betaFeedback

WorksMathematics

Boundaries of groups with compact-by-amenable decomposition

Abstract

We look at the Liouville property (triviality of harmonic functions) of topological groups G relative to a compact subgroup K. We show that Gelfand pairs have this property. Furthermore, we show that this property holds if and only if G admits a product decomposition G = K S for an amenable subgroup S. We identify the Poisson and Furstenberg boundaries of these groups. This applies in particular to reductive algebraic groups over local fields.

Introduction

Let μ\mu be a probability measure on a locally compact group GG, and let K≤GK \leq G be a compact subgroup. We say μ\mu has the Liouville property relative to KK if the following condition holds.

For any bounded left-KK-invariant function ff on GG, the equality f=∫Gf(⋅h) dμ(h)f = \int_G f(\cdot h)\,d\mu(h) implies that ff is constant.

Similarly, we say that (G,K)(G,K) has the Liouville property if there is such an admissible probability measure μ\mu. We show that this property holds whenever (G,K)(G,K) is a Gelfand pair, in the following sense.

Theorem A. Let GG be a locally compact second countable group, and let K≤GK \leq G be a compact subgroup such that the associated Hecke algebra Cc(K\G/K)C_c(K\backslash G/K) is commutative. Then all admissible, bi-KK-invariant probability measures have the Liouville property relative to KK.

For K={e}K=\{e\} the above recovers the Choquet–Deny theorem [5]. Our direct motivation comes from Furstenberg’s celebrated work [7] on the boundaries of semisimple Lie groups GG, in which a result of the above type for the maximal compact subgroup K≤GK \leq G plays a crucial role in the identification of the Poisson boundary of GG through analysis of random walks on the symmetric space G/KG/K.

Indeed, the relative Liouville property allows us to get a strong structural property for locally compact groups and their Poisson boundaries closely analogous to semisimple Lie groups. More precisely, we have the following results.

Theorem B. Let GG be a locally compact second countable group, and let K≤GK \leq G be a compact subgroup.

(1) If G=KSG=KS holds for some closed amenable subgroup S≤GS \leq G, then there exists an admissible probability measure on GG that has the Liouville property relative to KK.

(2) If there exists an admissible probability measure μ\mu on GG that has the Liouville property relative to KK, then G=KSG=KS for a closed, amenable subgroup S≤GS \leq G, and the Poisson boundary of (G,μ)(G,\mu) is (G/S,ν)(G/S,\nu) for some GG-quasi-invariant probability measure ν\nu.

For K={e}K=\{e\} the above recovers the main result of [17], which states that amenability of GG is characterized by the triviality of some Poisson boundary of GG. More recently, Monod proved that Gelfand pairs (G,K)(G,K) admit an Iwasawa decomposition [13]. The combination of Theorems A and B gives a random-walk proof of his result.

Finally, we also have a description of the Furstenberg boundary for groups GG satisfying the conclusion of the above theorem, as follows.

Theorem C. Let GG be a locally compact second countable group such that some Poisson boundary of GG is isomorphic to (G/S,ν)(G/S,\nu) where SS is a cocompact, amenable subgroup and ν\nu a GG-quasi-invariant probability measure. Then G/SG/S is the Furstenberg boundary of GG.

A prominent class of locally compact groups satisfying the assumption of Theorem A comes from connected reductive algebraic groups G\mathcal{G} over non-Archimedean local fields FF, by taking the group of FF-points G=G(F)G=\mathcal{G}(F). Using Theorems B and C, we identify the Furstenberg boundary of such groups GG, see § 7.1.

As a concrete example, we can consider the determinant-one matrices over the pp-adic numbers G=SLn(Qp)G=\mathrm{SL}_n(\mathbb{Q}_p) and identify its Furstenberg boundary as G/BG/B, where BB is the Borel subgroup of upper triangular matrices. This is a direct analogue of the well-known result in the context of semisimple Lie groups established in Furstenberg’s foundational work [Fur73].

While this paper was in the final stage of preparation, Monod announced [14]. His work contains closely overlapping results particularly on the Furstenberg boundary of groups GG as above, although the proof strategy seems somewhat different from ours.

Acknowledgements. Part of this work was carried out during the first-named author’s stay at Institut Henri Poincaré for the Operator algebras: approximation, rigidity and dynamics program in the fall of 2026. We thank Sven Raum for a stimulating discussion. M.Y.’s research was funded, in part, by The Research Council of Norway [project 324944].

Declaration on AI usage. After developing the main ideas for this paper, we used a large language model for part of the brainstorming process to refine formulations.

Preliminaries

Let GG be a locally compact, second countable group. Throughout the paper we fix a left Haar measure on GG, denoted mGm_G.

Poisson boundary

Poisson boundary. Let μ\mu be a probability measure on GG. We say that:

  1. μ\mu is spread out if μ∗n≪mG\mu^{*n}\ll m_G for some n≥1n\geq1, that is, some positive convolution power of μ\mu is absolutely continuous relative to the Haar measure on GG.

  2. μ\mu is adapted if supp⁡μ\operatorname{supp}\mu generates GG as a semigroup, that is, G=⋃n≥0(supp⁡μ)nG=\bigcup_{n\geq0}(\operatorname{supp}\mu)^n.

  3. μ\mu is admissible if μ\mu is spread out and adapted.

Let XX be a standard Borel GG-space. For a probability measure η\eta on XX, our convention of the convolution is μ∗η=∫Gη(g−1⋅) dμ(g)\mu*\eta=\int_G\eta(g^{-1}\cdot)\,\mathrm{d}\mu(g), which is a new probability measure on XX. We say η\eta is μ\mu-stationary if η=μ∗η\eta=\mu*\eta. By [15 Lemma 1.1], if μ\mu is admissible, then η\eta is GG-quasi-invariant, that is η(g⋅)\eta(g\cdot) and η\eta have the same null sets on XX for all g∈Gg\in G.

