Minimal version number for scipy about pyfvtool HOT 3 CLOSED

Thanks for taking care of this (a03e999).

It is a very good thing that pyfvtool only depends on the standard scientific packages numpy, scipy and matplotlib. As such it will readily work with any reasonably up-to-date scientific Python installation. For me, this is important: it makes it easy to share with students and colleagues without running into dependency problems and elaborate installation recipes.

We have the luxury with PyFVTool of being able to 'simply' use a direct method for solving the sparse matrix equation using scipy.sparse.linalg.spsolve (I believe it uses the UMFPACK code which is also used by Matlab -- or it uses SuperLU?). For more demanding systems, there is a choice of iterative solvers already in scipy.sparse (although I wouldn't know how to properly use those...). Farther beyond, I think we would be in FiPy territory and switch to unstructured meshes and high-performance solvers such as PETSc. Probably out of the intended scope of PyFVTool. Your choice to use standard spsolve is an excellent choice for pyfvtool.

A few years ago, scipy.sparse was not in optimal shape. I understand that it is now rapidly improving and even considered a priority (https://scientific-python.org/grants/sparse_arrays/). Better use recent versions of scipy.

from pyfvtool.

From version 1.8.0 onwards, SciPy has made significant improvements to the sparse array implementation, it seems. Another good reason to define a minimal version number for scipy.

from pyfvtool.

That is a very good point. I'll be sure to take care of it in a moment. When I started the development, I had no idea which sparse package to use. I think I read an article about the new improvements in csar_array and how it will be more consistent with numpy arrays so I just used it. I also saw a nice trick by @gmweir that could make pyfvtool work with the older scipy versions.

from pyfvtool.