Decision Optimization

Decision Optimization

Delivers prescriptive analytics capabilities and decision intelligence to improve decision-making.


#Analytics
#DecisionOptimization
#DecisionOptimization
 View Only
Back to discussions
Expand all | Collapse all

fixing variables and adding inequalities improves optimal objective value - repeat presolve issue?

  • 1.  fixing variables and adding inequalities improves optimal objective value - repeat presolve issue?

    Posted 04/22/17 08:29 AM

    Originally posted by: EXCT_RALF_GOLLMER


    Hi,

    in tracking down a mysterious effect in my decomposition program for stochastic integer optimization I found that the problem is not on the side of my code, but in the results of CPLEX 12.7:

    During an outer B&B I add inequalities (cuts from other scenarios and branching decisions, which in this BIP fix variables) -

    and the result is a smaller 'integer optimal' objective value for the minimization

    n0.sol:   objectiveValue="8941.3604459455964"
    n4.sol:   objectiveValue="8936.9080239019822"

    - which is simply impossible. n0.lp is the starting problem, n4 the one from the 4th node of the outer B&B.

    A diff shows that variables F_15 and F_51 were fixed and 13 inequalities were added to n4.lp in comparison to n0.lp.

    Could you reproduce that? Is there any explanation?

    Best regards

    Ralf

     

    P.S.: The effect is not due to the added inequalities, but the tightest possible simplex tolerances make the difference.

    The attached cplex.log just uses the n0.lp from the tar file and shows different results with default simplex tolerances (where 8.9369080239e+03 is reached), afterward with both feasibility and optimality set to 1e-9 (resulting in 8.9413604459e+03), and finally fixing F_15 to 1, which again gives 8.9369080239e+03 with the tight simplex tolerances.

    Looking at the log I concluded that it might still be the repeat presolve issue. The second Log start in the cplex.log proved: switching repeat presolve off gives the probably correct result also with the tight simplex bounds. But repat presolve works with the default simplex tolerances - no idea why

     


    #CPLEXOptimizers
    #DecisionOptimization


  • 2.  Re: fixing variables and adding inequalities improves optimal objective value - repeat presolve issue?

    Posted 04/24/17 10:09 AM

    Thanks a lot for your report.

    I reproduced the issue here. The problem is not with represolve itself but with some stuff that eventually shows up when doing represolve. I filed this as a bug. If you want to know more details, please send a message to daniel(dot)junglas(at)de(dot)ibm(dot)com.

    The best workaround for this issue is to disable represolve.


    #CPLEXOptimizers
    #DecisionOptimization


Related Content