Runtime Analysis of Evolutionary Algorithms

Runtime analysis asks how many fitness evaluations evolutionary algorithms need to reach an optimum or approximation, and how that cost depends on the problem and algorithm settings. Researchers use drift and fitness-level arguments, hitting-time and probability methods, and Markov models to analyze algorithms such as the (1+1) EA, genetic algorithms, and multi-objective methods on benchmarks including OneMax, Jump, and Pareto-front problems. The work examines mutation and crossover, population size and diversity, adaptive parameters, and settings such as dynamic or noisy objectives. A central challenge is that performance can change sharply with problem structure and parameter choices, while the complex population dynamics that may explain practical behavior remain difficult to bound tightly.