by Jim Pitman
Introduction
When I first met Marc Yor around 1977, the theory of excursions of
Brownian motion was at an early stage of development, with few applications.
The theory had been initiated by Lévy
[◊],
and developed further by Itô-McKean
[◊],
Section 2.9.
David Williams
[◊]
made the key observation that a Brownian excursion of length
In the early years of our collaboration, I provided expertise in Markovian excursion theory, while Marc was the master of martingale calculus. There was some friendly competition across the Atlantic, whether the laws of Brownian functionals were best computed by stochastic calculus or excursion theory. But it was not long before Marc was the master and I was the apprentice in both frameworks.
The perspectives of excursion theory and of martingale calculus are very different. In excursion theory, the Brownian path is viewed as a concatenation of a countable collection of random path fragments, each with its own internal dynamics, with timing of the concatenation by a local time clock. The fragments are dotted around some space of excursion paths, in which they can readily be subjected to various operations such as time-reversal, time-changes or other transformations. By comparison, to apply martingale calculus, you start with a given filtration, work with processes adapted to that filtration, enlarging the filtration when necessary to accommodate random times that are not stopping times. It is not so easy in this framework to cut up pieces of path and reassemble them, and have any idea what process has been created. Still, as Marc showed over and over again, each of these frameworks has much to offer the other. The key to many of Marc’s deepest contributions to the theory of Brownian motion and related processes was his mastery of both martingale calculus and excursion theory.
This article offers some higlights of Marc Yor’s work in the theory of excursions of Brownian motion and related processes, especially Bessel processes. See the articles of Bertoin [◊] and Le Gall [◊] for brief introductions to the context of this work. I follow quite closely the presentations of Pitman–Yor [◊] and [◊]. For much more extensive accounts of this work, see these articles and the monographs Revuz–Yor [◊], Yor [◊], Yen–Yor [◊], and Mallein–Yor [◊].
The agreement formula
In his fundamental paper
[◊],
Itô showed how to construct
a Poisson point process of excursions of a strong Markov process
at
Each point of the excursion process is a path in a suitable space of
possible excursions of starting at time called
the lifetime of the excursion.
The intensity measure of the Poisson process of excursions is a
Description I: Conditioning on the lifetime:
First pick a lifetimeaccording to the -finite density on ; then givenrun a, bridge from 0 to 0 over time .
Description II: Conditioning on the maximum:
First pick a maximum valueaccording to the -finite density on ; then givenjoin back to back two independent, processes, each started at 0 and run till it first hits .
Pitman–Yor
[◊]
generalized the equivalence of these two descriptions of Itô’s law of positive Brownian excursions as follows. The equivalence
involving
Description I: Conditioning on the lifetime:
First pick a lifetimeaccording to the -finite density on ; then given , run a bridge from 0 to 0 over time .
Description II: Conditioning on the maximum:
First pick a maximum valueaccording to the -finite density on ; then givenjoin back to back two independent, processes, each started at 0 and run till it first hits .
The measures
For all
For
In Pitman–Yor
[◊]
we established Theorem 1 for all
Note especially the most important instance of the agreement formula of and the domain of
Brownian excursions and the Riemann Zeta function
Riemann
[◊]
showed that his zeta function, initially defined by the series
Arcsine laws and interval partitions
For
Why there should be an identity in law between
To provide an adequate explanation of this identity in terms of
Brownian excursions, Pitman–Yor
[◊]
considered the sequence
While for the range of a stable subordinator, the law of relative ranked lengths
The two-parameter Poisson–Dirichlet distribution
This unexpected appearance of size-biased sampling in the analysis of excursion lengths of Brownian motion, and more generally for any Markov process whose zero set is the
range of a stable subordinator, led to a sequence of papers which explored variations and extensions of this idea.
The size factor proportional to
Last exit times
Hand in hand with excursion theory, a recurring theme of Marc’s work was the study of last exit times, especially for path fragments of Brownian motion and
one-dimensional diffusions. His early work on last exits relied on his balayage formula for semi-martingales, which is closely related to the
Azéma–Yor
[◊]
solution of Skorokhod’s problem. This problem is to find, for a suitable probability distribution
In Pitman–Yor
[◊],
Theorem 6.1,
we used martingale calculus to find a formula for the probability density of the last exit time of an upwardly transient diffusion process,
and indicated a number of applications to Bessel processes and properties of Bessel functions. In principle, the idea is very simple. For a discrete time transient Markov chain, with transition probabilities
En guise de conclusion
From “The Last Time” by Mick Jagger and Keith Richards (1965), a refrain we enjoyed for many years, until finally the last time came:
Well this could be the last time
This could be the last time
Maybe the last time
I don’t know. Oh no. Oh no