Asymptotic Analysis of Probabilistic Programs: When Expectations Do Not Meet Our Expectations

Michal Ajdarów, Antonín Kucera, Petr Novotný

Abstract: The computational complexity of a probabilistic program C is traditionally measured by the expected values of certain random variables defined over the runs of C. However, in some cases, this approach may lead to misleading conclusions about the actual runtime behavior of the program. Furthermore, the analysis of expected values is not compositional in general. In this paper, we propose alternative complexity measures for probabilistic programs that overcome some of these difficulties.