1: %\documentclass[amsmath,preprint]{revtex4}
2: \documentclass[showpacs,amsmath,pre]{revtex4}
3: \usepackage{graphicx}
4: \usepackage{amsmath}
5: %\usepackage{dcolumn}
6:
7: \usepackage{epsfig}
8: \usepackage{graphics}
9: \usepackage{latexsym}
10: \usepackage{amsfonts}
11: \usepackage{amssymb}
12:
13: \usepackage{plain}
14:
15:
16: \begin{document}
17:
18: \title{Gas adsorption/desorption in silica aerogels: a theoretical study of scattering properties}
19:
20: \author{F. Detcheverry, E. Kierlik, M. L. Rosinberg, and G. Tarjus}
21: \affiliation{Laboratoire de Physique Th\'eorique de la Mati\`ere Condens\'ee, Universit\'e Pierre et
22: Marie Curie, 4 place Jussieu, 75252 Paris Cedex 05, France}
23:
24: %\date{\today}
25:
26: \begin{abstract}
27: We present a numerical study of the structural correlations associated to gas
28: adsorption/desorption in silica aerogels in order to provide a theoretical
29: interpretation of
30: scattering experiments. Following our earlier work, we use a coarse-grained lattice-gas description
31: and determine the nonequilibrium behavior of the adsorbed gas within a local mean-field analysis.
32: We focus on the differences between the adsorption and desorption mechanisms and
33: their signature in the fluid-fluid and gel-fluid structure factors as
34: a function of temperature. At low temperature, but still in the regime where the isotherms are continuous, we find that the adsorbed
35: fluid density, during both filling and draining, is correlated over distances that may be much larger than the
36: gel correlation length.
37: In particular, extended fractal
38: correlations may occur during desorption, indicating the existence of a ramified cluster of vapor filled
39: cavities. This also induces an important increase of the scattering intensity at small wave
40: vectors. The similarity
41: and differences with the scattering of fluids in other porous solids such as Vycor are discussed.
42:
43:
44:
45: \end{abstract}
46: \pacs{ 64.70.Fx, 75.60.Ej, 61;20.Gy,67.70.+n}
47: \maketitle
48:
49:
50:
51: \section{Introduction}
52:
53: In a series of recent papers\cite{DKRT2003,DKRT2004,DKRT2005}, we have presented a
54: comprehensive theoretical study of gas adsorption/desorption in
55: silica aerogels, revealing the microscopic mechanisms that underly the
56: changes in the morphology of the hysteresis loops with temperature and gel porosity.
57: In particular, we have shown that the traditional capillary condensation scenario
58: based on the independent-pore model\cite{GS1982} does not apply to aerogels, as a
59: consequence of the ``open" nature and interconnectedness of their microstructure. We have
60: found, on the other hand, that nonequilibrium phase transitions (that differ on adsorption
61: and desorption) are at the origin of the very steep isotherms
62: observed with $^4$He in high porosity gels at low temperature\cite{C1996,TYC1999}. In this work, we complete our
63: study by investigating the correlations within the adsorbed fluid and computing the
64: fluid-fluid and solid-fluid structure factors that can be measured
65: (at least indirectly) in scattering experiments. Scattering methods
66: (using x-rays, neutrons and visible light) are now frequently combined with thermodynamic
67: measurements for extracting information on the structure and the dynamics of the
68: adsorbed molecules and understanding the influence of solid microstructure
69: on fluid properties\cite{H2004}. In the case of $^4$He in aerogel, both small-angle x-ray scattering (SAXS)\cite{LMPMMC2000}
70: and light scattering measurements\cite{LGPW2004} have been recently performed along the sorption isotherms.
71: However, the interpretation of the scattered
72: intensity is always complicated by the fact that it contains several contributions that
73: cannot be resolved without assuming some mechanism for the sorption
74: process. For instance, in the case of a low porosity solid like Vycor, the evolution of the scattered intensity along the capillary rise
75: is usually interpreted in the framework of an independent-pore model, with the gas condensing
76: in pores of increasing size that are (almost) randomly distributed throughout the material\cite{LRHHI1994,PL1995,KKSMSK2000}. This
77: explains that long-range correlations are not observed during adsorption. On the other hand, we have shown that large-scale
78: collective events occur in aerogels, and this may have a significant influence on the scattering properties. Indeed, we shall
79: see in the following that scattering curves at low temperature may not reflect the underlying
80: microstructure of the gel. More generally, our main objective is to understand
81: how the different
82: mechanisms for adsorption and desorption reflect in the scattering properties as the temperature is changed. (Note that there has been
83: a recent theoretical study of this problem that is closely related to the present one\cite{SG2004}; there are, however,
84: significant differences that will be commented in due place.)
85: Particular attention will be paid to the `percolation invasion' regime that is predicted
86: to occur during the draining of gels of medium porosity (e.g. $87\%$) and that
87: manifests itself by the presence of fractal correlations. Such correlations have been
88: observed in Vycor\cite{LRHHI1994,PL1995,KKSMSK2000} and xerogel\cite{H2002}, but no experiment has been carried out so far to
89: detect a similar phenomenon in aerogels. We therefore hope that the present work will
90: be an incentive for such a study.
91: On the other hand, the influence of gel porosity on scattering properties will only be
92: evoked very briefly. In particular, for reasons that will be
93: explained below, the correlations along the steep (and possibly discontinuous due to nonequilibrium phase transitions)
94: isotherms observed in high porosity gels at low temperature are not investigated.
95:
96:
97: The paper is organized as follows. In section 2, the model and the theory
98: are briefly reviewed and the cart.texomputation of the correlation functions and the corresponding
99: structure factors is detailed. The numerical results are presented in
100: section 3. The relevance of our results to existing and future scattering experiments is
101: discussed in section 4.
102:
103: \section{Model and correlation functions}
104:
105: \subsection{Lattice-gas model}
106:
107: As discussed in previous papers\cite{DKRT2003,DKRT2004,DKRT2005},
108: our approach is based on a coarse-grained lattice-gas description
109: which incorporates the essential physical ingredients of
110: gel-fluid systems. The model Hamiltonian is given by
111: \begin{align}
112: {\cal H} = -&w_{ff}\sum_{<ij>} \tau_{i}\tau_{j} \eta_i \eta_j
113: -w_{gf}\sum_{<ij>
114: }[\tau_{i}\eta_i (1-\eta_j)+\tau_{j}\eta_j (1-\eta_i)]-\mu \sum_i \tau_i\eta_i
115: \end{align}
116: where $\tau_i=0,1$ is the fluid occupation variable ($i=1...N$)
117: and $\eta_i=1,0$ is the quenched random variable that describes the solid microstructure (when $\eta_i=0$, site $i$
118: is occupied by the gel; $\phi=(1/N)\sum_i\eta_i$ is thus the gel porosity). Specifically, we adress the case of base-catalyzed silica
119: aerogels (typically used in helium experiments) whose structure is well accounted for by a
120: diffusion limited cluster-cluster aggregation algorithm (DLCA)\cite{M1983}. In the Hamiltonian,
121: $w_{ff}$ and $w_{gf}$ denote respectively the fluid-fluid and
122: gel-fluid attractive interactions, $\mu $ is the fluid chemical
123: potential (fixed by an external reservoir), and the double summations run over all distinct pairs of
124: nearest-neighbor (n.n.) sites.
125:
126: Fluid configurations along
127: the sorption isotherms are computed using local mean-field theory (i.e. mean-field density
128: functional theory), neglecting
129: thermal fluctuations and activated processes (the interested reader is referred to
130: Refs.\cite{DKRT2003,DKRT2004,DKRT2005} for a detailed presentation of the
131: theory). As $\mu$ varies, the system visits a sequence of
132: metastable states which are local minima of the following grand-potential functional
133: \begin{align}
134: \Omega(\{\rho_i\})&=k_BT \sum_i[\rho_i\ln \rho_i+(\eta_i-\rho_i)\ln(\eta_i-\rho_i)]
135: -w_{ff} \sum_{<ij>}\rho_i\rho_j
136: -w_{gf}\sum_{<ij>}[\rho_i(1-\eta_j)+\rho_j(1-\eta_i)] -\mu\sum_i\rho_i\
137: \end{align}
138: where $\rho_i(\{\eta_i\})=<\tau_i\eta_i>$ is the thermally averaged fluid density at site $i$.
139: Earlier work has shown that this approach reproduces qualitatively
140: the main features of adsorption phenomena in disordered porous solids\cite{KMRST2001,SM2001}.
141:
142:
143: All calculations presented below were performed on a body-centered cubic lattice of
144: linear size $L=100$ ($N=2L^3$) with periodic boundary conditions in all directions (the
145: lattice spacing $a$ is taken as the unit length).
146: $L$ is large enough to faithfully describe gels with porosity $\phi\leq 95\%$, but
147: an average over a significant number of gel realizations is required to obtain a
148: good description of the correlation functions. In the following,
149: we use $500$ realizations. $w_{ff}$ is taken as the energy unit and temperatures are expressed in the reduced unit $ T^*=T/T_c$, where $T_c$ is the critical
150: temperature of the pure fluid ($T_c=5.195K$ for helium and $kT_c/w_{ff}=2$ in the theory).
151: The interaction ratio $y=w_{gf}/w_{ff}$ is equal
152: to $2$ so as to reproduce approximately the height of the hysteresis loop
153: measured with $^4$He in a $87\%$ porosity aerogel at $T=2.42K$\cite{DKRT2003,TYC1999}.
154: \subsection{Correlation functions and structure factors}
155:
156: In a scattering experiment performed in conjunction with gas adsorption/desorption, one
157: typically measures a (spherically averaged)
158: scattered intensity $I(q)$ which is proportional to a combination of the
159: three partial structure factors $S_{gg}(q), S_{ff}(q)$ and $S_{gf}(q)$, where $g$ and $f$
160: denote the gel and the fluid, respectively.
161: Whereas a macroscopic sample is usually considered as
162: isotropic and statistically homogeneous, our calculations are
163: performed in finite samples and on a lattice, and some work is
164: needed to obtain structure factors that can be possibly compared to experimental data.
165:
166: Let us first consider the gel structure factor. The calculation of
167: $S_{gg}(q)$ proceeds in several steps. As in the case of an off-lattice DLCA
168: simulation\cite{H1994}, we first compute the two-point
169: correlation function $g_{gg}({\bf r})=h_{gg}({\bf r})+1$ by performing a double
170: average over the lattice sites and the gel realizations,
171: \begin{align}
172: \rho_g^2 g_{gg}({\bf r}) = \frac{1}{N}\sum_{i,j} \overline{(1-\eta_i)(1-\eta_j)}
173: \delta_{{\bf r},{\bf r}_{ij}}-\rho_g \delta_{{\bf r},{\bf 0}}
174: \end{align}
175: taking care of the periodic boundary conditions. Here, ${\bf r}_{ij}={\bf r}_{i}-{\bf r}_{j}$, $\rho_g=1-\phi$ is the lattice fraction occupied by the gel, and the second term in the right-hand side
176: takes into account the fact that there are only one particle per site, which
177: yields the (point) hard-core condition, $g_{gg}({\bf r=0})=0$.
178: In this expression and in the following, the overbar denotes an average over different gel
179: realizations produced by the DLCA algorithm. The computation of $g_{gg}({\bf r})$
180: is efficiently performed by introducing the Fourier transform
181: $\rho_g({\bf q})=\sum_i (1-\eta_i)\exp(-2i\pi{\bf q.r_i}/N)$
182: where ${\bf q}$ is a vector of the reciprocal lattice, and by
183: using the fast Fourier transform (FFT) method\cite{MLW1996,SG2004}. This reduces the
184: computational work to $O(N \ln N)$ instead of $O(N^2)$ when the direct real-space
185: route is used. (The same method is applied to the other correlation functions.)
186:
187: In a second step, we ``sphericalize" the correlation function by collecting the values having same
188: modulus of the argument ${\bf r}$,
189: \begin{align}
190: g_{gg}(r)=\frac{ \sum_{{\bf r}'} g_{gg}({\bf r}') \delta_{r,r'}}
191: { \sum_{{\bf r}'} \delta_{ r,r'}} \ .
192: \end{align}
193: Finally, instead of storing the values of $g_{gg}(r)$ for all possible distances $r_{ij}$ on the lattice between
194: $d=a\sqrt{3}/2$, the
195: nearest-neighbor distance, and $L/2$, we bin the data with a spacing $\Delta r=0.05 $ and
196: interpolate linearly between two successive points (the restriction to
197: $r<L/2$ avoids boundary artefacts). Moreover, we impose the
198: ``extended" hard-core condition $g_{gg}(r)=0$ for $r<d$, in line with our interpretation of
199: a gel site as representing an impenetrable silica particle\cite{DKRT2003}.
200: (In Ref. \cite{SG2004}, in contrast, the model is thought of as a discretization of
201: space into cells of the size of a fluid molecule and the gel particle ``radius" is varied
202: from $2$ to $10$ lattice spacings.)
203: Of course, the interpolation procedure does not completely erase
204: the dependence on the underlying lattice structure, especially at short distances.
205: Following Ref.\cite{H1994}, the structure factor is then computed using
206: \begin{align}
207: S_{gg}(q)= 1+ 4\pi\rho_g \int_0^{L/2} r^2 ( h_{gg}(r)-h_{gg}) \frac{\sin(qr)}{qr} dr
208: \end{align}
209: where $h_{gg}$ is a very small parameter adjusted such that
210: $S_{gg}(q)\rightarrow 0$ as $q\rightarrow 0$. Indeed, since the
211: DLCA aerogels are built in the ``canonical" ensemble with
212: a fixed number of gel particles $N_g=\rho_g N$, the following sum-rule holds:
213: \begin{align}
214: \rho_g\sum_{{\bf r}}g_{gg}({\bf r}) = N_g-1
215: \end{align}
216: which readily yields $S_{gg}(0)=1+\rho_g\sum_{{\bf r}} h_{gg}({\bf r})=0$ in Fourier space.
217: This trick allows one to obtain a reasonable continuation of $S_{gg}(q)$
218: below $q_{min}=2\pi/L$\cite{H1994}.
219:
220:
221: Similarly, the fluid-fluid two-point correlation function
222: $g_{ff}({\bf r})=1+h_{ff}({\bf r})$ is defined as
223: \begin{align}
224: \rho_f^2 g_{ff}({\bf r})=\frac{1}{N}\sum_{i,j}\overline{\langle\tau_i\eta_i\tau_j\eta_j\rangle}
225: \delta_{{\bf r},{\bf r}_{ij}}-\rho_f \delta_{{\bf r},{\bf 0}}
226: \end{align}
227: where $\rho_f=(1/N)\sum_i\overline{\langle\tau_i\eta_i\rangle}=(1/N)\sum_i\overline{\rho_i}$ is the
228: average fluid density and the sum is performed again over all lattice
229: sites to improve the statistics (for notational
230: simplicity, we have dropped the overbar on $\rho_f$).
231: Because of the double average over thermal fluctuations and over disorder, there
232: are two distinct contributions to $h_{ff}({\bf r})$, which are usually called ``connected"
233: and ``blocking" or ``disconnected" \cite{GS1992,RTS1994}, and which, in the present case, are given by
234: the expressions,
235: \begin{align}
236: &\rho_f^2 h_{ff,c}({\bf r})+\rho_f \delta_{{\bf r},{\bf 0}}=
237: \frac{1}{N}\sum_{i,j}\overline{[\langle\tau_i\eta_i\tau_j\eta_j\rangle
238: -\rho_i\rho_j]} \ \delta_{{\bf r},{\bf r}_{ij}}\\
239: &\rho_f^2 h_{ff,d}({\bf r})=\frac{1}{N}\sum_{i,j}[\overline{\rho_i\rho_j}-\rho_f^2]\
240: \delta_{{\bf r},{\bf r}_{ij}} \ .
241: \end{align}
242: In the pure fluid, $h_{ff,c}({\bf r})$ is just the standard connected pair
243: correlation function whereas $h_{ff,d}({\bf r})$ has no equivalent. It turns out, however, that only
244: $h_{ff,d}({\bf r})$ can be
245: computed along the sorption isotherms. Indeed, the quantity $\langle\tau_i\eta_i\tau_j\eta_j\rangle$ cannot be obtained in the framework of mean-field theory, and the only available route to $h_{ff,c}({\bf r})$ is via the ``fluctuation" relation\cite{RTS1994}
246: \begin{align}
247: \rho_f^2 h_{ff,c}({\bf r}_{ij})+\rho_f\delta_{{\bf r}_{ij},{\bf 0}}=
248: \frac{\partial^2 \beta \Omega}
249: {\partial (\beta \mu_i )\partial (\beta \mu_j)}=\frac{\partial\rho_i}{ \partial (\beta \mu_j)}
250: \end{align}
251: where $\mu_i$ is a site-dependent chemical potential\cite{note2}.
252: However, this relation only holds at equilibrium (like the Gibbs adsorption equation
253: $\rho_f=-(1/N)\partial \Omega/ \partial \mu$ discussed in Ref.\cite{DKRT2003}) and therefore
254: it cannot be applied along the hysteresis loop where the system jumps from one
255: metastable state to another. (In a finite sample, the grand potential changes discontinuously
256: along the adsorption and desorption desorption branches\cite{DKRT2003}.) We are thus forced to approximate
257: $h_{ff}({\bf r})$ by its disconnected part, $h_{ff,d}({\bf r})$\cite{note1}.
