Spatial structures and dynamics of kinetically constrained models for glasses
Extracted source text ยท arXiv cond-mat/0306746
\documentclass[aps,prl,twocolumn,groupedaddress]{revtex4}
\pagestyle{empty}
\usepackage{amsmath}
\usepackage{graphicx}
\newcommand{\be}{\begin{equation}}
\newcommand{\ee}{\end{equation}}
\newcommand{\bea}{\begin{eqnarray}}
\newcommand{\eea}{\end{eqnarray}}
\newcommand{\ba}[1]{\begin{array}{*{#1}{c}}}
\newcommand{\ea}{\end{array}}
\newcommand{\pr}{{\rm Prob}}
\newcommand{\pd}[2]{\frac{\partial #1}{\partial #2}}
\newcommand{\la}{\langle}
\newcommand{\ra}{\rangle}
\newcommand{\va}{v}
\newcommand{\s}{s}
\pagestyle{empty}
\begin{document}
\title{Spatial structures and dynamics of kinetically constrained models of glasses }
\author{Cristina Toninelli}
\email[]{cristina.toninelli@roma1.infn.it}
\affiliation{Dip. Fisica, Universita' La Sapienza, 00165 P.le A. Moro 5, Roma}
\author{Giulio Biroli}
\email[]{biroli@spht.saclay.cea.fr}
\affiliation{Service de Physique Th{\'e}orique, CEA/Saclay-Orme des Merisiers,
F-91191 Gif-sur-Yvette Cedex, FRANCE }
\author{Daniel S. Fisher}
\email[]{fisher@physics.harvard.edu}
\affiliation{Lyman Laboratory of Physics, Harvard University, Cambridge, MA 02138, USA}
\begin{abstract}
Kob and Andersen's simple lattice models for the dynamics of structural
glasses are analyzed. Although the particles have only hard core
interactions, the imposed constraint that they cannot move if surrounded by
too many others causes slow dynamics.
On Bethe
lattices a dynamical transition to a partially frozen phase occurs. In
finite dimensions there exist rare mobile elements that destroy the
transition. At low vacancy density, $v$, the spacing,
$\Xi$, between mobile elements diverges exponentially or faster
in $1/v$. Within the mobile elements, the dynamics is intrinsically cooperative and the characteristic time scale diverges faster than any power of $1/v$ (although
slower than $\Xi$). The tagged-particle diffusion coefficient vanishes roughly as $\Xi^{-d}$.
%More general implications are discussed briefly.
\end{abstract}
\pacs{}
\keywords{}
\maketitle
In many liquids dramatic slowing down occurs on cooling, equilibrium
cannot be achieved, and structural relaxation becomes complicated and spatially heterogeneous
\cite{DeBenedettiStillinger,ErdigerWeitz,Glotzer}. Despite a
great deal of effort, these remarkable phenomena associated with the
``glass transition" are still far from understood. Indeed, the most basic issues are unresolved: Is the rapid slowing down due to proximity to an equilibrium phase transition, albeit in a restricted part of phase space? Or is the underlying cause entirely dynamical? In either case, are there characteristic length scales that grow substantially near the glass transition? \\
Theoretical developments have been hampered by a shortage of models that capture essential features yet are simple enough to be analyzed. Exceptions are {\it kinetically constrained} models \cite{Jackle,ReviewKLG} based on
the Ansatz
that the glass transition is caused by geometrical constraints on
dynamical rearrangements with static correlations playing no role
(see e.g. \cite{ChandlerGarrahanKLG}).
These are also models for
granular media (see e.g. \cite{Sellitto})
for which slow dynamics occurs already at densities well below close-packing \cite{Jaeger}.\\
We focus on one of the simplest kinetically constrained models, the Kob-Andersen model (KA) \cite{KA}. KA is a single component lattice gas with no static interactions other than hard core exclusion and dynamics given by a continuous time stochastic process. Each particle attempts, at a fixed rate, to move to a randomly chosen empty neighboring site; {\it but} the jump
is allowed only if both before and after the move the particle has
no more than some number, $m$, neighboring particles.
