1: \documentclass[aps,superscriptaddress,amsmath,twocolumn,amssymb,showpacs,
2: floatfix]{revtex4} \usepackage{epsfig}
3:
4: \begin{document}
5:
6: \title{Heterogeneity and growing lengthscales in the dynamics of
7: kinetically constrained lattice gases in two dimensions}
8:
9: \author{Albert C. Pan}
10:
11: \affiliation{Department of Chemistry, University of California,
12: Berkeley, CA 94720-1460}
13:
14: \author{Juan P. Garrahan}
15:
16: \affiliation{School of Physics and Astronomy, University of
17: Nottingham, Nottingham, NG7 2RD, UK}
18:
19: \author{David Chandler}
20:
21: \affiliation{Department of Chemistry, University of California,
22: Berkeley, CA 94720-1460}
23:
24:
25: \date{\today}
26:
27: \begin{abstract}
28: We study dynamical heterogeneity and growing dynamical lengthscales in
29: two kinetically constrained models, namely, the one- and two-vacancy
30: assisted triangular lattice gases. One of the models is a strong
31: glassformer and the other is a fragile glassformer. Both exhibit
32: heterogeneous dynamics with broadly distributed timescales as seen in
33: the distribution of persistence times. We show that the
34: Stokes-Einstein relation is violated, to a greater degree in the
35: fragile glassformer, and show how this violation is related to dynamic
36: heterogeneity. We extract dynamical lengthscales from structure
37: factors of mobile particles and show, quantitatively, the growth of
38: this lengthscale as density increases. We comment on how the scaling
39: of lengths and times in these models relates to that in facilitated
40: spin models of glasses.
41: \end{abstract}
42:
43: \pacs{05.20.Jj, 05.70.Jk, 64.70.Pf}
44:
45: \maketitle
46:
47: \twocolumngrid
48:
49: \section{Introduction}
50:
51: The dramatic dynamical slowdown accompanying the formation of a glass
52: is a remarkable phenomenon
53: \cite{Ediger-et-al,Angell,Debenedetti-Stillinger}. One explanation
54: for the underlying microscopic cause of this slowdown relies on the
55: presence of local steric constraints on the movement of particles
56: which make themselves felt to an increasing degree as the temperature
57: is lowered (or the concentration of particles is increased).
58: Kinetically constrained lattice gas models
59: \cite{Kob-Andersen,Jackle,Ritort-Sollich} are simple caricatures of
60: glassformers which employ local steric constraints as their sole means
61: to glassiness in the {\em absence} of any non-trivial static
62: correlations between particles (for alternative thermodynamic views of
63: the glass transition see e.g. \cite{Tarjus-Kivelson,Xia-Wolynes}).
64: These constrained models have been extensively studied (see e.g.
65: \cite{Kob-Andersen,Jackle,Sellito-et-al,Toninelli-et-al,Marinari-Pitard}).
66: Our purpose here is to extend these studies to focus on the idea of
67: dynamical heterogeneity \cite{Ediger,Glotzer,Harrowell-et-al} as a
68: manifestation of excitation lines in space-time and to attempt
69: identification of scaling and universality classes in the dynamics of these
70: models
71: \cite{Garrahan-Chandler,Berthier-Garrahan,Jung-et-al,Whitelam-et-al,Berthier-et-al}.
72:
73: The paper is organized as follows. Section \ref{Models} describes the
74: two models we use as well as details of the computer simulations used
75: to study them. Section \ref{dpt} looks at heterogeneous dynamics in
76: our models via the distribution of persistence times. Section
77: \ref{slowdown} presents the scaling of the structural relaxation time
78: and diffusion constant, the implications of which lead to a discussion
79: of the breakdown of the Stokes-Einstein relation in section
80: \ref{SEBreakdown}. Section \ref{dynlength} discusses the emergence of
81: a dynamical lengthscale and gives a quantitative characterization of
82: this length by analyzing structure factors of mobile particles.
83: Finally, we end with a discussion of our results in Section
84: \ref{Discussion}.
85:
86: \section{Models and Computational Details}
87: \label{Models}
88:
89: We present results for two kinetically constrained triangular lattice
90: gas (TLG) models introduced by J\"ackle and Kr\"onig \cite{Jackle}.
91: These two-dimensional models are variants of lattice models proposed
92: by Kob and Andersen \cite{Kob-Andersen}. Each site of the triangular
93: lattice has six nearest neighbor sites and can hold at most one
94: particle. A particle at site ${\bf i}$ is allowed to move to a
95: nearest neighbor site, ${\bf i}'$, if (i) ${\bf i}'$ is not occupied
96: and (ii) the two mutual nearest neighbor sites of ${\bf i}$ and ${\bf
97: i}'$ are also empty. These rules coincide with a physical
98: interpretation of steric constraints on the movement of hard core
99: particles in a dense fluid \cite{Jackle}. We call the model with
100: these rules the (2)-TLG because both mutual nearest neighbors need to
101: be empty in order to facilitate movement. We also present results for
102: the (1)-TLG where the constraints are more relaxed: movement is
103: allowed as long as either of the mutual nearest neighbors is empty.
104: As with other kinetically constrained lattice gas models, the TLG has
105: no static interactions between particles other than those that
106: prohibit multiple occupancy of a single lattice site. Therefore,
107: initial configurations can be generated by random occupation of empty
108: lattice sites by particles until the desired density is reached.
109:
110: In the computer simulations, we investigated particle densities,
111: $\rho$, between 0.01 and 0.80 for the (2)-TLG and between 0.01 and
112: 0.996 for the (1)-TLG. The density $\rho$ = 1 corresponds to the
113: completely full lattice in both cases. For the (2)-TLG, we used a
114: lattice with edge length $L$ = 128 for all densities. There exists
115: the possibility in the (2)-TLG of initial configurations containing an
116: unmoveable structure which percolates throughout the system called a
117: backbone \cite{Kob-Andersen, Jackle}. Since the dynamics obey
118: detailed balance, these backbones could never be destroyed in the
119: course of the simulation. For the densities studied here, however,
120: $L$ = 128 is sufficiently large such that the probability of having
121: such a configuration is vanishingly small (see \cite{Jackle}). For
122: the (1)-TLG, we used $L$ = 128 to $L$ = 2048. A pair of vacancies in
123: the (1)-TLG can always diffuse freely \cite{Jackle} so this version of
124: the TLG does not suffer from the problem of backbones in the same way
125: as the (2)-TLG. At higher densities, however, one still needs to
126: ensure that there are a sufficient number of potentially mobile
127: particles in the system such that the typical dynamics of the model
128: are observed. For the (1)-TLG, the number of potentially mobile
129: particles at higher densities is approximately equivalent to the
130: number of vacancy pairs and therefore can be estimated as
131: $(1-\rho)^2L^2$. The system sizes at the various densities for the
132: (1)-TLG simulations were chosen such that the number of potentially
133: mobile particles estimated in this way was always approximately 100.
