cond-mat0604546/pre.tex
1: %\documentclass[11pt,emtex]{article}
2: \documentclass[debug,overfull]{epl} %[a4,twocolumn]
3: \usepackage{epsfig}
4: \usepackage{changebar}
5: \usepackage{graphicx}
6: \usepackage{epsfig}
7: \usepackage{subfigure}
8: \usepackage{amsmath}
9: \usepackage{rotating}
10: \usepackage{amsfonts}
11: \usepackage{amssymb} 
12: \newlength {\oldtextheight}
13: \newlength {\oldheadsep}
14: \setlength {\oldtextheight}{\textheight}
15: \setlength {\oldheadsep}{\headsep}
16: \newcommand {\longpage}{\enlargethispage{\baselineskip}}
17: \newcommand {\shortpage}{\enlargethispage{-\baselineskip}}
18: \psfigdriver{dvips}
19: 
20: \begin{document}
21: \title{Aging  and intermittency  in a    p-spin model of a glass} 
22: \shorttitle{Aging of p-spin model} 
23: \author{Paolo Sibani}
24: \institute{Institut for Fysik og Kemi, SDU, DK5230 Odense M, Denmark}
25: \pacs{65.60.+a}{Thermal properties of amorphous solids and glasses} 
26: \pacs{05.40.-a}{Fluctuation phenomena, random processes, noise, and Brownian 
27: motion}
28: \pacs{61.43.Fs }{Glasses}
29: \date{\today}
30: \maketitle
31: \begin{abstract} 
32: We numerically analyze the statistics of the heat flow between 
33:  an aging system and its thermal bath, following a method 
34:   proposed and tested for a spin-glass
35: model  in a recent Letter (P. Sibani and H.J. Jensen, Europhys. Lett.69, 563 (2005)).
36: The present  system, which    lacks quenched randomness,   consists of Ising  
37: spins  located  on a cubic lattice,  with each plaquette
38: contributing  to the total energy  the  product of 
39: the four spins located at its corners. 
40: Similarly to our previous findings, 
41:   energy leaves the system in rare but large,  
42: so called  intermittent, bursts 
43: which are embedded in reversible and equilibrium-like   
44: fluctuations of zero average. 
45: The intermittent bursts, or  quakes,   dissipate 
46: the excess energy trapped in the
47: initial state at a rate which falls off with the inverse of the
48: age.  This strongly heterogeneous dynamical picture is  explained using  
49: the idea that quakes are triggered by energy fluctuations of record
50: size,  which occur  independently  within a number of thermalized domains.
51: From the temperature dependence of the width of the
52: reversible heat fluctuations we surmise  that  these domains have
53: an exponential density of states. Finally, we show 
54: that  the   heat flow consists of 
55:   a  temperature  independent term  and  
56: a term with an Arrhenius temperature dependence.      
57: Microscopic  dynamical and structural information can thus 
58: be extracted from numerical  intermittency data. This type of  analysis 
59: seems now within the reach  
60: of  time resolved micro-calorimetry techniques. 
61: \end{abstract} 
62: 
63: \section{Introduction} 
64: Following the  change of an external parameter, 
65: e.g. \ a temperature quench,  complex  
66: systems  unable to  re-equilibrate within 
67: observational  time scales undergo a  so-called  aging process.
68: During aging,  physical quantities slowly change as a function of the
69: time elapsed since the quench, a time conventionally denoted by the
70: term   `age'.  
71: In   \emph{mesoscopic}  sized   systems~\cite{Bissig03,Buisson03}
72:  aging   manifests itself through   a sequence of  large, so-called  
73: intermittent,   configurational re-arrangements. These   
74: events  generate     non-Gaussian tails 
75:  in the probability density function (PDF) of  configurational  probes 
76:  such as  colloidal particle displacement~\cite{Kegel00,Weeks00} and
77:  correlation~\cite{Cipelletti03a} 
78:  or voltage noise fluctuations in glasses~\cite{Buisson03a}.
79: The   intermittent signal   contains valuable dynamical information
80: which is  straightforwardly collected   in numerical  simulations
81:  and  which  now  seems  within  reach  of  highly sensitive
82:   calorimetric techniques~\cite{Fon05}. 
