[Bonmin] Understanding IPOPT

[reischl maxi]

Hello all,

i have a basic question on how the IPOPT Solver behaves when calculating nonconvex (MINLP) Problems using Branch and Bound.
I know that the IPOPT Solver finds locally optimal solutions, based on a randomly chosen starting point.

In my Case Bonmin finds Solutions (i did 4000 examples of different parameter values) that are very accurate. (99.82% using a MIP-Gap of 1%).

But there are runaway values. In one example i only found a solution that is 60% of the optimal value (I used full enumeration to determine the optimal solution).

Now i am asking myself how this could happen.
I suspect that due to a unluckyness BONMIN found a feasable solution very fast and accepted it, without solving alot of nodes.

Now my question:
Does Bonmin decrease the lower bound (if one is found when solving different nodes) or is the lower bound fixed? 
With fixed i mean that the first solution found by ipopt (where every integer is relaxed) is never changed.

In my opinion the following could happen:
BONMIN forces decision variables, then IPOPT finds solutions outside the local optimum (found in step one).
Then the lower bound would decrase. Is that the case?

In the case where the found solution was 60% BONMIN could have been unlucky and wasnt able to solve enough nodes to force the lower bound out of the local optimum?

I hope i could make myself understandable.
Thank you very much!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/bonmin/attachments/20130801/417ce5c0/attachment.html>