Discovery of low thermal conductivity compounds with first-principles anharmonic lattice dynamics calculations and Bayesian optimization (1506.06439v1)

Abstract: Compounds of low lattice thermal conductivity (LTC) are essential for seeking thermoelectric materials with high conversion efficiency. Some strategies have been used to decrease LTC. However, such trials have yielded successes only within a limited exploration space. Here we report the virtual screening of a library containing 54,779 compounds. Our strategy is to search the library through Bayesian optimization using for the initial data the LTC obtained from first-principles anharmonic lattice dynamics calculations for a set of 101 compounds. We discovered 221 materials with very low LTC. Two of them have even an electronic band gap < 1 eV, what makes them exceptional candidates for thermoelectric applications. In addition to those newly discovered thermoelectric materials, the present strategy is believed to be powerful for many other applications in which chemistry of materials are required to be optimized.

• The paper introduces an innovative integration of first-principles anharmonic lattice dynamics with Bayesian optimization to screen and discover low lattice thermal conductivity compounds.
• The methodology employs DFT-based phonon analysis and the Boltzmann transport equation to accurately predict LTC values, identifying compounds with values as low as 0.9 and 0.2 W/m·K.
• The paper demonstrates that combining computational efficiency with machine learning accelerates the discovery of promising materials for advanced thermoelectric applications.

Overview of Low Thermal Conductivity Compound Discovery Using First-Principles Calculations and Bayesian Optimization

The paper presents a paper focused on the discovery of low lattice thermal conductivity (LTC) compounds critical for developing high-efficiency thermoelectric materials. The authors employ an innovative approach that integrates first-principles anharmonic lattice dynamics calculations with Bayesian optimization to screen a substantial compound library.

Research Methodology

The methodology is anchored on an initial set of 101 compounds for which LTC values were meticulously calculated using first-principles methods. These compounds feature three distinct prototype structures: rocksalt, zincblende, and wurtzite. Phonon properties, crucial for LTC calculations, were derived from force constants obtained via density functional theory (DFT). This groundwork enabled the authors to utilize the Boltzmann transport equation under a single-mode relaxation-time approximation to attain LTC values with experimental accuracy.

Following these calculations, Bayesian optimization was employed using kriging, enabled by Gaussian process regression (GPR). Here, volume and density served as the primary predictors for a virtual screening of 54,779 compounds from available materials databases. The screening revealed 221 new compounds with exceptionally low LTC, standing out as promising candidates for thermoelectric applications.

Numerical Results

The paper highlights several key findings. Among the 101 initially evaluated compounds, PbSe in the rocksalt structure emerged with the lowest LTC of 0.9 W/m·K at 300 K. Bayesian optimization identified compounds such as PbRbI$_3$, PbIBr, and PbRb$_4$Br$_6$, which demonstrated considerably lower LTC values. Specifically, 5 compounds showed LTC below 0.2 W/m·K, significantly lower than the previously noted 0.9 W/m·K for PbSe. Remarkably, K$_2$CdPb and Cs$_2$[PdCl$_4$]I$_2$ were identified with low LTCs and narrow electronic band gaps, suggesting excellent thermoelectric potential.

Implications and Future Developments

The implications of this research are substantial, not only for the field of thermoelectrics but also for the broader scope of materials science where LTC is a key parameter. The combination of first-principles LTC calculation with Bayesian optimization exemplifies a practical approach to expanding exploration spaces without relying on the conventional empirical restrictions of material selection, thereby enhancing the discovery rate of new materials.

Theoretically, this paper underscores the efficacy of combining machine learning techniques like Bayesian optimization with first-principles calculations to navigate vast compositional and configurational spaces. Practically, the identified low LTC compounds, particularly those with favorable electronic band gaps, could play a pivotal role in the creation of high figure-of-merit thermoelectric materials.

Looking forward, such a methodological framework could be tailored and applied to optimize other material properties across different applications, including but not limited to catalysis, battery technology, and superconductivity. Further methodological refinements could involve expanding the training dataset with additional descriptors or adapting advanced machine learning models for even more nuanced predictions.

In conclusion, the integration of predictive modeling with accurate computational methods marks a significant step toward rational materials design. This research demonstrates how computational efficiency and predictive power can collectively advance the discovery pipeline, paving the way for future innovations in materials science and engineering.