CODEWITHSUNDEEP
cplexminizinc
suresh_chinthy
suresh_chinthy

Reputation: 377

Non linear optimization Minizinc and Cplex

I have a simple non linear code built in minizinc as follows.

array[1..3] of var 1..10:x;
array[1..3] of var 10.0..20.0:y;
var float:z;
constraint sum(i in 1..3)(x[i]*y[i])=z;
solve minimize z;

If I use Cplex solver in minizinc background I have an error of Float_times.

unable to create linear formulation for the `float_times` constraint. This model instance cannot be solved using a linear solver.

If I convert the same code to OPL and use them in IBM ILOG Cplex IDE It provide the answer.Sample code as follows.

dvar int x[ 1..3]  in 1..10;
dvar float y[ 1..3]  in 10.0..20.0;
dvar float z;
minimize z;
subject to {
cons1:
sum(i in 1..3)x[i]*y[i]==z;
}

Could you someone explain why it is happening even though I used the same solver in both backends.

Upvotes: 0

Views: 629

Answers (1)

Alex Fleischer
Alex Fleischer

Reputation: 10059

your constraint is not linear. When you solve with OPL the presolve did everything but if you remove presolve then you need to change optimality target:

execute
{
  cplex.preind=0;
  cplex.optimalitytarget=3;
}

dvar int x[ 1..3]  in 1..10;
dvar float y[ 1..3]  in 10.0..20.0;
dvar float z;
minimize z;
subject to {
cons1:
sum(i in 1..3)x[i]*y[i]==z;
}

works fine

So you could either set that setting "optimality target" through minizinc

Or rely on logical constraints / indicators

execute
{
  cplex.preind=0;
  //cplex.optimalitytarget=3;
}

dvar int x[ 1..3]  in 1..10;
dvar float y[ 1..3]  in 10.0..20.0;
dvar float z;
dvar float xy[1..3];
minimize z;
subject to {
cons1:
sum(i in 1..3)xy[i]==z;

forall(i in 1..3) forall(j in 1..10) (x[i]==j) => xy[i]==y[i]*j;
}

works fine too

Upvotes: 2

Related Questions