How sauddy/grinn works
Project type and purpose
• This is an academic/research codebase (“open-source tool”) for solving hydrodynamical PDEs (with self-gravity) using two approaches: traditional finite-difference (FD) solvers vs. Physics-Informed Neural Networks (PINNs).
• The README confirms the goal: study Jeans instability in self-gravitating gas and compare wave propagation and instability growth between FD and PINN methods.
File structure
• Top‐level directories correspond to experimental tracks:
• FD_vs_PINN – Jupyter notebooks comparing FD vs. PINN in 1D & 2D.
• HYDRO-FINITE-DIFF – pure FD implementations (LAX scheme in 1D, 2D, 3D).
• GRINN_torch_code – core Python modules (LAX.py, Plotting.py) plus a demo notebook using PyTorch (“torch_GHD_1D.ipynb”).
• PINNS-Poissons, PINNs-Hydro-Gravity, PINNs-Hydro – PINN experiments on Poisson, hydro+gravity, and hydro-wave equations.
• paper-case_studies – extended 3D wave / hydro‐wave case studies.
• Support artifacts:
– data files (train.dat, test.dat, loss.dat) in PINNS-Poissons
– output folders (figures_*, 1D_fig, 2D_fig_)
– Jupyter checkpoint directories
• No deployment or CI; this is a research repo—no microservices or MVC pattern; instead a “library + experiment notebooks” structure.
README for additional insights
• Confirms use of Physics-Informed Neural Networks (likely PyTorch under the hood).
• Shows animated outputs, indicating a data‐flow: PDE definitions → solver → result → visualization.