Machine-learning-assisted space-transformation accelerates discovery of high thermal conductivity alloys

We study the thermal conductivity distribution of hypothetical graphene-like materials composed of carbon and heavy carbon atoms. These materials are representative of alloys and disordered materials, which are relatively unexplored for thermal properties owing to their large configuration spaces. Since the full thermal conductivity calculations using the Boltzmann transport equation based solutions are computationally prohibitive for each of the 2(32) considered configurations, we employ regularized autoencoders, a class of generative machine learning models that transform the configuration space to the latent space in which materials are clustered according to the target property. Such conditioning allows selective sampling of high thermal conductivity materials from the latent space. We find that the model is able to learn the underlying thermal transport physics of the system under study and is able to predict superlattice-like configurations with high thermal conductivity despite their higher mass. Published under license by AIP Publishing.