Originally posted by: Moslem Zamani

I introduce an algorithm for non-convex quadratic programs. I would like to check its efficiency compared to CPLEX. I have two stopping criteria, max CPU time and absolute tolerance gap. Unfortunately, I could not find options related to tolerance gap for quadratic programs. I only saw this option for MIP which quadratic programming was not sensitive to it or there were other stopping criteria, which I do not know how to set them. My code is as follows:

 n=size(H,1);
   m=size(A,1);
   cplex = Cplex('TEST');
   cplex.Param.optimalitytarget.Cur=3;
   cplex.Param.timelimit.Cur=1000;
   cplex.Param.mip.tolerances.absmipgap.Cur=10^-3;
   cplex.Param.mip.tolerances.integrality.Cur=0;
   cplex.Model.sense = 'minimize';
   cplex.addCols(2*f, [], zeros(size(A,2),1), Inf*ones(size(A,2),1));
   cplex.Model.Q   = H;
   cplex.addRows(-inf*ones(m,1), A, b);
   cplex.solve();

I wonder if you could help me. Moreover, if CPLEX provides lower bound for quadratic programs.

Originally posted by: EdKlotz

The mip gap does apply to nonconvex quadratic programs, even when they have all continuous variables.   Since the spatial algorithm used involves branching (albeit a more generalized form), we just use MIP parameters rather than add further clutter to the list of parameters by having separate parameters for the spatial branching algorithm.

Originally posted by: Mark L. Stone

Ed, I think the documentation can be made clearer in this regard. Perhaps documentation of applicable parameters suffers for global QP solving due to it being the new kid on the block.

Originally posted by: EdKlotz

Thanks Mark.  I just had a quick look at the user guide and tend to agree with you.   I shall bring it up with the relevant people momentarily.