Excited to unveil OCx24, a two-year effort with @UofT and @VSParticle! We've synthesized and tested in the lab hundreds of metal alloys for catalysis. With 685 million AI-accelerated simulations, we analyzed 20,000 materials to try and bridge simulation and reality. Paper:.

RT @Spec__Tech: @jehad__abed aims to accelerate the clean energy transition by providing the community with large lab datasets to enable AI….

Huge shout out to all the amazing people that helped make this work happen - Jiheon Kim, @mshuaibii, Brook Wander, @suhasm, @johnkitchin, @tedsargentNU_TO, @SintonLab, Jason Hattrick-Simpers, @zackulissi, Larry Zitnick, @AIatMeta, @OpenCatalyst, @VSPARTICLE, @UofT.

We anticipate the availability of experimental data generated specifically for AI training, such as OCx24, will significantly improve the utility of computational models in selecting materials for experimental screening.

Evaluating the models with leave-one composition group out (LOCO) cross-validation tests their ability to predict novel compositions. The results were poorer, highlighting the gap between simple descriptors and real-world chemistry, and underscoring the need for more experimental

Evaluating CO2 reduction models is complex due to multiple products like CO, H2, CH4, and C2H4. Prediction results showed fair correlations for H2 and CO, but challenges remain in capturing full reaction complexity.

The correlation between experimental results and computation descriptors improves with training dataset size. From this analysis we project that increasing the dataset size to 10^4 or 10^5 will.allow for significantly more predictive models to be built.

Using a fitted linear model on our experimental data, we run inference on 19,406 materials, identifying a Sabatier volcano with platinum near the apex, even without Pt alloys in the training data. This analysis reveals hundreds of potential HER catalysts, many made from low-cost

By combining the experimental and computational results, we built predictive models for hydrogen evolution reactions. Models were trained to predict the cell voltage at 50 mA/cm^2 production rate using the adsorption energies of H and OH as features.

We prepared 441 samples on gas diffusion electrodes, including replicates, and evaluated them using zero-gap electrolyzers. This testing focused on CO2 conversion and hydrogen production, operating at current densities up to 300 mA/cm² similar to real-world conditions.

Each sample is characterized by X-ray fluorescence and diffraction for precise composition and structure information. We identified phases with higher purity and structural alignment to target materials using an automated XRD multiphase identification pipeline.

The dataset features 572 samples synthesized using both wet and dry methods. We used chemical reduction and spark ablation: the former reduces metal salts, while the latter fragments metal rods into nanoparticles. Both ensure the creation of a diverse training data for ML

OCx24 bridges the gap between computational models and experimental results with a dataset of diverse materials, featuring both positive and negative outcomes, tested under industrial conditions. We focus on metal alloy nanoparticles, requiring precise synthesis control.

Catalyst discovery is slow and based on trial and error, hindered by time-intensive analysis and complex reproducibility issues. AI has accelerated this, akin to advances in protein folding and language models, but often struggles to translate predictions to the real-world.

This is a continuation of our @AIatMeta Open Catalyst Project, aiming to discover new catalysts to tackle climate change. By understanding atomic interactions, we can convert CO2 into valuable chemicals, create sustainable fuels, and more.

RT @bwood_m: Our team at FAIR is looking for research interns in 2025. We work on a range of AI for chemistry topics from applied projects….

RT @VGharakhanyan: One of the largest materials datasets and SOTA ML potentials are open-sourced by FAIR Chemistry. Immensely proud to be p….

RT @SamMBlau: What an incredible resource for the community! While Google and Microsoft report on closed-source models trained on closed-so….

Proud to be part of the team with a mission to open source science. Can’t wait to see what the community build with this 🚀🚀.

RT @mshuaibii: Excited to share OMat24 - the latest dataset to join our team's family of open datasets: OC20, OC22, ODAC23. Accelerating pr….