PHYSICAL REVIEW E 66, 031605 共2002兲

Process of irreversible nucleation in multilayer growth. I. Failure of the mean-field approach

Paolo Politi1,* and Claudio Castellano2,†
1
Istituto Nazionale per la Fisica della Materia, Unità di Firenze, Via G. Sansone 1, 50019 Sesto Fiorentino, Italy
2
Istituto Nazionale per la Fisica della Materia, Unità di Roma I, and Dipartimento di Fisica, Università di Roma ‘‘La Sapienza,’’
Piazzale Aldo Moro 2, 00185 Roma, Italy
共Received 29 October 2001; published 27 September 2002兲
The formation of stable dimers on top of terraces during epitaxial growth is investigated in detail. In this
paper we focus on mean-field theory, the standard approach to study nucleation. Such theory is shown to be
unsuitable for the present problem, because it is equivalent to considering adatoms as independent diffusing
particles. This leads to an overestimate of the correct nucleation rate by a factor N, which has a direct physical
meaning: on average, a visited lattice site is visited N times by a diffusing adatom. The dependence of N on
the size of the terrace and on the strength of step-edge barriers is derived from well-known results for random
walks. The spatial distribution of nucleation events is shown to be different from the mean-field prediction, for
the same physical reason. In the following paper we develop an exact treatment of the problem.

DOI: 10.1103/PhysRevE.66.031605 PACS number共s兲: 64.60.Qb, 68.55.Ac, 68.35.Fx, 05.40.Fb

I. INTRODUCTION ters of j atoms ( j clusters, j⭓2). If only adatoms are mobile,

A crystal can be produced artificially with different tech-
niques: we can pull it from the melt, grow it from a solution
or obtain it via deposition from a gas or vapor phase onto a
suitable substrate. One of the key mechanisms of the growth
process is the formation of supercritical nuclei, that is the
nucleation—via diffusion and aggregation—of crystalline
clusters whose growth rate exceeds the decay rate.
In this paper we devote our attention to epitaxial growth
by atomic or molecular beams 关1兴: particles travel ballisti-
cally towards the growing surface where they undergo a ther-
mally activated diffusion process. The size i * of the critical
nucleus is typically a few units and its actual value depends
on several factors: the substrate and the type of adatoms
deposited determine the activation barriers for the different
atomistic processes, while the temperature and the flux de-
termine what processes are really relevant on the time scale
of the experiment. Here we consider the simplest case: nucle-
ation is irreversible (i * ⫽1), i.e., once two adatoms meet
they form a stable dimer. After nucleation, the stable nucleus
grows by capturing other adatoms.
We start by explaining in general terms the role of nucle-
ation in the different stages of epitaxial growth and by dis-
cussing how rate equations and mean-field theory deal with
it. The focus of the rest of the paper will be on nucleation on
top of terraces bound by descending steps 共often called sec-
ond layer nucleation兲.
In the submonolayer regime diffusion takes place on the
substrate: adatoms are deposited randomly and they diffuse
until they meet another wandering adatom or a growing clus-
ter.
the spatial density ␳ j of j clusters varies in time according to
the relation 关3兴 ⳵ t ␳ j ⫽U j⫺1 ⫺U j , where U j is the net rate of
the process ( j cluster)⇒( j⫹1 cluster). U j is the sum of a
‘‘growth term’’ and a ‘‘decay term’’: the growth term repre-
sents the aggregation of an atom into a j cluster and it has the
form ␴ j D ␳␳ j , where ␳ is the adatom density, D the adatom
diffusion constant, and ␴ j is an adimensional capture num-
ber; the decay term represents the thermal detachment of an
atom from a ( j⫹1) cluster and it has the form
⫺(4D/a 20 ) ␳ j⫹1 exp(⫺⌬E/T), where a 0 is the lattice constant
and ⌬E is the energy difference between a j cluster 共plus a
free adatom兲 and a j⫹1 cluster. If i * ⫽1 such term is absent,
because all j clusters are stable and the nucleation rate is
␻ RE⫽ ␴ 1 D ␳ 2 共the subscript standing for ‘‘rate equations’’兲.
The capture factor ␴ j is defined through the flux ⌽ j of
adatoms attaching to the j cluster: ⌽ j ⬅D ␴ j ␳ , and it ac-
counts for the different adatom densities surrounding islands
of different size. In a mean-field 共MF兲 approach such density
is taken as a constant, ␴ ⫽2d 关4兴 and the nucleation rate per
lattice site reads ␻ MF⫽2dD¯␳ 2 .
By monitoring the adatom density ¯␳ and the total density
of islands ¯␳ TOT⫽ 兺 j⭓2¯␳ j one realizes the existence of a re-
gime characterized by ¯␳ TOT almost constant in time, prior to
the coalescence regime 关5兴. The average distance ᐉ D between
islands can therefore be defined as ᐉ D d
⫽1/¯␳ TOT , d being the
dimensionality of the substrate: usually d⫽2, but in the fol-
lowing we will consider d⫽1 as well. ᐉ D is called the dif-
fusion length and it also gives the typical linear distance
traveled by an adatom before being incorporated in an island.
During the time 1/F, necessary for the deposition of a
monolayer, there is on average a nucleation event per area ᐉ D d

Rate equations 关2兴 are widely used to describe the pro- 关6兴. Therefore, according to mean-field theory, ␻ MFᐉ D/F d

cesses of adatom capture and adatom detachment from clus- ⬃1, i.e., D¯␳ 2 ᐉ D d
/F⬃1 with 共see Appendix A1兲 ¯␳
⬃(F/D)ᐉ D . So we obtain the result ᐉ D⬃(D/F) ␥ , with ␥
2

