Accelerating Multicomponent Phase-Coexistence Calculations with Physics-informed Neural Networks

Michael Webb,Shengli Jiang,Satyen Dhamankar

DOI: https://doi.org/10.26434/chemrxiv-2024-rwrd1

2024-09-23

Abstract:Phase separation in multicomponent mixtures is of significant interest in both fundamental research and technology. Although the thermodynamic principles governing phase equilibria are straightforward, practical determination of equilibrium phases and constituent compositions for multicomponent systems is often laborious and computationally intensive. Here, we present a machine-learning workflow that simplifies and accelerates phase-coexistence calculations. We specifically analyze capabilities of neural networks to predict the number, composition, and relative abundance of equilibrium phases of systems described by Flory-Huggins theory. We find that incorporating physics-informed material constraints into the neural network architecture enhances the prediction of equilibrium compositions compared to standard neural networks with minor errors along the boundaries of the stable region. However, introducing additional physics-informed losses does not lead to significant further improvement. These errors can be virtually eliminated by using machine-learning predictions as a warm-start for a subsequent optimization routine. This work provides a promising pathway to efficiently characterize multicomponent phase coexistence.

Chemistry

What problem does this paper attempt to address?

The paper aims to address the efficiency and accuracy issues in the computation of phase coexistence in multicomponent systems. Specifically: - **Background Problem**: In multicomponent mixtures, phase separation phenomena are of significant importance in both fundamental research and technological applications. Although the thermodynamic principles describing phase equilibrium are relatively simple, determining the equilibrium phases and their compositions in multicomponent systems in practice is often very time-consuming and computationally intensive. - **Specific Objectives**: The paper proposes a machine learning-based workflow to simplify and accelerate the computation of phase coexistence in multicomponent systems. In particular, it utilizes neural networks to predict the number of phases, compositions, and relative abundances of systems described by the Flory-Huggins theory at equilibrium. By incorporating physical information constraints into the neural network architecture, the accuracy of predicting equilibrium compositions is improved, especially with smaller errors at the boundaries of stable regions. - **Methodological Contributions**: The study compares the performance of different neural network architectures (including models with physical information loss functions) and evaluates the errors in classification (number of phases) and regression (phase composition and abundance) tasks. Additionally, the study suggests that machine learning prediction results can be used as initial guesses for subsequent optimization processes to further enhance accuracy. In summary, the goal of this paper is to provide an efficient, accurate, and scalable method to characterize multicomponent phase behavior.

Accelerating Multicomponent Phase-Coexistence Calculations with Physics-informed Neural Networks

Phase behavior and morphology of multicomponent liquid mixtures

Phase Separation in Multicomponent Liquid Mixtures

Models for predicting multi-component phase equilibrium ratio based on artificial neural networks

Physics-informed neural networks for phase-field method in two-phase flow

A Multiscale Method for Two-Component, Two-Phase Flow with a Neural Network Surrogate

A Data-Driven Phase Change Model for Injection Flow Modeling

Fast Prediction of the Quasi Phase Equilibrium in Phase Field Model for Multicomponent Alloys Based on Machine Learning Method

Gradient Flow Based Phase-Field Modeling Using Separable Neural Networks

$\textit{Ab initio}$ construction of full phase diagram of MgO-CaO eutectic system using neural network interatomic potentials

Neural Network Representations of Multiphase Equations of State

PINNs-MPF: A Physics-Informed Neural Network Framework for Multi-Phase-Field Simulation of Interface Dynamics

LeapFrog: Getting the Jump on Multi-Scale Materials Simulations Using Machine Learning

Multi-Fidelity Physics-Constrained Neural Network and Its Application in Materials Modeling

Inference of Constitutive Relation of Phase‐Separated Polymers by Integrating Physics‐Informed Neural Networks and Symbolic Regression

Accelerated Solutions of Coupled Phase-Field Problems using Generative Adversarial Networks

Theory of Nonequilibrium Multicomponent Coexistence

Phase separation in fluids with many interacting components

Neural density functional theory of liquid-gas phase coexistence

Accelerating CALPHAD-based Phase Diagram Predictions in Complex Alloys Using Universal Machine Learning Potentials: Opportunities and Challenges

Multiphase flow prediction with deep neural networks