Stochastic phase-field simulations of symmetric alloy solidification
Extracted source text ยท arXiv cond-mat/0409713
%%Fluctuation and Noise Letters -- (FNL)
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%
%% Updated by Greg Harmer, 13/1/2001
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Choose the appropriate paper size
%\documentclass[twoside,10pt,letterpaper]{newFNLstyle}
\documentclass[twoside,10pt,a4paper]{newFNLstyle}
\usepackage{graphics}
\usepackage{cite}
\begin{document}
\volnumpagesyear{4}{3}{L?}{2004}
\dates{8 June 2004}{30 July 2004}{5 August 2004}
\title{Stochastic phase-field simulations of symmetric alloy solidification}
\authorsone{R. Ben\'{\i}tez and L. Ram\'{\i}rez-Piscina}
\affiliationone{Department de F\'{\i}sica Aplicada,
Universitat Polit\`ecnica de Catalunya}
\mailingone{Doctor Mara\~n\'on 44, E-08028 Barcelona, Spain.}
%% \authorstwo{FOURTH AUTHOR}
%% \affiliationtwo{Full affiliation}
%% \mailingtwo{}
% Add author/affliation/mailing sets as required
\maketitle
\markboth{Stochastic Phase-Field Simulations of Symmetric Alloy Solidification}{R. Ben\'{\i}tez and
L. Ram\'{\i}rez-Piscina}
\pagestyle{myheadings}
% Comment this out to remove the running heads
\keywords{Phase-field models; fluctuations; solidification; transient stages.}
% Keywords have to before the abstract I'm afraid.
\begin{abstract}
We study initial transient stages in directional solidification
by means of a non-variational phase field model with fluctuations.
This model applies for the symmetric solidification of dilute binary solutions
and does not invoke fluctuation-dissipation theorem to account for the
fluctuation statistics.
We devote our attention to the transient regime during which concentration
gradients are building up and fluctuations act to destabilize the interface. To
this end, we calculate both the temporally dependent growth rate of each mode and the
power spectrum of the interface evolving under the effect of fluctuations.
Quantitative
agreement is found when comparing the phase-field simulations with theoretical
predictions.
\end{abstract}
\section{Introduction}
\label{sec:1}
%
Solidification processes constitute a primary subject of research in materials
science.
One main objective
is to achieve the prediction and control of the final
macroscopic properties (mechanical, electrical, etc) of solids grown from their
melts.
These properties largely
depend on the microstructure,
which in turn is result of the
conditions at which the solid was grown \cite{annualreview}.
For example, relevant features at the microscopic scale such as compositional
inhomogeneities or the presence of grain boundaries are direct consequences of
instabilities and subsequent dynamics of the solidification front.
Of particular interest for the metallurgical community has long been the
directional solidification of mixtures and alloys, of direct relevance in
processes of zone melting and for the Bridgman method \cite{kurz}.
This configuration is also an archetypical model system in nonlinear physics to
study pattern selection and complex dynamics \cite{langerrmp}.
In this context it is known that the final wavelength of the dendritic array
depends on the history \cite{wl-1}, and specifically on the initial destabilization
of the solidification front \cite{wl-2,laure-caroli}.
This is a non-steady noise amplification process occurring during the early solute
redistribution transients, for which considering both fluctuations and transient
effects are of key importance. Experimental work in this regime can be found
in \cite{figue1,figue2}. We are interested here in dealing with
these aspects in numerical simulations of a directional solidification experiment,
by using a phase field model supplemented with fluctuations.
During the last decade, the phase-field model has emerged
as a quantitative simulation technique to study complex interfacial
morphologies \cite{CincaBook04}. Recently, the phase field approach
has deserved much attention because it can easily incorporate effects like
system anisotropies, kinetic attachment or equilibrium
fluctuations which might be difficult to include with other
simulation techniques. This method avoids the tracking of the
moving front by introducing a continuous order
parameter $\phi(\vec{r},t)$ (the phase field),
which takes different constant values at the bulk phases
separated by a smooth interface of width $W$.
The model then consists in a set of coupled equations for $\phi(\vec{r},t)$
and for the diffusion field which drives the interfacial dynamics, and that are
constructed so that it reproduces the physical dynamics in the limit $W \rightarrow 0$.
