Understanding the role of defects in the lattice transport properties of half-Heusler compounds: a machine learning analysis

TL;DR

This study uses machine learning and DFT to analyze how intrinsic defects influence phonon behavior and thermal conductivity in half-Heusler compounds, improving agreement with experimental data.

Contribution

It introduces a combined DFT and machine-learning approach to efficiently study defect effects on lattice and electronic properties in thermoelectric materials.

Findings

Defects lower lattice thermal conductivity by creating localized phonon modes.

Intrinsic defects reduce the electronic band gap, aligning theory with experiments.

Fe interstitials and antisite defects are most likely to form in TaFeSb.

Abstract

While the effect of intrinsic defects on the electronic properties of half-Heusler compounds has been extensively discussed in literature, their effect on the lattice vibrations has received much less attention, due to the prohibitive computational demands. This may lead to an erroneous description of the lattice thermal conductivity, which plays a crucial role in the thermoelectric efficiency, and for which there exists a significant discrepancy between ideal theoretical values and available experimental measurements. In this article, we employ a combination of density-functional theory (DFT) and machine-learning interatomic potentials (MLIPs) to investigate how intrinsic defects affect the phonon spectra and lattice thermal conductivity of TaFeSb, alongside its electronic structure. The calculation of the formation energies of various defects identifies Fe interstitial atoms sitting at the vacant side of the HH crystal structure as the most likely to form, immediately followed by Sb substitution at Ta sites and by other antisite configurations. Phonon calculations illustrate that both defects generate localized phonon modes that significantly lower the lattice thermal conductivity, especially around room temperature. This reduction aligns the calculated values with available measurements, underscoring the critical role of intrinsic defects in reconciling the existing discrepancies between theory and experiment. Our findings also reveal that these defects introduce localized electronic states, effectively reducing the theoretical electronic band gap and bringing it closer to the experimentally observed values. Finally, our analysis demonstrates the efficiency and effectiveness of machine-learning-based approaches to investigate defect-induced properties in complex materials for thermoelectric applications.