83: 
84: The present work  confirms the expected wider applicability 
85: of the method previously introduced in a Letter~\cite{Sibani05},
86: and extends the analysis to include the temperature dependence of the 
87: rate of energy flow. Secondly, 
88: it features  a more extensive  discussion of  
89: record dynamics~\cite{Sibani03} as a paradigm 
90: for non-equilibrium  aging and intermittency. 
91: 
92: We consider the   plaquette model~\cite{Lipowski00,Swift00}, which
93: belongs to a class of Ising   models   
94: with multiple spin interactions~\cite{Lipowski00,Swift00,Garrahan00,Davison01,Mezard02},
95: known to   possess central  
96: features of glassiness, e.g. a metastable super-cooled phase 
97: as well as  an aging phase~\cite{Lipowski00,Swift00}. 
98: 
99: In the sequel, a  short summary is given of the relevant 
100: properties of the p-spin model and of the
101: Waiting Time Algorithm~\cite{Dall01} (WTM) used in the simulations.
102: As a check of this   algorithm, we   calculate the known 
103:  equilibrium and metastability
104:  properties  of the model before   turning to 
105:  the    heat flow analysis. We show that  irreversible 
106:    intermittent events, so called \emph{quakes},  
107:  can be disentangled  from  the reversible and
108: equilibrium-like   fluctuations of zero average which
109: occur around  metastable configurations.  The age  and temperature dependences 
110: of the statistics of both types of events are described in detail.  
111: In the final section, our   findings are summarized and placed in a broader context. 
112: \section{The  model}
113: \label{model} 
114: We consider a set of   $N=16^3$   Ising spins, $\sigma_i = \pm 1$, placed   on a cubic
115:  lattice with periodic boundary conditions. The  spins 
116:   interact via the  `plaquette' Hamiltonian
117: \begin{equation}
118: {\cal H} = -\sum_{{\cal P}_{ijkl}} \sigma_i\sigma_j \sigma_k \sigma_l,
119: \label{hamilt}
120: \end{equation}
121: where  the    sum  runs   over  all the  elementary  plaquettes of the lattice,  
122: each  contributing the  product of the  four  spins   located at its  corners. 
123: 
124: By inspection, a `crystalline' ground 
125: state of the model is   the fully polarized state  with energy 
126: per spin $\epsilon_G= -3$. Other ground states can be 
127: obtained by  inverting  the   spins  in  any  plane orthogonal 
128: to  the $x$, $y$ or $z$ axes. These  symmetries lead    to a    
129: degeneracy $\approx 2^{3L}$, where $L$ is the linear lattice size. 
130: 
131: The simulations of Lipowski and Johnston~\cite{Lipowski00} show  a disordered 
132: equilibrium state above  $T \approx 3.6$, and, between $T=3.6$ 
133: and $T \approx 3.4$, 
134: the presence of a metastable  `super-cooled liquid'  phase. According to  
135: Swift et al.~\cite{Swift00}    the  energy autocorrelation function  decays in this regime  as a stretched
136: exponential, with no age dependence. The duration of the 
137: metastable plateau  diverges  as $T \downarrow 3.4$. Below 
138: this   temperature  aging sets in.  
139: \section{The WTM algorithm}
140: \label{WTM}
141: Monte Carlo algorithms   reproduce  central  
142:  features of aging  dynamics, 
143: and their use to study aging systems 
144: is  thus empirically, if not rigorously,  justified.    
145: As  in ref.~\cite{Sibani05}, the present simulations rely on  the  WTM~\cite{Dall01},  
146:  a rejectionless (event driven)  MC algorithm~\cite{Newman99d}, 
147: which  obeys detailed balance. Unlike the Metropolis algorithm, the WTM  relies on  
148:  an intrinsic or `global'  time variable, which is 
149:  akin to the clock time of  a physical system, modulo
150:  an overall scale factor of   little significance 
151:  for aging behavior, where most   quantities of interest depend on
152:   time ratios. 
153:   
154:    Let $b_i$ denote the energy change associated with the
155:    flipping of spin $i$ with fixed neighbors.  For a given initial configuration, 
156:    random waiting times $w_i$ are  drawn from an  exponential distribution  with average 
157:  $\tau_i =\exp(b_i/2T)$. 
