Fusing generative AI and ML-FFs for inverse design and high-fidelity dynamic simulation of fused-ring aromatic hydrocarbons

Rational design of fused-ring aromatic hydrocarbons is hindered by vast chemical space, costly quantum simulations, and a disconnect between generation and dynamic validation. Here, we report an integrated computational platform that unifies physics-constrained generative AI, interpretable multi-objective screening, and high-fidelity machine-learning force fields (ML-FFs) to enable inverse design and dynamic simulation of fused-ring aromatic hydrocarbons. A physical consistency module incorporating the BrownLadner method and iterative molecular weight inversion enforces chemical and structural feasibility from the outset, generating diversity, novelty, and theoretically synthesizable structures validated against experimental benchmarks, while XGBoost predictors accurately forecast key electronic properties. We further developed DeepAsphalt-26, ML-FFs specifically optimized for aromatic systems, which attains density-functional-theory-level accuracy while accelerating molecular dynamics by two orders of magnitude, accurately reproducing π-π stacking dynamics and free energy landscapes. By bridging generative design and high-fidelity simulation, this workflow offers an end-to-end in silico route to engineering carbonaceous materials with targeted properties.