On Mon, Oct 10, 2022 at 5:37 PM Alexander Lindsay <[email protected]> wrote:
I know that PETSc has native support for ASPIN. Has anyone tried MSPIN? I wouldn't be surprised if someone has implemented it in user code. Wondering what the barriers would be to creating an option like `-snes_type mspin` ?
David Keyes, LuLu Liu, and collaborators have several papers on MSPIN. It does work well in many circumstances. ASPIN is easy for PETSc because it only involves generating sub-Jacobians. MSPIN needs nonlinear subsystems. This is not possible with the traditional PETSc SNES callback interface. It is just barely possible using all the experimental stuff in Plex. You need the ability to subset the domain, setup a nonlinear problem with the same equations and boundary conditions (and normally homogeneous Dirichlet on the internal boundary), and usually linearizations of this subproblem. This means abstractions for the mesh, equations, boundary conditions, and linearizations that can be transferred onto a new subdomain. Thanks, Matt -- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>