Re: ITSOL interface

[Kai Torben Ohlhus]

On 10 July 2013 15:41, c. <address@hidden> wrote:

Hi Kai,

Actually I think it is a good idea to not wait until the very end of the project to start looking at the test cases.
Which makes me think that I had promised you to provide a simple way to generate automatically matrices for testing your preconditioners,
so here is a quick example:

-------------------------------------
pkg load bim

n = 50;

%% set beta = 0 for SPD matrix
beta = 1e4;

msh = msh3m_structured_mesh (n, n, n, 1, 1:6);
inodes = setdiff (1:columns (msh.p), bim3c_unknowns_on_faces (msh, 1:6));
x = msh.p(1,:) .';
A = bim3a_advection_diffusion (msh, 1, 1, 1, beta*x);
b = bim3a_rhs (msh, 1, 1);

%% the matrix to factorize is the one below
A = A(inodes, inodes);

%% solve the system by GE forcomparison
%% n = 80 is already to large to do this on my system;
b = b(inodes);
u = 0*x;
u(inodes) = A \ b;
-------------------------------------

for beta = 0 the matrix will be SPD and its condition number should scale as n^2
otherwise it will be unsymmetric.

n = 80 is already too large to handle via mldivide on my system it'd be very
interesting in to know the maximum size of system you can deal with by ilu+pcg
or ilu+gmres.

Thank you for the example c., but while I was trying to make use of the package installer of octave-forge I found there was a little formatting problem in the description file of "fpl" a dependency of "bim", it should be fixed fast. I created a bug report with more details: https://savannah.gnu.org/bugs/index.php?39464

 

Also, what's your plan about ichol?

c.

After the midterm I want to implement the function myself... AFAIK there is no package containing this functionality. Maybe Ruipeng or Saad have a good implementation, but you told me you had an M-File containing some implementation, too?

Best,

Kai