General Screening of Surface Alloys for Catalysis

Abstract

<jats:p>Intensive research in catalysis has resulted in design parameters for many important catalytic reactions; however, designing new catalysts remains difficult, partly due to the time and expense needed to screen a large number of potential catalytic surfaces. Here, we create a general, efficient model that can be used to screen surface alloys for many reactions without any quantum-based calculations. This model allows the prediction of the adsorption energies of a variety of species (explicitly shown for C, N, O, OH, H, S, K, F) on metal alloy surfaces that include combinations of nearly all of the d-block metals. We find that a few simple structural features, chosen using data-driven techniques and physical understanding, can be used to predict electronic structure properties. These electronic structure properties are then used to predict adsorption energies, which are in turn used to predict catalytic performance. This framework is interpretable and gives insight into how underlying structural features affect adsorption and catalytic performance. We apply the model to screen more than 10&lt;sup&gt;7&lt;/sup&gt; unique surface sites on approximately 10&lt;sup&gt;6&lt;/sup&gt; unique surfaces for 7 important reactions. We identify novel surfaces with high predicted catalytic performance, and demonstrate challenges and opportunities in catalyst development using surface alloys. This work shows the utility of a general, reusable model that can be applied in new contexts without requiring new data to be generated.&lt;br&gt;</jats:p>

Bayesian learning of chemisorption for bridging the complexity of electronic descriptors

“…Compared with a full quantum-mechanics treatment of many-body systems, the simplicity of physics-inspired descriptors comes at a cost of limited generalization, particularly for high-throughput materials screening. Incorporation of multifidelity site features into reactivity models with machine learning (ML) algorithms has shown early promise for the prediction of adsorption energies, with an accuracy comparable to the typical error (~0.1−0.2 eV) of density functional theory (DFT) calculations [8][9][10][11][12][13][14][15][16] . However, the approach is largely black-box in nature, prohibiting its physical interpretation.…”

When More Is Less: Nonmonotonic Trends in Adsorption on Clusters in Alloy Surfaces

“…The p-band center was calculated only including states up to the Fermi energy, while the d-band center was calculated by including states up to 0.3 eV above the Fermi energy, as in previous work. 21,24,26 The d-band peak was found by finding the energy with the maximum PDOS, after applying some mild smoothing. The center and peak energies are defined relative to the Fermi energy.…”

“…Broadly, these calculations suggest that the behavior of these systems is controlled by properties of the dopant's p-band and d-band, as suggested by previous work on transition metal alloy surfaces. 21,24,26 However, the importance of these two bands changes for different numbers of dopants.…”