134: For both models, periodic boundary conditions were used.
135:
136: \begin{figure}[t]
137: \epsfig{file=fig1.eps, width = 2.5in, clip}
138: \caption{\label{Dta} Distribution of local persistence times for (a)
139: the (2)-TLG model (from left to right) $\rho$ = 0.40, 0.50, 0.65,
140: 0.70, 0.77 and 0.79; and (b) the (1)-TLG model (from left to right)
141: $\rho$ = 0.20, 0.60, 0.70, 0.75, 0.80, 0.85, 0.92, 0.95 and 0.97. The
142: abscissae are given on a scale of logarithm base 10.}
143: \end{figure}
144:
145: At each density, several hundred independent trajectories of lengths
146: 10-100 times $\tau_{\alpha}$, where $\tau_{\alpha}$ is the time for
147: the self-intermediate scattering function at ${\bf q} = (\pi, 0)$
148: \cite{qnote} to reach $1/e$ of its initial value (see below), were
149: run. Trajectories were stored logarithmically for later analysis
150: (i.e. configurations were saved after 1, 2, 4, 8, 16, 32,
151: etc. sweeps). At each state point, between 128 and 256 independent
152: trajectories were acquired. Time was measured in Monte Carlo sweeps.
153: During each sweep, particles were chosen randomly and a move was
154: attempted. For the higher density runs in both models, a continuous
155: time algorithm was used for greater efficiency \cite{Newman-Barkema}.
156: This algorithm involved making and updating a list of only those
157: particles which have the possibility of moving and choosing from among
158: those exclusively during every move. The total time (in units of
159: Monte Carlo sweeps) was then updated accordingly by adding to it the
160: inverse of the number of mobile particles available during that
161: continuous time step. The continuous time algorithm resulted in a
162: speed up of our simulations by as much as 1-2 orders of magnitude for
163: the highest density runs in both the (1)-TLG and the (2)-TLG.
164: Finally, for the distribution of site persistence times (see below),
165: statistics were gathered over runs of very large systems ($L = 1024$
166: and $2048$).
167:
168: \begin{figure}[b]
169: \epsfig{file=fig2.eps, width=2.5in, clip}
170: \caption{\label{fskpi} Self-intermediate scattering function at
171: wavevector ${\bf q} = (\pi, 0)$ for (a) the (2)-TLG (from left to right)
172: $\rho$ = 0.01, 0.05, 0.10, 0.20, 0.30, 0.40, 0.50, 0.60, 0.65, 0.70,
173: 0.75, 0.77, 0.79, 0.80; and (b) the (1)-TLG (from left to right)
174: $\rho$ = 0.20, 0.30, 0.40, 0.50, 0.60, 0.70, 0.75, 0.80, 0.85, 0.88,
175: 0.90, 0.92, 0.95, 0.97, 0.98, 0.99, 0.993, 0.995, 0.996. The unit of
176: length is the lattice spacing. }
177: \end{figure}
178:
179:
180: \section{Heterogeneous dynamics and the Distribution of persistence times}
181: \label{dpt}
182:
183: \begin{figure*}[t]
184: \epsfig{file=fig3.eps, width=6in, clip}
185: \caption{\label{ang} (a) Structural relaxation time as a function of
186: temperature. Circles and squares correspond to the (2)-TLG and
187: (1)-TLG, respectively, throughout the paper. The inset shows the same
188: data for the (2)-TLG plotted versus $T - T_c$ and a power law fit
189: (solid line) to a portion of the data. (b) Structural relaxation as a
190: function of scaled reciprocal temperature. Here $T_g$ is such that
191: $\tau_\alpha(T_g) = 10^7$. The lines are fits to the relaxation times
192: at low temperatures. For the (1)-TLG we use the Arrhenius form
193: $\ln{\tau_\alpha} \propto \Delta / T$, with $\Delta \approx 2.29$.
194: For the (2)-TLG we use a double exponential \cite{Toninelli-et-al},
195: $\ln{\tau_\alpha} \propto \exp{(a / T)}$, with $a \approx 1.76$. }
196: \end{figure*}
197:
198: \begin{figure}[t]
199: \epsfig{file=fig4.eps, width=2.5in, clip}
200: \caption{\label{Dmsd} The mean-squared displacement: (a) the
201: (2)-TLG, $\rho = 0.01$ to $0.80$; (b) the (1)-TLG, $\rho = 0.01$ to
202: $0.996$.}
203: \end{figure}
204:
205: A central phenomenon behind our perspective of glasses is dynamical
206: heterogeneity \cite{Ediger,Glotzer}. A direct measure of
207: heterogeneous dynamics in glassy systems is the idea of persistence
208: times \cite{Berthier-Garrahan, Jung-et-al}. Fig.\ \ref{Dta}(a) and
209: Fig.\ \ref{Dta}(b) show the distribution of site persistence times,
210: $\pi(\log \tau)$, in the (2)-TLG and (1)-TLG at various densities;
211: that is, the distribution of times, given an initial configuration of
212: the lattice, before the first change at a particular site occurs,
213: either due to an empty site being filled or a filled site becoming
214: empty. These distributions are multi-point functions because they
215: depend not only on the state of a lattice site at the initial time and
216: the time when it changes, but also on all intervening points in time.
