Minisymposium Presentation

The development of robust and efficient multi-physics block preconditioners is crucial for drastically improving the computational performance of large-scale, thermo-mechanically coupled ice-sheet models. The primary challenge in simulating ice sheets at scale is solving the linear system associated with a thin, high-aspect ratio mesh. This problem is exacerbated by the coupling of first-order Stokes, a nonlinear elliptic problem for ice velocity, with the enthalpy formulation, an advection dominated problem. In this presentation, we show the performance of various multi-physics preconditioning strategies to assess a viable strategy for simulating the large, coupled ice-sheet model on large GPU supercomputers such as Perlmutter and Frontier.