Therefore given an admissible probability measure μ\mu on GG, for every standard Borel GG-space (X,η)(X,\eta) where η\eta is a μ\mu-stationary measure, we have a well-defined map between von Neumann algebras

Pη:L∞(X,η)⟶L∞(G,mG),Pη(F)(g)=∫XF(gx) dη(x).P_\eta:L^\infty(X,\eta)\longrightarrow L^\infty(G,m_G),\qquad P_\eta(F)(g)=\int_X F(gx)\,\mathrm{d}\eta(x).

Automatically, PηP_\eta is linear, unital, positive, normal, contractive and GG-equivariant. The image of PηP_\eta can be as small as C1G\mathbb{C}1_G, and by evaluating indicator functions one finds that PηP_\eta is isometric if and only if ess sup⁡g∈Gη(g−1A)=1\operatorname*{ess\,sup}_{g\in G}\eta(g^{-1}A)=1 holds for all Borel set A⊆XA\subseteq X satisfying η(A)>0\eta(A)>0.

By [8 Theorem 3.1], there exists a standard Borel GG-space (Πμ,μ^)(\Pi_\mu,\widehat{\mu}), called the Poisson boundary of (G,μ)(G,\mu), which has the property that Pμ^P_{\widehat{\mu}} is isometric and the image of Pμ^P_{\widehat{\mu}} is the set of the μ\mu-harmonic functions {f∈L∞(G,mG):f=f∗μ}\{f\in L^\infty(G,m_G): f=f*\mu\}. Here f∗μ=∫Gf(⋅h) dμ(h)f*\mu=\int_G f(\cdot h)\,d\mu(h) is an established but unfortunate notation, because f∗(μ1∗μ2)=(f∗μ2)∗μ1f*(\mu_1*\mu_2)=(f*\mu_2)*\mu_1 reverses the order of convolution, where μ1,μ2\mu_1,\mu_2 are admissible measures on GG.

There is a separable, metrizable topological GG-space that recovers Πμ\Pi_\mu by taking its Borel σ\sigma-algebra, see for example [22 Theorem 2.1.19].

Furstenberg boundary

Let XX be a compact space with a continuous GG-action. We say that:

  1. XX is minimal if it has no closed GG-invariant subspaces, except ∅\varnothing and XX.

  2. XX is strongly proximal if for every probability measure η\eta on XX there is a point mass contained in the weak∗^*-closure of the GG-orbit GηG\eta.

  3. XX is a GG-boundary if it is minimal and strongly proximal.

By [Fur73, Proposition 4.6] there exists a Furstenberg boundary or universal boundary ∂FG\partial_FG, that is an up to isomorphism unique GG-boundary with the universal property that for any GG-boundary XX there exists a GG-equivariant surjection ∂FG↠X\partial_FG\twoheadrightarrow X.

Liouville measures from Gelfand pairs

In this section we prove Theorem A. We recall that (G,K)(G,K) is a Gelfand pair, that means the Hecke algebra Cc(K\G/K)C_c(K\backslash G/K) is commutative, if and only if all bi-KK-invariant measures on GG commute. For any a∈Ga\in G the measure mKaK=mK∗δa∗mKm_{KaK}=m_K*\delta_a*m_K is an example of a bi-KK-invariant measure, where mKm_K denotes the Haar probability measure on the compact subgroup KK.

The only observation we need to make is that whenever μ\mu is a bi-KK-invariant probability measure and ff is a μ\mu-harmonic function, also f∗mKaKf*m_{KaK} are μ\mu-harmonic for all a∈Ga\in G:

(f∗mKaK)∗μ=f∗(μ∗mKaK)=f∗(mKaK∗μ)=(f∗μ)∗mKaK=f∗mKaK.(f*m_{KaK})*\mu=f*(\mu*m_{KaK})=f*(m_{KaK}*\mu)=(f*\mu)*m_{KaK}=f*m_{KaK}.

Also, for all bi-KK-invariant ν\nu on GG we may write

f∗ν=∫Gf(⋅h) dν(h)=∫G(f∗δh) dν(h)=∫G(f∗mKhK) dν(h).f*\nu=\int_G f(\cdot h)\,d\nu(h)=\int_G(f*\delta_h)\,d\nu(h)=\int_G(f*m_{KhK})\,d\nu(h).

Proof of Theorem A. Since μ\mu is spread out, there exists ρ∈L1(G,mG)\rho\in L^1(G,m_G) for some m≥1m\geq1 such that μ∗m=ρ⋅mG\mu^{*m}=\rho\cdot m_G. So any μ\mu-harmonic f∈L∞(G,mG)f\in L^\infty(G,m_G) has a continuous representative

f(g)=(f∗μ∗m)(g)=∫Gf(gh)ρ(h) dmG(h)=∫G(f⋅λgρ) dmG.f(g)=(f*\mu^{*m})(g)=\int_G f(gh)\rho(h)\,dm_G(h)=\int_G(f\cdot\lambda_g\rho)\,dm_G.

Also, we recall that μ\mu-harmonic ff is automatically right-KK-invariant because μ=δk∗μ\mu=\delta_k*\mu.

Consider

{f∣bounded, left-K-invariant, μ-harmonic}=span⁡C\{f\mid\text{bounded, left-}K\text{-invariant, }\mu\text{-harmonic}\}=\operatorname{span} C

where

C={f:G→[0,1]∣∀k∈K:f=λkf and f=f∗μ}⊆L∞(G,mG)C=\{f:G\to[0,1]\mid\forall k\in K:f=\lambda_kf\text{ and }f=f*\mu\}\subseteq L^\infty(G,m_G)

is convex and bounded. Because λk\lambda_k and f↦f∗μf\mapsto f*\mu are weak∗^*-continuous, the latter being the Banach-space adjoint of a contraction on L1(G,mG)L^1(G,m_G), we find that CC is weak∗^*-closed. By the Banach–Alaoglu theorem CC is weak∗^*-compact. So by the Krein–Milman theorem, it suffices to show the extreme points of CC are constant.

