Benchmark dataset Turbulent Channel Flow for Chaos Meets Attention: Transformers for Large-Scale Dynamical Prediction

This dataset serves as a benchmark for 3D turbulent channel flows, based on simulations performed using a high-fidelity lattice Boltzmann method (LBM) solver, as described in Xue et al., Phys. Fluids, 34,5, 2022.

It comprises 240 trajectories generated from 3D periodic turbulent channel flow simulations with a fixed relaxation time, $\tau = 0.5025$. We extract the central cross-section of the domain along the streamwise ($x$) direction with 3 coordinate components. The spatial resolution is $192 \times 192$, and the friction Reynolds number is set to $Re_{\tau} = 180$, equivalent to $Re = 3250$. The dataset is split into 192 training, 24 validation, and 24 test trajectories, all provided in .npy format.

This dataset is designed to facilitate machine learning research in dynamical systems, especially in the challenging context of high-dimensional, turbulent flow regimes.