1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: % INSTITUTE OF PHYSICS PUBLISHING %
3: % %
4: % `Preparing an article for publication in an Institute of Physics %
5: % Publishing journal using LaTeX' %
6: % %
7: % LaTeX source code `ioplau2e.tex' used to generate `author %
8: % guidelines', the documentation explaining and demonstrating use %
9: % of the Institute of Physics Publishing LaTeX preprint files %
10: % `iopart.cls, iopart12.clo and iopart10.clo'. %
11: % %
12: % `ioplau2e.tex' itself uses LaTeX with `iopart.cls' %
13: % %
14: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
15: \documentclass[10pt]{iopart}
16: \newcommand{\gguide}{{\it Preparing graphics for IOP journals}}
17: %Uncomment next line if AMS fonts required
18: \usepackage{epsfig}
19: \begin{document}
20:
21: \title[A random walk description of the heterogeneous glassy
22: dynamics of attracting colloids]{A random walk description of the
23: heterogeneous glassy dynamics of attracting colloids}
24:
25: \author{Pinaki Chaudhuri$^1$, Yongxiang Gao$^2$, Ludovic Berthier$^1$,
26: Maria Kilfoil$^2$, and Walter Kob$^1$}
27:
28: \address{$^1$ Laboratoire des Collo{\"\i}des, Verres
29: et Nanomat{\'e}riaux, UMR 5587, Universit{\'e} Montpellier II and CNRS,
30: 34095 Montpellier, France}
31:
32: \address{$^2$ Department of Physics,
33: McGill University, Montr\'eal, Canada H3A 2T8}
34:
35: \ead{berthier@lcvn.univ-montp2.fr}
36:
37: \begin{abstract}
38: We study the heterogeneous dynamics of attractive colloidal particles
39: close to the gel transition using confocal microscopy experiments
40: combined with a
41: theoretical statistical analysis.
42: We focus on
43: single particle dynamics and show that the
44: self part of the van Hove distribution function is not the
45: Gaussian expected for a Fickian process,
46: but that it reflects instead the existence,
47: at any given time, of colloids with widely
48: different mobilities. Our confocal microscopy
49: measurements can be described well by a simple analytical model
50: based on a conventional continuous time random walk
51: picture, as already found in several other glassy materials. In particular,
52: the theory successfully accounts for the presence of broad tails in the
53: van Hove distributions that exhibit exponential,
54: rather than Gaussian, decay at large distance.
55: \end{abstract}
56:
57: %Uncomment for PACS numbers title message
58: \pacs{64.70.Pf, 05.20.Jj}
59: % Keywords required only for MST, PB, PMB, PM, JOA, JOB?
60: %\vspace{2pc}
61: %\noindent{\it Keywords}: Article preparation, IOP journals
62: % Uncomment for Submitted to journal title message
63: %\submitto{\JPA}
64: % Comment out if separate title page not required
65: %\maketitle
66:
67: \section{Dynamic heterogeneity in colloidal gels}
68:
69: There are many systems in nature whose dynamics become slow
70: in some part of their phase diagram, because they undergo
71: a transition from a fluid to a disordered solid phase ---
72: like in a sol-gel transition, a glass transition, or a jamming
73: transition. These systems are generically called ``glassy materials'',
74: %We think here of
75: examples of which are simple or polymeric liquids,
76: colloidal particles with soft-core or hard-core interactions, grains, etc.
77: As physicists, we would like to have a microscopic understanding of the slow
78: dynamics of these materials and would like to answer, in particular,
79: an apparently very simple question: How do particles
80: move in a glassy material close to the fluid-solid transition?
81: To answer this question
82: directly, one needs to resolve the dynamics of individual
83: particles. In experiments, this is a particularly hard task
84: for molecular liquids, although some techniques are now
85: available~\cite{single,afm}
86: but becomes much easier in the colloidal and granular worlds, where
87: direct visualization is
88: possible~\cite{kegel,weeks,marty,virgile,laura,durian,gao,dibble}.
89: Of course, resolving
90: single particle dynamics is trivial in computer simulations
91: where, for each particle in the system, the equations of motion
92: are directly integrated.
93:
94: Hence, single particle dynamics have
95: now been well documented, both numerically and experimentally,
96: in a wide variety of materials.
97: A most striking feature emerging from these studies
98: is the existence of dynamic heterogeneity~\cite{ediger}.
99: In terms of single particle trajectories, dynamic
100: heterogeneity implies the
101: existence of relatively broad distributions of mobilities
102: inside the system. It is
103: therefore an important task to suggest a framework
104: to describe and interpret those data, and hopefully understand
105: the physical content carried by single particle displacements.
106:
107: In this work, we study an assembly of
108: moderately attractive colloidal particles (attraction depth
109: $U \approx 3 k_BT$,
110: where $k_BT$ is the thermal energy)
111: that undergo dynamic arrest at an ``intermediate'' volume fraction,
112: $\phi_c \sim 0.44$~\cite{gao}.
113: The system is in fact intermediate between fractal gels
114: made of very strongly attractive particles ($U \gg k_B T$)
115: at very low volume fraction, and hard sphere glasses
116: obtained with no attraction ($U \approx 0$)
117: at a much higher volume fraction, $\phi \approx 0.6$.
118: Although experiments clearly detect the presence
119: of an amorphous phase with arrested dynamics, the nature
120: of the transition towards this ``dense gel'' (or low density glass!)
121: remains unclear~\cite{zac}. The transition seems too far from the so-called
122: ``attractive glass'' obtained at higher volume fraction
123: in colloids with very short-range attraction (sticky particles), so that
124: other phenomena are usually invoked. A popular hypothesis is
125: that gelation is in fact a non-equilibrium phenomenon
126: due to a kinetically arrested phase separation~\cite{zac,dave}.
