cond-mat0101093/prl.tex
1: \documentstyle[prl,amsfonts,epsfig,twocolumn,aps]{revtex}
2: 
3: \begin{document}
4: \wideabs{
5: 
6: \title{Capillary condensation in disordered porous materials:
7: hysteresis versus equilibrium behavior.}
8: 
9: 
10: \author{E. Kierlik\( ^{1} \), P. Monson\( ^{2} \), M. L. Rosinberg\( ^{1} \), L. Sarkisov\( ^{2} \), and
11: G. Tarjus\( ^{1} \)}
12: 
13: \address{ $^{1}$Laboratoire de Physique Th{\'e}orique des Liquides\cite{AAAuth}, Universit{\'e}
14:  Pierre et Marie Curie, 4 Place Jussieu,\\ 75252 Paris Cedex 05, France \\
15: $ ^{2}$Department of Chemical Engineering, University of Massachussetts, Amherst,
16: MA 01003, USA}
17: 
18: 
19: \maketitle
20: 
21: \begin{abstract}
22: We study the interplay between  hysteresis and equilibrium behavior in
23: capillary condensation of  fluids  in mesoporous disordered  materials
24: via a mean-field density functional theory of a disordered lattice-gas
25: model.   The  approach   reproduces  all   major  features    observed
26: experimentally.  We show that  the simple  van   der Waals picture  of
27: metastability fails  due to the appearance   of a complex free-energy
28: landscape with  a  large number  of metastable states.  In particular,
29: hysteresis can occur  both with and  without an underlying equilibrium
30: transition, thermodynamic  consistency  is  not  satisfied  along  the
31: hysteresis   loop,  and  out-of-equilibrium   phase  transitions   are
32: possible.
33: \end{abstract}
34: \pacs{Pacs numbers: 05.50.+q,75.10.Nr,64.60.-i}}
35: 
36: 
37: 
38: Capillary condensation of  a gas inside a mesoporous   material
39: refers to  the rapid change  to a liquid-like state  that  occurs at a
40: pressure (or chemical potential) lower  than the bulk saturation value
41: \cite{1}\cite{2}. This phenomenon  is   often thought of as  a
42: shifted gas-liquid transition.  Theoretical studies of fluids confined
43: in single pores  of  ideal geometry  have clarified the  mechanism for
44: such a shifted transition   and  introduced the concept  of  capillary
45: criticality  that describes the  fact  that the liquid-vapor  critical
46: point    in a  pore   occurs  at  a  lower   temperature  than  in the
47: bulk\cite{2}\cite{3}.   However,  the situation   in  real  mesoporous
48: materials, such as porous glasses and silica  gels, that consist of an
49: interconnected network of  winding  pores of varying shape,  curvature
50: and  size is not as clear.   With the possible  exception of fluids in
51: aerogels of  very high porosity  ($95-98 \%$)  for  which a {\it  bona
52: fide}  liquid-gas phase transition has been   reported with no sign of
53: hysteresis\cite{4}, there is no direct evidence of a first-order phase
54: transition characterized,  say, by a   jump in the adsorbed  amount of
55: fluid, nor of  true criticality signaled   by density fluctuations  on
56: very large length scales.  The typical experimental observation is the
57: presence of  a hysteresis loop in  the sorption isotherms, a loop that
58: is reproducible and vanishes at a  temperature lower than the critical
59: temperature  of  the  bulk   fluid\cite{1}\cite{2}\cite{5}.   The main
60: questions  raised by  this phenomenology are:  is there  a true  phase
61: transition associated   with   capillary condensation   in  disordered
62: materials?  What  is the connection  between  this  transition and the
63: observed hysteresis?  What are the sources of hysteretic behavior?  No
64: satisfactory  answers  have been  provided  so far.   Indeed, in these
65: systems    that     are  locally   heterogeneous,  but macroscopically
66: (statistically) homogeneous,  the statistical-mechanical methods  that
67: have proved very  efficient in dealing with bulk  fluids and fluids in
68: simple external fields, namely computer simulations, integral-equation
69: theory   and   density  functional    theory,  are plagued     by many
70: difficulties.
