Matrix Unloaded: GPU-Boosted Solvers for Diffusion Physics


Home News

Developed by LLNL and Portland State University researchers, innovative matrix-free solvers offer performance gains for complex multiphysics simulations. For faster computation, these solvers subdivide physical systems into a finite element space known as H(div). This type of discretization increases the number of finite elements while decreasing their size—think of a 3D mesh breaking down into smaller and smaller squares while retaining a constant overall volume. H(div) is one of four finite element spaces in a set of differential equations known as the de Rham complex, all of which are incorporated into the Livermore-led MFEM (Modular Finite Element Methods) software library. Read more at LLNL Computing.