{"id":"arxiv/2610.27500","url":"https://arxiv.org/abs/2610.27500","tags":["math.NA","cs.LG","math.AP","nlin.PS"],"score":0,"title":"Soliton Profiles in 1D: Enhanced Neural-Network Solvers and Constrained Deep Ritz Methods","domain":"math.NA","source":"arxiv","authors":["Chandler Haight","Alex D Rodriguez","Svetlana Roudenko"],"reviews":0,"subject":"math.NA","summary":"We develop and compare enhanced neural-network solvers for one-dimensional ground-state profiles, which are solutions of $$ Q^{\\prime\\prime}=bQ-γQ^{p},\\quad Q>0,\\quad p>1, \\quad b, γ>0, $$ the stationary profile equation for the nonlinear Schrödinger, generalized Korteweg-de Vries, and nonlinear Klein-Gordon equations. Its explicit sech-type solutions serve as exact reference for error computation and benchmark hyperparameter selection. We improve on the standard Adam-trained physics-informed neural network (PINN) baseline by introducing higher-order optimization via…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-10-20","reactions":{}}