The “D” solvers GAMS/ConoptD, GAMS/ConvertD, GAMS/CplexD and GAMS/LGOD started as research and development versions of the production solvers GAMS/Conopt, GAMS/Convert, GAMS/Cplex and GAMS/LGO. So these “D” solvers offer some interesting new features that eventually will migrate into the production version. As of distribution 23.6 all new features of GAMS/ConoptD and GAMS/LGOD have been migrated to the production version and both GAMS/ConoptD and GAMS/LGOD are merely an alias for GAMS/Conopt and GAMS/LGO. We generally recommend to work with the production version of the solver and only cautiously use the “D” version if the new feature is essential to the model. Backward compatibility is of great concern to us, but with the “D” solvers, the backward compatibility requirements are relaxed (e.g. the ConoptD solver will soon disappear from the list of solvers).
The following describes the differences between the research and development “D” solvers and the production version as of distribution 24.0.
The ConvertD “solver” lacks most Convert options. The only Convert options implemented in ConvertD are:
Analyze: Generates three text files for rows columns and matrix
AnalyzeS: Generates short form of Analyze
Baron: Generates Baron input file
CplexMPS: Generate CPLEX MPS format input file
Dict: Generate Convert to GAMS Dictionary
FixedMPS: Generate fixed format MPS file
Gams: Generate GAMS scalar model. This is the default conversion format used.
Jacobian: Writes GDX version of current point - Jacobian
OSiL: Generates Optimization Services instance Language (OSil) file
The former Convert option
Hessian is now only available in ConvertD and the
Jacobian option has additional information. Moreover a few options are exclusively available in ConvertD:
LocalSolver: Generate LocalSolver input file
DictMap: Generate Convert to GAMS Dictionary Map
Moreover, in-core model communication (
Solvelink=%Solvelink.LoadLibrary%) is supported by ConvertD only.
Similar to Convert/ConvertD, the CplexD link to Cplex lacks a lot of functionality. For example, sensitivity analysis and support for SOS variables and indicator constraints are absent from the CplexD link. But CplexD offers a few facilities of interest to a small community: