EigenWave: An Optimal O(N) Method for Computing Eigenvalues and Eigenvectors by Time-Filtering the Wave Equation

TL;DR

EigenWave is an efficient, optimal O(N) algorithm that computes eigenvalues and eigenvectors of elliptic problems by filtering solutions of a wave equation, enabling spectrum-wide eigenpair computation without matrix inversion.

Contribution

The paper introduces EigenWave, a novel wave-equation-based method that computes eigenpairs efficiently for any spectrum region, integrating with Arnoldi and using implicit time-stepping for optimal scaling.

Findings

Achieves linear O(N) complexity with multigrid acceleration.

Successfully computes eigenpairs in complex geometries in 2D and 3D.

Demonstrates high accuracy with second and fourth-order schemes.

Abstract

An algorithm named EigenWave is described to compute eigenvalues and eigenvectors of elliptic boundary value problems. The algorithm, based on the recently developed WaveHoltz scheme, solves a related time-dependent wave equation as part of an iteration. At each iteration, the solution to the wave equation is filtered in time. As the iteration progresses, the filtered solution generally contains relatively larger and larger proportions of eigenmodes whose eigenvalues are near a chosen target frequency (target eigenvalue). The ability to choose an arbitrary target frequency enables the computation of eigenvalues anywhere in the spectrum, without the need to invert an indefinite matrix, as is common with other approaches. Furthermore, the iteration can be embedded within a matrix-free Arnoldi algorithm, which enables the efficient computation of multiple eigenpairs near the target frequency. For efficiency, the time-dependent wave equation can be solved with implicit time-stepping and only about $10$ time-steps per-period are needed, independent of the mesh spacing. When the (definite) implicit time-stepping equations are solved with a multigrid algorithm, the cost of the resulting EigenWave scheme scales linearly with the number of grid points $N$ as the mesh is refined, giving an optimal $O(N)$ algorithm. The approach is demonstrated by finding eigenpairs of the Laplacian in complex geometry using overset grids. Results in two and three space dimensions are presented using second-order and fourth-order accurate approximations.