Title:Accelerated Discovery of Vanadium Oxide Compositions: A WGAN-VAE Framework for Materials Design

Abstract:The discovery of novel materials with tailored electronic properties is crucial for advancing modern device technologies, yet traditional approaches rely on time-consuming empirical methods. We present an inverse design framework combining enhanced Wasserstein Generative Adversarial Network (WGAN) with specialized Variational Autoencoder (VAE) to accelerate stable vanadium oxide (V-O) discovery for electronic applications. Our framework introduces a WGAN architecture with integrated stability constraints and formation energy predictions, plus a refined VAE utilizing voxel-based representation that captures atomic positions and lattice parameters while maintaining chemical validity. Applying this framework, we generated 451 unique V-O compositions, with 91 structures identified as stable and 44 as metastable under rigorous thermodynamic criteria, including novel V2O3 configurations below the Materials Project convex hull. Electronic structure analysis using spin-polarized DFT+U calculations revealed distinct electronic behaviors, including promising half-metallic characteristics for spintronics and quantum computing applications. Our approach achieves approximately 20% stability rate under strict criteria compared to previous benchmarks, demonstrating its effectiveness for accelerated materials discovery.