Fluctuations of balanced urns with infinitely many colours (Q6165988)

The authors establish limit theorems for measure-valued Pólya processes (i.e, Pólya urns with infinitely many colours). A measure-valued Pólya process is a Markov processes \((\mathfrak{m}_n)_{n\geq0}\) taking values in the set of positive measures on the measurable space \(E\) of colours, parameterised by an initial composition measure \(\mathfrak{m}_0\) at time zero, and a replacement kernel \((R_x)_{x\in E}\), a family of positive measures on \(E\). At time step \(n=1,2,\ldots\), a random colour \(Y_n\) is drawn from \(E\) according to the probability distribution \(\mathfrak{m}_{n-1}/\mathfrak{m}_{n-1}(E)\). The state of the process is then incremented according to the replacement kernel: \(\mathfrak{m}_n=\mathfrak{m}_{n-1}+R_{Y_n}\). Throughout the paper the authors assume the balance condition that, for all \(x\in E\), we have \(R_x(E)=1\) almost surely. Other technical assumptions are also made to control the moments of relevant quantities and ensure the existence of an appropriate limit. Under these assumptions, convergence and fluctuation results are established, giving first- and second-order limits as \(n\to\infty\). Convergence is established almost surely and in \(L^2\). The second-order results cover cases of asymptotically Gaussian and non-Gaussian fluctuations, and particular attention is paid to establishing cases in which the limit is degenerate. Proofs use martingale methods and operator-theoretic techniques. The paper concludes with four applications: the out-degree distribution in the random recursive tree, the heat kernel on the square, a branching random walk, and reinforced processes on a countable state-space.