Active-Learning-Based Generative Design for the Discovery of Wide-Band-Gap Materials

Active learning has been increasingly applied to screening functional materials from existing databases with desired properties. However, the number of known deposited in popular such as ICSD and Materials Project is extremely limited consists just a tiny portion vast chemical design space. Herein, we present an active generative inverse method that combines deep autoencoder neural network adversarial model discover new target property whole The application this allowed us thermodynamically stable high band gap (SrYF5) semiconductors specified ranges (SrClF3, CaClF5, YCl3, SrC2F3, AlSCl, As2O3), all which are verified by first-principles DFT calculations. Our experiments show while itself may sample chemically infeasible candidates, these samples help train effective models for filtering out properties hypothetical created model. effectiveness our approach. source code can be freely downloaded https://github.com/glard/Active-Generative-Design.