Fixed-population and evolving-population multialgorithms for optimization lie at the confluence of traditional optimization methods, evolutionary algorithms and MIMD parallel computers. They hold the promise of more effective, adaptive computer routines than are available today.
The basic formulation of a multialgorithm, which is rooted in fundamental concepts of evolutionary biology, is first considered within a very specific setting (nonlinear conjugate gradients). A detailed numerical illustration is also provided. Then the underlying ideas are discussed in full generality, in particular with regard to their broad-ranging implications for algorithm formulation, effective use of MIMD computers, comparative testing of algorithms, development of adaptive software libraries, and the two-way traffic between multialgorithms and evolutionary biology. Applications to several classes of traditional optimization algorithms, which naturally give rise to multialgorithms, are considered in the discussion.