PHYSlCA

ELSEVIER

Physica A 239 (1997) 467-476

Reentrant wetting transition of a rough wall

Gilberto Giugliarelli a, Attilio L. Stella b,*
Dipartimento di Fisiea and INFN (Gruppo Colleyato di Udine, Sezione di Trieste),
Universitd deyli Studi di Udine, 1-33100 Udine, Italy
b Istituto Nazionale di Fisica della Mater&, Dipartimento di Fisica e [stituto Nazionale di Fisica
Nuck, are (Sezione di Padova), Universitd di Padova, 1-35131 Padova, Italy

Received 2 October t996; revised 18 October 1996

Abstract

A 2D model describing depinning of an interface from a rough, self-affine substrate, is studied

by transfer matrix methods. The phase diagram is determined for several values of the roughness
exponent, ~s, of the attractive wall. For all ~s > 0 the following scenario is observed. In first
place, in contrast to the case of a flat wall (~s = 0), for wall attraction energies between zero
and a ~s-dependent positive value, the substrate is always wet. Furthermore, in a small range of
attraction energies, a dewetting transition first occurs as T increases, followed by a wetting one.
This unusual reentrance phenomenon seems to be a peculiar feature of self-affine roughness, and
does not occur, e.g., for periodically corrugated substrates.
P A CS: 68.45.Gd

1. I n t r o d u c t i o n

Wetting phenomena occur when, e.g., a layer o f liquid phase coexisting with its
vapor grows macroscopically over an attractive solid substrate [1]. While wetting in
pure systems is relatively well understood by now, the effects o f disorder in interfacial
phenomena [2-5] pose many challenging issues and are the object o f active research.
O f particular interest is the effect o f geometric surface disorder (roughness) on the location and the nature o f the wetting transition. A n extensively studied type o f roughness,
also in view o f its experimental realizability [6], is that occurring when the average
height fluctuation in a sample o f longitudinal linear size L, wL, scales like wc ~ L ;~'.
This self-affine scaling disorder is globally characterized by the roughness exponent

* Corresponding author. E-mail: stella@padova.infn.it.

0378-4371/97/$17.00 Copyright @ 1997 Elsevier Science B.V. All rights reserved

PII S03 78-4371 (97)0001 6-2

468

G. Giu#liarellL A.L. Stella/Physica A 239 (1997) 46~476

~s, which can be expected as the only relevant substrate parameter possibly affecting
universal features of the wetting transition.
The relevance of ~s for both complete and critical wetting transitions has been
extensively studied in the last years [7,8]. As far as critical wetting is concerned,
a recent study by the present authors [9] has put in evidence the fact that self-affine
roughness can produce a change from continuous to first-order transition in systems
with short-range forces, when the roughness of the substrate, ~s, exceeds the intrinsic
roughness of the interface in the bulk, if0. On the other hand, for ffs < ~0, no change is
expected in the nature and universality class of the continuous wetting transition, with
respect to that on fiat substrate.
The possibility that substrate roughness drives a continuous wetting transition firstorder has been subsequently discussed in the context of Landau-type mean field
approaches [10].
A possible modification of the nature of the wetting transition is not the only effect
of self-affine substrate roughness on wetting. As we show here, roughness produces
modifications of the phase diagram of the interface, which can be quite dramatic and
important in experiments and applications.
In the present paper we study systematically the wetting phase diagram of a generalization [9] of the Chui-Weeks model [1 1] with a rough attractive boundary in 2D.
By this study we produce evidence of some remarkable and definitely unusual features
of the phase diagram, which, at a qualitative level, should be considered as generic for
wetting on self-affine rough substrates. The most notable feature of the phase diagram is
a reentrance phenomenon for wetting in the whole range of roughnesses (0 < ~s < 1).
This reentrance, which implies a dewetting followed by a wetting transition as the
temperature is raised, occurs both in regimes when the transitions (both dewetting and
wetting) are critical, and when they are first-order.
This paper is organized as follows. In the next section we introduce the model and
describe the transfer matrix method we apply. In Section 3 we discuss how the phase
diagram is calculated and illustrate in detail the main results. The last section is devoted
to further general considerations and to conclusions.

2. Model and transfer matrix

We consider here the same generalization of the Chui-Weeks interfacial model discussed in Ref. [9]. Let us denote by x and y the integer coordinates of points on
a square lattice. The interface configurations are self-avoiding paths (partially) directed in the x direction (see Fig. 1). Each interface step parallel to the x axis is
located by giving the ordinate y = hx of its left-hand extremity. In this way the configuration is determined by giving hx, Vx. We impose the following extra restrictions on the set {hx}. First of all hx>~Sx where {Sx} represents the directed profile of the substrate wall. {S~} are sets of integer ordinates determining the wall
configuration with the same conventions applied to {hx}. An extra constraint on

G. Giugliarelli, A.L. Stella/Physica A 239 (1997) 467 476

i1

469

h~

i.....

l-I

IJU

LJ

Ll

Fig. I. Example of rough substrate wall (continuous path) and interface configuration (dotted path).

possible interface configurations is given by hx+l - h x = 0, 1, while for St we choose

S ~ + l - S~ = 1. We impose such constraints for computational convenience: their removal or modification would not change the main qualitative features of the phase
diagram.
The sets {St} are generated by a random sampling procedure inspired by
Mandelbrot [12] and described in the appendix. This procedure produces directed paths
in 2D obeying the restrictions described above and satisfying the scaling relation
I S x + A x - Sxl ~ IAxl ~ .

(l)

In the last equation the bar indicates statistical average with respect to the above
mentioned sample of {S~}. Averages over wall configurations, i.e. over different {S~},
will be considered quenched and denoted by overbars below.
The interface Hamiltonian is of the form
X

.*~ = ) - ~ B(I + qzx - Z x _ , +sxd) - u,t,.0]

(2)

with zx = h x - S x and Sx = S x - S x - l . In Eq. (2) e (e > 0) is the energy cost of any interface step and - u (u > 0) is the energy gain of an interface contact with the attracting
wall. The sum in (2) is performed up to X, which represents the length of the interface
projection on the x axis. It should be noted that in our model only the horizontal steps
of the interface paths in contact with the wall are prized by energies u. This choice
is again not mandatory. Different conventions would not change the basic qualitative
results.
At a finite temperature T the fugacities so = e ,:..'r and k = e " v are associated with
each (horizontal or vertical) step of the path, and to each horizontal step on the wall,
respectively. Consequently, given a wall profile, the partition function associated to all
the interface configurations covering a distance X can be written in the form
YX = Z exp
{:, t

= ~x Z ~ o , , k , , , '

{--,}

(l+]z~-zx_l+Sxl)-u

<~:,,o

x
(3)

G. Giugliarelli, A.L. Stella/Physica A 239 (1997) 467-476

470

where the sum is done over the ensemble of all the directed paths (determined here by
{Zx}) compatible with the chosen wall. n and nc are the number of vertical steps of
the interfacial path and the number of horizontal steps on the wall, respectively. Note
that the total length of a path is given by L = X + n.
The partition function (3) can be more conveniently expressed in the form
~ex _- ~ x Z

Ts,

l,z

qS0(z),

(4)

l,z

where the Ts~ are transfer matrices defined as

(Ts)m,n = [6m,n-s q- ~O(6m,n--s--t q- Jm, n-s+l )]k 6'

(5)

with m,n>~O. The function qS0 in Eq. (4) can be used to enforce particular initial
conditions for the interface paths; if we choose paths with an extremity on the wall,
we put ~b0(z)= 6z,0.
With the above definitions, a wall profile corresponds to a particular sequence of
factors Ts, in the product of Eq. (4). Correspondingly, for asymptotically large systems,
the partition function ~ex can be expressed in terms of the largest Lyapunov eigenvalue
~max a s ~ x <'w(O')2max)x, for X ---+c~, where [13]

ma:'im E

q'/x

/'OJ

with 114'11=
4'(z)
For walls with some periodic geometry with period Xp the matrix product in (6) can
be written as [3-xp]x/xp ( X assumed an integer multiple of Xp) with Yx~, = 1-Ix=l
x~, Ts,.
In this case the calculation of 2max is equivalent to the calculation of the largest
eigenvalue, Axp, of the matrix Yx~,. Consequently, 2max = [Axp] 1/X~'.
In numerical calculations it is useful to introduce the normalized vectors Cx defined
by the recursion relation
~x = 1 T s , ~

(7)

with nx = IITs, q~x_~ II and ~0 = 4'0. It is straightforward to see that the zth component
of the vector ~'x corresponds to the probability that the path at x is at a distance z
from the wall [2,8]. This suggests our definition of II~0xll, For a given wall profile, the
above definitions allow to express the Lyapunov eigenvalue (6) as

2m~ =

lim
x---+ec

n~

= exp

lim --1
lnn~
x---,ec X x=l

(8)

l Note that our definition of norm is legitimate because ~bo(z)and the elements of the Ts,'S matrices are ~>0.

G. Giugliarelli, A.L. Stella/Physica A 239 (1997) 46~-476

The quenched, dimensionless free energy density given by l i m x ~

written in the following form

471

-ln

dex/X can be

./7 = _ In CO2m,,x= - In oJ - xlim

~ X Z

In n~.

(9)

x--I

If X is chosen large enough, the average over quenched wall disorder for the second
term on the right hand side requires only a rather limited sample of wall configurations.
This is due to a self-averaging property of f , which clearly manifests itself in the
numerical results.

3. Depinning transition and phase diagram

In the discussion below we implicitly make u and T dimensionless by dividing them
by s. The wetting transition occurs because, e.g., at a given T, the interface can be
bound to the wall of the substrate only for sufficiently high values of u. In the case of
ordered flat walls, i.e. {Sx = constant Vx}, the value of u above which the interface is
pinned can be easily calculated [14,11,2] to be
uc(T) = Tln[(l +

2exp(-1/T))/(1 + e x p ( - 1 / T ) ) ]

(10)

with uc(0)= 0. We denote by P0 the average fraction of interface horizontal steps on

the wall,
P0 = lim

X--* w:

(no)IX,

(11)

with brackets indicating canonical thermal average. One can check that P0 vanishes
continuously and linearly when the line u = u c ( T ) is approached from above.
When dealing with our random substrates the calculation of Po or f for each {S~ }
cannot be done exactly in a semi-infinite geometry, and truncations must be made.
To minimize effects of the finite size of the transfer matrices, in our calculations we
only considered matrix sizes much larger than the mean square perpendicular width
of the self-affine walls. This width can be defined as AS = ( ST - ~2)j,2 ~ X ~ , , with
S 2 = ( l / X ) Y~x $2 and S = ( l / X ) Y'~-xS~. In practice we used transfer matrices as large
as 104 104 in the roughest case, corresponding to ~s = In 12/ln 32-~0.717. With this
roughness X = 105 was reached.
In addition one has to average over different {Sx} in order to get P00 and .f. In
practice we could sample at most 10 or 15 different {Sx} in the most favorable cases,
due to the large X ' s needed to extract precisely P0. Fortunately, as mentioned above,
our large X values, of the order of 10 5, lead to a very high degree of self-averaging
in quantities like f and P0.
At fixed T, as the transition is approached from above (viz. u > u~(T)) the interface
free energy density (9) is negative and increasing, with decreasing u; at u = u~.(T) it
matches the bulk interface free energy density fbulk = --In o9(1 +2o~). Thus, it is useful

G. Giuyliarelli, A.L. StellalPhysica A 239 (1997) 467-476

472

to consider the interface excess free energy density, A f = y

of Eq. (9), can be expressed as

fbulk, which, by means

A f = - In 2mux + In( 1 + 209) = -- In 2tmux .

(12)

This last equation defines 2mux as the Lyapunov eigenvalue of transfer matrices defined
as I~ = [1/(1 + 2og)]Ts, which were directly used in our calculations.
Thus the position of the depinning transition corresponds to the vanishing of A f .
/
Also the interface contact probability can be evaluated using )~max, taking into account
that

The calculation of P0 offers an alternative way of locating dewetting or wetting transitions, by identifying the conditions under which the quantity first becomes zero.
A numerical study of how P0 approaches zero can also provide informations on whether
these transitions are continuous or discontinuous [9]. We used this criterion, together
with an analysis of the way in which A f vanishes to clarify the first or second order
nature of the transitions.
Fig. 2 summarizes the results we have obtained by a systematic calculation of f as
function of T and u, for 5 different values of the ffs. The curves in the figure represent
the behavior of uc versus T. u,. was determined numerically as the value of u below
which I A f l ~<0.0001 as u was changed in steps of 0.001. By using the definition of
1.5

1.2

....
---------

~s=0.2
~s=0. 4
~s=0.517
~s=0.6
~s=0.717

,t
i, I
js

Uc

_.1

11

f~

oJ
11 . ' ' ~
i , ' / J
J" ..-//

O.75

101

/',,::,'7

1 1]
#S,"
-

0.5

0.25

0,5

1.5

T
Fig. 2. Interface phase diagram for rough self-affine walls in the T-u plane (e = 1 is assumed). The curves
u=uc(T) are shown for five different values of the roughness exponent (s. The light continuous line
corresponds to u = uc(T) for a flat wall as given by Eq. (10).

G. Giugliarelli, A.L. Stella/Physica A 239 (1997) 467 476

473

~o and k the u=uc(T) curves can be converted into the curves k=k~.(~o). Note that
u,::(0) # 0 implies a power-law divergence of kc(~o) for ~o--+ 0: k~ ~_ co -~,l).
Looking at the curves for rough walls in Fig. 2, we note two main differences from
the flat case:
(a) It is not possible to pin an interface to a rough wall with a vanishing contact
energy. As T - - , 0 the minimal contact energy to pin an interface, uc(0), is finite and
increases as the wall roughness, i.e. ~s, increases.
(b) For each ~s, there is a temperature, TR(~s), below which u~. is a decreasing
function o f T. Note also that for all ~s < 1 u~(T) approaches uc(0) with zero slope.
A surprising consequence o f (b) is that if, as in an experiment, we monitor interface
behavior at fixed u, by varying T, we have to distinguish among three kinds of regimes:
(I) For u < uc(Tn) interface pinning is impossible, no matter how low T is. The
substrate is wet at all T, and no transitions take place.
(II) for u > uc(0) as the temperature is increased the interface passes from a pinned
to a depinned state at some Tw. Thus, the substrate is partially wet for T < T~ and
wet for T > T~f,,, as for smooth walls.
(111) for uc(Tn) < u < Uc(0), as the temperature is increased, the interface undergoes
two transitions: (1) at some temperature TD < TR we find an unexpected dewetting:
the interface which is depinned for T < Tz~ becomes pinned at T = Try; (2) at some
T,, > TR a more usual depinning transition follows. Thus, the substrate is wet at very
low T, then it dewets, and eventually it wets again at high T.
Unfortunately, a detailed study o f the behavior of TR for ~s approaching zero is not
feasible due to the necessity of generating extremely long walls in order to distinguish,
e.g., ~s = 0.1 from ~s = 0. However, our results suggest rather clearly that TR approaches
zero for both ~s --~ 0 and ~s --~ 1. Thus, in these two limits the reentrance disappears.
Another interesting aspect o f the phase diagram is that connected to the nature
o f the transitions involved. The continuous or discontinuous character o f the wetting
transitions upon varying ~s was already discussed in Ref. [9] by analyzing the way
in which P0 approaches zero for u - ~ u,.(T). There we found that when ~s ~
< 1/9._ the
transition remains second-order and most likely belongs to the same universality class
as the flat case, i.e. F 0 0 ~ ( u - uc) 4', with i f = 1. When (s exceeds 1/2 there is clear
evidence that depinning occurs discontinuously. ~0 = 1/2 is the roughness exponent o f
the interface in the bulk and it makes sense that this is the precise border value of ~,s.
separating the two regimes.
A natural question is whether, in the range uc(Te)< u < u,.(0), the dewetting transition occurring at low temperature is of the same kind as its wetting counterpart at
higher T. Following the lines o f Ref. [91] we made a systematic study of the dewetting
transition for two ~s values, respectively below and above ~s = 1/2. In the first case
(~s = 0.4) we found evidence o f a continuous dewetting, while in the latter (~s = 0,6)
it appeared discontinuous. Thus, in spite of the fact that the dewetting transition occurs
only for some u ranges, and at much lower temperatures, it seems that its character
could be the same as that of the corresponding wetting transition. O f course, a more
precise definition o f the limits within which the above identity o f transition orders

474

G. GiugliarellL A.L. Stella/Physica A 239 (1997) 46~476

applies would require an extensive systematic exploration, which is beyond the scope
of the present work.

4. Conclusions
The results presented in the previous section are somewhat unexpected and are worth
discussing further.
The first important fact is that uc(O)> 0 for ~s > 0. At T = 0, in order to decide
whether the interface is pinned or not, we need only to compare the ground state
energy in the bulk with the lowest energy of a state in which the interface is bound
to the substrate. In the bulk the state of lowest possible energy is clearly given by
a straight configuration (n = 0 ) . A bound state will have an energy relative to this
unbound ground state equal to n -unc (assuming e, = 1). n and nc of course depend
on the wall configuration to which this bound state refers. Clearly uc(0) is determined
by the condition under which this energy difference between the two states vanishes:
Uc(0)= l i m x ~ n/nc. The fact that bound ground state configurations satisfy this
limit condition with u~(0)> 0 is a nontrivial property of self-affine substrates. On a
periodically corrugated substrate with average horizontal orientation (like that sketched
in Fig. 3) this limit property would not be satisfied. In that case, for u very close to
zero, the bound ground state configuration is one in which n = 0, corresponding to
a straight interface touching the attractive tips of the periodically corrugated wall (see
Fig. 3). Thus, we would have n = 0 and n c 0, and, consequently, uc(0)= 0, like in
a fiat case. We conclude that a remarkable property of self-affine substrates is that they
can support ground state interface configurations with l i m x ~ n/nc > O.
Notice further that for a periodically corrugated substrate the reentrance phenomenon
with dewetting preceeding wetting is not possible, since in that case the curve u = u,(T)
cannot be decreasing in the neighborhood of T = 0. We verified by explicit calculations
that for a wall as in Fig. 3, u~(T) is in fact never decreasing on the whole T axis.
An even more remarkable property of the self-affine substrate, for which uc(O)> 0
is clearly a necessary but not sufficient condition, is the monotonically decreasing
character of the curve u=u~(T) in the interval (0, TR). This feature implies that, as
soon as T > 0, an interface can be more easily bound to the rough substrate. This
clearly shows that there is a very nontrivial energy-entropy interplay in the pinning
mechanism when self-affine roughness is involved.

<

Xp

Fig. 3. Example of a periodically corrugated wall on a square lattice (heavy line). The wall period is Xp = 10.
The light straight horizontal line corresponds to the bound interface ground state at T = 0 and u > 0.

G. Giugliarelli, A.L. Stella/Physica A 230 (1997) 46~476

475

We also compared the nature of dewetting transitions to the corresponding highT wetting ones. We got only preliminary results on this issue. These results suggest
the possibility that, once a first-order character prevails for wetting, the same applies
to dewetting as well. This would mean that geometry alone is the crucial factor in
determining the nature of transitions on rough substrates.
The model calculations we presented here are of course limited to 2D and to strictly
short-range forces. An extension in 3D is computationally unfeasible, and also the inclusion of long-range potentials would pose serious additional difficulties in our calculations. In 3D we expect that the main features of the phase diagram would persist. Of
course a major difference in 3D would be the character of the transitions. In 3D the interface has roughness (0 = 0 in the bulk [2]. Thus, following the conclusions of Ref. [9],
we should probably expect first-order dewetting and wetting transitions for all ~s > 0.
Concerning the effect of long-range forces, which should certainly be included in
more realistic calculations to compare with experiments, we can only conjecture that
they would not modify the main result obtained here, i.e. the reentrance. However, there
is at least one instance, that of interfaces in superconductors [15], in which a short
range description is fully adequate.

Acknowledgements
This work was partially supported by CNR within the CRAY Project of Statistical Mechanics. We are very grateful to T.L. Einstein for a critical reading of the
manuscript.

