{"id":"arxiv/2604.02840","url":"https://arxiv.org/abs/2604.02840","tags":["math.NA"],"score":0,"title":"Iterative Refinement for Diagonalizable Non-Hermitian Eigendecompositions","domain":"math.NA","source":"arxiv","authors":["Takeshi Terao"],"reviews":0,"subject":"math.NA","summary":"This paper develops matrix-multiplication-based iterative refinement for diagonalizable non-Hermitian eigendecompositions. The main theory concerns simple eigenvalues and distinguishes two input regimes. In the right-only regime, where only approximate right eigenvectors and eigenvalues are available, a first-order derivation selects the update and the resulting post-update residual identity is exact, yielding a quadratic residual bound. In the left-right regime, where approximate left and right eigenvectors are both available, the computable driving matrix is an exact perturbation of the…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-04-03","reactions":{}}