Search All of the Math Forum:
Views expressed in these public forums are not endorsed by
NCTM or The Math Forum.


Anna
Posts:
4
Registered:
4/10/13


Re: Fmincon: nonlinear binary problem
Posted:
Apr 12, 2013 6:16 AM


> > I was just thinking that all these 3 constraints can be written as one > > > > x_ik+x_jk>=2*z_ijk, right? > > > > This condition does not exclude the case that z_ijk=0 although both x_ik and x_jk are equal to 1. So z_ijk is not equal to x_ik*x_jk. It depends on your objective function if z_ijk will automatically be chosen to be 1 if x_ik=1 and x_jk = 1.
True!
> > > "Torsten" wrote in message <kk60l8$l9i$1@newscl01ah.mathworks.com>... > > > Because you substitute x_ik*x_jk = min(x_ik,x_jk) , I think you will need z_ijk: > > > z_ijk <= x_ik > > > z_ijk <= x_jk > > > z_ijk >= x_ik + x_jk  1 > > > > Good point Torsten. Theoretically, I would need to index z over k but from other constraints I also know that there may be at most one k for which z_ijk=1 and I don't need to know the detailed information which k it is, but only the indication whether it happens at all. Would indexing z_ij be correct in that case? It's important to me to limit the size of the variables as well. > > I don't think that this is possible. Imagine there are indices k1 and k2 such that, for fixed i and j, x_ik1*x_jk1 = 1 and x_ik2*x_jk2 = 0. Then, if you only introduce z_ij, your problem becomes infeasible.
The original nonlinear objective was obj=sum(i) sum (j) sum(k) x_ik*x_jk*p_ij
Maybe my explanation was not clear, I intend to replace sum(k) x_ik*x_jk=z_ij for every i and j. Would that be fine? It seems to me that all the previously proposed constraints should hold. Thank you for all your comments.
Regards, Anna



