A foundation model for atomistic materials chemistry

2023/12/29 by Ilyes Batatia, Batatia, Ilyes, Philipp Benner +182 · 5 voices · 200 citations
Decision Sciences · Materials Science · Physics and Astronomy · #physics.chem-ph #cond-mat.mtrl-sci

paper · pdf · doi:10.48550/arxiv.2401.00096

Abstract

Atomistic simulations of matter, especially those that leverage first-principles (ab initio) electronic structure theory, provide a microscopic view of the world, underpinning much of our understanding of chemistry and materials science. Over the last decade or so, machine-learned force fields have transformed atomistic modeling by enabling simulations of ab initio quality over unprecedented time and length scales. However, early ML force fields have largely been limited by: (i) the substantial computational and human effort of developing and validating potentials for each particular system of interest; and (ii) a general lack of transferability from one chemical system to the next. Here we show that it is possible to create a general-purpose atomistic ML model, trained on a public dataset of moderate size, that is capable of running stable molecular dynamics for a wide range of molecules and materials. We demonstrate the power of the MACE-MP-0 model - and its qualitative and at times quantitative accuracy - on a diverse set of problems in the physical sciences, including properties of solids, liquids, gases, chemical reactions, interfaces and even the dynamics of a small protein. The model can be applied out of the box as a starting or "foundation" model for any atomistic system of interest and, when desired, can be fine-tuned on just a handful of application-specific data points to reach ab initio accuracy. Establishing that a stable force-field model can cover almost all materials changes atomistic modeling in a fundamental way: experienced users get reliable results much faster, and beginners face a lower barrier to entry. Foundation models thus represent a step towards democratising the revolution in atomic-scale modeling that has been brought about by ML force fields.

Citations

• Fast Parallel Algorithms for Short-Range Molecular Dynamics
• Effect of the damping function in dispersion corrected density functional theory
• Vapor–liquid equilibria of mixtures containing alkanes, carbon dioxide, and nitrogen
• Self-Consistent Equations Including Exchange and Correlation Effects
• SchNet – A deep learning architecture for molecules and materials
• Halide Perovskite Photovoltaics: Background, Status, and Future Prospects
• Coordinatively unsaturated metal sites (open metal sites) in metal–organic frameworks: design and applications
• Quantum ESPRESSO toward the exascale
• CP2K: An Electronic Structure and Molecular Dynamics Software Package -- Quickstep: Efficient and Accurate Electronic Structure Calculations
• Rationally designed indium oxide catalysts for CO 2 hydrogenation to methanol with high activity and selectivity
• Accurate and Numerically Efficient r2SCAN Meta-Generalized Gradient Approximation
• Gaussian Process Regression for Materials and Molecules
• Kokkos 3: Programming Model Extensions for the Exascale Era
• LAMMPS - a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales
• Evolution of water structures in metal-organic frameworks for improved atmospheric water harvesting
• Software update: The ORCA program system—Version 5.0
• E(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials
• A universal graph deep learning interatomic potential for the periodic table
• First-principles Phonon Calculations with Phonopy and Phono3py
• Implementation strategies in phonopy and phono3py
• q-pac : A Python package for machine learned charge equilibration models
• Scaling deep learning for materials discovery
• Atomate2: modular workflows for materials science

