Let G be a semi-group acting continuously on a compact metric space X. The action G↷X is called Glasner if for every infinite set A⊆X and every ε>0, there exists g∈G such that gA is ε-dense in X. Let K(X) denote the space of nonempty compact subsets of X. A G-invariant Borel probability measure on K(X) is called an invariant random compact (IRC), and the action is IC-rigid if δX is the only IRC supported on infinite compact subsets of X. In this paper we give a pair-correlation criterion implying both IC-rigidity and the Glasner property. As an application, we prove both these properties for finitely generated Ad-Zariski-dense semi-groups acting on compact homogeneous spaces G/Λ, where G is connected and noncompact with simple Lie algebra and Λ<G is a cocompact lattice; in particular, this applies to finitely generated Zariski-dense semi-groups of SLd(R) and PSLd(R). We also derive a Fourier-analytic criterion for semi-group actions on Td and use it to prove IC-rigidity for actions arising from suitable matrix-valued polynomials.
Introduction
Let G be a semi-group acting continuously on a compact space X. The action G↷X is called Glasner if for every infinite set Y⊂X and every ε>0 there exists g∈G such that gY is ε-dense in X. The terminology originates from a theorem of Glasner [7], which states that the multiplicative action of N on T=R/Z has this property. This result has since been extended in several directions. Berend and Peres [3] proved that for every infinite A⊆T, the dilates nA converge to T along a density-one set of integers. Dong [5, 6] established analogous results for higher-dimensional linear actions and actions on compact homogeneous spaces, and more recently Bulinski and Fish [1] obtained a quantitative version of Glasner results for irreducible linear actions with Zariski-connected Zariski closure.
A closely related notion of IC-rigidity was introduced by Kra and Schmieding in [12]. Both properties concern the behavior of infinite subsets under the action: the Glasner property requires that every infinite set has arbitrarily dense images, while IC-rigidity requires that the only invariant measure supported on infinite compact subsets is δX. We now recall the necessary background and notation to formulate the notion of IC-rigidity.
Let K(X) denote the space of nonempty compact subsets of X with Hausdorff metric. We assume that the action is surjective and finite-to-one. The action of G on X induces an action of G on K(X). In [12] Kra and Schmieding introduced the notion of invariant random compact. A Borel probability measure μ on K(X) which is invariant under the action of G on K(X) is called an invariant random compact or an IRC (see [12 Definition 3.1]). An IRC μ for which μ-almost every compact subset of X is infinite is called nonfinitary.
Assuming X is infinite, every infinite invariant compact set Y⊂X yields a nonfinitary IRC δY, in particular there always exists a nonfinitary IRC, δX. Kra and Schmieding call an action of G on X to be IC-rigid if the only nonfinitary IRC is δX (see [12 Definition 3.7]). In this paper we provide a pair-correlation condition for the action G↷X that implies IC-rigidity and the Glasner property. Let (μN)N≥1 be a sequence of probability measures on G. For f∈C(X;C) define
Key words and phrases. IC-rigidity, invariant random compact, Glasner property, homogeneous actions, Zariski-dense semi-groups. pair-correlation kernel as follows.
KN,f(x,y):=∫Gf(gx)f(gy)dμN(g),x,y∈X.(1)
Our main result is the following.
Theorem 1.1. Let m be a fully supported Borel probability measure on X. Assume that, for every f∈C(X;C) satisfying ∫Xfdm=0, every η>0, and every x∈X, the set
{y∈X∖{x}:N→∞limsup∣KN,f(x,y)∣>η}
is finite. Then:
(a) The action G↷X has the Glasner property: for every infinite set A⊆X and every ε>0, there exists g∈G such that gA is ε-dense in X.
(b) The action G↷X is IC-rigid. Equivalently, δX is its only nonfinitary invariant random compact.
In the context of the Glasner property for actions on compact homogeneous spaces, the best known result is due to Dong [6]. Dong established the Glasner property for actions of epimorphic subgroups, in particular minimal parabolic subgroups, on compact homogeneous spaces. As an application of Theorem 1.1, we prove both the Glasner property and IC-rigidity for finitely generated Ad-Zariski-dense semi-groups acting on compact homogeneous spaces.
Theorem 1.2. Let G be a connected noncompact Lie group with simple Lie algebra, and let Λ<G be a cocompact lattice. Let μ be a finitely supported probability measure on G whose support generates an Ad-Zariski-dense semi-group Γ⊆G. Then the action Γ↷G/Λ has the Glasner property and is IC-rigid.
As a direct corollary of Theorem 1.2, we show that finitely generated Zariski-dense semi-groups in SLd(R) or PSLd(R) act on their compact homogeneous spaces with the Glasner property and IC-rigidity.
Corollary 1.3. Let d≥2, let
G=SLd(R)orG=PSLd(R),
and let Λ<G be a cocompact lattice. Let μ be a finitely supported probability measure whose support generates a Zariski-dense semi-group Γ<G. Then the action Γ↷G/Λ has the Glasner property and is IC-rigid.
In a recent preprint [11], Kra and Schmieding proved IC-rigidity for the action of the multiplicative semi-group {nr:n∈N}, for r≥1, on the torus T. They also proved IC-rigidity for the action on Td of a subgroup Γ<SLd(Z) whose Zariski closure is Zariski-connected and whose representation on Rd is irreducible (see [11 Theorem 1.2, Theorem 1.4]). Building upon Theorem 1.1, we provide a Fourier-analytic criterion (see Theorem 1.4 below) for semi-group actions on the d-dimensional torus, d∈N, which implies both IC-rigidity and the Glasner property. Our Fourier-analytic criterion also recovers the IC-rigidity results of Kra and Schmieding [11 Theorem 1.2, Theorem 1.4], through an independently developed but closely related approach.
Recall that by a semi-group action on a torus we mean the action of a semi-group G on Td=Rd/Zd by continuous group endomorphism. As before let (μN)N∈N be a sequence of probability measures on G. The averaged orbit probability measure of z∈Td (corresponding to μN), is given by
σN,z:=∫GδgzdμN(g).
For t∈R, we set e(t):=e2πit. For m∈Zd, the m-th Fourier coefficient of the probability measure σN,z is given by
We show that Theorem 1.1, together with a density argument, gives the following.
Theorem 1.4. Assume that there exists a sequence of probability measures (μN)N≥1 on G such that, for every m∈Zd∖{0} and every δ>0, the set
{z∈Td:N→∞limsup∫Ge(⟨m,gz⟩)dμN(g)≥δ}(2)
is finite. Then:
(a) The action G↷Td has the Glasner property: for every infinite set A⊆Td and every ε>0, there exists g∈G such that gA is ε-dense in Td.
(b) The action G↷Td is IC-rigid. Equivalently, δTd is its only nonfinitary invariant random compact.
In [2] Bulinski and Fish proved Glasner property for actions arising from matrix-valued polynomials. They showed that for A(x)∈Md(Z[x]) the condition
v⋅(A(x)−A(0))w=0
for every v,w∈Zd∖{0} is equivalent to the set {A(n):n∈Z} having the uniform Glasner property on Td. As an application of Theorem 1.4 (b), we prove IC-rigidity for the multiplicative semi-group generated by such a matrix-valued polynomial. In private communication, Bryna Kra and Scott Schmieding informed us that this result can also be obtained using their methods in [11].
Theorem 1.5. Let A(x)∈Md(Z[x]) satisfy
v⋅(A(x)−A(0))w=0
for every v,w∈Zd∖{0}, and suppose that detA(x)≡0. Let GA be the multiplicative semi-group generated by
{A(n):n∈Z,detA(n)=0}.
The action GA↷Td by endomorphisms, given by
g⋅z=gz(modZd),
is IC-rigid. Equivalently, δTd is the only nonfinitary invariant random compact for this action.
Open questions
We conclude with a few open questions. For relatively prime integers p,q≥2, let
Sp,q:={paqb:a,b≥0}
be the multiplicative semi-group generated by p and q, acting on T by dilations. It is natural to ask whether this action has the Glasner property. A closely related question concerning IC-rigidity was posed by Kra and Schmieding in [12]. In a recent preprint [Sch26], Schmieding proved that if μ is an IRC for this action and μ-almost every compact set Y has positive Hausdorff dimension, then μ-almost every Y is equal to T. In other words, μ=δT.
Question 1.6. Let p,q≥2 be relatively prime. Does the action Sp,q↷T have the Glasner property? Is this action IC-rigid?
Another question posed by Kra and Schmieding in [12], asks whether there exists an IC-rigid action of an abelian group, and in particular an IC-rigid Z-action. Recently, Kanigowski, Kasprazak and Lorenzo-Laguano [KKLL26] answered this question positively by proving that mixing rank-one Z-actions are IC-rigid. The IC-rigid Z-actions constructed there are realized as subshifts, and hence the underlying spaces are totally disconnected. On the other hand, Dong [5 Theorem 3.1] proved that no Z-action on a locally connected compact metric space has the Glasner property. These results motivate the following questions.
Question 1.7. (a) Does there exist, for some d≥2, a Zd-action on an infinite connected compact metric space with the Glasner property?
(b) Does there exist, for some d≥1, an IC-rigid Zd-action on an infinite connected compact metric space?
Structure of the paper
In § 2, we provide the basic background and definitions. In Section 3, using a fundamental result concerning the density of measures, we prove the pair-correlation criterion, Theorem 1.1. In § 4, we recall an equidistribution result for homogeneous dynamics due to Benoist–Quint [4] (see Theorem 4.1). Using this equidistribution result together with Theorem 1.1, we prove Theorem 1.2. In the same section, as a corollary we establish the Glasner property and IC-rigidity for Zariski-dense semi-groups of SLd(R) and PSLd(R) acting on compact homogeneous spaces, as stated in Theorem 1.3. In § 5, using a density argument together with Theorem 1.1, we prove the Fourier-analytic criterion for torus actions, Theorem 1.4. Finally in the same section, we use Theorem 1.4 to prove Theorem 1.5.
Acknowledgement. The authors would like to thank Bryna Kra and Scott Schmieding for communicating their results and for their encouragement. Both authors were supported by the Australian Research Council through grant DP240100472.
AI statement. AI was used in the paper to obtain the relevant version of the equidistribution theorem for homogeneous spaces due to Benoist–Quint [4] (see Theorem 4.1) in a form suitable for our purposes. Apart from this, AI was used to audit the paper, identify relevant references, and correct editorial typos.
Background and notation
Let (X,d) be a compact metric space and G be a countable semi-group or group. By (X,G) we denote a topological dynamical system where G acts on X continuously. In the case where G is a semi-group, we assume that the action is surjective and finite-to-one.
Definition 2.1. A continuous action of a semi-group G on a compact space X is called Glasner if for every infinite set Y⊂X and ε>0 there exists g∈G such that gY is ε-dense in X.
Following the work of Glasner [7], this property has been extensively studied in many different contexts. We refer our interested readers to [3, 10, 5, 6].
As mentioned before, we denote by K(X) the space of nonempty compact subsets of X which is endowed with Hausdorff metric. For K,L∈K(X) set
dH(K,L)=max{x∈Ksupd(x,L),y∈Lsupd(y,K)},(3)
where d(x,L):=infy∈Ld(x,y). The action of G on X induces an action of G on K(X) given by g⋅K=g(K). Using the notation of Kra and Schmieding for k∈N, we denote by K≤k(X) the collection of all subsets of X of size at most k and set
Kfin(X)=k=1⋃∞K≤k(X),Kinf(X)=K(X)∖Kfin(X).
Since K≤k(X) is a closed set for each k∈N, it follows that Kfin(X) is a Fσ set, hence it is Borel. This implies that Kinf(X) is also Borel. Observe that Kfin(X) is G-invariant and the finite-to-one property of the action implies that Kinf(X) is G-invariant as well. Now we formally define:
Definition 2.2. (see [12 Definition 3.1]) An invariant random compact (IRC) for a topological dynamical system (X,G), is a Borel probability measure μ on K(X) which is invariant under the induced action of G on K(X).
Observe that an invariant random compact μ can be decomposed into its finitary and nonfinitary parts i.e. μ=μF+μI, where the measures μF,μI on K(X) are defined as,
μF(A)=μ(A∩Kfin(X))andμI(A)=μ(A∩Kinf(X))
for a Borel set A⊆K(X).
Definition 2.3. (see [12 Definition 3.7]) The action of G on X is infinite compact rigid (or IC-rigid) if the only nonfinitary IRC is δX. In other words for every IRC μ, the decomposition of μ into its finitary and nonfinitary parts has the form
μ=μF+c⋅δX
for some c≥0.
Proof of thm:main
In this section we prove Theorem 1.1. To do so we first need an elementary result showing that small averages of finitely many mean-zero continuous functions force dense support.
Proposition 3.1. Let m be a fully supported Borel probability measure on X. For every ε>0, there exist q∈N, zero-mean functions f1,…,fq∈C(X;C), and constants c1,…,cq>0 such that the following holds. If ν is a Borel probability measure on X satisfying
∫Xfjdν<cjfor every 1≤j≤q,
then suppν is ε-dense in X.
Proof. Choose y1,…,yq∈X such that the open balls
Uj:=B(yj,ε/2),j=1,…,q,
cover X. For every j∈{1,…,q}, choose a nonzero non-negative bump function ϕj supported in Uj. Since m is fully supported,
aj:=∫Xϕjdm>0.
For every j∈{1,…,q}, set fj:=ϕj−aj and cj:=2aj. Using the fact that m is a probability measure, for every j∈{1,…,q}, we get ∫Xfjdm=0. Let ν be a Borel probability measure on X such that
∫Xfjdν<cjfor every 1≤j≤q.
Then
∫Xϕjdν=aj+∫Xfjdν>2aj>0.
Since ϕj vanishes outside Uj and ∫Xϕjdν>0, it follows that suppν∩Uj=∅ for every j∈{1,…,q}. Since Uj are balls of radius ε/2 that cover X, it follows that suppν is ε-dense in X. □
Proof of Theorem 1.1. We first establish a pair-correlation estimate that will be used in proving both part (a) and (b) of the statement. Fix ε>0, and let q∈N, zero-mean functions f1,…,fq∈C(X;C), and constants c1,…,cq>0 be as given by Theorem 3.1. Let A⊆X be infinite and let η>0. Set C:=∑j=1q∥fj∥∞2 and δ:=2qη. Choose M∈N such that M>η2C. For 1≤j≤q and x∈X, let
Rj,δ(x):={y∈X∖{x}:N→∞limsup∣KN,fj(x,y)∣>δ}.
By the assumption of Theorem 1.1, Rj,δ(x) is finite for every j and every x∈X. Note that, for any x=y∈X, we have KN,fj(y,x)=KN,fj(x,y) and thus ∣KN,fj(y,x)∣=∣KN,fj(x,y)∣. Thus the expression
N→∞limsup∣KN,fj(x,y)∣>δ
is symmetric in x and y. Since A is infinite and each set Rj,δ(x) is finite, we can choose distinct points x1,…,xM∈A, inductively so that, for every distinct i,ℓ∈{1,…,M} and every 1≤j≤q,
N→∞limsup∣KN,fj(xi,xℓ)∣≤δ.
We define
τM:=M1i=1∑Mδxi.
Note that τM is the probability measure on X obtained by averaging the Dirac measures at x1,…,xM. For g∈G, the pushforward of τM is given by
g∗τM=M1i=1∑Mδgxi.
For 1≤j≤q, set
Fj(g):=∫Xfjd(g∗τM)=M1i=1∑Mfj(gxi).