Consider an extreme point f∈Cf \in C. For ν=μ∗n\nu= \mu^{*n} and t=ν(A)∈(0,1)t = \nu(A) \in(0,1) we have

f=f∗ν=t(1ν(A)∫A(f∗mKaK) dν(a))+(1−t)(1ν(G∖A)∫G∖A(f∗mKaK) dν(a))f=f*\nu=t\left(\frac{1}{\nu(A)}\int_A(f*m_{K a K})\,d\nu(a)\right)+(1-t)\left(\frac{1}{\nu(G\setminus A)}\int_{G\setminus A}(f*m_{K a K})\,d\nu(a)\right)

Because f∗mKaK∈Cf*m_{K a K}\in C, and limits of convex combinations remain in CC, extremality implies

f=1ν(A)∫A(f∗mKaK) dν(a),ν(A)>0.f=\frac{1}{\nu(A)}\int_A(f*m_{K a K})\,d\nu(a),\qquad\nu(A)>0.

Note that for ν(A)=1\nu(A)=1 this holds trivially because f=f∗νf=f*\nu. By continuity, we may evaluate both sides at ee and use the bi-KK-invariance of ff in the form (f∗mKaK)(e)=f(a)(f*m_{K a K})(e)=f(a) to obtain that f(e)=1ν(A)∫Af(a) dν(a)f(e)=\frac{1}{\nu(A)}\int_A f(a)\,d\nu(a) for all ν(A)>0\nu(A)>0. This implies both {f>f(e)}\{f>f(e)\} and {f<f(e)}\{f<f(e)\} must be disjoint from supp⁡ν\operatorname{supp}\nu. Since μ\mu is adapted, this shows ff is constant on all of GG. □

Remark 3.1. In the case when KK is a compact, open subgroup of GG, the countable set measures mKaKm_{K a K} over double-cosets KaK∈K\G/KK a K\in K\backslash G/K forms a hypergroup under convolution. Combining results of [20] and [10], commutative hypergroups with dimension function can be shown to be weakly amenable in the sense of [11], which is witnessed by an ergodic probability measure. By an adaptation of this proof, on the group GG one can show that in fact all absolutely continuous, bi-KK-invariant probability measures on GG with full support have the Liouville property relative to KK.

As we remarked in the introduction, connected reductive algebraic groups over totally disconnected local fields satisfy the assumption for Theorem A, see § 7.1.

Liouville measure from Iwasawa decomposition

In this section, we prove the part of Theorem B that constructs an admissible probability measure with the Liouville property from a compact-by-amenable decomposition. We closely follow the proof of [KV83, Theorem 4.3]. The only additional ingredient is the following:

Lemma 4.1. Assume G=KSG=KS with KK compact and SS closed and amenable. Then for every compact subset C⊆GC\subseteq G and ε>0\varepsilon>0 there exists a probability measure α\alpha on GG such that

∥mK∗δg∗α−mK∗α∥<ε,g∈C.\lVert m_K * \delta_g * \alpha-m_K * \alpha\rVert<\varepsilon,\qquad g\in C.

Here ∥⋅∥\lVert\cdot\rVert is always the total variation norm for measures on GG.

Proof of Lemma 4.1. The set Q=S∩KCQ=S\cap KC is compact. By Reiter’s property, using the amenability of SS, there exists a probability density φ\varphi on SS such that α=φmS\alpha=\varphi m_S satisfies

∥δs∗α−α∥=∫S∣φ(s−1t)−φ(t)∣ dmS(t)<ε\lVert\delta_s *\alpha-\alpha\rVert=\int_S\lvert\varphi(s^{-1}t)-\varphi(t)\rvert\,dm_S(t)<\varepsilon

for all s∈Qs\in Q. Here mSm_S is the Haar measure on SS.

Now for all g∈Cg\in C we have a decomposition g=ksg=ks where k∈Kk\in K and s=k−1g∈S∩KC=Qs=k^{-1}g\in S\cap KC=Q, and mK∗δg=mK∗δsm_K*\delta_g=m_K*\delta_s by the KK-invariance of the Haar measure on KK. Hence

∥mK∗δg∗α−mK∗α∥=∥mK∗(δs∗α−α)∥≤∥δs∗α−α∥<ε.\lVert m_K *\delta_g *\alpha-m_K *\alpha\rVert=\lVert m_K*(\delta_s *\alpha-\alpha)\rVert\leq\lVert\delta_s *\alpha-\alpha\rVert<\varepsilon.

□

To construct a Liouville measure relative to KK we only have to construct μ\mu such that

lim⁡n→∞∥mK∗δg∗μ∗n−mK∗μ∗n∥=0,for all g∈G.(1)\lim_{n\to\infty}\lVert m_K *\delta_g *\mu^{*n}-m_K *\mu^{*n}\rVert=0,\qquad\text{for all }g\in G. \tag*{(1)}

This is because for all left-KK-invariant, μ\mu-harmonic ff we have f(g)=(mK∗δg∗μ∗n)(f)f(g)=(m_K*\delta_g*\mu^{*n})(f). So for all bounded, left-KK-invariant, μ\mu-harmonic functions ff, we get an estimate

∣f(g)−f(e)∣=∣(mK∗δg∗μ∗n−mK∗μ∗n)(f)∣≤∥f∥∞∥mK∗δg∗μ∗n−mK∗μ∗n∥⟶0\lvert f(g)-f(e)\rvert=\lvert(m_K *\delta_g *\mu^{*n}-m_K *\mu^{*n})(f)\rvert\leq\lVert f\rVert_\infty\lVert m_K *\delta_g *\mu^{*n}-m_K *\mu^{*n}\rVert\longrightarrow0

for all g∈Gg\in G, from (1).

There is a generic argument from the proof of [KV83, Theorem 4.3] which shows how to combine probability measures α1,α2,…\alpha_1,\alpha_2,\ldots obtained from Lemma 4.1 into a single probability measure

μ=p0α0+p1α1+⋯ ,\mu=p_0\alpha_0+p_1\alpha_1+\cdots,

for certain weights pk>0p_k>0 with ∑k≥0pk=1\sum_{k\geq0}p_k=1 such that (1) holds. Choosing α0\alpha_0 with strictly positive density relative to mGm_G gives μ\mu full support, and all α\alpha from Lemma 4.1 can be chosen absolutely continuous by convolving on the right with some fixed probability μ0≪mG\mu_0\ll m_G.

