fig3

Figure 3. Generate-accelerate-validate workflow for nanomaterial reconstruction under operating conditions. Generative models propose condition-dependent candidate structures, MLIPs accelerate configurational sampling, and DFT/AIMD together with experimental observations validate and refine the structural ensemble. The validated information is fed back to improve subsequent sampling and ensemble inference through active learning. AI: Artificial intelligence; MLIPs: machine learning interatomic potentials; DFT: density functional theory; AIMD: ab initio molecular dynamics.