1:
2:
3: \documentclass[twocolumn,showpacs,amsmath,pre]{revtex4}
4: %\documentclass[showpacs,amsmath,preprint,pre]{revtex4}
5: \usepackage{graphicx}
6: %\usepackage{amsmath}
7: %\usepackage{dcolumn}
8:
9: \begin{document}
10: \title{Local mean-field study of capillary condensation in
11: silica aerogels}
12:
13: \author{F. Detcheverry, E. Kierlik, M. L. Rosinberg, and G. Tarjus}
14: \affiliation{Laboratoire de Physique Th{\'e}orique des Liquides, Universit{\'e} Pierre et
15: Marie Curie, 4 place Jussieu, 75252 Paris Cedex 05, France}
16:
17: \date{\today}
18:
19: \begin{abstract}
20: We apply local mean-field (i.e. density functional) theory to a lattice model of a
21: fluid in contact with a dilute, disordered gel network. The gel
22: structure is described by a diffusion-limited cluster
23: aggregation model. We focus on the influence of porosity on both the
24: hysteretic and the equilibrium behavior of the fluid as one varies the
25: chemical potential at low temperature. We show that the shape of the
26: hysteresis loop changes from smooth to rectangular as the porosity
27: increases and that this change is associated to disorder-induced out-of-equilibrium
28: phase transitions that differ on adsorption and on desorption. Our
29: results provide insight in the behavior of $^4$He in silica aerogels.
30:
31: \end{abstract}
32: \pacs{64.60.-i,68.45.Da,75.60.Ej}
33: \maketitle
34:
35: \def\be{\begin{equation}}
36: \def\ee{\end{equation}}
37: \def\bea{\begin{eqnarray}}
38: \def\eea{\end{eqnarray}}
39:
40: \section{Introduction}
41:
42: The influence of quenched disorder on phase transitions and critical phenomena continues to be the focus
43: of intensive experimental and theoretical research activity. Major effects are expected and actually
44: observed when the disorder couples linearly to the order parameter of the
45: system, a situation that is realized when a fluid or a fluid mixture is confined within a porous glass or
46: is in contact with the interconnected strands of a gel. A dilute
47: rigid network is a particularly interesting medium from a theoretical perspective because exclusion (i.e., confinement) effects do not play a dominant role, so that the random-field Ising model (RFIM) may be a useful framework to interpret the experimental observations\cite{BG1983}.
48:
49: A striking example of the influence of a gel network on fluid phase behavior
50: is provided by the thermodynamic studies of Chan and co-workers
51: on $^4$He in silica aerogels of varying porosity. Silica
52: aerogels are highly porous, fractal-like solids made of a tenuous
53: network of SiO$_2$ strands interconnected at random sites. In a $95\%$ porosity aerogel, specific
54: heat and adsorption (vapor-pressure isotherm) measurements\cite{WC1990} performed
55: in the vicinity of the critical temperature of the pure fluid
56: ($T_c$=5.195K) show evidence of a phase separation between a low-density ``vapor'' phase
57: (presumably composed of $^4$He vapor plus a liquid film around the silica strands\cite{LMPMMC2000}) and a high-density ``liquid'' phase filling the whole pore space. The first-order transition appears to
58: terminate at a sharply defined critical point that is only 31mK below
59: $T_c$, which suggests that the system is in a weak random-field
60: regime. The coexistence boundary in the presence of
61: aerogel is, however, much narrower than in the pure system. Subsequently, similar results were obtained
62: with N$_2$ in the same aerogel, using light scattering and vapor-pressure isotherms\cite{WKGC1993}.
63: Since out-of-equilibrium and hysteretic effects due to domain
64: formation are generally observed in random-field systems below the
65: critical temperature of the pure system (as illustrated by the
66: behavior of binary mixtures in aerogels\cite{BFC1997} and diluted
67: antiferromagnets in a magnetic field\cite{B1998}), it is noteworthy
68: that no hysteresis is present in the measurements performed in
69: Refs.\cite{WC1990,WKGC1993}. The gel-fluid system has thus reached
70: equilibrium within the time scale of the experiments, which is
71: indeed found much longer than the characteristic time of activated dynamics\cite{WKGC1993}.
72: The situation changes, however, at lower temperatures and the adsorption of $^4$He in a $98\%$ porosity aerogel is clearly hysteretic at 3.60K and 2.34K\cite{C1996,TYC1999}. Both adsorption and desorption isotherms display a vertical step at well-defined pressures, but draining occurs at a lower pressure than filling\cite{note2}.
73: The shape of the hysteresis loop, moreover, depends on the porosity of
74: the aerogel: in an aerogel of 87\% porosity, $^4$He adsorbs and desorbs gradually at 2.42K and there is
75: no signature of a ``liquid-vapor'' phase coexistence (see Fig. 4(c) in
76: Ref.\cite{TYC1999}). Such hysteretic behavior is reminiscent of capillary
77: condensation in a low-porosity solid like Vycor glass where one
78: observes a rapid increase of the adsorbed quantity at a pressure below
79: the bulk saturated vapor pressure, but no sharp vertical step in
80: the adsorption isotherms\cite{E1967}. The mechanism for hysteresis in
81: porous substrates has prompted much discussion in the literature and
82: various explanations have been proposed, focusing either on
83: single-pore metastability {\it {\`a} la} van der Waals or on network pore-blocking effects\cite{BE1989}.
84: Since both models seem completely inadequate to describe light
85: aerogels, the $^4$He experiments raise several
86: questions: what is the scenario for the change in the shape
87: of the hysteresis loops\cite{note3}? Do filling and draining obey different mechanisms? What is the true equilibrium behavior when hysteresis is present?
88:
89: These questions are addressed in the present work where we build on
90: our earlier studies of capillary condensation in disordered
91: porous solids\cite{K2001,K2002,R2003}, but focusing on aerogels. We are mainly
92: interested in the influence of the porosity on the hysteretic behavior of sorption isotherms. This is partly
93: motivated by theoretical studies of the zero-temperature
94: RFIM which predict the existence of a disorder-induced out-of-equilibrium
95: phase transition in the hysteresis loop\cite{S1993}. The
96: first experimental observations of this phase transition have been
97: recently reported in the literature\cite{B2000}, and we argue that the
98: change in the $^4$He adsorption isotherm from sharp to smooth can be
99: interpreted within the same framework. We propose a different scenario
100: for draining and we relate the observed behavior to an
101: out-of-equilibrium phase transition
102: associated to the depinning of the liquid-vapor interface.
103: The approach we develop is similar
104: to that used previously for calculating the irreversible behavior of spin
105: glasses, random-field ferromagnets and diluted
106: antiferromagnets\cite{SLG1983}; it consists in studying the evolution of
107: the free-energy surface (more precisely, the grand-potential surface
108: $\Omega$) as the external driving field (here, the pressure $P$ of the
109: external vapor or, equivalently, the chemical potential $\mu$) is changed.
110: $\Omega$ is a functional of the local fluid
111: density and is treated in the mean-field approximation. At low
112: temperatures this multidimensional free-energy landscape is characterized by a large
113: number of local minima in which the system may be trapped. The main physical assumption underlying our
114: description is that thermally activated processes play a negligible role on the time scale
115: of the experiments (in other words, the dynamics is similar to that at $T=0$).
116: The evolution of the system then proceeds either continuously by the
117: deformation of the local minimum in which the system is trapped or
118: when this latter becomes unstable by a jump to another minimum
119: (avalanche); the response to the driving field is then discontinuous and irreversible. Since it is impossible to perform
120: such a study for a continuous model because of computational limitations
121: (especially when finite-size scaling analysis is required near phase transitions), we
122: adopt a lattice-gas description that incorporates at a coarse-grained
123: level the geometric and energetic disorder of the gel-fluid mixture\cite{KKRT2001}.
124: Previous work has shown that many of
125: the phenomena observed in experiments on fluids in
126: disordered porous solids can be reproduced qualitatively by
127: such simple lattice models\cite{K2001,K2002,R2003,WSM2001}. As in other
128: studies of phase transitions in aerogel\cite{UHR1995,STC1999}, we
129: model the solid by a
130: fractal structure obtained by diffusion-limited
131: cluster-cluster aggregation.
132:
133: The paper is organized as follows. In Sec. II, we introduce the
134: lattice model and discuss some important features of the aerogel structure. In Sec. III, we
135: present the local mean-field theory and describe the numerical
136: procedure. The results for the adsorption, desorption, and equilibrium
137: isotherms in $87\%$ and $95\%$ porosity aerogels at $T/T_c=0.45$ are given in Sec. IV. The final section presents a summary and conclusions.
138:
139: \begin{figure}[h]
140: \begin{center}
141: \resizebox{8cm}{!}{\includegraphics{fig1.ps}}
142: %\vspace{.4cm}
143: \caption{Three-dimensional realization of a 98\% porosity DLCA aerogel on a
144: bcc lattice of size $L=100$ with periodic boundary conditions.}
145: \end{center}
146: \end{figure}
147:
148: \section{Lattice model and aerogel structure}
149:
150: The lattice model used in this work describes the solid as a
151: collection of fixed impurities that exert a random yet correlated (by
152: the connectivity of the strands) external field on the atoms of the
153: fluid. The Hamiltonian is given by\cite{PRST1995,KRTP1998}
154:
155: \bea
156: {\cal H} = &-&w_{ff}\sum_{<ij>} \tau_{i}\tau_{j} \eta_i \eta_j -\mu \sum_i \tau_i
157: \eta_i\nonumber \\
158: &-&w_{sf}\sum_{<ij>
159: }[\tau_{i}\eta_i (1-\eta_j)+\tau_{j}\eta_j (1-\eta_i)]
160: \eea
161: where $\tau_i=0,1$ is the usual fluid occupation variable ($i=1...N$)
162: and $\eta_i=1,0$ is a quenched random variable that characterizes the
163: presence of gel particles on the lattice (when $\eta_i=0$, site $i$
164: is occupied by the gel); $w_{ff}$ and $w_{sf}$ denote, respectively, the fluid-fluid and
165: solid-fluid attractive interactions, $\mu $ is the fluid chemical
166: potential, and the double summations run over all distinct pairs of
167: nearest-neighbor (n.n.) sites. One can thus vary
168: the gel porosity, $p=(1/N) \sum_i \eta_i$, by changing the number of solid sites or modify the ``wettability'' of the solid surface
169: by changing the ratio $y=w_{sf}/w_{ff}$. For $y=1/2$,
170: the model reduces to a site
171: diluted Ising model, as can be seen by transforming
172: the fluid occupation variable $\tau_i$ to an Ising spin variable,
173: $s_i=2\tau_i-1$\cite{KRTP1998}: preferential adsorption of the liquid
174: phase onto the gel is thus modeled by $y>1/2$. Random fields are
175: generated in the system when $y \neq 1/2$, and the fluctuating part of the
176: field acting on spin $s_i$ is proportional to the number of solid
177: particles that occupy the nearest neighbors of site
178: $i$\cite{KRTP1998}. This a discrete random variable that can take
179: the values $0,1,...c$, where $c$ is the
180: coordination number of the lattice, and whose probability distribution is strongly porosity-dependent.
181:
182: Gel configurations (i.e., sets $\{\eta_i\}$) are generated by a standard on-lattice diffusion-limited
183: cluster-cluster aggregation (DLCA) algorithm\cite{M1983} adapted to a body-centered cubic (bcc) lattice
184: with periodic boundary conditions. The choice of the bcc lattice will
185: be motivated in Sec. IV.B.
186: The DLCA algorithm mimics the growth process
187: of base catalyzed aerogels used in helium experiments, and it has
188: been shown to reproduce the main structural features of these
189: aerogels measured from scattering experiments\cite{H1994}.
190: A typical exemple of a $98\%$ porosity DLCA aerogel on a lattice of linear size
191: $L=100$ is shown in Fig. 1 (from now on, we take the lattice
192: constant $a$ as the unit length; the total number of sites in the lattice is thus
193: $N=2 L^3$). One can clearly see
194: the fractal-like character of the gel network that results from the aggregation mechanism.
195:
196:
197: \begin{figure}[t]
198: \begin{center}
199: \resizebox{8cm}{!}{\includegraphics{fig2.ps}}
200: %\vspace{.4cm}
201: \caption{Log-log plot of the aerogel correlation function $g(r)$ for
202: $p=0.99, 0.98, 0.97, 0.95, 0.92$ and $0.87$ (for top to
203: bottom). The solid line is a fit with slope $-1.13$ that
204: corresponds to the fractal regime for $p=0.99$.}
205: \end{center}
206: \end{figure}
207:
208: More quantitative information on the structural properties of the gel
209: can be extracted from the two-point correlation function $g(r)$. Each
210: curve shown in Fig. 2 corresponds to a different porosity $p$ and
211: results from an average over several simulations in a box of size $L=100$
212: (for $p=0.87, 0.92, 0.95$) or $L=200$ (for $p=0.97, 0.98, 0.99$).
213: Only the most dilute samples exhibit a true intermediate regime
214: described by the power-law behavior $g(r) \simeq r^{-(3-d_f)}$
215: revealing the fractal character of the intra-cluster density-density
216: correlations\cite{note4}. For $p=0.99$, one finds $d_f\approx 1.87 $, a
217: value that lies in the range $1.7-1.9$ expected for DLCA in three
218: dimensions. Note, however, that the fractal dimension decreases with increasing
219: porosity\cite{L1998}, so that the asymptotic value must be somewhat
220: smaller. Another estimation of $d_f$ can be obtained from the average
221: cluster size $\xi_G$ by assuming that $\xi_G$ varies as
222: $\rho_G^{-1/(3-d_f)}$, where $\rho_G=1-p$ is the gel concentration. As
223: suggested in Ref.\cite{H1994}, $\xi_G$ can be estimated from the location of the
224: first minimum of $g(r)$ (which is hardly visible on the scale of
225: Fig. 2). The plot of $\xi_G$ versus $\rho_G$ shown in Fig. 3 yields $d_f\approx 1.85$.
226: Since the correlation length is
227: in the range $650-1300\AA$ for a $98\%$ base-catalyzed
228: aerogel\cite{LMPMMC2000,LKMP2000} and $\xi_G(\rho_G=0.02)\approx 21$ in the simulation, one can
229: estimate that the lattice constant $a$ corresponds to about $30-60\AA$. This is consistent with the
230: coarse-grained picture of a gel site representing a Si0$_2$ particle
231: with a diameter of about $30\AA$.
232:
233: Another relevant length scale is the size of the largest
234: cavity in the aerogel\cite{PP1999}. Fig. 4 shows the distribution $P(n)$ of nearest distances $n$
235: from an empty site to the aerogel as a function of porosity (the integer
236: ``distance'' $n$ is defined here as the length of the shortest path on the
237: lattice from an empty site to
238: a gel site; $n=1$ means that the empty site is n.n. of a gel site). The
239: plot is the normalized histogram of these distances and
240: $\sum_1^nP(n')$ gives the probability of being closer than $n$ to the
241: aerogel. One can see that the size of the largest cavity is $n\approx 32$
242: for a $99\%$ aerogel and decreases to $n\approx 10$ for a
243: $95\%$ aerogel and to $n\approx 5$ for a
244: $87\%$ aerogel. It is noticeable that the distribution changes
245: significantly when decreasing the porosity from $95\%$ to $87\%$. In the former case,
246: $P(n)$ has its maximum at $n=2$ and there is a significant proportion of empty
247: sites that are not in the immediate vicinity of the gel. In the latter
248: case, $P(n)$ is monotonic and strongly peaked at $n=1$, which
249: indicates that most of the empty sites are very close to the gel. We shall see in section IV.A that these differences
250: have important implications for the behavior of the fluid during adsorption.
251:
252: \begin{figure}[t]
253: \begin{center}
254: \resizebox{8cm}{!}{\includegraphics{fig3.eps}}
255: %\vspace{.4cm}
256: \caption{Average cluster size $\xi_G$ (estimated from the location
257: of the first minimum of $g(r)$) as a function of the gel concentration
258: $\rho_G=1-p$. The dashed line corresponds to the power law fit $\xi_G=0.707
259: \rho_G^{-0.868}$.}
260: \end{center}
261: \end{figure}
262:
263: \begin{figure}
264: \begin{center}
265: \resizebox{8cm}{!}{\includegraphics{fig4.eps}}
266: %\vspace{.4cm}
267: \caption{Distribution $P(n)$ of nearest distances from an empty site to
268: the gel (see
269: text). From right to left:
270: $p=0.99, 0.98, 0.97, 0.95, 0.92,0.87$ (the dashed lines are guides
271: for the eye).}
272: \end{center}
273: \end{figure}
274:
275: It is clear from Fig. 3 that the minimum system size necessary to describe
276: correctly collective effects occuring inside the aerogel on long length scales (like a sharp condensation event in
277: the {\it whole} pore space) depends strongly on the porosity. For instance,
278: a box of linear size $L=100$ is not large enough to represent the
279: whole pore space of a $99\%$ aerogel because it
280: does not contain a sufficient number of connected fractal aggregates
281: (a problem that cannot be cured by merely increasing the number of realizations). Within
282: the framework of local mean-field theory, it is however
283: impossible to simulate much larger lattices because the computational
284: time and the memory storage requirement become
285: prohibitive (at finite temperature, the local fluid densities are continuous variables, which forbids
286: the use of bits algorithms).
287: In the present work, we consider $95\%$ and $87\%$ aerogels
288: for which we can investigate the statistics of collective events
289: while working with lattices of reasonable size (between $L=25$ and $L=100$).