This corresponds to {\it vacancies} moving only if the initial and final
sites have at least $s=z-m-1$ neighboring vacancies, with $z$ the
coordination number of the lattice.
Since this dynamics satisfies detailed
balance, the
trivial distribution that is uniform over all configurations with a fixed number of particles is stationary:
there can thus be no equilibrium transition.
Nevertheless, the dynamics is
sluggish and heterogeneous at high density \cite{KA,FranzKA}.
For a three dimensional cubic-lattice with $s=2$,
fits of the self diffusion coefficient of a tagged-particle, $D_S$, strongly suggest a dynamical
glass
transition
at $\rho_c\simeq 0.881$, above which $D_S$
appears to vanish and the structural relaxation time to diverge
\cite{KA}.\\
%Is this a real dynamical transition? More generally, does one exists for some KA models? If not, what is the
%mechanism for the rapid slowing down? Are there collective
%processes involved? What are the characteristic length scales,
%and how are the time scales related to these? What causes dynamical heterogeneities? And, most
%importantly, what can one learn about
%more realistic models? We will answer all but the last of these on which we
%speculate in the conclusions.\\
Our analysis shows that there are four classes of behavior for general KA models. These can be characterized by the dependence of $D_S$ on the {\it vacancy density}, $v$: ({\cal N}) {\it normal} with $D_S\sim \va^q$ as $\va\to 0$; ({\cal C})
{\it collective freezing} with $D_S\to 0$ faster than any power of
$\va$ as $\va\to 0$ ; ({\cal T}) {\it dynamical transition} with $D_S\to 0$ as $\va \searrow \va_c$ with a non-zero critical $\va_c=1-\rho_c$; and ({\cal F}) {\it always frozen} with a finite fraction of the particles never moving for any $v$.
{\it Normal} behavior occurs if a finite cluster of $q$ vacancies can move through an otherwise-totally filled lattice, e.g. a triangular lattice with $s=1$ in which vacancy pairs ($q=2$) are mobile. This yields $D_S$ proportional to the cluster concentration, $v^q$. At the opposite extreme, a square lattice with $s=2$ is {\it always frozen} as a four particle square can never move.
We focus on the interesting intermediate cases, {\cal C} and {\cal T}.
We show that a
{\it dynamical transition}
takes place on (tree-like) Bethe lattices --- even with finite size loops. But such a transition {\it cannot}
occur on finite dimensional lattices: if neither normal nor always frozen, these exhibit {\it collective freezing} with the dominant dynamical processes involving a number of vacancies that diverges as $v\to 0$. This class ({\cal C}) includes the original $s=2$ cubic lattice.
We first analyze Bethe lattices: infinite tree-like graphs with fixed connectivity $z$
which crudely approximate high--dimensional or high--coordination--number lattices.
Monte-Carlo simulations
% of KA on Bethe lattices
of $10^{4}$ sites
with $z=4$ and $s=1$
(crudely mimicking the $s=1$ square lattice)
suggest a dynamical correlation length (as defined in \cite{FranzDH}) that diverges at a
critical density $\rho_{c}\simeq 0.89$. \cite{long}
% Approaching the transition from low density the
% dynamical susceptibility $\chi_{4} (t)$, introduced in \cite{FranzDH}
% to unveil
% in glass forming liquids,
Approaching this apparent dynamical transition the local--density correlation function displays two-step relaxation, as for supercooled liquids. \cite{ReviewOE}
% which gradually evolves an infinite plateau for $\rho >\rho_{c}$.