127: Dynamic heterogeneity in such systems has been analyzed before
128: in just a few systems, both experimentally~\cite{gao,dibble} and
129: numerically~\cite{puertas,pablo}.
130:
131: In this paper, we analyze single particle dynamics on the approach
132: to the glassy phase and show that the
133: self-part of the van Hove distribution function is not the
134: Gaussian expected for a Fickian process,
135: but that it reflects instead the existence,
136: at any given time, of colloids with widely
137: different mobilities: Our system is dynamically heterogeneous.
138: We then show that the simple analytical model proposed
139: in Ref.~\cite{pinaki} to describe data in a variety of systems
140: close to glass and jamming transitions also describes our experimental
141: data in a satisfactory manner.
142:
143: This paper is organized as follows. In Sec.~\ref{exp}
144: we describe the system, experimental techniques,
145: and the results obtained for the van Hove function. In
146: Sec.~\ref{modelsec} we describe the model used to fit the experimental
147: data and discuss the results. We conclude the paper in
148: Sec.~\ref{conclusion}.
149:
150: \section{Measuring single particle dynamics using confocal microscopy}
151: \label{exp}
152:
153: \subsection{Experimental system and techniques}
154:
155: The experimental system under study is a suspension of colloidal
156: particles interacting through a hard-core repulsion and a softer
157: attractive interaction, induced by depletion
158: by adding polymers. Details of the system
159: have been presented in Ref.~\cite{gao}.
160: The dynamics of this system is observed using confocal fluorescence microscopy.
161: The strength of the inter-particle attractive interaction, $U$, is
162: determined by the concentration
163: of polymers in the suspension. We present data for a sample at a
164: moderate
165: interaction strength of $U \approx 2.86 k_B T$.
166: We work at constant temperature $T$, so that our
167: control parameter is the volume fraction of the particles, $\phi$.
168: We find that the system becomes a gel when $\phi$ is increased,
169: with a transition close to $\phi_c \approx 0.442$~\cite{gao}.
170: Measurement of different relevant statistical quantities are carried
171: out at different $\phi < \phi_c$.
172:
173: Our procedure to vary slowly the volume fraction uses
174: particle sedimentation.
175: The relative buoyancy of the colloids is
176: $\Delta \rho = 0.011$~g/cm$^3$, corresponding to a
177: gravitational height of $h = k_B T /(\frac{4}{3}\pi a^3 \Delta \rho g)
178: \approx 40$ particle radii $a$, where $g$ is the acceleration due
179: to gravity. Therefore, the gravitational
180: field is small enough that it induces a very slow
181: densification of the system. The densification is slow enough
182: that microscopic dynamics of the colloids remains
183: controlled by the interplay between attraction
184: and steric hindrance, rather than by sedimentation itself.
185: Moreover, the large asymmetry between polymer coil diffusion
186: time, $\approx 0.3$~s,
187: and particle sedimentation time over one particle, $\approx 260$~s,
188: ensures that polymers are uniformly distributed, maintaining
189: the interaction strength $U$ constant in the course
190: of the experiment.
191:
192: \begin{figure}
193: \begin{center}
194: \psfig{file=snap4.ps,width=9.cm}
195: \vspace*{0.2cm}
196: \caption{Three-dimensional confocal microscopy rendered
197: image of a typical particle
198: configuration at volume fraction $\phi=0.429$.}
199: \label{snap429}
200: \end{center}
201: \end{figure}
202:
203: The colloidal particles are polymethyl-methacrylate (PMMA) spheres
204: of diameter $1.33~\mu$m, sterically stabilized by
205: chemically grafted poly-12-hydroxystearic acid,
206: dyed with the electrically neutral fluorophore 4-chloro-7-nitrobenzo-2
207: oxa-1,3-diazole (NBD), and suspended in a solvent mixture of
208: decahydronaphthalene (decalin), tetrahydronaphthalene (tetralin),
209: and cyclohexyl bromide (CXB)
210: that allows for independent control of the refractive index
211: and buoyancy matching with the particles.
212: Polystyrene polymers
213: (molecular weight $11.4{\times}10^6$~g/mol) are added at $1.177$ mg/ml
214: to induce a depletion attraction at a range estimated by
215: $\Delta = 2 R_g = 0.28a$, where $R_g$ is the polymer radius of
216: gyration.
217:
218: Using confocal microscopy,
219: we collect stacks of images at fixed time intervals ranging
220: from $12$ to $1500$~s at different $\phi$ to access short and
221: long time dynamics during the approach to gelation.
222: From the stacks of images we extract the particle positions of
223: $10^3$ particles in three dimensions and track their
224: positions at better than 10~nm resolution over time.
225: A three-dimensional rendering of a typical
226: particle configuration from a stack of images
227: at $\phi=0.429$ is illustrated in Fig.~\ref{snap429}.
228:
229: \subsection{Non-Gaussian distributions of single particle displacements}
230:
231: In an earlier work~\cite{gao}, we analyzed some structural
232: and dynamical properties of the system for different volume fractions.
233: In particular, we analyzed in some detail
234: the distinct part of the van Hove function, finding
235: dynamic signatures typical of gel systems. We only presented
236: briefly some preliminary results concerning the self-part
237: of the van Hove function. It is the latter that we investigate
238: in more detail here. It is defined by
239: \begin{equation}
240: G_s(x,t) = \frac{1}{N} \sum_{i=1}^N \delta (x - [x_i(t)-x_i(0)] ),
241: \label{gsdef}
242: \end{equation}
243: where $x_i(t)$ denotes the position of particle $i$ at time $t$
244: along one of the horizontal directions.
