How can we know what you did that now causes the problem? Since we do not see your code, we cannot read your mind. You changed something, and it no longer works. Surprise! What, EXACTLY did you change?
Changing the optimizer parameters as you have is generally not (almost never) the solution.
You might use the debugger to put a break point inside your objective function. That will stop it so you can check the current parameters at each call. Or just dump out the values of the parameters, as well as the current objective function produced.
Note that doing so will generate lots of calls at essentially the same location when it computes a gradient. So it will help to understand how an optimizer works, so you understand what it is doing at any point in time.
Finally, there are always problems when you try to use an optimizer on top of another tool like an ODE solver or an integration, or another optimization. The internal computation is only accurate to within some tolerance. So when you change the parameters slightly to compute a gradient, it uses a finite difference approximation to compute the gradient. But if there is error in the result, that gradient computation will HIGHLY magnify any error. Remember that a derivative is essentially delta_y/delta_x, as delta_x approaches zero. But if y becomes essentially a random variable, with a significant noise variance, then the gradient as computed becomes worthless.
The point is that these computations will often be nasty. Essentially your objective function is not differentiable at a fine scale. But fmincon (and other optimizers) assume that it is differentiable.
