Decision Optimization

Decision Optimization

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


#Analytics
#DecisionOptimization
#DecisionOptimization
 View Only
  • 1.  CPLEX issues with MIQCP

    Posted 08/09/10 04:17 PM

    Originally posted by: SystemAdmin


    Hi all,
    I am getting several errors with solving a MIQCP using CPLEX 12.2.
    (Reduced MIQCP has 1173 rows, 2737 columns, 388 binaries, 1043 indicators, 526 quadratic constraints.)

    1. It works just fine for a while - finds good feasible solutions and moves the best bound. Then suddenly

    CPLEX Error 3019: Failure to solve MIP subproblem.
    • CPLEX Error 3019: Failure to solve MIP subproblem.

    Error solving MIP subproblem.
    Solution available but not proven optimal due to numerical difficulties.
    Returning a primal only solution to GAMS (marginals all set to 0.0).

    Unrecoverable error (integer solution available).

    • Tried different options - miqcp strategy, complementarity tolerance, baralg. no luck.
    increasing complementarity tolerances seems to stabilize the problem a little, but I am not comfortable with this.

    2. Trying solution polishing and mipstart

    Found a feasible solution and polished.
    Solved the same problem using mipstart.
    Mipstart succeded, but the second polishing failed - 'trouble polishing mip start.'

    • The first polishing gives a better solution than the one I get from doing B&B until it crashes.
    so I am trying to repeat this multiple times.

    Did anyone have these problems previously? It would be really great if someone can give me directions.
    Thanks.
    #CPLEXOptimizers
    #DecisionOptimization


  • 2.  Re: CPLEX issues with MIQCP

    Posted 08/09/10 05:38 PM

    Originally posted by: SystemAdmin


    I haven't seen that problem -- possibly because I don't do quadratic anything (my world is entirely linear) -- but you might try turning on the numerical emphasis parameter (its name in CPLEX is NumericalEmphasis in most APIs, and you want to set it to 1). It sounds as though you have some numerical instability issues in one (or more likely a few) of the subproblems.

    /Paul

    Mathematicians are like Frenchmen: whenever you say something to them, they translate it into their own language, and at once it is something entirely different. (Goethe)
    #CPLEXOptimizers
    #DecisionOptimization