No, it doesn't depend on human analysis. It depends on the algorithms you use for solving it. E.g. Calculate one solution. Variate coordinate. Calculate again.

No, it doesn't depend on human analysis. It depends on the algorithms you use for solving it. E.g. Calculate one solution. Variate coordinate. Calculate again. If better optimum, mark this as top choice for now. Calculate the improvement to past value. Iterate until improvement change is negligible or answer is good enough sufficiently. Of course, this can take a lot of time and CPU power (even too much), and can end up in a local minimum if you don't variate the starting point and coordinates enough or include some random changes as well so that the whole ensemble space is well covered. This may (and sometimes strongly does) also depend on the random number generator's biases and it's quality of true randomness that you use in variating the coordinates. You can keep a list of several local minima, maybe even see to it so that they are not "close" to each other (or equal to each other) when the algorithm goes through the area, so that it avoids already known local minima and tries to just seek more.