(342aj) Predicting Rate Constants for Catalyzed Reactions with Machine Learning

Catalyzed reactions lie at the core of our fight against climate change, from their use to provide new renewable energy sources to the removal of CO2 from the atmosphere. However, computing catalytic reactivity atomistically is unfeasible for many reactions of interest as computational cost scales exponentially with system size [1]. Supervised machine learning has been proposed as a method to increase the speed with which predictions of reaction rates can be made [2]. However, one of the critical roadblocks to developing useful machine learning models to predict reactivity is developing a dataset to train the model with [3]. We are currently generating a dataset of reaction rate constants for ground state organic chemistry reactions using density functional theory CI-NEB calculations of minimum energy paths and aim to extend it to include reactions in the presence of catalysts. In this context, we are working to design input features which come at a low computational cost yet include information on the dynamics of the reaction and lead to a low loss in prediction. Here we will show our recent work on identifying and designing these features using our current dataset and previously published datasets [2, 4].

References

[1] Friesner, R. A. Ab initio quantum chemistry: Methodology and applications. PNAS 10, 6648–6653 (2005)

[2] Komp, E. & Valleau, S. Machine Learning Quantum Reaction Rate Constants. J. Phys. Chem. A 124, 8607–8613 (2020).

[3] Goldsmith, B. R., Esterhuizen, J., Liu, J. X., Bartel, C. J. & Sutton, C. Machine learning for heterogeneous catalyst design and discovery. AIChE J. 64, 2311–2323 (2018).

[4] Winther, K. T., Hoffmann, M. J., Boes, J. R., Mamun, O., Bajdich, M., & Bligaard, T. (2019). Catalysis-Hub.org, an open electronic structure database for surface reactions. Scientific Data, 6, 1–10 (2019).