On the evaluation of hydrogen evolution reaction performance of metal-nitrogen-doped carbon electrocatalysts using machine learning technique

Abstract

Single-atom catalysts (SACs) introduce as a promising category of electrocatalysts, especially in the water-splitting process. Recent studies have exhibited that nitrogen-doped carbon-based SACs can act as a great HER electrocatalyst. In this regard, Adaptive Neuro-Fuzzy Inference optimized by Gray Wolf Optimization (GWO) method was used to predict hydrogen adsorption energy (ΔG) obtained from density functional theory (DFT) for single transition-metal atoms including Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Zr, Nb, Mo, Tc, Ru, Rh, Pd, Ag, Cd, Hf, Ta, W, Re, Os, Ir, Pt, and Au embedded in N-doped carbon of different sizes. Various descriptors such as the covalent radius, Zunger radius of the atomic d-orbital, the formation energy of the single-atom site, ionization energy, electronegativity, the d-band center from − 6 to 6 eV, number of valence electrons, Bader charge, number of occupied d states from 0 to − 2 eV, and number of unoccupied d states from 0 to 2 eV were chosen as input parameters based on sensitivity analysis. The R-squared and MSE of the developed model were 0.967 and 0.029, respectively, confirming its great accuracy in determining hydrogen adsorption energy of metal/NC electrocatalysts.

Rational Designing of Bimetallic/Trimetallic Hydrogen Evolution Reaction Catalysts Using Supervised Machine Learning

“…Performing the DFT calculation for such large combinations is practically impossible even with the fastest ab initio tools and is the primary cause for using ML in this work. Previously, the adsorption energy has been observed as a prominent descriptor for the catalyst’s activity. − As the HER has a single electrochemical step and only one intermediate, the catalytic activity of a catalyst for HER is very much dependent on the adsorption energy of the catalyst. Moreover, adsorption energy descriptors are based on linear scaling relationships and linearly correlated with the activation energy through the Brønsted-Evans–Polanyi (BEP) relationships. − Hence, it is a powerful tool for predicting trends in catalytic activity and screening new potential catalyst materials.…”

“…Recent studies have demonstrated the use of ML or deep learning (DL) methods to discover new HER catalysts like two-dimensional materials, 33 single-atom catalysts (SACs), 34 and nanoclusters. 35 In this work, we sought to intimately predict electrocatalyst surface material with the inclusion of coverage effect on the catalytic surface by exploiting density functional theory (DFT) along with ML methodologies to anticipate nonhazardous and economic heterogeneous catalyst alloys for HER.…”

Rhenium-Based Electrocatalysts for Water Splitting

“…The work of Nørskov demonstrated the importance of computational descriptions to reduce the trial and error that has historically been the norm for experimental HER electrocatalysis. Since then, computational methods have predicted many candidates’ catalytic activity, saving time and costs. − As in the HER, the interaction of the OER intermediates OH*, O*, and OOH* can be successfully calculated by DFT . This scheme has accelerated OER catalysts’ screening and performance prediction. − …”

Data-driven design of electrocatalysts: principle, progress, and perspective

“…Baghban et al verified the effectiveness of a special set of algorithms by using the single-atom system. 116 In this research, the H adsorption energy of the single atoms anchored on the N-doped carbon matrix was calculated by using the gray wolf optimization (GWO). The study involved 27 kinds of transition metal atoms and various sizes of the carbon matrix.…”