258: However, this may not be a bad approximation at low temperature
259: because the
260: local fluid densities $\rho_i$ are then very close to zero or one\cite{DKRT2003}, which likely implies that
261: $h_{ff,c}({\bf r})$ is a much smaller quantity than $h_{ff,d}({\bf r})$\cite{note4}.
262:
263: We then apply to $h_{ff}({\bf r})$ the same procedure as for $g_{gg}({\bf r})$, taking
264: radial averages and then performing a binning of the data and a linear interpolation.
265: There are no ``extended" hard-core in this case. Indeed, since the scale of
266: the coarse-graining is fixed by the size of a gel particle (typically, a few nanometers), a
267: lattice cell may contain several hundreds of fluid molecules which may be thus
268: considered as point particles.
269: $h_{ff}(r)$ is then also interpolated between $r=0$ and $r=d$.
270:
271: In a grand-canonical calculation, the number of fluid particles fluctuates from sample to sample, which implies the following sum-rule for the disconnected pair correlation function $h_{ff,d}({\bf r})$\cite{RTS1994} (and thus for $h_{ff}({\bf r})$ in our approximation)
272: \begin{equation}
273: \sum_{\bf r} h_{ff}({\bf r}) \simeq\sum_{\bf r}h_b({\bf r}) =N \frac{\overline{{\rho_f^2(\{\eta_i\})}}-\rho_f^2}{\rho_f^2}
274: \end{equation}
275: where $\rho_f(\{\eta_i\})=(1/N)\sum_i \rho_i$ is the average fluid density for a
276: given gel realization. This sum-rule can be used to also extrapolate $S_{ff}(q)$ below $q<2\pi/L$, using
277: \begin{align}
278: S_{ff}(q)= 1+ 4\pi\rho_f \int_0^{L/2} r^2 ( h_{ff}(r)-h_{ff}) \frac{\sin(qr)}{qr} dr
279: \end{align}
280: where $h_{ff}$ is adjusted so that
281: $S_{ff}(0)=1+ N [\overline{\rho_f^2(\{\eta_i\})}-\rho_f]/\rho_f$.
282:
283: Finally, the gel-fluid two-point correlation function $g_{gf}({\bf r})=1+h_{gf}({\bf r})$ is computed from
284: \begin{align}
285: \rho_g \, \rho_f \, g_{gf}({\bf r})=\frac{1}{N}\sum_{i,j}\overline{(1-\eta_i)\langle\tau_j\eta_j\rangle}
286: \delta_{{\bf r},{\bf r}_{ij}}
287: \end{align}
288: and then sphericalized, binned, and linearly interpolated (taking $g_{gf}(r)=0$
289: for $0<r<d/2$ since no fluid molecule can be found at a distance less than $d/2$
290: from the centre of a gel particle).
291: The cross structure factor $S_{gf}(q)$ is then obtained from
292: \begin{align}
293: S_{gf}(q)= 4\pi \sqrt{\rho_g\rho_f} \int_0^{L/2} r^2 ( h_{gf}(r)-h_{gf}) \frac{\sin(qr)}{qr} dr
294: \end{align}
295: where $h_{gf}$ is adjusted so as to satisfy the sum-rule $S_{gf}(q\rightarrow0)=0$
296: which again results from the absence of fluctuations in the number of gel particles.
297:
298:
299:
300: \section{Numerical results}
301:
302:
303: \subsection{Aerogel structure}
304:
305: We first concentrate on the case of the empty aerogel. We have already presented in Ref.\cite{DKRT2003} the pair correlation function $g_{gg}(r)$ for
306: several porosities between $87\%$ and $99\%$. These curves exhibit a shallow minimum that strongly depends on $\phi$
307: and whose position gives an estimate of the gel correlation lenght $\xi_g$, as
308: suggested in Ref.\cite{H1994}. A DLCA gel can
309: indeed be sketched as a disordered packing of ramified blobs with average size
310: $\xi_g$. For instance, $\xi_g$ varies approximately from $4$ to $10$ lattice spacings
311: as the porosity increases from
312: $87\%$ to $95\% $ (this is much smaller than the box size $L$, which ensures that
313: there are no finite-size artifacts in the calculation of the gel structure\cite{note5}).
314: Only the highest-porosity samples exhibit a significant power-law regime
315: $g_{gg}(r)\sim r^{-(3-d_f)}$
316: that reveals the fractal character of the intrablob correlations.
317: \begin{figure}[hbt]
318: \includegraphics*[width=9cm]{fig1.eps}
319: \caption{Gel structure factors $S_{gg}(q)$ obtained with the DLCA algorithm for different
320: porosities. From left to right: $\phi=0.99, 0.98, 0.97, 0.95, 0.92, 0.90, 0.87$. The dashed line has a slope $-1.9$. The arrow indicates the wavevector $q=2\pi/L\approx 0.063$. (Color on line).}
321: \end{figure}
322: Although we shall essentially focus in the following on the case of the $87\%$ porosity gel,
323: for the sake of completeness we show in Fig. 1
324: the evolution of the simulated gel structure factor with porosity.
325:
326: The curves closely resemble those obtained with the continuum model\cite{H1994}.
327: In particular, they exhibit the same damped oscillations at large $q$ that result from
328: the ``extended" hard-core condition $g_{gg}(r)=0$ for $r<d$ (the oscillations, however, are more significant in the continuum model). The range of the
329: linear fractal regime increases with porosity (it is almost non-existent in the $87\%$ gel) and corresponds asymptotically to
330: a fractal dimension $d_f\approx 1.9$ (the value $d_f\approx 1.87$ was obtained
331: in Ref.\cite{DKRT2003} from the $g_{gg}(r)$ plot for the $99\%$ aerogel). A characteristic feature of the curves is the existence of a
332: maximum at smaller wavevectors whose location $q_{m}$ decreases with porosity and correlates well with
333: $1/\xi_g$ ($q_{m} \sim 2.6/\xi_g$). This maximum is thus the Fourier-space signature of
334: the shallow minimum observed in $g_{gg}(r)$. (Note that varying the size $L$ has only a weak influence on the small $q$
335: portion of the curves for the $87\%$ and $95\%$ gels, which validates the continuation procedure used in Eq. 5.)
336:
337: To compute the scattering intensity $I(q)$ and compare to the results of small-angle
338: x-rays or neutron experiments, it is necessary to introduce a form factor $F(q)$
339: for the gel particles. One can use for instance the form factor of spheres with radius $R=d/2$,
340: $F(q)=3 [\sin(qR)-qR\cos(qR)]/(qR)^3$. The curve $I(q)=S_{gg}(q)F(q)^2$ then
341: differs from $S_{gg}(q)$ in the large-$q$ regime ($q/2\pi>d^{-1}$) where it follows the
342: Porod law $I(q)\sim q^{-4}$\cite{P1982}. On the other hand, the intermediate
343: ``fractal" regime ($\xi_g^{-1}<q/2\pi<d^{-1}$) where $S_{gg}(q)\sim q^{-d_f}$, and
344: the location $q_{m}$ of the maximum are essentially unchanged. By comparing the
345: value of $q_m$ in Fig. 1 with the actual value in the experimental curves
346: (displayed for instance in Ref.\cite{H1994}), we can thus fix approximately the
347: scale of our coarse-graining. For $\phi=95\%$, $q_m\approx 0.01 $\AA{}$^{-1}$ which
348: yields $a\approx 3$ nm, a reasonable value for base-catalysed silica aerogels which is in agreement
349: with the estimation of Ref.\cite{DKRT2003} obtained from the gel correlation
350: length $\xi_g$.
351: It is worth noting that the DLCA simulated structure factors present a more pronounced maximum at $q_m$ than the experimental curves $I(q)$, as already noticed in the literature\cite{H1994,OJ1995}. There are obviously large scale inhomogeneities in actual aerogels that are not reproduced in the simulations. Moreover, as emphasized in Refs.\cite{H1994,OJ1995}, there are some significant details that are neglected in the DLCA model, such as the rotational diffusion of the aggregates, their polydispersity and irregular form, and all kinds of possible restructurating effects.
352:
353: \subsection{Fluid structure during adsorption}
354: \begin{figure}[hbt]
355: \includegraphics*[width=8cm]{fig2.eps}
356: \caption{Average hysteresis loops in a $87\%$ porosity aerogel at $T^*=0.5$ and $0.8$
357: (from left to right).
358: The points along the adsorption and
359: desorption isotherms at $T^*=0.5$
360: indicate the values of the chemical potential for which the correlation
361: functions are computed. The desorption isotherm has been computed either in presence of an external reservoir (solid line) or by using the procedure described in section IIIC (dashed line) (Color on line).}
362: \end{figure}
363: As shown
364: in Refs.\cite{DKRT2003,DKRT2004,DKRT2005}, the elementary condensation events
365: (avalanches) that occur in 87\% porosity aerogel as the chemical potential is slowly varied
366: are always of microscopic size, whatever the temperature. This implies that the
367: adsorption isotherms are smooth in the thermodynamic limit or when averaging over
368: a large number of finite samples, as illustrated in
369: Fig. 2.
370: We have computed the correlation functions and the corresponding structure factors
371: for a number of points along the $T^*=0.5$ and $T^*=0.8$ isotherms, as indicated in
372: the figure. We first consider the lowest temperature.
373:
374: Figs. 3 and 4 show the evolution of the correlation functions $h_{ff}(r)$ and
375: $h_{gf}(r)$ with chemical potential.
376: One can see that the curves change significantly as $\mu$ increases
377: (note that the vertical scale in Fig. 3(a) is expanded so as to emphasize the presence of
378: a shallow minimum in the curves; accordingly, the values of $h_{ff}(r)$ near zero are not visible).
379: For very low values of the chemical potential (e.g. $\mu=-6$), $h_{ff}(r)$ looks
380: very much like
381: $h_{gg}(r)$, apart from a shift towards larger values of $r$ by a distance of about
382: $2$ lattice spacings. Indeed, as shown in our
383: earlier work\cite{DKRT2003,DKRT2004}, in the early stage of the adsorption process,
384: the adsorbed fluid forms a liquid film that coats the aerogel strands and whose
385: thickness is approximately one lattice spacing at low temperature. In consequence, the distribution
386: of the fluid particles follows the spatial arrangement of the aerogel, a feature
387: already observed in a continuum version of the model\cite{KKRT2001}.
388: The existence of the liquid film also reflects in the rapid decrease of $h_{gf}(r)$
389: with $r$, which indicates that the fluid is only present in the vicinity of the gel
390: particles (the fact that $h_{gf}(d)<h_{gf}(a)$ may be ascribed to the connectivity
391: of the gel: around a gel particle, there are always other gel particles - $2.5$ in average in the first shell - and the probability to find a fluid particle is thus suppressed).
392: \begin{figure}[hbt]
393: \includegraphics*[width=12cm]{fig3.eps}
394: \caption{ Fluid-fluid correlation function $h_{ff}(r)$ along the adsorption isotherm
395: in a $87\%$ porosity aerogel at $T^*=0.5$. (a): From top to bottom, the curves correspond to points $1$ to $8$ in Fig. 2; the dashed line is the gel correlation function $h_{gg}(r)$. (b): Magnification of (a) showing the evolution of the minimum as $\mu$ varies from $-4.55$ to $-4$ (points $5$ to $8$ in Fig. 2) (Color on line).}
396: \end{figure}
397: \begin{figure}[hbt]
398: \includegraphics*[width=12cm]{fig4.eps}
399: \caption{ Same as Fig. 3 for the cross correlation function $h_{gf}(r)$ (Color on line).}
400: \end{figure}
401: As $\mu$ increases, the magnitude of the fluid-fluid correlations decreases at small $r$
402: (the contact value decreases from $4.45$ to $0.15$) and the depth of the minimum
403: in $h_{ff}(r)$ decreases as it shifts to larger values of $r$
404: (its location varies from $5$ to $24$ as $\mu$ increases
405: from $-6$ to $-4.47$). The minimum disappears as the last voids
406: in the gel fill with liquid (note the difference in the vertical
407: scales of Figs. 3(a) and 3(b)), and finally, as one reaches saturation
408: (at $\mu_{sat}=-4$), the shape of the gel-gel correlation function $h_{gg}(r)$
409: is approximately recovered, in accordance with
410: Babinet principle\cite{P1982,note6}. A similar behavior is observed
411: in the cross correlation function $h_{gf}(r)$ in Fig. 4, but the minimum occurs at a
412: smaller distance (its location is approximately $r\approx 12$ when it disappears).
413:
414: Like for $g_{gg}(r)$, we may associate to the location of the mimimum in $h_{ff}(r)$ a length $\xi_{f}$ that
415: characterizes the correlations within the adsorbed fluid. The fact that $\xi_{f}$
416: becomes significantly larger than $\xi_g$ as the adsorption proceeds
417: shows that the fluid develops its own complex structure that does not reflect anymore
418: the underlying gel structure. This is likely in relation with
419: the fact that some of the condensation events studied in our
420: previous works\cite{DKRT2004,DKRT2005} extend much beyond the largest voids in
421: the aerogel\cite{note9}. It is worth noting that
422: these large avalanches (with a radius of gyration
423: $R_g\approx 12$\cite{DKRT2004,DKRT2005})
424: occur approximately in the same range of chemical
425: potential ($-4.5\leq\mu \leq-4.4$) where $\xi_f$ reaches its maximum (this also corresponds to the steepest portion of the
426: isotherm).
427:
428: \begin{figure}[hbt]
429: \includegraphics*[width=10cm]{fig5.eps}
430: \caption{ Fluid-fluid structure factor $S_{ff}(q)$ along the adsorption isotherm
431: in a $87\%$ porosity aerogel at $T^*=0.5$. The numbers refer to points $1$ to $8$
432: in Fig. 2. The dashed line is the gel structure factor $S_{gg}(q)$ and the dotted line
433: illustrates the influence of the continuation procedure for $q<2\pi/L$
434: (see \cite{note7}) (Color on line).}
435: \end{figure}
436: \begin{figure}[hbt]
437: \includegraphics*[width=10cm]{fig6.eps}
438: \caption{Same as Fig. 5 for the cross structure factor $S_{gf}(q)$. Note that the vertical
439: scale is not logarithmic because $S_{gf}(q)$ has negative values (Color on line).}
440: \end{figure}
441:
442: The corresponding structure factors $S_{ff}(q)$ and $S_{gf}(q)$ are shown in
443: Figs. 5 and 6, respectively\cite{note7}. The main feature in $S_{ff}(q)$ is the presence of a broad peak that grows and moves towards smaller wavevector as the fluid condenses in the
444: porous space. This peak is clearly associated to the minimum in $h_{ff}(r)$ (its location is approximately proportional
445: to $\xi_f^{-1}$) and it thus tells us the same story: the growing of a characteristic
446: length scale in the fluid along the capillary rise. The peak disappears in the last
447: stage
448: of the adsorption process and is then replaced by a plateau (see curve 7 in Fig. 5). Finally, at
449: $\mu=\mu_{sat}$, one recovers a structure factor that can be deduced from $S_{gg}(q)-1$
450: by a linear transformation, in accordance with
451: Babinet principle (there are no oscillations in $S_{ff}(q)$, however, because
452: the fluid-fluid hard-core diameter is zero).
453:
454: The evolution of the gel-fluid cross structure factor $S_{gf}(q)$ is somewhat different.
455: The peak is more prominent, as a consequence of the `no-fluctuation' condition
456: $S_{gf}(q=0)=0$ (this
457: feature may not be so pronounced in actual systems because of large-scale
458: fluctuations), and it is located at a larger wavevector than in $S_{ff}(q)$ (in line with
459: the corresponding locations of the minima in $h_{ff}(r)$ and $h_{gf}(r)$). The most substantial
460: difference with $S_{ff}(q)$ is that the amplitude of the peak starts to decrease much before
461: the end of the adsorption
462: process. The negative correlation observed at saturation is again due to
463: complementarity with the gel-gel structure\cite{SG2004}.
464:
465: We have repeated these calculations at $T^*=0.8$ in order to investigate the
466: influence of temperature. In the $87\%$ gel, $T^*=0.8$ is just below $T_h$,
467: the temperature at which the hysteresis loop disappears (see Fig. 2).
468: $h_{ff}(r)$ and $h_{gf}(r)$ still exhibit a minimum that moves towards larger $r$
469: upon adsorption. However, the characteristic length $\xi_f$, associated to the
470: minimum of $h_{ff}(r)$, does not exceed $14$ lattice spacings, indicating that
471: the size of the inhomogeneities in the fluid decreases with increasing $T$. A similar observation
472: was made in Ref.\cite{DKRT2004,DKRT2005} concerning the size of the avalanches
473: which become more compact at higher temperature and often
474: correspond to a condensation event occuring in a single cavity of the aerogel.
475: The shape of the corresponding structure factors does not change significantly with respect to the $T^*=0.5$ case,
476: but the amplitude is significanly reduced: the maximal amplitudes of the peaks in $S_{ff}(q)$ and $S_{gf}(q)$
477: are divided approximately by $5$ and $2$, respectively.
478:
479: As shown in Refs.\cite{DKRT2003,DKRT2004,DKRT2005}, temperature has a much more
480: dramatic influence on the adsorption process in gels of higher porosity. In particular,
481: at low enough temperature ($T<T_c(\phi)$ with $T_c^*(\phi)\approx 0.5$ in the $95\%$ gel\cite{DKRT2005}), a macroscopic
482: avalanche occurs at a certain value of the chemical potential, with the whole sample
483: filling abruptly, which results in a discontinuous isotherm
484: in the thermodynamic limit. In a finite system, the signature of a macroscopic avalanche is
485: a large jump in the fluid density whose location in $\mu$ fluctuates from sample to sample,
486: which results in a steep but smooth isotherm after an average over the gel realizations (one then has
487: to perform a finite-size scaling study to conclude on the proper behavior in the thermodynamic limit\cite{DKRT2003,DKRT2005}).
