Poisson Boundary in Random Walks

• Poisson Boundary is a measure-theoretic object that encapsulates the asymptotic behavior of random walks on countable groups through bounded μ-harmonic functions.
• It plays a crucial role in understanding when all bounded harmonic functions are constant (the Liouville property) versus when nontrivial asymptotic behavior emerges.
• Diverse methodologies—including entropy criteria, strip methods, and operator-algebraic techniques—provide geometric, algebraic, and noncommutative models of the Poisson boundary.

The Poisson boundary of a random walk $(G,\mu)$ on a countable group is a measure-theoretic object that captures the asymptotic behavior of typical trajectories. Analytically, it is the space of bounded $\mu$-harmonic functions, that is, bounded functions $f:G\to\mathbb{R}$ satisfying

$f(g)=\sum_{h\in G} f(gh)\mu(h).$

The boundary is trivial precisely when all bounded $\mu$-harmonic functions are constant, a condition usually called the Liouville property (Erschler et al., 2023). Across current research, Poisson boundaries appear in random walks on groups, geometric group theory, planar graphs, Lorentzian diffusions, operator algebras, and tensor categories; at the same time, recent work shows that they are fundamentally measure-dependent and need not admit any nontrivial universal topological realization (Chawla et al., 16 Jun 2025).

1. Analytic formulation and boundary representations

In the group-theoretic setting, the Poisson boundary is the receptacle for bounded harmonic functions. In concrete identifications, a bounded harmonic function is recovered from a boundary function by an integral formula of Poisson type. For random walks on $\mathrm{Out}(F_N)$, once the boundary is identified with the space $\mathcal{FI}$ of simplices of free and arational trees equipped with the hitting measure $\nu$, every bounded $\mu$-harmonic function factors through the boundary as

$$f(g)=\int_{\mathcal{FI}} \widehat f(x)\,d(g_*\nu)(x),$$

which expresses the general principle that the boundary encodes exactly the asymptotic data needed to reconstruct bounded harmonic observables (Horbez, 2014).

The same theme persists for locally compact groups. If $\mu$0 is a countable dense subgroup of a locally compact second countable group $\mu$1, and $\mu$2 is supported on $\mu$3, one can consider both $\mu$4-harmonic functions on $\mu$5 and $\mu$6-harmonic functions $\mu$7 satisfying

$\mu$8

for almost every $\mu$9. In that setting, the $f:G\to\mathbb{R}$0-Poisson boundary can be constructed from the $f:G\to\mathbb{R}$1-boundary as the quotient $f:G\to\mathbb{R}$2, where $f:G\to\mathbb{R}$3 is the $f:G\to\mathbb{R}$4-Poisson boundary and $f:G\to\mathbb{R}$5 acts diagonally by $f:G\to\mathbb{R}$6 (Brofferio, 2015).

These formulations already show two persistent features. First, the Poisson boundary is intrinsically measurable rather than purely topological. Second, boundary identifications are normally stated relative to a specific measure $f:G\to\mathbb{R}$7, often under entropy or moment hypotheses.

2. Recognition criteria and proof methods

A large part of modern Poisson-boundary theory consists of criteria for proving either maximality of a candidate boundary or triviality of the boundary. One recurring method is the entropy criterion of Kaimanovich–Vershik: for the upper-triangular and Baumslag-group results discussed below, nontriviality is deduced from linear growth of asymptotic entropy, and triviality corresponds to vanishing asymptotic entropy (Erschler et al., 2023). In the theory of group extensions, this was refined by new $f:G\to\mathbb{R}$8-restriction entropy estimates and a comparison criterion that transfers nontriviality between extensions with the same quotient projection (Erschler et al., 2022).

A second major method is the strip criterion. For $f:G\to\mathbb{R}$9, it is used with strips built from axes in outer space associated to pairs of arational trees; subexponential growth of these strips yields maximality of the boundary $f(g)=\sum_{h\in G} f(gh)\mu(h).$0 (Horbez, 2014). For groups acting on CAT(0) cube complexes, intervals between boundary ultrafilters produce strips $f(g)=\sum_{h\in G} f(gh)\mu(h).$1, and polynomial growth of these strips is the key input in proving that the compact space $f(g)=\sum_{h\in G} f(gh)\mu(h).$2 realizes the Poisson boundary (Nevo et al., 2011).