217: As has been shown in spin-facilitated models \cite{Berthier-Garrahan},
218: three distinct dynamical regimes are observed: (i) at low densities,
219: there is a single peak at small relaxation times indicating
220: homogeneous fast dynamics; (ii) at intermediate densities, two peaks
221: develop and co-exist, one at faster times and the other at slower
222: times, indicating heterogeneous, fluctuation dominated, dynamics;
223: (iii) as density is increased even higher, the peak at faster times
224: becomes suppressed relative to the peak at slower times and the
225: dynamics again become homogeneous and slow. In region (ii), or the
226: crossover region, the dynamics are broadly distributed over several
227: orders of magnitude in time. Following \cite{Berthier-Garrahan},
228: we can define, qualitatively, an onset
229: density, $\rho_o$, for both models where the dynamics begin to feel
230: the influence of dynamical heterogeneity and thereby lose their
231: mean-field character, as well as a crossover density $\rho_c$ where
232: slow processes begin to dominate. For the (2)-TLG, $\rho_o = 0.50$
233: and (we anticipate) $\rho_c$ = 0.80 and for the (1)-TLG, $\rho_o =
234: 0.60$ and $\rho_c$ = 0.85.
235:
236: In the next section, we turn to two two-point functions which have
237: been the more conventional measures of glassy dynamics: the
238: self-intermediate scattering function and the mean-squared
239: displacement.
240:
241:
242: \section{Dynamical slowdown}
243: \label{slowdown}
244:
245: \begin{figure}
246: \epsfig{file=fig5.eps, width=2.5in, clip}
247: \caption{\label{Dmsd-scal} Self-diffusion constant as a function of
248: inverse temperature.}
249: \end{figure}
250:
251: A basic ingredient of a glassformer is a precipitous dynamical
252: slowdown over a narrow range of temperatures or densities. This
253: characteristic can already be seen qualitatively in the distribution
254: of persistence times by looking at the movement of the mean of the
255: distributions as $\rho$ increases. For example, in the (2)-TLG, from
256: $\rho$ = 0.70 to $\rho$ = 0.77, the mean persistence time increase 2-3
257: orders of magnitude. The traditional measure of this slowdown is the
258: self-intermediate scattering function, $F_s(q, t)$ = $\langle e^{i{\bf
259: q}\cdot({\bf r}_i(t) - {\bf r}_i(0))}\rangle$, particularly its decay
260: at wavevector ${\bf q} = (\pi, 0)$, Fig.\ \ref{fskpi}(a) and Fig.\
261: \ref{fskpi}(b). Here, ${\bf r}_i(t)$ denotes the position of particle
262: $i$ at time $t$. The angled brackets, $\langle\cdots\rangle$, denote
263: an average over different pairs of configurations along a trajectory
264: separated by a given time interval. The decay of the scattering
265: function to $1/e$ at this wavevector is typically defined to be
266: $\tau_{\alpha}$, or the structural relaxation time, as it gives a
267: sense of how density fluctuations relax at relatively short
268: lengthscales.
269:
270: Experiments report the scaling of viscosity versus inverse temperature
271: \cite{Angell}. Therefore, for the particular case of a kinetic
272: lattice gas, one would like to make connections between structural
273: relaxation time and viscosity, and density and inverse temperature.
274: In both experiments and computer simulations
275: \cite{RichterFrickFarago,OnukiPRE}, it has been shown that the
276: structural relaxation time scales like the viscosity. If one imagines
277: that the concentration of vacancies, $c$, in the TLG models are like
278: excitations or fluctuating regions of high energy, then it is
279: reasonable to define a reduced temperature, $T$, such that $c$
280: $\equiv$ $(1-\rho)$ $\equiv$ $e^{-1/T}$ or:
281: \begin{equation}
282: -\ln(1-\rho)\equiv 1/T.
283: \end{equation}
284: The structural relaxation time plotted versus $T$ and $1/T$ is
285: shown in Fig.\ \ref{ang}(a) and Fig.\ \ref{ang}(b), respectively.
286: The inset of Fig.\ \ref{ang}(a) shows a plot of $\tau_{\alpha}$
287: versus $T-T_c$ for the (2)-TLG where $T_c$ is taken to be the
288: temperature at the crossover density, $\rho_c$ (as defined in section
289: \ref{dpt}), and the black line is a power law fit to a portion of the
290: data in the manner of mode coupling theory (MCT) \cite{mct}. In MCT,
291: $T_c$ is a critical temperature where time scales diverge. We see
292: that the MCT fit is valid for almost five orders of magnitude in time
293: even though there is no dynamical arrest at $T_c$, as the relaxation
294: time of the (2)-TLG diverges only at $\rho=1$ \cite{Toninelli-et-al}.
295: A similar MCT power law fit (not shown) can be made for the (1)-TLG,
296: valid for about 2 orders of magnitude.
297:
298: Fig.\ \ref{ang}(b) shows the relaxation time on a reduced temperature
299: scale in the manner proposed by Angell \cite{Angell}. Here, $T_g$ is
300: defined as the temperature at which $\tau_{\alpha} = 10^7$. We see
301: that the relaxation time of the (1)-TLG is Arrhenius growing as
302: $\tau_{\alpha} \sim c^{-\Delta}$ with $\Delta = 2.29$, as $\rho
303: \rightarrow 1$. The relaxation of the (2)-TLG in contrast is
304: super-Arrhenius, and the low temperature data can be fit with a double
305: exponential in $1/T$ \cite{Toninelli-et-al}. In this context, the
306: (1)-TLG is a strong glassformer whereas the (2)-TLG is a fragile one.
307:
308: \begin{figure}
309: \epsfig{file=fig6.eps, width=2.5in, clip}
310: \caption{\label{SE} Stokes-Einstein violation in the (a) (2)-TLG
311: and (b) (1)-TLG. }
312: \end{figure}
313:
314: \begin{figure}
315: \epsfig{file=fig7.eps, width=2.6in, clip}
316: \caption{\label{fSE} Fractional Stokes-Einstein exponent: scaling of
317: the self-diffusion constant with relaxation time. The dashed lines
318: are fits to the data at longer times.}
319: \end{figure}
320:
321: Interestingly, the time exponent $\Delta \approx 2.3$ of the (1)-TLG
322: is the same as that for the one-spin facilitated Fredrickson-Andersen
323: (FA) model in dimension $d=2$, obtained in renormalization group (RG)
324: calculations and observed numerically \cite{Whitelam-et-al}. This
325: suggests that the (1)-TLG may be in the universality class of the FA
326: model, the prototypical kinetically constrained spin model for a
327: strong glassformer \cite{Trivial}. We will examine more of these
328: dynamic scaling relations below.
