The largest table in Chinese restaurant processes (Q1042976)

The authors compute the expectation of the size of the largest table in an \((\alpha, \theta)\)-Chinese restaurant process. To be more precise, suppose that in a Chinese restaurant there are countably many tables labeled by \(1,2,\dots \), and there are \(n\) customers occupying first \(k\) tables with \(n_i\) customers sitting at the \(i\)-th table. The next customer who arrives sits at the \(i\)-th occupied table with probability \((n_i-\alpha)/(n+\theta)\), \(i=1,2,\dots, k\), or at the first unoccupied table with probability \((\theta+k\alpha)/(n+\theta)\). Here either \(\alpha<0\) and \(\theta=-m\alpha\) for an integer \(m\geq 2\), or \(0\leq \alpha<1\) and \(\theta >-\alpha\). Such random arrangement is called an \((\alpha,\theta)\)-Chinese restaurant process. Denote by \(|\Pi_n^i|\) the number of customers at the \(i\)-th table at stage \(n\), and let \(L_{1,n}\geq L_{2,n}\geq \dots\) be its size-ordered rearrangement. The main result of the paper is an explicit formula for \[ \lim_{n\to \infty}{\mathbb E}\left(\frac{L_{r,n}}{n}\right)^p\, , \] where \(p\) and \(r\) are any positive integers. The proof relies on a relation between the Chinese restaurant process and the number of jumps in a certain Poisson process.