cond-mat0605244/chi.tex
1: \documentclass[twocolumn,aps,pre,showpacs]{revtex4}
2: 
3: \usepackage{graphicx}
4: \begin{document}
5: 
6: \title{Four-point susceptibility of a glass-forming binary mixture: 
7: Brownian dynamics}
8: \author{Grzegorz Szamel and Elijah Flenner}
9: \affiliation{Department of Chemistry,
10: Colorado State University, Fort Collins, CO 80523}
11: 
12: \date{\today}
13: 
14: \pacs{64.70.Pf}
15: 
16: \begin{abstract}
17: We study the four-point dynamic susceptibility $\chi_4(t)$ 
18: obtained from Brownian dynamics computer simulations 
19: of the Kob-Andersen Lennard-Jones mixture.
20: We compare the results of the simulations with 
21: qualitative predictions of the mode-coupling theory. In addition, we
22: test an estimate of the four-point susceptibility recently proposed
23: by Berthier \textit{et al.} [Science, \textbf{310}, 1797 (2005)]. 
24: \end{abstract}
25: \maketitle
26: 
27: 
28: \section{\label{intro}Introduction}
29: 
30: The origin of the extreme slowing down of liquids' dynamics upon approaching 
31: the glass transition and the very nature of this transition are still
32: hotly debated despite many experimental, simulational, and theoretical 
33: studies performed in the last two decades \cite{DebStil}.
34: One of the fundamental difficulties
35: is that the slowing down appeared to be a local, small
36: scale phenomenon which is not accompanied by a growing correlation length.
37: No long-range correlations have ever been found in any static quantity upon 
38: approaching the glass transition. The attention has 
39: recently shifted to dynamic correlations \cite{BB}, and there is evidence that 
40: there is a dynamic correlation length that slowly grows upon
41: approaching the glass transition. Unfortunately this dynamic correlation 
42: length cannot be easily obtained from experimental data.
43: 
44: The dynamic correlation length is often obtained
45: from the four-point dynamic susceptibility $\chi_4(t)$. This  
46: susceptibility is related to a space integral of a four-point density 
47: correlation function that quantifies correlations between relaxation 
48: processes at different points in space. It is assumed that 
49: the increase of the dynamic susceptibility signals the growth of the 
50: range of the correlations between relaxation processes. 
51: Furthermore, the four-point dynamic susceptibility can be 
52: used to discriminate between different theoretical approaches to 
53: glassy dynamics. For example, in a recent paper \cite{Toninelli4p} 
54: Toninelli \textit{et al.} discussed predictions for $\chi_4(t)$ 
55: obtained from various theoretical approaches to glassy dynamics and compared 
56: these predictions with two different simulations of 
57: atomistic models of glass forming liquids. 
58: 
59: A problem with the four-point susceptibility is that it is  
60: difficult to extract from experiments. This problem has been addressed
61: in another recent paper \cite{BerthierScience}. 
62: Berthier \textit{et al.}~argued that the four-point susceptibility can be
63: estimated using derivatives of an intermediate scattering function
64: with respect to thermodynamic parameters such as temperature or density. 
65: Since the intermediate scattering function can be easily obtained from 
66: experiments, Ref.~\cite{BerthierScience} introduced a way to experimentally 
67: investigate the existence of a growing correlation length. 
68: 
69: The goal of this contribution is twofold. First, we analyze 
70: the four-point susceptibility of the Kob-Andersen
71: \cite{KobAndersen} Lennard-Jones binary mixture undergoing
72: Brownian dynamics. Second, we test the approximate
73: estimate for the four point susceptibility proposed by Berthier \textit{et al.}
74: \cite{BerthierScience}. 
75: 
76: The paper is organized as follows: in Sec. \ref{model} we briefly review
77: the details of the simulation; in Sec. \ref{chi4} we define the 
78: four-point susceptibility and analyze the simulation results for this
79: quantity; in Sec. \ref{estimate} we compare the four-point susceptibility 
80: obtained directly from computer simulations 
81: to the approximate expression derived by Berthier \textit{et al.};
82: and in Sec. \ref{conclusion} we discuss our findings.
83: 
84: \section{\label{model} Simulation details} 
85: 
86: We simulated a binary mixture that was introduced by
87: Kob and Andersen which consists of $N_A=800$ particles of type A and
88: $N_B=200$ particles of type B. The interaction potential is
89: $V_{\alpha \beta}(r) = 4\epsilon_{\alpha \beta}[
90: ({\sigma_{\alpha \beta}}/{r})^{12} - ({\sigma_{\alpha \beta}}/{r})^6]$,
91: where $\alpha, \beta \in \{A,B\}$, $\epsilon_{AA} = 1.0$,
92: $\sigma_{AA} = 1.0$, $\epsilon_{AB} = 1.5$, $\sigma_{AB} = 0.8$,
93: $\epsilon_{BB} = 0.5$, and $\sigma_{BB} = 0.88$.  The interaction potential 
94: was cut off 
95: at $2.5\ \sigma_{\alpha \beta}$. We used a cubic simulation cell with 
96: the box length of $9.4\ \sigma_{AA}$ with periodic boundary conditions.
