Systematic softening in universal machine learning interatomic potentials
Lawrence Berkeley National Laboratory · University of California, Berkeley · +1 more institution
Abstract
Abstract Machine learning interatomic potentials (MLIPs) have introduced a new paradigm for atomic simulations. Recent advancements have led to universal MLIPs (uMLIPs) that are pre-trained on diverse datasets, providing opportunities for universal force fields and foundational machine learning models. However, their performance in extrapolating to out-of-distribution complex atomic environments remains unclear. In this study, we highlight a consistent potential energy surface (PES) softening effect in three uMLIPs: M3GNet, CHGNet, and MACE-MP-0, which is characterized by energy and force underprediction in atomic-modeling benchmarks including surfaces, defects, solid-solution energetics, ion migration…
Citation impact
- FWCI
- 42.07
- Percentile
- 100%
- References
- 56
Authors
9- BDBowen DengCorresponding
Lawrence Berkeley National Laboratory, University of California, Berkeley
- YCYunyeong Choi
Lawrence Berkeley National Laboratory, University of California, Berkeley
- PZPeichen Zhong
Lawrence Berkeley National Laboratory, University of California, Berkeley
- JRJanosh Riebesell
University of Cambridge
- SAShashwat Anand
Lawrence Berkeley National Laboratory
Topics & keywords
- Softening
- Interatomic potential
- Materials science
- Chemical physics
- Computer science
- Statistical physics
- Physics
- Chemistry
- Affordable and clean energy
Funding
- NSNational Science FoundationAward: ACI1053575
- UDU.S. Department of EnergyAwards: KC23MP, DE-AC0205CH11231
- NENational Energy Research Scientific Computing CenterAward: DE-AC0205CH11231
- OOOffice of ScienceAward: DE-AC0205CH11231
- BEBasic Energy SciencesAwards: KC23MP, DE-AC0205CH11231
- NRNational Renewable Energy Laboratory
- LBLawrence Berkeley National LaboratoryAward: DE-AC0205CH11231