Universal Machine-Learned Interatomic Potentials
Force fields learned from first-principles calculations that simulate materials across the periodic table far faster than DFT.
Tier Grades
Tier A: Frontier AI labs
feasible
Microsoft and Meta train potentials on large first-principles datasets.
Tier B: Indonesian AI labs
stretch
Training data comes from large DFT campaigns, but released potentials can be run locally.
Tier C: Institution units such as ITB
feasible
Running a released potential for local materials is a cheap replacement for DFT in a university lab.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
stretch
Screening candidate materials with a released potential is a possible proof of technology for chemical or mining clients.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeMatterSim: compute not stated in the abstract; Matbench Discovery ranks potentials by a combined score.
- DataMatterSim: actively learned from large-scale first-principles computations covering 0 to 5000 K and up to 1000 GPa.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 22 authors on the MatterSim paper (arXiv 2405.04967); 10 on Matbench Discovery.
- Time horizonMonths.
- Shelf-life signalsMatbench Discovery keeps an online leaderboard for new submissions, so the best potential changes as entries arrive.
- Pipeline roleTier A releases universal potentials; Tier C runs them on local materials questions; Tier D could screen candidates in a proof of technology.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Microsoft AI
Meta
EPFL
Harvard University
Orbital Materials
Seoul National University
DeepModeling
Institute of Computing Technology, Chinese Academy of Sciences
ICAMS Bochum
Massachusetts Institute of TechnologyLinked models
PET-OAM-XLmaterialsexemplareSEN-30M-OAMmaterialsexemplarMatterSimmaterialsexemplarDPA4materialsproduced byEquiformerV3materialsproduced bySevenNet-Omni-i12materialsproduced byTACE-v1-OAM-Mmaterialsproduced byNequip-OAM-XLmaterialsproduced byMatRIS-10M-OAMmaterialsproduced byGRACE-2L-OAM-Lmaterialsproduced byNequip-OAM-Lmaterialsproduced byEquFlashmaterialsproduced byORB v3materialsproduced byLinks
R&D TopicUniversal Machine-Learned Interatomic Potentials
ModelsMatterSimMicrosoft AI, materialseSEN-30M-OAMMeta, materialsPET-OAM-XLEPFL, materialsDPA4DeepModeling, materials+9 more
MatterSimMicrosoft AI, materialseSEN-30M-OAMMeta, materialsPET-OAM-XLEPFL, materialsDPA4DeepModeling, materials+9 moreUse Casesnone recorded
Who Runs Itnone recorded
Models 13
DPA4DeepModeling, materialsproduced byEquiformerV3Massachusetts Institute of Technology, materialsproduced byeSEN-30M-OAMMeta, materialsexemplarGRACE-2L-OAM-LICAMS Bochum, materialsproduced byMatRIS-10M-OAMInstitute of Computing Technology, Chinese Academy of Sciences, materialsproduced byMatterSimMicrosoft AI, materialsexemplarNequip-OAM-LHarvard University, materialsproduced byNequip-OAM-XLHarvard University, materialsproduced byORB v3Orbital Materials, materialsproduced byPET-OAM-XLEPFL, materialsexemplarSevenNet-Omni-i12Seoul National University, materialsproduced byEquFlashmaterialsproduced byPapers
- UMA: A Family of Universal Models for AtomsBrandon M. Wood et al., arXiv preprint, 2025 state of the art
- A foundation model for atomistic materials chemistryIlyes Batatia et al., arXiv preprint, 2023 state of the art
- Matbench Discovery -- A framework to evaluate machine learning crystal stability predictionsJanosh Riebesell et al., arXiv preprint, 2023 state of the art
- A Universal Graph Deep Learning Interatomic Potential for the Periodic TableChi Chen et al., arXiv preprint, 2022 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2405.04967 https://arxiv.org/abs/2405.04967
- arxiv.org/abs/2308.14920 https://arxiv.org/abs/2308.14920
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Universal Machine-Learned Interatomic Potentials", https://aurion.semai.id/topics/universal-machine-learned-interatomic-potentials/, records of October 2026.