290:
291: Since it has been shown previously that the interface between the porous solid and the bulk gas
292: may have a dramatic effect on the desorption process\cite{R2003}, two different setups are considered. In the first one, the system is periodically replicated in all directions so that no interfacial effects are taken into account. In the second setup an interface is created by placing a slab of vapor of width $L_b=10$ (the gas ``reservoir'') in
293: contact with one of the $[100]$ faces of the simulation box. Periodic boundary conditions
294: are then imposed in the $y$ and $z$ directions and reflective boundary
295: conditions in the $x$ direction (obtained by reflecting the lattice at the boundaries).
296:
297: In order to completely specify the model, one must also fix the value of the interaction
298: parameter $y$. As shown previously\cite{R2003}, the shape of the
299: hysteresis loops changes with $y$. Since it is
300: meaningless in a coarse-grained picture to compute $y$ from the actual values of the fluid-fluid and solid-fluid
301: van der Waals interactions, we have chosen its value
302: so as to reproduce approximately the height of the hysteresis loop in the $87\%$ aerogel at low temperature.
303: Specifically, the results presented in this work are calculated with $y=2$, a value
304: for which the filling of the $87\%$ aerogel at the lower closure point of the
305: hysteresis loop is roughly the same as in the experiment at $T=2.42$K
306: (i.e., at $T/T_c \approx 0.45$). This corresponds to about $1/3$ of the
307: filling reached
308: at the plateau just before saturation, as can be seen in Fig. 4(c)
309: of Ref.\cite{TYC1999} (see also Fig. 5(a) below). All calculations
310: are done at this single reduced temperature (recall that
311: $kT_c/w_{ff}=c/4=2$ in the mean-field approximation).
312:
313: \section{Local mean-field theory (LMFT) and numerical procedure}
314:
315: For the present model, the LMFT consists in solving
316: the self-consistent equations for the thermally averaged fluid densities $\rho_i(\{\eta_i\})=<\tau_i\eta_i>_T$
317: obtained from minimizing the mean-field grand-potential functional\cite{K2001}
318:
319: \bea
320: \Omega(\{\rho_i\})&=&k_BT \sum_i[\rho_i\ln \rho_i+(\eta_i-\rho_i)\ln(\eta_i-\rho_i)]\nonumber\\
321: &-&w_{ff} \sum_{<ij>}\rho_i\rho_j -\mu\sum_i\rho_i\nonumber\\
322: &-&w_{sf}\sum_{<ij>}[\rho_i(1-\eta_j)+\rho_j(1-\eta_i)] \ .
323: \eea
324: The variational procedure $\delta \Omega/\delta \rho_i=0$ gives a set of $N$ coupled non-linear equations
325: \be
326: \rho_i=\frac{\eta_i}{[1+e^{-\beta (\mu+w_{ff}\sum_{j/i}[\rho_j+y(1-\eta_j)]) }]}\ ,
327: \ee
328: where $\beta=1/(k_BT)$ and the sum runs over the $c$ nearest neighbors of
329: site $i$. At low temperature, these equations may have several
330: solutions,
331: and the grand potential corresponding to solution $\{\rho_i^{\alpha}\}$ is given by\cite
332: {K2002}
333: \be
334: \Omega^{\alpha}=k_BT \sum_i \eta_i\ln(1-\frac{\rho_i^{\alpha}}{\eta_i})+w_{ff} \sum_{<ij>}\rho_i^{\alpha}\rho_j^{\alpha} \ .
335: \ee
336: By using an iterative method to solve Eqs. (3), one only finds
337: solutions that are only local minima of the grand-potential surface, i.e., metastable states\cite{SLG1983}.
338:
339: For a given realization of the aerogel, the sorption isotherms (i.e.,
340: the curves $\rho_f=(1/N) \sum_i \rho_i(\{\eta_i\},\mu)$) are obtained by increasing
341: or decreasing the chemical potential in small steps $\delta \mu$. At each
342: subsequent $\mu$, the converged solution at $\mu-\delta\mu$ (on adsorption) or at
343: $\mu+\delta\mu$ (on desorption) is used to start the iterations. The
344: isotherms are then averaged over a number of gel realizations depending on the system size.
345: In order to determine the equilibrium isotherms, we have also searched for additional solutions of Eqs. (3) inside the hysteresis loop by performing scanning trajectories, as explained below
346: in section IV.C.
347:
348: To accelerate the convergence of the numerical procedure, the
349: iterations have been updated (i.e., the new value of $\rho_i$ is substituted into the r.h.s. of Eqs. (3)
350: before waiting for the rotation through the indices $i$ to be
351: complete) and the evolution of each site density $ \rho_i$ has been monitored in the following way. Fluid sites are divided into two categories, active and passive.
352: At the beginning
353: of the calculation, all sites are active and the procedure stops when all sites are passive. At each
354: iteration only active sites are considered and their new density is
355: calculated using Eqs. (3). If
356: $\mid\rho_i^{(n)}-\rho_i^{(n-1)}\mid/\rho_i^{(n-1)}<10^{-6}$, where $n$ denotes the
357: $n$th iteration, the site becomes
358: passive; otherwise its remains active and its nearest-neighors become (or remain)
359: active (some sites can thus be passive during a few iterations and become active
360: again).
361: \begin{figure}[t]
362: \begin{center}
363: \resizebox{8cm}{!}{\includegraphics{fig5.eps}}
364: %\vspace{.4cm}
365: \caption{Hysteresis loops in the $87\%$ (a) and $95\%$ (b) model aerogels
366: calculated at $T/T_c=0.45$. Equilibrium isotherms are indicated by the dotted lines.}
367: \end{center}
368: \end{figure}
369: The advantage of this algorithm
370: is that the number of active sites may quickly become a small fraction of the
371: total number of sites, which of course significantly reduces the overall
372: computation time. For instance, the number of active sites is only
373: $O(L^2)$ when a planar liquid-vapor interface propagates through the
374: system. The algorithm is useful in
375: dilute aerogels because of the presence of large cavities.
376: We have checked that the numerical results are not different (to the precision of the calculations) from those obtained without updating or with using a more standard convergence criterion, like in our previous studies of fluid adsorption in random matrices\cite{K2001,K2002,R2003}.
377: Convergence (to an accurracy of $10^{-6}$) typically requires between $10^{2}$ to $10^{3}$
378: iterations, and several CPU hours on a $2.4$ GHz workstation
379: are needed to calculate a single adsorption isotherm in a system
380: of linear size $L=100$ (using a step
381: $\delta\mu/w_{ff}=10^{-2}$ or $10^{-3}$ to increment the chemical potential). The
382: search for equilibrium isotherms is much more time consuming (see
383: below in Sec. IV.C) and has been made possible by using parallel
384: computation on a Beowulf cluster of $24$ processors.
385:
386:
387: \section{Results and discussion}
388:
389: The results of the present study are summarized in Fig. 5 that shows
390: the adsorption, desorption, and equilibrium isotherms calculated for
391: the $87\%$ and $95\%$ model aerogels for $y=2$ and $T/T_c=0.45$ (from now on, $\mu$ is in units of $w_{ff}$, the fluid-fluid interaction parameter).
392: These curves result from the detailed numerical analysis described
393: in the following and correspond to the thermodynamic
394: limit. The
395: most striking feature is the change in the shape of the hysteresis
396: loop from smooth to rectangular as the porosity increases. This change
397: is similar to that observed experimentally by Chan and
398: co-workers (see Figs. 4(b) and 4(c) in Ref.\cite{TYC1999}). As is explained below,
399: we predict that the three isotherms (adsorption, desorption, and
400: equilibrium) are discontinuous in the $95\%$ aerogel (at least within mean-field theory where thermal
401: fluctuations are neglected). The underlying
402: physical mechanisms are, however, quite different.
403:
404: \subsection{Adsorption }
405:
406: \begin{figure}[t]
407: \begin{center}
408: \resizebox{8cm}{!}{\includegraphics{fig6.eps}}
409: %\vspace{.4cm}
410: \caption{Representative adsorption isotherms in $87\%$ (a) and $95\%$ (b)
411: aerogels. System sizes are $L=50$ (a) and $L=100$ (b). Notice the
412: change in the scale of the $x$ axis between the two figures.}
413: \end{center}
414: \end{figure}
415:
416: We first examine the adsorption process. Calculations performed in the
417: presence and in the absence of a gel/reservoir interface yield isotherms that are
418: almost indistinguishable (except for the smallest system sizes),
419: confirming the conclusion reached in our previous work\cite{R2003} that adsorption does not depend on the existence of a free surface. This is also in line with the results of previous
420: calculations done in single slit-like or cylindrical pores\cite{MVS1989}.
