TY - JOUR
T1 - When do short-range atomistic machine-learning models fall short?
AU - Yue, Shuwen
AU - Muniz, Maria Carolina
AU - Andrade, Marcos F.Calegari
AU - Zhang, Linfeng
AU - Car, Roberto
AU - Panagiotopoulos, Athanassios Z.
N1 - Funding Information:
The authors would like to thank Thomas Gartner III and Pablo Piaggi for insightful discussions and feedback on the manuscript, Marc Riera for important instructions on the second virial coefficients calculation, and Michael Webb for helpful guidance on the normal mode calculation. This work was supported as part of the Computational Chemical Sciences Program funded by the U.S. Department of Energy, Office of Science, Basic Energy Sciences, under Award Nos. DE-SC0002128 and DE-SC001934, Chemistry in Solution and at Interfaces (CSI) Center. The computing resources were provided by the Princeton Institute for Computational Science and Engineering.
Publisher Copyright:
© 2021 Author(s).
PY - 2021/1/21
Y1 - 2021/1/21
N2 - We explore the role of long-range interactions in atomistic machine-learning models by analyzing the effects on fitting accuracy, isolated cluster properties, and bulk thermodynamic properties. Such models have become increasingly popular in molecular simulations given their ability to learn highly complex and multi-dimensional interactions within a local environment; however, many of them fundamentally lack a description of explicit long-range interactions. In order to provide a well-defined benchmark system with precisely known pairwise interactions, we chose as the reference model a flexible version of the Extended Simple Point Charge (SPC/E) water model. Our analysis shows that while local representations are sufficient for predictions of the condensed liquid phase, the short-range nature of machine-learning models falls short in representing cluster and vapor phase properties. These findings provide an improved understanding of the role of long-range interactions in machine learning models and the regimes where they are necessary.
AB - We explore the role of long-range interactions in atomistic machine-learning models by analyzing the effects on fitting accuracy, isolated cluster properties, and bulk thermodynamic properties. Such models have become increasingly popular in molecular simulations given their ability to learn highly complex and multi-dimensional interactions within a local environment; however, many of them fundamentally lack a description of explicit long-range interactions. In order to provide a well-defined benchmark system with precisely known pairwise interactions, we chose as the reference model a flexible version of the Extended Simple Point Charge (SPC/E) water model. Our analysis shows that while local representations are sufficient for predictions of the condensed liquid phase, the short-range nature of machine-learning models falls short in representing cluster and vapor phase properties. These findings provide an improved understanding of the role of long-range interactions in machine learning models and the regimes where they are necessary.
UR - http://www.scopus.com/inward/record.url?scp=85099774786&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85099774786&partnerID=8YFLogxK
U2 - 10.1063/5.0031215
DO - 10.1063/5.0031215
M3 - Article
C2 - 33499637
AN - SCOPUS:85099774786
SN - 0021-9606
VL - 154
JO - Journal of Chemical Physics
JF - Journal of Chemical Physics
IS - 3
M1 - 034111
ER -