A hidden strain memory programs C2+ product branching in reconstructed copper electrocatalysts
Authors list
Yahui Wu Zhe Zhao Xin Wei Yaru Sheng Nannan Wang Xinbo Gao Pei Zhang Xinchen Kang Zhe Li Bing An Buxing HanThis article is a preprint. Preprints have not been peer-reviewed.
You can read more about preprints.
Abstract
The lattice-scale structural basis of ethylene-to-C2+ alcohol branching in copper-catalyzed CO2 electroreduction remains difficult to quantify with conventional probes. Here, we show that weakly supervised, attention-based multiple-instance learning, trained on TEM image sets labeled by catalyst-state rather than product-selectivity or pixel-level annotations, reveals lattice features beyond manual inspection or ensemble-averaged characterization. The model decodes an electroreduction-imprinted strain memory, where diffuse εyy distortion contracts into localized islands, and its normalized attention-gated area fraction predicts ethylene-to-C2+ alcohol branching in out-of-sample validation (R2=0.93). Reciprocal-space analysis, operando spectroscopy, and DFT calculations link this memory to strain-bearing high-index step motifs that favor oxygen-retaining intermediates. Annealing erases the memory and shifts branching toward ethylene while largely preserving total geometric current density, whereas precursor and reduction protocols tune its retention to favor alcohol formation. These findings establish catalyst history as a programmable structural handle for C2+ product branching.
Details
Archive
chemRxiv
Available online
Publication date
Keywords
Type of publication