329:
330: \begin{figure*}
331: \epsfig{file=fig8.eps, width = 5in, clip}
332: \caption{\label{diffq} (Top) Self-intermediate scattering function for
333: the (2)-TLG at various wavevectors: (a) $\rho = 0.20$, (b) $\rho =
334: 0.77$. For both graphs, from left to right: $q$ = $(\pi, 0)$,
335: $(\pi/2,0)$, $(\pi/4, 0)$, $(\pi/8, 0)$, $(\pi/16, 0)$, $(\pi/32, 0)$,
336: $(\pi/64, 0)$. (Bottom) $D_s(q) \equiv 1/\tau(q)q^2$ as a function of
337: $q$ at various densities: (c) (2)-TLG, (from top to bottom) $\rho$ =
338: 0.20 to 0.80; (d) (1)-TLG (from top to bottom), $\rho$ = 0.40 to
339: 0.996. The higher density curves for the (1)-TLG include smaller $q$
340: values because of larger system sizes (see section II).}
341: \end{figure*}
342:
343: The mean-squared displacement, $\langle|\Delta {\bf r}_i(t)|^2
344: \rangle$ = $\langle|{\bf r}_i(t) - {\bf r}_i(0)|^2\rangle$, is shown
345: in Fig.\ \ref{Dmsd}. The self-diffusion coefficient, $D_s$, is
346: defined as $D_s = \lim_{t\rightarrow\infty}\langle|\Delta {\bf
347: r}_i(t)|^2\rangle/4t$. We see that at low
348: densities, $D_s$ for the (1)-TLG and (2)-TLG coincide.
349: At higher densities, $D_s$ for the (1)-TLG is
350: Arrhenius, and scales as $D_s \sim c^2$ as $\rho \rightarrow 1$, see
351: Fig.\ \ref{Dmsd-scal}. This result is in agreement with the
352: analytical prediction of \cite{Toninelli-et-al}. Moreover, this
353: is also the scaling of the diffusion constant for a probe molecule
354: coupled to the FA model in any dimension \cite{Jung-et-al}, further
355: evidence that the (1)-TLG is in the FA model universality class. On
356: the other hand, $D_s$ for the (2)-TLG is super-Arrhenius (see Fig.\
357: \ref{Dmsd-scal}). The behavior of $D_s$ is similar, qualitatively, to
358: that of $\tau_{\alpha}$. The quantitative difference in their scaling
359: with density, however, is significant, and is an indication that
360: relaxation behaviors at short and long lengthscales are not the same.
361: We turn to this issue now in both the (1)-TLG and the (2)-TLG.
362:
363: \begin{figure}[b]
364: \epsfig{file=fig9.eps, width=2.5in, clip}
365: \caption{\label{scaling} Data from Fig.\ \ref{diffq} collapsed onto a
366: master curve. The filled symbols correspond to the (2)-TLG and the
367: open symbols correspond to the (1)-TLG. The straight line is $q^2$.}
368: \end{figure}
369:
370: \section{The breakdown of the Stokes-Einstein Relation}
371: \label{SEBreakdown}
372:
373: An important ramification of broadly distributed heterogeneous
374: dynamics is the breakdown of mean-field dynamical relations such as
375: the much studied Stokes-Einstein (SE) relation. This relation says
376: that diffusion scales inversely with the relaxation time,
377: $D_s\tau_{\alpha} \sim \mbox{constant}$. It is a quantitative
378: statement of the expectation that the dynamical behavior of normal
379: liquids should be similar at all but the smallest lengthscales. In supercooled liquids
380: this simple mean-field approximation fails
381: \cite{Chang-Sillescu,Swallen-et-al}, and, given the results of
382: previous sections, we would expect a similar violation of the SE
383: relation in the TLG models. We see from Fig.\ \ref{SE} that the SE
384: relation is indeed violated for both the (2)-TLG and the (1)-TLG, the
385: effect being more pronounced in the fragile case. Moreover, we see
386: that the densities at which the product $D_s\tau_{\alpha}$ begins
387: deviating from constancy coincide with the onset densities,
388: $\rho_o$'s, extracted from the distribution of persistence times.
389: This observation reinforces the idea that it is the fluctuation
390: dominated nature of the dynamics that leads to the SE breakdown
391: \cite{Jung-et-al}.
392:
393: SE violation implies that the self-diffusion constant does not scale
394: with the structural relaxation time as $\tau_{\alpha}^{-1}$. One
395: possibility is that it obeys a fractional SE law, $D_s \sim
396: \tau_{\alpha}^{-\xi}$ where $\xi < 1$. This is observed in
397: experiments \cite{Swallen-et-al}, and is obtained theoretically for
398: probe diffusion in the FA and East models
399: \cite{Jung-et-al} (see also \cite{Schweizer-Saltzman}). Figure
400: \ref{fSE} shows that the diffusion constant also
401: obeys a fractional SE law in the TLG models. The SE exponent is $\xi \approx 0.88$ for
402: the (1)-TLG, which is the value expected for the FA model in $d=2$,
403: $\xi \approx 2/2.3$ \cite{Jung-et-al}. In the case of the (2)-TLG,
404: despite the fact that $D_s$ and $\tau_{\alpha}$ are both
405: super-Arrhenius, we find that the scaling exponent is temperature
406: independent at large densities, $\xi \approx 0.58$. The deviation of
407: this exponent from 1 is
408: larger than that for both the FA and East models in two dimensions
409: \cite{Jung-et-al}. It indicates a larger violation of the SE law,
410: consistent with the fact that the (2)-TLG is more fragile than either
411: of those models \cite{Toninelli-et-al}.
