HAL Id: jpa-00247233
https://hal.archives-ouvertes.fr/jpa-00247233
Submitted on 1 Jan 1996
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Decay of Grooves Cut in a Surface with Singular Orientation when the Neighbouring Orientations are
Unstable
Christophe Duport, Anna Chame, W. Mullins, Jacques Villain
To cite this version:
Christophe Duport, Anna Chame, W. Mullins, Jacques Villain. Decay of Grooves Cut in a Surface
with Singular Orientation when the Neighbouring Orientations are Unstable. Journal de Physique I,
EDP Sciences, 1996, 6 (8), pp.1095-1125. �10.1051/jp1:1996118�. �jpa-00247233�
Decay of Grooves Cut in
aSurface with Singular Orientation when the Neighbouring Orientations
areUnstable
Christophe Duport (~*),
Anna Chame(~,**)
,
W-W- Mullins
(~)
andJacques Villain (~)
(~) CEA,
Département
de Recherche Fondamentale sur la MatièreCondensée,
SPSMS, 38054 Grenoble Cedex 9, France(~)
Carnegie-Mellon University, Department
of Materials Science andEngineering, Pittsburgh,
PA 15213, USA
(Received
19February
1996,accepted
10April 1996)
PACS.68.35.Ja Surface and interface
dynamics
and vibrationsPACS.05.40.+j Fluctuation
phenomena,
random processes, and Brownian motion PACS.68.35.Bs Surface structure andtopography
Abstract. The
decay
of agrooved profile
when trie average orientation issingular
is treated in theparticular
case when the orientations close to thesingular
one are unstable. The system is assumed toexchange
atoms with its vapour. The step fluctuations, which allow theprofile decay,
are treatedby
apartially
exact transfer matrix method. The time T topeel
the topmostlayer
is obtainedas a function of the width 1of a terrace: f c~
iexp(~i),
where the constant ~depends
on the temperature.Il. Introduction
~rhe
smoothing
of groovesartific1ally
made in acrystal
surface is a classicalproblem (Blakely il ],
130nzel [2] and Preuss[3])
solvedby
Mullins [4] in the case of anon-singular
orientation. If the final(or average)
orientation of the surface issingular,
the non-linearproblem
which arises isontrovers1al.
Indeed,
it is not clear whether the facets whichtemporarily
anse inexperiments
<ire a result of a miscut as claimed
by Lançon
and Villain[Si,
or intrinsicproperties
as claimed1)y
Spohn
[6] andHager
et ai. [7]. Theproblem
is somewhatsimpler
if the orientations close la thesingular
one areunstable,
because in that case thetopmost layer
ispeeled
withoutperturbing
too much the lower ones. Thepresent
work is aninvestigation
of this situation.At
equihbrium,
theparts
withsingular
orientations form facets withsharp edges.
It isi easonable to assume that the
top
and the bottomparts
of thedecaying profile
also have facets withsharp edges (Fig. l). Indeed,
thesloping parts
of the surface are formed ofsteps
whichinteract
through
an attractive interaction(in
contrast with the usual case discussedby
Bonzel<md Preuss [2] and then
by
Rettori and Villain [8] and thenby others).
The
exchange
of matter(atom
peratom)
may be eitherby
diffusion on thesurface,
inside the solid orby exchange
with the vapour [4]. The last process, called SALK(surface
attachmentlimited
kinetics),
is muchsimpler
than the other ones. We shall restrict our attention io this(*)
Author forcorrespondence le-mail: [email protected])
(**)
Permanent address: Universidade Federal Fluminense, Instituto de Fisica, 24210-340, Niterôi RJ, BrazilQ
LesÉditions
dePhysique
1996- 1+
Fig.
1. Initial state of the profile.Fig.
2. Atypical
structure withmeandering
topmost step. The lowerlayers
are not disturbed.case. This process is
supposed
toapply
notonly
to trueevaporation/condensation
of acrystal
in contact with a vapour
phase,
but also in the case of surface diffusion on thecrystal
surface(no exchange
withvapour)
whenexchange
of the mobile speciesje.
g.adatoms)
with sources and sinksje-g-
kinks andsteps)
is so dilficult as to be ratelimiting.
Then the mobile species may be considered to have an uniform chemicalpotential
~1 as if it were a vapour
bathing
the surface. The
exchange
of matter between this mobilelayer
and the sources or sinks is assumedproportional
to the difference between~land
the local chemicalpotent1al
~1+ ô~l determinedby
surface energychanges resulting
from surfacedisplacements.
It is not clear whether thislimiting
caseactually
exists, but it has been usedextensively
as a basis for modelcalculations [9].
In a realistic model of the
profile
consideredhere,
the two topmost steps should be assumed to fluctuate between their lowest energypositions.
The other stepsmight
be assumed to bestraight. However,
thissimplified
model isfairly complicated,
and we willstudy
ayet simpler
model with asingle fluctuating step (Fig. 2).
The system will be assumed toexchange
atoms with its vapour, so that it is sulficient to consider asingle ridge.
In Sections 2 and
3,
the relaxation mechanism isanalyzed
and theproblem
is reduced to theequilibrium
statistical mechanics of asingle fluctuating
fine attractedby
a fixed fine. Thisproblem
is a dassical one and itsthermodynamical properties
will be recalled in Section 4. Itsdynamical behaviour, however,
requires aspecial analysis
which is given in Section5,
where the relaxation time is estimated. In Section 6 theprincipal
results arepresented.
Adictionary
ai
t=tb)
~_~2
Cl t=~£
1?ig. 3. Successive states of the
peehng
of trie topmostlayer: ai
at time ii ajunction
with the immobile step occurs and the two stepsloops
createdbegin
to recoil(trie
topmostlayer begin
to<~vaporate), b)
at time t2 anotherjunction
occurs andci finally
at T two oppositerecoiling loops
merge~nd the
layer
ispeeled.
>Jf the notations used is at the end of the text. The most
complicated part
of thealgebra
is,zontained in the
appendices.
InAppendix
A and B someequilibrium properties
are evaluatedasing
the transfer matrix method and inAppendix C,
thedynamic problem
is treated usingan alternative
approach:
a Fokker-Planckequation
is derived and solvedapproximately,
and thetypical
times involved are estimated.2. Characteristic Time and
Length
If the
topmost
step is in itsground
state(or nearly soi
at time t= 0 and
begins
tofluctuate,
one has to wait a
typical
time fi until it makesjunctions
with the immobilestep (Fig. 3a).
After a
junction
has takenplace,
both ends of thetopmost
terraceevaporate
and recoil at thevelocity
~, which isonly
a function of the terrace width1,
and of the fine tensionWI la
of thestep.
If one considers one of therecoiling
steploops,
it will merge(Fig. 3c)
after a time f with anotherloop recoiling
at theopposite velocity,
which has been created at time t2(Fig. 3b).
Let
Li
be the distance between thepoints
where thejunctions
at times fi and t2 have takenplace.
Let pi be theprobability
per unitlength
and time to have ajunction.
The timefi
is the time one has to wait before
observing
ajunction
in alength
of orderLi Therefore, piLiti
Ci 1. But(assuming
t2 >fil
the time t2 fi is also the time one has to watt beforeobserving
ajunction
in alength
of orderLi
Therefore piLi(t2 fi)
CÎ 1. Thus:fi CÎ t2 Ci t2 fi CÎ
j (2.1)
Pi 1 The
quantity (f fi)
iscomprised
betweenLi/~ (if
t2=
f)
andLi/(2~) (if
t2=
fi).
Therefore
f fi CÎ
~~ (2.2)
~
The left hand side f fi of this
equation
is of courselarger
than t2 fi, but cannot be muchlarger; otherwise,
this wouldimply
that theprobability
of ajunction
in alength Li during
a time muchlarger
than t2 fi isnegligible,
and this would be in contradiction with the definition of piTherefore,
f fi CÎ t2 tiComparison
with(2.1)
and(2.2) yields
Li
fi CÎ
(2.3)
Now, (2.2)
reads f cf fi +Li /~
orinserting (2.3) jnd (2.1):
f Qf
~~
Qf
(~ ~)
~
~Î~Î
Then,
it is clear that1Ci
(2 5)
fi
Thus,
the time f necessary topeel
thetopmost layer
will be known if we are able to calculate~ and pi The
velocity
~ isproportional
to the difference between theevaporation
rate a of the curvedstep
and theevaporation
rate où of thestraight step.
This difference is itselfproportional
to the local excess ô~l of the chemicalpotential.
Theprecise
formulae are~ =
a(a où)
=aaoflô~l (2.6)
where a is the interatomic distance and
fl
is1/(kBT).
The excess chemicalpotential
is theexcess free energy per
partiale.
Since aloop
of width 1andlength
~dt has an energyWi~dt la
and contains
i~dtla~ particles,
it followsô~l= Wia Ii
and(2.6)
reads~ oe
a~aofl ~ (2 7)
3. The Junction
Probability
piIi)
The
junction probability
can be evaluatedignoring
thenon-fluctuating step.
Inreality
when thefluctuating
fine touches the immobile one, the process becomes irreversible and the twoseparated
parts of the topmostlayer begin
to recoil. In our calculation we willsimply
obtainan estimate of the
typical
time after which thefluctuating
fine touches the other for the first time.Consider a
part
oflength
L of thefluctuating step.
At time t=
0,
it issupposed
to bestraight
atposition xiv, 0)
= 0 for any value of y. At time
t,
the maximum valueh(t)
ofxiv, fi
will be called theelongation.
We want to know thetypicaltime Tilt, L)
after which1
- ~
a) L
(h -'- y
~)
------'
L
x
~y -~-~---~~à-~-
i~
- -W
c) L
1?ig. 4.
Typical
excursion(of elongation h)
of the top of theprofile
on alength
Lai explicitly iepresenting
the lower steps and the nonfluctuating
step andb) ignoring
them. A excursion withelongation larger
thon1, the size of the terrace, is shown inci.
For thisconfiguration
we can definelhe distance Y
along
which all the points have a position x >1.h(t)
reaches the value1for the first time(Fig. 4a).
