Different ("optimal") solutions between 0.4.8. and 0.5.0/0.5.6

[svigerske]

Issue created by migration from Trac.

Original creator: coniglio

Original creation time: 2016-07-18 14:43:09

Assignee: @merraksh

Version:

Hi, I'm witnessing two solutions being reported as optimal by two different versions of Couenne (0.4.8 vs 0.5.0./0.5.6; the latter two produce the same result). I'm attaching the output for the three cases and a self-ontained AMPL file containing both the model and the run commands.

[svigerske]

Attachment couenne-ampl-ticket.zip by coniglio created at 2016-07-18 14:44:21

[svigerske]

Comment by coniglio created at 2016-07-18 14:46:00

What I meant with "two solutions" was two different solutions with different objective function value: -30.40816326530603 vs. -30.15646258503444.

[svigerske]

Comment by coniglio created at 2016-07-18 16:21:10

Extra bit of info: Baron 16.4.4 finds the same solution as Couenne 0.4.8:

=========================================================================== BARON 16.4.4 (2016.04.04): 8687 iterations, optimal within tolerances. Objective 30.40816327 Objective lower bound = 30.4081632654, upper bound = 30.4081936736 barstatus = 1, modelstatus = 1 max nodes in memory = 1498 optimum found at node 8687 Baron run time (excluding setup) = 50.02 seconds