97: 
98: We performed Brownian dynamics simulations.  The equation of
99: motion for the position of the $i_{th}$ particle of type $\alpha$, 
100: $\vec{r}\,_{i}^{\alpha}$, is
101: \begin{equation}\label{Lang}
102: \dot{\vec{r}}\,_{i}^{\alpha} = \frac{1}{\xi_0} \vec{F}_i^{\alpha}
103: + \vec{\eta}_i(t) ,
104: \end{equation}
105: where $\xi_0$ is the friction coefficient of an isolated particle, 
106: $\xi_0 = 1.0$, and $\vec{F}_i^{\alpha}$
107: is the force acting on the $i_{th}$ particle of type $\alpha$,
108: \begin{equation}\label{force}
109: \vec{F}_i^{\alpha}= - \nabla_i^{\alpha} \sum_{j} \sum_{\beta=1}^2
110: V_{\alpha \beta}\left(\left|\vec{r}\,_i^{\alpha} - 
111: \vec{r}\,_j^{\beta} \right| \right)
112: \end{equation}
113: with $\nabla_i^{\alpha}$ being the gradient operator with 
114: respect to $\vec{r}\,_i^{\alpha}$ (note that the term with 
115: $\beta=\alpha$ and $i=j$ has to be excluded from the double sum in 
116: Eq. (\ref{force})). 
117: In Eq. (\ref{Lang})
118: the random noise $\vec{\eta}_i$ satisfies the fluctuation-dissipation
119: theorem,
120: \begin{equation}\label{fd}
121: \left\langle \vec{\eta}_i(t) \vec{\eta}_j(t') \right\rangle =
122: 2 D_0 \delta(t-t') \delta_{ij} \mathbf{1}.
123: \end{equation}
124: Here $D_0$ is the diffusion coefficient of an isolated 
125: particle, $D_0 = k_B T/\xi_0$, where
126: $k_B$ is Boltzmann's constant. Furthermore, in Eq. (\ref{fd}) $\mathbf{1}$ is
127: the unit tensor.  The equations of motion (\ref{Lang}-\ref{fd}) 
128: allow for diffusive motion of the center of mass, thus 
129: all the results will be presented relative to the
130: center of mass (\textit{i.e.}\ momentary positions of all the particles are 
131: always relative to the momentary position of the center of mass \cite{remark}).
132: The results are presented in terms of the reduced units
133: with $\sigma_{AA}$, $\epsilon_{AA}$, $\epsilon_{AA}/k_B$,
134: and $\sigma_{AA}^2 \xi_0/\epsilon_{AA}$ being the units of length, energy,
135: temperature, and time, respectively. In these units the short-time 
136: self-diffusion coefficient is proportional to the temperature, thus
137: the time is rescaled to a reduced time equal to $t D_0/\sigma_{AA}^2$
138: to facilitate comparisons with theoretical approaches that often
139: assume temperature-independent short-time dynamics.
140: 
141: The equations of motion were solved using a Heun algorithm 
142: with a small time step of $5 \times 10^{-5}$.  We simulated a broad range of
143: temperatures $0.44 \le T \le 5.0$. Here we present results for the 
144: following temperatures: $T = 0.45$, 0.47, 0.50, 0.55, 0.60, 0.65, 0.80, and 1.0.
145: We ran equilibration runs and 4-6 production runs.
146: The equilibration runs were typically twice to four times 
147: shorter than the production runs, and the latter
148: were up to $1.2 \times 10^9$ time steps long for the lowest temperature
149: discussed here, $T=0.45$. 
150: The results presented are averages over the production runs.
151: 
152: \section{\label{chi4}Four-point dynamic susceptibility}
153: 
154: In the context of the dynamics of 
155: supercooled liquids, the four-point susceptibility has been 
156: introduced by Glotzer and collaborators \cite{Sharon}.
157: The four-point susceptibility that we discuss in this paper is 
158: slightly different from that defined in Refs.~\cite{Sharon}. 
159: We consider here the susceptibility which was extensively  
160: analyzed by Toninelli \textit{et al.}~\cite{Toninelli4p}, which 
161: is defined as the variance
162: of the fluctuations of the self-intermediate scattering function.
163: 
164: We start with the definition of the self-intermediate scattering
165: function, $F_s(k;t)$,
166: \begin{equation}\label{Fs}
167: F_s(k;t) = \left<  \frac{1}{N} \sum_i \cos \vec{k}\cdot 
168: [\vec{r}_i(t) - \vec{r}_i(0)] \right>.
