request of a FGMRES krylov solver
I would like to have a FGMRES Krylov solver in scipy. FGMRES is a variant of the GMRES method with right preconditioning that enables the use of a different preconditioner at each step of the Arnoldi process. A possibility could be to "borrow" the fgmres function from pyamg (which as a compatible license) http://code.google.com/p/pyamg/source/browse/trunk/pyamg/krylov/_fgmres.py Including this code into scipy should be straightforward. Télesphore Bigorneault
2011/1/12 Thélesphonse Bigorneault <bigorneault@gmail.com>
I would like to have a FGMRES Krylov solver in scipy. FGMRES is a variant of the GMRES method with right preconditioning that enables the use of a different preconditioner at each step of the Arnoldi process.
A possibility could be to "borrow" the fgmres function from pyamg (which as a compatible license)
http://code.google.com/p/pyamg/source/browse/trunk/pyamg/krylov/_fgmres.py
Including this code into scipy should be straightforward.
Télesphore Bigorneault
Bump.
2011/1/12 Thélesphonse Bigorneault <bigorneault@gmail.com>:
I would like to have a FGMRES Krylov solver in scipy. FGMRES is a variant of the GMRES method with right preconditioning that enables the use of a different preconditioner at each step of the Arnoldi process.
I vote for this ! Christine
On Wed, Jan 12, 2011 at 5:23 PM, Thélesphonse Bigorneault < bigorneault@gmail.com> wrote:
I would like to have a FGMRES Krylov solver in scipy. FGMRES is a variant of the GMRES method with right preconditioning that enables the use of a different preconditioner at each step of the Arnoldi process.
A possibility could be to "borrow" the fgmres function from pyamg (which as a compatible license)
http://code.google.com/p/pyamg/source/browse/trunk/pyamg/krylov/_fgmres.py
Including this code into scipy should be straightforward.
Télesphore Bigorneault
Hi Télesphore, The only obstacle to moving the PyAMG solvers like into to SciPy is a handful of C++ routines here [1]. In principle we could add this C++ code to scipy.sparsetools, but I'd strongly prefer that the Krylov solvers be pure-Python codes for simplicity. It shouldn't be too hard to implement reasonably fast versions of the routines in [1] using NumPy/SciPy. So, if you could contribute implementations of those routines that were close in performance to the existing C++ I'd migrate the solvers from PyAMG to SciPy. [1] http://code.google.com/p/pyamg/source/browse/trunk/pyamg/amg_core/krylov.h -- Nathan Bell wnbell@gmail.com http://www.wnbell.com/
participants (3)
-
Darcoux Christine -
Nathan Bell -
Thélesphonse Bigorneault