Structure-preserving Fourier Neural Operators for Long-time Cahn-Hilliard Dynamics under Coarse Temporal Supervision
My rating
This paper addresses long-time prediction of Cahn-Hilliard phase-field dynamics, with particular attention to the computationally demanding late-stage coarsening regime. It introduces a structure-preserving Fourier Neural Operator trained under coarse temporal supervision to learn an efficient surrogate map for advancing the dynamics. The method targets the accumulation of small prediction errors that occurs when a learned map is repeatedly applied over long physical times.
Tags: