cond-mat0501521/sel2.tex
1: \documentclass[twocolumn,prl,floatfix]{revtex4}
2: %\documentclass[showpacs,prl,floatfix,preprint]{revtex4}
3: 
4: \usepackage{color}
5: \usepackage{epsfig}
6: 
7: \begin{document}
8: 
9: \title{Velocity selection problem for combined motion 
10: of melting and solidification fronts}
11: \author{Efim A. Brener and D. E. Temkin}
12: \affiliation{Institut f\"ur Festk\"orperforschung, Forschungszentrum J\"ulich,
13: D-52425 J\"ulich, Germany} 
14: 
15: \begin{abstract}
16: We discuss a free boundary problem for two moving solid-liquid 
17: interfaces that strongly interact 
18: via the diffusion field in the liquid layer between them. This problem arises 
19: in the context of liquid film migration (LFM) during the partial melting of solid alloys.
20: In the LFM mechanism the system chooses 
21: a  more efficient kinetic path which is controlled by  diffusion in the liquid film, 
22: whereas the  process with only one melting front would be controlled
23:  by the very slow diffusion in the mother solid phase.
24: The relatively weak coherency strain energy 
25: is the effective driving force for LFM.
26: As in the classical dendritic growth problems, also in this case  
27: an exact family of steady-state solutions  
28: with two parabolic fronts and an arbitrary velocity exists  
29: if capillary effects are neglected
30:  \cite{temkin2005}.  
31: We develop a velocity selection theory for this problem,
32: including  anisotropic surface tension effects.
33: The strong diffusion interaction and coherency strain 
34: effects in the solid near the melting
35: front  lead to substantial changes compared to classical dendritic growth. 
36: \end{abstract}
37: 
38: 
39: \maketitle
40: 
41: The early observation of liquid film migration (LFM) were made
42: during sintering in the presence  of liquid phase \cite{yoon1} 
43: or during partial melting of alloys \cite{musch1}  
44: (see  \cite{yoon2} for a review).  Nowadays LFM
45: is a well established phenomenon of great practical importance.  
46: In LFM one  crystal is melted 
47: and another one is solidified.  Both solid-liquid interfaces 
48: move  together with the same velocity. In the investigated 
49: alloys systems the migration velocity is of the order of $10^{-6}-10^{-5}$ 
50: cm/s and it is controlled  by the solute diffusion through 
51: a thin liquid layer between the two interfaces \cite{4}. 
52: The migration 
53: velocity is much smaller than the characteristic velocity of 
54: atomic kinetics at the interfaces. Therefore,  both solids should at the 
55: interfaces  be  locally in thermodynamic equilibrium with the  
56: liquid phase. On the other hand,  these local equilibrium 
57: states should be different for the two interfaces 
58: to provide the driving force for the process. 
59:  It is by now well accepted  (see, for example, \cite{yoon2,4}) 
60: that the difference of the equilibrium states at the melting and solidification fronts 
61: is due to the coherency strain energy,  
62: important only at the melting front because of the sharp concentration profile  
63: ahead the moving melting front (diffusion in the solid phase is very slow and the 
64: corresponding diffusion length is very small). 
65: Thus, the liquid composition
66: at the melting front, which depends on the coherency strain energy 
67: and on the curvature of the front, differs from the liquid composition 
68: at the unstressed and curved solidification front. This leads to the 
69: necessary gradient of the concentration across the liquid film and the process is 
70: controlled by the diffusion in the film.
71: 
72: If only the melting front existed,  
73: the melting process would be controlled by the very slow diffusion 
74: in the mother solid phase 
75: and  elastic effects would be irrelevant. In the LFM mechanism the system chooses 
76: a  more efficient kinetic path  which 
77:  is controlled by the much faster diffusion in the liquid film. However, 
78: in this case 
79: the relatively weak coherency strain energy is  involved as  effective driving 
80: force for this process. In this respect the LFM mechanism is similar to  other 
81: well-known phenomena such as diffusion induced grain boundary migration and 
82: cellular precipitation \cite{yoon2}. In these processes a relatively fast diffusion 
83: along the grain boundaries controls the kinetics and the coherency strain energy  
84: plays also a controlling role.
