Machine Learning in Screening High Performance Electrocatalysts for CO₂ Reduction

Abstract

Converting CO2 into carbon‐based fuels is promising for relieving the greenhouse gas effect and the energy crisis. However, the selectivity and efficiency of current electrocatalysts for CO2 reductions are still not satisfactory. In this paper, the development of machine learning methods in screening CO2 reduction electrocatalysts over the recent years is reviewed. Through high‐throughput calculation of some key descriptors such as adsorption energies, d‐band center, and coordination number by well‐constructed machine learning models, the catalytic activity, optimal composition, active sites, and CO2 reduction reaction pathway over various possible materials can be predicted and understood. Machine learning is now realized as a fast and low‐cost method to effectively explore high performance electrocatalysts for CO2 reduction.

Machine learning accelerated calculation and design of electrocatalysts for CO2 reduction

“…In addition to failing to make predictions accurately, a failing model will have a high bias (underfitting) if it finds insufficient data for applicable rules, or the algorithm is not flexible enough to reflect the relationship between input and output data. High variance (overfitting) occurs if the model is too complex and has too many parameters 11 …”

“…High variance (overfitting) occurs if the model is too complex and has too many parameters. 11 Sargent et al 37 performed DFT calculations on the optimal ML-predicted structures to characterize the changes of reaction energies in the main steps of CO 2 reduction. Benefitting from the introduction of Al, the reaction energy of the C−C bonding rate-determining step is reduced from 1.4 to 0.6 eV for Cu (111) and from 0.6 to 0.4 eV for Cu (100).…”

“…With the fast development of ML in past decades, ML has received much more attention from researchers in materials science in recent years, and has been used for various applications, such as in photocatalytic 10 and electrocatalytic 11 CO 2 reduction, nitrogen reduction, 12 water splitting, [13][14][15] fuel cells, 16,17 photovoltaics, [18][19][20] and soft materials. 21 Some reviews have discussed the application strategies of ML for energy materials, 22 ML in DFT, 6 ML for the porous material, 23 ML in 2D materials, 24 and ML in experimental synthesis.…”

Avoiding Pyrolysis and Calcination: Advances in the Benign Routes Leading to MOF‐Derived Electrocatalysts

“…This can culminate in a predictive modeling-guided blueprint complemented well by state-of-the-art machine learning approaches. [101] With the benefit of hindsight, we envisage that a two-way iterative feedback loop between experimental observations (complemented by machine learning tools) and computational simulations (augmented by machine learning tools) will serve as a strong prognostic indicator, [102] making headway in handpicking commercialization-ready MOF electrocatalysts. [65] Our present assessment, herein, not only offers a perspective on the energy-efficient, environmentally benign approaches of developing MOF electrocatalysts (thus avoiding pyrolysis and/or calcination derived carbonized hybrids), it also widely underpins the general understanding of MOF electrocatalysts.…”

Galvanic‐Cell Deposition Enables the Exposure of Bismuth Grain Boundary for Efficient Electroreduction of Carbon Dioxide

“…[10] In this respect, metals including indium (In), tin (Sn), cadmium (Cd), lead (Pb), and bismuth (Bi) with high hydrogen evolution overpotential have been reported to be selective toward formate. [11][12][13][14] With the distinct merits of low cost and environmental benignity, metallic Bi holds great promise in reducing CO 2 into formate. [15] In this regard, numerous research efforts have been devoted to tailoring the nanoarchitectures, morphologies, compositions, and defects of Bi-based catalysts toward efficient and stable electroreduction of CO 2 into formate.…”