English

Block subspace expansions for eigenvalues and eigenvectors approximation

Numerical Analysis 2024-12-17 v1 Numerical Analysis Functional Analysis

Abstract

Let ACn×nA\in\mathbb C^{n\times n} and let XCn\mathcal X\subset \mathbb C^n be an AA-invariant subspace with dimX=d1\dim \mathcal X=d\geq 1, corresponding to exterior eigenvalues of AA. Given an initial subspace VCn\mathcal V\subset \mathbb C^n with dimV=rd\dim \mathcal V=r\geq d, we search for expansions of V\mathcal V of the form V+A(W0)\mathcal V+A(\mathcal W_0), where W0V\mathcal W_0\subset \mathcal V is such that dimW0d\dim \mathcal W_0\leq d and such that the expanded subspace is closer to X\mathcal X than the initial V\mathcal V. We show that there exist (theoretical) optimal choices of such W0\mathcal W_0, in the sense that θi(X,V+A(W0))θi(V+A(W))\theta_i(\mathcal X,\mathcal V+A(\mathcal W_0))\leq \theta_i(\mathcal V+A(\mathcal W)) for every WV\mathcal W\subset \mathcal V with dimWd\dim \mathcal W\leq d, where θi(X,T)\theta_i(\mathcal X,\mathcal T) denotes the ii-th principal angle between X\mathcal X and T\mathcal T, for 1iddimT1\leq i\leq d\leq \dim \mathcal T. We relate these optimal expansions to block Krylov subspaces generated by AA and V\mathcal V. We also show that the corresponding iterative sequence of subspaces constructed in this way approximate X\mathcal X arbitrarily well, when AA is Hermitian and X\mathcal X is simple. We further introduce computable versions of this construction and compute several numerical examples that show the performance of the computable algorithms and test our convergence analysis.

Cite

@article{arxiv.2411.14578,
  title  = {Block subspace expansions for eigenvalues and eigenvectors approximation},
  author = {Francisco Arrieta Zuccalli and Pedro Massey and Demetrio Stojanoff},
  journal= {arXiv preprint arXiv:2411.14578},
  year   = {2024}
}
R2 v1 2026-06-28T20:08:27.410Z