#paperOfTheDay : "Sampling Pfaffian point processes and the symplectic Arnoldi method" from 2026.
A Pfaffian is an algebraic object similar to a determinant: It eats a matrix and produces a number. The name is from Johann Friedrich Pfaff, who was one of the first to use it, and who was also the doctoral advisor of Karl Friedrich Gauss (also known as the prince of
#mathematics ). A point process is a random process that consists of a discrete sequence of numbers. The concrete numbers are random, but they follow certain distributions. A classical point process is the Poisson process, which is characterized by the numbers being independent from one another. For example, at any time, a certain number of people write posts on Mastodon independently. These posts do not appear e.g. one per second, but sometimes many at once, then none for a few seconds. The total number of posts that exist is then a Poisson process.
For any point process, one can ask how the individual numbers are correlated, this gives the n-point correlation functions, or moments. A Poisson process with rate r has the special property that the n-point function is simply r^n. A much more general class of processes are Pfaffian point processes, where the n-point function is given by a Pfaffian of some explicitly known matrix (which in general depends on variables). In one paper of the day a few weeks ago, we met the distribution of zeros of random power series, this is a Pfaffian point process.
The present paper develops an
#algorithm to generate random samples of a Pfaffian process, i.e. how to produce an instance of the process where the numbers have the required n-point functions of a given correlation matrix.
https://arxiv.org/abs/2605.01202