Fine-Tuning ML Potentials Corrects Phase Diagrams
Researchers propose a top-down fine-tuning strategy for foundational machine-learning potentials using Differentiable Trajectory Reweighting to directly correct phase transition temperatures to experimental values. Demonstrated on pure titanium, the method matches experimental phase diagrams within tenths of kelvins up to 5 GPa and improves liquid-state diffusion; it is model-agnostic and applicable to multi-component systems and other experimental properties.
Key Points
- 1Introduces top-down fine-tuning for ML potentials using Differentiable Trajectory Reweighting.
- 2Corrects phase-transition temperature biases, achieving experimental agreement within tenths of kelvins.
- 3Enables accurate phase diagrams and improved diffusion predictions for materials at pressures up to 5 GPa.
Scoring Rationale
Novel, practical method with strong material results; limited scope and single-study credibility pending peer review.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