⫽ 15 in d⫽1 and ␥ ⫽ 61 in d⫽2. This relation is wrong in one

*Present address: Istituto di Fisica Applicata ‘‘Nello Carrara,’’ dimension, but essentially correct in two dimensions 关6,7兴. In
Consiglio Nazionale delle Ricerche, Via Panciatichi 56/30, 50127 Sec. IV A it will become clear that mean-field theory 共MFT兲
Firenze, Italy. Email address: politi@ifac.cnr.it gives the correct result for ␥ in d⫽2 because step-edge bar-
†
Email address: castella@pil.phys.uniroma1.it riers at island edges play no role in the submonolayer re-
gime.
1063-651X/2002/66共3兲/031605共14兲/$20.00 66 031605-1 ©2002 The American Physical Society
PAOLO POLITI AND CLAUDIO CASTELLANO
If adatoms freshly attached to a growing nucleus are im-
mobile the resulting island is normally fractal, but if they can
diffuse along the step-edge the island gets compact. Compact
islands are therefore obtained at not too low temperatures 共so
that thermally activated edge diffusion actually occurs兲
whilst nucleation is irreversible at not too high temperatures
共so that a dimer is thermally stable兲. In some experimental
systems both conditions are fulfilled in a given temperature
range: Pt共111兲/Pt in the range 350– 430 K 关8兴, Ag共100兲/Ag in
the interval 200–300 K 关9兴, and Fe共100兲/Fe between 300 and
500 K 关10兴 are some examples.
Once island coalescence has set off, most nucleation
events take place on top terraces. The obvious reason is that
even for layer-by-layer growth a new atomic layer starts to
form before the completion of the previous one and conse-
quently the growing surface is made up of a certain number
of exposed layers. Terraces can be classified as top (T), vici-
nal (V), and bottom 共B兲 terraces according to the type of
steps surrounding them: in general only top terraces attain a
size large enough to have a considerable probability of
nucleating a new island. Since this probability grows
abruptly from zero to one with increasing terrace size 关11兴, it
is possible to introduce a critical nucleation length ᐉ nuc 关12兴.
In the presence of step-edge barriers 关13兴, hindering inter-
layer transport, such length differs for the three types of ter-
races: ᐉ nuc
T
goes to zero with increasing barriers while ᐉ nucV
and ᐉ nuc remain finite in such limit 关12,14兴. Nucleations gen-
B
erally occur on top terraces because the size of the vicinal
and bottom terraces hardly reaches the nucleation length.
However, nucleation is a stochastic phenomenon and there-
fore dimer formation may also occur on vicinal terraces ev-
ery now and then, while on bottom terraces it is an exceed-
ingly rare event. These occasional ‘‘vicinal nucleations’’ are
an important stabilizing mechanism of the surface, even if
their relative weight decreases as barriers increase 关15,16兴.
The mean-field approach to evaluate the nucleation rate
for 共reversible and irreversible兲 nucleation on top terraces
has been worked out by Tersoff, van der Gon, and Tromp in
Ref. 关11兴 and since then their results have been widely used
to analyze experimental data and extract the values of step-
edge barriers. The mean-field approximation can also be
used in ‘‘mesoscopic’’ models of growth 关17兴 for implement-
ing the rule for the spatial distribution P(x) of nucleation
events, via the relation P(x)⬀ ␳ 2 (x).
Recently, several authors 关18 –20兴 have started to review
critically the MF approach. In Refs. 关18,20兴 authors are
mainly interested in the validity of MFT for different values
of the critical size i * : this check has been done through
scaling analysis and kinetic Monte Carlo simulations. A
deeper investigation of the irreversible case (i * ⫽1) is done
in Ref. 关19兴 in the limit of strong step-edge barriers, through
a quantitative approach based on the different time scales
involved in the nucleation process.
The inaccuracy of MF theory in dealing with the problem
of second layer nucleation has therefore already been pointed
out in the literature. In this work we analyze thoroughly the
validity of the MF theory for irreversible second layer nucle-
ation, we rigorously prove its inadequacy both for the nucle-
ation rate ␻ and the spatial distribution P(x) of nucleation
031605-2
PHYSICAL REVIEW E 66, 031605 共2002兲
sites and understand the physical origin of its failure. In the
following paper 关21兴 we compute exactly ␻ and P(x), pro-
viding the correct expressions that must be used in place of
the MF approximations.
We have organized this paper as follows. The three rel-
evant time scales for the nucleation process and the associ-
ated regimes are introduced and discussed in the following
section. In Sec. III we explain how we can get rid of the
stochastic nature of the deposition process 共Sec. III A兲 and
we introduce the quantities of interest in the paper, the spatial
distribution of nucleation events 共Sec. III B兲, and the nucle-
ation rate 共Sec. III C兲. In Sec. IV we show the equivalence of
MF theory with a model of noninteracting particles, both for
the nucleation rate 共Sec. IV A兲 and for the spatial distribution
共Sec. IV B兲. The final section contains a critical summary of
the results.
A short report of this work has been published in Ref.
关22兴.
II. TIME SCALES
We consider a top terrace of fixed linear size L, subject to
a flux F of particles. Once on the terrace, each particle moves
with a diffusion constant D until it meets another particle or
it leaves the terrace. At step edges an additional energy bar-
rier, usually called Ehrlich-Schwoebel 共ES兲 barrier 关13兴, re-
duces the rate D ⬘ of interlayer transport and the ES length
ᐉ ES⫽ 冉 冊
D
D⬘
⫺1 a 0 共1兲
measures the asymmetry between D and D ⬘ . In the follow-
ing the lattice constant a 0 will be taken as unit length (a 0
⫽1).
In general, throughout the paper we will consider discrete
space and time, i.e., particles moving on a lattice 共a square
lattice in d⫽2), at fixed time steps. However, we will some-
times use a continuum notation as well. The matching be-
tween discrete and continuum is straightforward: if t c is a
continuous time and t d its discrete counterpart 共number of
time steps兲, then t c ⫽t d ⌬t, where ⌬t⫽1/(2dD) is the time
an atom remains on each lattice site.
Let us now discuss the three time scales involved in the
problem 关19兴.
共1兲 The traversal time ( ␶ tr ) is the average time needed by
a diffusing particle to reach the terrace boundaries. In the
large L limit
␶ tr ⫽ ␤ L 2 /D, 共2兲
where ␤ is a numerical prefactor depending on the dimen-
sion d and the shape of the terrace. Its value is discussed
after Eq. 共4兲.
共2兲 The residence time ( ␶ res ) is the average time a particle
spends on the terrace. It is related to the average density ¯␳ of
adatoms via the relation 关19兴 ¯␳ ⫽F ␶ res , where the density ␳
共and therefore its average value ¯␳ ) can be determined 共see
Appendix A1兲 by solving the stationary diffusion equation in
the presence of a constant flux F,
PROCESS OF IRREVERSIBLE . . . . I . . . .
Dⵜ 2 ␳ ⫹F⫽0. 共3兲
Boundary conditions depend on the strength of the ES barrier
at descending steps. If the equilibrium adatom density—due
to thermal detachment from steps—is neglected, the bound-
␳ Depending on the relative size of ␶ res with respect to the
ary condition for ␳ is ⳵⬜ ␳ 兩 step⫽ 兩 , where ⳵⬜ is the
ᐉ ES step
local derivative in the direction perpendicular to the step
共directed inward the terrace兲. In a discrete picture 共for ex-
ample in d⫽1) if n⫽1 is a lattice site at the edge of the
terrace and n⫽0 is its fictitious neighbor outside the terrace,
we have ␳ (1)⫺ ␳ (0)⫽ ␳ (0)/ᐉ ES , i.e., ␳ (0)⫽ᐉ ES /(1
⫹ᐉ ES) ␳ (1)⬅a ␳ (1).
The solution of Eq. 共3兲 共see Appendix A1兲 gives the fol-
lowing result, valid both in one and two dimensions: ¯␳
⫽(F/D)( ␤ L⫹ ␣ ᐉ ES)L, where ␣ is another numerical factor
depending on the dimension d and on the shape of the ter-
race. We can finally write
␶ res ⫽ 共 ␤ L⫹ ␣ ᐉ ES兲 L/D. 共4兲
In the absence of ES barriers (ᐉ ES⫽0) ␶ res and ␶ tr are equal.
From Eqs. 共A7兲, 共A8兲, 共A18兲, and 共A50兲 we infer that in d
⫽1, ␤ ⫽1/12 and ␣ ⫽1/2. In d⫽2, for a circular terrace of
radius L, ␤ ⫽1/8 and ␣ ⫽1/2, while for a square terrace, ␤
⯝32/␲ 6 and ␣ ⫽1/4.
In the discrete picture ␶ res is clearly equal to the total
number of sites (N all) visited by an atom during its diffu-
sional motion on the terrace. Since the adatom stays on a
lattice site a time 1/(2dD) we have ␶ res ⫽N all /(2dD).
Hence, the residence time ␶ res , the average density ¯␳ and the
number N all of all sites visited by the adatom carry the same
piece of physical information, once F and D are set. The
quantity N all should not be confused with the number of
distinct sites (N dis) visited by an adatom: a given lattice site,
visited k times, contributes for 1 to N dis and for k to N all .
共3兲 The deposition time ( ␶ dep ) is the average time be-
tween a deposition event and the next one. For a terrace of
area A⫽L d ,
1 1
␶ dep ⫽ ⫽ . 共5兲
FA FL d
Physically sensible values for F, L and D imply that ␶ tr
Ⰶ ␶ dep . This relation indeed is L 2 /DⰆ1/FL d , i.e., D/F
ⰇL d⫹2 : we can now recall the diffusion length 关23兴 intro-
duced in Sec. I, ᐉ D⬃(D/F) ␥ and measuring the ‘‘maximal’’
size of a terrace in the absence of step-edge barriers. For
irreversible nucleation the exponent ␥ is equal to 关23兴
1/2(d⫹1) so that we obtain the condition ᐉ D 2(d⫹1)/(d⫹2)
ⰇL, i.e., ᐉ D ⰇL in d⫽1 and ᐉ D ⰇL in d⫽2. Smooth
4/3 3/2
growth requires that ᐉ DⰇ1; furthermore L is at most of order
ᐉ D if ᐉ ES⫽0, but for finite barriers it is 共much兲 smaller. We
conclude that the above conditions are fulfilled and that we
can safely suppose that ␶ tr Ⰶ ␶ dep .
This inequality has a consequence of primary importance:
processes involving more than two adatoms at a time can be
fully neglected. Once two adatoms are simultaneously on the
031605-3
PHYSICAL REVIEW E 66, 031605 共2002兲
terrace, they meet—if they do—on the time scale of the tra-
versal time ␶ tr . This fact is intuitively clear and it is proven
in Ref. 关21兴. The probability that a third atom lands in the
meanwhile is ␶ tr / ␶ dep , negligibly small. Irreversible nucle-
ation is therefore the result of two-adatoms processes only.
other two time scales, three different regimes may occur:
␶ tr ⯝ ␶ res Ⰶ ␶ dep zero or weak barriers 关 regime 共i兲兴 , 共6兲
␶ tr Ⰶ ␶ res Ⰶ ␶ dep strong barriers 关 regime 共ii兲兴 , 共7兲
␶ tr Ⰶ ␶ dep Ⰶ ␶ res infinite barriers 关 regime 共iii兲兴 . 共8兲
The difference between the three regimes is easily under-
stood. A nucleation may occur only if a new adatom is de-
posited before the previous one leaves the terrace. If ␶ res
Ⰶ ␶ dep 关regimes 共i兲 and 共ii兲兴 this is a rare event. When it
happens, the second atom finds the first one with a spatial
distribution that differs in cases 共i兲 and 共ii兲 共see Appendix B兲.
If ᐉ ESⰆL 关regime 共i兲兴 when the adatom reaches the edge of
the terrace, it gets off. Steps act as absorbing boundaries and
the adatom density vanishes there: ␳ has a parabolic shape
with a maximum in the middle of the terrace 共see Appendix
A1兲. If ᐉ ESⰇL 关regime 共ii兲兴 the adatom is pushed back from
the terrace edge several times before being able to descend.
Steps act as 共imperfect兲 reflecting walls and ␳ is approxi-
mately uniform over the whole terrace. In regime 共iii兲 when
a new adatom is deposited it always finds the previous one
still on the terrace, they both have a flat distribution and they
will certainly meet.
III. GENERAL FORMALISM
In the present and in the following paper 共Ref. 关21兴兲 we
are going to use a discrete formulation for particle dynamics,
both in space and in time. In this section, n indicates the
whole set of d integer numbers specifying the position of a
particle on the terrace. A nucleation event is assumed to oc-
cur when two adatoms are on the same lattice site, rather
than on neighboring lattice sites: this definition avoids use-
less mathematical complications, but retains all the physics
of the nucleation process.
A. Reduction to two particles deposited simultaneously