421:
422: In Figs. 6(a) and 6(b) are shown some
423: adsorption isotherms calculated in $87\%$ and $95\%$ porosity samples of linear
424: size $L=50$ and $L=100$, respectively (this corresponds to roughly the
425: same ratio $L/\xi_G \approx 10$). In both cases, we focus on the region where
426: the adsorption is the steepest (see Fig. 5). At lower coverages, the isotherms look gradual and
427: smooth in the two systems, but as $\mu$ increases, they consist of
428: little steps of varying sizes. In the $87\%$ aerogel, the size of these
429: ``avalanches'' remains small all the way up to the slowly increasing plateau that extends to
430: saturation. On the other hand, in the $95\%$ aerogel, the size of the
431: jumps tends to increase with $\mu$ and in most of the samples there is a large final avalanche after which the filling is almost complete (there are however some rare
432: realizations where the last jump is not much larger that the preceeding ones, as illustrated by one of the isotherms in Fig. 6(b)).
433:
434: \begin{figure}[t]
435: \begin{center}
436: \resizebox{6cm}{!}{\includegraphics{fig7.ps}}
437: %\vspace{.4cm}
438: \caption{Cross-section of a $87\%$ aerogel sample on adsorption at
439: $\mu=-5, -4.64, -4.5$ and $-4.35$ (L=100). Gel sites are shown in black
440: and fluid density is shown as various degrees of grey. As explained in
441: the text, one can actually distinguishes only two regions corresponding
442: respectively to $\rho_i\leq 0.05$ (light grey) and $\rho_i\geq 0.95$ (dark grey).}
443: \end{center}
444: \end{figure}
445:
446: In order to better visualize the underlying microscopic mechanism, we
447: show in Figs. 7 and 8 some cross-sections of typical $87\%$ and $95\%$
448: porosity samples for different values of the chemical potential along the
449: isotherms (here, the size of both systems is $L=100$). One can see
450: that in both aerogels the first stage of the adsorption process is
451: the formation of a liquid layer that coats the aerogel strands. Then,
452: as $\mu$ increases, the film thickens and condensation occurs in the
453: smallest crevices defined by neighboring gel strands. In the $87\%$
454: aerogel, it is difficult to discriminate between these two filling
455: processes because the available empty space is small, as illustrated by Fig. 4.
456: For the same reason, the vapor bubbles that remain in the system at $\mu=-4.5$
457: are isolated, and, as they
458: shrink in size, the adsorption continues gradually until the solid is completely filled with liquid. This is precisely the scenario described in Ref.\cite{TYC1999} from the experimental observations
459: with $^4$He. This is also similar to what happens in a low-porosity
460: glass like Vycor\cite{PLAHDW1995}. On the contrary, in the $95\%$
461: sample, one clearly distinguishes the small capillary condensation events that occur in
462: some regions of the aerogels (compare the figures for
463: $\mu=-4.16$ and $\mu=-4.13$) and the major final event that
464: corresponds to the filling of a large void space spanning the
465: whole sample (as $\mu$ is increased from $ -4.13$ to $-4.125$). Note
466: that the local ``liquid-vapor'' interfaces inside the
467: gel are very
468: sharp at this low temperature, which explains why
469: one can only distinguishes two different regions in Figs. 7 and 8. Indeed, as
470: illustrated in Fig. 9 for the lighter aerogel, it is found that
471: the distribution of the $\rho_i$'s is bimodal, with
472: most of the fluid sites having a density lower than $0.05$ or larger than
473: $0.95$ (the distribution is similar in the $87\%$
474: porosity aerogel with just a little more intermediate
475: densities).
476:
477: \begin{figure}[t]
478: \begin{center}
479: \resizebox{6cm}{!}{\includegraphics{fig8.ps}}
480: %\vspace{.4cm}
481: \caption{Same as Fig. 7 for a $95\%$ sample at $\mu=-4.5, -4.16, -4.13$
482: and $-4.125$ (L=100). This is one of
483: the samples shown in Fig. 5.b}
484: \end{center}
485: \end{figure}
486:
487: \begin{figure}[b]
488: \begin{center}
489: \resizebox{6cm}{!}{\includegraphics{fig9.eps}}
490: \caption{Histogram of the local fluid densities $\rho_i$ in the $95\%$ aerogel sample of Fig. 8 at $\mu=-4.13$.}
491: \end{center}
492: \end{figure}
493: From these results (see for instance the cross-section of the $95\%$
494: aerogel at $\mu=-4.13$), there is no indication that the radius of
495: curvature of the liquid-vapor interface is concave and uniform
496: throughout the sample just before the major condensation event, as was
497: suggested in Ref.\cite{TYC1999}.
498: This makes unlikely any interpretation in terms of a traditional
499: capillary condensation model based for instance on the
500: application of the Kelvin equation\cite{LMPMMC2000,GBBC2003}. Indeed,
501: condensation in the remaining void space is itself triggered by a
502: condensation event that occurs in one (or several) small region of the aerogel and that induces further
503: condensation in the system.
504: This collective behavior in the form of avalanches of varying sizes is similar to that of
505: the Gaussian RFIM at $T=0$ proposed by Sethna and co-workers to
506: describe the Barkhausen effect in low-T ferromagnetic
507: materials\cite{S1993}. Increasing slowly the chemical potential of the
508: adsorbed fluid is equivalent to increasing adiabatically the applied
509: field in a ferromagnet, and changing the porosity of the aerogel
510: modifies the amount of quenched disorder in the system. In the RFIM,
511: the amount of disorder is controlled by the width of the random-field
512: distribution: only small avalanches are observed for large disorder,
513: resulting in a smooth hysteresis loop, whereas one macroscopic avalanche produces a discontinuity in the magnetization for small disorder\cite{S1993,PV2003}. The transition between these two regimes for a certain value of the disorder corresponds to a critical out-of-equilibrium phase transition at which the distribution of avalanches follows a power law.
514:
515: \begin{figure}[t]
516: \begin{center}
517: \resizebox{8cm}{!}{\includegraphics{fig10.eps}}
518: %\vspace{.4cm}
519: \caption{Average adsorption isotherms in $87\%$ (a) and $95\%$ (b)
520: aerogels for different system sizes. The number of gel realizations
521: ranges from $500$ ($L=35$) to $20$ ($L=100$) for the $87\%$ aerogel,
522: and from $2000$ ($L=35$) to $300$ ($L=100$) for the $95\%$ aerogel}
523: \end{center}
524: \end{figure}
525: The existence of a major condensation event in most of the $95\%$
526: samples of size $L=100$ therefore suggests that there is one macroscopic (i.e., infinite)
527: avalanche in the thermodynamic limit, corresponding to a finite jump
528: in the adsorption isotherm. However, this can be only confirmed by
529: performing a finite-size scaling analysis of the isotherms obtained
530: after averaging over many gel realizations. We indeed observe
531: significant sample-to-sample fluctuations both in the
532: location and the height of the largest jump. Such average
533: isotherms are shown in Fig. 10. For the $87\%$ aerogel, there is obviously
534: no size-dependence and we can safely conclude that the adsorption is gradual
535: in the thermodynamic limit. On the contrary, in the
536: lighter aerogel, the isotherms look steeper and steeper
537: as the system size is increased and the results suggest that there is indeed a
538: discontinuity when $L\to \infty$ (the isotherm corresponding to $L=35$ is
539: shown here for completeness, but this size is probably too small to
540: describe properly a $95\%$ aerogel). As we discussed elsewhere\cite{K2002}
541: in the case of fluid equilibrium behavior in purely random solids, one
542: expects that the maximal slope of the
543: isotherms should scale as $L^{3/2}$ at a first-order transition
544: (assuming that the location of the largest jump fluctuates around its mean value $\overline{\mu}_t(L)$ with a
545: variance $\delta \mu_t(L)^2 \propto L^{-3}$). As shown in Fig. 11, one can obtain a reasonably good collapse of
546: the $L=50, 70$ and $100$ irreversible ``compressiblity'' curves $d\rho_f/d\mu$
547: by using the scaling variable $L^{\chi}\{\mu-
548: \overline{\mu}_t(L)\}$ with $\chi\approx 1.22$.
549:
550: There are at least two possible explanations for the discrepancy with the expected value $\chi=3/2$. Firstly,
551: the system sizes could still be too small so that too many gel
552: realizations would have a non-typical behavior, with several avalanches
553: of similar heights (note that in a coarse-grained picture of a $95\%$
554: gel-fluid system where a region of size $\xi_G$ is represented by a
555: single effective spin, there would be only about $10^3$ such
556: effective spins in a sample of size
557: $L=100$). Secondly, one may be close to the critical value $p_c$ of the
558: disorder, i.e., the critical value of the porosity for $y=2$ and $T/T_c=0.45$, for which the
559: infinite avalanche first appears. Studies of the $T=0$
560: RFIM have moreover shown that the critical region is unusually
561: large\cite{S1993}. The value $\chi=1.22$ is compatible with the
562: predictions of Sethna and co-workers, with $\chi=\beta\delta/\nu\simeq2-\eta$\cite{note5}, although one cannot also exclude that
563: the exponents differ from those of the conventional RFIM
564: because of the presence of impurities\cite{T1996}.
565: In any case, it seems reasonable, on the basis of the present
566: calculations, to predict the existence of a phase transition with
567: most probably a
568: discontinuity in the adsorption isotherm in the thermodynamic limit,
569: as shown in Fig. 5(b).
570:
571:
572: \begin{figure}[t]
573: \begin{center}
574: \resizebox{8cm}{!}{\includegraphics{fig11.eps}}
575: %\vspace{.4cm}
576: \caption{Scaling plot of the compressibility curves $d\rho_f/d\mu$ during adsorption in
577: the $95\%$
578: aerogel with $\chi=1.22$ ($\overline{\mu}_t(L)=-4.123, -4.126, -4.128$ for $L=50, 70,
579: 100$, respectively; extrapolation to $L\to \infty$ gives $\mu_t=$lim$_{L\to \infty}\overline{\mu}_t(L)\simeq -4.131$).}
580: \end{center}
581: \end{figure}
582:
583:
584: \subsection{Desorption }
585: \begin{figure}[t]
586: \begin{center}
587: \resizebox{8cm}{!}{\includegraphics{fig12.eps}}
588: %\vspace{.4cm}
589: \caption{Desorption isotherms in $87\%$ (a) and $95\%$ (b)
590: aerogels. System sizes are $L=50$ (a) and $L=100$ (b).}
591: \end{center}
592: \end{figure}
593:
594: As discussed elsewhere\cite{R2003}, fluid desorption in disordered
595: porous solids can take place via several different
596: mechanisms, depending on the temperature and on the
597: structural and energetic properties of
598: the solid (in Ref.\cite{R2003}, however, only the influence
599: of the interaction parameter $y$, i.e., of the wetting properties of
600: the adsorbed fluid, was described). In
601: a first mechanism, desorption is due to the appearance of vapor
602: bubbles in the bulk of the material, bubbles that grow, coalesce, and
603: eventually extend over the whole pore space. The mass adsorbed
604: then decreases continuously. In
605: the other mechanisms, draining of the solid starts from the surface, and the
606: desorption is associated to the penetration of a
607: vapor-liquid interface which was previously pinned by the
608: irregularities of the solid structure for $\mu$ larger than some threshold
609: value $\mu_c$ (the ``depinning''threshold). The desorption curve is then
610: either gradual or discontinuous,
611: depending on whether the growth of the vapor domain is isotropic and
612: percolation-like (self-similar), or compact (self-affine). In those cases the transition is however always critical because the scale of the rearrangements of the interface diverges as $\mu \to \mu_c^+$. This
613: is very similar to the physics of fluid invasion in porous media\cite{CR1988}, although we are considering here a single compressible fluid, and of field-driven domain wall motion in disordered magnets\cite{KR2000}.
614:
615: In order to find what are the relevant mechanisms in the
616: $87\%$ and $95\%$ aerogels for $y=2$ and $T/T_c=0.45$, we have
617: studied the two systems
618: in the presence and in the absence of the interface with the gas reservoir.
619: In the latter case, we find that emergence of vapor bubbles in the bulk of
620: the solid only occurs at low values of the chemical potential ($\mu
621: \simeq -5.27$ and $\mu\simeq -5.25$ for $87\%$ and $95\%$ aerogels, respectively),
622: very close to the value $\mu_{spi}=-5.249$ corresponding to the liquid (mean-field) spinodal
623: of the bulk fluid at $T/T_c=0.45$. This shows that the perturbation induced by the
624: solid is too small to displace significantly the liquid spinodal. On
625: the other hand, the contact with the ambiant vapor at the surface of the gel has a
626: major influence, as illustrated in Fig. 12 by typical desorption
627: isotherms calculated in the presence of a free surface. One can see that when decreasing the chemical potential
628: all the curves exhibit a pronounced drop much before vapor
629: bubbles appear in the bulk of the material. This is a clear
630: indication that the mechanism of desorption is due to the surface.
631:
632: \begin{figure} [t]
633: \begin{center}
634: \resizebox{6cm}{!}{\includegraphics{fig13.eps}}
635: %\vspace{.4cm}
636: \caption{Invading vapor domain $V_t$ (in dark) in a $87\%$
637: aerogel at $\mu=-4.706$ ($L=100$). $10\%$ of the fluid has drained out. The gas
638: reservoir located at the bottom of the box and the aerogel are not shown.}
639: \end{center}
640: \end{figure}
641:
642: \begin{figure}[b]
643: \begin{center}
644: \resizebox{6cm}{!}{\includegraphics{fig14.eps}}
645: %\vspace{.4cm}
646: \caption{Same as Fig. 13 but for a $95\%$ aerogel at $\mu=-4.220$. }
647:
648: \end{center}
649: \end{figure}
650:
651: The isotherms shown in Fig. 12 consist of many steps of varying size,
652: but it also appears that one step is significantly
653: larger than the other ones in most of the $95\%$ samples,
654: a feature that is not present in the isotherms of the $87\%$ aerogel.
655: This suggests that the growth morphology of the invading
656: vapor domain may change with the porosity. This is confirmed by the snapshots
657: displayed in Figs. 13 and 14 that show the invading vapor domain
658: in two typical $87\%$ and $95\%$ samples when about $10\%$ of the fluid has
659: drained out of the aerogel.
660: This is just after the onset of the sharp drop in the isotherms. As in the case of adsorption, we find that
661: the distribution of the $\rho_i$'s is bimodal in
662: this range of chemical potentials, with
663: most of the fluid sites having a density lower than $0.05$ or larger than
664: $0.95$. One can thus identify unambiguously the emptied
665: sites. Although the volume occupied by the vapor is
666: the same in the two samples (about $200000$ sites), the morphology of the domain is remarkably
667: different. In the $87\%$ aerogel, the vapor domain exhibits an intricate isotropic
668: structure that resembles that produced in
669: invasion percolation. In contrast, the domain looks compact with a self-affine
670: interface in the lighter aerogel. This is very similar to the two regimes observed in the $T=0$ RFIM when an interface separating two magnetic domains is driven by an external field\cite{JR1992}: when the disorder is large, the interface forms a self-similar pattern with a large-scale structure characteristic of percolation, whereas the growth is compact and the domain wall forms a self-affine fractal surface at intermediate degrees of disorder (note that the use of the body-centered cubic lattice has allowed us to suppress the faceted growth regime that is observed at $T/T_c=0.45$ in the $95\%$ porosity aerogel on the simple cubic lattice\cite{note6}; this regime is an artifact of the lattice description).
671:
672:
673: In order to confirm the existence of two growth regimes and to determine
674: the actual behavior in the thermodynamic limit, a finite-size scaling
675: analysis of the desorption isotherms is required. Average isotherms
676: calculated for different system sizes are shown in Fig. 15. By analogy
677: with the problem of a driven interface in the $T=0$ RFIM, one expects
678: that the total volume $V_t(\mu)$ of the invading vapor domain shows a
679: power law divergence at the depinning threshold $\mu_c$. Then, assuming
680: that the only relevant length scales near $\mu_c$ are the sytem size
681: $\L$ and a single correlation length $\xi\sim[(\mu -\mu_c)/\mu_c]^{-\nu}$, the
682: dependence of $V_t$ on sytem size should be described by the scaling
683: form $L^{D_f}g(x)$ where $D_f$ is the fractal dimension
684: characterizing the domain and $g$ is a universal function of the
685: scaling variable $x=L^{1/ \nu}( \mu -\mu_c)/\mu_c$\cite{JR1992}.
686:
687: In both aerogels, one observes important finite-size
688: effects (see Fig. 15), but, unfortunately, they are not only due to the existence of a diverging
689: correlation length in the system but also to boundary effects. There is indeed an initial regime in which
690: desorption is due to the draining of large crevices at the surface of the gel where the fluid is in direct
691: contact with the ambiant vapor. For a given porosity, the number of
692: these crevices is proportional to $L^2$ and the contribution to the fluid
693: density is thus proportional to $1/L$. It turns out that this initial regime
694: extends to rather low values of the chemical potential ($\mu \simeq -4.6$ and
695: $\mu \simeq -4.15$ for the $87\%$ and $95\%$ aerogels, respectively), so that
696: the scaling region around the depinning threshold is too small to be
697: studied properly and to extract the values of the critical exponents (such an effect is not present
698: in the numerical studies of domain growth in the standard $T=0$
699: RFIM\cite{KR2000} because there is a different random field on {\it
700: each} site of the lattice and no equivalent of the crevices). We
701: note, however, that the curves in Fig. 15(b) have a common
702: intersection at $\mu\simeq -4.24$, in contrast with those in Fig. 15(a). This is
703: consistent with the scaling ansatz for the volume of the invading
704: vapor domain with $D_f=3$ for the $95\%$ aerogel and
705: $D_f<3$ for the $87\%$ aerogel. We thus conclude that the
706: desorption seems to be discontinuous in the first case and gradual
707: (percolation-like) in the second one.
