Originally posted by: MarekGrzes

Hello,

I would like to use OPL/CPLEX to solve an optimization problem with non-convex quadratic objective subject to linear constraints. I was reading other posts on IBM forums and it looks that I can solve such a problem approximately (local optima) setting solutiontarget=2. I am trying to run a tiny example to check if this works, but the following code does not work:

dvar float x;
dvar float y ;

execute PARAMS {
  solutiontarget=2;
}

minimize x * y;

subject to {
  0.5 <= x <= 1;
  0.5 <= y <= 1;
}

CPLEX says:

### ENGINE exception: CPLEX Error  5002: Q in objective is not positive semi-definite.


<<< no solution
 

Is there anything else that I need to change to obtain an approximate solution? Obviously, Q is not PSD, but if CPLEX can search for an approximation, it should be able to ignore this requirement.

Kind Regards,

Marek

Hi

you should write

execute PARAMS {
  cplex.solutiontarget=2;
}
 

regards

Originally posted by: MarekGrzes

Thanks!

Originally posted by: S6R9_Dua_Weraikat

hi,

i have a question: what does it mean that:

cplex.solutiontarget=2;

???

Hi

have you had a look at

http://pic.dhe.ibm.com/infocenter/cosinfoc/v12r6/index.jsp?topic=%2Filog.odms.studio.help%2FCPLEX%2FReleaseNotes%2Ftopics%2Freleasenotes126%2FnewNonconvexMIQP.html

?

regards

Originally posted by: S6R9_Dua_Weraikat

Thanks for reply. But for my problem it is not integer. I have linear constraints and for the objective function: I have 3 terms each is the product of two decision variables. (Maximize X1*X2+ X3*X4+X5*X6- x1-2X3...)

I want to know if I can use the same cplex.solutiontarget=2;

to solve my model.

I use cplex studio 12.3

Hi

have you had a look at the example qcpex1.mod in opl\examples\opl\qcpex1 ?

regards

Originally posted by: S6R9_Dua_Weraikat

yes I did.. Thank you. So it is just local optimal not global optimal.