Extrapolating Phase-Field Simulations in Space and Time with Purely Convolutional Architectures

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Bonneville, Christophe, Bieberdorf, Nathan, Robbe, Pieterjan, Asta, Mark, Najm, Habib N., Capolungo, Laurent, Safta, Cosmin
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914055989493760
author Bonneville, Christophe
Bieberdorf, Nathan
Robbe, Pieterjan
Asta, Mark
Najm, Habib N.
Capolungo, Laurent
Safta, Cosmin
author_facet Bonneville, Christophe
Bieberdorf, Nathan
Robbe, Pieterjan
Asta, Mark
Najm, Habib N.
Capolungo, Laurent
Safta, Cosmin
contents Phase-field models of liquid metal dealloying (LMD) can resolve rich microstructural dynamics but become intractable for large domains or long time horizons. We present a conditionally parameterized, fully convolutional U-Net surrogate that generalizes far beyond its training window in both space and time. The design integrates convolutional self-attention and physics-aware padding, while parameter conditioning enables variable time-step skipping and adaptation to diverse alloy systems. Although trained only on short, small-scale simulations, the surrogate exploits the translational invariance of convolutions to extend predictions to much longer horizons than traditional solvers. It accurately reproduces key LMD physics, with relative errors typically under 5% within the training regime and below 10% when extrapolating to larger domains and later times. The method accelerates computations by up to 16,000 times, cutting weeks of simulation down to seconds, and marks an early step toward scalable, high-fidelity extrapolation of LMD phase-field models.
format Preprint
id arxiv_https___arxiv_org_abs_2509_20770
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Extrapolating Phase-Field Simulations in Space and Time with Purely Convolutional Architectures
Bonneville, Christophe
Bieberdorf, Nathan
Robbe, Pieterjan
Asta, Mark
Najm, Habib N.
Capolungo, Laurent
Safta, Cosmin
Computational Engineering, Finance, and Science
Computer Vision and Pattern Recognition
Machine Learning
Numerical Analysis
Phase-field models of liquid metal dealloying (LMD) can resolve rich microstructural dynamics but become intractable for large domains or long time horizons. We present a conditionally parameterized, fully convolutional U-Net surrogate that generalizes far beyond its training window in both space and time. The design integrates convolutional self-attention and physics-aware padding, while parameter conditioning enables variable time-step skipping and adaptation to diverse alloy systems. Although trained only on short, small-scale simulations, the surrogate exploits the translational invariance of convolutions to extend predictions to much longer horizons than traditional solvers. It accurately reproduces key LMD physics, with relative errors typically under 5% within the training regime and below 10% when extrapolating to larger domains and later times. The method accelerates computations by up to 16,000 times, cutting weeks of simulation down to seconds, and marks an early step toward scalable, high-fidelity extrapolation of LMD phase-field models.
title Extrapolating Phase-Field Simulations in Space and Time with Purely Convolutional Architectures
topic Computational Engineering, Finance, and Science
Computer Vision and Pattern Recognition
Machine Learning
Numerical Analysis
url https://arxiv.org/abs/2509.20770