Numerical approximation for the mean curvature flow with nucleation using implicit time-stepping: an adaptive algorithm (Q1282123)

This paper presents a new self-adaptive finite element algorithm for the mean curvature flow. For the time discretization of the equation of motion, that includes a forcing term, an implicit scheme was proposed by several authors and this is the starting point for this work. At each time level the finite element approximation minimizes a suitable regularized convex functional which is dealt with elsewhere by the author. As to the main point in the paper, i.e. adaptivity, which is justified by the need to use fine meshes to ensure convergence of the discrete flow to the continuous one, the use of locally refined meshes generated by convenient error indicators tracking the evolution is proposed. The three ingredients employed to obtain a successful procedure are as follows. The refinement/coarsening operations are derived from bisection algorithms formulated by \textit{E. Bänsch} [IMPACT Comput. Sci. Eng. 3, No. 3, 181-191 (1991; Zbl 0744.65074)]. The a posteriori error estimators are based on quite involved error estimates that are obtained following similar arguments developed by \textit{R. Verfürth} [Math. Comput. 62, No. 206, 445-475 (1994; Zbl 0799.65112)] for approximate solutions of nonlinear problems. Finally, an implicit adaptive strategy again by \textit{E. Bänsch} [Computational Mechanics Publications, 47-76 (1993; Zbl 0804.76050)] is the model applied in the present context. The numerical results presented in two sections deal with simulations of isotropic and anisotropic evolutions, and with phase transition in nucleation phenomena. In both cases the accompanying figures speak for themselves.