fenics team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Tue, Dec 20, 2005 at 12:38:01AM +0100, Garth N. Wells wrote:
> On Mon, 2005-12-19 at 17:07 -0600, Anders Logg wrote:
> > > 
> > > Setting up PETSc GMRES solver: (rtol, atol, dtol, maxiter) = (1.0e-05, 1.0e-50,
> > > 1.0e+04, 10000).
> > > 
> > > *** Error: GMRES solver did not converge. [GMRES.cpp:71: solve()]
> > > 
> > >  
> > > 
> > > My postulate would be that PETSc with the above configuration somehow gets put
> > > into single precision mode and the atol=1.0e-50 doesn’t stand a change.
> > > However, I can’t find this information (yet) in the PETSc log.
> > > 
> > >  
> > > 
> > > So, any known issues that explain the two issues here?
> > 
> 
> At one stage, not so long ago, the preconditioner was turned off (it's
> on again).
> This lead to a some of the demos not converging, particularly
> when the mesh was refined. Could be an explanation.
> 
> Garth 

Make sure you use the latest version (0.5.11) which should work.

/Anders