Differentiable Thermodynamic Phase-Equilibria for Machine Learning
arXiv:2603.11249v1 Announce Type: new Abstract: 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
arXiv:2603.11249v1 Announce Type: new Abstract: 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.
Executive Summary
The article introduces DISCOMAX, a differentiable algorithm for phase-equilibrium calculation that ensures thermodynamic consistency in machine learning models. This approach enables physics-consistent end-to-end learning of neural models for predicting phase equilibria, outperforming existing surrogate-based methods. The algorithm is based on statistical thermodynamics and uses a discrete enumeration with masked softmax aggregation of feasible states. The method demonstrates strong performance on binary liquid-liquid equilibrium data and offers a general framework for learning from various kinds of equilibrium data.
Key Points
- ▸ Introduction of DISCOMAX, a differentiable algorithm for phase-equilibrium calculation
- ▸ Ensures thermodynamic consistency in machine learning models
- ▸ Outperforms existing surrogate-based methods on binary liquid-liquid equilibrium data
Merits
Improved Accuracy
DISCOMAX provides a more accurate prediction of phase equilibria by incorporating thermodynamic structure into neural networks.
Generalizability
The algorithm offers a general framework for learning from different kinds of equilibrium data, making it a versatile tool for various applications.
Demerits
Discretization Limitation
The method is subject to a user-specified discretization, which may introduce limitations in certain scenarios.
Computational Complexity
The algorithm's computational complexity may be higher than existing methods, potentially affecting its scalability.
Expert Commentary
The introduction of DISCOMAX marks a significant advancement in the field of machine learning for phase equilibria prediction. By ensuring thermodynamic consistency, this algorithm addresses a long-standing challenge in chemical engineering. The method's ability to generalize to different types of equilibrium data makes it a valuable tool for various applications. However, further research is needed to fully explore the algorithm's limitations and potential applications. As machine learning continues to play a larger role in chemical engineering, the development of physics-consistent models like DISCOMAX will be crucial for driving innovation and improvement in the field.
Recommendations
- ✓ Further evaluation of DISCOMAX on a wider range of equilibrium data
- ✓ Investigation into potential applications of the algorithm in other fields, such as materials science or biology