Machine-Learning-Augmented Chemisorption Model for CO₂ Electroreduction Catalyst Screening

Abstract

We present a machine-learning-augmented chemisorption model that enables fast and accurate prediction of the surface reactivity of metal alloys within a broad chemical space. Specifically, we show that artificial neural networks, a family of biologically inspired learning algorithms, trained with a set of ab initio adsorption energies and electronic fingerprints of idealized bimetallic surfaces, can capture complex, nonlinear interactions of adsorbates (e.g., *CO) on multimetallics with ∼0.1 eV error, outperforming the two-level interaction model in prediction. By leveraging scaling relations between adsorption energies of similar adsorbates, we illustrate that this integrated approach greatly facilitates high-throughput catalyst screening and, as a specific case, suggests promising {100}-terminated multimetallic alloys with improved efficiency and selectivity for CO2 electrochemical reduction to C2 species. Statistical analysis of the network response to perturbations of input features underpins our fundamental understanding of chemical bonding on metal surfaces.

Machine-learning prediction of the d-band center for metals and bimetals

“…[20][21][22][23][24] Considering that first-principles calculations are too time-consuming to explore the full spectrum of possibilities, and on the other hand, a great amount of data is being generated and accumulated in the field, ML methods can give a fast and high-precision alternative to the first-principles models. However, ML methods in catalysis [25][26][27][28][29][30][31][32][33][34] are still in their infancy.…”

Subsurface Oxygen in Oxide-Derived Copper Electrocatalysts for Carbon Dioxide Reduction

“…1,2 Among all pure metals, copper is the only one that gives a rich gamut of single-and multi-carbon products, [3][4][5] which can be explained by an optimal binding energy of CO to the surface. 6,7 However, the explanation becomes more complicated when dealing with nanostructured copper materials derived from oxidized precursors, which show both higher activity and greatly enhanced selectivity towards multicarbon products. [8][9][10][11][12][13][14][15][16] It has been hypothesized that this could be related for instance to an increase of local pH, [17][18][19] grain boundaries, 20 undercoordinated sites, 21 or residual oxides.…”

“…23,25 However, if the increased binding energy is caused by a simple shift of the dband in the catalyst (such as if we had nickel instead of copper), scaling relations would actually predict a higher overpotential for CO-CO-coupling on Cu(100). 7 Furthermore, an experimental study of different transition metals also suggests that an increase in CO binding energy actually lowers CO 2 RR activity. 5 In contrast, if the increased binding energy has a different origin, such an understanding could become a future strategy to overcome scaling relations.…”

Application of Artificial Neural Networks for Catalysis: A Review

“…By estimating the energy barriers using the Brønsted-Evans-Polanyi (BEP) relationship (the linear or nearly-linear relationships between the energy barrier and reaction energy under the same reaction mechanism [74]) [75,76], and predicting the binding energies of reaction intermediate species using the scaling relationship [77,78], they modeled the general trends of the catalytic activities of heterogeneous catalysts, which significantly boosted the subsequent industrial community for new catalyst design. Interestingly, they found that the heterogeneous catalytic activities always correlate well with one or two binding energies of the reaction species (e.g., CO and O for CO oxidation [73], CO for CO 2 electroduction [79], O and OH for oxygen reduction [71], and H for hydrogen evolution [72]). Using these binding energies as the reaction descriptors, a large number of new mono-and multimetallic catalysts have been discovered [80,81].…”

“…To predict the more complicated system, the alloy catalysts, Ma et al used an ANN model to screen the CO binding energies on various bimetallic models. Based on the ANN-predicted results, they found that Cu 3 Y−Ni@Cu and Cu 3 Sc−Ni@Cu had the desired CO binding energies for CO 2 electroreduction [79]. Their screening results are shown in Figure 7.…”