245: The function $G_s(x,t)$ measures the
246: probability that a given particle has undergone a displacement $x$
247: in a time interval of duration $t$.
248:
249: Once the distribution (\ref{gsdef}) is known, several
250: quantities can be determined. Perhaps the simplest one
251: is the mean squared displacement, $\langle x^2 \rangle$,
252: where the average is taken over the distribution $G_s(x,t)$,
253: which contains quantitative information about the average
254: mobility of the colloidal particles. In particular, its
255: long-time limit yields values for the self-diffusion
256: constant $D_s$ of the particles through $\langle x^2 \rangle \sim
257: 2 D_s t$ for large $t$. Such a measurement, however, tells nothing about
258: the possible presence of dynamic heterogeneity in the system.
259:
260:
261: \begin{figure}
262: \begin{center}
263: \psfig{file=fit1.ps,width=10.cm,angle=-90}
264: \end{center}
265: \caption{\label{fit} The self-part of the van Hove function,
266: Eq.~(\ref{gsdef}),
267: measured using confocal microscopy upon approaching the colloidal
268: gel transition by increasing the volume fraction $\phi$. For each
269: $\phi$ we show the distribution for a time corresponding to average
270: particle displacements being close to the particle radius.
271: %approximately to the peak of the non-Gaussian parameter.
272: Dashed lines represent Gaussian fits to the center of the distribution.
273: The full lines through the data are fits obtained from the
274: model and parameters
275: described in Sec.~\ref{modelsec}, showing very good agreement with the data.}
276: \end{figure}
277:
278:
279: In our earlier study~\cite{gao}, we had measured $\langle x^2 \rangle$;
280: although we observed a slowing down of the dynamics, the heterogeneous
281: nature remains hidden and could only be seen by measuring $G_s(x,t)$.
282: In Fig.~\ref{fit}, we have
283: plotted $G_s(x,t \sim t^\star)$ for four different volume
284: fractions ranging from $0.37$ to $0.44$, where $t^\star$
285: corresponds to the time when $\sqrt{\langle{x^2}\rangle} \approx 0.2a$
286: ($a$ being the particle radius) -
287: which is a meaningful measure of the timescale
288: for structural relaxation~\cite{gao}. We can clearly see that $G_s(x,t)$,
289: at this timescale, is very non-Gaussian and therefore, the
290: particle trajectories do not correspond to Fickian dynamics.
291:
292:
293: It can be easily seen that although, in all four cases,
294: most of the statistical weight of the functions is carried
295: by particles which have barely moved,
296: $x < 0.5 \mu$m, there is a pronounced tail extending to
297: distances that are much larger than what is expected
298: for the Gaussian prediction shown as dashed lines \cite{gao}.
299: The small $x$ behavior, however, is not far from a Gaussian distribution,
300: corresponding to quasi-harmonic vibrations in the cage formed by neighbouring
301: particles, but at large distances the decay is well described
302: by an exponential, rather than a Gaussian, decay.
303: Thus, the single particle motion for our
304: experimental system at timescales corresponding to $t^\star$
305: is strongly non-diffusive. Such a
306: non-Gaussian behaviour has been observed in other
307: glass-forming systems, both in simulations~\cite{kob,stariolo,odagaki}
308: and experiments~\cite{kegel,weeks,marty,virgile,laura}.
309:
310: Such non-Gaussianity is related
311: to the presence of heterogeneity in the dynamics of the
312: particles in the system. Most of the particles simply undergo
313: vibrational motion around their initial position ---
314: this corresponds to the central
315: Gaussian part in $G_s(x,t)$. Additionally,
316: a small fraction of the particles gets
317: the opportunity to explore larger distances during the
318: observational time and contributing to the
319: non-trivial tail of $G_s(x,t)$~\cite{kob}. Moreover, Fig.~\ref{fit}
320: shows that, with increasing volume fraction, the
321: width of the central Gaussian of $G_s(x,t)$ at time $t^\star$
322: decreases while
323: the tail gets more pronounced. This implies that even though the
324: volume available for the quasi-harmonic vibrations decreases with
325: $\phi$, some particles still find pathways to travel large distances
326: and allow the structural relaxation of the system.
327:
328: Several attempts have been previously made to empirically fit the
329: non-Gaussian shape of $G_s(x,t)$ with known functional forms.
330: Weeks {\it et al.}~\cite{weeks} have tried to fit their experimentally
331: measured $G_s(x,t)$ with
332: a stretched exponential function, in order to fit both the broad
333: tails and the narrow center. Attempts have also been made
334: to fit both components
335: of the van Hove function as the sum of two different
336: Gaussian functions~\cite{kegel,gao}.
337: However, neither attempts seem to give
338: satisfactory results since the shape of the distribution changes with time.
339: Basing their analysis on numerical
340: simulations of a Lennard-Jones system, Stariolo and Fabricius~\cite{stariolo}
341: recognized that the tails are probably
342: better fitted with an exponential function in some time window.
343: Using extensive data, it has recently been shown~\cite{pinaki} that the
344: $G_s(x,t)$, for different glass-formers, colloidal hard spheres
345: and granular materials close to jamming are better represented by a
346: superposition of a central Gaussian along with an
347: exponential tail for the large distances, which crosses over,
348: at large times, to a Gaussian form. The model therefore allows one
349: to describe the data at different times without changing
350: the fitting formula in the middle of the game.
351: As we show below, the exponential tail is interpreted as the
352: direct consequence of the occurrence of
353: rare events of particles undergoing large displacements
354: that are statistically distributed.
355:
356: \section{Random walk analysis}
357: \label{modelsec}
358:
359: \subsection{Modeling single particle dynamics}
360:
361: There have been several attempts to map the
362: heterogeneous single particle
363: dynamics of gels and glass-formers
364: to some stochastic process.