Proof of part (1) in Theorem B. We pick compact sets e∈C1⊆C2⊆⋯e\in C_1\subseteq C_2\subseteq\cdots with ⋃k≥1Ck=G\bigcup_{k\geq1}C_k=G. Pick pk,εk>0p_k,\varepsilon_k>0 with ∑k≥0pk=1\sum_{k\geq0}p_k=1 and εk\varepsilon_k decreasing to 00 and such that (p0+⋯+pn−1)n<εn(p_0+\cdots+p_{n-1})^n<\varepsilon_n for all n≥1n\geq1. Now, for any k≥1k\geq1, given α0,…,αk−1\alpha_0,\ldots,\alpha_{k-1} consider the finite set of probability measures

Fk={αw1∗⋯∗αwm∣0≤wj<k for all 0≤j≤m<k}.F_k=\{\alpha_{w_1}*\cdots*\alpha_{w_m}\mid0\leq w_j<k\text{ for all }0\leq j\leq m<k\}.

We may select (e.g. among the CkC_k) a large enough compact set DkD_k such that for all β∈Fk\beta\in F_k it holds β(G∖Dk)<εk\beta(G\setminus D_k)<\varepsilon_k. Now finally, we pick using Lemma 4.1 a probability measure αk\alpha_k such that

∥mK∗δh∗αk−mK∗αk∥<εk,h∈(CkDk).\lVert m_K*\delta_h*\alpha_k-m_K*\alpha_k\rVert<\varepsilon_k,\qquad h\in(C_kD_k).

Now for μ=∑k≥0pkαk\mu=\sum_{k\geq0}p_k\alpha_k write pw=pw1⋯pwnp_w=p_{w_1}\cdots p_{w_n} and αw=αw1∗⋯∗αwn\alpha_w=\alpha_{w_1}*\cdots*\alpha_{w_n} for words w∈Nnw\in\mathbb{N}^n. Then

μ∗n=∑w∈Nnpwαw=∑w∈{0,…,n−1}npwαw+∑w∈Nn∖{0,…,n−1}n⇒(∃k≥n and β∈Fn)pw(β∗αk∗γ).\mu^{*n}=\sum_{w\in\mathbb{N}^n}p_w\alpha_w =\sum_{w\in\{0,\ldots,n-1\}^n}p_w\alpha_w+ \sum_{\substack{w\in\mathbb{N}^n\setminus\{0,\ldots,n-1\}^n\\ \Rightarrow\,(\exists k\geq n\text{ and }\beta\in F_n)}}p_w(\beta*\alpha_k*\gamma).

Here the first summand contains only words where all the nn letters are less than nn. The second summand contains the rest of all words, where there is some letter kk that is at least nn and we choose the first such occurrence. So there αw=β∗αk∗γ\alpha_w=\beta*\alpha_k*\gamma where β∈Fn⊆Fk\beta\in F_n\subseteq F_k and γ\gamma is some probability measure made of the remaining factors αwj\alpha_{w_j}. For all g∈Gg\in G it holds

∥mK∗δg∗μ∗n−mK∗μ∗n∥≤∑w∈Nnpw∥mK∗δg∗αw−mK∗αw∥≤2∑w∈{0,…,n−1}npw+∑the restpw∥mK∗δg∗β∗αk−mK∗β∗αk∥∥γ∥≤2εn+sup⁡k≥n, β∈Fn∫G∥mK∗δg∗δx∗αk−mK∗δx∗αk∥ dβ(x).\begin{aligned} \lVert m_K*\delta_g*\mu^{*n}-m_K*\mu^{*n}\rVert &\leq\sum_{w\in\mathbb{N}^n}p_w\lVert m_K*\delta_g*\alpha_w-m_K*\alpha_w\rVert\\ &\leq2\sum_{w\in\{0,\ldots,n-1\}^n}p_w+\sum_{\text{the rest}}p_w\lVert m_K*\delta_g*\beta*\alpha_k-m_K*\beta*\alpha_k\rVert\lVert\gamma\rVert\\ &\leq2\varepsilon_n+\sup_{\substack{k\geq n,\ \beta\in F_n}}\int_G\lVert m_K*\delta_g*\delta_x*\alpha_k-m_K*\delta_x*\alpha_k\rVert\,d\beta(x). \end{aligned}

Here is the only place we use the choice of pkp_k, to ensure the first sum over the nnn^n words is small, where we have used the crude bound 22 for the difference of probability measures.

For the integral expression, we have

∫G∥mK∗δgx∗αk−mK∗δex∗αk∥ dβ(x)≤∫G∖Dk2 dβ+∫Dk2εk dβ≤4εk\int_G\lVert m_K*\delta_{gx}*\alpha_k-m_K*\delta_{ex}*\alpha_k\rVert\,d\beta(x) \leq\int_{G\setminus D_k}2\,d\beta+\int_{D_k}2\varepsilon_k\,d\beta\leq4\varepsilon_k

since the case x∈Dkx\in D_k allows us to replace mK∗δgx∗αkm_K*\delta_{gx}*\alpha_k by mK∗αkm_K*\alpha_k with an εk\varepsilon_k-error, while the case x∉Dkx\notin D_k has small β\beta-measure bounded by εk\varepsilon_k again. So we have shown ∥mK∗δg∗μ∗n−mK∗μ∗n∥<6εn\lVert m_K*\delta_g*\mu^{*n}-m_K*\mu^{*n}\rVert<6\varepsilon_n for every g∈Cng\in C_n. □

Poisson boundary from a Liouville measure

We are now ready to give applications to boundary theory. In this section, we prove the part of Theorem B that constructs the Iwasawa decomposition and Poisson boundary.

