Introduction
Let Γ \Gamma Γ be a countable group, L ( Γ ) L(\Gamma) L ( Γ ) its group von Neumann algebra with the canonical trace τ \tau τ , canonical unitaries { λ ( g ) } g ∈ Γ \{\lambda(g)\}_{g\in\Gamma} { λ ( g ) } g ∈ Γ , and ∥ x ∥ 2 = τ ( x ∗ x ) 1 / 2 \lVert x\rVert_2=\tau(x^*x)^{1/2} ∥ x ∥ 2 = τ ( x ∗ x ) 1/2 . We denote by Sub ( Γ ) \operatorname{Sub}(\Gamma) Sub ( Γ ) the space of all subgroups of Γ \Gamma Γ endowed with the compact metrizable Chabauty topology and by SA ( L ( Γ ) ) \operatorname{SA}(L(\Gamma)) SA ( L ( Γ )) the space of all von Neumann subalgebras of L ( Γ ) L(\Gamma) L ( Γ ) endowed with the Effros–Maréchal topology [26 , 48 , 34 , 28 ] . Γ \Gamma Γ acts naturally on both these spaces. We will use the following two Γ \Gamma Γ -equivariant maps going in opposite directions, with L L L continuous and J J J Borel.
Sub ( Γ ) → L SA ( L ( Γ ) ) SA ( L ( Γ ) ) → J Sub ( Γ ) Δ ↦ L ( Δ ) M ↦ Γ M : = { γ ∣ λ ( γ ) ∈ M } \begin{aligned}
\operatorname{Sub}(\Gamma)&\xrightarrow{\ L\ }\operatorname{SA}(L(\Gamma))&
\operatorname{SA}(L(\Gamma))&\xrightarrow{\ J\ }\operatorname{Sub}(\Gamma)\\
\Delta&\mapsto L(\Delta)&
M&\mapsto\Gamma_M:=\{\gamma\mid\lambda(\gamma)\in M\}
\end{aligned} Sub ( Γ ) Δ L SA ( L ( Γ )) ↦ L ( Δ ) SA ( L ( Γ )) M J Sub ( Γ ) ↦ Γ M := { γ ∣ λ ( γ ) ∈ M }
Clearly J L ( Δ ) = Δ JL(\Delta)=\Delta J L ( Δ ) = Δ and L J ( M ) ≤ M LJ(M)\leq M L J ( M ) ≤ M . Typically, L L L is far from being surjective, but for invariant subalgebras the situation is sometimes different. A group Γ \Gamma Γ has the ISR property if every invariant M ∈ SA ( L ( Γ ) ) M\in\operatorname{SA}(L(\Gamma)) M ∈ SA ( L ( Γ )) is of the form L ( N ) L(N) L ( N ) for some normal subgroup N ∈ Sub ( Γ ) Γ N\in\operatorname{Sub}(\Gamma)^\Gamma N ∈ Sub ( Γ ) Γ . This rigidity was first established by Kalantar and Panagopoulos [42 Theorem 1.1] for irreducible lattices in higher rank semisimple Lie groups. A series of recent works [9 ] , [21 ] , [41 ] , [25 ] introduces this notion and establishes the ISR property in various settings. The study of invariant subalgebras has developed into a theory of its own; see also [5 , 3 ] .
Our goal in this note is to generalize such results to the setting of invariant random subgroups and subalgebras. Invariant random subgroups (IRSs for short) and their younger analogues invariant random algebras (IRAs) are Γ \Gamma Γ -invariant probability measures on Sub ( Γ ) \operatorname{Sub}(\Gamma) Sub ( Γ ) and on SA ( L ( Γ ) ) \operatorname{SA}(L(\Gamma)) SA ( L ( Γ )) respectively. Invariant subalgebras are the simplest examples of invariant random algebras: the Dirac measure δ M \delta_M δ M is an IRA exactly when M M M is Γ \Gamma Γ -invariant. Thus the study of invariant subalgebras can be viewed as the study of IRAs that are Dirac measures. IRSs were introduced by Abért, Glasner and Virág [2 ] . The notion of IRS and the measure theoretic methods that it brought with it into group theory turned out to be very fruitful, and the hope is that IRAs might play a similar role in the von Neumann algebra setting.
One can push forward any ν ∈ IRS ( Γ ) \nu\in\operatorname{IRS}(\Gamma) ν ∈ IRS ( Γ ) on Γ \Gamma Γ to an IRA L ∗ ν L_*\nu L ∗ ν . We call a countable group Γ \Gamma Γ IRA rigid if every IRA μ \mu μ of Γ \Gamma Γ is of the form L ∗ ν L_*\nu L ∗ ν for an IRS ν \nu ν of Γ \Gamma Γ . Our first two main theorems highlight situations for which this is the only way.