365: Closely related to our approach are the ones of
366: Refs.~\cite{odagaki,monthus,epl}.
367: Odagaki and Hiwatari~\cite{odagaki} have studied the
368: dynamics of atoms near the glass transition of simple classical
369: liquids, on the basis of a mesoscopic stochastic-trapping diffusion model,
370: and calculated various dynamical quantities such as the
371: mean squared displacement, non-Gaussian parameter and intermediate
372: scattering functions.
373: Monthus and Bouchaud~\cite{monthus} have looked at various models of
374: independent particles hopping between energy traps which have
375: relaxation functions similar to glass-formers. They also show
376: that diffusion in trap models can be described
377: using the formalism of the continuous time random walk (CTRW)~\cite{mw},
378: widely used in many different areas of physics.
379: Finally, Berthier {\it et al.}~\cite{epl} also proposed to describe
380: the process of self-diffusion in glass-forming materials in terms
381: of a CTRW picture, and they base their analysis on the study
382: of spin facilitated models. In this context, the CTRW picture
383: directly follows quite generically
384: from the spatially heterogeneous nature of the
385: dynamics~\cite{jung1} .
386:
387: \begin{figure}
388: \begin{center}
389: \psfig{file=msd2.ps,width=7cm}
390: \psfig{file=fig3b.ps,width=8.5cm}
391: \end{center}
392: \caption{Temporal evolution of squared displacement from an initial
393: position for (top) particles in a binary Lennard Jones liquid
394: at low temperature, $T=0.435$ (top),
395: and (bottom) for the attractive colloidal particles at $\phi=0.429$.
396: While some particles rattle around their mean position, others perform one
397: or several quasi-instantaneous jumps.
398: Occurrence of jumps occur randomly in time
399: and are random in size.
400: The straight line in the plots
401: corresponds to the mean-squared displacement.}
402: \label{ctrwfig}
403: \end{figure}
404:
405: In fact, a convincing empirical
406: rationale for this type of approaches stems from a
407: visual inspection of particle trajectories in materials
408: with slow dynamics, such as the ones
409: shown in both panels of Fig.~\ref{ctrwfig},
410: which represent examples of particle displacements for a Lennard-Jones
411: supercooled liquid~\cite{lj} and for the present colloidal system.
412: Direct visualization reveals that, when observed on a timescale
413: comparable to $t^\star$,
414: most of the particles simply perform a large number
415: of localized vibrations around their initial position,
416: just as in a disordered solid. However,
417: the particles that contribute to the tail of the van Hove function
418: undergo one or several quasi-instantaneous jumps separating long periods of
419: localized vibrations, as can be seen in Fig.~\ref{ctrwfig}.
420: For both systems, we observe that these jumps occur randomly in
421: time and also have distributed amplitudes. Therefore,
422: a continuous time random walk~\cite{mw} should be a
423: good coarse-grained stochastic model
424: for single particle trajectories of these systems as suggested
425: before~\cite{odagaki,monthus,epl,jung1}.
426:
427: The case of very low density gels is peculiar,
428: since in these systems there is a well-defined network of quasi-immobile
429: particles existing along with the free particles. Therefore,
430: the system can truly be decomposed into two dynamically distinct
431: families, whose properties directly follow from the heterogeneous
432: nature of the stucture of these gels.
433: Indeed, a two-family dynamical model has been shown to fit the
434: van Hove distribution functions obtained in
435: computer simulations of gel systems for a wide
436: window of parameters~\cite{pablo}. However, for denser systems,
437: we have no structural basis to assume that such a distinction
438: can be made, although this has been done in other studies~\cite{langer}.
439: For supercooled liquids, it can even be quantitatively
440: established by simulations~\cite{rob}
441: that dynamic heterogeneity at the particle scale
442: has no such structural origin.
443: Since the present system lies somewhat in between
444: low density gels and dense glasses, it is not obvious
445: {\it a priori} whether
446: we should adopt a glass (one family)
447: or a gel (two families) description.
448: In fact, we will show that the strong hypothesis of a two-family
449: model is not necessary to account for our measurements. Therefore we will
450: model the system as a collection of indistinguishable
451: particles undergoing continuous time random walks and we
452: prove below that such a modeling accounts well for the data presented
453: in Fig.~\ref{fit}. Obviously, a two-family model would also
454: fit our data very well since one can always artificially
455: separate one group of particles into two distinct subgroups,
456: (the reverse is not necessarily true).
457:
458: \subsection{A simplified CTRW model}
459:
460: We now describe the CTRW model, introduced
461: in Ref.~\cite{pinaki}, which will be used
462: to fit the experimental data. We consider particles
463: undergoing a stationary, three-dimensional, isotropic
464: random walk process, as in the
465: original Montroll-Weiss CTRW model~\cite{mw}, and add to the process
466: localized vibrations occuring on a fast timescale in between the jumps.
467: We assume that vibrations are Gaussian and distributed
468: according to
469: $f_{\rm vib}(r) =
470: (2 \pi \ell^2)^{-3/2}
471: \exp(-r^2/2 \ell^2)$, so that $\ell^2$ represents
472: the variance of the size of vibrations.
473: %the root mean size of the vibrations.
474: We also assume that the jump size is distributed according to
475: $f_{\rm jump}(r) = (2 \pi d^2)^{-3/2}
476: \exp(-r^2/2 d^2)$, introducing $d^2$,
477: the variance in the size of the jumps.
478: %the root mean size of the jumps.
479: The last ingredient needed to define the CTRW model is the distribution
480: of times between jumps, called the waiting time distribution~\cite{mw},
481: which we denote as $\phi_2(t)$,
482: for reasons that will become clear in a moment.
