[BCP] Loss of Primal Fesibility when Branchig

[Francois Margot]

Hi Sepher:

>
> During the implementation of my branch-and-price algorithm I have
> encountered a problem when while doing branching, in some nodes,  the
> primal feasibility is being lost. I have tried to check the status by
> using
>
> getLpProblemPointer()->lp_solver->initialSolve() and it seems that
> initialSolve() is ending with an optimal solution where it is not the case
> when getLpProblemPointer()->lp_solver->resolve() is used. In the latter
> case, the optimal basis is not avaialble and the primal feasibility is
> being lost, for example, I get the following messages
>
> LP: **** Processing NODE 7 on LEVEL 4 (dived) ****
> LP: *** Starting iteration 1 ***
> Clp0006I 0  Obj 1789 Primal inf 0.9999998 (2)
> Clp0006I 1  Obj 1801
> Clp0001I Primal infeasible - objective value 1801
> LP:   Matrix size: 81 vars x 16 cuts
> LP:   Solution value: 1801.0000 / 36 , 1
> optimal basis is available ()  = 0
> I should note that the during baranchig, I change the upper bounds of some
> of variables, based on my branching decision and also I am not using
> strong
> branching. Also, it is not possible for my relaxed master problems to have
> be primal infeasible because of the artificial variables that I have added
> in the restricted master problem for having a feasibl solution in each
> node
> after branching.
> The LP Solver that I am using is CLP and I am using Bcp-1.3.4 verision.

Can you try to run this with Cplex for example instead of Clp? If the
issue appears only with Clp, it might be related with the Clp tickets

https://projects.coin-or.org/Clp/ticket/54
https://projects.coin-or.org/Clp/ticket/55

Francois