Automated linear solver selection for simulation of multiphysics processes in porous media
This paper describes a machine learning-based approach to tuning the linear solvers and their parameters during the simulation.
Main content
Special attention is given to dynamic linear solver switching, which is guided by the dominant physical phenomena in the model that change with time. The numerical examples demonstrate that the automated linear solver selection approach surpasses the preselected configurations, regardless of the exploration phase, in which the algorithm purposefully selects suboptimal solver configurations to learn about their performance.
Automated linear solver selection for simulation of multiphysics processes in porous media
https://doi.org/10.1016/j.cma.2024.117031
Yury Zabegaev, Eirik Keilegavlen, Einar Iversen, Inga Berre