Expanding the square and integrating over G with respect to μN gives
Thus, for every infinite A⊆X, every finite collection of zero-mean functions f1,…,fq, and every η>0, there is a finite set of distinct points x1,…,xM∈A for which (4) holds.
We now show that the action G↷X has the Glasner property. Fix an infinite set A⊆X and ε>0. Let q∈N, zero-mean functions f1,…,fq∈C(X;C), and constants c1,…,cq>0 be as given by Theorem 3.1, and set
c:=1≤j≤qmincj>0.
For η=c2/2, choose x1,…,xM∈A such that, for the corresponding τM and Fj, (4) holds. Thus, for a sufficiently large N,
∫Gj=1∑q∣Fj(g)∣2dμN(g)<c2.
Since μN is a probability measure, there exists g∈G such that ∑j=1q∣Fj(g)∣2<c2 otherwise the integral would be at least c2. Hence
∫Xfjd(g∗τM)=∣Fj(g)∣<c≤cjfor every 1≤j≤q.
Thus Theorem 3.1 implies that
supp(g∗τM)={gx1,…,gxM}⊆gA
is ε-dense in X. Hence gA is also ε-dense in $X. This proves the Glasner property.
We now show that the action G↷X is IC-rigid. Let ν be a nonfinitary invariant random compact. For ε>0, set
Kε:={K∈K(X):dH(K,X)<ε}.
Note that the map K↦dH(K,X) is continuous hence Kε is an open subset of K(X). We show that for every infinite K∈K(X),
N→∞limμN{g∈G:gK∈Kε}=1.(5)
Fix such a K and let q∈N, zero-mean functions f1,…,fq∈C(X;C), and constants c1,…,cq>0 be as given by Theorem 3.1. We set c:=min1≤j≤qcj>0. For any η>0, choose x1,…,xM∈K such that, for the corresponding τM and Fj, (4) holds. Thus,
N→∞limsup∫Gj=1∑q∣Fj(g)∣2dμN(g)<η.(6)
As we noted earlier, if ∑j=1q∣Fj(g)∣2<c2, then
∫Xfjd(g∗τM)=∣Fj(g)∣<c≤cjfor every 1≤j≤q.
Hence, by Theorem 3.1, supp(g∗τM)={gx1,…,gxM}⊆gK is ε-dense in X. In other words, gK∈Kε. Thus,
{g∈G:gK∈/Kε}⊆{g∈G:j=1∑q∣Fj(g)∣2≥c2}.
Taking measure μN of both sides and using Markov’s inequality we conclude
μN{g∈G:gK∈/Kε}≤c21∫Gj=1∑q∣Fj(g)∣2dμN(g).
Taking limsupN→∞ of both sides and using (6), we get
Since μN is a probability measure, 0≤HN,ε(K)≤1 for K∈K(X). Also, using the fact that ν is nonfinitary, (5) implies that HN,ε(K)→1 for ν-almost every K∈K(X). Hence, by the dominated convergence theorem we get,
N→∞lim∫K(X)HN,ε(K)dν(K)=1.(7)
Using Fubini’s theorem and the G-invariance of ν, we get
Letting N→∞ on both sides and using (7), we get ν(Kε)=1, for every ε>0. In particular,
ν(n=1⋂∞K1/n)=1.
But
n=1⋂∞K1/n={K∈K(X):dH(K,X)=0}={X}.
Hence ν({X})=1, thus ν=δX. Thus the action G↷X is IC-rigid.
Zariski-dense actions on compact homogeneous spaces
In this section, we prove Theorem 1.2. Let X=G/Λ and let m denote the G-invariant probability measure on X. For a semi-group Γ⊆G, we denote by
ΓΔ:={(γ,γ):γ∈Γ}⊆G×G,
the diagonal semi-group associated with Γ. To prove Theorem 1.2, we will apply an equidistribution result of Benoist–Quint [4] to the action ΓΔ↷X×X where Γ⊆G is as in the statement and ΓΔ denotes the closure of ΓΔ in G×G. Using the equidistribution result, we will show that the action Γ↷X satisfies the hypothesis of Theorem 1.1, hence it is Glasner and IC-rigid.
Equidistribution result of Benoist–Quint
We now recall an equidistribution result due to Benoist–Quint that will be used in the proof of Theorem 1.2. Let H be a real Lie group, let L<H be a cocompact lattice, and let ϑ be a compactly supported Borel probability measure on H. Let
Σ:=⟨suppϑ⟩+⊆H
be the closed sub-semi-group of H generated by suppϑ. Thus Σ is a compactly generated closed sub-semi-group of H. Assume that the Zariski closure of Ad(Σ) is semisimple with no compact factor. For k≥1, let ϑ∗k denote the k-fold convolution of ϑ, that is, the distribution of a product hk⋯h1 of k independent elements of H chosen according to ϑ. For completeness we set ϑ∗0:=δe, where e is the identity element of H. For z∈H/L and k≥0, define
ϑ∗k∗δz:=∫Hδhzdϑ∗k(h).
Thus ϑ∗k∗δz is a probability measure on H/L. It is the k-step random-walk distribution starting from z. Recall that a closed subset Y of a homogeneous space H/L is called a finite-volume homogeneous subspace if its stabilizer HY acts transitively on Y and there exists an HY-invariant Borel probability measure νY on Y. We call νY the homogeneous probability measure on Y. We use the following consequence of results of Benoist–Quint [4 Theorems 1.1, 1.2 and 1.7], stated in the form adapted to the probability measure ϑ above.
