Refining Machine Learning Potentials through Thermodynamic Theory of Phase Transitions

2025-12-03

AI Summary

Machine learning interatomic potentials (MLIPs) exhibit systematic biases of several hundred kelvins in predicting phase transition temperatures, severely undermining their reliability in thermodynamic simulations. To address this, we propose a top-down fine-tuning framework grounded in thermodynamic phase transition theory: leveraging differentiable trajectory reweighting within molecular dynamics simulations, we directly optimize the free energy difference to enable end-to-end calibration against experimental phase diagrams. This approach requires no modification or reconstruction of the underlying potential function and is applicable to multicomponent systems and diverse transitions—including solid–solid and solid–liquid. For pure titanium across pressures up to 5 GPa, the predicted phase transition temperatures achieve an error of less than ±0.1 K relative to experiment, while liquid-phase diffusion coefficients are also significantly improved. This work represents the first demonstration of MLIP-based phase transition temperature predictions reaching experimental uncertainty levels, establishing a generalizable calibration paradigm for high-accuracy materials thermodynamic modeling.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: ML for personalized search and recommendations
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Foundational Machine Learning Potentials can resolve the accuracy and transferability limitations of classical force fields. They enable microscopic insights into material behavior through Molecular Dynamics simulations, which can crucially expedite material design and discovery. However, insufficiently broad and systematically biased reference data affect the predictive quality of the learned models. Often, these models exhibit significant deviations from experimentally observed phase transition temperatures, in the order of several hundred kelvins. Thus, fine-tuning is necessary to achieve adequate accuracy in many practical problems. This work proposes a fine-tuning strategy via top-down learning, directly correcting the wrongly predicted transition temperatures to match the experimental reference data. Our approach leverages the Differentiable Trajectory Reweighting algorithm to minimize the free energy differences between phases at the experimental target pressures and temperatures. We demonstrate that our approach can accurately correct the phase diagram of pure Titanium in a pressure range of up to 5 GPa, matching the experimental reference within tenths of kelvins and improving the liquid-state diffusion constant. Our approach is model-agnostic, applicable to multi-component systems with solid-solid and solid-liquid transitions, and compliant with top-down training on other experimental properties. Therefore, our approach can serve as an essential step towards highly accurate application-specific and foundational machine learning potentials.

Problem

• Refine machine learning potentials to correct phase transition temperature deviations
• Improve predictive accuracy for material phase diagrams using experimental data
• Enable model-agnostic fine-tuning for solid-solid and solid-liquid transitions

Innovation

• Fine-tuning via top-down learning corrects predicted transition temperatures
• Uses Differentiable Trajectory Reweighting to minimize free energy differences
• Model-agnostic approach applicable to multi-component phase transitions