Interpretable machine learning for chiral induced symmetry breaking of spin density boosting hydrogen evolution

The integration of machine learning and electrocatalysis presents notable advancements in designing and predicting the performance of chiral materials for hydrogen evolution reactions (HER). This study utilizes theoretical calculations and machine learning techniques to assess the HER performance of both chiral and achiral M-N-SWCNTs (M = In, Bi, and Sb) single-atom catalysts (SACs). The stability preferences of metal atoms are dependent on chirality when interacting with chiral SWCNTs. The HER activity of the right-handed In-N-SWCNT is 5.71 times greater than its achiral counterpart, whereas the left-handed In-N-SWCNT exhibits a 5.12-fold enhancement. The calculated hydrogen adsorption free energy for the right-handed In-N-SWCNT reaches as low as -0.02 eV. This enhancement is attributed to the symmetry breaking in spin density distribution, transitioning from C-2v in achiral SACs to C-2 in chiral SACs, which facilitates active site transfer and enhances local spin density. Right-handed M-N-SWCNTs exhibit superior alpha-electron separation and transport efficiency relative to left-handed variants, owing to the chiral induced spin selectivity (CISS) effect, with spin-up alpha-electron density reaching 3.43 x 10(-3) e/Bohr(3) at active sites. Machine learning provides deeper insights, revealing that the interplay of weak spatial electronic effects and appropriate curvature-chirality effects significantly enhances HER performance. A weaker spatial electronic effect correlates with higher HER activity, larger exchange current density, and higher turnover frequency. The curvature-chirality effect underscores the influence of intrinsic structures on HER performance. These findings offer critical insights into the role of chirality in electrocatalysis and propose innovative approaches for optimizing HER through chirality. (c) 2024 Science Press and Dalian Institute of Chemical Physics, Chinese Academy of Sciences. Published by Elsevier B.V. and Science Press. All rights are reserved, including those for text and data mining, AI training, and similar technologies.