30-Jan-2024
Happy new year! We hope you had a great 2023 and wish you a pleasant and productive 2024. In this KIM quarterly update, we would like to inform you of some recent developments in OpenKIM and related projects, as well as some upcoming events.
Recent Developments
- We are excited to introduce ColabFit, a new project under the KIM Initiative that provides open access to curated and standardized first-principles datasets suitable for fitting IPs. The database can be accessed at https://colabfit.org/ and already contains 337 datasets! In addition to using the datasets available, we encourage you to submit or recommend datasets for ingestion.
- KIM REVIEW, a journal containing commentaries and discussion on seminal papers in molecular simulation, has been launched at https://kimreview.org/. Readers are encouraged to discuss the published commentaries at https://matsci.org/kimreview/. See the announcement here: https://openkim.org/news/2022-08-01/.
- We have developed Deep Citation, a machine learning-based framework that finds and analyzes all papers citing the primary source of an interatomic potential to determine whether the potential was used or merely cited for background. It is now integrated into every model page on OpenKIM. For more information, see https://openkim.org/doc/usage/deep-citation/.
- We have switched to Conda as the primary distribution platform for KIM API binaries. The kim-api package on conda-forge has been updated to be able to install individual models, so Conda users can use any model as soon as it is live on OpenKIM.org, as well as being able to install custom models locally. For more details, see https://openkim.org/news/2024-01-29/.
- The KIM Developer Platform (KDP), a Docker image for developing KIM models and tests, has been updated to version 1.3.0. For more information see https://openkim.org/doc/evaluation/kim-developer-platform/.
- The KDP is the basis of the new OpenKIM Binder sandbox, accessible here or from the OpenKIM.org front page. You can use it to follow the included tutorials and experiment with OpenKIM features in a pre-configured environment in your browser!
- OpenKIM values the developers of OpenKIM content, and we have overhauled our developer attribution metadata. You can now browse models by developer at https://openkim.org/browse/models/by-developer. If you have developed a model that is hosted on OpenKIM, you can claim your developer profile and edit it by clicking on your name in the above link.
Upcoming Events
- OpenKIM is co-hosting a symposium at the 2024 MACH Conference titled "Systematic Discovery and Characterization of Materials Across Compositions and Structures". Abstract submission is closed, but we encourage you to attend nonetheless. The conference is being held on Apr. 3-5 in Baltimore, MD. For more information, see https://machconference.org/about/topics/. Stay tuned to OpenKIM or MACH announcements for registration information.
- OpenKIM and ColabFit will be giving a short course titled "Training and Deploying Physics-Based and Machine Learning Interatomic Potentials for Advanced Materials Applications" at the 2024 World Congress on Computational Mechanics (WCCM). The Congress is taking place in Vancouver, BC on July 21-26. A description can be found here: https://www.wccm2024.org/W24-11. You can register for the short course alongside your conference registration, see https://www.wccm2024.org/fees.
Coming Soon
- We are actively developing Crystal Genome, an upgrade to the KIM Property Testing Framework that will cover ALL known crystal structures, categorized using the AFLOW prototype label (http://aflow.org/).
- We are introducing support for bonded force fields such as IFF into OpenKIM by leveraging the new LAMMPS Type Labels feature (https://docs.lammps.org/Howto_type_labels.html).
- As part of an integrated machine learning push involving Colabfit and the KIM-based Learning-Integrated Fitting Framework (KLIFF), (https://github.com/openkim/kliff) we are developing a model driver that can support ANY machine learning model that can be expressed in TorchScript.
How to stay up-to-date