A Fast Propagation Method for the Helmholtz equation

TL;DR

This paper introduces a hierarchical, domain decomposition-based fast method for solving high frequency Helmholtz equations efficiently, especially suitable for large-scale problems on parallel computers.

Contribution

It extends the source transfer domain decomposition method with a hierarchical approach, reducing computational complexity for high frequency Helmholtz problems.

Findings

Achieves $O(n^{3/2} \, \log n)$ complexity in 2D

Can solve problems with half a billion unknowns efficiently

Demonstrates scalability on massively parallel machines

Abstract

A fast method is proposed for solving the high frequency Helmholtz equation. The building block of the new fast method is an overlapping source transfer domain decomposition method for layered medium, which is an extension of the source transfer domain decomposition method proposed by Chen and Xiang \cite{Chen2013a,Chen2013b}. The new fast method contains a setup phase and a solving phase. In the setup phase, the computation domain is decomposed hierarchically into many subdomains of different levels, and the mapping from incident traces to field traces on all the subdomains are set up bottom-up. In the solving phase, first on the bottom level, the local problem on the subdomains with restricted source is solved, then the wave propagates on the boundaries of all the subdomains bottom-up, at last the local solutions on all the subdomains are summed up top-down. The total computation cost of the new fast method is $O(n^{\frac{3}{2}} \log n)$ for 2D problem. Numerical experiments shows that with the new fast method, Helmholtz equations with half billion unknowns could be solved efficiently on massively parallel machines.