488: Within a grand canonical calculation, there is unfortunately no way to study the evolution of the structural
489: properties of the fluid during a macroscopic avalanche as this would require to consider intermediate fluid densities that
490: are inaccessible\cite{note8} (the situation would be different if the fluid density
491: was controlled instead of the chemical potential, as is done frequently in experiments\cite{TYC1999,WC1990}). All we can do is to study
492: the $95\%$ gel at a higher temperature where the adsorption is still gradual, for instance at $T^*=0.8$. In this case, no
493: qualitative change is found with respect to the case of the $ 87\%$ gel at $T^*=0.5$. Indeed, as emphasized in Ref.\cite{DKRT2005},
494: adsorption proceeds similarly in a high-porosity gel at high temperature and in a lower-porosity gel at low temperature. The
495: correlation length $\xi_f$ is somewhat larger in the $95\%$ gel (beyond $30$ lattice spacings) so that finite-size effects come into play (in particular,
496: it becomes problematic to extraplotate $S_{ff}(q)$ to $q=0$ so as to simulate the infinite-size limit). To go to lower temperatures, it
497: would be thus necessary to use a much larger simulation box, which would increase considerably the computational work. Note that one expects $h_{ff}(r)$ to decay algebraically at the critical temperature $T_c(\phi)$. Indeed, according to the analogy with the $T=0$ nonequilibrium random-field Ising model (RFIM), there should be only one important length scale in the system close to criticality, length scale which is proportional to the average linear extent of the largest avalanches\cite{DS1996}. At criticality, this correlation length diverges.
498:
499: \subsection{Fluid structure during desorption}
500:
501: As discussed in detail in Refs.\cite{DKRT2003,DKRT2004}, different mechanisms may be responsible for gas desorption in aerogels, depending
502: on porosity and temperature. In contrast with adsorption, the outer surface of the material where the adsorbed
503: fluid is in contact with the external vapor may play an essential role.
504: For instance, in the $87\%$ aerogel at $T^*=0.5$, the theory predicts
505: a phenomenon akin to percolation invasion : as $\mu$ is decreased from saturation,
506: some gas ``fingers" enter the sample and grow until they percolate at a certain
507: value of $\mu$, forming a fractal, isotropic cluster. The desorption then proceeds gradually via the growth of the gaseous domain. Accordingly, in the thermodynamic limit, the isotherm shows a cusp at the percolation threshold followed by a steep but continuous decrease
508: (the cusp is rounded in finite samples).
509:
510: The simulation of the desorption process thus requires the use of an explicit external reservoir adjacent to the gel sample, which of course introduces a severe anisotropy in the model and makes it difficult to calculate radially averaged correlation functions. To circumvent this problem, we have used another procedure where the desorption is not initiated by the interface with an external reservoir but triggered by the presence of
511: gas bubbles inside the material. We have indeed observed in our previous studies (see Fig. 16 in Ref.\cite{DKRT2003}) that the last desorption scanning curves (obtained by stopping the adsorption just before saturation and then decreasing the chemical potential) look very much like
512: the desorption isotherms obtained in presence of a reservoir (when averaging the
513: fluid density deep inside the aerogel, which gives a good estimate of the isotherm
514: in the thermodynamic limit). Near the end of the adsorption process, the remaining gaseous domain is composed of isolated bubbles which obviously play the same role as an external reservoir when the chemical potential is decreased. The advantage of initiating the desorption with these last
515: bubbles is that one can use the same geometry as during adsorption, with
516: periodic boundary conditions in all directions (moreover, using
517: small bubbles instead of a planar interface of size $L^2$ considerably suppresses
518: finite-size effects).
519:
520: In practice, the calculation has been performed by keeping five bubbles in each
521: sample. This implies that the chemical potential at which desorption is
522: initiated is slightly different in each sample (if one chooses the same $\mu$ in
523: all samples, some of them may be already completely filled with liquid). This number of bubbles results from a compromise: on the one hand, keeping a single bubble may not be sufficient to trigger the desorption process (in some samples, the growth of the bubble is hindered by the neighboring gel particles and the desorption occurs at a much lower value of the chemical potential); on the other hand,
524: keeping too many bubbles results in a too rapid growth of the gas domain.
525: As can be seen in Fig. 2, the isotherm obtained with this procedure is indeed very
526: close to the isotherm calculated in presence of an
527: external reservoir.
528:
529: The fluid-fluid and solid-fluid correlation functions computed at several points
530: along the desorption branch are shown in Figs. 7 and 8.
531: \begin{figure}[hbt]
532: \includegraphics*[width=10cm]{fig7.eps}
533: \caption{ Fluid-fluid correlation function $h_{ff}(r)$ along the desorption isotherm
534: in a $87\%$ porosity aerogel at $T^*=0.5$. The numbers refer to the points
535: in Fig. 2; the dashed line is the gel correlation function $h_{gg}(r)$. The vertical scale is expanded in the inset,
536: showing the very slow decrease of $h_{ff}(r)$ towards zero in the steepest portion of the isotherm (Color on line).}
537: \end{figure}
538: \begin{figure}[hbt]
539: \includegraphics*[width=10cm]{fig8.eps}
540: \caption{Same as Fig. 7 for the cross correlation function $h_{gf}(r)$ (Color on line). }
541: \end{figure}
542:
543: One can see that $h_{ff}(r)$ is dramatically changed with respect to
544: adsorption: from saturation down to $\mu\approx-4.7$ (curves $9$ to $16$), the function is
545: monotonically decreasing, showing no minimum. Moreover, as shown in the inset of Fig. 7, a long-range
546: tail is growing and $h_{ff}(r)$ may differ significantly
547: from zero at $r=L/2$\cite{note10}.
548: Although it is difficult to
549: characterize this decrease by a unique and well-defined correlation length (the curve cannot
550: be fitted by a simple function such as an exponential), it is clear that the range of the correlations
551: is small at the beginning of
552: the desorption process, then increases considerably, goes through a maximum in the
553: steeppest portion of the isotherm (corresponding approximately to point $13$
554: in Fig.2), and eventually decreases. As the hysteresis loop closes and the adsorbed phase
555: consists again of a liquid film coating the areogel strands, a shallow minimum
556: reappears in the curve, which is reminiscent of the underlying gel structure.
557:
558: In contrast, the changes in the cross-correlation function $h_{gf}(r)$ between
559: adsorption and desorption are very small (the function has only a
560: slightly longer tail during
561: desorption). It appears that the gel-fluid correlations depend
562: essentially on the average fluid density: for a given value of $\rho_f$,
563: they are almost the same on the two branches of the hysteresis loop.
564:
565: The calculation of the fluid-fluid structure factor $S_{ff}(q)$
566: is complicated by the fact that $h_{ff}(r)$ decreases very slowly to zero, and one cannot
567: use anymore the continuation procedure that forces the sum-rule, Eq. 11, to be satisfied
568: (the results for $q<2\pi/L$ change considerably with $L$, which shows that the resulting curve is not a good approximation of the
569: infinite-size limit).
570: Instead, we have only subtracted from $h_{ff}(r)$ its value at $r=L/2$ (setting $h_{ff}=h_{ff}(r=L/2)$
571: in Eq. 12) so as to avoid the large but spurious oscillations in $S_{ff}(q)$ that result
572: from the discontinuity at $L/2$ (there is still a discontinuity in the slope of
573: $h_{ff}(r)$ at $L/2$
574: that induces small oscillations in some of the curves of Fig. 9). It is clear that finite-size effects are
575: important in this case and the results for $q<2\pi/L$ must be considered with caution.
576: \begin{figure}[hbt]
577: \includegraphics*[width=10cm]{fig9.eps}
578: \caption{ Fluid-fluid structure factor $S_{ff}(q)$ along the desorption isotherm
579: in a $87\%$ porosity aerogel at $T^*=0.5$. The numbers and arrows refer to points $9$ to $18$
580: in Fig. 2. The dashed line is the gel structure factor $S_{gg}(q)$ (Color on line).}
581: \end{figure}
582:
583: The resulting fluid-fluid structure factors along the desorption isotherm are shown in Fig. 9. (We do not present the curves
584: for $S_{gf}(q)$ as they look very much like those in Fig. 6 with only a slightly broader peak.)
585: As could be expected, the structure factors computed just before and after the knee
586: in the isotherm (for $\mu>-4.67$)
587: are very different from those obtained during adsorption.
