Machine Learning Study On Surface Structures And Adsorption Characteristics Of Precious Metal Based Alloys

Alloys have been widely used in various fields,due to the unique physical and chemical performance as well as flexible tunability brought about by alloying effect.Understanding the structure-property relationship of alloys is crucial for the design of high-performance alloys,but their diverse surface active sites and complex local environment pose challenges to the research.The intrinsic descriptors based on valence electron properties provide support for solving this problem,as their parameters are accessible and they can reflect physical pictures of the surface structures and adsorption characteristics of alloys.On the other hand,the efficiency of conventional trial-and-error experiment and theoretical methods is too low,making them difficult to quickly screen and evaluate a large number of candidate alloys.In contrast,machine learning techniques can effectively process and analyze large datasets,thereby quickly extracting valuable information to solve complex problems.The schemes combining intrinsic descriptors with machine learning algorithms can significantly improve the efficiency of material screening and development.In this thesis,we systematically study the surface stability and adsorption characteristics of alloys based on the intrinsic descriptors,and establish the analytical models as well as the machine learning models to quantitatively predict the surface properties of complex precious metal based alloys.The detailed contents are as follows:Firstly,we utilize several accessible intrinsic descriptors for describing carbon dioxide reduction catalysts,including transition metals,near-surface alloys and single-atom alloys.Considering the electronic and geometric descriptors of the alloy substrates,as well as the characteristic parameters of adsorbates,we comprehensively capture the interaction between substrates and adsorbates.By combining machine learning methods,we develop a model that can simultaneously accurately predict adsorption energies of various carbon-terminated intermediates and can quickly screen and evaluate in the huge phase space of alloy catalysts.In addition,this machine learning scheme can simultaneously describe the adsorption energies of reactants,intermediates,and products in the given reaction process at the equal footing,and analyze the different coupling mechanisms between adsorbates and substrates.This universal design scheme for carbon dioxide reduction reaction catalysts can provide strong assistance for the screening of alloy catalysts,rendering the design methodology cost-effective and efficient.Secondly,based on the average effect of surface electrons on alloys,we establish a simple analytical model to quantitatively describe the non-local surface environment of high-entropy alloys(HEAs).Based on the layered surface electronic descriptor,surface energies of complex HEAs can be linearly determined with tiny errors,reflecting the mean-field effect of surface energy corresponding to the surface chemical distribution.In addition,the entropy state of bulk atoms will cause fine-tuning of the linear relationship between surface energy and the descriptor.This analytical model effectively quantifies the alloying effect among different elements,facilitating the understanding of the structure-property relationship and the prediction of surface properties of complex HEAs.Due to the accessibility of all parameters,this scheme can quickly screen the desired HEAs in a large phase space,and can also be conveniently used to fine-tune various desired surface properties of HEAs,which is helpful for the design of HEAs with unique mechanical and physical properties.Finally,based on intrinsic descriptors and machine learning methods,we construct models that can accurately predict surface segregation energies and *CO adsorption energies of HEAs with various facets under different conditions.We investigate the atomic segregation trend,the influencing factors of *CO adsorption strength,the effect of surface segregation on *CO adsorption,and the *CO adsorption induced surface segregation mechanism on HEA surfaces.Understanding the coupling mechanism between segregation and adsorption,we capture the core determinants of atomic segregation trends and *CO adsorption behavior on HEA surfaces.Through in-depth analysis of the segregation and adsorption behavior on HEA surfaces,we reveal the physical origin of surface complexity and unique catalytic performance of HEAs.These findings are helpful for the design and optimization of high-performance HEAs in subsequent research,and lay a theoretical foundation for understanding the structure-property relationship of surface properties and adsorption characteristics of HEAs.In summary,this thesis focuses on the surface stability and adsorption characteristics of precious metal based alloys.Based on the intrinsic descriptors of alloys,the analytical models and machine learning models can not only quantitatively predict the surface properties of alloys,but also deeply analyze the structure-property relationship.Our models reveal underlying physical pictures of surface properties and performances of alloys,providing solid theoretical basis and universal research schemes for the screening,design and optimization of high-performance alloys.