A Novel Proper Generalized Decomposition (PGD) Based Approach for Non-Matching Grids

Abstract

Proper Generalized Decomposition (PGD) is a recent model reduction technique, successfully employed to solve many multidimensional problems. This method is able to circumvent, or at least alleviate, the curse of dimensionality. This method is based on the use of separated representations. By avoiding the exponential complexity of standard grid based discretization techniques, the PGD circumvents the curse of dimensionality in a variety of problems. With the PGD, the problem’s usual coordinates (e.g. space, time), but also model parameters, boundary conditions, and other sources of variability can be viewed globally as coordinates of a high-dimensional space wherein an approximate solution can efficiently be computed at once. Non-matching grids are very common in advanced scientific computing (e.g. contact problems, sub-domains coupling,).In this framework, approximate solutions from one grid to a non-matching second grid needs to be projected. This approach poses substantial numerical complexity which increases when going from one to higher dimensional interfaces. In this paper, we try to simulate a domain, which has a coarse mesh on one side and a fine mesh on other side by PGD. We show that PGD can handle these non -matching grids by using a smooth transition of the separated representation description.