A two-grid method with dispersion matching for finite-element Helmholtz problems
By: Christiaan C. Stolk
Potential Business Impact:
Solves hard math problems much faster.
This work is about a new two-level solver for Helmholtz equations discretized by finite elements. The method is inspired by two-grid methods for finite-difference Helmholtz problems as well as by previous work on two-level domain-decomposition methods. For the coarse-level discretization, a compact-stencil finite-difference method is used that minimizes dispersion errors. The smoother involves a domain-decomposition solver applied to a complex-shifted Helmholtz operator. Local Fourier analysis shows the method is convergent if the number of degrees of freedom per wavelength is larger than some lower bound that depends on the order, e.g. more than 8 for order 4. In numerical tests, with problem sizes up to 80 wavelengths, convergence was fast, and almost independent of problem size unlike what is observed for conventional methods. Analysis and comparison with dispersion-error data shows that, for good convergence of a two-grid method for Helmholtz problems, it is essential that fine- and coarse-level dispersion relations closely match.
Similar Papers
Multigrid methods for the ghost finite element approximation of elliptic problems
Numerical Analysis
Solves hard math problems faster on weird shapes.
Convergence theory for two-level hybrid Schwarz preconditioners for high-frequency Helmholtz problems
Numerical Analysis
Makes computer simulations of sound waves faster.
A meshless method for computational electromagnetics with improved dispersion properties
Computational Physics
Makes computer simulations of electricity more accurate.