Bogotá - Ciencias - Doctorado en Ciencias - Matemáticas · 2023
On generalized multiscale methods for flow in complex porous media and their applications
In this document, the Generalized Multiscale Finite Element Method (GMsFEM) is studied, which deals with constructing multiscale spectral basis functions designed for high-contrast multiscale problems. The multiscale basis functions are built from the product of the eigenvectors, computed from a local spectral problem and a partition of unity over the study domain. The eigenvalues detect essential features of the solutions that are not captured by the initial multiscale basis functions. This document reviews the general convergence study where the error estimates are written in terms of the eigenvalues associated with the eigenvectors not used in the construction. Error analysis involves local and global norms that measure the convergence speed of the expansion of the solution in terms of local eigenvectors; this is achieved with a careful choice of the initial multiscale basis functions and the configuration of the eigenvalue problems. Two novel important numerical applications are presented: the first is the free-boundary dam problem posed on a heterogeneous high-contrast medium, where we introduce a fictitious time variable that motivates an adequate time discretization that can be understood as a fixed-point iteration. For the steady-state solution, we use the duality method to deal with the multivalued nonlinear terms involved; then, efficient approximations of pressure and saturation are calculated using the GMsFEM method. The second application is the solution of a parabolic equation. Here implementing time discretizations, such as finite differences or exponential integrators in the presence of a high contrast coefficient, it may not be practical in because each time iteration one needs the computation of matrix operators involving very large and extremely ill-conditioned sparse matrices. The GMsFEM is essential since it allows obtaining the solution of the problem more simply, allowing to combine the GMsFEM with the method of exponential integrators in time to get a good approximation of the final temporary solution. (Texto tomado de la fuente)