}{2p_i^+}-\frac{p^2_{i'}}{2p^+_{i'}}) \langle \tilde{\Psi}(P_1^-;p_1^+,p_1,p^+_{1'},p_{1'}) \tilde{\Psi} (P_2^-;p^+_2,p_2,p^+_{2'},p_{2'})\cdots \rangle \ .\end{aligned}$$ We note that the correlation functions appearing in this construction are not the correlation functions of conformal current operators $J_{--\cdots-}$. As Maldacena and Zhiboedov have discussed, the Ward identities based on currents provide a reconstruction of correlation functions for bi-local operators of the form $\mathcal{B}(x^+; (x_1^-,x_2^-); x_1=x_2)$. Since these are bi-local in $x$ but local in the other coordinates one is not in a position to consider the above defined $S$-matrix.
Our evaluation of the $S$-matrix proceeds as follows. Using the time-like quantization we will evaluate the 3 and 4-point scattering amplitude corresponding to associated Witten diagrams. In momentum space, in terms of the bi-local fields $$\begin{aligned}
\eta(t;\vec{x}_1,\vec{x}_2) &=& \int d\vec{k}_1 d\vec{k}_2\frac{1}{\sqrt{2\omega_{k_1}\omega_{k_2}}} \left(e^{+i(\vec{k}_1 \cdot \vec{x}_1+\vec{k}_2 \cdot \vec{x}_2)}\alpha_{\vec{k}_1\vec{k}_2}+h.c. \right) \\
\pi(t;\vec{x}_1,\vec{x}_2) &=& i \int d\vec{k}_1 d\vec{k}_2 \sqrt{\frac{\omega_{k_1}\omega_{k_2}}{2}} \left(e^{-i(\vec{k}_1 \cdot \vec{x}_1+\vec{k}_2 \cdot \vec{x}_2)}\alpha^\dagger_{\vec{k}_1\vec{k}_2}-h.c. \right)\end{aligned}$$ the cubic (\[cubic\]) and quartic (\[quartic\]) interaction potentials take the form $$\begin{aligned}
H^{(3)} &=& \frac{\sqrt{2}}{\sqrt{N}}\int \prod_{i=1}^3 d\vec{k}_i \Bigl[-\frac{\omega_{k_1 k_2 k_3}}{3}\alpha_{\vec{k}_1 \vec{k}_2}\alpha_{-\vec{k}_2 \vec{k}_3}\alpha_{-\vec{k}_3-\vec{k}_1}+\omega_{k_2} \alpha_{\vec{k}_1 \vec{k}_2}\alpha_{-\vec{k}_2 \vec{k}_3}\alpha^\dagger_{\vec{k}_3 \vec{k}_1}+h.c. \Bigr] \label{cubicosc} \\
H^{(4)} &=& \frac{1}{N} \int \prod_{i=1}^4 d\vec{k}_i \; \frac{\omega_{k_1 k_2 k_3 k_4}}{4} \Bigl[\alpha_{\vec{k}_1 \vec{k}_2}\alpha_{-\vec{k}_2 \vec{k}_3}\alpha_{-\vec{k}_3 \vec{k}_4}\alpha_{-\vec{k}_4-\vec{k}_1}+4\alpha_{\vec{k}_1 \vec{k}_2}\alpha_{-\vec{k}_2 \vec{k}_3}\alpha_{-\vec{k}_3 \vec{k}_4}\alpha^\dagger_{\vec{k}_4 \vec{k}_1}+h.c. \cr
&& \qquad\qquad\qquad\qquad+4\alpha_{\vec{k}_1 \vec{k}_2}\alpha_{-\vec{k}_2 \vec{k}_3}\alpha^\dagger_{\vec{k}_3 \vec{k}_4}\alpha^\dagger_{-\vec{k}_4 \vec{k}_1}+2\alpha_{\vec{k}_1 \vec{k}_2}\alpha^\dagger_{\vec{k}_2 \vec{k}_3}\alpha_{\vec{k}_3 \vec{k}_4}\alpha^\dagger_{\vec{k}_4 \vec{k}_1} \Bigr]
\label{quarticosc}\end{aligned}$$ where we used the notation $\omega_{k_1k_2\cdots k_i} \equiv \omega_{k_1}+\omega_{k_2}+\cdots+\omega_{k_i}$ and $h.c.$ means taking the hermitian conjugate of [*only*]{} the terms ahead of it.
For the three-dipole scattering ($1+2 \to 3$), the amplitude is given by $$\langle 0 \vert \alpha_{\vec{p}_{3} \vec{p}_{3'}} \,
T \exp \left[ -i \int_{-\infty}^\infty dt \, H^{(3)} (t) \right]
\alpha^\dagger_{\vec{p}_{2} \vec{p}_{2'}} \alpha^\dagger_{\vec{p}_{1} \vec{p}_{1'}} \vert 0 \rangle$$ where $T$ means [*time-ordered*]{}. Using the explicit form of the cubic interaction given in (\[cubicosc\]), the only surviving $1/\sqrt{N}$ contribution to the scattering amplitude is $$-\frac{i\sqrt{2}}{\sqrt{N}} \int d\vec{k}_i \; \omega_{k_2}
\langle 0 \vert \alpha_{\vec{p}_{3} \vec{p}_{3'}}
\alpha_{\vec{k}_1 \vec{k}_2} \alpha_{-\vec{k}_2 \vec{k}_3} \alpha^\dagger_{\vec{k}_3 \vec{k}_1}
\alpha^\dagger_{\vec{p}_{2} \vec{p}_{2'}} \alpha^\dagger_{\vec{p}_{1} \vec{p}_{1'}} \vert 0 \rangle \ .
\label{vev3}$$ Graphically, this corresponds to evaluate the Feynman diagram shown in Figure \[threescattering\].
![Scattering of three dipoles.[]{data-label="threescattering"}](three.eps){width="30.00000%"}
The evaluation can be most easily performed in the [*interaction picture*]{}. Using the bi-local propagator symmetrized over the momenta $$\begin{aligned}
\langle 0\vert T \alpha_{\vec{p}_{1} \vec{p}_{1'}}(t_1)
\alpha^\dagger_{\vec{p}_{2} \vec{p}_{2'}}(t_2)\vert 0 \rangle
=\int dE\frac{ie^{-iE(t_1-t_2)}}{E-\omega_{p_1}-\omega_{p_{1'}}}
& \frac{1}{2}[\delta(\vec{p}_1-\vec{p}_2)\delta(\vec{p}_{1'}-\vec{p}_{2'}) \cr
& +\delta(\vec{p}_1-\vec{p}_{2'})\delta(\vec{p}_{1'}-\vec{p}_2)] \ ,\end{aligned}$$ the integral over the vev in (\[vev3\]) is calculated to be $$\begin{aligned}
& \frac{i^3}{8} \delta(E_1+E_2-E_3) [(\omega_{p_{1'}}+\omega_{p_2}) \delta(\vec{p}_1-\vec{p}_3) \delta( \vec{p}_{2'}-\vec{p}_{3'}) \delta (\vec{p}_{1'}+\vec{p}_2) + \text{7 more terms} ] \cr
=& \frac{i^3}{8} (E_1+E_2-E_3) \delta(E_1+E_2-E_3) [\delta(\vec{p}_1-\vec{p}_3) \delta( \vec{p}_{2'}-\vec{p}_{3'}) \delta (\vec{p}_{1'}+\vec{p}_2) + \cdots ]\end{aligned}$$ where in the last step we have used energy conservation and the delta functions. The seven more terms are due to the symmetrization of $(1 \leftrightarrow 1'),(2 \leftrightarrow 2'),(3 \leftrightarrow 3')$. Combining all the pre-factors, we get the final result $$\begin{aligned}
S(1+2\rightarrow 3)=&-\frac{\sqrt{2}}{8\sqrt{N}} (E_1+E_2-E_3) \; \delta(E_1+E_2-E_3) \cr
&\{\delta(\vec{p}_1-\vec{p}_3)\delta(\vec{p}_{2'}-\vec{p}_{3'})\delta(\vec{p}_{1'}+\vec{p}_2)+\text{7 more terms} \}\end{aligned}$$ so that the result $S_3=0$ follows.
Next for the four-dipole scattering ($1+2 \to 3+4$), the calculation is similar. The scattering amplitude is given by $$\langle 0\vert \alpha_{\vec{p}_{3}\vec{p}_{3'}}\alpha_{\vec{p}_{4}\vec{p}_{4'}} T \exp \left[
-i\int^\infty_{-\infty} dt \, \left( H^{(3)} (t) + H^{(4)} (t) \right) \right]
\alpha^\dagger_{\vec{p}_{1} \vec{p}_{1'}}\alpha^\dagger_{\vec{p}_{2} \vec{p}_{2'}}\vert 0 \rangle$$ where $H^{(4)}$ is explicitly given in (\[quarticosc\]). The $1/N$ contributions to the $S_4$ scattering amplitude are collected as follows $$\begin{aligned}
&&-\frac{2}{9N}\int d\vec{k}_i d\vec{l}_j \, \omega_{k_1k_2k_3}\omega_{l_1l_2l_3}\langle 0 \vert \alpha_{\vec{p}_{3} \vec{p}_{3'}}\alpha_{\vec{p}_{4} \vec{p}_{4'}}\alpha_{\vec{k}_1 \vec{k}_2}\alpha_{-\vec{k}_2 \vec{k}_3}\alpha_{-\vec{k}_3-\vec{k}_1}\alpha^\dagger_{\vec{l}_1 \vec{l}_2}\alpha^\dagger_{-\vec{l}_2 \vec{l}_3}\alpha^\dagger_{-\vec{l}_3-\vec{l}_1}\alpha^\dagger_{\vec{p}_{1} \vec{p}_{1'}}\alpha^\dagger_{\vec{p}_{2} \vec{p}_{2'}}\vert 0 \rangle \cr
&&-\frac{2}{N}\int d\vec{k}_i d\vec{l}_j \, \omega_{k_2}\omega_{l_2}\langle 0 \vert \alpha_{\vec{p}_{3} \vec{p}_{3'}}\alpha_{\vec{p}_{4} \vec{p}_{4'}}\alpha_{\vec{k}_1 \vec{k}_2}\alpha_{-\vec{k}_2 \vec{k}_3}\alpha^\dagger_{\vec{k}_3 \vec{k}_1}\alpha^\dagger_{\vec{l}_1 \vec{l}_2}\alpha^\dagger_{-\vec{l}_2 \vec{l}_3}\alpha_{\vec{l}_3 \vec{l}_1}\alpha^\dagger_{\vec{p}_{1} \vec{p}_{1'}}\alpha^\dagger_{\vec{p}_{2} \vec{p}_{2'}}\vert 0 \rangle \cr
&&-\frac{i}{N}\int d\vec{k}_i \, \omega_{k_1k_2k_3k_4}\langle 0\vert \alpha_{\vec{p}_{3} \vec{p}_{3'}}\alpha_{\vec{p}_{4} \vec{p}_{4'}}\alpha_{\vec{k}_1 \vec{k}_2}\alpha_{-\vec{k}_2 \vec{k}_3}\alpha^\dagger_{\vec{k}_3 \vec{k}_4}\alpha^\dagger_{-\vec{k}_4 \vec{k}_1}\alpha^\dagger_{\vec{p}_{1} \vec{p}_{1'}}\alpha^\dagger_{\vec{p}_{2} \vec{p}_{2'}}\vert 0 \rangle \cr
&&-\frac{i}{2N}\int d\vec{k}_i \, \omega_{k_1k_2k_3k_4}\langle 0 \vert \alpha_{\vec{p}_{3} \vec{p}_{3'}}\alpha_{\vec{p}_{4} \vec{p}_{4'}}\alpha_{\vec{k}_1 \vec{k}_2}\alpha^\dagger_{\vec{k}_2 \vec{k}_3}\alpha_{\vec{k}_3 \vec{k}_4}\alpha^\dagger_{\vec{k}_4 \vec{k}_1}\alpha^\dagger_{\vec{p}_{1} \vec{p}_{1'}}\alpha^\dagger_{\vec{p}_{2} \vec{p}_{2'}}\vert 0 \rangle \ .
\label{fourlines}\end{aligned}$$ The first line of (\[fourlines\]) has only $s$-channel contributions shown in Figure \[fourchanneelsca\], while the second line of (\[fourlines\]) has all $s,t,u$-channel contributions. The $s$-channel diagrams and their twisted ones (due to the symmetrization of propagators) are summed to be $$\begin{aligned}
\frac{i}{8 N} \delta(E_1+E_2-E_3-E_4) \bigl[ & (\omega_{p_{2'}}+\omega_{p_3}) \delta(\vec{p}_1-\vec{p}_3) \delta(\vec{p}_{1'}+\vec{p}_2) \cr
& \delta( \vec{p}_{2'}-\vec{p}_{4'}) \delta (\vec{p}_{3'}+\vec{p}_4) + \text{15 more terms} \cr
+ & (\omega_{p_{1'}}+\omega_{p_{3}}) \delta(\vec{p}_2-\vec{p}_3) \delta(\vec{p}_1+\vec{p}_{2'}) \cr
& \delta( \vec{p}_{1'}-\vec{p}_{4'}) \delta (\vec{p}_{3'}+\vec{p}_4) + \text{15 more terms} \bigr] \ .\end{aligned}$$
![The $s$-channel diagram of four-dipole scattering.[]{data-label="fourchanneelsca"}](fourchannel.eps){width="35.00000%"}
It is also convenient to calculate the $t,u$-channel diagrams together, with their twisted diagrams, they are summed to be $$\begin{aligned}
\frac{i}{8 N} \delta(E_1+E_2-E_3-E_4) \bigl[ & (\omega_{p_{1'}}+\omega_{p_2}) \delta(\vec{p}_1-\vec{p}_3) \delta(\vec{p}_{1'}+\vec{p}_2) \cr
& \delta( \vec{p}_{2'}-\vec{p}_{4'}) \delta (\vec{p}_{3'}+\vec{p}_4) + \text{15 more terms} \cr
+ & (\omega_{p_1}+\omega_{p_{2'}}) \delta(\vec{p}_2-\vec{p}_3) \delta(\vec{p}_1+\vec{p}_{2'}) \cr
& \delta( \vec{p}_{1'}-\vec{p}_{4'}) \delta (\vec{p}_{3'}+\vec{p}_4) + \text{15 more terms} \bigr] \\
+ \frac{i}{16 N} \delta(E_1+E_2-E_3-E_4) \bigl[ & (\omega_{p_{1}}+\omega_{p_{1'}}+\omega_{p_2}+\omega_{p_{2'}}) \delta(\vec{p}_1-\vec{p}_3) \delta(\vec{p}_{1'}-\vec{p}_4) \cr
& \delta( \vec{p}_{2'}-\vec{p}_{4'}) \delta (\vec{p}_2-\vec{p}_{3'}) + \text{15 more terms} \cr
+ & (\omega_{p_{1}}+\omega_{p_{1'}}+\omega_{p_2}+\omega_{p_{2'}}) \delta(\vec{p}_2-\vec{p}_3) \delta(\vec{p}_{2'}-\vec{p}_4) \cr
& \delta( \vec{p}_{1'}-\vec{p}_{4'}) \delta (\vec{p}_1-\vec{p}_{3'}) + \text{15 more terms} \bigr] \ .
\label{cancel1}\end{aligned}$$
The third line of (\[fourlines\]) is the cross-shaped diagram shown in Figure \[fourcrosssca\], which gives the result $$\begin{aligned}
-\frac{i}{8 N} \delta(E_1+E_2-E_3-E_4) \bigl[ & (\omega_{p_{1'}}+\omega_{p_{2'}}+\omega_{p_3}+\omega_{p_4}) \delta(\vec{p}_1-\vec{p}_3) \delta(\vec{p}_{1'}+\vec{p}_2) \cr
& \delta( \vec{p}_{2'}-\vec{p}_{4'}) \delta (\vec{p}_{3'}+\vec{p}_4) + \text{15 more terms} \cr
+ & (\omega_{p_{1}}+\omega_{p_{2}}+\omega_{p_{3'}}+\omega_{p_{4'}}) \delta(\vec{p}_2-\vec{p}_3) \delta(\vec{p}_1+\vec{p}_{2'}) \cr
& \delta( \vec{p}_{1'}-\vec{p}_{4'}) \delta (\vec{p}_{3'}+\vec{p}_4) + \text{15 more terms} \bigr] \ .\end{aligned}$$
![The cross-shaped diagram of four-dipole scattering.[]{data-label="fourcrosssca"}](fourcross.eps){width="30.00000%"}
The calculation of the fourth line is similar to the third line, which gives the result $$\begin{aligned}
-\frac{i}{16 N} \delta(E_1+E_2-E_3-E_4) \bigl[ & (\omega_{p_{1}}+\omega_{p_{1'}}+\omega_{p_2}+\omega_{p_{2'}}) \delta(\vec{p}_1-\vec{p}_3) \delta(\vec{p}_{1'}-\vec{p}_4) \cr
& \delta( \vec{p}_{2'}-\vec{p}_{4'}) \delta (\vec{p}_2-\vec{p}_{3'}) + \text{15 more terms} \cr
+ & (\omega_{p_{1}}+\omega_{p_{1'}}+\omega_{p_2}+\omega_{p_{2'}}) \delta(\vec{p}_2-\vec{p}_3) \delta(\vec{p}_{2'}-\vec{p}_4) \cr
& \delta( \vec{p}_{1'}-\vec{p}_{4'}) \delta (\vec{p}_1-\vec{p}_{3'}) + \text{15 more terms} \bigr] \ .
\label{cancel2}\end{aligned}$$
Summing all the diagrams, it is easy to see (\[cancel1\]) and (\[cancel2\]) cancel each other, while the rest diagrams give the final result $$\begin{aligned}
S(1+ & 2\rightarrow 3+4)= \frac{i}{8 N} \delta(E_1+E_2-E_3-E_4) \cr
& \bigl[ (\omega_{p_2}-\omega_{p_4}) \delta(\vec{p}_1-\vec{p}_3) \delta(\vec{p}_{1'}+\vec{p}_2) \delta( \vec{p}_{2'}-\vec{p}_{4'}) \delta (\vec{p}_{3'}+\vec{p}_4) + \text{15 more terms} \cr
+ & (\omega_{p_{2'}}-\omega_{p_{3'}}) \delta(\vec{p}_2-\vec{p}_3) \delta(\vec{p}_1+\vec{p}_{2'}) \delta( \vec{p}_{1'}-\vec{p}_{4'}) \delta (\vec{p}_{3'}+\vec{p}_4) + \text{15 more terms} \bigr] \cr
=& \frac{i}{16 N} (E_1+E_2-E_3-E_4) \delta(E_1+E_2-E_3-E_4) \cr
& \bigl[ \delta(\vec{p}_1-\vec{p}_3) \delta(\vec{p}_{1'}+\vec{p}_2) \delta( \vec{p}_{2'}-\vec{p}_{4'}) \delta (\vec{p}_{3'}+\vec{p}_4) + \text{15 more terms} \cr
+ & \; \delta(\vec{p}_2-\vec{p}_3) \delta(\vec{p}_1+\vec{p}_{2'}) \delta( \vec{p}_{1'}-\vec{p}_{4'}) \delta (\vec{p}_{3'}+\vec{p}_4) + \text{15 more terms} \bigr] \ ,\end{aligned}$$ which implies $S_4=0$.
It is clear that the direct evaluation can be continued to higher point scattering with the conjectured result $S_{n \ge 5}=0$. One can describe the nonlinear collective field theory in the following way: its nonlinearity, and higher point vertices are precisely such that they reproduce the boundary correlators through bi-local (Witten) diagrams. These same diagrams however give vanishing results in the on-shell evaluation as described above.
In general quantum field theory, one has the equivalence theorem. Consequently a vanishing $S$-matrix implies that there should exist a (nonlinear) field transformation which linearizes the theory. For the present case this concerns the linearization of bulk $G_N=1/N$ interactions. We will in the next section describe such a field transformation.
Since we view the collective construction to represent a gauge fixed description of Vasiliev’s HS theory, analogous statements are expected to hold there. Finally it is also clear that one can expect that any change of boundary conditions will result in non-trivial $S$-matrix.
Field Transformation {#sec:Field}
====================
We have concluded in the previous section that the $S$-matrix equals $1$ for the bi-local theory of the free UV fixed point. The theory is nonlinear with a sequence of $1/N$ vertices which are needed to reproduce arbitrary $n$-point correlators (and the “boundary $S$-matrix”). By correspondence Vasiliev’s HS theory has the same properties. As suggested in section \[sec:CM\], this implies that there should be a field transformation that linearizes the $G_N=1/N$ interactions. We will now describe such a procedure for deducing the transformation. The procedure is based on considering an algebraic description of the bi-local system. We will be able to show that the bi-local pseudo-spin algebra has among other two representations: one equalling the nonlinear collective field theory and another in which the Hamiltonian becomes quadratic.
For the free theory in question one has exact creation operators for the singlet sector of the theory. They are given by the bi-local operators $$\begin{aligned}
A(\vec{p}_1,\vec{p}_2)&=&\frac{1}{\sqrt{2N}}\sum_i a^i(\vec{p}_1)a^i(\vec{p}_2) \ , \\
A^\dagger(\vec{p}_1,\vec{p}_2)&=&\frac{1}{\sqrt{2N}}\sum_i a^{i\, \dagger}(\vec{p}_1)a^{i\, \dagger}(\vec{p}_2) \ , \\
B(\vec{p}_1,\vec{p}_2)&=&\frac{1}{2}\sum_i a^{i\,\dagger}(\vec{p}_1)a^i(\vec{p}_2) \ .\end{aligned}$$ In terms of these collective variables the Hamiltonian is $$H=\int d^{d-1}\vec{p} \; {\cal H}(\vec{p},\vec{p}) \ , \qquad
{\cal H}(\vec{p},\vec{p}) = 2\omega_{\vec{p}} \, B(\vec{p},\vec{p})
+ \frac{N}{2}\omega_{\vec{p}} \, \delta(\vec{0}) \ .
\label{Hamiltonian}$$ The above operators (representing bi-local pseudo-spin variables) close an algebra $$\begin{aligned}
\big[ A(\vec{p}_1,\vec{p}_2),A^\dagger(\vec{p}_3,\vec{p}_4)\big]=
& \, \frac{1}{2}\left(\delta_{\vec{p}_2,\vec{p}_3}\delta_{\vec{p}_4,\vec{p}_1}+\delta_{\vec{p}_2,\vec{p}_4}\delta_{\vec{p}_3,\vec{p}_1}\right)
+\frac{1}{N}\bigl[\delta_{\vec{p}_2,\vec{p}_3}B(\vec{p}_4,\vec{p}_1) \cr
&+\delta_{\vec{p}_2,\vec{p}_4}B(\vec{p}_3,\vec{p}_1)
+\delta_{\vec{p}_1,\vec{p}_3}B(\vec{p}_4,\vec{p}_2)
+\delta_{\vec{p}_1,\vec{p}_4}B(\vec{p}_3,\vec{p}_2) \bigr] \ , \\
\big[ B(\vec{p}_1,\vec{p}_2),A^\dagger(\vec{p}_3,\vec{p}_4)\big]=
& \, \frac{1}{2}\bigl( \delta_{\vec{p}_2,\vec{p}_3}A^\dagger(\vec{p}_1,\vec{p}_4)+
\delta_{\vec{p}_2,\vec{p}_4}A^\dagger(\vec{p}_1,\vec{p}_3)\bigr) \ , \\
\big[ B(\vec{p}_1,\vec{p}_2),A(\vec{p}_3,\vec{p}_4)\big]=
&-\frac{1}{2}\bigl(\delta_{\vec{p}_1,\vec{p}_3}A(\vec{p}_2,\vec{p}_4)+
\delta_{\vec{p}_1,\vec{p}_4}A(\vec{p}_2,\vec{p}_3)\bigr) \ .\end{aligned}$$
We note that the theory based on this algebra was studied in detail by Berezin [@Berezin:1978sn]. In the $O(N)$ case one finds the quadratic (Casimir) constraint $$-\frac{8}{N}A^\dagger\star A +\left(1+\frac{4}{N}B\right)\star \left(1+\frac{4}{N}B\right) = \text{$\mathbb{I}$} \ .$$ The importance of the Casimir constraint is that it implies that the above non-commuting set of bi-local operators is not independent. In particular the bi-local pseudo-spin algebra has representations in terms of canonical pairs of variables.
The canonical collective theory based on the equal-time bi-local field and its conjugate provides one specific representation of the above algebra. Explicitly, one can show $$\begin{aligned}
A(\vec{x}_1,\vec{x_2})&=&\int d\vec{p}_1d\vec{p}_2d\vec{y}_1d\vec{y}_2e^{i\vec{p}_1\cdot(\vec{x}_1-\vec{y}_1)}e^{i\vec{p}_2\cdot(\vec{x}_2-\vec{y}_2)}
\Bigl[\frac{-2}{\sqrt{\omega_{p_1} \omega_{p_2}}}\Pi(\vec{y}_1,\vec{z}_1)\star\Psi(\vec{z}_1,\vec{z}_2)\star\Pi(\vec{z}_2,\vec{y}_2)\cr
&&-i\sqrt{N}\sqrt{\frac{\omega_{p_2}}{\omega_{p_1}}}\Psi(\vec{y}_2,\vec{z}_1)\star\Pi(\vec{y}_1,\vec{z}_1)
-i\sqrt{N}\sqrt{\frac{\omega_{p_1}}{\omega_{p_2}}}\Psi(\vec{y}_1,\vec{z}_1)\star\Pi(\vec{y}_2,\vec{z}_1)\cr
&&-\frac{N}{8}\frac{1}{\sqrt{\omega_{p_1} \omega_{p_2}}}\frac{1}{\Psi}(\vec{y}_1,\vec{y}_2)
+\frac{N\sqrt{\omega_{p_1} \omega_{p_2}}}{2}\Psi(\vec{y}_1,\vec{y}_2) \Bigr] \ .\end{aligned}$$ Transforming it to momentum space and expanding in $1/N$ we generate an infinite series $$\begin{aligned}
A(\vec{k}_1,\vec{k}_2)=& \, \alpha_{\vec{k}_1\vec{k}_2}-\frac{1}{\sqrt{2N}}\Bigl[
\alpha_{\vec{k}_1\vec{k}_3}\star\alpha_{-\vec{k}_3\vec{k}_2}
-\alpha_{\vec{k}_1\vec{k}_3}^\dagger\star\alpha_{-\vec{k}_3\vec{k}_2}^\dagger \cr
& -\alpha_{\vec{k}_1\vec{k}_3}\star\alpha_{\vec{k}_3\vec{k}_2}^\dagger
-\alpha_{\vec{k}_1\vec{k}_3}\star\alpha_{\vec{k}_3\vec{k}_2}^\dagger\Bigr]+O(\alpha^3) \ , \\
B(\vec{k}_1,\vec{k}_2)=& \, \frac{1}{2}\left[
\alpha_{\vec{k}_1\vec{k}_3}\star\alpha_{\vec{k}_3\vec{k}_2}^\dagger
+\alpha_{\vec{k}_1\vec{k}_3}^\dagger\star\alpha_{\vec{k}_3\vec{k}_2}\right]
+\sqrt{\frac{2}{N}}\Bigl[\alpha_{\vec{k}_1\vec{k}_3}\star\alpha_{-\vec{k}_3\vec{k}_4}
\star\alpha_{-\vec{k}_4\vec{k}_2} \cr
&+\alpha_{\vec{k}_1\vec{k}_3} \star\alpha_{\vec{k}_3\vec{k}_4}^\dagger\star\alpha_{\vec{k}_4\vec{k}_2}
-\alpha_{\vec{k}_1\vec{k}_3}^\dagger\star\alpha_{\vec{k}_3\vec{k}_4}\star\alpha_{\vec{k}_4\vec{k}_2}^\dagger \cr
&-\alpha_{\vec{k}_1\vec{k}_3}^\dagger\star\alpha_{-\vec{k}_3\vec{k}_4}^\dagger
\star\alpha_{-\vec{k}_4\vec{k}_2}^\dagger \Bigr]+O(\alpha^4) \ .\end{aligned}$$ The key to our arguments is the fact that one can write another realization of the algebra in terms of an oscillator $\beta (\vec{p}_1,\vec{p}_2)$ obeying $$\begin{aligned}
\beta(\vec{p}_1,\vec{p}_2) &=& \left(1+\frac{2}{N} B\right)^{-\frac{1}{2}}(\vec{p}_1,\vec{p})
\star A(\vec{p},\vec{p}_2) \label{invbetabilocalrep} \\
\beta^\dagger(\vec{p}_1,\vec{p}_2) &=& A^\dagger(\vec{p}_1,\vec{p})
\star \left(1+\frac{2}{N}B \right)^{-\frac{1}{2}}(\vec{p},\vec{p}_2)\end{aligned}$$ which has two important properties that $$\begin{aligned}
B(\vec{p}_1,\vec{p}_2) &=& \beta^\dagger(\vec{p}_1,\vec{p})\star \beta(\vec{p},\vec{p}_2) \\
\left[\beta (\vec{p}_1,\vec{p}_2),\beta^\dagger (\vec{p}_3,\vec{p}_4)\right] &=& \delta_{\vec{p}_1,\vec{p}_4}\delta_{\vec{p}_2,\vec{p}_3} \ .\end{aligned}$$ We see that in this realization the Hamiltonian is quadratic due to . Furthermore, using one can generate the transformation between the fields $$\begin{aligned}
\beta(\vec{k}_1,\vec{k}_2)=& \, \alpha_{\vec{k}_1\vec{k}_2}-\frac{1}{\sqrt{2N}}\Bigl[
\alpha_{\vec{k}_1\vec{k}_3}\star\alpha_{-\vec{k}_3\vec{k}_2}
-\alpha_{\vec{k}_1\vec{k}_3}^\dagger\star\alpha_{-\vec{k}_3\vec{k}_2}^\dagger \cr
&-\alpha_{\vec{k}_1\vec{k}_3}\star\alpha_{\vec{k}_3\vec{k}_2}^\dagger
-\alpha_{\vec{k}_1\vec{k}_3}\star\alpha_{\vec{k}_3\vec{k}_2}^\dagger \Bigr]+O(\alpha^3) \ .\end{aligned}$$
In conclusion we have presented a construction of the field transformation (in bi-local space) that linearizes the nonlinear $1/N$ Hamiltonian. Under this transformation the correlation functions change but the $S$-matrix does not. This represents the working of the Coleman-Mandula theorem in the large $N$ dual associated with the free field CFT. As such it complements the Maldacena-Zhiboedov argument for these theories.
Conclusion {#sec:Con}
==========
We have discussed some features of the Higher Spin AdS correspondence involving free $O(N)$ fields. The existence of an (infinite) sequence of higher symmetries in these theories raises the question regarding the implementation of the Coleman-Mandula theorem. Our focus was the question regarding the nonlinear $1/N$ theory which reproduces the (boundary) correlators. We argued that in these theories we are able to define a genuine $S$-matrix representing the scattering of collective dipoles. The $S$-matrix is specified with the standard LSZ procedure as an on-shell limit of (bi-local) correlation functions.
For the theory based on the free correspondence i.e. the UV fixed point of the vector model we have evaluated the $S$-matrix showing the result $S=1$ as claimed in the title. This represents the consequence of the Coleman-Mandula theorem for the associated Higher Spin theory and complements the results of Maldacena and Zhiboedov. As we have discussed it implies that the nonlinear Higher Spin theory can be linearized through nonlinear field transformations. We have explicitly constructed such a transformation in the bi-local framework. We have also emphasized that a change of boundary conditions will change the above conclusion, namely one expects a nontrivial $S$-matrix. Based on the present results and the earlier $c=1$ case it is plausible to conclude that these features will characterize any large $N$ correspondence based on free fields.
Acknowledgments {#acknowledgments .unnumbered}
===============
We would like to thank Sumit Das, Igor Klebanov, Soo-Jong Rey, Juan Maldacena and Suvrat Raju for relevant and constructive comments. The work of AJ and QY is supported by the Department of Energy under contract DE-FG-02-91ER40688. RdMK is supported by the South African Research Chairs Initiative of the Department of Science and Technology and National Research Foundation. The work of KJ is supported in part by the Swiss National Science Foundation.
Some of the results were presented by AJ at “CQUeST Spring Workshop on Higher Spins and String Geometry, Seoul” and “ESI Workshop on Higher Spin Gravity, Vienna”. He would like to thank the organizers for their hospitality. KJ would also like to thank the Erwin Schrödinger Institute, Vienna.
[99]{}
I. R. Klebanov and A. M. Polyakov, “AdS dual of the critical O(N) vector model,” Phys. Lett. B [**550**]{}, 213 (2002) \[hep-th/0210114\]. E. Sezgin and P. Sundell, “Massless higher spins and holography,” Nucl. Phys. B [**644**]{}, 303 (2002) \[Erratum-ibid. B [**660**]{}, 403 (2003)\] \[hep-th/0205131\]. S. R. Das and A. Jevicki, “Large-N collective fields and holography,” Phys. Rev. D [**68**]{}, 044011 (2003) \[arXiv:hep-th/0304093\]. S. Giombi and X. Yin, “Higher Spin Gauge Theory and Holography: The Three-Point Functions,” JHEP [**1009**]{}, 115 (2010) \[arXiv:0912.3462 \[hep-th\]\]. S. Giombi and X. Yin, “Higher Spins in AdS and Twistorial Holography,” JHEP [**1104**]{}, 086 (2011) \[arXiv:1004.3736 \[hep-th\]\]. R. d. M. Koch, A. Jevicki, K. Jin and J. P. Rodrigues, “AdS$_4$/CFT$_3$ Construction from Collective Fields,” Phys. Rev. D [**83**]{}, 025006 (2011) \[arXiv:1008.0633 \[hep-th\]\]. M. R. Douglas, L. Mazzucato and S. S. Razamat, “Holographic dual of free field theory,” Phys. Rev. D [**83**]{}, 071701 (2011) \[arXiv:1011.4926 \[hep-th\]\]. A. Jevicki, K. Jin and Q. Ye, “Collective Dipole Model of AdS/CFT and Higher Spin Gravity,” J. Phys. A [**44**]{}, 465402 (2011) \[arXiv:1106.3983 \[hep-th\]\]. S. H. Shenker and X. Yin, “Vector Models in the Singlet Sector at Finite Temperature,” arXiv:1109.3519 \[hep-th\]. J. Maldacena and A. Zhiboedov, “Constraining Conformal Field Theories with A Higher Spin Symmetry,” arXiv:1112.1016 \[hep-th\]. X. Bekaert, E. Joung and J. Mourad, “Comments on higher-spin holography,” arXiv:1202.0543 \[hep-th\]. M. A. Vasiliev, “Holography, Unfolding and Higher-Spin Theory,” arXiv:1203.5554 \[hep-th\]. J. Maldacena and A. Zhiboedov, “Constraining conformal field theories with a slightly broken higher spin symmetry,” arXiv:1204.3882 \[hep-th\].
M. Henneaux and S. J. Rey, “Nonlinear $W_{\infty}$ as Asymptotic Symmetry of Three-Dimensional Higher Spin Anti-de Sitter Gravity,” JHEP [**1012**]{}, 007 (2010) \[arXiv:1008.4579 \[hep-th\]\]. A. Campoleoni, S. Fredenhagen, S. Pfenninger and S. Theisen, “Asymptotic symmetries of three-dimensional gravity coupled to higher-spin fields,” JHEP [**1011**]{}, 007 (2010) \[arXiv:1008.4744 \[hep-th\]\]. M. R. Gaberdiel and R. Gopakumar, “An AdS$_3$ Dual for Minimal Model CFTs,” Phys. Rev. D [**83**]{}, 066007 (2011) \[arXiv:1011.2986 \[hep-th\]\]. M. R. Gaberdiel, R. Gopakumar, T. Hartman and S. Raju, “Partition Functions of Holographic Minimal Models,” JHEP [**1108**]{}, 077 (2011) \[arXiv:1106.1897 \[hep-th\]\]. C. M. Chang and X. Yin, “Higher Spin Gravity with Matter in AdS$_3$ and Its CFT Dual,” arXiv:1106.2580 \[hep-th\]. M. Ammon, P. Kraus and E. Perlmutter, “Scalar fields and three-point functions in D=3 higher spin gravity,” arXiv:1111.3926 \[hep-th\].
M. Gary, D. Grumiller and R. Rashkov, “Towards non-AdS holography in 3-dimensional higher spin gravity,” JHEP [**1203**]{}, 022 (2012) \[arXiv:1201.0013 \[hep-th\]\]. M. R. Gaberdiel and R. Gopakumar, “Triality in Minimal Model Holography,” arXiv:1205.2472 \[hep-th\].
D. Anninos, T. Hartman and A. Strominger, “Higher Spin Realization of the dS/CFT Correspondence,” arXiv:1108.5735 \[hep-th\]. G. S. Ng and A. Strominger, “State/Operator Correspondence in Higher-Spin dS/CFT,” arXiv:1204.1057 \[hep-th\].
M. A. Vasiliev, “Higher-spin gauge theories in four, three and two dimensions,” Int. J. Mod. Phys. D [**5**]{}, 763 (1996) \[arXiv:hep-th/9611024\]. M. A. Vasiliev, “Nonlinear equations for symmetric massless higher spin fields in (A)dS(d),” Phys. Lett. B [**567**]{}, 139 (2003) \[hep-th/0304049\]. X. Bekaert, S. Cnockaert, C. Iazeolla and M. A. Vasiliev, “Nonlinear higher spin theories in various dimensions,” hep-th/0503128. R. R. Metsaev, “Cubic interaction vertices of massive and massless higher spin fields,” Nucl. Phys. B [**759**]{}, 147 (2006) \[hep-th/0512342\]. R. R. Metsaev, “Cubic interaction vertices for fermionic and bosonic arbitrary spin fields,” Nucl. Phys. B [**859**]{}, 13 (2012) \[arXiv:0712.3526 \[hep-th\]\]. D. Polyakov, “Higher Spins and Open Strings: Quartic Interactions,” Phys. Rev. D [**83**]{}, 046005 (2011) \[arXiv:1011.0353 \[hep-th\]\]. C. Iazeolla and P. Sundell, “Families of exact solutions to Vasiliev’s 4D equations with spherical, cylindrical and biaxial symmetry,” JHEP [**1112**]{}, 084 (2011) \[arXiv:1107.1217 \[hep-th\]\].
G. Mack, “D-independent representation of Conformal Field Theories in D dimensions via transformation to auxiliary Dual Resonance Models. Scalar amplitudes,” arXiv:0907.2407 \[hep-th\]. G. Mack, “D-dimensional Conformal Field Theories with anomalous dimensions as Dual Resonance Models,” arXiv:0909.1024 \[hep-th\]. J. Penedones, “Writing CFT correlation functions as AdS scattering amplitudes,” JHEP [**1103**]{}, 025 (2011) \[arXiv:1011.1485 \[hep-th\]\]. A. L. Fitzpatrick, J. Kaplan, J. Penedones, S. Raju and B. C. van Rees, “A Natural Language for AdS/CFT Correlators,” JHEP [**1111**]{}, 095 (2011) \[arXiv:1107.1499 \[hep-th\]\]. M. F. Paulos, “Towards Feynman rules for Mellin amplitudes,” JHEP [**1110**]{}, 074 (2011) \[arXiv:1107.1504 \[hep-th\]\]. D. Nandan, A. Volovich and C. Wen, “On Feynman Rules for Mellin Amplitudes in AdS/CFT,” arXiv:1112.0305 \[hep-th\].
D. J. Gross and I. R. Klebanov, “S = 1 for c = 1,” Nucl. Phys. B [**359**]{}, 3 (1991). R. Gopakumar and D. J. Gross, “Mastering the master field,” Nucl. Phys. B [**451**]{}, 379 (1995) \[hep-th/9411021\]. M. R. Douglas, “Stochastic master fields,” Phys. Lett. B [**344**]{}, 117 (1995) \[hep-th/9411025\].
R. de Mello Koch and J. P. Rodrigues, “Systematic 1/N corrections for bosonic and fermionic vector models without auxiliary fields,” Phys. Rev. D [**54**]{}, 7794 (1996) \[arXiv:hep-th/9605079\].
R. R. Metsaev, “Light cone form of field dynamics in anti-de Sitter spacetime and AdS/CFT correspondence,” Nucl. Phys. B [**563**]{}, 295 (1999) \[arXiv:hep-th/9906217\].
Y. M. Makeenko, “Conformal Operators In Quantum Chromodynamics,” Sov. J. Nucl. Phys. [**33**]{}, 440 (1981) \[Yad. Fiz. [**33**]{}, 842 (1981)\]. V. M. Braun, G. P. Korchemsky and D. Mueller, “The Uses of conformal symmetry in QCD,” Prog. Part. Nucl. Phys. [**51**]{}, 311 (2003) \[hep-ph/0306057\].
S. R. Das and A. Jevicki, “String Field Theory and Physical Interpretation of D = 1 Strings,” Mod. Phys. Lett. A [**5**]{}, 1639 (1990). K. Demeterfi, A. Jevicki and J. P. Rodrigues, “Perturbative results of collective string field theory,” Mod. Phys. Lett. A [**6**]{}, 3199 (1991).
F. A. Berezin, “Models of Gross-Neveu Type as Quantization of Classical Mechanics with Nonlinear Phase Space,” Commun. Math. Phys. [**63**]{}, 131 (1978).
[^1]: We mention that this is analogous to an earlier situation involving the $c=1$ matrix model with 2d string correspondence where one had the statement “$S=1$ for $c=1$" demonstrated in [@Gross:1991qp]. The only difference is that the collective boson (representing fluctuations above the fermion surface) is now replaced by the collective dipole.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The sensitivity of inertial confinement fusion implosions of the type performed on the National Ignition Facility (NIF) [@lindl:339] to low-mode flux asymmetries has been investigated numerically. It is shown that large-amplitude, low-order mode shapes (Legendre polynomial $P_4$), resulting from associated low order flux asymmetries, cause spatial variations in capsule & fuel momentum that prevent the DT “ice” layer from being decelerated uniformly by the hot spot pressure. This reduces the transfer of kinetic to internal energy of the central hot spot, thus reducing neutron yield. Furthermore, synthetic gated x-ray images of the hot spot self-emission indicate that $P_4$ shapes may be unquantifiable for DT layered capsules. Instead the positive $P_4$ asymmetry “aliases” itself as an oblate $P_2$ in the x-ray self emission images. Correction of this apparent $P_2$ distortion can further distort the implosion while creating a round x-ray image. Long wavelength asymmetries may be playing a significant role in the observed yield reduction of NIF DT implosions relative to detailed post-shot 2D simulations.'
author:
- 'R.H.H. Scott'
- 'D.S. Clark'
- 'D.K. Bradley'
- 'D.A. Callahan'
- 'M.J. Edwards'
- 'S.W. Haan'
- 'O.S. Jones'
- 'B.K. Spears'
- 'M.M. Marinak'
- 'R.P.J. Town'
- 'P.A. Norreys'
- 'L.J. Suter'
title: 'Numerical Modeling of the Sensitivity of X-Ray Driven Implosions to Low-Mode Flux Asymmetries'
---
[^1]
Indirect-drive inertial confinement fusion (ICF) [@Basov:1991uq; @NUCKOLLS:1972fk; @lindl:339] uses lasers to heat the inside of a cavity (or hohlraum). The absorbed laser energy is re-emitted as approximately black-body radiation in the soft x-ray regime. These x-rays heat the outer surface of a hollow, spherical, $\sim 2$ mm diameter, shell that contains a $\sim 70$ m thick layer of cryogenically frozen Deuterium and Tritium fuel (“DT fuel” or “DT layered capsules”). The heated outer shell ablates, which creates a reaction force, accelerating the remaining shell spherically inwards at extremely high velocity ($\sim 350$ km/s). During the implosion, spherical convergence causes the pressure in the gaseous void (or hot spot) within the shell to rise rapidly. This pressure decelerates the shell, simultaneously compressing the solid fuel and converting the shell’s kinetic energy into hot spot internal energy. If this conversion rate exceeds loss rates due to thermal conduction and bremsstrahlung radiation, the hot spot will heat, initiating DT fusion reactions. Provided the hot spot areal density is sufficient, $\alpha$-particles created by the fusion reactions will redeposit their energy locally, further heating the hot spot, resulting in bootstrap heating, ignition, and propagation of burn into the surrounding cold fuel. Numerical modeling indicates that the National Ignition Facility (NIF) can, for the first time, initiate inertial fusion ignition in the laboratory [@haan:051001; @clark:052703; @1742-6596-112-2-022021].
In this Letter, the effects of large, low-mode asymmetries in the x-ray drive are examined numerically. The non-uniformity of the x-ray flux incident upon the shell and the resultant non-spherical shell shapes can be described mathematically as a series of Legendre polynomials [@Abramowitz:1972kx]. It is shown that a large-amplitude $P_4$ implosion asymmetry, that might result from low-order hohlraum generated flux asymmetries, causes spatial variations in the capsule & fuel momentum. This can inhibit the DT fuel from being decelerated uniformly by the hot spot pressure, reducing the efficient transfer of implosion kinetic energy to hot spot internal energy thus significantly reducing the capsule performance. Furthermore, simulated gated x-ray images of the hot spot self-emission show reduced sensitivity to the $P_4$ mode, instead the images appear to have a pronounced oblate $P_2$ shape. Reducing the amplitude of the oblate $P_2$ shape (as measured from the x-ray image) further reduces the sensitivity to the $P_4$ mode such that no quantitative evaluation of the hot spot $a_4$ (where $a_4$ is the amplitude of the $P_4$ mode) can be made, furthermore the x-ray images are circular despite the capsule shape being highly distorted. Comparisons are made between key physical properties of the implosion, synthetically generated experimental observables, and NIF experimental data.
![(a) Applied radiation temperature vs time. Colors depict when during the drive pulse the various flux asymmetries were applied as shown on subsequent plots. (b) The relationship between applied flux asymmetry and hot spot $a_4$ at x-ray bangtime (the time at which peak emission occurs). Each point represents a 2D Hydra run. (c) Burn averaged hot spot + fuel + ablator $\rho r$ vs hot spot $a_4$ at x-ray bangtime: large dots are spatially averaged $\rho r$, while the smaller points with the same color and $a_4$ are the maxa and minima of the spatially averaged value; large variations in $\rho r$ occur due to $P_4$. (d) The burn averaged energy partition as a function of hot spot $a_4$; increasing $P_4$ perturbations prevent kinetic energy (black) from being converted to both hot spot (red) and solid fuel (blue) internal energy during stagnation. (e) Burn averaged hot spot pressure as a function of hot spot $a_4$. (f) Total thermonuclear neutron yield as a function of hot spot $a_4$; yield varies by a factor of 15. []{data-label="fig2"}](./trad_time.png "fig:") ![(a) Applied radiation temperature vs time. Colors depict when during the drive pulse the various flux asymmetries were applied as shown on subsequent plots. (b) The relationship between applied flux asymmetry and hot spot $a_4$ at x-ray bangtime (the time at which peak emission occurs). Each point represents a 2D Hydra run. (c) Burn averaged hot spot + fuel + ablator $\rho r$ vs hot spot $a_4$ at x-ray bangtime: large dots are spatially averaged $\rho r$, while the smaller points with the same color and $a_4$ are the maxa and minima of the spatially averaged value; large variations in $\rho r$ occur due to $P_4$. (d) The burn averaged energy partition as a function of hot spot $a_4$; increasing $P_4$ perturbations prevent kinetic energy (black) from being converted to both hot spot (red) and solid fuel (blue) internal energy during stagnation. (e) Burn averaged hot spot pressure as a function of hot spot $a_4$. (f) Total thermonuclear neutron yield as a function of hot spot $a_4$; yield varies by a factor of 15. []{data-label="fig2"}](./DTp4_bt_fluxmyC.png "fig:") ![(a) Applied radiation temperature vs time. Colors depict when during the drive pulse the various flux asymmetries were applied as shown on subsequent plots. (b) The relationship between applied flux asymmetry and hot spot $a_4$ at x-ray bangtime (the time at which peak emission occurs). Each point represents a 2D Hydra run. (c) Burn averaged hot spot + fuel + ablator $\rho r$ vs hot spot $a_4$ at x-ray bangtime: large dots are spatially averaged $\rho r$, while the smaller points with the same color and $a_4$ are the maxa and minima of the spatially averaged value; large variations in $\rho r$ occur due to $P_4$. (d) The burn averaged energy partition as a function of hot spot $a_4$; increasing $P_4$ perturbations prevent kinetic energy (black) from being converted to both hot spot (red) and solid fuel (blue) internal energy during stagnation. (e) Burn averaged hot spot pressure as a function of hot spot $a_4$. (f) Total thermonuclear neutron yield as a function of hot spot $a_4$; yield varies by a factor of 15. []{data-label="fig2"}](./DTrhor_DTp4myC.png "fig:") ![(a) Applied radiation temperature vs time. Colors depict when during the drive pulse the various flux asymmetries were applied as shown on subsequent plots. (b) The relationship between applied flux asymmetry and hot spot $a_4$ at x-ray bangtime (the time at which peak emission occurs). Each point represents a 2D Hydra run. (c) Burn averaged hot spot + fuel + ablator $\rho r$ vs hot spot $a_4$ at x-ray bangtime: large dots are spatially averaged $\rho r$, while the smaller points with the same color and $a_4$ are the maxa and minima of the spatially averaged value; large variations in $\rho r$ occur due to $P_4$. (d) The burn averaged energy partition as a function of hot spot $a_4$; increasing $P_4$ perturbations prevent kinetic energy (black) from being converted to both hot spot (red) and solid fuel (blue) internal energy during stagnation. (e) Burn averaged hot spot pressure as a function of hot spot $a_4$. (f) Total thermonuclear neutron yield as a function of hot spot $a_4$; yield varies by a factor of 15. []{data-label="fig2"}](./DTkeshell_DTp4myC.png "fig:") ![(a) Applied radiation temperature vs time. Colors depict when during the drive pulse the various flux asymmetries were applied as shown on subsequent plots. (b) The relationship between applied flux asymmetry and hot spot $a_4$ at x-ray bangtime (the time at which peak emission occurs). Each point represents a 2D Hydra run. (c) Burn averaged hot spot + fuel + ablator $\rho r$ vs hot spot $a_4$ at x-ray bangtime: large dots are spatially averaged $\rho r$, while the smaller points with the same color and $a_4$ are the maxa and minima of the spatially averaged value; large variations in $\rho r$ occur due to $P_4$. (d) The burn averaged energy partition as a function of hot spot $a_4$; increasing $P_4$ perturbations prevent kinetic energy (black) from being converted to both hot spot (red) and solid fuel (blue) internal energy during stagnation. (e) Burn averaged hot spot pressure as a function of hot spot $a_4$. (f) Total thermonuclear neutron yield as a function of hot spot $a_4$; yield varies by a factor of 15. []{data-label="fig2"}](./DTprhs_DTp4myC.png "fig:") ![(a) Applied radiation temperature vs time. Colors depict when during the drive pulse the various flux asymmetries were applied as shown on subsequent plots. (b) The relationship between applied flux asymmetry and hot spot $a_4$ at x-ray bangtime (the time at which peak emission occurs). Each point represents a 2D Hydra run. (c) Burn averaged hot spot + fuel + ablator $\rho r$ vs hot spot $a_4$ at x-ray bangtime: large dots are spatially averaged $\rho r$, while the smaller points with the same color and $a_4$ are the maxa and minima of the spatially averaged value; large variations in $\rho r$ occur due to $P_4$. (d) The burn averaged energy partition as a function of hot spot $a_4$; increasing $P_4$ perturbations prevent kinetic energy (black) from being converted to both hot spot (red) and solid fuel (blue) internal energy during stagnation. (e) Burn averaged hot spot pressure as a function of hot spot $a_4$. (f) Total thermonuclear neutron yield as a function of hot spot $a_4$; yield varies by a factor of 15. []{data-label="fig2"}](./DTnyield_DTp4myC.png "fig:")
The indirect-drive approach to ICF smooths high mode spatial non-uniformities in the x-ray flux incident on the capsule, however the spatial distribution of the cones of laser beams which illuminate the hohlraum means that low mode x-ray flux non-uniformities can occur [@lindl:339], these are considerably lower mode than those recently examined by Thomas *et al* [@PhysRevLett.109.075004]. The growth of Legendre polynomial $P_4$ capsule shapes was investigated using the radiation-magnetohydrodynamics code Hydra [@marinak:2275]. A frequency and time dependent x-ray source was developed to drive these capsule-only simulations. The initial x-ray drive was taken from an integrated hohlraum simulation and then adjusted to match the shock timing data obtained using the VISAR diagnostic [@barker:4669] from NIF shot N110521, and the capsule implosion trajectory [@hicks:102703] measured on NIF shot N110625. A $90^{\circ}$ ‘wedge’ of the capsule ($2\pi$ Sr) was modeled using two-dimensional (2D) cylindrically-symmetric geometry with $256\times312$ cells. Doubling and quadrupling the cell resolution demonstrated convergence. In all runs the Quotidian Equation of State [@more:3059] was used with tabular opacities and multi-group radiation diffusion. The effects of Legendre polynomial $P_4$ hohlraum flux asymmetries were investigated by perturbing the tuned x-ray drive with spatially varying flux asymmetries of the form: $fds(\theta,t)=(a_0P_0 + a_4P_4(\theta))*fds(t)$ where $fds$ is the energy density of the tuned photon frequency dependent x-ray drive source, $a_n$ is the amplitude of the $n^{th}$ Legendre polynomial, $a_0=1$, $a_4= (\pm0.10, \pm0.05, \pm0.025, \pm0.01, \pm0.005, \pm0.0025)$, $\theta$ the angle between the equatorial plane and polar axis, and $t$ time. Hydra modeling of the hohlraum & capsule for nominal implosions suggests the flux asymmetry incident on the capsule would be expected to vary by $<3$% except for in the first $\sim 2$ ns of the laser pulse were it can vary by up to 10% [@Jones:2012]. The flux asymmetries were applied 100 m from the capsule ablation front during discrete time intervals (see fig. \[fig2\](a)), creating a database of $>200$ 2D modeling runs of both DT layered implosions and DHe$^3$ gas filled capsules with a surrogate inner CH layer of equal mass to a DT fuel layer (symmetry capsules). Time resolved synthetic gated x-ray images of the hot spot self-emission $>6$ keV, including its attenuation by the compressed fuel and ablator, were created from both polar and equatorial directions by post processing each Hydra run. The images were blurred in order to reproduce the $11$ m resolution of the diagnostic. The key implosion performance metrics (neutron yield, hot spot pressure, mass, volume, density, ion & electron temperatures, the effective ion temperature computed from the FWHM of the DT neutron spectrum, fuel and ablator areal density ($\rho r$) and kinetic energy) were extracted from the simulations. The hot spot shape was evaluated as a function of time by performing a Legendre polynomial decomposition (modes 1-10) of the appropriate contour. For DT layered capsules the hot spot contour is defined for each angular ‘strip’ of cells $j$ as the minimum radius where $T_{e_j} > \frac{1}{2}T_{e_{j_{max}}}$ and $\rho_j < \frac{1}{2}\rho_{j_{max}}$ where $T_{e}$ is the electron temperature and $\rho$ the mass density, ‘max’ denotes the maximum value within the $j^{th}$ strip. This has been found to produce a robust definition of the hot spot even for highly distorted implosions. The 17% contour of the gated x-ray diagnostic (GXD) is used both for the synthetic GXD and experimentally, as previous studies have shown this provides a faithful representation of the hot spot shape for small departures from sphericity.
![Axis of rotational symmetry is horizontal at Radius $=0$ m. (a) DT layered capsule density plot at x-ray bangtime showing a positive Legendre polynomial $P_4$ shape. This simulation had a 10% flux asymmetry applied from 11.5-14 ns. Black arrows indicate the mass flows which occur during stagnation. After bangtime ‘fingers’ of fuel continue to flow inwards (red arrows). White dots depict the hot spot contour. (b) Synthetic gated x-ray image of the hot spot self emission from (a), white dots show the 17% contour. (c) Fig. (a) 100 ps later: due to burn truncation with large $a_4$ this is the neutron bangtime for an equivalent spherical implosion. (d) The synthetic GXD from (c), showing a large negative $P_2$ and almost zero $a_4$ despite the obvious $P_4$ in (c).[]{data-label="picture"}](./density_paper.png "fig:") ![Axis of rotational symmetry is horizontal at Radius $=0$ m. (a) DT layered capsule density plot at x-ray bangtime showing a positive Legendre polynomial $P_4$ shape. This simulation had a 10% flux asymmetry applied from 11.5-14 ns. Black arrows indicate the mass flows which occur during stagnation. After bangtime ‘fingers’ of fuel continue to flow inwards (red arrows). White dots depict the hot spot contour. (b) Synthetic gated x-ray image of the hot spot self emission from (a), white dots show the 17% contour. (c) Fig. (a) 100 ps later: due to burn truncation with large $a_4$ this is the neutron bangtime for an equivalent spherical implosion. (d) The synthetic GXD from (c), showing a large negative $P_2$ and almost zero $a_4$ despite the obvious $P_4$ in (c).[]{data-label="picture"}](./btgxd_paper.png "fig:") ![Axis of rotational symmetry is horizontal at Radius $=0$ m. (a) DT layered capsule density plot at x-ray bangtime showing a positive Legendre polynomial $P_4$ shape. This simulation had a 10% flux asymmetry applied from 11.5-14 ns. Black arrows indicate the mass flows which occur during stagnation. After bangtime ‘fingers’ of fuel continue to flow inwards (red arrows). White dots depict the hot spot contour. (b) Synthetic gated x-ray image of the hot spot self emission from (a), white dots show the 17% contour. (c) Fig. (a) 100 ps later: due to burn truncation with large $a_4$ this is the neutron bangtime for an equivalent spherical implosion. (d) The synthetic GXD from (c), showing a large negative $P_2$ and almost zero $a_4$ despite the obvious $P_4$ in (c).[]{data-label="picture"}](./density2_paper.png "fig:") ![Axis of rotational symmetry is horizontal at Radius $=0$ m. (a) DT layered capsule density plot at x-ray bangtime showing a positive Legendre polynomial $P_4$ shape. This simulation had a 10% flux asymmetry applied from 11.5-14 ns. Black arrows indicate the mass flows which occur during stagnation. After bangtime ‘fingers’ of fuel continue to flow inwards (red arrows). White dots depict the hot spot contour. (b) Synthetic gated x-ray image of the hot spot self emission from (a), white dots show the 17% contour. (c) Fig. (a) 100 ps later: due to burn truncation with large $a_4$ this is the neutron bangtime for an equivalent spherical implosion. (d) The synthetic GXD from (c), showing a large negative $P_2$ and almost zero $a_4$ despite the obvious $P_4$ in (c).[]{data-label="picture"}](./gxd+100ps_paper.png "fig:")
The applied Legendre $P_4$ flux asymmetries induce $P_4$ hot spot shapes at stagnation (see Figs. \[picture\] (a) & (c)), the sign of which is dependent on the timing of the applied flux asymmetry. If the asymmetry is present only during the shock compression phase (the first $\sim 18$ ns), shocks created in regions of the capsule exposed to higher flux propagate faster, these faster shocks break out of the inner DT ice layer earlier, causing these regions to move ahead of those exposed to less flux. This also causes ablator mass to flow laterally, away from the high flux region. Consequently during peak drive the regions initially exposed to high flux are at smaller radii, meaning they are accelerated less efficiently by the hohlraum flux and gain less total momentum. They can also have less ablator aerial density. The net effect is that the regions experiencing high flux during shock compression will protrude outwards at stagnation. Conversely if the flux asymmetry is applied during peak drive, the regions of the capsule exposed to more flux gain more momentum, and protrude inwards at stagnation. Regardless of the timing of the applied asymmetry, during the stagnation phase of the implosion, pressure within the lower density hot spot decelerates the higher density fuel from peak velocity, making this interface Rayleigh-Taylor unstable [@Rayleigh:1900kx; @Taylor:1950vn]. The instability will accentuate any shape imperfections during deceleration, as indicated by the significant simulated growth shown in figs. \[picture\](a) & (c).
The scalings of some important DT layered capsule implosion parameters as a function of hot spot $a_4$ measured at x-ray bangtime are summarized in figure \[fig2\]. Fig. \[fig2\](b) shows the relationship between applied $P_4$ flux perturbation amplitude and the resulting shape $a_4$ at x-ray bangtime. Fig. \[fig2\](c) depicts the ‘burn averaged’ $\rho r$ (the burn average of a quantity $Q_b = (\sum_{t=0}^{t=\infty} Q_t Eprodr dt)/\int_{t=0}^{t=\infty}Eprodr\ dt$ where $Q_t$ is Q at time $t$ and $Eprodr$ the thermonuclear energy production rate in time $dt$) as a function of hot spot $a_4$. Although the spatially averaged $\rho r$ is relatively constant, the lateral mass flows caused by the $P_4$ can create large spatial variations in $\rho r$. The regions with higher momentum continue to propagate radially inwards; fig. \[fig2\](d) depicts the remaining capsule kinetic energy as a function of $a_4$ and the partition of that energy into hot spot and solid fuel internal energy. For large $a_4$ less of the implosion kinetic energy is converted into hot spot internal energy and the hot spot pressure is reduced (see fig. \[fig2\](e)). The reduction in neutron yield can be as large as 15$\times$ for hot spot $a_4=20$ m (flux asymmetry $\sim 10\%$) as shown in fig. \[fig2\](f)).
![(a) DT layered capsule hot spot $a_4$ plotted against the synthetic GXD $a_4$; particularly for large positive $a_4$ the GXD is unable to effectively measure the amplitude of the $P_4$ mode. (b) DT layered capsule hot spot $a_4$ plotted against the synthetic GXD $a_2$; the GXD measures a significant $P_2$ mode amplitude despite the DT layered capsule hot spot $a_2$ being $0\pm1$ m (not shown).[]{data-label="fig3"}](./DTp4myC_DTp4gxd17pc.png "fig:") ![(a) DT layered capsule hot spot $a_4$ plotted against the synthetic GXD $a_4$; particularly for large positive $a_4$ the GXD is unable to effectively measure the amplitude of the $P_4$ mode. (b) DT layered capsule hot spot $a_4$ plotted against the synthetic GXD $a_2$; the GXD measures a significant $P_2$ mode amplitude despite the DT layered capsule hot spot $a_2$ being $0\pm1$ m (not shown).[]{data-label="fig3"}](./DTp4myC_DTp2gxd17pc.png "fig:")
Analysis of synthetic GXD images created from the 2D Hydra runs suggest that the $a_4$ measured experimentally with the GXD is not a true representation of the hot spot $a_4$, particularly for large positive $a_4$ amplitudes. Fig. \[fig3\](a) depicts the relationship between the DT layered capsule “hot spot $a_4$” (as previously defined) and that from the 17% contour of the synthetic GXD (the synthetic GXD $a_4$), both were extracted at x-ray bangtime (the principal value used for analysis of experimental data). $a_4$ measured from the synthetic GXD is consistently lower than that of the hot spot. The insensitivity to positive hot spot $a_4$ is caused by lateral ablator mass flows which accumulate at $\sim 45 ^{\circ}$ at the expense of ablator material near the equator and poles (see Fig. \[picture\] (a)). The ablator material is rotationally symmetric about the horizontal axis, so the accumulated material absorbs the x-rays emitted from the polar-lobes of the hot spot (left and right), while allowing x-rays to more readily pass through the equatorial regions (top and bottom). Consequently the polar-lobes of the hot spot which are visible in the density plots of Fig. \[picture\] as dark regions (the hot spot is the central region of low density) are almost completely invisible in the GXD plots compared to the emission through the equator. This causes the x-ray image to have a pronounced negative $P_2$ shape (oblate or “pancaked”). As the hot spot $a_2=0\pm 1$ m ($a_2$ is the amplitude of the $P_2$ mode) for all these pure $P_4$ modelling runs, the $P_2$ inferred from the x-ray image is a “false” negative $P_2$ mode. This suggests that a negative $P_2$ mode measured from the self-emission x-ray image may in fact be a signature of a positive $P_4$ mode, although it does not, of course, preclude the presence of a true $P_2$ mode. This is potentially important for interpretation of x-ray images from DT implosions, which often exhibit oblate (negative) $P_2$ modes [@Glenzer05032010].
NIF experiments also use low convergence, DHe$^3$ gas filled “symmetry capsules” which have a surrogate CH fuel mass. Using Hydra, symmetry capsule & DT layered capsule pairs of runs were created by applying identical x-ray drives to both capsules. When realistic experimental noise is applied to the synthetic GXD images, symmetry capsules, which have better signal to noise ratios and larger stagnation diameters, enable a far better measurement of the $P_4$ mode than equivalent DT layered capsules. Nevertheless, these calculations indicate symmetry capsules also show reduced sensitivity to $a_4$, exhibit a “false” $P_2$, and are quantitatively very similar to those shown in Figs. \[fig3\] (a) & (b) respectively.
![(a) Density plot of a DT layered capsule run with both $P_2$ and $P_4$ flux modes applied. Axis of rotational symmetry is horizontal at Radius $=0$ m. (b) The equatorial synthetic GXD image of (a) at the same time, dotted line shows 17% contour. Despite the highly non-spherical density distribution, the equatorial GXD image is almost perfectly round. (c) As the $P_2$ flux amplitude is increased in order to make the GXD image look round (reducing $a_2$) the sensitivity to $a_4$ is reduced towards zero.[]{data-label="fig5"}](./10pcP2_10pcP4_DT_density_at_xray_bt.png "fig:") ![(a) Density plot of a DT layered capsule run with both $P_2$ and $P_4$ flux modes applied. Axis of rotational symmetry is horizontal at Radius $=0$ m. (b) The equatorial synthetic GXD image of (a) at the same time, dotted line shows 17% contour. Despite the highly non-spherical density distribution, the equatorial GXD image is almost perfectly round. (c) As the $P_2$ flux amplitude is increased in order to make the GXD image look round (reducing $a_2$) the sensitivity to $a_4$ is reduced towards zero.[]{data-label="fig5"}](./10pcP2_10pcP4_DTequatorial_gxd_at_xray_bt.png "fig:") ![(a) Density plot of a DT layered capsule run with both $P_2$ and $P_4$ flux modes applied. Axis of rotational symmetry is horizontal at Radius $=0$ m. (b) The equatorial synthetic GXD image of (a) at the same time, dotted line shows 17% contour. Despite the highly non-spherical density distribution, the equatorial GXD image is almost perfectly round. (c) As the $P_2$ flux amplitude is increased in order to make the GXD image look round (reducing $a_2$) the sensitivity to $a_4$ is reduced towards zero.[]{data-label="fig5"}](./p2p4DTgxdSensitivity.png "fig:")
DT implosions on the NIF currently have yields $\sim 3 - 10 \times$ below detailed 2D post-shot Hydra simulations [@Clark:2012fk] that match the measured shock timing, implosion velocity, and capsule and ice surface roughnesses. In comparison to experimentally measured or inferred values [@Springer:2011uq], these simulations have similar hot spot temperatures, however the hot spot volumes are reduced while the hot spot mass is increased, causing a $2-3\times$ discrepancy in the hotspot density. $P_4$ shape perturbations offer one mechanism which may explain these experimental observations in particular bringing the yield and ion temperature relationship into better agreement. In these simulations, the DT fuel and hot spot do not mix; clear boundaries still exist (note these simulations use smooth capsules, but when nominal realistic capsule surface roughness [@clark:082701] was employed and modes up to 200 resolved, no significant implosion degradation occurred for the full range of $a_4$). Consequently unlike high mode ‘mix’ [@lindl:339] (where the hot spot can be radiatively cooled by high $Z$ impurities), the simulated ion temperature inferred from the neutron spectrum remains unaffected at $3.9\pm 0.05$ keV for all $a_4$. The large $a_4$ does however truncate the thermonuclear burn, moving both the neutron and x-ray bangtimes earlier in time, therefore as the capsule is still converging at bangtime, the hot spot size and volume are increased. The hot spot mass decreases with positive $a_4$, bringing Hydra simulations approximately in line with experimental data, as shown in Table \[table1\]. This compares NIF experimental data [@Springer:2011uq] with two Hydra implosions; one is perfectly spherical while the other has a hot spot $a_4$ of +20 m. Notable features are the significantly reduced yield, reduced pressure, reduced hot spot mass, unchanged ion temperature and increased hot spot volume. We must emphasize, however, that this should not be interpreted as conclusive evidence that a $P_4$ asymmetry is responsible for the observed reduced NIF capsule performance. Although this study has concentrated on the $P_4$ mode, it is likely that all low modes would reduce the conversion of capsule kinetic energy into hot spot pressure, and may result in similar ambiguity in the shape of the x-ray emission from the hot spot [@B.K.:2012uq]. To explore the issue of low mode asymmetries further, experiments using x-ray backlighters are currently being conducted on NIF to measure the implosion shape both in-flight [@hicks:102703] and at stagnation using Compton radiography [@tommasini:056309].
-------------------------------- ------------------------- --------------- ---------------- -- -- -- -- --
NIF expt. Hydra Hydra
range[@Springer:2011uq] ($a_4 = 0$ m) ($a_4 = 20$ m)
Pressure (GBar) 57-81 348 115
Mass (g) 2-6.4 8 5.5
Density (gcm$^{-3}$) 22-35 136 69
Volume ($\times 10^{-7}cm^3$) 0.9-1.9 0.6 1.0
Tion (keV) 3.3-4.4 3.9 3.9
Fuel $\rho r$ (gcm$^{-2}$) 0.77-0.98 0.7 0.72
Yield (neut. $\times 10^{14}$) 1.9-6.0 74 5.3
-------------------------------- ------------------------- --------------- ---------------- -- -- -- -- --
: A comparison of NIF DT layered capsule experimental data from 4 shots N110608-N110908 with two Hydra implosions, one spherical ($a_4=0$ m, and another with $a_4 = 20$ m. Large positive $P_4$ brings the modeled implosion observables approximately in line with the experimental data.[]{data-label="table1"}
As discussed, implosions with a significant $P_4$ asymmetry can have a very apparent but “false” $P_2$ asymmetry in GXD images. We find that attempting to correct this “false” $P_2$ by increasing laser power to the hohlraum waist (capsule equator) [@Glenzer05032010] can lead to a round GXD image even though the correction actually produces a more distorted DT fuel ice layer. This is depicted in fig. \[fig5\] for the case of a DT layered capsule where we applied and empirically adjusted a $P_2$ flux asymmetry, in addition to the original $P_4$, in order to make the synthetic GXD image appear round. Fig. \[fig5\](c) quantifies a related effect; as the applied $P_2$ flux is increased in order to reduce the “false” GXD $a_2$ towards zero, there is a marked additional reduction in sensitivity to $a_4$ (relative to that shown in Fig. \[fig3\]). This suggests that attempts to tune the hohlraum to eliminate a “false” $P_2$ can have the unintended consequence of exacerbating overall asymmetry. Other information, such as comparison of the widths of images taken from both the polar and equatorial lines of sight [@benedetti:2012] need to be taken into consideration. These simulations show that when a hotspot has a positive but pure $P_4$ asymmetry the equatorial image width is larger than the width in the polar image (for negative $P_4$ this is reversed). This could be used to identify an implosion where the measured $P_2$ may be caused by a dominant $P_4$ asymmetry. However, our simulations also show us that the empirically but incorrectly tuned implosion of fig. \[fig5\] would have a polar image width that is equal to the equatorial image width, further misleading us into thinking that we had engineered an approximately spherical implosion. A corollary of figure \[fig5\], is that it is possible to create imploded configurations which appear to be symmetric in the GXD but, in fact, are significantly asymmetric and have greatly reduced performance in comparison to equivalent spherical implosions because a large fraction of the imploding shell’s kinetic energy remains unstagnated.
In summary, numerical simulations have been used to examine the sensitivity of implosions similar to those currently taking place on NIF to low-mode flux asymmetries. It is shown that Legendre polynomial $P_4$ flux modes induce $P_4$ shape modes at the time of capsule stagnation. The largest $P_4$ amplitudes can cause up to 50% of the capsule kinetic energy to remain unconverted to hot spot and DT ice internal energy, in turn reducing the neutron yield by up to $15\times$. Simulated x-ray images of the hot spot self-emission show reduced sensitivity to the positive $P_4$ mode, instead the images appear to have a pronounced oblate $P_2$ shape. Attempting to correct for this apparent $P_2$ distortion can further distort the implosion while creating x-ray images which appear round & self-consistent from both equatorial and polar directions. This also further reduces the sensitivity to the $P_4$ mode such that that no quantitative evaluation of the hot spot $a_4$ can be made. Long wavelength asymmetries may be playing a significant role in the observed yield reduction of NIF DT implosions relative to detailed post-shot 2D simulations.
[10]{}
J. D. Lindl *et al.*, Physics of Plasmas **11**, 339 (2004).
N. Basov, Edward teller lectures (1991).
J. NUCKOLLS *et al.*, Nature **239**, 139 (1972).
S. W. Haan *et al.*, Physics of Plasmas **18**, 051001 (2011).
D. S. Clark *et al.*, Phys. Plamas **17**, 052703 (2010).
D. A. Callahan *et al.*, Journal of Physics: Conference Series **112**, 022021 (2008).
M. Abramowitz and I. A. Stegun, *Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables*, volume pp. 331-339 and 771-802, Dover (1972).
V. A. Thomas and R. J. Kares, Phys. Rev. Lett. **109**, 075004 (2012).
M. M. Marinak *et al.*, Physics of Plasmas **8**, 2275 (2001).
L. M. Barker and R. E. Hollenbach, Journal of Applied Physics **43**, 4669 (1972).
D. G. Hicks *et al.*, Physics of Plasmas **17**, 102703 (2010).
R. M. More *et al.*, Physics of Fluids **31**, 3059 (1988).
O. S. Jones, Private Communication, (2012).
L. Rayleigh, *Scientific Papers*, Cambridge University Press, Cambridge, England (1900).
G. I. Taylor, *The Instability of Liquid Surfaces when Accelerated in a Direction Perpendicular to their Planes. I*, Proc. R. Soc. London, Ser. A (1950).
S. H. Glenzer *et al.*, Science **327**, 1228 (2010).
D. Clark *et al.*, Physics of Plasmas *(Submitted)* (2012).
P. Spinger and C. Cerjan, in [*IFSA Proceedings*]{} Bordeaux, France (2011).
B.K. Spears, Private Communication, (2012).
R. Tommasini *et al.*, Physics of Plasmas **18**, 056309 (2011).
L.R. Benedetti, [*APS DPP Proceedings*]{}, Providence, RI, United States (2012).
[^1]: Also visiting scientist at Lawrence Livermore National Laboratory, Livermore, CA 94551, United States of America, & Department of Physics, The Blackett Laboratory, Imperial College London, Prince Consort Road, London, SW7 2AZ, United Kingdom. The authors thank M.H. Key and C. Cerjan for useful discussions, the staff of NIF and Livermore Computing.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Tree transducers are formal automata that transform trees into other trees. Many varieties of tree transducers have been explored in the automata theory literature, and more recently, in the machine translation literature. In this paper I review T and xT transducers, situate them among related formalisms, and show how they can be used to implement rules for machine translation systems that cover all of the cross-language structural divergences described in Bonnie Dorr’s influential article on the topic. I also present an implementation of xT transduction, suitable and convenient for experimenting with translation rules.'
author:
- |
Alex Rudnick\
School of Informatics and Computing, Indiana University\
Bloomington, Indiana, USA\
[[email protected]]{}
bibliography:
- 'transducers.bib'
title: 'Tree Transducers, Machine Translation, and Cross-Language Divergences'
---
Introduction
============
Word-based approaches to statistical machine translation, starting with the work from IBM in the early 1990s [@DBLP:journals/coling/BrownPPM94] have been successful both in use in production translation systems and in invigorating MT research. Since then, newer phrase-based MT techniques such as the alignment template model [@DBLP:journals/coling/OchN04], and hierarchical phrase-based models [@chiang:2005:ACL] have made significant improvements in SMT translation quality.
Despite their sophistication and apparent complexity, many word-based and phrase-based SMT models can be implemented entirely in terms of finite-state transducers. This allows researchers to make use of the rich automata literature for finding clean and efficient algorithms; it is also useful from a software engineering perspective, making it possible to do experiments quickly, using generic toolkits for programmatically manipulating finite-state transducers. Several such packages are freely available, such as OpenFST [@openfst] and the WRTH FSA Toolkit [@kanthak-ney:2004:ACL].
However, since they make no attempt to explicitly model the syntax of either involved language, and typically use simple n-gram models to guide generation, the output of word-based SMT systems can be syntactically incoherent, especially in light of long-distance dependencies. Additionally, word-based SMT models have difficulties encoding word order differences across languages.
So we have seen new methods in MT that explicitly model syntax, where typically the grammar of a language, and the relationships between the grammars of two languages, can be learned from treebanks. There are many different available theoretical frameworks for describing syntax and transformations over syntactic representations. Both from a theoretical standpoint, and as MT implementors, we would like a framework that is clean and general, and is suitably expressive for explicitly capturing syntactic structures and the divergences across languages. We would also like one for which there are efficient algorithms for training rules and performing transduction (i.e., decoding at translation time), and ideally one for which a good software toolkit is freely available.
Not all syntactic relationships can be cleanly represented with every syntactic formalism; each formalism has its own expressive power. Bonnie Dorr provides us with an excellent test bed of seven cross-language divergences that may occur when we want to perform translation, even between languages as closely related as English, Spanish and German [@DBLP:journals/coling/Dorr94]. While these divergences do not totally describe the ways in which languages can differ in their typical descriptions of an event, they provide a concrete starting point, and are easily accessible.
In this paper, I specifically investigate T and xT transducers, situate them in the space of formalisms for describing syntax-based translation, and demonstrate that xT transducers are sufficient for modeling all of the syntactic divergences identified by Dorr. I also present `kurt`, a small software toolkit for experimenting with these transducers, which comes with sample translation rules that handle each of Dorr’s divergences.
In the rest of this paper, we will discuss some relevant grammar and transducer formalisms, including a more in-depth look at T and xT transducers; go through the linguistic divergences discussed by Dorr and explain why they might cause difficulties for MT systems; show how xT transducers can be used to address each of these divergences; present the software that I have built; review some of the related work that has informed this paper; and finally, suggest future possible directions for work with tree transducer-based MT.
Grammars and Transducers
========================
Here we contrast several kinds of formalisms over strings, trees, and pairs of strings and trees; please see Figure \[glossary\] for a glossary of different kinds of automata and grammars that will be referenced in this paper. A grammar describes a single set of strings or trees, and consists of a finite set of rules that describes those strings or trees. Familiar formalisms for grammars that describe sets of strings include context-free grammars and the other members of the Chomsky Hierarchy. Some grammars describe sets of trees, and these will be the main focus of the rest of this paper; when discussing grammars over strings, I will specifically mention it. For example, regular tree grammars (RTG) is the class of grammars corresponding to context-free grammars but describing trees; they describe the trees whose *yield* (string concatenation of the symbols at the leaves) is a context-free grammar [@KnightGraehlOverview].
Contrastingly, *synchronous* grammars describe sets of pairs of objects; here again, we are mostly concerned with synchronous grammars that describe trees. Formally, a synchronous grammar over trees establishes a mathematical relation over two sets of trees, and allows us to answer the question of whether, for a given pair of trees, that pair is in the relation. The production rules of a synchronous grammar do not just describe one language, but have pairs of production rules $<r_1,r_2>$, such that when $r_1$ is used to derive a string in language $L_1$, $r_2$ must be used in the derivation of a string in $L_2$.
Thus synchronous grammars can be used for several kinds of tasks, such as parsing parallel texts, generating parallel text, or most intuitively useful for a machine translation setting, parsing text in one language while jointly generating parse trees that yield text the other. All of these operations are described for synchronous context-free grammars in David Chiang’s tutorial [@Chiang06anintroduction]. In his tutorial, Chiang describes some of the limitations of using synchronous CFGs; notably, they cannot rearrange parts of parse trees that are not sisters. Of particular interest in this work is raising and lowering elements; Chiang gives the example of swapping subjects and objects, as in the example of translation between English and French in Figure \[missesmary\]. Chiang points out that, for syntax-aware MT, we would like to be able to use some more powerful formalism that can perform transformations like this. Synchronous tree substitution grammars, for example, are able to describe transformations of this form, but not the transformation from cross-serial dependencies in subordinate clauses in Dutch to the nested clause structure of English. This latter transformation would require more formal power, which is offered by tree-adjoining grammars.
TAG and Related Formalisms {#sec:tagfamily}
--------------------------
Tree adjoining grammar, introduced by Joshi [@Joshi:1975:TAG:1739967.1740303], has been a popular formalism for describing grammars over trees. It provides additional expressive power not available in regular tree grammars, handling some, but not all context-sensitive languages. TAG can cleanly describe many of the non-context-free features observed in human languages, such as the cross-serial dependencies in Dutch. TAG is thus called “weakly context-sensitive", and has been shown formally equivalent to several other syntactic formalisms, such as Combinatory Categorial Grammar (CCG) and Linear Indexed Grammars [@vw94].
The operations of TAG are substitution and adjunction, which combine the two different kinds of elementary trees present in a given TAG grammar, initial trees and auxiliary trees. The substitution operation takes two trees, one with a leaf that is an unresolved nonterminal $\alpha$, and produces a new tree in which that node has been replaced with an entire subtree (copied from another initial tree in the grammar, or one that has already been derived) whose root node is also $\alpha$. For example, an initial tree may have an unresolved nonterminal that wants to have an NP attached to it (it has a leaf labelled NP); the substitution operation attaches an existing subtree whose root is NP, producing a new tree where that nonterminal is now resolved. The adjunction operation takes an existing tree and an auxiliary tree, which has a special node marked as the “foot", and grafts the auxiliary tree in place in the middle of the existing tree, attaching the tree material at the target location to the foot node of the auxiliary tree. For a very clear tutorial on TAG with good examples, please see [@vannoord93], Section 4.2.4. Synchronous TAG has also been investigated, and its use in machine translation has been advocated by Shieber, who argues that its expressive power may make up for its computational complexity [@textscshieber:2007:SSST].
Restricted versions of TAG and their synchronous analogues have also been investigated. These do not provide the full expressive power of TAG, but can be parsed and trained more efficiently. The two limited versions of TAG that are most prominently discussed in the literature are tree substitution grammars (TSG) and tree insertion grammars (TIG). TSG only provides the substitution operation, and does not have auxiliary trees or adjunction [@eisner:2003:ACL-companion]. TIG, on the other hand, includes both the substitution and adjunction operations, but places constraints on the permissible shapes of auxiliary trees: their foot nodes must be at the leftmost or rightmost edge of the frontier, and a given derivation may not adjoin “left" auxiliary trees into “right" ones, or vice-versa. These restrictions are sufficient to limit the weak generative capacity of TIGs to that of CFGs, but they also ensure that algorithms on TIGs can run more efficiently. While parsing with a TAG takes in the general case $O(n^6)$ complexity, TIG (like the general case for CFGs) can be parsed in $O(n^3)$ [@Nesson:2006:IPS]. Both STIG and STSG have seen use in machine translation; for example, probabilistic STIG is used in [@Nesson:2006:IPS], and STSG has been notably used in [@eisner:2003:ACL-companion].
Tree Transducers
----------------
While synchronous grammars provide a *declarative* description of a relation that holds between two sets of trees, tree transducers more explicitly describe the process by which a tree may be converted into other trees. Like finite-state transducers, which operate over strings, tree transducers typically describe nondeterministic processes, so for a given input tree, there is a set of possible output trees; that set may (for example) be described by a regular tree grammar.
Tree transducers and synchronous grammars both describe mathematical relations over trees, so we can sensibly ask about their comparative formal expressive power, and use them to compute similar queries. For example, with either a synchronous grammar or a transducer, we may ask, for a given tree, what are the other trees that are in the mathematical relation with it [@Chiang06anintroduction]. There are transducer varieties with the same formal expressive power as certain synchronous grammars. For example, synchronous tree substitution grammars (STSG) have the same formal power as xLNT transducers [@Maletti:2010:WST:1857999.1858129], which will be described in more detail in the next section.
While there are very many kinds of possible tree transducers, the ones used in NLP applications typically fall into one of two classes, T transducers, which operate “top-down", and “B" transducers, which operate “bottom-up".
- synchronous grammar: a grammar over two languages simultaneously, where rules are given in pairs and must be used together
- probabilistic grammar: a grammar where rules have associated weights, which defines a probability distribution over derivations licensed by that grammar
- TAG: tree adjoining grammar, mildly context-sensitive grammar formalism over trees, with substitution and adjunction operations
- TIG: tree insertion grammar: a TAG wherein rules have certain restrictions, described in Section \[sec:tagfamily\].
- TSG: tree substitution grammar; similar to TAG without the adjoining operation
- RTG: regular tree grammar, the tree analogue of context-free grammar
- finite-state transducers: transducers over strings; finite-state automata with the added ability to produce output
- tree transducers: automata that define relations over trees procedurally
- T transducers: “top down" tree transducers
- R transducers: the same as T transducers, a name used in earlier work. “R" stands for “Root to Frontier"
- (L)T transducers: “linear" T transducers, constrained such that their rules are non-copying, and a variable appearing on the left-hand side of a rule must appear at most once in the right-hand side
- (N)T transducers: “nondeleting" T transducers, constrained such that a variable appearing on the left-hand side of a rule must appear at least once in the right-hand side
- (x)T transducers: T transducers with “extended" pattern matching, allowing for complex finite patterns to appear in the left-hand side of rules.
