The expense of quantum chemistry calculations hinders thorough large-scale catalyst screening studies. As the number of studied systems increases, the number of needed calculations scales. This is especially problematic when investigating the differing kinetics of catalyzed processes on numerous materials compositions in varying reaction environments. To address this problem, we are assessing the performance of computational alchemy, a method that has promise in accelerating catalyst discovery.

With this method, we can efficiently approximate descriptors for hundreds of catalysts using information from a single set of DFT calculations. In the first publication, we benchmark computational alchemy in predicting oxygen reduction reaction intermediate binding energies (BEs) on transition metal alloys and explain the accuracies and limitations of this method. We also investigated, in a second publication, this method’s performance with carbide, nitride, and oxide catalysts. There we found that computational alchemy is most reliable with metallic systems. Our current work includes developing open-source Python packages to let this method easily be used among others in the computational catalysis community, improving areas with shortcomings, and applying computational alchemy to new systems.


Lingyan Zhao, Charles Griego, Karthi Saravanan

Saravanan et al., J. Phys. Chem. Lett., 2017, 8, 5002-5007, DOI: 10.1021/acs.jpclett.7b01974.
Griego et al., Adv. Theory Simul., 2019, 2, 1800142, DOI: 10.1002/adts.201800142.

We are currently pursing error correction of alchemy with supervised machine learning.