412:
413:
414: \section{Dynamical Lengthscales}
415: \label{dynlength}
416:
417: \subsection{Indication of a dynamical lengthscale from a two-point function}
418:
419: \begin{figure}[t]
420: \resizebox{\columnwidth}{!}{
421: \includegraphics{fig10.eps}}
422: \caption{Growth of mobile particle regions as a function of
423: observation time $\Delta t$ at $\rho = 0.77$ in the (2)-TLG. Black
424: and grey regions indicate the location of particles and white
425: regions indicate empty lattice sites. Particles colored
426: in black have moved at least one lattice spacing in a time $\Delta
427: t$ whereas particles colored in gray have not. (Top, from left to right:
428: $\Delta t = 10^3$ and $10^4$; bottom, from left to right: $\Delta t = 10^5$
429: and $10^6$; $\tau_{\alpha} \sim 10^5$ at this density).}
430: \label{2tlg}
431: \end{figure}
432:
433: Since the growth in timescales and the violation of the
434: Stokes-Einstein relation in the TLG models are clearly not tied to a growth in static
435: lengthscales, we turn now to the discussion of dynamical lengthscales.
436: Such a lengthscale can be inferred from examining the relaxation
437: behavior of the self-intermediate scattering function over more than
438: one wavevector at different densities. One can appreciate this fact
439: qualitatively by looking at the decay of $F_s(q, t)$ for the (2)-TLG
440: over several values of $q$ at low and high density, Fig.\ \ref{diffq}
441: (top), \cite{Berthier-et-al}. At low density, the decay of the
442: various curves looks similar at all wavevectors (except for the
443: largest wavevector) whereas at high density, even the curves at
444: intermediate wavevector differ greatly from the simple exponential
445: form seen at smaller wavevectors. The high density curves bunch up at
446: intermediate to large ${\bf q}$ indicating that the relaxation
447: behavior at these lengthscales is different (i.e. slower) than one
448: would expect from the behavior at larger lengthscales. Similar
449: behavior has also been observed in the Kob-Andersen kinetic lattice
450: gas model and kinetically constrained spin models
451: \cite{Kob-Andersen,Berthier-et-al}.
452:
453: \begin{figure}[t]
454: \resizebox{\columnwidth}{!}{
455: \includegraphics{fig11.eps}}
456: \caption{Same as Fig.\ \ref{2tlg} for the (1)-TLG. Here, $\rho$ =
457: 0.95. (Top, from left to right: $\Delta t = 10$ and $10^2$; bottom,
458: from left to right: $\Delta t = 10^3$ and $10^4$; $\tau_{\alpha} \sim
459: 10^3$ at this density).}
460: \label{1tlg}
461: \end{figure}
462:
463: To quantify the above behavior, we proceed as in
464: \cite{Berthier,Ediger-et-al-2}. In the hydrodynamic regime, we have
465: $\lim_{q \rightarrow 0} F_s(q, t) \sim \exp(-D_s q^2 t)$, and one
466: expects the product $D_s \tau(q) q^2$ to be independent of ${\bf q}$, where
467: $\tau(q)$ is the time when the intermediate scattering
468: function at wavevector $\bf{q}$ decays to $1/e$. In Fig.\ \ref{diffq} (bottom), we
469: plot the quantity $D_s(q) \equiv 1/\tau(q) q^2$ as a function of $q$ at
470: various densities. A flat line independent of $q$ indicates normal
471: diffusive behavior whereas a downward bend signifies a change to
472: sub-diffusive behavior. As density increases, the curves begin to
473: bend at a smaller and smaller wavevector, $q^{*}$. This behavior is
474: indicative of a growing dynamical lengthscale as density is increased
475: \cite{Berthier,Ediger-et-al-2,Berthier-et-al}.
476:
477: Following the prescription of \cite{Berthier-et-al}, we extract a
478: lengthscale, $\ell^*$, from Fig.\ \ref{diffq} as $\ell^* \sim
479: \sqrt{D_s(q \rightarrow 0) \tau_{\alpha}}$. This lengthscale determines
480: the onset of Fickian diffusion. Using $\ell^*$ to rescale
481: space, and using $\tau(q)q^2$ to rescale time, the data from Fig.\
482: \ref{diffq} can be collapsed onto a master curve, Fig.\ \ref{scaling}.
483:
484: \subsection{Direct observation and quantification of a dynamical
485: heterogeneity lengthscale}
486:
487: \begin{figure*}[t]
488: \epsfig{file = fig12.eps, width=5in, clip}
489: \caption{\label{xi} (Top) Growth of dynamical heterogeneity length as
490: a function of observation time, $\Delta t$, for (a) the (2)-TLG,
491: $\rho$ = 0.70 to 0.80; and (b) the (1)-TLG, $\rho$ = 0.95 to 0.996.
492: (Bottom) (c) Scaling of the maximum value of the dynamical heterogeneity
493: length, $\ell_{max}$, from (a) and (b) with relaxation time.
494: The inset shows $\ell_{max}$ versus $c$ for the (1)-TLG, on a
495: log-log scale to highlight power law scaling.}
496: \end{figure*}
497:
498: We can study dynamical lengthscales in the TLG models directly by
499: observing a trajectory over a time $\Delta t$ and coloring particles
500: which have moved at least one lattice spacing. Snapshots of applying
501: this procedure to trajectories of the (2)-TLG and the (1)-TLG at high
502: particle densities over progressively longer $\Delta t$'s are shown in
503: Fig.\ \ref{2tlg} and Fig.\ \ref{1tlg}. Mobility is indeed correlated:
504: mobile particles are clustered and the clusters of mobility at earlier
505: $\Delta t$ act as seed particles from which subsequent mobility grows.
506: Moreover, there is a qualitative difference in the shape of the
507: clusters in the (2)-TLG and the (1)-TLG. In the fragile model, the
508: clusters are smooth and more anisotropic, indicating a directed growth
509: of mobile regions. In the strong model, the clusters are rugged and
510: isotropic. These observations can be understood as arising from the
511: difference in the local constraints of both models. The strict two
512: site facilitation rule of the (2)-TLG requires cooperative,
513: hierarchical rearrangement of particles for movement whereas the one
514: site facilitation rule of the (1)-TLG allows for the random diffusion
515: of vacancy pairs \cite{Jackle}. These same correlations between
516: fragility and the smoothness of interfaces and between slow and fast
517: dynamically heterogeneous regions are present in other facilitated
518: models \cite{Garrahan-Chandler,Whitelam-et-al,NEF}.
