{"id":"arxiv/2607.00237","url":"https://arxiv.org/abs/2607.00237","tags":["math.AP","math.AC"],"score":0,"title":"Tropical Geometry as a Restricted Architecture for Physics-Informed Neural Networks: Applications in Nonlinear Fluid-Structure Examples","domain":"math.AP","source":"arxiv","authors":["Carla Valencia-Negrete","Cristhian Garay-Lopez","Marco Favela-Rodriguez","Alonso Andapia-Viveros"],"reviews":0,"subject":"math.AP","summary":"Nonlinear algebraic (polynomial) differential equations that govern fluid-structure interactions, such as those modeling vortex-induced vibrations, and shock waves, often lack analytical solutions, creating significant challenges to efficient prediction and control. While Physics-Informed Neural Networks (PINNs) offer a mesh-free numerical alternative, they frequently suffer from convergence stagnation when optimizing over chaotic landscapes or stiff singularities. This paper introduces a hybrid methodology that integrates tropical differential algebraic geometry with deep learning. Using…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-06-30","reactions":{}}