T Transducers
=============
Let us now describe T transducers in more detail. T transducers transform trees into other trees via sets of production rules. Many production rules may apply at a given step in a derivation, so the transductions are usually nondeterministic, relating a given input tree to many possible output trees. Thus a T transducer, like a synchronous tree grammar, defines a *relation* over sets of trees.
Intuitively, transduction begins with an input tree, where its root node is in the initial state $q_0$. Each node in a tree may be in one of the states in $Q$ (the set of possible states), or in no state at all. Transduction proceeds by finding all the transduction rules that can apply to an existing tree, or subtrees of an existing tree. A rule applies when the root of its left-hand side matches a node in the tree, and the state of the node matches the state of the rule. When a rule matches a subtree (call it $t$), then a new tree is produced and added to the set of current trees by replacing the subtree that matched the rule with the right-hand side of the rule, save that the variables in the right-hand side of the rule have been replaced by the corresponding subtrees of $t$. Additionally, a rule may specify that subtrees of the new tree being produced should be in states as well, indicating that more transduction work must be done on them before the derivation is finished. A complete, successful transduction in a T transducer begins with the root node being in the initial state, then states propagating down the tree to the leaves, until the entire tree has been transduced. See Figure \[exampletransduction\] for an illustrative example, adapted from [@DBLP:journals/coling/GraehlKM08].
Rule 1: $\longrightarrow$
Rule 2: $\longrightarrow$
Input tree:
Tree after application of Rule 1:
Tree after subsequent application of Rule 2:
A new tree is produced by replacing the subtree that matched the rule with the right-hand side of the rule, with its variables filled in with the appropriate subtrees. The new tree is then added to the inventory of current trees in the usual way for production systems. A transduction is complete for a tree in the inventory when all of its nodes are no longer in states; at this point, the states will have propagated all the way from the top of the tree to the leaves, and then be resolved; in the case of translation, the symbols in the tree will be words in the output language. The transduction process is nondeterministic; many rules may apply to a given tree in the inventory, and even the same rule may apply to different subtrees. To do a complete search for all possible transductions, we apply each rule to every subtree where it is applicable, and produce every possible resulting tree; beam search may also be done, where search paths with low probabilities are pruned.
Formally, a T transducer has the following elements.
- an input alphabet $\Sigma$
- an output alphabet $\Delta$
- a set of *states* $Q$
- an *initial state*, typically denoted $q_0$
- transition rules, which are tuples of the form $(q \subset Q,\sigma \subset \Sigma,tpat, p)$
The transition rule tuples specify the state that a given node must be in, and the symbol from the input language that the subtree must have, (state $q$ and symbol $\sigma$, respectively), in order for this rule to match. They also specify a tree *pattern* that forms the right-hand side of the rule, and a weight $p$ for this rule. The tree pattern is a tree where some of the elements in the tree may be variables, which refer to subtrees of the left-hand side under consideration.
xT Transducers
--------------
The “extended" variation of T transducers, indicated with an “x" prefix, adds the capability for rules to check whether a potentially matching subtree matches a certain pattern of finite size, in addition to the given state and value of the node. The tree pattern in the left-hand side of an xT transduction rule may contain literal symbols as well as variables, which allows for lexicalized rules that only apply when certain words are in a subtree. The tree patterns also make it possible for the rules to reference material finitely far into a subtree, which makes local rotations straightforward; see Figure \[missesanybody\] for example xT rules that perform a local rotation and also use finite lookahead to produce Francophone names. In the notation common in the literature, a state for a node is written next to that node in the tree structure.
1\. $\longrightarrow$
2\. $\longrightarrow$
3\. $\longrightarrow$
While T transducers are not as expressive as synchronous TSG [@Shieber:2004:SGT], xT transducers are as expressive, and can even be used to simulate synchronous TAG in some cases [@maletti:2010:ACL]. In addition to their formal expressive power, xT transducers are much more convenient for rule authors; some finite lookahead can be simulated with the standard T transducers, as shown in [@DBLP:journals/coling/GraehlKM08], but it is somewhat tedious. The use of xT transducers makes writing rules to rearrange material in a tree much more convenient.
Restricted Versions of T and xT Transducers
-------------------------------------------
For computational efficiency purposes, we may also consider placing certain restrictions on the rules in a T or xT transducer. Options that have been explored include requiring that a transducer be *linear*, which means that any variable occurring in the left-hand side of a rule should appear no more than once in the right-hand side, and *nondeleting*, which means that a variable in the left-hand side must appear at least once in the right-hand side. Linear transducers are given the prefix “L", and nondeleting transducers the prefix “N", so for example, extended linear non-deleting top-down transducers are described as “xLNT". This particular combination of options has been used several times in the literature, including [@galley-EtAl:2004:HLTNAACL]. Also note that the transducer in Figure \[exampletransduction\] is not nondeleting, since Rule 2 does not reference its variables in its right-hand side.
Among the benefits of adding these constraints on rules are that, LT and LNT transducers are *compositional*, meaning that a relation that can be expressed by a cascade of two LT transducers can also be expressed by a single LT transducer, and that the composition of those two transducers can be computed. However this is not possible with any other members of the T-transducer family; even xLNT transducers are non-compositional [@DBLP:journals/mt/Knight07].
Linguistic Divergences
======================
Bonnie Dorr, in [@DBLP:journals/coling/Dorr94], enumerates several different kinds of structural divergences that we might see in translation between languages. These divergences occur when translating from English to closely related languages, Spanish and German, all of which have fairly similar word orders. These are not the only kinds of syntactic differences that there can be in a translation. They do not, for example, cover the more large-scale reorderings that we see when translating between SVO and SOV or VSO languages. However, each of these divergences require something more than simple word substitution or reordering the children of a given node: many of these require raising and lowering tree material (performing “rotations", in the terminology of [@Shieber:2004:SGT]), and nested phrases that are present in one language are often not present in the other. Many of these divergences may appear in a given pair of translated sentences. The following subsections describe Dorr’s seven kinds of divergence.
Thematic Divergence
-------------------
Different languages may express a situation by assigning different thematic roles to the participants of an action, swapping (for example) the subject and object. For example, translating from English to Spanish, we see:
- I like Mary
- María me gusta a mí
In Spanish it is more common to say that “X pleases Y“ than that ”Y wants/likes X". The Spanish verb *querer* has the same structure as the English “like", but the meaning of “gustar" is closer to the English “to like".
Promotional Divergence
----------------------
A modifier in one language may be the head in another language.
- John usually goes home
- Juan suele ir a casa
Here in English, an adverb modifies the verb to indicate that it is habitual, but in Spanish we use the verb “soler" (which inflects as “suele" for third-person singular), to express this. It has an infinitive as a dependent.
Demotional Divergence
---------------------
The demotional divergence is similar to a demotional divergence, viewed in the other direction; in cases of demotional divergence, a head in one language is a modifier in the other. In [@DBLP:journals/coling/Dorr94], a formal distinction is made between the two because in Dorr’s MT system, they would be triggered in different circumstances, but for our purposes they are effectively analogous.
- I like eating
- Ich esse gern
In this example, while English uses the verb “to like", German has an adverb. The sentence has a literal translation of “I eat likingly".
Categorial Divergence
---------------------
In cases of categorial divergence, the meaning of a word with a certain part of speech in one language is expressed with a different part of speech in the other.
- I am hungry
- Ich habe Hunger.
The German sentence here translates literally as “I have hunger."
Structural Divergence
---------------------
In cases of structural divergence, there are phrases in one language not present in the other.
- John entered the house
- Juan entró en la casa
While the English sentence has the destination of the motion verb as an object, in the Spanish we see the prepositional phrase “en la casa" (“in the house").
Lexical Divergence
------------------
In cases of lexical divergence, the two languages involved have different idiomatic phrases for describing a situation.
- John broke into the room
- Juan forzó la entrada al cuarto
While “break into" is a phrasal verb in English, in Spanish it is more idiomatic to “force entry to". This example also includes a structural divergence, as “al cuarto" is a prepositional phrase not present in the English.
Conflational Divergence
-----------------------
The meaning of the sentence may be distributed to different words in a different language; the meaning of a verb, for example, may be carried by a verb and its object after translation.
- I stabbed John
- Yo le di puñaladas a Juan
Here the Spanish sentence means literally “I gave John knife wounds". The words “le" and “a" are both required, but for different reasons: the verb “dar" (to give) requires the personal pronoun beforehand, and whenever a human being is the object of a verb in Spanish, we add the “personal a" beforehand.
Implementation
==============
In the course of this project, I have produced a small, easily-understandable toolkit named `kurt` (the Keen Utility for Rewriting Trees), for experimenting with weighted xT tree transducers. It is implemented in Python 3 and makes use of the NLTK tree libraries [@nltkbook]. `kurt` has been released as free software, and is available online [^1].
The software can perform tree transduction in general for weighted xT transducers: given a tree, it applies xT transduction rules and produces a list of output trees. The implementation is fairly naïve, and proceeds as a simple production system. Partial solutions are matched against every rule in the transducer, then each matching rule is applied to the partial solution, producing a new generation of partial solutions. Eventually, the derivation either succeeds by producing at least one tree with no nodes in a state, or it fails if the input tree cannot be completely transduced by the given rules. The system returns all possible output trees, and the complete solutions are printed out at the conclusion of the program.
The xT rules are straightforward to write, and are stored in YAML files. I have also provided example xT rules that translate the examples of divergences given by Dorr; these are described in more detail in Section \[sec:dorr-transducers\].
A complete and useful MT system based on this software – such that the rules and their weights were not completely the product of human knowledge engineering – would require the implementation of a few more algorithms described in [@DBLP:journals/coling/GraehlKM08], particularly their EM training algorithm to calculate weights for a given set of transduction rules, which depends on their transduction algorithm that produces the more compact representation of a transduction, a RTG. Decoding would require beam search over tree transduction, or perhaps over generation using this compact RTG representation. Additionally, some clever algorithm for extracting tree transducer rules from parallel treebanks would be useful for the case where parallel treebanks are available; some candidate techniques for this last problem are discussed in Section \[sec:extraction\].
Using the Software
------------------
Transducers are stored in YAML files, with one xT transducer per file; each rule is specified as an entry in that YAML file, and contains the following entries.
- `state`: (required) The name of the state that a node at the root of a subtree must be in to match this rule
- `lhs`: (required) The left-hand side of the rule: a tree pattern, typically with variables (tokens starting with `?`) that must unify with a subtree in order for that subtree to match this rule
- `rhs`: (required) The right-hand side of the rule: another tree pattern, which is filled in when this rule is applied. It may contain variables, in which case all of the variables must also be present in the left-hand side of the rule.
- `newstates`: (optional) Specifies the locations of transduction states in the subtree produced by this rule. There may be many states specified in the new subtree. They are given in the form `[location, statename]`, where location is a bracketed list that describes the path down the tree from the root of the subtree, with 0-indexed children. For example, to put the second child of the leftmost child of the root in state `foo`, a rule would have a `newstates` member `[[0,1], foo]`.
- `weight`: (optional) The weight for this rule. If unspecified, it defaults to 1.0.
Given a file with these entries for each rule of a transducer, say called `translation.yaml`, a Python 3 program can use `kurt` to do tree transductions in the following way, assuming the libraries are all in the `$PYTHONPATH` or the current working directory.
from loadrules import loadrules
from translate import translate
rules = loadrules("translation.yaml")
tr = Tree("""(S (NP (PRP I))
(VP (VB am)
(JJ hungry)))""")
## print all valid transductions
translate(tr, rules)
Simple Topicalization Example
-----------------------------
In Figure \[topicalization\], we see a toy example of xT rules realized with the system. This is a complete running example that exercises many features of the software; it translates an English sentence into “LOLcat" Internet slang, which features more prominent topicalization [^2]. For simplicity, the syntactic structure of the parse tree is elided. The initial rule matches a sentence in the initial state `q`, containing “let me show you my $x_0$" and produces a new sentence where “my $x_0$" has been moved to the front . The rule also specifies that the (0-indexed) child of the S node at index $1$ is in the state `respell`. The second rule matches the word “Pokémon" when it is in the state `respell`, replacing it with the slang spelling of “Pokemans". The third rule is for generalization, allowing words other than “Pokémon" to be translated in this position. Due to both the second and third rules applying to the subtree, both spellings are produced in the output, but the translation with the slang spelling is given a higher weight.
## lolcat topicalization (fronting)
- state: q
lhs: (S let me show you my ?x0)
rhs: (S my ?x0 , let me show you them)
newstates:
- [[1], respell]
- state: respell
lhs: Pokémon
rhs: Pokemans
weight: 0.9
- state: respell
lhs: ?x0
rhs: ?x0
weight: 0.1
xT Transducers for Linguistic Divergences {#sec:dorr-transducers}
=========================================
I wrote xT transduction rules for the software toolkit that handle each of Dorr’s divergence examples. Most of the work involved was constructing parse trees for the source- and target-language sentences; I then converted the trees into templates for the desired trees, at which point they were effectively xT transduction rules. Some examples are included in Figures \[translationGerman\] and \[translationSpanish\], but the complete set of rules are in `german.yaml` and `spanish.yaml`, included with the software. Most of the transformations required to implement these rules are instances of local rotations, as described by [@Shieber:2004:SGT].
# handle <pronoun> like <gerund>
- state: q
lhs: (S (NP ?x0) (VP (VB like) ?x1))
rhs: (S (NP ?x0) (VP ?x1 (RB gern)))
newstates:
- [[0, 0], lookup]
- [[1, 0], gerundtotensed]
# handle I am <adj>
- state: q
lhs: (S (NP ?x0) (VP (VB am) ?x1))
rhs: (S (NP ?x0) (VP (VB habe) ?x1))
newstates:
- [[0, 0], lookup]
- [[1, 1], adjtonoun]
## simple lookups for known phrases
- state: lookup
lhs: (PRP I)
rhs: (PRP ich)
## POS changes.
- state: gerundtotensed
lhs: (VBG eating)
rhs: (VB esse)
- state: adjtonoun
lhs: (JJ hungry)
rhs: (NN hunger)
# handle <pronoun> like <name>
- state: q
lhs: (S (NP ?x0) (VP (VB like) (NP ?x1)))
rhs: (S (NP ?x1) (VP (NP ?x0) (VB gusta) ?x0))
newstates:
- [[0, 0], lookup]
- [[1, 0, 0], objectivize]
- [[1, 2], tothisperson]
- state: tothisperson
lhs: (PRP I)
rhs: (PP (A a) (PRP mí))
# handle usually -> soler
- state: q
lhs: (S (NP ?x0) (VP (RB usually) ?x1))
rhs: (S (NP ?x0) (VP (VBZ suele) ?x1))
newstates:
- [[0,0], lookup]
- [[1,1], unconjugate]
# handle entered-object -> entró en ...
- state: q
lhs: (S (NP ?x0) (VP (VBD entered) ?x1))
rhs: (S (NP ?x0) (VP (VBD entró) (PP (IN en) ?x1)))
newstates:
- [[0,0], lookup]
- [[1,1,1], lookup]
# handle broke-into X -> forzó la entrada a X
- state: q
lhs: (S (NP ?x0) (VP (VBD broke) (PP (IN into) ?x1)))
rhs: (S (NP ?x0) (VP (VBD forzó)
(NP (DT la) (NN entrada) (PP (IN a) ?x1))))
newstates:
- [[0,0], lookup]
- [[1,1,2,1], lookup]
- [[1,1,2], al]
# handle I stabbed X -> le di puńaladas a
- state: q
lhs: (S (NP ?x0) (VP (VBD stabbed) ?x1))
rhs: (S (NP ?x0) (VP (PRP le) (VBD di)
(NP (NN puñaladas)) (NP (A a) (NNP Juan))))
newstates:
- [[0,0], lookup]
Related Work {#sec:relatedwork}
============
In addition to the work on tree-based MT, some very sophisticated string-based MT algorithms have been framed in terms of finite-state transducers. Not long after the introduction of modern word-based SMT, Knight and Al-Onaizan showed that IBM Model 3 could be expressed with a cascade of FSTs [@DBLP:conf/amta/KnightA98]. Since string transducers can be composed, decoding in this case becomes one enormous beam search over a single state machine. Similarly, Shankar Kumar and William Byrne expressed the phrase-based alignment template model as FSTs [@DBLP:conf/naacl/KumarB03]. The last part of the decoding process in Chiang’s hierarchical phrase-based model can also be described in terms of FSTs [@iglesias-EtAl:2009:NAACLHLT09]; Iglesias et al. use finite-state techniques to traverse a lattice of possible translations once chart parsing with an SCFG has completed.
For tutorials and related algorithms, Chiang provides an excellent introduction to synchronous grammars in [@Chiang06anintroduction]. My understanding of TAG was greatly aided by the TAG section in [@vannoord93]; it is referenced in the TAG Wikipedia page. For overviews of different applications of T-family tree transducers and their various properties, in a very approachable style, [@DBLP:journals/mt/Knight07] and [@KnightGraehlOverview] are very helpful. Additionally [@DBLP:journals/coling/GraehlKM08] contains excellent examples for understanding xT transduction (one of which is in this paper in simplified form, though the original example is worth working through and understanding fully), along with a set of algorithms that can be computed over xT transducers, including an EM procedure that can be used to estimate the weights for an xT grammar given a parallel treebank.
Conclusions and Future Work
===========================
Here I have described the “T" family of tree transducers and situated them among the various formalisms for describing relations over strings and trees; I have also demonstrated that xT transducers are sufficient for handling translation across the linguistic divergences described by Dorr. I have presented a software package suitable for experimentation with xT transducers, which comes with example translation rules that perform translations over each of the divergences.
There remains significant work to be done on the topic; for example, to my knowledge, there is no easily available end-to-end MT system based on tree transducers, either commercial or Open Source. There are many more questions that I would like to answer; as far as I know, these are open problems in the field.
Transducers, Disambiguation, and Language Models
------------------------------------------------
While weighted synchronous grammars and xT transducers provide generative models of translation, the probabilities that they assign to a given rule are set ahead of time, and are not conditioned on features of the surrounding context. It may be fruitful to try using discriminative approaches (i.e., classifiers) to help a transducer-based MT system make decisions about which rules are the most likely to apply in a given context, either based on the surrounding tree material, or on the surface words in the source-language sentence. It may turn out that there is a more principled way to achieve the same benefits, perhaps by adding more conditions on the probabilities in a generative model. However, cross-language phrase-sense disambiguation with classifiers, like in the work of Carpuat and Wu [@carpuatpsd], has proved useful for phrase-based SMT. For phrase-based SMT in general, discriminative approaches such as Minimum Error-Rate Training (MERT) [@DBLP:conf/acl/Och03] have become quite typical.
Another guide for the tree transduction process could be language models, either flat n-gram models or structured ones, which would have the added benefit that they could be trained on larger corpora than those used to produce the tree transduction rules in the first place.
Extraction and Training Transducers {#sec:extraction}
-----------------------------------
Thus far, it seems as though there is no agreed-upon best approach for extracting a set of tree transduction rules from a parallel treebank, such that a tree-to-tree MT system could be constructed. While parallel treebanks are not abundant, with sufficiently good monolingual parsers, parallel trees can be created from bitext, and hopefully these could be used to induce transduction rules for tree-to-tree MT systems. Other work has presented methods for learning tree-to-string transduction rules, for example [@galley-EtAl:2004:HLTNAACL] and [@deneefe-knight:2009:EMNLP]. These approaches for learning tree-to-string transducers, if I understood them more completely, might turn out to generalize easily to the tree-to-tree case, but if so, it is not yet obvious to me how to do this.
One proposed approach for learning relations over trees is given in [@eisner:2003:ACL-companion], in which Eisner presents algorithms for both extracting an STSG grammar and training its weights; STSGs can then be expressed as xT transducers as described by Maletti in [@maletti:2010:ACL]. Additionally, approaches for leaning tree transduction rules have been suggested for tasks other than machine translation, particularly in the summarization work of Cohn and Lapata [@cohn-lapata:2007:EMNLP-CoNLL2007], [@cohn-lapata:2008:PAPERS], who work with a corpus that not only has parse trees for both source and target languages (in their case, pairs of longer and paraphrased sentences, both in English), but has also been word-aligned. The word alignments inform their grammar extraction. Cohn and Lapata use a very small training paraphrase corpus (480 sentences), which suggests that perhaps their methods would be useful for MT with low-resourced languages. They also use of discriminative methods for training and decoding. Both their algorithm for rule extraction and the tree transducers with discriminative methods may have been used in tree-to-tree MT system, but I have not yet found work that describes this; if it has not yet been tried, someone should explore it.
XDG as Transducers
------------------
Given that many grammar formalisms are expressible in terms of tree transducers, one wonders if constraint-based dependency frameworks, such as Extensible Dependency Grammar [@Debusmann06], which has been used by Michael Gasser for machine translation [@gasser:2011:freerbmt], could be expressed in terms of tree transducers. Transducers over dependency trees have already been used for machine translation, for example by Ding and Palmer [@ding-palmer:2005:ACL]. However, XDG defines not just one layer of dependency analysis for a language, but several. Its analysis of a sentence in a given language is a multigraph with multiple dimensions of analysis, with constraints describing permissible structures on each dimension, as well as the relationships between dimensions. This suggests that perhaps XDG could be expressed as a cascade of transducers, with each layer in the cascade describing the relation between one XDG dimension and the next.
A problem with this interpretation is that not all layers of an XDG multigraph are tree structures. This might mean that XDG cannot be cleanly expressed in this way at all, or perhaps that another kind of transducer that operates on graphs more generally could be used. Alternatively, perhaps XDG could be tweaked such that every layer has a tree structure.
If it is in fact possible to express XDG translation rules as a cascade of transducers, then this would present a clear path for integrating machine learning into the largely rule-based system, making use of the training algorithms already present in the literature. As a fairly modest step, given small numbers of parallel training sentences, one could use EM to train the weights of the transduction rules that implement the XDG grammar. More ambitiously, one could perhaps extract grammar rules from example translation pairs, although the XDG parse graphs would have to be provided by an expert, for each layer in the analysis. This could be done either simply on demand, when the existing grammar fails to parse and translate a sentence, or using active learning to select sentences for human annotation.
One problem not addressed at all in the literature that I have seen is how to translate, either into or out of, morphologically rich languages using tree transducers. It seems as though morphological analysis and lemmatization would be an important first step in a transducer-based MT system, to limit the number of rules that the system needs to consider, but then the morphological information should be used to help the system make choices during transduction (decoding). Perhaps morphological features would be useful to classifiers trained to help make syntactic disambiguation decisions.
[^1]: http://github.com/alexrudnick/kurt
[^2]: Readers may or may not be familiar with the moderately popular catchphrase “My Pokemans, let me show you them".
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Zecheng Zhang, Xiaoxiao Wu, Naijing Zhang, Siyuan Zhang, and Edgar Solomonik[^1]\
Department of Computer Science\
University of Illinois at Urbana-Champaign
bibliography:
- 'paper.bib'
title: 'Enabling Distributed-Memory Tensor Completion in Python using New Sparse Tensor Kernels'
---
Introduction {#sec:intro}
============
Tensor Completion {#sec:cmpl}
=================
New Sparse Tensor Kernels {#sec:krnl}
=========================
Python Interface and Implementation {#sec:prog}
===================================
Experimental Evaluation {#sec:exp}
=======================
Related Work {#sec:rel}
============
Conclusion {#sec:cnc}
==========
Acknowledgements {#sec:ack}
================
[^1]: Contact via [email protected].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Based on BHHRS [@BHHRS] and further sharpened by discussions with Gerry Brown that I had in October 2008, we [@footnote-we-i] arrive at the present assessment of the dilepton saga, namely, that dileptons become “blind" to changes in the vacuum structure of chiral symmetry (such as, e.g., BR scaling [@BR91]) at high temperature and/or at high density and hence are [*not*]{} an appropriate probe for a signal for partial or complete chiral restoration, contrary to what has been widely believed. There, however, are a variety of indirect indications that the scaling notion is qualitatively, if not quantitatively, valid and should work in various low-energy nuclear phenomena, and it is fair to conclude that while there is no direct evidence for the scaling notion, there is [*none against*]{} it either, in disagreement with the claim made in [@cern; @green]. I will touch briefly on certain observables that could give a clear-cut litmus signal for the vacuum structure of chiral symmetry modified by temperature and/or density.'
author:
- Mannque Rho
title: 'Dileptons Get Nearly “Blind" to Mass-Scaling Effects In Hot and/or Dense Matter'
---
[*$\bullet$ The upshot*—]{} There seems to be a general consensus in the heavy-ion community [@cern] (1) that the dilepton spectra of the recent refined measurements as NA60 and others can be described more or less fully by [*conventional hadronic many-body effects*]{} alone [@footnote1], and (2) that these results imply there can be little, if any, “mass-shift" of hadrons caused by the change of vacuum structure of chiral symmetry (CVCS for short) in hot and/or dense matter. We agree with the first statement but disagree strongly with the second. Our assertion is that the dileptons measured in the experiments performed so far are [*nearly blind*]{} to those vector mesons which carry information on chiral symmetry properties of hot and/or dense medium. I would like to clarify what our (Gerry’s and my) points are as they stand now.
As in BHHRS [@footnote2], we suggest that dileptons are [*strongly*]{}, though perhaps not completely, suppressed in the temperature regime defined as “hadronic freedom" (HF for short) between the chiral transition temperature $T_c$ and the flash temperature $T_{flash}$ in which the CVCS effect should be discernable and hence whatever carries information on chiral symmetry is swamped by the mundane effects that I shall simply refer to as “garbages" [@footnote2p]. So within the largely uncontrollable uncertainty in theory as well as in dealing with experimental data (such as evolution code, “cocktails" etc), the dilepton measurements so far performed could not properly single out the CVCS that one is looking for. It would be grossly unjustified if not totally wrong to conclude that the CVCS was absent in the measured process.
[*$\bullet$ CVCS in nuclei*—]{} There are a variety of evidences in nuclear physics that the CVCS is present and operative. But they are all indirect. In some sense, the situation is very similar to meson-exchange currents in nuclei or at the most fundamental level, to the Lamb shift in QED. In order to “see" the effect in question, all possible mundane nuclear effects that are not [*directly*]{} [@footnote2pp] connected to the CVCS (i.e., “garbages") have to be accurately identified and subtracted out. In the case of meson-exchange currents, this required that all possible nucleonic effects involving many-body dynamics be accurately calculated within one well-defined and consistent theoretical scheme (e.g., chiral perturbation theory) before meson-exchange effects could be singled out of the same calculation [@MR]. It took many decades since Yukawa’s work on meson theory for this effect to be confirmed.
Let me just cite a couple of recent cases to illustrate my point. Other examples are given in review articles by Gerry and me [@BR-PR1; @BR-PR2; @BR-PR3] and also in my recent book [@MR-book].
1. The C-14 dating: This is one of the most remarkable cases where a simple or even what I would consider a “naive" accounting of BR scaling is revealed rather spectacularly. It comes in the tensor force contributed by the $\rho$ exchange between two nucleons in nuclear matter. That nuclear tensor forces get a substantial suppression in nuclear matter was pointed out in 1990 [@BR-tensor]. The mechanism is primarily that the $\rho$ tensor force which comes with the sign opposite to that of the pion tensor force gets enhanced effectively by the factor $(g^*/m_\rho^*)^2$ (here $g$ is the $\rho NN$ coupling related to hidden gauge coupling in hidden local symmetry (HLS) [@HY]) where the asterisk denotes in-medium quantity. Now it has been established that $g^*\simeq g$ up to $n\sim n_0$ where $n_0$ is the normal nuclear matter density, whereas $m_\rho^*/m_\rho\simeq f_\pi^*/f_\pi\approx 1-\kappa (n/n_0)$ where $\kappa$ can be somewhere between 0.1 and 0.2. This effect is used by Holt et al [@holt1] to explain the striking effect in the C-14 dating which exploits that the process occurs below but very near the nuclear matter density $n_0$. Interestingly it has been suggested that the C-14 dating could also be explained by three-body forces [@holt2]. I will suggest below that these mechanisms are related.
2. The nuclear symmetry energy $S$ crucial for neutron star structure: It has been suggested that the symmetry energy that fits the $\pi^-/\pi^+$ ratio of FOPI/GSI [@FOPI] could be understood in terms of the same BR-scaled $\rho$ tensor force that figures in the C-14 dating [@BAL]. To be certain, it would require further confirmation, both experimental and theoretical, that the present FOPI/GSI does indeed constrain $S$ in the way claimed in [@BAL1]. The verdict will presumably come from FAIR/GSI or even from the RIB accelerators in construction [@BAL-private]. But let me assume that it’s OK, focusing on qualitative aspects involved. Remarkably, as suggested in [@BAL], it can also be explained by the three-body forces of the sort that figure in the C-14 dating. The FOPI data appear to constrain $S$ only up to about $\sim$ 1.3$n_0$, and beyond that density, nothing is precisely known. Now in Ref.[@BAL1], the authors assume that $(g^*/m_\rho^*)^2\simeq (g/m_\rho^*)^2$ for [*all*]{} $n$ beyond $n_0$. We have suggested – based on hidden local symmetry with vector manifestation (HLS/VM for short) [@footnote1] – that $(g^*/m_\rho^*)\approx {\rm const.}$ for $n > n_{flash}$ where $n_{flash}$ density must be a bit above the maximum density measured by the FOPI data. In terms of three-body forces, this feature would correspond to the [*suppression*]{} of three-body force effects at some density above $\sim 1.3n_0$. This means that the analysis made in [@BAL1] could very well be invalidated by this HLS/VM effect. My guess is that the non-Newtonian gravity effect considered in [@BAL1] will not be indispensable for reconciling the symmetry energy $S$ with neutron-star data once HLS/VM scaling is taken into account.
[*$\bullet$ BR scaling vs. three-body forces*—]{} At this point, let me make a conjecture. It’s only a conjecture for the moment but it needs to be made more rigorous.
The conjecture is that doing chiral perturbation calculation including up to three-body forces with a chiral Lagrangian whose parameters are determined in the vacuum is equivalent – in some given channels – to doing chiral perturbation calculation [*without*]{} three-body forces with the same chiral Lagrangian whose parameters are determined at the Landau Fermi liquid fixed point, i.e., BR scaling fit to the pion decay constant at $n\simeq n_0$. Song [@song] has shown how this connection via what’s called “rearrangement terms" is required for thermodynamic consistency of many-body dynamics. Now what about for densities $n > n_0$? I would suggest that the HLS/VM effect discussed above in the two-body case, that is, $m_\rho^*$ scaling along with $g^*$, captures the mechanism of the three-body forces getting [*suppressed*]{} at high density. One could see this in terms of $\omega$-exchange three-body forces accounting for the contact three-body term with the coefficient $C_E$ in [@holt1]. There is a similar 5D dual interpretation of this suppression in holographic QCD [@hashimoto]. It is shown there that at short distances, one can use flat geometry in 5D in which case the ADHM multi-instanton solution is directly applicable for many-nucleon dynamics. One can therefore analytically calculate N-body forces and show that they are suppressed as $V_N \sim {\cal O}(N_c/\lambda^{N-1})$ where $\lambda$ is the ’t Hooft constant. Since in hadronic world, $N_c/\lambda\sim {\cal O} (10^{-1})$, N-body forces will be strongly suppressed [*at short distances*]{}. I think this can be translated into the VM in 4D.
[*$\bullet$ CVCS in dilepton production*—]{} Let me consider the following HLS scenario. Consider an HLS theory in which baryons are introduced via topology, i.e., skyrmions, and scalars are introduced via the trace anomaly as recently discussed in [@HKL-MR], namely in terms of the soft dilaton $\chi_s$. Let me endow the parameters of this Lagrangian, which I will just call ${\cal L}_{HLS}^\prime$, with a suitable scaling (like BR scaling for the mass and coupling as alluded above). Now compute the $\rho$ spectral function with this Lagrangian taking into account thermal and dense loops including all relevant degrees of freedom. In computing this spectral function, I would have to do the thermal evolution from, say, $T_c$, to $T_{flash}$ at which $\sim 90\%$ on-shell masses are recovered, and the strong interactions recover nearly full on-shell strengths. Let’s follow the temperature evolution going down from $T_c$. I will not consider what happens above $T_c$.
1. From $T_c$ to $T_{flash}$ (“hadronic freedom" (HF) regime), there will be BR-scaled sharp low-mass dileptons as dictated by HLS/VM, but we argue that they are [*strongly*]{} suppressed by HLS/VM as in BHHRS. [@footnote4]
2. At $T_{flash}$, all hadrons entering in the process, as discussed in connection with the STAR $\rho^0/\pi^-$ ratio [@rho-pi], go on-shell with their strong coupling strengths recovered. All the variety of different nuclear processes, such as couplings to sobars, collisional broadening etc. taken into account in various different conventional many-body hadronic approaches (generically referred to as $nonVM$ [@footnote1]) are essentially getting the same thing since the parameters of all the $nonVM$ models, though different in specific contents, are [*fit*]{} to on-shell data. We don’t have our own numerical codes constructed [*in consistency*]{} with ${\cal L}_{HLS}^\prime$ etc, so cannot compute numerically, and compare with, what the experimentalists call the $\rho$ spectral function. However if our theory ${\cal L}_{HLS}^\prime$ is to give on-shell quantities correctly – which it certainly should [@HY], then it must describe correctly the spectral function [*appropriate for the flash point*]{}. I don’t see how HLS/VM can go wrong on-shell including many-body processes. We in fact looked at the NA60 spectrum – modulo absolute normalization – predicted by ${\cal L}_{HLS}^\prime$ and we were satisfied that the $\rho$ spectrum, including the shoulder below the $\rho$ peak, can be understood entirely (at least qualitatively) by mundane processes taking place on-shell at and below $T_{flash}$ assuming that the dileptons coming from the $\rho$ mesons produced in the HF regime are highly suppressed. Furthermore, whatever dileptons that may have been produced from the $\rho$ mesons in the HF regime (hence BR scaled) will appear below the $\rho$ peak, but, highly suppressed, they will be swamped by the on-shell production (involving width broadened by mundane nuclear effects, p-wave pion-pion scattering etc.) and will be buried in the error bars there. They will indeed be “needles" in the haystack as we have been saying all along.
What could make a $nonVM$ calculation numerically different from what we will get from our scenario is, however, that in the $nonVM$ scenario, dileptons are produced vector-dominated with on-shell conditions [*throughout*]{} the evolution, [*including the narrow HF regime.*]{} From $T_{flash}$ downwards, there will be no essential differences between the two theories since it’s the on-shell action that will figure for which our model should give, when fully calculated, the same result as any good phenomenological model which is fit to data in the matter-free vacuum would do.
So where would the difference show up? The answer is that it will be in the absolute normalization.
[*If it can be shown without any doubt that the missing strength between the temperature range between $T_c$ and $T_{flash}$ implied by HLS/VM is inconsistent with the experimental absolute normalization, then that will be a clear falsification of the HLS/VM scenario*]{}: In HLS/VM, the dilepton yield from the $\rho$ within the HF regime will be mostly missing due to the suppression in that regime.
However in the presence of so many free parameters and fudge factors in the phenomenological fitting, it is highly doubtful that whatever differences there may be in the absolute normalization due to what happens in a narrow temperature regime where CVCS figures – which is expected to be insignificant – could be discerned. In this sense, there cannot be any noticeable differences between what the $nonVM$-type calculation gets and what we predict with a consistent ${\cal L}_{HLS}^\prime$ theory. Thus as it stands, [*the dileptons measured so far can say nothing about CVCS!!*]{}
[*$\bullet$ “Seeing" BR scaling*—]{} What can one do with the measured dilepton yields to “see" the effect of CVCS?
Here is one way. Subtract away all “garbages" from the dilepton yields: That is, subtract not only all the usual cocktails but also all dileptons emitted from “on-shell" vector mesons. For the latter, both HLS/VM with ${\cal L}^\prime_{HLS}$ and realistic phenomenological models – there may be more than one in the literature – in the $nonVM$ class will do equally well. What result will then be (a) the dileptons directly coupled to pions that populate the HF regime, this coupling being present due to the strong in-medium violation of vector dominance in HLS/VM, plus (b) those dileptons with (suppressed) coupling to the sharp BR-scaled $\rho$ mesons expected in HLS/VM, namely, those “needles" in the haystack with the haystack removed. The dilepton coupling to pions in the HF regime will be point-like, so will be structureless. This feature will be totally absent in the $nonVM$ scenario where HLS/VM or BR scaling is missing. Perhaps this feat is much too daunting a task to perform.
[*$\bullet$ “Smoking gun" signal*—]{} The argument that dileptons are “blind" to BR-scaling vector mesons applies equally to dense medium. In fact, as stressed elsewhere [@BR-PR2; @BR-PR3], the HF regime sets in a lot more precociously in dense matter than in hot medium. This is already manifest in that vector dominance in the sense of Sakurai’s is badly violated in the nucleon form factor already in matter free space. It fact the photon couples to nucleon roughly half of the time via vector mesons and the other half directly. In holographic QCD, this is understood by the fact that the nucleon form factor is vector-dominated by the [*infinite tower*]{} of vector mesons – and not by the lowest members – and it is thus a bad approximation for nucleon form factors to couple the photon 100% to the lowest vector mesons $\rho$ and $\omega$ [@HRYY]. This will be more so in dense matter: Density will make dileptons [*even more blind*]{} to BR-scaling vector mesons than temperature does. Thus the CEBAF’s negative result is of no great surprise.
But then is there any process that can show a clear evidence for HLS/VM scaling?
Here is one smoking-gun evidence for or against HLS/VM. Measure the pion velocity $v_\pi$ at or near $T_c$. If there is no massless or no near massless $\rho$ meson, then one should find $v_\pi\approx 0$ as predicted by Son and Stephanov [@son-stephanov]. $v_\pi\approx 0$ will kill HLS/VM and hence BR scaling. However if there is massless $\rho$ near $T_c$, then one should find $v_\pi\approx 1$ [@HKRS]. $v_\pi\approx 1$ will then be an unequivocal support for HLS/VM as well as for BR scaling. This is a night-and-day difference. There may be other signals and they should be looked for.
[*Acknowledgments*—]{} This work was supported by the WCU project of Korean Ministry of Education, Science and Technology (R33-2008-000-10087-0).
[99]{} G. E. Brown, M. Harada, J. W. Holt, M. Rho and C. Sasaki, “Hidden local feld theory and dileptons in relativistic heavy ion collisions,” Prog. Theor. Phys. [**121**]{} (2009) 1209 \[arXiv:0901.1513 \[hep-ph\]\]. In what follows, “we" will stand for Gerry Bown and myself. Otherwise “I" will be used throughout.
G. E. Brown and M. Rho, “Scaling effective Lagrangians in a dense medium,” Phys. Rev. Lett. [**66**]{} (1991) 2720. For an up-to-date account, see S. Damjanovic, R. Shahoyan and H.J. Specht, CERN Courier [**October 30**]{} (2009) 31. I am puzzled at how the green curve labeled “Brown-Rho" in Fig. 2 in [@cern] could possibly be associated with BR scaling. As far as I can see, it has [*absolutely nothing*]{} to do with the scaling we proposed in [@BR91], explained clearly in our subsequenct review artlces and given more precision in [@BHHRS].
The class of theoretical descriptions that are anchored on “conventional hadronic many-body interactions" [*essentially*]{} differ from our approach (referred to as HLS/VM) based on hidden local symmetry (HLS) field theory [@HY] – the only field theory so far known that provides a theoretical basis for Brown-Rho scaling – in that the former does not have the “vector manifestation" that the latter has near the chiral phase transition. The vector manifestation – which is the unique feature of the HLS theory – encodes matching to QCD with RG flow and is the only theory known that incorporates both symmetries and dynamics approrpriate for the vacuum structure of spontaneoulsy broken chiral symmetry and scale invariance near the critical point. I shall refer to the former approaches generically as “$nonVM$" and the latter as “HLS/VM."
M. Harada and K. Yamawaki, “Hidden local symmetry at loop: A new perspective of composite gauge boson and chiral phase transition,” Phys. Rept. [**381**]{} (2003) 1 \[arXiv:hep-ph/0302103\]. Here I focus on temperature effect but a similar argument holds for density. I will also deal with the chiral limit although symmetry breaking effects are not always ignorable.
Or broadly, part of the background to the phenomenon one is to zero in.
Of course [*all*]{} light-quark hadronic properties, be they elementary or many-body, are in a variety of ways connected to chiral symmetry. This is clear already in the intricate relation [@BR-PR2; @BR-PR3] between the anomalous nuclear orbital gyromagnetic ratio $\delta g_l$ and, via Landau Fermi-liquid theory, the BR scaling $\Phi\approx (m_V^*/m_V)$ and as suggested below, also between many-body forces and BR scaling etc.
M. Rho, “Exchange currents from chiral Lagrangians,” Phys. Rev. Lett. [**66**]{} (1991) 1275. G. E. Brown and M. Rho, “Chiral restoration in hot and/or dense matter,” Phys. Rept. [**269**]{} (1996) 333 \[arXiv:hep-ph/9504250\].
G. E. Brown and M. Rho, “On the manifestation of chiral symmetry in nuclei and dense nuclear matter,” Phys. Rept. [**363**]{} (2002) 85 \[arXiv:hep-ph/0103102\].
G. E. Brown and M. Rho, “Double decimation and sliding vacua in the nuclear many-body system,” Phys. Rept. [**396**]{} (2004) 1 \[arXiv:nucl-th/0305089\]; M. Rho, [*Chiral Nuclear Dynamics: From Quarks to Nuclei to Compact Stars*]{} (World Scientific Publishing, Singapore, 2008).
G.E. Brown and M. Rho, “In-medium stiffening of the nucleon nucleon spin-isospin interaction," [Phys. Lett.]{} [**B237**]{} (1990) 3.
J. W. Holt, G. E. Brown, T. T. S. Kuo, J. D. Holt and R. Machleidt, “Shell model description of the $^{14}$C dating beta decay with Brown-Rho-scaled NN interactions,” Phys. Rev. Lett. [**100**]{} (2008) 062501 \[arXiv:0710.0310 \[nucl-th\]\]. W. Reisdorf [*et al.*]{} \[FOPI Collaboration\], “Systematics of pion emission in heavy ion collisions in the 1A GeV regime,” Nucl. Phys. A [**781**]{} (2007) 459 \[arXiv:nucl-ex/0610025\].
