数据驱动的高熵碲化物组分高通量理论设计与性能预测OA

To address the vast compositional space and the inefficiency of traditional"trial-and-error"methods in the development of high-entropy telluride thermoelectric materials,this study establishes a theoretical data-driven design framework that integrates machine learning with first-principles calculations.By constructing a high-dimensional descriptor space that encompasses physical features such as mixing enthalpy and atomic size difference,a gradient boosting regression model was trained to accurately predict the formation energy of multiprincipal-element systems.SHAP(SHapley Additive exPlanations)interpretability analysis elucidated the competitive mechanisms between thermodynamic and geometric parameters,leading to the formulation of a quantitative criterion for phase formation.Using this criterion,a high-throughput theoretical screening of the entire PbTe-SnTe-GeTe ternary space was conducted.An optimal high-entropy composition featuring both high thermodynamic stability(-46.31 kJ/mol)and ultra-low lattice thermal conductivity(0.85 W/(m·K))was theoretically identified.Electronic structure calculations further demonstrated that the introduction of the transition metal Mn induces convergence of the conduction band and strong hybridization of p-d orbitals,which effectively decouples the electrical and thermal transport parameters.Theoretical evaluations suggest that this high-entropy composition could achieve a thermoelectric figure of merit(zT)of 1.60 at 850 K,representing an approximate 67%improvement over the PbTe matrix.This work preliminarily verifies the feasibility of data-driven strategies for rational theoretical design of complex thermoelectric materials and provides a clear frame-work for future experimental synthesis and validation.It is important to note that all results herein are theoreti-cal predictions,and subsequent experimental synthesis and performance evaluations are ongoing.