This time shouldsatisfy
pijtiL~ijt,Li
~ ij3.ii
Insertion of this
equation
into(2.5) gives
usy ~
@13.21
The immobile
step
has nopart
in the calculation ofTi
Therefore,
it can beignored.
Wenow consider a
single
step(Fig. 4b).
Rather than
calculating Tilt, L),
it may be easier to calculate thetypical
time ToIi) during
which the
elongation
h remainslarger
than 1 if it islarger
than 1 at a given time. It tutus out that bothquantities
are relatedby
asimple
relation which uses anequilibrium property, namely
the
probability p)(L,1)
that theelongation
at aparticular
time islarger
than aparticular
value1mjch larger
than the. thermal averageelongation. Indeed,
theperiods
in which theelongation~is larger
than 1have atypical
duration ToIi)
and areseparated by
a time interval of orderTilt, L).
ThereforePllL'i)
~~)j~j~ 13'3)
1 ,
The
quantity p) IL,1)
is not easy to calculateexactly.
In order to uselin
anapproximate
way)
the exact data which areavailable,
we introduce theequilibrium probability pj(L,
Y,h)
that,
on a fine of givenlength L,
a numberYla (and
nomore)
ofpoints
have aposition
x > h.Obviously
p)(L,1)
=
~j pj(L,Y,1) (3.4)
Y>o
On the other
hand,
consider a numberôÎ
ofreplicas
of the system(1.e.
a fine oflength L)
taken from the canonical ensemble with the canonical
probability.
The number ofpoints
of agiven
ordinate y which have aposition
z > h isôlp)(h),
wherePI (h)
=
~j p(z) (3.5)
~>h
is the
equilibrium probability
that for agiven
y,x(y)
> h. Theprobability p(h)
is knownexactly
as recalled in the next section. The total number ofpoints
of the fine which have aposition
x > h is obtainedby multiplying
the number of points of agiven
ordinate y which have aposition
x > hby
the number Lla,
and therefore it isequal
to(Lla)ôlp((h).
On theother
hand,
the number of points of a given ordinate y which have aposition
x > h can also be written asN
£
~'pj (L, Y, h) (3.6)
y~o a
Equating
this toIL la)ôlp) (hi,
one obtains£ Ypj(L, Y, hi
=
Lp) (3.7)
Y>o
This formula may be
exploited
in anapproximate
way if Y isreplaced by
its average value1:
P £ pj(L, Y, hi
=
Lp) (3.8)
Y>o Insertion into
(3.4)
nowyields
p)(L, hi
=
~~~~~~
(3 9)
Y
Now, using (3.2), (3.3)
and(3.9)
we obtain the basic formula for thepeehng
time:T ~
fiÎ 13.i°)
where ~ is
given by (2.7)
aspointed
out before andp) ii)
isexactly
known. Its value will berecalled in Section 4.
Before
estimating
Y and ToIi)
we define moreprecisely
our model and recall some well-knownequilibrium properties.
4.
Equilibrium Properties
of aFluctuating Step
In the
previous sections,
theproblem
has been reduced to the statistical mechanics of a fine which can fluctuate around a fixed fine which attracts it. As will be seen, an"unbinding"
transition can occur for this
system:
the two fines cari be bound(localized solution)
if the tem-perature
issulficiently
low or, for a fixed temperature, if the attractivepotential
issulficiently strong.
Thefirit
exacttheory
ofbinding
of two watts in d= 2 was done
by
Abraham[loi iising
the transfer matrix method. Chui and Weeks [11] treated the sameproblem
under asolid-on-solid condition. A different
derivation,
with an extension to d= 3 was
proposed by
Vallade andLajzerowicz [12].
The
problem
treated inAppendix
A is thestudy
of the fluctuations of astep
which has astiffness
+~
(defined by (5.1) below)
and is boundby
an attractivepotential -Wo
to astraight
fine.
We will assume
throughout
the paper that the distance between kinksalong
thestep,
which isof the order of +~a~
/kBT,
islarge
withrespect
to a, that is to say +~a »kBT.
Thisassumption
,ùlows one to use
equation (A.3c),
thatis,
to consideronly
transitions of onestep
away orloward the
binding
fine.In
Appendix A, by using
the transfer matrixmethod,
it is shown that thedensity
ofproba- lJility p(x)
to find apoint
of the step at a distance x of the borderline is, for x# 0,
Plx)
=iaoi~exPl-21~x) 14.1)
where ~ is a real number.
Since the
integral
of theprobability
isunity,
the constant in(4.1)
isao[~
oe 2~(4.2)
The
probability
to find apoint
of thestep
at a distancelarger
than x of the borderline isjjiven by integrating (4.1):
p)(x)
ceexp(-2~x) (4.3)
The width of the
topmost layer
is assumed to belarge
withrespect
to the size of the thermal fluctuations which is of the orderof1/~,
then i~ » 1.Moreover,
the continuum limitimplies
that ~a < 1.Thus,
the relative order of thelength
scales is 1»1/~
» aIt
might
beargued
that theassumption
~a < 1 is veryspec1al,
since it will be validonly
on a very small range of
temperatures.
Thisassumption imphes
that a fluctuation oflarge elongation
would berelatively
easy. If alarge
value ofWo (say Wi/2)
ischosen,
which would bephysically reasonable,
it could make ~acomparable
tounity
and would restrict allowedfluctuations to a few times a.