483:
484: With these definitions, one can express the van Hove function as~\cite{mw}
485: \begin{equation}
486: G_s(r,t) = \sum_{n=0}^{\infty} p(n,t) f(n,r),
487: \label{sum}
488: \end{equation}
489: where $p(n,t)$ is the probability to make $n$ jumps
490: in a time $t$, and $f(n,r)$ is the probability to move a distance
491: $r$ in $n$ jumps~\cite{mw}. These probabilities involve convolutions
492: and are more easily expressed in the Fourier-Laplace domain,
493: ($r,t$) $\to$ ($q,s$). The sum in Eq.~(\ref{sum}) is geometric and can
494: be performed easily to yield the well-known result~\cite{tunaley}:
495: \begin{equation}
496: G_s(q,s) = f_{\rm vib}(q) \frac{1-\phi_1(s)}{s} +
497: f(q) f_{\rm vib}(q) \frac{\phi_1(s)}{s} \frac{1-\phi_2(s)}{1-\phi_2(s)
498: f(q)} ,
499: \label{model}
500: \end{equation}
501: where we defined $f(q) \equiv f_{\rm vib}(q) f_{\rm jump}(q)$
502: and the distribution $\phi_1(t)$ is related
503: to the waiting time distribution $\phi_2(t)$ through
504: the Feller relation~\cite{tunaley}:
505: \begin{equation}
506: \phi_1 (t)=\frac{\int_t^\infty{dt'}\phi_2 (t')}{
507: \int_0^\infty dt' \phi_2(t') t'}.
508: \label{feller}
509: \end{equation}
510: Physically, $\phi_1(t)$ represents the distribution
511: of the time, $t$, a walker takes to undergo a jump starting
512: from an arbitrary initial condition at time $t=0$.
513: Note that $\phi_1$ becomes equal to $\phi_2$ when the distribution
514: of waiting time is a simple exponential, while it also follows
515: that the moments of $\phi_1 (t)$ are larger than those of
516: $\phi_2 (t)$ if the distributions are broader than
517: exponential~\cite{barkai1, barkai2, jung2}.
518: In a measurement of the van Hove function, $\phi_1(t)$ represents the
519: distribution of the time to the first observed jump,
520: as seen in the first term in the right hand side of Eq.~(\ref{model}),
521: corresponding to $n=0$ in Eq.~(\ref{sum}). The distribution
522: $\phi_2(t)$ quantifies the time between subsequent jumps
523: and contributes to the second term in Eq.~(\ref{model}) which
524: contains the contribution of all the terms with $n>0$ in the sum
525: (\ref{sum}).
526:
527: The importance of the distinction
528: between the first and subsequent jumps
529: in order to derive the
530: correct expression of the van Hove function
531: was emphasized long ago by Tunaley~\cite{tunaley}, and is crucial
532: when the distribution of waiting time becomes broad.
533: As noted by Monthus and Bouchaud~\cite{monthus},
534: and by Barkai and coworkers~\cite{barkai1,barkai2}, this first term
535: is in fact directly responsible of the aging dynamics observed
536: in CTRW characterized by ``fat'' waiting time distributions.
537: In Ref.~\cite{barkai2},
538: Barkai {\it et al.} even provide an example of a system for which
539: the average time to the first jump is infinite, while
540: the average time between jumps is finite: Eq.~(\ref{model})
541: then shows that in that case
542: particles never leave their initial positions.
543: Jung {\it et al.}~\cite{jung1,lutz}
544: refer to the two distributions as ``persistence'' and ``exchange''
545: and relate them to the decoupling
546: phenomena observed in supercooled liquids.
547:
548: To proceed further and use Eq.~(\ref{model}) to fit
549: experimental or numerical data, one needs input about the
550: waiting time distribution. It has been claimed by Odagaki and
551: Hiwatari ~\cite{odagaki}
552: %in the beginning of the 90's
553: that waiting time
554: distributions in a binary mixture of soft spheres
555: becomes fat with power law tails
556: at low temperature, as in the trap models studied
557: of Monthus and Bouchaud~\cite{monthus}.
558: Garrahan and coworkers
559: performed extensive studies of waiting time distributions
560: both in kinetically constrained glass models~\cite{jung1,jung2}
561: and more recently
562: using molecular dynamics simulations~\cite{lutz}. Their results clearly
563: confirm that waiting time distributions in glass-forming
564: systems are not trivial. In particular, they report measurements
565: of various moments of the distributions $\phi_1$ and $\phi_2$
566: and confirm that they evolve differently with temperature~\cite{jung2,lutz},
567: establishing the complex nature of the waiting time distributions for
568: glass-formers.
569:
570: Using these insights, we have suggested~\cite{pinaki} the following
571: simplification to make the use of Eq.~(\ref{model})
572: much more practical. In the absence of
573: definite information on the detailed shape of $\phi_2(t)$,
574: we characterize $\phi_1(t)$ and and $\phi_2(t)$ in Eq.~(\ref{model})
575: by their respective first moments, $t_1$ and $t_2$, and we generally
576: expect that
577: \begin{equation}
578: t_2 \leq t_1.
579: \label{ineq}
580: \end{equation}
581: We assume that
582: the distributions $\phi_1 (t)$ and $\phi_2 (t)$ are
583: exponential, $\phi_1 (t) = t_1^{-1} \exp(-t/t_1)$ and
584: $\phi_2 (t) = t_2^{-1} \exp(-t/t_2)$, and that they are
585: independent from one another. The real link between them
586: in the Feller relation (\ref{feller}) and their complex shapes
587: are now hidden in the inequality (\ref{ineq}).