The tree-like structure of Bethe lattices enables
analytic study by iteration. Arranging the tree with $k=z-1$ branches going
up from each node and one going down, we focus on the following
events for a chosen {\it node} and rearrangements {\it restricted} to the {\it subtree above it} :
(${\cal A}$) node is occupied by a particle which cannot move up
even if the site
below it is empty; (${\cal B}$) node is empty but a particle can never move onto it from below; (${\cal C}$)
node is occupied by a particle which can only move up if the node below it is empty; (${\cal D}$) none of the above. Self-consistent equations for the probabilities of these events can be found by combining $k$ branches together with the site
below them to which they are all linked \cite{long}.
For $s=1$ with $z=k+1=4$,
this yields a critical density,
$\rho_c \cong 0.888825$ in agreement with simulations.
Below $\rho_{c}$, $\pr[{\cal A}]=\pr[{\cal B}]=0$; particles are mobile and diffusive at long times.
At $\rho_c$, $\pr[{\cal A}]=A_c\neq 0$ and the fraction of immobile particles jumps discontinuously; so does the {\it extensive configurational entropy},
$S_C=\ln[\# \mbox{different ergodic components}]$. \cite{long}
Above $\rho_c$, $\pr[{\cal A}]-A_c \sim \sqrt{\rho-\rho_c}$, the singularity indicative of diverging length and time scales. %The diverging length scale is purely dynamical
%and indeed the dynamical susceptibility $\chi_{4} (t)$ --- related in equilibrium to the density-density
% correlation function \cite{FranzDH} --- diverges at the transition.
The KA dynamical transition on this Bethe lattice (like $1/r^2$ percolation in one-dimension \cite{lr-perc})
thus has both critical and ``first order" characteristics.
Interestingly,
infinite-range {\it quenched random} p-spin models, conjectured to be related
to the glass transition, also exhibit
a dynamical transition with similar
features \cite{ReviewOE}.
The basic features found above hold for
generic KA models on Bethe lattices with $s\neq 0,k$,
including those with {\it finite size loops}
\cite{long}, i.e. for the so called pure Husimi trees \cite{Monroe}
and for Bethe lattices where single sites are replaced by $L\times L$
squares (these interpolate among the Bethe lattice, $L=1$, and the
square lattice, $L=\infty$).
The only exceptions are cases in which particles with two
occupied neighbors cannot move ($m=1$): the transition
is then continuous and $\rho_c=1/k$, the critical density for conventional
site percolation. This equivalence is due to the fact that the dynamical
transition is associated with the emergence of an infinite cluster of
permanently blocked particles. In general a frozen phase will only exist
if an infinite cluster remains after all the particles mobile under KA rules have been iteratively removed. This is
a type of {\it bootstrap percolation}, already introduced
in this \cite{KA} and broader \cite{Adler, Leath,AizenmanLebowitz}
contexts. For Bethe lattices, an exact solution
\cite{long} (see \cite{Leath} for a simpler example)
yields the same critical density found above.
We now turn to finite dimensions and show that the dynamical transition is destroyed by exponentially rare
processes which will only occur in sufficiently large systems. Nevertheless, at high densities the dynamics is intrinsically collective and extremely slow. Furthermore, the ghost of the Bethe lattice transition may cause a sharp crossover at an apparent critical density.
We focus initially on the simple $s=1$ square lattice case.
First, let us define a configuration as {\it framed} if all its boundary sites are
empty (see for example the square on the right of fig. 1); and as {\it
frameable} if, by an allowed sequence of moves, a {\it framed}
configuration can be reached (see for example the square on the left of fig. 1).
A key observation is that the lines of vacancies on the boundaries of a framed square can be
shifted (moving vacancies starting from the corner) so
that any nearest neighbor pair of particles inside the square can be sandwiched
between two lines of vacancies.
The
position of this pair of particles can then be exchanged
and the vacancy lines afterwards shifted back to their original
positions. The net result is a pair exchange of particles
with all other particles returned to their original positions.
By combining such pair exchanges all frameable configurations with
the same density can be connected. Thus, these form an irreducible
component $\cal Z_\rho$ of the configuration space.
We must now show that almost all random configurations with a given $\rho<1$ belong to $\cal Z_\rho$ in the
thermodynamic limit.