Planar-graph work introduced a different recognition mechanism, formulated in terms of sharp harmonic functions. A candidate boundary is a realization of the Poisson boundary if and only if it is faithful to all sharp harmonic functions; this criterion underlies the identification of square-tiling boundaries with Poisson boundaries for large classes of planar graphs (Georgakopoulos, 2013).

In Lorentzian diffusion theory, the dominant techniques are coupling and dévissage. For relativistic diffusions in Robertson–Walker and model space-times, the analysis passes through subdiffusions with trivial boundary, then uses equivariance and almost sure convergence of the remaining coordinates to show that the full invariant sigma-field is generated by the limiting geometric data (Angst, 2014, Angst et al., 2018).

In noncommutative settings, the corresponding proof technology is operator-algebraic. For unital completely positive maps on von Neumann algebras, dilation theory yields the relevant boundary algebra and a canonical product on peripheral eigenvectors (Bhat et al., 2022).

3. Geometric realizations

Many of the most explicit Poisson-boundary identifications are geometric. For $f(g)=\sum_{h\in G} f(gh)\mu(h).$3, if $f(g)=\sum_{h\in G} f(gh)\mu(h).$4 has finite first logarithmic moment and finite entropy and its support generates a nonelementary subgroup, then almost every sample path converges in Culler–Vogtmann outer space to a simplex of a free, arational tree, and the space $f(g)=\sum_{h\in G} f(gh)\mu(h).$5 with its hitting measure is the Poisson boundary. Via the Bestvina–Reynolds and Hamenstädt description of $f(g)=\sum_{h\in G} f(gh)\mu(h).$6, the Gromov boundary of the free factor complex with the same hitting measure is also a Poisson boundary model (Horbez, 2014).

For groups acting cocompactly on finite-dimensional, locally finite CAT(0) cube complexes, a natural compact metric space $f(g)=\sum_{h\in G} f(gh)\mu(h).$7 is constructed as the closure of the nonterminating ultrafilters in the Roller compactification. The action on $f(g)=\sum_{h\in G} f(gh)\mu(h).$8 is minimal and strongly proximal, $f(g)=\sum_{h\in G} f(gh)\mu(h).$9 admits a unique stationary measure for any generating probability measure, and when the measure has finite logarithmic moment, $\mu$0 is a compact metric model of the Poisson boundary (Nevo et al., 2011).

Planar graphs admit especially concrete boundary models. For bounded-degree plane triangulations, the Poisson boundary coincides with the boundary of the square tiling and with the boundary of the circle packing; the same work proves that the square-tiling boundary also coincides with the Martin boundary (Hutchcroft et al., 2015). More generally, for any planar, uniquely absorbing graph with bounded degrees, the boundary of the square tiling of a cylinder realizes the Poisson boundary (Georgakopoulos, 2013). These results extend to graphs rough-isometric to bounded-degree planar graphs, which suggests that the boundary behavior is stable under coarse planar geometry when the hypotheses of those theorems apply (Hutchcroft et al., 2015).

Lorentzian examples exhibit a close but not always exact relation between stochastic and geometric compactifications. In a spatially flat, fast expanding Robertson–Walker space-time, the relativistic diffusion has $\mu$1 almost surely, and the Poisson boundary is generated by $\mu$2, hence identified with the future causal boundary (Angst, 2014). In the broader class of model manifolds of constant scalar curvature and Robertson–Walker space-times, the Poisson boundary may coincide with the conformal or causal boundary, but in Anti-de Sitter space and in several infinite-horizon Robertson–Walker cases it is strictly richer, involving pointed light circles, asymptotic directions, or intercept parameters (Angst et al., 2018).

An additional geometric construction appears for locally compact closures of discrete groups. If $\mu$3 is viewed as a dense subgroup of $\mu$4, then under moment assumptions and when the action on $\mu$5 is not contracting, the Poisson boundary of the ambient group is the $\mu$6-solenoid

$\mu$7

obtained from the discrete boundary by the quotient construction above (Brofferio, 2015).

