Introducing the AQCat25-EV2 Models
Webinar
Accelerating Materials Discovery: Launching AQCat25-EV2
Co-Presented with NVIDIA
A New Era in AI-Accelerated Catalyst Innovation
AQCat25-EV2 is the world’s first family of machine learning models for heterogeneous catalysis that incorporates spin polarization and covers every industrially relevant element. Trained on the AQCat25 dataset (with 13.5 million high-fidelity DFT calculations), the AQCat25-EV2 models deliver DFT-level accuracy at up to 20,000X the speed, making high-throughput AI-accelerated virtual screening practical across industry applications for the first time.
Powering the Next Industrial Revolution
Catalysis drives the global economy, from the fuels that power our world to the materials that shape it. With AQCat25-EV2, industries can now simulate, screen, and optimize catalysts virtually with physics-based accuracy, unlocking performance and sustainability breakthroughs at unprecedented scale.
Endless applications in materials innovation:
Examples: Accelerate CO₂-to-fuels, Water splitting, fuel cells, and methane-to-methanol conversion.
Materials
Examples: syngas-to-ethanol, and next-generation ammonia synthesis, depolymerization and circular-economy chemistry.
Mobility
Examples: Cleaner combustion, advanced fuels, and exhaust-gas conversion.
Consumer Goods
Examples: sustainable pathways for fertilizers and commodity chemicals.
AQCat25-EV2 de-risks R&D pipelines across all major sectors dependent on catalysis innovation; transforming discovery from an art of iteration to a science of prediction.
Quantum Accuracy at Machine Learning Speed
Trained on High Fidelity Data: Explicitly includes magnetic effects (Fe, Ni, Co etc.) and improved DFT settings for realistic energetics across the periodic table.
Mitigate Catastrophic Forgetting: Integrates AQCat25 with Meta FAIR’s OC20 dataset to accurately represent spin-polarized catalyst materials without sacrificing generalizability.
Architectural Modifications to Maximize Performance: Conditioning EquiformerV2 models with spin and fidelity flags improves model accuracy and generalizability.
Improved Data Generation and Model Training Efficiency: Minimizing data redundancy enables rapid prototyping and testing of different architectures.
Breakthrough AI Trained on Cutting-Edge Data
The AQCat25-EV2 family of machine learning interatomic potentials (MLIPs) was trained on the previously-released AQCat25 dataset. AQCat25 is a large‑scale, publicly available dataset of 13.5 million density functional theory (DFT) calculations covering 47,000 intermediate-catalyst systems. Built to train large quantitative models (LQMs), AQCat25 brings unprecedented accuracy within this domain to a new design space, making in-silico discovery and optimization of new catalysts feasible at scale.
13M+
Data Points
47K
Adsorbate slab pairs
500K
GPU hours to compute
Unlike prior datasets, AQCat25:
- Explicitly models spin polarization, which is critical to correctly model metals such as Fe, Ni, and Co.
- Raises the plane‑wave cutoff to 500 eV. This gives superior accuracy in energetics.
- Captures 20 transition state structures to enable calculation of reaction rates.
- Adds six new elements (Ba, Ce, F, Li, La, Mg) not included in other heterogeneous catalyst datasets, which unlocks exploration of new, important potential catalysts.
Together, these key features provide reliable reaction and barrier energies across a broader chemical space, enabling fast and accurate screening and design of catalysts.
Build with AQCat25 Model and Data
The AQCat25 Dataset and AQCat25-EV2 model are available free for non‑commercial research under the CC BY-NC-SA 4.0 license on Hugging Face. For commercial applications or to explore partnership opportunities to discover novel catalysts, please contact us directly. Our team will be more than happy to assist with licensing, support, and integration.
For more information:
Catalysts are Critical for Turning Raw Materials into Value
More than 80% of all manufactured goods globally and almost all (>90%) chemicals produced worldwide rely on catalysis in their production. Better catalysts reduce energy use, cut emissions, and unlock new pathways for critical materials, including green hydrogen, sustainable aviation fuel, and more efficient ammonia for food production.
AQCat25 dataset and AQCat25-EV2 models provide the depth and breadth needed to model these processes accurately, allowing R&D teams to screen candidates in silico before committing to costly synthesis and testing.
rely on catalysis in their production
Contact us about AQCat25
If you’re interested to learn more about the AQCat25 dataset and models, or to see how they might be expanded to include use cases of special interest to your business, we’d love to hear from you. Contact us today to talk to one of our experts.