J. W. Holt, N. Kaiser and W. Weise, “Chiral three-nucleon interaction and the carbon-14 dating beta decay,” Phys. Rev. C [**79**]{} (2009) 054331 \[arXiv:0901.4750 \[nucl-th\]\]. C. Xu and B. A. Li, “Why is the nuclear symmetry energy so uncertain at supra-saturation densities?,” arXiv:0910.4803 \[nucl-th\]. D. H. Wen, B. A. Li and L. W. Chen, “Super-soft symmetry energy encountering non-Newtonian gravity in neutron stars,” Phys. Rev. Lett. [**103**]{} (2009) 211102 \[arXiv:0908.1922 \[nucl-th\]\]. I am grateful for discussions on this issue with Byungsik Hong, Hyun Kyu Lee and Bao-An Li.
C. Song, “Dense nuclear matter: Landau Fermi-liquid theory and chiral Lagrangian with scaling,” Phys. Rept. [**347**]{} (2001) 289 \[arXiv:nucl-th/0006030\].
K. Hashimoto, N. Iizuka and T. Nakatsukasa, “N-body nuclear forces at short distances in holographic QCD,” arXiv:0911.1035 \[hep-th\]).
H. K. Lee and M. Rho, “Dilatons in hidden local symmetry for hadrons in dense matter,” Nucl. Phys. A [**829**]{} (2009) 76 \[arXiv:0902.3361 \[hep-ph\]\]. Note, however, that in this HF regime, due to VD violation, dileptons of various invariant masses will be produced by direct point-like coupling to pions which will not enter in the proper $\rho$ spectral function. In holographic dual description in 5D, the point-like coupling corresponds to the infinite tower above the ground state vector mesons. I will come back to this point below.
G. E. Brown, C. H. Lee and M. Rho, “Vector manifestation of hidden local symmetry, hadronic freedom, and the STAR $\rho^0 / \pi^-$ ratio,” Phys. Rev. C [**74**]{} (2006) 024906.
Given a consistent Lagrangian field theory as our ${\cal L}^\prime_{HLS}$, these could be calculated in a systematic and consistent way, at least in principle. Whatever theoretical ingredients enter in the variety of different calculations that have claimed to “explain" the experimental data, what matters is that they all are manufactured to fit on-shell experimental data in one way or other and therefore with a sufficient number of parameters to adjust, they all could, as they claim, describe what happens on-shell. The differences that may be present between different authors could be just variations in their fitting procedures with little significance. In this sense, our ${\cal L}^\prime_{HLS}$ should give more or less the [*same on-shell spectral function.*]{} D. K. Hong, M. Rho, H. U. Yee and P. Yi, “Nucleon form factors and hidden symmetry in holographic QCD,” Phys. Rev. D [**77**]{} (2008) 014030 \[arXiv:0710.4615 \[hep-ph\]\]. D.T. Son and M.A. Stephanov, “Pion propagation near the QCD chiral phase transition, [Phys. Rev. Lett.]{} [**88**]{} (2002) 202302.
M. Harada, Y. Kim, M. Rho and C. Sasaki, “The pion velocity at chiral restoration and the vector manifestation,” Nucl. Phys. A [**730**]{} (2004) 379 \[arXiv:hep-ph/0308237\].
| {
"pile_set_name": "ArXiv"
} |
---
address: 'DAMTP, Centre for Mathematical Sciences, Cambridge University, Wilberforce Rd., Cambridge CB3 0WA, UK '
author:
- 'John D. Barrow and Janna Levin'
title: A Test of a Test for Chaos
---
psfig.sty
Introduction
============
A promising new binary test for chaos has been proposed by Gottwald and Melbourne [@gm]. The test returns 0 for a non-chaotic system and 1 for a chaotic system. It does not offer any quantitative information and so cannot quantify the degree of irregularity in the way a Lyapunov exponent might. However, it can provide a quick and useful diagnostic for chaos.
Consider a base dynamical system ${\bf \dot{x}}={\bf f}({\bf x})$ of any dimension with orbits ${\bf x}(t)$. The proposed method to determine if it is chaotic is based on a Euclidean ${\bf E}(2)$ group extension of the underlying base dynamics . The dynamics is explicitly extended to include two new variables $(p,q)$ defined through $$\begin{aligned}
\dot{p} &=&\phi ({\bf x})\cos (\omega _{0}t), \nonumber \\
\dot{q} &=&\phi ({\bf x})\sin (\omega _{0}t),\end{aligned}$$ where $\phi ({\bf x})$ is any observable of the base dynamics and $\omega
_{0}\neq 0$ is an arbitrary constant frequency that is needed to damp off any linear growth that may be common to both non-chaotic and chaotic orbits. If the observable $\phi $ is drawn from a non-chaotic system, then the $(p,q)$ subspace will be bounded. But if the observable $\phi $ is drawn from a chaotic system, then the motion in the $(p,q)$ subspace will be Brownian and unbounded. The extended variables diffuse through the subspace as the observables of the base dynamics jump around unpredictably.
Gottwald and Melbourne introduced a binary test for chaos by defining a mean-square displacement $$M(t)=\lim_{T\rightarrow \infty }{\frac{1}{T}}\int_{0}^{T}\left( p(t+\tau
)-p(\tau )\right) ^{2}d\tau$$ and characterizing the behaviour of $M(t)$ through $$K=\lim_{t\rightarrow \infty }{\frac{\log M(t)}{\log t}}\ \ .$$ If there is no chaos then the motion in $p(t)$ is bounded and $K\rightarrow 0$. If the base dynamics is chaotic then $p(t)$ will exhibit Brownian diffusion so that $\Delta p\rightarrow t^{1/2}$ as $t\rightarrow \infty$ and $K\rightarrow 1$. The $K$-test can be used for both continuous and discrete dynamical systems even when the precise underlying base dynamics is unknown [@gm].
This test has certain advantages of simplicity and was shown to confirm the results of testing for chaos using Lyapunov exponents for the forced van-der-Pol system [@gm]. We show here that the $K$-test also confirms the trend of the Lyapunov exponents for the simple Hamiltonian system of Henon and Heiles [@hh]. So the $K$-test is applicable to Hamiltonian as well as dissipative systems.
While the test works well as a diagnostic of the transition from regularity to chaos we express some reservations about the use of the test on an orbit by orbit basis. Most importantly, we found the method difficult to interpret for systems that experienced chaotic transients in contrast to the easy interpretation of dissipative systems near a strange attractor, or in highly chaotic Hamiltonian systems. To illustrate this we use the Lorenz model [@loref] to show that transient chaos, which can be found in the Lorenz system by other means, goes undetected by the $K$-test. Orbits that enter highly chaotic regions of phase space and then depart into regular regions are not Anosov and may not show the diffusive motion in phase space required to yield a $K\rightarrow 1$ result from the test. These transient systems may not satisfy the criteria for a system within the remit of the $K$-test. This signals one of the possible limitations of the K-test.
Henon-Heiles
============
The Henon-Heiles system is an ideal Hamiltonian system on which to evaluate the utility of the $K$-test. It provides a simple and well understood model for the motion of stars in a galactic potential as well as for the motion of non-linearly-coupled molecules. The transition from regularity to chaos as the energy is increased is well documented and can be identified with Poincaré surfaces of section, Lyapunov exponents, and the Painlevé property.
The Henon-Heiles Hamiltonian is $$H={\frac{1}{2}}\left( p_{x}^{2}+p_{y}^{2}+x^{2}+y^{2}\right) +x^{2}y-
{\frac{1}{3}}y^{3}$$ with equations of motion $$\begin{aligned}
\dot{x} &=&p_{x} \nonumber \\
\dot{y} &=&p_{y} \nonumber \\
\dot{p}_{x} &=&-x-2xy \nonumber \\
\dot{p}_{y} &=&-y-x^{2}+y^{2}\ \ . \label{henon}\end{aligned}$$ We add the Euclidean extension $$\begin{aligned}
\dot{p} &=&x\cos (\theta ) \nonumber \\
\dot{q} &=&x\sin (\theta ) \nonumber \\
\dot{\theta} &=&\omega _{0} \label{e2}\end{aligned}$$ where we have chosen $\phi =x$. We evolve these equations numerically and check that the Hamiltonian remains conserved throughout the simulation. Our results are summarized in Figs. \[m\] and \[l\]. The orbits selected for display have identical initial conditions up to the energy $H$ which was varied from orbit to orbit. The value of $K(t)$ is measured as the slope of the line $ln(M(t))$ versus $ln(t)$ and is plotted in Fig. \[m\] as a function of the energy $H$. The transition from non-chaotic orbits ($K\rightarrow 0 $) to chaotic orbits ($K\rightarrow 1$) as the energy increases is clearly demonstrated. The transition confirms the break up of tori in the phase space for the base dynamics (and therefore the non-integrability of the base dynamics) as the energy grows. We also confirm the onset of chaotic motion for these orbits with a numerical determination of the principal Lyapunov exponent shown as a function of $H$ in Fig. \[l\].
A few comments on Fig. \[m\] should be made. The convergence of the value of $K$ towards either $0$ or $1$ can be improved if the numerical simulations are allowed to run far longer and a more precise extrapolation method is used so as to avoid any early transients. Of course, this renders the application more numerically intensive and slower without yielding a very great improvement in convergence. There is also a clear dependence of the motion in the Euclidean subspace on the specific value of $\omega _{0}$ employed. However, the overall trend in $K$ is typically the same for various values of $\omega _{0}$.
While Fig. \[m\] does show the required transition from 0 to 1, it is important to stress how the conservative system differs from the dissipative system. For a dissipative system all trajectories, regardless of initial conditions, are drawn onto the same attractor. Therefore the $K$-test should show no dependence on initial conditions. This is not true for a Hamiltonian system. There can be a mixture of regular and irregular orbits and so the test result can depend on initial conditions. Only for a completely ergodic system will $K\rightarrow 1$ independent of initial conditions. Also, since chaotic transients can often arise in conservative systems, the $K$-test may not always give such a crisp transition. Chaotic transients in dissipative systems are discussed further in the next section.
Lorenz
======
The standard Lorenz system demonstrates a transition from no chaos to transient chaos and then further to a full chaotic attractor. The standard system is $$\begin{aligned}
\dot{X} &=&-\sigma (X-Y), \nonumber \\
\dot{Y} &=&-XZ+rX-Y, \nonumber \\
\dot{Z} &=&XY-bZ,\end{aligned}$$ with $\sigma =10$ and $b=8/3$ and we vary the constant $r$. The transition to chaos is summarized in Ref. [@ott]. For $r<1$ there is one fixed point at $X=Y=Z=0$ and no chaos. For $1<r \mathrel{
\hbox to 0pt{\lower
3pt\hbox{$\mathchar"218$}\hss}\raise 2.0pt\hbox{$\mathchar"13C$}}13.96$ there are two additional attractors, one at $$\begin{aligned}
X_{R} &=&(b(r-1))^{1/2} \nonumber \\
Y_{R} &=&(b(r-1))^{1/2} \nonumber \\
Z_{R} &=&r-1,\end{aligned}$$ and another at $(X_{L},Y_{L},Z_{L})=(-X_{R},-Y_{R},Z_{R}),$ but still no chaos. A non-chaotic orbit drawn onto the simple fixed point $(X_{R},Y_{R},Z_{R})$ with $(\dot{X},\dot{Y},\dot{Z})\rightarrow 0$ is shown in Fig. \[simple\]. A closely neighbouring orbit will follow a similar path onto the same attractor since there is no exponential sensitivity to initial conditions, no positive Lyapunov exponent, and no chaos.
However, for values of $r\mathrel{
\hbox to 0pt{\lower
3pt\hbox{$\mathchar"218$}\hss}\raise 2.0pt\hbox{$\mathchar"13E$}}13.96$ a non-attracting form of transient chaos develops around a homoclinic orbit in phase space, first noted by Kaplan and Yorke [@ky]. Two orbits undergoing transient chaos are shown in Fig. \[transient\]. One orbit winds around a bit before being drawn onto the regular attracting point $(X_{L},Y_{L},Z_{L})$ while its very near neighbour ends up on $(X_{R},Y_{R},Z_{R})$.
We choose to use the method of fractal basin boundaries to locate regions of transient chaos. An initial slice in phase space is color-coded according to whether the orbit eventually lands on the fixed point $(X_{R},Y_{R},Z_{R})$ (black) or the fixed point $(X_{L},Y_{L},Z_{L})$ (white). In the absence of chaos all the basins will have smooth boundaries. In the presence of chaos the boundaries become fractal, demonstrating both extreme sensitivity to initial conditions and a chaotic mixing of orbits. Such fractal basin boundaries are illustrated in Fig. \[fbb\]; the orbits shown in Fig. \[transient\] are drawn from along the fractal. They have nearly identical initial values but divergent outcomes, a characteristic of chaotic transients.
5truept
5truept
5truept
For $r
\mathrel{
\hbox to 0pt{\lower
3pt\hbox{$\mathchar"218$}\hss}\raise 2.0pt\hbox{$\mathchar"13E$}}24.06$ the transient chaos gives way to chaos on attractors which merge into the famous Lorenz strange attractor beyond $r
\mathrel{
\hbox to 0pt{\lower
3pt\hbox{$\mathchar"218$}\hss}\raise 2.0pt\hbox{$\mathchar"13E$}}24.74$ [@ott]. An orbit that drifts onto the strange Lorenz attractor is shown in Fig. \[attractor\].
We apply the $K$-test to the Lorenz system and show that it effectively marks the transition from non-chaotic motion to chaos on a strange attractor at $r\simeq 24.74$ where we see $K$ rising from $0$ to $1$. However, the test is unable to pick up the chaotic transient behaviour for values of $13.96
\mathrel{
\hbox to 0pt{\lower
3pt\hbox{$\mathchar"218$}\hss}\raise 2.0pt\hbox{$\mathchar"13E$}}r
\mathrel{
\hbox to 0pt{\lower
3pt\hbox{$\mathchar"218$}\hss}\raise
2.0pt\hbox{$\mathchar"13E$}}24.06$. Notice that a similar random scan through the Lyapunov exponents in Fig. \[lexplor\] also misses the transient episodes although short time exponents along the fractal basin boundary can be isolated.
Specific orbits of Fig. \[klor\] can be isolated to illustrate the behaviour in the $(p,q)$ subspace explicitly. The orbit at $r=18$ corresponds to the right-most orbit in Fig. \[transient\]. The orbit winds around before drifting onto the fixed point $(X_{R},Y_{R},Z_{R})$. This is reflected in the $(p,q)$ subspace by the stray steps taken before the orbit moves onto the smooth, bounded ring of the regular attractor as shown in Fig. \[sample1\]. This results in a regular oscillation in the mean-square displacement $M(t)$ that will eventually die away to give $K\rightarrow 0,$ as shown in Fig. \[klor\]. The $K$-test reflects the regularity of the attractor and is unable to detect the subtle chaotic transient in the early motion.
By contrast, an orbit at $r=30$ in Fig. \[klor\] is drawn onto the Lorenz strange attractor and its strongly chaotic behaviour is detected by the $K$-test. The chaotic motion is well reflected by the Brownian diffusion evidenced in the $(p,q)$ subspace shown in Fig. \[sample2\]. This orbit leads to $K\rightarrow 1$ as expected.
Summary
=======
We have confirmed that the $K$-test provides a simple and easy diagnostic for the transition from regularity to chaos for Hamiltonian as well as dissipative systems. For the test to be effective, an orbit must spend sufficient time on the hyperbolic chaotic attractor in a dissipative system, or be confined to a highly chaotic region of phase space if the system is Hamiltonian. However, for chaotic transients which move into a chaotic region of phase space and then out into a regular region of phase space the motion is not consistently Brownian. Consequently, the test can yield ambiguous and confusing results although a qualitative look over the $(p,q)$ subspace can provide guidance as to the transient irregularity of the base dynamics. Additionallly the K-test is not ideal for relativistic settings since it depends on the time coordinate used and so, like the Lyapunov exponents, is not a covariant indicator of chaos. These limitations of the $K$-test are not unexpected: no one probe of chaos can suit every scenario. Nor are they fatal to its utility. Rather, we hope that they will help to map out the territory over which this simple test is a reliable guide to the presence of chaos
Acknowledgements {#acknowledgements .unnumbered}
================
We are especially grateful to both G. Gottwald and I. Melbourne for answering questions regarding their method. JL thanks N.J. Cornish for useful conversations. JL is supported by a PPARC Advanced Fellowship and an award from NESTA.
G. A. Gottwald and I. Melbourne, nlin.CD/0208033.
M. Henon and C. Heiles, Ap. J. [**69**]{} 73 (1964).
M. Nicol, I. Melbourne, and P. Ashwin, Nonlinearity [**14**]{} 275 (2001).
M.J. Field, I. Melbourne, and A. Török, Ergod. Th. & Dynam. Sys. in press (2002).
E. Lorenz, J. Atmos. Phys. [**20**]{} 130 and 148 (1963).
E. Ott, [*Chaos in dynamical systems*]{}, (Cambridge University Press, Cambridge, 1993).
J.J. Kaplan and J.A. Yorke, Comm. Math. Phys. [**67**]{} 93 (1979).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A simple coherent-imaging method due to Paganin [*et al.*]{} is widely employed for phase–amplitude reconstruction of samples using a single paraxial x-ray propagation-based phase-contrast image, provided the sample-to-detector distance is sufficiently small for the associated Fresnel number to be large compared to unity. The algorithm is particularly effective when employed in a tomographic setting, using a single propagation-based phase-contrast image for each projection. Here we develop a simple extension of the method, which improves the reconstructed contrast of very fine sample features. This provides first-principles motivation for boosting fine spatial detail associated with high Fourier frequencies, relative to the original method, and was inspired by several recent works employing empirically-obtained Fourier filters to a similar end.'
author:
- 'David M. Paganin'
- 'Vincent Favre-Nicolin, Alessandro Mirone, Alexander Rack, Julie Villanova'
- 'Margie P. Olbinado'
- Vincent Fernandez
- 'Julio C. da Silva'
- Daniele Pelliccia
bibliography:
- 'GPM2020.bib'
title: 'Boosting spatial resolution by incorporating periodic boundary conditions into single-distance hard-x-ray phase retrieval '
---
Introduction
============
In 2002 a simple algorithm was published for reconstructing the projected thickness of a single-material sample given a single propagation-based phase contrast image obtained in the small-defocus regime [@Paganin2002]. In this method, the ratio of the real part of the projected refractive index decrement and the projected linear attenuation coefficient is assumed to be both known and constant. The method assumes paraxial coherent radiation or matter waves (e.g. x-rays, visible light, electrons or neutrons), plane-wave illumination of known intensity, and an object-to-detector propagation distance that is sufficiently small for each structure in the sample to produce no more than one Fresnel fringe (more precisely, the object-to-detector distance is assumed to be small enough to make the corresponding Fresnel number large compared to unity). Within its domain of validity (single-material sample and small object-to-detector propagation distance for paraxial radiation or matter waves), the method may be viewed as providing a computationally-simple unique closed-form deterministic solution to the twin-image problem of inline holography [@Gabor1948], since propagation-based phase contrast images are synonymous with inline holograms [@Pogany1997].
The 2002 algorithm has been widely utilised, particularly for propagation-based x-ray phase contrast imaging. Its advantages, bought at the price of the previously stated strong assumptions, include simplicity, speed, significant noise robustness even for strongly absorbing samples, and the ability to process time-dependent images frame-by-frame. Efficient computer implementations are available in the following software packages: ANKAphase [@Weitkamp2011], X-TRACT [@Gureyev2011], pyNX [@FavreNicolin2011], PITRE [@Chen2012], Octopus [@Boone2012; @Dierick2004], pyHST2 [@mirone2014], TomoPy [@Gursoy2014; @Pelt2016], SYRMEP Tomo Project [@Brun2017] and HoloTomo Toolbox [@Lohse2020]. While most applications to date have employed x-rays, the method was originally developed with a broader domain of applicability in mind, including but not limited to electrons, visible light and neutrons [@Paganin2002]. Subsequently, the method has now also been applied to out-of-focus contrast images obtained using electrons [@Liu2011], visible light [@Poola2017] and neutrons [@Paganin2019].
When the method of @Paganin2002 (PM) is utilised in a tomographic context [@Mayo2003], its domain of utility broadens since many objects may be viewed as locally composed of a single material of interest, in three spatial dimensions, even though they cannot be described as composed of a single material in projection [@Beltran2010; @Beltran2011]. Early examples of applications of the PM in a tomographic setting include the imaging of paper [@Mayo2003], polymer micro-wire composites [@Mayo2006], high-Weber-number water jets [@Wang2006], self healing thermoplastics [@Mookhoek2010], paint-primer micro-structure [@Yang2010], partially open foams [@Blankenburg2017], sandstone micro-structure [@Yang2013], granite [@Denecke2011], melting snow [@Uesugi2012], anthracite coal [@Wang2013], evolving liquid foams [@Mokso2013], iron oxide particles in mouse brains [@Marinescu2013; @Rositi2013], rat brains [@Beltran2011], mouse lungs [@Lovric2013], rabbit lungs [@Beltran2011], mouse tibiae [@Stevenson2010], crocodile teeth [@Enax2013], mosquitoes [@Weitkamp2011], fly legs [@Mayo2003], high speed [*in vivo*]{} imaging of a fly’s flight motor system [@Schwyn2013], wood [@Mayo2006], dynamic crack propagation in heat treated hardwood [@Gilani2013], rose peduncles [@Matsushima2012], amber-fossilised spiders [@McNeil2010; @Penney2012], amber-fossilised centipedes [@Edgecombe2012], fossilised rodent teeth [@Rodrigues2012], fossil bones [@Sanchez2012], ancient cockroach coprolites [@Vrskansky2013], fossilised early-animal embryos [@Yin2013], fossil muscles of primitive vertebrates [@Trinajstic2013; @Sanchez2013] and the vertebral architecture of ancient tetrapods [@Pierce2013]. Publications from 2014 onwards to the present are too numerous to list [^1].
The present work was inspired by several publications that incorporate unsharp masking or related techniques to boost fine spatial detail in reconstructions obtained using the PM. This includes the deconvolution filter incorporated into the ANKAphase [@Weitkamp2011] version 2.1 implementation of the PM, Sanchez [*[et al.]{}*]{}’s incorporation of an unsharp mask into the pyHST2 implementation of the PM [@Sanchez2012; @mirone2014], and the work of @Irvine2014 utilising the measured phase contrast image as a physical unsharp mask. These extensions of the method all suppress high spatial-frequency information by a factor less than that given by the Fourier-space Lorentzian [^2] filter that is employed in the PM. Notable also is the work of @Yu2017, which enhances fine spatial detail by adapting the PM to a multi-image setting. The resulting improvements, most particularly in fine spatial detail obtained via tomographic reconstructions utilising the method, are clearly evident in the previously cited publications. These publications [@Weitkamp2011; @Sanchez2013; @mirone2014; @Irvine2014; @Yu2017] provide impetus to revisit the theory underpinning the PM, thereby seeking a first-principles justification for reduced suppression of high spatial frequency information, relative to the Fourier filter in the original form of the method.
The remainder of the paper is structured as follows. Section II derives a generalised form of the PM (“GPM”), showing how it reduces to the original form of the single-image phase-retrieval algorithm for low spatial frequencies in the input phase-contrast image data. Simulated x-ray data is considered in Sec. III, comparing the GPM to the PM, and exploring the limits of both approaches. Section IV discusses the domain of applicability for both the GPM and the PM, together with the effective high-pass filter to the PM that is implied by the GPM. Section V indicates some possible possible avenues for future work. We conclude with a brief summary in Sec. VI.
Incorporation of periodic boundary conditions into the PM
=========================================================
For a monochromatic scalar x-ray wavefield with intensity $I(x,y,z)$ and phase $\varphi(x,y,z)$ that is paraxial with respect to an optical axis $z$, the associated continuity equation is known as the transport-of-intensity equation [@Teague1983] (TIE): $$\begin{aligned}
\nabla_{\perp}\cdot[I(x,y,z)\nabla_{\perp}\varphi(x,y,z)]=-k\frac{\partial I(x,y,z)}{\partial z}.
\label{eq:A}\end{aligned}$$ Here, $(x,y)$ denote Cartesian coordinates in planes perpendicular to the optical axis, $\nabla_{\perp}$ denotes the gradient operator in the $xy$ plane and $k=2\pi/\lambda$ is the wavenumber corresponding to the vacuum wavelength $\lambda$. A validity condition for this equation is that the Fresnel number $N_{\textrm{F}}$ obey $$\begin{aligned}
N_{\textrm{F}} \equiv \frac{W^2}{\lambda\Delta} \gg 1.
\label{eq:Fresnel_number}\end{aligned}$$ Here, $W$ is the characteristic transverse length scale for the wavefield being propagated, and $\Delta\ge 0$ is the distance from (i) the planar exit surface $z=0$ over which the unpropagated wavefeld is specified, to (ii) the parallel planar surface $z=\Delta$ over which the intensity of the propagated wavefield is registered using a pixellated position-sensitive detector (see Fig. \[fig:BasicDiagram\].)
Following @Paganin2002, consider a single-material object lying immediately upstream of the plane $z=0$, whose $z$-projection of thickness is given by $T(x,y)$ – see Fig. \[fig:BasicDiagram\]. The projection approximation [@Paganin2006] gives the usual Beer–Lambert law for the intensity $I(x,y,z=0)$ at the exit surface $z=0$ of the object, for the case where the object is illuminated with $z$-directed monochromatic plane waves having uniform intensity $I_0$: $$\begin{aligned}
I(x,y,z=0)=I_0 \exp[-\mu T(x,y)].
\label{eq:B}\end{aligned}$$ Here, $\mu$ is the linear attenuation coefficient of the object. The projection approximation also gives an expression for the transverse phase distribution over the exit surface of the object [@Paganin2006]: $$\begin{aligned}
\varphi(x,y,z=0)=-k\delta T(x,y),
\label{eq:C}\end{aligned}$$ where $1-\delta$ is the real part of its complex refractive index $$n=1-\delta+i\beta$$ and $$\mu=2k\beta.
\label{eq:Definition_for_mu}$$ Note that the single-material object may be generalised to the case of variable mass density $\rho(x,y,z)$, the requirement then being that its complex refractive index have the form $$n(x,y,z)=1-A\rho(x,y,z),$$ where $A$ is a fixed complex constant at fixed energy, and ${\textrm{Re}}A>0$ [@Paganin2004a].
![A single-material object of projected thickness $T(x,y)$ is illuminated by normally incident $z$-directed monochromatic scalar plane waves of uniform intensity $I_0$, where $(x,y)$ are Cartesian coordinates perpendicular to the optical axis $z$. The resulting paraxial exit-surface wavefield, over the plane $z=0$, propagates in vacuum through a distance $z=\Delta$. The associated propagation-based phase contrast image has intensity distribution $I(x,y,z=\Delta)$.[]{data-label="fig:BasicDiagram"}](Diagram1.pdf){width="50.00000%"}
Assume vacuum to lie in the half space $z\ge 0$ downstream of the object. Assume the exit-surface wavefield over the plane $z=0$ to propagate through a distance $\Delta > 0$ downstream of the object, with this distance being sufficiently small for the Fresnel number to be much greater than unity. We may then make the following forward-finite-difference approximation to the longitudinal intensity derivative on the right side of Eq. (\[eq:A\]), using the propagation based phase contrast image $I(x,y,z=\Delta)$ of the single-material object in tandem with the estimate for the contact image given by Eq. (\[eq:B\]): $$\begin{aligned}
\nonumber \left. \frac {\partial I(x,y,z)}{\partial z} \right|_{z=0} \approx \frac{I(x,y,z=\Delta)-I_0\exp[-\mu T(x,y)]}{\Delta}. \\
\label{eq:D}\end{aligned}$$ If Eqs (\[eq:B\]), (\[eq:C\]) and (\[eq:D\]) are substituted into Eq. (\[eq:A\]), re-arrangement yields the screened Poisson equation [@Paganin2002]: $$\begin{aligned}
\frac{I(x,y,z=\Delta)}{I_0}=\left( 1-\frac{\delta\Delta}{\mu}\nabla_{\perp}^2\right) \exp[-\mu T(x,y)].
\label{eq:E}\end{aligned}$$ The manner in which this has been previously solved is to notice that Fourier transformation turns this partial differential equation into an algebraic equation, via the Fourier derivative theorem. This leads immediately to the PM [@Paganin2002]: $$\label{eq:SPEX}
T(x,y)=-\frac{1}{\mu}\log_e\left(\mathcal{F}^{-1}
\left\{\frac{\mathcal{F}\left[I(x,y,z=\Delta)/I_0\right]}
{1+(\delta\Delta/\mu)(k_x^2+k_y^2)}\right\}
\right).$$ Here $\mathcal{F}$ denotes Fourier transformation with respect to $x$ and $y$ in any convention for which $\nabla_{\perp}$ transforms to $(i k_x,i k_y)$, $\mathcal{F}^{-1}$ is the corresponding inverse Fourier transformation, and $(k_x,k_y)$ are Fourier-space spatial frequencies corresponding to $(x,y)$. The Fourier-space filter, in the above expression, has the previously-mentioned Lorentzian form.
When Eq. (\[eq:SPEX\]) is directly applied to experimental propagation-based x-ray phase contrast images that are sampled over a Cartesian mesh, and the discrete Fourier transform used to approximate the (continuous) Fourier transform integral, there is an implicit assumption that the object does not contain appreciable spatial frequency information in the vicinity of the Nyquist limit of the mesh. While this assumption was once typically quite reasonable in most coherent-x-ray-imaging contexts, the exquisitely detailed structures that are now routinely imaged in contemporary x-ray phase-contrast-tomography applications imply that this implicit assumption may now be becoming somewhat less broadly applicable—see e.g. @Sanchez2012. For such applications, and as the following argument will demonstrate, Eq. (\[eq:SPEX\]) overly strongly filters the highest spatial-frequency information that is present in the data.
With a view to extending the validity of the PM out to the Nyquist limit of the data sampled on a typical pixellated imaging-detector array, recall the following five-point approximation for the transverse Laplacian [@Press1996; @AbramowitzStegun]:
$$\begin{aligned}
\nabla_{\perp}^2 h(x_m,y_n) = \frac{h(x_{m-1},y_n)+h(x_{m+1},y_n)+h(x_m,y_{n-1})+h(x_m,y_{n+1})-4h(x_m,y_n)}{W^2}.
\label{eq:F}\end{aligned}$$
Here, $h(x,y)$ is a twice-differentiable continuous single-valued function, sampled over a mesh in which each grid element is a square of physical width $W$ metres by $W$ metres. Hence the mesh locations are given by $$(x_m,y_n)=(Wm, Wn),$$ where $m$ and $n$ are integers (mesh indices) that are restricted to the ranges $0\le m\le N_1-1$ and $0\le n \le N_2-1$, with $N_1$ being the number of sample points in the $x$ direction, and $N_2$ being the number of sample points in the $y$ direction. The key point, here, is that while the fundamental-calculus definition of the transverse derivative considers the mesh step-size $W$ to tend to zero, when working with a discrete grid we are not justified in taking $W$ to be any smaller than the pixel size of the mesh.
With the specified mesh of pixel locations $(x_m,y_n)$ in place, the function $h(x_m,y_n)$ may be expressed in terms of its discrete Fourier transform $H(k_{x,p},k_{y,q})$ [@Press1996]:
$$\begin{aligned}
h(x_m,y_n)=\frac{1}{N_1 N_2}\sum_{p=0}^{N_1-1} \sum_{q=0}^{N_2-1} \exp\left(-\frac{2\pi i m p}{N_1}\right) \exp\left(-\frac{2\pi i n q}{N_2}\right)H(k_{x,p},k_{y,q}).
\label{eq:G}\end{aligned}$$
Here, the discreteness of the sampling grid restricts the allowed spatial frequencies $(k_x,k_y)$ to the mesh $$(k_{x,p},k_{y,q})=\left(\frac{2\pi p}{N_1 W},\frac{2\pi q}{N_2 W}\right),$$ with $p$ lying in the range $-\tfrac{1}{2}N_1, \cdots, \tfrac{1}{2}N_1$, and $q$ lying in the range $-\tfrac{1}{2}N_2,\cdots,\tfrac{1}{2}N_2$.
Motivated by the form of the differential operator in Eq. (\[eq:E\]), we can show by direct substitution of Eq. (\[eq:G\]) into Eq. (\[eq:F\]) that (cf. @Freischlad1986, @Ghiglia1994 and @arnison2004):
$$\begin{aligned}
\nonumber (1-\alpha\nabla_{\perp}^2)h(x_m,y_n)=\frac{1}{N_1 N_2}\sum_{p=0}^{N_1-1} \sum_{q=0}^{N_2-1} \exp\left(-\frac{2\pi i m p}{N_1}\right)\exp\left(-\frac{2\pi i n q}{N_2}\right)H(k_{x,p},k_{y,q}) \\ \times \left\{1-\frac{2\alpha}{W^2}\left[\cos\left(\frac{2\pi p}{N_1}\right)+\cos\left(\frac{2\pi q}{N_2}\right)-2\right]\right\},
\label{eq:H}\end{aligned}$$
where $\alpha$ is a constant having dimensions of squared length. Set this constant to the real non-negative number: $$\alpha=\frac{\delta\Delta}{\mu}.
\label{eq:Definition_for_alpha}$$ Equation (\[eq:H\]) then implies that Eq. (\[eq:E\]) may be solved for the projected thickness $T(x,y)$ of the single-material sample, over the lattice of points $(x_m,y_n)$, via the following generalised form of the PM (termed the “GPM” henceforth):
$$\begin{aligned}
\nonumber T(x_m,y_n)=-\frac{1}{\mu}\log_e\left(\text{IDFT}^{p\rightarrow m}_{q\rightarrow n}
\left\{\frac{{\text{DFT}}^{m\rightarrow p}_{n\rightarrow q} [I(x_m,y_n,z=\Delta)/I_0]}{1-\frac{2\alpha}{W^2}\left[\cos\left(Wk_{x,p}\right)+\cos\left(Wk_
{y,q}\right)-2\right]}\right\}\right), \quad \alpha=\frac{\delta\Delta}{\mu},\quad |Wk_{x,p}|,|Wk_{y,q}| \le \pi. \\
\label{eq:I}\end{aligned}$$
Here, ${\text{DFT}}^{m\rightarrow p}_{n\rightarrow q}$ is the discrete Fourier transform operator, which maps a function $h(x_m,y_n)$ sampled on the real-space lattice $(x_m,y_n)$ to its discrete Fourier transform $H(k_{x,p},k_{y,q})$ sampled on the Fourier-space lattice $(k_{x,p},k_{y,q})$, and $\text{IDFT}^{p\rightarrow m}_{q\rightarrow n}$ is the corresponding inverse discrete Fourier transform (cf. Eq. (\[eq:G\]); cf. @Ghiglia1994, who write a similar expression in the context of the Poisson equation). Note that the key parameter in the above equation is the dimensionless constant: $$\label{eq:DimensionlessAlphaConstant}
\frac{\alpha}{W^2}=\frac{\delta/\beta}{4\pi N_{\textrm{F}}}.$$
Several plots of the low-pass Fourier-space filter in Eq. (\[eq:I\]) are given in Fig. \[fig:NewFilter\], as a function of the single dimensionless parameter $\alpha/W^2$. Note the transition from (i) near-rotational-symmetry and near-Lorentzian form close to the origin of Fourier space, to (ii) the symmetry of a square at the edges of Fourier space. The filter obeys periodic boundary conditions at the edges of the Fourier-space mesh, with each mesh value along the mesh’s edges corresponding to a Nyquist frequency, namely: $$W k_{x,p}^{\textrm{max}}, Wk_{y,q}^{\textrm{max}}=\pm\pi.$$ An evident trend is that, while the GPM filter clearly differs from the (rotationally symmetric) PM filter for the smaller values of $\alpha/W^2 \le 1$ given in Figs. \[fig:NewFilter\](a,b,c), for the larger values of $\alpha/W^2 > 1$ in Figs. \[fig:NewFilter\](d,e,f) the GPM filter has essentially converged to the PM filter. This relation between the two filters will be explored in further detail below. On a different note, another general trend is evident in Fig. \[fig:NewFilter\], namely the fact that the Fourier-space filtration becomes progressively stronger as the dimensionless parameter $\alpha/W^2$ increases. This latter trend makes intuitive physical sense, since (i) the stronger the propagation-based phase contrast effects in the image recorded over the plane $z=\Delta$, the greater will be the amplification of encoded high-spatial-frequency detail in the projected thickness of the sample, and hence the greater the degree of Fourier-space filtration that is required in order to decode the phase-contrast signal so as to give the required projected thickness; (ii) the phase-contrast signal is directly proportional to $\delta/\beta$ (see Eq. (\[eq:E\])) and inversely proportional to the Fresnel number $N_{\textrm{F}}$ (as larger Fresnel numbers correspond to smaller propagation distances $\Delta$ and/or more slowly varying transverse structure in the wavefield at the exit surface of the sample), which is precisely the dependence on the right side of Eq. (\[eq:DimensionlessAlphaConstant\]).
{width="100.00000%"}
In the low-spatial-frequency limit, we may make the second-order Taylor-series approximation $$\begin{aligned}
\nonumber \cos(Wk_{x,p})\approx 1 - \frac{1}{2}(Wk_{x,p})^2, \\
\cos(Wk_{y,q})\approx 1 - \frac{1}{2}(Wk_{y,q})^2.\end{aligned}$$ Equation (\[eq:I\]) then reduces to the PM with its rotationally-symmetric discrete-Fourier-transform representation of Eq. (\[eq:SPEX\]) and the associated Lorentzian filter:
$$\begin{aligned}
T(x_m,y_n)\longrightarrow -\frac{1}{\mu}\log_e\left(\text{IDFT}^{p\rightarrow m}_{q\rightarrow n}
\left\{\frac{{\text{DFT}}^{m\rightarrow p}_{n\rightarrow q} [I(x_m,y_n,z=\Delta)/I_0]}{1+{\alpha\left(k_{x,p}^2+k_{y,q}^2\right)}}\right\}\right), \quad \alpha=\frac{\delta\Delta}{\mu},\quad |Wk_{x,p}|,|Wk_{y,q}| \ll 1.\quad
\label{eq:J}\end{aligned}$$
While Eq. (\[eq:I\]) (GPM) reduces to Eq. (\[eq:J\]) (PM) for low Fourier spatial frequencies, the GPM is less strongly suppressing of high Fourier spatial frequencies. This key point will be explored further in a subsequent section. However, before considering this point in further detail, it is useful first to consider an indicative numerical illustration of the difference between the GPM and PM filters. It is to this latter topic that we now turn.
Computer simulations
====================
{width="100.00000%"}
The efficiency of the generalised phase-retrieval reconstruction method can be asserted by simulating paraxial x-ray propagation through a suitably-high-resolution object, and reconstructing it using both the PM and the GPM. To perform this, we chose a spatially random binary object (Fig. \[fig:simulation\]a) with an x-ray wavelength of $0.5\ \text{\normalfont\AA}$, $\beta=10^{-9}$, $\delta=5 \times 10^{-7}$, a pixel size of $10\ \mu$m, and a thickness for the object of $40\ \mu$m. A simulated unit-amplitude plane-wave was transmitted through this object and propagated by a distance of $\Delta=0.1$ m using a near-field propagator. In order to avoid aliasing effects due to the discrete Fourier transform used, the propagation was performed on a $2 \times$ oversampled object (where the binary random pattern pixels had a $2 \times 2$ size), and the propagated intensity was rebinned (averaging the intensity over $2 \times 2$ pixels) before being back-propagated. Thus the object-plane image was oversampled and the corresponding propagated intensity subsequently downsampled to compensate for the initial oversampling of the object.
Figures \[fig:simulation\]b and \[fig:simulation\]c show the intensity of the propagated waves reconstructed using the PM and GPM, respectively, where the obtained thickness maps (normalised to the starting object thickness of $40\ \mu$m) are compared. There is a significant improvement in the reconstructed images obtained with the GPM relative to the images reconstructed with the PM. Figure \[fig:simulation\]d displays line profiles across the images which indicate improvement in the contrast. Since the low-frequency reconstructed images can be approximated as due to a convolution-induced blurring of the original image, we also performed a Richardson–Lucy deconvolution [@Richardson1972; @Lucy1974] using the original pattern as a reference, which allows us to estimate the point-spread-function kernel relating the reconstructed and original arrays. Figure \[fig:simulation\]e shows that the GPM yields a sharper kernel.
Discussion
==========
As stated in the introduction, the work of the present theoretical paper was inspired by several experimental investigations [@Weitkamp2011; @Sanchez2012; @mirone2014; @Sanchez2013; @Irvine2014] that phenomenologically employ unsharp masks or deconvolution to boost high-spatial-frequency information in x-ray phase contrast tomograms whose reconstructions are obtained with the assistance of the PM. This phenomenological modification to the PM often very significantly improves the level of fine-detail clarity in the reconstructions. Is there a fundamental explanation, derivable from an optical-imaging-physics perspective, that casts some light on why the phenomenological high-frequency-boost strategy is so successful? While we do not claim to give a complete answer to this still-open question, below we argue that the PM-to-GPM transition goes partway to addressing it.
Consider the ratio $R(k_x,k_y)$ of the GPM and PM Fourier-space filters, as given by Eqs (\[eq:I\]) and (\[eq:J\]): $$\begin{aligned}
\nonumber R(k_{x},k_{y})=\frac{1+\alpha (k_{x}^2+k_{y}^2)}{1-\frac{2\alpha}{W^2} [\cos(W k_{x})+\cos(W k_{y})-2]}, \\ -\pi/W \le k_{x}, k_{y} \le \pi/W. \quad\quad
\label{eq:K}\end{aligned}$$ This ratio may be viewed as a form of deconvolution mask, here derived from first principles, whose application transforms the PM into the GPM. As mentioned earlier in the present paper, such a deconvolution-mask viewpoint may be compared to (and was indeed inspired by) previously-published work which introduced such masks from a phenomenological perspective [@Weitkamp2011; @Sanchez2012; @mirone2014; @Irvine2014].
The ratio in Eq. (\[eq:K\]) is always greater than or equal to unity, implying that the GPM filter (Eq. (\[eq:I\])) suppresses each Fourier component of the measured phase contrast signal, by an amount that is never more than the degree of suppression based on the PM filter (Eq. (\[eq:J\])). To illustrate this point, see Fig. \[fig:Ratios\] for a series of contour plots of the Fourier-filter-ratio in Eq. (\[eq:K\]), for the same range of $\alpha/W^2$ values that was used in Fig. \[fig:NewFilter\]. The form of these plots—which give a GPM-to-PM filter ratio of unity near the Fourier-space origin, and boost high spatial frequencies by taking a value that is progressively greater than unity the further we move away from the Fourier-space origin—gives a partial first-principles justification for the previously cited works boosting high spatial frequencies of tomographic reconstructions based on the PM. However, we must emphasise that the degree of high-spatial-frequency boost, in the previously cited works [@Weitkamp2011; @Sanchez2012; @mirone2014; @Sanchez2013; @Irvine2014], is typically significantly larger than the degree of boost that can be justified using the arguments developed in the present paper. It is for this reason that we describe our first-principles justification as “partial”: the GPM gives a reconstruction of fine spatial detail that is superior to that obtained with the PM, but the GPM yields reconstructions that are inferior to those obtained by applying the previously-cited phenomenological approaches [@Weitkamp2011; @Sanchez2012; @mirone2014; @Sanchez2013; @Irvine2014] to improve the PM by suitably boosting high spatial frequencies. We conjecture there to be an additional factor or factors that can be used to derive additional high-frequency boosts from first principles, but the nature of these factors remains unanswered by the present investigation.