Theorem 4.1.[4] Let H, L, ϑ and Σ be as above and for z∈H/L, set Yz:=Σz. Then Yz is a Σ-invariant, Σ-ergodic finite-volume homogeneous subspace. Let νz denote the homogeneous probability measure on Yz. Then
N1k=0∑N−1ϑ∗k∗δz⟶νz
in the weak-* topology. If, in addition, the centralizer of Σ in H is discrete, then for every sequence (zn)n∈N in H/L, the corresponding homogeneous subspaces Yzn:=Σzn have a subsequence (Yznm)m∈N converging to a Σ-invariant, Σ-ergodic finite-volume homogeneous subspace Y. Moreover, Yznm⊆Y for all sufficiently large m∈N.
Proof of Theorem 1.2. Let μ be a finitely supported probability measure on G whose support generates an Ad-Zariski-dense semi-group Γ⊆G. For k≥1, let μ∗k denote the k-fold convolution of μ, that is, the distribution of a product gk⋯g1 of k independent elements chosen according to μ. For N≥1, define
μN:=N1k=1∑Nμ∗k.
Thus μN is a probability measure on Γ. We will show that the sequence (μN)N≥1 satisfies the hypothesis of Theorem 1.1. We set
H:=G×G,L:=Λ×Λ,
and identify H/L=X×X=(G×G)/(Λ×Λ). Since Λ is cocompact in G, the lattice L is cocompact in H. Let Δ be the diagonal map given by
Δ:G⟶G×G,Δ(g):=(g,g).
Let ϑ:=Δ∗μ be the pushforward of the measure μ under the diagonal map. Since μ is finitely supported probability measure on G, ϑ is a compactly supported probability measure on H. Moreover, its support generates the diagonal semi-group ΓΔ. Hence, using the notation of Theorem 4.1, we have Σ=ΓΔ. Observe that, for k∈N, ϑ∗k=Δ∗(μ∗k). In other words ϑ∗k is the pushforward of μ∗k.
Let g:=Lie(G), which is simple by assumption. Hence it follows that Lie(H)=g⊕g and for (γ,γ)∈H, AdH(γ,γ)=(Ad(γ),Ad(γ)). Since Σ=ΓΔ, the Zariski closure of AdH(Σ) is the diagonal copy of the Zariski closure of Ad(Γ). Since Γ is Ad-Zariski dense in G, this is the diagonal copy of the Zariski closure of Ad(G). Since g is simple and G is noncompact, the Zariski closure of AdH(Σ) is semisimple with no compact factor. Hence the Zariski-closure hypothesis of Theorem 4.1 is satisfied.
We claim that the centralizer of Σ in H is discrete. Let (U,V)∈Lie(ZH(Σ)), since (U,V) belongs to the Lie algebra of the centralizer it is fixed under the adjoint action of every element of Σ, that is, AdH(σ)(U,V)=(U,V) for every σ∈Σ. In particular, since ΓΔ⊆Σ, for every γ∈Γ we have
AdH(γ,γ)(U,V)=(Ad(γ)U,Ad(γ)V)=(U,V).
Hence Ad(γ)U=U and Ad(γ)V=V for every γ∈Γ. By the Ad-Zariski density of Γ, the same holds for every g∈G. Now, for Z∈g, taking g=exp(tZ) gives
Ad(exp(tZ))U=U,Ad(exp(tZ))V=V.
for every t. Thus Ad(exp(tZ))U is constant in t. Differentiating at t=0, and using the fact that
dtdt=0Ad(exp(tZ))U=[Z,U]
we get [Z,U]=0 for every Z∈g. Similarly we get [Z,V]=0 for every Z∈g. In other words U and V belong to the center of g. Since the center of the simple Lie algebra g is trivial, U=V=0. Therefore Lie(ZH(Σ))=0. Since ZH(Σ) is a closed subgroup of the Lie group H, this implies that ZH(Σ) is discrete. Hence the hypothesis of Theorem 4.1 holds.
Now fix f∈C(X;C) satisfying ∫Xfdm=0, and for u,v∈X, set Φf(u,v):=f(u)f(v) for convenience. For x,y∈X, set
Yx,y:=Σ(x,y)
and let νx,y denote its homogeneous probability measure. By Theorem 4.1,
Here we used the fact that averages from k=1 to N have the same limit as (8), since the two averages differ only in the k=0 and k=N terms, each divided by N. Fix x∈X and η>0. We claim that
Ef,η(x):={y∈X∖{x}:N→∞limsup∣KN,f(x,y)∣>η}
is finite. To see this assume by contradiction that Ef,η(x) is infinite. For n∈N, choose distinct yn∈Ef,η(x) and set
Yn:=Yx,yn=Σ(x,yn),νn:=νx,yn.
Here for n∈N, νn denotes the homogeneous probability measure on Yn. Then by definition of Ef,η(x) and (9) we have
∫YnΦfdνn>ηfor every n.
