Scaling deep learning for materials discovery

2023/11/29 by Amil Merchant, Simon Batzner, Samuel S. Schoenholz +3 · 196 citations
Energy · Engineering · Materials Science · #Advanced Photocatalysis Techniques #Fuel Cells and Related Materials #Machine Learning in Materials Science

paper · pdf · doi:10.1038/s41586-023-06735-9

openalex publication_date 2023/11/29 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/04

Abstract

Abstract Novel functional materials enable fundamental breakthroughs across technological applications from clean energy to information processing 1–11 . From microchips to batteries and photovoltaics, discovery of inorganic crystals has been bottlenecked by expensive trial-and-error approaches. Concurrently, deep-learning models for language, vision and biology have showcased emergent predictive capabilities with increasing data and computation 12–14 . Here we show that graph networks trained at scale can reach unprecedented levels of generalization, improving the efficiency of materials discovery by an order of magnitude. Building on 48,000 stable crystals identified in continuing studies 15–17 , improved efficiency enables the discovery of 2.2 million structures below the current convex hull, many of which escaped previous human chemical intuition. Our work represents an order-of-magnitude expansion in stable materials known to humanity. Stable discoveries that are on the final convex hull will be made available to screen for technological applications, as we demonstrate for layered materials and solid-electrolyte candidates. Of the stable structures, 736 have already been independently experimentally realized. The scale and diversity of hundreds of millions of first-principles calculations also unlock modelling capabilities for downstream applications, leading in particular to highly accurate and robust learned interatomic potentials that can be used in condensed-phase molecular-dynamics simulations and high-fidelity zero-shot prediction of ionic conductivity.

Citations

• Projector augmented-wave method
• From ultrasoft pseudopotentials to the projector augmented-wave method
• Rationale for mixing exact exchange with density functional approximations
• VESTA 3 for three-dimensional visualization of crystal, volumetric and morphology data
• Constructing high‐dimensional neural network potentials: A tutorial review
• Efficient iterative schemes forab initiototal-energy calculations using a plane-wave basis set
• Deep Learning Scaling is Predictable, Empirically
• Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point clouds
• A Performance and Cost Assessment of Machine Learning Interatomic Potentials
• Accurate and Numerically Efficient r2SCAN Meta-Generalized Gradient Approximation
• An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
• Materials challenges and opportunities for quantum computing hardware
• Highly accurate protein structure prediction with AlphaFold
• E(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials
• A universal graph deep learning interatomic potential for the periodic table

Cited by

• Simulating Subjects: The Promise and Peril of Artificial Intelligence Stand-Ins for Social Agents and Interactions
• Universally Accurate or Specifically Inadequate? Stress‐Testing General Purpose Machine Learning Interatomic Potentials
• Artificial intelligence-driven design of inorganic materials for biomedical applications
• Peer Review and AI: Your (Human) Opinion Is What Matters
• Symmetry and Generalisation in Neural Approximations of Renormalisation Transformations
• Rational Design of Covalent Organic Framework Catalysts for Photocatalytic Carbon Dioxide Fixation
• Towards a Generalizable AI for Materials Discovery: Validation through Immersion Coolant Screening
• AQCat25: Unlocking spin-aware, high-fidelity machine learning potentials for heterogeneous catalysis
• Normal Dirac Semimetal Phase and Zeeman-Induced Topological Fermi Arc in PtSr5
• Towards end-to-end automation of AI research
• A Unified Perspective on Optimization in Machine Learning and Neuroscience: From Gradient Descent to Neural Adaptation
• HyperDiffusionFields (HyDiF): Diffusion-Guided Hypernetworks for Learning Implicit Molecular Neural Fields
• Efficient small-cell sampling for machine-learning potentials of multi-principal element alloys
• Prediction-Augmented Trees for Reliable Statistical Inference
• On Foundation Models for Temporal Point Processes to Accelerate Scientific Discovery
• MatSciBench: Benchmarking the Reasoning Ability of Large Language Models in Materials Science
• Optimizing Cross-Domain Transfer for Universal Machine Learning Interatomic Potentials
• GraphTracer: Graph-Guided Failure Tracing in LLM Agents for Robust Multi-Turn Deep Search
• Rise of the Robochemist
• exaPD: A highly parallelizable workflow for multi-element phase diagram (PD) construction
• Are diffusion models ready for materials discovery in unexplored chemical space?
• Mephisto: Self-Improving Large Language Model-Based Agents for Automated Interpretation of Multi-band Galaxy Observations
• Near-limit quantum control beyond analytic tractability in many-body spin systems
• Ambient pressure high temperature superconductivity in RbPH₃ facilitated by ionic anharmonicity
• GyroSwin: 5D Surrogates for Gyrokinetic Plasma Turbulence Simulations
• Autonomous interpretation of atomistic scattering data
• Fundamental Limits of Crystalline Equivariant Graph Neural Networks: A Circuit Complexity Perspective
• Hybrid Sequential Quantum Computing
• Training Variation of Physically-Informed Deep Learning Models
• exa-AMD: An Exascale-Ready Framework for Accelerating the Discovery and Design of Functional Materials
• Machine Learning Prediction of Charged Defect Formation Energies from Crystal Structures
• Efficient E(3)-equivariant framework for universal charge density prediction
• Nondestructive characterization of laser-cooled atoms using machine learning
• Are neural scaling laws leading quantum chemistry astray?
• Fingerprinting Organic Molecules for the Inverse Design of Two-Dimensional Hybrid Perovskites with Target Energetics
• Agentic Exploration of Physics Models
• High-entropy nanomaterials by candlelight
• From Edge to HPC: Investigating Cross-Facility Data Streaming Architectures
• Space Group Conditional Flow Matching
• Autonomous laboratories in China: an embodied intelligence-driven platform to accelerate chemical discovery
• Accelerating Crystal Structure Prediction Using Data-Derived Potentials: High-Pressure Binary Hydrides
• CrystalGym: A New Benchmark for Materials Discovery Using Reinforcement Learning
• Optimizing the non-Clifford-count in unitary synthesis using Reinforcement Learning
• Scalable Foundation Interatomic Potentials via Message-Passing Pruning and Graph Partitioning
• The future of machine learning for small-molecule drug discovery will be driven by data
• Artificial Intelligence Empowered New Materials: Discovery, Synthesis, Prediction to Validation
• Quantum machine learning interatomic potential: Application of variational quantum algorithm
• Can AI Follow In Einstein's Footsteps?
• Material dynamics analysis with deep generative model
• Learning Crystallographic Disorder: Bridging Prediction and Experiment in Materials Discovery
• Lessons from complex systems science for AI governance
• Will AI accelerate or delay the race to net-zero emissions?
• Prussian Blue Analogues as Anode Materials for Battery Applications: Complexities and Horizons
• Application of pretrained universal machine-learning interatomic potential for physicochemical simulation of liquid electrolytes in Li-ion batteries
• El Agente Sólido: A New Age(nt) for Solid State Simulations
• A foundation model for atomistic materials chemistry
• Accelerated Discovery of Topological Conductors for Nanoscale Interconnects
• Higher-order, generically complete, continuous, and polynomial-time isometry invariants of periodic sets
• Inverse Design of Amorphous Materials with Targeted Properties
• Mutual Information Surprise: Rethinking Unexpectedness in Autonomous Systems
• All that structure matches does not glitter
• 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
• Facet: highly efficient E(3)-equivariant networks for interatomic potentials
• Decoratypes: An Extensible Crystal Taxonomy for Machine Learning-Guided Materials Discovery
• The Need for Verification in AI-Driven Scientific Discovery
• Multitask Battery Management with Flexible Pretraining
• Challenges in Non-Polymeric Crystal Structure Prediction: Why a Geometric, Permutation-Invariant Loss is Needed
• Self-organized learning emerges from coherent coupling of critical neurons
