Sorry, if you have a decent initialization to your search, you will find the local minimum much faster than some random guess. Even a coarse approximation to the local minimum will get you a lot of the way, and help you avoid other local minimums. Copying works, even if you don't get the final product, you get close enough to it so that the last bit is much easier. Of course, this all depends on the local smoothness of the function, but I would guess for something as complex as radar return, the function would be very bumpy on a large scale and only a little bit smoother as scale decreases. If your copy is good enough to get you to a place where the gradient is pointing towards the minimum, you're in luck.