Ab Initio Study On Solid-solution And Thermal Transport Properties Of Mg₃Sb₂ Based Thermoelectric Materials

In recent years,Mg₃Sb₂-based thermoelectric（TE）materials have gained attention due to their high TE performance and environmentally friendly features,as they comprise abundant,nontoxic,and low-cost constituent elements.Although the thermoelectric properties of Mg₃Sb₂-based TE materials have been extensively studied,reports on their solid-solution properties are limited.In scientific research and industry,the simulation of the thermal transport properties of materials is crucial,but the thermal transport simulation of solid solutions is a challenge in current theoretical research.In this dissertation,we investigated the thermodynamic properties of isostructural and hetero-structural solid-solution systems of Mg₃Sb₂-based TE materials using the approaches of density functional theory（DFT）,cluster expansion（CE）,Monte Carlo（MC）,and machine learning.We used the solid-solution properties of isostructural Mg₃Sb₂-Mg₃Bi₂ and hetero-structural Mg₂Sn-Mg₃Sb₂ systems as examples for application.Furthermore,the feasibility of using machine learning interatomic potentials（MLIPs）in the thermal transport simulation of complex Mg_2-δ(Sn_1-x Sbx)solid solutions were studied.The research contents and main conclusions are listed as follows:In the isostructural Mg₃Sb₂-Mg₃Bi₂ system,the formation energies of Mg₃(Sb_1-x Bix)₂ solid solutions were calculated by first-principles method.Based on the Gibbs free energy of the solid solutions,we obtain the phase diagram for the Mg₃Sb₂-Mg₃Bi₂ system.Our findings reveal the presence of a miscibility gap in the Mg₃(Sb_1-x Bix)₂ solid solutions.A method of calculating solute activity in multi-component alloys by combining first-principles calculations and thermodynamic analysis associated with defects is developed.Based on the model,the activities of Mg in Mg₃(Sb_1-x Bix)₂ alloy were estimated.Our estimated activities reproduce the available experimental results well.According to the criterion of phase equilibrium,we estimate the solubility limits of Mg in Mg₃(Sb_1-x Bix)₂ and construct the isothermal section phase diagram of Mg-Mg₃Sb₂-Mg₃Bi₂.We discuss the effect of Mg on the thermoelectric performance of Mg₃(Sb_1-x Bix)₂alloy and find that Mg interstitials are the dominant defects in Mg₃(Sb_1-x Bix)₂ alloy,responsible for the n-type thermoelectric under Mg excess conditions.Hetero-structural solid solutions,composed of parent compounds with distinct crystal structures,have garnered significant interest in the exploration of new materials,but a comprehensive and systematic theoretical research is still lacking.To overcome the calculation difficulties of energy and configurational entropy,we developed a general approach that combines first-principles calculations,CE,and MC simulations to evaluate the energetics and configuration evolution of complex hetero-structural solid solutions.We applied this approach to the Mg₂Sn-Mg₃Sb₂ system and found that the Gibbs free energy results indicate that Mg_2-0.5x Sn_1-x Sbx is stabilized at the Mg₂Sn-based phase in the range of0<x≤～0.47 and becomes metastable in the range of～0.47<x≤0.75,while the Mg₃Sb₂-based phase is unstable in the range of 0.75<x<1.0,which is consistent with experimental observations.Large-scale MC simulations reveal that Mg_2-0.5x Sn_1-x Sbx solid solutions have natural hierarchical microstructures,which contain primarily nanoscale Mg-deficiency and Sb-rich clusters and a homogeneous Mg₂Sn-based matrix.Compositional analysis of the microstructures shows that the composition ratio of Mg/（Sn+Sb）for the nanocluster region is close to the atomic ratio of 3:2 in Mg₃Sb₂,and the ratio for the matrix region is close to2:1 in Mg₂Sn,which reveals the overall inhomogeneous nature of the hetero-structural Mg_2-0.5x Sn_1-x Sbx solid solutions.The electronic structures of the microstructures of Mg_2-0.5x Sn_1-x Sbx manifest an inhomogeneous coexistence of p-type nanoclusters and n-type matrix.Selective doping can realize the n-type/p-type alteration of microstructures,providing a theoretical basis and method for the optimization and design of hetero-structural solid solutions.The application of machine learning interatomic potentials（MLIPs）is widespread,but obtaining reliable MLIPs for solid solutions with complex and diverse microstructures,particularly hetero-structural solid solution systems,has always been challenging.We propose that,based on the hierarchical microstructure results from DFT-CE-MC simulations,a reasonable structure training dataset can be established,and reliable MLIPs can be obtained by taking into account the wide sampling space containing both compositional fluctuations and hierarchical microstructure changes.Using DFT and MD calculations,we constructed MLIPs of hetero-structural Mg_2-δSn_1-x Sbx solid solutions by employing a dual adaptive sampling method to generate training datasets in the configuration space with a wide temperature range of 50-800 K,a wide composition range of 0≤x<0.8,and complex microstructures.The MLIPs predicted energies,forces,atomic motive trajectories,lattice constants,formation energy,and thermal transport properties of Mg₂Sn and Mg_1.875Sn_0.75Sb_0.25 agree well with the results of DFT calculations,indicating the MLIPs’good reliability and accuracy.Finally,we evaluated the lattice thermal conductivity and its temperature dependence for Mg2-δSn_0.8Sb_0.2 solid solutions using the Green-Kubo method based on the MLIPs.