519:
520: To quantify the above ideas, we extract a dynamical lengthscale,
521: $\ell(\Delta t)$, from structure factors of the mobile particles
522: \cite{Glotzer,Garrahan-Chandler,Lacevic-et-al,Berthier}. Motivated by
523: the mobility criterion described in the previous paragraph and
524: depicted in Fig.\ \ref{2tlg} and Fig.\ \ref{1tlg}, we consider the binary field $n_{\bf
525: r}(t; \Delta t) = p_{\bf r}(t)[1-p_{\bf r}(t+\Delta t)]$, which gives
526: a signal only when there is particle motion at ${\bf r}$ over the
527: range of time between $t$ and $t+\Delta t$. Its structure factor is a
528: four point function: it measures a correlation function which depends
529: on two points in time, $t$ and $t + \Delta t$, and two points in
530: space, ${\bf r}$ and ${\bf r'}$.
531:
532: We define the structure factor for the mobility field as the following
533: normalized correlation function:
534: \begin{equation}
535: \label{skeq}
536: S({\bf q}; \Delta t) = \frac{1}{L^2}\frac{\langle \delta n_{\bf q}(t;
537: \Delta t) \delta n_{-\bf q}(t; \Delta t)\rangle}{\langle\delta
538: n_{\bf r}(t; \Delta t)^2\rangle}
539: \end{equation}
540: where $\delta n_{\bf r}(t; \Delta t)$ = $n_{\bf r}(t; \Delta t) -
541: \langle n_{\bf r}(t; \Delta t)\rangle$ is the deviation of $n_{\bf r}$
542: from its average value and $\delta n_{\bf q}(t; \Delta t)$ is the
543: Fourier transform of $\delta n_{\bf r}(t; \Delta t)$:
544: \begin{equation}
545: \delta n_{\bf q}(t; \Delta t) = \sum_{\bf r} \exp\left(\frac{2\pi
546: i}{L^2}{\bf r} \cdot {\bf q}\right)\delta n_{\bf r}(t; \Delta t).
547: \end{equation}
548: The angled brackets, $\langle\cdots\rangle$, denote an average over
549: different pairs of configurations along a trajectory separated by a
550: given time interval $\Delta t$. We then define the lengthscale,
551: $\ell(\Delta t)$, to be proportional to the inverse of the first
552: moment, $\bar{q}_{\Delta t}$, of the circularly averaged structure
553: factor, $\tilde{S}(q_n; \Delta t)$ \cite{Amar-et-al}. That is,
554: $\ell(\Delta t) \sim 1/ \bar{q}_{\Delta t} $ where:
555: \begin{equation}
556: \bar{q}_{\Delta t} = \sum_{n}q_n\tilde{S}(q_n; \Delta
557: t)/\sum_{n}\tilde{S}(q_n; \Delta t).
558: \end{equation}
559: Here, $q_n$ = $2\pi n/L$ and $n$ = 0, 1, 2, ..., $L/2$. The
560: lengthscale was normalized such that $\ell$ extracted from the
561: structure factor of a random configuration of particles on the lattice
562: (i.e. an ideal gas) was unity.
563:
564: Fig.\ \ref{xi}(a) and Fig.\ \ref{xi}(b) show $\ell(\Delta t)$ at
565: various densities for the (2)-TLG and the (1)-TLG. The basic shape of
566: these curves is as expected: as $\Delta t \rightarrow 0$, mobility is
567: sparse and uncorrelated so $\ell$ approaches unity and as $\Delta t
568: \rightarrow \infty$, everything becomes mobile and $\ell$ once again
569: tends towards unity. In between, as the pictures in Fig.\ \ref{2tlg}
570: and Fig.\ \ref{1tlg} suggest, mobility clusters together and grows.
571: Looking at the maximum of these different curves, $\ell_{max}(\Delta
572: t_{max})$, a growing lengthscale is clearly evident as $\rho$
573: increases. It is important to note that, in general, $\ell_{max} \neq
574: \ell^*$ \cite{Berthier-et-al}.
575:
576: Fig.\ \ref{xi}(c) shows the value of $\ell_{max}$ plotted versus the
577: structural relaxation time for both the (2)-TLG and the (1)-TLG. At
578: short relaxation times,
579: the curves merge and approach the ideal gas value of one.
580: As relaxation times increases, the dynamical heterogeneity
581: lengthscale for the strong version of the model is always larger than that of
582: the fragile version at a
583: fixed value of $\tau_{\alpha}$ \cite{Garrahan-Chandler}. We also find that the
584: observation time, $\Delta t_{max}$, at which the maximum lengthscale,
585: $\ell_{max}$, occurs, scales with the structural relaxation time,
586: $\tau_{\alpha}$, for both models (not shown).
587:
588: If the (1)-TLG is in the universality class of the FA model, then we
589: would expect $\ell_{max}$ to scale as a power of both the excitation
590: concentration, $\ell_{max} \sim c^{-\nu}$, and of the relaxation time,
591: $\ell_{max} \sim \tau_{\alpha}^{1/z}$. This appears to be the case,
592: as shown in Fig.\ \ref{xi}(c). For the correlation and dynamic
593: exponents we find $\nu \approx 0.89$ and $1/z \approx 0.36$, in
594: reasonable agreement with Ref.\ \cite{Whitelam-et-al}, $\nu \approx
595: 0.7$ and $1/z = \nu/\Delta \approx 0.3$ \cite{Trivial2}. The $z$ exponents shown
596: in Fig.\ \ref{xi}(c) are what we would expect from Fig.\ \ref{xi}(b)
597: where a range of $\ell(\Delta t)$ curves at different densities
598: merge at early times and display power law scaling with
599: similar exponents.
600:
601: As alluded to earlier, the dynamics of the (1)-TLG at high densities
602: is controlled by the motion of vacancy pairs. The physics of these
603: vacancy pairs is similar to excitations in the FA model. Vacancy
604: pairs have the ability to interact with other lattice vacancies in
605: order to branch and coalesce. It is important to note that evidence
606: of these interactions can only be seen in simulations of large enough
607: system sizes where the number of vacancy pairs is approximately
608: 50-100. As mentioned in section \ref{Models}, this requirement leads
609: to system sizes, for example, of $L=2048$ for $\rho = 0.995$.
