A new article by Mingjian Wen, Stephen M. Whalen, Ryan S. Elliott and Ellad B. Tadmor has been published in Modeling and Simulation in Materials Science and Engineering on interatomic models that include functions represented as tables of discrete values with spline interpolation. It is shown that the nature of the interpolation affects model predictions. This highlights the importance of archiving the interatomic model implementation along with model parameters as done in OpenKIM.
From the abstract:
"Empirical interatomic potentials are widely used in atomistic simulations due to their ability to compute the total energy and interatomic forces quickly relative to more accurate quantum calculations. The functional forms in these potentials are sometimes stored in a tabulated format, as a collection of data points (argument—value pairs), and a suitable interpolation (often spline-based) is used to obtain the function value at an arbitrary point. We explore the effect of these interpolations on the potential predictions by calculating the quasi-harmonic thermal expansion and finite-temperature elastic constant of a one-dimensional chain compared with MD simulations. Our results show that some predictions are affected by the choice of interpolation regardless of the number of tabulated data points. Our results clearly indicate that the interpolation must be considered part of the potential definition, especially for lattice dynamics properties that depend on higher-order derivatives of the potential. This is facilitated by the Knowledgebase of Interatomic Models (KIM) project, in which both the tabulated data (“parameterized model”) and the code that interpolates them to compute energy and forces (“model driver”) are stored and given unique citeable identifiers. We have developed cubic and quintic spline model drivers for pair functional type models (EAM, FS, EMT) and uploaded them to the OpenKIM repository (https://openkim.org)."
The full article is available here.