Optimization on Rugged Landscapes
This chapter reviews the notion of a landscape, a discussion of types and properties of landscapes, a description of some models for rugged landscapes, and finally a review of some novel optimization methods for finding the highest or lowest point on a rugged landscape. Rugged landscapes are a common underlying feature of many complex systems. They are studied from various viewpoints in the physics of glasses and spin glasses, in the biophysics of macromolecules, in the computer science of combinatorial optimization problems, and in the interdisciplinary field of neural networks. Similarity might be defined in terms of similar fitness, or on the basis of physical or chemical substitutability. The landscape may even be self-similar, so that a magnified image of a part would be much like the whole surface, with structure on every scale. The traditional tools of statistical mechanics, and even the new ones developed for complex systems, do not let us optimize on a particular rugged landscape.
