Single-atom alloy catalysts (SAACs) offer a promising approach to efficient catalysis by optimizing reactant dissociation and intermediate binding. However, identifying effective SAACs is hampered by the vast number of potential candidates and the difficulty of predicting their catalytic properties. Traditional trial-and-error methods are time-consuming and inefficient. This study proposes a novel approach combining first-principles calculations with advanced data analytics to overcome this challenge. The research focuses on developing a computationally efficient and accurate predictive model for SAAC catalytic performance, enabling high-throughput screening of a large number of potential candidates. The successful development of such a model would significantly accelerate the discovery of novel and highly efficient SAACs for various industrially important reactions, reducing reliance on expensive and time-consuming experimental methods. The importance of this research lies in its potential to revolutionize the design and discovery process of SAACs, leading to more efficient and sustainable catalytic technologies.

Previous research has demonstrated the high efficiency and selectivity of SAACs in various reactions, particularly hydrogenation. Studies using techniques like desorption measurements and scanning tunneling microscopy have shown that isolated atoms of platinum group metals on inert metal hosts significantly reduce energy barriers for hydrogen uptake and desorption. However, computational approaches for SAAC design often rely on trial-and-error, hindered by synthesis challenges and experimental characterization limitations. While descriptor-based models have been explored, limitations arise from the complex, nonlinear relationships between catalyst properties and surface reactions. Existing linear correlations, such as the Brønsted-Evans-Polanyi relationship and d-band center theory, offer simplified approximations but lack the accuracy needed for reliable prediction, and often neglect the crucial aspect of catalyst stability.

This research combines first-principles calculations (DFT with RPBE exchange-correlation functional) with a compressed-sensing data-analytics method called SISSO (sure independence screening and sparsifying operator). DFT calculations provide a large dataset of hydrogen binding energies (BEH), hydrogen dissociation energies (Eb), and segregation energies (SE) for over 300 SAACs on various metal surfaces. SISSO identifies key descriptors—material features easily evaluated and correlated with catalytic properties—from a vast pool of potential candidates. These descriptors, based solely on host and guest atom properties, significantly reduce computational time compared to first-principles calculations, enabling high-throughput screening. The model's predictive power is assessed using 10-fold cross-validation. To understand the underlying mechanisms, subgroup discovery (SGD) is applied to qualitatively analyze the complex SISSO models and identify local patterns in the data that maximize a chosen quality function, providing insight into the relative importance of different features and their influence on catalytic properties. The high-throughput screening uses the computationally efficient SISSO models to evaluate thousands of SAAC candidates, considering both activity and stability criteria at different temperatures and pressures. The stability criteria are crucial in identifying practical, robust catalysts. An activity-stability map is constructed using the free energy of H adsorption and energy barriers to visualize and identify optimized candidates.

The study successfully developed accurate and reliable SISSO models for predicting BEH, Eb, and SE for SAACs. The models show strong correlation with experimental data for existing SAACs. High-throughput screening identified over 160 promising flat-surface SAACs at low temperature (200 K) and over 100 at high temperature (700 K). Several SAACs exhibiting improved stability and performance compared to experimentally known ones were discovered. The subgroup discovery method provided insights into the underlying mechanisms governing SAAC activity and stability. Analysis reveals the importance of the d-band center, cohesive energies of guest and host metals, and ionization potential of guest atoms in influencing these properties. Specifically, the SGD analysis helped clarify the role of cohesive energy differences between host and guest materials and the ionization potential of the guest atom in influencing catalyst stability. The inclusion of stability considerations, particularly segregation energy, significantly improved the accuracy of predicting catalytic trends. An activity-stability map highlights new candidates with superior activity compared to existing SAACs.

The findings address the research question by demonstrating the feasibility of using a combined first-principles and data-analytics approach for efficient and accurate prediction of SAAC catalytic properties. The high-throughput screening and the identification of superior SAAC candidates significantly advance the field of catalysis. The results highlight the power of data analytics in overcoming limitations of traditional methods and providing insights into complex catalyst-surface interactions. The identification of new, highly efficient and stable SAACs opens avenues for developing improved catalysts for various industrial applications, potentially leading to more efficient and sustainable chemical processes. The novel application of subgroup discovery provides valuable interpretability of the complex predictive models, improving our understanding of the crucial factors governing SAAC performance. This integrated approach promises a more rational and efficient design process for advanced catalysts.

This study successfully combined first-principles calculations and data analytics to develop accurate and efficient models for predicting the catalytic performance of SAACs. The high-throughput screening identified numerous promising candidates with superior properties compared to existing SAACs. The novel use of subgroup discovery provided valuable insights into the underlying mechanisms governing SAAC behavior. This integrated approach significantly advances the field of catalyst design, enabling faster and more efficient discovery of high-performance catalysts for various applications. Future research could focus on exploring other catalytic reactions, expanding the database of SAACs, and refining the data-analytics models to further enhance their predictive power and interpretability.

The accuracy of the predictive models relies on the accuracy of the DFT calculations used to generate the initial dataset. The choice of the RPBE functional might influence the results, although the study notes that main trends are likely robust to functional changes. The subgroup discovery method, while providing valuable qualitative insights, is sensitive to the choice of quality function and could reveal different patterns with alternative choices. The study focuses primarily on hydrogenation reactions; the generalizability of the models to other reaction types needs further investigation.