Professional Documents
Culture Documents
Geometrical Interface: Models Evolution
Geometrical Interface: Models Evolution
Richard C. Brower
Lyman Laboratory of Physics, Haruard Uniuersity, Cambridge, Massachusetts 02138
David A. Kessler
Los Alamos National Laboratory, Los Alamos, New Mexico 87544
patterns evolve. —/
The outline of the remainder of this paper is as follows. g ~
Bs
In Sec. II, we describe the kinematics of geometrical evo-
lution (sometimes referred to as contour dynamics ). In and its derivative (which must be parallel to the unit nor-
Sec. III, we specialize to a class of local models which are mal n) defines the curvature
motivated by the solidification problem. We present some
simple solutions and consider their (linear) stability. Fi- K= —Pl ' 8 x
$2
nally, Sec. IV is devoted to the dendritic steady-state solu-
tion. An Appendix contains some computational details. In two dimensions it is convenient to describe the curve by
the complex position z=x+iy. In this notation, the
tangent and normal vectors are written as
II. KINEMATICS FOR GEOMETRICAL
EVOLUTION
~=z'=age'@'I n = '
ie'@—
where 8(s) is the angle between the tangent and the x axis.
Imagine a system evolving dynamically in d dimensions The curvature is just the rate at which 8 changes,
with a d —1-dimensional interface separating two K=d8/ds. Then differentiating the evolution equation
domains. The interfacial surface can be specified parame- z=nU, we have
trically by the position vector x (t, o ) where
tr, 1 &a &d —1 is a set of parameters and t is the time. (n U)
To find the surface at a later time we need to know the Bo'
projection of the velocity in the normal direction whose real and imaginary parts yield
where we have used d/dt =8/Bt ~, +sB/Bs. As in most As in the two-dimensional case, the curvature tensor
gauge problems, a gauge-invariant formulation introduces must satisfy some constraints in order to correspond to a
nonlocality. closed surface, but these are of two forms. First, there are
%'e can extend these considerations to the general case integrability conditions for the set of partial-differential
of a surface evolving in d dimensions, using standard re- equations giving h p in terms of derivatives of x. These
sults of tensor calculus. If a =(a', . . . , o" ') is an ar- conditions, the Gauss-Codazzi equations, are
bitrary set of coordinates on the surface, we have a set of —I —I
tangent vectors a~=Ox/Bo. which from (3) obey the evo- Qyh~p- —Bph~y p bye y hp5 (19)
lution equation and when they hold, the surface can be found locally.
Again, one can show that these equations are satisfied at
all times if they hold at t =0. Second, there are global
conditions analogous to (9) which require the resulting
On the surface, there is a metric tensor surface to be closed. The analog of (9a) is the Gauss-
Bonnet theorem involving the surface integral of the
gap +a +p Gaussian curvature, but in place of (9b) one h'as an infinite
" set of equations, one for each closed curve on the surface.
and a curvature tensor, or "second fundamental form, Let us return to the single closed curve in two dimen-
defined analogously to (S) by sions for which
hp ——nB~p. z= e ds +zo
Using the standard convention of raising indices with g p, As t — +co, the arclength will go to infinity in many
the inverse of the metric tensor, we can alternately write growth models. To test for the possibility that the inter-
(13a) as face becomes a fractal with structure on an increasing
8 n=h p~p. range of length scales, we define a series of exponents
(13b)
which characterize the long-time behavior of the curve.
Combining (10), (11), and (13) the metric evolves accord- First, we write
ing to
. 2h .
g p —— pU (14)
for the length of the curve. The area enclosed by the
The time evolution of the normal can be found from ma- curve is readily shown to be
nipulation of the identity n r =0 and (11) to be I. L
s s--' 'e Sg(g
I ) —Eg(g)
e
dn 2i
dt +il)) ') —i8t
)l (21)
and from (12) we have 4i
interfacial surface geometry. For now, we will assume ro- of time-dependent radius i(t)=V(llr). For a=P=O,
tational invariance (we will return to this point later) and this just gives r(t)=[2t+. r (0)]'~. To see the meaning
so we must construct U from the curvature and its covari- of y, we perform a stability analysis around this solution.
