Research testing of numerics

# how should we test functions, that yield similar, but not equal arrays?

I have a feeling that this might be a bit complicated. Some research is needed. I saw something like this in a blog post on http://software-carpentry.org