71: 
72: In this letter, we present a mean-field density functional theory of a
73: disordered lattice-gas  model.   This model, introduced  in a previous
74: work\cite{6},     incorporates    the    main  physical    ingredients
75: characterizing fluids   in disordered mesoporous   media: preferential
76: adsorption of one phase of the fluid,  connectivity of the pore space,
77: geometric and energetic disorder, exclusion effect  due to the matrix.
78: (Note that the model could  also be used  to describe the demixing  of
79: binary   mixtures in   mesoporous  materials\cite{2}.)  The    present
80: approach allows us to study  for the first  time the interplay between
81: out-of-equilibrium  (hysteresis) and equilibrium behavior   associated
82: with capillary condensation in a disordered  matrix.  We show that the
83: theory  captures  the main aspects of   the phenomenology of capillary
84: condensation in real    systems.   The combination   of  disorder  and
85: interconnectivity of the pore network  drastically alters the  picture
86: of    capillary   condensation   built    upon  the   independent-pore
87: model\cite{1}\cite{2}\cite{7} or the simple  van der Waals description
88: of metastability because it  generates a complex free-energy landscape
89: characterized by a large number of  metastable states.  In particular,
90: we show that (i)  hysteresis can occur {\it with}  or {\it without} an
91: underlying equilibrium phase transition, (ii) the disappearance of the
92: hysteresis loop  is  not associated with  capillary criticality, (iii)
93: thermodynamic consistency is not  satisfied along the hysteresis loop,
94: and (iv) collective  out-of-equilibrium phenomena such as  macroscopic
95: ``avalanches'' are possible.
96: 
97: The model is described by the following Hamiltonian,
98: 
99: \begin{eqnarray}
100: \label{eq:1}
101: \cal{H}=&-& w_{ff}\sum _{<ij>}\tau _{i}\eta _{i}\tau _{j}\eta _{j}
102: -w_{mf}\sum _{<ij>}[ \tau _{i}\eta _{i}( 1-\eta _{j})\nonumber\\
103: &+&\tau _{j}\eta _{j}( 1-\eta _{i})] ,
104: \end{eqnarray}
105: where \( \tau _{i}=0,1 \) and \(  \left( 1-\eta _{i}\right) =0,1 \) denote
106: the  fluid and matrix occupancy  variables, respectively, and the sums
107: run  over distinct pairs of   nearest-neighbor (n.n.)  sites. Although
108: different   disordered   microstructures of     the matrix   could  be
109: considered\cite{8}, we choose  here the  simplest nontrivial  case, a
110: random matrix. The model   is thus specified   by two parameters,  the
111: average matrix density \( \rho _{m} \) that fixes the porosity (equal to
112: \(    1-\rho _{m}  \))  and   the ratio of    the matrix-fluid  over the
113: fluid-fluid  interactions, \( y=w_{mf}/w_{ff}  \), that determines the
114: ``wettability'' .     In the following  we   only  consider attractive
115: matrix-fluid interactions\cite{9} and we set
116: \( \rho _{m}=0.25 \).  We use a $3$D bcc lattice with linear size
117: $L=8,12,16,24,48$ for  which we consider many different configurations
118: of the random matrix, from $1400$ for $L=8$ to  $5$ for $L=48$.  Most
119: of the  results  are illustrated here for   $L=48$ and  typical matrix
120: configurations, but when necessary,  e.g.,  to check the existence  of
121: phase  transitions, we have    performed a finite-size  analysis after
122: averaging over all matrix configurations.
123: 
124: The  mean-field  density  functional  theory  (or  equivalently for  a
125: lattice, the local mean-field theory) starts with the formulation of a
126: free-energy functional of the fluid  density field for a given  matrix
127: realization \(  \left\{  \eta _{i}\right\}   \):
128: 
129: \begin{eqnarray}
130: &F&(\{\rho_i\})=\frac{1}{\beta}                                \sum_i[\rho_i\ln
131: \rho_i+(\eta_i-\rho_i)\ln(\eta_i-\rho_i)]\nonumber\\  &-&w_{ff}\sum_{<ij>}\rho_i\rho_j
132: - w_{mf}\sum_{<ij>}[\rho_i(1-\eta_j)+\rho_j(1-\eta_i)] \ ,
133: \end{eqnarray}
134: where $\beta=1/(k_BT)$ and $\rho   _{i}=<\eta  _{i}\tau _{i}>$ is   the average
135: fluid density at site \emph{i}. (The overall fluid  density is then \(
136: \rho _{f}=\left(  1/N\right) \sum _{i}\rho _{i}  \), where $N=2 L^3$  is the
137: total number of sites.) For  a given   chemical
138: potential
139: \( \mu  \), minimization of $\Omega(\{\rho_i\})=F(\{\rho_i\})-\mu \sum_i \rho_i$ w. r.  t.