169: \end{equation}
170: Next, we define the fluctuation of the instantaneous value of the 
171: scattering function, $\delta F_s(\vec{k};t)$,
172: \begin{equation}\label{deltaFs}
173: \delta F_s(\vec{k};t) = \frac{1}{N} \sum_i \cos \vec{k}\cdot 
174: [\vec{r}_i(t) - \vec{r}_i(0)] - F_s(k;t).
175: \end{equation}
176: The four-point susceptibility, $\chi_4(t)$, is then defined as
177: \begin{equation}\label{chi4b}
178: \chi_4(t) = 
179: N \left< \delta F_s(\vec{k};t) \delta F_s(\vec{k};t) \right>.
180: \end{equation}
181: The four-point susceptibility depends on the
182: wave vector $\vec{k}$ (and it is sometimes denoted by 
183: $\chi_{\vec{k}}(t)$). This wave vector is customarily fixed at the position of
184: the maximum of the static structure factor \cite{Chandler}.
185: 
186: The system considered in this paper is a two-component mixture, thus
187: instead of one self-intermediate scattering function $F_s(k;t)$ 
188: we could introduce
189: two different scattering functions involving particles $A$ and $B$. All the
190: results presented in this paper concern the $A$ particles only, thus,
191: \textit{e.g.} sums in Eqs. (\ref{Fs}-\ref{deltaFs}) run only over the
192: $A$ particles. Since we 
193: are not presenting any results for the $B$ particles we do not introduce 
194: additional sub-scripts or super-scripts indicating particle labels. The magnitude
195: of the wave vector
196: $\vec{k}$ is fixed at the  position of the maximum of the partial structure 
197: factor of the $A$ particles, $|\vec{k}|= 7.25$.
198: 
199: The mode coupling theory of the glass transition 
200: was formulated by G\"otze and collaborators
201: for Newtonian systems \cite{Goetze}, and was later extended 
202: by Szamel and L\"owen to Brownian systems
203: \cite{SL}. The theory makes predictions 
204: for the self- and collective intermediate scattering functions. 
205: Biroli and Bouchaud \cite{BBEPL} recently 
206: argued that the mode coupling theory could
207: be understood as a dynamic mean-field theory, and that the usual
208: mode coupling equations are saddle point equations obtained from an
209: action functional. This new interpretation made it possible to calculate 
210: fluctuations of the order parameter (\textit{i.e.}\ fluctuations of a 
211: two-point dynamic correlation function) 
212: from the inverse of the second derivative of the action functional. The
213: details of the calculation have not been published, but the main predictions
214: have already been discussed and compared with Newtonian dynamics simulations
215: \cite{Toninelli4p}. According to Biroli and Bouchaud, on the $\beta$ 
216: relaxation time scale the four-point susceptibility grows with time 
217: as a power law, $\chi_4(t) \propto t^{\mu}$, with exponents equal to the 
218: standard mode coupling exponents $a$ and $b$ in the 
219: early and late $\beta$ regime, respectively. 
220: Furthermore,  $\chi_4(t)$ reaches its maximum value, $\chi_4(t^*)$, 
221: on the time scale of the $\alpha$ relaxation time, 
222: $t^* \sim \tau_{\alpha}$, and the maximum value diverges upon approaching
223: the mode coupling temperature, $\chi_4(t^*)\propto (T-T_c)^{-\gamma_1}$
224: with $\gamma_1=1$.
225: 
226: The authors of Refs.~\cite{Toninelli4p,BerthierScience} emphasized 
227: that these predictions are valid for Newtonian systems in the microcanonical
228: (NVE) ensemble and hinted that they may be different in the canonical
229: (NVT) ensemble. 
230: We present results obtained from Brownian dynamics
231: simulations. In such simulations constant temperature is maintained 
232: automatically by the equations of motion. Thus, in principle it is not
233: clear whether the predictions of Biroli and Bouchaud \cite{BBEPL}
234: are relevant for our simulation results. However, there are subtle theoretical
235: arguments that suggest that the reverse is true \cite{BBprivinfo}. 
236: In addition, at the level of the mode coupling equations, there is no
237: difference between Newtonian systems in the NVE or NVT ensemble, and
238: Brownian systems (beyond the microscopic time scale, \textit{i.e.}
239: on the time scales of the $\beta$ relaxation and longer) \cite{SL}. 
240: These reasons encouraged us
241: to compare our results obtained from Brownian dynamics
242: simulations to the predictions of Biroli and Bouchaud \cite{BBEPL}.
243: 
244: In Fig. \ref{chi4fig} we show the general shape of the four-point 
245: susceptibility, $\chi_4(t)$, for several temperatures. The time 
246: dependence of $\chi_4(t)$ is similar to that obtained for Newtonian
247: systems, see Figs. 4 and 5 of Ref. \cite{Toninelli4p}. In particular,
248: it can be argued that there is power law like growth in time of the four-point 
249: susceptibility as it approaches it's maximum value. 
