order of nonlinear inequality constraints when using fmincon

When using fmincon, I have found that the order in which I build up the nonlinear inequality (cIneqVec) constraint vector impacts the resulting value of the objective function.

Is the order in which I build up cIneqVec important?

Does fmincon somehow prioritize the constraints based on their element position in the vector?

Answer 1

[Matt J]

The order of the constraints would probably cause their gradients and Hessians to be summed in a different order, which in turn would introduce small differences in floating point approximation errors into the results. My guess is that the differences you see are due to that.