This can be done iteratively by observing that an $l$ by $l$ frameable
configuration that has at least two vacancies externally adjacent to each of its sides is
also an $l+2$ by $l+2$ frameable configuration, see fig. 1. Starting from a two by two frameable
``nucleus" of vacancies, one can thus grow an $L$ by $L$
frameable configuration if the requisite vacancies are present in each concentric shell. The probability that this occurs is
$P_{L} (\rho )=(1-\rho )^{4}\prod_{l=2}^{L/2}
[1-\rho ^{2l}-2l\rho ^{2l-1}(1-\rho)]^{4}$
%$B_{L}(\rho )=(1-\rho )^{4}\prod_{l=1}^{(L-2)/2}
%(1-\rho ^{2l}-2l\rho ^{2l-1}(1-\rho))^{4}$.
%\begin{equation}
%P_{L}(\rho )=(1-\rho )^{4}\prod_{l=1}^{(L-2)/2}
%(1-\rho ^{2l}-2l\rho ^{2l-1}(1-\rho))^{4}
%\end{equation}
which converges to a non-zero probability,
$P_\infty(\rho) $. Thus for all $\rho$ infinite frameable squares exist with all frameable configurations within them being reachable. For small vacancy densities, $P_\infty(\rho) \simeq
e^{-2 \hat{K}_F/v} $
with $\hat{K}_{F}\simeq 4.48$ .
Moreover, $P_L(\rho)$ depends weakly on
$L$ for $L>\xi(\rho) \sim\ln (1/\va )/\va$; $\xi$ is
thus the {\it core} size of frameable regions. Note that $P_L$ is the probability that a frameable square can be constructed around a nucleus at a {\it fixed} position.
On the other hand the probability, $F_L(\rho)$, that a large square
is frameable is roughly given by considering the $O(\left[L/\xi\right]^2)$ possible positions of a nucleus within the square. Although the probabilities of a configuration being frameable around each of these possible nuclei are not independent, if the nuclei are separated by more than a core diameter they are roughly independent. Indeed, it can be
proven that $F_L$ asymptotes to one for $L\gg \Xi$ with the long crossover length scale $\Xi(\rho)$ given, as expected from the above argument, by
$\Xi^2 P_\xi(\rho)\sim \Xi^2 P_\infty(\rho) \sim \xi^{2}$.
We have shown that at any density almost all configurations of sufficiently large systems can be reached from one another. This {\it excludes} the possibility of a dynamical transition \cite{footnoteergodicity}.
% and is one in thermodynamic limits
The phase space of systems with linear size, $L$, larger than the crossover length $\Xi$ is covered almost entirely by a single ergodic
component, while the phase space for $L<\Xi$ is typically decomposed into many disjoint parts. The framing analysis yields an {\it upper
bound} for $\Xi$.
A lower bound for $\Xi$ can be obtained from
bootstrap percolation arguments.
>From a random configuration with density $\rho $ iteratively
remove all particles with less than three occupied neighbors. If
any particles remain they must form a system-spanning cluster of frozen
particles and the configuration space will thus be broken up
into many disjoint pieces. This happens with high probability for $\ln L< K_F/\va$ with $K_F=\pi^{2}/18$ \cite{AizenmanLebowitz,new-bootstrap},
yielding a lower bound for $\Xi$ with the same density dependence as the above upper bound, but a different constant $K_F$ less than $\hat{K}_F$ of the upper bound.
However, a better definition of
framed configurations yields a $\hat{K}_F$ which is identical to
$K_{F}$ \cite{long}.
The corresponding frame is {\it minimal} and
the asymptotic form of the crossover length $\Xi \sim
e^{\pi^2/18\va}$ {\it exact} (up to subdominant factors).
\begin{figure}[bt]
\centerline{\includegraphics[width=0.9\columnwidth]{fraprl.eps}}
\caption{
On the left there is a $4$ by $4$ framed configuration with two vacancies (white circles) adjacent
to each side. After moving the external vacancies at the
corners the framed square can be expanded to the $6$ by $6$ framed
configuration at right.}
\label{fig}
\end{figure}
%In fact one has to
%consider the optimally framed configurations of a
%$L\times L$ square (with $L$ even) as those with
%$L+1$ vacancies arranged along alternate sites of two perpendicular
% sides with an additional one in the corner, plus any number of additional vacancies.
All other finite-dimensional KA models can be analyzed by generalizing the notion of frameable configurations, we thereby
rule out {\it any} dynamical transitions. For $d$-dimensional hyper-cubic lattices $s=0$ and $s>d-1$ correspond respectively to {\cal N} and {\cal F} while all the non-trivial cases, $1\le s \le
d-1$, are class {\cal C}, collective freezing. \cite{long}
%The generalization of the previous frameable configurations are those in which hyperedges of dimension $s$ are totally vacant. Again,
%within a framed hyper-cube, any pair can be exchanged.
Analogous expansion arguments yield upper bounds on their crossover lengths:
$\Xi< exp^{\circ s} \left( K_U (s,d)/v^{\frac{1}{d-s}}\right)$ with $exp^{\circ s}$ the
exponential function iterated $s$ times and $K_{U}(s,d)$ a positive
constant given in \cite{long}. Bootstrap percolation yields a lower bound of the same form
but with a different constant $K_{L} (s,d)$ \cite{cerfcirillo}
% in some cases, these can be made equal by finding minimal framing}.
The scale, $\xi$, above which the probability that a nucleus is expandable
saturates, is generally of order $\ln \Xi$, up to log-log factors.
%As an example, for cubic lattices with $s=1$, such framed cubes have a linear
%frame of vacancies along their {\it edges} and $\Xi \sim \exp K/\sqrt{\va}$.
%For the tightly constrained $s=2$ case originally considered
%in \cite{KA}, framed cubes have all their boundary {\it faces}
%vacant. For such a cube to be expandable
%a frameable square is needed adjacent to one of its faces, yielding a double exponential $\Xi \sim \exp \exp K/\va$.\\
We now turn to the dynamics,
focusing again on the $s=1$ square lattice model at high density.
Vacancies will typically be far
apart or in small clusters that cannot move. However vacancies can move within the cores of large
frameable regions. And these cores of size $\xi \sim \ln (1/v)/ v$ are themselves mobile because they are likely to find vacancies on {\it all} the $\xi$ successive line segments needed for them
to move in a given direction.
A generic particle cannot move substantial distances except when a
mobile core passes by; assuming independent core motion a tagged particle will thus diffuse with
$D_S\sim D_Mn_M$, where $n_M\sim 1/\Xi^2$ is the density of the mobile cores
which diffuse with $D_{M}\simeq
\xi^{2}/\tau_\xi$, $1/\tau_\xi$ being the typical relaxation rate of a core. Using our framing analysis and a generalization of the
technique of \cite{Spohn}, we have proved that, indeed, $D_S>0$ for all $\rho$. \cite{long}
The
relaxation time, $\tau_\xi $, of a core is dominated by the time to reach the most severe {\it bottleneck}
in configuration space. Since all configurations are equiprobable
this is proportional to $\exp (\Delta S)$,
the ratio of the number of accessible (frameable)
configurations of the core to the number in the bottleneck \cite{long}.
The worst case scenario in which there is a {\it single} bottleneck configuration corresponds to $\Delta S=S_{\rm total}\simeq \ln
\xi!$, yielding an upper bound $\tau_\xi < \xi!$.
