[newton] fixed bugs with residual being exactly 0

[HenKlei]

Addresses #927.

This merge request fixes a bug that occurred when the residual in the Newton algorithm becomes exactly 0.

[codecov]

Codecov Report

Merging #932 into master will not change coverage.
The diff coverage is n/a.

Impacted Files	Coverage Δ
src/pymor/algorithms/newton.py	86.53% <ø> (ø)

[sdrave]

Thanks for the fix, @HenKlei. Maybe it would be more elegant to require atol/rtol to be >= 0 and to let both default to 0? What do you think?

[HenKlei]

Yes, I think that's better. That was also my first idea. Then we can remove the first if-condition, I think. As far as I can see, the second one is required nevertheless.

[sdrave]

As far as I can see, the second one is required nevertheless.

Don't think so. If error_squence[-1] is not the current but the last error. And if it was zero, then we would have aborted the loop?!

[HenKlei]

Don't think so. If error_squence[-1] is not the current but the last error. And if it was zero, then we would have aborted the loop?!

Yeah, you're absolutely right. Having 0 as default value for atol and rtol, that should work.

[HenKlei]

I think this is the simplest solution now. The test is also working as expected. Thus, merging is fine with me.
@sdrave Do you have any objections or further comments?

[sdrave]

Looks good to me! I will merge as is.