Numerical Linear Algebra with Applications Applied Mathematics

A Robust Chunked Block Lanczos Method With Adaptive Shift Selection for Large‐Scale Generalized Eigenvalue Problems in Structural Dynamics

Anatoly Vershinin
Wiley · 2026-10-01 · DOI 10.1002/nla.70125

Abstract

ABSTRACT The computation of many low‐frequency modes of large finite element models is constrained by memory growth, clustered spectra, and rigid‐body modes. A block Lanczos method with bounded active Krylov storage is presented for symmetric generalized eigenvalue problems in structural dynamics. The method advances through automatically selected spectral intervals while preserving numerical clusters at internal boundaries and reusing a fixed active workspace. Adaptive shifts and inertia‐based eigenvalue counts provide consistency checks for spectral continuation, while a calibrated parameter‐selection procedure balances computational work and available memory. Numerical experiments assess the accuracy and robustness of the computed modes, examine parameter sensitivity, and compare the tested complete solution workflows with established eigensolvers. The results support the use of this approach for large sparse structural‐dynamics problems in which direct factorization can be amortized over many eigenpairs.

Read at publisher →