Optimizing Cross-Domain Transfer for Universal Machine Learning Interatomic Potentials

2025/10/13 by Jaesun Kim, Kim, Jaesun, You, Jinmu +22 · 3 citations
Materials Science · Computer Science · Chemistry · #Machine Learning in Materials Science #Computational Drug Discovery Methods #Inorganic Chemistry and Materials

paper · pdf · doi:10.48550/arxiv.2510.11241

Abstract

Accurate yet transferable machine-learning interatomic potentials (MLIPs) are essential for accelerating materials and chemical discovery. However, most universal MLIPs overfit to narrow datasets or computational protocols, limiting their reliability across chemical and functional domains. We introduce a transferable multi-domain training strategy that jointly optimizes universal and task-specific parameters through selective regularization, coupled with a domain-bridging set (DBS) that aligns potential-energy surfaces across datasets. Systematic ablation experiments show that small DBS fractions (0.1%) and targeted regularization synergistically enhance out-of-distribution generalization while preserving in-domain fidelity. Trained on fifteen open databases spanning molecules, crystals, and surfaces, our model, SevenNet-Omni, achieves state-of-the-art cross-domain accuracy, including adsorption-energy errors below 0.06 eV on metallic surfaces and 0.1 eV on metal-organic frameworks. Despite containing only 0.5% r²SCAN data, SevenNet-Omni reproduces high-fidelity r²SCAN energetics, demonstrating effective cross-functional transfer from large PBE datasets. This framework offers a scalable route toward universal, transferable MLIPs that bridge quantum-mechanical fidelities and chemical domains.

Citations

• A Computational Study for Screening High-Selectivity Inhibitors in Area-Selective Atomic Layer Deposition on Amorphous Surfaces
• Reaction dynamics of lithium-mediated electrolyte decomposition using machine learning potentials
• Graph atomic cluster expansion for foundational machine learning interatomic potentials
• Universal Machine Learning Potentials under Pressure
• Universal Machine Learning Potential for Systems with Reduced Dimensionality
• The Open DAC 2025 Dataset for Sorbent Discovery in Direct Air Capture
• Open Molecular Crystals 2025 (OMC25) Dataset and Models
• MOFSimBench: Evaluating Universal Machine Learning Interatomic Potentials In Metal--Organic Framework Molecular Modeling
• MP-ALOE: An r2SCAN dataset for universal machine learning interatomic potentials
• UMA: A Family of Universal Models for Atoms
• Massive Atomic Diversity: a compact universal dataset for atomistic machine learning
• A Graph Neural Network for the Era of Large Atomistic Models
• The Open Molecules 2025 (OMol25) Dataset, Evaluations, and Models
• NEP89: Universal neuroevolution potential for inorganic and organic materials across 89 elements
• High-performance training and inference for deep equivariant interatomic potentials
• Orb-v3: atomistic simulation at scale
• A Foundational Potential Energy Surface Dataset for Materials
• Learning Smooth and Expressive Interatomic Potentials for Physical Property Prediction
• An Efficient Sparse Kernel Generator for O(3)-Equivariant Deep Networks
• Application of pretrained universal machine-learning interatomic potential for physicochemical simulation of liquid electrolytes in Li-ion battery
• Taming Multi-Domain, -Fidelity Data: Towards Foundation Models for Atomistic Scale Simulations
• Orb: A Fast, Scalable Neural Network Potential
• Data-efficient multi-fidelity training for high-fidelity machine learning interatomic potentials
• Data-Efficient Construction of High-Fidelity Graph Deep Learning Interatomic Potentials
• MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures
• Hierarchy of Exchange-Correlation Functionals in Computing Lattice Thermal Conductivities of Rocksalt and Zincblende Semiconductors
• A foundation model for atomistic materials chemistry
• MACE-OFF: Transferable Short Range Machine Learning Force Fields for Organic Molecules
• Scaling deep learning for materials discovery
• The Open DAC 2023 Dataset and Challenges for Sorbent Discovery in Direct Air Capture
• First-principles Phonon Calculations with Phonopy and Phono3py
• Symmetry-based computational search for novel binary and ternary 2D materials
• SPICE, A Dataset of Drug-like Molecules and Peptides for Training Machine Learning Potentials
• Testing the r²SCAN density functional for the thermodynamic stability of solids with and without a van der Waals correction
• e3nn: Euclidean Neural Networks
• MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields
• The Design Space of E(3)-Equivariant Atom-Centered Interatomic Potentials
• A universal graph deep learning interatomic potential for the periodic table
• LAMMPS - a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales
• E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials
• Accurate and numerically efficient r²SCAN meta-generalized gradient approximation
• Machine Learning in Python: Main developments and technology trends in data science, machine learning, and artificial intelligence
• Grain Boundary Properties of Elemental Metals
• The atomic simulation environment—a Python library for working with atoms
• <i>ω</i>B97M-V: A combinatorially optimized, range-separated hybrid, meta-GGA density functional with VV10 nonlocal correlation
• Effect of the damping function in dispersion corrected density functional theory
• Improved adsorption energetics within density-functional theory using revised Perdew-Burke-Ernzerhof functionals
• From ultrasoft pseudopotentials to the projector augmented-wave method

Cited by

• Catalyst Diffusion Transformer: Generative Inverse Design of Heterogeneous Catalysts
• Breaking the Training Barrier of Billion-Parameter Universal Machine Learning Interatomic Potentials
• Fast and Accurate Foundation Models for Equivariant Machine-Learned Interatomic Potentials

Related

• Taming Multi-Domain, -Fidelity Data: Towards Foundation Models for Atomistic Scale Simulations
• Cross-functional transferability in universal machine learning interatomic potentials
• Cross Learning between Electronic Structure Theories for Unifying Molecular, Surface, and Inorganic Crystal Foundation Force Fields
• Data-efficient multi-fidelity training for high-fidelity machine learning interatomic potentials
• Generalization of Long-Range Machine Learning Potentials in Complex Chemical Spaces