A foundation model for atomistic materials chemistry

Ilyes Batatia,Philipp Benner,Yuan Chiang,Alin M. Elena,Dávid P. Kovács,Janosh Riebesell,Xavier R. Advincula,Mark Asta,Matthew Avaylon,William J. Baldwin,Fabian Berger,Noam Bernstein,Arghya Bhowmik,Samuel M. Blau,Vlad Cărare,James P. Darby,Sandip De,Flaviano Della Pia,Volker L. Deringer,Rokas Elijošius,Zakariya El-Machachi,Fabio Falcioni,Edvin Fako,Andrea C. Ferrari,Annalena Genreith-Schriever,Janine George,Rhys E. A. Goodall,Clare P. Grey,Petr Grigorev,Shuang Han,Will Handley,Hendrik H. Heenen,Kersti Hermansson,Christian Holm,Jad Jaafar,Stephan Hofmann,Konstantin S. Jakob,Hyunwook Jung,Venkat Kapil,Aaron D. Kaplan,Nima Karimitari,James R. Kermode,Namu Kroupa,Jolla Kullgren,Matthew C. Kuner,Domantas Kuryla,Guoda Liepuoniute,Johannes T. Margraf,Ioan-Bogdan Magdău,Angelos Michaelides,J. Harry Moore,Aakash A. Naik,Samuel P. Niblett,Sam Walton Norwood,Niamh O'Neill,Christoph Ortner,Kristin A. Persson,Karsten Reuter,Andrew S. Rosen,Lars L. Schaaf,Christoph Schran,Benjamin X. Shi,Eric Sivonxay,Tamás K. Stenczel,Viktor Svahn,Christopher Sutton,Thomas D. Swinburne,Jules Tilly,Cas van der Oord,Eszter Varga-Umbrich,Tejs Vegge,Martin Vondrák,Yangshuai Wang,William C. Witt,Fabian Zills,Gábor Csányi

2024-03-01

Abstract:Machine-learned force fields have transformed the atomistic modelling of materials by enabling simulations of ab initio quality on unprecedented time and length scales. However, they are currently limited by: (i) the significant computational and human effort that must go into development and validation of potentials for each particular system of interest; and (ii) a general lack of transferability from one chemical system to the next. Here, using the state-of-the-art MACE architecture we introduce a single general-purpose ML model, trained on a public database of 150k inorganic crystals, that is capable of running stable molecular dynamics on 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 problems in the physical sciences, including the 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 and as a starting or "foundation model" for any atomistic system of interest and is thus a step towards democratising the revolution of ML force fields by lowering the barriers to entry.

Chemical Physics,Materials Science

What problem does this paper attempt to address?

This paper introduces a baseline model called MACE-MP-0, which aims to address two main issues in atomistic materials chemistry: the development and validation of machine learning (ML) potentials for specific systems requires significant computational and human resources, and the current ML potentials generally lack transferability across different chemical systems. By utilizing the state-of-the-art MACE architecture, the MACE-MP-0 model is trained solely on a public database containing 150,000 inorganic crystals and is capable of performing stable molecular dynamics simulations on molecules and materials. The paper demonstrates the wide applicability of the MACE-MP-0 model in various problems within the physical sciences, including solid state, liquid state, gas state, chemical reactions, interfaces, and protein dynamics. This model can be directly applied as a baseline model for any atomic system, reducing the entry barrier for ML force field revolution and democratizing material modeling. The research mentions that although current ML potential models have improved accuracy in certain applications, they often require a large amount of customized training data and are not applicable to a wide range of chemical and structural space. The MACE-MP-0 model demonstrates its generality and robustness by showcasing qualitative and quantitative accuracy in various chemical systems, enabling simulation of complex systems that are beyond the reach of the largest computing resources. The paper showcases the capabilities of the MACE-MP-0 model through a series of examples, such as properties of water and water-containing systems, prediction of catalytic reactions, and performance prediction of metal-organic framework materials. These examples demonstrate the excellent extrapolation performance of the MACE-MP-0 model even for chemical and structural spaces not covered in the training set.

A foundation model for atomistic materials chemistry

Crash Testing Machine Learning Force Fields for Molecules, Materials, and Interfaces: Model Analysis in the TEA Challenge 2023

Evaluation of the MACE Force Field Architecture: from Medicinal Chemistry to Materials Science

Crash Testing Machine Learning Force Fields for Molecules, Materials, and Interfaces: Molecular Dynamics in the TEA Challenge 2023

MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures

MACE-OFF23: Transferable Machine Learning Force Fields for Organic Molecules

BIGDML—Towards accurate quantum machine learning force fields for materials

Machine learned force-fields for an Ab-initio quality description of metal-organic frameworks

Accurate Machine Learned Quantum-Mechanical Force Fields for Biomolecular Simulations

Accurate interatomic force field for molecular dynamics simulation by hybridizing classical and machine learning potentials

Fast and flexible range-separated models for atomistic machine learning

Efficient Machine Learning Force Field for Large-Scale Molecular Simulations of Organic Systems

Roadmap on Machine Learning in Electronic Structure

Machine Learning Interatomic Potentials as Emerging Tools for Materials Science

Accurate Force Fields for Atomistic Simulations of Oxides, Hydroxides, and Organic Hybrid Materials up to the Micrometer Scale

On the design space between molecular mechanics and machine learning force fields

Thermal Conductivity Predictions with Foundation Atomistic Models

Accurate machine learning force fields via experimental and simulation data fusion