By Theorem 4.1, after passing to a subsequence, there exists a Σ-invariant, Σ-ergodic finite-volume homogeneous subspace Y⊆X×X such that Yn⊆Y for all sufficiently large n, and νn⟶νY in the weak-* topology. This implies
∫YΦfdνY≥η.(10)
Since (x,yn)∈Yn⊆Y for all sufficiently large n, the fiber F:=Y∩({x}×X) is infinite. Let HY be the stabilizer of Y in H. Since Y is a finite-volume homogeneous subspace, HY acts transitively on Y and preserves the homogeneous probability measure νY. We set
Gx:={g∈G:gx=x}.
Since X=G/Λ, the group Gx is conjugate to Λ, and hence is discrete. The group
K:=HY∩(Gx×G)
acts transitively on F. Assume that K were discrete. Since H is a second-countable Lie group, K would be countable. As K acts transitively on F, the set F would be countable and homogeneous under the action of K. Moreover, F=Y∩({x}×X) is compact, hence F would be a countable compact homogeneous space. Recall that, by the Baire category theorem, every countable compact homogeneous space is finite. Hence we reach a contradiction. Thus K is not discrete. Since K is a closed subgroup of the Lie group H, it is itself a Lie group, hence Lie(K)=0. Since Gx is discrete, Lie(Gx)=0, and K⊆HY∩(Gx×G), we get
0=Lie(K)⊆hY∩(0⊕g),
where hY:=Lie(HY). In other words, there exists Z∈g∖{0}, such that (0,Z)∈hY∩(0⊕g). Thus the set
i:={Z∈g:(0,Z)∈hY}
is nonzero. Since Y is a finite-volume homogeneous subspace and is Σ-invariant, Σ-ergodic, by definition Σ is a sub-semi-group of the stabilizer HY and acts ergodically on (Y,νY). In particular, Σ⊆HY. Hence it follows that, (γ,γ)∈HY for every γ∈Γ. Thus the adjoint action of (γ,γ) preserves the Lie algebra hY. In other words, if Z∈i, then (0,Z)∈hY and
AdH(γ,γ)(0,Z)=(0,Ad(γ)Z)∈hY.
Hence Ad(γ)Z∈i for every γ∈Γ, and therefore i is invariant under Ad(Γ). By the Ad-Zariski density of Γ, i is invariant under Ad(G). Differentiating the adjoint action gives
[g,i]⊆i.
Thus i is an ideal of g. Since g is simple and i=0, we get i=g. Hence by definition of i, we have
Since the homogeneous measure νY is invariant under the action of HY, it follows that νY is invariant under the action of {e}×G. We define a finite complex measure ρ on X as follows. Let
∫Xψ(v)dρ(v):=∫X×Xf(u)ψ(v)dνY(u,v),
for every ψ∈C(X). This defines ρ by the Riesz representation theorem. The invariance of νY under {e}×G implies that ρ is G-invariant. Since every finite G-invariant complex measure on G/Λ is a scalar multiple of m, it follows that ρ=cm for some c∈C. Taking ψ=f, we obtain
This contradicts (10), hence Ef,η(x) is finite. Thus the sequence (μN)N≥1 satisfies the hypothesis of Theorem 1.1. Hence Theorem 1.1 implies that Γ↷X has the Glasner property and is IC-rigid. □
Proof of Theorem 1.3. For G=SLd(R) or G=PSLd(R), the group G is a connected noncompact Lie group and
Lie(G)=sld(R),
which is simple for d≥2. Since Γ is Zariski dense in G, and the adjoint representation is an algebraic homomorphism it follows that Ad(Γ) is Zariski dense in Ad(G). Thus Γ is Ad-Zariski dense in G. Therefore all the hypotheses of Theorem 1.2 are satisfied. Hence the action Γ↷G/Λ has the Glasner property and is IC-rigid. □
Glasner Property and IC-rigidity of Torus actions
In this section we prove Theorem 1.4 which is a Fourier-analytic condition that implies IC-rigidity and the Glasner property for semi-group actions on d-dimensional torus. Using this criterion we prove Theorem 1.5, i.e. the matrix-valued polynomial actions on the d-dimensional torus are IC-rigid. The Glasner property for these actions was previously established by Bulinski and Fish [2], who in fact proved the stronger uniform Glasner property.
To do so we need an elementary result showing that small Fourier coefficients for a measure on the torus force dense support (see Theorem 5.1). Note that this result can be derived from Theorem 3.1 but we provide a simple standalone proof here. Recall that for a Borel probability measure μ on Td, and m∈Zd, we denote the m-th Fourier coefficient of μ by
μ(m):=∫Tde(⟨m,x⟩)dμ(x).
Proposition 5.1. For every ε>0, there exist a constant cε>0 and a finite nonempty set Λε⊆Zd∖{0} such that the following holds. If μ is a Borel probability measure on Td satisfying
∣μ(m)∣<cεfor every m∈Λε,
then suppμ is ε-dense in Td.
Proof. Let λ denote normalized Haar measure on Td. Choose y1,…,ys∈Td such that the collection of open balls {Uj:=B(yj,ε/2):1≤j≤s}, cover Td. For each 1≤j≤s, choose a non-negative continuous function fj such that suppfj⊆Uj and ∫Tdfjdλ=1. Such a function can be obtained by taking a non-negative bump function supported in a smaller ball around yj and dividing by its integral. Using the fact that trigonometric polynomials are uniformly dense in C(Td), for every 1≤j≤s, we can choose a trigonometric polynomial Pj such that
∥Pj−fj∥∞<81.
