Phase-field approximation of the Willmore flow (Q2043048)

Let \(T>0\) and \(\Omega\) be a bounded domain of \(\mathbb{R}^N\) with smooth boundary \(\partial\Omega\). Setting \(f(z) =(z^2-1)^2/4\) for \(z\in\mathbb{R}\), formal asymptotics reveals that, as \(\varepsilon\to 0\), the zero level set at time \(t\) of the solution \(\phi_\varepsilon(t)\) to the phase-field system \begin{align*} \varepsilon^3 \partial_t \phi_\varepsilon - \varepsilon^2 \Delta\mu_\varepsilon + f''(\phi_\varepsilon) \mu_\varepsilon & = 0\,, \qquad (x,t)\in \Omega\times (0,T)\,, \\ - \varepsilon^2 \Delta\phi_\varepsilon + f'(\phi_\varepsilon) & = \varepsilon\mu_\varepsilon\,, \qquad (x,t)\in \Omega\times (0,T)\,, \end{align*} approaches an hypersurface \(\Gamma_t^0\) which is such that \(\Gamma^0 = \bigcup_{t\in [0,T)} \Gamma_t^0 \times \{t\}\) evolves according to the Willmore flow; that is, \[ V = \Delta_{\Gamma^0} H + H|A|^2 - \frac{H^3}{2}\,, \] where \(V\), \(H\), and \(A\) are the normal velocity, the scalar mean curvature, and the second fundamental form of \(\Gamma^0\), respectively, the quantity \(|A|^2\) corresponding to the sum of the squares of the principal curvatures. A rigorous justification of this convergence is provided here when \(N\in\{2,3\}\), and the phase-field system is supplemented with boundary conditions \(\phi_\varepsilon=1\) and \(\nabla\phi_\varepsilon\cdot \mathbf{n}=0\) on \(\partial\Omega\times (0,T)\) and initial condition. More precisely, it is shown that, given a smooth solution \(\Gamma^0\) to the Willmore flow defined up to some time \(T>0\), \(k\ge 10\), and \(\varepsilon>0\) sufficiently small, there is an approximate solution \((\phi_{a,\varepsilon},\mu_{a,\varepsilon})\) to the phase-field system in \(\Omega\times (0,T)\) with an error of order \(\varepsilon^{k-1}\) which satisfies \((\phi_{a,\varepsilon}(t),\mu_{a,\varepsilon}(t)) = (\pm 1,0)\) outside a neighbourhood of \(\Gamma^0_t\) in the following sense: introducing the signed distance \(d^{(0)}(x,t)\) from \(x\) to \(\Gamma_t^0\), \((\phi_{a,\varepsilon}(x,t),\mu_{a,\varepsilon}(x,t)) = (\pm 1,0)\) when \(\pm d^{(0)}(x,t)\ge \delta\). Furthermore, if the initial condition \(\phi_\varepsilon(0)\) is sufficient close to \(\phi_{a,\varepsilon}(0)\) and satisfies \(\|\phi_{\varepsilon}(0)-\phi_{a,\varepsilon}(0)\|_{L^2(\Omega)} \le C_{in} \varepsilon^{7/2}\) for a sufficiently small constant \(C_{in}\), then this error propagates and \(\|\phi_\varepsilon(t)-\phi_{a,\varepsilon}(t)\|_{L^2(\Omega)} \le C \varepsilon^{7/2}\) for \(t\in [0,T]\).