158: A  WTM move identifies   the shortest waiting time,
159:  adds it to the  `global'  time variable,
160: and  flips the corresponding spin. 
161: Spins whose $b$ values  are affected  by the current  
162: move have their waiting times  recalculated.
163: The others do not, since the statistical properties of their 
164:  (memoryless) waiting process would be unchanged by a renewed evaluation.     
165: In    unstable  configurations,   where negative $b_i$ values  are present, 
166:  the WTM generates     short  waiting times, leading to    
167:  the nucleation of flip avalanches, which can be followed in detail.  
168: In  Metropolis simulations, 
169:   the spins  considered for update are 
170: randomly selected, and  events 
171: below the shortest  time unit of  one   MC sweep cannot be resolved. 
172: Thus,  while all MC algorithms with  detailed balance converge, under weak conditions,    
173:  to the   same  (Boltzmann) equilibrium state, they   may   do so 
174:  in   different ways,  see e.g.\ ref.~\cite{Gotcheva05}. 
175:  For  our model, a  comparison of the time dependence of the
176: average energy   to   results previously obtained with the Metropolis algorithm 
177: only shows   minor differences.  
178: \begin{figure}[t]
179: \centering
180: \mbox{
181: \subfigure[Energy relaxation]{\epsfig{figure=aging_energies.eps,width=.47\textwidth}} \quad 
182: \subfigure[Heat flow PDF]{\epsfig{figure=drift.eps,width=.47\textwidth}}} \quad 
183: \caption{{\em (a)}:  The energy per spin is plotted vs. the system age   for several 
184: temperatures, showing, from top to bottom, the presence of  thermal equilibrium,  metastability  and  aging. 
185: Data points are shown with symbols, the dotted lines are guides to the eye. 
186: In each panel, the temperatures  corresponding to  the five curves shown are, 
187: listed  from  top to the bottom:  $T=4.75,4.25,3.75, 3.57, 3.50$ (upper panel), $T=3.475,3.45,3.40,3.25,3.00$ 
188: (middle panel) and $T=1.25,1.50,1.75,2.25,2.50$ (lower panel). 
189: {\em (b)}: the  PDF of the heat exchanged  over an interval $\delta t$ 
190: at $T=1$  is  sampled in the interval  $[t_w, 3 t_w]$. The three  data sets shown have  $t_w=300, 1000$ and $5000$
191: respectively, with a  fixed
192: ratio   $\delta t/t_w =0.01$.  The  insert shows similarly collected  data for $t_w=5000$
193: and $T=3.475$. The line is   a fit  of the data to a function which equals 
194:  an exponential for sufficiently negative $H$ value and a Gaussian for the rest of the range.
195:  }
196: \label{aging_of_energy}
197: \end{figure}    
198: % made with program  /home/paolo/bose/WTM/4SPINS/plot_en.m 
199: % and with  /home/paolo/bose/WTM/4SPINS/RUN_DIR/PDFscalefit.m
200: \section{Results}
201: \label{meanenergy}  
202:  Fig.~\ref{aging_of_energy}\emph{(a)} demonstrates 
203:  the three  known relaxation regimes of  this  model:
204:  For approximately   $T\geq 3.500$,  (top section), 
205: the energy  immediately  equilibrates at a high  value.
206:  (The  fast initial equilibration stage is omitted). 
207: In the interval  $3.00< T< 3.475$ (middle section),  the 
208: final  equilibrium  near 
209: the  ground state energy is reached after  dwelling in  
210:  a metastable  plateau. The  duration of the plateau   grows with   the
211: temperature, reaching its maximum near  $T=3.475$, i.e. near 
212: the temperature where metastability  disappears. The equilibrium 
213: mean energy correspondingly undergoes an abrupt change between $T=3.475$ and $T=3.5$,
214: in a way  reminiscent of a first order transition\cite{Lipowski00}. 
215: At the low temperature end of the metastable regime, the plateau gradually
216: softens, giving rise to a nearly  logarithmic decay of the energy. 
