Accurate and efficient machine learning models for predicting hydrogen evolution reaction catalysts based on structural and electronic feature engineering in alloys

Abstract

Predictive materials design of high-performance alloy electrocatalysts is a grand challenge in hydrogen production via the water electrolysis. The vast combinatorial space of element substitutions in alloy electrocatalysts offers a...

Analyzing the Active Site and Predicting the Overall Activity of Alloy Catalysts

“…As is well-known, the features play a crucial role in determining the predictive accuracy of ML . Two major types of widely used features can be identified: geometric structure features, including coordination numbers, bond angles, bond lengths, elemental composition of specific sites, and electronic structure features, such as electron affinity, electronegativity, and the first ionization energy of specific atoms. ,− Therefore, it is imperative to engage in feature selection and data set construction before using most ML methods. Additionally, scientists aim to capture accurate structural information on active sites by designing diverse structural features according to their types, including top, bridge, and hollow sites on the (111) crystal facet of a face-centered cubic (FCC) lattice or M-N 4 single-atom sites .…”

“…8 Zhang et al accurately predicted the HER performance of the 2290 alloy using a feature matrix consisting of electronic and geometrical information. 9 Especially, the crystal graph convolutional neural network (CGCNN) 10 model, which utilizes graph theory to describe crystal structure features, has been successfully applied to the screening of alloy catalysts. 11 It is worth noting that the aforementioned studies all consider intermetallic compounds in material databases as potential screening targets.…”

“…For instance, Tran and Ulissi employed an active learning strategy and identified hundreds of promising alloy surfaces for CO 2 reduction reaction (CO 2 RR) and hydrogen evolution reaction (HER) . Zhang et al accurately predicted the HER performance of the 2290 alloy using a feature matrix consisting of electronic and geometrical information . Especially, the crystal graph convolutional neural network (CGCNN) model, which utilizes graph theory to describe crystal structure features, has been successfully applied to the screening of alloy catalysts .…”

Local descriptors-based machine learning model refined by cluster analysis for accurately predicting adsorption energies on bimetallic alloys

“…44–46 Among the most relevant local chemical descriptors proposed to date, those focused on correlating the nearest neighboring surface atoms to the adsorption site have gained significant attention. These descriptors include the average of elemental properties, 47 atom-specific fingerprints derived from elemental properties, 48,49 and use ML models based on neural networks, 50–52 or graph neural networks. 53–56 However, these descriptors have difficulties adapting to other adsorbates or ML-based model architectures, hindering their transferability to other systems.…”

Identifying Stable Electrocatalysts Initialized by Data Mining: Sb2WO6 for Oxygen Reduction

“…Data mining from the open‐source computational databases can contribute to the search of unexplored promising materials (e.g., thermoelectric materials, [ 16 ] optoelectronic semiconductors, [ 17 ] and batteries [ 18 ] ), which could also be a way‐out to effectively search promising catalysts. [ 19 , 20 ] The Materials Project database, [ 21 ] the largest computational bulk material database reported to date, has included various physical properties of >140 000 materials, among which the thermodynamic phase diagrams [ 22 , 23 , 24 ] and bulk Pourbaix phase diagrams [ 25 , 26 , 27 ] can help assess the thermodynamic and aqueous stability of potential electrocatalysts.…”