Originally posted by: NiuX-HNU

Hello everyone:
   I'm using cplex to do some research on MIP problem, but when I try to execute this script I have the error "cplex error 5002: Q in objective is not positive semi-definite."

The model:
  
   int T=...;      
   int nw=...;     
   int ns=...;      
   int np=...; 

   range time=1..T;    
   range wind=1..nw;   
   range solar=1..ns;  
   range price=1..np;  

   float gbmax=...;

   dvar float+ gc[wind,solar,price,time];           
   dvar float+ gch[wind,solar,price,time];
   dvar float+ gb[wind,solar,price,time];
   dvar boolean u[wind,solar,price,time];   
   dvar boolean v[wind,solar,price,time];

   int Cbat=...;
   int At=...;
 
   float effic=...; 
   float effid=...;
   float climt=...;
   float dlimt=...;

  sum(w in wind,s in solar,p in price,t in time)
      pip[p]*(gb[w,s,p,t]/gbmax)*(gch[w,s,p,t]+gd[w,s,p,t])*Cbat/At;  

  subject to{
     ct1
      forall(w in wind,s in solar,p in price,t in 2..T)
      gb[w,s,p,t]==gb[w,s,p,t-1]+effic*gch[w,s,p,t]-gd[w,s,p,t]/effid;

      ct2:
    forall(w in wind,s in solar,p in price,t in time)
     gch[w,s,p,t]<=climt*u[w,s,p,t];      
  
    forall(w in wind,s in solar,p in price,t in time)
     gd[w,s,p,t]<=dlimt*v[w,s,p,t];
          
    ct3:
    forall(w in wind,s in solar,p in price,t in time)
      u[w,s,p,t]+v[w,s,p,t]<=1;
   
    ......
}

       I guess (gb[w,s,p,t]/gbmax)*(gch[w,s,p,t]+gd[w,s,p,t])*Cbat/At' in the objective function may be the reason, because gb[w,s,p,t], gch[w,s,p,t] and gd[w,s,p,t] are all decision variables, that makes the objective function quadratic.

       So I'm wondering how to solve this problem? Are there any methods to linearize this formulation? Thanks very much!

I suggest to search for this error message (cplex error 5002)  in this Forum. Questions very similar to yours have been asked and answered many times.

Originally posted by: NiuX-HNU

Thank you for your generous help. I had add  cplex.solutiontarget=3 before my objective function. Now my script can execute without error, however the computational speed is too slow. Please tell me how to acclerate the computatianal speed.

Am I correct that you started a new thread about improving performance here?

Originally posted by: EdKlotz

Running with solutiontarget = 3 only makes since if you have a truly nonconvex quadratic objective and want so solve the associated MIQP.

Is that the case?   If so, this thread is now a performance issue as Daniel said.   However, if you believe your quadratic objective is actually

convex (or concave and a maximization problem), then the 5002 error typically indicates that the quadratic objective you have created is not

the one you have intended, and you need to identify the part of the associated Q matrix that is indefinite.    I see you are using OPL.   You can

export a SAV file of the model, then run the diagnostic program (written in C or Python, so you'll need access to a C compiler or Python interpreter

and the CPLEX Callable Library). available at http://www-01.ibm.com/support/docview.wss?uid=swg21400047.

Ed