VOLUME 51 26 SEPTEMBER 1983 NUMBER 1)

Geometrical Approach to Moving-Interface Dynamics

Richard C. Brower
Iyman Laboratory of Physics, Harvard University, Cambridge, Massachusetts 02138

and
David A. Kessler
Theoretical Division, Los A. lamos National Laboratory, I.os Alamos, New Mexico 87544

Joel Koplik and Herbert Levine

Schllrnbexge~-Doll Research, gidgefield, Connecticut 06877
(Received 7 March 1983)
A general class of models is introduced which relate the motion of a phase boundary to
properties of the local interfacial geometry. These systems can undergo successive de-
stabilizations as they grow, possibly giving rise to nonequilibrium spatial patterns. This
formalism has applications to a wide variety of physical problems, especially including
dendritic solidification.
PACS numbers: 05.70. Ln

Many problems of physical interest involve the the interfacial temperature by an amount propor-
question of how a nonlinear system dynamically tional to the curvature. This problem has been
evolves in time. This is a particularly fascinat- studied extensively by Langer and co-workers"
ing issue whenever the initial state is prepared over the past few years. There exists evidence
in a configuration which has domains that are not for the operation of a pattern-selection mecha-
absolute minima of the free energy. In this case, nism wherein the growing dendrite undergoes a
the interface motion is necessarily a nonequilib- series of successive destabilizations and subse-
rium problem which has the possibility of giving quent restabilizations by the emission of "side
rise to either chaotic final states or to intricate branches. " The associated hypothesis that this
patterns. A characteristic feature of these prob- sort of nonlinear limit cycle causes the system
lems is the competition between long-wavelength to operate at the marginal stability point seems
instabilities and a stabilization mechanism acting '
consistent with experimental results. However,
at the shortest scales. the detailed workings of this mechanism as well
A prototypical situation is the formation of as side-branch spacings and global symmetries
dendritic ice crystals from a supercooled melt. are not understood. Some other systems that ex-
The equations which govern the growth of the sol- hibit similar competition between stabilizing and
id phase are the heat-diffusion equation for which destabilizing forces are directional solidifica-
the interface motion, through emission of latent tion, ' fluid flow in a porous material, and bio-
heat, acts as a source, and the thermodynamic logical growth. '
boundary condition that surface tension reduces The purpose of this work is to introduce a sim-

1983 The American Physical Society

VOLUME 517 NUMBER 13 PH YSICAL REVIEW LETTERS 26 SEPTEMBER 1983

pie class of models in which we believe some of the generalization to three dimensions will be
the essential features of the moving-interface presented elsewhere. '
problem may be isolated and explored. The key In order to motivate a choice for E in (2), we
idea is to reduce the complicated dynamics of turn to the physics of the solidification problem.
the entire system which is responsible for pro- It is easy to argue that the growth rate will be
ducing the domain structure to an equation for enhanced by increasing the curvature of the sur-
the interface motion which is a function only of face. Schematically, a local outward bulge will
the local geometry of the interface itself. In what allow latent-heat diffusion to occur more rapidly
follows, we will show that this simplification to than a more planar interface. Moreover, an ab-
a geometrical surface dynamics still preserves solutely planar interface cannot move at all, oth-
a rich and varied phenomenology. In particular, er than for certain special values of the under-
we examine a subset of models whose behavior cooling. Thus, I' must vanish at the origin. The
bears a striking resemblance to several features undercooling leads to an asymmetry between posi-
of the full solidification problem. tive and negative curvature, suggesting a positive
Specifically, we assume equations of the gener- term quadratic in I(.". Finally, the presence of a
al form minimum nucleation size constrains Ito become
negative at large ~. This leads us to suggest that
n Bx/Bt
~
=F(a, Vg. . . ), solidification could be modeled by our geometri-
where x is the interface position, ~ is the local cal surface dynamics with F of the form
curvature, n is the outward unit normal, and V
E(K) = K + GK —pK (4)
is the covariant derivative on the surface. F is
a general function of tc and its covariant deriva- We view this resulting model as a suggestion in
tives. For simplicity, we will restrict the dis- the spirit of a I andau-Ginzburg mean-field ap-
cussion here to the two-dimensional problem proach.
where the phase boundary is just a curve. Then Our model allows for analytic treatment direct-
V is just 8/Bs, where s is the arclength. An at- ly parallelto that applied to the solidification
tractive choice for 5 is problem, though with significantly greater ease.
As there, an exact solution of the above model is
F =E(~) +y 8'~/Bs', a circle whose radius obeys
where the second term stabilizes the system at
short wavelengths for y &0.
Equation (1) has the important property of re- Because of our choice of I", large enough circles
parametrization invariance. This means that the will always expand. Consider, then, the stability
curve x can be arbitrarily parametrized without of such a solution. To leading order, we can set
affecting the physical content of the time evolu-
tion. In fact, the phase boundary is a purely geo-
R(O) =R, +g„a (t) cos(me),
metrical, stringlike entity, totally independent where the initial circle is parametrized by the
of any coordinate grid used to describe it. One polar angle 0. A standard analysis yields the in-
consequence of this is that the tangential compo- stantaneous growth rate
nent of Bx/Bt is not determined by the physics (as
contained in 5), but instead may be chosen for
convenience. An often useful choice is the or-
'
thogonal gauge, Bx(t, o)/Bt =nF, where o param- With y = 0, stability depends on the sign of I '.
etrizes the curve. This is our analog of the Mullins-Sekerka insta-
It is useful to rewrite the time-evolution equa- bility. 'Nonzero y acts as a short-distance cutoff
tion (1) in terms of the curvature and the are- on the fluctuation spectra, making the dominant
length. The arclength is related to the pa. ra, m- instabilities occur at intermediate length scales.
etrization a by ds =g'~ do, where the metric g A comparison of this expression with the results
=x' ~ x', and the prime means derivative with re- of the full solidification problem shows that y is
spect to o. Defining the curvature by K =n ' 8 x/ analogous to the surface tension.
8s', one can easily derive The full solidification problem is known to have
—(8'/Bs'+ v')S (z), a family of exact solutions in the limit of zero
ic = g = 2g~r. surface tension. ' These dendritelike structures
A full discussion of these equations as well as translate at a constant velocity related to their

