aiqbench analysis
This research introduces DISCOMAX, a differentiable algorithm that enforces thermodynamic consistency in neural networks for predicting phase equilibria. By integrating discrete enumeration with masked softmax aggregation, the method enables end-to-end learning of Gibbs energy models while outperforming traditional surrogate approaches on liquid-liquid equilibrium data. The work advances physics-informed machine learning by bridging statistical mechanics and deep learning architectures to solve complex chemical engineering problems.
Differentiable Thermodynamic Phase-Equilibria for Machine Learning
arXiv:2603.11249v4 Announce Type: replace 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 combines discrete enumeration of feasible phase states with masked softmax aggregation in the backward pass, with the propagation of the true equilibrium state in the forward pass, using a straight-through gradient estimator to enable physics-consistent end-to-end learning of neural \gls{gE}-models. We show that this approach bears analogy to statistical thermodynamics, and we evaluate it on binary liquid-liquid equilibrium data where it outperforms existing surrogate-based methods, while offering a general framework for learning from different kinds of equilibrium data.
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 combines discrete enumeration of feasible phase states with masked softmax aggregation in the backward pass, with the propagation of the true equilibrium state in the forward pass, using a straight-through gradient estimator to enable physics-consistent end-to-end learning of neural \gls{gE}-models. We show that this approach bears analogy to statistical thermodynamics, and we evaluate it on binary liquid-liquid equilibrium data where it outperforms existing surrogate-based methods, while offering a general framework for learning from different kinds of equilibrium data.
Machine Learning (cs.LG)
arXiv:2603.11249 [cs.LG]
(or arXiv:2603.11249v4 [cs.LG] for this version)
https://doi.org/10.1021/acs.jcim.6c00728 DOI(s) linking to related resources
From: Karim K. Ben Hicham [view email] [v1] Wed, 11 Mar 2026 19:14:04 UTC (3,563 KB) [v2] Mon, 16 Mar 2026 13:34:37 UTC (3,563 KB) [v3] Thu, 11 Jun 2026 16:31:43 UTC (12,632 KB) [v4] Mon, 17 Aug 2026 10:54:06 UTC (12,969 KB)
Full text under the publisher's open licence · 214 words
Related coverage
The Geometry of Semantic Space: A Continuous Geometric Framework for the Transformer Architecture
arXiv · Jul 21, 2026
Synthesis Superintelligence: from Semiconductors to Superconductors — Periodic Labs’ Liam Fedus and Ekin Dogus Cubuk
Latent Space · Oct 8, 2026
Universal Thermodynamic Interatomic Potentials for Crystalline Materials
arXiv · Aug 17, 2026
News
Quantum Computing Report · Jun 7, 2026