ant derivatives. Restricting ourselves to two dimensions, It suffices to consider perturbations normal to the original
we assume curve, since any tangential perturbation can be eliminated
by a suitable gauge transformation (2). (In the Appendix,
= V(z)+y 8 K
in a different example, we show explicitly that tangential
U
S2
perturbations drop out of the linear stability analysis. ) We
First we discuss how one chooses the "potential" V and then have
then return to the importance of y. V must embody much
x(t, 80) =r[r(t)+e(t, 8o)], (26)
of the macroscopic features of the particular system under
investigation. Again, Iet us turn to solidification. A stan- where 80 is the polar angle of the original circle. We
dard calculation" shows that the growth rate of a spheri- derive
cal crystal (circular in two dimensions) behaves as
V-1/R for large R. For Eq. (1) to incorporate this I ——
1
--d
2+1 e
behavior, V-a for small a. A planar interface fails to dHo
move at all, except at extremely special values of the un-
dercooling, so V should not contain a constant term. 1 d d
+1
Next, a very small or highly curved piece of solid will r d80 d80
contract due to surface tension or, in other words, there is ~m'
a minimum bubble size for nucleation. We include this Using the eigenfunctions e~ (t, 80) =me cosm 80, the
effect via a term of the form — Pa. , where P '~ is pro- growth rate is
portional to the minimum bubble size. Finally, in the soli-
dificatio process there is an asymmetry between freezing
and remelting of the dynamic interface. This arises sim-
ply because the heat flow is outward from the phase boun-
dary to the heat sink maintaining the liquid at the under- The m =1 mode corresponds to a uniform translation of
cooled temperature. This then suggests the addition of a the circle and is neutrally stable. To see this, note that for
final piece +a~, where u is some measure of the under- ~(8) = ~cos8„
cooling. We arrive then at the evolution equation x
cos8o+ e cos 2 Oo
3' —sin0o+ e cosos»t9o
dx— = 8 K r r
n V(~)+y (24a)
dt Bs and if we define 8O 80+ sin80, then
via 80 ——
V(v) = a + ax —Pa (24b) —
"=cos6, +., ~ =»ne,
,
r r
where we have chosen the unit of length to flx the coeffi-
cient of the linear term in V. Equations similar to (16) which is clearly a circle shifted by e with polar angle Oo.
have been studied in the context of spiral crystal growth This trivial point serves to remind us that reparametriza-
by Muller-Krumbhaar, Burkhardt, and KroH, ' and a tion invariance is very important when dealing with an ar-
mathematical study of curvature-driven growth has been bitrary, extrinsic parametrization such as Oo. This will be
given by Brakke.
' important later in the discussion of the dendrite solution.
Our growth Eq. (24) is reminiscent of a time-dependent The salient feature of Eq. (27) is that the first term is
Landau-Ginsburg model of critical phenomena. Thinking destabilizing for all m if V & 0. Since V-i~ for sufficient-
of the curvature as an order parameter that distinguishes ly large r, eventually this will always occur. This effect is
between regular growth (constant a) and irregular or pat- directly analogous to the Mullins-Sekerka instability" in
terned growth (nonconstant x. ), one might hypothesize the the usual solidification equations. Its origin in the physi-
evolution equation cal system is that an outward kink on an otherwise flat in-
terface will radiate heat more effectively and therefore
will grow faster than its surroundings. The second term is
always a stabilizing influence. Depending on the form of
with a free-energy functional V and the value of R, it will act to limit the range of m
values which are unstable. In this regard, the coefficient
y plays the role of surface tension in the full-growth equa-
E[a]= J ds 2 Bs
+ w(a) (25b) tions.
Based on the form of Eq. (24), one might attempt to
This approach yields an equation similar (but not identi- find stationary solutions of the form d x/dt = 0, which re-
cal) to (24). From this point of view, the terin in y gUll CS
represents the usual interaction of neighboring points on
the curve, and 8' is a potential to be fixed by the physical y-dk = —V(a. ) .
situation. Bs
Equation (16) admits solutions corresponding to a circle This equation can be thought of as a particle of mass y
GEOMETRICAL MODELS GF INTERFACE EVOLUTION
moving in a potential mell 8'= V s. Then a solution common with the aforementioned dendrites. These con-
would correspond to possible periodic orbits in 8'. Recall figurations are solutions of the evolution equation at y =0
that there are three equations of constraint that must be which, as we have already argued, is analogous to taking
obeyed by the curvature in order for it to represent a the limit of zero surface tension. Initially, our analysis
closed curve. The first is that the integral of the curvature will parallel the standard treatment —
we will carry out the
equals 2m. This can be satisfied by choosing the "energy'* linear stability analysis both excluding and including the y
of the particle such that I
ads over a single "primitive"
orbit equal to 2~/n, and by having the closed curve con-
term and also compute the perturbative shape correction
due to nonzero y. Afterwards, we will argue that for any
sist of n traverses of the periodic orbit. It is then easy to nonzero y, the family of steady-state solutions ceases to
show that since exist. We will then comment on how this last fact causes
us to reinterpret the marginal stability hypothesis both for
~
i Hd ~ ~
2mik/n
orbit
i Hd
9
our local system and for the full solidification equations.
