{"id":"arxiv/2610.20285","url":"https://arxiv.org/abs/2610.20285","tags":["math.NA"],"score":0,"title":"Scaled Companion Linearizations for Coefficientwise Backward Stability","domain":"math.NA","source":"arxiv","authors":["Javier Pérez"],"reviews":0,"subject":"math.NA","summary":"A central goal in solving a polynomial eigenvalue problem is to compute eigenpairs that are exact for a nearby polynomial, with each coefficient perturbation bounded by a modest multiple of unit roundoff times that coefficient's norm. A backward stable eigensolver for a linearization does not automatically provide this coefficientwise relative guarantee. Scaling can improve accuracy, but a single scaling may not work across a widely spread spectrum. We construct a family of scaled companion linearizations that covers the nonzero complex plane. Each linearization is paired with a region and…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-10-14","reactions":{}}