708:
709: This leads to the isotherms shown in Fig. 5. Their shape resembles that of the experimental curves
710: in Figs. 4(b) and 4(c) of Ref.\cite{TYC1999}. The minor differences
711: can be rationalized:
712: the experimental isotherm in the $87\%$ aerogel (Fig. 4(c) in
713: Ref.\cite{TYC1999}) does not exhibit a
714: sharp kink at $\mu_c$, but this rounding may be due to the activated
715: processes that are neglected in the present treatment;
716: Fig. 4(b) in Ref.\cite{TYC1999} corresponds to a $98\%$ aerogel, which
717: probably explains why the shape of the hysteresis is more rectangular
718: than in the present Fig. 5(b).
719:
720:
721: \begin{figure}[t]
722: \begin{center}
723: \resizebox{8cm}{!}{\includegraphics{fig15.ps}}
724: %\vspace{.4cm}
725: \caption{Average desorption isotherms in $87\%$ (a) and $95\%$ (b) aerogels for different system sizes. The number of gel realizations
726: ranges from $1000$ ($L=25$) to $100$ ($L=100$) for the $87\%$ aerogel,
727: and from $1000$ ($L=35$) to $200$ ($L=100$) for the $95\%$ aerogel. The
728: arrow in (b) indicates the common intersection of the curves.}
729: \end{center}
730: \end{figure}
731:
732:
733: \subsection{Equilibrium }
734:
735: In order to determine the equilibrium isotherms, one has to find,
736: for each value of $\mu$, the lowest lying state(s) among all the
737: metastable states obtained from
738: Eqs. (3). A complete enumeration of these states is, however, an
739: impossible numerical task, and like in previous
740: work\cite{K2001,K2002}, we have only
741: calculated a limited number of states that, hopefully, can
742: provide a good approximation of the equilibrium isotherm. This can be
743: checked {\it a posteriori} by using the Gibbs adsorption equation
744: $\rho_f=-\partial(\Omega/N)/\partial\mu$ which is only satisfied by the equilibrium
745: curve\cite{K2001}. In Refs.\cite{K2001,K2002}, we searched for metastable states inside the
746: hysteresis loop by starting the iterative procedure with initial
747: configurations corresponding to uniform fillings of the lattice (with $\rho_i^{(0)}=\rho$
748: varying between $0$ and $1-p$). This method, however, does not
749: converge in dilute aerogels because the structure is very
750: inhomogeneous and correlated, at least for distances smaller than
751: $\xi_G$. It is then likely that all metastable fluid configurations are
752: also very inhomogeneous and cannot be reached iteratively from
753: initial uniform fillings.
754: We have thus searched for metastable states by calculating desorption and adsorption scanning curves, i.e., by performing incomplete filling or draining of the aerogel and then reversing the sign of the evolution of the chemical potential.
755: For convenience, these (very long) calculations are done
756: in systems with periodic boundary conditions in all directions, i.e.,
757: in the absence of an interface with the reservoir\cite{note7}.
758:
759: \begin{figure}[t]
760: \begin{center}
761: \resizebox{8cm}{!}{\includegraphics{fig16.ps}}
762: %\vspace{.4cm}
763: \caption{Typical desorption scanning curves (points) and equilibrium isotherms
764: (solid lines) in $87\%$ (a) and $95\%$ (b) aerogels (L=70).}
765: \end{center}
766: \end{figure}
767:
768: Some typical desorption scanning curves are shown in Figs. 16(a) and
769: (b). In both cases, the top curve is the major desorption branch, and by
770: comparing with the isotherms shown in Figs. 12 or 15, it is clear that
771: the part of this branch extending to low $\mu$'s and corresponding to
772: liquid-like metastable states that are isolated from the other states
773: is an artifact coming from the absence of the interface with the
774: reservoir. We also notice that there are no scanning curves in the
775: upper part of the hysteresis loop for the $95\%$ sample. This is
776: because there is a jump in the adsorption isotherm and, thus, there are no intermediate states from which
777: desorption can be started. Therefore the states contained in this region of the
778: plane $(\rho_f,\mu)$ cannot be reached by performing scanning
779: trajectories\cite{note8}. (Further work is clearly needed to determine if there are no metastable
780: states at all in this region or if the states can be obtained by other
781: means, for instance by changing the temperature.)
782:
783: \begin{figure}[t]
784: \begin{center}
785: \resizebox{8cm}{!}{\includegraphics*{fig17.eps}}
786: %\vspace{.4cm}
787: \caption{Check of thermodynamic consistency along the adsorption,
788: desorption, and equilibrium isotherms for a $87\%$ porosity aerogel
789: ($L=70$). (a) adsorption and desorption (b) equilibrium. Symbols:
790: average fluid density $\rho_f$. Dashed curves: related quantity obtained
791: by differentiating the corresponding grand potential with respect to
792: the chemical potential.}
793: \end{center}
794: \end{figure}
795:
796: \begin{figure}[b]
797: \begin{center}
798: \resizebox{8cm}{!}{\includegraphics*{fig18.eps}}
799: %\vspace{.4cm}
800: \caption{Same as Fig. 17 but for a $95\%$ aerogel.}
801: \end{center}
802: \end{figure}
803:
804: Fig. 16 also shows the approximate equilibrium isotherms obtained by
805: selecting, for each value of $\mu$, the state $\alpha$ that gives the lowest value of the grand
806: potential, as calculated from Eq. (4). We have checked that taking into
807: account the additional metastable states
808: obtained from the adsorption scanning curves did
809: not change significantly the results (as in Ref.\cite{K2002}, we have also
810: checked that keeping all solutions with a weighting factor equal to the
811: Boltzmann factor gives the same isotherms). As illustrated
812: in Figs. 17 and 18, the Gibbs adsorption equation is very well
813: verified along these curves. This indicates that we have indeed obtained a good
814: approximation of the true equilibrium isotherms. In contrast,
815: thermodynamic consistency is strongly violated along the adsorption
816: and desorption branches. The presence of (delta) peaks
817: in $\partial\Omega/\partial \mu$ also shows that the grand potential $\Omega$ changes
818: discontinuously (as the fluid density $\rho_f$) during adsorption and desorption in
819: a single finite sample. On the other hand, $\Omega$ is continuous (but
820: $\rho_f$ is discontinuous) along the equilibrium isotherm, as already noticed in Ref.\cite{K2002}.
821:
822: One can see from Figs. 16(a) and 16(b) that the equilibrium behavior
823: is different in the $87\%$ and $95\%$ porosity aerogels.
824: In particular, there is a large final jump in the isotherm
825: of Fig. 16(b). The same feature
826: exists in all samples, but in order to conclude on the actual behavior
827: in the infinite system one has again to analyze the size-dependence of
828: the average isotherms. This is shown in Fig. 19. For the $87\%$ aerogel, there is almost no
829: size-dependence and it is clear that no
830: transition occurs when $L\to \infty$. On the other hand, for
831: the $95\%$ aerogel, the isotherms become steeper as $L$ increases and there is
832: a rather well defined common intersection at $\mu_t\simeq -4.17$. This is a
833: clear indication that a phase transition does occur in the
834: thermodynamic limit. However, we have not
835: succeeded in obtaining a satisfactory collapse of the
836: isotherms using the scaling reduced variable
837: $L^{3/2}[\mu-\mu_t(L)]/\mu_t(L)$ as was done in Ref.\cite{K2002} in the case
838: of a random solid. The origin of this
839: problem is still unclear, but we suspect that the system sizes are
840: again too small. Indeed, we note that the maximal slope increases
841: rather weakly between $L=35$ and $L=70$, but has a size-dependence
842: consistent with the exponent $3/2$ between $L=70$ and
843: $L=100$. We thus conclude that our results are consistent with a discontinuous jump in
844: the thermodynamic limit, as indicated in Fig. 5(b). This corresponds
845: to a true equilibrium liquid-vapor phase separation in the system.
