Machine learning based prediction of lattice thermal conductivity for half-Heusler compounds using atomic information

Abstract

Half-Heusler compound has drawn attention in a variety of fields as a candidate material for thermoelectric energy conversion and spintronics technology. When the half-Heusler compound is incorporated into the device, the control of high lattice thermal conductivity owing to high crystal symmetry is a challenge for the thermal manager of the device. The calculation for the prediction of lattice thermal conductivity is an important physical parameter for controlling the thermal management of the device. We examined whether lattice thermal conductivity prediction by machine learning was possible on the basis of only the atomic information of constituent elements for thermal conductivity calculated by the density functional theory in various half-Heusler compounds. Consequently, we constructed a machine learning model, which can predict the lattice thermal conductivity with high accuracy from the information of only atomic radius and atomic mass of each site in the half-Heusler type crystal structure. Applying our results, the lattice thermal conductivity for an unknown half-Heusler compound can be immediately predicted. In the future, low-cost and short-time development of new functional materials can be realized, leading to breakthroughs in the search of novel functional materials.

Lattice thermal conductivity of half-Heuslers with density functional theory and machine learning: Enhancing predictivity by active sampling with principal component analysis

“…The baseline and active sampling models use on average 5.7 and 6.3 features out of the 14 potential features, respectively. The relatively few features used is in line with the recent results of Miyazaki et al [70] showing that using a limited feature subset gives the best ML performance, as overfitting can arise with redundant features. In both the baseline and active sampling models, B and V are the most frequently selected features, which agrees with their high Spearman correlation with κ TDEP as discussed in the previous section.…”

“…While the performance gain when using more complex features and larger training sets has been demonstrated in earlier studies [20][21][22][23][70][71][72][73], this work clearly shows that rather modest training set sizes and low feature complexity can still give reliable predictions with the use of active sample selection. Finally, we note that the use of a semi-random selection, rather than one that is truly random do accentuate the performance gains of the active sampling model.…”

High-throughput computational discovery of 40 ultralow thermal conductivity and 20 highly anisotropic crystalline materials

“…The availability of such databases is facilitating the data-driven accelerated discovery of novel materials, for instance, using machine learning techniques. Owing to the high computational cost of phonon thermal conductivity prediction (10000-100000 cpu-hours compared to only 1-10 cpuhours for simpler properties), however, the available data on phonon thermal conductivity of materials is scarce; limited majorly to monoatomic and diatomic compounds with simple crystal structures [15][16][17][18]. Further, the available data is obtained using different computational packages employing diverse methodologies/simulation parameters [3,8].…”

“…Though such DFT driven thermal conductivity calculations are instrumental in correctly describing the material thermal transport physics and thus accelerating noble material discovery compared to the experimental trialerror approach by manifold, these calculations are computationally expensive and require, for instance, a single computer running for 10,000-100,000 hours to arrive at a thermal conductivity of a single material [6]. Thus, the application of such calculations is limited to simple material systems, and computational exploration of new materials is restricted to the simple substitution of one or two atomic species in known material systems [15][16][17][18].…”

“…Making use of such developments, in this work, we performed true high-throughput calculations to search for noble material systems with low phonon thermal conductivity and high thermal transport anisotropy. Unlike past studies where the search is restricted to a certain class of materials [15][16][17][18], we have performed full DFT-driven calculations on more than 230 ternary compounds spanning 43 chemical species across 32 spacegroups. Since phonon thermal conductivity follows simple 1/T dependence, where T is the temperature [23], all calculations are performed only at 300 K. All calculations are performed systematically using the same set of well-converged simulation parameters by ensuring thermodynamic and structural stability (no imaginary phonon modes) of all compounds.…”