588:
589: \subsection{The exponential tail}
590:
591: \begin{figure}
592: \begin{center}
593: \vspace*{0.45cm}
594: \psfig{file=fig3.ps,width=7.cm,angle=-90}
595: \end{center}
596: \caption{\label{fig3} Self-part of the van Hove function predicted by the
597: model in Eq.(\ref{model}) with parameters
598: $t_1=3 \times 10^5$~s, $t_2 = 10^4$~s,
599: $\ell = 0.08 {\mu}m$ and $d = 0.284 {\mu}m$ at different times $t$.
600: We show the data on an extended vertical scale
601: to show that the tail is indeed very close to being
602: exponential. Inset: Fitted slope $\lambda$ for different
603: precision measurements, $G_s(r,t){\approx}10^{-5}$ (top) and
604: $G_s(r,t){\approx}10^{-30}$ (bottom). The very slow growth simply reflects
605: crossover towards the long-time Gaussian form of the distribution
606: at fixed $x$.}
607: \end{figure}
608:
609:
610: The first term in Eq.~(\ref{model}), which corresponds to the particles
611: undergoing localized vibrations modulated by the waiting time
612: distribution for first jumps, controls
613: the shape of the central part of the van Hove function $G_s(x,t)$. The
614: second term in Eq.~(\ref{model}) is responsible for
615: the broad tail in $G_s(x,t)$ and stems from particles
616: which have performed one or several jumps
617: after a time $t$.
618: Using parameters relevant for our colloidal system (see below
619: for the details of the fitting procedure), we
620: present on an extended vertical scale, the predictions
621: of Eq.~(\ref{model}) concerning the shape of the van Hove
622: function and its evolution with time in Fig.~\ref{fig3}.
623: The van Hove functions
624: can clearly be described as the superposition of ``mobile''
625: and ``immobile'' particles with broad tails that are well fitted by
626: an exponential decay for large $x$:
627: \begin{equation}
628: G_s(x,t) \sim \exp \left( -\frac{x}{\lambda(t)} \right),
629: \end{equation}
630: which defines a new lengthscale $\lambda(t)$.
631:
632: In fact, a close to exponential decay of the van Hove function
633: is present in the original CTRW model~\cite{mw} when
634: distances outside the realm
635: of central limit theorem are considered.
636: Using a saddle-point calculation, we have proved~\cite{pinaki}
637: analytically that Eq.~(\ref{model}) generically leads
638: to broad distributions that indeed decay exponentially
639: (with logarithmic corrections).
640: Interestingly this expansion can be obtained
641: independently of the actual shape of the distributions,
642: establishing its universality.
643: We have also shown~\cite{pinaki}
644: that these tails simply become enhanced in glassy materials, and
645: are therefore more easily measured using typical experimental
646: accuracy.
647:
648: Using the exact solution from Eq.~(\ref{model}) shown in
649: Fig.~\ref{fig3}, we fit
650: the decay of $G_s(x,t)$ with an exponential
651: function for two measurements of different precisions
652: corresponding to $G_s$ levels of $10^{-5}$ as in typical
653: experiments, and of $10^{-30}$, which is obviously not accessible
654: experimentally. We find that the lengthscale
655: $\lambda(t)$ slowly increases with time,
656: the growth being slower for the most asymptotic measurements.
657: This suggests that if one were to measure $\lambda(t)$ for even
658: lower values of $G_s(x,t)$, $\lambda(t)$ would be almost constant,
659: in agreement with the saddle-point calculation. In fact, most
660: of the time dependence of $\lambda(t)$ observed through fitting
661: is due to the distribution
662: crossing over, at fixed $x$ and increasing $t$, to its long-time
663: Gaussian limit. We conclude therefore that probably the ``growing lengthscale''
664: $\lambda(t)$ does not carry any deep physical information.
665:
666: Finally we remark that, quite often,
667: the quantities ${4\pi}r^2 G_s(r,t)$ or even $P(\log_{10}r,t)
668: \propto r^3 G_s(r,t)$ are measured in simulations~\cite{puertas,kob},
669: % plotted in numerical work
670: and the appearance of a secondary
671: peak in $r$ at low temperature is given a large significance, supposedly
672: signalling the change towards an ``activated'' dynamics
673: with ``hopping'' processes.
674: We would like to inform that within our CTRW model (which is a
675: purely ``hopping'' model), a secondary peak is not necessarily present.
676: Although the functions $r^2 \exp(-r/\lambda)$ and
677: $r^3 \exp(-r/\lambda)$ describing the tails
678: have a maximum
679: at some value of $r$, this peak is sometimes buried below the
680: Gaussian central part
681: of the van Hove function, so that only a shoulder (instead
682: of a secondary maximum) is observed.
683: A peak emerges, for instance, when the ratio between times
684: $t_1$ and $t_2$ is large enough, the precise limiting value
685: depending also on the parameters $d$ and $\ell$.
686: Therefore, we believe that the observation of such peaks
687: is not in general indicative of a deep change in the physical
688: behaviour of the system.