However, experimentally,
for various reasonsje-
g.feasibility),
we are interestedby
macro- or at leastmesoscopié profiles.
Theprofile
width 1 then must be muchlarger
than a.Therefore,
theproduct
~a should besmall,
otherwise theproduct
~i will behuge
and the relaxation time too(as
will be seenlater).
On the otherhand,
since ~changes rapidly
withT,
thetemperature
range where ~i is small is very narrow. In this way the
assumption
1 »1/~
» a isreally physical.
In this
connection,
inFigure
5 aplot
of ~ versusWo
ispresented (obtained by solving Eq.
(A.25) ),
in which ~ goes to zero asWo approaches
thelimiting
valuegiven by equation (A.26b) (as
anequality).
Since1/~
is the order of the averageelongation,
thisplot gives
the reader afeel for the
dilficulty
of afluctuating step
to cross the top terrace.The transfer matrix method demonstrates the existence of a transition at a certain tempera- ture
Tc
at which the constant ~ vanishes. Above thistemperature,
thestep prefers
to increase,
/
' /
/
/
,
0.8 /
/
' ,
/ /
/
~'~ Î /
/
~Î t'
~ / ~~Î
~~0.4 ' / ~~ ~'~
/
/
fiw =1.5i
/
fiwi =20°'~
' /
fiw
=2.5,
/
fiw = 3.0
0
0 0.2 0A 0.6 0.8
)W
Fig.
5.~ as function of
Wo,
thebinding potential.
The order of the averageelongation
of an excursion is1/~.
The solidcurve
(flwi
=
o-à)
may be unreliable becauseexp(-flwi)
would be toobig.
its
entropy
rather than itspotential
energy, and thus is not bound to the terraceedge.
This transition is calledunbinding
transition ordelocalisation
transition. When T >Tc,
there areno forbidden orientations
in the
vicinity
of the average surfaceorientation,
and this is the case discussed in references [5] to[8].
In the presentwork,
the condition T <Tc
is assumed. Thenin the continuum
hmit,
whichcorresponds
to ~a <1,
we obtain the relations(see
inAppendix A, Eqs. (A.23)
and(A.25),
which define ~, andEq. (A.46))
Wo
<kBT (4.4a)
~~~~'~~~
~ÎÎ~Î~ ~~'~~~
The transition
temperature
isgiven by
the condition ~= 0 and therefore
kBTc
"@~ (4.4c)
5. Estimate of
1
Dur next task is to calculate Y and To. In this section an estimate for
1
is obtainedusing
asimple
and intuitiveargument
and inAppendix
B it is evaluatedthrough
a detailed calculation.The evaluation ofTo will also be
performed
in two ways: in Section 6through
asimple argument
and inAppendix
C.A
simple
way to estimate Y isthrough
theevaluation,
for twolimiting
cases, of the averagequadratic
difference(ôx~ )y
between the horizontalpositions
of twopoints
of the hneseparated
by
a distance y.If y <
1,
then(ôx~)y
must beapproximately
the same as in the absence of an attractivepotential, namely
jôx2j~
=j jà.ii
On the other
hand,
if y»1,
the attractivepotential
must beconsidered,
and(ôx~)y
=/ ~ Î~ lx x')~pja;nt(y;
x,x')dxdx' (5.2)
where pja;nt
(vi
x,x')
is thejoint probability
thatx(0)
= x and
xiv)
=
x'
simultaneously.
It canlJe written as the sum of an uncorrelated
probability
and acorrection, namely
Pjo>nul, X>
X')
"AIX)piX')
+ôpi§,
X,X') 15.31
]Ify »1,
the correlation term isnegligible
andm m
~~~~~~
Î
~~
~'~~PixiPix')dxdx~ j~,~~
which can be written as
(ôx~)y
= 2/~ x~p(x)dx
2/~ p(x)dx)
~(5.5)
o o
Using (4.1)
and(4.2),
thisyields
~~~~~~
12
~~'~~V
can then be defined as the crossover value between the behaviours(5.1)
and(5.6),
so thatY
=
~( (5.7)
2~
A more elaborated evaluation
of1
is done in detail inAppendix
B and the final result is ,£most the same as theprevious
one:~ Y
=
0.45~) (5.8)
The above result is consistent with the
assumption1
»$
in the hmit~a <
1,
that is to iay, this result is vahd when1
is muchgreater
than the vertical distance betweenkinks,
andthis occurs for
large
values of theelongation.
Since1
is definedas
(see Eq. (3.9)
forinstance)
the averagelength along
which thepoints
of the fine have
positions
x > h for an excursion withelongation larger
than h(where
h is muchlarger
than the thermal averageelongation),
the fact that1
does notdepend
on h issurprising.
Ingeneral1
woulddepend
onh,
however the result(5.8)
shows that1
does notdepend
on h if theelongation
issulficiently large.
6. The
Typical
Time TOThe last unknown
quantity
is TOIi).