It is clear that the problem of dimer formation on a terrace
involves the study of the diffusion of two particles deposited
at different times: the spatial and temporal distributions of
landing events play therefore a prominent role. The incoming
flux of particles is supposed to be spatially and temporally
uniform 关24兴: a particle arrives on each lattice site with uni-
form probability p U n ⬅1/L and the interarrival time ␶ be-
d
tween two deposition events decays exponentially 关25兴,
P dep共 ␶ 兲 ⫽exp共 ⫺ ␶ / ␶ dep 兲 / ␶ dep . 共9兲
Let us now consider any ‘‘two-particle’’ quantity O, i.e.,
any quantity depending on the initial distributions of particle
1 and 2 and on their interarrival time ␶ .
PAOLO POLITI AND CLAUDIO CASTELLANO
Let p n (0)⫽p U n ⫽1/L be the initial uniform distribution
d
of an atom and p n ( ␶ ) its dynamical evolution at time ␶ 共in
the absence of other particles兲. If particle 1 is deposited at
time zero and particle 2 a time ␶ later, we call
n ( ␶ ),p n (0) 其 the resulting physical quantity. O might
O兵 p (1) (2)
be, for example, the probability p nuc that a deposited particle
nucleates a dimer before getting off the terrace 关see Eq.
n ( ␶ ), p n (0) 其 is known, one should evalu-
共19兲兴. Once O兵 p (1) (2)
ate its average over ␶ ,
⬁
兺
Ô⫽ P dep共 ␶ 兲 O兵 p (1)
n 共 ␶ 兲, p n 共 0 兲其.
(2)
␶ ⫽0
The crucial point is that if O is linear in the initial distribu-
tions p (1,2)
n of the two atoms 共as all quantities discussed in the
paper are兲, the above equation can be rewritten as
Ô⫽O兵 p eff
n , p n 共 0 兲其
(2)
in such a way that the average over ␶ is now included in an
effective initial distribution
⬁
n ⫽
p eff 兺
␶ ⫽0
P dep共 ␶ 兲 p (1)
n 共 ␶ 兲.
We can make more explicit the physical content of the
above reasoning, which is based on the linearity with respect
to the initial distribution of the two particles. Atom 2 is de-
posited with probability P dep( ␶ ) a time ␶ after atom 1, which
means that atom 2 has the probability P dep( ␶ ) to find atom 1
distributed according to p (1)n ( ␶ ): on average—and it is now
that linearity comes into play—atom 2 finds atom 1 with the
effective distribution given in Eq. 共12兲.
In this way we have reduced the problem of evaluating Ô
to the evaluation of O for two particles deposited simulta-
neously. Thus we can ignore the stochasticity of the deposi-
tion process and assume that atoms 1 and 2 land at the same
time, but the actual initial distribution for atom 1 共the uni-
form distribution兲 is replaced by p eff
n .
The next task is then the determination of p eff
n . The func-
tion p n ( ␶ ) 共discussed in Appendix A2兲 is the distribution of
the first adatom at time ␶ , i.e., the solution of the diffusion
equation for a single particle with initial condition
1 strate and the size of islands is made as uniform as possible
p n 共 ␶ ⫽0 兲 ⫽p U
n⫽ . 共13兲
Ld
The sum of p n ( ␶ ) over all times ␶ is the solution of the
stationary diffusion equation 共3兲 共of its discretized version,
actually, see Appendix A1兲, whose normalized form will be
indicated with p Sn (S standing for stationary兲. It has, in gen-
eral, a parabolic form, and in particular in d⫽1, p Sn
1 ‘‘system-dependent.’’ Hence, the growth law L( ␶ ) of the ter-
⫽ 关 ᐉ ESL⫹(L⫹1)n⫺n 2 兴 /[ᐉ ESL 2 ⫹ L(L⫹1)(L⫹2)].
6 race size must be supplied beyond ␻ (L), and it depends on
In Appendix B it is shown that in all dimensions we can
write
PHYSICAL REVIEW E 66, 031605 共2002兲
␶ res
n ⫽
p eff pS . 共14兲
␶ dep ⫹ ␶ res n
The physical content of Eq. 共14兲 is readily understood. For
infinite barriers 关regime 共iii兲兴 p eff n ⫽ p n ⫽1/L : the first par-
S d
ticle cannot escape from the terrace and its distribution is
still uniform and normalized when the second one lands. For
strong but finite barriers 关regime 共ii兲兴 p eff n ⫽( ␶ res / ␶ dep )p n
S
⫽( ␶ res / ␶ dep ) (1/L ): most of the particles that arrive on the
d
terrace leave it before another particle lands, but the distri-
bution of the first particle remains practically uniform be-
共10兲 cause many attempts are needed to overcome the ES barrier.
In the limit of zero or weak barriers 关regime 共i兲兴 p eff n
⫽( ␶ res / ␶ dep )p Sn and p Sn vanishes on the edges, reflecting the
presence of the absorbing boundaries.
B. The spatial distribution of nucleation events
共11兲 In the preceding section we have explained how to trans-
form the original problem into the new problem of two at-
oms deposited at the same time, with normalized distribu-
tions p Sn 共the first兲 and p m
U
共the second兲. We can now define
the probability R(n,t) that a nucleation event occurs on site
n at time t and introduce the following quantities:
共12兲
P共 n 兲⫽ 兺t R 共 n,t 兲 , 共15兲
W⫽ 兺n P 共 n 兲 . 共16兲
P(n) is the spatial distribution of nucleation events and W
is the probability that two atoms, both on the terrace at time
zero, meet before leaving the terrace. It is useful to consider
the normalized spatial distribution P (N) (n)⫽ P(n)/W as
well.
C. The nucleation rate
The nucleation rate ␻ is defined as the number of nucle-
ation events per unit time on the whole terrace of size L,
irrespective of the spatial location of the meeting point. This
quantity is of great importance because it is related to the
probability of second layer nucleation. In a classical experi-
ment 关26兴 a fraction of a monolayer is deposited on the sub-
through an annealing procedure. Starting from this template
a second dose of atoms is deposited and nucleation on top of
existing islands is monitored. ␻ (L) enters in the interpreta-
tion of this experiment because the probability P(t) that a
nucleation event has occurred on a terrace by time t is P(t)
⫽1⫺exp兵⫺兰t0d␶␻关L(␶)兴其. The rate ␻ (L) is defined and
evaluated for a constant terrace size L: in the experiment
instead, L grows in time and the time dependence of L is
the specific morphology of the surface and the experimental
setup. In other words, the nucleation rate—on the one
031605-4
PROCESS OF IRREVERSIBLE . . . . I . . . .
hand—has a very general and basic meaning, but—on the
other hand—it can hardly be measured directly. This means
that, despite our results for ␻ (L) being exact, the evaluation
of L(t) introduces some approximations in the interpretation
of experimental results, whose accuracy depends on the de-
tail of the system considered. In addition, some secondary
effects, as steering and nonuniform barriers, may further
complicate the problem.
We now connect ␻ to P(n) and W. In Sec. II we ex-
plained that only processes involving two adatoms are rel-
evant for studying irreversible nucleation, because ␶ tr
Ⰶ ␶ dep . If we define the nucleation probability per atom,
p̂ nuc , we can write the nucleation rate as the number of at-
oms landing on the terrace per unit time (FL d ⫽1/␶ dep ) times
the nucleation probability per atom,
␻ ⫽FL d p̂ nuc .
The quantity p̂ nuc is the probability that a deposited particle
nucleates a dimer before getting off the terrace and it can be
written as
⬁
p̂ nuc⫽ 兺
␶ ⫽0
P dep共 ␶ 兲 p nuc兵 p (1)
n 共 ␶ 兲, p n 共 0 兲其,
(2)
where the dependence on the initial distributions of atoms 1
and 2 has been made explicit. We stress that the dependence
on the initial distributions occurs via the full diffusion pro-
cess. For example, for independent particles in one dimen-
sion, the explicit form of p nuc is
L ⬁
n 共 ␶ 兲 ,p n 共 0 兲 其 ⫽
p nuc兵 p (1) (2)
兺 兺
m⫽1 t⫽0
n 共 ␶ 兲, p n 共 0 兲其,
p m,m 共 t 兲 兵 p (1)
where p m,m (t) is the solution of the diffusion equation in two
dimensions with the initial condition given by the product
p (1) ( ␶ )p (2) (0) 共see Ref. 关21兴 for more details兲.
Because of the linearity of p nuc , we have
p̂ nuc⫽p nuc兵 p eff
n , p n 共 0 兲其.
(2)
The nucleation probability per atom can be thought as the
probability that atom 1 is still on the terrace when atom 2 is
deposited, times the probability they meet before getting off
the terrace. This is exactly what emerges from Eq. 共20兲 once
expression 共14兲 for the effective distribution p eff
n is inserted,
␶ res
p̂ nuc⫽ p 兵 p S , p (2) 共 0 兲 其 .
␶ dep ⫹ ␶ res nuc n n
The normalization factor of p eff n is the probability that
atom 1 is still on the terrace when the next one shows up; for
infinite ES barriers ( ␶ res Ⰷ ␶ dep ) such a probability is trivially
1, while for weak and strong barriers ( ␶ res Ⰶ ␶ dep ) it is
␶ res / ␶ dep . The remaining quantity on the right-hand side
(p nuc兵 p Sn ,p (2)
n (0) 其 ) is the probability that two atoms, both on
PHYSICAL REVIEW E 66, 031605 共2002兲
the terrace at time zero 关 p Sn and p (2)
n (0) are normalized兴,
meet before descending. Therefore it coincides with W and
we finally obtain
␶ res
␻ ⫽FL d W. 共22兲
␶ dep ⫹ ␶ res
D. Noninteracting particles
We are considering a system such that once adatoms come
together an immobile dimer is formed irreversibly: adatoms
stop diffusing and the dimer does not dissociate. It turns out
to be of great help to consider also an artificial model, with
adatoms treated as independently diffusing particles: even if
they meet on the same lattice site they go on diffusing and
therefore they can cross each other several times before leav-
ing the terrace. We consider all these meetings as ‘‘fictitious
共17兲 nucleations,’’ and define also for noninteracting particles the
quantities mentioned above: the nucleation rate ␻ NI , the spa-
tial distribution P NI(n), and the total number W NI of nucle-
ation events, the subscript NI standing for ‘‘noninteracting.’’
IV. EQUIVALENCE OF MEAN-FIELD THEORY
AND NONINTERACTING PARTICLES MODEL
共18兲
A. The nucleation rate
We have introduced the nucleation rate in Sec. III C and
obtained Eq. 共22兲. W is the nucleation probability between
two atoms that are both on the terrace at time zero. For the
model of noninteracting adatoms W should be replaced by
W NI , the average number of meetings between the two inde-
pendent particles. Of course W NI can be larger than 1.
The simplest and less interesting case is the regime 共iii兲 of
(2) infinite barriers. In such a case W is trivially 1 and ␻
⫽FL d ⫽1/␶ dep , i.e., any particle deposited on a terrace does
共19兲 form a dimer. In a sense, this limit is unphysical for mean-
field theory because ¯␳ and ␻ MF⫽2dDL d¯␳ 2 diverge when
ᐉ ES→⬁.
In the other two regimes 共weak and strong barriers兲, Eq.
共22兲 becomes
␶ res
␻ ⫽FL d W⫽FL 2d¯␳ W, 共23兲
共20兲
␶ dep
where we have used the relations ␶ dep ⫽(FL d ) ⫺1 and F ␶ res
⫽¯␳ . We can repeat the same procedure for noninteracting
particles and obtain
␶ res
␻ NI⫽FL d W ⫽FL 2d¯␳ W NI . 共24兲
␶ dep NI
共21兲
It is possible to relate W and W NI to single-particle quan-
tities, N dis and N all 共see Sec. II兲. They are the number of
distinct (N dis) and all (N all) sites visited by a single walker
diffusing on the terrace 关27兴. Let us assume one of the two
adatoms fixed on site s. w(s) is the probability that the dif-
fusing adatom visits site s before getting off the terrace. W is
then the average value of w(s), W⯝ 兺 s w(s)/L d . The quan-
tity 兺 s w(s) is nothing but the total number of distinct sites
031605-5
PAOLO POLITI AND CLAUDIO CASTELLANO
FIG. 1. 共a兲 Log-log plot of W 共empty circles兲, N dis /L 2 共full
circles兲, W NI 共empty squares兲, N all /L 2 共full squares兲 versus ᐉ ES for
d⫽2 and L⫽20. 共b兲 The same quantities plotted versus L for d
⫽2 and ᐉ ES⫽0.
N dis visited by the diffusing adatom, so that W⯝N dis /L d .
The same argument for noninteracting particles gives W NI
⯝N all /L d .
The relations
N dis N all
W⯝ and W NI⯝
L d
Ld
have been derived under the assumption that one atom is
immobile. In Fig. 1 we compare numerically the values of
single-particles quantities (N dis /L d ,N all /L d ) with two-
particle quantities (W,W NI). The former have been calcu-
lated via Monte Carlo simulations and the latter through the
numerical solution of the discrete diffusion equation for two
atoms on a terrace 共discussed in detail in Ref. 关21兴兲: it comes
out that relations 共25兲 are well satisfied, so that assuming one
atom as immobile is perfectly reasonable for the evaluation