250: This power law-like dependence 
251: will be further analyzed at the end of this section.
252: 
253: \begin{figure}
254: 	\includegraphics[scale=0.25]{chi4.eps}
255: \caption{\label{chi4fig} Time dependence of  the four-point susceptibility 
256: $\chi_4(t)$ plotted \textit{vs.} $t D_0/\sigma^2$ for $T=$1.0, 0.80,
257: 0.60, 0.55, 0.50, 0.47, and 0.45 (left to right).}
258: \end{figure}
259: 
260: In Fig.~\ref{peaketc} we present the results of some quantitative analysis of $\chi_4(t)$. 
261: First, in Fig.~\ref{peaketc}a we show the temperature dependence of the
262: its maximum value, $\chi_4(t^*)$. We plot the maximum value \textit{vs.}
263: $\epsilon=(T-T_c)/T_c$ where $T_c=0.435$ is the standard value
264: of the mode coupling temperature for the Kob-Andersen Lennard-Jones binary 
265: mixture \cite{KobAndersen}. In an intermediate range
266: of temperatures, $\chi_4(t^*)$ grows as a power law 
267: with decreasing $(T-T_c)/T_c$, $\chi_4(t^*)\propto \epsilon^{-\gamma_1}$. 
268: The exponent obtained from the fit, 
269: $\gamma_1=0.995 \pm 0.05$, 
270: is very close to the theoretical prediction of Biroli 
271: and Bouchaud, $\gamma_1^{th}=1$. 
272: It should be noted that the range of reduced temperatures 
273: for which the power law dependence of $\chi_4(t^*)$ is observed coincides
274: with the range of reduced temperatures for which mode coupling theory was
275: found to correctly describe the time evolution of the self-intermediate
276: scattering function and the mean square displacement \cite{FS}.
277: In Fig.~\ref{peaketc}b we compare the temperature dependence of the 
278: time at which $\chi_4(t)$ reaches its maximum value, $t^*$, with that
279: of the $\alpha$ relaxation time, $\tau_{\alpha}$. The latter time
280: is defined as the time at which the self-intermediate 
281: scattering function decays to $e^{-1}$ of its initial value,
282: $F^s(q,\tau_{\alpha}) = e^{-1}$. We find that 
283: $t^*$ and $\tau_{\alpha}$ are very close and have the same temperature
284: dependence. In particular, in the same intermediate range
285: of temperatures these times grow according to power laws with decreasing 
286: $(T-T_c)/T_c$, $t^*\propto \epsilon^{-\gamma^*}$, 
287: $\tau_{\alpha}\propto \epsilon^{-\gamma}$.  
288: The exponents obtained from the fits, 
289: $\gamma^*=2.27 \pm 0.04$ and $\gamma=2.31 \pm 0.02$ are very close. The numerical solution
290: of the mode coupling equations predicts a slightly higher value of the 
291: scaling exponent for $\tau_{\alpha}$, $\gamma^{th} = 2.46$ \cite{FS}. 
292: Finally, in Fig.~\ref{peaketc}c we plot $\chi_4(t^*)$ \textit{vs.}
293: $t^*$. We find that slight deviations from power laws that are 
294: visible in Figs.~\ref{peaketc}a and \ref{peaketc}b are magnified in 
295: Fig.~\ref{peaketc}c.
296: Nevertheless we can still fit a power law $\chi_4(t^*) \propto 
297: (t^*)^{1/\gamma_2}$ with the exponent $\gamma_2=2.31 \pm 0.21$. 
298: 
299: \begin{figure}
300: 	\includegraphics[scale=0.25]{chi4peak.eps}\\[0.25cm]
301: 	\includegraphics[scale=0.25]{tautstar.eps}\\[0.25cm]
302: 	\includegraphics[scale=0.25]{peaktstar.eps}
303: \caption{\label{peaketc}
304: (a) Temperature dependence of the maximum value of the four-point 
305: susceptibility: $\chi_4(t^*)$ plotted \textit{vs.} $\epsilon=(T-T_c)/T_c$.
306: Solid line is the power law fit $\chi_4(t^*) \propto \epsilon^{-\gamma_1}$
307: with $\gamma_1$=0.995. 
308: (b) Temperature dependence of the time at which $\chi_4(t)$ reaches 
309: its maximum value, $t^*$ (circles), and the $\alpha$ relaxation time,
310: $\tau_{\alpha}$ (squares). Both times are plotted \textit{vs.} 
311: $\epsilon=(T-T_c)/T_c$. The solid line and the dashed line are power law fits
312: $t^*\propto \epsilon^{-\gamma^*}$ with $\gamma^*$=2.27 and 
313: $\tau_{\alpha}\propto \epsilon^{-\gamma}$ with $\gamma$=2.31 , respectively.
