PICT is a differentiable pressure-implicit solver developed in PyTorch with GPU support for fluid dynamics simulations.
The simulator's accuracy in forward simulation and gradients is validated in various benchmarks before demonstrating its ability to learn turbulence models in 2D and 3D.
Both supervised and unsupervised training regimes are applied, along with physical priors, to match flow statistics and learn a stable sub-grid scale model for 3D turbulent channel flow.
PICT is open source and available at https://github.com/tum-pbs/PICT for leveraging its capabilities in optimization and learning in physics simulations.