217: Approximately below $T = 2.50$ (bottom section) the energy value
218: reached at any stage   decreases with increasing temperature, an   extreme  
219: non-equilibrium situation typical for  the aging regime.  In summary, the WTM 
220: and the Metropolis algorithm used in other  works agree on the thermodynamical 
221: (average)  behavior, with the hardly significant  exception of  the  duration of the 
222:   metastable plateau. The latter  attains its maximum value at  $T=3.40$ under the Metropolis
223:    update~\cite{Swift00}, and at a slightly higher temperature in the WTM case. 
224:  
225: To investigate  the  energy flow  statistics we  choose a   
226: time interval $\delta t$ and partition the observation
227: time $t$ into $[t/\delta t]$ intervals of equal duration. The heat   $H$ exchanged
228: over $\delta t$  arises as    the  difference between the energies of the two 
229: configurations at the endpoints of the interval,  
230: \begin{equation}
231: H_n = E(t_w +(n+1)\delta t) - E(t_w +n\delta t), \quad n=1,2, \ldots [t/\delta t]
232: \label{H_definition}
233: \end{equation} 
234: From ref.~\cite{Sibani05}, we expect   the rate of energy flow
235: to fall off as $1/t_w$ (pure aging),  leading to  a     
236:  $\delta t/t_w$ scaling of the PDF. For the present model, 
237: the collapse of three PDF's shown  in the main 
238:  plot of Fig.~\ref{aging_of_energy}, panel \emph{(b)} confirms 
239: this expectation.
240: The PDF's    are  collected at   $T=1.0$,  
241: i.e.  deep in the aging regime, during time  intervals
242:   $[t_w,3 t_w]$, with $t_w=300, 1000$ and $5000$, 
243: (squares, diamonds and polygons, respectively), keeping  the  ratio  
244:  $\delta t/t_w$ constant and equal to $0.01$.
245:  The insert shows data taken at   $T=3.475$,  i.e.  at the upper edge of the 
246: metastable region. The calculations are repeated for $200$ 
247: independent trajectories, giving a total of  $20000$ data points
248: for each PDF. 
249:  
250: The full line is a least square fit of the data
251: to a continuous function comprising  an  exponential tail, $f_<(H)\propto e^{H/q}$ for $H<d<0$, and a Gaussian 
252: part centered at zero, $f_> \propto e^{-H^2/2{\sigma^2}_{rev}} $ for $H>d$. The  position,   $d$, 
253: of the joining point  is  itself a fitting parameter. As the overall scale of the PDF 
254:  is arbitrary,   the    three parameters  of physical significance are  
255:  $q, \sigma_{rev}$ and $d$.   
256: Figure~\ref{aging_of_energy} \emph{(b)} already   
257: indicates that  the  exponential  intermittent tail carries the  
258: bulk of  the energy flow. This also   applies 
259:  to the    $T=3.475$  data shown in the insert, even though   
260: at this   temperature, which is outside   the aging regime, 
261: the intermittent events are  swamped  by  the  reversible Gaussian fluctuations. 
262: \begin{figure}[t]
263: \centering
264: \mbox{
265: \subfigure[Age dependence of intermittency]{\epsfig{figure=heat_PDFsT1.eps,width=.45\textwidth}} \quad 
266: \subfigure[Rate of energy flow]{\epsfig{figure=rate_vs_age.eps,width=.45\textwidth}}} \quad 
267: \caption{{\em (a)}: The  PDF of the heat exchanged 
268: between system and thermal bath over a time $\delta t=5$. Negative values correspond to an
269: energy  outflow.
270: The data are based on $200$ independent 
271: runs,    taken at temperature $T=1$ in the intervals  $[t_w,t_w+500]$ for
272:  $t_w=1000$ (diamonds),  $t_w=2000$ (squares)  $t_w=4000$ (polygons) 
273: $t_w=8000$  (hexagons) and $t_w=16000$ (circles). 
274: {\em (b)}: The average rate of flow of the energy is plotted versus the age
275: for the five    temperatures shown. 
276: The full line has the form $y = c t_w^{-1}$, with the proportionality constant $c$ estimated as
277: the mean of $t_w r_E$. 
