{"id":"arxiv/2610.38790","url":"https://arxiv.org/abs/2610.38790","tags":["math.OC","math.NA"],"score":0,"title":"Safeguarded Neurally-Accelerated Preconditioning (SNAP) framework for Nonlinear Optimization","domain":"math.OC","source":"arxiv","authors":["Youngkyu Lee","George Em Karniadakis","Alena Kopaničáková"],"reviews":0,"subject":"math.OC","summary":"We introduce the Safeguarded Neurally-Accelerated Preconditioning (SNAP) framework for hybridizing nonlinear optimization methods with pre-trained neural operators (NOs). SNAP targets parametric optimization problems arising from variational formulations of nonlinear partial differential equations (PDEs), where many parameter instances require repeated solves. Inspired by multilevel line-search methods, SNAP operates in cycles, combining classical optimization steps with a search direction generated by a pre-trained NO. Since NO-generated search directions carry no a priori descent…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-10-28","reactions":{}}