{"id":"arxiv/2601.04510","url":"https://arxiv.org/abs/2601.04510","tags":["cs.CE","cs.AI","cs.CV","cs.LG","math.NA"],"score":0,"title":"Towards Spatio-Temporal Extrapolation of Phase-Field Simulations with Convolution-Only Neural Networks","domain":"cs.CE","source":"arxiv","authors":["Christophe Bonneville","Nathan Bieberdorf","Pieterjan Robbe","Mark Asta","Habib Najm","Laurent Capolungo","Cosmin Safta"],"reviews":0,"subject":"cs.CE","summary":"Phase-field simulations of liquid metal dealloying (LMD) can capture complex microstructural evolutions but can be prohibitively expensive for large domains and long time horizons. In this paper, we introduce a fully convolutional, conditionally parameterized U-Net surrogate designed to extrapolate far beyond its training data in both space and time. The architecture integrates convolutional self-attention, physically informed padding, and a flood-fill corrector method to maintain accuracy under extreme extrapolation, while conditioning on simulation parameters allows for flexible time-step…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-01-08","reactions":{}}