Trimming a Tree and the Two-Sided Skorohod Reflection

Abstract: The $h$-trimming of a tree is a natural regularization procedure which consists in pruning the small branches of a tree: given $h\geq0$, it is obtained by only keeping the vertices having at least one leaf above them at a distance greater or equal to $h$. The $h$-cut of a function $f$ is the function of minimal total variation uniformly approximating the increments of $f$ with accuracy $h$, and can be explicitly constructed via the two-sided Skorohod reflection of $f$ on the interval $[0,h]$. In this work, we show that the contour path of the $h$-trimming of a rooted real tree is given by the $h$-cut of its original contour path. We provide two applications of this result. First, we recover a famous result of Neveu and Pitman, which states that the $h$-trimming of a tree coded by a Brownian excursion is distributed as a standard binary tree. In addition, we provide the joint distribution of this Brownian tree and its trimmed version in terms of the local time of the two-sided reflection of its contour path. As a second application, we relate the maximum of a sticky Brownian motion to the local time of its driving process.