The expensive evaluation justifies several other techniques that might not be worthwhile in more conventional applications. To improve convergence we implement not only a GA crossover operator, but also several operators that are based on standard practices and “rules of thumb” from domain experts. We do not attempt to apply these intelligently, as a skilled engineer would do. We simply allow the GA to apply these operators as additional ways of evolving the population. We might think of these from an evolutionary point of view as “genetic engineering” rather than random mutation and crossover.