4. Algebraic classifications for solvable, linear, and extension-type groups

Recent work has substantially sharpened the picture for amenable and solvable groups. A central example is the torsion-free Baumslag group in dimension $\mu$8: every finite entropy irreducible random walk on $\mu$9 has nontrivial Poisson boundary, in contrast with torsion cases where simple random walks behave as in corresponding lamplighter groups (Erschler et al., 2023). This result is obtained by realizing the group as a linear group and embedding it into a class of upper-triangular groups $\mathrm{Out}(F_N)$0 defined from Laurent polynomials. For $\mathrm{Out}(F_N)$1, nontriviality holds for all finite entropy irreducible random walks whenever the polynomial $\mathrm{Out}(F_N)$2 has the spaced polynomial property; the polynomial $\mathrm{Out}(F_N)$3 satisfies this property, and $\mathrm{Out}(F_N)$4 is isomorphic to $\mathrm{Out}(F_N)$5 (Erschler et al., 2023).

For finitely generated linear groups over fields of characteristic $\mathrm{Out}(F_N)$6, the situation is much more rigid. An amenable linear group has nontrivial Poisson boundary for a simple random walk if and only if one of its basic blocks contains $\mathrm{Out}(F_N)$7 as a subgroup, and in this class the Stability Problem has a positive answer: nontriviality does not depend on the choice of nondegenerate finite-entropy random walk (Erschler et al., 2022). In characteristic $\mathrm{Out}(F_N)$8, the same paper proves a sufficient condition for triviality that is independent of the walk and conjectures that this condition is also necessary (Erschler et al., 2022).

Wreath products provide another class with explicit descriptions. For $\mathrm{Out}(F_N)$9, if $\mathcal{FI}$0 has finite entropy and lamp configurations stabilize almost surely, then the Poisson boundary is $\mathcal{FI}$1, where $\mathcal{FI}$2 is the Poisson boundary of the projected walk on $\mathcal{FI}$3. If the projection to $\mathcal{FI}$4 is Liouville, the Poisson boundary reduces to the space of limit lamp configurations endowed with the hitting measure. In particular, for $\mathcal{FI}$5, $\mathcal{FI}$6, and finite first moment measures, this resolves the question of Kaimanovich and of Lyons–Peres by showing that the boundary is exactly the limit lamp configuration (Frisch et al., 2023).

A related but distinct phenomenon occurs in lampshuffler groups $\mathcal{FI}$7. If $\mathcal{FI}$8 has finite first moment and the projected walk on $\mathcal{FI}$9 is transient, then the permutation coordinate stabilizes pointwise almost surely. For $\nu$0, this limit permutation completely describes the Poisson boundary (Silva, 2023).

An explicit recent computation appears for a Schreier-chain random walk associated with Thompson’s group $\nu$1. Projecting the simple symmetric random walk on $\nu$2 to the dyadic orbit point $\nu$3 yields a walk whose full Poisson boundary is the skeleton end boundary of the dyadic Schreier graph. Via Kaimanovich’s coding, this boundary is identified with the odd $\nu$4-adic integers $\nu$5, and the hitting measure is a biased Bernoulli product measure with

$\nu$6

The measure is singular with respect to Haar measure, has full topological support, and is exact-dimensional (Mönch, 22 Jun 2026).

5. Noncommutative, operator-algebraic, and categorical extensions

The phrase Poisson boundary also has established noncommutative meanings. For a normal unital completely positive map $\nu$7 on a von Neumann algebra, the classical noncommutative Poisson boundary is the fixed-point space $\nu$8 equipped with the Choi–Effros product. The peripheral Poisson boundary enlarges this by taking the norm-closed span of all peripheral eigenspaces

$\nu$9

and endowing it with a $\mu$0-algebra structure obtained through dilation theory. For eigenvectors $\mu$1 and $\mu$2, the product is

$\mu$3

This peripheral boundary is invariant under powers, $\mu$4, and $\mu$5 acts on it by automorphisms (Bhat et al., 2022).

For II$\mu$6 factors, Poisson boundaries can be defined relative to hyperstates extending the trace. The resulting boundary $\mu$7 is a von Neumann algebra containing the factor $\mu$8, and it is always injective. This framework supports operator-algebraic analogues of Furstenberg theory, including double ergodicity and entropy. Within this setting, all finite factors satisfy the MV-property, and property $\mu$9 factors exhibit an entropy gap (Das et al., 2020).

On full Fock space, the noncommutative Poisson boundary associated with

$$f(g)=\int_{\mathcal{FI}} \widehat f(x)\,d(g_*\nu)(x),$$0

is the von Neumann algebra generated by the right creation operators,

$$f(g)=\int_{\mathcal{FI}} \widehat f(x)\,d(g_*\nu)(x),$$1

and it is an injective factor of type III for any choice of $$f(g)=\int_{\mathcal{FI}} \widehat f(x)\,d(g_*\nu)(x),$$2. In finite dimension, the factor is completely classified by Connes’ $$f(g)=\int_{\mathcal{FI}} \widehat f(x)\,d(g_*\nu)(x),$$3-invariant: it is of type III$$f(g)=\int_{\mathcal{FI}} \widehat f(x)\,d(g_*\nu)(x),$$4 or type III$$f(g)=\int_{\mathcal{FI}} \widehat f(x)\,d(g_*\nu)(x),$$5, with $$f(g)=\int_{\mathcal{FI}} \widehat f(x)\,d(g_*\nu)(x),$$6 belonging to a restricted class of algebraic numbers (Bhat et al., 2021).

There is also a categorical version. Given a rigid $$f(g)=\int_{\mathcal{FI}} \widehat f(x)\,d(g_*\nu)(x),$$7-tensor category $$f(g)=\int_{\mathcal{FI}} \widehat f(x)\,d(g_*\nu)(x),$$8 with simple unit and a probability measure $$f(g)=\int_{\mathcal{FI}} \widehat f(x)\,d(g_*\nu)(x),$$9 on $\mu$00, one defines a Poisson boundary category $\mu$01 whose morphisms are bounded $\mu$02-harmonic natural transformations, together with a unitary tensor functor $\mu$03. When $\mu$04 has simple unit, $\mu$05 is universal among unitary tensor functors realizing the amenable dimension function on $\mu$06, thereby linking Poisson-boundary theory with amenability of tensor categories, quantum groups, and subfactors (Neshveyev et al., 2014).

6. Measure dependence, stability, and non-realizability

A longstanding theme in the subject is the extent to which the Poisson boundary depends on the chosen random walk. For finitely generated linear groups in characteristic $\mu$07, the recent classification by basic blocks gives a positive answer to the Stability Problem: within that class, nontriviality of the Poisson boundary is a property of the group rather than of the specific simple random walk, provided one restricts to nondegenerate finite-entropy measures (Erschler et al., 2022).

Outside such stable settings, however, the dependence on $\mu$08 can be drastic. A major 2025 result proves that for any countable group $\mu$09 with probability measure $\mu$10, there exists a randomized stopping time $\mu$11 such that $\mu$12 has a strictly larger space of bounded harmonic functions than $\mu$13, unless this space is trivial for all measures on $\mu$14. In particular, there exists an irreducible probability measure on the free group $\mu$15 whose Poisson boundary is strictly larger than the geometric boundary equipped with the hitting measure. As a corollary, there is never a nontrivial universal topological realization of the Poisson boundary for any countable group (Chawla et al., 16 Jun 2025).

This directly rules out a common expectation suggested by many positive identification theorems: that a single canonical geometric boundary should uniformly realize all random-walk boundaries on a fixed group. The available evidence now points in two directions at once. On the one hand, under suitable entropy, moment, or stabilization assumptions, the Poisson boundary often admits precise geometric or algebraic models. On the other hand, recent non-realizability results show that these models are inherently measure-specific, and that the full theory cannot be reduced to a universal topological boundary (Chawla et al., 16 Jun 2025).

The modern literature therefore presents the Poisson boundary as a unifying but highly flexible notion. In some contexts it coincides with a causal boundary, a square-tiling circle, a space of arational trees, a lamp-configuration space, a $\mu$16-adic end space, a von Neumann algebra, or a tensor category. In others, it strictly exceeds the most natural geometric candidate. This suggests that the Poisson boundary should be viewed less as a single geometric compactification than as the maximal asymptotic sigma-field appropriate to a specified stochastic dynamics.