Density functional theory (DFT) calculations have been conducted to investigate the mechanism of cobalt(II) tetraamino phthalocyanine (CoPc-NH 2 ) catalyzed electro-reduction of CO 2 . Computational results show that the catalytically active species 1 ( 4 [Co II (H 4 L)] 0 ) is formed by a four-electron-four-proton reduction of the initial catalyst CoPc-NH 2 . Complex 1 can attack CO 2 after a one-electron reduction to give a [Co III −CO 2 2− ] − intermediate, followed by a protonation and a one-electron reduction to give intermediate [Co II −COOH] − (4).Complex 4 is then protonated on its hydroxyl group by a carbonic acid to generate the critical species 6 (Co III −L •− −CO), which can release the carbon monoxide as an intermediate (and also as a product). In parallel, complex 6 can go through a successive four-electron-four-proton reduction to produce the targeted product methanol without forming formaldehyde as an intermediate product. The high-lying π orbital and the low-lying π* orbital of the phthalocyanine endow the redox noninnocent nature of the ligand, which could be a dianion, a radical monoanion, or a radical trianion during the catalysis. The calculated results for the hydrogen evolution reaction indicate a higher energy barrier than the carbon dioxide reduction. This is consistent with the product distribution in the experiments. Additionally, the amino group on the phthalocyanine ligand was found to have a minor effect on the barriers of critical steps, and this accounts for the experimentally observed similar activity for these two catalysts, namely, CoPc-NH 2 and CoPc.

Catalytic reduction of pyridines to N−H 1,4dihydropyridines is exceptionally challenging because they are essential intermediates to form tetrahydropyridines. Using a facile dihydrogen source H 3 N•BH 3 to activate the pyridine ring in situ, we have achieved selective transfer hydrogenation of nicotinate derivatives to N−H 1,4-dihydropyridines by cobalt-amido cooperative catalysis. The reactions operate smoothly under mild conditions to produce a variety of N−H 1,4-dihydropyridines with high chemoand regioselectivity. This catalytic method also provides a practical protocol to regenerate Hantzsch analogues after delivery of H 2 .

Density functional theory (DFT) calculations were conducted to investigate the cobalt porphyrin‐catalyzed electro‐reduction of CO 2 to CO in an aqueous solution. The results suggest that Co II −porphyrin (Co II −L) undertakes a ligand‐based reduction to generate the active species Co II −L⋅ − , where the Co II center antiferromagnetically interacts with the ligand radical anion. Co II −L⋅ − then performs a nucleophilic attack on CO 2 , followed by protonation and a reduction to give Co II −L−COOH. An intermolecular proton transfer leads to the heterolytic cleavage of the C−O bond, producing intermediate Co II −L−CO. Subsequently, CO is released from Co II −L−CO, and Co II −L is regenerated to catalyze the next cycle. The rate‐determining step of this CO 2 RR is the nucleophilic attack on CO 2 by Co II −L⋅ − , with a total barrier of 20.7 kcal mol −1 . The competing hydrogen evolution reaction is associated with a higher total barrier. A computational investigation regarding the substituent effects of the catalyst indicates that the CoPor−R3 complex is likely to display the highest activity and selectivity as a molecular catalyst.

Due to the limited scale and quality of video-text training corpus, most visionlanguage foundation models employ image-text datasets for pretraining and primarily focus on modeling visually semantic representations while disregarding temporal semantic representations and correlations. To address this issue, we propose COSA, a COncatenated SAmple pretrained vision-language foundation model. COSA jointly models visual contents and event-level temporal cues using only image-text corpora. We achieve this by sequentially concatenating multiple image-text pairs as inputs for pretraining. This transformation effectively converts existing image-text corpora into a pseudo long-form video-paragraph corpus, enabling richer scene transformations and explicit event-description correspondence. Extensive experiments demonstrate that COSA consistently improves performance across a broad range of downstream tasks, including long-form/short-form videotext tasks and image-text tasks such as retrieval, captioning, and question answering. Notably, COSA achieves state-of-the-art results on various competitive benchmarks. Code and model are released at https://github.com/TXH-mercury/COSA.