85: 
86: Thus, a theoretical description of LFM requires the solution
87: of a free boundary problem for two combined moving solid-liquid 
88: interfaces with a liquid film in between. In Ref.\cite{temkin2005}  
89: this problem was considered for simplified boundary conditions: 
90: the temperature and the chemical composition along each interface 
91: were kept constant. Their values are different for the melting and 
92: solidification fronts and differ from those far from the migrating 
93: liquid film. This means that any capillary, kinetic and crystallographic 
94: effects at the interfaces were neglected. It was found that under these simplified 
95: boundary conditions two co-focal parabolic fronts can move together 
96: with the same velocity. The situation is rather similar to a steady-state  motion 
97: of one parabolic solidification front into a supercooled melt \cite{ivan1,ivan2} 
98: or one parabolic melting front into a superheated solid. 
99: In this approximation the Peclet numbers were found, but the steady-state velocity 
100: remained undetermined at that stage. Thus, the problem of velocity selection arises.
101: 
102: Solvability theory has been very successful in predicting certain properties of 
103:  pattern selecting in dendritic growth and a number of related phenomena 
104: (see, for example, \cite{saito96,kessler88,brener91}).
105: In the two-dimensional dendritic case, the basic approach is as follows. 
106: One attempts to model the dendritic tip by a needle crystal, that is, a shape-preserving 
107: steady-state growth shape which is a solution of the equation of motion governing 
108: diffusion in the neighborhood of a solidification front. This needle crystal 
109: is assumed to be close in shape to the parabolic Ivantsov solution. If anisotropic 
110: capillary effects are included,  a single dynamically stable solution is found 
111: for any external growth conditions. This theory has been extended to the 
112: three-dimensional case \cite{ben93,brener93}. We note that capillarity is 
113: a singular perturbation and the anisotropy of the surface energy is a prerequisite 
114: for the existence of the solution. 
115: 
116: The main purpose of this Letter is to develop a velocity selection theory for LFM, 
117: including in the consideration anisotropic surface tension effects.
118: We note that this is not a just 
119: routine extension of the existing  theory because the diffusion interaction between 
120: two interfaces changes the problem substantially. 
121: 
122: \begin{figure}
123: \begin{center}
124: \epsfig{file=fig1.eps, width=6cm}
125: \caption{Schematic presentation of two  moving nearly parabolic fronts;
126: $S1$ and $S2$ are the melting and growing solids, and $L$ is the liquid film.}
127: \end{center}
128: \end{figure}
129: 
130: We discuss  the two-dimensional problem of the steady-state motion of a 
131: thin liquid film during the process of 
132: isothermal melting of a binary alloy, see Fig.1. 
133: We assume that the diffusion in 
134: the solid phases is very slow and the concentration $c$ in the liquid film obeys the 
135: Laplace equation. We introduce the normalized concentration $C=(c-c_L)/(c_L-c_S)$
136: with $c_L$ and $c_S$ being the  liquidus and solidus 
137: concentrations of the equilibrium phase diagram at a given temperature.
138: Then the equilibrium concentration and the mass balance conditions
139: at the solidification front read
140: \begin{equation}
141: C=d_2K_2, \qquad  V_n=-D\partial C/\partial \bf{n}.
142: \end{equation}
143: At the melting front the equilibrium concentration is changed 
144: by the presence of the elastic coherency strain energy \cite{yoon2} 
145: and  also the diffusional flux changes 
146:  because in the solid ahead of the melting front 
147: the concentration is $c_0$ which is different from $c_S$:
148: \begin{equation}
149: C=-b\Delta^2-d_1K_1, \qquad  V_n(1-\Delta)=-D\partial C/\partial \bf{n}.