610:
611: Fig.\ \ref{skscaled} shows the structure factors of the mobility
612: field, equation (\ref{skeq}), measured at the structural relaxation
613: time, $\Delta t = \tau_{\alpha}$ for the (2)-TLG and the (1)-TLG.
614: The curves are scaled in a manner suggestive of a coarsening
615: process. The collapse of the various structure factors
616: \cite{largek} implies that, in the glassy regime, increasing density
617: corresponds to a coarsening of dynamic heterogeneity fields in
618: space-time. An appreciation for the self-similarity of dynamic
619: heterogeneity fields at different temperatures has already been
620: noted in spin-facilitated models \cite{Garrahan-Chandler,
621: Whitelam-et-al}.
622:
623: Both sets of scaled structure factors in Fig.\ \ref{skscaled} have an
624: intermediate power law regime going as $k^{-2.3}$ for the (2)-TLG and
625: $k^{-2}$ for the (1)-TLG. One explanation for the difference
626: in exponents could be the following.
627: Porod law scaling, $k^{-(d+1)}$, arises from a
628: system which is extensive in interfaces. In two dimensions, this
629: implies that a system with a scaling exponent closer to 3 would have
630: smoother interfaces. This interpretation is consistent with the
631: snapshots of the
632: dynamic heterogeneity fields in Fig.\ \ref{2tlg} and Fig.\ \ref{1tlg}.
633:
634: \begin{figure}[t]
635: \epsfig{file = fig13.eps, width=2.5in, clip}
636: \caption{Structure factors of the mobility field, equation
637: (\ref{skeq}), measured at $\Delta t = \tau_{\alpha}$,
638: for various densities in the (a) (2)-TLG and the
639: (b) (1)-TLG. The axes are scaled as indicated.}
640: \label{skscaled}
641: \end{figure}
642:
643: The structure factors for the (1)-TLG imply a value of the dynamical
644: exponent
645: $\eta$ very close to zero, $S(k) \sim 1/k^{2-\eta}$ as
646: $k\rightarrow\infty$.
647: This gives a prediction for the exponent
648: $\gamma$ via the scaling relation $\gamma = (2-\eta)\nu_{\perp}$ of
649: $\gamma \approx 1.8$. The $\gamma$ exponent controls the scaling of
650: the dynamic susceptibility, $\chi \equiv S(q\rightarrow 0; \rho,
651: \tau_{\alpha})$ with the concentration of excitations.
652: The inset to Fig.\ \ref{skscaled}(b) shows that
653: this expectation is approximately satisfied \cite{Whitelam-et-al}.
654:
655: \section{Discussion}
656: \label{Discussion}
657:
658: Despite their simplicity, the constrained lattice gas models we have
659: studied show the essential features of glass forming liquids, such as
660: a precipitous dynamical slowdown and dynamical heterogeneity. They
661: are intermediate between the fully coarse grained kinetically
662: constrained spin models such as the FA and East models, and atomistic
663: models such as the binary Lennard-Jones mixture. We find a broad
664: distribution of persistence times, especially in the (2)-TLG
665: (Fig.\ \ref{Dta}).
666: From the scaling of
667: the structural relaxation time it follows that the (2)-TLG is a
668: fragile model and the (1)-TLG is a strong one, consistent with the
669: predicted scaling of the self-diffusion constant in \cite{Toninelli-et-al}.
670: Fragile behavior
671: versus non-fragile behavior coincides with hierarchical versus
672: diffusive propagation of excitations \cite{Jackle,Palmer-et-al}, and
673: the former follows from directional persistence
674: \cite{Garrahan-Chandler} as evident from the patterns of dynamic
675: heterogeneity seen in Fig.\ \ref{2tlg} and Fig.\ \ref{1tlg}.
676: Dynamic heterogeneity produces length-time
677: scaling and decoupling phenomena. Dynamic heterogeneity is present in
678: both strong and fragile materials, not only in the latter. This is
679: consistent with recent molecular dynamics simulation on silica
680: \cite{Vogel-Glotzer} and earlier theoretical predictions
681: \cite{Garrahan-Chandler,Berthier-Garrahan}.
682:
683:
684:
685: \acknowledgments
686:
687: We are grateful to Robert Jack for important discussions. This work
688: was supported by the US National Science Foundation, by the US
689: Department of Energy grant no.\ DE-FE-FG03-87ER13793, by EPSRC grants
690: no.\ GR/R83712/01 and GR/S54074/01, and University of Nottingham grant
691: no.\ FEF 3024. A.C.P. is an NSF graduate research fellow. This
692: research used resources of the National Energy Research Scientific
693: Computing Center, which is supported by the Office of Science of the
694: U.S. Department of Energy under Contract No. DE-AC03-76SF00098.
695:
696:
697: \begin{thebibliography}{99}
698:
699: \bibitem{Ediger-et-al} M.D. Ediger, C.A. Angell and S.R. Nagel,
700: J. Phys. Chem. {\bf 100}, 13200 (1996).
701:
702: \bibitem{Angell} C.A. Angell, Science {\bf 267}, 1924 (1995).
703:
704: \bibitem{Debenedetti-Stillinger} P.G. Debenedetti and F.H. Stillinger,
705: Nature {\bf 410}, 259 (2001).
706:
707: \bibitem{Kob-Andersen} W. Kob and H.C. Andersen, Phys. Rev. E {\bf
708: 48}, 4364 (1993).
709:
710: \bibitem{Jackle} J. J\"ackle and A. Kr\"onig, J. Phys. Condens. Matter
711: {\bf 6}, 7633 (1994); 7655 (1994).
712:
713: \bibitem{Ritort-Sollich} F. Ritort and P. Sollich, Adv. Phys. {\bf
714: 52}, 219 (2003).
715:
716: \bibitem{Tarjus-Kivelson} D. Kivelson, S.A. Kivelson, X.L. Zhao,
717: Z. Nussinov and G. Tarjus, Physica A {\bf 219}, 27 (1995).