of W and W NI .
If we insert the relations 共25兲 into Eqs. 共23兲 and 共24兲 we
obtain
␻ ⯝FL d¯␳ N dis ,
PHYSICAL REVIEW E 66, 031605 共2002兲
FIG. 2. Plot of the correction factor ␻ MF / ␻ and of N
⬅N all /N dis versus ᐉ ES /L, for a square terrace of size L⫽20.
␻ NI⯝FL d¯␳ N all . 共27兲
Since N all is related to the residence time by ␶ res
⫽N all /(2dD), we can write
␻ NI⯝2dFL d¯␳ D ␶ res ⫽2dDL d¯␳ 2 ⫽ ␻ MF . 共28兲
In this way we have shown that for the nucleation rate the
mean-field treatment is equivalent to considering particles as
noninteracting, i.e., counting also meeting events following
the first one. For this reason the mean-field value is an over-
estimate of the correct nucleation rate. Furthermore we have
proven that
␻ MF W NI N all
⯝ ⯝ ⬅N. 共29兲
␻ W N dis
In Fig. 2, the comparison of ␻ MF / ␻ , computed exactly in
the companion paper 关21兴 with N all /N dis , evaluated numeri-
cally, shows clearly that Eq. 共29兲 is valid with great accuracy.
The correction factor N depends on well-known proper-
ties of single particles performing a random walk. The nu-
merator N all is just 共see Sec. II兲: N all⫽2dL( ␤ L⫹ ␣ ᐉ ES). The
value of the denominator N dis is well known 关28兴 in the
absence of step-edge barriers, being of order L in d⫽1 and
共25兲
of order L 2 /lnL in d⫽2, and it is trivial in the limit of
infinite barriers, being exactly equal to L d . Hence in d⫽1
we obtain N⬃(L⫹ ␣ ᐉ ES), for all ᐉ ES .
In d⫽2 we have the limiting expressions N⬃lnL for
weak barriers and N⬃ᐉ ES /L for strong ones. For intermedi-
ate barriers it is possible to find a simple approximate ex-
pression for N dis and therefore an interpolation between the
two limits. The atom performs on average a number N tr
⫽ ␶ res / ␶ tr of traversals of the island. During a single tra-
versal each site has a probability p 1 ⬃(1/lnL) to be visited.
After all N tr traversals the probability p s that a generic site
has been visited at least once is given by 1⫺p s ⫽(1
⫺p 1 ) N tr . Hence we can estimate the number of distinct vis-
ited sites as
共26兲 N dis⫽L 2 p s ⫽L 2 关 1⫺ 共 1⫺ p 1 兲 ␶ res / ␶ tr 兴 . 共30兲
031605-6
PROCESS OF IRREVERSIBLE . . . . I . . . .
FIG. 3. Log-log plot of N all /N dis versus ᐉ ES for d⫽2 and L
⫽20.
This expression assumes all traversals to be independent,
which is clearly not strictly correct. However, it gives the
right values in the limits ᐉ ES⫽0 and ᐉ ES⫽⬁ and for inter-
mediate barriers its accuracy can be tested numerically. In
Fig. 3 we have plotted the ratio N all /N dis as a function of
ᐉ ES , for L⫽20. The picture shows a reasonable agreement
between the analytical estimate and the numerical simula-
tion.
B. The spatial distribution
In the preceding section we have shown that MF theory
overestimates the nucleation rate by the quantity N because
it counts all meetings between two noninteracting adatoms.
We are now going to prove that the identification of the
mean-field approach as a theory of noninteracting particles is
valid for the spatial distribution of nucleation events as well.
We adopt a continuum notation so that a single proof is suf-
ficient to demonstrate that P NI(x) and ␳ 2 (x) are proportional
in any dimension, for any value of the ES barrier and for any
terrace shape. In the regimes of strong and infinite ES barrier
this result is trivial because both ␳ 2 (x) and P NI(x) are con-
stant.
We face the problem of determining the quantity P NI(x)
for a pair of adatoms, one with initial distribution p (2) (x,0)
⫽p U (x) and the other with the effective distribution
p (1) (x,0)⫽ p eff(x)⫽ 兺 t p (2) (x,t)⫽ ␳ (x).
We can consider the coordinates (x 1 ,x 2 ) of the two atoms
(x 1,2 are vectors in a d-dimensional space兲 as defining the
position x⫽(x 1 ,x 2 ) of a single particle in a space of dimen-
sionality d ⬘ ⫽2d. This particle moves according to the dif-
fusion equation ⳵ t p⫽ (D/2) ⵜ 2 p. The factor 21 appears be-
cause of the different time step ⌬t employed in describing a
single walker (⌬t⫽1/2dD) or two walkers (⌬t⫽1/2d ⬘ D)
on a terrace.
Integrating in time and defining P(x)⫽ 兰 ⬁0 dtp(x,t) one
finds
D 2
ⵜ P 共 x兲 ⫽⫺p 共 x,0兲 . 共31兲
2
An interchange of the two particles 关 p (1) (x,0)⫽p U (x)
031605-7
PHYSICAL REVIEW E 66, 031605 共2002兲
and p (2) (x,0)⫽ ␳ (x)] is equally legitimate and it is useful to
use a symmetrized form for p(x,0):
1
p 共 x,0兲 ⫽ 关 ␳ 共 x 1 兲 p U 共 x 2 兲 ⫹ p U 共 x 1 兲 ␳ 共 x 2 兲兴 . 共32兲
2
Notice that ␳ (x) is also the solution of the equation
Dⵜ 2 ␳ (x)⫽⫺p U (x). Therefore,
D 2 D
ⵜ P 共 x兲 ⫽ 关 ␳ 共 x 1 兲 ⵜ 22 ␳ 共 x 2 兲 ⫹ ␳ 共 x 2 兲 ⵜ 21 ␳ 共 x 1 兲兴 , 共33兲
2 2
where ⵜ 2i acts on x i only and ⵜ 2 ⬅ⵜ 21 ⫹ⵜ 22 . Hence,
ⵜ 2 P 共 x兲 ⫽ⵜ 2 关 ␳ 共 x 1 兲 ␳ 共 x 2 兲兴 , 共34兲
i.e., the function ␹ (x)⫽ P(x)⫺ ␳ (x 1 ) ␳ (x 2 ) is harmonic. It is
easy to show 关29兴 that ␹ (x) must be identically zero. Hence,
P 共 x兲 ⫽ ␳ 共 x 1 兲 ␳ 共 x 2 兲 . 共35兲
If we set x 1 ⫽x 2 ⫽x, the left-hand side is just the nucle-
ation probability P NI(x) at point x between two noninteract-
ing adatoms, and the right-hand side is the mean-field pre-
diction. Notice that we have not used the explicit form of
p U (x). Hence, the proof holds for any initial spatial distri-
bution p U (x), so that the equivalence between MF theory
and the noninteracting particles model is true even if atoms
are not deposited uniformly.
So far we have rigorously shown that the MF result for
P(n) is not exact. However, one may wonder whether the
error introduced by taking into account all meeting events
following the first one is expected to be large or negligible.
We address this issue by evaluating the relative weight of
successive encounters for noninteracting particles.
Let us consider noninteracting particles and define ␮ j as
the fraction of times the jth meeting event actually occurs.
Clearly ␮ 0 ⫽1 and ␮ j ⭓ ␮ j⫹1 . Let us also define the normal-
ized distribution for the jth nucleation event P NIj (n). Notice
that P NI
1 (n)⫽ P (N)
(n), the distribution for interacting par-
ticles.
The total distribution of nucleation sites is simply
⬁
P NI共 n 兲 ⫽ 兺 ␮ j P NI
j 共n兲 , 共36兲
j⫽1
and the quantities W and W NI are given by
W⫽ ␮ 1 , 共37兲
⬁
W NI⫽ 兺n P NI共 n 兲 ⫽ 兺
j⫽1
␮j . 共38兲
If we now introduce the normalized distribution of all
fictitious nucleations following the first one,
PAOLO POLITI AND CLAUDIO CASTELLANO
⬁ TABLE I. We report the nucleation rate ␻ and the correction
兺 ␮ j P NIj 共 n 兲
j⫽2
P̂ NI共 n 兲 ⫽ ⬁ ,
兺
j⫽2
␮j
we can write
⬁
P NI共 n 兲 ⫽ ␮ 1 P NI
1 共 n 兲⫹ 兺
j⫽2
␮ j P NI
j 共 n 兲 ⫽W P
(N)
共 n 兲 ⫹ P̂ NI共 n 兲
⫻共 W NI⫺W 兲 .
For weak barriers, in d⫽1, the weight W of the first term is
constant, while the second one diverges as L. As a conse-
quence, for large L the distribution P NI(n) is dominated by
the contribution of the fictitious successive nucleations. In
d⫽2 the first term goes as 1/lnL while the second is con-
stant. Again, for large L, the contribution of first nucleation
events becomes negligible. For strong and infinite barriers,
W⫽1 while W NI is infinite, so P NI(n) coincides with
P ⬁NI(n).
In all cases the MF expression for the spatial distribution
of nucleation sites 关 P NI(n) 兴 is dominated for large L by the
contribution of the fictitious nucleations following the first
one. A priori there is no reason for supposing that the distri-
bution of the jth nucleation event is equal to the distribution
of the first one, so we expect that the difference between the
MF spatial distribution and the exact result persists for all
values of L. This will be checked and confirmed in the fol-
lowing paper 关21兴.
V. CONCLUSIONS
This paper has been devoted to an accurate investigation
of the mean-field approach to the problem of irreversible
nucleation. The main outcome is the proof that MFT is
equivalent to a model where particles do not interact and all
their meetings are counted as fictitious nucleations.
In the regime of infinite ES barriers, MFT simply breaks
down because it predicts a diverging nucleation rate, in con-
trast to the correct value ␻ ⫽FL d . In the other, physically
more interesting, regimes the equivalence of MFT with the
model of noninteracting particles implies that ␻ MF overesti-
mates the correct nucleation rate by the factor N
⫽N all /N dis . This ratio has a clear meaning: a diffusing ada-
tom passes N times on a visited site. It depends on single-
particle quantities (N all ,N dis) whose expressions are well
known from the theory of random walks.
In Table I we summarize the value of the correction factor
N in regimes 共i兲 and 共ii兲 and we report the approximate ex-
pressions for the nucleation rate ␻ . They are approximate in
the sense that numerical prefactors are neglected, but they
scale correctly with L,D,D ⬘ ,a 0 . The lattice constant a 0 has
been reintroduced in order to give dimensionally correct
espressions. Also, we have made explicit the dependence of
ᐉ ES on D and D ⬘ , so that only basic quantities appear.
The expression ␻ ⬃F 2 L 5 a 0 /D ⬘ , valid in two dimensions
for strong step-edge barriers has already been given in Ref.
PHYSICAL REVIEW E 66, 031605 共2002兲
factor N in the two regimes of weak 共i兲 and strong 共ii兲 step-edge
barriers and in one and two dimensions. The conditions defining
共39兲 regimes 共i兲 ( ␶ res ⯝ ␶ tr ) and 共ii兲 ( ␶ tr Ⰶ ␶ res Ⰶ ␶ de p ) are written in
terms of the basic quantities L,D,D ⬘ ,a 0 .
共i兲 共ii兲
D⫺D ⬘ L D L
Ⰶ Ⰷ D ⬘ ⰇFL d⫹1 a 0
D⬘ a0 D⬘ a0
␻ N ␻ N
2 4
d⫽1 F L L F 2L 3a 0 D
共40兲 D a0 D⬘ D⬘
d⫽2 F 2L 6 ln(L/a0) 2 5
F L a0 Da 0
Dln(L/a0) D⬘ D ⬘L
关19兴. It is worth repeating that the nucleation rate in this limit
does not depend on the diffusion constant D so that the
nucleation rate cannot be promoted by using surfactants.
Application of MFT is acceptable only in the regime of
vanishing barriers in two dimensions, because in this case
the correction factor 关 N⫽ln(L/a0)兴 is a small number, for
realistic terrace sizes.
In order to obtain exact expressions for ␻ it is necessary
to have an accurate estimate of W, or equivalently of N dis . W
is a function of the terrace size L and of the ES length ᐉ ES :
for strong barriers W⫽1, while for weak barriers W⬃1 in
d⫽1 and W⬃1/lnL in d⫽2. So, for realistic values of L, W
depends on L and ᐉ ES much more weakly than the other
quantities appearing in ␻ ⫽FL 2d¯␳ W. However, its depen-
dence is not fully negligible: Figs. 1共a兲 and 1共b兲 show that
共in d⫽2) for L⫽20, W varies by a factor 10 by changing
ᐉ ES from zero to infinity and for ᐉ ES⫽0, W varies by a factor
3 by changing the terrace size from L⫽4 to L⫽32
(ln32/ln4⫽2.5). For comparison, the quantity ¯␳ , which ap-
pears along with W in the expression for ␻ , varies by a
factor 50 by changing ᐉ ES /L from 0 to 6 and by a factor 64
by changing L from 4 to 32, for ᐉ ES⫽0. The problem of the
exact determination of W will be tackled again in the follow-
ing paper 关21兴.
A last comment on rate equations and mean-field approxi-
mation is in order here. According to the former, the nucle-
ation rate is written ␻ RE⫽ ␴ 1 D ␳ 2 and the latter corresponds
to taking ␴ 1 as a constant. In general, ␴ 1 is defined through
the relation ⌽ 1 ⫽D ␴ 1 ␳ , where ⌽ 1 is the flux of atoms at-
taching to an adatom. The resulting relation ␻ RE⫽⌽ 1 ␳ is
exact, if ⌽ 1 is evaluated correctly; for example, we can solve
the diffusion equation for a single walker on the terrace,
where a second walker is taken as an absorbing sink. Since
we have shown that the nucleation rate can be evaluated
assuming an atom as immobile, such treatment is essentially
correct. In other words, if the capture number is not taken as
a constant, the expression ␻ RE⫽ ␴ 1 D ␳ 2 may give correct
results, but this method has nothing to do with the usual
mean-field approach.
Finally, with regard to the spatial distribution, we have
provided a very general demonstration of the equivalence
between mean field and noninteracting particles. We have
031605-8
PROCESS OF IRREVERSIBLE . . . . I . . . . PHYSICAL REVIEW E 66, 031605 共2002兲

also shown that the difference between P(n) and P NI(n) is In a continuum formalism, the equation is D ⳵ 2x ␳ ⫹F⫽0 and
not an effect of the finite size of the terrace and it remains the solution in d⫽1 is
true for large L. The full computation of the spatial distribu-
tion of nucleation events requires the solution of the dynami- F
␳共 x 兲⫽ 关 ᐉ L⫹Lx⫺x 2 兴 . 共A5兲
cal problem of two interacting atoms diffusing on a terrace. 2D ES
This problem will be solved analytically and/or numerically
in the following paper 关21兴. In d⫽2 the solution of the continuum equation is as easy
as in d⫽1 if we specialize to a circular terrace. If L now
denotes the radius, the solution for generic ᐉ ES is
ACKNOWLEDGMENTS

Authors gratefully acknowledge useful discussions and F

␳共 r 兲⫽ 关 L 2 ⫹2Lᐉ ES⫺r 2 兴 . 共A6兲
correspondance with Jim Evans, Thomas Michely, and 4D
Filippo Colomo.
We finally evaluate the average density ¯␳ on the terrace
APPENDIX A: SINGLE PARTICLE ON A TERRACE
F
¯␳ ⫽ L 共 L⫹6ᐉ ES兲 d⫽1, 共A7兲
In this appendix we summarize the behavior of a single 12D
particle on a terrace for all values of the Ehrlich-Schwoebel
length ᐉ ES . F
¯␳ ⫽ L 共 L⫹4ᐉ ES兲 d⫽2 关circular terrace兴. 共A8兲
8D
1. The stationary adatom density
The discrete evolution equation for a particle in a cubic 2. The dynamical problem in one dimension
d-dimensional space is
We now summarize the dynamical behavior of a single
1 particle on a one-dimensional terrace. The two-dimensional
p n 共 t⫹1 兲 ⫽
2d 兺␦ p n⫹ ␦共 t 兲 , 共A1兲 case is treated in the following section.
The discrete evolution equation for the particle is
where n⫹ ␦ indicates a neighbor of the site n. If we sum over p n 共 t⫹1 兲 ⫽ 21 关 p n⫹1 共 t 兲 ⫹ p n⫺1 共 t 兲兴 , 共A9兲
⬁
t and define the quantity ␳ n ⫽ 兺 t⫽0 p n (t) we obtain

冋兺 册
with the usual boundary conditions p 0 (t)⫽ap 1 (t) and
p L⫹1 (t)⫽a p L (t). The solution is found by separating the
␳ n⫹ ␦ ⫺2d ␳ n ⫹2dp n 共 0 兲 ⫽0. 共A2兲
␦ space and time variables, p n (t)⫽X(n)F(t),

The terms in square brackets give the discrete Laplacian F 共 t⫹1 兲 X 共 n⫹1 兲 ⫹X 共 n⫺1 兲
⫽␭⫽ , 0⬍␭⬍1.
of ␳ n ; therefore the sum ␳ n ⫽ 兺 t p n (t) is simply the solution F共 t 兲 2X 共 n 兲
of the stationary diffusion equation in the presence of the 共A10兲
flux 2dp n (0).
In d⫽1, for constant p n (0), it is possible to find the exact The temporal part is F(t)⫽␭ t F(0). The spatial part has the
discrete solution for any value of ᐉ ES , once we remark that general form
the general solution of the homogeneous equation is ␳ n ⫽c 0
X 共 n 兲 ⫽Asin共 n ␾ 兲 ⫹Bcos共 n ␾ 兲 , 共A11兲
⫹c 1 n and a particular solution of the nonhomogeneous
equation is ␳ n ⫽⫺c 2 n 2 关the factor c 2 depending on the con- which gives ␭⫽cos ␾. The boundary conditions determine
stant term in Eq. 共A2兲兴. the values of A, B, and ␾ .
Boundary conditions are ␳ 0 ⫽a ␳ 1 and ␳ L⫹1 ⫽a ␳ L , where In particular, by imposing the boundary condition in n
a⫽ᐉ ES /(1⫹ᐉ ES) goes from 0 to 1 as the Ehrlich-Schwoebel ⫽0 one obtains B⫽bA with b⫽asin ␾/(1⫺acos ␾). Using
length ᐉ ES varies from 0 to ⬁. this relation and imposing the other boundary condition in
The explicit expression of ␳ n is n⫽L⫹1 one obtains
1 共 a 2 ⫺1 兲 sin ␾
␳ n ⫽ 关 ᐉ ESL⫹ 共 L⫹1 兲 n⫺n 2 兴 . 共A3兲 tan共 L ␾ 兲 ⫽ . 共A12兲
L 共 1⫹a 2 兲 cos ␾ ⫺2a
Its normalized version is This equation has L solutions that we label as ␾ k with k
⫽1, . . . ,L. Then the general solution is
␳n 1
p Sn ⬅ ⫽ L
L 共 L⫹1 兲共 L⫹2 兲
兺n ␳ n ᐉ ESL 2 ⫹
6
p n共 t 兲 ⫽ 兺
k⫽1
cost 共 ␾ k 兲 X k 共 n 兲 , 共A13兲

⫻ 关 ᐉ ESL⫹ 共 L⫹1 兲 n⫺n 2 兴 . 共A4兲 with

031605-9
PAOLO POLITI AND CLAUDIO CASTELLANO PHYSICAL REVIEW E 66, 031605 共2002兲

X k 共 n 兲 ⫽A k sin共 ␾ k n 兲 ⫹B k cos共 ␾ k n 兲 . 共A14兲 k␲

冉 冊
冋 册
sin
6 2 Lk ␲

冋 册
Given p n (t) one can compute S(t), the probability that an A k ⬅A Sk ⫽ sin .
adatom is still on the terrace at time t after deposition 共sur- L 共 L⫹1 兲 共 L⫹2 兲
2
3
k␲ 2 共 L⫹1 兲
sin
vival probability兲, 2 共 L⫹1 兲
L
共A23兲
S共 t 兲⬅ 兺
n⫽1
p n共 t 兲 . 共A15兲 As shown in Appendix A1, p Sn is the normalized version
of ␳ n ⫽ 兺 ␶⬁⫽0 p n ( ␶ ) where p n ( ␶ ) is the solution of the diffu-
Another important quantity is the residence time, defined as sion equation with uniform initial condition p U n . Writing ex-
plicitly the sum we obtain
⬁

␶ res ⬅ 兺 t 关 S 共 t⫺1 兲 ⫺S 共 t 兲兴
冉 冊 冉 冊
共A16兲 ⬁ ⬁ L
k␲ nk ␲
t⫽1
兺 p n共 ␶ 兲 ⫽ ␶兺
␶ ⫽0
兺 A Uk cos␶
⫽0 k⫽1 L⫹1
sin
L⫹1
because S(t⫺1)⫺S(t) is the probability that the particle 共A24兲
stays on the terrace exactly a time t. It is easy to check that

冋兺 册 冉 冊
L
AU nk ␲
⬁ ⬁
兺
冉 冊
L L k
⫽ sin . 共A25兲
␶ res ⫽ 兺 S 共 t 兲 ⫽ 兺 p n共 t 兲 ⫽ 兺 ␳n . 共A17兲 k⫽1 k␲ L⫹1
t⫽0 n⫽1 t⫽0 n⫽1 1⫺cos
L⫹1
Recalling Appendix A1, for the initial distribution p n (0) Hence
n ⫽1/L we have, for any value of the ES barrier,
⫽p U
AU AU
共 L⫹1 兲共 L⫹2 兲
冉 冊 冋 册
k k
␶ res ⫽ ⫹ᐉ ESL. 共A18兲 A Sk ⬀ ⫽ , 共A26兲
6 k␲ k␲
1⫺cos 2 sin2
L⫹1 2 共 L⫹1 兲
In order to pass to a continuous time we have to multiply
it by ⌬t⫽1/2D. For large L, ␶ res ⫽L/D(L/12⫹ᐉ ES/2). This as can be easily verified by comparing Eq. 共A22兲 with Eq.
result agrees with the relation ¯␳ ⫽F ␶ res 关see Eq. 共A7兲兴. 共A23兲.
Unfortunately it is not possible to solve explicitly Eq. If we sum p n (t) over n 关see Eq. 共A20兲兴 we obtain the
共A12兲 for generic values of a: we now discuss the two limit survival probability

冉 冊 冉 冊 冉 冊
cases where an explicit solution is possible. L
k␲ L k␲ k␲
a. Zero barriers
S共 t 兲⫽ 兺 A k cost
k⫽1 L⫹1
sin
2 L⫹1
sin
2
For ᐉ ES⫽0 (a⫽0), the allowed values of ␾ k are

␾ k⫽
k␲
共A19兲
⫻cosec 冉 冊1 k␲
2 L⫹1
. 共A27兲
共 k⫽1, . . . ,L 兲
L⫹1
The distribution p n (t) is in general a sum of exponential
and the general solution is decays

冉 冊 冉 冊 冉 冊 冋 冉 冊册
L
L
k␲ nk ␲ nk ␲ k␲
p n共 t 兲 ⫽ 兺 A k cos t
L⫹1
sin
L⫹1
共A20兲 p n共 t 兲 ⫽ 兺
k⫽1
A k sin
L⫹1
exp tln cos
L⫹1
.
k⫽1
共A28兲
with
It can be considered as a single exponential when the second

冉 冊
L
2 nk ␲ slowest decaying exponential is negligible. For LⰇ1 this
A k⫽
L⫹1 兺 p n共 0 兲 sin
n⫽1 L⫹1
. 共A21兲 means

In particular, two forms of p n (0) are most interesting to

us. For a uniform distribution p n (0)⫽p U n ⫽1/L the coeffi-
exp ⫺冋冉 冊 册 2␲
L
2
t
2
Ⰶ1 ⇒ tⰇ
L2
2␲2
⯝ ␶ tr . 共A29兲

cients are
Hence, for tⰇ ␶ tr ,

A k ⬅A U
k⫽
2
L 共 L⫹1 兲
sin
L k␲
2 L⫹1冉sin
k␲
2
cosec 冊 冉 冊 冉
1 k␲
2 L⫹1
. 冊 p n 共 t 兲 ⯝A 1 sin 冉 冊 冋冉冊 册
n␲
L⫹1
exp ⫺
␲
L
2
t
2
共A22兲
For the distribution p n (0)⫽p Sn ⫽6/ 关 L(L⫹1)(L⫹2) 兴 n(L
⫹1⫺n) 关see Eq. 共A4兲兴, the explicit solution is
⯝A 1 sin 冉 冊 冉 冊
n␲
L⫹1
exp ⫺
t
␶ res
. 共A30兲

031605-10
PROCESS OF IRREVERSIBLE . . . . I . . . . PHYSICAL REVIEW E 66, 031605 共2002兲

For the same reason, for tⰇ ␶ tr , 2 2

exp关 tln cos共 ␾ 2 兲兴 Ⰶ1 ⇒ tⰇ L ⯝ ␶ tr . 共A40兲
␲2

S 共 t 兲 ⯝A 1
2 共 L⫹1 兲
␲
exp ⫺
t
冉 冊
␶ res
. 共A31兲
Hence for times larger than ␶ tr one can write

冉 冑 冊 冉冑 冊
b. Infinite barriers
2 2
For ᐉ ES⫽⬁ (a⫽1) the allowed values of ␾ k are p n 共 t 兲 ⫽B 1 cos n cost
Lᐉ ES Lᐉ ES

冉 冊
k␲
␾ k⫽ 共 k⫽0, . . . ,L⫺1 兲 共A32兲 t
L ⯝B 1 exp ⫺ , 共A41兲
Lᐉ ES
and A k ⫽B k tan(k ␲ /2L), so that the general solution is

