If you print pressure_rhs.l2_norm() in the solution section you get a
number in the order of hundreds (~400) and you are using this as
tolerance factor. I think this number is used for the algorithm as exit
condition, when the difference between 2 iterations is less than this
number. I set the number to 1.e-10 and it converges to the right
solution (i mean, to the boundary value). Unfortunately I have no
experience working with discontinuous functions so I cannot give an
opinion in that.


On mar, 2012-07-03 at 10:01 +0200, Wolfgang Bangerth wrote: 
> > Thank you very much for looking at this.  Yes, the correct solution
> > is uniform at the same value as that on the boundary.  The reason the
> > solver tolerance values are so low, is that otherwise no solution can
> > be found, as the computer is trying to give a discontinuous solution
> > for a problem with continuous FE space.
> 
> This can't be right. If you set the tolerance to system_rhs.l2_norm() 
> then even the zero vector is a sufficiently accurate solution. In any 
> case, you are mixing discretization error (that's what you talk about 
> when you consider approximating a discontinuous function with continuous 
> finite elements) and iteration error (that's when you talk about solving 
> a linear system Ax=b inexactly using an iterative solver). The two are 
> independent, i.e., you can solve a linear system accurately even if the 
> solution you get this way is a poor approximation of the continuous 
> solution.
> 
> If you can't solve the linear system to high accuracy, then you need to 
> find out why not. Just papering over it by setting the tolerance high is 
> hiding the problem.
> 
> Best
>   W.
> 
> ------------------------------------------------------------------------
> Wolfgang Bangerth               email:            [email protected]
>                                  www: http://www.math.tamu.edu/~bangerth/
> 
> _______________________________________________
> dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii


_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii

Reply via email to