Training a PyTorch neural network on cell-centered OpenFOAM fields

This post covers the use of the PyTorch C++ API for approximating cell-centered fields (pressure, temperature, density, etc.) in OpenFOAM using a Neural Network (NN). The training is done in an OpenFOAM application, the code can be generalized and ported into a library as described in this post. The unstructured Finite Volume method in OpenFOAM averages values at cell centers to ensure second-order accuracy on unstructured meshes.