Existence of nonnegative solutions to stochastic thin-film equations in two space dimensions (Q2171951)

This paper treats the existence of martingale solutions to stochastic thin-film equations in the physically relevant space dimension \(d=2\). Namely, \[ du=-\operatorname{div}\{m(u)\nabla(\varDelta u-F'(u))\}dt+\operatorname{div}\{\sqrt{m(u)}dW\}\tag{1} \] on a space-time cylinder \(\mathcal{O} \times (0, T]\). Here the mobility \(m(\cdot)\) may be chosen as \(m(u) = u^3 + \beta u^2\). Roughly speaking, the authors rely on a stochastic Faedo-Galerkin approach using tensor-product linear finite elements in space. Augmenting the physical energy on the approximate level by a curvature term weighted by positive powers of the spatial discretization parameter \(h\), they succeed in deriving stochastic counterparts of the energy and entropy estimates known from the deterministic setting, by combining Itô's formula with inverse estimates and appropriate stopping time arguments. In the limit \(h \searrow 0\), they prove that strictly positive finite elements (solutions) converge towards nonnegative martingale solutions by making use of compactness arguments based on Jakubowski's theorem and subtle exhaustion arguments to identify third-order spatial derivatives in the flux terms. More precisely, the authors consider the existence of martingale solution to the stochastic thin-film equation of the form: \[ du=-\operatorname{div}\{m(u)\nabla (\varDelta u-F'(u))\}dt+\sum_{\alpha\in\{x,y\}}\sum_{k,\ell\in\mathbb{Z}}\partial_{\alpha}(\sqrt{m(u)}\lambda_{k\ell}^{\alpha} g_{k \ell}) d \beta_{k \ell}^{\alpha}\text{ on }\mathcal{O} \times [0, \infty).\tag{2} \] Assume that the effective interface potential \(F\) has continuous second-order derivatives on \(\mathbb{R}^+\). \(W\) is a \(Q\)-Wiener process on a probability space \((\Omega, \mathcal{F}, \mathbb{P})\), adapted to the usual filtration \((\mathcal{F}_t)_{t \geq 0}\) which admits a decomposition of the form: \[ W = \sum_{\alpha \in \{x, y \}} \sum_{k, \ell \in \mathbb{Z}} \lambda_{k \ell}^{\alpha} g_{k \ell}\mathbf{b}_{\alpha} \beta_{k \ell}^{\alpha} \] for independent sequences of i.i.d. Brownian motions \(\beta_{k \ell}^{\alpha}\), \((\alpha \in \{x, y \})\) and a sequence of sufficiently smooth basis functions \(g_{k \ell}\). Here, \(\mathbf{b}_x\) denotes the standard Cartesian basis vectors in \(\mathbb{R}^2\). Its components are given by \[ W_{\alpha} := \sum_{k, \ell \in \mathbb{Z}} \lambda_{k \ell}^{\alpha} g_{k \ell} \beta_{k \ell}^{\alpha}, \quad \alpha \in \{x, y \}. \] The corresponding components of \(Q\) is denoted by \(Q_x\), for instance. The noise \(W\) is colored in the sense that, for a positive constant \(C > 0\), \[ \sum_{k, \ell = 1}^{\infty} \{(\lambda_{k \ell}^x)^2 + (\lambda_{k \ell}^y)^2 \} \Vert g_{k \ell} \Vert_{W^{2, \infty}(\mathcal{O})}^2 \leqslant C \] holds. The regularization parameter \(\varepsilon\) (which is used to define the regularized discrete energy and the discrete entropy) is small enough such that there exists a constant \(\rho > 0\) such that \[ 1 > \frac{2}{p} + \frac{\varepsilon}{2} +\frac{\rho}{2p}, \] where \(p\) is the exponent associated with the growth of \(F\). Moreover, \(\mathcal{O} \subset \mathbb{R}^2\) is a bounded rectangular spatial domain. \(W^{k,p}(\mathcal{O})\) denotes the space of \(k\)-times weakly differentiable functions with weak derivatives in \(L^p(\mathcal{O})\), and in particular for the case \(p=2\), it will be denoted by \(H^k(\mathcal{O})\). And also the subscript ``per'' indicates the subspace of \(\mathcal{O}\)-periodic ones. Let \(\Lambda\) be a probability measure on \(H_{per}^2(\mathcal{O})\). A triple \(((\tilde{\Omega}, \tilde{\mathcal{F}}, (\hat{\mathcal{F}}_t), \tilde{\mathbb{P}}), \tilde{u}, \tilde{\mathbb{W}})\) is called a weak martingale solution to the stochastic thin-film equation (2) with initial data \(\Lambda\) on the time interval \([0, T]\) provided that the solution \(\tilde{u}\) is an element of \[ L^q(\tilde{\Omega}; L^{\infty} (0, T; H_{per}^1(\mathcal{O}))) \cap L^2(\tilde{\Omega}; L^2(0, T; H_{per}^2(\mathcal{O}))) \cap L^{\sigma} (\tilde{\Omega} ; C^{1/4} ([0,T]; (H_{per}^1(\mathcal{O}))')) \] for all \(q < \infty\) and \(\sigma < 8/5\) such that \(\sqrt{m (\tilde{u})} \nabla (\varDelta \tilde{u} - F'(\tilde{u})) \in L^2(\{\tilde{u} > 0 \})\), and also provided that there exists an \(\tilde{\mathcal{F}}_0\)-measurable \(H_{per}^2(\mathcal{O}; \mathbb{R}^+)\)-valued random variable \(\tilde{u}_0\) such that \(\Lambda = \tilde{\mathbb{P}} \circ (\tilde{u}^0)^{-1}\), and the equation \[ \begin{multlined} \int_{\mathcal{O}} \tilde{u}(t) \phi dx dy = \int_{\mathcal{O}} \tilde{u}^0 \phi dx dy + \int \int_{\{\tilde{u} > 0 \}} m(\tilde{u}) \nabla (\varDelta \tilde{u} - F'(\tilde{u})) \cdot \varDelta \phi dx dy ds \\ - \sum_{\alpha \in \{x,y \}} \sum_{k, \ell \in \mathbb{Z}} \lambda_{k \ell}^{\alpha} \int_0^t \int_{\mathcal{O}} \sqrt{m(\tilde{u})} g_{k \ell} \partial_{\alpha} \phi dx dy d \tilde{\beta}_{k \ell}^{\alpha} \end{multlined} \] holds \(\tilde{\mathbb{P}}\)-a.s. for all \(t \in [0, T]\) and all \(\phi \in W_{per}^{1, q^*} (\mathcal{O})\) with \(q^* > 2\). Here is the main theorem for the existence of weak martingale solutions to the equation (2). \textbf{Theorem.} With some additional assumptions, there exists a stochastic basis \((\tilde{\Omega}, \tilde{\mathcal{F}}, (\tilde{\mathcal{F}}_t), \tilde{\mathbb{P}})\) and processes \(\tilde{u}_h\), \(\tilde{J}_h^x\), \(\tilde{J}_h^y\), and \(\tilde{u}\) such that the following holds: \begin{itemize} \item[(i)] The processes \(\tilde{u}_h\), \(\tilde{J}_h^x\), and \(\tilde{J}_h^y\) have the same law as the processes \(u_h\), \[ J_h^x := \mathcal{I}_h^y \{\sqrt{[ G''(u_h) ]_x^{-1}} \partial_x p_h \} \quad \text{and} \quad J_h^y := \mathcal{I}_h^x \{\sqrt{[ G''(u_h) ]_y^{-1}} \partial_y p_h \} \] with interpolation operators \(\mathcal{I}_h^{\{\cdot \}}\) and \(G(u) = \int_1^u \int_1^s m(r)^{-1} dr ds\); \item[(ii)] for an appropriate subsequence, the \(\tilde{\mathbb{P}}\)-almost sure convergences \[ \tilde{u}_h \to \tilde{u}\quad \text{ strongly in }C([0, T]; L^q(\mathcal{O})) \cap L^2(0, T: W_{per}^{1,q}(\mathcal{O})) \quad (1 \leqslant q < \infty), \] \[ \tilde{J}_h^x \rightharpoonup \tilde{J}^x \quad \text{ weakly in } L^2(0, T; L^2(\mathcal{O})), \] which can be identified with \(- \tilde{u} \partial_x (\varDelta \tilde{u} - F'(\tilde{u}))\) on \([ \tilde{u} > 0 ]\), and \[ \tilde{J}_h^y \rightharpoonup \tilde{J}^y \quad \text{ weakly in } L^2(0, T; L^2(\mathcal{O})), \] which can be identified with \(- \tilde{u} \partial_y (\varDelta \tilde{u} - F' (\tilde{u}))\) on \([ \tilde{u} > 0 ]\). \end{itemize} Furthermore, \(\tilde{u}\) is a weak martingale solution to the stochastic thin-film equation (2) in the aforementioned sense. In the above, \(h\) is a parameter of the practical numerical simulation for the given stochastic equation, and \(u_h\) and \(p_h\) are solutions to the regularized semi-discrete version of the equation (2). The ussage of the symbol \(\tilde{u}\) instead of ordinary \(u\) indicates the fact that in the compactness argument the authors employed Jakubowski's theorem [\textit{A. Jakubowski}, Teor. Veroyatn. Primen. 42, No. 1, 209--216 (1997; Zbl 0923.60001)] which is a generalization of Skorokhod's theorem [\textit{A. V. Skorokhod}, Teor. Veroyatn. Primen. 1, 289--319 (1956; Zbl 0074.33802)]. For other related works, see, e.g., [\textit{K. Dareiotis} et al., Arch. Ration. Mech. Anal. 242, No. 1, 179--234 (2021; Zbl 1471.35345)] for non-negative martingale solutions to the stochastic thin-film equation with nonlinear gradient noise; [\textit{J. Fischer} and \textit{G. Grün}, SIAM J. Math. Anal. 50, No. 1, 411--455 (2018; Zbl 1395.76013)] for the existence of positive solutions to stochastic thin-film equations; [\textit{B. Gess} and \textit{M. V. Gnann}, Stochastic Processes Appl. 130, No. 12, 7260--7302 (2020; Zbl 1454.60092)] for the stochastic thin-film equation: existence of nonnegative martingale solutions.