689:
690: \subsection{Fitting the data}
691:
692: \begin{table}
693: \begin{center}
694: \begin{tabular}{|c|c|c|c|c|c|c|}
695: \hline
696: \hline
697: $\phi$ & $\ell$ & $d$ & $t_1^{\rm th}$ &
698: $t_2^{\rm th}$ & $t_1^{\rm exp}$ & $t_2^{\rm exp}$ \\
699: \hline
700: 0.37 & 0.14 & 0.195 & 300 & 70 & 59 & 36 \\
701: 0.403 & 0.10 & 0.251 & 4000 & 800 & 955 & 448 \\
702: 0.429 & 0.08 & 0.279 & 60000 & 5000 & 11050 & 4047 \\
703: 0.440 & 0.06 & 0.284 & 300000 & 10000 & 22300 & 7086 \\
704: \hline
705: \hline
706: \end{tabular}
707: \caption{\label{table} Fitting parameters used to get the fits
708: shown in Fig.~\ref{fit}. Timescales are in seconds,
709: lengthscales in microns. }
710: \end{center}
711: \end{table}
712:
713: We have used the model, given by Eq.~(\ref{model}), with the four fitting
714: parameters $\{d, \ell, t_1, t_2 \}$ described above to fit
715: the van Hove function $G_s(x,t)$ measured
716: in our experimental system. Like in
717: our previous work with different materials showing
718: slow dynamics~\cite{pinaki}, suitable choice of the fitting
719: parameters results in very good fits of the
720: experimental data, as can be recognized from Fig.~\ref{fit}.
721: The fitting parameters we have used are presented in Table~\ref{table}.
722:
723:
724:
725: To confirm that the good agreement obtained from the model
726: is not due to a large number of free parameters that would allow to fit
727: any set of data, we have tried to compare our choice for the waiting times,
728: $t_1$ and $t_2$, with the same quantities being measured directly
729: from the observed trajectories.
730:
731: To do so, we must determine the ``jumps'' from our trajectories.
732: In our experiments, we say that a particle undergoes a jump
733: if the magnitude of its displacement between two successive
734: experimental frames is larger than a threshold, $x_{\rm cut}$.
735: Here we consider displacements only in one dimension, and
736: use $x_{cut} = 0.1~{\mu}$m, which is slightly larger than
737: the typical lengthscale for the vibrations, $x_{\rm cut} >
738: \ell$.
739: Similar to the CTRW model, we measure two distinct timescales
740: associated with the jumps and separately record
741: timescales to the first jump from an arbitrary initial condition,
742: and timescales between jumps. Given that our experimental
743: trajectories have a finite duration, we observe particles
744: which do not jump, meaning that we probably
745: underestimate both timescales.
746: Moreover, it needs to be noted that
747: in the present experiment, only trajectories where at least
748: two jumps have occured are being recorded meaning that
749: $t_1$ is slightly more underestimated than $t_2$ in our measurements.
750: From the statistics of the observed events, we obtain two
751: time distributions, from which we compute the first moments,
752: which we label as $t_1^{\rm exp}$ and $t_2^{\rm exp}$, respectively.
753:
754: \begin{figure}
755: \begin{center}
756: \vspace*{0.6cm}
757: \psfig{file=tau1.eps,width=8.5cm}
758: \end{center}
759: \caption{\label{taudat}
760: The times $t_1$ and $t_2$ obtained directly from experiments ``exp''
761: are compared to the timescales obtained through the fitting procedure ``th''.
762: The $\phi$ dependence of both sets of data is similar, and the
763: inequality $t_1 > t_2$ is strong in both cases, indicative of
764: a broad distribution of waiting times $\phi_2(t)$.}
765: \end{figure}
766:
767: We can then compare the experimental data
768: to the results obtained through the fitting procedure,
769: which we label as $t_1^{\rm th}$ and $t_2^{\rm th}$, as shown
770: in Fig.~\ref{taudat}. We find that the average waiting times,
771: measured directly from
772: experiments and by using the CTRW fitting procedure, show
773: very similar trends when volume fraction is varied.
774: This good agreement gives evidence that our modeling of
775: the dynamics is physically correct, and that our fitting procedure
776: of the van Hove function indeed yields a detailed
777: statistical information on the particle trajectories.
778:
779: However, the numbers for $t_1$ and $t_2$ obtained from the
780: CTRW model are higher than the numbers extracted from
781: experimental measurements by a factor 2 and 10, respectively.
782: There can be several reasons for this mismatch, which might originate
783: from the model or from the experimental determination
784: of waiting times, or from both.
785: The waiting time distributions are perhaps far more complex
786: than the exponential distributions that are used in our model.
787: But, as mentioned above, we have good reasons to believe
788: that waiting times are slightly underestimated in our
789: experimental analysis,
790: $t_1$ more than $t_2$, a trend compatible with Fig.~\ref{taudat}.
791: One could also imagine the presence of back and forth motions,
792: as seen in Fig.~\ref{ctrwfig}, and that would erroneously be
793: counted as jumps, again biasing the experimental waiting times
794: towards small values, in agreement with the results
795: presented in Fig.~\ref{taudat}.
796: Given these possible sources of discrepancy,
797: we conclude that the agreement reported
798: in Fig.~\ref{taudat} is quite satisfactory.
799:
800: \section{Conclusion}
801: \label{conclusion}
802:
803: In this paper, we have analyzed the heterogeneous dynamics
804: of a colloidal system which undergoes dynamical arrest
805: at a volume fraction intermediate between low density
806: gels and dense glasses.
807: We have focused our attention on single particle
808: trajectories and have analyzed in detail the
809: self-part of the van Hove distribution functions. These
810: distributions are strongly non-Gaussian with tails
811: that are broad and decay close to exponentially
812: with distance. We have shown that a simple
813: continuous time random walk analysis proposed
814: in the context of glass and jamming transitions
815: describes the experimental data in a very satisfactory
816: manner, showing that the present experimental system
817: shares deep similarities with other glassy systems.
818:
819: \ack We thank J. R\"ottler and M. Kennett for inviting us to
820: participate to the workshop ``Mechanical behaviour in glassy behaviour'',
821: Vancouver, July 21-23, 2007, which led to the present collaborative work.
822: We would also like to thank David Reichman and Andreas Heuer
823: for useful correspondence.
824: Financial support from the Joint Theory Institute
825: (Argonne National Laboratory and University of Chicago),
826: CEFIPRA Project 3004-1, and ANR Grants TSANET and DYNHET
827: is acknowledged.
828:
829: \section*{References}
830:
831: \begin{thebibliography}{10}
832:
833: \bibitem{single}
834: A. N. Adhikari, N. A. Capurso, and D. Bingemann,
835: J. Chem. Phys. {\bf 127}, 114508 (2007).
836:
837: \bibitem{afm} E. Vidal Russell and N. E. Israeloff, Nature
838: {\bf 408}, 695 (2000).
839:
840: \bibitem{kegel}
841: W. K. Kegel and A. van Blaaderen, Science {\bf 287}, 290 (2000).
842:
843: \bibitem{weeks}
844: E. R. Weeks, J. C. Crocker, A. C. Levitt, A. Schofield, and
845: D. A. Weitz, Science {\bf 287}, 627 (2000).
846:
847: \bibitem{marty} G. Marty and O. Dauchot, Phys. Rev. Lett. {\bf 94},
848: 015701 (2005).
849:
850: \bibitem{virgile}
851: P. Bursac, G. Lenormand, B. Fabry, M. Oliver, D. A.Weitz,
852: V. Viasnoff, J. P. Butler, and J. J. Fredberg,
853: Nat. Mater. {\bf 4}, 557 (2005).
854:
855: \bibitem{laura}
856: L. J. Kaufman and D. A. Weitz, J. Chem. Phys. {\bf 125},
857: 074716 (2006).
858:
859: \bibitem{durian} A. S. Keys, A. R. Abate, S. C. Glotzer, and D. J. Durian,
860: Nature Phys. {\bf 3}, 260 (2007).
861:
862: \bibitem{gao} Y. Gao and M. Kilfoil, Phys. Rev. Lett. {\bf 99},
863: 078301 (2007).
864:
865: \bibitem{dibble}
866: C. J. Dibble, M. Kogan, and M. J. Solomon, Phys. Rev. E {\bf 74}, 041403
867: (2006).
868:
869: \bibitem{ediger} M. D. Ediger,
870: Annu. Rev. Phys. Chem. {\bf 51}, 99 (2000).
871:
872: \bibitem{zac} E. Zaccarelli,
873: J. Phys.: Condens. Matter {\bf 19}, 323101 (2007).
874:
875: \bibitem{dave} S. Manley, H. M. Wyss, K. Miyazaki, J. C. Conrad,
876: V. Trappe, L. J. Kaufman, D. R. Reichman, and D. A. Weitz,
877: Phys. Rev. Lett. {\bf 95}, 238302 (2005).
878:
879: \bibitem{puertas} A. M. Puertas, M. Fuchs, and M. E. Cates,
880: J. Chem. Phys. {\bf 121}, 2813, (2004).
881:
882: \bibitem{pablo}
883: P. Hurtado, L. Berthier, and W. Kob,
884: Phys. Rev. Lett. {\bf 98}, 135503 (2007).
885:
886: \bibitem{pinaki}
887: P. Chaudhuri, L. Berthier, and W. Kob,
888: Phys. Rev. Lett. {\bf 99}, 060604 (2007).
889:
890: \bibitem{kob}
891: W. Kob, C. Donati, S. J. Plimpton, P. H. Poole, and S. C. Glotzer,
892: Phys. Rev. Lett. {\bf 79}, 2827 (1997).
893:
894: \bibitem{stariolo}
895: D. A. Stariolo and G. Fabricius,
896: J. Chem. Phys. {\bf 125}, 064505 (2006).
897:
898: \bibitem{odagaki} T. Odagaki and Y. Hiwatari,
899: Phys. Rev. A {\bf 41}, 929 (1990).
900:
901: \bibitem{monthus}
902: C. Monthus and J.-P. Bouchaud, J. Phys. A {\bf 29}, 3847
903: (1996).
904:
905: \bibitem{epl} L. Berthier, D. Chandler, and J. P. Garrahan,
906: Europhys. Lett. {\bf 69}, 320 (2005).
907:
908: \bibitem{mw}
909: E. W. Montroll and G. H. Weiss, J. Math. Phys. (N.Y.) {\bf 6},
910: 167 (1965).
911:
912: \bibitem{jung1} Y. Jung, J.P. Garrahan, and
913: D. Chandler, Phys. Rev. E {\bf 69}, 061205 (2004).
914:
915: \bibitem{lj} L. Berthier and W. Kob, J. Phys.: Condens. Matter
916: {\bf 19}, 205130 (2007).
917:
918:
919: \bibitem{langer} J. S. Langer and S. Mukhopadhyay, arXiv:0704.1508.
920:
921: \bibitem{rob} L. Berthier and R. L. Jack,
922: Phys. Rev. E {\bf 76}, 041509 (2007).
923:
924: \bibitem{tunaley} J. K. E. Tunaley,
925: Phys. Rev. Lett. {\bf 33}, 1037 (1974).
926:
927: \bibitem{barkai1} E. Barkai and Y.-C. Cheng, J. Chem. Phys. {\bf 118}, 6167
928: (2003).
929:
930: \bibitem{barkai2} E. Barkai, V. Fleurov, and J. Klafter,
931: Phys. Rev. E {\bf 61}, 1164 (2000).
932:
933: \bibitem{jung2}
934: Y. Jung, J. P. Garrahan, and D. Chandler,
935: J. Chem. Phys. {\bf 123}, 084509 (2005).
936:
937: \bibitem{lutz} L. O. Hedges, L. Maibaum, D. Chandler,
938: and J. P. Garrahan, J. Chem. Phys. {\bf 127}, 211101 (2007).
939:
940:
941: \end{thebibliography}
942:
943: \end{document}
944:
945: