Double-Atom Catalysts Featuring Inverse Sandwich Structure for CO2 Reduction Reaction: A Synergetic First-Principles and Machine Learning Investigation

Electrocatalytic CO2 reduction reactions (CO2RR) based on scalable and highly efficient catalysis providean attractivestrategy for reducing CO2 emissions. In this work, we combinedfirst-principles density functional theory (DFT) and machine learning(ML) to comprehensively explore the potential of double-atom catalysts(DACs) featuring an inverse sandwich structure anchored on defectivegraphene (gra) to catalyze CO2RR to generate C-1 products. We started with five homonuclear M-2 & BOTTOM;gra(M = Co, Ni, Rh, Ir, and Pt), followed by 127 heteronuclear MM & PRIME;& BOTTOM;gra(M = Co, Ni, Rh, Ir, and Pt, M & PRIME; = Sc-Au). Stable DACswere screened by evaluating their binding energy, formation energy,and dissolution potential of metal atoms, as well as conducting first-principlesmolecular dynamics simulations with and without solvent water molecules.Based on DFT calculations, Rh-2 & BOTTOM;gra DAC was foundto outperform the other four homonuclear DACs and the Rh-based single-and double-atom catalysts of noninverse sandwich structures. Out ofthe 127 heteronuclear DACs, 14 were found to be stable and have goodcatalytic performance. An ML approach was adopted to correlate keyfactors with the activity and stability of the DACs, including thesum of radii of metal and ligand atoms (d (M-M & PRIME;), d (M-C), and d (M & PRIME;-C)), the sum and difference of electronegativityof two metal atoms (P (M) + P (M & PRIME;), P (M) - P (M & PRIME;)), the sum and difference of firstionization energy of two metal atoms (I (M) + I (M & PRIME;), I (M) - I (M & PRIME;)), the sumand difference of electron affinity of two metal atoms (A (M) + A (M & PRIME;), A (M) - A (M & PRIME;)), andthe number of d-electrons of the two metal atoms (N (d)). The obtained ML models were further used to predict154 potential electrocatalysts out of 784 possible DACs featuringthe same inverse sandwich configuration. Overall, this work not onlyidentified promising CO2RR DACs featuring the reportedinverse sandwich structure but also provided insights into key atomiccharacteristics associated with high CO2RR activity.

Automated summary

In this work, a combination of first-principles density functional theory (DFT) and machine learning (ML) was used to explore the potential of double-atom catalysts (DACs) on defective graphene for CO2 reduction reactions. Stable DACs were identified based on their binding energy, formation energy, and dissolution potential, and machine learning models were employed to predict potential electrocatalysts.