Solver drifts away from chi square minimum

Created by: revoltek

@darafferty and I noticed that, with a higher tolerance, solutions show less outliers (the opposite of the expectation) - this has been tested on DDE solver diag+rot, 1 direction. In the image attached left is with tolerance 0.1, right is with tolerance 0.001. Also the problem outlined in #81 (closed) seems less prominent in this case. It seems that the solver drift away from the correct minimum if it has enough iterations to do it.