Abstract
Double half‐Heusler alloys are a new class of materials with low lattice thermal conductivity, promising for thermoelectric applications. Theoretical predictions of thermal conductivity based on phonon spectra are limited as they require significant computational resources. However, recently, universal machine learning interatomic potentials (UMLIPs) have emerged, which allow fast and accurate prediction of interatomic forces and, consequently, the phonon spectrum. In this study, phonon spectra, Grüneisen parameters, and lattice thermal conductivities of double half‐Heusler alloys Hf 2 FeNiSb 2 and Ti 2 FeNiSb 2 are investigated using established ab initio techniques and the recently developed UMLIP MACE. The results from various methods for calculating lattice conductivity, including solving the linearized Boltzmann transport equation, the Debye–Callaway method, and the Slack formula, are compared. It is demonstrated that a pretrained UMLIP based on the MACE architecture can accurately predict phonon spectra and lattice thermal conductivity of double half‐Heusler alloys without requiring significant computational resources. It is shown that acoustic phonons in Hf 2 FeNiSb 2 and Ti 2 FeNiSb 2 exhibit strong anharmonicity, which causes low lattice thermal conductivities (1.4 W (m·K) −1 for Ti 2 FeNiSb 2 and 1.9 W (m·K) −1 for Hf 2 FeNiSb 2 at 900 K).
Funding Information
  • Ministry of Science and Higher Education of the Russian Federation (FEUZ‐2023‐0015)
  • Ministry of Science and Higher Education of the Russian Federation (IMP‐UrB‐RAS)

This publication has 34 references indexed in Scilit: