Unstructured VMS Mesh CFD Trajectories Dataset for Physics-Constrained GNN

This dataset contains reference 2D numerical simulation trajectories of unsteady incompressible fluid flows around circular cylinders with varying diameters and spatial positions. The data was generated using a Variational Multiscale (VMS) finite-element formulation for 150 geometric configurations split across three nominal Reynolds numbers (Re_nom in {10^2, 10^3, 10^4}).

It is structured for graph neural network (GNN) surrogate modeling on unstructured meshes (e.g., PyTorch Geometric, DGL) and accompanies the manuscript: "Physics-Constrained Graph Neural Networks for Autoregressive Prediction of Unsteady Cylinder Wakes".