Closed AndreyBychkov closed 1 year ago
Looks like a bug to me: one extra derivative of an input should be always sufficient.
The funny thing, actually, is that both versions are correct. The problem lies in the branching strategy that does not select the correct variable in quadratization (w2^2
) if the upper bound is infinite. So, the actual solution would be to use the domination upper bound or the Hemery-Fages-Soliman M-set.
In the last commit in master it was enough to have
T''
in the system. However, in the next commit in dev with the rework of conditions we must haveT'''
with no obvious reason. I need to prove whetherT'''
is actually the correct result or the rework was messed up somewhere. Below I attach the outputs for both:Master:
Dev: