CCDCGAN: Inverse design of crystal structures

Autonomous materials discovery with desired properties is one of the ultimate goals of materials science. We have developed constrained crystal deep convolutional generative adversarial networks (CCDCGAN), which can be used to design unreported (meta-)stable crystal structures using encoded 2D latent space.¹ Such a latent space can also be applied to forwardly predict physical properties like the formation energy. Correspondingly, it is demonstrated that the optimization of physical properties in the latent space can be integrated into the generative model as on-top screening or backwards propagator, both with their own advantages. The CCDCGAN has been successfully applied on a specific binary (i.e., Bi-Se) system and multicomponent systems (i.e., all binary and ternary compounds in the Materials Project database). It is observed that the crystal structures distinct from the known cases can be obtained covering the whole composition range. We suspect that CCDCGAN can be extended to multi-objective optimization, such as band gap and mechanic properties, which paves the way to achieve the inverse design of crystalline materials with optimal properties.

1. Noh, J. et al., Inverse Design of Solid-State Materials via a Continuous Representation, Matter 1, 1370–1384 (2019).