There are several approaches to the genetic algorithm (GA) technique (e.g. variations in parameters, birth-death timing, encodings) and these differences typically affect the performance and outcome of the technique. There are some cases where the appropriateness of one variation is clearly better than others, but many more cases of a technique being chosen merely by convention or because something still must be done when there are no differentiating reasons. Part of the problem is that the differences generated by the variations in GA approach are things which are very difficult to measure with existing metrics: dynamical properties relating to the number and location of attractors, their relative basin sizes, the speed of convergence, etc. These are precisely the sorts of properties that I developed the tipping point methodology to be able to measure and so bringing it to bear on a few GA variations of a common problem may yield some valuable insights.

The first step is to find a problem that is hot right now in the GA community. Ideally this would be a problem that researchers in the field have already applied several variations of GA techniques to. This way I could procure the data to run my analysis without having to recreate the problem and solution search myself. It would also be great of there already exists a debate regarding which is the more appropriate technique for the problem space under consideration. That way I could anchor the discussion as evaluating the claims made by each side. And even better would be if that problem had a great deal of practical import so that the gains made from applying my technique can be translated into easily identifiable progress and benefit. But if there is no such problem then I'll just build the varieties of GA analysis for some famous problem(s) and hype the results for their generality to lots of other problem spaces.

Among the important variants that are known to have an effect on the output of a genetic algorithm are 1) generational vs. continuous population dynamics, 2) mutation vs. crossover, and 3) termination criteria. By capturing the changes in dynamics across these variations the tipping points technique (still looking for a better name) will allow us to access scale-free measures of their performance and general characteristics of their operation. These results may allow us to organize setting combinations into classes of equivalent operation (where the variation is irrelevant) and splitting classes where output features are variation dependent. The final result, therefore, will be improved solutions overall and especially improved fit between the GA technique used and the problem to be solved.