314: (c) The maximum value of the four-point 
315: susceptibility, $\chi_4(t^*)$, plotted \textit{vs.} $t^*$. The solid line is
316: the power law fit $\chi_4(t^*)\propto (t^*)^{1/\gamma_2}$ with $\gamma_2$ = 
317: 2.31.}
318: \end{figure}
319: 
320: In Fig.~\ref{power} we address the question of the power law dependence of
321: the four-point susceptibility on time. Instead of trying to fit power laws
322: to $\chi_4(t)$ over some specific time intervals, 
323: we have numerically calculated the 
324: derivative of $\ln \chi_4(t)$ with respect to $\ln t$, 
325: $d \ln \chi_4(t) / d\ln t = 
326: (t/\chi_4(t)) d\chi_4(t)/dt$. For any time interval over which $\chi_4(t)$
327: depends on $t$ in a power law fashion,  $d \ln \chi_4(t) / d\ln t$ should have
328: a constant value (a plateau). On the basis of the predictions of 
329: Ref.~\cite{BBEPL} we expect two different plateaus in our data.
330: A plateau in the early-$\beta$ regime and another one in the late-$\beta$ 
331: regime, \textit{i.e.} on approaching the maximum of $\chi_4(t)$ (additional
332: plateaus are predicted for shorter times, and we do not analyze them here). 
333: However, the simulations agree with mode coupling predictions
334: only over a restricted range of temperatures. It is not clear whether 
335: these two time scales are well separated over this temperature range, 
336: thus it is not clear if we could see two different plateaus. 
337: 
338: Indeed, Fig. \ref{power} does not show well developed plateaus. 
339: However, we notice that $\chi_4(t)$ grows with $t$ with an exponent of 
340: approximately $b = 0.8$ upon approaching it's maximum for all
341: temperatures. 
342: This exponent is comparable to the exponents 
343: obtained from Newtonian dynamics simulations
344: \cite{Toninelli4p}. It is quite a bit higher than
345: the exponent obtained from the mode coupling theory $b^{th} = 0.62$
346: \cite{bremark}.
347:   
348: \begin{figure}
349: 	\includegraphics[scale=0.25]{pexp5.eps}
350: \caption{\label{power} Derivative of $\ln \chi_4(t)$ w.r.t $\ln t$,
351: $d \ln \chi_4(t) / d\ln t = (t/\chi_4(t)) d\chi_4(t)/dt$,
352: plotted \textit{vs.} $t/t^*$ for $T=$ 0.45, 0.47, 0.50, and 0.55
353: (left to right).}
354: \end{figure}
355: 
356: \section{\label{estimate}Test of the estimate of the four-point susceptibility
357: proposed by Berthier \textit{et al.}}
358: 
359: The four-point susceptibility can be easily obtained from 
360: computer simulations, but it is not readily determined from experiments.
361: To address this problem Berthier \textit{et al.}~\cite{BerthierScience}
362: proposed an approximate estimate for the four-point susceptibility 
363: in terms of the derivative of the self-intermediate scattering function
364: with respect to temperature, $\chi_T(t)$,
365: \begin{equation}\label{chiT}
366: \chi_T(t) = \frac{\partial F_s(k;t)}{\partial T}.
367: \end{equation}
368: 
369: The starting point of their argument was a fluctuation-dissipation relation
370: (Eq. (2) of Ref. \cite{BerthierScience}).
371: If we naively adopted this relation to our system 
372: (\textit{i.e.} Brownian dynamics, NVT ensemble) we would get 
373: \begin{equation}\label{FDT}
374: k_B T^2 \chi_T(t) = 
375: \left< \delta F_s(\vec{k};t) \delta V(0) \right>.
376: \end{equation} 
377: Here $\delta V(t)$ denotes the instantaneous fluctuation of the total
378: potential energy,
379: \begin{equation}\label{V} 
380: V(t) = \sum_{i,j}\!'\sum_{\alpha\beta} 
381: V_{\alpha \beta}\left(\left|\vec{r}\,_i^{\alpha}(t) - 
382: \vec{r}\,_j^{\beta}(t) \right| \right),
383: \end{equation} 
384: \begin{equation}\label{deltaV} 
385: \delta V(t) = V(t) - \left<V(t)\right>.
386: \end{equation} 
387: The derivative with respect to the temperature in
388: Eq.~(\ref{estimate}), Eq.~(\ref{FDT}) and in the remainder of this section 
389: have to be calculated while keeping the short-time diffusion
390: coefficient $D_0$ constant since our short-time diffusion 
391: coefficient is proportional to the temperature. 
392: 
393: The main analytical result obtained by Berthier \textit{et al.} was an exact 
394: lower bound for $\chi_4(t)$ in terms of $\chi_T(t)$
395: (Eq. (5) of Ref.~\cite{BerthierScience}). Naively applying this 
396: result to our Brownian system we would get
397: \begin{equation}\label{bound}
398: \chi_4(t) \ge \frac{k_B}{c^{pot}_V} T^2 \chi_T^2(t).