A lower bound is obtained by noting \cite{long} that to equilibrate a frameable square of size $\xi$ one has to pass through
%configurations in which a minimally frameable square of size of order
%(at least) $1/\va \approx \xi/\ln\xi$ has its nucleus in a {\it
%corner} of the square (line segments of lengths less than $1/\va $, such as the edge of such a square, are typically
%completely filled) .
configurations with the nucleus in a {\it
corner} of the square. Using a transfer matrix technique we obtain for
$\ell\times\ell$ minimally frameable squares an entropy difference
$\Delta S\approx \Upsilon\sqrt{\ell} + \alpha \ln \ell + C$ with
$\Upsilon= 2\sqrt{6+\sqrt{22}}-2\sqrt{3}\cong 3.075$ and $\alpha$
computable in principle. \cite{long} Simulations of frameable squares with $\ell=4$ to $\ell=16$ yield equilibration times with $\log \tau_{\ell}$ increasing
slower than $\ell $ and in good agreement
with the square root law with the {\it predicted} coefficient $\Upsilon$
plus logarithmic corrections. We thus conjecture that the entropic bottlenecks for moving the core of a mobile region are equivalent, up to subdominant factors, to
the entropy loss associated with forcing the nucleus of
%a minimal frameable square of size $1/\va$
of such core to a corner. % One can show that creating frameable squares from a nucleus by expansion (as discussed above) typically only entails deviations of intermediate rectangles from square by of order the square root, $r$, of their perimeter. Associated with this, there is an order-unity difference in entropy between the general and corner nucleus cases for each expansion by of order $r$. For $\ell\times\ell$ minimally frameable squares, this yields an entropy difference of $\Delta S\approx \Upsilon\sqrt{\ell} + \alpha \ln \ell + C$ with $\Upsilon= 2\sqrt{6+\sqrt{22}}-2\sqrt{3}\cong 3.075$ and $\alpha$ computable in principle. Running Montecarlo simulations we have measured equilibration times for frameable squares of
%In any case, the corner-entropic barrier provides a lower bound for the relaxation time of a mobile core: $\ln \tau_\xi > U \sqrt{\xi/\ln\xi}$ with $U$ related to $\Upsilon$. The actual behavior may not have the $\ln\xi$ factor.
%With the dynamics of cores dominated by the corner-entropy barriers,
%the diffusion coefficient of the mobile cores, $D_M\sim1/\tau_\xi$,
%will be very small. We conjecture that $D_M$ will control the {\it
%density} diffusion on long time scales. \\
As long as bottlenecks for motion of the mobile cores are not too much larger than the above estimates, $\tau_\xi\ll\Xi^2$ and
the dominant contribution to $D_S\sim D_Mn_M$ will arise from the low density of mobile regions rather than the long time needed for motion within them.
Thus, to leading order in $\va^{-1}$, $\ln D_S\approx 2K_F/\va$
for the square lattice model with $s=1$.
Numerical simulations at high densities
%(from 0.65 to 0.92 and for
%$400$ by $400$ squares)
fit this form well, see fig. 2: in particular
$\lim_{v\rightarrow 0}v\ln D=2K_F\cong
0.9-0.95$ {\it cf} the predicted $2K_F\cong 1.1$.
%(Note that the proportionality between $D_{S}$ and
%the density of holes found in \cite{Dawson} works astonishingly
%well in a large density region but at very high density $D_{S}$
%decreases much more rapidly and is well described by the exponential
%law.)
\begin{figure}[bt]
\centerline{
\includegraphics[width=0.9\columnwidth]{d.eps}}
\caption{$(\ln D)^{-1}$ as a function of $\rho $ for
$400$ by $400$ square lattice with $s=1$.
Straight line shows $\ln D_S\propto 1/v$ behavior.}
\label{fig2}
\end{figure}
For the other KA models, we obtain similar relationship between $D_S$ and $n_M$.
For the $s=2$ cubic case theis yelds $\ln \ln D^{-1}_{S}\propto
1/v$ , a result consistent with a recent finite size scaling analysis \cite{Berthier}.
Thus far we have focused on the high density limit.
Yet simulations on KA and other kinetically constrained models suggest a transition at a non-trivial critical density \cite{KA}. This is a natural consequence of the existence of a dynamical transition on Bethe lattices: in finite dimensions, the transition will be replaced by a crossover which, in some cases, could be quite sharp. In the $s=1$ square lattice case a three--vacancy element can move
along a network of other vacancies provided these are linked no more weakly than via second
neighbors (along axes or diagonals).