Proof of part (2) in Theorem B. Let P=supp⁡μ^⊆ΠμP = \operatorname{supp}\hat{\mu} \subseteq\Pi_{\mu}. It is closed, and GG-invariant because μ^\hat{\mu} was GG-quasi-invariant. Because Πμ∖P\Pi_{\mu} \setminus P is a null set, we have a GG-equivariant isomorphism of measure spaces

(Πμ,μ^)≅(P,μ^∣P).(\Pi_{\mu},\hat{\mu}) \cong(P,\hat{\mu}|_{P}).

Similarly as in [7 Lemma 5.1], for every bounded continuous ϕ∈Cb(Πμ)\phi\in C_b(\Pi_{\mu}), we consider the function

ϕ′(ξ)=∫Kϕ(kξ) dmK(k),ξ∈Πμ.\phi'(\xi) = \int_K \phi(k\xi)\,dm_K(k), \qquad\xi\in\Pi_{\mu}.

It is bounded, and KK-invariant by right invariance of mKm_K (compact groups are unimodular). By continuity of the action and dominated convergence, ϕ′\phi' is continuous. Then Pμ^(ϕ′)P_{\hat{\mu}}(\phi') is a bounded, left-KK-invariant μ\mu-harmonic function. By the Liouville property of μ\mu it is constant, say Pμ^(ϕ′)=c∈CP_{\hat{\mu}}(\phi') = c \in\mathbb{C}. Because Pμ^P_{\hat{\mu}} is linear, unital and injective, ϕ′=c\phi' = c holds μ^\hat{\mu}-a.e. So the set {ϕ′≠c}\{\phi' \ne c\} is an open null set. By definition, it must be disjoint from PP. This shows

∀ϕ∈Cb(Πμ): ϕ′ is constant on P.(2)\forall\phi\in C_b(\Pi_{\mu}):\ \phi' \text{ is constant on } P. \tag*{(2)}

Fix x∈Px \in P. By GG-invariance of PP we have Kx⊆PKx \subseteq P. Suppose there is y∈Py \in P with Ky≠KxKy \ne Kx. Then the KK-orbits KxKx and KyKy are disjoint compact sets, and we find ϕ∈Cb(Πμ)\phi\in C_b(\Pi_{\mu}) with ϕ∣Kx=0\phi|_{Kx} = 0 and ϕ∣Ky=1\phi|_{Ky} = 1. Because the topology on Πμ\Pi_{\mu} is metrizable, one does not even need Urysohn’s lemma but can construct ϕ\phi concretely. In any case ϕ′(x)=0\phi'(x) = 0 and ϕ′(y)=1\phi'(y) = 1, contradicts (2), so P=KxP = Kx. Since PP is GG-invariant, we have Gx⊆P=Kx⊆GxGx \subseteq P = Kx \subseteq Gx, so we have shown

P=Gx=Kx.P = Gx = Kx.

Define

S=Gx={g∈G∣gx=x}.S = G_x = \{g \in G \mid gx = x\}.

For all g∈Gg \in G we know that gx∈Kxgx \in Kx, so gx=kgxgx = k_gx for some kg∈Kk_g \in K, that is kg−1g∈Sk_g^{-1}g \in S, i.e. g∈KSg \in KS. That means G=KSG = KS. Importantly, the compact G/SG/S is Hausdorff because SS is closed. Now the orbit map

ψ:G/S→P,gS↦gx,\psi: G/S \to P,\qquad gS \mapsto gx,

is a well-defined GG-equivariant bijection. It is continuous by the universal property of the quotient topology, and it is actually a homeomorphism because continuous bijections from compact spaces onto Hausdorff spaces are closed. So we get a GG-equivariant isomorphism of measure spaces

(Πμ,μ^)≅(P,μ^∣P)→ψ−1(G/S,ν),ν=(ψ−1)∗(μ^∣P).(\Pi_{\mu},\hat{\mu}) \cong(P,\hat{\mu}|_{P}) \xrightarrow{\psi^{-1}} (G/S,\nu),\qquad\nu= (\psi^{-1})_*(\hat{\mu}|_{P}).

By equivariance, again ν\nu is GG-quasi-invariant.

The action G↷(Πμ,μ^)G \curvearrowright(\Pi_{\mu},\hat{\mu}) is Zimmer-amenable [21 Theorem 5.2]. The Zimmer-amenability does not feel null sets or equivariant isomorphisms of measure spaces, so G↷(G/S,ν)G \curvearrowright(G/S,\nu) is Zimmer-amenable. By [22 Proposition 4.3.2] this is equivalent to SS being amenable. □

The following is essentially the computation leading up to [7 Theorem 5.4].

Corollary 5.1. In the setting of Theorem B part (2), if in addition μ\mu is left-KK-invariant, then ν\nu is the unique KK-invariant probability measure on G/SG/S. Namely, ν=mK∗δeS\nu= m_K * \delta_{eS}.

Proof. If μ\mu is left-KK-invariant, then μ^\hat{\mu} is KK-invariant, so also ν\nu is KK-invariant. Let ν′\nu' be any KK-invariant probability measure on G/SG/S and f∈C(G/S)f \in C(G/S). By KK-invariance and Fubini

ν′(f)=∫K∫G/Sf(kz) dν′(z) dmK(k)=∫G/S∫Kf(kz) dmK(k) dν′(z).\nu'(f) = \int_K \int_{G/S} f(kz)\,d\nu'(z)\,dm_K(k) = \int_{G/S} \int_K f(kz)\,dm_K(k)\,d\nu'(z).

Every z∈G/Sz \in G/S is of the form z=k0Sz=k_0S with k0∈Kk_0 \in K, because G=KSG=KS. By right-KK-invariance of mKm_K (compact groups are unimodular) we have ∫Kf(kk0S) dmK(k)=∫Kf(kS) dmK(k)\int_K f(kk_0S)\,\mathrm{d}m_K(k)=\int_K f(kS)\,\mathrm{d}m_K(k). Hence ν′(f)=∫Kf(kS) dmK(k)\nu'(f)=\int_K f(kS)\,\mathrm{d}m_K(k).

