Cross-functional transferability in universal machine learning interatomic potentials

2025/04/07 by Xu Huang, Bowen Deng, Huang, Xu +9 · 3 citations
Computer Science · Materials Science · Physics and Astronomy · #Advanced Graph Neural Networks #Benchmarking #Bridging (networking) #Energy (signal processing) #Energy transfer #FOS: Computer and information sciences #FOS: Physical sciences #Machine Learning (cs.LG) #Machine Learning in Materials Science #Materials Science (cond-mat.mtrl-sci) #Online machine learning #Quantum many-body systems #Scaling #Transfer of learning #Transferability

paper · pdf · doi:10.48550/arxiv.2504.05565

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2025/04/07 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05

Abstract

The rapid development of universal machine learning interatomic potentials (uMLIPs) has demonstrated the possibility for generalizable learning of the universal potential energy surface. In principle, the accuracy of uMLIPs can be further improved by bridging the model from lower-fidelity datasets to high-fidelity ones. In this work, we analyze the challenge of this transfer learning problem within the CHGNet framework. We show that significant energy scale shifts and poor correlations between GGA and r²SCAN pose challenges to cross-functional data transferability in uMLIPs. By benchmarking different transfer learning approaches on the MP-r²SCAN dataset of 0.24 million structures, we demonstrate the importance of elemental energy referencing in the transfer learning of uMLIPs. By comparing the scaling law with and without the pre-training on a low-fidelity dataset, we show that significant data efficiency can still be achieved through transfer learning, even with a target dataset of sub-million structures. We highlight the importance of proper transfer learning and multi-fidelity learning in creating next-generation uMLIPs on high-fidelity data.

Citations

• Learning charges and long-range interactions from energies and forces
• Open Catalyst Experiments 2024 (OCx24): Bridging Experiments and Computational Models
• Orb: A Fast, Scalable Neural Network Potential
• Data-efficient multi-fidelity training for high-fidelity machine\n learning interatomic potentials
• Data-Efficient Construction of High-Fidelity Graph Deep Learning Interatomic Potentials
• Discovering High-Entropy Oxides with a Machine-Learning Interatomic Potential
• Adaptive energy reference for machine-learning models of the electronic density of states
• MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures
• Systematic assessment of various universal machine-learning interatomic potentials
• Cartesian atomic cluster expansion for machine learning interatomic potentials
• A foundation model for atomistic materials chemistry
• Scaling deep learning for materials discovery
• Assessing r2SCAN meta-GGA functional for structural parameters, cohesive energy, mechanical modulus and thermophysical properties of 3d, 4d and 5d transition metals
• Learning Together: Towards foundational models for machine learning interatomic potentials with meta-learning
• Transfer learning on large datasets for the accurate prediction of material properties
• Testing the r²SCAN density functional for the thermodynamic stability of solids with and without a van der Waals correction
• A Universal Graph Deep Learning Interatomic Potential for the Periodic Table
• E(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials
• Learning properties of ordered and disordered materials from multi-fidelity data
• The Open Catalyst 2020 (OC20) Dataset and Community Challenges
• The Joint Automated Repository for Various Integrated Simulations (JARVIS) for data-driven materials design
• A critical examination of compound stability predictions from machine-learned formation energies
• Uncovering vacuum level in infinite solid by real-space potential-unfolding
• Performance of the strongly constrained and appropriately normed density functional for solid-state materials
• Deep Potential Molecular Dynamics: a scalable model with the accuracy of quantum mechanics
• Strongly Constrained and Appropriately Normed Semilocal Density Functional
• Density Functionals that Recognize Covalent, Metallic, and Weak Bonds
• Python Materials Genomics (pymatgen): A robust, open-source python library for materials analysis
• AFLOW: An automatic framework for high-throughput materials discovery
• First-principles prediction of redox potentials in transition-metal compounds withLDA+U
• A linear response approach to the calculation of the effective interaction parameters in the LDA+U method
• Real-Space Mesh Techniques in Density Functional Theory
• Generalized Gradient Approximation Made Simple
• Self-interaction correction to density-functional approximations for many-electron systems
• Momentum-space formalism for the total energy of solids
• Matbench Discovery -- A framework to evaluate machine learning crystal stability predictions
• Induction of decision trees

Cited by

• Predicting Crystal Structures and Ionic Conductivities in Li₃YCl_6-xBr_x Halide Solid Electrolytes Using a Fine-Tuned Machine Learning Interatomic Potential
• LeMat-Traj: A Scalable and Unified Dataset of Materials Trajectories for Atomistic Modeling
• Understanding multi-fidelity training of machine-learned force-fields

Related

• Optimizing Cross-Domain Transfer for Universal Machine Learning Interatomic Potentials
• Comparing the latent features of universal machine-learning interatomic potentials
• Data-efficient multi-fidelity training for high-fidelity machine\n learning interatomic potentials
• Global Universal Scaling and Ultra-Small Parameterization in Machine Learning Interatomic Potentials with Super-Linearity
• Training Data Selection for Accuracy and Transferability of Interatomic Potentials