A priori analysis of a tensor ROM for parameter dependent parabolic problems (Q7030698)

Tensor methods are applied as intrusive reduced order models (ROMs) for parametric parabolic PDEs. The singular value decomposition (SVD) is replaced with a low-rank tensor decomposition (LRTD) of the snapshot tensor, which allows to recover information about the variation of optimal reduced spaces with respect to parameters. Furthermore, parameter-specific reduced spaces are constructed for out-of-sample parameter values. By exploiting the local smoothness of the solution manifold, global characteristics such as Kolmogorov \(n\)-widths are avoided. Existing a priori error analyses of Galerkin ROMs are extended to the parametric setting, covering out-of-sample parameter values. Approximation estimates are validated by numerical experiments for the heat equation, and advection-diffusion equation.