Permutation invariance in last-passage percolation and the distribution of the Busemann process

TL;DR

This paper provides an explicit, exact description of the joint distribution of Busemann functions in exponential last-passage percolation, revealing permutation invariance and enabling efficient sampling.

Contribution

It introduces a new explicit joint distribution formula for Busemann functions in all directions and edges, extending previous results and employing novel invariance and coupling techniques.

Findings

Derived a joint distribution for Busemann functions in all directions and edges.

Established permutation invariance of inhomogeneous last-passage times.

Provided a method for exact sampling from the joint distribution.

Abstract

In i.i.d. exponential last-passage percolation, we describe the joint distribution of Busemann functions, over all edges and over all directions, in terms of a joint last-passage problem in a finite inhomogeneous environment. More specifically, the Busemann increments within a $k\times\ell$ grid, and associated to $d$ different directions, are equal in distribution to a particular collection of last-passage increments inside a $(k+d-1)\times(\ell+d-1)$ grid. The joint Busemann distribution was previously described along a horizontal line by Fan and the fourth author, using certain queuing maps. By contrast, our new description explicitly gives the joint distribution for any collection of edges (not just along a horizontal line) using only finitely many random variables. Our result thus provides an exact and accessible way to sample from the joint distribution. In the proof, we rely on one-directional marginal distributions of the inhomogeneous Busemann functions recently studied by Janjigian and the second and fourth authors. Another innovation of our argument is a joint invariance of inhomogeneous last-passage times under permutation of the inhomogeneity parameters. This result does not appear to be contained among the last-passage invariances recently described by Dauvergne. Our proof of invariance is also rather different, using the Burke property instead of the RSK correspondence, and leading to an explicit coupling of the weights before and after the permutation of the parameters.