{width="100.00000%"}
In light of the above comments, let us make some additional remarks regarding the Fourier-filter-ratio plots in Fig. \[fig:Ratios\]. Near the origin of Fourier space, corresponding to coarser spatial detail in the reconstruction, the plots of Fig. \[fig:Ratios\] have a plateau of values near unity. Again, this is consistent with the GPM reducing to the PM at sufficiently coarse spatial resolution. The maximum value $R_{\textrm{max}}$ of the ratio $R(k_x,k_y)$, attained at the corners of the Fourier-space mesh, is given by $$R_{\textrm{max}}=\frac{1+2\pi^2\Upsilon}{1+8\Upsilon},$$ where $$\Upsilon\equiv\alpha/W^2=(4\pi N_{\textrm{F}})^{-1}\delta/\beta.$$ When $\Upsilon \gg 1$ (e.g. for sufficiently small $W$), we see that the maximal Fourier-space boost $R_{\textrm{max}}$ (of the GPM Fourier filter relative to the PM Fourier filter) is by a factor of $\pi^2/4\approx 2.5$. Conversely, when $\Upsilon \ll 1$ (e.g. for sufficiently large $W$), then $R_{\textrm{max}}$ tends to unity.
If we Taylor expand Eq. (\[eq:K\]) to fourth order in spatial frequency, which will be a fair approximation for spatial frequencies that are not too large in magnitude relative to the Nyquist frequency, we obtain $$R(k_x,k_y) \approx 1+\tfrac{1}{12}\alpha W^2 (k_x^4 + k_y^4).$$ The fact, that the smallest non-constant term in this expansion is quartic in spatial-frequency coordinates, is consistent with the plateau of values close to unity, exhibited by the plots in Fig. \[fig:Ratios\] near the Fourier-space origin. This aligns with the idea that the usual form of the PM works well for coarser spatial detail, but needs the GPM (or some other suitable approach) for better treatment of higher-spatial-frequency detail. The above result also implies that the GPM Fourier filter is approximately equal to $R(k_x,k_y) \approx 1+\frac{1}{12}\alpha W^2 (k_x^4 + k_y^4)$ multiplied by the PM Fourier filter, at least for sufficiently small Fourier frequencies; the corresponding real-space unsharp mask is therefore proportional to the result of applying the non-rotationally-symmetric fourth-order differential operator $\partial^4/\partial x^4 + \partial^4/\partial y^4$ to the PM image, rather than the more usual unsharp mask proportional to the result of applying the rotationally-symmetric second-order differential operator $-(\partial^2/\partial x^2 + \partial^2/\partial y^2)$ to the PM image.
It is interesting to further examine the conditions under which the GPM differs significantly from the PM. Take the ratio of Fourier filters in Eq. (\[eq:K\]), and evaluate this ratio at the maximum (Nyquist) $x$ and $y$ spatial frequencies $$k_x^{\textrm{max}}=k_y^{\textrm{max}}=\frac{\pi}{W}.$$ This gives the condition $$R\left(k_x=\frac{\pi}{W},k_y=\frac{\pi}{W}\right)=R_{\textrm{max}}=\frac{1+\frac{2\alpha\pi^2}{W^2}}{1+\frac{8\alpha}{W^2}}\geq 1 +\aleph$$ for the GPM to be significantly different in comparison to the PM. Here, $\aleph$ is a lower bound on the maximum relative difference, between the ratio of the two filters and unity, which is considered to be “significant”. Next we (i) make use of Eqs. (\[eq:Definition\_for\_mu\]) and (\[eq:Definition\_for\_alpha\]); (ii) incorporate both the definition of the Fresnel number and its associated requirement as given in Eq. (\[eq:Fresnel\_number\]). Hence we obtain the following material-dependent parameter domain for which the GPM is both (i) a significant improvement upon the PM, and (ii) still within the domain of validity for the underpinning transport-of-intensity equation: $$\frac{\delta}{\beta}\left(\frac{\frac{\pi}{2}-\frac{2}{\pi}}{\aleph}-\frac{2}{\pi}\right)\ge N_{\textrm{F}} \gg 1.
\label{eq:EPMCondition}$$ Thus e.g. if we choose $\aleph=0.1$, and round numerical factors to the nearest order of magnitude, the above inequalities become the material-dependent conditions: $$10 \frac{\delta}{\beta} \ge N_{\textrm{F}} \gg 1.
\label{eq:EPMCondition2}$$ Some sample numerical values may be used to illustrate the above expression: if $\delta/\beta=500$, $\lambda=0.5~{\mbox{\normalfont\AA}} =0.5\times 10^{-10}$ m and $W=10~\mu$m$=10^{-5}$ m, then Eq. (\[eq:EPMCondition2\]) will be satisfied if $0.4$ mm $\le \Delta \ll 2$ m. If the pixel size is halved to $W=5~\mu$m, leaving all other parameters unchanged, we instead obtain $0.1~{\textrm{mm}} \le \Delta \ll 50$ cm. If the condition in Eq. (\[eq:EPMCondition\]) is violated then more sophisticated methods than either the PM or the GPM will need to be employed, including but not limited to (i) holotomography [@Cloetens1999], (ii) approaches based on the first Born and Rytov approximations [@Gureyev--RytovBorn], (iii) approaches based on the contrast-transfer-function formalism [@Gureyev2004b--LinearAlgorithmsPaper; @Turner2004], and (iv) the variety of approaches that are both reported upon and compared in @Yu2018.
One further point should be made regarding unsharp masks and deconvolution. The GPM will still benefit from additional unsharp masking or deconvolution, to further boost high-spatial-frequency detail, since—among other reasons that are beyond the scope of the present investigation, such as truncation of the effects of Fresnel diffraction to ignore the presence of multiple Fresnel-diffraction fringing—the GPM does not explicitly take source-size-induced blurring into account. The degree of sharpening required for GPM-reconstructed images will necessarily be less pronounced than that which has been needed for PM-reconstructed images. In this context we point out that the image-blurring effect due to finite source size may be modelled by making the following replacement in Eq. (\[eq:E\]) [@beltran2018]: $$\frac{\delta}{\mu}\longrightarrow \frac{\delta}{\mu}-\frac{2 S^2}{\Delta}.
\label{eq:Beltran2018}$$ Here, $S$ is the radius of the effective incoherent point-spread function at the detector plane, that is due to source-size blurring (cf. @Gureyev2004). The above replacement transforms Eq. (\[eq:E\]) into a Fokker–Planck form [@Risken1989; @MorganPaganin2019; @PaganinMorgan2019] for which $2 S^2/\Delta$ plays the role of an effective diffusion coefficient [^3]. This simple algebraic replacement may be carried through all of the calculations of the present paper, thereby incorporating a partial source-size deconvolution into the analysis.
We close this discussion by observing that we can readily introduce a real parameter $\tau$, which lies between zero and unity inclusive, that may be used to continuously deform the PM algorithm in Eq. (\[eq:J\]) ($\tau = 0$), into the GPM algorithm in Eq. (\[eq:I\]) ($\tau = 1$), via:
$$\begin{aligned}
T(x_m,y_m)=-\frac{1}{\mu}\log_e\left(\text{IDFT}^{p\rightarrow m}_{q\rightarrow n}
\left\{\frac{{\text{DFT}}^{m\rightarrow p}_{n\rightarrow q} [I(x_m,y_n,z=\Delta)/I_0]}{1+{\alpha\left(k_{x,p}^2+k_{y,q}^2\right) - \frac{2\alpha\tau}{W^2} \Phi(k_{x,p},k_{y,q})}}\right\}\right), ~ 0\le\tau\le 1,
\label{eq:TUNABLE}\end{aligned}$$
where $$\begin{aligned}
\nonumber
\Phi(k_{x,p},k_{y,q})\equiv\cos(Wk_{x,p})+\cos(Wk_{y,q})-2 \\ +\tfrac{1}{2}(Wk_{x,p})^2+\tfrac{1}{2}(Wk_{y,q})^2. \end{aligned}$$
Some avenues for future work
============================
Since @Beltran2010 [@Beltran2011] reported signal-to-noise ratio (SNR) boosts of up to 200 in utilising the PM in a tomographic setting, relative to absorption contrast, there has been some interest in the “SNR boosting” properties of the PM [@Nesterets2014; @Gureyev2014; @Kitchen2017; @Gureyev2017]. Of particular note is the result that the SNR boost has 0.3 $\delta/\beta$ as an approximate upper bound under the assumption of Poisson statistics [@Nesterets2014; @Gureyev2014], with the SNR boost being even more favourable for very low sample-exposure times [@Kitchen2017; @Clark2019]. Since dose is proportional to the square of SNR, dose reductions of $300^2 =90,000$ or more are in principle possible with the PM [@Kitchen2017]. This implies that tomographic analyses are possible using much less dose (or, equivalently, much lower acquisition times) than previously required for a single two-dimensional projection. This reduced dose is of importance in the context of medical imaging, while in an industrial-inspection product-quality-control context (or security screening context) it enables significant increases in throughput speed due to the associated increase in source effective-brilliance; cf. e.g. the recent achievement of over 200 x-ray phase-contrast tomograms per second [@200tps], incorporating PM-based data processing. In light of the above comments, it would be interesting to see how the previously published analyses for SNR boost and associated dose reduction are altered by passage from the PM to the GPM. It appears likely that SNR boosts will be reduced somewhat if the GPM is used in place of the PM, in accord with the tradeoff between noise and spatial resolution [@GureyevNRU; @GureyevNRU2020].
Another interesting avenue for future work begins with the previously mentioned observation that, from version 2.1 onwards, the ANKAphase [@Weitkamp2011] implementation of the PM incorporates a deconvolution mask to boost fine spatial detail in reconstructions. This deconvolution filter $R_{\textrm{ANKA}}(k_x,k_y)$ takes the Fourier-space form [^4] $$\label{eq:ANKAphase_filter}
R_{\textrm{ANKA}}(k_x,k_y)=\frac{1+c}{c+\exp[-\pi\sigma^2(k_x^2+k_y^2)]},$$ where $c$ is a dimensionless constant and $\sigma$ is a characteristic width. In light of the findings of the present paper, it would be interesting to consider replacing the ANKAphase deconvolution filter with $$\label{eq:ANKAphase_filter_revised}
\tilde{R}_{\textrm{ANKA}}(k_x,k_y)=\frac{1+c}{c+\exp[-\sigma^4(k_x^4+k_y^4)]},$$ since the fourth-order Taylor expansion of the deconvolution filter would then agree with the fourth-order Taylor expansion of $R(k_x,k_y)$, provided that $$\frac{12\sigma^4}{1+c}=W^2\alpha.$$
One more avenue for future research would be to replace Eq. (\[eq:F\]), which formed a starting point for the analysis of the present paper, with a more sensitive expression that utilises higher-order discrete approximations to the transverse Laplacian operator. For example, Eq. (25.3.30) of @AbramowitzStegun makes it clear that, while we have a Taylor-series truncation error on the order of $W^2$ in the five-point approximation for the Laplacian that is given in Eq. (\[eq:F\]), their nine-point approximation in Eq. (25.3.31) [@AbramowitzStegun] gives a better estimate whose truncation error is on the order of $W^4$. Use of such higher-order approximations may lead to improved forms of Eq. (\[eq:I\]).
We end these indications of possible avenues for future work, by noting that both (i) the two-image TIE-based phase-retrieval method of @Paganin1998 (see also Sec. 4.5.2 of @Paganin2006), which does not make the assumption of a single-material object, and (ii) the single-image differential-phase-contrast version of the PM [@Paganin2004b], which does make such an assumption, would benefit from an analogous treatment of differential operators under the discrete Fourier transform, to that used in the passage from Eqs. (\[eq:J\]) to (\[eq:I\]). Perhaps the former method (namely that which is based on Ref. [@Paganin1998]) may become of increasing utility in an x-ray imaging setting, given both recent advances in semi-transparent detectors, and the fact that this method does not need the single-material assumption upon which the PM relies.
Conclusion
==========
A simple extension was given for the method of @Paganin2002 (PM), for phase–amplitude reconstruction of single-material samples using a single paraxial x-ray propagation-based phase contrast image obtained under the conditions of small object-to-detector propagation distance. This improves the reconstructed contrast of very fine sample features, using an approximation that is derived from a first-principles perspective. This theoretical investigation was motivated by, and partially explains from a fundamental perspective, the success of several papers incorporating unsharp masking or related techniques to boost fine spatial detail in reconstructions obtained using the PM [@Weitkamp2011; @Sanchez2012; @mirone2014; @Sanchez2013; @Irvine2014].
Acknowledgements {#acknowledgements .unnumbered}
================
Financial support by the Experiment Division of the ESRF for DMP to visit in 2017 and 2018 is gratefully acknowledged. DMP acknowledges fruitful discussions with Mario Beltran, Carsten Detlefs, Timur Gureyev, Marcus Kitchen, Kieran Larkin, Thomas Leatham, Kaye Morgan, Tim Petersen, James Pollock, Manuel Sánchez del Río and Paul Tafforeau.
[^1]: For a partial list of additional references that use the ANKAPhase implementation [@Weitkamp2011] of the phase-retrieval algorithm of @Paganin2002, see e.g. `http://www.alexanderrack.eu/ANKAphase/ ankaphase\textunderscoreusers.html`.
[^2]: We use the term “Lorentzian” to refer to functions of the form $f(x,a)=1/(a^2 + x^2)$, where $a$ is a real non-zero constant and $x$ is real variable. Note, however, that such functions are also often referred to as Breit–Wigner distributions or Cauchy distributions.
[^3]: See, in particular, the special case of Eq. (59) in the paper by @PaganinMorgan2019, in which their “effective diffusion coefficient” $D_{\textrm{eff}}(x,y)$ is considered to be independent of transverse coordinates $(x,y)$. The resulting Fokker–Planck equation is mathematically identical in form to the screened Poisson equation given when Eq. (\[eq:E\]) of the main text is modified by the replacement that is indicated in Eq. (\[eq:Beltran2018\]).
[^4]: See `https://imagej.nih.gov/ij/plugins/ankaphase/ ankaphase-userguide.html` for an updated form of the ANKAphase software, which extends the form reported in @Weitkamp2011, to incorporate the optional image-restoration deconvolution filter that is reproduced in Eq. (\[eq:ANKAphase\_filter\]) of the main text.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We report on our result of third order Coulomb correction to the cross section $\sigma(e^+ e^-\rightarrow t\bar{t})$ near threshold. Analytic expression for the Coulomb energy and wave function at the origin are obtained. We discuss the significance of the Coulomb correction to the threshold cross section and heavy quarkonium phenomenology.'
author:
- |
Y. Kiyo [^1]\
[*Institut für Theoretische Physik E, RWTH Aachen,\
D – 52056 Aachen, Germany*]{}
title: 'Third order Coulomb correction to $t\bar{t}$ threshold cross section [^2]'
---
Threshold cross section
=======================
The $t\bar{t}$ threshold cross section has the following schematic form in the conventional perturbative expansion of $\alpha_s$, $$\begin{aligned}
\sigma(e^+e^-\rightarrow t\bar{t}) \sim \sigma_{\rm Born}
\large[1+\frac{\alpha_s}{v}+ \left(\frac{\alpha_s}{v}\right)^2 +
\cdots \,\large]\, ,\end{aligned}$$ where $\sigma_{\rm Born}$ is a Born cross section, and $v=\sqrt{1-4m_t^2/s}$ is speed of produced $t,\bar{t}$. The combination of $(\alpha_s/v)^n$ appears to all order in the perturbation theory, known as Coulomb singularity. The perturbative expansion in $\alpha_s$ is not applicable to the threshold cross section because $v$ is of order $\alpha_s$ near the $t\bar{t}$ threshold $\sqrt{s}\sim 2m_t$, and the Coulomb singularity $(\alpha_s/v)^n$ dominates the cross section. To obtain meaningful cross section the Coulomb singularity has to be summed to all order in $\alpha_s$. The physical origin of the Coulomb singularity is instantaneous gluon exchange between $t$ and $\bar{t}$ which has a small spatial momentum $|\vec{q}|\sim v \sim 0$. This is called potential gluon because of its propagator $$\widetilde{V}_C({\mbox{\boldmath ${q}$}}) = -\frac{4\pi C_F\alpha_s}{{\mbox{\boldmath ${q}$}}^2} \, ,
\label{Cpot}$$ where $C_F=4/3$, which is the Coulomb potential $V_C =-C_F\alpha_s/r$ in coordinate space.
This understanding leads to a quantum mechanical description which is equivalent to resummation of the Coulomb singularity. People had been looking for an effective field theory (EFT) which makes the resummation systematic. Finally a non-relativistic version of QCD was derived, called pNRQCD/vNRQCD [@Pineda-Soto97; @Luke-Manohar-Rothstein99]. The EFT makes the resummation systematic based on nonrelativistic power counting $v \sim \alpha_s\ll 1$ in Lagrangian level. Nowadays we understand the resummation using the EFT, and higher order corrections are taken into account in the EFT framework systematically. The NNLO calculation for the total cross section was completed by several groups [@TopWGR], and now we are working on the NNNLO corrections using the EFT.
The EFT classifies the corrections into three classes:\
\
$\bullet$ Hard corrections included in Wilson coefficients of (composite) operators.\
$\bullet$ Potential corrections, which are non-local (in space) 4-Fermi operators but local in time.\
$\bullet$ Dynamical gluon corrections called ultra-soft gluon in the EFT.\
\
The hard corrections are related to re-normalization of the operators in the EFT. The ultrasoft correction appears at NNNLO calculation, and the corresponding energy level correction is know by Kniehl-Penin [@kniehl-penin99]. However complete NNNLO correction to the total cross section is not known, yet. We discuss a part of the potential corrections in this report, which has the following form in the EFT Lagrangian, $$\begin{aligned}
&& \hspace{-.6cm} \delta {\cal L}(x)=\int d^3{{\mbox{\boldmath ${r}$}}}
\big[\psi^\dag \psi \big](x+{\mbox{\boldmath ${r}$}}) V({\mbox{\boldmath ${r}$}}) \big[
\chi^\dag\chi\big](x)\, ,
\label{EFTLag}\end{aligned}$$ where $\psi^\dag, \chi$ is a creation operator of heavy quark and anti-quark, respectively. We will report on the result of the Coulomb correction.
We parameterize the momentum space Coulomb potential as follows $$\begin{aligned}
\widetilde{V}({\mbox{\boldmath ${q}$}}) = \widetilde{V}_C({\mbox{\boldmath ${q}$}})\,
\big[1+ \widetilde{\cal V}_{1}({\mbox{\boldmath ${q}$}})
+\widetilde{\cal V}_{2}({\mbox{\boldmath ${q}$}})
+\widetilde{\cal V}_{3}({\mbox{\boldmath ${q}$}}) \big], \label{VCparametrization}\end{aligned}$$ where ${\cal V}_{n}({\mbox{\boldmath ${q}$}})$ is n-th order correction to the pure Coulomb potential $\widetilde{V}_C$, induced by loop diagrams when the EFT is derived from QCD. Explicit form of the potentials were derived at NNNLO (except $a_3$) in ref. [@kniehl-penin-smirnov-steinhauser00], and the Coulomb potential reads $$\begin{aligned}
&&\hspace{-.7cm} \widetilde{\cal V}_1 = \frac{\alpha_s}{4\pi}
\big[a_1+\beta_0 \, l_q \, \big]
\nonumber \\
&& \hspace{-.7cm} \widetilde{\cal V}_2 =
\left(\frac{\alpha_s}{4\pi}\right)^{\!2} \big[a_2 + \big(2
a_1\beta_0+\beta_1\big) \, l_q + \beta_0^2 \,l_q^2 \,\big]
\nonumber \\
&&\hspace{-.7cm} \widetilde{\cal V}_3 =
\left(\frac{\alpha_s}{4\pi}\right)^{\!3} \big[a_3 + 8\pi^2
C_A^3\,l_q(\nu) + \big(3 a_2\beta_0+2 a_1\beta_1
\nonumber\\
&&\hspace{-.5cm} +\beta_2\big)\,l_q + \,\big(3
a_1\beta_0^2+\frac{5}{2}\beta_0\beta_1\big) \, l_q^2 + \beta_0^3 \,
l_q^3 \, \big] ,
\label{CalV}\end{aligned}$$ where $l_q=\ln(\mu^2/{\mbox{\boldmath ${q}$}}^2)$ and $l_q(\nu)=\ln(\nu^2/{\mbox{\boldmath ${q}$}}^2)$, $\beta_i$ are the coefficients of QCD $\beta$-function. The scale $\mu$ is QCD renormalization scale, and the $\nu$ in ${\cal V}_3$ is a scale introduced to separate the ultrasoft and potential modes in the EFT. Physical quantities are scale independent if all the corrections at given order are taken into account (see for instance [@brambilla-pineda-soto-vairo99]). Now our task is to calculate the threshold cross section using the EFT Lagrangian eq.(\[EFTLag\]) with the Coulomb potential eqs.(\[VCparametrization\]), (\[CalV\]).
Green function method
=====================
We use the optical theorem to calculate the threshold total cross section which tells us that the cross section can be obtained by taking the imaginary part of the correlation function of production currents $$\begin{aligned}
&& \hspace{-0.6cm} \sigma(e^+ e^- \rightarrow t\bar{t}) =\frac{18\pi
e_t^2}{m_t^2}
{\rm Im} G(E+i \Gamma_t), \nonumber \\
&& \hspace{-0.6cm} G=\frac{i}{N_c}\int d^4 x e^{iqx}\langle {\rm
vac}|[\psi^\dag\chi](x) [\chi^\dag \psi](0)|{\rm vac}\rangle ,\end{aligned}$$ where $N_c=3$ and $m_t$ is the quark pole mass. Using the EFT one can show that the matrix element $G(E)$ can be expressed by the quantum mechanical Green function $$\begin{aligned}
&& \hspace{-.7cm} G(E)=\langle 0| \hat{G}(E) |0\rangle =\langle 0|\,
\frac{1}{\frac{{\mbox{\boldmath ${p}$}}\,^2}{m_t}+V({\mbox{\boldmath ${r}$}})-E}\,|0\rangle\, ,
\label{GreenFunc}\end{aligned}$$ where $|0\rangle$ denotes a quantum mechanical position eigenstate at the origin ${\mbox{\boldmath ${r}$}}=0$. At this stage we perform quantum mechanical perturbation theory by expanding the denominator of the Green function with respect to the higher order Coulomb potentials. The third order corrections read $$\begin{aligned}
&& \hspace{-.6cm} \delta_3\hat{G}=
-\,\hat{G}_0\delta V_3 \hat{G}_0 + \, 2\hat{G}_0\delta V_1
\hat{G}_0\delta V_2 \hat{G}_0
\nonumber \\
&&\hspace{-.0cm}
- \, \hat{G}_0\delta V_1\hat{G}_0\delta V_1\hat{G}_0\delta V_1
\hat{G}_0 \, .
\label{expandedGF}\end{aligned}$$ Here $G_0=({\mbox{\boldmath ${p}$}}^2/m_t-E)^{-1}$ is the zeroth order Green function, $\delta V_n \equiv [{\cal V}_n\, V_C](r)$ is the n-th order Coulomb potential. We calculated the expanded Green function semi-analytically and obtained double sum representations, which were evaluated numerically [@beneke-kiyo-schuller05].
The Green function $G(E)$ has a single pole at the bound-state energy level $E=E_n$, $$G(E) \stackrel{E\to E_n}{=}\frac{|\psi_n(0)|^2}{E_n-E-i\epsilon}
+\mbox{non-singular} \, ,
\label{nearpole}$$ where $|\psi_n(0)|^2$ and $E_n$ is the bound-state wave function squared at the origin and energy level, respectively, which has series expansion in $\alpha_s$ $$\begin{aligned}
&&\hspace{-.7cm} E_n=E_n^{(0)}\, \big[1+\sum_{i=1}^3
\left(\frac{\alpha_s}{4\pi}\right)^i e_i \, \big]\, ,
\nonumber\\
&&\hspace{-.7cm} |\psi_n(0)|^2=|\psi_n^{(0)}(0)|^2\,
\big[1+\sum_{n}^3 \left(\frac{\alpha_s}{4\pi}\right)^i f_i \, \big]\,,
\label{energy_wf}\end{aligned}$$ where $E_n^{(0)}, |\psi_n(0)|^2$ is the zeroth order result for the energy and wave function. By performing matching between expanded Green function eq.(\[expandedGF\]) and the pole structure of exact Green function eq.(\[nearpole\]), we obtained analytical result [@beneke-kiyo-schuller05] for $e_i$ and $f_i$ for the S-wave bound state at NNNLO. The expression is too lengthy to show here. In the next section we discuss phenomenological significance of the NNNLO Coulomb corrections using the obtained result.
Numerical analysis and Conclusion
=================================
Our formalism up to now is not specific to the top quark, actually it is applicable to bottom quark system replacing $m_t$ and $n_f=5$ by $m_b$ and $n_f=4$ ($n_f$ exists in the $\alpha_s$ and $\beta$-function in the Coulomb potential.) We investigate the quarkonium energy levels and $t\bar{t}$ threshold cross section in the following.
In the phenomenological analysis we use the potential-subtracted (PS) mass [@Beneke98] to make our prediction infrared renormalon free. The relation between the pole and PS masses is given by $$\begin{aligned}
&&\hspace{-.6cm}
m = m_{\rm PS}(\mu_f) -\frac{1}{2}\int_{q \leq
\mu_f}
\frac{d^3 {{\mbox{\boldmath ${q}$}}}}{(2\pi)^3}\,
\tilde{V}({\mbox{\boldmath ${q}$}})|_{\nu=\mu_f}\, ,
\label{PSmass}\end{aligned}$$ where $\mu_f$ is infrared cutoff of order $m \alpha_s$. We take into account non-Coulomb corrections for quarkonium energy levels known from literatures [@kniehl-penin99; @kniehl-penin-smirnov-steinhauser00; @penin-steinhauser02; @penin-smirnov-steinhauser05]. So the results are complete NNNLO as far as energy level is concerned.
Bottomonium masses
------------------
Using the analytical expression for the $e_i, f_i$, we obtain the relation between mass of bottomonium and bottom quark. It might be instructive to show the results using pole and PS masses: $$\begin{aligned}
&&\hspace{-.6cm } M_{\Upsilon(1S)}=2 m_b+E_1^{(0)}\big[ 1+
1.09_{\rm NLO}
\nonumber \\
&&\hspace{-.6cm}
+\big(1.42+0.36_{nC}\big)_{\rm N^2LO}
+\big(2.29+0.28_{nC}\big)_{\rm N^3LO} \big]
\nonumber \\
&&\hspace{-.6cm} = 2 m_{b,\rm PS}+E_{1,\rm PS}^{(0)} \big[ 1+
0.19_{\rm NLO} \nonumber \\
&&\hspace{-.6cm} +\big(0.07-0.23_{nC}\big)_{\rm N^2LO}
+\big(0.09-0.19_{nC}\big)_{\rm N^3LO} \big], \nonumber\end{aligned}$$ where $m_b=5$ GeV, $m_{\rm PS}(2GeV)=4.6$ GeV. The numbers are given separately for Coulomb and non-Coulomb to show numerical dominance of the former (in the pole scheme). One can see the presence of anomalously large Coulomb correction (IR renormalon) in the pole-mass scheme, while in the PS mass scheme this behavior is improved and convergence of the series became better.
We use the mass relation between $\Upsilon(1S)$ and $m_{b,{\rm PS}}$ to extract the bottom PS mass from the experimental value $M_{\Upsilon(1S)}|_{exp.}=9.460$ GeV. We obtained $$\begin{aligned}
&&\hspace{-.6cm}
m_{b,\rm PS}(2\,\mbox{GeV})
=(4.57 \pm 0.03_{\rm pert.}
\nonumber \\
&&\hspace{1cm} \pm 0.01_{\alpha_s} \pm 0.07_{\rm non-pert.}) \,\,
\mbox{GeV},
\label{finalres}\end{aligned}$$ where the subscripts denote the source of errors.
In Fig.\[fig:PSmass\] we show a scale dependence of the extracted PS mass $m_{b,{\rm PS}}(2{\rm GeV})$. Using extracted PS mass we are able to predict the masses of excited states of the spin triplet S-wave $\Upsilon$ family. In Fig.\[fig:mb2S\] we show the $\Upsilon(2S)$ mass using $m_{b,PS}(2{\rm GeV})=4.57$ GeV as a function of the renormalization scale $\mu$. One can see that the large NNNLO corrections are preferable for scale $\mu > 2$ GeV, however the prediction overshoots the experimental value $M_{\Upsilon(2S)}=10.023$ GeV at the lower scale. The naively expected natural scale for the bottomonium is $\mu=C_F\alpha_s m_{b,\rm PS}/n$ ($n$ is the principle quantum number), which is $\mu=1.23$ GeV for $\Upsilon(2S)$. This may indicate break down of perturbative computation for the excited $\Upsilon$ family. Similar behavior is observed for $\Upsilon(3S)$.
Toponium
--------
In future linear colliders remnant of toponium $1S$ resonance should be observed as a peak position of the $t\bar{t}$ cross section. This enables us to measure the $M_{t\bar{t}(1S)}$ and extract the top quark mass from the data. Here we perform an exercise, how precisely we can predict the 1S toponium mass when we fix the top quark mass as an input parameter. Adopting $m_{t,\rm PS}=175\,$GeV and $\mu=32.6\,$ GeV ($=C_F\alpha_s m_{t,\rm PS}$), we obtain $$\begin{aligned}
&& \hspace{-.6cm} M_{t\bar t(1S)}=(350+0.85+0.05-0.13+0.01)\,\,{\rm
GeV}
\nonumber \\
&&\hspace{.6cm} = 350.78\,\,\mbox{GeV}.\end{aligned}$$ Scale variation between $15 < \mu < 60 {\rm GeV}$ changes the total number only by $60$ MeV. The small higher order correction implies that precise top quark mass extraction is possible in principle, in the total cross section measurement. (There are several issues in the total cross section measurement at linear collider experiments, see for instance [@snowmass05]). To obtain [$\overline{\mathrm{MS}}$ ]{}mass, being more commonly used in high energy processes, from $M_{t\bar{t}(1S)}$ we need to know the relation between the PS and [$\overline{\mathrm{MS}}$ ]{}masses at 4-loop order, which is currently unknown. Analysis using [*large*]{}-$\beta_0$ approximation to 4-loop [$\overline{\mathrm{MS}}$ ]{}- pole mass relation and direct extraction of [$\overline{\mathrm{MS}}$ ]{}mass from $M_{t\bar{t}(1S)}$ is available from ref.[@kiyo-sumino03], which is consistent with our result.
The Coulomb wave function at the origin and Green function
----------------------------------------------------------
In this subsection we discuss the Coulomb wave function and the Green function. Since the complete NNNLO non-Coulomb corrections and Wilson coefficient of the production current in EFT are unknown, we shall discuss only the Coulomb corrections. As we demonstrated in the previous section, applicability of this method to bottomonium system is doubtful due to large NNNLO correction and slow convergence. Thus we focus on the case of toponium Coulomb wave function and Green function.
We show numerical formula for the $1S$ toponium Coulomb wave function at the origin $$\begin{aligned}
&&\hspace{-.6cm}
\big|\psi_1(0)\big|^2_C
=
\frac{(m_t C_F\alpha_s)^3}{8\pi}
\big[
1
+\alpha_s \big(-0.4333
\nonumber \\
&&\hspace{-0.6cm}
+ 3.661\,L\big)
+\alpha_s^2 \big(5.832-5.112\,L + 8.933\,L^2\big)
\nonumber \\
&&\hspace{-.6cm}
+\,\alpha_s^3 \big(-13.73
+6.446\,\ln\big(\frac{\nu}{m_t C_F \alpha_s}\big)
+ 39.72\,L
\nonumber \\
&&\hspace{-.6cm}
- 22.91\,L^2 + 18.17\,L^3\big)\,
\big]\, ,
\label{toppsi1}\end{aligned}$$ where $L=\ln(\mu/C_F\alpha_s m_t)$. To draw Fig.\[fig:wf\] we rewrote the eq.(\[toppsi1\]) using $m_{t,
PS}(20 {\rm GeV})=175$ GeV and took into account the mass correction eq.(\[PSmass\]) at given order of nonrelativistic expansion consistently. The figure shows that perturbative corrections to the Coulomb wave function is reasonably small for $\mu > 25$ GeV and the scale dependence is mild, while in lower scale $\mu < 25$ GeV the corrections are too large so the perturbative expansion is unreliable in the lower region.
In Fig.\[fig:cs\_order\] we show the $t\bar{t}$ threshold cross section as a function of CM energy for $m_{t,\rm PS} (20{\rm GeV})=175$ GeV including only the Coulomb correction successively from LO to NNNLO. The line denoted as “NNNLO exact” is a cross section obtained by numerically solving shröding equation for the Green function using NNNLO Coulomb potential. The result show a convergence of the perturbative approximation to the NNNLO-exact line.
In Fig.\[fig:cs\_mu\] we show the scale dependence of NNNLO cross section for the perturbative Green function with $\mu=15,30, 60$ GeV, and $\mu=15$ GeV for NNNLO-exact. The exact Green function is stable against scale variation (so we showed only the case $\mu=15$ GeV), while perturbative Green function (cross section) is unstable against scale variation from 15 to 30 GeV (moderate change from $\mu=30$ to 60 GeV). This is consistent with wave function analysis, where the higher order corrections were large for $\mu <25$ GeV, so we may conclude that the perturbative expansion is not reliable in lower scale region. The NNNLO-exact contains higher order insertions of the Coulomb potential to all order in eq.(\[expandedGF\]). The perturbative cross section agrees well with NNNLO-exact for large scale where it is supposed to be reliable from the wave function analysis. We believe that the NNNLO-exact cross section is reliable in wider range of $\mu$, and the perturbative cross section is reliable only in the region $\mu > 25$ GeV. Indeed we find that the multiple insertions of the Coulomb potential give large contributions to the perturbative Green function, and is slowly converging at small scale. Thus we conclude that the “correct” scale choice for the perturbative (Coulomb) cross section is $ \mu>25$, while choice of small scale may lead to misleadingly large uncertainties. We estimated yet unknown higher order Coulomb corrections should be less than 5 %.
0.4cm[**Acknowledgements**]{}\
Y.K. would like to thank M. Beneke and K. Schuller for useful discussion, and for reading this manuscript. Y.K. thanks J. Kodaira for invitation to the RADCOR05 conference.
[99]{}
M. Beneke, Y. Kiyo and K. Schuler, Nucl. Phys. [**B714**]{} (2005) 67. A. Pineda and J. Soto, Nucl. Phys. Proc. Suppl. [**64**]{} (1998) 428 \[hep-ph/9707481\]. M. E. Luke, A. V. Manohar and I. Z. Rothstein, Phys. Rev. [**D 61**]{} (2000) 074025. A. H. Hoang et al., Eur. Phys. J. direct [**C 2**]{} (2000) 1, and references therein.
B. A. Kniehl and A. A. Penin, Nucl. Phys. [**B 563**]{} (1999) 200. B.A. Kniehl, A.A. Penin, V.A. Smirnov and M. Steinhauser, Nucl. Phys. [**B 635**]{} (2002) 357. N. Brambilla, A. Pineda, J. Soto and A. Vairo, Nucl. Phys. [**B 566**]{} (2000) 275. M. Beneke, Phys. Lett. B [**434**]{} (1998) 115. A. A. Penin and M. Steinhauser, Phys. Lett. [**B 538**]{} (2002) 335. A. A. Penin, V. A. Smirnov and M. Steinhauser, Nucl. Phys. [**B 716**]{} (2005) 303. A. Juste et al., Report of the 2005 Snowmass Top/QCD Working Group \[hep-ph/0601112\].
Y. Kiyo and Y. Sumino, Phys. Rev. [**D 67**]{} (2003) 071501
[^1]: Supported by the DFG Sonderforschungsbereich/Transregio 9 “Computer-gestützte Theoretische Teilchenphysik”.
[^2]: PITHA 06/03. Talk based on Ref.[@beneke-kiyo-schuller05]. To appear in the proceedings of the 7th International Symposium on Radiative Corrections (RADCOR05), Shonan Village, Japan Oct. 2005
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We report a detailed study of the temperature and composition dependence of the resistivity and thermopower for a series of bismuth chalcogenides Bi$_2$Te$_{3-x}$Se$_x$. The temperature dependence of the thermopower can be semi-quantitatively described by a simple model for an extrinsic semiconductor. We show that, by substituting selenium for tellurium, the Fermi level can be continuously tuned from the valence band into the conduction band. The maximum values of the thermopower, bulk band gap as well the activation energy are found for $x \approx 1$.'
author:
- Ana Akrap
- Alberto Ubaldini
- Enrico Giannini
- László Forró
bibliography:
- 'BiTeSe.bib'
title: 'Following the Fermi level in the Bi$_2$Te$_{3-x}$Se$_x$ series through thermopower'
---
Layered bismuth chalcogenides Bi$_2$Te$_{3-x}$Se$_x$ are a family of narrow-band semiconductors, well known since the 1950’s for their exceptional thermoelectric properties.[@goldsmid64] Bi$_2$Te$_3$ has the highest known thermoelectric figure of merit $zT$ at room temperature, and is widely used in room-temperature thermoelectric applications.
Recently, bismuth chalcogenides were rediscovered within the novel context of topological insulators. They were among the first identified three-dimensional topological insulators.[@zhang09; @wang11] The surface states in Bi$_2$Te$_{3-x}$Se$_x$ are simple since only one Dirac cone crosses the band gap.[@hasan09; @valla12] However, in spite of a relatively large band gap ($200 - 300$ meV), bulk conductivity of known bismuth chalcogenides is too high and hinders the observation of topological surface states, even in the most compensated compound Bi$_2$Te$_2$Se.[@ren10; @akrap12] Throughout this series, the Fermi level is found either within the conduction or valence band.
In this paper we study the detailed thermoelectric properties of bismuth chalcogenides to clarify the relation between the composition and position of the Fermi level. We investigate ten different compositions of the Bi$_2$Te$_{3-x}$Se$_x$ series, where Se content is tuned from $x=0$ to 3. We find that thermopower $S(T)$ may be semi-quantitatively explained by the standard extrinsic semiconductor model.[@goldsmid64] Most importantly, thermopower shows that tuning Te/Se content leads to a steady shift of chemical potential, from the valence band in the p-type Bi$_2$Te$_3$, to the conduction band in the n-type Bi$_2$Se$_3$. The highest value of the thermopower is found at $x=1$ and it corresponds to the lowest Fermi temperature.
Single crystals of Bi$_2$Te$_{3-x}$Se$_x$ were grown by the floating zone method from the stoichiometric ratio of metallic bismuth and chalcogenide elements.[@Alberto_growth] The unit cell of a Bi$_2$Te$_{3-x}$Se$_x$ compound consists of quintuple Te/Se–Bi–Te/Se–Bi–Te/Se layers stacked along the [*c*]{}-axis direction.[@nakajima63; @bayliss91] The quintuple layers are bound by weak van der Waals interaction. For most compositions, crystals readily cleave and expose shiny surfaces. We investigated ten different compositions, given by the Se content $x=0,0.6,0.9,0.95,1,1.3,1.5,2$ and 3. The resistivity and thermopower were measured simultaneously using a custom setup. Crystals were cleaved and cut into thin bars of approximately $1.5 \times 0.5 \times 0.02$ mm$^3$. The resistivity was measured using a standard four-probe technique. Thermopower was determined using a $dc$ method.The sample was fixed to a ceramic surface with a small heater attached next to it, and the thermal gradient was determined using a chromel-constantan differential thermocouple. The bulk band gap was determined through transmission and reflection measurements on cleaved thin flakes, using a Bruker Hyperion microscope.[@Alberto_growth]
![(Color online) (a) Resistivity at high and low temperature for varying composition, from Bi$_2$Te$_3$ to Bi$_2$Se$_3$. A maximum $\rho$ occurs close to Bi$_2$Te$_2$Se. (b) Temperature dependence of resistivity for several different compositions (Selenium stoichiometries), $x$. (c) Bulk band gap is shown for the whole series. \[fig:rho\_xT\]](Rho_all.pdf){width="\columnwidth"}
The behavior of the resistivity throughout the series is shown in Fig. \[fig:rho\_xT\](a) at room temperature and at 10 K. A pronounced maximum in $\rho$ occurs around $x=0.9$. A more detailed temperature dependence of the resistivity is shown in Fig. \[fig:rho\_xT\](b) in an Arrhenius plot, for several different compositions $x$. While most samples exhibit a weakly metallic resistivity, we observe an activated behavior for $x=0.9,0.95,1$ and 1.3. This is precisely in the $x$ range where $\rho(x)$ has maximum value. However, the activation energies are small and range from $\sim 4.5$ meV for $x=0.95, 1$ and 1.3, to $\sim 40$ meV for $x=0.9$.
The stoichiometry dependence of the bulk band gap is shown in Fig. \[fig:rho\_xT\](c). It was determined through the optical reflection and transmission measurements by following the energy of the onset of the Fabry-Perot interferences, similar to the previous high pressure studies of bismuth chalcogenides.[@vilaplana11a; @segura12; @akrap12] The details of the measurement will be published elsewhere.[@Alberto_growth] The bulk band gap spans from approximately 200 meV in Bi$_2$Te$_3$ to 300 meV in Bi$_2$Se$_3$, and reaches a maximum of 325 meV in Bi$_2$Te$_{2.1}$Se$_{0.9}$. It is interesting to note that the maximum band gap happens for the same stoichiometry ($x\approx 1$) for which resistivity is maximum. While the band gap is determined by the band structure, the transport gap is linked to the defects which lead to impurity doping. Recent theoretical work by Skinner [*et al.*]{} shows there is a relation between these two quantities, and the activation energy of the resistivity in a compensated topological insulator is proportional to the band gap.[@skinner12]
The [*dc*]{} transport activation energy never exceeded 40 meV, which points to the fact that Bi$_2$Te$_{3-x}$Se$_x$ compounds are always doped by impurities. The impurity levels are particularly evident at low temperatures, where $\rho(T)$ shows very little temperature dependence (Fig. \[fig:rho\_xT\](b)). Small activation energies accompanied by an almost constant $\rho$ at low temperatures suggest that the donor or acceptor levels within the bulk gap are important contributors to the conduction. Impurity levels broaden into a band which overlaps with the valence or conduction band. This impurity band probably lies at the edge of the conduction or valence band.[@goldsmid12]
There are several reasons why impurity doping inevitably occurs in this series. Bismuth and tellurium have a similar Pauling electronegativity.[@gaudin94]ÊBismuth atoms therefore occupy tellurium sites and make for a high concentration of negatively charged defects, which is why Bi$_2$Te$_3$ crystals are most often $p$-type.[@kulbachinskii01] Selenium, on the other hand, has a high fugacity, and the accumulation of selenium vacancies causes the $n$-type behavior of Bi$_2$Se$_3$.[@plechacek02] For the intermediate stoichiometries, both of the above types of disorder, vacancies, and antisite defects, are present and lead to intrinsic doping. In reality, when the the composition of the compound is tuned from Bi$_2$Te$_{3+\delta}$ to Bi$_2$Se$_{3-\delta}$, one passes from a system with excess Te, towards a Se-deficient compound. Around the composition $x=1$, the Se/Te sublattice is expected to be ordered which significantly reduces the number of defects in the structure, decreasing conductivity.