Asimple
way to estimate thisquantity
is to consider the recoil of anexceptional
excursion ofamplitude
h.Let a be the
evaporation probability
from thefluctuating
curve per site and persecond,
and letrp
be theadsorption probability
from the vapour when this vapour has adensity
P " Po
exp(fl~l) (6.1)
The
quantity ~lis
the vapour chemicalpotential
counted from anappropriate
origin. Atequilibrium, adsorption
and emissionmutually compensate
anda =
rpo expjp~) j6.2)
In
reality,
an excursion oflarge amplitude
h is not atequilibrium
with the vapour which hasa fixed
density
po, so that theadsorption
rate per site has a fixed valueoù "
rpo (6.3)
On the
contrary,
the emission rate per site has a valuegiven by (6.2)
as a function of the chemicalpotential ~1(h)
that the vapour would have if it were inequilibrium
with the excursion ofamplitude
h:a = où
exp(fl~l) (6.4)
This value will be assumed to be uniform on the whole excursion.
In this way, the
velocity
of recoil of the excursion isgiven by
(
"
°o(1
eXPlfl~))a (6.5)
Assuming
thatfl~l
<1,
one can write ôh1 °°~~~
~~'~~The chemical
potent1al
~1 can be written as
(+~a~)/R,
where R is the radius of curvature ofthe excursion. The
quantity
which appear in thisexpression
for ~1 is the hne stiffness+~ rather than the fine tension
Wi la
as in(2.7)
because thisexpression
for~1 is valid for weak
slope
fluctuations while(2.7)
is related to a macroscopicloop.
To estimate R we recall a result fromAppendix A,
where thetypical
extension of a fluctuation ofelongation
h is obtained. One finds that~~~~ àÎ~ ~~'~~
Assuming
the excursion is an arc of ciEcle orparabola
of baselength
andamplitude h,
one finds R=
(fl~+~~h) /~~ multiplied by
a constant of orderunity.
Insertion into(6.6) yields
ôh
ao~~a~
j 16p~h (6.8)
The fact that h > 1 at t
= 0
imphes
that thetypical elongation [
that exceeds 1lies between 1 and 1+) Therefore,
thetypical
time To necessary for[
to become smaller than is1/(~ () ).
Replacing () by
its expression(6.8),
one obtains thatT0(h)
t~j~
~
j6_gj
ao~ a
7. Results
In this
section,
theassumptions
and theprincipal
results of thepreceding
sections arerecalled,
in order to obtain a final
expression
for the time f topeel
thetopmost layer.
We assumed that the
step
which fluctuates is bound to thehmiting
borderline of thetopmost layer,
that is to say, thestep
is below theunbinding
transition: the wavevector of the boundstate is ~.
The
topmost layer
is éissumed to belarge
withrespect
to the size of the thermal fluctuations which is of the orderof1/~,
then i~ » 1.Moreover,
the continuum limitimplies
that ~a < 1.In this way the relative order of the
length
scales is1»1/~
» a.Finally,
we havesupposed
that thetypical length
between kinksalong
astep,
fl+~a~, islarger
than the interatomic distance a and that we are near the
unbinding
transition in order thatthe
large
excursions are not tooexceptional.
In this way +~a »kBT
»Wo.
The
quantity
we want to calculate is thepeeling
time of thetopmost layer,
which isgiven by (3.10).
It is sulficient to insert into that formula theexpressions (2.7), (4.3), (5.7)
and(6.9)
to obtain the final result
~ ~
210a2
fi~~~~'~~~
~~'~~where ~ is defined
by (4.4b).
All these results refer to the case when the
topmost
terrace is limitedby
afluctuating
step and arigid
one. If it is limitedby
twofluctuating
steps, a formulaanalogous
to(3.10)
shouldapply, namely
" ~
i~ ~~~~
where
p)o(h)
is theprobability that,
for agiven
y, the distance between bothtopmost steps
is
greater
than h. Thisprobability
isgiven by integrating
theprobability pro(h)dh that,
for agiven
y, the distance between bothtopmost steps
is between h and h + dh. NowPoolhl
"
/ Plh qlPlqldq (7.3j
or
using (4.1)
and(4.2):
pro(h)
=
4~~hexp(-2~h) (7A)
Integration
nowyields
m cc
p(o(1)
=pro(h)dh
=4~~
hexp(-2~h)dh
=(2~i
+1) exp(-2~i) (7.5)
Equations (2.7), (5.7)
and(6.9),
whichrespectively yield
~,i
and To, are stillexpected
to hold.Inserting
these relations and(7.5)
into(7.2),
one obtains thepeeling
time of alayer
in the case of twofluctuating steps
asy~ r-
2~t3aoa2 fiexpj~ti) ah 17.6)
Acknowledgments
This work was initiated in Jülich
(Germany).
W-W- Mullinsacknowledges
thehospitality
of the Institute für Grenzflàchen undVakuumphysik
andsupport
from the von HumboldtSitftung,
and J. Villain the
hospitality
of theHochleistungrechenzentrum.
A. Chameacknowledges
agrant
from the Brazilian National Council for the Scientific andTechnologica1Development (CNPq)
and thehospitality
of the Centre d'Etudes Nucléaires de Grenoble. We thank Prof.Lajzerowicz
for fruitful discussions.Appreciation
isexpressed
topreparation
ofFigure
5.Notation
Dictionary
. a: interatomic distance.