冉 冊 where B 1 ⯝1/L and ␶ res ⫽Lᐉ ES . This value of ␶ res , multi-

L⫺1
k␲
p n共 t 兲 ⫽ 兺
k⫽0
A k cost
L
X k共 n 兲 , 共A33兲 plied by ⌬t⫽1/2D coincides with its continuum counterpart
¯␳ /F⫽Lᐉ ES /(2D).
where

冋 冉 冊 冉 冊 冉 冊册
3. The dynamical problem in two dimensions
k␲ nk ␲ nk ␲
X k 共 n 兲 ⫽ tan sin ⫹cos . 共A34兲 It is useful to summarize here some results for a single
2L L L
particle on a two-dimensional terrace. The general solution is
The coefficients A k depend on the initial condition

冋 冉 冊
through the relation L
1 k␲
1
L p m,n 共 t 兲 ⫽ 兺 Akj 2t cos
L⫹1
兺
k, j⫽1
A k⫽ p n共 0 兲 X k共 n 兲 , 共A35兲

冉 冊册
Nk n⫽1
j␲ t
⫹cos X k共 m 兲 X j共 n 兲 , 共A42兲
where ( ␦ k0 is the Kronecker symbol兲 L⫹1

N k⫽
L
2 冋
1⫹tan2
k␲
2L 冉 冊册 共 1⫹ ␦ k0 兲 . 共A36兲 where the coefficients A k j are

p n (t) is the sum of a constant 共the term for k⫽0) and expo- L
1
nentially decaying terms (k⬎0). For p n (0)⫽1/L, the only
nonvanishing coefficient is A 0 ⫽1/L and this implies for all
Ak j⫽
N kN j 兺
m,n⫽1
p m,n 共 0 兲 X k 共 m 兲 X j 共 n 兲 . 共A43兲

times

1 For zero barriers X k (n)⫽sin关nk␲/(L⫹1)兴 and N k ⫽(L

p n共 t 兲 ⫽ . 共A37兲 ⫹1)/2. For a uniformly distributed adatom, p m,n
U
⫽1/L 2 and
L
the coefficients are
In the general case of nonconstant p n (0), the exponential
decays are negligible when exp关⫺(␲/L)2t/2兴 Ⰶ1, that is to
say tⰇ(2/␲ 2 )L 2 ⯝ ␶ tr .

c. Strong barriers
k j ⫽A k A j ⫽
AU U U
冋 2
册 冉 冊 冉 冊 冋 册
L 共 L⫹1 兲
2
sin
k␲
2
sin
j␲
2
sin
Lk ␲
2 共 L⫹1 兲

Let us consider now the case of finite but large ᐉ ES (ᐉ ES

ⰇL). The solution of Eq. 共A12兲, with a→1 and large but
⫻sin 冋 L j␲
2 共 L⫹1 兲 册 冋 册 冋 册
csc
k␲
2 共 L⫹1 兲
csc
j␲
2 共 L⫹1 兲
. 共A44兲
fixed L, yields, for the two smallest ␾ k ,

␾ 1⯝ 冑 2 共 1⫺a 兲
L
⫽ 冑 2
Lᐉ ES
, 共A38兲
S
We indicate as p m,n the normalized solution of the station-
ary diffusion equation in the presence of a constant flux.
Differently from what occurs in the one-dimensional case,
S
␲ the explicit form of p m,n is not known exactly for a square
␾ 2 ⫽ ⫹O 共 1⫺a 兲 . 共A39兲 terrace. However, the expression of its coefficients A Sk j can
L
be obtained by exploiting the property 共see Appendix A 1兲
The slowest decays in the general solution are therefore
S
that p m,n ⫽N 兺 ␶⬁⫽0 p m,n ( ␶ ) where p m,n ( ␶ ) is the solution of
exp(⫺␾21t/2) and exp(⫺␾22t/2). For finite values of L we can the diffusion equation with uniform initial condition p m,n U

neglect the second exponential for times such that ⫽1/L and N⫽1/␶ res is a normalization factor
2

031605-11
PAOLO POLITI AND CLAUDIO CASTELLANO PHYSICAL REVIEW E 66, 031605 共2002兲

冋 冉 冊 冉 冊册 冉 冊 冉 冊
⬁ ⬁ L ␶
1 k␲ j␲ mk ␲ n j␲
兺 p m,n共 ␶ 兲 ⫽ ␶兺
␶ ⫽0
兺 Akj 2␶
⫽0 k, j⫽1
cos
L⫹1
⫹cos
L⫹1
sin
L⫹1
sin
L⫹1

冉 冊 冉 冊
L
Akj mk ␲ n j␲
兺
冋 冉 冊 冉 冊册
⫽ sin sin . 共A45兲
k, j⫽1 1 k␲ j␲ L⫹1 L⫹1
1⫺ cos ⫹cos
2 L⫹1 L⫹1

Hence The sum 兺 ␶ p n ( ␶ ) has been shown in Appendix A1 to be

equal to the solution ␳ n of the stationary diffusion equation,
AU which always has a parabolic shape. Its normalized version

冋 冉 冊 冉 冊册
kj
A Sk j ⫽N . 共A46兲
1 k␲ j␲ p Sn is
1⫺ cos ⫹cos
2 L⫹1 L⫹1
␳n ␳n
p Sn ⫽ ⫽ , 共B3兲
The numerical prefactor N can be determined by imposing L
␶ res
S
that the sum over m and n of p m,n is 1, that is 兺
n⫽1
␳n

兺 兺 A Sk j sin
m,n k, j
冉 冊 冉 冊
mk ␲
L⫹1
sin
n j␲
L⫹1
⫽1, 共A47兲 so that

␶ res S
n ⫽
p eff 关共i兲 and 共ii兲兴. 共B4兲
which implies p
␶ dep n
1
冋
L 共 L⫹1 兲
册兺 2
共AU
kj兲
2

冋 冉 冊 冉 冊册
⫽ . This equation corresponds to Eq. 共14兲 in the limit ␶ res
N 2 1 k␲ j␲
k, j
1⫺ cos ⫹cos Ⰶ ␶ dep .
2 L⫹1 L⫹1 For strong and infinite barriers 关regimes 共ii兲 and 共iii兲兴, the
共A48兲 contribution of times shorter than ␶ tr is smaller than ␶ tr / ␶ dep
and therefore negligible. Accordingly, we can evaluate the
In the limit of large L,
sum 共B1兲 using the expression for p n ( ␶ ) that is valid in the

冋 册 limit ␶ Ⰷ ␶ tr 关see Eq. 共A41兲, in d⫽2 the generalization is

U 2
1 L 共 L⫹1 兲 2
共 A 11 兲 27 2

冉 冊
␶ res ⫽ ⯝ ⯝ L . trivial兴, p n ( ␶ )⫽(1/L d )exp(⫺␶/␶res) and obtain
N 2 ␲ ␲6

冋 冉 冊册
1⫺cos ⬁
L⫹1 1 1 1
共A49兲 n ⫽
p eff
L ␶ dep
d 兺 exp ⫺ ␶
␶ ⫽0
⫹
␶ dep ␶ res
. 共B5兲

Hence in the continuum ␶ res ⯝(32/␲ 6 )L 2 /D and

Converting the sum over discrete times into an integral, we
32 have
␤⯝ 6 . 共A50兲
␲
␶ res 1 ␶ res
In the limit of strong but finite barriers one finds ␶ res n ⫽
p eff d⫽ pS , 共B6兲
␶ res ⫹ ␶ dep L ␶ res ⫹ ␶ dep n
⫽Lᐉ ES /(4D), so that ␣ ⫽1/4.
where—as usual—p Sn is the normalized solution of the sta-
APPENDIX B: THE EFFECTIVE DISTRIBUTION tionary diffusion equation. Thus, formula 共14兲,
We want to evaluate the effective distribution ␶ res
n ⫽
p eff pS 共B7兲
⬁ ␶ res ⫹ ␶ dep n
n ⬅
p eff 兺
␶ ⫽0
P dep共 ␶ 兲 p n 共 ␶ 兲 共B1兲
can be used in all the different regimes.
In the continuum it is possible to work out a more rigor-
introduced in Sec. III A. Since p n ( ␶ ) decays to zero after a ous approach and determine p effn as the solution of a differ-
time of order ␶ res , for regimes 共i兲 and 共ii兲 共where ␶ res
ential equation, which is the generalization of the stationary
Ⰶ ␶ dep ) P dep( ␶ ) can be taken as a constant. Hence diffusion equation 共3兲. We start with the diffusion equation
⬁ for p(x,t), ⳵ t p⫽Dⵜ 2 p, where x is a d-dimensional vector.
1
n ⫽
p eff
␶ dep 兺
␶ ⫽0
p n共 ␶ 兲 关共i兲 and 共ii兲兴. 共B2兲 If we multiply both sides by P dep(t) and integrate in time, we
obtain

031605-12
PROCESS OF IRREVERSIBLE . . . . I . . . . PHYSICAL REVIEW E 66, 031605 共2002兲

冕 ⬁
dt P dep共 t 兲 ⳵ t p 共 x,t 兲 ⫽Dⵜ 2 冕 ⬁
dt P dep共 t 兲 p 共 x,t 兲 . 共B8兲
where I 0 and I 1 are the modified Bessel functions of order
zero and one, respectively. The arguments of the Bessel func-
tions are at most equal to L/ 冑D ␶ dep ⫽ 冑8 ␶ tr / ␶ dep , a small
0 0

The right-hand side is just Dⵜ 2 p eff(x) while the left-hand quantity. An expansion of the Bessel functions gives
side is

