20.6 Major release (May 25, 2002)

GAMS System


  • We have two new models (clearlak and srkandw) that demonstrate the use of scenred, an interesting MIP model for scheduling TV commercials, and an example that shows the use of some GDX utilities.


  • This utility converts MPS files into a GAMS program making use of the GAMS GDX facility. This replaces the contributed utility mps2gams.


  • Starting with this distribution the GAMS system for Windows includes a collection of Posix utilities which are usually available for the different Unix systems and therefore help to write platform independent scripts. More information here.


  • ScenRed (a new addition to the GAMS system) allows GAMS users easy access to the scenario reduction algorithms found in [ScenRed] (http://www.mathematik.hu-berlin.de/%7Eromisch/projects/GAMS/scenred.html). Given the event tree for a stochastic program, ScenRed determines a subset of scenarios and the optimal redistribution of probabilities for the preserved scenarios. This is useful when the stochastic program that results from using the original (complete) event tree is too large to solve. Making use of the new execution-time GAMS data interface (execute_load/execute_unload), GAMS/ScenRed takes the original tree from the modeler, along with parameters controlling the reduction, and returns a reduced tree for use in subsequent solves or data manipulation.



  • There are two new options (dfsstay and acceptnonopt) in SBB that can help to find good solutions more quickly as well as handle almost optimal subproblems which are ignored by default.


  • The GAMS/XA link has been modified to work with the new generation (Version 13) of XA libraries. All XA-supported architectures are now using this new version, and so will include a Newton-barrier capability.

Beta Solvers


  • Complete primal and dual solutions values are now reported by using CONOPT as a post processor. The preprocessing has been enhanced to allow free (=N=) equations.


  • MOSEK from MOSEK ApS is a large scale system for solving problems of the following classes:
    • Linear optimization
    • Mixed integer linear optimization
    • Convex quadratic optimization
    • Conic quadratic optimization
  • The released version of MOSEK will also solve
    • Quadratically constrained convex optimization
    • Convex optimization
  • More information about MOSEK can be found at [www.mosek.com] (http://www.mosek.com).


  • OQNLP from Optimal Methods, Inc is a solver for global optimization of smooth constrained problems with either all continuous variables or a mixture of discrete and continuous variables. This multi start method combines mathematical programming approaches with meta heuristics like tabu search.


  • NLPEC is a solver for MPEC models that works by reformulating the MPEC model as an NLP, solving the NLP using one of the GAMS NLP solvers, and then extracting the MPEC solution from the NLP solution. All of this happens automatically, although it is possible to access the intermediate NLP model. Like the CONVERT solver, the reformulated models NLPEC produces are in scalar form. Many different reformulations (currently around 20) are supported by the NLPEC solver. MPEC models are notorious for their difficultly, but the combination of different reformulations and NLP solvers give users a good chance to solve them.


  • PATHNLP is now set up to allow the PATHLIB presolver to perform additional model reductions.
  • PATHNLP now uses the same libraries as PATH. Previous versions used an experimental version of PATHLIB (ver 5.X). Since the supported PATHLIB is now 4.6, it looks like PATHNLP now uses an older PATHLIB; this is not the case.