399: \end{equation}
400: Here $c^{pot}_V$ is the potential contribution to the constant volume
401: specific heat per particle.
402: 
403: Berthier \textit{et al.} found that the difference
404: between the right and left sides of relation
405: (\ref{bound}) diminishes with decreasing temperature. On this basis, they 
406: proposed using the right-hand-side of Eq.~(\ref{bound}) as an approximate 
407: estimate for the four-point susceptibility. Furthermore, they showed that this
408: estimate can be easily calculated using either experimental or simulational
409: results for the self-intermediate scattering function. 
410: 
411: Both the fluctuation-dissipation relation 
412: and the bound derived by Berthier \textit{et al.}~are valid only
413: if the equations of motion do not involve the 
414: temperature. The only place where the temperature can enter is the 
415: initial condition which is given by the canonical distribution. 
416: This is true neither for the usual NVT computer simulations,
417: where the temperature enters into the equations of motion 
418: \textit{via} a thermostat, nor for our Brownian dynamics 
419: simulations, where the temperature enters \textit{via} the noise strength. 
420: Thus we have no arguments in favor of either the 
421: fluctuation-dissipation relation (\ref{FDT}) or the bound (\ref{bound}) for
422: our system.
423: In fact, we show explicitly in Fig. \ref{chiTcomp} that the relation
424: (\ref{FDT}) is violated in Brownian systems. On the other hand
425: we do not have strong numerical evidence for the violation of
426: the inequality (\ref{bound}). Figure \ref{chi4comp}c suggests that
427: this inequality is violated, but the extent of the violation is smaller
428: than the error bars. 
429: 
430: We should mention that there is a related, exact bound for the four-point 
431: susceptibility that follows from the Cauchy-Schwarz inequality 
432: (see footnote 22 of Ref. \cite{BerthierScience}):
433: \begin{equation}\label{boundCS}
434: \chi_4(t) \ge \frac{\left< \delta F_s(\vec{k};t) \delta V(0) \right>^2}{k_B T^2 c^{pot}_V}.
435: \end{equation}
436: However, as we show in Fig. \ref{chi4comp} this bound does not lead
437: to a useful estimate for the four-point susceptibility.
438: 
439: \begin{figure}
440: 	\includegraphics[scale=0.25]{chiTdFcomp10n.eps}\\[0.25cm]
441: 	\includegraphics[scale=0.25]{chiTdFcomp055n.eps}\\[0.25cm]
442: 	\includegraphics[scale=0.25]{chiTdFcomp045n.eps}
443: \caption{\label{chiTcomp}
444: Comparison of $k_B T^2 \chi_T(t)= k_B T^2 \partial F(k;t)/\partial T$ 
445: (closed circles) with  $\left< \delta F_s(\vec{k};t) \delta V(0) \right>$
446: (open squares) at three representative temperatures: 
447: (a) T=1.00,
448: (b) T=0.55 and 
449: (c) T=0.45.}
450: \end{figure}
451: 
452: Berthier \textit{et al.} also gave a different, general 
453: argument leading to an estimate for the four-point susceptibility in terms of
454: the right-hand-side of inequality (\ref{bound}).
455: This argument can be easily adopted for our system.  If we assume that 
456: the main source of fluctuations of the instantaneous expression for the
457: scattering function is the potential energy, we get
458: \begin{equation}\label{fluct}
459: \delta F_s(\vec{k};t) \approx 
460: \frac{\partial F_s(k;t)}{\partial T}
461: \frac{\delta V(0)}{N c^{pot}_V}.
462: \end{equation}
463: Eq. (\ref{fluct}) leads immediately to 
464: \begin{equation}\label{est}
465: \chi_4(t) \approx \frac{k_B}{c^{pot}_V} T^2 \chi_T^2(t).
466: \end{equation}
467: 
468: \begin{figure}
469: 	\includegraphics[scale=0.25]{chi4dFcomp10n.eps}\\[0.25cm]
470: 	\includegraphics[scale=0.25]{chi4dFcomp055n.eps}\\[0.25cm]
471: 	\includegraphics[scale=0.25]{chi4dFcomp045n.eps}
472: \caption{\label{chi4comp}
473: Comparison of $\chi_4(t)$ (open circles) with 
474: $k_B T^2 \chi_T^2(t)/c_V^{pot}$ 
475: (closed circles) and  
476: $\left< \delta F_s(\vec{k};t) \delta V(0) \right>^2/(k_B T^2 c^{pot}_V)$
477: (open squares) at three representative temperatures: 
478: (a) T=1.00,
479: (b) T=0.55 and 
480: (c) T=0.45.}
481: \end{figure}
482: 
483: In Fig. \ref{chi4comp}
484: we show the time dependence of both sides of the 
485: relation (\ref{est}) at three representative temperatures. 