Furstenberg boundary from homogeneous Poisson boundary

Let us next prove Theorem C, which gives an application to the Furstenberg boundary. From [6 Proposition 6.4] we know that GG has a cocompact amenable subgroup if and only if GG acts transitively on the Furstenberg boundary. By [16 Proposition 10 and the preceding discussion], in the setting of Theorem C we know already that the Furstenberg boundary of GG must be of the form G/S~G/\widetilde{S} for some maximal cocompact, relatively amenable subgroup S~\widetilde{S} that is unique up to conjugation.

In order to prove Theorem C, namely that S~\widetilde{S} can be taken to be SS, we have to show:

Proposition 6.1. Suppose some Poisson boundary of GG is isomorphic to (G/H,ν)(G/H,\nu) with some closed, cocompact subgroup H≤GH\leq G and GG-quasi-invariant ν\nu. Then G/HG/H is a GG-boundary.

Theorem C will follow from this proposition: One may use that SS is amenable, to invoke the fixed-point property and find some SS-invariant probability measure η\eta on ∂FG\partial_FG. Since G/SG/S is compact, the GG-orbit of η\eta already contains a point mass, which forces η=δx\eta=\delta_x for some x∈∂FGx\in\partial_FG. Then the map α:G/S↠∂FG\alpha:G/S\twoheadrightarrow\partial_FG, gS↦gxgS\mapsto gx is continuous and GG-equivariant. By the universal property we have ∂FG↠G/S\partial_FG\twoheadrightarrow G/S a GG-equivariant surjection in the other direction. But on every GG-boundary XX, the identity map id⁡X\operatorname{id}_X is the unique continuous, GG-equivariant self map. So α\alpha and β\beta are inverses of each other, and G/S≅∂FGG/S\cong\partial_FG.

Lemma 6.2. Let ν\nu be GG-quasi-invariant. If ν(A)=0\nu(A)=0 then ∀x∈G/H\forall x\in G/H: mG({g∣gx∈A})=0m_G(\{g\mid gx\in A\})=0.

Proof. Let X=G/HX=G/H and A⊆XA\subseteq X Borel. If ν(A)=0\nu(A)=0, then by Fubini and GG-quasi-invariance

∫XmG({g∣gx∈A}) dν(x)=∫Gν(g−1A) dmG(g)=0.\int_X m_G(\{g\mid gx\in A\})\,\mathrm{d}\nu(x)=\int_G \nu(g^{-1}A)\,\mathrm{d}m_G(g)=0.

This shows the claim for some x∈Xx\in X. By GG-transitivity and ΔG>0\Delta_G>0 we get it for all x∈Xx\in X.

In particular, Lemma 6.2 reminds that quasi-invariant ν\nu have full support: One just notes that for all nonempty, open subsets U⊆G/HU\subseteq G/H again {g∣gx∈U}⊆G\{g\mid gx\in U\}\subseteq G is nonempty, open. And nonempty, open subsets of GG have positive Haar measure, so ν(U)=0\nu(U)=0 would be a contradiction. So supp⁡ν=G/H\operatorname{supp}\nu=G/H because its complement is ⋃U open: ν(U)=0U=∅\bigcup_{U\text{ open}:\,\nu(U)=0}U=\varnothing.

Proof of Proposition 6.1. Write X=G/HX=G/H. Being GG-transitive, XX is minimal, so we only have to show strong proximality. In fact, showing some point-measure is in the GG-orbit of a given probability measure η\eta on XX will imply that all point masses are contained. So we can assume η\eta and x∈Xx\in X are given, and we have to construct a net gi∈Gg_i\in G such that giη→δxg_i\eta\to\delta_x in the weak* sense.

Having that (X,ν)(X,\nu) is a Poisson boundary, the Poisson transform PνP_\nu is isometric, so for every open neighborhood UU of xx we have ess sup⁡g∈Gν(g−1U)=1\operatorname*{ess\,sup}_{g\in G}\nu(g^{-1}U)=1. Since XX is metrizable, we can choose a sequence of neighborhoods Un⊆XU_n\subseteq X with radius 1/n1/n for some fixed metric, and elements gn∈Gg_n\in G such that ν(X∖gn−1Un)<1/2n\nu(X\setminus g_n^{-1}U_n)<1/2^n. Then by the Borel-Cantelli lemma the following is ν\nu-null:

N=lim sup⁡n→∞(X∖gn−1Un)=X∖⋃n≥1⋂k≥n(gk−1Uk).N=\limsup_{n\to\infty}(X\setminus g_n^{-1}U_n)=X\setminus\bigcup_{n\geq1}\bigcap_{k\geq n}(g_k^{-1}U_k).

So for ν\nu-a.e. y∈Xy \in X, namely for all y∈X∖Ny \in X \setminus N, we know that gny→xg_n y \to x. This already means we have gnν→δxg_n\nu\to\delta_x weak*ly. Using the quasi-invariance, this computation transfers to all measures η\eta on the homogeneous space, after possibly applying a shift:

Lemma 6.2 gives mG({h∈G∣hy∈N})=0m_G(\{h \in G \mid hy \in N\}) = 0 for all y∈Xy \in X. By Fubini

∫Gη(h−1N) dmG(h)=∫XmG({h∈G∣hy∈N}) dη(y)=0.\int_G \eta(h^{-1}N)\,dm_G(h) = \int_X m_G(\{h \in G \mid hy \in N\})\,d\eta(y) = 0.

So there exists h∈Gh \in G with (hη)(N)=0(h\eta)(N) = 0. That means gny→xg_ny \to x for (hη)(h\eta)-almost every y∈Xy \in X. This shows (gnh)η→δx(g_nh)\eta\to\delta_x in the weak* sense using dominated convergence theorem. □

In practice, one may read the combination of Theorems B and C as a way to compute Poisson and Furstenberg boundaries for compact-by-amenable groups. However, starting from a factorization G=KHG = KH with closed, amenable HH one obtains a potentially larger amenable subgroup SS with G=KSG = KS where G/SG/S identifies the boundaries. The following makes precise how much larger the subgroup SS becomes:

Corollary 6.3. Let GG be locally compact, second countable, with compact subgroup K≤GK \leq G. If G=KHG = KH with closed, amenable subgroup HH, then the Furstenberg boundary of GG can be computed:

∂FG=G/S,S={k∈K∣⟨H∪{k}⟩ is amenable}⋅H.\partial_F G = G/S,\qquad S = \{k \in K \mid\langle H \cup\{k\}\rangle\text{ is amenable}\}\cdot H.

Here ⟨⋅⟩\langle\cdot\rangle means “the smallest closed subgroup of GG containing (⋅)(\cdot)”.

Proof. We can find a cocompact amenable subgroup SS such that ∂FG≅G/S\partial_F G \cong G/S where H⊆SH \subseteq S, and such that SS is maximal in the sense that if any other amenable AA contains HH, then SS contains AA.

We have S=(S∩K)⋅HS = (S \cap K)\cdot H. The inclusion “⊇\supseteq” holds because H⊆SH \subseteq S. The inclusion “⊆\subseteq” holds because G=KHG = KH. Namely, if S∋s=khS \ni s = kh with k∈Kk \in K, h∈Hh \in H, then already k=sh−1∈Sk = sh^{-1} \in S.

We have S∩K={k∈K∣⟨H∪{k}⟩ amenable}S \cap K = \{k \in K \mid\langle H \cup\{k\}\rangle\text{ amenable}\}. The inclusion “⊆\subseteq” holds because H⊆SH \subseteq S and because closed subgroups of amenable groups are amenable. The inclusion “⊇\supseteq” holds because A=⟨H∪{k}⟩A = \langle H \cup\{k\}\rangle, k∈Kk \in K contains HH. So if AA is amenable then A⊆SA \subseteq S, so k∈Sk \in S. □

Example 6.4. One may consider G=SL2(R)G = \mathrm{SL}_2(\mathbb{R}). It contains copies of the compact circle group

K={(cos⁡(θ)−sin⁡(θ)sin⁡(θ)cos⁡(θ))∣0≤θ<2π}K = \left\{\left(\begin{array}{cc} \cos(\theta) & -\sin(\theta) \\ \sin(\theta) & \cos(\theta) \end{array}\right)\mathrel{\big|} 0 \leq\theta< 2\pi\right\}

and the upper half-plane

H={(yx01/y)∣y>0},H = \left\{\left(\begin{array}{cc} y & x \\ 0 & 1/y \end{array}\right)\mathrel{\big|} y > 0\right\},

which is a semi-direct product of abelian groups, so solvable and hence amenable. One can write down a unique factorization G=KHG = KH with K∩H={I}K \cap H = \{I\}.

Now one computes the Furstenberg boundary as ∂FG=G/S\partial_F G = G/S where S={±I}⋅HS = \{\pm I\}\cdot H: In fact, for k=±Ik = \pm I one checks that (H∪−H)(H \cup-H) is closed and still solvable, while for all other k≠±Ik \ne\pm I it is not difficult to write down a formula that expresses a generic element of GG in terms of only kk, k−1k^{-1} and elements of HH, but ⟨H,k⟩=G\langle H,k\rangle= G is famously non-amenable.

Examples

Reductive algebraic groups over local fields

7.1. Reductive algebraic groups over local fields. Let GG be a connected reductive algebraic group over an infinite non-Archimedean local field FF. We consider the locally compact group G=G(F)G = G(F). Then, by the Bruhat–Tits theory, we obtain a compact open subgroup K⊂GK \subset G as the stabilizer of an appropriately chosen point in the enlarged Bruhat–Tits building for GG [3, 19]. In the case of split semisimple groups, we can take K=G(O)K = \mathcal{G}(O) for the ring of integers O⊂FO \subset F.

By a result of Satake [18] (see also [4 Theorem 4.1]), the spherical Hecke algebra Cc(K\G/K)C_c(K \backslash G/K) is commutative.

Proposition 7.1. Let P⊂G\mathcal{P} \subset G be the minimal parabolic subgroup defined by a choice of root data for GG. Then the subgroup S=P(F)S = \mathcal{P}(F) is maximally amenable in GG.

Proof. The groups GG and SS become part of a Tits system, see for example [1 paragraph IV.14.15 and Theorem V.21.15]. The intermediate subgroups S⊂P⊂GS \subset P \subset G are classified by subsets II of the set of positive simple roots [2 Théorème IV.2.5.3]. If II is not empty, the corresponding parabolic subgroup PIP_I is indeed nonamenable, since it contains a closed subgroup locally isomorphic to SL2(F)\mathrm{SL}_2(F) in its Levi factor. □\square

From any initial cocompact amenable subgroup BB, [Fur73, Proposition 4.4] determines that all GG-boundaries are of the form G/HG/H for some intermediate amenable subgroup B⊆H⊆GB \subseteq H \subseteq G.

Corollary 7.2. In the above setting, the Furstenberg boundary of GG is G/SG/S.

Type A groups over p-adic numbers

For the type A groups G=SLn(Qp)G = \mathrm{SL}_n(\mathbb{Q}_p), n≥2n \geq2 over the p-adic numbers Qp\mathbb{Q}_p there are more elementary ways to compute the Furstenberg boundary.

It will be enough to show that G/BG/B is strongly proximal, with BB the subgroup of upper triangular matrices of GG. Because when G/BG/B is a GG-boundary, from the universal property of ∂FG\partial_F G, we obtain an equivariant surjection ∂FG↠G/B\partial_F G \twoheadrightarrow G/B. When ∂FG≅G/H\partial_F G \cong G/H for B⊆HB \subseteq H then from the equivariance we get also H⊆gBg−1H \subseteq gBg^{-1} for g∈Gg \in G with ϕ(eH)=gB\phi(eH) = gB. But the upper triangular subgroup BB does not contain any conjugate except itself, so g−1Bg⊆H⊆Bg^{-1}Bg \subseteq H \subseteq B implies g=eg = e, and thus H=BH = B, so ∂FG≅G/B\partial_F G \cong G/B.

To show the strong proximality of G↷G/BG \curvearrowright G/B, one may consider the “large” open subset

W={gB∣all top-left subminors of g∈G are invertible}⊆G/B.W = \{gB \mid\text{all top-left subminors of } g \in G \text{ are invertible}\} \subseteq G/B.

