This dataset contains filtered data (spectral cut) from three DNS simulations (train, tests, decay) of 3-dimensional homogeneous isotropic turbulence in a \(512^3\) periodic domain. More precisely, the following fields are available: Filtered velocities Filtered transported (passive) scalar Divergence of the SGS term from the transport equation obtained from DNS SGS fluxes (in the three directions) from the transport equation obtained from DNS The scalar is forced on the high spectral wavenumbers, such that filtered data is not impacted in train and tests, while the scalar forcing is removed in the decay simulation. Note that all three simulations are forced on the velocities with an Alvelius-type scheme. The dataset also provide with three different filter sizes: 8, 16 and 32 times from the initial DNS resolution, which give domain sizes of \(64^3, 32^3, 16^3\) respectively. This dataset has been used to train NN models available : https://github.com/hrkz/SubgridTransportNN.

{"references": ["Frezat, Hugo et al. (2020). Physical invariance in neural networks for subgrid-scale scalar flux modeling (arXiv:2010.04663)"]}