fmincon ignoring lower function value during convergence

I am trying to optimize (minimize) a non-linear function using fmincon using 2 variables with bounds and non-linear constraints. I noticed in the iterative display of results that fmincon ignored a clearly lower function value, continued iterations and converged to higher function value.

First-order Norm of

Iter F-count f(x) Feasibility optimality step

0 3 1.812411e+03 0.000e+00 9.241e+00

1 6 1.736525e+03 0.000e+00 8.528e+00 8.473e+00

2 9 1.676578e+03 0.000e+00 8.056e+00 7.125e+00

3 13 1.676428e+03 0.000e+00 5.824e+00 1.868e-02

4 16 1.675054e+03 0.000e+00 3.055e+00 1.116e+00

5 19 1.665816e+03 0.000e+00 1.189e+00 8.565e+00

6 22 1.628995e+03 4.332e-05 1.123e+00 3.430e+01

7 25 1.439375e+03 2.384e-03 9.349e-01 1.936e+02

8 28 1.373926e+03 2.591e-03 7.666e-01 7.575e+01

9 32 1.373768e+03 2.586e-03 3.914e-01 2.064e-01

10 35 1.374922e+03 1.155e-03 1.239e-01 4.145e+00

11 38 1.376012e+03 1.310e-07 1.374e-03 3.172e+00

12 41 1.376011e+03 0.000e+00 2.000e-04 1.435e-02

13 44 1.376011e+03 0.000e+00 2.000e-06 1.316e-03 ---------> final function value taken as minimum

I haven't come across such behaviour. I assumed it had something to do with the variables I was using (I was initially using 4 to 5 variables), and so I reduced it down to 2 simple variables but the problem persists.

Could somebody explain what is going on? because I am clueless.

Answer 1

[Alan Weiss]

Notice that the feasibility of the iteration you like is about 2e-3, meaning the constraints are not met at that point. fmincon went on to reduce the infeasibility to zero while not increasing the objective very much.

Alan Weiss

MATLAB mathematical toolbox documentation

[Archana Beena]

OH!!! Okayyy!

Thank you so much!

I get it now. Will keep that information in mind from now on.