Dataset for "COAST-AR: A Physics-aware Autoregressive Transformer for Coagulation-Driven Aerosol Evolution"

The dataset is generated using physics-based sectional aerosol simulations under controlled thermodynamic and aerosol parameter conditions. A uniform sampling strategy is adopted across the input parameter space to ensure balanced coverage and to avoid bias in model training. The key input variables include Temperature (T), Density (D), Pressure (P), initial count median diameter (CMD$_{nm}$), initial total particle number concentration ($N_{tot}$), and initial geometric standard deviation (GSD). Each simulation is performed for 15 discrete timesteps to capture the temporal evolution of aerosol moments under coagulation dynamics. In total, approximately 94,000 simulations are generated, resulting in a large time-resolved sequential dataset suitable for training deep learning models. The uniformity of sampling across the parameter space is verified through statistical analysis and histogram-based diagnostics.