150: \end{equation}
151: Here $V_n$ is the normal velocity; 
152: $D$ is the diffusion coefficient in the liquid film; $K$ is the curvature assumed 
153: to be negative for the interfaces in Fig.1; $\Delta=(c_0-c_S)/(c_L-c_S)$ is the 
154: dimensionless driving force; $b=Y\Omega(da/dc)^2/a^2f_L''$ 
155: is the dimensionless constant which describes the 
156: coherency strain energy \cite{4}, $\Omega$ is the atomic volume, 
157: $Y$ is the bulk elastic modulus,  $a$ is the atomic constant,
158: $f_L(c)$ is the free energy of the liquid phase per atom,
159: $f_L''$ is the second derivative of $f_L(c)$ at $c=c_L$; 
160: $d_i$ are the anisotropic chemical capillary lengths:
161: \begin{equation}
162:  d_i(\theta)=d_0[1-\alpha\cos 4(\theta-\theta_i)],\quad 
163: d_0=\gamma\Omega / f_L''(c_L-c_S)^2
164: \label{anis}
165: \end{equation}
166: with the isotropic part of the surface energy $\gamma$ and 
167: the anisotropy parameter $\alpha<<1$; 
168: $\theta$ is the angle between the normal to the interface $\bf{n}$
169: and the direction of motion,
170: $\theta_i$ is the direction of the minimum of $d_i(\theta)$ for each of the 
171: interfaces.
172: 
173: We measure all lengths in the units of the radius of curvature of the Ivantsov parabolic
174:  solidification front, $R_2$. Introducing  a parabolic coordinates system,
175: \begin{equation}
176: y=(\eta^2-\beta^2)/2, \quad \quad x=\eta\beta,
177: \end{equation} 
178: we look for a solution of the Laplace equation for the concentration 
179: field $C(\eta,\beta)$
180: in the 
181: liquid phase with the following boundary conditions: 
182: $$C(\eta=\eta_2(\beta),\beta)=d_2(\beta)K_2(\beta),$$
183: \begin{equation}
184: C(\eta=\eta_1(\beta),\beta)=-b\Delta^2-d_1(\beta)K_1(\beta).
185: \end{equation}
186: The mass balance conditions at the interfaces are:
187: $$
188: 2P(\eta_2(\beta)+\beta\eta'_2(\beta))=
189: -(\partial C/\partial\eta-\eta'_2\partial C/\partial\beta)
190: %|_{\eta=\eta_2(\beta)},
191: $$
192: \begin{equation}
193: 2P(1-\Delta)(\eta_1(\beta)+\beta\eta'_1(\beta))
194: =-(\partial C/\partial\eta-\eta'_1\partial C/\partial\beta)
195: %|_{\eta=\eta_2(\beta)},
196: \end{equation}
197: where the Peclet number is $P=VR_2/2D$ and $V$ is the steady-state velocity in the 
198: $y$-direction. 
199:  We present the two moving interfaces in the form,
200: \begin{equation}
201: \eta_2=1+\epsilon_2(\beta),
202: \quad \quad
203:  \eta_1=\eta_0+\epsilon_1(\beta)
204: \end{equation}
205: and the concentration field in the form,
206: \begin{equation}
207: C(\eta,\beta)=C_0(\eta)+u(\eta,\beta),
208: \end{equation}
209: where $C_0(\eta)=-b\Delta^2(\eta-1)/(\eta_0-1)$ is the solution without 
210: surface tension ($u=\epsilon_1=\epsilon_2$=0) which satisfies 
211: the boundary conditions: $C_0(1)=0$ and $C_0(\eta_0)=-b\Delta^2$. The balance equations
212: at the interfaces,
213: $2P=2P(1-\Delta)\eta_0=-\partial C_0/\partial\eta$ lead to the relations for the 
214: Peclet number and $\eta_0$: 
215: \begin{equation}
216:  P=\frac{VR_2}{2D}=\frac{b\Delta^2}{2(\eta_0-1)},
217: \quad  \eta_0=\sqrt{\frac{R_1}{R_2}}=\frac{1}{1-\Delta}.
