(174h) Understanding the Segregation Energy Behavior of Single Atom Alloys in the Presence of Ligands.

Over the years, single atom alloys (SAAs) have received tremendous attention due to their tunable properties that allow for improved catalytic performance, such as high activity and selectivity. The stability of SAAs is dictated by the formation of an isolated dopant on the surface of the alloy (i.e., surface segregation and aggregation). Additionally, surface segregation is governed by many factors¹, one of which is the presence of an adsorbate. Current studies focused on investigating the effect of CO and H on the stability of SAAs. Despite these advances in the SAA field, there is still a knowledge gap in the literature in understanding of how ligands used in colloidal nanoparticle (NP) synthesis affect metal segregation in the NP surface. In this work, we used Density Functional Theory (DFT) to investigate how ligands, such as H₃C-NH₂ and H₃C-S, affect the stability of SAAs. We also extended our study to include H₃C-NH to understand how a reaction intermediate (from H₃C-NH₂ dissociation) can affect surface segregation. Then, we applied machine learning to develop a model that can efficiently screen through different SAA catalysts in the presence of adsorbates. Collecting a gamut of DFT-calculated segregation energy (E_seg) in the presence of adsorbates data, we developed an accurate four-feature neural network: multilayer perceptron regression (NN MLP) model on d⁸- (Pt, Pd, Ni) and d⁹- (Ag, Au, Cu) based SAAs on low-index surfaces such as (111) and (100). The model captures the underlying physics behind surface segregation in the presence of adsorbates by incorporating features that describe the thermodynamic stability (metal bulk cohesive energy) while accounting for the coordination number of the dopant, adsorbate effects (binding strength of the adsorbate to the metals and the coordination number of the adsorbate), strain effects (Wigner-Seitz radius), and electronic effects (charge transfer), such as the electron affinity. We found that the adsorption configuration and the binding strength of the adsorbate to the SAA surface alter the E_seg trends. Furthermore, we developed an accurate NN MLP model that predicts E_seg in the presence of adsorbates to find thermodynamically stable SAAs, eliminating trial-and-error in experimentation and minimizing the use of computationally expensive DFT. Overall, this work allows for accelerated predictions across a large space of different SAA metal combinations in the presence of adsorbates.

1. Salem, M.; Cowan, M. J.; Mpourmpakis, G. Predicting Segregation Energy in Single Atom Alloys Using Physics and Machine Learning. ACS Omega 2022, 7 (5), 4471–4481. https://doi.org/10.1021/acsomega.1c06337.