hi, i was wondering how other ppl deal with optimizing functions that are unbounded...
Let say i have a function: lim f(a) = inf or lim f'(a)=inf. we can further assume that the problem is convex (so there is always only one solution: either at lower bound or not)
i have come across two situations in which i had to prevent the optimizer to try to evaluate f(a) or f'(a). basically both limit x to 'a + small number' (say 1e8).
1) if i use fmincon anyway, it just put a lower bound on x (for example: lb=x+1e8).
2) if f'(a) is inf, and i use fsolve for f(x,Q)=b where Q is a parameter, then first i shift the function to the left, so g(x,Q) = f(x+1e8,Q). then i calculate the boundary of Q: g(a,Q_bnd) = b so that for Q<=Q_bnd the solution is "x=a". Then i solve g(x,Q) = b for Q>Q_bnd.
But i am interested in other methods as well... so if anybody has anything to say about it, please do, because i dont feel very comfortable with this type of problem...
