Guessing at a basis

 Top  Previous  Next

One does not have to use GAMSBAS and may not be able to for the very first solve of a model.  One can try to guess at a solution by specifying

•Non-zero marginals for the constraints that are felt to be binding.
•Non-zero levels for the variables that are felt to be non-zero in the solution as follows.
•Non-zero marginals for variable not felt to be in the solution.

For example,

 

OBJECTIVE.m        =    1 ;

RESOURCE.m ("R1")  =    1 ;

RESOURCE.m ("R2")  =    1 ;

RESOURCE.m ("R3")  =    0 ;

OBJFUN.l           =    1 ;

X.l ("X1")         =    1 ;

X.l ("X2")         =    1 ;

X.m ("X3")         =    1 ;

 

This basis if included would cause the simple problem to solve in zero iterations.  But making this kind of guess for a large model would be hard.  Guess in the NLP case must be done with care as the variable levels are more important and provide the starting point for evaluating the Jacobian.