218: \label{peclet}
219: \end{equation}
220:  These relations can also be obtained in 
221: the proper limit of Eqs.(27)-(28) of Ref. \cite{temkin2005} and give the expressions 
222:  for the radii of curvature for the two interfaces, $R_1$ and $R_2$, for a given 
223: velocity $V$ which remains undetermined at this stage.
224:  
225: The surface tension plays a  
226: crucial role in the velocity selection problem. 
227: In order to find the small corrections to these 
228: solutions due to the small (but singular) 
229: surface tension effects we  should first  find the 
230: small correction $u$ to the diffusional 
231: field in the linear approximation with respect to $\epsilon$ and $dK$ (see, for example,
232: \cite{brener91}). In this approximation, the field $u$ satisfies the Laplace equation 
233: and  the boundary conditions:
234: $$
235: u(\eta=1,\beta)=\Psi_2(\beta)=2P\epsilon_2+d_2K_2,   
236: $$
237: \begin{equation}
238: u(\eta=\eta_0,\beta)=\Psi_1(\beta)=2P\epsilon_1-d_1K_1.
239: \label{bounde}
240: \end{equation}
241:  The balance equations read:
242: $$
243: 2P(\epsilon_2+\beta\epsilon'_2)=-\partial u/\partial\eta|_{\eta=1}
244: $$
245: \begin{equation}
246:  2P(1-\Delta)(\epsilon_1+\beta\epsilon'_1)=-\partial u/\partial\eta|_{\eta=\eta_0}.
247: \label{fluxe}
248: \end{equation}
249: The Laplace equation with the boundary conditions, Eqs.(\ref{bounde}), is easily 
250: solved by  Fourier transform, with respect to the variable $\beta$, which we define as
251: $$
252: \overline{g(\lambda)}=
253: \frac{1}{\sqrt{2\pi}}\int_{-\infty}^{\infty}d\beta\exp(-i\lambda\beta)g(\beta)
254: $$
255: $$
256: g(\beta)=\frac{1}{\sqrt{2\pi}}\int_{-\infty}^{\infty}
257: d\lambda\exp(i\lambda\beta)\overline{g(\lambda)}.
258: $$
259: Then,  
260: \begin{equation}
261: \overline{u(\eta,\lambda)}=
262: \frac{\overline{\Psi_1}
263: \sinh [\lambda(\eta-1)]}{\sinh [\lambda(\eta_0-1)]}-
264: \frac{\overline{\Psi_2}
265: \sinh [\lambda(\eta-\eta_0)]}{\sinh [\lambda(\eta_0-1)]}.
266: \end{equation}
267: Finally, Eqs.(\ref{fluxe}) read
268: $$
269: 2P\frac{d\overline{\epsilon_2}}{d\lambda}=
270: \frac{\overline{\Psi_1}}{\sinh[\lambda(\eta_0-1)]}-
271: \frac{\overline{\Psi_2}\cosh[\lambda(\eta_0-1)]}{\sinh[\lambda(\eta_0-1)]}$$
272: \begin{equation}
273: 2P(1-\Delta)\frac{d\overline{\epsilon_1}}{d\lambda}=
274: \frac{\overline{\Psi_1}\cosh[\lambda(\eta_0-1)]}{\sinh[\lambda(\eta_0-1)]}
275: -\frac{\overline{\Psi_2}}{\sinh[\lambda(\eta_0-1)]}.
276: \end{equation}
277: Eliminating $\overline{\epsilon_1}$ from the first of these two equations,
278: $$
279: 2P\overline{\epsilon_1}= 2P(\overline{\epsilon_2})'\sinh[\lambda(\eta_0-1)]
280: +2P\overline{\epsilon_2}\cosh[\lambda(\eta_0-1)]+$$
281: \begin{equation}
282: \overline{d_1K_1}+
283: \overline{d_2K_2}\cosh[\lambda(\eta_0-1)],
284: \end{equation}
285: we find from the second one,
286: $$2P(\overline{\epsilon_2})''-2P \overline{\epsilon_2}=\overline{d_2K_2} $$
287: \begin{equation}
288: -\left[\frac{(\overline{d_1K_1})'}
289: {\sinh[\lambda(\eta_0-1)]}
290: +(\overline{d_2K_2})'\frac{\cosh[\lambda(\eta_0-1)]}
291: {\sinh[\lambda(\eta_0-1)]}\right ]
292: \label{fur2}
293: \end{equation}
294: Here $'$ denotes derivatives with respect to $\lambda$.
295: 
296: These equations are convenient for the further analysis. 
297: We are interested in the motion of the thin film, which means that $(\eta_0-1)\approx 
298: \Delta << 1$. In this case we can expand the hyperbolic functions 
299: for small values of their argument almost everywhere. This leads to $\epsilon_1\approx
300: \epsilon_2$, $K_1\approx K_2$ and then, from Eq.(\ref{fur2}) we find  in  
301: direct space:
302:  \begin{equation}  
303:  \frac{d}{d\beta}\left[(1+\beta^2)\epsilon_2\right]=\frac{d_1+d_2}{2P\Delta}\beta K_2
304: \label{local} 
305: \end{equation}   
306: A few remarks are in order.
307: First of all, this is somewhat the expected result 
308: because for the thin films the resulting 
309: equation should be local and, in principal, can be derived in the framework of boundary 
310: layer techniques. However,  the used approximation for small 
311: arguments of hyperbolic functions breaks down in the small vicinity 
312: (of the order of $\Delta$) near the 
313: singular points in the complex plane, $\beta=\pm i$. We will discuss this point 
314: later. 
315: Finally,  we can return to  Cartesian coordinates because the 
316: expression for the curvature is much simpler in this representation. 
317: In the linear approximation $(1+\beta^2)\epsilon_2\approx \zeta_2(x)$ where $\zeta_2(x)$ 
318: is the correction to the parabolic solution in the Cartesian coordinates.
319: Then Eq.(\ref{local}) reads:
320:   \begin{equation}  
321:  \frac{d\zeta_2}{dx}=\frac{d_1+d_2}{2P\Delta}x K_2
322: \label{localc} 
323: \end{equation}
324: One finds the first regular corrections to the parabolic 
325: shape  by replacing $d_1=d_2=d_0$ and 
326: $K_2=-1/[R_2(1+x^2)^{3/2}]$ for the parabola:
327: \begin{equation}
328: \zeta_2(x)= \sigma/[\Delta (1+x^2)^{1/2}], \quad \Delta<<1,
329: \label{reg}
330: \end{equation}
331: where $\sigma=d_0/(PR_2)$ is the usual stability parameter which appears 
332: in such  problems.
333: 
334: In the vicinity of the  point $x=-i$ where the curvature becomes singular and anisotropy,
335: Eq.(\ref{anis}),
336: becomes  important we rescale the variables \cite{brener91}:
337: \begin{equation}
338: x=-i(1-\sqrt{\alpha}z), \quad \zeta_2=\alpha F,  \quad \tau=z+dF/dz,
339: \label{resc}
340: \end{equation} 
341: with a small anisotropy parameter, $\alpha<<1$.
342: Then Eq.(\ref{localc}) reads
343: \begin{equation}
344: \frac{dF}{dz}=- \frac{\sigma}{\Delta\alpha^{5/4}}A(\tau)\frac{1+d^2F/dz^2}{(2\tau)^{3/2}}
345: \label{sing1}
346: \end{equation}
347: where the anisotropy factor is $A(\tau)=1 -[\exp(-4i\theta_1)+\exp(-4i\theta_2)]/\tau^2$.
348: This equation is similar to the limit of the kinetic dendrite in 
349: Eq.(9.1) of Ref.\cite{brener91}.  This similarity is purely formal because the real 
350: interface kinetic effects 
351: are not included in our description. Instead,  the strong diffusion interaction leads 
352: to such a selection problem. The same arguments as in \cite{brener91} give 
353: the selection condition
354: \begin{equation} 
355: \sigma=\sigma^{\star}\sim \Delta\alpha^{5/4}, \qquad \Delta<< \sqrt{\alpha}.
356: \label{sel1}
357: \end{equation}
358: Moreover, the selected growth direction lies in between  
359:  the directions of  the minima 
360: of $d_1(\theta)$ and $d_2(\theta)$ for the two interfaces, 
361: $\theta_1=-\theta_2$, because the coefficient
362:  appearing in $A(\tau)$ must be real \cite{brener91}. 
363: 
364: While the regular correction given by Eq.(\ref{reg}) is valid for $\Delta <<1$, the 
365: restriction for the validity of inner Eq.(\ref{sing1}) 
366: and for the scaling relation, Eq.(\ref{sel1}), 
367: is much stronger, $\Delta << \sqrt{\alpha}<<1$.
368: % when the important  scale of the inner region, 
369: %$\sqrt{\alpha}$, is larger than $\Delta$.
370: In the opposite limit, $\Delta >> \sqrt{\alpha}$, we can derive another local inner 
371: equation where the important scale $|x+i|\sim\sqrt{\alpha}$
372: is much smaller then $\Delta$. The argument of the 
373: hyperbolic functions becomes large and we can neglect the term $K_1$ in Eq.(\ref{fur2}).
374: Then, in the vicinity of the singular point $x=-i$ in the complex plane,  
375: this equation reads:
376: \begin{equation}
377: \zeta_2(x)=-d_2K_2/P,
378: \end{equation}
379: which is the usual inner equation for a single dendrite in the one-sided model 
380: \cite{brener91}. With the same rescaling as in  Eq.(\ref{resc}), we find    
381: \begin{equation}
382: F=\frac{\sigma}{\alpha^{7/4}}A(\tau)\frac{1+d^2F/dz^2}{(2\tau)^{3/2}},
383: \label{sing2}
384: \end{equation}
385: where the anisotropy factor is $A(\tau)=1 -2\exp(-4i\theta_2)/\tau^2$.
386: This corresponds to the selection condition
387: \begin{equation} 
388: \sigma=\sigma^{\star}\sim \alpha^{7/4}, \qquad \Delta>>\sqrt{\alpha}
389: \label{sel2}
390: \end{equation}
391: and the selected growth direction corresponds to the minimum of $d_2(\theta)$, 
392: $\theta_2=0$. Whereas  Eq.(\ref{sing2}) and 
393: the selection condition Eq. (\ref{sel2}) are valid not only for small $\Delta$ 
394: \cite{brener91},  
395: the regular correction in the form of Eq.(\ref{reg}) requires 
396: $\Delta<<1$. 
397: 
398: The selection conditions, Eqs.(\ref{sel1},\ref{sel2}), together with the relations 
399: for the Peclet numbers, Eq.(\ref{peclet}), solve the posed problem of pattern formation 
400: for two combined nearly parabolic fronts:
401: \begin{equation}
402: V=\frac{2D}{d_0}P^2(\Delta)\sigma^{\star}
403: \label{vel}
404: \end{equation}
405: and 
406: \begin{equation}
407: R_2=d_0/[P(\Delta)\sigma^{\star}], \quad R_1=R_2/(1-\Delta)^2.
408: \label{R}   
409: \end{equation}
410: 
411: While these results are formally similar to the free dendritic case, 
412: the selected stability
413: parameter $\sigma^{\star}$ scales as $\alpha^{7/4}$ only for $\Delta>>\sqrt{\alpha}$.
414: Otherwise, the other scaling relation, Eq.(\ref{sel1}), holds. In principal, it could 
415: be conceivable that the strong diffusional interaction between the front leads to 
416: the selection even without anisotropy. However, our analysis of the inner equation 
417: does not support this hypothesis. The Peclet number is 
418: $P(\Delta)\sim b\Delta$ which reflects the fact that the coherency 
419: strain energy plays a crucial role  
420: in the LFM mechanism. This parameter $b$ is usually small but the melting process by 
421: the LFM mechanism is controlled by  the fast diffusion in the liquid, whereas the  
422: the process with only one melting front would be controlled
423:  by the very slow diffusion in the mother solid phase. 
424: As we already noted, with the help of 
425: the LFM mechanism the system chooses  a more efficient kinetic path to relax to 
426: the equilibrium state.
427: 
428: Finally, we estimate the velocity $V$ from Eq.(\ref{vel}) and the thickness of the film
429: $R_2\Delta$  from Eq.(\ref{R})
430: using  characteristic values of the parameters: $D\sim 10^{-5}$cm$^2$/s, 
431: $d_0\sim 10^{-7}$cm, $b\sim 0.05$,  $\Delta\sim 0.05$ and $\sigma^{\star}\sim 10^{-2}$. 
432: This leads to $V\sim 10^{-5}$cm/s and $R_2\Delta\sim 10^{-4}$cm, 
433: which qualitatively agree with typical values in LFM experiments.
434: 
435: In conclusion, we developed a selection theory for the process of liquid film migration 
436: where the strong diffusion interaction between melting and solidification fronts plays
437: a crucial role. This process is very important in practical applications, in particular 
438: during sintering  in the presence 
439: of the liquid phase \cite{yoon1,yoon2}. However, despite  its practical importance, 
440: experimental investigations of this process are  far from the level of accuracy 
441: of the model experiments in classical dendritic growth. 
442: %Perhaps, this is due to the fact 
443: %that practically important alloys are not very suitable for accurate testing of 
444: %the pattern formation process, while the most detailed and 
445: %elegant experiments in dendritic 
446: %growth were performed with model transparent materials. The other reason could be 
447: %that the selection theory developed for dendritic growth was not  applied to 
448: %LFM so far.
449: Our approach extends selection theory developed for dendritic growth to LFM and  
450: we hope that our results will stimulate  further theoretical and 
451: experimental investigations in this very interesting field.
452: From the theoretical side it would be a challenge to attack this problem   by a direct 
453: numerical approach, for example, by means of the phase-field model.
454: 
455: 
456: We acknowledge the support by the Deutsche Forschungsgemeinschaft under 
457: project SPP 1120.
458: 
459: \begin{thebibliography}{99}
460: \bibitem{yoon1} D. N. Yoon, and W. J. Hupmann, Acta Metall {\bf 27}, 973 (1979). 
461: \bibitem{musch1} T. Muschik T, W. A. Kaysser,  and T. Hehenkamp, 
462: Acta Metall {\bf 37}, 603 (1989).
463: \bibitem{yoon2} D. N. Yoon, Int. Mater. Rev. {\bf 40}, (1995). 
464: \bibitem{4} D. N. Yoon, J. W. Cahn,  C.A.  Handwerker,  J. E.Blendell,
465:   and Y. J. Baik, In: Interface Migration and Control of Microstrucutres.
466:   Am Soc. Metals. Park. Ohio (1985), pp. 19-31. 
467: \bibitem{temkin2005} D.E. Temkin , to be published in Acta Materialia.
468: \bibitem{ivan1} G. P. Ivantsov, Dokl. Akad. Nauk SSSR {\bf 58}, 567 (1947).
469: \bibitem{ivan2} G. P. Ivantsov, Dokl. Akad. Nauk SSSR {\bf 83}, 573 (1952).
470: \bibitem{saito96} Y. Saito, Statistical Physics of Crystal Growth, World
471:   Scientific Publishing, Singapore, 1996.
472: \bibitem{kessler88} D. Kessler, J. Koplik, and H. Levine, 
473: Adv. Phys. {\bf{37}}, 255 (1988).
474: \bibitem{brener91}  E. A. Brener, and V. I. Mel'nikov, Adv. Phys. {\bf 40} 53 (1991).
475: \bibitem{ben93} M. Ben Amar, and E.A. Brener,  Phys. Rev. Lett {\bf{71}}, 589 (1993).
476: \bibitem{brener93} E.A. Brener,  Phys. Rev. Lett. {\bf{71}}, 3653 (1993).
477: \end{thebibliography}
478: 
479: \end{document}
480: