Retired mathematician-theoretical physicist living on a tropical island---Foldit is a nice hobby when staying out of the sun.

The maximization techniques used for foldit are basically locallized gradient searches in a constrained path-space---each perturbation of the protein leads to a valuation of a "directional derivative"----the real problem is that the constrained path-space has infinite dimension and if the local extrema are isolated and "far" from eachother it can be very frustrating to try to improve the score by minor perturbations.

Just think of being on a mountain-top---blindfolded with a long extendible cane. You want to go "Up-hill"---you reach out with the cane horizontally and if you touch something you move in that direction---else just try another direction. If the cane is too short then you have to extend it(i.e. by using larger perturbations).

Good Folding to All!