冕 0
⬁
dt P dep共 t 兲 ⳵ t p 共 x,t 兲 ⫽ P dep共 t 兲 p 共 x,t 兲 兩 ⬁0 p eff共 r 兲 ⫽
1 L 2 ⫹2ᐉ ESL⫺r 2
␲ L 2 4D ␶ dep ⫹L 共 L⫹2ᐉ ES兲
. 共B12兲

⫺ 冕0
⬁
dt 关 ⳵ t P dep共 t 兲兴 p 共 x,t 兲 By using the results 共A6兲 and 共A8兲, after some algebra we
obtain the final expression

⫽⫺
1
␶ dep
1
⫹
1
A ␶ dep
冕 0
⬁
dt P dep共 t 兲 p 共 x,t 兲
␶ res

冉 冊
共B9兲 p eff共 r 兲 ⫽ p S共 r 兲 , d⫽2. 共B13兲
L L ᐉ ES
␶ dep ⫹ ⫹
and Eq. 共B8兲 becomes D 4 2
1
Dⵜ 2 p eff共 x 兲 ⫺ p eff共 x 兲 ⫹F⫽0. 共B10兲
␶ dep The calculation in d⫽1 leads to the result

It differs from the stationary diffusion equation Dⵜ 2 ␳

⫹F⫽0 because of the presence of a ‘‘desorption’’ term ␶ res

冉 冊
关 ⫺p eff(x)/ ␶ dep 兴 that is the responsible of the saturation of p eff共 x 兲 ⫽ p S共 x 兲 , d⫽1. 共B14兲
L L ᐉ ES
p eff(x) at large ᐉ ES . As a matter of fact, in the limit ᐉ ES ␶ dep ⫹ ⫹
→⬁, ␳ is known to diverge as (F/D)Lᐉ ES 关see Eqs. 共A7兲 D 8 2
and 共A8兲兴 while the above equation clearly shows that
p eff(x) goes to the constant F ␶ dep ⫽1/A.
The quantity L/D(•••) appearing on the right-hand side
The exact solution of Eq. 共B10兲 can be found both in d
in the denominator does not coincide with ␶ res because the
⫽1 and in d⫽2 for a circular terrace and the proof that
term L 2 /D has a prefactor 41 instead that 81 in d⫽2 and a
p eff(x)⫽ ␶ res /( ␶ res ⫹ ␶ dep )p S (x) works much in the same
prefactor 18 instead that 121 in d⫽1. Nonetheless such quan-
way in the two cases. We give here some more details for the
tity differs from ␶ res for a quantity of order ␶ tr which can be
bidimensional case. The solution of Eq. 共B10兲 with the usual
safely neglected with respect to ␶ dep 共always appearing in
boundary condition ⳵ r p eff(r)⫽⫺p eff(r)/ᐉ ES evaluated for r
the denominator兲 so that, in the limit ␶ tr Ⰶ ␶ dep 共a limit ap-

冋 冊册
⫽L 共the radius of the circular terrace兲 is

冉 冊
plied throughout the paper兲 we can conclude that the relation
r
I0
1 冑D ␶ dep ␶ res

冉 冊 冉
p eff共 r 兲 ⫽ 1⫺ , p eff共 x 兲 ⫽ p S共 x 兲 共B15兲
␲L2 L ᐉ ES L ␶ dep ⫹ ␶ res
I0 ⫹ I1
冑D ␶ dep 冑D ␶ dep 冑D ␶ dep
共B11兲 is always valid.

关1兴 M.A. Herman and H. Sitter, Molecular Beam Epitaxy 关7兴 A. Pimpinelli, J. Villain and D.E. Wolf, Phys. Rev. Lett. 69,
共Springer, Heidelberg, 1996兲. 985 共1992兲.
关2兴 J.A. Venables, G.D.T. Spiller, and M. Hanbücken, Rep. Prog. 关8兴 P.J. Feibelman and T. Michely, Surf. Sci. 492, L723 共2001兲; M.
Phys. 47, 399 共1984兲. Kalff, P. Smilauer, G. Comsa, and T. Michely, ibid. 426, L447
关3兴 J.A. Venables, Introduction to Surface and Thin Film Pro- 共1999兲; Thomas Michely 共private communication兲.
cesses 共Cambridge University Press, Cambridge, 2000兲. 关9兴 C.-M. Zhang, M.C. Bartelt, J.-M. Wen, C.J. Jenks, J.W. Evans,
关4兴 It is easy to realize that ␻ MF should read ␻ MF⫽(1/⌬t)¯␳ 2 , and P.A. Thiel, Surf. Sci. 406, 178 共1998兲; L. Bardotti, C.R.
where ⌬t⫽1/2dD is the time an atom spends on each lattice Stoldt, C.J. Jenks, M.C. Bartelt, J.W. Evans, and P.A. Thiel,
site (d is the substrate dimension兲. Phys. Rev. B 57, 12 544 共1998兲.
关5兴 J.G. Amar, F. Family, and P.-M. Lam, Phys. Rev. B 50, 8781 关10兴 J.A. Stroscio, D.T. Pierce, and R.A. Dragoset, Phys. Rev. Lett.
共1994兲. 70, 3615 共1993兲; J.A. Stroscio and D.T. Pierce, Phys. Rev. B
关6兴 J. Villain, A. Pimpinelli and D.E. Wolf, Comments Condens. 49, 8522 共1994兲.
Matter Phys. 16, 1 共1992兲. 关11兴 J. Tersoff, A.W. Denier van der Gon, and R.M. Tromp, Phys.

031605-13
PAOLO POLITI AND CLAUDIO CASTELLANO
Rev. Lett. 72, 266 共1994兲.
关12兴 I. Elkinani and J. Villain, J. Phys. I 4, 949 共1994兲.
关13兴 G. Ehrlich and F.G. Hudda, J. Chem. Phys. 44, 1039 共1966兲;
R.L. Schwoebel and E.J. Shipsey, J. Appl. Phys. 37, 3682
共1966兲; R.L. Schwoebel, ibid. 40, 614 共1969兲; K. Kyuno and
G. Ehrlich, Surf. Sci. 383, L766 共1997兲.
关14兴 See also Appendix B in: P. Politi, G. Grenet, A. Marty, A.
Ponchet, and J. Villain, Phys. Rep. 324, 271 共2000兲.
关15兴 P. Politi and J. Villain, Phys. Rev. B 54, 5114 共1996兲.
关16兴 This has also been tested directly by changing the Zeno Model
关12,15兴 in such a way as to switch on/off nucleation on vicinal
terraces 关P. Politi and J. Villain 共unpublished兲兴.
关17兴 C. Ratsch, M.F. Gyure, S. Chen, M. Kang and D.D. Vveden-
sky, Phys. Rev. B 61, R10598 共2000兲.
关18兴 J. Rottler and P. Maass, Phys. Rev. Lett. 83, 3490 共1999兲; S.
Heinrichs, J. Rottler, and P. Maass, Phys. Rev. B 62, 8338
共2000兲.
关19兴 J. Krug, P. Politi, and T. Michely, Phys. Rev. B 61, 14037
共2000兲.
关20兴 J. Krug, Eur. Phys. J. B 18, 713 共2000兲.
关21兴 P. Politi and C. Castellano, following paper, Phys. Rev. E 66,
031606 共2002兲.
关22兴 C. Castellano and P. Politi, Phys. Rev. Lett. 87, 056102 共2001兲.
关23兴 A. Pimpinelli and J. Villain, Physics of Crystal Growth 共Cam-
bridge University Press, Cambridge, 1998兲.
关24兴 This is a very common assumption, but it may not be always
appropriate. In fact the interaction between the landing atom
and surface steps can steer the atom trajectory and create spa-
031605-14
PHYSICAL REVIEW E 66, 031605 共2002兲
tial inhomogeneities in the incoming flux of adatoms. See, S.
van Dijken, L.C. Jorritsma, and B. Poelsema, Phys. Rev. Lett.
82, 4038 共1999兲.
关25兴 In the discrete representation, if p 0 is the deposition probabil-
ity per time step we have P dep(t)⫽p 0 (1⫺p 0 ) t . The definition
of ␶ de p as the average time of deposition implies the relation
␶ de p ⫽(1⫺p 0 )/p 0 , so that P dep(t)⫽(1⫹ ␶ de p ) ⫺1 关 (1
⫺1 ␶ dep ⫺t/ ␶ dep
⫹ ␶ de p ) 兴 . If ␶ de p Ⰷ1 关i.e., ␶ de p Ⰷ1/(2dD) in the con-
tinuum picture兴 the prefactor is ␶ ⫺1 de p and the quantity in square
brackets is just e. Hence P dep(t) coincides with Eq. 共9兲.
关26兴 K. Bromann, H. Brune, H. Röder, and K. Kern, Phys. Rev.
Lett. 75, 677 共1995兲.
关27兴 We take the single walker to be distributed uniformly at t⫽0
关 p n (0)⫽p U n 兴 , but a different choice changes N dis and N all only
slightly. For this reason the relations between W are W NI and
N dis and N all are given by the symbol ⯝.
关28兴 B.D. Hughes, Random Walks and Random Environments
共Clarendon Press, Oxford, 1995兲.
关29兴 The function ␹ has the maximum and the minimum on the
border of the terrace 关see V.I. Smirnov, A Course of Higher
Mathematics 共Oxford, Pergamon Press, 1964兲, Vol. 2兴. If ␹ is
not identically equal to zero, let us suppose the maximum is
positive 共otherwise we can suppose the minimum negative兲
and apply the boundary condition ⳵⬜ ␹ ⫽ ␹ /ᐉ ES at the maxi-
mum. If ᐉ ES⫽0 it implies that ␹ increases in the inward direc-
tion and therefore it cannot be the maximum. If ᐉ ES⫽0 and we
apply the boundary condition at the maximum and the mini-
mum, we find that they both vanish.