Cited by

• Sustainable Materials Discovery in the Era of Artificial Intelligence
• Thermal Equilibrium Vacancy Concentration in an Alloy with Chemical Short-Range Order
• Materealize: a multi-agent deliberation system for end-to-end material design and synthesis
• A Combined Tight Binding with Machine Learning Potential Model for Magnesium Compounds
• PhononBench-MP40: a spectrum-resolved benchmark dataset for phonon stability
• Benchmarking Universal Machine Learning Interatomic Potentials on Elemental Systems
• How accurate are foundational machine learning interatomic potentials for heterogeneous catalysis?
• Ultra-Fast Machine-Learned Interatomic Potential for MoS2 Enabling Non-Equilibrium Molecular-Dynamics Simulation of Epitaxial Growth
• Computational tuning of the elastic properties of low- and high-entropy ultra-high temperature ceramics
• A Software Package for Generating Robust and Accurate Potentials using the Moment Tensor Potential Framework
• A probabilistic framework for crystal structure denoising, phase classification, and order parameters
• Transport Novelty Distance: A Distributional Metric for Evaluating Material Generative Models
• Statistical physics for artificial neural networks
• Benchmarking Universal Machine Learning Interatomic Potentials for Supported Nanoparticles: Decoupling Energy Accuracy from Structural Exploration
• Comparing the latent features of universal machine-learning interatomic potentials
• Revealing interstitial energetics in Ti-23Nb-0.7Ta-2Zr gum metal base alloy via universal machine learning interatomic potentials
• Beyond Adam: Disentangling Optimizer Effects in the Fine-Tuning of Atomistic Foundation Models
• Platonic representation of foundation machine learning interatomic potentials
• LeMat-GenBench: A Unified Evaluation Framework for Crystal Generative Models
• Refining Machine Learning Potentials through Thermodynamic Theory of Phase Transitions
• Evaluation of Foundational Machine Learned Interatomic Potentials for Migration Barrier Predictions
• LitMOF: An LLM Multi-Agent for Literature-Validated Metal-Organic Frameworks Database Correction and Expansion
• Evaluating Mechanical Property Prediction across Material Classes using Molecular Dynamics Simulations with Universal Machine-Learned Interatomic Potentials
• Generative models for crystalline materials
• Machine-Learned Interatomic Potentials for Structural and Defect Properties of YBa₂Cu₃O_7-δ
• Accelerated Discovery of Crystalline Materials with Record Ultralow Lattice Thermal Conductivity via a Universal Descriptor
• Bridging Atomistic and Mesoscale Lithium Transport via Machine-Learned Force Fields and Markov State Models
• Signatures of coherent phonon transport in frequency dependent lattice thermal conductivity
• Electrochemical Interfaces at Constant Potential: Data-Efficient Transfer Learning for Machine-Learning-Based Molecular Dynamics
• MOCLIP: A Foundation Model for Large-Scale Nanophotonic Inverse Design
• Predicting the Thermal Behavior of Semiconductor Defects with Equivariant Neural Networks
• Li-P-S Electrolyte Materials as a Benchmark for Machine-Learned Interatomic Potentials
• Origin of metal-insulator transition in rare-earth Nickelates
• Intrinsic structure of relaxor ferroelectrics from first principles
• Maximizing Efficiency of Dataset Compression for Machine Learning Potentials With Information Theory
• Guiding Generative Models to Uncover Diverse and Novel Crystals via Reinforcement Learning
• Magnitude-Modulated Equivariant Adapter for Parameter-Efficient Fine-Tuning of Equivariant Graph Neural Networks
• Fine-Tuning Unifies Foundational Machine-learned Interatomic Potential Architectures at ab initio Accuracy
• Connecting the concepts of quantum state tomography and molecular representations for machine learning
• Stability and Dynamics of Sn-based Halide Perovskites: Insights from MACE-MP-0 and Molecular Dynamics Simulations
• Atomistic Simulations of Oxide-Water Interfaces using Machine Learning Potentials
• Benchmarking Simulacra AI's Quantum Accurate Synthetic Data Generation for Chemical Sciences
