Determinantal point processes

Placeholder notes for a type of point process, with which I am unfamiliar, but about which I am incidentally curious.

Wikipedia says:

Let \(\Lambda\) be a locally compact Polish space and \(\mu\) be a Radon measure on \(\Lambda\). Also, consider a measurable function \(K:\Lambda^2\rightarrow \mathbb{C}\).

We say that \(X\) is a determinantal point process on \(\Lambda\) with kernel \(K\) if it is a simple point process on \(\Lambda\) with a joint intensity/Factorial_moment_densityorcorrelation function (which is the density of its factorial moment measure) given by

\[ \rho_n(x_1,\ldots,x_n) = \det[K(x_i,x_j)]_{1 \le i,j \le n} \]

for every \(n\geq 1\) and \(x_1,\dots, x_n\in \Lambda.\)

The most popular tutorial introduction to this topic seems to be (Kulesza and Taskar 2012). I found it unhelpful as it is rooted in discrete-space problems which is precisely where I do not work. For continuous state space, (Møller and Waagepetersen 2007; Møller and Waagepetersen 2017) and Terry Tao’s summary of (J. Ben Hough et al. 2006) are good.

Interesting property: The zeros random polynomials with Gaussian coefficients are apparently to be distributed as DPPs (John Ben Hough et al. 2009; Krishnapur 2006).

One idea these processes provoke is use as a source of random low-discrepancy samples for quadrature, which I have seen suggested by Richard Xu Qiao et al. (2016) and Belhadji, Bardenet, and Chainais (2019).

• Introduction to the Determinantal Point Process — Wray Buntine

See also its cousin, the permanental point process.