846:
847:
848: \begin{figure}[t]
849: \begin{center}
850: \resizebox{8cm}{!}{\includegraphics{fig19.eps}}
851: %\vspace{.4cm}
852: \caption{Average equilibrium isotherms in $87\%$ (a) and $95\%$ (b) aerogels for different system sizes. The number of gel realizations
853: ranges from $500$ ($L=25$) to $20$ ($L=100$) for the $87\%$ aerogel,
854: and from $2000$ ($L=35$) to $200$ ($L=100$) for the $95\%$ aerogel.}
855: \end{center}
856: \end{figure}
857:
858: \section{Summary and conclusion}
859:
860: In this paper, we have proposed an interpretation of the hysteretic
861: behavior observed in the experiments of $^4$He adsorption in light silica
862: aerogels. The overall shape of the experimental hysteresis
863: loops is well described by our theoretical model and we have been
864: able to reproduce the dramatic influence of porosity. In this interpretation, the disordered character of
865: the aerogel structure plays an essential role, the porosity $p$ being the tunable parameter that controls
866: the amount of disorder in the system. The history-dependent behavior is thus associated to the presence of
867: many metastable states in which the system may be trapped and that prevent thermal equilibration
868: at low temperatures. The most
869: important conclusion of our study is that adsorption and desorption
870: obey to different mechanisms and may be gradual or discontinuous,
871: depending on the porosity. Adsorption is insensitive to the presence of the interface between the solid and the
872: external vapor, and the change in the shape of the isotherm from smooth
873: to discontinuous as $p$ increases from $87\%$ to $95\%$ has been related to the appearance
874: of a infinite avalanche occuring in the bulk of the system, similar to
875: what happens in the $T=0$ RFIM below the critical
876: disorder\cite{S1993}. In contrast, desorption
877: is triggered by the presence of the outer surface and an associated
878: depinning transition. The change in the shape of the
879: isotherm is then related to a change in the morphology of the
880: invading vapor domain from percolation-like to compact (note however
881: that the mechanism
882: may be different in aerogels of lower porosity\cite{D2003}).
883:
884:
885: One must keep in mind that our calculations are based on local-mean
886: field theory in which disorder-induced fluctuations are properly
887: accounted for but thermal
888: fluctuations are neglected. This appears to be a reasonable assumption
889: at very low temperatures where the time scale associated to thermally
890: activated processes is much larger than the experimental time
891: scale. In this case, both the adsorption and desorption branches are
892: metastable and we have shown that the true equilibrium isotherm
893: (which probably cannot be reached experimentally) is
894: somewhere in between: this isotherm also changes from gradual to
895: discontinuous as $p$ increases. The situation is different at higher temperature (in
896: the vicinity of $T_c$) since true equilibrium behavior without
897: hysteresis has been
898: observed experimentally\cite{WC1990,WKGC1993}.
899: It will be therefore of interest to study the influence of temperature
900: on the hysteretic behavior and to understand how the dynamics of relaxation
901: towards equilibrium changes with $T$ (see, e.g., Ref.\cite{WM2003} for
902: a recent study of the relaxation behavior associated to capillary
903: condensation in a lattice model of Vycor).
904:
905: Finally, we suggest that the scenario for filling and draining
906: described in this work could be tested in more detail by
907: performing experiments in a series of aerogels of gradually increased
908: porosity. If our interpretation of the adsorption process in
909: terms of avalanches is correct, there should exist a value of
910: the porosity for which the isotherm becomes critical and the
911: distribution of avalanche sizes follows a power law with well-defined
912: critical exponents\cite{S1993}. Using superfluid instead of normal
913: $^4$He could perhaps allow to observe distinct avalanche events in
914: the aerogel and to study their statistical properties. Such a study has
915: been performed recently in the nanoporous material Nucleopore by a
916: capacitance technique\cite{LH2001}. One could also
917: check that the draining of the aerogel starts from the surface and
918: that the growth of the invading vapor domain obeys different
919: regimes. Related studies have been for instance carried out in Vycor using
920: ultrasonic attenuation and scattering
921: techniques\cite{PLAHDW1995,K2000}. Such experiments would give a
922: definite answer to the long-standing question about the nature of
923: hysteresis in fluid adsorption in disordered porous media.
924:
925:
926:
927: \acknowledgments
928: The Laboratoire de Physique Th\'eorique des Liquides is the UMR 7600 of
929: the CNRS. We thank R. Jullien for providing us with his lattice DLCA algorithm
930: to build the model aerogel.
931:
932: \begin{thebibliography}{10}
933: %\begin{references}
934: \bibitem{BG1983} F. Brochard and P.G. de Gennes, J. Phys. Lett. {\bf 44}, L785 (1983); P. G. de Gennes,
935: J. Phys. Chem. {\bf 88}, 6469 (1984).
936: \bibitem{WC1990} A. P. Y. Wong and M. H. W. Chan, Phys. Rev. Lett. {\bf 65}, 2567 (1990).
937: \bibitem{LMPMMC2000} L. B. Lurio, M. Mulders, M. Paerkau, M. Lee, S. G. J. Mochrie, and M. H. W. Chan, J. Low Temp. Phys. {\bf121}, 591 (2000).
938: \bibitem{WKGC1993} A. P. Y. Wong , S. B. Kim , W. I. Goldburg, and M. H. W. Chan, Phys. Rev. Lett. {\bf 70}, 954 (1993).
939: \bibitem{BFC1997} A. E. Bailey, B. J.Frisken, and D. S. Cannell, Phys. Rev. E {\bf 56}, 3112 (1997).
940: \bibitem{B1998} see, e.g., D. P. Belanger in {\it Spin Glasses and
941: Random Fields}, edited by A. P. Young (World Scientific, Singapore, 1998), p. 251.
942: \bibitem{C1996} M. H. W. Chan, Czech J. of Phys. {\bf suppl. S6}, 2915 (1996).
943: \bibitem{TYC1999} D. J. Tulimieri, J. Yoon, and M. H. W. Chan, Phys. Rev. Lett. {\bf 82}, 121 (1999).
944: \bibitem{note2} It was recently reported, using a capacitive technique
945: (J. R. Beamish and T. Herman, Proceedings of LT23, to appear in Physica
946: B), that
947: hysteresis is present in the sorption isotherms of $^4$He in a 95\% porosity
948: aerogel up to $T=$5.155K and that the adsorption remained
949: gradual at all temperatures below $T_c$. It is not yet clear whether
950: these results, which challenge those of Ref.\cite{WC1990}, are due to
951: an insufficient equilibration of the system (very slow thermal
952: relaxation in the coexistence region was indeed observed) or to a
953: different preparation process of the aerogel. Hysteretic behavior in
954: the same range of temperatures has also been found using a low-frequency
955: mechanical oscillator (G. Gabay {\it et al.}, J. Low Temp. Phys. {\bf 121}, 585 (2000)).
956: \bibitem{E1967} D. H. Everett, in {\it The Solid-Gas Interface}, edited by E. A. Flood (Marcel Dekker, New York,
957: 1967), Vol. {\bf 2}, p. 1055.
958: \bibitem{BE1989} P. C. Ball and R. Evans, Langmuir {\bf 5} 714 (1989).
959: \bibitem{note3} We consider here hysteretic cycles that are observed
960: at very slow flow rates on filling and on draining when the system
961: exhibits no explicit time-dependence (after some transient time). This corresponds to the so-called rate-independent limit.
962: \bibitem{K2001} E. Kierlik, P. A. Monson, M. L. Rosinberg,
963: L. Sarkisov, and G. Tarjus, Phys. Rev. Lett. {\bf 87}, 055701 (2001).
964: \bibitem{K2002} E. Kierlik, P. A. Monson, M. L. Rosinberg, and G. Tarjus, J. Phys.: Condens. Matter. {\bf 14}, 9295 (2002).
965: \bibitem{R2003} M.L. Rosinberg, E. Kierlik, and G. Tarjus, Europhys. Lett. {\bf 62}, 377 (2003) .
966: \bibitem{S1993} J. P. Sethna, K. Dahmen, S. Kartha, J. A. Krumhansl, B. W. Roberts, and J. D. Shore, Phys. Rev. Lett. {\bf 70}, 3347 (1993); O. Perkovic, K. Dahnen, and J. P. Sethna, Phys. Rev. Lett. {\bf 75}, 4528 (1995) and cond-mat/9609072 (unpublished); K. Dahmen and J. P. Sethna, Phys. Rev. B {\bf 53}, 14872 (1996); O. Perkovic, K. Dahnen, and J. P. Sethna, Phys. Rev. B {\bf 59}, 6106 (1999).
967: \bibitem{B2000} A. Berger, A. Inomata, J. S. Jiang, J. E. Pearson, and S. D. Bader, Phys. Rev. Lett. {\bf 85}, 4176
968: (2000); J. Marcos, E. Vives, L. Manosa, M. Acet, E. Duman, M. Morin, V. Novak, and A. Planes, Phys. Rev. B {\bf 67}, 2244XXX (2003).
969: \bibitem{SLG1983} C. M. Soukoulis, K. Levin, and G. S. Grest,
970: Phys. Rev. B {\bf 28}, 1495 (1983); H. Yoshizawa and D. P. Belanger, Phys. Rev. B {\bf
971: 30}, 5220(1984); C. Ro, G. S. Grest, C. M. Soukoulis, and K. Levin, Phys. Rev. B {\bf
972: 31}, 1682 (1985); G. S. Grest, C. M. Soukoulis, and K. Levin, Phys. Rev. B {\bf
973: 33}, 7659 (1986); E. P. Raposo and M. D. Coutinho-Filho, Phys. Rev. B {\bf
974: 57}, 3495 (1998).