• Cross Learning between Electronic Structure Theories for Unifying Molecular, Surface, and Inorganic Crystal Foundation Force Fields
• Considering the ethics of large machine learning models in the chemical sciences
• Transient Detour and Cooperative Oxygen Exchange in the Polarization Switching of Ferroelectric Hf0.5Zr0.5O2
• Thermodynamics-Informed Machine Learning for Energy Materials Discovery
• Understanding Mobile Particles in Solid-State Materials: From the Perspective of Potential Energy Surfaces
• Universally Accurate or Specifically Inadequate? Stress‐Testing General Purpose Machine Learning Interatomic Potentials
• Predicting Reactivity with Machine Learning
• QM9star, two Million DFT-computed Equilibrium Structures for Ions and Radicals with Atomic Information
• A Simple and Scalable Kernel Density Approach for Reliable Uncertainty Quantification in Atomistic Machine Learning
• DeFecT-FF: Accelerated Modeling of Defects in Cd-Zn--Te-Se-S Compounds Combining High-Throughput DFT and Machine Learning Force Fields
• AQCat25: Unlocking spin-aware, high-fidelity machine learning potentials for heterogeneous catalysis
• LacMaterial: Large Language Models as Analogical Chemists for Materials Discovery
• Atomic cluster expansion potential for the Si-H system
• How Accurate Are DFT Forces? Unexpectedly Large Uncertainties in Molecular Datasets
• Prospects for Using Artificial Intelligence to Understand Intrinsic Kinetics of Heterogeneous Catalytic Reactions
• Unravelling the Catalytic Activity of Dual-Metal Doped N6-Graphene for Sulfur Reduction via Machine Learning-Accelerated First-Principles Calculations
• SimplySQS: An Automated and Reproducible Workflow for Special Quasirandom Structure Generation with ATAT
• Cross-disciplinary perspectives on the potential for artificial intelligence across chemistry
• Challenges and opportunities for machine learning potentials in transition path sampling: alanine dipeptide and azobenzene studies
• Advancing AI-Driven Analysis in X-ray Absorption Spectroscopy: Spectral Domain Mapping and Universal Models
• SimPoly: Simulation of Polymers with Machine Learning Force Fields Derived from First Principles
• Reciprocal Space Attention for Learning Long-Range Interactions
• Continuous Uniqueness and Novelty Metrics for Generative Modeling of Inorganic Crystals
• Optimizing Cross-Domain Transfer for Universal Machine Learning Interatomic Potentials
• Are diffusion models ready for materials discovery in unexplored chemical space?
• Transferable Generative Models Bridge Femtosecond to Nanosecond Time-Step Molecular Dynamics
• Autonomous interpretation of atomistic scattering data
• Comparing fine-tuning strategies of MACE machine learning force field for modeling Li-ion diffusion in LiF for batteries
• Transformers Discover Molecular Structure Without Graph Priors
• Active learning and explicit electrostatics enable accurate modeling of electrolytes
• Learning from the electronic structure of molecules across the periodic table
• Fine-Tuning Bulk-oriented Universal Interatomic Potentials for Surfaces: Accuracy, Efficiency, and Forgetting Control
• Extrapolation of Machine-Learning Interatomic Potentials for Organic and Polymeric Systems
• ADAPT: Lightweight, Long-Range Machine Learning Force Fields Without Graphs
• Accelerating Crystal Structure Prediction Using Data-Derived Potentials: High-Pressure Binary Hydrides
• Scalable Foundation Interatomic Potentials via Message-Passing Pruning and Graph Partitioning
• MLIP Arena: Advancing Fairness and Transparency in Machine Learning Interatomic Potentials via an Open, Accessible Benchmark Platform
• Quantum machine learning interatomic potential: Application of variational quantum algorithm
• Fast and Accurate Foundation Models for Equivariant Machine-Learned Interatomic Potentials
• Mapping the influence of symmetry breaking in structure-property relationships of ABO₃ perovskites
• Toward accelerating rare-earth metal extraction using equivariant neural networks
• Equivariant Many-body Message Passing Interatomic Potentials for Magnetic Materials
• The loss landscape of powder X-ray diffraction-based structure optimization is too rough for gradient descent
• El Agente Sólido: A New Age(nt) for Solid State Simulations
• Capacity-weighted figures-of-merit for battery transport metrics
• Highly efficient path-integral molecular dynamics simulations with GPUMD using neuroevolution potentials: Case studies on thermal properties of materials
• Graph atomic cluster expansion for foundational machine learning interatomic potentials
• OMNI-P2x universal neural network potential for excited-state simulations
• General Learning of the Electric Response of Inorganic Materials
• Bayesian Neural Networks versus deep ensembles for uncertainty quantification in machine learning interatomic potentials
• The Good, the Bad, and the Ugly of Atomistic Learning for "Clusters-to-Bulk" Generalization
• Spectral/Spatial Tensor Atomic Cluster Expansion with Universal Embeddings in Cartesian Space
• A universal machine learning model for the electronic density of states
• Resolving the Body-Order Paradox of Machine Learning Interatomic Potentials
• Inverse Design of Amorphous Materials with Targeted Properties
• Thermal Conductivity Limits of MoS₂ and MoSe₂: Revisiting High-Order Anharmonic Lattice Dynamics with Machine Learning Potentials
• The Quantum Method of Planes -- Local Pressure Definitions for Machine Learning Potentials
• Teachers that teach the irrelevant: Pre-training machine learned interaction potentials with classical force fields for robust molecular dynamics simulations
• Cluster Models for Next-Generation, Machine-Learning-Based Energy Functions for Molecular Simulations
• OpenCSP: A Deep Learning Framework for Crystal Structure Prediction from Ambient to High Pressure
• Benchmarking CHGNet Universal Machine Learning Interatomic Potential Against DFT and EXAFS: Case of Layered WS2 and MoS2
• Visualizing phonon edge states on molybdenum disulphide
• Facet: highly efficient E(3)-equivariant networks for interatomic potentials
• Accelerating first-principles molecular-dynamics thermal conductivity calculations for complex systems
• Towards Accurate and Scalable High-throughput MOF Adsorption Screening: Merging Classical Force Fields and Universal Machine Learned Interatomic Potentials
• Global properties of the energy landscape: a testing and training arena for machine learned potentials
• Medium-range structural order in amorphous arsenic
• Node-Equivariant Message Passing for Efficient and Accurate Machine Learning Interatomic Potentials
• Training a Foundation Model for Materials on a Budget
• Decoding local framework dynamics in the ultra-small pore MOF MIL-120(Al) CO2 sorbent with Machine Learned Potentials
• LeMat-Traj: A Scalable and Unified Dataset of Materials Trajectories for Atomistic Modeling
• MC3D: The Materials Cloud computational database of experimentally known stoichiometric inorganics
• Equivariant Electronic Hamiltonian Prediction with Many-Body Message Passing
• Expanding the search space of high entropy oxides and predicting synthesizability using machine learning interatomic potentials
• aims-PAX: Parallel Active eXploration for the automated construction of Machine Learning Force Fields
• Performance of universal machine-learned potentials with explicit long-range interactions in biomolecular simulations
• Multi-head committees enable direct uncertainty prediction for atomistic foundation models
• Machine Learning Phonon Spectra for Fast and Accurate Optical Lineshapes of Defects
• Autonomous thermodynamically informed database generation for machine-learned interatomic potentials and application to magnesium
• Deep learning methods for 2D material electronic properties
• A literature-derived dataset of migration barriers for quantifying ionic transport in battery materials
• Leveraging transfer learning for accurate estimation of ionic migration barriers in solids
• On-the-Fly Machine Learning of Interatomic Potentials for Elastic Property Modeling in Al-Mg-Zr Solid Solutions
• TorchSim: An efficient atomistic simulation engine in PyTorch
• RAFFLE: active learning accelerated interface structure prediction
• Model Accuracy and Data Heterogeneity Shape Uncertainty Quantification in Machine Learning Interatomic Potentials
• The Open DAC 2025 Dataset for Sorbent Discovery in Direct Air Capture
• Autonomous Inorganic Materials Discovery via Multi-Agent Physics-Aware Scientific Reasoning
• Expert-Guided LLM Reasoning for Battery Discovery: From AI-Driven Hypothesis to Synthesis and Characterization
• A Foundation Model for Material Fracture Prediction
• Investigating CO Adsorption on Cu(111) and Rh(111) Surfaces Using Machine Learning Exchange-Correlation Functionals
• Heterogeneous Ensemble Enables a Universal Uncertainty Metric for Atomistic Foundation Models
• Gradient-based grand canonical optimization enabled by graph neural networks with fractional atomic existence
• Physics-grounded generative design of inherently stable, novel and controllable crystal structures
• Atomistic Generative Diffusion for Materials Modeling
• Dataset: Statistics makes a difference: Machine learning adsorption dynamics of functionalized cyclooctyne on Si(001) at DFT accuracy
• Advancing our Understanding of Optoionic Effects for the Design of Solar Batteries: A Theoretical Perspective
• Comparing classical and machine learning force fields for modeling deformation of metal–organic frameworks relevant for direct air capture
• A framework to evaluate machine learning crystal stability predictions
• Modeling the Behavior of Complex Aqueous Electrolytes Using Machine Learning Interatomic Potentials: The Case of Sodium Sulfate
• Corner-Sharing PS₄-BS₄ Modes Facilitate Fast Ion Conduction in Lithium Thioborophosphate Iodide Glassy Solid Electrolytes
• Phase Stability and Transformations in Lead Mixed Halide Perovskites from Machine Learning Force Fields
• MP-ALOE: An r2SCAN dataset for universal machine learning interatomic potentials
• Beyond Scaling Curves: Internal Dynamics of Neural Networks Through the NTK Lens
• System-performance and cost modeling of Large Language Model training and inference
• A Scalable and Quantum-Accurate Foundation Model for Biomolecular Force Field via Linearly Tensorized Quadrangle Attention
• XtalOpt Version 14: Variable-Composition Crystal Structure Search for Functional Materials Through Pareto Optimization
• UMA: A Family of Universal Models for Atoms
• Band-Gap Tunability in Anharmonic Perovskite-like Semiconductors Driven by Polar Electron-Phonon Coupling
• Fine-Tuning Universal Machine-Learned Interatomic Potentials: A Tutorial on Methods and Applications
• A Survey of AI for Materials Science: Foundation Models, LLM Agents, Datasets, and Tools
• A Study on the Fine-Tuning Performance of Universal Machine-Learned Interatomic Potentials (U-MLIPs)
• Beyond Scaling: Chemical Intuition as Emergent Ability of Universal Machine Learning Interatomic Potentials
• Machine Learning Potentials for Alloys: A Detailed Workflow to Predict Phase Diagrams and Benchmark Accuracy
• An efficient forgetting-aware fine-tuning framework for pretrained universal machine-learning interatomic potentials
• An Atomic Cluster Expansion Potential for Twisted Multilayer Graphene
• Understanding multi-fidelity training of machine-learned force-fields
• Machine learning potentials for modeling alloys across compositions
• GEARS H: Accurate machine-learned Hamiltonians for next-generation device-scale modeling
• Fast and Fourier Features for Transfer Learning of Interatomic Potentials
• A Graph Neural Network for the Era of Large Atomistic Models
• Benchmarking Universal Machine Learning Interatomic Potentials for Real-Time Analysis of Inelastic Neutron Scattering Data
• A Foundation Model for Non-Destructive Defect Identification from Vibrational Spectra
• Machine-learning interatomic potentials from a users perspective: A comparison of accuracy, speed and data efficiency
• Elucidating Ion Capture and Transport Mechanisms of Preyssler Anions in Aqueous Solutions Using Biased MACE-Accelerated MD Simulations
• Machine-Learned Potentials for Solvation Modeling