140: the \(  \rho  _{i}  \)'s  provides  the  grand
141: potential $\Omega$ of the adsorbed fluid.  The  corresponding equations for the
142: fluid density on each site are
143: 
144: \begin{equation}
145: \label{eq:3}
146: \rho _{i}=\frac{\eta _{i}}{1+\exp \left[-\beta\left[ \mu +\sum _{j/i}\left( w_{ff}\rho _{j}+w_{mf}\left( 1-\eta _{j}\right) \right) \right]\right] },
147: \end{equation}
148: where the sum  is over all n.n.  of  site \emph{i}.  The  above set of
149: nonlinear coupled   equations  has been solved  by  means  of a simple
150: iteration algorithm according to two different protocols: (1) to mimic
151: the experimental procedure, we follow continuously the solutions under
152: small variations of the chemical potential \( \mu \) (typically, $\Delta \mu
153: =10^{-3} $)  and (2)  to  search  more  exhaustively the  solutions of
154: Eq. (3) for a given \( \mu \), we repeat the iteration procedure with a
155: large number of initial conditions (typically, $10^2$) corresponding to
156: uniform  fillings of the lattice  at different overall fluid densities
157: (i.e., $\rho_{i}^{(0)}=\rho^{(0)}$).
158: 
159: We  show   in  Fig.  1    typical  isotherms obtained    by increasing
160: continuously  \(  \mu  \)  from  \(   -\infty \) (adsorption   branch)  and
161: decreasing continuously \(  \mu     \) from  \( +\infty     \)  (desorption
162: branch). They look qualitatively similar to the experimental isotherms
163: of  fluids     adsorbed    in    porous    glasses     and      silica
164: gels\cite{1}\cite{2}\cite{5},  with an   asymmetric  hysteresis   loop
165: (corresponding  to   the   H2  shape  in  the  IUPAC   classification)
166: characterized  by a steep  desortion  branch
167:  
168: \begin{figure}
169: \begin{center}
170: \epsfxsize= 200pt
171: \epsffile{fig1.ps}
172: \vspace{.4cm}
173: \caption{Theoretical sorption isotherms  for  $y=1.5  $  at  $T=0.6,0.8, 0.95, 1.1$   and
174: $1.25$. (The mean-field bulk critical point is at $T=2$.) }
175: \end{center}
176: \end{figure}
177: 
178: and  a    smoothly increasing    adsorption  branch.   The   so-called
179: ``hysteresis phase diagrams'' that are constructed  from the points at
180: which the loop  opens and closes are  shifted to lower $T$ and  higher
181: $\rho_f$   compared to  the   bulk-fluid coexistence  curve,   and their
182: (asymmetric) shape is  also quite comparable  to what is obtained from
183: experimental data\cite{1}\cite{2}\cite{5}.
184: 
185: An  important test   for  the  relevance of  any   model  of capillary
186: condensation in  disordered  porous  materials is  the  shape of   the
187: so-called ``scanning curves'', i.    e., curves that are obtained   by
188: reversing  the sign of  the evolution of  \( \mu \) at different points
189: along the adsorption or  the desorption branches\cite{7}. It  has been
190: stressed that  these curves are not  properly reproduced by the widely
191: used  independent-pore model\cite{1}\cite{2}\cite{7}.  An illustration
192: of the scanning curves  on  adsorption and desorption  obtained within
193: the present theory is shown in Fig. 2. They look strikingly similar to
194: those observed experimentally, with an  upward curvature on adsorption
195: and  a downward curvature on   desorption.  When varying the  chemical
196: potential up and down  along different  paths we  find a  hierarchy of
197: inner scanning curves that display the ``return-point memory'' effect,
198: in close analogy to the hysteretic behavior displayed  by a variety of
199: systems\cite{10}\cite{11}\cite{12}.
