Machine learning based data-driven discovery of nonlinear phase-field dynamics

TL;DR

This paper introduces machine learning architectures to discover nonlinear phase-field equations from data, successfully modeling complex dynamics like Allen--Cahn and Cahn--Hilliard, even without physical assumptions.

Contribution

It presents novel ML-based methods for data-driven discovery of PDEs governing phase-field models, including approaches with and without physical guidance.

Findings

Effective learning of time derivatives from data

Accurate time propagation using learned PDEs

Models work for both guided and black-box scenarios

Abstract

One of the main questions regarding complex systems at large scales concerns the effective interactions and driving forces that emerge from the detailed microscopic properties. Coarse-grained models aim to describe complex systems in terms of coarse-scale equations with a reduced number of degrees of freedom. Recent developments in machine learning (ML) algorithms have significantly empowered the discovery process of the governing equations directly from data. However, it remains difficult to discover partial differential equations (PDEs) with high-order derivatives. In this paper, we present new data-driven architectures based on multi-layer perceptron (MLP), convolutional neural network (CNN), and a combination of CNN and long short-term memory (CNN-LSTM) structures for discovering the non-linear equations of motion for phase-field models with non-conserved and conserved order parameters. The well-known Allen--Cahn, Cahn--Hilliard, and the phase-field crystal (PFC) models were used as the test cases. Two conceptually different types of implementations were used: (a) guided by physical intuition (such as local dependence of the derivatives) and (b) in the absence of any physical assumptions (black-box model). We show that not only can we effectively learn the time derivatives of the field in both scenarios, but we can also use the data-driven PDEs to propagate the field in time and achieve results in good agreement with the original PDEs.