TY - JOUR

T1 - Lack of robustness and accuracy of many numerical schemes for phase-field simulations

AU - Xu, Jinchao

N1 - KAUST Repository Item: Exported on 2023-06-09
Acknowledgements: This work is partially supported by the Center for Computational Mathematics and Applications at Pennsylvania State University and KAUST Baseline Research Fund. In addition, the second author was partially supported by the Computational Materials Sciences Program funded by the US Department of Energy, Office of 11Science, Basic Energy Sciences, under Award Number DE-SC0020145, when he was studying at Pennsylvania State University.

PY - 2023/5/16

Y1 - 2023/5/16

N2 - In this paper, we study the stability, accuracy and convergence behavior of various numerical schemes for phase-field modeling through a simple ODE model. Both theoretical analysis and numerical experiments are carried out on this ODE model to demonstrate the limitation of most numerical schemes that have been used in practice. One main conclusion is that the first-order fully implicit scheme is the only robust algorithm for phase-field simulations while all other schemes(that have been analyzed ) may have convergence issue if the time step size is not exceedingly small. More specifically, by rigorous analysis in most cases, we have the following conclusions:
(i) The first-order fully implicit scheme converges to the correct steady state solution for all time step sizes. In the case of multiple solutions, one of the solution branches always converges to the correct steady state solution.
(ii) The first-order convex splitting scheme, which is equivalent to the first-order fully implicit scheme with a different time scaling, always converges to the correct steady state solution but may seriously lack numerical accuracy for transient solutions.
(iii) For the second-order fully implicit and convex splitting schemes, for any time step size δt>0, there exists an initial condition u0, with |u0|>1, such that the numerical solution converges to the wrong steady state solution.
(iv) For |u0|≤1, all second-order schemes studied in this paper converge to the correct steady state solution although severe numerical oscillations occur for most of them if the time step size is not sufficiently small.
(v) An unconditionally energy-stable scheme (such as the modified Crank–Nicolson scheme) is not necessarily better than a conditionally energy-stable scheme (such as the Crank–Nicolson scheme).
Most, if not all, of the above conclusions are expected to be true for more general Allen–Cahn and other phase-field models.

AB - In this paper, we study the stability, accuracy and convergence behavior of various numerical schemes for phase-field modeling through a simple ODE model. Both theoretical analysis and numerical experiments are carried out on this ODE model to demonstrate the limitation of most numerical schemes that have been used in practice. One main conclusion is that the first-order fully implicit scheme is the only robust algorithm for phase-field simulations while all other schemes(that have been analyzed ) may have convergence issue if the time step size is not exceedingly small. More specifically, by rigorous analysis in most cases, we have the following conclusions:
(i) The first-order fully implicit scheme converges to the correct steady state solution for all time step sizes. In the case of multiple solutions, one of the solution branches always converges to the correct steady state solution.
(ii) The first-order convex splitting scheme, which is equivalent to the first-order fully implicit scheme with a different time scaling, always converges to the correct steady state solution but may seriously lack numerical accuracy for transient solutions.
(iii) For the second-order fully implicit and convex splitting schemes, for any time step size δt>0, there exists an initial condition u0, with |u0|>1, such that the numerical solution converges to the wrong steady state solution.
(iv) For |u0|≤1, all second-order schemes studied in this paper converge to the correct steady state solution although severe numerical oscillations occur for most of them if the time step size is not sufficiently small.
(v) An unconditionally energy-stable scheme (such as the modified Crank–Nicolson scheme) is not necessarily better than a conditionally energy-stable scheme (such as the Crank–Nicolson scheme).
Most, if not all, of the above conclusions are expected to be true for more general Allen–Cahn and other phase-field models.

UR - http://hdl.handle.net/10754/692506

UR - https://www.worldscientific.com/doi/10.1142/S0218202523500409

U2 - 10.1142/S0218202523500409

DO - 10.1142/S0218202523500409

M3 - Article

SN - 1793-6314

SP - 1

EP - 26

JO - MATHEMATICAL MODELS & METHODS IN APPLIED SCIENCES

JF - MATHEMATICAL MODELS & METHODS IN APPLIED SCIENCES

ER -