Mortar upscaling for multiphase flow in porous media (Q1611741)

A novel multiblock mortar methodology for modeling complex multiphysics phenomena occurring in energy and environmental applications allows for coupling different physical processes and different time discretizations in a single simulation. This is achieved by decomposing the physical domain into a series of subdomains (blocks) and using independently constructed numerical grids and possibly different discretization techniques in each block. Physically meaningful matching conditions can be imposed on block interfaces in a numerically stable and accurate way using mortar finite element spaces. In this paper the authors establish a close connection between mortar methodology and some recent upscaling procedures often referred to as subgrid-scale modeling which treat linear steady state or linearized transient problems. The motivation for these formulations is that fine scale features are often numerically unresolvable on practical finite element meshes. An advantage of mortar interface formulation is that it provides the flexibility to adaptively vary the number of mortar degrees of freedom. It is shown herein that finer mortar grids provide better accuracy while coarser mortar-grids lead to easier solutions of algebraic problems. In highly heterogeneous large variation problems, however, it may be necessary to use finer mortar grids in parts of the domain for better flow resolution. Natural way of measuring the mortar upscaling error is to compute the flux jump in a finer mortar space. This can be used as an estimator for adapting the mortar grids. The tests conducted by the authors indicate that, with proper adaptivity, the increase in computational cost is only a fraction of the increase in soluiton quality. The ability to measure the mortar upscaling error and to adaptively account for it provides an advantage to standard upscaling methods where the upscaling error is often difficult to estimate. It is further shown that substantial computational savings can be achieved without sacrificing accuracy by appropriately choosing different physical models (possibly non-matching) local grids, and by different numerical methods in different subdomains.