Id: arxiv/2606.21828 Url: https://arxiv.org/abs/2606.21828 Tags: math.NA, cs.LG Score: 0 Title: Spectrally Safe Neural Operator Warm-Starts for Large-Scale Newton Solvers Domain: math.NA Source: arxiv Authors: Jaemin Oh, Youngkyu Lee, Jerome Darbon, George Em Karniadakis Reviews: 0 Subject: math.NA Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-06-20 Page: https://delphi.science/r/arxiv/2606.21828 Neural operators are increasingly used to warm-start Newton solvers for nonlinear PDEs, on the premise that a low test error places the initial guess inside the basin of attraction. We show that this premise is unreliable. An operator trained to the relative \(L^2\) error \(O(10^{-3})\) can still produce an initial state in which the discrete Jacobian is indefinite, because the mean-squared training controls error on average while leaving localized pointwise violations of the underlying physics. For a nearly incompressible hyperelasticity problem, we trace this to the predicted volume…