1112
VOLUME 51, NUMBER 13 PHYSICAL REVIEW LETTERS 26 SEPTEMBER 1983

tip radius. Similarly, for y =0 our equations al- less, we believe it still may be useful, even for
so possess such solutions. For the case I =~, nonzero y, to consider the linear mode analysis
the profile of the solution is given by around the y = 0 solution. It is straightforward to
show that there is a critical value of y/p' above
x=p8, y =pin(cos8), (8)
which all modes become stable. This point of
where the parameter 8 is just the angle that the marginal stability may be relevant for the long-
normal vector makes with the y axis. Near the time behavior of dendrites in our model, in anal-
tip, this curve looks parabolic with arbitrary ogy with Langer's suggestions for the solidifica-
radius p. The velocity of translation along the tion problem. ' Perhaps the lack of a steady-state
y axis is given by 1/p. A stability analysis along solution provides a new mechanism to drive the
the lines of Muller-Krumbhaar and Langer' leads system to the marginally stable point.
to the eigenvalue equation One of the most useful features of our approach
—cos'8(5" + 5') =u&5 is the ease of doing computer experiments. The
full nonlocal diffusion-limited boundary motion is
for the distortion 5(8) normal to the interface. extremely difficult to treat numerically, and this
This problem can be solved exactly by transform- has been carried out to only a limited extent. "
of a kink potential. "
ing it to a Schrodinger equation in the presence
The eigenfunctions are giv-
Our models, on the other hand, immediately lend
themselves to simulations. We have done this by
en by 5 - (cos8)"'P„,'"(sin8), ~ = 4+ k', where P integrating the curvature equations (3) and then
is an associated Legendre polynomial. All modes reconstructing the curve. Our programs were
grow in time, representing the instability of the checked by varying the number of points and also
dendrite to tip splitting in the absence of a sur- by comparing the results to direct integration of
face-tension term. &q. (1). We will discuss the detailed results of
For nonzero y, the situation is very different. our simulation studies elsewhere, ' but we wish
Perturbation theory seems to predict a relatively to point out here that this class of models can in-
insignificant steady-state shape correction. How- deed give rise to interesting structures, includ-
ever, a direct integration of Eqs. (4) shows that ing dendritic growth patterns. In Fig. .1, for ex-
perturbation theory breaks down at large distanc-
"
es from the tip. This manifests itself in the
ample, the initial pattern stabilizes to a circle
as it grows, but is then destabilized by (comput-
presence af increasingly rapid oscillations in e(8) er) noise and emits irregular dendrites. Figure
as 8 approaches + 2m. In the strict mathematical 2 exhibits tip splitting and in Fig. 3 the dendrite
sense, there cannot exist uniformly translating initiates a side-branch excitation. We emphasize
solutions. This leads us to conjecture that a sim- that (a) these curves represent the initial stages
ilar breakdown occurs in the solidification prob-
lem, contrary to earlier expectations. Nonethe- " of growth and that (b) we do not claim that all ob-
served features of solidification are reproduced
by our specific choices of F and its parameters.

10-
——t = 13.00
- 10—
—.—t = 2.50
-- —t = 5.06 """-t = 17.00 t=0
~~~~
1.50 . :"". ~" """t=3418
~~

~ ~~
~ ~ ~ ~ ~ I

0— , ""-"/-"
~ ~ t~
P
Y
'. (
0- 4 I ;/
~~
~ ~~ ' \» ~ ~
/
i j x r' 5
ba
0
~

)'.
~ ~
~
~ ~~ ~
~ ~ ~
~~

~~ ~~
~o
-5— 1

(
i
& ~ ~ ~ ~

~ t
~~~ ~ ~ ~~

—10 l I I

-10 -5 0 10 -10' l I s
X -10 -5 0 10
X
FIG.1. Emergence of dendrites: solution of (4) with
a=0, P=3, m=0. 2. FIG. 2. Tip splitting; n = 1, P = 0.25, y = 0.25.

1113
VOLUME 51, NUMBER 13 PH YSI GAL RE VIEW LETTERS 26 SEPTEMBER 1983

10— convinced us that there is much to be understood

——t = 3.00
-

-t = 1.50..., ~'
,. t = 3.75 here in the way of nonlinear phenomena. While
this class of models cannot hope to reproduce all
of the intricate details inherent in a full nonlocal
~

approach (such as the lack of self-intersections),

~

;j::.
we are optimistic that a detailed study of these
('
~ ~ ~ ~,~ ~ ~

phenomena will provide much needed insight into

0— :::.
I .j::
. : the process of pattern formation.
&-~ (
~~ ~
~ ~
~ M
.-i&
':I' We would like to thank Jim Langer for gener-
ously sharing with us his insight into the physics
/
I of these systems as well as some of his unpub-
~o
~ ~ ~ s~z +go

~
~~ ~ ~ ~
lished results on models which are quite similar
(and in some cases identical) to the ones studied
~~ ~~

-10 -10 I

-5
I
here.
10

FIG. 3. Side branching; o. = 1, P = 0.25, y = 0.35.

'J. S. Larger and H. Muller-Krumbhaar, Acta Metall.

In fact, a crucial question in solidification is the 26, 1681, 1689, 1697 (1978).
importance of crystalline anisotropy, which may J. S. Langer, Rev. Mod. Phys. 52, 1 (1980), and
be required for repeated side branching. Phys. Rev. A 27, 499 (1983); Q. Dee and J. S. Langez,
The ice-crystal problem led us to consider F's Phys. Rev. Lett. 50, 383 (1983).
of the form given in (2). If we turn instead to the 3J. S. Larger, Phys. Rev. Lett. 44, 1023 (1980);
oscillations of a bubble of one liquid immersed G. Dee and R. Mathur, Institute for Theoretical Physics
Report No. NS F-ITP 82-112 (unpublished) .
in a second, immiscible liquid (e.g. , oil in water), 4R. Chandler, J. Koplik, K. Lerman, and J. F. Wil-
we must incorporate the fact that the area of the lemsen, J. Fluid Mech. 119, 249 (1982).
enclosed liquid cannot change, at least insofar 5See, for example, Mathematical Models of Molecular
as the fluid is incompressible. This will indeed and Cellula~ Biology, edited by L. A. Segel (Cambridge
be the case if we choose 5 of the form BV (tc)/&s Univ. Press, Cambridge, Ergland, 1980).
+y 8'x/Bs', where V is some function of curva- 6There is an extremely large literature on reparam-
ture. Likewise, a system whose phase boundary etrization invariance in the context of the string model.
For an introduction, see C. Rebbi, Phys. Rep. 12C, 1
moves even for planar interfaces, such as a wa-
(19743.
ter flood displacing oil in a reservoir, '4 would VR. C. Brower, D. A. Kessler, J. Koplik, and H. Le-
necessitate a term in F which would be nonzero vine, to be published.
at ~ =0. Finally, one could attempt to formulate W. W. Mullins and R. F. Sekerka, J. Appl. Phys.
more realistic models by allowing at least some 35, 444 (1964).
nonlocality in F. Stable crystal growth has been ~H. Muller-Krumbhaar and J. S. Larger, Acta Me-
studied in this manner, "
and recently a model of tall. 29, 145 (1981).
"P. M. Morse and H. Feshbaeh, Method of Theoretical
this sort for dendritic growth has been indepen-
dently proposed by Langer. " Physics {McGraw-Hill, New York, 1953), Vol. 2, Sec.
12 3.
~

In summary, our geometrical surface dynam- "We thank J. Langer for pointing out this possibility.
ics can be used as a starting point for investigat- "G. E. Nash and M. E. Qlicksman, Acta Metall. 22,
ing the question of nonequilibrium pattern forma- 1283 {1974).
tion in a variety of systems. We have demon- '3R. C. Brower, D. A. Kessler, and H. Levine, un-
strated that for solidification, these models re- published; J. B. Smith, J. Comput. Phys. 39, 112
produce the Mullins-Sekerka instability and the (1981).
'4For an introduction to this topic, see A. E. Scheideg-
tip-splitting behavior for the dendrite. We also
ger, The Physics of Elom in Porous Media (Univ. of
conjecture that the lack of a steady-state solution Toronto Press, Toronto, 1974), 3rd ed.
for nonzero y could lead to a new explanation for ' H. Muller-Krumbhaar, T. Burkhardt, and D. Kroll,
the marginal-stability-selection mechanism. Fi- J. Cryst. Growth 38, 13 (1977).
nally, some simple numerical experiments have ' J. S. Langer, private communication.

1114