A Combined First-Principles and Machine Learning Study of Phase Stability and Thermoelectric Properties in ZrSiPt

Abstract

ZrSiPt is an interesting ternary compound that has two different crystalline phases: a cubic half-Heusler structure and an orthorhombic TiNiSi-type structure. This study offers a comprehensive computational analysis that integrates first-principles density functional theory (DFT) with machine learning (ML) to thoroughly comprehend and anticipate its phasedependent characteristics. Our DFT calculations, utilising the GGA-PBE functional and the TB-mBJ potential for precise electronic structure, indicate that the cubic phase is an indirectgap semiconductor (~1.40 eV) with significant thermoelectric potential, whereas the orthorhombic phase is a semimetal characterised by enhanced mechanical stiffness and anisotropic electronic dispersion. Thermodynamic modelling indicates a phase transition occurring around 300 K, while Boltzmann transport calculations forecast a significant figure of merit (ZT ~0.9) for the cubic phase at room temperature. We created a strong machine learning framework using Random Forest (RF), Support Vector Machine (SVM), and Neural Network (NN) models to add to these ideas. These models, which were trained on a synthetic dataset made from our DFT results, can accurately predict the crystal phase (about 90% of the time)and key thermoelectric properties like the Seebeck coefficient, electrical conductivity, and ZT.The RF model was the most accurate and easy to understand. The feature importance analysis showed that the lattice constant and band gap were the main factors that determined phase stability, which is in line with our DFT-based physical understanding. The models can predict the Seebeck coefficient (R² ≈ 0.8) and ZT (R² ≈ 0.7-0.8) with high accuracy, showing how powerful ML can be at finding complex structure-property relationships. This synergistic DFT+ML approach identifies ZrSiPt as a versatile, tunable material and offers a broadly applicable framework for expediting the design of phase-specific materials for use in thermoelectrics, optoelectronics, and high-strength components.