Note that there exists a common finite set Λε⊆Zd∖{0} such that for every 1≤j≤s we can write
Pj(x)=aj,0+m∈Λε∑aj,me(⟨m,x⟩)
for x∈Td. We denote by
C:=1≤j≤smaxm∈Λε∑∣aj,m∣,andcε:=4(1+C)1.
Since λ is a Haar measure, for m∈Zd∖{0}, ∫Tde(⟨m,x⟩)dλ(x)=0. Thus for every 1≤j≤s,
∫TdPjdμ−∫TdPjdλ=m∈Λε∑aj,mμ(m).
Using the fact that ∣μ(m)∣<cε it follows that for every 1≤j≤s,
Thus ∫Tdfjdμ>1/2 for every j∈{1,…,s}. Since fj vanishes outside Uj and ∫Tdfjdμ>0, it follows that suppμ∩Uj=∅ for every j∈{1,…,s}. Since Uj are balls of radius ε/2 that cover Td, it follows that suppμ is ε-dense in Td.
Proof of Theorem 1.4. Let λ denote normalized Haar measure on Td. Fix ε>0 and let cε>0 and Λε⊆Zd∖{0} be as in Theorem 5.1. We set q:=∣Λε∣ and enumerate Λε={m1,…,mq}. For every 1≤j≤q we define fj(x):=e(⟨mj,x⟩). Since mj=0, it follows that ∫Tdfjdλ=0, for every 1≤j≤q. In other words f1,…,fq∈C(Td;C), is a collection of zero-mean functions. Note that for every Borel probability measure ν on Td,
∫Tdfjdν=∫Tde(⟨mj,x⟩)dν(x)=ν(mj).
Hence, by Theorem 5.1, if
∫Tdfjdν<cεfor every 1≤j≤q,
then suppν is ε-dense in Td. Thus the functions f1,…,fq, with the common constant cε, have precisely the ε-density property used in the proof of Theorem 1.1. Recall that in the proof of Theorem 1.1, for each ε>0, we used Theorem 3.1 to obtain a finite collection of zero-mean functions and positive constants which detect ε-density. Here Theorem 5.1 provides such a collection, namely f1,…,fq, with common constant cε. The remainder of the proof of Theorem 1.1 uses the pair-correlation condition only for this finite collection. We now verify the pair-correlation condition for this finite collection of functions using the hypothesis of Theorem 1.4. For 1≤j≤q and x,y∈Td we have
Thus Rj,η(x) is finite for every 1≤j≤q. Hence the proof of Theorem 1.1, applied with the finite collection of functions above, gives that the action G↷Td has the Glasner property and is IC-rigid. □
Matrix-valued polynomial actions are IC-rigid
Proof of Theorem 1.5. Let A(x)∈Md(Z[x]) be as in the statement of the theorem. We consider the action of the semi-group
GA={A(n1)A(n2)⋯A(nℓ):ℓ≥1,n1,…,nℓ∈Z,detA(ni)=0 for all 1≤i≤ℓ}.
on Td by endomorphisms. Since detA(x) is a nonzero polynomial, there are only finitely many integers n for which detA(n)=0. Thus, for all sufficiently large N, let
IN:={n∈Z:∣n∣≤N,detA(n)=0}
and define a probability measure on GA by
μN:=∣IN∣1n∈IN∑δA(n).
For the finitely many remaining N, define μN arbitrarily. Since only finitely many integers are excluded, it follows that 2N+1∣IN∣→1. For z∈Td, the average orbit probability measure σN,z (corresponding to μN) is given by
σN,z:=∫GAδgzdμN(g)=∣IN∣1n∈IN∑δA(n)z.
For m∈Zd, the m-th Fourier coefficient of σN,z is given by
To prove that the action is IC-rigid, by Theorem 1.4 (b), it is enough to show that Em,δ is finite.
We first consider a non-torsion point z∈Td. For m∈Zd let mt denote the corresponding row vector. We write
mt(A(x)−A(0))=(P1(x),…,Pd(x)),
where P1,…,Pd∈Z[x]. The polynomials P1,…,Pd are linearly independent over Z. Indeed, otherwise there would exist w∈Zd∖{0} such that
m⋅(A(x)−A(0))w=0,
contrary to the hypothesis of the theorem. Since z=(z1,…,zd) is non-torsion, at least one zi is irrational. We claim that the polynomial ∑i=1dziPi(x) has an irrational coefficient. To see this note that, since P1,…,Pd are linearly independent over Z they are also linearly independent over Q. Let D be the maximum of the degrees of the entries of A(x). Since Pi(0)=0 and degPi≤D for every i, we may write Pi(x)=∑j=1Dbjixj, where bji∈Z.
Let B∈MD×d(Z) be the matrix whose i-th column is the coefficient vector of Pi. Since P1,…,Pd are linearly independent over Q, the matrix B has full column rank over Q. Assume by contradiction that polynomial ∑i=1dziPi(x) has only rational coefficients then Bz∈QD. Since B has full column rank over Q, this would imply that z∈Qd, contradicting the assumption that z is non-torsion. Hence ∑i=1dziPi(x) has an irrational coefficient. Since Pi(0)=0 for every i, this coefficient is necessarily non-constant. Note that
and the first term is constant in x. Hence ⟨m,A(x)z⟩ has at least one irrational non-constant coefficient. By Weyl’s equidistribution theorem, we get
N→∞lim2N+11∣n∣≤N∑e(⟨m,A(n)z⟩)=0.
Removing the finitely many n for which detA(n)=0 does not affect the limit. Therefore
N→∞lim∣IN∣1n∈IN∑e(⟨m,A(n)z⟩)=0.
Thus every point of Em,δ is a torsion point. For a torsion point z∈Td, let
z=qa+Zd,a=(a1,…,ad)∈Zd,gcd(a1,…,ad,q)=1
be its reduced form. Since A(x) has integer polynomial entries and m,a∈Zd, the quantity m⋅A(x)a is a polynomial in Z[x] of degree at most D. Write
m⋅A(x)a=b0+b1x+⋯+bDxD,
where b0,…,bD∈Z. Using the fact that m⋅A(0)a=b0, we get
(mtA(x)−mtA(0))a=b1x+⋯+bDxD.(12)
On the other hand, since mt(A(x)−A(0))=(P1(x),…,Pd(x)) the left-hand side in (12) is equal to ∑i=1daiPi(x). By the definition of B, the coefficient vector of this polynomial is Ba. Therefore
Ba=(b1,…,bD)t.
Since B has full column rank over Q, there exists a matrix L∈Md×D(Q) such that LB=Id. Clearing denominators, there exist C∈Md×D(Z) and an integer Qm≥1, depending only on A(x) and m, such that CB=QmId. Let r:=gcd(b1,…,bD,q). Since r divides every entry of Ba and also divides q, it follows from CBa=Qma that r∣gcd(Qma1,…,Qmad,q). Since gcd(a1,…,ad,q)=1, we have gcd(Qma1,…,Qmad,q)∣Qm. Hence r≤Qm. We set
q0:=rq,bj′:=rbj,1≤j≤D.
Then
gcd(b1′,…,bD′,q0)=1.(13)
Since m⋅A(x)a has integer coefficients, m⋅A(n+q)a−m⋅A(n)a is divisible by q. Hence the sequence
e(⟨m,A(n)z⟩)=e(qm⋅A(n)a)
is periodic in n with period q. Since only finitely many terms are removed from the symmetric averages, we obtain
The first factor is independent of n and has absolute value one. Note that the numerator of the second factor is a polynomial with integer coefficients, hence the second factor is periodic in n with period q0. Since q=rq0, the sum over 1≤n≤q consists of r identical blocks of length q0. Therefore
Let η=1/(2D). Since k≤D, we have −k1+2D1≤−2D1. Absorbing the finitely many possible values of k into a constant CD, and applying (14) we get
q1n=1∑qe(qm⋅A(n)a)≤CDq0−2D1.
Since q0=q/r and r≤Qm,
q1n=1∑qe(qm⋅A(n)a)≤CD(qQm)2D1.
Thus, if z∈Em,δ, then δ≤CD(qQm)2D1 and consequently
q≤Qm(δCD)2D.
Hence the order of every point in Em,δ is bounded in terms of m, δ, and A(x). There are only finitely many torsion points of Td of bounded order, and therefore Em,δ is finite. The hypothesis of Theorem 1.4 is satisfied. Hence, by Theorem 1.4 (b), the action GA↷Td is IC-rigid. □
References
School of Mathematics and Statistics, University of Sydney, NSW 2006, Australia
[1]Kamil Bulinski and Alexander Fish. Glasner property for linear group actions and their products. Math. Z., 303(4):Paper No. 99, 12, 2023.arxiv.org/abs/2210.16973
[2]Kamil Bulinski and Alexander Fish. Glasner property for unipotently generated group actions on tori. Israel J. Math., 255(1):109–122, 2023.arxiv.org/abs/2103.08912
[3]Daniel Berend and Yuval Peres. Asymptotically dense dilations of sets on the circle. J. London Math. Soc. (2), 47(1):1–17, 1993.DOI
[4]Yves Benoist and Jean-François Quint. Stationary measures and invariant subsets of homogeneous spaces (III). Ann. of Math. (2), 178(3):1017–1059, 2013.
[5]Changguang Dong. On density of infinite subsets I. Discrete Contin. Dyn. Syst., 39(5):2343–2359, 2019.arxiv.org/abs/1709.05591
[6]Changguang Dong. On density of infinite subsets II: Dynamics on homogeneous spaces. Proc. Amer. Math. Soc., 147(2):751–761, 2019.arxiv.org/abs/1709.05593
[7]Shmuel Glasner. Almost periodic sets and measures on the torus. Israel J. Math., 32(2-3):161–172, 1979.DOI
[8]L. K. Hua. Additive theory of prime numbers. Translations of Mathematical Monographs, Vol. 13. American Mathematical Society, Providence, RI, 1965.
[10]Michael Kelly and Thái Hoàng Lê. Uniform dilations in higher dimensions. J. Lond. Math. Soc. (2), 88(3):925–940, 2013.DOI
[11]Bryna Kra and Scott Schmieding. IC-rigidity and generalizations of a theorem of Mahler, 2026. arXiv:2611.05607.arxiv.org/abs/2611.05607
[12]Bryna Kra and Scott Schmieding. Invariant random compacts, 2026. arXiv:2605.03993.arxiv.org/abs/2605.03993