Topic: Knowledgebase of Interatomic Models (KIM)
Minnesota Team: Daniel Karls, Mingjian Wen, John Hooper, Adam Ludvik, Ryan Elliott, Ellad Tadmor
Cornell Team: Matt Biebaum, Junhao Li, James Sethna
Collaboration: Chandler Becker (NIST), Noam Bernstein (NRL), Ronald Miller (Carleton), Jakob Schiøtz (DTU), Steve Stuart (Clemson)
Funding: National Science Foundation (CDI and CDS&E programs)
Description: The Knowledgebase of Interatomic Models (KIM) is a cyberinfrastructure located at https://openkim.org, which is associated with the collective endeavor of the computational materials community to rationalize, standardize, and characterize interatomic models (IMs). An IM is a parameterized functional form used to approximate the energy of a collection of atoms. Molecular dynamics (MD) using IMs is applied routinely to simulate millions and even billions of atoms on today's parallel supercomputers for technological applications addressing many of the world's pressing problems (including climate change, energy generation and distribution, world health, and terrorism). The KIM project seeks to address the main limitations of MD (and other molecular simulation techniques), which include a lack of ability to reproduce simulations since IMs have traditionally been distributed in an uncontrolled fashion; a difficulty in reusing IMs because each molecular simulation tool has a different way of implementing IMs; and an inability to select rigorously the appropriate IM for a given application. As such, the KIM project has the following main objectives: (1) long-term archival storage of IMs; (2) development of an application programming interface (API) for IMs allowing them to work seamlessly with any simulation code supporting the API; (3) development of measures for assessing IM transferability (i.e. the ability of an IM to describe behavior it was not fitted to reproduce).
The KIM project was originally funded through the NSF Cyber-enabled Discovery and Innovation (CDI) program and has recently been renewed through the NSF Computational and Data-Enabled Science (CDS&E) program. Led by Ellad B. Tadmor (PI) and Ryan S. Elliott at the University of Minnesota and James P. Sethna at Cornell University, the project has seriously engaged roughly a hundred investigators in the standards development process, and over fifty investigators in adapting nearly three hundred interatomic models to the KIM API standard. KIM currently has about 400 members in the US and internationally. The KIM team has developed a technologically sophisticated web portal, a processing pipeline for automatically applying diagnostic and predictive KIM tests, and visualization tools to allow users to rapidly and realistically assess the relative speed and utility of a variety of possible IMs for their application. Reproducibility of atomistic simulations is made possible by the introduction of unique "KIM IDs" for all KIM content. All content on KIM is open source. The KIM API is supported by a growing number of major molecular simulation codes including LAMMPS, IMD, DL_POLY, ASE/ASAP, and GULP.