Poisson boundary

In mathematics, the Poisson boundary is a probability space associated to a random walk.

Uses your system voice; voice and rate in settings.

Snapshot

≈ 7 min read

Names in other languages

No names in other languages are held for this entry yet.

Article bodies are English and are not machine-translated. A name in another language is not a translation of the article.

In mathematics, the Poisson boundary is a probability space associated to a random walk. It is an object designed to encode the asymptotic behaviour of the random walk, i.e. how trajectories diverge when the number of steps goes to infinity. Despite being called a boundary, it is in general a purely measure-theoretical object and not a boundary in the topological sense.

However, in the case where the random walk is on a topological space, the Poisson boundary can be related to the Martin boundary, which is an analytic construction yielding a genuine topological boundary. Both boundaries are related to harmonic functions on the space via generalisations of the Poisson formula. A related construction is the Furstenberg boundary of a semisimple Lie group.

The Poisson formula states that given a positive harmonic function f on the unit disc \mathbb D = \{ z \in \mathbb C : |z| (that is, \Delta f = 0 where \Delta is the Laplace–Beltrami operator associated to the Poincaré metric on \mathbb D ) there exists a unique measure \mu on the boundary \partial \mathbb D = \{ z \in \mathbb C : |z| = 1\} such that the equality f(z) = \int_{\partial \mathbb D} K(z, \xi) \, d\mu(\xi) where K(z, \xi) = \frac {1 - |z|^2}{|\xi - z|^2} is the Poisson kernel, holds for all z \in \mathbb D . One way to interpret this is that the functions K(\cdot, \xi) for \xi \in \partial \mathbb D are up to scaling all the extreme points in the cone of nonnegative harmonic functions. This analytical interpretation of the set \partial \mathbb D leads to the more general notion of minimal Martin boundary (which in this case is the full Martin boundary).

This fact can also be interpreted in a probabilistic manner. If W_t is the Brownian motion on \mathbb D with the Poincaré Riemannian metric, then the process f(W_t) is a continuous-time martingale, and as such converges almost everywhere to a function on the Wiener space of possible (infinite) trajectories for W_t . Thus the Poisson formula identifies this measured space with the Martin boundary constructed above, and ultimately to \partial \mathbb D endowed with the class of Lebesgue measure (note that this identification can be made directly since a path in Wiener space converges almost surely to a point on \partial \mathbb D ). This interpretation of \partial \mathbb D as the space of trajectories for a Markov process is a special case of the construction of the Poisson boundary.

Finally, the constructions above can be discretised, i.e. restricted to the random walks on the orbits of a Fuchsian group acting on \mathbb D . This gives an identification of the extremal positive harmonic functions on the group, and to the space of trajectories of the random walk on the group (both with respect to a given probability measure), with the topological/measured space \mathbb D . Let G be a discrete group and \mu a probability measure on G , which will be used to define a random walk X_t on G (a discrete-time Markov process whose transition probabilities are p(x, y) = \mu(x^{-1}y) ); the measure \mu is called the step distribution for the random walk. Let m be another measure on G , which will be the initial state for the random walk.

The space G^{\mathbb N} of trajectories for X_t is endowed with a measure \mathbb P_m whose marginales are m*\mu^{*n} (where * denotes convolution of measures; this is the distribution of the random walk after n steps). There is also an equivalence relation \sim on G^{\mathbb N} , which identifies (x_t) to (y_t) if there exists n, m \in \mathbb N such that x_{t+n} = y_{t+m} for all t \ge 0 (the two trajectories have the same "tail").

The Poisson boundary of (G, \mu) is then the measured space \Gamma obtained as the quotient of (G^{\mathbb N}, \mathbb P_m) by the equivalence relation \sim . If \theta is the initial distribution of a random walk with step distribution \mu then the measure \nu_\theta on \Gamma obtained as the pushforward of \mathbb P_\theta .

It is a stationary measure for (G, \mu) , meaning that for every measurable set A in the Poisson boundary \int_G \nu(g^{-1}A) \mu(g) = \nu_\theta(A). It is possible to give an implicit definition of the Poisson boundary as the maximal G -set with a (G, \mu) -stationary measure \nu , satisfying the additional condition that (X_t)_*\nu almost surely weakly converges to a Dirac mass. Let f be a \mu -harmonic function on G , meaning that \sum_{h\in G} f(gh)\mu(h) = f(g) . Then the random variable f(X_t) is a discrete-time martingale and so it converges almost surely.

Denote by \hat f the function on \Gamma obtained by taking the limit of the values of f along a trajectory (this is defined almost everywhere on G^{\mathbb N} and shift-invariant). Let x \in G and let \nu_x be the measure obtained by the construction above with \theta = \delta_x (the Dirac mass at x ). If f is either positive or bounded then \hat f is as well and we have the Poisson formula: f(x) = \int_\Gamma \hat f(\gamma) \, d\nu_x(\gamma).

This establishes a bijection between \mu -harmonic bounded functions and essentially bounded measurable functions on \Gamma . In particular the Poisson boundary of (G, \mu) is trivial, that is reduced to a point, if and only if the only bounded \mu -harmonic functions on G are constant. \mathcal G(x, y) = \sum_{n \ge 1} p_n(x, y). If the walk is transient then this series is convergent for all x, y . Fix a point o \in G and define the Martin kernel by: \mathcal K_o(x, y) = \frac{\mathcal G(x, y)}{\mathcal G(o, y)} .

The embedding y \mapsto \mathcal K_o( \cdot, y) has a relatively compact image for the topology of pointwise convergence, and the Martin compactification is the closure of this image. A point \gamma \in \Gamma is usually represented by the notation \mathcal K(\cdot, \gamma) .

The Martin kernels are positive harmonic functions and every positive harmonic function can be expressed as an integral of functions on the boundary, that is for every positive harmonic function there is a measure \nu_{o, f} on \Gamma such that a Poisson-like formula holds: f(x) = \int \mathcal K_o(x, \gamma) \, d\nu_{o, f}(\gamma). The measures \nu_{o, f} are supported on the minimal Martin boundary, whose elements can also be characterised by being minimal.

A positive harmonic function u is said to be minimal if for any harmonic function v with 0 \le v \le u there exists c \in [0, 1] such that v = cu . There is actually a whole family of Martin compactifications.

Define the Green generating series as \mathcal G_r(x, y) = \sum_{n \ge 1} p_n(x, y)r^n. Denote by R the radius of convergence of this power series and define for 1\leq r \leq R the r -Martin kernel by \mathcal K_{o,r}(x, y) = \frac{\mathcal G_r(x, y)}{\mathcal G_r(o, y)} . The closure of the embedding y \mapsto \mathcal K_{o,r}( \cdot, y) is called the r -Martin compactification.

Text derived from https://en.wikipedia.org/w/index.php?oldid=1358296403 under CC BY-SA 4.0; modifications have been made.

Sources cited in this revision

No external citation was extracted for this revision. That is a statement about the source, not a claim that none exists.