dolfin team mailing list archive

[Nuno David Lopes <ndl@xxxxxxxxxxxxxx>]

On Thursday 25 September 2008, kent-and@xxxxxxxxx wrote:
> > On Thursday 25 September 2008, Matthew Knepley wrote:
> >> This is definitely wrong. Dave and I have been doing this all week here
> >> and those times are way way off.
> >
> > Sorry the time i show is the total time from the beggining of the
> > process. I get the following times:
> >
> > ----------------------------------------------------------------------
> > Solving linear system of size 566769 x 566769 (Krylov solver).
> > PETSc Krylov solver (gmres, sor) converged in 35 iterations.
> > Summary of timings  |  Average time  Total time  Reps
> > -----------------------------------------------------
> > ASSEMB_BC            |         24.42       24.42     1
> > Assemble cells         |          5.33       10.66     2
> > Build sparsity           |         6.305       12.61     2
> > Delete sparsity         |         0.125        0.25     2
> > GMRES_SOR            |         35.86       35.86     1
> > Init dof map             |          0.66        2.64     4
> > Init tensor                |          0.16        0.32     2
> > End Process after 69.400000 seconds.
> > --------------------------------------------------------------------
>
> It seems a bit strange that only 35 iterations are required for such a
> fine mesh.
I'm testing with a large relative error 10^-5.
> Do you have a really good start vector?
That was my first doubt,
but for this simple  case zero should be a good start vector, Ax=f
where f is nonzero only on the inflow boundary 

> Or does the initial error only 
> contain
> high frequency errors?
> Kent
>

-- 
Nuno David Lopes

e-mail:ndl@xxxxxxxxxxxxxx        (FCUL/CMAF)
           nlopes@xxxxxxxxxxxxxxx    (ISEL)
http://ptmat.ptmat.fc.ul.pt/%7Endl/ 

Thu Sep 25 18:26:06 WEST 2008