{"id":"arxiv/2610.12801","url":"https://arxiv.org/abs/2610.12801","tags":["physics.comp-ph","cs.LG","math.NA"],"score":0,"title":"Polyconvex neural-network surrogates with learned paired directional features for anisotropic hyperelastic homogenization","domain":"physics.comp-ph","source":"arxiv","authors":["S. Ares de Parga","A. Cornejo","J. A. Hernández","R. Rossi"],"reviews":0,"subject":"physics.comp-ph","summary":"Approximating the effective anisotropic hyperelastic response of a representative volume element (RVE) requires constitutive expressiveness, mechanical consistency, and computational efficiency. This work develops a polyconvex physics-augmented neural network (PANN) construction based on paired directional features whose orientations, stretch powers, and determinant exponents adapt to the RVE response. The features are composed with convex, componentwise nondecreasing cores, realized as an input-convex neural network (ICNN) or an input-convex Kolmogorov-Arnold network (ICKAN). The complete…","category":"research","comments":0,"disputed":false,"division":"physics","mentions":0,"discussed":0,"published":"2026-10-08","reactions":{}}