[Carpet] Reductions in new version of Carpet

[Thomas Radke]

Luca Baiotti wrote:
> Hallo,
> 
> while trying to fix some testsuites (all carpet testsuites that I am
> aware of fail with the new version), I found that (after adding the
> required activation for InitBase and adding its necessary parameter)
> the results of the norm1 and norm2 reductions are very different from
> before (e.g. norm of the lapse was ~1 now it is ~0.5).
> 
> Is there a problem with the reductions in the new version of carpet?
> What is everyone's experience with testsuites applied to the new version?

I believe all norms which depend on the number of grid points are 
miscalculated in the stable-3 version (alias the former experimental 
version) because they did not take the weight into account properly.

For example, the average norm is defined as sum / count where sum is 
implemented as $$sum := \sum_i w_i a_i$$ in both Carpet versions but 
count is $$count := n$$ in stable-3 and $$count := \sum_i w_i$$ in the 
development version (also in stable-2).

I couldn't find which patch of Erik's actually fixed this though.

-- 
Cheers, Thomas.