Exploring Heat-Shielding Nanoparticle-Based Materials via First-Principles Calculations and Transfer Learning

From an energy-saving perspective, blocking infrared (IR) radiations in the solar spectrum has garnered increased attention, and several heat-shielding nanoparticle-based materials have been employed for this purpose. However, they are composed of precious metals and rare-earth elements, and therefore, alternative materials are preferred. Herein, heat-shielding nanoparticle-based materials were explored via first-principles calculations and transfer learning. We evaluated the visible light absorption from the imaginary part of the bulk dielectric function. However, to calculate the dielectric function with high accuracy, first-principles calculations using hybrid functionals are required, but their computational costs are high. Therefore, it is difficult to obtain reasonable data to train ML models to create a prediction model using machine learning techniques. We created a prediction model from "small data" obtained via the hybrid functional method using data from the generalized-gradient-approximation method and transfer learning. Using this model, heat-shielding materials were comprehensively investigated, and potential candidates, such as RE2BiO2 (RE = Ho, Dy, Tb, and Gd), Ba-containing materials, KAgF3, RbAgCl3, and Sr–Cu–O systems that were not previously considered heat-shielding materials, were identified. The absorption cross section for the material nanoparticles was calculated, and the influence of the nanoparticle shape on the heat-shielding properties was also analyzed. Because synthesizing nanoparticles of the abovementioned materials is cheaper than synthesizing nanoparticles of conventional heat-shielding nanoparticle-based materials, we expect that these nanoparticle-based materials will be used in heat-shielding windows.