. kB~ Boltzmann constant.
. i: terrace width.
. L:
length
of anarbitrary part
of thesystem.
.
Lit
distance between thepoints
where thejunctions
at times fi and t2 have takenplace.
.
L(A):
number ofadsorption
sites.. N: number of lattice sites in the x direction.
. pli
junction probability
per unitlength
and unit time..
pj(L,Y, h): equilibrium probability that,
on a fine ofgiven length L,
a numberY/b
ofpoints
have aposition
x > h..
p(h): equilibrium probability
that for agiven
y,x(y)
=
h, independent
of y.. fil time one has to watt before
observing
ajunction
in alength
of orderLi
.
p((h): equilibrium probability
that for agiven
y,xiv)
> h..
p((L,1): equilibrium probability
that theelongation
at aparticular
time islarger
thana
particular
value 1 muchlarger
than the thermal average value of x..
p(y;
x,x') joint probability
thatx(0)
= x and
xiv)
= x'.
.
p(~~ ~(y,
x,x'):
conditionalprobability
thatxiv)
= x' if
x(0)
= x.
.
pro(h): probability that,
for agiven
y, the distance between bothtopmost steps
is h..
p)o ii) probability that,
for a given y, the distance between bothtopmost steps
isgreater
than 1..
PL (h, t): probability
that theelongation
at time t of a finesegment
oflength
L is h.. T:
temperature.
. ~: recoil
velocity
of thetopmost
terrace..
Woi
attractivepotential
due to thebinding
line(Wo
>0).
.
Wi step
free energy per atom.Wi la
is also the line tension at lowtemperature (Wi
>0).
.
i(h):
average value of thelength Y, along
which the points havepositions
x > h for anexcursion of
elongation larger
thanh,
where h is muchlarger
than the thermal averageelongation.
~
p,
1kBT'
. ~t: line stifIness.
. r:
adsorption probability
from the vapour per site and per second when this vapour hasa
density
p= 1.
. e:
exp(-flwi).
. < defined
by equation (A.25).
.
À(x):
extension of a fluctuation ofelongation
x; vertical distance after which the line has lost the memory and come back to its thermalposition.
. /~; vapour chemical
potential.
. pot vapour
deisity (fixed).
. a:
evaporation probability
from thefluctuating
curve per site and per second.. où
adsorption probability
from the vapour per site and per second.. f: time necessary to
peel
thetopmost layer
when one of the topmoststeps
is unable to fluctuate.. f21 time necessary to
peel
the topmostlayer,
when bothtopmost steps
can fluctuate..
Ti(1, L):
time interval between theperiods
in which theelongation
of aparticular step segment
oflength
L islarger
than 1.. To
Ii): typical
timeduring
which theelongation
remainslarger
than1 if it islarger
than 1 at agiven
time.Appendix
AUnbinding
llkansition in 1+1 Dimensions We consider(Fig. 3a)
a fixedstraight
wallix
=
0)
and a hne which is at theposition xiv)
> 0.The line has a stifIness which maintains it
straight
at thetemperature
T =0,
and it is attractedby
the fixed wall due to apotential V(x)
< 0 for x= 0.
It is easier to suppose that x and y are discrete
variables,
xla
and yla being integers,
where a is the interatomic distance. It is also easier to useperiodic boundary
conditions. The energy isjN-1)a jN-1)a
U((X), (VI)
#
~j Vl(X(§))
+~j V2(X(iÎ)> X(iÎ
+~)) (~.~)
y=0 y=0
where
Aix)
=
-woôo,~ (A.2)
with
Wo
> 0 the attractivepotential
and V2corresponds
to the stifIness. We can writeV~(x,x)
= 0
(A.3a)
V2(x,
x +a)
=V2(x
+ a,x)
=
Wi
> 0(A.3b)
and for the other
values,
V2(x, x')
= oo
(A.3c)
Equation (A.3c)
means thatx(y) x(y -1)
canonly
have the values0,
a or -a, otherwise the Boltzmann factor isequal
to zero.A.I. THE PARTITION FUNCTION. The
partition
function isZ =
£ exp[-flvi (xo
Ilexp[- flV2 (xo,
xi
Il exp[- flvi (xi
Il~o,~i, ,xN-i
x
exp[-
flV2(xi,
x2Ilexp[- flV2 (xN-i,
xoIl(AA)
where xp =
x(pa).
It is useful to introduce the notation(x[8[x')
= exp
-~Vi xii
exp[-fl%(x,x')]
exp(-~Vi(x')j (A.5)
2 2
Now
(AA)
can be writtenZ "
L iXoi8iXiiiXii8iX2i. IXPIÔIXP+ii. IXN-iiÔiXoi iA.61
The
quantity (x[8 lx')
can be considered as an element of a real Hermitian matrix 8. For themoment we do not
give
the vector1almeaning
to the notationxi.
We will do itlater,
but nowthis notation is
only
a part of thequantity (xl
8lx'),
which we can call8xx,. Equation (A.6)
is written
Z = Tr
8~ IA.?)
