Time discretization of a nonlinear phase field system in general domains

Abstract

This paper deals with the nonlinear phase field system \begin{equation*} \begin{cases} \partial_t (\theta +\ell \varphi) - \Delta\theta = f & \mbox{in}\ \Omega\times(0, T), \\[1mm] \partial_t \varphi - \Delta\varphi + \xi + \pi(\varphi) = \ell \theta,\ \xi\in\beta(\varphi) & \mbox{in}\ \Omega\times(0, T) \end{cases} \end{equation*} in a general domain $\Omega\subseteq\mathbb{R}^N$. Here $N \in \mathbb{N}$, $T>0$, $\ell>0$, $f$ is a source term, $\beta$ is a maximal monotone graph and $\pi$ is a Lipschitz continuous function. We note that in the above system the nonlinearity $\beta+\pi$ replaces the derivative of a potential of double well type. Thus it turns out that the system is a generalization of the Caginalp phase field model and it has been studied by many authors in the case that $\Omega$ is a bounded domain. However, for unbounded domains the analysis of the system seems to be at an early stage. In this paper we study the existence of solutions by employing a time discretization scheme and passing to the limit as the time step $h$ goes to $0$. In the limit procedure we face with the difficulty that the embedding $H^1(\Omega) \hookrightarrow L^2(\Omega)$ is not compact in the case of unbounded domains. Moreover, we can prove an interesting error estimate of order $h^{1/2}$ for the difference between continuous and discrete solutions.