At low density, there will always be a percolating vacancy network of such type. As the density increases, diffusion slows down as the vacancy percolation cluster, on which mobile elements rely, sparsens.
Beyond the second-neighbor vacancy percolation transition at $\rho_{2P}$, the {\it simple} mobile elements can no longer move. But by {\it growing}, mobile elements can still form: they have to become large enough to find a percolating cluster on which objects of their size can move. At high densities, such mobile elements will be the cores of (minimal) frameable regions that we have
discussed above.
If the crossover from small mobile objects and behavior similar to that on Bethe lattices is sharp enough, a substantial range of ``critical" behavior of $D_S$ --- and of relaxation times --- might be observed near to an {\it apparent} transition at
$\rho_G\approx\rho_{2P}$.
This reasoning can be generalized to other KA models --- indeed a sharper crossover should occur in higher dimensions --- and could be
an explanation of the apparent dynamical transition found for the
original cubic lattice KA \cite{KA}.
We have shown how simple kinetically constrained models exhibit concretely some of
the qualitative features conjectured to be the
cause of the dramatic slowing down in structural glasses. In
particular, finite-dimensional models can ``almost'' have a
a dynamical phase transition; such exists on Bethe lattices and
percolation arguments suggest it is replaced by a
crossover --- possibly very sharp --- in finite dimensions. The dominant dynamical processes involve cooperative rearrangements of regions whose size diverges in the high density limit; nevertheless, these yield a non-vanishing
self-diffusion coefficient at all densities. But the
length and time scales of these processes grow extremely
rapidly at high densities. This is reminiscent of the super-Arrhenius increase of the relaxation time as fragile liquids are cooled.\\
%%THE FOLLOWING HAS BEEN CUT TO SAVE SPACE
%Several significant issues concerning KA and related kinetically
%constrained models remain open, including the generality of our results. That some similar
%behavior has been found for other
%kinetically constrained models \cite{Jackle,ReviewKLG}, suggests substantial generality. Other questions include:
%How can the sharpness of the crossover at the approximate transition be quantified? How do spatio-temporal correlations behave on time scales larger than the typical relaxation time?
%What, if anything beyond very qualitative aspects, do various
%scenarios
%proposed (for example that of Kurchan et
%al. \cite{Kurchanetal} or the mode coupling theory \cite{Gotze} that
%motivated KA) have to do with the behavior of the KA or related models
% found here?
%More important, of course, is the question of connections to real systems, either structural glasses, or shaken granular media. Can one, for example, construct more physical models in continuous space that exhibit similar behavior? As for real glasses, one would expect that, in the continuum, when one class of processes becomes too sluggish, another, previously negligible, takes over; does the behavior of the KA models suggest mechanisms by which one might understand such crossovers? We leave these, and much else, for further study.
\begin{acknowledgments}
We thank L. Berthier, L. Bertini, J-P. Bouchaud, K. Dawson, J-M. Luck and
M. M{\'e}zard for interesting discussions.
CT thanks COFIN MIUR-2002027798 for financial support.
DSF thanks the National Science Foundation for support via DMR-9976621 and
DMR-0229243.
We thank Les Houches summer school LXXVII where this collaboration
initiated.
\end{acknowledgments}
\begin{thebibliography}{99}
\bibitem{DeBenedettiStillinger}
Recent reviews: P.G. De Benedetti and F.H. Stillinger ,
Nature, {\bf 410}, (2001) 267; C.A. Angell, Science,
{\bf 267}, 1924 (1995); P. G. De Benedetti, {\it Metastable liquids},
Princeton
University Press (1997); G.Tarjus and D. Kivelson, in {\it Jamming and
Rheology}, A.J.Liu and S. R. Nagel eds., Taylor and Francis, N.Y., 2001.