588: Firstly, the peak that was associated to the minimum in $h_{ff}(r)$ has now
589: disappeared and the small-$q$ intensity saturates to a value that is considerably
590: larger than the maximum value obtained during
591: adsorption (compare the vertical scales in Figs. 5 and 9). Secondly, as $\mu$ varies
592: from $-4.65$ to $-4.67$ (curves $11$ to $14$), there is a linear portion
593: in $S_{ff}(q)$ whose maximal extension is about one decade on a
594: log-log scale. On the other hand,
595: when $\mu$ is decreased further, the peak in $S_{ff}(q)$ is recovered and the curves become more similar to
596: the ones computed on the adsorption branch.
597:
598:
599: The linear regime in $S_{ff}(q)$ strongly suggests the presence of fractal
600: correlations\cite{note3}. However,
601: according to our previous studies\cite{DKRT2004}, it is only the {\it gaseous} domain that
602: should be a fractal object at the percolation threshold, as illustrated by the isotropic and strongly
603: ramified structure shown in Fig. 10.
604: \begin{figure}[hbt]
605: \includegraphics*[width=7cm]{fig10.ps}
606: \caption{Snapshot of the vapor domain in a $87\%$ gel sample during
607: desorption at $T^*=0.5$ and $\mu=-4.63$ (Color on line).}
608: \end{figure}
609: The correlation function $h_{ff}(r)$, on the other
610: hand, does not discriminate between a site representing a gaseous region
611: ($\rho_i\approx 0$ at low
612: temperature) and a site representing a gel particle ($\rho_i\equiv 0$). In order to
613: really show the existence of fractal correlations within the gas domain, one must consider either the `gas-gas'
614: correlation function (defining the quantity $\rho^{gas}_i=\eta_i-\rho_i$ which is
615: equal to $1$ only in the gas phase) or the complementary function that measures the correlations within
616: the dense (solid or liquid) phase (one then defines $\rho^{dense}_i=1-\rho^{gas}_i$).
617: The corresponding structure factor $S_{dd}(q)$ is the quantity that is measured experimentally
618: when using the `contrast matching' technique\cite{H2004}. It is related to
619: $S_{ff}(q)$ and $S_{gf}(q)$ by
620: \begin{equation}
621: (\rho_g+\rho_f)(S_{dd}(q)-1)=\rho_f(S_{ff}(q)-1)+\sqrt{\rho_g\rho_f}S_{gf}(q)+\rho_g(S_{gg}(q)-1) \ .
622: \end{equation}
623: $S_{dd}(q)$ is shown in Fig. 11 for $\mu=-4.65$\cite{note11}, in the region of the
624: knee in the desorption isotherm (point $11$ in Fig. 2). It clearly contains a linear
625: portion over almost one decade and can be very well represented in this range of wavevectors
626: by the fit\cite{FKS1986}
627: \begin{equation}
628: S_{dd}(q)\sim \frac{\sin\left[ (d_f-1) \tan^{-1}(ql)\right]}
629: { q \left[ l^{-2}+q^2 \right]^{(d_f-1)/2}}
630: \end{equation}
631: with $d_f=2.45$ and $l=17$, where $l$ is a crossover length that limits the fractal
632: regime at large distances. Note that the linear portion itself
633: has a slope $\approx -2.1$ (the above formula reproduces the right slope only
634: when $l$ is very large\cite{H1994}). An accurate determination of $d_f$ would
635: therefore require a much larger system and, at this stage, it is not possible to
636: decide if the fractal dimension is consistent with that of random percolation. In any
637: case, these results strongly
638: suggest that the gas domain exhibits fractal correlations during desorption, correlations which have
639: no relation with the underlying gel microstructure (we recall that there is almost no fractal regime in
640: the $87\%$ aerogel, as can be seen in Fig. 1).
641: \begin{figure}[hbt]
642: \includegraphics*[width=8cm]{fig11.ps}
643: \caption{Structure factor of the dense phase (see text), $S_{dd}(q)$, during desorption at $T^*=0.5$ and
644: $\mu=-4.65$. The dashed-dotted curve is the best fit according to Eq. 16 and the straight dashed
645: line has a slope $-2.1$.}
646: \end{figure}
647:
648: Raising the temperature to $T^*=0.8$ has a dramatic effect, as shown in Fig.12.
649: The maximum value of $S_{ff}(q)$ has dropped by two orders of magnitude and
650: there is no significant region with a fractal-like power-law behavior. Indeed, $h_{ff}(r)$ has no
651: more a long-range tail and the correlations are very
652: similar during adsorption and desorption, as could be expected from the very
653: thin shape of hysteresis loop.
654: This is the signature that the desorption mechanism has changed, in agreement with the
655: analysis of Refs.\cite{DKRT2003,DKRT2004}.
656: It is now due to a cavitation phenomenon in which gas bubbles first appear
657: in the largest cavities of the gel and then grow and coalesce until the whole
658: void space is invaded\cite{note13}.
659: \begin{figure}[hbt]
660: \includegraphics*[width=9cm]{fig12.eps}
661: \caption{ Fluid-fluid structure factor $S_{ff}(q)$ along the desorption isotherm
662: in a $87\%$ porosity aerogel at $T^*=0.8$. (Color on line).}
663: \end{figure}
664:
665: We have not studied the correlations during desorption in the $95\%$ porosity gel.
666: At very high temperature ($T^*\gtrsim 0.9$), desorption is expected to be due again
667: to cavitation\cite{DKRT2004}, and the results shoud be similar to those in the $87\%$
668: that have just been described. On the other hand, at low temperature (e.g. $T^*=0.5$),
669: the theory predicts a depinning transition in which a self-affine interface sweeps through the whole
670: sample, resulting in a discontinuous desorption isotherm\cite{DKRT2004}. Therefore, like
671: in the case of the macroscopic avalanche during adsorption, the correlations along the isotherm cannot
672: be studied within the framework of a grand-canonical calculation. At intermediate temperatures,
673: one could probably observe again extended fractal correlations associated to a percolating cluster of gas, but
674: this study requires the use of larger systems so as to probe smaller values of $q$ and discriminate
675: the effects due to the own fractal structure of the gel.
676:
677: \section{Scattered intensity and comparison with experiments}
678:
679: As mentionned in the introduction, there have been two recent scattering studies of gas condensation in aerogels, both with
680: $^4$He\cite{LMPMMC2000,LGPW2004}.
681: In Ref.\cite{LGPW2004}, light scattering is used to study adsorption and desorption in a $95\%$ porosity gel
682: at several temperatures between $4.47K$ ($T^*\approx0.86$) and $5.08K$ ($T^*\approx0.98$). These experimental results
683: cannot be directly compared to our theoretical predictions, and our system size is too small to investigate the large-scale
684: inhomogeneities that are seen in the experiments (some of them are visible to the eye). However, there are two key observations that
685: appear to be in agreement with our predictions:
686: i) at the lowest temperature studied, the optical signal due to helium adsorption is larger than if the fluid density was simply correlated to the density of silica, indicating that the correlations within the fluid extend beyond the aerogel correlation length, and ii) the aerogel is much brighter during desorption, indicating that the characteristic size of the density fluctuations is much larger than during adsorption.
687:
688: This latter conclusion was also reached from the small-angle x-ray scattering measurements (SAXS) performed in a
689: $98\%$ porosity aerogel at $3.5K$ ($T^*\approx0.67$)\cite{LMPMMC2000}. SAXS is particularly well suited for observing the structural features associated to fluid adsorption, and in order to compare more easily to experiments we shall push further our calculations and compute the resulting scattered intensity. Of course, the predictions must be taken with a grain of salt, considering the limitations in the model and the theory.
690: Since the scattered intensity is proportional to the Fourier transform of the electron density fluctuations, one has
691: \begin{align}
692: I(q)\propto \rho_g F(q)^2S_{gg}(q) + 2\sqrt{\rho_g\rho_f}\alpha F(q) S_{gf}(q) + \rho_f \alpha^2S_{ff}(q)
693: \end{align}
694: where $F(q)$ is the form factor of silica particles (see section IIIA) and $\alpha$ is the ratio of the electron density in the adsorbed fluid to that in the solid. As an additional approximation, we shall take $F(q)\approx 1$, restricting the study to the range $2\pi/L\leq q\leq 2$ where this is presumably a reasonable approximation (in real units this corresponds to $0.02\lesssim q \lesssim 0.7$ nm$^{-1}$, taking $a=3$nm). Assuming that the adsorbed liquid has the same density as the bulk liquid at $P_{sat}$, and using the tabulated densities of helium and silica, one finds that $\alpha$ varies from $6.55 \ 10^{-2}$ at $T^*=0.5$ to
695: $5.75 \ 10^{-2}$ at $T^*=0.8$.
696:
697: The theoretical scattered intensities during adsorption and desorption in the $87\%$ gel at $T^*=0.5$ are shown in Figs. 13 and 14. Like in the experiments\cite{LMPMMC2000}, we plot the ratio $R(q)=I(q)/I_e(q)$
698: where $I_e(q)$ is the contribution of the empty aerogel (the first term in the right-hand side of Eq. 17) in order to accentuate the effects of the adsorbed gas, especially in the initial stage of adsorption. Moreover, we hope that this also partially corrects the small-$q$ defects due to the absence of large-scale fluctuations in the DLCA gel structure factor.
699: \begin{figure}[hbt]
700: \includegraphics*[width=10cm]{fig13.eps}
701: \caption{ Theoretical ratio $R(q)$ of the scattered intensity $I(q)$ to the scattered intensity $I_e(q)$ of the empty aerogel during helium adsorption in a $87\%$ aerogel
702: at $T^*=0.5$. The dashed line coresponding to $R=1$ is shown for reference (Color on line).}
703: \end{figure}
704: \begin{figure}[hbt]
705: \includegraphics*[width=10cm]{fig14.eps}
706: \caption{ Same as Fig. 13 during desorption (Color on line).}
707: \end{figure}
708:
709: The main features of the curves displayed in Fig. 13 are the following: i) At the very beginning of adsorption, $R(q)$ slightly increases but
710: remains almost independent of $q$. This is the signature of the $^4$He film coating the aerogel. In this regime, the main contribution to the scattered intensity comes from the gel-fluid correlations (the second term in the right-hand side of Eq. 17). ii) As $\mu$ increases, the scattering grows in intensity at small $q$, reflecting the presence of the broad peak in $S_{ff}(q)$ that moves towards smaller wavevector with filling (see Fig. 5). iii) As the aerogel fills further, $R(q)$ decreases until it becomes again almost flat at complete filling. The total intensity is then reduced with respect to that of the empty aerogel.
711: Direct comparison with the experimental results of Ref.\cite{LMPMMC2000} is again problematic: the adsorption isotherm in the
712: $98\%$ gel is indeed very steep at $3.5K$, suggesting that one may be in the regime of a macroscopic avalanche. However, the behavior of the experimental $R(q)$ is remarkably similar to what have just been described (preliminary measurements in a $86\%$ aerogel\cite{M2005}
713: also show the same trends). The results of Ref.\cite{LMPMMC2000} were interpreted according to a model of two-phase coexistence, with a
714: `film' phase in equilibrium with a filled `pore' phase. This is at odds with the theoretical scenario discussed in Refs.\cite{DKRT2003,DKRT2004,DKRT2005}
715: which emphasizes the nonequilibrium character of the transition. The present results seem to show that this approach can also
716: elucidate (at least qualitatively) the behavior of the scattered intensity.
717:
718: During desorption, the most characteristic feature of the curves shown in Fig. 14 is the very significant increase of the ratio $R(q)$ at small $q$ with respect to adsorption (note the logarithmic scale on the vertical axis). This is related to the corresponding increase in $S_{ff}(q)$ shown in Fig. 9 and is clearly due to the presence of long-range correlations within the fluid. As the desorption proceeds, $R(q)$ goes through a maximum and then decreases until it becomes flat again. Remarkably, no power-law fractal regime is visible in $R(q)$ in the range $0.06\lesssim q\lesssim 1$ as was the case with $S_{dd}(q)$ in Fig. 11. It is the small value of $\alpha$ (due to the small electron density of He), and not the division by $I_e(q)$, which is responsible for this unfortunate disappearance ($I(q)$ becomes proportional to $S_{dd}(q)$ when $\alpha=1$, which is only the case in a contrast matching experiment). In the measurements of Ref.\cite{LMPMMC2000}, this increase of $R(q)$ at small $q$ is not mentionned, but the analysis of the data shows that the charateristic size of the inhomogeneities is much larger than during adsorption, as already mentionned, and that it decreases rapidly in the last stage of desorption.
719: \begin{figure}[hbt]
720: \includegraphics*[width=10cm]{fig15.eps}
721: \caption{ Same as Fig. 13 during desorption at $T^*=0.8$ (Color on line).}
722: \end{figure}
723:
724: Not surprisingly, the theoretical scattered intensity in the $87\%$ aerogel is considerably smaller at high temperature, as illustrated in Fig. 15
725: for $T^*=0.8$. The intensity ratio $R(q)$ has been divided by about $40$.
726: We therefore conclude that the magnitude of the scattered intensity can indicate that
727: the nature of the desorption process has changed.
728: We leave to our colleague experimentalists the challenge of checking the presence of a fractal-regime during desorption, as was done in Vycor\cite{LRHHI1994,PL1995,KKSMSK2000} and xerogel\cite{H2002}.
729:
730:
731:
732:
733: \acknowledgments
734: We are grateful to N. Mulders for very useful discussions and communication of unpublished results.
735: The Laboratoire de Physique Th\'eorique de la Mati\`ere Condens\'ee is the UMR 7600 of
736: the CNRS.
737:
738: \begin{thebibliography}{10}
739:
740: \bibitem{DKRT2003} F. Detcheverry, E. Kierlik, M. L. Rosinberg, and G. Tarjus, Phys. Rev. E {\bf 68}, 61504 (2003).
741: \bibitem{DKRT2004} F. Detcheverry, E. Kierlik, M. L. Rosinberg, and G. Tarjus, Langmuir {\bf 20}, 8006 (2004).
742: \bibitem{DKRT2005} F. Detcheverry, E. Kierlik, M. L. Rosinberg, and G. Tarjus, arxiv
743: cond-mat/0508267 (to appear in Phys. Rev. E).
744: \bibitem{GS1982} S. J. Gregg and K. S. W. Sing, {\it Adsorption, Surface Area, and
745: Porosity}, Academic Press, New York (1982).
746: \bibitem{C1996} M. H. W. Chan, Czech. J. Phys. {\bf S6}, 2915 (1996).
747: \bibitem{TYC1999} D. J. Tulimieri, J. Yoon, and M. H. W. Chan, Phys. Rev. Lett. {\bf 82}, 121 (1999).
748: \bibitem{H2004} For a recent review, see e.g. E. Hoinkis, Part. Part. Syst. Charact. {\bf 21}, 80 (2004).
749: \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).
750: \bibitem{LGPW2004} T. Lambert, C. Gabay, L. Puech, and P.E. Wolf, J. Low Temp. Phys. {\bf 134}, 293 (2004); T. Lambert, Ph.D. thesis, Universit\'e Joseph Fourier, Grenoble I (2004).
751: \bibitem{LRHHI1994} J. C. Li, D. K. Ross, L. D. Howe, K. L. Stefanopoulos, J. P. A. Fairclough, R. Heenan, and
752: K. Ibel, Phys. Rev. B {\bf 49}, 5911 (1994).
753: \bibitem{PL1995} J. H. Page, J. Liu, B. Abeles,
754: E. Herbolzheimer, H. W. Deckman, and D. A. Weitz, Phys. Rev. E {\bf 52}, 2763 (1995).
755: \bibitem{KKSMSK2000} A. Ch. Mitropoulos, J. M. Haynes, R. M. Richardson, and N. K. Kanellopoulos,
756: Phys. Rev. B {\bf 52}, 10035 (1995); E. S. Kikkinides, M. E. Kainourgiadis, K. L.
757: Stefanopoulos, A. Ch. Mitropoulos, A. K. Stubos, and N. K. Kanellopoulos,
758: J. Chem. Phys. {\bf 112}, 9881 (2000).
759: \bibitem{SG2004} R. Salazar and L. D. Gelb, Mol. Phys. {\bf 102}, 1015 (2004).
760: \bibitem{H2002} E. Hoinkis and B. R\"ohl-Kuhn, in {\it Fundamental of Adsorption 7} (Eds. K. Kaneko, H. Kanoh, and
761: Y. Hanzawa), IK international, Japan (2002).
762: \bibitem{M1983} P. Meakin, Phys. Rev. Lett. {\bf 51}, 1119
763: (1983); M. Kolb, R. Botet, and R. Jullien, Phys. Rev. Lett. {\bf 51},
764: 1123 (1983).
765: \bibitem{KMRST2001} E. Kierlik, P. A. Monson, M. L. Rosinberg, L. Sarkisov, and
766: G. Tarjus, Phys. Rev. Lett. {\bf 87}, 055701 (2001); E. Kierlik, P. A. Monson,
767: M. L. Rosinberg, and G. Tarjus, J. Phys.: Condens. Matter. {\bf 14}, 9295 (2002);
768: M.L. Rosinberg, E. Kierlik, and G. Tarjus, Europhys. Lett. {\bf 62}, 377 (2003); F. Detcheverry, E. Kierlik, M. L. Rosinberg, and G. Tarjus, Adsorption {\bf 11}, 115 (2005).
769: \bibitem{SM2001} H. J. Woo, L. Sarkisov, and P. Monson, Langmuir {\bf 17}, 7472 (2001); L. Sarkisov and P. Monson, Phys. Rev. E {\bf 65}, 011202 (2001); H. J. Woo and P. Monson, Phys. Rev. E {\bf 67}, 041207 (2003).