Assume the curve is translating uniformly in the y
direction. If we use the angle made by normal with the y
the other constraints will be automatically obeyed for all axis as a (time-dependent) curve parametrization (24a)
n&1. As we will discuss in Sec. IV, there also exist reduces to
steady-state solutions wherein the curve maintains its
shape as a function of time but nevertheless translates in dx
n —
=U cosO= V(v) .
space.
Finally, let us comment on simple extensions of our ab-
This equation can be inverted to find a(8) and the latter
solutely local growth dynamics. %'e can imagine remov-
integrated to find x(8). Solutions exist for all velocities.
ing the restriction regarding rotational isotropy, and allow
For the simple example a=P=O in (24), x =cosO. This
some dependence on either 8= f
a ds, or removing
translation invariance by allowing dependence on the an-
can be integrated to give the curve (in the frame where the
tip at 8=0 is instantaneously at the origin)
gle of the curve position measured from some special
point (a physical origin). These extensions might be neces-
x =pO, y =p ln cosO (30)
sary to model effects such as crystalline anisotropy (in,
say, the surface tension) or growth on an underlying sub- with p= 1/U. Near the tip, this is approximately parabolic
strate. In fact, one of the major goals of the study of with radius p. In this two-dimensional solution, y goes to
these simple systems is to pinpoint how important these infinity at a finite value of x. It is amusing to note that
ostensibly small effects are in controlling the long-time the analogous system in three dimensions, V(v, ~g ) =~
development of an interface. Another possible extension gives rise to a dendrite which asymptotically approaches a
is the addition of a forcing term inta Eq. (24) which can paraboloid of revolution. This is because such a surface
be used to represent the interaction of an external, decou- has a incan curvature proportional to (cosO+cos 8) which
pled field with the interface motion. If we then couple for large distances (8-m/2) is approximately cosO, there-
this field back to the interface itself by adding an equation by solving Eq. (29). One can also ask the question in re-
for its motion too, we arrive at a sort of "boundary layer" verse, i.e., which potential V will give rise to an exact par-
model. This type of system has recently been introduced abola in two dimensions. A simple calculation shows that
by Ben-Jacob et al. to study dendritic crystal growth. the answer is V(a)-~'~ .
Let us turn to the stability analysis, working with the
IV. DENDRITIC GROWTH a=P=O solution for analytical ease. We use the extrinsic
(i.e., non-gauge-invariant) parametrization Oo, where Oo is
The solution of the full, nonlocal solidification equa- =
just 8 of the unperturbed solution. Then, x x o
tions which appear to have the most relevance for the ac-
tual time development of the system is the dendrite crys-
+ no5(Oo, t) implies (see Appendix for details)
tal. ' As discussed in great detail by Langer and Muller- 5"
Krumbhaar, ' there exists a mode selection problem for K=Kp — + gp
2 6 —Kp5 ~
gp 2gp
dendritic crystal growth. In the absence of surface ten-
sion, there exists a complete family of dendritic solutions,
with the velocity determined by the radius of the parabolic 1/2
tip. Experimentally, one observes a unique velocity' go=(&o +3'o ) =psecOo (31b)
which may be related to the point of marginal stability of x. =u cosOo — cos
U Oo(5" +5)+U cosOosinOo5' . (32)
the linear fluctuations around the dendrite. The latter is
the value of the velocity below which the steady-state The perturbation of the left-hand side of the evolution
growth of the dendrite becomes linearly unstable. A selec- equation is just
tion mechanism of this type has been analyzed in detail in
some one-dimensional toy equations. ' However, in the
n. dx—=5 i e, +5'Oo .
full theory, marginal stability remains only a hypothesis
consistent with the data and has not yet been convincingly
demonstrated. p=Uxp =U cospsinp
In this section, we will consider the family of steady-
state configurations of our local models that have much in [which follows from (29)] we find the final equation
1 340 BROKER, KESSLER, KGPLIK, AND LEVINE 29
The solutions of (33) are where p 2 —p 3 is thedegenerate root. The marginal stabili-
ty point is then defined by finding the value of y such
5( OQ, r ) = + cosO+ '&
&z ( sin OQ )e "r, that Reco = 0. A simple calculation shows that this condi-
(34)
co = ( —,' + k )U
tion along with the first part of (38) leads to
~~4
60—
6
' -7 v cosO=K(8)+y
8 K(8)
$2
~& =p —7p —Ep (37) This can then be integrated to give the change in the arc-
length and curve position,
(where co& —co&/U, y =yu ) for the "sidebranch" oscilla-
tions on the side of the dendrite. x (8) =pO — sin(28),
A fourth-order equation of exactly this form was re- 2p (42)
' Their
cently studied by Langer and Muller-Krumbhaar.
analysis proceeds by rewriting the fourth-order relation as y (8) =p ln cosO 1+ [cos(28) —1] .
P 2p
4.
We can express this in terms of the extrinsic parameter
Pp' p" +ip —
—~, = Q (p —p ) by recalling that it is fixed by allowing deformations
00
to
occur normal to the original surface. Specifically,
A numerical analysis of (37) shows that the maximum x(OQ) =pOQ+ y sinOQO'(OQ),
value of Reco occurs when two roots p; coincide. This
leads, in our case, to the condition y (OQ) =p ln COSOQ+ y cosOQo (OQ),
29 GEOMETRICAL MODELS OF INTERFACE EVOLUTION 1341
where o(80) is the shift caused by y to the steady-state shape correction. These calculations are directly analo-
solution. Assuming 8=80+yp(Oo), we derive the two gous to standard results in the full solidification problem.
equations If these features are enough to give stable dendrite growth
at a selected velocity, our model should also exhibit this
pp(80) —1 sin280 —sin80cr(80), growth when simulated on the computer. If not, there
2p (43) may be additional features, crystalline anisotropy for ex-
—P SI8p
p(80)+ —ln cos80 — [cos(280 —1)] ample, which can be added to the local model, to then en-
1 1
cosep P 2p sure dendritic growth. Finally, it may be the case that
some feature of dendrites really depends on the global na-
= cos80cr(80) . ture of the self-coupling of the interface via the heat dif-
These equations then determine the new shape, as given by fusion field. This work is in progress and we hope to
o(80). If one wishes, one can then try to take o.(80) into know the answer soon.
account in the previous stability analysis. Since its effect We also showed that the steady-state solutions actually
is only to change the quantitative result for y', we have cease to exist at finite y. This would a priori seem to be a
not done this. real distinction between our system and the true solidifica-
The above perturbative treatment seems to show that a tion equations. However, there is no evidence that these
full family of solutions continues to exist at finite y. solutions do exist in the full solidification system also.
However, this turns out to be illusory and nonperturba- Claims to this effect in the literature ' fail to take into ac-
tively and the solution breaks down. The simplest way to count the existence of sidebranching modes which in fact
see this is to use the definition are the most important excitation in the system. The
whole question of existence may be irrelevant, anyway,
K since the steady-state solutions act only as a starting point
E=
Bs for deriving the behavior of the modes of (29). If this is
to write the steady-state equation as the case though, we need to reinterpret the marginal sta-
bility hypothesis of Langer and Muller-Krumbhaar'5 to
e
(cosO —
1 refer to stability around the asymptotic y=O solution.
K) . (44)
Bs This may affect the way in which the system approaches
If we supplement this by the relation a=88/Bs, the the unique operating velocity. These issues will be dis-
steady-state equation can be thought of as a nonlinear cussed in detail elsewhere.
dynamical system autonomous in the arclength s. The
points @=a=0, 8=+m/2 are critical points of this sys- ACKNOWLEDGMENTS
tem. Near 8= — m. /2, linear excursions lead to the stabili-
We would like to thank J. Langer for communicating to
ty matrix [in the variables (n /2+O, ~, e)] us several results prior to publication, and we ac-
knowledge useful discussions with E. Ben-Jacob and N.
Goldenfeld.
APPENDIX
We wish to derive the equation for small deformations
of a surface translating at constant velocity, and in partic-
with eigenvalues co satisfying
ular show that the equation involves only the component
2 1 of deformation along the normal. The unperturbed sur-
N CO+ face is given as xo(t, o (t)) in a time-dependent parametri-
y y
zation, has normal n and tangents w =Bxo/Bo, and sat-
which for small y approach +1, —1/2+i /v y. Two tra- isfies
jectories spiral into the fixed point and only one emerges.
Near 8=+m. /2, the exact opposite occurs two trajec- — dxp
=nU(v, ),
tories leave the fixed point and only one is attracted. dt
Since a dendrite is a solution which must go from the
8= — m/2 to 8=+ n. /2 fixed point, it will only exist if the
where the a, are the principal curvatures of the surface.
Consider the general displacement
single trajectory emerging is the exact same one that is at-
tracted at the other side. For arbitrary y, there is no x = xp+ pl6+ v~E' (A2)
reason for this to occur and a numerical integration shows
that this match fails to occur by an amount proportional The normal component of the time derivative of x is
— const/y
term 0
(e ra), this makes no contribution to (A3) to lead- Now we equate this to the perturbed U
ing order because of (Al). If the unperturbed curve aU(~. )
translates with velocity vp then Bxp/Bt=vp, and from U(tc, +5tc, )=U(tc, )+ 8Kb
(8 tcse +bstcs) (A7)
(Al)
dx where hycb is the change in curvature linear in y. Com-
0= r . = ra (vp+ rpo ~) parison of (A6) and (A7) yields the desired result
aU
or 5=
Ka
(S~.—H t'a. 5a~. ) (A8)
n = U(a., )+5 —(H ~BQ —d') —Bate, . It is easy to verify that for the case considered in the text,
BK~ this set of equations reduces to (31) and (33).
For a review, see J. S. Langer, Rev. Mod. Phys. 52, 1 (1980). (1964).
For an introduction to this topic see A. E. Scheidigger, The H. Muller-Krumbhaar, T. W. Burkhardt, and D. Kroll, J.
Physics of Flow in Porous Media (University of Toronto Press, Cryst. Growth 38, 13 (1977).
Toronto, 1974). i3K. Brakke, The Motion of a Surface by its Mean Curvature
G. I. Sivashinsky, Acta Astronantica 4, 1177 (1977). (Princeton University Press, Princeton N. J., 1978).
4G. F. Oster, in Proceedings of the Los Alamos Conference on ~4E. Ben-Jacob, N. Goldenfeld, J. S. Langer, and G. Schon,
Fronts, Interfaces and Patterns [Physica D (to be published}]. Phys. Rev. Lett. 51, 1930 (1983).
5E. Overman and N. Zabusky, Phys. Rev. Lett. 45, 1693 (1980); ~5U. Nakaya, Snow Crystals (Harvard University Press, Cam-
G. Trygvasson and H. Aref, Numerical Studies of an Inter- bridge, Mass. , 1954).
face in a Hele-Shaw Cell (unpublished). J. S. Langer and H. Muller-Krumbhaar, Acta Metall. 26,
J. Koplik, D. Wilkinson, and J. F. Willemsen, Percolation and 1681 (1978); 26, 1689 (1978); 26, 1697 (1978).
Capillary Fluid Displacement, Schlumberger-Doll Research S. C. Huang and M. E. Glicksman, Acta. Metall. 29, 701
Center, Ridgefield, unpublished. (1981); 29, 717 (1981).
7P. G. Saffman and G. I. Taylor, Proc. R. Soc. London, Ser. A 8G. Dee and J. S. Langer, Phys. Rev. Lett. 50, 383 (1983).
245, 312 (1958); R. L. Chuoke, P. van Meurs, and C. van der J. S. Langer and H. Muller Krumbhaar, Phys. Rev. A 27, 499
Poel, Petrol. Trans. AIME 216, 188 (1959). (1983).
J. B. Smith, J. Comput. Phys. 39, 112 (1981); R. Brower, D. This possibility was first suggested by J. S. Langer (private
Kessler, and H. Levine (unpublished). communication).
9L. P. Eisenhart, An Introduction to Differential Geometry G. E. Nash and M. E. Glicksman, Acta Metall. 22, 1283
(Princeton University Press, Princeton, N. J., 1947). (1974).
R. C. Brower, D. A. Kessler, J. Koplik, and H. Levine, Phys. 22R. G. Brower, D. Kessler, J. Koplik, and H. Levine (unpub-
Rev. Lett. 51, 1111 (1983). lished).
W. W. Mullins and R. F. Sekerka, J. Appl. Phys. 35, 444