抄録
This paper discusses equilibrium distributions of a large number of agents by their types, their choices or some other characteristics such as the shares of algorithms when the decision makers are classified according to the rules they use in some optimization or choice situations. Models with a large number of types are considered. We describe the distribution of the order statistics of the fractions of agents as the average of multinomial distributions, conditional on the vector of random fractions governed by Dirichlet distributions. The result is the same as the Ewens sampling formula well known in the field of population genetics. The interaction processes of a large number of agents are modeled as Markov processes on the set of exchangeable random partitions of a set of integers. Patterns of compositions or shares evolve as random partitions of agents. By introducing some simple transition rates for the types of agents a master equation or a diffusion equation approximation describes the time evolution of the population compositons of agents. The characterization given in this paper of distributions may be useful in designing simulation studies or interpreting results of such studies.