770: \bibitem{H1994} A. Hasmy, E. Anglaret, M. Foret, J. Pelous, and R. Jullien,
771: Phys. Rev. B {\bf 50}, 6006 (1994).
772: \bibitem{MLW1996} A. Meroni, D. Levesque, and J. J. Weis, J. Chem. Phys. {\bf 105}, 1101 (1996).
773: \bibitem{GS1992} J. A. Given and G. Stell, J. Chem. Phys. {\bf 97}, 4573 (1992).
774: \bibitem{RTS1994} M. L. Rosinberg, G. Tarjus, and G. Stell, J. Chem. Phys. {\bf 100}, 5172 (1994).
775: \bibitem{note2} From Eq. (2) (with $\mu$ replaced by $\mu_i$), one obtains the equation
776: ${\cal G}_{ff}({\bf r}_{ij})=
777: \rho_i (\eta_i-\rho_i)/\eta_i [\delta_{{\bf r}_{ij},{\bf 0}}+\beta\sum_{k/i}{\cal G}_{ff}({\bf r}_{kj})]$ where
778: ${\cal G}_{ff}({\bf r}_{ij})\equiv \partial^2 \beta \Omega^{eq}
779: /(\partial \beta \mu_i \partial \beta \mu_j)=\rho_f^2 h_{ff,c}({\bf r}_{ij})+\rho_f\delta_{{\bf r}_{ij},{\bf 0}}$, and the sum runs over the nearest neighbors of site $i$.
780: This equation must be solved by iteration for every ${\bf r}_{ij}$, a very demanding
781: numerical task in general (see, e.g., in a different context, D. Lancaster, E. Marinari, and G. Parisi, J. Phys. A: Math. and Gen. {\bf 28}, 3959 (1995)).
782: \bibitem{note1} Note that this issue is not discussed in Ref.\cite{SG2004}.
783: \bibitem{note4} Strictly speaking, this is only true at equilibrium, where Eq. 10 holds (see the equation just above in \cite{note2}).
784: In particular, the fluid isothermal compressibility $\chi$ given by $k_BT\rho_f \chi=1+\rho_f\sum_{{\bf r}}h_{ff,c}({\bf r})$\cite{RTS1994} is very small at low temperature. It would be worth checking by Monte Carlo simulations that $h_{ff,d}({\bf r})$ is the most important contribution to $h_{ff}({\bf r})$ along the sorption isotherms too. Note that in random-field-like systems these disorder-induced fluctuations are expected to be also predominant near criticality, with $h_{ff,d}({\bf r})$ diverging more strongly than $h_{ff,c}({\bf r})$ (see, e.g., E. Pitard,
785: M. L. Rosinberg, G. Stell, and G. Tarjus, Phys. Rev. Lett. {\bf 74}, 4361 (1995)).
786: \bibitem{note5} On the other hand, a box of size $L=200$ has been used to simulate
787: the aerogels with porosity $\phi\geq0.97$.
788: \bibitem{P1982} G. Porod, {\it Small Angle X-Ray Scattering}, Academic Press, London (1982).
789: \bibitem{OJ1995} N. Olivi-Tran and R. Jullien, Phys. Rev. B {\bf 52}, 258 (1995).
790: \bibitem{KKRT2001} V. Krakoviack, E. Kierlik, M. L. Rosinberg, and G. Tarjus, J. Chem. Phys. {\bf 115}, 11289 (2001).
791: \bibitem{note6} According to Babinet principle, two `complementary' objects have
792: the same diffraction properties. When the gel is completely filled with a liquid
793: whose density is approximately uniform and close to $1$ (which is the case
794: at low temperature), Babinet
795: principle (in real space) implies that
796: $h_{ff}(r)\approx (1-\phi)^2/\phi^2 \ h_{gg}(r)$ and $h_{gf}(r)\approx
797: -(1-\phi)/\phi \ h_{gg}(r)$. This is not quite visible in Figs. 3(b) and 4(b) because
798: of the drastic change in the vertical scales.
799: \bibitem{note9} The size of a so-called 'void' or `cavity' in the lattice aerogel may be estimated from the radius of the largest sphere that can be inserted
800: in the cavity without overlapping with a gel site\cite{SG2004} (see also L. D. Gelb and K. E. Gubbins, Langmuir {\bf 15}, 305 (1999)). Alternatively, one can compute an integer `distance' $n$ associated to the length of the shortest path from an empty site to a gel site\cite{DKRT2003}. The two procedures gives similar results. For a $87\%$ aerogel, the size of the largest cavity is $n\approx 5$ which is comparable to the correlation length $\xi_g$.
801: \bibitem{note7} We recall that $S_{ff}(q)$ has been computed so as to satisfy Eq. 11,
802: the correct $q=0$ limit of $h_{ff,d}(q)$. Then $S_{ff}(q)$
803: is expected to represent more faithfully the structure factor of the infinite
804: system. In fact, setting $h_{ff}=0$ or $\neq 0$ in Eq. 12 only changes $S_{ff}(q)$
805: for $q\lesssim2\pi/L$ and the differences are small, as illustrated by the
806: dotted curve in Fig. 5. The same is true for $S_{gf}(q)$.
807: \bibitem{note8} One could imagine to study separately the samples in which the jump has not yet occured and those which are already in
808: the high-density phase, and then to use a sort of `lever' rule to compute the structural properties at intermediate densities. However, this would be an incorrect procedure: the two fluid configurations before and after the avalanche are not in equilibrium (they are both metastable and the grand-potential has decreased during the avalanche) and one cannot apply to this problem the standard methods of phase equilibria.
809: \bibitem{WC1990} A. P. Y. Wong and M. H. W. Chan, Phys. Rev. Lett. {\bf 65}, 2567 (1990).
810: \bibitem{DS1996} K. Dahmen and J. P. Sethna, Phys. Rev. B {\bf 53}, 14872 (1996); J. P. Sethna, K. Dahmen, and O. Perkovi\'c in {\it The Science of Hysteresis}, edited by G. Bertotti and I. Mayergoyz, Elsevier (2004).
811: \bibitem{note3} This important feature has not been observed in the study of Ref.\cite{SG2004}. This is perhaps due to the fact
812: that the system size is too small: a cubic lattice of size $L=256$ is used, but since the gel particle radius is at least equal to $2$ lattice spacings,
813: the ratio $L/d$ is smaller than in our simulations.
814: \bibitem{note10} This can be understood by noting that Eq. 10 can be
815: rewritten as $\rho_f^2 h_{ff,d}({\bf r})=(1/N)\sum_{i,j}
816: [\overline{\delta\rho_i\delta\rho_j}
817: +\overline{\rho_f^2(\{\eta_i\})}-\rho_f^2]\
818: \delta_{{\bf r},{\bf r}_{ij}}$ where $\delta\rho_i=\rho_i-\rho_f(\{\eta_i\})$
819: represents the deviation of $\rho_i$ from the average density in the sample. The term $\overline{\rho_f^2(\{\eta_i\})}-\rho_f^2$ is
820: of order $1/N$ but it may be not negligible in a finite system when
821: the number of fluid particles fluctuates significantly from one sample to another.
822: It may then dominate the long-range behavior of $h_{ff,d}({\bf r})$.
823: This is precisely what happens in the steepest
824: portion of the desorption isotherm where
825: some samples are still filled with liquid whereas other ones are already significantly
826: drained.
827: \bibitem{note11} The structure factor $S_{dd}(q)$ shown in Fig.11 has been computed from
828: the Fourier transform of the `dense-dense' correlation function $h_{dd}({\bf r})$,
829: using a fit of the form $h_{dd}({\bf r})\sim \exp (-r/l)/r^{3-d_f}$ to extrapolate the
830: function to $r>L/2$. This procedure is used to avoid the small oscillations that
831: come from $S_{ff}(q)$ and $S_{gf}(q)$ when $S_{dd}(q)$ is directly computed from Eq. 15.
832: \bibitem{FKS1986} T. Freltoft, J. K. Kjems, and S. K. Sinha, Phys. Rev. B {\bf 33}, 269 (1986).
833: \bibitem{note13} Accordingly, the outer surface of the material does not play a role
834: anymore\cite{DKRT2003,DKRT2004} and one can study the correlation functions along the
835: desorption isotherm without introducing
836: an explicit external reservoir or using the last remaining gas bubbles. One can also safely extrapolate $S_{ff}(q)$ towards $q=0$ since there are no long-range correlations.
837: \bibitem{M2005} N. Mulders, private communication
838:
839:
840: \end{thebibliography}
841:
842:
843: \end{document}
844:
845: \bibitem{GF} A. Guinier and G. Fournet Small-Angle Scattering of
846: X-Rays Chapman and Hall, Ltd., London, 1955; A. Guinier, G. Fournet: Small-Angle Scattering of X-Rays, J. Willey, NY 1955; Small-Angle X-Ray Scattering, O. Glatter and O. Kratky, Eds., Academic Press, London 1982