No abstract

A finite electronic band gap is a standard filter in high-throughput screening of materials using density functional theory (DFT). However, because of the systematic underestimation of band gaps in standard DFT approximations, a number of compounds may be incorrectly predicted metallic. In a more accurate treatment, such materials may instead appear as low band gap materials and could have good thermoelectric properties if suitable doping is feasible. To explore this possibility, we performed hybrid functional calculations on 1093 cubic materials listed in the Materials Project database with four atoms in the primitive unit cell, spin-neutral ground state, and a formation energy within 0.3 eV of the convex hull. Out of these materials, we identified eight compounds for which a finite band gap emerges. Evaluating electronic and thermal transport properties of these compounds, we found the compositions MgSc2Hg and Li2CaSi to exhibit promising thermoelectric properties. These findings underline the potential of reassessing band gaps and band structures of compounds to identify additional potential thermoelectric materials.

Low thermal conductivity is an important materials property for thermoelectricity. The lattice thermal conductivity (LTC) can be reduced by introducing sublattice disorder through partial isovalent substitution. Yet, large-scale screening of materials has seldom taken this opportunity into account. The present study aims to investigate the effect of partial sublattice substitution on the LTC. The study relies on the temperature-dependent effective potential method based on forces obtained from density functional theory. Solid solutions are simulated within a virtual crystal approximation, and the effect of grain-boundary scattering is also included. This is done to systematically probe the effect of sublattice substitution on the LTC of 122 half-Heusler compounds. It is found that substitution on the three different crystallographic sites leads to a reduction of the LTC that varies significantly both between the sites and between the different compounds. Nevertheless, some common criteria are identified as most efficient for reduction of the LTC: The mass contrast should be large within the parent compound, and substitution should be performed on the heaviest atoms. It is also found that the combined effect of sublattice substitution and grain-boundary scattering can lead to a drastic reduction of the LTC. The lowest LTC of the current set of half-Heusler compounds is around 2 W/Km at 300 K for two of the parent compounds. Four additional compounds can reach similarly low LTC with the combined effect of sublattice disorder and grain boundaries. Two of these four compounds have an intrinsic LTC above ∼15 W/Km, underlining that materials with high intrinsic LTC could still be viable for thermoelectric applications.

Low lattice thermal conductivity is essential for high thermoelectric performance of a material. Lattice thermal conductivity is often computed based on density functional theory calculations, but such calculations carry a high computational cost and machine learning is therefore increasingly being used to estimate lattice thermal conductivity at a much lower computational expense. With the ability to asses larger sets of materials, machine learning could offer an effective procedure to identify low lattice thermal conductivity compounds. However, such compounds can be quite rare and distinct from typical compounds in a given training set. This can be problematic as standard machine learning methods lack the ability to precisely interpret properties of compounds with features differing significantly from those in the training set. By computing the lattice thermal conductivity of 122 half-Heusler compounds using the temperature dependent effective potential method, we generate a data set sufficient in span to explore this issue. We show that random forest regression can fail to identify low lattice thermal conductivity compounds with random selection of training data. However, if the choice of training data is instead guided using feature and principal component analysis, it can drastically improve the ability to identify low lattice thermal conductivity compounds as well as model performance.