718:
719: \bibitem{Xia-Wolynes} X. Xia and P.G. Wolynes,
720: Proc. Natl. Acad. Sci. USA {\bf 97}, 2990 (2000); J.-P. Bouchaud and
721: G. Biroli, J. Chem. Phys. {\bf 121}, 7347 (2004).
722:
723: \bibitem{Sellito-et-al} J. Kurchan, L. Peliti and M. Sellito,
724: Europhys.\ Lett.\ {\bf 39}, 365 (1997).
725:
726: \bibitem{Toninelli-et-al} C. Toninelli, G. Biroli and D.S. Fisher,
727: Phys. Rev. Lett. {\bf 92}, 185504 (2004); C. Toninelli and G. Biroli,
728: J. Stat. Phys. {\bf 117}, 27 (2004)
729:
730: \bibitem{Marinari-Pitard} E. Marinari and E. Pitard,
731: Euro. Phys. Lett. {\bf 69}, 235 (2005).
732:
733: \bibitem{Ediger} M.D. Ediger, Annu. Rev. Phys. Chem. {\bf 51}, 99
734: (2000).
735:
736: \bibitem{Glotzer} S.C. Glotzer, J. Non-Cryst. Solids, {\bf 274}, 342
737: (2000).
738:
739: \bibitem{Harrowell-et-al} D.N. Perera and P. Harrowell,
740: J. Chem. Phys. {\bf 111}, 5441 (1999).
741:
742: \bibitem{Garrahan-Chandler} J.P. Garrahan and D. Chandler,
743: Phys. Rev. Lett. {\bf 89}, 035704 (2002); Proc. Natl. Acad. Sci. USA
744: {\bf 100}, 9710 (2003).
745:
746: \bibitem{Berthier-Garrahan} L. Berthier and J.P. Garrahan,
747: J. Chem. Phys. {\bf 119}, 4367 (2003); Phys. Rev. E {\bf 68}, 041201
748: (2003).
749:
750: \bibitem{Jung-et-al} Y.J. Jung, J.P. Garrahan and D. Chandler,
751: Phys. Rev. E {\bf 69}, 061205 (2004).
752:
753: \bibitem{Whitelam-et-al} S. Whitelam, L. Berthier and J.P. Garrahan,
754: Phys. Rev. Lett. {\bf 92}, 185705 (2004); Phys. Rev. E {\bf 71},
755: 026128 (2005).
756:
757: \bibitem{Berthier-et-al} L. Berthier, D. Chandler and J.P. Garrahan,
758: Euro. Phys. Lett. {\bf 69}, 320 (2005)
759:
760: \bibitem{qnote} The wavevector dependent quantity, $F_s(q, t)$, was
761: calculated as a Fourier transform on the square lattice. A
762: correspondence can be made between the reciprocal lattice vectors of
763: the square and triangular lattices as in \cite{Jackle}.
764:
765: \bibitem{Newman-Barkema} M.E.J. Newman and G.T. Barkema, {\em Monte
766: Carlo Methods in Statistical Physics} (Oxford University Press,
767: Oxford, 1999).
768:
769: \bibitem{OnukiPRE}
770: R. Yamamoto and A. Onuki, Phys. Rev. E {\bf 58}, 3515 (1998).
771:
772: \bibitem{RichterFrickFarago} D. Richter, B. Frick and B. Farago,
773: Phys. Rev. Lett. {\bf 61}, 2465 (1988).
774:
775: \bibitem{mct} W. G\"{o}tze and L. Sj\"{o}gren,
776: Rep. Prog. Phys. {\bf 55}, 55 (1992).
777:
778: \bibitem{Trivial} An alternative explanation is that relaxation in the
779: (1)-TLG can simply be accounted for by the free diffusion of vacancy
780: pairs, with interactions playing no role asymptotically. In that
781: case $\tau_\alpha$ would scale as $c^{-2}$ and the deviations
782: measured numerically would correspond to logarithmic corrections
783: characteristic of two-dimensional diffusion. This is the situation,
784: for example, in the square-plaquette spin model \cite{PQ}.
785:
786: \bibitem{PQ} R. Jack, L. Berthier and J.P. Garrahan, cond-mat/0502120.
787:
788: \bibitem{Swallen-et-al} S.F. Swallen, P.A. Bonvallet, R.J. McMahon and
789: M.D. Ediger, Phys. Rev. Lett. {\bf 90}, 015901 (2003).
790:
791: \bibitem{Chang-Sillescu} I. Chang and H. Sillescu, J. Phys. Chem. B
792: {\bf 101}, 8794 (1997).
793:
794: \bibitem{Schweizer-Saltzman} K.S. Schweizer and E.J. Saltzman,
795: J. Phys. Chem. B {\bf 108}, 19729 (2004)
796:
797: \bibitem{Berthier} L. Berthier, Phys. Rev. E \textbf{69}, 020201(R)
798: (2004).
799:
800: \bibitem{Ediger-et-al-2} M.D. Ediger et al., unpublished.
801:
802: \bibitem{NEF} L. Berthier and J.P. Garrahan, J. Phys. Chem. B {\bf
803: 109}, 3578 (2005).
804:
805: \bibitem{Lacevic-et-al} N. Lacevic, F.W. Starr, T.B. Schr{\o}der and
806: S.C. Glotzer, J. Chem. Phys. {\bf 119}, 7372 (2003).
807:
808: \bibitem{Trivial2} If interactions play no role in the (1)-TLG, then
809: we would expect to observe $\nu=1$ and $z=2$.
810:
811: \bibitem{Palmer-et-al} R. G. Palmer, D. L. Stein, E. Abraham, and
812: P. W. Anderson, Phys.\ Rev.\ Lett.\ {\bf 53}, 958 (1984).
813:
814: \bibitem{Vogel-Glotzer} M. Vogel and S.C. Glotzer,
815: Phys. Rev. Lett. {\bf 92}, 255901 (2004).
816:
817: \bibitem{Amar-et-al} J.G. Amar, F.E. Sullivan, and R.D. Mountain,
818: Phys. Rev. B {\bf 27}, 196 (1988).
819:
820: \bibitem{largek} The discrepancy at large $k$ is due to probing the
821: granularity of the lattice.
822:
823: \end{thebibliography}
824:
825: \end{document}
826:
827:
828:
829: