ARKI Consulting & Development A/S

ARKI Consulting & Development A/S, the developer of CONOPT, has released the first version of SBB = Simple Branch and Bound, a new GAMS solver for MINLP models. The new solver is included on the latest CD from GAMS labeled "Distribution 19.5".

Modelers with an interest in MINLP models can get a free 3-month demonstration license. Contact your GAMS Distributor or GAMS Development Corp.

Modelers with large and/or difficult MINLP models are encouraged to submit test-models to adrud@arki.dk. These test-models will be used in future development and testing of the new SBB and of the existing DICOPT solvers.

SBB has been designed to complement the exiting GAMS/DICOPT solver. SBB applies a branch and bound approach directly to the nonlinear model whereas DICOPT handles the combinatorial aspect by solving a sequence of linearized MIP models. SBB seems to work well on models where the nonlinear aspects are significant and DICOPT has difficulties with feasibility. DICOPT seems better suited to models with a significant combinatorial part where SBB has to solve too many NLP sub-models.

The initial version of SBB uses CONOPT or CONOPT2 as an NLP solver. Future versions will also be able to use MINOS and SNOPT.

Additional information on SBB can be found in the GAMS/SBB Documentation.