First-principles And Deep Learning Studies On Catalytic Surface Microstructure

Heterogeneous catalysis is vital to industrial production because of its stability and convenient separation.Development of efficient,stable and environmentally friendly heterogeneous catalysts can meet the growing material needs of humankind and reduce the impact of catalysis on the environment.Through deep learning and the combination of theoretical calculation and experiment,this thesis explores the influence of the surface microstructure on the catalytic performance from the three aspects of reaction interface,metal-support interface,and the regulation of surface fine structure,which provides a good reference for the design of catalysts and high-throughput screening.Firstly,the effect of reaction interface on electrochemical nitrogen reduction（NRR）is investigated.The results show that N₂ completely occupies the top sites in the way of vertical adsorption after N₂ is enriched on the surface of Pd catalyst,and the H adsorption site is isolated by N₂.Therefore,the adjacent H cannot be close to form H₂,which inhibits hydrogen evolution（HER）.According to the calculation results,we propose a three-phase reaction interface of gas reactant-solid catalyst-liquid electrolyte,in which N₂ surrounds Pd in the form of bubbles,thereby promoting a large amount of N₂ to preferentially occupy the Pd active center and increasing the N₂ coverage on Pd surface in aqueous solution.Under mild conditions,the Faraday efficiency（FE）of Pd catalyst in acidic aqueous solution is as high as 97%.The strategy of three-phase interface is also applicable to other catalysts with strong H adsorption capacity,and their FE can be increased to over 90%.In addition to the reaction interface,metal-support interface can also affect the performance of the catalyst.Next,how the metal support interaction（MSI）affects the performance of the catalyst for semi hydrogenation of 2-Methyl-3-butyn-2-ol（MBY）is uncovered.For the small-size Pd nanoparticles（NPs）,although it contains more active sites,the proportion of edge and corner sites is high,which can lead to over hydrogenation,resulting it difficult to achieve excellent activity and selectivity simultaneously.By manufacturing oxygen vacancy and controlling MSI,small-size Pd NPs are flattened on the support with strong MSI.Infrared data shows that on strong MSI supports,the proportion of planar sites on small NPs is comparable to that on large NPs.Theoretical simulations also show that with the increase of MSI,the atoms on Pd clusters tend to transfer to the interface,resulting in the flattening of small-size Pd NPs.Flattened small Pd NPs have more active sites and fewer corner sites at the same time,enabling MBY semi hydrogenation reaction to achieve high activity while achieving selectivity comparable to large NPs（96%）.The fine structures of the surface have great impacts on the catalytic performance of structural sensitive reactions.High-throughput（HT）screening and machine learning（ML）are considered to be efficient for exploring the hidden rules of the impacts.However,there is currently no protocol for constructing an interpretable ML framework that is sensitive to the fine structure.Finally,we created a data augmented convolutional neural network（CNN）ML framework called GLCNN,which combines"global+local"features.This framework can capture the original fine structures without complicated encoding methods by transforming catalytic surfaces and adsorption sites into two-dimensional grids and one-dimensional descriptors,respectively.The GLCNN framework accurately predicted and distinguished the adsorption energies of OH on a set of analogous carbon-based transition metal single-atom catalysts with a mean absolute error of less than 0.1 e V,ranking the best result of popular models trained on large datasets so far.Unlike conventional CNN and descriptor-based ones with one-sided feature extraction,this fine-structure sensitive ML framework can extract the key factors that affect catalytic performance from both geometric and chemical/electronic features,such as symmetry and coordination elements,through unbiased interpretable analysis.This framework provides a feasible solution for high-precision HT screening of heterogeneous catalyst with a broad physical and chemical space.