• Apax: A Flexible and Performant Framework For The Development of Machine-Learned Interatomic Potentials
• DistMLIP: A Distributed Inference Platform for Machine Learning Interatomic Potentials
• Roadmap for warm dense matter physics
• Defining Foundation Models for Computational Science: A Call for Clarity and Rigor
• Interpretable machine learned predictions of adsorption energies at the metal--oxide interface
• High-Entropy Solid Electrolytes Discovery: A Dual-Stage Machine Learning Framework Bridging Atomic Configurations and Ionic Transport Properties
• RamanPFN: learning from Raman spectral structure with a tabular foundation model
• Tailored Vapor Deposition Unlocks Large-Grain, Wafer-Scale Epitaxial Growth of 2D Magnetic CrCl3
• FlashMD: long-stride, universal prediction of molecular dynamics
• Computational screening and experimental validation of promising Wadsley-Roth Niobates
• Efficient Parallelization of Message Passing Neural Network Potentials for Large-scale Molecular Dynamics
• Generation of magnetic metal-organic frameworks
• Inverse Design of Inorganic Compounds with Generative AI
• Prediction and Experimental Verification of Electrolyte Solvation Structure from an OMol25-Trained Interatomic Potential
• Differentiable inverse design of short-range order in high-entropy alloys: from target sro to target property
• Pushing the limits of unconstrained machine-learned interatomic potentials
• Accurate Modeling of Interfacial Thermal Transport in van der Waals Heterostructures via Hybrid Machine Learning and Registry-Dependent Potentials
• Quantum annealing for materials
• MLIPilot: LLM-Driven Auto-Research for Machine-Learned Interatomic Potentials
• NEP89: Universal neuroevolution potential for inorganic and organic materials across 89 elements
• Materials Database from All-electron Hybrid Functional DFT Calculations
• Egret-1: Pretrained Neural Network Potentials for Efficient and Accurate Bioorganic Simulation
• Data-Driven Thermal and Mechanical Modeling of Defective Covalent Organic Frameworks
• Towards Faster and More Compact Foundation Models for Molecular Property Prediction
• LAMBench: A Benchmark for Large Atomistic Models
• Predicting neutron experiments from first principles: A workflow powered by machine learning
• AccelNet: Exact backward-compatible acceleration of polynomial angular descriptors through Cartesian moment factorization
• Crystal structure prediction with host-guided inpainting generation and foundation potentials
• Calculations of the Krypton Phase Diagram and Novel Plasticity
• Accurate Machine Learning Interatomic Potentials for Polyacene Molecular Crystals: Application to Single Molecule Host-Guest Systems
• Optimizing Data Distribution and Kernel Performance for Efficient Training of Chemistry Foundation Models: A Case Study with MACE
• Orb-v3: atomistic simulation at scale
• Cross-functional transferability in universal machine learning interatomic potentials

Discussions

• I'm biased as a co-author, but this is a must-read on how good ML interatomic potentials can be on materials and applications beyond the limits of their training set. Open the SI. I dare you. 😉 arx [bsky, 9 points, 1 comments]
• a well spent winter holiday.. arxiv.org/abs/2401.00096 [bsky, 3 points, 0 comments]
• #compchem Good read: A foundation model for atomistic materials chemistry arxiv.org/abs/2401.00096 [bsky, 2 points, 0 comments]
• Thursday morning started with another session about machine learning force fields and their many applications. Gábor Csányi took us on a journey through the history of MLFFs and presented the latest [bsky, 1 points, 1 comments]
• A foundation model for atomistic materials chemistry [hn, 1 points, 0 comments]

Related

• Foundation Models for Atomistic Simulation of Chemistry and Materials
• The Role of Reference Points in Machine-Learned Atomistic Simulation Models
• PFD: Automatically Generating Machine Learning Force Fields from Universal Models
• BLAST: Bridging Length/time scales via Atomistic Simulation Toolkit
• Beyond potentials: integrated machine-learning models for materials