Ensuring Reliability, Reproducibility and Transferability In Atomistic Simulations: The Knowledgebase of Interatomic Models (openKIM.org)

Atomistic simulations in materials science play a key role in realistic scientific and industrial applications. However, the predictive capability of these approaches hinges on the accuracy of the model (potential or force field) used to describe the atomic interactions.  Modern models are optimized to reproduce electronic structure results for a dataset of representative atomic configurations. However, no standardized approach currently exists for quantifying the range of applicability of an interatomic model or estimating the accuracy of its predictions. This makes it difficult or even impossible to select an appropriate model for a given application. A second difficulty is that since computer implementations of interatomic models do not subscribe to coding standards, it is hard to connect a given model to a desired simulation tool.  This talk will describe a current NSF-CDI funded effort to create an open source online tool that addresses these limitations: the Knowledgebase of Interatomic Models (http://openKIM.org).  KIM will allow users to compare model predictions with reference data, to generate new predictions by uploading simulation test codes, and to download models conforming to application programming interface (API) standards which are being developed in collaboration with atomistic simulation community. Inthis talk I will give an overview of the KIM project and describe recent developments.