A very reliable method to follow the progression from $p$ to $n$-type conductor is by measuring thermopower. This is particularly simple in a situation where there is only one relevant electronic band. Fig. \[fig:TEP\](a) shows the dependence of thermopower on Se content, $x$, at several different temperatures. There is a very sharp sign change taking place between $x=0.95$ and $x=1.0$. For $x \geq 1$ the thermopower becomes negative, meaning that there is a transition from $p$-type to $n$-type conduction. This transition is illustrated within inset of Fig. \[fig:TEP\](a) cartoon.
![(Color online) The dependence of thermopower on Selenium stoichiometry is shown in (a), at five different temperatures. (b) The experimental temperature dependence for nine different stoichiometries (given by Selenium content $x$). (c) Results of a simple model calculation given by Eq. \[eq:TEP\]. The Fermi level is varied between 50 K and 900 K in steps of 50 K (from 4.3 to 77.6 meV). Note the different scales in (b) and (c). \[fig:TEP\]](TEP_all.pdf){width="\columnwidth"}
For thermoelectric applications, it is important to optimize the figure of merit $zT$. In applications that do not require maintaining a temperature gradient, the power factor $S^2/\rho$ should be maximized. The power factor ranges between 10 to 50 $\mu$W/(K$^2$cm) for the series of the samples we study here. These values are similar to those of Sokolov [*e*t al]{}, and display a similar doping dependence.[@sokolov07] Maximum power factor of 50 $\mu$W/(K$^2$cm) takes place for $x=2$, and minimum 10 $\mu$W/(K$^2$cm) occurs for $x=1$ where both the thermopower and the resistivity are maximum.
Fig. \[fig:TEP\](b) shows a detailed temperature dependence of thermopower $S(T)$ for nine different compositions $x$. All of the samples are either $n$ or $p$ type. The only exception is the $x=0.9$ sample, which is $p$-type at high temperatures, and $n$-type below $\sim 50$ K. Leaving the $x=0.9$ composition aside, let us focus on the temperature dependence of thermopower in the other nine compounds.
The thermopower of a heavily-doped semiconductor on the metallic side of the metal-insulator transition is given by:[@goldsmid64] $$S=\pm \frac{k_B}{e} \left[ \eta_F - \frac{(r+5/2)F_{r+3/2}(\eta_F)}{(r+3/2)F_{r+1/2}(\eta_F)} \right]
\label{eq:TEP}$$ Here $\eta_F=E_F/(k_B T)$ is the reduced Fermi energy; the parameter $r$ describes the energy dependence of the scattering time, and $$F_{n}=\int_0^\infty {\rm d} \eta \frac{\eta^n}{1+ \exp \left( \eta - \eta_F \right)}
\label{eq:TEP}$$ is the Fermi integral. $E_F$ is measured from the bottom of the conduction band for the $n$-type, or the top of the valence band in the case of a $p$-type semiconductor. To describe the energy dependence of the scattering time parameter, we take the standard value[@goldsmid64] $r=-0.5$ for acoustic phonon scattering and calculate $S(T)$ assuming a set of temperature-independent values for $E_F$. A more correct method would be to determine $E_F$ from the charge balance constraint,[@sales10] but since the charge doping comes from unintentional impurities, a precise amount of carriers is unknown.
For Bi$_2$Se$_3$, the carrier density $n$ may be estimated by comparing the resistivity data in Fig. \[fig:rho\_xT\] to the results reported by Butch [*et al.*]{}[@butch10], which gives a result of $n\sim 10^{18} -10^{19}$ cm$^{-3}$.
Results for the calculated temperature dependence of thermopower are shown in Fig. \[fig:TEP\](c) for $E_F/k_B$, ranging from 50 K to 900 K (corresponding to the $E_F$ between 4.3 and 77.6 meV), in steps of 50 K. This ballpark for $E_F$ seems to be in fair agreement with ARPES results from some members of the series. For example, Hor [*et al.*]{}[@hor09] observed a Fermi level of 50 meV in Bi$_2$Se$_3$ and Chen [*et al.*]{}[@chen09] saw $E_F\approx45$ meV in Bi$_2$Te$_3$.
The measurements of quantum oscillations for Bi$_2$Se$_3$ show that the fundamental state of the system is metallic and that carrier density can be tuned by several orders of magnitude via controlling the stoichiometry.[@fauque12] When changing the carrier density from $10^{17}$ to $10^{18}$ cm$^{-3}$, the Fermi temperature moves from 80 K to 1100 K. This is in good agreement with the approximate value of 900 K suggested for our Bi$_2$Se$_3$ sample by the calculated series in Fig. \[fig:TEP\](c).
The simple semiconductor model given by the Eq. \[eq:TEP\] captures our experimental results quite well. The higher values of $E_F$ lead to a thermopower almost linear in temperature, very similar to the one measured in Bi$_2$Te$_3$ and Bi$_2$Se$_3$. A smaller $E_F$ leads to a thermopower with a distinct change in slope for $k_B T<E_F$. Such a shape of $S(T)$ is similar to what we measure for $x\approx 1$. However, the calculated values of $S$ are smaller than the measured thermopower. This may be due to additional contributions to the thermopower not taken into account by the model, such as the phonon drag, or because the parameter $r$ is not simply $-0.5$, but has a temperature and composition dependence.[@kulbachinskii12; @kulbachinskii97]
In the calculated $S(T)$ curves, the temperature slope changes at temperatures $T \sim E_F/k_B$. With this in mind, one might conclude from Fig. \[fig:TEP\](b) that samples with $x=0.6, 0.9$, and 1.0 all have $E_F\sim 4$ meV. On the contrary, Fermi levels in pure Bi$_2$Te$_3$ and Bi$_2$Se$_3$ should be above room temperature since no clear change in slope is visible from our thermopower data.
For $x=0.9$ (Bi$_2$Te$_{2.1}$Se$_{0.9}$), $S(T)$ seems to be an exception. The temperature dependence is different than in the rest of the series. $S$ rises above 400 $\mu$V/K at 100 K, and then precipitously drops below 100 K. This is the only composition for which the sample is $p$-type at high temperatures and $n$-type at low temperatures, suggesting that the charge carriers exist in more than one band. This sample also has the highest resistivity within the series, as shown in Fig. \[fig:rho\_xT\](b), and among the highest bulk band gaps ($\sim 325$ meV). This more complex behavior may possibly be linked to additional impurity bands.
To summarize, we have presented a systematic study of the composition and temperature dependence of thermopower within the bismuth chalcogenide Bi$_2$Te$_{3-x}$Se$_x$ series. The maximum resistivity, thermopower, and bulk band gap occur with compositions where $x \approx 1$. For this composition, the Te/Se sublattice is maximally ordered, which reduces the defects in the crystal lattice and diminishes impurity doping.[@ren10]
The value of thermopower at a given temperature can be tuned by changing the stoichiometry. Starting from the $p$-type Bi$_2$Te$_3$, upon replacing a part of Te atoms with Se, thermopower is strongly enhanced, and with the composition Bi$_2$Te$_2$Se thermopower changes its sign. As the amount of Se is increased, the value of thermopower decreases. The temperature dependence of thermopower may be semi-quantitatively described by a simple model for an extrinsic semiconductor. Here, the Fermi level $E_F$ decreases from Bi$_2$Te$_3$ towards Bi$_2$Te$_{2.05}$Se$_{0.95}$, reaching a minimum of approximately 4 meV. When more Se is replaced for Te, the Fermi level $E_F$ starts increasing again. Both the bulk band gap and the Fermi level can be tuned by controlling Se/Te stoichiometry.
We would like to thank Kamran Behnia, Brian C. Sales and Nathaniel Miller for helpful advice and comments. Research was supported by the Swiss NSF through grant No. 200020-135085 and its NCCR MaNEP. A.A. acknowledges funding from “Boursières d’Excellence” of the University of Geneva.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present a short atlas illustrating the unusual Ca [ii]{} K-line profiles in upper main sequence stars with anomalous abundances. Slopes of the profiles for 10 cool, magnetic chemically peculiar (CP) stars change abruptly at the very core, forming a deep “nib." The nibs show the same or nearly the same radial velocity as the other atomic lines. The near wings are generally more shallow than in normal stars. In three magnetic CP stars, the K-lines are too weak to show this shape, though the nibs themselves are arguably present. The Ca [ii]{} H-lines also show deep nibs, but the profiles are complicated by the nearby, strong H$\epsilon$ absorption. The K-line structure is nearly unchanged with phase in $\beta$ CrB and $\alpha$ Cir. Calculations, including NLTE, show that other possibilities in addition to chemical stratification may yield nib-like cores.'
author:
- 'C. R. Cowley,'
- 'S. Hubrig,'
- 'I. Kamp'
title: |
An Atlas of K-line Spectra for Cool Magnetic CP Stars\
The Wing-Nib Anomaly (WNA)
---
Introduction
============
Abundance studies of chemically peculiar (CP) stars of the upper main sequence are usually based on classical, plane-parallel atmospheres. Until recently, there was little evidence that this assumption would lead to significant errors, apart from the obvious case of spectrum variables. Even in these cases, it was common to assume plane-parallel structure for localized regions (abundance patches) of the photosphere.
Recent work on the ionization equilibrium [@ryab04] and the Balmer profiles [@cowl01] of cooler CP stars in the magnetic sequence has drawn attention to significant departures from classical atmospheric structure.
Decades ago, spectroscopists had noted another indication of departures from a classical atmosphere in the Ca [ii]{} K-lines. [@baba58] describes very complicated K-line profiles in a number of CP stars. In a few cases, variable profiles suggested to him “ejection of ionized clouds or streams such as the sun produces." Analytical spectroscopists did little more than note these remarks at the time. Most of the interest in CP star focused on their peculiar abundances and attempts, for example, by @conti65 and @vveer66 to explain the abundance anomalies in Am stars by unorthodox models had been unsuccessful.
New, high-resolution observations make it appropriate to take a closer look at the K-lines of CP stars. We confine ourselves to spectra of cooler magnetic stars with minimal rotational broadening. We find that the K-lines of these stars generally exhibit a “wing-nib anomaly” (WNA).
Two forms for the WNA are shown schematically in Fig. \[fig:one\]. In the upper part of the figure, the wings show a gently changing slope, which appears to nearly level off prior the deep minimum. In the lower part of the figure, the wings show an approximately constant slope beyond the deep nib. In both cases, the region outside of the nib has a higher intensity than is observed for normal stars, or the Am star HR 1353. Individual examples are cited in the captions of figures to follow.
The Ca II H-line also shows a sharp, deep nib, but the structure is complicated by the presence of the strong Balmer H$\epsilon$ line. We shall not discuss the H-line in this paper.
Previous K-line Work
====================
Complications in the core of the Ca [ii]{} K-line (as well as the H-line) are well known in stellar astronomy. The most common phenomena are single and double reversals (e.g. K$_2$, and K$_3$). These reversals are common among F-type stars, and may be seen, weakly, in the central cores of several non-CP stars discussed later in the paper. Additionally, there are numerous examples of sharp, interstellar components, which typically have different radial velocities than the stellar lines.
It is likely that [@babb58] saw the WNA among a variety of other phenomena that influence the K-line structure in CP stars. However, his papers give verbal descriptions, not detailed profiles.
An important observational step was taken by Babel (1994). He measured two K-line “equivalent widths,” a $W_{0.3}$ within $\pm 0.3$ Å of the line center, and a $W_{10}$ within $\pm 10$ Å. He showed that for a given $W_{10}$, the $W_{0.3}$ was smaller for Ap stars than for normal stars. In other words, the core absorption of Ap stars was smaller than for normal stars with comparable overall K-line strengths. Note this does not of itself indicate the presence of a nib. Clearly nibs, when present, increase the $W_{0.3}$ absorption rather than reduce it. The $W_{0.3}$/$W_{10}$ trend found by Babel persists in spite of the nibs rather than because of them.
The current atlas elaborates these findings for the small class of Ap stars. Our UVES spectra have twice or more typically four times the resolution employed by Babel. Finally, because we deal with profiles rather than equivalent widths, we show in detail how the K-line absorption in peculiar [*and normal*]{} stars differs.
Ca K-line nibs may indeed be seen for three stars in the papers by Babel (1992, 1994) and @ryab02 53 Cam, $\gamma$ Equ, and $\beta$ CrB. None of these papers contrast the nib-structure with the Ca K [*profiles*]{} of normal stars.
Observations
============
All spectra were obtained with the VLT UV-Visual Echelle Spectrograph UVES at UT2. The major part of the spectra of magnetic CP stars was obtained in the course of our study of chemical abundances of CP stars (ESO programmes Nos. 70.D-0470, 072.D-0414, and 074.D-0392). A few other spectra of CP stars were retrieved from the ESO UVES archive (ESO programme No. 68.D-0254). Spectra of some other stars were downloaded from the UVESPOP web site [@bag04]. Details are summarized in Table \[tab:dates\]; spectral types are from @ren91 and SIMBAD. The stars are all bright, and signal-to-noise ratios greater than 200 were typical. In a few cases where multiple scans were available, the spectra were averaged, weighted by the number of counts.
All spectra were observed with Dichroic standard settings at the resolving power of $\lambda{}/\Delta{}\lambda{} \approx 0.8\times10^5$ and have been reduced by the UVES pipeline Data Reduction Software, which is an evolved version of the ECHELLE context of MIDAS.
The 13 CP stars presented here were selected on the basis of (1) their chemical peculiarity, (2) their low values of the projected rotational velocity ($v\cdot\sin(i)$), and (3) the availability of high-resolution, high signal-to-noise spectra. All stars have effective temperatures between approximately 6600 and 9600K. Of these, 10 show the WNA. The three stars without a WNA have Ca K profiles that are too weak to have extended wings. In the case of two of these stars, HR 5623 and HR 7552, the stars are too hot to have strong K-lines. The K-line is also too weak to show the WNA in Przybylski’s star. Przybylski’s star is the coolest in our sample, but the K-line is weak because of the low calcium abundance.
The magnetic Ap stars in Figs. \[fig:cold\] and \[fig:hot\] are arranged roughly in order of effective temperature. The bottom spectra of Fig \[fig:hot\] are of the Am star, HR 1353 (above) and the F2 V $\sigma$ Boo (below). Among these spectra, Figs. 3A and especially 3B most closely resemble the upper schematic profile of Fig. 1. Figs. 2C, 2E, and 2F are examples of the lower profile of Fig. 1.
Fig. \[fig:misc\] shows two hotter magnetic Ap stars (A and B), and the cool roAp, Przybylski’s star (C). In all three cases, there is a deep, sharp, core, but the wings are too weak to provide the contrast we refer to as the wing-nib anomaly. In the case of Przybylski’s star, low-dispersion spectra show broad, shallow absorption in the region of the H+H$\epsilon$ and the K-line. This absorption is not obvious in Fig. \[fig:misc\]. Nevertheless, one may say that Przybylski’s star also shows a WNA, though the proportions of the near wing and nib are quantitatively different from those of the ten other cool magnetic CP stars illustrated in Figs. 2 and 3.
HD 965 (Fig. 4D) has an intermediate temperature, probably near 8000K. @bord03 discuss the difficulties in finding a temperature for this star. The lower plot (Fig. 4E) is an LTE calculation using Michigan codes for a star with an effective temperatureof 6750K, $\log(g) = 4.0$, and enhanced abundances as in a magnetic CP star.
Full nib widths, as defined in Fig. \[fig:one\], are given in Table \[tab:nibwidths\]. The average in angstroms is $0.42\pm 0.06$. Provisional calculations show these nibs are formed several thousand kilometers above $\tau_{\rm 5000} = 1$.
Phase Variations
================
The WNA would be washed out in the spectrum of a star with significant rotational velocity. Thus, the spectra presented have all had sharp lines, and the stars themselves relatively long rotational periods. In Figs. \[fig:acir\_op\] and \[fig:bcrb\_op\] we present overplots of the K-line region for two stars with relatively shorter periods, $\alpha$ Cir and $\beta$ CrB. Small variations are seen in both spectra, but the general shape of the WNA persists.
The relative constancy of these profiles might be compared with those of Babel’s (1992) Fig. 5 for 53 Cam, where considerable phase variation is seen. The 53 Cam spectra appear to show WNA at phases 0.17 and 0.29. These fundamental differences in phase behaviour show the importance of having bigger samples for the study of the WNA. Any model that attempts to explain this anomalous line profiles has to account for variable as well as non-variable line profiles.
Non-CP stars
============
We present K-line region spectra of several non-CP stars in Fig. \[fig:non\], to make clear the distinction of WNA profiles. The upper spectrum is of the well-studied subdwarf, HD 140283. The star has an effective temperature close to that of the sun, but the lines are weak because of the very low metal-to-hydrogen ratios. Spectrum B is of $\gamma$ Dor, a pulsating variable related to the $\delta$ Scuti stars, but surrounded by a dust envelope [@balon94]. The bottom three spectra are of stars with MK types, though all are metal poor; 68 Eri is F2 V, 53 Vir is F6 V, and $\gamma$ Pav is F6 V.
Discussion
==========
It is our purpose to present an observational atlas and not a new interpretation of the wing-nib anomaly. We do not dispute stratification models. They are demonstrably capable of producing nibs. Nevertheless, a few remarks are appropriate.
The wing-nib and core-wing (Balmer lines) anomalies show that the atmospheres of at least some of the cool CP stars depart significantly from the classical models. This was suggested decades ago as an alternate explanation to abundance anomalies—a way to explain the peculiar spectra with normal abundances. The objections were dismissed forcefully by @sarg66. Briefly, he argued that the CP stars had normal colors, curves of growth, ionization temperatures, and Balmer profiles. His arguments were valid at that time though they are untenable today. Nevertheless, his conclusion is widely accepted, that the photospheres of magnetic CP stars are chemically anomalous.
Traditional abundance studies were based on classical, plane-parallel atmospheres, with a uniform chemical composition. @bab92 showed that a stratified abundance structure would produce a Ca K-line nib while Ryabchikova et al. (2002) used a chemically stratified model similar to Babel’s. They also obtained a K-line nib, though they did not fit it precisely. They found certain other atomic lines that could be fit more satisfactorily with the assumption of abundance stratification than with a traditional model.
Much new work has been devoted to this phenomena. Indeed, @dwor05 described the session of IAU Symposium 224 [@zver05] on diffusion by saying “the keword ...was ‘stratification.’" @kochu02 showed that the core-wing anomaly of the Balmer lines might be explained by an ad hoc variation in the $T$–$\tau$ relation. We have made LTE calculations to show that wing-nib structure of Ca [ii]{} K may also be obtained with a modified $T$–$\tau$.
Most of the calculations relevant for the WNA have been made in LTE. We have begun calculations with the Kiel non-LTE code (Steenbock & Holweger 1984) that will be reported in more detail elsewhere. We use the updated version of the Ca model atom by @wat85 [@stur93]. The core of the Ca [ii]{} K-line forms in layers of the atmosphere, where both involved levels show only very moderate deviations from LTE ($<0.05$ dex). Preliminary results show that these small quantitative differences from LTE in the K-line core vary somewhat with effective temperature. As an example, we present here a cool atmosphere (T$_{\rm eff}=6750$ K, $\log g=4.0$) with an ad-hoc modified $T$–$\tau$ relation (see inset of Fig. \[fig:bugger\]). The atmosphere temperature was raised to $T\sim 6000$ K over the optical depth range $-4.5 < \log \tau < -0.5$. Fig. \[fig:bugger\] shows the resulting LTE (solid line) and NLTE (dashed line) Ca[ii]{}K profiles. NLTE leads to a $\sim$ 5% weaker core, but does not affect the main profile and the wings; thus, it would not lead to significant abundance differences. At present it seems that a fully non-LTE approach could modify details of a workable model, but would not require qualitative changes.
Neither of the above modifications takes account of the influence of magnetic pressure on atmospheric structure. Thus, it is reasonable to explore a third plausible departure from a classical model atmosphere, based on a consideration of magnetohydrodynamical effects. We lack the information necessary to take such effects explicitly into account. One possibility is that they simulate a model with gravity that decreases outward [@conti65; @valy04]. Since a model with reduced gravity in the upper layers would have lower pressured and therefore reduced atomic absorption, the effect should resemble that of stratified models. We have made provisional calculations showing that in LTE, such a model is indeed capable of yielding a K-line nib. Thus far, the modified-gravity models that give K-line nibs have not yielded core-wing anomalous Balmer profiles.
CRC acknowledges useful conversations with H. A. Abt, W. P. Bidelman, D. J. Bord, and P. Hughes. We thank an anonymous referee for calling our attention to the important studies of Babel. Thanks are also due to the ESO staff for the UVESPOP public data archive. This research has made use of the SIMBAD database, operated at CDS, Strasbourg, France.
[*Facility:*]{}
Babcock, H. W. 1958a, , 128,228. Babcock, H. W. 1958b, ApJS, 3,141. Babel, J. 1992, A&A, 258, 449. Babel, J. 1994, A&A, 283, 189. Bagnulo, S., Jehin, E., Ledoux, C., et al. & the ESO Paranal Science Operations Team 2004, http://www.sc.eso.org/santiago/uvespop/ (see also ESO Messenger, 114, 10, 2003). Balona, L. A., Krisciunas, K. & Cousins, A. W. J. 1994, , 270, 905. Bord, D. J., Cowley, C. R., Hubrig, S., & Mathys, G. 2003, , 35, 1357. Conti, P. S. 1965, , 11, 47. Cowley, C. R., Hubrig, S., Ryabchikova, T. A., et al. 2001, , 367, 939. Dworetsky, M. M. 2005, in [*The A-Star Puzzle*]{}, IAU Symp. 224, ed. J. Zverko, J. Žižňovský, S. J. Adelman, and W. W. Weiss (Cambridge: University Press), p. 499. Kochukhov, O., Bagnulo, S., & Barklem, P. S. 2002, , 578, 75. Kurtz, D. W. 1989, , 238, 261. Kurtz, D. W., Sullivan, D. J., Martinez, P. & Tripe, P. 1994, , 270, 674. Moore, C. E. 1945, Cont. Princeton Univ. Obs., No. 20. Renson, P., Gerbaldi, M., & Catalano, F. A. 1991, Supp., 89, 429. Ryabchikova, T., Nesvacil, N., Weiss, W. W., Kochukhov, O., & Stütz, Ch. 2004, , 423, 705. Ryabchikova, T., Piskunov, N., Kochukhov, O., Tsymbal, V., Mittermayer, P., & Weiss, W. W. 2002, , 384, 545. Sargent, W. L. W. 1966, in [*Abundance Determinations in Stellar Spectra*]{}, (ed.) H. Hubenet, IAU Symposium, 26. Steenbock, W., & Holweger, H. 1984, , 130, 319. Stürenburg, S. 1993, , 277, 139. Valyavin, G., Kochukhov, O., and Piskunov, N. 2004, A&A, 420, 993. van’t Veer, C. 1966, in [*Abundance Determinations in Stellar Spectra*]{}, IAU Symposium 26, ed. H. Hubenet (London: Academic Press), p. 261. Note the remarks by Cayrel following this paper. Watanabe, T.,& Steenbock, W. 1985, A&A, 149, 21. Zverko, J., Žižňovský, J., Adelman, S. J., and Weiss, W. W. 2005, [*The A-Star Puzzle*]{}, IAU Symposium 224 (Cambridge: University Press).
-- -------------- ------------ ------ -----------
BD +0 21 18/09/2002 A8p Sr
HR 1217 13/03/2001 A9 SrEuCr
$\gamma$ Dor 19/09/2002 F4 III
HR 1353 14/08/2002 A3m
68 Eri 21/09/2002 F2 V
Przybylski’s 11/01/2003 Fp REE
53 Vir 27/07/2001 F6 V
BD -17 3829 06/02/2003 F0 SrCrEi
BD +6 2827 21/01/2002 F0p
$\sigma$ Boo 13/03/2001 F2 V
$\alpha$ Cir 25/01/2005 0.94 A3-A9
27/01/2005 0.19
01/02/2005 0.40
05/02/2005 0.50
HR 5623 26/02/2002 A0 SrCr
BD -13 4081 26/02/2002 F2 SrEuCr
$\beta$ CrB 03/02/2004 0.30 A9 SrEuCr
03/26/2004 0.60
03/27/2004 0.65
03/29/2004 0.76
04/01/2004 0.92
33 Lib 20/09/2002 F0p SrEuCr
BD -10 4149 08/07/2001 sdF3
10 Aql 08/10/2001 A6 Sr
HR 7552 07/10/2001 A0 EuCrSi
HR 7575 08/10/2001 A6 SrCrEu
$\gamma$ Pav 06/12/2002 F6 V
BD -12 6357 08/10/2001 A7 SrCrEu
CD =45 14901 21/09/2002 A5 SrEuCr
-- -------------- ------------ ------ -----------
: Dates of Observations[]{data-label="tab:dates"}
star(HD) width\[Å\] star(HD) width\[Å\]
---------- ------------ ---------- ------------
965 0.40 137949 0.45
24712 0.39 176232 0.33
116114 0.40 188041 0.35
122970 0.46 216018 0.39
137909 0.45 217522 0.53
: Measured Nib (Full) Widths[]{data-label="tab:nibwidths"}
![Schematic profiles for the cool, magnetic CP stars (heavy line). Specific examples of both kinds of profiles are illustrated in subsequent figures. Vertical lines on either side of the v-shaped “nib” indicate the nib width. Numerical values are given in Table \[tab:nibwidths\] below. The underlying stellar spectrum is of the F2 V star $\sigma$ Boo. []{data-label="fig:one"}](f1.eps){width="30.00000%"}
![The K-line spectrum of $\alpha$ Cir at phases 0.19, 0.40, 0.50, and 0.94. These are calculated from the rotational period of 4.463 days [@kurtz94]. Zero phase is JD 2449111.6852. []{data-label="fig:acir_op"}](f5.eps){width="30.00000%"}
![The K-line spectrum of $\beta$ CrB at phases 0.30, 0.60, 0.65, 0.76, and 0.92. These are calculated from the rotation period of 18.4868 days [@kurtz89]. Zero phase is JD 2434204.7. []{data-label="fig:bcrb_op"}](f6.eps){width="30.00000%"}
| {
"pile_set_name": "ArXiv"
} |
psfig =cmr12 [H]{} §[[S]{}]{} = 1
[**Section 0: Introduction.**]{} One strategy for understanding a dynamical system is to first isolate invariant sets that are dynamically indecomposable. One then studies the structure of these pieces and how they fit together to give the global dynamics. This idea goes back at least to Birkhoff and has a particularly clear expression in Conley’s Morse decompositions.
There are many notions of dynamical indecomposibility in the literature. In this paper we consider a fairly strong one that uses both topology and measure. An invariant set is called [*strictly ergodic*]{} if it is both minimal (every orbit is dense) and uniquely ergodic (existence of a unique, invariant Borel probability measure). These properties are preserved under topological conjugacy but [*not*]{} measure isomorphism.
The simplest such invariant sets are periodic orbits, and there are many theorems concerning their existence. The next simplest strictly ergodic systems are probably rigid rotations on the circle with irrational rotation number and the closely related Denjoy minimal sets. Elements of these invariant sets are sometimes called (generalized) quasi-periodic points. The models for Denjoy minimal sets are the minimal sets in nontransitive circle homeomorphisms with irrational rotation number. An abstract dynamical system is called a [*Denjoy minimal set*]{} if it is topologically conjugate to such a model. One of the questions that motivated this paper is what kind of properties of periodic orbits are also true for more general strictly ergodic invariant sets, in particular, for Denjoy minimal sets?
One way to begin to address this question is to collect these invariant sets into spaces. For a fixed homeomorphism $f$ of a compact metric space $X$, let $\S(X,f)$ denote the set of all strictly ergodic $f$-invariant subsets of $X$. Since different minimal sets are of necessity disjoint, each point in $\S(X,f)$ represents a minimal set that is disjoint from every other minimal set. A strictly ergodic set supports a unique invariant Borel probability measure, so we may use these measures with the weak topology to put a topology on $\S(X,f)$. If $\D(X,f)$ denotes the set of $f$-invariant subsets that are Denjoy minimal sets, then $\D(X,f)\subset\S(X,f)$, so we may use the weak topology on $\D(X,f)$ also.
In (\[M\]), Mather shows that for a area-preserving monotone twist map of the annulus, $f:A\rightarrow A$, the nonexistence of an invariant circle with a given irrational rotation number implies the existence of numerous Denjoy minimal sets with that rotation number. More precisely, using the notation just introduced, $\D (A, f)$ contains topological balls of every finite dimension. From one point of view this is a very surprising result. One has an arbitrarily large dimensional family of minimal sets embedded in a two-dimensional dynamical system. Another question that motivated this paper is how common is this kind of phenomenon in dynamics on finite dimensional manifolds?
It is important to note that even for a smooth system, $\S(M,f)$ can be empty. One example of this is Furstenberg’s $C^\omega$-diffeomorphism of the two torus that is minimal but [*not*]{} strictly ergodic (\[F\]). However, for the full shift on two symbols $(\Sigma_2, \sigma)$ one has:
The basic tool in the proof of this theorem is the main construction. This construction takes a certain type of open set in the circle (a regular one) and produces a compact, invariant set in the full two-shift. The construction uses the open set to produce itineraries with respect to a rigid rotation on the circle by an irrational angle. This process is somewhat analogous to using a Markov partition to produce a symbolic model for a system. Another analogous process is used in the kneading theory of unimodal maps of the interval. The difference here is that the chosen open set, in general, has no relation to the dynamics. The Hilbert cube of strictly ergodic sets is obtained by showing that the invariant sets constructed in the two-shift have unique invariant probability measures that depend continuously on the regular open sets in the appropriate topologies.
The main construction is a generalization of Morse and Hedlund’s construction of Sturmian minimal sets as described on page 111 of \[G-H\]. Such generalizations are a standard tool in topological dynamics. In particular, the main construction is a special case of the almost automorphic minimal extensions of Markley and Paul given in \[M-P\]. Also of particular relevance are pages 234-241 of \[A\] and \[H-H1\].
For any regular open set, the main construction yields a minimal set in the shift. If the open set is a finite union of intervals, it gives a Denjoy minimal set. When the open set is more complicated, the resulting minimal set is more complicated. In particular, it follows from \[M-P\] that for certain open sets the construction gives minimal sets that have positive topological entropy and are not uniquely ergodic (see Remark 3.4 below).
The full two-shift is frequently embedded in the iterates of a complicated dynamical system. (In fact, this is one definition of a “complicated” dynamical system.) In view of Theorem 0.1 one would therefore expect that that $\S(X,f)$ will frequently contain a Hilbert cube. In the following corollary, the first sentence is a consequence of Theorem 0.1 and the Birkhoff-Smale theorem (a particularly suitable statement of which can be found on page 109 of \[Rl\]). The second sentence follows from the first and a theorem of Katok (\[K\]).
[**Corollary 0.2.**]{} [*If $f:M\rightarrow M$ is a diffeomorphism of the compact manifold $M$ that has a transverse homoclinic orbit to a hyperbolic periodic point, then $\S(M,f)$ contains a subspace homeomorphic to the Hilbert cube. In particular, this is the case when $M$ is two-dimensional, $f$ is $C^{1 + \alpha}$ and has positive topological entropy.* ]{}
As was the case with Mather’s theorem, one has a large dimensional family of minimal sets (in this case an infinite dimensional family) embedded in finite dimensional dynamics. We shall see in Remark 3.7 below that in many cases this can be viewed as a manifestation of the fact that the Hilbert cube is the continuous, surjective image of the Cantor set. There is an invariant Cantor set $\hat{\Lambda}$ embedded in the dynamics. The orbit closure of each point in $\hat{\Lambda}$ supports a unique invariant probability measure. When the measures are given the weak topology, the map that takes the point to the measure is a continuous surjection of the Cantor set $\hat{\Lambda}$ onto the Hilbert cube.
There are two important examples that illustrate the necessity of the smoothness and dimension in the second sentence of Corollary 0.2. In \[R\] Rees constructs a homeomorphism of the two torus that is minimal and has positive topological entropy. Herman gives a $C^\omega$-difeomorphism of a $4$-manifold that is also minimal with positive topological entropy (\[Hm\]). Neither example is uniquely ergodic, so in these cases $\S(M,f)$ is empty. The second sentence of Corollary 0.2 also raises the question of a converse. Specifically, if $\S(M,f)$ contains a subspace that is homeomorphic to the Hilbert cube, does $f$ have positive topological entropy? Proposition 3.1 shows that this is false on manifolds of dimension bigger than three.
It is an easy exercise to show that periodic orbits are dense in the full two-shift. A somewhat deeper result due to Parthasarathy says that the invariant probability measures supported on period orbits are weakly dense in the set of all shift-invariant probability measures, $\Ms$ (\[P\]). The next proposition gives the analog of these results for Denjoy minimal sets.
This paper is organized as follows. Section 1 gives basic definitions, background information and the main construction. Section 2 contains the statement and proof of the main theorem. This theorem describes continuity properties of the main construction and the structure of resulting invariant sets. Section 2 also contains the proof of Theorem 0.3. The proof of Theorem 0.1 is given in Section 3, as is the example that shows that the converse of Corollary 0.2 is false in dimensions three and greater. The last section examines the relationship between the intrinsic rotation number of a Denjoy minimal set and its “extrinsic” rotation number when it is embedded in a map of the annulus. It is also shown that any Denjoy minimal set in the two-shift can be generated from a regular open set in the circle using the main construction.
[**Acknowledgments:**]{} The author would like to thank B. Kitchens, N. Markley, B. Weiss and S. Williams for useful comments and references.
[**Section 1: Preliminaries.**]{} This section introduces assorted notation and definitions and recalls some basic facts from topology, ergodic theory and topological dynamics. Many of these facts are stated without proof or references. In such cases, the facts are either elementary exercises or can be found in Walters’ book \[W\].
For a set $X, Cl(X), Int(X), X^c$ and $Fr(X)$ denote the closure, interior, complement and frontier of the set, respectively. The operator $\dcup$ is the disjoint union. Thus $A\dcup B$ represents the union of the two sets, but conveys the added information that the sets are disjoint. The indicator function of a set $X$ is denoted $\Chi_X$. Thus $\Chi_X(x) = 1$ if $x\in X$, and is $0$ otherwise. The circle is $S^1 = \reals/\Z$ and $R_\eta : S^1\rightarrow S^1$ is rigid rotation by $\eta$, [*i.e*]{} $R_\eta(\theta) =
\theta + \eta\ \hbox{mod} \ 1$. Haar measure on the circle is denoted by $m$.
A nonempty, proper subset $U\subset S^1$ is called a [*regular open*]{} set if $Int(Cl(U)) = U$. The set of all regular open sets is $$\R = \{ U\subset S^1 : U \hbox{ is a regular open set} \}.$$ Given an open set $U$, its $\ast$[*-dual*]{} is the interior of its complement and is denoted by $U^\ast = Int(U^c)$. Note that $U$ is regular open if and only if $S^1$ can be written as the disjoint union of three nonempty sets, $S^1 = U \dcup F \dcup U^\ast$ with $F = Fr(U) = Fr(U^\ast)$. In consequence, $U\in\R$ if and only if $U^\ast\in\R$.
The set $\R$ of regular open sets will be topologized using the symmetric difference of sets. For $U, V \in \R$, their symmetric difference is $U \d V = (U \cap V^c)\dcup (U^c\cap V)$ and the distance between them is $d(U,V) = m(U\d V)$. If $U$ and $V$ are regular open, when $U\d V$ is nonempty it contains an interval. In particular, $d(U,V) = 0$ if and only if $U = V$. Since $d(U,V) =
\int |\Chi_U - \Chi_V| dm = \|\Chi_U - \Chi_V\|_1$, $\R$ maybe thought of as a subspace of $L^1(S^1, m)$. This makes it clear that $d$ gives a metric on $\R$.
If the frontiers of either $U$ or $V$ have positive measure, it could happen that $d(U,V) \not = d(U^\ast, V^\ast)$. To avoid this and related situations it is sometimes necessary to restrict attention to the set of regular open sets whose frontiers have measure zero, $$\Rz = \{ U\in \R : m(Fr(U))= 0\}.$$ A metric that controls both regular open sets and their $\ast$-duals is given by $$\ds(U,V) = (d(U,V) + d(U^\ast, V^\ast))/2.$$ Unless otherwise noted, the topology on $\R$ will be that given by the metric $\ds$. Note that when restricted to $\Rz$, $d$ and $\ds$ give the same metric.
It will also be useful to identify regular open sets that are equal after a rigid rotation of $S^1$. More precisely, say $U\sim V$ if there exists an $\eta \in S^1$ with $V = R_\eta(U)$. Denote the quotient spaces by $\Rp = \R\;/\!\sim$ and $\Rzp = \Rz\;/\!\sim$. Note that the topology generated by the projection $\R\rightarrow\Rp$ can be viewed as being generated by the metric $d^\prime([U], [V]) = \inf \{ \ds(U, R_\eta(V)) : \eta\in S^1\}$, where $[U]$ denotes the equivalence class of $U$ under $\sim$.
A related notion is that of a symmetric set. A set $U\in\R$ is called [*symmetric*]{} if there exists an $\eta\not = 0 $ with $R_\eta(U) = U$. Because $U$ is open, such an $\eta$ will always be a rational number.
In this paper a [*dynamical system*]{} means a pair $(X,h)$ where $X$ is a compact metric space and $h$ is a homeomorphism. Given a point $x\in X$, its orbit is $o(x,h) = \{ \dots, h^{-1}(x), x, h(x), \dots\}$. A finite piece of the forward orbit is denoted $o(x,h, N) = \{ x, h(x), \dots, h^N(x)\}$. If $(X,h) \ra (Y,g)$ is a continuous semiconjugacy, then $(X,h)$ is called an [*extension*]{} of $(Y,g)$, and $(Y,g)$ is a [*factor*]{} of $(X,h)$. When the semiconjugacy is one to one on a dense $G_\delta$ set, the extension is termed [*almost one to one*]{}.
The pair $(X,h)$ is called a [*minimal set*]{} if every orbit is dense. The pair is [*uniquely ergodic*]{} if there exists a unique invariant Borel probability measure. A useful characterization is: $(X,h)$ is uniquely ergodic if and only if the sequence of functions $(\sum_{i=0}^N f \circ h^i)/ (N +1) $ converges uniformly for all $f\in C(X, \reals)$. A pair that is both minimal and uniquely ergodic is called [*strictly ergodic*]{}. Note that the property of being minimal, uniquely ergodic or strictly ergodic is preserved under topological conjugacy. Also, if an extension is strictly ergodic, then so is its factor.
A compact $h$-invariant set $Y\subset X$ is called minimal, uniquely ergodic or strictly ergodic if $h$ restricted to $Y$ has that property. In a slight abuse of notation, this situation is described by saying that $(Y,h)$ is minimal, etc.
Perhaps the simplest nontrivial strictly ergodic system is $(S^1,R_\alpha)$ for an irrational $\alpha$. A homeomorphism $g:S^1\rightarrow S^1$ that has an irrational rotation number and the pair $(S^1, g)$ is [*not*]{} minimal is called a [*Denjoy example*]{}. Such examples are classified up to topological conjugacy in \[My\]. The two classifying invariants are the rotation number and the set of orbits that are “blown up” into intervals. A Denjoy example always has a unique minimal set $Y\subset S^1$ with $(Y,g)$ strictly ergodic.
An abstract dynamical system $(X,h)$ is called a [*Denjoy minimal set*]{} if it is topologically conjugate to the minimal set in a Denjoy example. Such an $(X,h)$ is always strictly ergodic. Mather points out in \[M\] that a Denjoy minimal set $(X,h)$ always has a well defined [*intrinsic rotation number*]{}, [*i.e.*]{} if $(X,h)$ is topologically conjugate to the minimal sets in two Denjoy examples $(S^1, g_1)$ and $(S^1, g_2)$, then either $g_1$ and $g_2$ have the same rotation number or else $g_1$ and $g_2^{-1}$ do. If $(X,h)$ is a Denjoy minimal set with intrinsic rotation number $\alpha$, it is an almost one to one extension of $(S^1,R_\alpha)$.
A general dynamical system $(Z,h)$ can have many invariant subsets that are Denjoy minimal sets or strictly ergodic. These subsets are collected together in the spaces $$\D(Z,h) = \{ Y\subset Z: (Y,h) \hbox{ is a Denjoy minimal
set} \}$$ and $$\S(Z,h) = \{ Y\subset Z: (Y,h) \hbox{ is strictly ergodic} \}.$$
To topologize these spaces we recall the weak topology on measures. Given a dynamical system $(Z,h)$, the set of all its invariant, Borel probability measures is denoted $\M(Z,h)$. The weak topology on $\M$ can be defined by saying that the measures $\mu_n\rightarrow\mu_0$ weakly if and only if $\int f d\mu_n\rightarrow\int f \mu_0$ for all $f \in C(Z, \reals)$. Note that $\M(Z,h)$ with this topology is compact, and when viewed as a subspace of the dual space to $C(Z, \reals)$, it is convex with extreme points equal to the ergodic measures. Since a strictly ergodic system supports a unique invariant probability measure, there is a natural inclusion $\S(Z,h)\subset\M(Z,h)$. This inclusion induces a topology on $\S(Z,h)$ that will be called the weak topology. The fact that $\D(Z,h)\subset\S(Z,h)$ allows us to use the weak topology on $\D(Z,h)$ also.
In the absence of unique invariant measures we use the Hausdorff metric to measure the distance between compact invariant subsets. Given a compact space $X$, the space consisting of the closed subsets of $X$ with the Hausdorff topology is denoted $\H(X)$. Note that if $X$ is compact metric then so is $\H(X)$. A map $\Phi :E\rightarrow \H(X)$ is called [*lower semicontinuous*]{} if for all closed subsets $Y\subset X$, the set $\{ e\in E : \Phi(e) \subset Y\}$ is closed in $E$. We will need the fact that the following property implies that $\Phi$ is lower semicontinuous: When $e_n\rightarrow e$ in $E$ and for some subsequence $\{n_i\}$, $\Phi(e_{n_i})\ra K$ in $\H(X)$ then $\Phi(e)\subset K$. Informally, $\Phi$ is lower semicontinuous if when you perturb $e$, $\Phi(e)$ may get suddenly larger, but never suddenly smaller.