• Migration as a Probe: A Generalizable Benchmark Framework for Specialist vs. Generalist Machine-Learned Force Fields
• Search for thermodynamically stable ambient-pressure superconducting hydrides in GNoME database
• Memorization in Graph Neural Networks
• Universal Machine Learning Potential for Systems with Reduced Dimensionality
• Equivariant Electronic Hamiltonian Prediction with Many-Body Message Passing
• Molecular representation learning: cross-domain foundations and future Frontiers
• BLIPs: Bayesian Learned Interatomic Potentials
• "DIVE" into Hydrogen Storage Materials Discovery with AI Agents
• aims-PAX: Parallel Active eXploration for the automated construction of Machine Learning Force Fields
• CrystalDiT: A Diffusion Transformer for Crystal Generation
• Open science falling behind in the era of artificial intelligence
• Multi-head committees enable direct uncertainty prediction for atomistic foundation models
• Autonomous thermodynamically informed database generation for machine-learned interatomic potentials and application to magnesium
• Subsampling Factorization Machine Annealing
• Chi-Geometry: A Library for Benchmarking Chirality Prediction of GNNs
• Bottom-up Domain-specific Superintelligence: A Reliable Knowledge Graph is What We Need
• Enhancing Corrosion Resistance of Aluminum Alloys Through AI and ML Modeling
• Discovery Learning accelerates battery design evaluation
• Text-mined dataset of solid-state syntheses with impurity phases using Large Language Model
• Deep learning methods for 2D material electronic properties
• Hybrid Quantum--Classical Machine Learning Potential with Variational Quantum Circuits
• EAC-Net: Predicting real-space charge density via equivariant atomic contributions
• Viability of perturbative expansion for quantum field theories on neurons
• Artificial Intelligence and Generative Models for Materials Discovery -- A Review
• 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
• Superconductivity in atom-intercalated quaternary hydrides under ambient pressure
• Towards trustworthy AI in materials mechanics through domain-guided attention
• Iterative Pretraining Framework for Interatomic Potentials
• Energy Underprediction from Symmetry in Machine-Learning Interatomic Potentials
• 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
• Accelerated Data‐Driven Discovery of Dual‐Functional Ionic Liquid Passivation for FAPbI 3 Perovskite Solar Cells Using Graph Neural Network
• Materials for thermochemical energy storage and conversion: attributes for low-temperature applications
• Constructing material network representations for intelligent amorphous alloys design
• Graph Learning Metallic Glass Discovery from Wikipedia
• Perovskite-R1: A Domain-Specialized LLM for Intelligent Discovery of Precursor Additives and Experimental Design
• MOFSimBench: Evaluating Universal Machine Learning Interatomic Potentials In Metal--Organic Framework Molecular Modeling
• Accelerated data-driven materials science with the Materials Project
• Advancing our Understanding of Optoionic Effects for the Design of Solar Batteries: A Theoretical Perspective
• A framework to evaluate machine learning crystal stability predictions
• HSG-12M: A Large-Scale Benchmark of Spatial Multigraphs from the Energy Spectra of Non-Hermitian Crystals
• Virtual Laboratories: Domain-agnostic workflows for research
• MolSight: A Graph-Aware Vision-Language Model for Unified Chemical Image Understanding
• Advancing network resilience theories with symbolized reinforcement learning
• Kinetic Langevin Diffusion for Crystalline Materials Generation
• SciGA: A Comprehensive Dataset for Designing Graphical Abstracts in Academic Papers
• XtalOpt Version 14: Variable-Composition Crystal Structure Search for Functional Materials Through Pareto Optimization
• AI-Assisted Rapid Crystal Structure Generation Towards a Target Local Environment
• Accelerated discovery and design of Fe-Co-Zr magnets with tunable magnetic anisotropy through machine learning and parallel computing
• Fine-Tuning Universal Machine-Learned Interatomic Potentials: A Tutorial on Methods and Applications
• XxaCT-NN: Structure Agnostic Multimodal Learning for Materials Science
• Analysis-ready VCF at Biobank scale using Zarr
• exa-AMD: A Scalable Workflow for Accelerating AI-Assisted Materials Discovery and Design
• 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
• Leveraging neural network interatomic potentials for a foundation model of chemistry
• Structural chirality and related properties in periodic inorganic solids: review and perspectives
• CLOUD: A Scalable and Physics-Informed Foundation Model for Crystal Representation Learning
• Over-squashing in Spatiotemporal Graph Neural Networks
• Machine learning potentials for modeling alloys across compositions
• SIMSHIFT: A Benchmark for Adapting Neural Surrogates to Distribution Shifts
• Equivariant Neural Diffusion for Molecule Generation
• GEARS H: Accurate machine-learned Hamiltonians for next-generation device-scale modeling
• Machine learning for in-situ composition mapping in a self-driving magnetron sputtering system
• chemtrain-deploy: A parallel and scalable framework for machine learning potentials in million-atom MD simulations
• Soft causal learning for generalized molecule property prediction: An environment perspective
• Industry’s Path to a Greener Future: A Perspective on Current Sustainable Practices and Areas of Opportunity
• Matter-of-Fact: A Benchmark for Verifying the Feasibility of Literature-Supported Claims in Materials Science
• A Graph Neural Network for the Era of Large Atomistic Models
• AI Scientists Fail Without Strong Implementation Capability
• Graph Neural Networks for Jamming Source Localization
• Beyond Atomic Geometry Representations in Materials Science: A Human-in-the-Loop Multimodal Framework
• BioCLIP 2: Emergent Properties from Scaling Hierarchical Contrastive Learning
• DistMLIP: A Distributed Inference Platform for Machine Learning Interatomic Potentials
• Machine Learning Interatomic Potentials: library for efficient training, model development and simulation of molecular systems
• RamanPFN: learning from Raman spectral structure with a tabular foundation model
• CIKT: A Collaborative and Iterative Knowledge Tracing Framework with Large Language Models
• Operator Learning for Schrödinger Equation: Unitarity, Error Bounds, and Time Generalization
• Improving Generative Inverse Design of Rectangular Patch Antennas with Test Time Optimization
• When May a Model Replace the Experiment? Audits, Licenses, and the Price of Trust in Surrogate-Driven Design
• Bridging Theory and Experiment in Materials Discovery: Machine-Learning-Assisted Prediction of Synthesizable Structures
• InvDesFlow-AL: active learning-based workflow for inverse design of functional materials
• Rejoining fragmented ancient bamboo slips with physics-driven deep learning
• Self-Optimizing Machine Learning Potential Assisted Automated Workflow for Highly Efficient Complex Systems Material Design
• BOOM: Benchmarking Out-Of-distribution Molecular Property Predictions of Machine Learning Models
• Intern-S1-Pro: Scientific Multimodal Foundation Model at Trillion Scale
• The evolution of machine learning potentials for molecules, reactions and materials
• Generation of magnetic metal-organic frameworks
• Artificial intelligence and downscaling global climate model future projections
• Towards a new paradigm of scientific discovery with socialized artificial intelligence
• The Precursor Genome: A Pairwise Reaction Dataset for Solid-State Synthesis
• BatteryMat: a hierarchical machine-learning and DFT framework for average-voltage screening of lithium-ion cathode materials
• InvDesMobility: a reliability-gated first-principles feedback framework for closed-loop materials discovery
• Materials discovery acceleration by using condition generative methodology
• Spatial statistics for screening molecular structures
• NOVA: Fundamental Limits of Knowledge Discovery Through AI
• A New Strategy for Artificial Intelligence: Training Foundation Models Directly on Human Brain Data
• NEP89: Universal neuroevolution potential for inorganic and organic materials across 89 elements
• Database and deep-learning scalability of anharmonic phonon properties by automated brute-force first-principles calculations
• Materials Database from All-electron Hybrid Functional DFT Calculations
• Accurate, Interdisciplinary and Transparent Structure-property Understanding with Deep Native Structural Reasoning
• Information Entropy Based Crystal Structure Prediction of Chemically Disordered Alloys via Graph Convolutional Neural Networks
• Large language model-enabled automated data extraction for concrete materials informatics
• Towards Automated Discovery: A Review of Generative Models, Multimodal Learning and Closed-Loop Workflows in Inverse Materials Design
• Physics Is All You Need? A Case Study in Physicist-Supervised AI Development of Scientific Software
• A Universal Spin-Orbit-Coupled Hamiltonian Model for Accelerated Quantum Material Discovery
• Predicting neutron experiments from first principles: A workflow powered by machine learning
• EvoMaster: A Foundational Evolving Agent Framework for Agentic Science at Scale
• Generative design of inorganic materials
• Computational search for materials having a giant anomalous Hall effect in the pyrochlore and spinel crystal structures
• AI-driven discovery of new materials: breaking data bottlenecks and transforming research paradigms
• How artificial intelligence reshapes materials design and its evolutionary path
• Towards Solving the Gilbert-Pollak Conjecture via Large Language Models
• Large Language Models to Accelerate Organic Chemistry Synthesis
• Crystal structure prediction with host-guided inpainting generation and foundation potentials
• Accurate and efficient protocols for high-throughput first-principles materials simulations
• High-performance training and inference for deep equivariant interatomic potentials
• Accelerated discovery of cost-effective photoabsorber materials for near-infrared (λ=1600 nm) photodetector applications
• Fine-Tuning Small Language Models for Reliable VASP INCAR Generation
• GenGA: Editable and Data-Grounded Graphical Abstract Generation for Academic Papers
• Towards Interpretable Deep Generative Models via Causal Representation Learning
• Accurate Machine Learning Interatomic Potentials for Polyacene Molecular Crystals: Application to Single Molecule Host-Guest Systems
• MatterTune: An Integrated, User-Friendly Platform for Fine-Tuning Atomistic Foundation Models to Accelerate Materials Simulation and Discovery
• CRYSIM: Prediction of Symmetric Structures of Large Crystals with GPU-based Ising Machines
• Cross-functional transferability in universal machine learning interatomic potentials

Related

• Large-scale machine-learning-assisted exploration of the whole materials space
• Data-model Coevolution as the Architectural Principle for AI-Native Materials Databases
• Crystalline Material Discovery in the Era of Artificial Intelligence
• Establishing baselines for generative discovery of inorganic crystals
• Deep learning in-depth analysis of crystal graph convolutional neural networks: A new era in materials discovery and its applications