One can show that for every probability measure η\eta on G/BG/B one may apply a shift η~=kη\widetilde{\eta} = k\eta, k∈Gk \in G such that η~\widetilde{\eta} is supported on WW. Furthermore, one can show the sequence anη~a^n\widetilde{\eta} converges weak∗^* to the point mass δeB\delta_{eB}, where a∈Ga \in G can be any diagonal matrix of the form

a=diag⁡(pλ1,…,pλn),λ1<⋯<λn,λ1+⋯+λn=0.a = \operatorname{diag}(p^{\lambda_1}, \ldots, p^{\lambda_n}), \qquad\lambda_1 < \cdots< \lambda_n,\qquad\lambda_1 + \cdots+ \lambda_n = 0.

For example, when n=2n = 2 the group G=SL2(Qp)G = \mathrm{SL}_2(\mathbb{Q}_p) acts on the projective line G/B≅P1(Qp)G/B \cong\mathbb{P}^1(\mathbb{Q}_p) via the Moebius transform

(abcd)(x:y)=(ax+by:cx+dy).\begin{pmatrix} a & b \\ c & d \end{pmatrix}(x:y) = (ax + by:cx + dy).

Here W=P1(Qp)∖{(0:1)}W = \mathbb{P}^1(\mathbb{Q}_p) \setminus\{(0:1)\}, and since Qp\mathbb{Q}_p is uncountable, every probability measure η\eta must vanish on some (x:y)(x:y), so one can translate its support into WW. Then positive powers of

a=(p−100p+1)a = \begin{pmatrix} p^{-1} & 0 \\ 0 & p^{+1} \end{pmatrix}

contract WW to {(1:0)}\{(1:0)\}.

References

References

  1. [1]Borel, A. Linear algebraic groups. Second. Vol. 126. Graduate Texts in Mathematics. Springer-Verlag, New York, 1991, pp. xii+288.
  2. [2]Bourbaki, N. Éléments de mathématique. Groupes et algèbres de Lie. Chapitre 4 à 6. Reprint of the 1981 (Masson) edition. Springer, Berlin-Heidelberg, 2007.
  3. [3]Bruhat, F. and Tits, J. “Groupes réductifs sur un corps local”. In: Inst. Hautes Études Sci. Publ. Math. 41 (1972), pp. 5–251.
  4. [4]Cartier, P. “Representations of p-adic groups: a survey”. In: Automorphic forms, representations and L-functions (Proc. Sympos. Pure Math., Oregon State Univ., Corvallis, Ore., 1977), Part 1. Vol. XXXIII. Proc. Sympos. Pure Math. Amer. Math. Soc., Providence, RI, 1979, pp. 111–155.
  5. [5]Choquet, G. and Deny, J. “Sur l’équation de convolution μ = μ ∗ σ”. In: C. R. Acad. Sci. Paris 250 (1960), pp. 799–801.
  6. [6]Caprace, P.-E., Kalantar, M., and Monod, N. “A type I conjecture and boundary representations of hyperbolic groups”. In: Proc. Lond. Math. Soc. (3) 127.2 (2023), pp. 447–486.arxiv.org/abs/2110.00190
  7. [7]Furstenberg, H. “A Poisson formula for semi-simple Lie groups”. In: Ann. of Math. (2) 77 (1963), pp. 335–386.DOI
  8. [8]Furstenberg, H. “Random walks and discrete subgroups of Lie groups”. In: Advances in Probability and Related Topics, Vol. 1. Dekker, New York, 1971, pp. 1–63.
  9. [10]Hayashi, T. “Harmonic function spaces of probability measures on fusion algebras”. In: Publ. Res. Inst. Math. Sci. 36.2 (2000), pp. 231–252.DOI
  10. [11]Hiai, F. and Izumi, M. “Amenability and strong amenability for fusion algebras with applications to subfactor theory”. In: Internat. J. Math. 9.6 (1998), pp. 669–722.DOI
  11. [13]Monod, N. “Gelfand pairs admit an Iwasawa decomposition”. In: Math. Ann. 378.1-2 (2020), pp. 605–611.arxiv.org/abs/1902.09497
  12. [14]Monod, N. Invariant measures, Gelfand pairs and integral representations. preprint. 2026.arxiv.org/abs/2611.14085
  13. [15]Nevo, A. and Zimmer, R. J. “Homogenous projective factors for actions of semi-simple Lie groups”. In: Invent. Math. 138.2 (1999), pp. 229–252.DOI
  14. [16]Ozawa, N. Lecture on the Furstenberg boundary and C*-simplicity. lecture note. 2014.
  15. [17]Rosenblatt, J. “Ergodic and Mixing Random Walks on Locally Compact Groups.” In: Mathematische Annalen 257 (1981), pp. 31–42.DOI
  16. [18]Satake, I. “Theory of spherical functions on reductive algebraic groups over p-adic fields”. In: Inst. Hautes Études Sci. Publ. Math. 18 (1963), pp. 5–69.DOI
  17. [19]Tits, J. “Reductive groups over local fields”. In: Automorphic forms, representations and L-functions (Proc. Sympos. Pure Math., Oregon State Univ., Corvallis, Ore., 1977), Part 1. Vol. XXXIII. Proc. Sympos. Pure Math. Amer. Math. Soc., Providence, RI, 1979, pp. 29–69.DOI
  18. [20]Yamagami, S. “Notes on amenability of commutative fusion algebras”. In: Positivity 3.4 (1999), pp. 377–388.DOI
  19. [21]Zimmer, R. J. “Amenable ergodic group actions and an application to Poisson boundaries of random walks”. In: J. Functional Analysis 27.3 (1978), pp. 350–372.DOI
  20. [22]Zimmer, R. J. Ergodic theory and semisimple groups. Vol. 81. Monographs in Mathematics. Birkhäuser Verlag, Basel, 1984, pp. x+209.DOI

Paper details

vlnc:W_x2xf42e6
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

a large language model

View statement
Looking for related papers…
Contents