The trace is easy to calculate if the
eigenvalues 0k
are known. Then Z =£ 0f (A.8)
k
The
interesting
case is the limit N= oo. In this case,
by supposing
that there is a gap between thegreatest eigenvalue
00 and the otherseigenvalues, equation (A.8)
is writtenZ =
0/ (A.9)
This
expression
is notuniversal,
because there is notalways
a gap, but isquite general.
The calculation of theeigenvalues, instead,
isspecific
of eachproblem.
A.2. EIGENVALUES. An
eigenvalue
0 is defined from theequation
jjxjejx'iijx'i
=
oijxi jA,ici
where we have omitted the indice k.
From
equations IA-à)
and(A.3c), equation (A.10)
reduces tolx181x)41x)
+lx181x
+a)41x
+a)
+lxie(z a)j(z ai
=
ojjz) jA.ii)
For x >
2a,
fromequations (A.5), (A.2)
and(A.3),
one can writtenequation (A.Il)
as#ix)
+e#ix
+a)
+e#ix a)
=0#ix) iA.12)
where
f = exp
i-flwi) iA.13)
Equation (A.12) obviously
has thefollowing
solution for x > a~k(X)
" akeXp(-kX) (A.14)
with the
eigenvalue
0k
" 1 +e[exp(ka)
+exp(-ka)] (A.15)
Since the matrix 8 is
Hermitian,
itseigenvalues
must be real and then there are threepossibilities:
1) k isreal, ii)
k ispurely imaginary, iii)
theimaginary
part of k isequal
to 17r.The third
possibility yields
aneigenvalue (A. là)
which is smaller than one and can not be thelargest eigenvalue.
We will see that k can be or not a real value(and only one).
To understand theimportance
of thisfact,
let us calculate the distribution ofpoints
of thefluctuating
line.A.3. DISTRIBUTION OF THE FLUCTUATING LINE. The
probability p(x)
thatx(y)
has avalue x is
independent
of y and isgiven by pjx)
=j expj- pvi ix)j expj-pV2 ix,
xi )iexpj-pvi (xi )i expj-pV21xN-1, x)j
~
xi,...,x~-i
where Z is
given by equation (AA). Transforming
thisequation
in the same way as we had transformedequation (AA),
we obtain thefollowing expression,
similar toequation (A.6):
Plx)
=j L lx181xi)lxi181x2). lxp181xp+1) lxN-i181x) lA.16)
To obtain a
compact
relation similar toequation IA.?)
it is suitable to introduce the vectors[xi
and theoperators
X defined fromjx'jxi
=
ôxx, jA.17ai
and
(x'[X[x)
=
ôxx,x (A.17b)
This is a little abstract. For a
physicist
familiar withquantum physics,
it is useful to considerthe vectors
[xi
asrepresenting
the states of apartiale
that movesalong
the x axissatisfying
the
equation
ofSchrôdinger IA.10).
Fromequation (A.16)
we obtainPlx)
=jlx18~ix) lA.18)
To utilize this
equation,
we have to introduce theeigenvectors [k)
of 8.Substituting
8 =
£ jk)0k(kl lA.191
k
in
equation (A.18),
we obtainPlx)
=j L0ilxlk)lklx) lA.20a)
If the condition of
validity
ofequation (A.9)
issatisfied, equation (A.20a)
can be writtenPlx)
=j0tlx10)101x)
=lx10)101x) lA.20b)
where we have used
equation (A.9)
to write the secondequation
and we have introduced theeigenvector loi
whichcorresponds
to thegreatest eigenvalue 00.
Thiseigenvector
is givenby equation (A.14),
where k has aparticular
value ~, that is to say,loi
=fl lx1401xl
=
aoDl0l
+ aoL [xl exPl-~xl lA.211
x x#o
where D is a constant.
Inserting equation (A.21)
intoequation (A.20b), using equation (A.17b),
and
supposing
~ to bereal,
weobtain,
forx diflerent from
0, equation (4.1).
If ~ ispositive,
the
fluctuating
fine is localized. In thisarticle,
this condition is assumed to be fulfilled.If ail the k were
imaginary,
thegeneral
term ofequation (A.20a)
isindependent
of x. Thefluctuating
fine is delocalized.A.4. SEARCH FOR A BOUND STATE. A bound state
corresponds
to the solution(A.14)
with real k.
Equation (A.14)
must also be solution ofequation (A.Il)
for x= 0 and x
= a,
that is to say,
A~#o10)
+eA#oia)
=
0#o10) jA.22a)
#oia)
+e#o12a)
+eA#o10)
=
0#oia) iA.22b)
where
A =
exp(flwo/2) (A.23)
By replacing
intoequation (A.22) #o(0), #o(a),
and#o(2a)
calculated fromequation (A.21),
we find
A~D
+ CAexp(-~a)
= 0D
(A.24a)
AD
= 1
(A.24b)
Inserting
the value for 0given
inequation (A.15),
we obtain thefollowing equation
of second order inexp(~a):
A~
+eA~ exp(-~a)
= +
e[exp(-~a)
+exp(~a)]
or
f(e~~) ~2
" e~~~ e~~
(A~ -1)
= 0
(A.25)
e
The solutions of this
equation
inexp(~a)
arealways
real and withopposite
sign. Thepositive
solution is theonly
one whichcorresponds
to a real value of k. Itcorresponds
to a localized solution if ~ ispositive,
that is to say ifexp(~a)
isgreater
than1,
that is to say iff(1)
< 0.Then
f(1)
=1- 1+f~ (A~ -1)
< 0or
A~ > 1 +
~ (A.26a)
or
+ 2
exp(-flwi
(A.26b)
~~P~~~°~
~l +
exp(-flwi
We observe that the two lines are bounded if the
temperature
issulficiently
low or for a fixedtemperature,
if the attractivepotential
issulficiently strong.