The [*full shift on two symbols*]{} is the the pair $(\Sigma_2, \sigma)$ consisting of the sequence space $ \Sigma_2 = \{0,1\}^\Z$ and the shift map $\sigma$. A [*symbol block*]{} $b$ is a finite sequence $b_0, b_1, \dots, b_{N-1}$ which each $b_i$ equal to $0$ or $1$. The [*length*]{} of the block $b$ is $N$ and the [*period*]{} is its period when considered as a cyclic word. A sequence $s\in\Sigma_2$ has [*initial block*]{} $b$ if $b_i = s_i$ for $i=0, \dots, N-1$. It is notationally convenient to view the topology on $\Sigma_2$ as being generated by a metric $d_\Sigma$ with $d_\Sigma(s, t) < 1/N$ if and only if $s_i = t_i$ for $|i|<N$. A [*cylinder set*]{} depends on a block $b$ and an integer $n$ and is a set of the form $$C^n_b = \{ s\in \Sigma_2 : s_{i + n} = b_i, \hbox{ for } i = 0,
\dots, length(b)-1 \}.$$ If $n = 0$, we write $C^0_b = C_b$.
Since cylinder sets are both open and closed, their indicator functions are continuous. In fact, the finite linear combinations of such indicator functions form a dense set in $C(\Sigma_2,\reals)$. This implies that the measures $\mu_n\rightarrow\mu_0$ weakly if and only if $\mu_n(C_b^n) \rightarrow\mu_0(C_b^n)$ for all cylinder sets $C_b^n$. Since the elements of $\Ms$ are shift invariant measures, any such measure $\mu$ satisfies $\mu(C_b^n) = \mu(C_b)$ for all $n$. Thus the topology on $\Ms$ is in fact generated by the metric $$d(\mu_1, \mu_2) = \sum | \mu_1(C_{b^{(n)}}) - \mu_2(C_{b^{(n)}}) | /
2^n$$ where the sum is over some enumeration $b^{(n)}$ of all possible blocks by the natural numbers $n$.
The main construction in this paper takes a regular open set in the circle and produces a compact invariant set in $(\Sigma_2, \sigma)$ along with an invariant measure. As noted in the introduction, it is closely related to the construction given in \[M-P\]. We are primarily interested here in the dependence of the construction on the open set and a “rotation number”. This dependence is encoded in two functions $\lambda:\Rz \times S^1\rightarrow \Ms$ and $\Lambda:\R \times S^1 \rightarrow \Hd$ defined as follows.
Fix $U\in \R$ and $r\in S^1$ Define $B\subset S^1$ as $$B = \{ x \in S^1 : o(x, R_r) \cap Fr(U) = \emptyset\}.$$ Since $U$ is regular open, $ Fr(U)$ is closed and nowhere dense, and thus since $B = \cap_{i\in \N} R^i_r(Fr(U)^c)$, $B$ is dense $G_\delta$. Now define $\phi: B \ra \Sigma_2$ so that $$(\phi(x))_i = \Chi_U(R^i_r(x)).$$ Thus for any point $x\in B$, the sequence $\phi(x)$ is the “itinerary” of $x$ under $R_r$ with respect to the set $U$, [*i.e.*]{} $\phi(x)$ has a $1$ in the i$^{th}$ place if $R_r^i(x)$ is in $U$ and $0$ if it is in $U^\ast$. It is easy to see that $\phi$ is continuous.
Now define $\Lambda(U,r) = Cl(\phi(B))$. If $U\in \Rz$, then $m(U\dcup U^\ast) = 1$ and so $m(B) = 1$. Thus we may define a probability measure $\lambda\in\Ms$ by $\lambda = \phi_\ast(m)$, where as usual this means that $\lambda(X) = m(\phi^{-1}(X))$ for a Borel set $X$.
In this construction, $B$ and $\phi$ depend on the choice of $U$ and $r$. If this dependence needs to be emphasized, we write $B = B_{U,r}$ and $\phi = \phi_{U,r}$. It is clear that for all $\eta \in S^1$ and $U\in \R $, $\Lambda(R_\eta(U), r) = \Lambda(U,r)$ and $\lambda(R_\eta(U), r) = \lambda(U,r)$. Thus the maps $\Lambda$ and $\lambda$ descend to maps on $\Rp \times S^1$ and $\Rzp \times S^1$ that will also be called $\Lambda$ and $\lambda$.
To make the last definition, we need to adopt the notation that $U_0 = U^\ast$ and $U_1 = U$. For a block of symbols $b$ of length $N+1$, define $$U_{b,r}= \bigcap_{i=0}^N R_r^{-i}(U_{b_i}).$$ The important property of these sets is that for $x\in B$, $x\in U_{b,r}$ if and only if $\phi(x)$ is in the cylinder set $C_b$. As a consequence, for $U\in\Rz$, $\lambda(U,r)[C_b] = \phi_\ast m(C_b) =
m(U_{b,r})$.
We first prove continuity of the map $\eta \mapsto d(U,R_\eta(U))$ at $\eta = 0$. Since $U\in S^1$ is open, we can find a countable set of disjoint intervals $\{I_n\}$ so that $U = \dcup I_n$. Now given $\epsilon > 0$, pick $M$ so that $\sum_{n>M} m(I_n) < \epsilon/4$ and assume $|\eta| < \epsilon/ (4 M)$. Now for each $n$, clearly $m(I_n \cap R_\eta(U)^c) < \eta$ and so $$\eqalign{m(U \cap R_\eta(U)^c)& < \sum_{n>M} m(I_n) +
\sum_{n\le M} m(I_n \cap R_\eta(U)^c)\cr
& < \epsilon/2.\cr}$$ Now since $m(U^c \cap R_\eta(U)) = m(R_{-\eta}(U)^c \cap U)$, we also get $m(U^c \cap R_\eta(U)) < \epsilon/2$ and so $d(U, R_\eta(U))
< \epsilon$.
What we have just shown also implies that $\eta \mapsto d(U^\ast,R_\eta(U^\ast))$ is continuous at $\eta = 0$, and thus $\eta \mapsto d_\ast(U,R_\eta(U))$ is also. Since $d(R_\eta(U), R_{\eta^\prime}(U)) = d(U, R_{\eta - \eta^\prime}(U))$, the continuity of $\eta \mapsto R_\eta(U)$ at all $\eta$ follows. Finally, since $d_\ast$ is a metric, and therefore a continuous function $\R \times \R \rightarrow \reals$, we get $\eta \mapsto d_\ast(U,R_\eta(U))$ continuous for all $\eta$.
[**Proof of (c).**]{} Given two finite collections of sets ${A_i}$ and ${B_i}$ with $i \in \{0, \dots N\}$ using the fact that $d(A, B) = \|\Chi_A-\Chi_B\|_1$ and standard integral inequalities it is easy to show that $|m(A) - m(B)| \le d(A,B) $ and $\ds(\cap A_i, \cap B_i ) \le \sum \ds(A_i, B_i)$.
The continuity of $U \mapsto m(U)$, follows from the fact that $\ds(U,V) \le \epsilon/2$ implies $\epsilon \ge d(U,V) \ge |m(U) - m(V) | $.
[**Proof of (d).**]{} If the length of the fixed block $b$ is $N+1$, then given $\epsilon > 0$ using (a), pick $\delta < \epsilon/(2N + 2)$ so that $|\eta| < \delta$ implies $d_\ast(U, R_\eta(U)) < \epsilon/(2N + 2)$.
We therefore have for $(V,s)\in \R \times S^1$ with $d(U,V) <\delta$ and $|r-s|<\delta/N$, $$\eqalign{ \ds( U_{b,r}, V_{b,s}) &
= \ds(\cap R_r^{-i}(U_{b_i}),\cap R_s^{-i}(V_{b_i}))\cr
&\le \sum d_\ast(R_r^{-i}(U_{b_i}), R_s^{-i}(V_{b_i}))\cr
&= \sum\ds(R_{i(s-r)}(U_{b_i}), V_{b_i})\cr
&\le \sum(\ds(R_{i(s-r)}(U_{b_i}), U_{b_i}) + \ds(U_{b_i},V_{b_i})\cr
&\le \epsilon.\cr}$$
[**Section 2: The main theorem.**]{} The main goal of this section is to prove the following theorem. For the reader interested in the quickest route to Theorem 0.1, we note that the lower semicontinuity of $\Lambda$ and the results in part (3) are not needed for that proof.
[**Theorem 2.** ]{}
*Let the maps $\lambda:\Rzp \times S^1\rightarrow \Ms$ and $\Lambda:\Rp \times S^1\rightarrow \Hd$ be as defined in Section 1.*
\(1) The map $\lambda$ is continuous and the map $\Lambda$ is lower semicontinuous.
\(2) Fix $\alpha \not\in \Q$.
=-18pt
\(a) For all $U\in \Rp$, $(\Lua, \sigma)$ is an almost one to one minimal extension of $(S^1, R_{n \alpha})$ for some natural number $n$.
\(b) If $U\in\Rzp$, then $(\Lua, \sigma)$ is uniquely ergodic.
\(c) If $Fr(U)$ is a finite set, then $(\Lua, \sigma)$ is a Denjoy minimal set with intrinsic rotation number $n \alpha$ for some natural number $n$.
\(d) For fixed $\alpha \not\in \Q$, when considered as a function of $U$, $\Lambda$ and $\lambda$ are injective.
\(3) Fix $p/q \in Q$ with $p$ and $q$ relatively prime.
=-18pt
\(a) For all $U\in \Rp$, $\Lambda(U,p/q)$ is a finite collection of periodic orbits whose periods divide $q$.
\(b) For fixed $p/q \in Q $, when considered as a function of $U$, the image of $\lambda$ is the convex hull of the probability measures supported on the periodic orbits whose periods divide $q$.
[**Proof of (1).**]{} Since $\Lambda(R_\eta(U), r) = \Lambda(U,r)$, it suffices to check the continuity of $\Lambda$ as a map defined on $\R$. A similar comment holds for $\lambda$.
As noted in the previous section, the weak topology on $\Ms$ is generated by the metric $d(\lambda_1, \lambda_2)
= \sum | \lambda_1(C_{b^{(n)}}) - \lambda_2(C_{b^{(n)}}) | /
2^n$ and $\lambda(U,r)[C_b] = m(U_{b, r})$. Thus to prove the continuity of $\lambda$ it suffices to check that for fixed $b$ the map $U \mapsto m(U_{b, r}) $ is continuous. This follows from Lemma 1 (c) and (d).
For the proof of the lower semicontinuity of $\Lambda$, begin by assuming that $(U^{(n)}, r^{(n)}) \rightarrow (U^{(0)}, r^{(0)})$. If for some subsequence $\{n_i\}$, $\Lambda(U^{(n_i)}, r^{(n_i)}) \rightarrow
K$ in the Hausdorff topology, then we will show that $\Lambda(U^{(0)}, r^{(0)})
\subset K$. As noted in the previous section, this implies the desired semicontinuity. Fix an $x_0 \in B^{(0)}$ and integer $N > 0$ and let $b$ be the initial block of length $N+1$ in $\phi(x_0)$. This certainly implies that $U^{(0)}_{b, r^{(0)}}$ is a nonempty open set and therefore has positive measure. Therefore by Lemma 1 (c) and (d) there exists an $M$ so that $n>M$ implies that $m(U^{(n)}_{b, r^{(n)}}) > 0$. In particular, for $n>M$, there exists $x_n \in B^{(n)}$ so that $\phi_n(x_n)$ has its initial block equal to $b$. There therefore exits a sequence $x_j \in B^{(j)}$ with $\phi_j(x_j) \rightarrow \phi_0(x_0)$.
Now assuming that for some subsequence $\{n_i\}$, $\Lambda(U^{(n_i)}, r^{(n_i)}) \rightarrow
K$ in the Hausdorff topology, then if $x_{n_i} \in B^{(n_i)}$ is the appropriate subsequence of the sequence constructed in the previous paragraph, then $\phi_{n_i}(x_{n_i}) \rightarrow \phi_0(x_0)$, so certainly $\phi_0(x_0) \in K$. But $x_0\in B^{(0)}$ was arbitrary, and so $\phi_0(B^{(0)})\subset K$ and since $\Lambda(U^{(0)}, r^{(0)})$ is the closure of the $\phi_0(B^{(0)})$, we have $\Lambda(U^{(0)}, r^{(0)})
\subset K$, as required.
[**Proof of (2).**]{} For the proof of (2), fix an $\alpha\not\in \Q$ and for the proof of (2a), (2b) and (2c) a $U\in\R$. We will suppress the dependence of various objects on $U$ and $\alpha$ and so $\Lambda = \Lambda(U, \alpha)$, etc.
[**(2a).**]{} To prove the minimality of $\Lambda $ we use the following characterization of minimality (\[O\]): If $f:X \rightarrow X$ is a homeomorphism of a compact metric space and $x \in X$, then $Cl(o(x,f))$ is a minimal set if and only if given $\epsilon > 0$, there exists an $N$ such that for all $n$, there exists an $i$ with $0 \le i \le N$ and $d(f^{n + i}(x), x) <
\epsilon$.
To apply this to the case at hand, first note that for $x \in B$, certainly $o(x, R_\alpha)$ is dense in $B$, and so $\Lambda = Cl(o(\phi(x), \sigma))$. Since $(S^1, R_\alpha)$ is minimal, the above property holds for $Cl(o(x,R_\alpha))$. Since $\phi$ is continuous, it also holds for $Cl(o(\phi(x), \sigma)) = \Lambda$, which is therefore minimal.
The proof of the semiconjugacy requires a new definition. Given $U, V \in \R$, define $\rho(U,V) = \sup\{ m(I) : I
\hbox{ is an interval contained in\ } U\d V\}$. Now $\rho$ will not satisfy the triangle inequality but it is easy to see that for fixed $U\in \R$, the map $\eta \mapsto \rho(U, R_\eta(U))$ is a continuous function $S^1 \rightarrow \reals$. Also, if $U$ is asymmetric, then $\rho(U, R_\eta(U))=
0$ if and only if $\eta = 0$.
The first step in the proof of the semiconjugacy is to show that $\phi$ is injective when $U$ is asymmetric. Assume that for $x_1, x_2 \in B$, $\phi(x_0) = \phi(x_1)$, and therefore for all $i$, $\Chi_U(R_\alpha^i(x_1)) = \Chi_U(R_\alpha^i(x_2))$. Thus if $x_2 = R_\eta(x_1)$, $\Chi_U = \Chi_U \circ R_\eta$ when restricted to the dense set $o(x_1, R_\alpha)$. In particular, $\rho(U, R_\eta(U))=
0$ and since $U$ is asymmetric, $d(x_1, x_2) = \eta = 0$.
Continuing with the assumption that $U$ is asymmetric, we show that $\phi^{-1}$ is uniformly continuous. Since $\phi(B)$ is certainly dense in $\Lambda$, this implies that we can extend $\phi^{-1}$ to a semiconjugacy from $(\Lambda, \sigma)$ to $(S^1, R_\alpha)$.
Since $S^1$ is compact and $\eta \mapsto \rho(U, R_\eta(U))$ is continuous, given $\epsilon > 0$ there exists a $\delta > 0$ so that $\rho(U, R_\eta(U)) <
\delta$ implies $|\eta| < \epsilon$. Pick $N >0$ so that for every $x\in S^1$, every interval of length $\delta$ contains a point of $o(x, R_\alpha, N)$. Now if $x_1, x_2 \in B$ satisfy $d_\Sigma(\phi(x_1), \phi(x_2)) < 1/N$ and if $x_2 = R_\eta(x_1)$, then $\Chi_U =
\Chi_U \circ R_\eta$ when restricted to the set $o(x, R_\alpha, N)$. Now if $\rho(U, R_\eta(U)) > \delta$ then $U\d R_\eta(U)$ will contain an interval of length $\delta$ and thus a point of $o(x, R_\alpha, N)$, a contradiction. Thus $\rho(U, R_\eta(U)) < \delta$ and so by the choice of $\delta$, $d(x_1, x_2) =| \eta | < \epsilon$, proving the uniform continuity of $\phi^{-1}$. Note that $\phi(B)$ is dense $G_\delta$ in $\Lambda$ so the extension is almost one to one.
Now assume that $U$ is symmetric. The group of numbers $r$ such that $R_r(U) = U$ has a rational generator, say $p/q$, with $0 < p/q <1 $ and $p$ and $q$ relatively prime. If $U^\prime = \pi(U)$ where $\pi:S^1 \rightarrow S^1/R_{p/q}$ is the projection, then $\Lambda(U,\alpha)$ has $\Lambda(U^\prime, q \alpha)$ as a $q$-fold factor (here we have identified $S^1/R_{p/q}$ with $S^1$). Since $U^\prime $ is asymmetric, $\Lambda(U^\prime,q \alpha)$ has $(S^1, R_{ q\alpha})$ as a factor, finishing the proof of (2a).
[**(2b).**]{} Let $\psi$ denote the extension of $\phi^{-1}$ to a continuous semiconjugacy from $(\Lambda, \sigma)$ to $(S^1,R_{\alpha q})$ and assume that $m(Fr(U)) = 0$. If $\lambda_1$ and $\lambda_2$ are two invariant Borel probability measures supported on $\Lambda$, then since $(S^1,R_{\alpha q})$ is uniquely ergodic, $\psi_\ast(\lambda_1) =
\psi_\ast(\lambda_2) = m$. If $X \subset \Lambda$ is a Borel set, then since $m(B) = 1$, for $i = 1,2$, $\lambda_i(X) =
\lambda_i(\psi^{-1}(B) \cap X)$. Now since $\psi$ is injective on $B$, this is equal to $\lambda_i(\psi^{-1}(B \cap\psi(X)) =
m(B\cap\psi(X))$ and so $\lambda_1=\lambda_2$.
[**(2c).**]{} Now assume $Fr(U)$ is a finite set. In this case, each $x \in B^c$ will have exactly two preimages under $\psi$, namely, the limit of $\phi(x_n)$ as $x_n \rightarrow x$ from the right and the limit of $\phi(x_n)$ as $x_n \rightarrow x$ from the left. This makes it clear that in this case $\Lambda$ is conjugate to the minimal set in the circle homeomorphism obtained by “blowing up” into intervals points on the orbits of each $x\in Fr(U)$.
[**(2d).**]{} When $U\in\Rz$, $\Lambda(U,\alpha)$ is the support of $\lambda(U,\alpha)$. Thus to prove (2d) it suffices to show that $\Lambda(U,\alpha)$ is an injective function of $U$. Assume that for some $U_1, U_2 \in \R$, $\Lambda(U_1, \alpha) = \Lambda(U_2, \alpha)$. Using (2a), $\phi(B_1)$ and $\phi_2(B_2)$ are dense $G_\delta$ in the compact metric space $\Lambda(U_1, \alpha) = \Lambda(U_2, \alpha)$. This implies that $\phi(B_1) \cap \phi_2(B_2)\not = \emptyset$, and so there exist $x_1, x_2 \in S^1$ with $\phi_1(x_1) = \phi_2(x_2)$. Thus if $R_\eta(x_1) = x_2$, then $\Chi_{U_1} = \Chi_{U_2} \circ
R_\eta$ when restricted to the dense set $o(x_1, R_\alpha)$. This implies that $U_1 \d R_\eta(U_2)$ contains no intervals. Since the $U_i$ are regular open sets, this means that $U_1 = R_\eta(U_2)$ and so $U_1$ and $U_2$ are in the same equivalence class in $\Rp$, as required.
[**Proof of (3).**]{} Fix $p/q\in\Q$ with $p$ and $q$ relatively prime. Since $R^q_{p/q} = Id$, it is clear that any $s \in \Lambda(U, p/q)$ will satisfy $\sigma^q(s) = s$ which implies (3a). Say a symbol block $b$ is [*prime*]{} if its length equals its period. For $U\in \Rzp$, by construction, $\lambda(U, p/q) = \sum m(U_{b, p/q}) \mu_b$ where $\mu_b$ is the probability measure supported on the periodic orbit with repeating block $b$ and the sum is over all prime blocks $b$ whose period divides $q$. With this formula in hand it is easy to construct a $U$ so that $\lambda(U, p/q)$ is any desired point in the convex hull given in the statement of (3b).
(a). A theorem of Parthasarathy says that the measures supported on periodic orbits are dense in $\Ms$ (\[P\]). Fix one such measure $\mu_0$, and assume it is supported on an orbit of period $q$. Using the formula given in the proof of Theorem 2 (3b), find a regular open set $U$ with $Fr(U)$ a finite set and a $p/q$ with $\lambda(U, p/q) = \mu_0$. Now pick irrationals $\alpha_n\ra p/q$. By Theorem 2 (1), $\lambda(U, \alpha_n)\ra \mu_0$, and by Theorem 2 (2c), each $\lambda(U, \alpha_n)$ is the unique measure supported on a Denjoy minimal set.
(b). It suffices to show that for any symbol block $b$, there exists an $s\in \Sigma_2$ which has initial block $b$ and $Cl(o(s, \sigma))$ is a Denjoy minimal set. Fix an irrational $\alpha$ and $x_0 \in S^1$. Choose a finite union of intervals $U$ so that $R_\alpha^i(x_0)\in U$ if and only if $b_i = 1$, for $i = 0, \dots, length(b) - 1$. Further, the open set $U$ should satisfy $o(x_0, R_\alpha) \cap Fr(U) = \emptyset$. If $U$ has these properties, Theorem 2 (2c) shows that $\Lambda(U,\alpha)$ is the desired Denjoy minimal set.
[**Section 3: The Hilbert cube of strictly ergodic sets.**]{} We begin with some definitions in preparation for the proof of Theorem 0.1. A copy of the Hilbert cube is given by the collection of sequences, $$\Hb = \{ \gamma\in\reals^\N : 0\le\gamma_i\le {1\over i+2} \hbox{ for all }
i\in\N\}.$$ A subspace of $\Hb$ that contains topological balls of all dimensions is $$\Hb_0 = \{ \gamma\in\Hb : \gamma_i = 0,
\hbox{ for all but finitely many } i\}.$$ For $\gamma\in\Hb$, define an asymmetric regular open set $U_\gamma$ by $$U_\gamma = \bigcup_{i\in\N} ({1\over i+2} - \gamma_i^3,
{1\over i+2} + \gamma_i^3).$$ Now define a map $\Gamma : \Hb \ra \Rzp$ via $\Gamma(\gamma) = [U_\gamma]$. It is clear that $\Gamma$ is continuous and injective. Since $\Hb$ is compact, $\Gamma(\Hb)$ is homeomorphic to $\Hb$. Fix an irrational $\alpha$. By Theorem 2 (2ab), the set $\Lambda(\Gamma(\Hb), \alpha)$ consists of strictly ergodic sets. Since $\Gamma(\Hb)$ is compact, using Theorem 2 (1) and (2d), we have that $\lambda(\Gamma(\Hb), \alpha)$ is homeomorphic to $\Gamma(\Hb)$ and therefore to $\Hb$. This proves the first statement in the theorem. To prove the second, note that Theorem 2 (2c) implies that $\lambda(\Gamma(\Hb_0), \alpha)$ consists of measures supported on Denjoy minimal sets. Since $\lambda(\Gamma(\Hb_0), \alpha)$ is homeomorphic to $\Hb_0$, it (and consequently, $\D(\Sigma_2, \sigma)$) contains topological balls of all dimensions.
[**Remarks.**]{}
[**(3.1)**]{} In Theorem 0.1 there is an obvious distinction between $\S(\Sigma_2, \sigma)$, which contains a copy of $\Hb$, and $\D(\Sigma_2, \sigma)$, which contains a copy of $\Hb_0$. This is because $\Lambda(\Gamma(\Hb), \alpha)$ contains minimal sets that are not Denjoy. In particular, if $\gamma\in\Hb - \Hb_0$ and for some $i\neq 0$, $R_\alpha^i(0)\in Fr(U_\gamma)$, then $\Lambda(U_\gamma, \alpha)$ is not a Denjoy minimal set. In the semiconjugacy from $(\Lambda(U_\gamma, \alpha), \sigma)$ to $(S^1, R_\alpha)$, the inverse image of $0$ consists of [*three*]{} points.
A Denjoy minimal set is obtained from an irrational rotation on the circle by replacing (or ‘blowing up’’) each element of a collection of orbits by a [*pair*]{} of orbits. For all $\gamma$ not of the type just described, $\Lambda(U_\gamma, \alpha)$ is a Denjoy minimal set. When $\gamma \in \Hb_0$, the number of orbits blown up is the same as the number of distinct orbits containing points of $Fr(U_\gamma)$. For $\gamma\in\Hb - \Hb_0$, if for all $i\neq 0$, $R_\alpha^i(0)\not\in Fr(U_\gamma)$, then $\Lambda(U_\gamma, \alpha)$ is a Denjoy minimal set with countably many orbits blown up. All the infinite dimensional families we could construct had the property that some minimal set was not Denjoy.
[**(3.2)**]{} Morse and Hedlund’s construction of Sturmian minimal sets corresponds to the special case $U = (0, \alpha)$. In this case, $\Lambda(U, \alpha)$ is a Denjoy minimal set with a single orbit blown up.
[**(3.3)**]{} Theorem 2 (1) states that $\Gamma$ is a lower semicontinuous function whose range is the set of closed subsets of a compact metric space. When such functions have a domain that is a Baire space, they are continuous on a dense, $G_\delta$ set (see page 114 of \[C\]). It seems unlikely that $\R$ is a Baire space, but since $\Gamma(\Hb)$ is homeomorphic to the Hilbert cube, we may apply this result to show that the map (for fixed $\alpha$) $$\Lambda(\cdot\ , \alpha) : \Gamma(\Hb)\ra\Hd$$ is continuous at a generic point of $\Lambda(\Hb)$. This result can also be obtained directly by showing that the map is, in fact, continuous at all points $\Gamma(\gamma)$ for which all points of $Fr(U_\gamma)$ are on disjoint orbits.
[**(3.4)**]{} As is perhaps obvious from Remark (3.1), when $Fr(U)$ is more complicated topologically, so is the structure of $\Lambda(U, \alpha)$ (for irrational $\alpha$). However, Theorem 2 (2b) says that for all $U\in \R_0$, $\Lambda(U, \alpha)$ is uniquely ergodic. It is in fact measure isomorphic to $(S^1, R_\alpha)$. To get minimal sets with more interesting measure theoretic properties we must have $m(Fr(U)) > 0$. In this case the set $B_{U,\alpha}$ from the main construction is a zero measure, dense $G_\delta$ set in the circle. This leads one to expect that $\Lambda(U, \alpha)$ could support more than one invariant probability measure.
The results of \[M-P\] show that this is frequently the case. The relevant construction from that paper begins with a Cantor $K$ in the circle. The complement of $K$ is the disjoint union of open intervals. One chooses a set of labels for these open sets with each open set labeled by zero or one. The set of labels is used to construct a minimal set in the two-shift as in the main construction. If $K$ has positive measure, then for most sets of labels (in the appropriate sense) the constructed minimal set is not uniquely ergodic and has positive topological entropy.
However, the constructed minimal set can be uniquely ergodic as the following example suggested by Benjamin Weiss shows. Let $(X, f)$ be a Denjoy minimal set with intrinsic rotation number $\alpha$. Note that $(X, f)$ is both measure isomorphic to and an almost one to one extension of $(S^1, R_\alpha)$. Using results of Jewett and Kreiger we may find a zero-dimensional strictly ergodic system $(Z,h)$ that is mixing and has positive entropy. Let $(Y, g)$ be the product of the two systems. Because $(Z,h)$ and $(X,f)$ are strictly ergodic and $(Z,h)$ is mixing and $(X,f)$ has pure point spectrum, $(Y,g)$ is strictly ergodic. Now think of Y as an extension of X. The main theorem and the remark following Theorem 4 in \[F-W\] imply that there is a minimal almost 1-1 extension of X , say $(\tilde{Y}, \tilde{g})$, which maps onto $(X,f)$ in such a way that the invariant measures of $(\tilde{Y}, \tilde{g})$ are in one to one correspondence with the $g$-invariant measures on $Y$. Thus $(\tilde{Y}, \tilde{g})$ is a strictly ergodic, positive entropy, almost 1-1 extension of rotation by alpha. Further, as a consequence of the method of construction in \[F-W\], since $X$, $Y$, and $Z$ are zero-dimensional, $\tilde{Y}$ is also.
Let $p:\tY\ra S^1$ denote the given semiconjugacy and let $\tB \subset\tY$ be the dense $G_\delta$ set on which $p$ is injective. Pick two sets, each open and closed, with $V_0 \dcup V_1 =\tY$. Note that $U =
(p(V_0))^c$ is a regular open set. Use the partition $\{ V_0, V_1\}$ in the usual way to get a symbolic model by defining $k:\tY\ra\Sigma_2$ so that $$(k(y))_i = \Chi_{V_1}(\tg^i (y)).$$ It is fairly straightforward to show that $\tB\subset
p^{-1}(B_{U,\alpha})$ and thus, $p = \psi\circ k$ where $\psi:\Lambda(U,\alpha)\ra S^1$ is the semiconjugacy constructed in the proof of Theorem 2 (2a). This implies that $\Lambda(U,\alpha)$ is a factor of $(\tY,\tg)$, and thus is strictly ergodic. Further, we may choose $V_0$ and $V_1$ so that $\Lambda(U,\alpha)$ has positive entropy. To finish, note that $m(Fr(U)) > 0$, for if not, $\Lambda(U,\alpha)$ would be measure isomorphic to the zero entropy system $(S^1, R_\alpha)$.
It would be interesting to have conditions on a regular open set with positive measure frontier that distinguish these two cases. More precisely, give necessary and sufficient conditions for the unique ergodicity of $\Lambda(U, R_\alpha)$. Another interesting question is the structure of the set of its invariant measures in the cases when $\Lambda(U,\alpha)$ is not uniquely ergodic ([*cf.*]{} \[Wm\]).
[**(3.5)**]{} Since each point in $\S(X,f)$ represents a disjoint minimal set, the size of $\S(X,f)$ should give some indication of the complexity of the dynamics of $f$. The topological entropy of $(X,f)$, denoted $h(X,f)$, is perhaps the most common way of measuring dynamical complexity. Corollary 0.2 shows that, at least in some cases, when the topological entropy is positive, $\S(X,f)$ is large. If the size of $\S(X,f)$ is to give a measure of dynamical complexity, the converse should be true. The next proposition shows that this is not the case, at least when the “size” of $\S(X,f)$ is measured by the maximal dimension of an embedded ball and $X$ is a manifold of dimension greater than two.
Fix an irrational $\alpha$ and let $T = S^1 \times \Hb$. Define $F:T\ra T$ as $F = R_\alpha\times Id$. We will do a construction analogous to the main construction, but now using the space $T$ and the map $F$. To get an open set in $T$ we use the open sets $U_\gamma$ constructed above to define $$\Uh= \bigcup_{\gamma\in \Hb} U_\gamma \times\{ \gamma\}.$$ Next let $$\Bh = \{ \beta\in T : o(\beta, F) \cap Fr(\Uh) = \emptyset \}$$ and define $\Phi: \Bh\ra \Sigma_2$ so that $$(\Phi(\beta))_i = \Chi_{\Uh}(F^i(\beta)).$$ Finally, let $\Lh = Cl(\Phi(\Bh))$.
Note that for fixed $\gamma$, $\Phi$ restricted to $(S^1\times\{ \gamma\})\cap\Bh$ is just $\phi_{U_\gamma, \alpha}$ from the main construction and that $$\Lh = Cl(\bigcup_{\gamma\in \Hb} \Lambda(U_\gamma, \alpha)).$$
Theorem 2 (2a) and (2d) imply that $\Phi$ is injective. Using an argument similar to one in the proof of Theorem 2 (2a), one gets that $\Phi^{-1}$ is uniformly continuous, and therefore has a continuous extension to a $\Psi:\Lh\ra T$ that satisfies $\Psi\circ\sigma = F\circ\Psi$.
The variational principle (see page 190 in \[W\]) implies that $h(\Lh, \sigma) = 0$ if all ergodic measures for $(\Lh, \sigma)$ have metric entropy zero. If $\eta$ is an ergodic, invariant Borel probability measure for $\Lh$, then $\Psi_\ast(\eta)$ is such a measure for $(T, F)$ and so $\Psi_\ast(\eta)$ is Haar measure on $S^1\times\{\gamma_0\}$ for some $\gamma_0$. This implies that $\eta$ is supported on $\Psi^{-1}(S^1\times\{\gamma_0\})$. Once again, using an argument virtually identical to one in the proof of Theorem 2 (2b), one obtains $\eta = \lambda(U_{\gamma_0}, \alpha)$. This measure with the shift is measure isomorphic to rotation on the circle by $\alpha$ and therefore has zero metric entropy, as required. Note that the argument just given also shows that $\S(\Lh, \sigma)$ is in fact homeomorphic to the Hilbert cube, $\Hb$.
[**Proof of (b).**]{} We first construct the map on the space $P=D^2\times [-1,1]$, where $D^2$ is a closed two-dimensional disk. Let $h:D^2\ra D^2$ be a Smale horseshoe, [*i.e.*]{} $h$ is a $C^\infty$-diffeomorphism whose nonwandering set consists of the union of a finite number of fixed points and a set $\Omega$ on which the dynamics are conjugate to the full two-shift. The compact invariant set $\Lh$ constructed in the proof of (a) is embedded in $\Omega$ by the conjugacy. Call this embedded set $\Lo$.
Next, let $h_t$ for $t\in [-1,1]$ be an isotopy with $h_{-1} = Id$, $h_0=h$, and $h_1 = Id$. Further, $h_t$ restricted to the boundary of $D^2$ should be the identity for all $t$. Now pick a $C^\infty$-function $w:P\ra\reals$ with $w\ge 0$ and $w^{-1}(0) = \partial P \dcup (\Lo \times \{0\})$. Let $g:P\ra P$ be the time one map of the flow generated by the vector field $w(u) {\partial\ \over\partial z}$, where $u = (x,y,z)$ is a point in $P$. Now let $f = g \circ (h_t\times Id)$. By construction, the nonwandering set of $f$ is $\partial P \dcup (\Lo \times \{0\})$ and thus $h(f) = 0$. Since each point on $\partial P$ is a fixed point for $f$, $\S(P,f)$ is homeomorphic to $\S(\Lo,\sigma)\dcup\partial P$, which in turn, is homeomorphic to $\Hb \dcup \partial P$.
To obtain the result on a general manifold of dimension three or higher, embed a copy of $(P,f)$ in it and extend $f$ by the identity on the rest of the manifold.
[**Remarks**]{}
[**(3.6)**]{} This proposition leaves open the possibility of a converse to Corollary 0.2 in dimension 2. In this dimension there are a number of results that show that the existence of certain types of zero entropy invariant sets can imply that a homeomorphism has positive topological entropy. For example, if an orientation-reversing homeomorphism of a compact surface of genus $g$ has periodic orbits with $g+2$ distinct odd periods, then it has positive entropy (\[B-F\], \[H\]). For orientation-preserving homeomorphisms there are restrictions on the periods that occur in zero entropy maps given in \[S\]. Even a single period orbit can imply positive entropy if the isotopy class on its complement is nontrivial (\[Bd\]). These results give credence to the conjecture that for a manifold $M$ of dimension $2$, if $f:M\ra M$ is a homeomorphism and $\S(M,f)$ contains a topological ball of dimension $3$, then $h(f)>0$.
[**(3.7)**]{} It was noted in the introduction that the existence of a Hilbert cube of strictly ergodic sets can often be viewed as a manifestation of a standard topological fact, namely, the Hilbert cube is the continuous surjective image of the Cantor set. For concreteness, let $f: M\ra M$ be a homeomorphism with an invariant set $\Lo$ with $(\Lo, f)$ conjugate to $(\Lh, \sigma)$, where $\Lh$ is the set constructed in the proof of Proposition 3.1 (b). Using the conjugacy, the proof of Proposition 3.1 (b), and Theorem 2 (2b) one gets that for each $x\in\Lo$, $Cl(o(x,f))$ supports a single invariant probability measure which is $c_\ast(\lambda(U_{\gamma(x)}, \alpha))$ for the appropriate $\gamma(x)$. Further, the map $x\mapsto c_\ast(\lambda(U_{\gamma(x)}, \alpha))$ is continuous. (More formally, this map is $$x\mapsto c_\ast(\lambda(\Gamma(\pi_2(\Psi(x))), \alpha))$$ where $\pi_2: S^1\times\Hb\ra\Hb$ is the projection). The domain of this map is the invariant Cantor set $\Lo$ and its image is $\lambda(\Gamma(\Hb),\alpha)$, which is homeomorphic to the Hilbert cube, $\Hb$.
[**(3.8)**]{} The construction in the proof of 3.1 (b) can be used to embed any compact shift invariant subset of $\Sigma_2$ as the only “interesting” dynamics in a three-dimensional diffeomorphism. It is reminiscent of Schweitzer’s construction of C$^1$-counterexample to the Seifert conjecture (\[Sc\]).
[**Section 4: Intrinsic and extrinsic rotation numbers.**]{} In the Section 1 it was noted that abstract Denjoy minimal sets have well-defined intrinsic rotation numbers. The next proposition specializes some previous results to the case of fixed intrinsic rotation number.
[**Proposition 4.1.**]{}
*Fix an irrational $\alpha$ and let $\D_\alpha(\Sigma_2, \sigma)$ denote the set of Denjoy minimal sets in the shift with intrinsic rotation number $\alpha$.*
=-18pt
\(a) When given the weak topology, the space $\D_\alpha(\Sigma_2, \sigma)$ contains topological balls of dimension $n$ for all natural numbers $n$.
\(b) The set of points that are members of Denjoy minimal sets with intrinsic rotation number $\alpha$ is dense in $\Sigma_2$.
\(c) If $(D,\sigma)$ is a Denjoy minimal set with intrinsic rotation number $\alpha$, then $ D = \Lambda(U,\alpha)$ for some regular open set $U$ with $m(Fr(U)) = 0$. Consequently, $\D_\alpha(\Sigma_2, \sigma)$ $ \subset \lambda(\Rz, \alpha)$.
When $U$ is asymmetric and $\Lambda(U, \alpha)$ is a Denjoy minimal set, it has intrinsic rotation number $\alpha$. This follows from Theorem 2 (2b) (and its proof). Thus to prove (a) we need only note that the proof of Theorem 0.1 began with a statement, “ Fix an irrational $\alpha$”. The proof of Proposition 0.3 (b) contains a similar statement, so that proof proves (b).
To prove (c), note that by definition, there exists a conjugacy $c:D\ra Y$ where $Y$ is the minimal set in a Denjoy example $g:S^1\ra S^1$ with rotation number $\alpha$. It is a standard fact that there exists a semiconjugacy $h$ of $(S^1, g)$ to $(S^1, R_\alpha)$ with the properties that $h$ is injective on a set that is dense in $Y$ and the lift of $h$ is weakly order preserving, [*i.e.*]{} $x < y$ implies $\tilde{h}(x)\le \tilde{h}(y)$.
Now let $p = h\circ c$ and $U=(p(C_0))^c$. Since $C_0$ is compact in $\Sigma_2$, $U$ is open. Further, the properties given above imply that $U^\ast = (p(C_1))^c$ and $p(C_0)\cap p(C_1) = Fr(U) = Fr(U^\ast)$. Thus using a fact from Section 1, $U$ is a regular open set, and by construction, $\Lambda(U,\alpha) = D$. Since $p(C_0)\cap p(C_1)$ is at most countable, $m(Fr(U)) = 0$.
These results, of course, also hold for homeomorphisms with a full two-shift embedded in their dynamics. In this case, however, one is perhaps more interested in [*extrinsic*]{} properties of invariant sets, [*i.e.*]{} properties associated with how the sets are embedded in the manifold. Perhaps the simplest such extrinsic property is the extrinsic rotation number, and the simplest case in which this can be defined is for a homeomorphism of the annulus.
If $f:A \ra A$ is a homeomorphism of the annulus and $z\in A$, define the rotation number of $z$ under $f$ as $$\rho(z) = \lim_{n\ra\infty}
{\pi_1(\tilde{f}^n(\tilde{z})) - \pi_1(\tilde{z}) \over n},$$ if the limit exists. Here $\tilde{f}:\reals\times [-1,1]\ra\reals\times [-1,1]$ and $\tilde{z}$ are lifts of $f$ and $z$, respectively, and $\pi_1:\reals\times [-1,1]\ra\reals$ is the projection. Note that the rotation number is only defined modulo $1$ as it depends on the choice of lift.
If $D\subset A$ is a Denjoy minimal set under $f$, then it is uniquely ergodic. Thus for all $z\in D$, $\rho(z)=\int r(z)\; d\mu$, where $\mu$ is the the unique invariant probability measure of $(D,f)$ and $r:S^1\ra \reals$ is the map that lifts to $\pi_1\circ\tilde{f} - \pi_1$. This number will be called the [*extrinsic rotation number*]{} of $(D,f)$.
The Denjoy minimal sets constructed by Mather in \[M\] have monotonicity properties that imply that their extrinsic and intrinsic rotation numbers are rationally related. For Denjoy minimal sets in a general homeomorphism of the annulus this will not be the case. As a specific example, we will consider homeomorphisms $f: A\ra A$ that have a [*rotary horseshoe*]{} ([*cf.*]{} \[H-H2\]) A picture of the lift of such a map is shown in Figure 1. The dotted vertical lines are the boundaries of fundamental domains.
Figure 1: The lift of a rotary horseshoe.
A map contains a rotary horseshoe if it has a compact invariant set $\Omega$ that is conjugate to the full two-shift. The conjugacy $c:\Omega\ra \Sigma_2$ is required to have the property that for $z\in \Omega$ the first element in $c(z)$ is $1$ if and only if $\tilde{f}$ moves $\tilde{z}$ (approximately) one fundamental domain to the right. More precisely, for $z\in\Omega$ it is required that $$\rho(z) = \lim_{N\ra\infty}
\sum_{i=0}^N {\Chi_{C_1}(\sigma^i(c(z)))\over (N+1)}.$$ Thus $\rho(z)$ is the asymptotic average number of ones in the sequence $c(z)$.
We are now almost in a position to state a result about the existence of Denjoy minimal sets with given intrinsic and extrinsic rotation number. For an annulus homeomorphism $f$, let $\D_{\alpha, \beta}(A,f)$ denote the set of all Denjoy minimal sets for $f$ with intrinsic rotation number $\alpha$ and extrinsic rotation number $ \beta$.
