Re: [Help-glpk] MIP bounds for cutting

[Brady Hunsaker]

On Tue, 2004-10-26 at 08:05, Joao Pedro Pedroso wrote:
> Dear GLPK specialists,
> 
> I am using the GLPK MIP solver in a situation where I know a primal
> feasible solution.  I would like to tell this solution's objective
> value to the solver, so that branch-and-bound may cut some extra
> branches.  Is there a way to do this?
> 
> I believe that the branch-and-bound is doing it through the variable 
>       double best;
>       /* incumbent objective value, that is the objective value which
>          corresponds to the best known integer feasible solution (it is
>          undefined if the flag found is not set); this value is a global
>          upper (minimization) or lower (maximization) bound for integer
>          optimal solution of the original problem being solved */
> in the structure MIPTREE; is it safe to change this variable directly?
> 
> Thank you,
> 
> Pedro

I don't think GLPK currently has an easy way to do this.

You are correct about the use of 'best' in the MIPTREE object.  I think
it might "work" to just change its value, but this could lead to
problems if the algorithm attempts to print the solution (not just the
objective value), since no solution exists.

A cleaner solution would be to add an api function that submits a
solution.  This routine could be modeled on record_solution in
glpmip2.c.  It should check feasibility and integrality and then record
the solution and objective value as in record_solution (tree->found,
tree->best, and tree->mipx[]).

Brady

-- 
Brady Hunsaker
Assistant Professor
Industrial Engineering
University of Pittsburgh
http://www.engr.pitt.edu/hunsaker/