## I have a MINLP problem which is going to solve. Before I solve this problem, I can somehow get a good feasible integer solution for the problem elsewhere. I anticipate that this feasible solution can be used as the starting point to accelerate the process of getting the optimal solution.

I understand one can use “var.l” to set the initial value of all variables, which is the general way to set starting point for any GAMS solver. However, since most MINLP solvers solve the RMINLP problem, which is NLP acutally, as the first step of the optimization, the integer part of the starting point is ignored by the solvers. For example, SBB uses the initial values of the continuous variables to solve the RMINLP problem and then takes it as the root node. These initial values of continuous variables do accelerate the optimization of RMINLP problem. But when the branch-and-bound process begines, the feasible integer solution as a whole is not used, because the “Best Integer” column is still empty. SBB needs a lot of iterations to get a even worse “Best Integer”, which is actually a waste of labor.

I have found an alternative but imperfect way to do what I anticipate. I use the “tryint” option. But the solver (SBB) still has to visit many nodes until the feasible integer solution I provide is “got” and shown in the “Best Integer” column.

I’d like to know if SBB or any other MINLP solver has the possibility to do exactly what I anticipate (Use the feasible integer solution as a whole, just like how CPLEX uses the feasible integer solution as the “Best Integer” from the beginning of branch-and-bound). Any experience or pointers to relevant literature would also be appreciated.

Thanks a lot!

\

To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.

To post to this group, send email to gamsworld@googlegroups.com.

Visit this group at http://groups.google.com/group/gamsworld.

For more options, visit https://groups.google.com/groups/opt_out.