486: While the approximation (\ref{est}) is inaccurate,
487: it becomes better, especially near the maximum value of $\chi_4(t)$,
488: when the temperature approaches the mode coupling temperature.
489: 
490: \begin{figure}
491: 	\includegraphics[scale=0.25]{peakcomp.eps}
492: \caption{\label{peakcomp} Temperature dependence of the maximum value of 
493: $k_B T^2 \chi_T^2(t)/c_V^{pot}$ 
494: (closed circles) compared to that of the maximum value of $\chi_4(t)$
495: (open circles).}
496: \end{figure}
497: 
498: Figure \ref{peakcomp} compares 
499: the temperature dependence of the maximum value of the
500: right-hand-side of the relation (\ref{est}) to the temperature
501: dependence of the maximum value of the  
502: four-point susceptibility. The difference between the two diminishes
503: with decreasing temperature. However, it is clear that 
504: the temperature dependence of these quantities is different. In particular,
505: one should be cautious when using
506: the temperature dependence of the estimate (\ref{est}) to obtain the
507: temperature dependence of the dynamic correlation length, at least for
508: temperatures above the mode coupling temperature.
509: 
510: Finally, Fig. \ref{chi4dercomp} compares the derivative of the logarithm 
511: of the four-point susceptibility, $\ln \chi_4(t)$,  
512: with respect to the logarithm of time, $\ln t$
513: and the derivative of the logarithm of the estimate (\ref{est}), 
514: $\ln k_B T^2 \chi_T^2(t)/c^{pot}_V = 2\ln \chi_T(t) + const.$, 
515: with respect to $\ln t$. We find that the estimate 
516: (\ref{est}) approaches its maximum value with a power law exponent that 
517: is approximately equal to 1 and thus somewhat higher than the power
518: law exponent describing $\chi_4(t)$'s approach to its maximum value. 
519: Moreover, comparing Figs. \ref{chi4dercomp}a and \ref{chi4dercomp}b 
520: we find that the
521: difference between exponents obtained from the estimate (\ref{est}) 
522: and from the four-point susceptibility
523: does not seem to diminish with decreasing temperature. 
524: 
525: \begin{figure}
526: 	\includegraphics[scale=0.25]{expchi4dF055n.eps}\\[0.25cm]
527: 	\includegraphics[scale=0.25]{expchi4dF045n.eps}
528: \caption{\label{chi4dercomp}
529: Comparison of the derivative 
530: $d\ln \chi_4(t)/d \ln t = (t/\chi_4(t))d\chi_4(t)/dt$ 
531: (open circles) and the derivative 
532: $d\ln k_B T^2 \chi_T^2(t)/c^{pot}_V/d\ln t = 2(t/\chi_T(t))d\chi_T(t)/dt$ 
533: (closed circles) at two temperatures: (a) T=0.55, and (b) T=0.45.}
534: \end{figure}
535: 
536: 
537: \section{\label{conclusion}Conclusions}
538: 
539: We performed a quantitative analysis of the four-point susceptibility
540: $\chi_4(t)$ of the Kob-Andersen Lennard-Jones binary mixture. We
541: compared the results of Brownian dynamics computer simulations 
542: to predictions obtained from a recent re-formulation of the  
543: mode coupling theory. We did not compare computer simulation results 
544: to other approaches to glassy dynamics (\textit{e.g.}, to the predictions 
545: obtained using facilitated kinetic Ising models \cite{fkIm1,fkIm2}) 
546: because these 
547: other approaches are more concerned with a temperature range 
548: lower than the one accessible in our simulations.
549: 
550: We found that some of the mode coupling predictions agree with our simulation
551: results. Most notably, the height of the maximum of the four-point 
552: susceptibility grows upon approaching the mode coupling temperature from above
553: in the way predicted by the theory. Moreover, the time at which $\chi_4(t)$
554: reaches its maximum value has the same temperature dependence as the
555: $\alpha$ relaxation time. For each of these two quantities, 
556: which are derived from 
557: a \emph{four}-point function, the power law  dependence on
558: $(T-T_c)/T_c$ is obeyed over the same temperature range as for quantities 
559: derived from \emph{two}-point functions, \textit{e.g.}
560: the $\alpha$ relaxation time and the self-diffusion coefficient.
561: 
562: On the other hand, we found some disagreement with the theory with
563: regard to the power law dependence of $\chi_4(t)$ on time. 
564: We found that upon approaching its maximum value $\chi_4(t)$ grows with
565: time with an effective exponent of about 0.8.
566: This exponent is quite a bit larger than the exponent predicted by the theory.
567: It should be recalled that the exponent predicted by the theory is
568: in turn larger than the one obtained by fitting a formula inspired by
569: the mode coupling theory to the (two-point) 
570: self-intermediate scattering function. Hence, our understanding of the 
571: connection between the time dependence of the four-point susceptibility 
572: and the time dependence of two-point functions seems incomplete.
573: 
574: Finally, we tested the approximate estimate of the four-point susceptibility
575: in terms of the temperature derivative of the self-intermediate scattering
576: function. We found that
577: the estimate becomes accurate around the peak of $\chi_4(t)$ upon approaching
578: the mode coupling temperature. However,
579: the temperature dependence of the maximum value of $\chi_4(t)$ is 
580: weaker than that of the approximate estimate. Moreover, the time
581: dependence of $\chi_4(t)$ differs from that of the estimate even
582: near the mode coupling temperature. 
583: 
584: We should emphasize that all our computer simulation results were obtained
585: using Brownian dynamics (and, therefore, NVT ensemble). It is possible
586: that results obtained from Newtonian dynamics computer simulations 
587: would agree better (or worse) with theoretical predictions. 
588: 
589: 
590: 
591: \section*{Acknowledgments}
592: We gratefully acknowledge the support of NSF Grant No.~CHE 0517709.
593: 
594: \begin{thebibliography}{99}
595: \bibitem{DebStil} See, \textit{e.g.}, P.G. Debenedettin and F.H.Stillinger,
596: Nature \textbf{410}, 259 (2001).
597: \bibitem{BB} J.-P. Bouchaud and G. Biroli, Phys. Rev. B \textbf{72}, 
598: 064204 (2005).
599: \bibitem{Toninelli4p} C. Toninelli, M. Wyart, L. Berthier, G. Biroli, and 
600: J.-P. Bouchaud, Phys. Rev. \textbf{71}, 041505 (2005).
601: \bibitem{BerthierScience} L. Berthier, G. Biroli, J.-P. Bouchaud,
602: L. Cipelletti, D. El Masri, D. L'Hote, F. Ladieu, and M.  Pierno,
603: Science \textbf{310}, 1797 (2005).
604: \bibitem{KobAndersen} W. Kob and H.C. Andersen, Phys. Rev. Lett. 
605: \textbf{73}, 1376 (1994); Phys. Rev. E \textbf{51}, 4626 (1995);
606: Phys. Rev. E \textbf{52}, 4134 (1995).
607: \bibitem{remark} The alternative used in other stochastic dynamics
608: simulations is to explicitly modify the noise in such a way that the
609: center of mass does not move (W. Kob, private communication).
610: \bibitem{Sharon} S. Franz, C. Donati, G. Parisi, and S.C. Glotzer,
611: Philos. Mag. B \textbf{79}, 1827 (1999); C. Donati, S. Franz, S.C. Glotzer,
612: and G. Parisi, J. Non-Cryst. Solids \textbf{307}, 215 (2002); S.C. Glotzer,
613: V.N. Novikov, and T.B. Schr\o der, J. Chem. Phys. \textbf{112}, 509 (2000).
614: \bibitem{Chandler} After this paper had been written a cond-mat report was
615: posted (D. Chandler, J.P. Garrahan, R.L. Jack, L. Maibaum, A.C. Pan, 
616: cond-mat/0605084) which discusses the dependence of the four-point 
617: susceptibility $\chi_{\vec{k}}(t)$ on the magnitude of the wave vector $\vec{k}$.
618: \bibitem{Goetze} W. G\"otze, in \textit{Liquids, Freezing and Glass
619: Transition}, J.P. Hansen, D. Levesque, and J. Zinn-Justin, eds.
620: (North-Holland, Amsterdam, 1991).
621: \bibitem{SL} G. Szamel and H. L\"owen, Phys. Rev. A \textbf{44}, 8215 (1991).
622: \bibitem{BBEPL} G. Biroli and J.-P. Bouchaud, Europhys. Lett. \textbf{67},
623: 21 (2004).
624: \bibitem{BBprivinfo} L. Berthier and G. Biroli, private communication.
625: \bibitem{FS} E. Flenner and G. Szamel, Phys. Rev. E \textbf{72}, 031508 (2005).
626: \bibitem{bremark} We obtained $b^{th}=0.62$ by fitting the asymptotic
627: functional form predicted by the mode coupling theory to the solution
628: of the full mode coupling equations \cite{FS}. This value is very close
629: to $b^{th} = 0.63$ quoted in \cite{Toninelli4p}. 
630: \bibitem{fkIm1} J.P. Garrahan and D. Chandler, Phys. Rev. Lett. \textbf{89},
631: 035704 (2002); Proc. Natl. Acad. Sci. U.S.A. \textbf{100}, 9710 (2003).
632: \bibitem{fkIm2} L. Berthier and J.P. Garrahan, J. Chem. Phys. \textbf{119},
633: 4367 (2003); Phys. Rev. E \textbf{68}, 041201 (2003).
634: 
635: \end{thebibliography}
636: \end{document}
637: 
638: