Method of the distance function to the Bence-Merriman-Osher algorithm for motion by mean curvature (Q2566915)

The level set formulation of mean curvature flow for hypersurfaces in \({\mathbb R}^n\) was introduced by \textit{S.~Osher} and \textit{J. A. Sethian} [J. Comp. Phys. 79, 12--49 (1988; Zbl 0659.65132)], and developed further by \textit{L. C.~Evans} and \textit{J.~Spruck} [J. Differ. Geom. 33, No. 3, 635--681 (1991; Zbl 0726.53029)] and \textit{Y.-G.~Chen, Y.~Giga} and \textit{S.~Goto} [J. Differ. Geom. 33, No. 3, 749--786 (1991; Zbl 0696.35087)]. Later [\textit{J.~Bence, B.~Merriman} and \textit{S.~Osher}, Diffusion generated motion by mean curvature, In: Computational Crystal Growers Workshop, Selected Lectures in Mathematics, Amer. Math. Soc., Providence, RI, 73--83 (1992; Zbl 0776.65002)] proposed a numerical algorithm for approximating mean curvature flow using the heat equation, repeatedly reinitialized after short time steps. Shortly afterwards \textit{L. C.~Evans} [Indiana Univ. Math. J. 42, 533--557 (1993; Zbl 0802.65098)] gave a rigorous justification of this algorithm using the theory of nonlinear semigroups. Different approaches and extensions were also developed by other authors. Here the authors provide a new, more direct, proof of the convergence of the Bence-Merriman-Osher algorithm. Taking into account the asymptotic behaviour of the approximate solutions \(u_k\) in this scheme in directions normal to the interface \(\{x:u_k(t,x)=0\}\), the authors construct an approximate signed distance function \(z_h\), and show that the limit distance function \(d=d(t,x)\) is a viscosity solution of the the mean curvature flow.