Machine learning models for the lattice thermal conductivity prediction of inorganic materials

Abstract

The lattice thermal conductivity (κ L ) is a critical property of thermoelectrics, thermal barrier coating materials and semiconductors. While accurate empirical measurements of κ L are extremely challenging, it is usually approximated through computational approaches, such as semi-empirical models, Green-Kubo formalism coupled with molecular dynamics simulations, and first-principles based methods. However, these theoretical methods are not only limited in terms of their accuracy, but sometimes become computationally intractable owing to their cost. Thus, in this work, we build a machine learning (ML)-based model to accurately and instantly predict κ L of inorganic materials, using a benchmark data set of experimentally measured κ L of about 100 inorganic materials. We use advanced and universal feature engineering techniques along with the Gaussian process regression algorithm, and compare the performance of our ML model with past theoretical works. The trained ML model is not only helpful for rational design and screening of novel materials, but we also identify key features governing the thermal transport behavior in non-metals.

Anharmonicity measure for materials

“…17 illustrates that σ A is a good descriptor for anharmonicity by itself because as a material's vibrational properties become more anharmonic, its phonon lifetimes, and therefore κ L , decrease. It is remarkable that even without explicitly including any of the other material properties that influence κ L , such as group velocities or heat capacities, we get a similar AFD as other semi-empirical models [42,49]. We note that a similar correlation is observed between κ L,exp and σ A for those three perovskites in our data set, for which experimental values of κ L,exp are available [50][51][52].…”

“…As an outlook, we present an additional finding in Fig. 17, in which the experimental lattice thermal conductivity κ L at 300 K is plotted against σ A (300 K) for those RS/ZB/WZ compounds with reliable measurements of κ L [49]. Fitting the data on log-log scale to a linear model, we get a slope of −4.79, implying an inverse power law between κ L and σ A , with an average factor difference (AFD) of 1.48.…”

“…Comparison of σ A with experimental lattice thermal conductivities of 6 wurtzite, 19 zincblende, and 22 rock salt materials (defect-forming noble metal halides were excluded) with thermal conductivity values taken from Ref. [49]. Values of σ A < 0.2 are obtained by one-shot sampling, values >0.2 by aiMD.…”

Guided patchwork kriging to develop highly transferable thermal conductivity prediction models

“…Using the descriptors related directly to the physics of κ l , a prediction model developed on 120 compounds gives RMSE of 0.21 [22]. A prediction model developed on 100 experimentally available κ l compounds, a train/test RMSE and R 2 of 0.17/0.21 and 0.93/0.93, respectively, has been reported [53].…”

Active learning for the power factor prediction in diamond-like thermoelectric materials

“…In 2018, Miller et al viewed diamond-like semiconductors from the perspective of carrier concentration range with ML method and quantified their dopabilities by linear regression 13 . In 2019, an ML model for predicting the κ L was proposed based on the experimentally measured κ L s of~100 inorganic materials 14 . In the same year, Tshitoyan et al employed the text mining method on the material literature and sought potential TE materials by their similarities with the word "thermoelectric" 15 .…”