278:  }
279: \label{heat_flow}
280: \end{figure} 
281: Further  properties of the heat flow at $T=1$    are illustrated by panel \emph{(a)} of Fig. \ref{heat_flow}.
282:  The five  PDF's shown  are  semilogarithmic plots
283: of the distribution  of $80000$ values of   $H$ taken  with  $\delta t=5$.
284:   The values are   sampled from  $800$ independent trajectories,  
285: each stretching over  an `observation' interval of duration $t=500$.   
286:     
287:  For sufficiently negative $H$ values, all the  PDFs   have an  exponential  intermittent tail, which turns into 
288:  a bell shaped,  nearly Gaussian distribution as $H$ increases toward and beyond zero.
289:  The strongest intermittency  observed in these data (top curve) is for
290:   $t_w=10^3$. Successively  doubling   $t_w$, up to   $t_w=1.6 \times 10^4$,  
291:  leads to  PDF's  approaching   a  Gaussian shape. This expresses the well known 
292:  fact that aging systems appear equilibrated, when  observed over a time shorter 
293:  than  their  age. 
294:  Importantly, the Gaussian part of the PDF has no conspicuous age dependence. 
295:  
296:  The full lines are, as mentioned,  obtained as least square fits of the PDF data
297:  to a continuous, piecewise differentiable curve. 
298:  The parameter $\sigma^2_{rev}$   estimates the  variance of the reversible fluctuations 
299:  and should not be confused with the full variance of $H$, which  is  mainly determined  by the  
300:  tail of the PDF. 
301: 
302: The  instantaneous  rate of energy flow, a  constant 
303: for time homogeneous dynamics, 
304: is  here inversely proportional to the age. 
305: Ideally, this rate is obtained  as the ensemble  average  of the ratio 
306:  $\frac{H(\delta t, t_w)}{\delta t}$, i.e., 
307:  \begin{equation}
308:  r_E(t_w) \stackrel{{\rm def}}{=}  \lim_{\delta t \rightarrow 0}\frac{\langle H(\delta t, t_w)\rangle}{\delta t} .
309:  \end{equation} 
310: In practice   we (mainly) use  an ensemble of  $1000$ 
311: independent  trajectories,
312: and  perform in  addition  a time average over a 
313: finite  observation interval $t$.  
314: We use   $t=30$ for $t_w \le 2500$,  $t=100$ for $t_w \le 40000$ and
315: $t=500$ for $t_w > 40000$. By varying  $\delta t$   over the  interval 
316:  $[0.1,23]$ we ascertain that  $\langle H \rangle$ 
317:  is  linearly dependent on $\delta t$. Finally, we estimate 
318:  $r_E$ as the arithmetic average of   $H(\delta t, t_w)/\delta t$,
319: over the approx. $5500$ data points available at each temperature.  
320: Panel \emph{(b)} of Fig. \ref{heat_flow} shows
321: in a log-log plot the negative  energy flow rate 
322: versus the system age,  for five selected temperatures. 
323: To avoid   clutter, the  data sets belonging to  each $T$ value 
324:  are artificially 
325: shifted in the vertical direction by  dividing, in order of increasing $T$,
326:  by  $1,3,9,27$ and $81$.  
327:  The    full lines of  Fig.\ref{heat_flow} are obtained by 
328:  fitting the data to the  form $r_E = c/t_w$, with  the chosen value of $c$ 
329:  minimizing   the RMS distance to the data.   
330:    
331:  
332: According to record dynamics~\cite{Sibani03,Sibani05},  the rate of quakes,
333: $r_q$ obeys 
334: \begin{equation}
335: r_q(t_w) = \frac{\alpha(N)}{t_w}, 
336: \end{equation}
337: where   $\alpha$ represents 
338: the   number of independent domains in the system, 
339: which is linearly dependent on the system size $N$, but independent 
340: of the temperature. The rate of energy flow  depends on the
341: rate of quakes and on the  amount of energy given off in each quake. 
342: The latter could in principle depend on both temperature and age. 
