In which cases does ga return 'no feasible point found'?

Ga returns 'no feasible point found.' after 1 generation but when I print the population and check its members myself, the bounds [.9,1] and nonlinear inequality equations are satisfied. When I widen the bounds to [.1,1], ga works and finds a solution inbetween the old bounds (around 0.95), but runs very long. I tried to enlarge the constraint tolerance; it works, but then the result is not usable because the constraints are violated.

Which other options can I change?

Or how can I access the best individual which was found by ga?

Answer 1

[Matt J]

You should probably show your nonlinear inequality constraint function. It appears very hard to satisfy. I don't think widening the bounds is helping in any systematic way. Changing the bounds probably just causes the initial population to be distributed differently, so that, purely by luck, a population member lands close to your small nonlinearly constrained region

Answer 2

[Dorothee]

Thanks for your comment. I am using simulation results in the inequality constraint function which are computed in the fitness function which is only called for feasible members after checking the constraints: therefor there were wrong values as constraints. I changed it now to be computed inside the constraint function itself and it works.