[**Proof of Proposition 4.2.**]{} If for a given $U\in\Rz$ and irrational $\alpha$, $\Lambda(U,\alpha)$ is a Denjoy minimal set, then the comments above Lemma 1 and unique ergodicity imply that for all $s\in \Lambda(U,\alpha)$, $$\lim_{N\ra\infty}\sum_{i=0}^N {\Chi_{C_1}(\sigma^i(s))\over
(N+1)}= \lambda(U,\alpha)[C_1]= m(U).$$ This implies that the corresponding Denjoy minimal set in the annulus has extrinsic rotation number equal to $m(U)$. To finish the proof, one need only imitate the proof of Theorem 0.1 using a family $U_\gamma$ that satisfies $m(U_\gamma) = \beta$, for all $\gamma$.
Note that the case of rational $\beta$ is included in this result. This means that large dimensional balls of Denjoy minimal sets with a given [*rational*]{} extrinsic rotation number are present in the dynamics.
REFERENCES
[\[A\]]{} Auslander, J., [*Minimal Flows and their Extensions*]{}, North Holland Mathematics Studies, vol. 153, 1988.
[\[B-F\]]{} Blanchard, P. and Franks, J., The dynamical complexity of orientation reversing homeomorphisms of surfaces, [*Inv. Math.*]{}, [**62**]{}, 1980, 333–339.
[\[Bd2\]]{} Boyland, P. , An analog of Sharkovski’s theorem for twist maps, [*Contemporary Math.*]{}, [**81**]{}, 1988, 119–133.
[\[C\]]{} Choquet, G., [*Lectures on Analysis*]{}, vol 2, W.A. Benjamin, 1969.
[\[F\]]{} Furstenberg, H., Strict ergodicity and transformations of the torus, [*Amer. J. Math.*]{}, [**83**]{}, 1961, 573–601.
[\[F-W\]]{} Furstenberg, H. and Weiss, B., On almost 1–1 extensions, [*Isr. J. Math.*]{}, [**65**]{}, 1989, 311–322.
[\[GH\]]{} Gottschalk, W. and Hedlund, [*Topological Dynamics*]{}, AMS Colloquium Pub., vol. 36, 1955.
[\[H\]]{} Handel, M. , The entropy of orientation reversing homeomorphisms of surfaces, [*Topology*]{}, [**21**]{}, 1982, 291–296.
[\[Hm\]]{} Herman, M., Construction d’un difféomorphisme d’entropie topologique non nulle, [*Ergod. Th. & Dynam. Sys.*]{}, [**1**]{}, 1981, 65–76.
[\[H-H1\]]{} Hockett, K. and Holmes. P., Bifurcation to rotating Cantor sets in maps of the circle, [*Nonlinearity*]{}, [**1**]{}, 1988, 603–616.
[\[H-H2\]]{} Hockett, K. and Holmes, P., Josephson’s junction, annulus maps, Birhoff attractors, horseshoes and rotation sets, [*Ergod. Th. & Dynam. Sys.*]{}, [**6**]{}, 1986, 205–239.
[\[K\]]{} Katok, A., Lyapunov exponents, entropy and periodic orbits for diffeomorphisms, [*Publ Math IHES*]{}, [**51**]{}, 1980, 137–173.
[\[M\]]{} Mather, J., More Denjoy minimal sets for area-preserving mappings, [*Comm. Math. Helv.*]{}, [**60**]{}, 1985, 508–557.
[\[My\]]{} Markley, N., Homeomorphisms of the circle without periodic orbits, [*Proc. Lond. Math. Soc.*]{}, [**20**]{}, 1970, 688–698.
[\[M-P\]]{} Markley, N. and Paul, M., Almost automorphic symbolic minimal sets without unique ergodicity, [*Isr. J. Math.*]{}, [**34**]{}, 259–272.
[\[O\]]{} Oxtoby, J., Ergodic sets, [*Bull. A.M.S.*]{}, [**58**]{}, 1952, 116–136.
[\[P\]]{} Parthasarathy, K., On the category of ergodic measures, [*Ill. J. Math*]{}, [**5**]{}, 1961.
[\[R\]]{} Rees, M., A minimal positive entropy homeomorphism of the two torus, [*J. London Math Soc*]{}, [**23**]{}, 1981, 537–550.
[\[Rl\]]{} Ruelle, D., [*Elements of Bifurcation Theory*]{}, Academic Press, 1989.
[\[Sc\]]{} Schweitzer, P. A., Counterexamples to the Seifert conjecture and opening closed leaves of foliations, [*Ann. of Math.*]{}, [**100**]{}, 1974, 386–400.
[\[S\]]{} Smillie, J., Periodic points of surface homeomorphisms with zero entropy, [*Ergod. Th. & Dynam. Sys.*]{}, [**3**]{}, 1983, 315–334.
[\[W\]]{} Walters, P., [*An Introduction to Ergodic Theory*]{}, Graduate Texts in Mathematics, vol. 79, Springer-Verlag, 1982.
[\[Wm\]]{} Williams, S., Toeplitz minimal flows that are not uniquely ergodic, [*Z. Wahrscheinlichkeitstheori verw. Geb.*]{}, [**67**]{}, 1984, 95–107.
| {
"pile_set_name": "ArXiv"
} |
Appendix {#sup_models}
========
A summary of all VAE architectures used in this paper can be seen in Tbl \[tbl\_models\]. Next we provide various auxiliary details for the different datasets.
Dataset Optimiser
------------------------------ --------------------------------- --------- ---------------------------------------------------------------------
2D shapes adagrad [@Duchi_etal_2011] encoder fc 4096-1200-1200-10 (ReLU)
decoder fc 10-1200-1200-1200-4096 (tanh)
3D shapes adam [@Kingma_Ba_2014] encoder conv 32x6x6 (2-1)-64x6x6 (2-1)-512-32 (tanh)
decoder deconv 32-512-32x4x4 (2-1)-64x4x4 (2-1)-128x4x4 (2-1)
Amoeba adagrad [@Duchi_etal_2011] encoder fc 16384-400-205-10 (ReLU)
decoder fc 10-400-8392-16384 (ReLU)
Atari (breakout) adagrad [@Duchi_etal_2011] encoder conv 3x48x80-64x6x6 (2)-32x6x6 (2)-32x5x5 (2)-30 (tanh)
decoder deconv 30-3840-SU(2)-64x5x5-SU(2)-64x5x5-SU(2)-3x5x5-3x48x80 (tanh)
Atari (other) adam [@Kingma_Ba_2014] encoder conv 32x6x6 (2)-64x6x6 (2-1)-64x6x6 (2-1)-512-various (ReLU)
decoder deconv reverse of encoder (ReLU)
3D chairs [@Aubry_etal_2014] rmsprop [@Tieleman_Hinton_2012] encoder conv 32x6x6 (2)-64x6x6 (2)-256-10 (ReLU)
decoder deconv reverse of encoder (ReLU)
3D game rmsprop [@Tieleman_Hinton_2012] encoder conv 3x64x64-32x4x4 (2)-32x5x5 (2)-64x5x5 (2)-64x4x4 (ReLU)
decoder deconv reverse of encoder (ReLU)
: Various VAE architectures and optimisers were used for different experiments to show robustness of our approach. For convolutional architectures the numbers in parenthesis indicate: (stride-padding). SU stands for spatial upsampling.[]{data-label="tbl_models"}
2D shapes dataset
-----------------
We trained the fully connected architecture in Tbl. \[tbl\_models\] with cross-entropy cost function using adagrad [@Duchi_etal_2011] with learning rate of 1e-2.
Factor change classification
----------------------------
In order to quantify the degree of disentanglement learnt by the models we generated factor change data according to the pseudocode shown in Algorithm \[alg\_factor\_change\]. We used a linear classifier to learn the identity of the generative factor that produced $z_{diff}$. We used a fully connected neural network mapping between input of size $z_{diff}$ to an output of size 4 corresponding to the 4 generative factors (position X, position Y, scale and rotation) with softmax output nonlinearity and cross-entropy cost function. The classifier was trained with adagrad [@Duchi_etal_2011] with learning rate of 1e-2 until convergence.
All factor change classification results reported in the paper were calculated in the following manner. Ten replicas of each VAE experiment was run, each with a different random seed. Each of the ten replicas was evaluated three times using the factor change classification algorithm, each time with a different random seed. We then discarded the bottom 50% of the thirty resulting scores and reported the remaining results.
$objId \gets \text{randomly sample object identity}$ $changeFactor \gets \text{randomly sample factor identity}$ $changeDir \gets \text{Randomly sample the direction of change (+/-)}$
$groundTruth_{start}^{factor} \gets \text{randomly sample factor value}$
$groundTruth_{end} \gets groundTruth_{start}$ $groundTruth_{end}^{changeFactor} \gets \text{randomly sample a new value in the direction of } changeDir$
$x_{start} \gets \text{pixel representation of } groundTruth_{start}$ $x_{end} \gets \text{pixel representation of } groundTruth_{end}$
$z_{start} \gets Enc(x_{start})$ $z_{end} \gets Enc(x_{end})$
$z_{diff}^n \gets \frac{|z_{start}^{\mu} - z_{end}^{\mu}|}{max(|z_{start}^{\mu} - z_{end}^{\mu}|)}$
Zero shot inference regression
------------------------------
In order to map $z_{orig}$ to $z_{new}$, we used a fully connected linear neural network with smooth L1 loss trained with adagrad [@Duchi_etal_2011] with learning rate of 1e-2 until convergence.
Amoeba dataset
--------------
We trained the fully connected architecture in Tbl. \[tbl\_models\] with binary cross-entropy criterion and adagrad [@Duchi_etal_2011] optimizer with learning rate of 1e-2.
3D shapes dataset
-----------------
We trained a convolutional VAE (see Tbl. \[tbl\_models\]) with learning rate 1e-4 on a dataset of three 3D objects (cylinder, cube and pyramid) with three factors of variation (6 scales, 60 out of plane rotations and 26 colours). The 3D objects were rotating around the z-axis over $2\pi$ using 60 equidistant steps. The objects were generated in Blender and the 6 scales and 6x6 position translations were generated for each object in each rotational position using ImageMagick. The full dataset contained 38,880 frames of size 64x64. The decoder had Gaussian outputs.
Atari dataset
-------------
We trained a convolutional VAE (see Tbl. \[tbl\_models\]) with learning rate 1e-4 on frames from the Atari games Breakout ($\vz$ size 30, $\beta=1$), SeaQuest ($\vz$ size 10, $\beta=5$), Frostbite ($\vz$ size 100, $\beta=5$) and Enduro ($\vz$ size 100, $\beta=1.75$). The Atari dataset consisted of 1 million frames collected from a trained DQN agent [@Mnih_etal_2015]. The frames were pre-processed as described in [@Mnih_etal_2015]. The continuity of the dataset enabled the VAE to learn disentangled representations of the independent factors in the data (see video visualisations at http://tinyurl.com/jgbyzke). The decoder had Gaussian outputs. The model was trained using adam [@Kingma_Ba_2014] optimizer with learning rate of 1e-4.
3D chairs dataset
-----------------
For the 3D chairs dataset [@Aubry_etal_2014] we trained a convolutional VAE (see Tbl. \[tbl\_models\]) on 82 chair identities. The images were cropped and downsampled to 100x100 pixels. The decoder had Gaussian outputs. The model was trained using rmsprop [@Tieleman_Hinton_2012] optimizer with learning rate of 1e-5.
3D first person view maze navigation game dataset
-------------------------------------------------
We also trained a convolutional VAE (see Tbl. \[tbl\_models\]) on frames from a first person view 3D first person maze navigation game environment. The game frames were made greyscale and downsampled to 84x84 pixels. The dataset contained 1 million frames. This environment shares many properties with the real world: it is continuous and the dynamics of visual scene changes are similar to those experienced in the real world. After training the VAE was able to learn disentangled representations of several factors of variation present in the 3D game world (see video visualisations at http://tinyurl.com/jgbyzke). For example, certain single latent units learnt to represent changes in light, forward/backward movement and rotational movement. The VAE also learnt to allocate single latent units to represent the change in score and the rotation of the little character head at the bottom of the screen. For this experiment $z$ size was set to 32 and $\beta=1$. The decoder had Gaussian outputs. The model was trained using rmsprop [@Tieleman_Hinton_2012] optimizer with learning rate of 1e-4.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Studies of systems with two fermionic bands with [*repulsive*]{} interaction strength $U$ have a long history, with the Periodic Anderson Model (PAM) being one of the most frequently considered Hamiltonians. In this paper, we use Quantum Monte Carlo to study analogous issues for [*attractive*]{} interactions. As in the Periodic Anderson Model, we focus on a case where one band is uncorrelated ($U=0$), and focus on the effect of hybridization $V$ between the bands on the pairing correlations. A key difference with the PAM is that there is no sign problem, so that we are able to explore the physics of [*doped*]{} multi-band attractive systems at low temperatures whereas ground state properties of repulsive models can be determined only at half-filling. For small $V$, pairing in the $U<0$ layer induces pairing in the $U=0$ layer. At larger $V$ the ground state of the coupled system loses its superconducting character. The Quantum Monte Carlo data are complemented by results obtained with the Bogoliubov-de Gennes approximation.'
author:
- 'Aleksander Zujev$^{1}$, Richard T. Scalettar$^{2}$, George G. Batrouni$^{3,4,5}$ and Pinaki Sengupta$^{1}$'
title: 'Pairing Correlations in the two-layer attractive Hubbard Model'
---
Introduction
============
Cuprate superconductors are characterized by CuO$_2$ planes which are rather isolated from each other by intervening rare earth atoms. The larger (“$c$-axis") separation of Cu atoms perpendicular to the planes compared to the in-plane lattice constants ($a,b$) has focused theoretical attention on the magnetic and superconducting properties of two-dimensional models, most notably the square lattice Heisenberg and Hubbard Hamiltonians. Indeed, one of the most fundamental theoretical questions which arose in the initial investigations of possible models of high temperature superconductivity concerned the nature of the magnetism in the ground states of these Hamiltonians: Were they spin-liquids or did they exhibit long range order? It was established numerically[@reger88; @hirsch89; @white89] that both the Heisenberg Hamiltonian and the half-filled Hubbard Hamiltonian have long range antiferromagnetic order at $T=0$. A still-unresolved question is the nature of pairing in the doped Hubbard Hamiltonian.
A natural extension of these single layer questions concerns the behavior of spin and pairing correlations in coupled planes. Interlayer connections are required to elevate the magnetic ordering, which is possible only at $T=0$ in two dimensions, to the finite temperature Neél transitions observed experimentally. Similarly, the superconducting transition temperature tends to increase with the number of adjacent CuO$_2$ layers between the charge resevoirs, getting steadily larger from single layered La$_{2-x}$Sr$_x$CuO$_4$, to bilayer YBaCu$_3$O$_{6+y}$ to the HgBaCaCuO sequence where $T_c$ peaks at about $130^\circ$ K for materials with $n=3$ layers [@shimizu12]. Antiferromagnetism and thus superconductivity exhibit the expected behavior that adding additional (non-frustrating) couplings, and increasing the dimensionality, enhances the tendency to order.
The route from 2D to 3D is, however, not completely straightforward. Quantum Monte Carlo (QMC) studies of coupled Heisenberg layers[@sandvik94], have shown that when the interplane coupling grows sufficiently large, there is a Quantum Phase Transition (QPT) from a ground state with long range magnetic order to a spin-gapped ground state, in which singlets form between pairs of spins in the two planes[@foot1]. A similar effect is seen in QMC studies of the two-layer half-filled Hubbard Hamiltonian [@scalettar94].
The basic phenomenon of singlet formation usurping magnetic order in itinerant fermion Hamiltonians does not crucially depend on having the same interaction strength $U$, or hopping $t$, for the two species. Indeed, a Hamiltonian with $U=0$ in one (conduction) band and hopping $t=0$ in another (localized) band, is the commonly encountered Periodic Anderson Model (PAM). At half-filling, for weak hybridization between the conduction and localized bands, antiferromagnetic order is present. As the hybridization grows bigger, however, “Kondo" singlets form and destroy antiferromagnetic (AF) order. Thus the PAM exhibits a similar QPT as for two identical Hubbard layers ($t_1=t_2$ and $U_1=U_2$). This close similarity of the PAM and Hubbard bilayers emphasizes that the physics of fermions in multiple layers can equivalently be interpreted in in terms of multiple orbitals.
The purpose of this paper is to study the possibility of analogous phenomena in the case when there is an [*attractive*]{} interaction. More specifically, we examine the nature of pairing correlations for coupled planes (orbitals) in situations when an attraction is present only in one plane (orbital). We use the $-U$ Hubbard Hamiltonian as an appropriate simple model, and employ a combination of Quantum Monte Carlo (QMC) and Bogoliubov-de Gennes (BdG) Mean Field Theory. We will focus on a case analogous to that of the PAM, namely when $U=0$ in one of the layers. We are interested in both how the interlayer hopping affects the $s$-wave pairing correlations in a layer with $U<0$, and also whether pairing can be induced by a “proximity effect" in the $U=0$ layer. Such behavior would be directly analogous to the “RKKY" spin polarization cloud which is induced in the conduction band by the presence of the local moments in the PAM. A particularly interesting issue is whether the QPT which occurs at commensurate filling also occurs in the doped case.
The remainder of this paper is organized as follows: In Sec. II we write down the precise model and give an overview of the QMC and BdG computational methodologies. Secs. III and IV present the BdG and QMC results respectively. Sec. V recaps our conclusions.
Model and Computational Methods
===============================
Our starting point is the two layer (orbital) attractive Hubbard Hamiltonian, $$\begin{aligned}
\label{eq:attr-hub-mod}
H=&-&t\sum_{<ij> l \sigma}
( c^{\dagger}_{i l \sigma}c^{\phantom{\dagger}}_{j l \sigma} +
c^{\dagger}_{j l \sigma}c^{\phantom{\dagger}}_{i l \sigma})
\nonumber \\
&-&V \sum_{i \sigma}
( c^{\dagger}_{i 1 \sigma}c^{\phantom{\dagger}}_{i 2 \sigma} +
c^{\dagger}_{i 2 \sigma}c^{\phantom{\dagger}}_{i 1 \sigma})
- \sum_{i l \sigma} \mu^{\phantom{\dagger}}_l c^{\dagger}_{i l \sigma}
c^{\phantom{\dagger}}_{i l \sigma}
\nonumber \\
&-&\sum_{i}\big|U^{\phantom{\dagger}}_{1}\big|
(n^{\phantom{\dagger}}_{i 1 \uparrow}-\frac12)
(n^{\phantom{\dagger}}_{i 1 \downarrow}-\frac12)\,\,\,.
\label{eq:ham}\end{aligned}$$ Here an intralayer kinetic energy term $c_{i l \sigma}^\dagger
c^{\phantom{\dagger}}_{j l \sigma}$ describes the creation of a fermion with spin $\sigma$ on site $i$ of layer $l$ and its destruction on site $j$ of the same layer. A second interlayer kinetic energy term $c^{\dagger}_{i l \sigma}c^{\phantom{\dagger}}_{i
l^\prime \sigma}$ hybridizes the two layers $l,l^\prime$. Fermions of different spin feel an attractive interaction $-|U_l|$ on layer $l$. The chemical potentials $\mu_l$ control the filling. Our lattice geometry consists of two coupled square lattice of linear size $L$. We choose $U_1<0$ in the ‘superconducting’ plane and $U_2=0$ in the metallic one.
At half-filling, a particle-hole transformation formalizes the similarity of the repulsive and attractive models. The vanishing of antiferromagnetic order with increasing $V$ for $U>0$ bilayers immediately implies that ground state pairing (and charge density wave order, with which it is degenerate) must be destroyed as $V$ grows for $U<0$. However, whereas AF order occurs only at half-filling (and only at $T=0$) in the 2D repulsive Hubbard Hamiltonian, superconductivity appears away from half-filling, and at a finite (Kosterlitz-Thouless) transition temperature [@scalettar89; @moreo91], in the attractive Hubbard Hamiltonian. This opens up fundamentally new issues in the attractive case. Whether the fermions on sites in adjacent layers will still lock into a local object and destroy long range order when the filling is incommensurate is an interesting question. The absence of a sign problem in the case of the attractive Hubbard model enables the study of low temperatures even when the filling is incommensurate, allowing access to ground state properties.
We perform “determinant" QMC [@blankenbecler82] simulations of the Hamiltonian, Eq. \[eq:ham\] by writing down a path integral for the partition function $Z$. The exponential of the interaction term $U_l$ is decoupled with a Hubbard-Stratonovich field which allows the trace of the remaining exponentials of quadratic forms of the fermion operators to be performed analytically. The result is a sum over configurations of the Hubbard-Stratonovich field with a weight which takes the form of the product of the determinants of two matrices of dimension the spatial lattice size, one for each spin species. In the case of attractive interaction the two matrices, and hence their determinants, are identical, so that the weight is a perfect square and there is no sign problem.
We present results here for the real space pair correlation functions in the two orbitals, $$\begin{aligned}
P_l(j) &=& \langle \Delta^{\phantom{\dagger}}_{i+j,l}
\Delta^\dagger_{i,l} \rangle
\nonumber \\
\Delta^\dagger_i &=& c^{\dagger}_{i l \uparrow} c^{\dagger}_{i l
\downarrow}
\label{eq:paircorrspace}\end{aligned}$$ and also their associated structure factors, $$\begin{aligned}
{\cal P}^l_s = \frac{1}{N} \sum_{j} P_l(j)
\label{eq:pairstructure}\end{aligned}$$ as a functions of interband hybridization $V$, on-site attraction $U_l$, density $\rho$, and temperature $T$. The long distance behavior of $P^{\phantom{\dagger}}_l(j)$ yields the square of the superconducting order parameter, as does the lattice size dependence of the structure factor.
Eq. \[eq:ham\] can also be studied in Mean Field Theory via the solution of the BdG equations. In this approach, the four fermion (interaction) term is decoupled in the Hamiltonian itself, $$\begin{aligned}
{\cal{H}}_{\rm eff}=&-&t\sum_{<ij>l\sigma}
(c^{\dagger}_{il\sigma}c^{\phantom{\dagger}}_{jl\sigma} +
c^{\dagger}_{jl\sigma}c^{\phantom{\dagger}}_{il\sigma})
\nonumber \\
&-&V \sum_{i \sigma}
( c^{\dagger}_{i 1 \sigma}c^{\phantom{\dagger}}_{i 2 \sigma} +
c^{\dagger}_{i 2 \sigma}c^{\phantom{\dagger}}_{i 1 \sigma})
- \sum_{i l \sigma}{\tilde\mu^{\phantom{\dagger}}_{i l}}
c^{\dagger}_{i l \sigma}c^{\phantom{\dagger}}_{i l \sigma}
\nonumber \\
&-& \sum_{i}
\big|U^{\phantom{\dagger}}_{1}\big|\big[
\Delta^{\phantom{\dagger}}_{i1}
c^{\dagger}_{i 1 \uparrow}c^{\dagger}_{i 1 \downarrow}
+\Delta^{*\phantom{\dagger}}_{i1}
c^{\phantom{\dagger}}_{i 1 \downarrow}
c^{\phantom{\dagger}}_{i 1 \uparrow}\big]\,\,\,.
\label{eq:HBdG}\end{aligned}$$ Here the gap $\Delta^{\phantom{\dagger}}_{il}=\big<c^{\phantom{\dagger}}_{il\uparrow}
c^{\phantom{\dagger}}_{il\downarrow}\big>$ and density $\big<n^{\phantom{\dagger}}_{il\sigma}\big>=\big<c^{\dagger}_{il\sigma}
c_{il\sigma}\big>$ are determined self-consistently by diagonalizing the quadratic BdG Hamiltonian and putting together the eigenvalues and eigenvectors appropriately to compute refined values which are inserted back in the Hamiltonian in an iterative process. A related BdG treatment of superconductivity in the presence of randomness (spatially varying $\mu_i$) is given in \[\]. ${\tilde\mu_{il}}=\mu+\big|U_{l}\big| \langle n_{il} \rangle/2$ includes a site-dependent Hartree shift with $\langle n_{il} \rangle
=\sum_{\sigma} \langle n_{il\sigma} \rangle$.
In the general case, when inhomogeneous terms are present in the Hamiltonian, or are expected to develop spontaneously, (like the charge and spin stripes of the doped, repulsive Hubbard model), the order parameter and densities are allowed to depend on site index and the diagonalization must be done numerically. If translation invariance is present, Eq. \[eq:HBdG\] can be diagonalized analytically by going to momentum space. The resulting momentum sums are typically still done numerically, but larger lattices can be studied. We have implemented both approaches in the work presented here.
Although inhomogeneous Hartree-Fock theory allows for spatial variation of the densities and superconducting order parameters, the expectation values in Eq. \[eq:HBdG\] are independent of imaginary time, unlike the fluctuating Hubbard-Stratonovich field in the QMC approach. This leads to a less accurate treatment of interparticle correlations, but a numerically much more simple problem. In particular, larger spatial lattices can be studied, and the BdG solution can exhibit broken symmetry so that $\Delta$ itself can be nonzero, rather than having to be extracted from the asymptotics of the correlation function Eq. 2. Here the combination of QMC and BdG will serve to provide complementary information.
In the case of the BdG calculations our focus will be on the value of the order parameter $\Delta$ as a function of the same parameters as varied in the QMC. As mentioned above, the asymptotic value of $P_l(j)$ is a measure of $\Delta^2$. The BdG approach, since it neglects fluctuations, yields larger values of $\Delta$ than those of the QMC, which is an exact method.
Bogoliubov-de Gennes Results
============================
The results of BdG calculations are summarized in Fig. \[bfig:bdg\]. We consider two layers with $U_1/t=-8$ and $U_2=0$, at a density of $\rho=0.8$ electrons per site and study the evolution of the pairing correlation, $\Delta_l$, in the two layers as a function inter-layer coupling, $V$. Pairing in the correlated layer is suppressed by coupling to the non-correlated layer and eventually destroyed completely at a critical $V_c$, analogous to the destruction of antiferromagnetism by singlet formation in the repulsive case. A further interesting feature of Fig. \[bfig:bdg\] is that the $l=2$ layer, with $U_2=0$, develops induced pairing with the onset of inter-layer coupling. The pairing amplitude, $P_2$, increases with increasing $V$ up to a maximum before decreasing and eventually vanishing at $V_c$. For $V> V_c$, the ground state consists of weakly interacting inter-layer dimers. Crucially, the pairing order parameter in [ *both layers*]{} remains non-zero over a finite range of inter-layer coupling – the induced pairing reported previously at half-filling extends to finite dopings. In the next section, we present extensive QMC results to confirm and complement the BdG results.
Quantum Monte Carlo Results
===========================
Half Filling
------------
We begin our discussion of QMC results at half-filling, $\rho_1=\rho_2=1$. As discussed previously, this corresponds to the particle-hole symmetric point $\mu_1=\mu_2=0$ where there exists an exact mapping to the repulsive Hubbard model and the results can be benchmarked against previous studies of the Hubbard model \[\]. Lattices of the form $2\times
L\times L$, with $4\leq L \leq 14$ were studied with periodic boundary conditions over a wide range of inter-layer hybridization, and on-site interaction for the interacting layer. An inverse temperature $\beta
t=L$ was found to be sufficient for the observables to have converged to their ground state values.
The spatial dependence of the pair correlations $P_l(j)$ at half-filling $(\rho=1)$, intersheet hybridization $V=0.75 t$, and interactions $U_1=-6, U_2=0$ is shown in Fig. \[fig:P1db16D\]. The separation $j$ follows a trajectory along the $x$ axis to maximal $x$ separation $(\frac{L}{2},0)$ on a lattice with periodic boundary conditions, and then to $(\frac{L}{2},\frac{L}{2})$ before returning to separation $(0,0)$. Results for lattices with $L=8, 12$ are shown in the figure. The interacting layer $l=1$ exhibits clear long range order with $P_1(j)$ nearly independent of $j$ beyond a separation of approximately $\frac{L}{4}$. The noninteracting layer $l=2$ also exhibits proximity-effect induced long range order, although the correlation function is roughly an order of magnitude smaller. This corresponds to an order parameter $\Delta_l$ which is approximately a factor of three smaller.
Increasing the inter-layer hybridization reveals that the pair structure factor for the interacting layer, $P^1_s$, decreases monotonically and eventually the intra-layer pairing is destroyed at some critical $V_c$. On the other hand, the pairing structure factor for the non-interacting layer $P^2_s$ varies non-monotonically with $V$. It is vanishingly small at small $V$. As $V$ increases, long-range pairing correlations increase, reach a maximum at an intermediate $V$ (which depends on the strength of on-site interaction in the interacting layer) and then decrease continuously to zero at $V_c$. For $V>V_c$, the ground state is dominated by [*inter-*]{}layer singlets. Fig. \[fig:PFSS1\] shows the finite size dependence of the pairing structure factor for the two layers at several values of $V$. For $V
\gtrsim1.6$, both $P^1_s$ and $P^2_s$ extrapolate to zero in the thermodynamic limit indicating the absence of intra-layer pairing in this regime.
Here and further, the finite size scaling was done by linear fit in the plane $1/L \; - \; P^l_s$. The scatter of the data in Fig. \[fig:PFSS1\] for different values of the interlayer hybridization $V$ is dominated by the error bars associated with the linear fit to the structure factor on different lattice sizes used in the extrapolation, as opposed to the statistical errors for an individual run. The data scatter serves as a stand-in to provide a visual indication of the uncertainty in the the calculation.
![The dependence of the ground state pair correlation functions, $P_{1,2}(j)$, on separation $j$ in a bilayer system comprised of one correlated layer coupled to an uncorrelated layer for two different lattice sizes $L$. The system is at half-filling, $\rho=1$, with interplane hybridization $V=0.8t$, and on-site interaction strength $U_1=-6t$ for the correlated layer. The correlation functions converge to non-zero value at large separations, providing clear evidence for long-range order. Despite the absence of interactions, there is proximity-effect induced long-range order in the uncorrelated layer, $l=2$, with an order parameter $\Delta_2$ approximately a factor of three lower than for the correlated layer. As in Fig. \[bfig:bdg\], finite size effects are modest. []{data-label="fig:P1db16D"}](P12t0.8db12D.eps){height="2.5in" width="3in"}
![The pair structure factor for a bilayer at half-filling for attractive on-site interaction in the correlated layer $U=-10t$. The main panels show the extrapolated structure factor as a function of the inter-layer coupling, $V$. The insets show the extrapolation of the data from finite-sized systems to the thermodynamic limit. Top panel: The pair structure factor in the correlated layer, $P_s^{1}$. For inter-layer hybridization $V \le 1.6$ the extrapolated structure factor is non-zero indicating the existence of long-range order (LRO). Bottom panel: The pairing structure factor $P_s^{2}$ in the non-interacting layer. $P_s^{2}$ increases from zero for $V>0$, signaling the onset of induced pairing. With further increase of $V$, LRO vanishes simultaneously in both layers. []{data-label="fig:PFSS1"}](P12tU1-80.eps){height="3.5in" width="3in"}
Incommensurate Filling
----------------------
We now turn to the case of doped planes. As mentioned in the Introduction, this is interesting for two reasons. First, there is a possibility of a finite temperature KT transition because the CDW-SC symmetry is broken. Second, simulations of coupled, doped repulsive layers at low $T$ are not possible. Hence there is no DQMC information available for the nature of magnetism in coupled layers away from half-filling. Our simulations address this issue, albeit for attractive on-site interactions.
The results in Fig. \[fig:PFSS2\] confirm unambiguously that induced pairing in the uncorrelated layer extends to finite dopings away from half-filling; indeed the pair correlation functions in the two layers are qualitatively similar to those at half-filling. Pairing in the correlated layer decreases monotonically to zero at a finite $V_c$ and that in the non-interacting layer increases rapidly from zero as the inter-layer hybridization is turned on, reaches a maximum value at intermediate $V$ and then decreases to zero. At intermediate to strong values of the on-site interaction ($|U| \gtrsim 6t$), the ground state of a system of coupled correlated and uncorrelated layers away from half-filling consists of intra-layer pair formation in [ *both*]{} layers over a finite non-zero range of inter-layer coupling. Eventually, at sufficiently strong inter-layer hybridization, pairing in both layers is destroyed and the ground state is dominated by singlet formation between the layers. Induced pairing is found to be absent for $|U| \lesssim 6t$. The simulation results are summarized in Fig. \[fig:PFSS2\] where the extrapolated pairing structure factor for the two layers are shown as a function of the inter-layer hybridization, $V$, for some representative values of the on-site interaction in the correlated layer at a fixed value of the density, $\rho = 0.8$. The maximum of the pairing structure factor in the uncorrelated layer and the critical value of inter-layer coupling for complete suppression of pairing, $V_c$, increase with $|U|$ . Our results are consistent with the disappearance of pairing in both the layers occurring simultaneously at $V_c$. The error associated with determining the extrapolated pairing structure factor in the thermodynamic limit limits the accuracy with which $V_c$ can be determined to about $\pm 0.2t$. In a marked departure from the BdG results, the QMC data shows unambiguously that a finite non-zero inter-layer coupling is required for the onset of induced pairing in the uncorrelated layer. This is a consequence of the mean field nature of the BdG results.
Fig. \[fig:PFSS2\] shows that $P_1$ increases systematically as $U$ changes from $U=-4$ to $U=-10$ and, indeed, shows no sign of saturation. This is at variance with the behavior of the magnetic response of the repulsive Hubbard Hamiltonian which first increases with $U$ but then reaches a maximum at $U \sim 8t$ before falling. This behavior is understood qualitatively from the fact that the superexchange $J=4t^2/U$ which provides the large $U$ magnetic energy scale declines with $U$. Thus, for example the Neél temperature of the 3D Hubbard model at half-filling is maximized at $U \sim 8t.$ The analog in the attractive Hubbard model is the fact that the pairs become heavy, with an effective hopping $t_{\rm eff} \sim t^2/U$ associated with the fact that a pair must be (temporarily) broken before it can hop. This analogy suggests the pairing response might also be maximal at an intermediate $U$. However, because of the sign problem, it is not known from DQMC whether such a non-monotonic behavior exists in the repulsive model away from half-filling. Our results here for the attractive case, which show no sign of saturation with increasing $U$, suggest that it might not.
![The pair structure factor for a bilayer away from half-filling, $\rho=0.8$, for three representative values of the attractive on-site interaction in the correlated layer: $U_1=-4t,
-6t, -10t$. For $|U_1|\lesssim 6t$, $P_s^{2}=0$ at all values of $V$, indicating the absence of induced pairing. At stronger on-site interaction ($|U_1|>6t$) in the correlated layer, induced pairing appears in the uncorrelated layer over a finite range of $V$. The critical $V_c$ for destruction of intra-layer pairing increases with $|U_1|$. Insets: Finite size scaling for $U_1=-10t$.[]{data-label="fig:PFSS2"}](P12tU1.eps){height="3.5in" width="3in"}
Having established the existence of induced pair formation at finite doping and the evolution of the associated structure factor with $V$, we investigate the dependence of the pairing phenomenon on the density of electrons in the layers. Our results (Fig. \[fig:PFSS3\]) show that induced pairing occurs generically at all densities. For all density values studied, the pairing structure factor behaves in a qualitatively similar manner, with the pairing strength optimally enhanced at $\rho=0.8$, as evidenced by the initial increase in peak height with decreasing density. Pairing appears to show a modest decline as the density is reduced further to $\rho=0.6$, although the effect is rather small given the scatter in the data. Such a non-monotonic doping dependence would be consistent with the behavior of the superconducting $T_c$ with doping for a single 2D sheet. $T_c$ rises abruptly from zero at half-filling where charge density order competes with pairing, but then shows a gradual decline below an optimal doping [@scalettar89].
![The pair structure factors at various filling values. Induced pairing occurs at all densities, with the amplitude of induced pairing increasing with doping away from half-filling. Insets: Finite size scaling for $\rho=0.8$. []{data-label="fig:PFSS3"}](P12tU1-10.eps){height="3.5in" width="3in"}
Conclusions
===========
In this paper we have used Quantum Monte Carlo and Bogoliubov-de Gennes mean field theory to elucidate the properties of a multi-orbital [ *attractive*]{} Hubbard Hamiltonian. Our key conclusions are the quantification of the induced pairing in a noninteracting orbital by superconductivity in a correlated orbital, and, conversely, the suppression of pairing in an orbital with an attractive interaction through its coupling to a noninteracting orbital.
As we have emphasized in the introduction, analogous issues for repulsive models have a long history. In fact, over the last 4-5 years, the question of ‘orbitally selective Mott transitions’, has been extensively explored. The fundamental objective has been an examination of the density of states to determine whether or not one subsystem can be metallic (zero energy gap at $E_{\rm F}$) and the other insulating (nonzero gap), or whether the coupling forces the transition to occur simultaneously in all orbitals (layers) [@liebsch04; @liebsch05; @arita05; @inaba06; @costi07; @koga04; @ferrero05; @demedici05; @ruegg05; @inaba05; @knecht05; @biermann05]. Our work suggested that for attractive interactions the destruction of superconductivity at large inter-orbital hybridization $V$ always occurs simultaneously. It is to be noted, however, that it appears that at weak $V$ pairing can exist in the correlated orbital before it is induced in the noninteracting one. That is, we find that for the attractive case at small interlayer coupling, SC exists only in the interacting layer despite the fact that at larger $V$ the vanishing of SC appears at a common $V_c$. Our work is related to several other recent determinant QMC and Lanczos studies of the attractive Hubbard model. Assman [*et.al*]{} [@assmann12] found that in an attractive Hubbard model with a smoothly varying chemical potential (such as would be present in a confined ultracold atomic cloud) pairing is significantly increased in the half-filled portion of the lattice through its contact with doped regions. That is, the suppression of superconductivity caused by the appearance of a degenerate charge density wave phase at $\rho=1$ is eliminated. Paiva [*et.al.*]{} explored a one-dimensional model of borocarbides in which $U<0$ and $U=0$ sites alternate [@paiva03]. Superconductivity is found to be possible only above a critical density.
Finally we note that it was recently shown in repulsive models with more than two layers that there can be further interesting and even surprising evolution of magnetic properties with the hopping across an interface between a Mott insulator and a metal [@euverte12]. It would be interesting to pursue related questions in the repulsive model and, in particular whether the formation of a layer of local pairs at the boundary can lead to a shielding of penetration effects.
We thank Simone Chiesa and Axel Euverte for very helpful dicsussions. This work was supported in part by an ARO Award W911NF0710576 with funds from the DARPA OLE Program, by the CNRS-UC Davis EPOCAL LIA joint research grant; and by the CNRS-National University of Singapore FSQL LIA joint research grant.
[99]{}
J. D. Reger and A. P. Young, Phys. Rev. [**B37**]{}, 5978 (1988).
J.E. Hirsch and S. Tang, Phys. Rev. Lett. [**62**]{}, 591 (1989).
S. R. White, D. J. Scalapino, R. L. Sugar, E. Y. Loh, Jr., J. E. Gubernatis, and R. T. Scalettar, , 506 (1989).
S. Shimizu, S. Iwai, S.I. Tabata, H. Mukuda, Y. Kitaoka, P.M. Shirage, H. Kito, and A. Iyo Phys. Rev. [**B83**]{}, 144523 (2011); S. Shimizu, S.I. Tabata, S. Iwai, H. Mukuda, Y. Kitaoka, P.M. Shirage, H. Kito, and A. Iyo Phys. Rev. [**B85**]{}, 024528 (2012); and references cited therein.
A.W. Sandvik and D.J. Scalapino, Phys. Rev. Lett. [**72**]{}, 2777 (1994).
This AF-singlet transition in coupled planes is closely analogous to the difference between the magnetic correlations of odd and even rung ladders. An odd number of coupled one dimensional chains is gapless and has power law decay of spin correlations, while even number of chains are gapped and have exponential decay. See, for example, E. Dagotto and T.M. Rice, Science [**271**]{}, 618 (1996).
R.T. Scalettar, J.W. Cannon, D.J. Scalapino, and R.L. Sugar, Phys. Rev. [**B50**]{}, 13419 (1994).
M. Vekic, J.W. Cannon, D.J. Scalapino, R.T. Scalettar, and R.L. Sugar, Phys. Rev. Lett. [**74**]{}, 2367 (1995).
R. Blankenbecler, D. J. Scalapino and R. L. Sugar, , 2278 (1981).
A. Ghosal, M. Randeria, and N. Trivedi, Phys. Rev. Lett. [**81**]{}, 3940 (1998).
A. Ghosal, M. Randeria, and N. Trivedi, Phys. Rev. [**B65**]{}, 014501 (2001).
A. Liebsch, Phys. Rev. [**B70**]{}, 165103 (2004).
A. Liebsch, Phys. Rev. Lett. [**95**]{}, 116402 (2005).
R. Arita and K. Held, Phys. Rev. [**B72**]{}, 201102(R) (2005).
K. Inaba and Akihisa Koga, Phys. Rev. [**B73**]{}, 155106 (2006).
T.A. Costi and A. Liebsch, Phys. Rev. Lett. [**99**]{}, 236404 (2007).
A. Koga, N. Kawakami, T.M. Rice, and M. Sigrist, Phys. Rev. Lett. [**92**]{}, 216402 (2004).
M. Ferrero, F. Becca, M. Fabrizio, and M. Capone, Phys. Rev. [**B72**]{}, 205126 (2005).
L. de’Medici, A. Georges, and S. Biermann, Phys. Rev. [**B72**]{}, 205124 (2005).
A. Rüegg, M. Indergand, S. Pilgram, and M. Sigrist, Eur. Phys. J. [**B48**]{}, 55 (2005).
K. Inaba, A. Koga, S.I. Suga, and N. Kawakami, Phys. Rev. [**B72**]{}, 085112 (2005).
C. Knecht, N. Blumer, and P.G.J. van Dongen, Phys. Rev. [**B72**]{}, 081103(R) (2005).
S. Biermann, L. de’Medici, and A. Georges, Phys. Rev. Lett. [**95**]{}, 206401 (2005).
R.T. Scalettar, E.Y. Loh, Jr., J.E. Gubernatis, A. Moreo, S.R. White, D.J. Scalapino, R.L. Sugar, and E. Dagotto, Phys. Rev. Lett. [**62**]{}, 1407 (1989).
A. Moreo and D.J. Scalapino. Phys. Rev. Lett. [**66**]{}, 946 (1991).
E. Assmann, S. Chiesa, G.G. Batrouni, H.G. Evertz, and R.T. Scalettar, Phys. Rev. [**B85**]{}, 014509 (2012).
T. Paiva M. El Massalami; and R.R. dos Santos, J. of Phys.-Cond. Mat. [**15**]{}, 7917 (2003).
A. Euverte, F. Hebert, G.G. Batrouni, S. Chiesa, and R.T. Scalettar, Phys. Rev. Lett., to appear.
| {
"pile_set_name": "ArXiv"
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.