[dune-fem] Strange initialisation of tolerance in CG stopping criterion

[Andreas Dedner]

On 27/05/14 16:44, Sacconi, Andrea wrote:
> ", sqr(Residuum):"
> >><< residuum << std::endl;
Sorry I missed the point here - but I think I still do (only differently):

The variable residuum contains the square and the verbose output states 
that:
sqr(Residuum)

Of course the text shouldn't be in German an probably
using ^2 instead of sqr would be more appropriate since it is not so 
easy to confuse with sqrt.
But there is not really a mistake here or am I missing something?

Btw: the stopping criteria does make sense in many cases, consider
a Ax = a b
with some constant real number a - you would like the cg method to 
compute x with the same
accuracy independent of the value of a. Of course there are cases where 
a relative error is
not a good idea and having some switch would be a good thing. Nobody has 
really worked on a
good concept for defining stopping criteria which could be usefully 
applied to all the different solvers
available in dune-fem. We have discussed it a few times but....
Andreas