large atomistic models (LAMs), material, crystal structure, AI for science, Computer engineering. Computer hardware, TK7885-7895, Electronic computers. Computer science, QA75.5-76.95

Inspired by the success of large language models, the development of large atomistic models (LAMs) has rapidly gained momentum in scientific computation. Since 2022, the Deep Potential team has been actively pretraining LAMs and launched the OpenLAM Initiative, which advocates for creating a community-driven platform aimed at accelerating the development of open-source foundation models by sharing curated datasets, algorithms, and relevant workflows. As part of the OpenLAM Initiative, the OpenLAM Challenges are competitions designed to benchmark atomic modeling methods, encourage community collaboration, and accelerate advances in machine learning-driven scientific discovery. The LAM Crystal Philately competition, as an example, aims to construct an open-source database of crystal structures by collecting unique configurations with arbitrary chemical compositions. During the competition, structures submitted by participants are validated by an LAM based on energy and force criteria, and their stabilities are assessed using the OpenLAM convex hull derived from all structures within the database. The first round of the LAM Crystal Philately competition has collected over 19.8 million valid structures, including approximately 350 000 on the OpenLAM convex hull, driving advancements in generative modeling and materials science applications.