Speaker
Description
Computational modeling and the simulation of catalytic processes, in particular of reactions like water electrolysis to produce hydrogen and CO or CO$_2$ hydrogenation to produce alkanes or higher oxygenates, have become key ingredients in developing new catalytic materials for the sustainable generation of synthetic fuels. The calculation of relevant quantities such as adsorption enthalpies and reaction barriers is often carried out using density functional theory (DFT). However, the computational cost of DFT remains a limiting factor, rendering the large-scale exploration of new catalysts difficult. Replacing the costly DFT calculations with computationally cheaper machine learning (ML) models is becoming increasingly popular in recent years. This approach has distinct advantages over fitting semi-empirical potentials such as the flexible tailoring of models and training data sets to arrive at the desired balance between cost and accuracy.
In recent work, we have focused on developing data-efficient and physically motivated ML models for the prediction of adsorption enthalpies $^{[1,2]}$. Given the lack of accurate descriptors for the reactivity of transition metal oxides (TMOs) toward water electrolysis, we employed a data-driven compressed sensing method $^{[3]}$ to explore representative multidimensional descriptors expressed as nonlinear functions of intrinsic properties of the clean surface. Our descriptors largely outperform previously highlighted descriptors in terms of accuracy and computational cost, and properties related to local charge transfer are identified to be crucial for the screening of promising TMO catalysts. Concerning the more complex adsorbates involved in CO or CO$_2$ hydrogenation with numerous mono- and multi-dentate adsorption modes, we propose a kernelized ML model with a physics-inspired graph representation, enabling prediction quality that is en par with DFT. Its good extrapolation ability and built-in uncertainty estimate to reliably capture outliers makes it promising for comprehensively exploring complex reaction networks on novel catalysts.
[1] W. Xu et al., ACS Catal. 11, 2, 734-742 (2021)
[2] W. Xu et al. (in preparation)
[3] R. Ouyang et al., Phys. Rev. Mater. 2, 083802 (2018)