200: 
201: The  above  results  confirm   that  the present approach   reproduces
202: qualitatively the phenomenology associated with capillary condensation
203: in porous glasses and silica   gels. This indicates that, despite  its
204: simplicity, the  model does   capture  the main physical  features  of
205: fluids adsorbed in real  mesoporous systems and that equilibration via
206: thermally   activated processes, processes     that are absent  in the
207: mean-field description, does not substantially modify the picture.
208: Indeed, the  implicit assumption behind
209: \begin{figure}
210: \begin{center}
211: \epsfxsize= 200pt
212: \epsffile{fig2.ps}
213: \vspace{0.4cm}
214: \caption{Theoretical desorption (a) and adsorption (b) scanning curves
215: for \( y=1.5 \) and \( T=0.8 \).}
216: \end{center}
217: \end{figure}
218: the direct comparison
219: of the theoretical sorption curves with the  experimental ones is that
220: the system does not have time to equilibrate at constant
221: \( T \)  and \(  \mu \)  and  that adsorption/desorption only  proceeds
222: under  the influence of   chemical-potential changes. (This assumption
223: may break down when the perturbation induced  by the matrix is small as
224: in very dilute aerogels.)
225: 
226: Hysteresis    and   scanning     curves    are   manifestations     of
227: out-of-equilibrium behavior.  This is well  understood for instance in
228: systems like  athermal  martensites   and low-\(  T   \) ferromagnetic
229: materials whose transformations are driven by an external field; these
230: systems are well described  by  the \( T=0  \)  limit of models   with
231: quenched disorder\cite{10} (models somewhat related to the lattice-gas
232: studied  here).   On   the other  hand,  in  the   case  of  capillary
233: condensation  one often assumes that hysteresis  is also the signature
234: of a  true equilibrium phase transition  with two  possible metastable
235: states, gas and liquid, as in the standard van der Waals loop for bulk
236: fluids and fluids confined in a single pore.  However, this simple van
237: der Waals picture fails for the present system  because this latter has (in the
238: region where   capillary condensation   occurs) a complex  free-energy
239: landscape with a large  number of  metastable states,  as can  be seen
240: from  studying  the solutions of the   local mean-field equations, Eq.
241: (3).    For  each  $T, \mu$   and each  matrix   realization,   we have
242: investigated $50$ to  $100$ different initial  conditions to Eq.  (3).
243: When plotted  as in Fig.  3 on  the \(  \rho _{f}-\mu  \)  diagram (for a
244: typical matrix configuration of the $L=48$ system), all solutions fall
245: inside the major hysteresis  loop that is  found to coincide with  the
246: curves of extremal solutions obtained  from an initially empty lattice
247: (lower branch) and an initially filled lattice (upper branch). We suspect
248:  that this picture corresponds to the situation encountered 
249: in real  disordered porous materials (a similar behavior
250: 
251: 
252: 
253: \begin{figure}
254: \begin{center}
255: \epsfxsize= 200pt
256: \epsffile{fig3.ps}
257: \vspace{0.4cm}
258: \caption{Multiplicity of local mean-field solutions obtained from solving
259: Eq. (3) with many different initial conditions for each value of $\mu$:
260: (a) \( y=1.5 \) and \( T=0.8
261: \), (b) \( y=1  \)  and  \( T= 1\).    The solid lines represent   the
262: equilibrium curves obtained     by connecting the  states  of   lowest
263: grand potential.}
264: \end{center}
265: \end{figure}
266: has been observed for other disordered microstructures\cite{8}).
267: 
268: The most  significant  finding   of   our study is   that   hysteresis
269: associated with  capillary  condensation  in  disordered  porous media
270: occurs with  as well   as without   an underlying   equilibrium  phase
271: transition, the existence of this latter  depending on the strength of
272: the perturbation   induced  by  the matrix.    We  have   computed the
273: equilibrium   properties  of the   model  by  means  of  the  weighted
274: mean-field  approach\cite{13}:  at each $\mu$    and $T$, densities  of
275: extensive quantities  such as the fluid  density $\rho_f$ are calculated
276: as  Boltzmann  weighted sums   over  the different   solutions of  the
277: mean-field equations (i.e., each solution $\alpha$  is weighted by $e^{-\beta
278: \Omega^{\alpha}}/\sum_{\alpha}  e^{-\beta  \Omega^{\alpha}}$).  The  average   over the various
279: matrix realizations  is  taken afterwards.   For a  large matrix-fluid
280: interaction,   e.g.,   for $y=1.5$, there     is  no equilibrium phase
281: transition: this is illustrated    in Fig.  3a where  the  equilibrium
282: isotherm at   $T=0.8$  is  perfectly smooth  (it   stays so   at lower
283: temperatures).  On the other hand, for $y=0.5$, the system undergoes a
284: true  thermodynamic transition at low  $T$ as it becomes isomorphic to
285: the site-diluted  Ising model\cite{9} ($\rho_m=0.25$ is above the percolation threshold).  For
286: the intermediate value $y=1$, equilibrium  isotherms may display steep
287: portions, as illustrated in Fig.  3b for $T=1$ (however, a finite-size
288: scaling analysis   of  the results  for  $L=8,12$  and  $16$  seems to
289: indicate that this  isotherm is  still supercritical).  The  capillary
290: critical point certainly occurs at  a  \( T\) significantly below  the
291: temperature  \( T_{hyst} \)  at which hysteresis  first appears.  This
292: result is at odds with the behavior predicted  for a fluid confined in
293: a single pore\cite{2}\cite{3}  and is a   consequence of the  quenched
294: disorder; on the other hand, the fact  that true capillary criticality
295: may still   occur in disordered   porous  media is  at odds  with  the
296: prediction based on the independent-pore model\cite{1}\cite{2}\cite{7}
297: and results from the connectivity of the void  space accessible to the
298: adsorbed fluid.
299: 
300: \begin{figure}
301: \begin{center}
302: \epsfxsize= 240pt
303: \epsffile{fig4.ps}
304: \vspace{0.4cm}
305: \caption{Check of thermodynamic consistency (\( \left( \partial \Omega /\partial \mu
306: \right) _{T}=-N\rho _{f} \))\textbf{} along the adsorption (a), 
307: desorption    (b),    and   equilibrium    (c)  isotherms    ($y=1.5$,
308: $T=0.8$).  Filled symbols:  average  fluid density  obtained from  the
309: mean-field  solutions.      Open    symbols:  quantity    obtained  by
310: differentiating the corresponding grand potentials.}
311: \end{center}
312: \end{figure}
313: 
314: The breakdown of the simple van  der Waals picture of metastability in
315: the present problem shows up distinctly when considering thermodynamic
316: consistency along the sorption  isotherms.  Since the theory  provides
317: both the  grand  potential and the fluid   density, one can  study the
318: validity of the Gibbs adsorption equation,
319: \( \left( \partial \Omega /\partial \mu \right) _{T}=-N\rho_{f} \), along particular
320: isotherms.  As shown in  Fig.  4 a,b, the   equation is not satisfied,
321: neither  along the adsorption isotherm  nor  along the desorption one.
322: This  results from the  fact that the system  often jumps from a given
323: grand potential minimum to another along these  two isotherms and then
324: looses thermodynamic  consistency.  An  important consequence  is that
325: the thermodynamic integration  procedure used for building equilibrium
326: phase  diagrams\cite{2}   is no  longer   valid for  this  model.  For
327: instance,  in the case  \( y=1.5 \),  this latter procedure predicts a
328: capillary phase diagram whereas,  as already discussed, no equilibrium
329: phase transition takes place.  We believe this  may also be  true more
330: generally  for     other models  of   fluids   in   disordered  porous
331: materials. Note that the Gibbs adsorption equation is obeyed along the
332: equilibrium isotherm,  as illustrated in  Fig.  4c. (Interestingly, in
333: the cases   studied here, the   isotherm calculated from  the weighted
334: mean-field approach is  virtually indistinguishable from that obtained
335: by taking  at   each  $\mu$ the  solution  with   the  lowest  grand
336: potential.)
