TY - JOUR
T1 - Augmentation of Universal Potentials for Broad Applications
AU - Pitfield, Joe
AU - Brix, Florian
AU - Tang, Zeyuan
AU - Slavensky, Andreas Møller
AU - Rønne, Nikolaj
AU - Christiansen, Mads Peter Verner
AU - Hammer, Bjørk
N1 - Publisher Copyright:
© 2025 American Physical Society.
PY - 2025/2
Y1 - 2025/2
N2 - Universal potentials open the door for DFT level calculations at a fraction of their cost. We find that for application to systems outside the scope of its training data, pretrained CHGNet [Deng et al., Nat. Mach. Intell. 5, 1031 (2023)] has the potential to succeed out of the box, but can also fail significantly in predicting the ground state configuration. We demonstrate that via fine-tuning or a Δ-learning approach it is possible to augment the overall performance of universal potentials for specific cluster and surface systems. We utilize this to investigate and explain experimentally observed defects in the Ag(111)-O surface reconstruction and explain the mechanics behind their formation.
AB - Universal potentials open the door for DFT level calculations at a fraction of their cost. We find that for application to systems outside the scope of its training data, pretrained CHGNet [Deng et al., Nat. Mach. Intell. 5, 1031 (2023)] has the potential to succeed out of the box, but can also fail significantly in predicting the ground state configuration. We demonstrate that via fine-tuning or a Δ-learning approach it is possible to augment the overall performance of universal potentials for specific cluster and surface systems. We utilize this to investigate and explain experimentally observed defects in the Ag(111)-O surface reconstruction and explain the mechanics behind their formation.
UR - https://www.scopus.com/pages/publications/85217498410
U2 - 10.1103/PhysRevLett.134.056201
DO - 10.1103/PhysRevLett.134.056201
M3 - Journal article
C2 - 39983164
AN - SCOPUS:85217498410
SN - 0031-9007
VL - 134
JO - Physical Review Letters
JF - Physical Review Letters
IS - 5
M1 - 056201
ER -