Early formulations of phase field models were variational, i.e. the model
equations were derived from functional derivatives of a single free energy
functional, but the convergence of such models was rather poor.
Recent advances in phase field formulations, either based in the so called thin
interface limit \cite{Karma96,Karma01} or in higher order expansions \cite{almgren-one-sided},
permitted to improve dramatically the convergence of the model, while
dealing with some specific realistic situations.
However the standard procedure to introduce fluctuations in phase field models is
restricted to
variational formulations \cite{Karma99,Pavlik99,Pavlik00}, analogous to the model
C of critical dynamics
\cite{Halperin74,wheeler1,wheeler2}, where the intensity of the fluctuations
can be determined by using a fluctuation-dissipation relation.
As the more recent phase field formulations do not maintain this variational
structure, the fluctuation-dissipation relation
cannot be used to infer the statistics of the noise appearing in
the equations. In the model and simulations presented here, we
rely on a recent calculus \cite{benitez} that projects
the dynamics of a generic stochastic phase field model to the motion of the
fluctuating interface. This procedure provides a prescription for the
intensity of the noise terms in the model, accounting for fluctuations of both
internal and external origin. This approach has previously been used in
equilibrium situations only \cite{benitez}, and our aim here is to employ it in
a transient, out of equilibrium, situation.
We will restrict ourselves to the study of the initial destabilization in the
linear regime, where comparison with theory is possible, as a first step to the
complete problem of considering the whole competition process in the presence of
fluctuations.
The validation of this approach should be also useful in other situations of
interfacial dynamics with fluctuations, both in non variational formulations of
phase field models and in considering more general sources of fluctuations, of
external origin.
\section{Stochastic Phase-Field Model for Symmetric Directional Solidification}
In a directional solidification experiment, the mixture sample is pulled at a
velocity $\tilde{v}_p$ in an externally imposed temperature gradient given by
$T(\tilde{z}) = T_M + \tilde{G} \tilde{z}$, where $T_M$ is the melting temperature
of the mixture and the tilde refers to physical units.
For pulling velocities higher than a certain critical value, the interface becomes
unstable \cite{mullins-sekerka} giving rise to cellular regimes.
The diffusive field which drives the
interface dynamics is the reduced solute concentration field
$u=\frac{C-C_\infty}{\Delta C_0}$,
where $C_\infty$ is the solute concentration of the sample
in the liquid bulk far from the solid-liquid interface
and $\Delta C_0= C_L^0-C_S^0$ is the miscibility gap
which is approximated to be constant in this model.
We are interested in the symmetric case for which the
diffusion coefficients in both phases are equal, situation which is typical in
dilute mixtures of liquid crystals.
We consider a modified version of
a non-variational phase-field model presented by Karma \cite{Karma98,Karma01}
for the directional solidification of a symmetric dilute alloy.
Scaling space and time with $l=D/\tilde{v}_p$, $\gamma=l^2/D$, being $D$ the
solutal diffusivity, we obtain the phase-field equations
%
\begin{eqnarray}
\alpha \varepsilon^2 \partial_{t} \phi &=& \varepsilon^2\nabla^2\phi-
f'(\phi)- \varepsilon \lambda g'(\phi) (u+\frac{z-t}{l_T}) +
\varepsilon^{\frac{3}{2}}\eta({\bf r},t)
\label{eq:pf-2}\\
\partial_{t} u &=& \nabla^{2}u+\frac{1}{2}\partial_{t} h(\phi) -
{\bf \nabla}\cdot {\bf q}({\bf r},t)\;,
\label{eq:u-2}
\end{eqnarray}
%
with
%
\begin{eqnarray}
\langle
\eta({\bf r},t) \eta({\bf r'},t')
\rangle &=&
2 \sigma_\phi^2 \delta({\bf r}-{\bf r'})
\delta(t-t')
\label{eq:eta-2}\\
\langle
q_i({\bf r},t)
q_j({\bf r'},t')
\rangle &=&
2 \sigma_u^2 \delta_{ij} \delta({\bf r}-{\bf r'})\delta(t-t')\;.
\label{eq:q-2}
\end{eqnarray}
%
%
In the last equations, $\varepsilon=W/l$ stands for the dimensionless
thickness of the interface, and $l_T=m \Delta C_0/G$ is the thermal length imposed
by the
external gradient, being $m$ the liquidus slope of the alloy.
Fluctuations have been included following the procedure
introduced in \cite{benitez} to account for the equilibrium statistics.
It can be shown that the
model defined by Eqs.~(\ref{eq:pf-2}), (\ref{eq:u-2}), (\ref{eq:eta-2}) and (\ref{eq:q-2})
recovers the classical moving boundary description of the problem
in the limit of $\varepsilon \rightarrow 0$ (small interface width)
and $\sigma_\phi,\sigma_u \ll 1$ (small noise intensities).
%
For this particular model, and performing a matching asymptotic
procedure up to first order in $\varepsilon$, we find the relations
$\lambda = \frac{I_1}{I_2}\frac{1}{d_0}$,
$\alpha = \frac{\beta + I_4\varepsilon}{d_0}$,
$\sigma_\phi^2 = \frac{\beta v_m I_1}{d_0^2 \Delta C_0 (1-k) N_A l^d}$, and
$\sigma^2_u = \frac{C^0_L v_m}{N_A \Delta C_0^2 l^d}$,
where $I_1$, $I_2$ and $I_4$ are integral constants
given by $I_1=2\sqrt{2}/3$, $I_2=16/15$ and $I_4=0.55$,
and we have chosen the particular set of functions
$f(\phi) = -\frac{1}{2}\phi^2 + \frac{1}{4}\phi^4$,
$g(\phi) = \phi- \frac{2}{3}\phi^3+\frac{1}{5}\phi^5$
and $h(\phi) = \phi$. In the last relations, $\beta$ is the
kinetic attachment, $k$ the segregation coefficient, $N_A$ the Avogadro's number,
$v_m$ the molar volume and $d_0=\sigma T_M / L m \Delta C_0$ the chemical
capillary length,
where $\sigma$ is the surface energy and $L$ the latent heat per unit volume.
%
\section{Simulation Results}
\label{sec:4}
%
As a quantitative benchmark of the results presented in \cite{benitez},
we perform numerical simulations of the
phase-field model and compare them with theoretical predictions obtained from the moving boundary
description of the front.
We will present results for both the stationary power spectrum of the interfacial
fluctuations and for the wavelength selection during the initial recoil transient.
The substance parameters correspond to the liquid crystal
4-n-octylcyanobiphenil (8CB), which has been used in several quantitative
experimental works \cite{oswald,simon,figue1}.
The substance parameters have been obtained from \cite{oswald,simon,figue1}, and
are given by $k=0.9$, $m=88.46$ K/mol,
$T_M=313.5$ K, $\sigma=2.2\times 10^{-4}$ J/m$^2$,
$\beta=113.04$ s mol/m, $D =4\times 10^{-10}$ m$^2$/s,
$L=2.2\times 10^{6}$ J/m$^3$,
%$c=5\times 10^{6}$ J/Km$^3$ (specific heat),
$Z=291.44$ g (molecular weight) and
$\rho=10^6$ g/m$^3$ (density). The experimental parameters chosen in the
simulations
are $C_\infty=0.012$ mol, $\tilde{v}_p=6\times10^{-5} m/s$ and
$\tilde{G}=2.3\times10^{3} K/m$.
%
%% %
\subsection{Stationary power spectrum}
\label{sec:4.2}
%
We have simulated a planar, stationary interface in order to
evaluate its spectral statistical properties. In particular, we consider
the mean-square fluctuation spectrum, which can be predicted analytically by
$S(k)=
\int \frac{dk'}{2\pi} \langle z_f(k) z_f(k') \rangle=
\frac{m v_m}{N_A (1-k) (G + md_0k^2)}$,
being $z_f(k,t)=\int dk \; z_f({\bf r},t)
\; e^{-i {\bf k} {\bf r}}$ the Fourier transform of the interface
position.
%
Figure \ref{fig:zz} compares the theoretical power spectrum $S(k)$ with the
power spectrum obtained by analyzing the spectral properties
of the phase-field interface. This simulation has been performed by using an
explicit finite differences scheme in a $N_x \times N_z=256 \times 100$
rectangular grid with a scaled time step of $\Delta t=6.75 \times 10^{-5}$
and a scaled grid spacing $\Delta z=\Delta x=0.03$.
The interface width is $\varepsilon=0.0375$, and the phase-field
simulations have been thermalized during $10^4$ time steps
before calculating any of the magnitudes presented in this work.
The interfacial fluctuation spectrum has been averaged among $10^4$ samples
taken every 10 simulation steps in order to avoid statistical time correlations.
%
\begin{figure}[t]
\vspace{0.2cm}
\centering{\resizebox{5cm}{!}{\includegraphics{zzdef.eps}}}
\caption{Mean-square interfacial fluctuation spectrum
obtained from phase-field simulations and compared with
the theoretical prediction.}
\label{fig:zz}
\end{figure}
%
The quantitative agreement indicates the correctness of the determination of
the model parameters. The scale associated to the
interfacial width $\varepsilon$, which constitutes the limit
of the descriptions of any phase-field model, is indicated by a vertical dashed
line
and determines the maximum wavelength for which the simulations results are
reliable.
%
\subsection{Transient power spectrum and wavelength selection}
\label{sec:4.3}
%
We next investigate the amplification of the
fluctuations during the initial redistribution transient.
The simulations are compared with a theoretical
prediction based on the Warren and Langer's work on
noise amplification~\cite{wl-2}.
In this reference, the time evolution of the interface correlations
is described by $S(k,t) = S_0(k) \int_{-\infty}^{t} dt'
\exp \left\{2 \int_{t'}^{t} \omega (k,t) dt' \right\}$
%
%
where $\omega(k,t)$ is the transient growth rate
of the symmetric model \cite{dresden}, and $S_0(k)$ is given by
$S_0(k)=\frac{|\omega^e| K_B T^2_M}
{LG(1+d^c_0 \;l_T \;k^2)}$,
being $\omega^e=\omega(k,t=\infty)$ the amplification
rate at $t \rightarrow \infty$.
Figure \ref{fig:spec} shows the amplification of the
interfacial fluctuation spectrum at three different physical times
during the initial transient ($\tilde{t}$= 0, $1.8$ s and $2.7$ s).
Dashed lines are obtained by integrating $S(k,t)$
in the adiabatic regime $\omega \ll k^2$.
In this case the interface thickness is $\varepsilon=0.15$, and
the simulations have been performed in a $500 \times 256$ grid
with $\Delta z=\Delta x=0.09$, $\Delta t=0.00135$. The resulting power spectrum
is noisy because the results are direct spectral properties
of the fluctuating interface at each time and no
average or filtering has been used.
As it can be observed, quantitative agreement is achieved between theory
and simulations in the early time amplification of fluctuations as well as
in the wavelength selection of the cellular pattern.
%
\begin{figure}[t]
\vspace{0.26cm}
\centering{\resizebox{6cm}{!}{\includegraphics{powtransdef.eps}}}
\caption{Noise amplification and wavelength selection
during the transient. Comparison between the theoretical prediction
and phase-field simulation results.}
\label{fig:spec}
\end{figure}
%
%
After the early stages where the growth of fluctuations is linear,
the system enters in a nonlinear regime and a cellular pattern with a
certain wavelength sets in the system. The end of the
linear regime can be defined as the time at which the amplitude
of the most unstable Fourier mode is comparable with its wavelength.
This condition defines a crossover time $t_0$ which can be
theoretically determined by the condition
$<\delta z^2> \sim \lambda_{max}$, where
$<\delta z^2> = \int \frac{d^2 k}{(2\pi)^2} S(k,t)$
is the mean-square fluctuation amplitude and $\lambda_{max}$ is the
wavelength of the largest Fourier mode. Figure \ref{fig:cross}
shows the evolution in time of these two magnitudes in our system, and
indicates the determination of the crossover time $\tilde{t}_0$ in physical units.
%
\begin{figure}[htbp]
\vspace{0.1cm}
\centering{\resizebox{5cm}{!}{\includegraphics{cross_graphdef.eps}}}
%\centering{\frame{\rule{2cm}{2cm}}}
\caption{Determination of the crossover time $\tilde{t}_0$.}
\label{fig:cross}
\end{figure}
%
%% \begin{table}[htbp]
%% \caption{Substance parameters for the 8CB liquid crystal obtained from
%% \cite{oswald,simon,figue1}}
%% \centering\footnotesize
%% \label{tab:subs}
%% \begin{center}
%% \begin{tabular}{|l|c|c|c|}
%% \hline
%% \rule[-1ex]{0pt}{3.5ex} {\bf Magnitude} & {\bf Symbol} & {\bf Value} & {\bf
%% units} \\
%% \hline
%% \rule[-1ex]{0pt}{3.5ex} Segregation Constant & k & 0.9 & dimensionless \\
%% \hline
%% \rule[-1ex]{0pt}{3.5ex} liquid slope & m & 88.46 & K/mol \\
%% \hline
%% \rule[-1ex]{0pt}{3.5ex} Melting temperature & $T_M$ & 313.5 & K \\
%% \hline
%% \rule[-1ex]{0pt}{3.5ex} Surface tension & $\sigma$ & $2.2\times 10^{-4}$ &
%% $J/m^2$ \\
%% \hline
%% \rule[-1ex]{0pt}{3.5ex} interface mobility & $\mu$ & $10^{-4}$ & $m/ K s$\\
%% \hline
%% \rule[-1ex]{0pt}{3.5ex} Diffusion constant & D & $4\times 10^{-10}$ & $m^2/s$
%% \\
%% \hline
%% \rule[-1ex]{0pt}{3.5ex} Latent heat & $L$ & $2.2\times 10^{6}$ & $J/m^3$ \\
%% \hline
%% \rule[-1ex]{0pt}{3.5ex} Specific heat & $c$ & $5\times 10^{6}$ & $J/K m^3$\\
%% \hline
%% \rule[-1ex]{0pt}{3.5ex} Moleculat weight & Pm & 291.44 & g \\
%% \hline
%% \rule[-1ex]{0pt}{3.5ex} density & $\rho$ & $10^6$ & $g/m^3$ \\
%% \hline
%% \end{tabular}
%% \end{center}
%% \end{table}
%
\section{Conclusions}
%
We have studied the initial transients of a directional
solidification experiment of a mixture of liquid crystals in the
context of the symmetric model with constant concentration gap.
Contrary to previous work on transients in fluctuating phase-field
simulations \cite{dresden}, results presented here have been
obtained by using a non-variational formulation of the stochastic
phase-field model, recently proposed \cite{benitez}.
Fluctuations have been implemented giving the correct
equilibrium statistics but without employing any
fluctuation-dissipation relation to determine the noise intensities in
the model. The agreement between predictions and simulations has been excellent,
thus confirming the applicability of this approach for the study of
fluctuations during transient stages. As a final conclusion, the employ of this
class
of efficient stochastic phase-field models opens
new possibilities for the quantitative study by simulation of
microstructure formation in solidification
processes.
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section*{Acknowledgements}
%
This work was financially supported by
%We acknowledge financial support from
Direcci\'on General de Investigaci\'on Cient\'{\i}fica y T\'ecnica
(Spain) (Project BFM2003-07850-C03-02) and
Comissionat per a Universitats i Recerca (Spain) (Project
2001/SGR/00221).
%We also acknowledge computing
%support from Fundaci\'o Catalana per a la Recerca through C4 and CESCA (Spain).
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% For BiBTeX users, just uncomment the following two lines
\bibliographystyle{FNL2}
\bibliography{benbib}
%% \begin{thebibliography}{10}
%% \end{thebibliography}
\end{document}
\begin{thebibliography}{10}
\bibitem{annualreview}
W.~J. Boettinger, S.~R. Coriell, A.~L. Geer, A.~Karma, W.~Kurz, M.~Rappaz and
R.~Trivedi {\em Acta. Mater.\/} {\bf 48} (2000) 43.
\bibitem{kurz}
W.~Kurz and D.~Fisher {\em Fundamentals of Solidification\/} (Trans Tech Pub.,
1989).
\bibitem{langerrmp}
J.~Langer {\em Rev. Mod. Phys.\/} {\bf 52} (1980) 1.
\bibitem{wl-1}
J.~A. Warren and J.~S. Langer {\em Phys. Rev. A\/} {\bf 42}, 6 (1990)
3518--3525.
\bibitem{wl-2}
J.~A. Warren and J.~S. Langer {\em Phys. Rev. A\/} {\bf 47}, 4 (1993)
2702--2712.
\bibitem{laure-caroli}
B.~Caroli, C.~Caroli and L.~Ram\'{\i}rez-Piscina {\em J. Crys. Growth\/} {\bf
132} (1993) 377--388.
\bibitem{figue1}
J.~M.~A. Figueiredo, A.~Vidal and O.~N. Mesquita {\em J. Crys. Growth\/} {\bf
166} (1996) 222.
\bibitem{figue2}
J.~M.~A. Figueiredo, M.~Santos, L.~Ladeira and O.~N. Mesquita {\em Phys. Rev.
Lett.\/} {\bf 71}, 26 (1993) 4397.
\bibitem{CincaBook04}
R.~Gonz\'alez-Cinca, R.~Folch, R.~Ben\'{\i}tez, L.~Ram\'{\i}rez-Piscina,
J.~Casademunt and A.~Hern\'andez-Machado in E.~Korutcheva and R.~Cuerno,
eds., {\em Advances in Condensed Matter and Statistical Mechanics\/} (Nova
Science Publishers, 2004).
\bibitem{Karma96}
A.~Karma and W.~J. Rappel {\em Phys. Rev. E\/} {\bf 53} (1996) R3017.
\bibitem{Karma01}
A.~Karma {\em Phys. Rev. Lett.\/} {\bf 87} (2001) 115701.
\bibitem{almgren-one-sided}
R.~F. Almgren {\em SIAM J. Appl. Math.\/} {\bf 59} (1999) 2086.
\bibitem{Karma99}
A.~Karma and W.-J. Rappel {\em Phys. Rev. E\/} {\bf 60} (1999) 3614.
\bibitem{Pavlik99}
S.~Pavlik and R.~Sekerka {\em Physica A\/} {\bf 268} (1999) 283.
\bibitem{Pavlik00}
S.~Pavlik and R.~Sekerka {\em Physica A\/} {\bf 277} (2000) 415.
\bibitem{Halperin74}
B.~Halperin, P.~Hohenberg and S.~Ma {\em Phys. Rev. B\/} {\bf 10} (1974) 139.
\bibitem{wheeler1}
A.~A. Wheeler, W.~J. Boettinger and G.~B. McFadden {\em Phys.\ Rev. A\/} {\bf
45} (1992) 7424.
\bibitem{wheeler2}
A.~A. Wheeler, W.~J. Boettinger and G.~B. McFadden {\em Phys.\ Rev. E\/} {\bf
47} (1993) 1893.
\bibitem{benitez}
R.~Ben\'{\i}tez and L.~Ram\'{\i}rez-Piscina submitted to Phys. Rev. Lett.
(2004).
\bibitem{mullins-sekerka}
W.~W. Mullins and R.~F. Sekerka {\em J. Appl. Phys.\/} {\bf 35} (1964) 444.
\bibitem{Karma98}
W.~Losert, D.~A. Stillman, H.~Z. Cummins, P.~Kopczy\'nski, W.~J. Rappel and
A.~Karma {\em Phys. Rev. E\/} {\bf 58} (1998) 7492.
\bibitem{oswald}
P.~Oswald, J.~Bechhoefer and A.~Libchaber {\em Phys. Rev. Lett.\/} {\bf 58}, 22
(1987) 2318.
\bibitem{simon}
A.~J. Simon, J.~Bechhoefer and A.~Libchaber {\em Phys. Rev. Lett.\/} {\bf 61},
22 (1988) 2574.
\bibitem{dresden}
R.~Ben\'{\i}tez and L.~Ram\'{\i}rez-Piscina vol.~32 of {\em Lecture Notes in
Computational Science and Engineering\/} 160--165 (Springer-Verlag, Berlin,
2003).
\end{thebibliography}