Appendix. Generating random self-aitine paths in 2D

The procedure we describe here is a random version of a deterministic algorithm
by Mandelbrot [12]. Given two even integers, p and q, with p < q , the procedure
allows us to construct iteratively a partially directed path with a roughness exponent
.(s = In p/ln q and with X = q " after n iterations.
First we consider a set {~t} of vectors with q components. For each vector in the
set (q + p)/2 components, chosen at random, are set equal to 1, while the remaining
( q - p ) / 2 are put equal to - 1 . Once the set {e} contains a sufficiently large number
of such elements, the construction of a wall profile proceeds iteratively. In the first
iteration we consider a vector fll with q components and, after choosing an element ~t
at random in the set {~t}, we set f i t - ~t.
In the second iteration we construct a vector f12 with q2 components. Once chosen
q vectors ~/ ( j = 1,2 . . . . . q) at random in {~t}, the components of f12 are determined
according to the rule:

f12((_/- l)q + i ) = f l l ( j ) ~ / ( i )
with i - - 1,2 .... , q and ~j (i) indicating the ith component of the vector ~t/.

(A1)

G. Giugliarelli, A.L. Stella/Physica A 239 (1997) 46~476

476

The last equation can o b v i o u s l y be iterated for the construction o f vectors fin with

qn components:
f l n ( ( j - 1)qn-1 + i ) = f l n - l ( j ) c t j ( i )
with j =

1,2,...,qn-I

(A2)

and i = 1,2 . . . . . q.

A t any iteration n a directed path defined by the sequence o f integers Sx, as explained
in Section 2, can be obtained f r o m fin by

Sx+l = Sx + fin(x),

(A3)

w h e r e we usually set SI = 0. L a r g e r n implies a large horizontal size X = qn o f the

path. The a c c u r a c y o f the self-affine average scaling o f the paths, as detectable on
the basis o f Eq. (1), increases with increasing n and the n u m b e r o f walls considered.
For q ranging b e t w e e n 8 and 32 w e h a v e seen that w h e n the longitudinal length o f the
paths X is o f the order o f 105 the c o m p u t e d roughness exponents o f a single generated
path coincides with the theoretical ~s = In p / l n q within 1%.

References
[1] For a wide review on wetting phenomena see: S. Dietrich, in: Phase Transitions and Critical Phenomena,
eds. C. Domb and J.L. Lebowitz, Vol. 12 (Academic Press, London, 1988).
[2] G. Forgacs, R. Lipowsky and ThM. Nieuwenhuizen, in: Phase Transitions and Critical Phenomena, eds.
C. Domb and J.L. Lebowitz, Vol. 14 (Academic Press, London, 1991).
[3] D.A. Huse and C.L. Henley, Phys. Rev. Lett. 54 (1985) 2708.
[4] M. Kardar, Phys. Rev. Lett. 55 (1985) 2235; M. Kardar and D.R. Nelson, Phys. Rev. Lett. 56 (1986)
472.
[5] R. Lipowski and M.E. Fisher, Phys. Rev. Lett. 56 (1986) 472.
[6] P. Pfeifer, Y.J. Wu, M.W. Cole and J. Kfim, Phys. Rev. Lett. 62 (1989) 1997; see also J. Krim,
I. Heyvaert, C. Van Haesendonck and Y. Bruynseraede, Phys. Rev. Lett. 70 (1993) 57.
[7] H. Li and M. Kardar, Phys. Rev. B 42 (1990) 6546; M. Kardar and J.O. lndekeu, Europhys. Lett.
12 (1990) 61.
[8] G. Giugliarelli and A.L. Stella, Physica A 212 (1994) 12.
[9] G. Giugliarelli and A.L. Stella, Phys. Rev. E 53 (1996) 5035.
[10] A.O. Parry, P.S. Swain and J.A. Fox, J. Phys.: Condensed matter 8 (1996) L659.
[11] S.T. Chui and J.D. Weeks, Phys. Rev. B 23 (1981) 2438.
[12] B.B. Mandelbrot, Phys. Scripta 32 (1985) 257.
[13] A. Crisanti, G. Paladin and A. Vulpiani, in: Products of Random Matrices in Statistical Physics,
ed. H.K. Lotsch, Springer Series in Solid State Sciences, Vol. 104 (Springer, Berlin, 1993).
[14] D. Abraham, Phys. Rev. Lett. 44 (1980) 1165.
[15] J.O. lndekeu and J.M.J. van Leeuwen, Phys. Rev. Lett. 75 (1995) 1618.