TurtLES: A Large-Scale Benchmark for Turbulent 3D Neural PDE Surrogates
Neural operators tested on TurtLES lose small-scale turbulent energy and struggle to remain stable over long rollouts.
The 14 TB dataset contains 400 simulated trajectories of turbulent 3D cylinder wakes, each spanning 400 time steps. Each frame has 3–11 million mesh points, with variations in geometry, Reynolds number and angle of attack. High-fidelity simulations capture an active energy cascade driven by vortex stretching. The benchmark tests models’ ability to predict successive flow states on large, irregular meshes at industrially relevant resolution.
HF Daily Papers' note
The 14 TB dataset contains 400 simulated trajectories of turbulent 3D cylinder wakes, each spanning 400 time steps. Each frame has 3–11 million mesh points, with variations in geometry, Reynolds number and angle of attack. High-fidelity simulations capture an active energy cascade driven by vortex stretching. The benchmark tests models’ ability to predict successive flow states on large, irregular meshes at industrially relevant resolution.
HF Daily Papers' note
score 5