343: However, since     panel \emph{(b)} of Fig. \ref{heat_flow} shows that    
344:  $r_E \propto 1/t_w$ to a good approximation, we  conclude that 
345: $r_E(t_w) = r_q(t_w) e(T)$,
346: where the    average  amount $e(T)$
347: of energy exchanged  in a single  quake is given by 
348: \begin{equation}
349: e(T) = \frac{r_E \cdot t_w}{\alpha},
350: \label{energyperquake} 
351: \end{equation} 
352: which  only depends on $T$.
353: \begin{figure}[t]
354: \centering
355: \mbox{
356: \subfigure[Reversible fluctuations]{\epsfig{figure=sigma_rev_vs_T.eps,width=.45\textwidth}} \quad 
357: \subfigure[Activated behavior of $r_E$]{\epsfig{figure=rate_vs_T.eps,width=.47\textwidth}}} \quad 
358: \caption{{\em (a)}: The standard deviation $\sigma_{rev}$ of the reversible energy fluctuations
359: is extracted from the Gaussian fit of the central part of the PDF data, and plotted 
360: versus the temperature. The dotted line is a fit to the  theoretical prediction
361: discussed in the main text. The data are based on $800$ independent 
362: runs,    taken  in the interval   $[1000,1500]$. 
363: {\em (b)}: The average rate of flow, shifted as indicated,  is 
364: plotted  logarithmically versus the inverse  temperature. The straight 
365: line part of the data corresponds to  Arrhenius activated behavior.  
366: These high precision data are based on $80000$ independent 
367: runs,  all  taken  in the interval   $[1000,1500]$.  
368:  }
369: \label{heat_flow_vs_T}
370: \end{figure} 
371: %figures data files  in  /home/paolo/bose/WTM/4SPINS/RUN_DIR/t1000-1500
372: %scripts in the directory: PDF fit.m makes panel a and  ratefit.m makes panel b.
373:  
374: Figure~\ref{heat_flow_vs_T} covers 
375: the temperature dependence of  the   heat transport PDF 
376: in the age interval $[1000-1500]$. Specifically, panel \emph{(a)}
377: covers the $T$ dependence of the Gaussian part of the PDF, 
378: as described by the  corresponding standard deviation
379: $\sigma_{rev}$, while panel \emph{(b)} covers the    $T$ dependence
380: of the intermittent tail, as described by the rate $r_E$.
381: 
382:  In panel \emph{(a)} the dots  are the standard deviation $\sigma_{rev}$  of the 
383:  reversible heat exchange fluctuations
384:  obtained from the Gaussian part of the PDF. 
385: The line is the  theoretical prediction 
386: obtained under   two  model assumptions:  
387: \emph{(i)}  reversible energy fluctuations are treated as  thermal equilibrium fluctuations,
388: which  \emph{(ii)}   occur independently within the previously introduced
389: number $\alpha$ of thermalized 
390: domains. These are   characterized by an exponential local density of states,
391: ${\cal D}(e) \propto \exp(e/\epsilon)$. These assumptions lead, via standard
392: thermal equilibrium calculations, to the expression  
393: \begin{equation}
394: \sigma_{rev} = \surd (2\alpha) \frac{\epsilon T}{\epsilon -T}; \quad T < \epsilon .
395: \label{reversible}
396: \end{equation}  
397:  The parameter  $\epsilon$   marks   
398:  the upper   limit of  the temperature range 
399:  where   the  exponential density  of states is physically 
400:  relevant (the attractor  is thermally   unstable and 
401:  physically irrelevant  for $T \ge  \epsilon$). 
402: The pre-factor $\surd 2$ arises since $H$ is the difference of two
403: energy values, taken $\delta t$ apart.  These values   are 
404: statistically independent, and the variance is correspondingly additive, 
405:  provided that   $\delta t$ is sufficiently large, as is the case   here. 
406:   Finally the pre-factor $\surd \alpha $ expresses  the linear dependence 
407: of the variance on the number $\alpha $ of    (supposedly) equivalent 
408: and independent domains.
409: The parameter values minimizing the RMS distance between data and Eq.~\ref{reversible}
410: are  $\alpha=25$  and $\epsilon=3.8$. The    domain 
411: volume  corresponding to $\alpha=25$ is  $v = 16^3/25 \approx  162$ spins, whence  the linear  
412: size  is,  for a roughly cubic domain,  close to $5$.  
