Title: Determinantal processes and stochastic domination
Speaker: Raghavendra Tripathi (IISc Mathematics)
Date: 30 April 2019
Time: 11:30 am
Venue: LH-1, Mathematics Department
The systematic study of determinantal processes began with the work of Macchi (1975), and since then they have appeared in different contexts like random matrix theory (eigenvalues of random matrices), combinatorics (random spanning tree, non-intersecting paths, measures on Young diagrams), and physics (fermions). A particularly interesting and well-known example of a discrete determinantal process is the Uniform spanning tree (UST) on finite graphs. We shall describe UST on complete graphs and complete bipartite graphs—in these cases it is possible to make explicit computations that yield some special cases of Aldous’ result on CRT.
The defining property of a determinantal process is that its joint intensities are given by determinants, which makes it amenable to explicit computations. One can associate a determinantal process with a finite rank projection on a Hilbert space of functions on a given set. Let $H$ and $K$ are two finite dimensional subspaces of a Hilbert space, and let $P$ and $Q$ be determinantal processes associated with projections on $H$ and $K$, respectively. Lyons showed that if $H$ is contained in $K$ then $P$ is stochastically dominated by $Q$. We will give a simpler proof of Lyons’ result which avoids the machinery of exterior algebra used in the original proof of Lyons and also provides a unified approach of proving the result in discrete as well as continuous case.
As an application of the above result, we will obtain the stochastic domination between the largest eigenvalue of Wishart matrix ensembles $W(N,N)$ and $W(N-1,N+1)$. It is well known that the largest eigenvalue of Wishart ensemble $W(M,N)$ has the same distribution as the directed last-passage time $G(M,N)$ on $Z^2$ with i.i.d. exponential weights. We, thus, obtain stochastic domination between $G(N,N)$ and $G(N-1,N+1)$ - answering a question of Riddhipratim Basu. Similar connections are also known between the largest eigenvalue of Meixner ensemble and directed last-passage time on $Z^2$ with i.i.d. geometric weights. We prove a stochastic domination result which combined with the Lyons’ result gives the stochastic domination between Meixner ensemble $M(N,N)$ and $M(N-1,N+1)$.