Strong convergence rate of finite difference approximations for stochastic cubic Schrödinger equations (Q2013151)

A Stratonovich stochastic Schrödinger equation \(\mathbf i\,du+(\Delta u+\lambda|u|^2u)\,dt=u\circ\,dW\) in \((0,T]\times(0,1)\) with an initial datum \(u(0)=u_0\) under the homogeneous Dirichlet boundary condition is considered. Here, \(\lambda\in\{-1,1\}\), \(T>0\) and \(W\) is a Wiener process in \(L^2(0,1)\) with covariance operator \(Q\). The authors first give conditions on \(u_0\), \(Q\), \(p\geq 2\) and \(\mathbf s\geq 2\) that guarantee the boundedness of the uniform \(p\)-moments \(\mathbb E\,[\sup_{t\in[0,T]}\|u(t)\|^p_{\mathbb H^{\mathbf s}}]\) of the solution in the Sobolev space \(\mathbb H^s\), and an analogous estimation of the \(p\)-moments \(\mathbb E\,[\sup_{t\in[0,T]}\|\delta_+u^h(t)\|^p_{\ell^2_h}]\) of the solution \(\{u^h(l)\}\) to a central difference scheme \[ du^h(l)=\left(\mathbf i\delta_+\delta_-u^h(l)+\mathbf i\lambda|u^h(l)|^2u^h(l)-\frac 12F_Q(l)u^h(l)\right)\,dt-\mathbf iu^h(l)\,dW(t,l), \] for a uniform partition of \([0,1]\), where \(\delta_+\) and \(\delta_-\) are the forward and the backward difference operators, respectively, and \(F_Q=\sum_k(Q^\frac12e_k)^2\) for an orthonormal basis \(\{e_k\}\) in \(L^2(0,1)\). Next, conditions on \(u_0\), \(Q\) and \(q\geq 1\) for the exponential integrability \[ \left\|\exp\left(\int_0^T\|u_0\|_{L^2}\|\nabla u\|_{L^2}\,dr\right)\right\|_{L^q(\Omega)}+\left\|\exp\left(\int_0^T\|u^h_0\|_{\ell^2_h}\|\delta_+u^h\|_{\ell^2_h}\,dr\right)\right\|_{L^q(\Omega)}\leq C \] for both the stochastic equation and its central difference scheme are established, as well as continuous dependence in \(L^p(\Omega;C([0,T];L^2(0,1)))\), i.e., \[ \|u-v\|_{L^p(\Omega;C([0,T];L^2(0,1)))}\leq C\|u(0)-v(0)\|_{L^{2p}(\Omega;L^2(0,1))}, \] of the solutions of the stochastic equation on the initial data for \(p=2\) or \(p\geq 4\). If the equations \(\mathbf i\,du^\varepsilon+(\Delta u^\varepsilon+\lambda|u^\varepsilon|^2u^\varepsilon)\,dt=\varepsilon u^\varepsilon\circ\,dW\), \(u^\varepsilon(0)=u_0\) are considered for \(\varepsilon\geq 0\), then the estimation \[ \|u^\varepsilon-u^0\|_{L^p(\Omega;C([0,T];L^2(0,1)))}\leq C|\varepsilon| \] is proved for \(p=2\) or \(p\geq 4\). Finally, the authors present several sets of assumptions on \(u_0\), \(p\geq 2\) and \(Q\) that guarantee the convergence of the solutions of the central difference scheme to the solution of the stochastic equation in the following sense: \[ \|u-u^h\|_{L^p(\Omega;C([0,T];\ell^2_h))}\leq Ch^2. \]