[BCP] Antw: Re: Primal and dual solution are not the same

[Laszlo Ladanyi]

Hi Andrea,

Just a quick question: what bounds do you define the algorithmic 
variables with? As >=0, or as having lower and upper bounds? In the 
latter case do you compute and use the dual values corresponding to the 
upper bound constraints on the variables?

--Laci

On 08/29/2012 07:47 AM, Andrea Trautsamwieser wrote:
> Hi,
>
> I still have the same problem. I tried different solvers (Clp, Cplex, Xpress) and all of them lead to the same results. I also installed lapack and blas for numerical stability of the LP solver and turned the presolve off in Xpress. However, nothing helps.
>
> The dual objective does not equal the primal objective in some of the iterations.
>
> The funny thing though is the following observation. The dual and the primal objective are different in one iteration. I generate an additional variable that is actually already generated due to false dual variables. However, in the next iteration I obtain the same solution (primal solution) and now the dual solution is equal and therefore the dual variables are correct.
>
> Has anybody any idea why this happens?
>
> Thanks, Andrea
>
>
>>>> "Andrea Trautsamwieser" <andrea.trautsamwieser at boku.ac.at> 14.08.2012 07:30 >>>
> Do you mean the fixing of variables due to reduced cost? I have turned these variables (BCP_DoReducedCostFixingAtZero, BCP_DoReducedCostFixingAtAnything) off.
>
> I have no access to Cplex. That's why I am using the other two solvers.
> Andrea
>
>>>> Sepehr Nemati Proon <sepehr.nemati at gmail.com> 13.08.2012 19:04 >>>
> Hi Andrea,
> Have you tried it when you turn off the fixing variables? Also, i woulld use CPLEX if i were you.
> Sepehr
>
> On Mon, Aug 13, 2012 at 5:10 AM, Andrea Trautsamwieser <andrea.trautsamwieser at boku.ac.at> wrote:
>
>
> Currently I am using Clp. But I also tried it with Xpress. It doesn't matter. I am using Bcp-1.2.3 and Bcp-1.3.4.
>
>
> Andrea
>
>>>> Sepehr Nemati Proon 12.08.12 14.21 Uhr >>>
>
> Which solver are you using?
> Sepehr NP
>
> On Fri, Aug 10, 2012 at 7:24 AM, Andrea Trautsamwieser <andrea.trautsamwieser at boku.ac.at> wrote:
>
>
> Hi,
> I want to use the dual variables for improving my algorithm, However, I found out that the dual solution does not equal the primal solution in each node, although it should.
> The termcode of the BCP_lp_result is 2, so it should be solved to optimality. I have tried to solve small test instances to check the error, however I have not find out what is going wrong exactly.
> The primal solution contains some of the new generated variables, but not all of the dual variables are adapted accordingly. It seems that the dual neglects at least one of the new algorithmic variables and so I obtain different objective values and the algorithmic variable is generated again in the next column generation iteration.
> Has anyone any idea to that problem?
> Thank you very much,
> Andrea
>
> _______________________________________________
> BCP mailing list
> BCP at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/bcp
>
>
>
>
>
>
>
> _______________________________________________
> BCP mailing list
> BCP at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/bcp
>

-- 
Laszlo Ladanyi
IBM ILOG CPLEX R&D