Beyond scaling relations for the description of catalytic materials (1902.07495v1)

Abstract: Computational screening for new and improved catalyst materials relies on accurate and low-cost predictions of key parameters such as adsorption energies. Here, we use recently developed compressed sensing methods to identify descriptors whose predictive power extends over a wide range of adsorbates, multi-metallic transition metal surfaces and facets. The descriptors are expressed as non-linear functions of intrinsic properties of the clean catalyst surface, e.g. coordination numbers, d-band moments and density of states at the Fermi level. From a single density-functional theory calculation of these properties, we predict adsorption energies at all potential surface sites, and thereby also the most stable geometry. Compared to previous approaches such as scaling relations, we find our approach to be both more general and more accurate for the prediction of adsorption energies on alloys with mixed-metal surfaces, already when based on training data including only pure metals. This accuracy can be systematically improved by adding also alloy adsorption energies to the training data.

Essay on "Beyond Scaling Relations for the Description of Catalytic Materials"

The paper by Andersen, Levchenko, Scheffler, and Reuter presents a significant advancement in the computational screening of catalytic materials by applying compressed sensing techniques to predict adsorption energies. The paper explores the identification of multi-dimensional descriptors that improve the reliability and accuracy of adsorption energy predictions across diverse adsorbates and transition metal surfaces, surpassing the capacities of traditional scaling relations.

In the field of catalysis, accurate prediction of adsorption energies is crucial because it directly influences the effectiveness and efficiency of catalytic reactions. The authors employ the SISSO (sure independence screening and sparsifying operator) method to determine powerful multidimensional descriptors that cover a wide range of transition metal surfaces and facets. This method leverages properties such as $d$-band moments and the density of states at the Fermi level derived from density-functional theory (DFT) calculations of clean catalyst surfaces. Notably, this approach enables predictions without necessitating closed-form physical equations, thus encapsulating complex material properties that scaling relations typically approximate.

The research articulates that traditional scaling relations, while useful, are limited by their linear nature and assumptions, resulting in greater prediction errors, particularly for alloys with mixed-metal configurations. The SISSO-derived descriptors, conversely, exhibit low root-mean-square errors (RMSEs) and substantial generality, demonstrating superior accuracy even when predicting on facets not included in the training dataset. The authors substantiate the superior performance of their 8D descriptor derived from SISSO ($\mathbf{\Phi}_3$ feature space), highlighting a significant reduction in RMSE compared to established scaling relations.

An essential contribution of the paper is the demonstration of the predictive capabilities of SISSO-based models to identify outliers to conventional scaling relations. This feature opens pathways to discovering promising catalytic materials that might have been overlooked using traditional methods. Furthermore, the paper emphasizes the negligible computational cost associated with applying these descriptors, thus facilitating large-scale screenings across material spaces for catalytic candidates.

From a practical standpoint, the implications of this research extend toward more efficient and targeted exploration of catalytic materials, particularly in identifying alloys with potentially enhanced catalytic properties. Theoretically, it integrates machine learning approaches into the domain of materials science, signaling a shift toward data-driven paradigms in catalyst design.

In conclusion, this paper showcases a robust method for enhancing the predictive accuracy of adsorption energies, transcending the limitations of scaling relations. By integrating SISSO in the identification of multi-dimensional descriptors, this paper injects a new level of precision and efficiency into the computational screening of catalytic materials. The methodologies and findings presented may be instrumental in driving future advancements in the field, ultimately contributing to the discovery of innovative catalysts with optimized properties for industrial applications. Future research can expand this framework to encompass a broader range of adsorption sites and material systems, further broadening its applicability in catalytic science.