Iterative and direct solutions of Hierarchical Poincaré-Steklov type discretizations for the 3D Helmholtz equation with variable coefficients.
When
We present two solvers for the Hierarchical Poincaré-Steklov (HPS) discretization of 3D variable coefficient Helmholtz problems appearing in geological prospection. An iterative approach uses a GMRES solver coupled with a leaf-wise block-Jacobi preconditioner. The preconditioner is built using two nested local solvers accelerated by local homogenization. Both the operator and preconditioner are implemented in a matrix-free fashion and with distributed memory. The solver can tackle problems approximately 50 wavelengths in each direction requiring more than a billion unknowns to get approximately 7 digits of accuracy half an hour. We compare with a the direct solver where matrix compression accelerate the solution and reduce memory footprint. We test both approaches and their performance with application examples. This is a work funded by Total Energies and with Adrianna Gillman.
Speaker will be on Zoom.
Place: Math, 402 and Zoom https://arizona.zoom.us/j/83758253931Password: applied