I happened across a project of sorts, called Body and Soul Mathematics
(bodysoulmath.org). It looks like there are 3 volumes of books
available commercially, and the publisher wants control, so there is
are only snippets of freely available at the project.
At the moment, Volume 4 is not fully revised, and so you can get a
complete volume 4 (400+ pages) for free. This volume is on solving
incompressible fluid flow problems.
Other volumes are planned.
These books revolve around the python supported FEnICS project, and
typically use some kind of Galerkin method for solving problems.
Debian has a fenics package, I assume most other distributions will as
If you have a "big" problem, you might think it nice if you could throw
your GPU at it. Parallel support for fenics is via PetSC (or however
it is capitalised). Something like CUSP? In 2013 the answer was, it
may not be worth it. Today? I don't know.
There is some kind of CUDA support, I didn't look at it. The other
support is OpenCL, which is multi-vendor. Unless you are running the
very top end GPUs, your GPU may only provide single precision floating
point. This doesn't mean it isn't useful, as some algorithms can
handle that. But it probably means you have to build petsc from source
(and the leading edge one), and then maybe build fenics from source.