Differentiable Thermodynamic Phase-Equilibria for Machine Learning
This work addresses a central problem in chemical engineering for accurate phase-equilibria prediction, offering a novel method for a known bottleneck in extending physics-consistent machine learning to equilibrium data.
The paper tackles the challenge of predicting phase equilibria in chemical engineering by introducing DISCOMAX, a differentiable algorithm that ensures thermodynamic consistency during training and inference, and demonstrates its superior performance over existing surrogate-based methods on binary liquid-liquid equilibrium data.
Accurate prediction of phase equilibria remains a central challenge in chemical engineering. Physics-consistent machine learning methods that incorporate thermodynamic structure into neural networks have recently shown strong performance for activity-coefficient modeling. However, extending such approaches to equilibrium data arising from an extremum principle, such as liquid-liquid equilibria, remains difficult. Here we present DISCOMAX, a differentiable algorithm for phase-equilibrium calculation that guarantees thermodynamic consistency at both training and inference, only subject to a user-specified discretization. The method is rooted in statistical thermodynamics, and works via a discrete enumeration with subsequent masked softmax aggregation of feasible states, and together with a straight-through gradient estimator to enable physics-consistent end-to-end learning of neural $g^{E}$-models. We evaluate the approach on binary liquid-liquid equilibrium data and demonstrate that it outperforms existing surrogate-based methods, while offering a general framework for learning from different kinds of equilibrium data.