Open Catalyst Project (opencatalystproject.org)

The Open Catalyst Project, a collaboration between Meta’s Fundamental AI Research (FAIR) and Carnegie Mellon University’s Chemical Engineering department, has released large open datasets and tools to accelerate AI-driven catalyst discovery for renewable-fuel production. The project published the OC20 and OC22 datasets—together containing about 1.3 million molecular relaxation trajectories derived from over 260 million density functional theory (DFT) calculations—and open-sourced baseline models, code, and a public leaderboard/evaluation server to let the community train and benchmark ML models that approximate expensive quantum simulations. This matters because DFT-level simulations, while accurate, are computationally costly and limit how many candidate catalyst structures can be screened for reactions like CO2 reduction or hydrogen generation. ML models trained on OC20/OC22 can predict energies, forces, and relaxation pathways orders of magnitude faster, enabling high-throughput screening (for example, modelling CO* adsorption on Zr sites in Zr3Sc) and guiding experimental prioritization. By lowering the computational barrier, the project aims to shorten design cycles, help find lower-cost, high-activity catalysts at scale, and democratize research through standardized datasets and benchmarks—while still requiring careful validation of ML predictions against quantum and experimental results.