413: The $\epsilon$ value is a theoretical  upper bound to the  
414:  thermal stability range, and safely exceeds  $T=2.5$, the empirical upper temperature limit 
415:  to the aging regime.
416:  Finally, using Eq.~\ref{energyperquake}, the value $\alpha \approx 25$ and the 
417:  available values of $r_E$,   the average energy 
418: exchanged in a quake, $e(T)$, turns out to    vary from $-8.8$ to $-15$ when $T$ varies from $0.65$  to $2.35$.
419: In all cases  $-e(T)>>T$, meaning that quakes are  
420: thermally irreversible on the time scale at which the occur, a necessary 
421: condition for record dynamics to apply~\cite{Sibani03,Sibani05,Sibani06}. 
422:  In summary, the  procedure just described   seems 
423: preferable to  e.g.  a parabolic fit, which has  nearly the same  quality,  but 
424:  requires one more  parameter, without  providing  a  physical interpretation. 
425:  
426: Panel \emph{(b)} of  Fig.~\ref{heat_flow_vs_T} shows the temperature dependence 
427: of   the  energy flow rate $r_E$ in the age  interval $[1000-1500]$.  
428: For fixed $t_w$  
429: and  $T\rightarrow 0$, $r_E$  approaches the    value  $\approx -0.172$. 
430: The logarithm of the deviation   $-r_E - 0.172$ 
431:  is plotted in an Arrhenius fashion versus the reciprocal temperature $1/T$ 
432:  (circles). 
433:  The full line  in  panel \emph{(b)} is  
434:  an Arrhenius fit  within the restricted range of $T$
435: obtained by excluding the lowest temperatures, as indicated. The fit yields
436: an  energy barrier of characteristic size  $b\approx 3$. The modest  activated contribution  shows that 
437:   once initiated, a quake can  cover  more ground  
438: the  higher temperature,  of course within  the boundary  of  the aging regime. 
439:  The activated contribution   explains   why, 
440:  within   a fixed amount of time,  aging reaches 
441: lower energies at higher temperatures (see e.g. Fig. \ref{aging_of_energy}).
442: In summary, the main  contribution to the energy release is $T$ independent, but
443:  the process is  nevertheless  enhanced  by 
444:  an increased ability to climb   small energy barriers.  
445:  \section{Discussion} 
446: The  heat exchange  in the plaquette model 
447: features  both equilibrium-like fluctuations, the Gaussian 
448: part of the PDF,  and quakes, the exponential tail.  
449: Crucially,  as the Gaussian  signal has  zero average, 
450: the energy outflow  is  exclusively due 
451: to the quakes, which we interpret~\cite{Sibani05,Sibani03}  as a 
452: change from one metastable configuration to another,
453: having  considerably lower energy.  
454: The interpretation is explicitely verified in  a numerical study of 
455:   the random orthogonal model~\cite{Crisanti04}, where energy changes  from one inherent 
456: state (IS)~\cite{Stillinger83}  to another
457: care monitored.  The  processes 
458: corresponding to the Gaussian and exponential parts of the PDF are 
459: there  called stimulated and spontaneous, respectively. 
460: 
461: As quakes are  localized events in time and space, 
462: record dynamics  treats   aging   as a strongly 
463: heterogeneous dynamical process. 
464: However, as emphasized by 
465: recent work~\cite{Mayer04,Appignanesi04,Pan05, Merolle05},
466: exponential  tails in the PDF's of relevant dynamical  variables 
467: also appear   outside the aging
468: regime (see also the insert of Fig.~\ref{aging_of_energy}), i.e. 
469: heterogeneity is not    exclusively  associated to aging.
470: 
471: Our  analysis  employs a record dynamics scenario~\cite{Sibani03,Sibani05}, 
472: where quakes are assumed to be  irreversible, an  assumption appropriate
473:  far from equilibrium, and explicitely verified in the present model.  
474:  The  scenario has been   used to design a   numerical exploration 
475:  technique~\cite{Dall03,Boettcher05},  melds, as we have seen,  
476:  real space (the mumber $\alpha$ of domains) 
477:  and `landscape' (the density of states of local attractors)  properties of glassy dynamics,  
478:  and has a number of   predictions 
479:  which can be tested numerically and/or  experimentally.  
480: The  physical adjustments  induced  
481: by each quake affect all  measurements. In principle, these aspects require a  
482: separate modeling effort, which in our case is limited to 
483: the  temperature dependence of the average energy $e(T)$ exchanged in a single
484: quake.  
485: More generally, by  assuming   that the induced changes are stochastic and 
486:  mutually   independent, and  that no rearrangement can occur   
487:  without    a quake,  approximate   closed-form expressions 
488:  for the rate of change of any  quantity of physical interest
489:  can be obtained~\cite{Sibani06,Sibani06a}. 
490: E.g.,  the configuration autocorrelation function
491: and the linear response between times $t_w$ and $t_w+t$ are predicted to scale with the  
492: logarithm of the ratio $t/t_w$, as indeed  observed~\cite{Sibani06,Castillo03,Chamon04}
493: in a number of cases. 
494:  
495: In the p-spin   model, the average energy decays in logarithmic fashion,  
496: a behavior also seen in      spin-glasses~\cite{Sibani05} 
497: and    Lennard-Jones glasses~\cite{Angelani03}. In the latter case,
498: the  rate of hopping between different minima basins 
499: (arguably similar to our quakes, see also ref.\cite{Crisanti04} ) 
500:  decreases as the reciprocal of the age. 
501: Broadly similar  behavior is observed, experimentally
502: or numerically,  in a large number 
503: of other complex systems~\cite{Lee05,Hannemann05,Oliveira05,Balankin05}.
504: 
505: 
506:  Record dynamics sees   aging as an    entrenchment
507:  into a hierarchy of metastable attractors with growing degree of stability.    
508: It thus implies  the existence of a hierarchy  
509: of the sort  explicitely built into  `tree' models
510: of glassy relaxation~\cite{Sibani89,Hoffmann90,Sibani91,Joh96,Hoffmann97}, 
511: models which  notably reproduce key features of glassy dynamics. 
512: For  such  models,  an entry in a hitherto unexplored   subtree  
513:  requires    a record-sized energy fluctuation. The converse statement,
514: which is assumed in record dynamics, 
515:   is  only   true for 
516:  a continuously branching tree, i.e. when the energy difference 
517: between  neighboring nodes   tends to zero. In spite of these 
518: mathematical difference, the physical picture behind record
519: dynamics and tree models is  nearly the same. Note  
520: that hierarchies    are specifically associated  
521: to    locally thermalized   and independently relaxing real space 
522: domains. 
523: The fit in Fig.~\ref{heat_flow_vs_T}  simplistically assumes  that,
524: within each of these, the number of configurations available  
525:  increases exponentially with the energy difference from the state of lowest
526:  energy within the domain. The same  property is    built in 
527:   tree models, and  leads~\cite{Grossmann85} to a
528:  thermal instability  when the temperature 
529: approaches from below the characteristic scale of the exponential growth
530: ($\epsilon$ in our notation). 
531: Nearly exponential local density of states have  been found by a variety of numerical 
532: techniques  in a  number of microscopic 
533: models~\cite{Sibani93,Sibani94,Schon98,Klotz98,Schon00,Schon02,Schubert05},
534: providing independent evidence of the physical relevance of the approximation. 
535:  
536: In conclusion,   intermittency data  offer  a unique 
537: window  into  the microscopic  dynamics of aging 
538: systems,  and  open the possibility   
539: of   investigating issues of central 
540: importance in the   statistical physics  of complex systems,
541: e.g.  barriers and attractor basins, using calorimetric experiments.
542:  
543: \section{Acknowledgments} Financial support  from the Danish Natural Sciences Research Council
544: is gratefully acknowledged. The author is  greatly indebted  
545: to Stefan Boettcher for suggesting 
546: this investigation  and for his useful comments,
547: and to Karl Heinz Hoffmann, Henrik J. Jensen and  Christian Sch\"{o}n for discussions.
548: \bibliographystyle{unsrt}
549: \bibliography{SD-meld}
550: \end{document}
551:  
552:   
553:  
554: