Ito excursion theory
/ Ito: EE-toh /
Between consecutive visits to zero, Brownian motion makes an excursion: it leaves 0, wanders (above or below), and returns to 0. Ito excursion theory is the magnificent decomposition of the whole path into its excursions, organised not by ordinary clock time but by LOCAL TIME at zero, with the family of excursions forming a Poisson point process governed by a single infinite measure — the Ito excursion measure.
The construction: let L_t = L_t^0 be local time at 0 and let tau_l = inf{ t : L_t > l } be its right-continuous inverse (a stable-1/2 subordinator). As local time l runs, the gaps of the closed zero set are the excursion intervals, and to each one attaches an excursion e_l — the path on that interval, shifted to start at 0 and recentred in time. Ito's theorem says the point process { (l, e_l) } of (local-time, excursion) pairs is a Poisson point process on (0, infinity) x (excursion space) with intensity dl x n(de), where n is a sigma-finite measure on excursion paths called the Ito excursion measure. The measure n is INFINITE — n(excursions of length > epsilon) is finite for each epsilon but n(all excursions) = infinity, because infinitely many tiny excursions occur in any local-time interval. Under n the excursion duration zeta has the stable-1/2-type tail n(zeta > t) = 1 / sqrt(2 pi t) (a one-sided density n(zeta in dt) proportional to t^(-3/2)), reflecting first-passage scaling; conditioned on its length, an excursion is a Brownian bridge-like object (a Bessel(3) bridge from 0 to 0).
Why it matters: excursion theory is the cleanest route to many classical Brownian facts — the arcsine laws (sum of signed excursion lengths), the distribution of the maximum within an excursion, the Ray-Knight theorems (local time as a function of level, assembled from excursions), and the construction of skew Brownian motion (flip each excursion's sign independently with probability p). It generalises to any nice Markov process via Ito-Maisonneuve theory and is the probabilistic backbone of the Williams path decompositions. The honest subtlety: excursions are indexed by local time, NOT real time — the map from local time to real time is the stable subordinator tau_l, which is pure-jump, and the infinitude of n is not a defect but the precise statement that the zero set is a measure-zero fractal accumulated from infinitely many vanishing excursions.
Color each Brownian excursion away from 0 independently: keep it positive with probability p, flip it negative with probability 1 - p. Because excursions form a Poisson process under the Ito measure, the result is again a strong Markov process — skew Brownian motion with parameter p.
Excursions away from 0 form a Poisson point process in local time, governed by the infinite Ito excursion measure.
Excursions are indexed by local time, not real time, and the Ito measure n is infinite (sigma-finite): infinitely many short excursions occur in every local-time interval — that is what builds the measure-zero zero set.