Requirements for evaluation function in evolutionary algorithms For a publication I'm looking for references for generic requirements of evaluation/fitness functions in evolutionary (more specifically genetic) algorithms.
I could come up with some requirements myself (efficient implementation, intuitive results, etc.) but I would like some references to existing work, since it should not be new.
Any reference, preferably to papers but if necessary to books, is appreciated.
 A: I found a (partial) answer to my question. Extensions and replies are welcome!
Could someone especially comment on the 'orthogonal' requirement? The others make sense but I could not find references that talk about orthogonality of metrics in multi-objective evolutionary algorithms.
I currently have the following four requirements:


*

*Efficient Implementation. This one is logical and Koza spends a whole appendix (appendix H) on it in his first genetic programming book [1]

*Intuitive Results. It is clear that best/worst candidates should also have best/worst values in a particular objective. And that the difference in better/worse should be expressed in the value of the metric [2]

*Clear specification. The way it is calculated (the formula) should be easy to understand for the educated reader, such that the values assigned to candidates can be verified

*Orthogonal. Different measures should be orthogonal to each other and should not both punish/reward for a particular aspect.


References:
[1] John R. Koza. Genetic Programming: On the Programming of Computers by Means
of Natural Selection. MIT Press, 1992.
[2] Wolfgang Banzhaf, Frank D. Francone, Robert E. Keller, and Peter Nordin. Genetic
programming: an introduction: on the automatic evolution of computer programs and
its applications. Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 1998
