Charting lattice thermal conductivity for inorganic crystals and discovering rare earth chalcogenides for thermoelectrics

Abstract

Thermoelectricity produced from usually negative-valued heat is a green and promising candidate on the future energy landscape. The most effective thermoelectric materials exhibit low thermal conductivity κ. However, less than...

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

“…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.…”

“…This randomness makes RF less prone to overfitting. RF has been shown to perform well in earlier ML studies involving the lattice thermal conductivity [23]. In the RF regression, a given input sample is sorted in each of the decision trees based on its features, so that in a given tree, the sample is assigned to a κ {i..} in the training set.…”

“…The computational cost arises because accounting for the phonon-phonon interactions due to the anharmonicity of the lattice vibrations requires obtaining third-order force constants extracted from a large number of supercell-based density functional theory (DFT) calculations [15][16][17]. For this reason, machine learning (ML) methods are increasingly supplementing first-principles based calculations for predicting κ [18][19][20][21][22][23][24][25], and pre-trained ML models have been used in a convenient web-based application [26].…”

Unsupervised machine learning for discovery of promising half-Heusler thermoelectric materials

“…Currently, Machine learning (ML) 19,20 and high throughput calculations 21,22 have gradually replaced the traditional trial-anderror approaches with the advantages of high efficiency and low cost. Among them, ML is a powerful tool for the exploration of the desired materials by employing algorithms to construct a statistical model based on the complicated patterns found in high dimensional spaces [23][24][25][26] , such as guiding the chemical synthesis 27 , assisting the multi-dimensional characterization 28 , analyzing the crystal structure 29 , and regulating the phase transition 30 and defects 31 , etc. Supervised learning 32 is the most widespread form of ML in materials science, which needs sufficient amount of relevant data, along with the known target properties and has been applied in the TE materials development and prediction of the Seebeck coefficient 33 (S), power factor 34 (PF = S 2 σ), lattice thermal conductivity 35 (κ L ), and zT values 36 .…”

Figure of merit enhancement in thermoelectric materials based on γ‐Ln0.8Yb0.2S1.5‐y (Ln = Gd, Dy) solid solutions

“…In recent years, material science is focused on the fabrication of high-temperature thermoelectric materials based on lanthanide chalcogenides. [21][22][23][24] For instance, the solid solutions based on the lanthanide sesquisulfides with Th 3 P 4 structural type are one of the promising classes of high-temperature thermoelectric materials. [25][26][27][28] Doping of γ-Dy 2 S 3 with Gd 3+ ions allows one to significantly reduce the total thermal conductivity due to the additional heat dissipation by the lattice deformations and results in the ZT increase.…”