Activity Origin and Design Principles for Oxygen Reduction on Dual-Metal-Site Catalysts: A Combined Density Functional Theory and Machine Learning Study

Abstract

Dual-metal-site catalysts (DMSCs) are emerging as a new frontier in the field of oxygen reduction reaction (ORR). However, there is a lack of design principles to provide a universal description of the relationship between intrinsic properties of DMSCs and the catalytic activity. Here, we identify the origin of ORR activity and unveil design principles for graphene-based DMSCs by means of density functional theory computations and machine learning (ML). Our results indicate that several experimentally unexplored DMSCs can show outstanding ORR activity surpassing that of platinum. Remarkably, our ML study reveals that the ORR activity of DMSCs is intrinsically governed by some fundamental factors, such as electron affinity, electronegativity, and radii of the embedded metal atoms. More importantly, we propose predictor equations with acceptable accuracy to quantitatively describe the ORR activity of DMSCs. Our work will accelerate the search for highly active DMSCs for ORR and other electrochemical reactions.

“More is Different:” Synergistic Effect and Structural Engineering in Double‐Atom Catalysts

“…combined comprehensive DFT calculations, microkinetic modeling, and ML to scrutinize possible ORR descriptors for TM 2 ‐N 6 ‐C DACs (Figure 11d). [ 52a ] Using DFT calculations, they found a linear scaling relationship between Δ G OH* and Δ G OOH* (Figure 11e) and also found that Δ G OH* can act as an independent descriptor to describe the ORR activity of DACs. Nevertheless, it would be more intuitive if the ORR activity for DACs can be predicted even prior to the intensive DFT calculations because of its high computational cost.…”

“…Owing to the relatively low computational cost, microkinetic modeling can open up new pathways for analyzing experimental results and provide predicting power; thus, it can act as a powerful tool for probing the catalytic activity of DACs. [ 52 ]…”

“…[ 53b ] In a pioneering work, Zhu and co‐workers applied the gradient boosted regression (GBR) ML algorithm to investigate the possible fundamental factors governing the catalytic performance of TM 1 /TM 2 ‐N 6 ‐C DACs for ORR. [ 52a ] The data set comprised fundamental parameters for TM atoms, such as pauling electronegativities, first ionization energies, and the number of d electrons. After evaluating these key features using the GRB algorithm, precise predictions of Δ G OH* (a descriptor for ORR) from ML, which is in good agreement with the DFT‐calculated results, were achieved.…”

“…[ 34,83 ] Metal dimers supported on 2D materials also consist of a large portion of DACs, but currently, experimental evidence is lacking. Recent theoretical investigations fill this gap by predicting that Co–Co on C 2 N, [ 84 ] Cu–Fe, Ni–Cu, [ 52a ] Co–Pt, [ 89 ] Fe–Co, [ 85 ] and Co–Zn [ 86 ] on NG and hydroxyl‐group‐modified metal (Ni, Co, and Fe) dimers on defected graphene [ 87 ] are potential candidates for ORR DACs. C 2 N monolayers with a size of 0.83 nm and well‐ordered vacancies terminated by sp 2 ‐bonded nitrogen atoms can provide enough space to hold metal dimers and are thus promising for constructing stable DACs.…”

Machine learning: Accelerating materials development for energy storage and conversion

“…Several early reviews have introduced the applications of ML to materials science, including materials discovery and design, catalysts, and structure prediction . Very recently, ML investigations on energy storage and conversion materials have rapidly increased, which have not been comprehensively summarized.…”