\bibitem{ErdigerWeitz}
M.D. Ediger Annu. Rev. Phys. Chem. {\bf 51} (2000) 99.
E.R. Weeks, J.C. Crocker,A.C. Levitt,A. Schofield and D.A. Weitz
, Science {\bf 287} (2000) 627.
\bibitem{Glotzer}
M.M. Hurley and P. Harrowell, Phys. Rev. E {\bf 52} (1995) 1694.
C. Donati, S. C. Glotzer, P. H. Poole, W. Kob, and S. J. Plimpton,
Phys. Rev. E {\bf 60}, (1999) 3107.
\bibitem{Jackle}
J. Jackle J. Phys. Condens. Matter {\bf 14}, (2002) 1423.
\bibitem{ReviewKLG}
A recent review is F. Ritort, P. Sollich {\it Glassy dynamics of
kinetically constrained models} cond-mat/0210382.
\bibitem{ChandlerGarrahanKLG}
J.P. Garrahan and D. Chandler, cond-mat/0301287.
\bibitem{Sellitto}
M. Sellitto, J-J. Arenzon, Phys. Rev. E {\bf 62}, (2000) 7793.
\bibitem{Jaeger}
H.M. Jaeger, J.B. Knight, R.P. Behringer:
{\em Granular solids, liquids, and gases}
Rev Mod. Phys {\bf 68} (1996) 1259.
\bibitem{KA}
W. Kob and H.C. Andersen, Phys. Rev. E {\bf 48} (1993) 4364.
\bibitem{FranzKA}
S. Franz, R. Mulet and G. Parisi, Phys. Rev. E {\bf 65}, (2002) 021506.
\bibitem{long}
C. Toninelli, G. Biroli and D.S. Fisher in preparation.
\bibitem{FranzDH}
S. Franz and G. Parisi J. Phys.: Condens. Matter {\bf 12} (2000) 6335.
C. Donati, S. Franz, G. Parisi, S.C. Glotzer,
J. Non-Cryst. Solids {\bf 307-310} (2002) 215.
\bibitem{lr-perc} M. Aizenman, J. T. Chayes, L. Chayes, and
C.M. Newman, J. Stat. Phys. {\bf 50}, (1988) 1.
\bibitem{ReviewOE}
J.P. Bouchaud, L.F. Cugliandolo, J. Kurchan and M. M{\'e}zard,
{\it Spin Glasses and Random Fields}, Singapore 1998. World Scientific.
L.F. Cugliandolo, cond-mat/0210312.
%\bibitem{Monroe} J.L. Monroe, J. Stat. Phys. {\bf 65} (1991) 255.
\bibitem{Monroe} J.W.Essam, M.E. Fisher, Rev. Mod. Phys. {\bf 42} (1970) 272
\bibitem{AizenmanLebowitz}
M. Aizenmann and J.L. Lebowitz, J. Phys. A
{\bf 21} (1988) 3801.
\bibitem{Adler}
J. Adler, Physica A {\bf 171}, (1991) 435.
\bibitem{Leath}
J. Chalupa, P.L. Leath and G.R. Reich, J. Phys. C {\bf 12} (1979) L31.
\bibitem{new-bootstrap}
A.E. Holroyd, Probab. Theory Rel {\bf 125} (2003) 195.
\bibitem{footnoteergodicity}
For the KA model one can also prove that irreducibility implies no ergodicity breaking in the thermodynamic limit (see \cite{long}).
\bibitem{cerfcirillo}
R.Cerf, E.Cirillo, Annals. of Probab. {\bf{27 n. 4}} (1999) 1837.
R. Cerf, F. Manzo, Stoch. Proc. Appl. {\bf 101} (2002) 69.
\bibitem{Spohn}
H. Spohn, J. Stat. Phys. {\bf 59}, (1990) 1227; Physica A {\bf 163},
(1990) 134.
\bibitem{Berthier}
L. Berthier, cond-mat/0303452.
\end{thebibliography}
\end{document}