Ga Results Worse with Each Generation

I am trying to use the ga function to optimize a function, but am noticing a strange behavior. My function has 5 variables that are bound to between 0 and -10 and there is a equality constraint that in the form of c(x) -1 = 0. I am using the following options:

    options = optimoptions('ga','UseVectorized',true,'PopulationSize',3000,'InitialPopulationMatrix',intpop2,'MaxGenerations',50,...
        'FunctionTolerance',1e-4,'ConstraintTolerance',1e-4,'Display','iter');
    
    

where the initial poptulaiton is a matrix of all combinations of the five variable (equally spaced along a grid between 0 and -10). The problem I am having is that I frequently notice that after several generation the ga function returns worse results. When exaiminging the output of each iteration, an acceptable answer seems to be found after 5 or so iterations; however the function continues and the results become worse. I have tried to change the function tolerance and constraint tolerance but the results are largely the same or become inconsistent (outside of the normal variations when using optimaztion functions). Below is an example of the output I frequently get.

    Generation  Func-count        Best f(x)    Max Constraint  Stall Generations
    1         162500    0.00106595      0.01563      0
    2         314182   0.000272166      0.01934      0
    3         473557    0.00190035      0.01894      0
    4         632932   0.000113225      0.01067      0
    5         792307   4.30226e-05     0.005452      0
    6         943989   0.000232475     0.001222      0
    7        1103364   1.50892e-06     0.001525      0
    8        1262739   1.50892e-06     0.001525      1
    9        1422114   1.50892e-06     0.001525      2
    10        1581489   1.50892e-06     0.001525      3
    11        1740864   1.50892e-06     0.001525      4
    12        1900239   6.42236e-05     0.002838      0
    13        2059614       5.37465          385      0
    14        2218989       3.37523        14.95      0
    15        2378364       3.19792            1      0
    16        2537739       3.25397            1      0
    17        2697114       2.60784            1      0
    18        2856489       3.26691       0.9944      0
    19        3015864       3.31657            1      0
    20        3175239       5.10486       0.5483      0

[John D'Errico]

Let me see if I understand this, in a context we might all recognize. It has been said that every generation thinks the world is going to hell, and that things were better in the old days. (I may have said that myself lately. If you don't believe me, ask your grandfather.) Should we be surprised if GA reports the same thing, that with every generation, things get worse?

Sorry. Somethings just need to be said. ;-)

[Alan Weiss]

I'm not sure that I understand your "equality constraint that in the form of"

c(x) - 1 = 0

Usually, I would call this a nonlinear equality constraint

ceq(x) = 1

I am not sure how you have implemented your nonlinear constraint function. Can you show us what inputs and outputs it gives?

Alan Weiss

MATLAB mathematical toolbox documentation

[agion]

Yes, my mistake I have it set to a nonlinear equality constraint of

ceq(x) = 1

I can't really show all of the inputs and output as it is part of a much larger application with a gui and user inputs that I am writting, but effectvely the input to the nonlinear constraint function is 3 values (a1, a2, and a3) that are calculated by the objective function and within the nonlinar constraint function (a1*a2)/(a3*K) = 1, where K is constant. I should also not that if I remove this constraint the ga function performs as expected.