A machine-learned expression for the excess Gibbs energy

2025-09-08

AI Summary

Predicting excess Gibbs energy (G^E) of multicomponent mixtures solely from molecular structures has long suffered from poor thermodynamic consistency and limited extrapolation capability. To address this, we propose HANNA: an end-to-end trainable neural network that intrinsically enforces fundamental thermodynamic laws—particularly the Gibbs–Duhem equation—as hard architectural constraints, and employs geometric projection for parameter-free extrapolation to arbitrary component numbers. Furthermore, we introduce a novel surrogate solver integrating liquid–liquid equilibrium (LLE) data to enrich training information beyond conventional vapor–liquid equilibrium (VLE) measurements. Trained exclusively on binary experimental data, HANNA outperforms state-of-the-art methods in prediction accuracy, thermodynamic consistency, and generalization across mixture composition and component count. The model, source code, and an interactive MLPROP platform are publicly released to accelerate phase-equilibrium prediction and molecular design in chemical engineering.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

The excess Gibbs energy plays a central role in chemical engineering and chemistry, providing a basis for modeling the thermodynamic properties of liquid mixtures. Predicting the excess Gibbs energy of multi-component mixtures solely from the molecular structures of their components is a long-standing challenge. In this work, we address this challenge by integrating physical laws as hard constraints within a flexible neural network. The resulting model, HANNA, was trained end-to-end on an extensive experimental dataset for binary mixtures from the Dortmund Data Bank, guaranteeing thermodynamically consistent predictions. A novel surrogate solver developed in this work enabled the inclusion of liquid-liquid equilibrium data in the training process. Furthermore, a geometric projection method was applied to enable robust extrapolations to multi-component mixtures, without requiring additional parameters. We demonstrate that HANNA delivers excellent predictions, clearly outperforming state-of-the-art benchmark methods in accuracy and scope. The trained model and corresponding code are openly available, and an interactive interface is provided on our website, MLPROP.

Problem

• Predicting excess Gibbs energy from molecular structures
• Ensuring thermodynamic consistency in neural network predictions
• Extrapolating binary mixture models to multi-component systems

Innovation

• Neural network with physical law constraints
• Surrogate solver for equilibrium data training
• Geometric projection for multi-component extrapolation