2021 Annual Meeting

(8f) Open Catalyst Project: Advancements and Challenges in Building a Flexible Machine Learning Potential

Authors

Zachary Ulissi - Presenter, Carnegie Mellon University
Lowik Chanussot, Facebook AI Research
Siddharth Goyal, Facebook AI Research
Thibaut Lavril, Facebook AI Research
Abhishek Das, Facebook AI
Morgane Riviere, Facebook AI Research
Kevin Tran, Carnegie Mellon University
Javier Heras-Domingo, Carnegie Mellon University
Caleb Ho, Facebook AI Research
Weihua Hu, Stanford University
Anuroop Sriram, Facebook AI Research
Junwoong Yoon, Carnegie Mellon University
Devi Parikh, Georgia Tech and Facebook AI Research
C. Lawrence Zitnick, Facebook AI Research
Machine learning potentials have shown considerable success in accelerating quantum mechanical calculations for computational catalysis applications. These applications are often focused on a particular chemistry or narrow combination of elements, limiting their ability to screen new catalysts. The Open Catalyst Project aims to develop new ML methods and models to accelerate the catalyst simulation process for renewable energy technologies and improve our ability to predict properties across catalyst composition. The initial release of the Open Catalyst 2020 (OC20) dataset presented the largest open dataset of thermochemical intermediates across multi-metallic surfaces, spanning 55 unique elements and 80 reaction intermediates. Since then, the development of state-of-the-art models and methods has been well underway by ourselves, collaborators, and members of both the catalysis and ML communities. I will introduce the Open Catalyst Project, the associated OC20 dataset, and recent updates since our initial release. Specifically I will discuss some of the progress in building state-of-the-art models and common challenges faced along the way.