(509dh) Predicting Segregation Energy in Single Atom Alloys Using Physics and Machine Learning

Over the years, single-atom alloys (SAAs) have emerged as promising heterogeneous catalysts for a broad range of reactions due to their tunable properties that allow for enhanced efficiency and selectivity. Understanding the thermodynamically preferred doping site of a heterometal atom (surface vs. bulk) is imperative to designing stable SAA catalysts. Surface segregation energy (E_seg) is a common metric used to probe the thermodynamic stability of SAAs. Typically, E_seg is determined with density functional theory (DFT) calculations; however, DFT is computationally expensive and time-consuming. To overcome this issue, we developed a model that accurately predicts E_seg in bulk systems (i.e. periodic slabs) using tabulated elemental features, insights from the Bond-Centric Model¹ (model that describes stability of bimetallic particles), and machine learning techniques. Our initial model was trained to capture E_seg for SAAs on (111), (100), (110), and (210) surface facets from FCC hosts and dopants. We then extended our training set to develop a model that captures E_seg of FCC SAAs with FCC, BCC, and HCP dopant metals. The extended model accurately captures the interactions between the metal host and dopant, including electronic and geometric effects. Lastly, we tested our developed model against Ir- and Pd- based SAA cuboctahedral nanoparticles, varying in size and heterometal dopant. We found that the model accurately predicts the E_seg of nanoparticles, supporting its wide application from the nanoscale to bulk. Overall, our model enables rapid screening of the broad SAA materials space towards improved catalyst design.

1. Yan, Z.; Taylor, M. G.; Mascareno, A.; Mpourmpakis, G., Size-, Shape-, and Composition-Dependent Model for Metal Nanoparticle Stability Prediction. Nano Lett. 2018, 18 (4), 2696-2704.