Dual processes and an invariance result for exchangeable models in population genetics (Q1079141)

The paper considers non-overlapping generations genetic haploid models with two alleles \((A_ 1\) and \(A_ 2)\) and no selection. Population size is fixed and mutation is allowed with given probabilities. In generation t, each individual i \((i\in I=\{1,2,...,N\})\) gives birth to \(\nu_ i\) offspring. The r.v.'s \(\nu_ i\) are exchangeable, independent of family sizes in other generations, with joint distribution independent of t, and with \(\nu_ 1+...+\nu_ N=N\). These offspring constitute generation \(t+1\) and are randomly labeled from 1 to N. A particular case is the Wright-Fisher model, where the distribution of the \(\nu_ i's\) is symmetrically multinomial. The set \(\zeta_ t\) of individuals of type \(A_ 1\) in generation t is a Markov chain with state space \(2^ I\); let X(t) be the number of its elements. This set is constructed using a graphical procedure [inspired by \textit{T. E. Harris}, Ann. Probab. 6, 355-378 (1978; Zbl 0378.60106)] and is associated with a random graph (the percolation substructure). A dual percolation substructure is obtained, which can be interpreted as an interactive particle system. For large population size N, using a suitable rescaling of time, it is shown that the dual processes converge to the same limiting dual process. The proof follows ideas of \textit{J. F. C. Kingman} in several 1982 papers. The limiting results allow approximate expressions for the transient behavior of the process, particularly heterozigosity and moments of X(t). Approximate results for the problem of line descent are obtained. The paper ends with diffusion approximations.