Engineering Papers⌕ Search
Home/Archive/DOE OSTI

DOE OSTI · 2528028

Systematic softening in universal machine learning interatomic potentials

Deng, Bowen [Univ. of California, Berkeley, CA (United States); Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:000000034085381X)·Choi, Yunyeong [Univ. of California, Berkeley, CA (United States); Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:0000000279464232)·Zhong, Peichen [Univ. of California, Berkeley, CA (United States); Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:0000000319211628)·Riebesell, Janosh [Univ. of Cambridge (United Kingdom). Cavendish Lab.] (ORCID:0000000152333462)·Anand, Shashwat [Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:0000000188051547)·Li, Zhuohan [Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:0000000153729450)·Jun, KyuJung [Univ. of California, Berkeley, CA (United States); Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:000000031974028X)·Persson, Kristin A. [Univ. of California, Berkeley, CA (United States); Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:0000000324955509)·Ceder, Gerbrand [Univ. of California, Berkeley, CA (United States); Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:0000000192753605)
View original source ↗Read source document ↗Export metadata ↓

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 barriers, phonon vibration modes, and general high-energy states. The PES softening behavior originates primarily from the systematically underpredicted PES curvature, which derives from the biased sampling of near-equilibrium atomic arrangements in uMLIP pre-training datasets. Our findings suggest that a considerable fraction of uMLIP errors are highly systematic, and can therefore be efficiently corrected. We argue for the importance of a comprehensive materials dataset with improved PES sampling for next-generation foundational MLIPs.

Explore related subjects

36 MATERIALS SCIENCE

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Deng, Bowen [Univ. of California, Berkeley, CA (United States); Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:000000034085381X), Choi, Yunyeong [Univ. of California, Berkeley, CA (United States); Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:0000000279464232), Zhong, Peichen [Univ. of California, Berkeley, CA (United States); Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:0000000319211628), Riebesell, Janosh [Univ. of Cambridge (United Kingdom). Cavendish Lab.] (ORCID:0000000152333462), Anand, Shashwat [Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:0000000188051547), Li, Zhuohan [Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:0000000153729450), Jun, KyuJung [Univ. of California, Berkeley, CA (United States); Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:000000031974028X), Persson, Kristin A. [Univ. of California, Berkeley, CA (United States); Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:0000000324955509), Ceder, Gerbrand [Univ. of California, Berkeley, CA (United States); Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:0000000192753605). 2025-01-10. Systematic softening in universal machine learning interatomic potentials. https://doi.org/10.1038/s41524-024-01500-6

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related reports

DOE OSTI2026

Cyclic moisture reactivation of calcium sorbents for long duration thermochemical energy storage

The transition to a flexible and reliable energy infrastructure, using electro-thermal energy generation technologies such as geothermal, concentrated solar power, and nuclear, usually demands simultaneous advancement of thermal energy storage (TES) to support on-demand electricity generation and industrial applications while mitigating the inherent intermittency of renewable energy sources and power outages from direct energy generation. Among TES technologies, thermochemical energy storage (TCES) based on calcium looping emerges as a compelling high-power energy storage candidate due to its high reaction enthalpy, compatibility with elevated operating temperatures, and abundance of low-cost materials. However, the long-term durability of calcium-based sorbents for TCES is hindered by surface sintering and particle aggregation, leading to performance degradation over repeated thermal cycles. This study explores a moisture hydration-based strategy to regenerate a degraded calcium sorbent and mitigate performance degradation for long duration TCES. The addition of moisture transforms calcium oxide into calcium hydroxide and produces intercalation water layers, associated with a regenerated surface area and reduced calcium oxide crystallite size. Both these effects are beneficial in restoring the sorbents' reactivity for carbonization. Additionally, an optimized hydration-assisted reactivation protocol balances the recovered energy storage capacity with heating penalty required for moisture removal from hydrated samples, resulting in an enhanced energy storage capacity up to 176% compared to benchmark sorbents that undergo cycling without reactivation after 60 cycles. In conclusion, these results highlight the potential of hydration-assisted reactivation to enhance the long-term performance of TCES, providing an effective pathway to advancing electro-thermal storage technologies.

36 MATERIALS SCIENCE↗