975: \bibitem{KKRT2001} see V. Krakoviack, E. Kierlik, M. L. Rosinberg, and G. Tarjus, J. Chem. Phys. {\bf 115}, 11289 (2001) for a different
976: treatment of liquid-vapor phase separation in aerogels using integral equation theory and the replica formalism
977: in the continuum. This method, however, does not allow the study of the hysteretic behavior.
978: \bibitem{WSM2001} H.-J. Woo, L. Sarkisov, and P. A. Monson, Langmuir
979: {\bf 17}, 7472 (2001).
980: \bibitem{UHR1995} K. Uzelac, A. Hasmy, and R. Jullien, Phys. Rev. Lett.
981: {\bf 74}, 422 (1995); C. Vasquez, R. Paredes, A. Hasmy, and R. Jullien, preprint cond-mat/0303470.
982: \bibitem{STC1999} R. Salazar, R. Toral, and A. Chakrabarti, J. Sol-Gel
983: Sci. and Technol. {\bf 15}, 175 (1999).
984: \bibitem{PRST1995} E. Pitard, M. L. Rosinberg, G. Stell, and G. Tarjus, Phys. Rev. Lett. {\bf 74}, 4361(1995).
985: \bibitem{KRTP1998} E. Kierlik, M. L. Rosinberg, G. Tarjus, and E. Pitard, Mol. Phys. {\bf 95}, 341 (1998).
986: \bibitem{M1983} P. Meakin, Phys. Rev. Lett. {\bf 51}, 1119
987: (1983); M. Kolb, R. Botet, and R. Jullien, Phys. Rev. Lett. {\bf 51},
988: 1123 (1983).
989: \bibitem{H1994} A. Hasmy, E. Anglaret, M. Foret, J. Pelous, and R. Jullien, Phys. Rev. B {\bf 50}, 6006 (1994).
990: \bibitem{note4} It is possible, however, that long-range
991: correlations in the connectivity of the gel network exist that are not
992: captured by the two-point function $g(r)$\cite{UHR1995}. This for
993: instance may explain that no crossover to bulk-like critical behavior
994: is observed in the superfluid transition of $^4$He in aerogel (Yoon {\it et al.},
995: Phys. Rev. Lett. {\bf 80}, 1461 (1998)).
996: \bibitem{L1998} M. Lach-hab, A. E. Gonzales, and E. Blaisten-Barojas,
997: Phys. Rev. E {\bf 57}, 4520 (1998).
998: \bibitem{LKMP2000} G. Lawes, S. C. J. Kingsley, N. Mulders, and J. Parpia,
999: Phys. Rev. Lett. {\bf 84}, 4148 (2000).
1000: \bibitem{PP1999} J. V. Porto and J. M. Parpia,
1001: Phys. Rev. B. {\bf 59}, 14583 (1999).
1002: \bibitem{MVS1989} U. Marini Bettolo Marconi, and F. van Swol,
1003: Phys. Rev. A {\bf 39}, 4109 (1989); A. Papadopoulou, F. van Swol, and
1004: U. Marini Bettolo Marconi, J. Chem. Phys. {bf 97}, 6942 (1992).
1005: \bibitem{PLAHDW1995} J. H. Page, J. Liu, B. Abeles, E. Herbolzheimer, H. W. Deckman, and D. A. Weitz, Phys. Rev. E {\bf 52}, 2763 (1995).
1006: \bibitem{GBBC2003} A. I. Golov, I. B. Berkutov, S. Babuin, and D. J. Cousins, to appear in Physica B (2003).
1007: \bibitem{PV2003} F. J. Perez-Reche and E. Vives, Phys. Rev. B {\bf 67}, 134421 (2003).
1008: \bibitem{note5} Strictly speaking, the disorder-induced critical
1009: transition studied by Sethna and co-workers\cite{S1993} only exists at
1010: $T=0$. In LMFT however, and more generally when thermal fluctuations
1011: can be neglected, this is again a well-defined transition and it is
1012: reasonable to assume that is characterized by the $T=0$ critical
1013: exponents. It is then easy to check that the finite-size scaling
1014: exponent $\chi$ identifies with the combination $\beta\delta/\nu$ of the exponents
1015: introduced in Ref.\cite{S1993}. This is based on the hypothesis that
1016: close to $p_c$ (for $p\geq p_c$) and $\mu_c=\mu_t(p_c)$ the discontinuous
1017: change in $\rho_f $ is described by the scaling form
1018: $\Delta\rho_f(L,p,\mu)=r^{\beta}F_{-}(rL^{1/\nu},|h|/r^{\beta\delta})$ where $r=(p-p_c)/p_c$,
1019: $h=\mu-\mu_c$ and $F_{-}$ is a universal scaling funtion. Defining a new
1020: universal scaling function ${\tilde F}_{-}(rL^{1/\nu},|h|L^{\beta \delta/
1021: \nu})=(rL^{1/\nu})^{\beta} F_{-}(rL^{1/\nu},|h|/r^{\beta\delta})$, this can be rewritten
1022: as $\Delta\rho_f(L,p,\mu)=L^{-\beta/ \nu}{\tilde F}_{-}(rL^{1/\nu},|h|L^{\beta \delta/ \nu})$.
1023: Hence, $\partial \Delta\rho_f(L,p,\mu)/ \partial \mu\propto L^{2-\eta}\partial {\tilde F}_{-}(rL^{1/\nu},|h|L^{\beta
1024: \delta/ \nu}) / \partial h$ with $2-\eta=(\beta\delta -\beta)/\nu$. According to Ref.\cite{S1993},
1025: one has $1/\nu=0.71 \pm 0.09$, $\beta=0.035 \pm 0.028$, $\beta\delta=1.81\pm 0.32$, and
1026: $\eta=0.73\pm 0.28$. Note that in Fig. 11, we have neglected the dependence
1027: on the distance to $p_c$ and used $\chi=\beta\delta/ \nu\simeq 2-\eta$. In fact, according
1028: to Ref.\cite{PV2003}, the scaling behavior of the magnetisation jump
1029: in the $T=0$ RFIM could be more complicated than the one predicted by Sethna and co-workers.
1030: \bibitem{T1996} B. Tadic, Phys. Rev. Lett. {\bf 77}, 3843 (1996).
1031: \bibitem{CR1988} M. Cieplak and M. O. Robbins, Phys. Rev. Lett. {\bf 60}, 2042 (1988).
1032: \bibitem{KR2000} see, e.g., B. Koiller and M. O. Robbins,
1033: Phys. Rev. {\bf B 62}, 5771 (2000) and references therein.
1034: \bibitem{note6} In the absence of gel, there is no
1035: threshold for the onset of motion of the initial liquid-vapor
1036: interface parallel to the $[100]$ plane of the bcc lattice, i.e., the interface moves
1037: along the $x$-axis for $\mu<\mu_{sat}=-4$ at all temperatures. The actual
1038: self-affine interface, however, has various orientations at short-length scale.
1039: We found numerically that, at $T/T_c=0.45$, a portion of the
1040: interface parallel to the $[110]$ plane starts to
1041: move at $\mu= -4.11$. Since the
1042: drop in the desorption isotherms of Fig. 12 occurs at lower
1043: values of $\mu$, the pinning of the interface is due to disorder and not
1044: to lattice effects.
1045: \bibitem{JR1992} H. Ji and M. O. Robbins, Phys. Rev. B {\bf 46}, 14519
1046: (1992).
1047: \bibitem{note7} In the thermodynamic limit, the boundary condition
1048: (presence or absence of an interface) is not expected to influence the
1049: equilibrium phase behavior in the bulk of the material. As we have seen
1050: before, this reasoning does not apply to metastable states and
1051: out-of-equilibrium behavior.
1052: \bibitem{note8} The presence of a jump in the adsorption isotherm also
1053: implies that there are no adsorption
1054: scanning curves in the upper part of the hysteresis loop. Indeed, in
1055: the absence of interface with the
1056: reservoir, these curves are obtained by re-adsorbing from the
1057: metastable states located along the last desorption scanning curve.
1058: \bibitem{D2003} F. Detcheverry, E. Kierlik, M.L. Rosinberg, and
1059: G. Tarjus, in preparation.
1060: \bibitem{WM2003} H.-J. Woo and P. A. Monson, Phys. Rev. E {\bf 67}, 041207 (2003).
1061: \bibitem{LH2001} M. P. Lilly and R. B. Hallock, Phys. Rev. B {\bf
1062: 64}, 024516 (2001); M. P. Lilly, A. H. Wooters, and R. B. Hallock,
1063: Phys. Rev. B {\bf 65}, 104503 (2002).
1064: \bibitem{K2000} E. S. Kikkinides, M. E. Kainourgiakis,
1065: K. L. Stefanopoulos, A. Ch. Mitropoulos, A. K. Stubos, and
1066: N. K. Kanellopoulos, J. Chem. Phys. {\bf 112}, 9881 (2000).
1067:
1068: \end{thebibliography}
1069: %\end{references}
1070:
1071: \end{document}
1072:
1073:
1074:
1075: