The paper addresses the problem of finding a low-rank approximation of a multi-dimensional tensor, $\Phi $, using a subset of its entries. A distinctive aspect of the tensor completion problem explored here is that entries of the $d$-dimensional tensor $\Phi$ are reconstructed via $C$-dimensional slices, where $C < d - 1$. This setup is motivated by, and applied to, the reduced-order modeling of parametric dynamical systems. In such applications, parametric solutions are often reconstructed from space-time slices through sparse sampling over the parameter domain. To address this non-standard completion problem, we introduce a novel low-rank tensor format called the hybrid tensor train. Completion in this format is then incorporated into a Galerkin reduced order model (ROM), specifically an interpolatory tensor-based ROM. We demonstrate the performance of both the completion method and the ROM on several examples of dynamical systems derived from finite element discretizations of parabolic partial differential equations with parameter-dependent coefficients or boundary conditions.
翻译:暂无翻译