Lettuce: PyTorch-Based Lattice Boltzmann Framework

The lattice Boltzmann method (LBM) is an efficient simulation technique for computational fluid mechanics and beyond. It based on a simple stream-and-collide algorithm Cartesian grids, which easily compatible with modern machine learning architectures. While it becoming increasingly clear that deep can provide decisive stimulus classical techniques, recent studies have not addressed possible connections between LBM. Here, we introduce Lettuce, PyTorch-based LBM code threefold aim. Lettuce enables GPU accelerated calculations minimal source code, facilitates rapid prototyping of models, integrating simulations PyTorch's automatic differentiation facility. As proof concept combining the LBM, neural collision model developed, trained doubly periodic shear layer then transferred to different flow, decaying turbulence. We also exemplify added benefit framework in flow control optimization. To this end, spectrum forced isotropic turbulence maintained without further constraining velocity field. freely available from https://github.com/lettucecfd/lettuce.