A.5. JOINT PROBABILITY. The
probability
thatx(y')
= x and
simultaneously x(y'+y)
= x'
is
p(y;
x,x')
=
£ ~j exp[-flvi lx)] exp[-flV2(x,
xi)1exp[-flvi (xi
~ )1
xi, ,xyla-i xyla+i, .,xN-i
~~~~~~Î~~~ÎÎ~~~~~~fI~~Î~~~~~~~ ~~~~"~~~~~~~~~~~~ ~~~~Y/~+~)Î
-,
or
piy;
x,x~i
=
j L ixieixiiixiieiz~i iz~/~-iieiz~i
xi,..,x~/~-i
£ l~'lÔl~yla+11" l~N-llôlX)
~yla+>;. ~N-1
or
Plv;x,x')
=
jlx18~/~lx')lx'l8~~Y/~lx)
For y much smaller than the size Na of the
system,
thisexpression
can be writtenplu; x,x')
=
fl ())
~~~lxlk)lklx'llx'101101xl lA.27al
or
fl Yla
pjy.
xx')
=
~j
~(x[k)(k[x')[ao[~ exp[-~(x
+x')] (A.27b)
~
flo~
An useful check is obtained for y
=
0, using equations (A.27a), (A.20b)
and the relation£~ [k)(k[
= 1:p(0;
x,x')
=
£(x[k)(k[x')(x'[0)(0[x)
k
=
(xix')(x'ΰ)(°Îx)
"ô~~'P(x) (A.28)
It is convenient to
distinguish
between the localized stateloi
and the unbound state[q).
Equation (A.27)
is writtenP(Yi
X,X')
"P(X)P(x')
+ôp(y;
x,x') (A.29)
with
ôPlv; x,x'l
=
L ())
~~~lxlqllqlx'llx'101101xl lA.30)
q#o °
The wave function of the unbounded state is written
[q)
=Dqoq[0)
+ oq£ sin(qx
+~q)
lx) (A.31)
~#0
where
Dq
and +~q are solutions of thesystem A~Dq
+ CAsin(qa
+~q) =
0Dq (A.32a)
sin(qa
+~q) + e
sin(2qa
+~q) +
eADq
= 0sin(qa
+~q)
(A.32b)
or,
replacing
0by
itsexpression IA-là),
A~Dq
+ CAsin(qa
+~q)=
il
+ 2ecos(qa)]Dq (A.33a)
ADq
= sin+~q(A.33b)
Eliminating Dq,
we obtainA~
+eA~[cos(qa) sin(qa)
cot+~q] = 1+
2ecos(qa) (A.34)
or
~2
1 +f
COS(qa)(À~ ~)
~~
~~~~~t +~~ =
~~2
sin(qa)
We see that
+~q is
proportional
to q if q is small.The other parameters are determined:
Dq,
fromequation (A.33b);
oq, from the normaliza- tion condition. Inshort,
for an infinite system in the xdirection,
ail values of q are allowed.However,
it may bepreferable
to assume alarge
but finite size L=
Na,
so that q takes discrete valueswhich,
asusual,
are theinteger multiples
of 27rIL. Then,
the normalisation condition takes the formÎ~QÎ~
l~
~~~~~~~ '~Q) ~~~2'~~j ~~'~~~)
~
~
which,
forlarge L,
reduces tolaql~
Cf) lA.36b)
By using (A.30)
and(A.31)
we obtaino Yla
ôPlY> x,
x')
=
lac
l~£ laq
l~
(() smjqz
~f~)
sinjqz'
~f~)expj-~jz
+z')j
~
o Yla
=
lac
l~£ laq
l~ ~
coslqjz z')i expl-~jz
+z')j
~
o
~ Yla
~l£X01~
£1£XQl~ (() C°Sl~lX
+
X')
2'fQl~~Pl~~lX
+ X')1lA.37él)
~
or,
substituting
the expression(A.15)
for 0ôp(y;
x,x')
=
[oo
Î~~j Îaq
[~
~ ~~
°~~(~~
~~~cos[q(x x')] exp[-~(x
+x')]
A37b)
~
l + 2e cosh
~a)
~
_~ ~ +
z'
j~~
j2j jaq
Î~i~/Î~ÎÎÎÎ~)
~~ ~~~~~~~ ~~'~
~'~~~~~~~
~It is
pretty hard,
from thisformula,
to recover(A.28)
for y= 0. As a matter of
fact,
formula(A.37b)
is not very convenient for small y.However,
if y islarge enough,
thequantity
Il +2ecos(qa) ~f~
+ 2e