337: 
338: As a final  remark, we mention that  the mean-field density functional
339: theory   also predicts  the   occurrence of  out-of-equilibrium  phase
340: transitions.   Similarly  to the macroscopic   avalanches observed  in
341: low-\( T    \) ferromagnetic   materials\cite{10}, the   adsorption and
342: desorption branches of  the  major hysteresis loop may   display sharp
343: jumps. This is   illustrated in Fig.  1 where   one can see that   the
344: desorption  isotherm has  a   discontinuity at the  lowest temperature
345: shown. Due to the asymmetry of  the random chemical potentials exerted
346: by the matrix on the fluid, the jump always occurs at a higher \( T \)
347: in the desorption branch than in the adsorption branch. Whether or not
348: this out-of-equilibrium  transition,   with  its associated   critical
349: behavior\cite{10}, can be observed  in capillary condensation of fluids
350: in   real  mesoporous  materials depends  on   the  efficiency of  the
351: thermally activated processes.
352: 
353: \begin{references}
354: \bibitem[*]{AAAuth} The Laboratoire de Physique Th{\'e}orique des Liquides is the UMR 7600 of the CNRS.
355: \bibitem{1}D. H. Everett, in \emph{The Solid Gas Interface},
356: E. A. Flood Ed. (Marcel Dekker, New York, 1967), Vol.  2, p.  1055;
357: S.  Gregg and K.  Sing, \emph{Adsorption,   Surface Area and   Porosity} (Academic,  New York,
358: 1982); C. Burgess \emph{et  al.},  Pure and Appl.  Chem.  \textbf{61},
359: 1845 (1989).
360: \bibitem{2}For a recent review, see L. Gelb \emph{et al.}, Rep. Prog. Phys. \textbf{62},
361: 1573 (1999).
362: \bibitem{3}R. Evans, J. Phys.: Condens. Matter \textbf{2}, 8989 (1990).
363: \bibitem{4}A. Wong and M. Chan, Phys. Rev. Lett. \textbf{65}, 2567
364: (1990); A. Wong {\it et al.}, Phys. Rev. Lett. \textbf{70}, 954 (1993).
365: \bibitem{5}W. Machin, Langmuir \textbf{10}, 1235 (1994); M. Thommes
366: and G.  Findenegg, Langmuir \textbf{10},  4270 (1994); S.  Gross \emph{et
367: al.}, preprint (2000).
368: \bibitem{6}E. Pitard \emph{et al.}, Phys. Rev. Lett. \textbf{74}, 4361 (1995); E. Kierlik
369: \emph{et al.}, Mol. Phys. \textbf{95}, 341 (1998).
370: \bibitem{7}P.C. Ball and R. Evans, Langmuir \textbf{5}, 714 (1989).
371: \bibitem{8}L. Sarkisov and P. Monson (in preparation).
372: \bibitem{9} The system has an overall symmetry w.r.t. \( y=1/2  \):
373: this case corresponds to the site-diluted Ising model\cite{6}.
374: \bibitem{10}J. P. Sethna \emph{et al.}, Phys. Rev.  Lett. \textbf{70},
375: 3347 (1993); O.  Dahmen and J. P. Sethna  , Phys. Rev.  \textbf{B 53},
376: 14872 (1996), and   references   therein;  E.   Vives   {\it  et  al.},
377: Phys. Rev. Lett.  \textbf{72}, 1694 (1994).
378: \bibitem{11}R. A. Guyer and K. R. McKall, Phys. Rev. \textbf{B 54}, 18
379: (1996)
380: \bibitem{12} M. P. Lilly
381: \emph{et al.}, Phys. Rev. Lett. \textbf{71}, 4186 (1993); M. P. Lilly \emph{et
382: al.}, Phys. Rev. Lett. \textbf{77}, 4222 (1996).
383: \bibitem{13} D. Lancaster {\it et al.}, J. Phys. A: Math. Gen., {\bf
384: 28}, 3959 (1995).
385: 
386: \end{references}
387: 
388: \end{document}
389: