I don't think anyone has thought about this yet, so I don't think a good
answer is known. Maybe someone else can chime in?
If you want to look into this yourself, one way to get some insight
might be to simply run the code and output the cuts that are generated.
Maybe a clear picture emerges from this that can then be verified by proof.
To do this, I would print the operator names of the operators in the
cut. A good place to do this is just before the line "cut.clear();" in
lm_cut_heuristic.cc.
Cheers,
Malte