Theorem 1.1. Let n ≥ 1 n\geq1 n ≥ 1 and let Γ i \Gamma_i Γ i be countable acylindrically hyperbolic groups with trivial amenable radical. Set Γ = ⨁ 1 ≤ i ≤ n Γ i \Gamma=\bigoplus_{1\leq i\leq n}\Gamma_i Γ = ⨁ 1 ≤ i ≤ n Γ i . Then every IRA μ \mu μ of Γ \Gamma Γ is of the form μ = L ∗ ν \mu=L_*\nu μ = L ∗ ν for a unique IRS ν \nu ν of Γ \Gamma Γ .
The first theorem concerns products of groups with rank one behaviour. We now turn to higher rank lattices.
Theorem 1.2. Let K K K be a number field, let S S S be a finite set of places containing all Archimedean places, and let G \mathbf{G} G be a connected, absolutely simple, adjoint, K K K -split algebraic group with ∑ v ∈ S rank K v ( G ) ≥ 2 \sum_{v\in S}\operatorname{rank}_{K_v}(\mathbf{G})\geq2 ∑ v ∈ S rank K v ( G ) ≥ 2 . Let Γ < G ( K ) \Gamma<\mathbf{G}(K) Γ < G ( K ) be any subgroup commensurable
with G ( O S ) \mathbf{G}(\mathcal{O}_S) G ( O S ) . Then every ergodic IRA μ \mu μ of Γ \Gamma Γ is of the form μ = L ∗ ν \mu=L_*\nu μ = L ∗ ν for an ergodic IRS ν \nu ν of Γ \Gamma Γ .
By ergodic decomposition, and since L ( Sub ( Γ ) ) L(\operatorname{Sub}(\Gamma)) L ( Sub ( Γ )) is closed, Theorem 1.2 implies that Γ \Gamma Γ is IRA rigid.
While the two theorems look similar, their proofs are different in nature. For products of acylindrically hyperbolic groups, both IRSs and IRAs are abundant and the proof relies only on general, geometric properties of the group and recurrence. For higher rank lattices, on the other hand, the proof relies on deep rigidity results that amount to a full classification of all ergodic IRSs given by the Nevo–Stuck–Zimmer theorem [53 ] , [50 ] , [24 Corollary 1.4] and then of all extremal characters [16 ] , [52 ] , [15 ] , [24 ] of these groups. In turn, combining Theorem 1.2 with these classification results yields a complete classification of all IRAs for such K K K -split higher rank lattices.
Let us now discuss the proof strategy of the above theorems. Our result for products of acylindrically hyperbolic groups follows from a more general classification of boomerang subalgebras of their group von Neumann algebras. Our motivation comes from boomerang subgroups.
Boomerang subgroups were introduced in [33 ] and [31 ] . Here we define, similarly, boomerang subalgebras. These are subalgebras M ∈ SA ( L ( Γ ) ) M\in\operatorname{SA}(L(\Gamma)) M ∈ SA ( L ( Γ )) that are recurrent points in the corresponding topological dynamical system ( SA ( L ( Γ ) ) , ⟨ γ ⟩ ) (\operatorname{SA}(L(\Gamma)),\langle\gamma\rangle) ( SA ( L ( Γ )) , ⟨ γ ⟩) with respect to every element γ ∈ Γ \gamma\in\Gamma γ ∈ Γ separately. More explicitly, for every e ≠ γ ∈ Γ e\ne\gamma\in\Gamma e = γ ∈ Γ there is an increasing sequence n i → + ∞ n_i\to+\infty n i → + ∞ such that λ ( γ ) n i M λ ( γ ) − n i → M \lambda(\gamma)^{n_i}M\lambda(\gamma)^{-n_i}\to M λ ( γ ) n i M λ ( γ ) − n i → M in the Effros–Maréchal topology. Boomerang subgroups are defined analogously.
It is a direct consequence of Poincaré recurrence (see [33 Example 2.9(5)] ) that every IRS gives full measure to the set of boomerang subgroups. The same application of Poincaré recurrence shows that every IRA gives full measure to the boomerang subalgebras (since SA ( L ( Γ ) ) \operatorname{SA}(L(\Gamma)) SA ( L ( Γ )) is a Polish space [48 ] ). Thus boomerang subgroups generalize IRSs and recent works [47 ] , [32 ] have shown that in free groups this is indeed a generalization as there are many boomerangs that do not come from IRSs, which makes the following theorem all the more interesting:
Theorem 1.3. Let n ≥ 1 n\ge1 n ≥ 1 and let Γ i \Gamma_i Γ i be countable acylindrically hyperbolic groups with trivial amenable radical. Set Γ = ⨁ 1 ≤ i ≤ n Γ i \Gamma=\bigoplus_{1\le i\le n}\Gamma_i Γ = ⨁ 1 ≤ i ≤ n Γ i . Then, every boomerang subalgebra M ∈ SA ( L ( Γ ) ) M\in\operatorname{SA}(L(\Gamma)) M ∈ SA ( L ( Γ )) is of the form L ( Δ ) L(\Delta) L ( Δ ) for a unique boomerang subgroup Δ ∈ Sub ( Γ ) \Delta\in\operatorname{Sub}(\Gamma) Δ ∈ Sub ( Γ ) .
While this directly implies Theorem 1.1 , it is important to note that the two theorems are of different nature. Boomerangs are deterministic objects, not measures. Consequently, Theorem 1.3 is not an almost everywhere statement, but rather a statement about every boomerang subalgebra. In the higher rank setting, boomerangs still play a crucial role, but unfortunately our proof falls short of establishing a similar generalization of Theorem 1.2 .
We do not know whether the ISR property, or the property that every boomerang subalgebra is a group von Neumann algebra, is preserved under finite direct products. Thus the product statement in Theorem 1.3 requires a separate argument.
The groups in Theorems 1.1 and 1.2 belong to the larger class of C ∗ C^* C ∗ -simple groups; see [22 Theorem 8.14] , [17 Theorem 1] and [23 Proposition 19(i)] . It is natural to ask whether there is a dynamical characterization of IRA rigid groups. We take a first step in this direction. For every nontrivial countable group Γ \Gamma Γ , the free product Γ ∗ ⟨ t ⟩ \Gamma* \langle t\rangle Γ ∗ ⟨ t ⟩ , where ⟨ t ⟩ ≅ Z \langle t\rangle\cong\mathbb{Z} ⟨ t ⟩ ≅ Z , is acylindrically hyperbolic (see, for example, [49 Corollary 2.2 and the paragraph following it] ). Its amenable radical is trivial, so it is IRA rigid by Theorem 1.1 . To obtain IRA rigidity for Γ \Gamma Γ from this larger group, we use mixed identity-freeness.
Recall that a nontrivial group Γ \Gamma Γ is mixed identity-free if every nonidentity element of Γ ∗ ⟨ t ⟩ \Gamma* \langle t\rangle Γ ∗ ⟨ t ⟩ survives under some homomorphism Γ ∗ ⟨ t ⟩ → Γ \Gamma* \langle t\rangle\to\Gamma Γ ∗ ⟨ t ⟩ → Γ fixing Γ \Gamma Γ pointwise, where ⟨ t ⟩ ≅ Z \langle t\rangle\cong\mathbb{Z} ⟨ t ⟩ ≅ Z [35 ] .
Theorem 1.4. Every finite product of countable nontrivial mixed identity-free groups is IRA rigid.
As a consequence, we obtain IRA rigidity for finite products of groups admitting the following actions.
Theorem 1.5. Let n ≥ 1 n \ge1 n ≥ 1 and let Γ i \Gamma_i Γ i be countable groups for 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n . Suppose that each Γ i \Gamma_i Γ i satisfies one of the following conditions:
(1) Γ i \Gamma_i Γ i admits a topologically free minimal extremely proximal action on a compact Hausdorff space with more than two points.
(2) Γ i \Gamma_i Γ i admits a faithful minimal action of general type on a tree, and the induced action on the boundary of the tree is topologically free.
Then Γ = ⨁ 1 ≤ i ≤ n Γ i \Gamma= \bigoplus_{1 \le i \le n}\Gamma_i Γ = ⨁ 1 ≤ i ≤ n Γ i is IRA rigid.
In particular, the Baumslag–Solitar groups BS ( m , n ) \operatorname{BS}(m,n) BS ( m , n ) with ∣ m ∣ , ∣ n ∣ ≥ 2 |m|, |n| \ge2 ∣ m ∣ , ∣ n ∣ ≥ 2 and ∣ m ∣ ≠ ∣ n ∣ |m| \ne|n| ∣ m ∣ = ∣ n ∣ , and their finite products, are IRA rigid (see Corollary 6.5 ).
Theorem 1.4 also gives amenable examples of IRA rigid groups beyond this class of C ∗ C^* C ∗ -simple groups. There exists a two-generated infinite elementary amenable IRA rigid group and, for each prime p p p , an infinite locally finite p p p -group which is IRA rigid. We also obtain continuum many pairwise nonisomorphic finitely generated nonamenable IRA rigid groups which contain no nonabelian free subgroup and have nontrivial amenable radical (see Corollary 6.6 ). Moreover, every countable amenable group embeds into a finitely generated amenable IRA rigid group, and the overgroup can be chosen elementary amenable when the given group is elementary amenable (see Corollary 6.7 ).
For a mixed identity-free group Γ \Gamma Γ , our proof uses positive-definite functions and their associated trace-preserving UCP maps to construct an IRA of Γ ∗ ⟨ t ⟩ \Gamma* \langle t\rangle Γ ∗ ⟨ t ⟩ from an IRA of Γ \Gamma Γ . We then apply the acylindrically hyperbolic case (Theorem 1.1) and recover the original IRA by intersecting with L ( Γ ) L(\Gamma) L ( Γ ) .
We also spend some time discussing the structural properties of Boomerang subalgebras and in particular deal with amenable Boomerang subalgebras or the lack thereof. Boomerang subgroups generalize normal subgroups (for trivial reasons). Since a big chunk of the class of groups considered here are C ∗ C^* C ∗ -simple which admit no nontrivial amenable normal subgroup [18 ] , it is natural to ask when the same conclusion holds for amenable boomerang subgroups, and more generally for amenable boomerang subalgebras.
We answer this question under some natural assumptions on the group, or when they admit certain actions. In particular, we prove that every amenable boomerang subalgebra is trivial when Γ \Gamma Γ has property P n a i P_{\mathrm{nai}} P nai , admits a faithful extremely proximal action on a compact Hausdorff space with more than two points, or admits an action with property P g e o P_{\mathrm{geo}} P geo (see Theorem 3.18 ). The properties P n a i P_{\mathrm{nai}} P nai and P g e o P_{\mathrm{geo}} P geo were introduced by Bekka–Cowling–de la Harpe [17 ] . These conclusions strengthen the corresponding results for invariant amenable subalgebras and amenable IRAs in [8 Theorems A and B] .
Organization of the paper. In addition to this introduction, this paper has five other sections. § 2 collects the preliminaries, including the correspondence between positive-definite functions and trace preserving UCP maps (Proposition 2.2 ). In § 3 , we study basic properties of boomerang subalgebras, their associated positive-definite functions, and finite dimensional and amenable boomerang subalgebras. In particular, we prove Theorem 3.18 here. § 4 is devoted to classifying the IRAs of K K K -split higher rank lattices. It should not come as a surprise then that we prove Theorem 1.2 in this section. In § 5 , we classify boomerang subalgebras of acylindrically hyperbolic groups and their finite products, and end up proving Theorem 1.1 and Theorem 1.3 . Finally, in § 6 , we prove IRA rigidity for mixed identity-free groups and their finite products. That is to say, we finish by proving Theorem 1.4 and Theorem 1.5 , and obtain the Corollaries 6.6 and 6.7 .
Recent Developments. While this draft was being readied, there were two recent developments that deserve particular mention.
The first is joint work of Amrutam, Klisse and Pagliero [12 ] , in which the relative ISR property was first established for all countable nontrivial mixed identity-free groups. In particular, this includes groups admitting topologically free minimal extremely proximal actions on compact Hausdorff spaces with more than two points, and the Baumslag–Solitar groups B S ( m , n ) \mathrm{BS}(m,n) BS ( m , n ) with ∣ m ∣ , ∣ n ∣ ≥ 2 |m|, |n| \ge2 ∣ m ∣ , ∣ n ∣ ≥ 2 and ∣ m ∣ ≠ ∣ n ∣ |m| \ne|n| ∣ m ∣ = ∣ n ∣ . The Baumslag–Solitar result answers the corresponding part of a question of Dudko and Jiang [25 Question 6.1] . The techniques there are different from those employed here: the proof uses tracial ultrapowers, and that argument does not directly extend to invariant random algebras.
The second is the recent work of Fima, Oppelmayer and Tao [30 ] , who introduce boomerang C ∗ C^* C ∗ -subalgebras using the Wijsmann topology. Among other things, they show that every boomerang subalgebra of C r ∗ ( Γ ) C^*_r(\Gamma) C r ∗ ( Γ ) comes from a boomerang subgroup when Γ \Gamma Γ is a torsion-free non-amenable hyperbolic group [30 Theorems 2.13 and 6.8] . These rigidity results are in the C ∗ C^* C ∗ -algebraic setting, whereas ours concern group von Neumann algebras, and the proofs use different techniques.
AI Declaration. The mathematical ideas, results, and proofs are due to the authors. ChatGPT (OpenAI) was used for grammatical corrections, assistance with exposition and references, and feedback on some of the authors’ arguments. The authors take full responsibility for the manuscript.
Acknowledgments. This work began in 2024 when T.A. was supported by the European Research Council (ERC) under the European Union’s Horizon Europe research and innovation programme, Grant agreement No. 101078193 and Israel Science Foundation grant 1175/18. Much water has flown under the bridge since then. T.A. is currently supported by National Science Centre, Poland Sonata, grant number 2025/59/D/ST1/03117.
Preliminaries and setting
Throughout this paper, Γ \Gamma Γ is a countable group, and τ 0 \tau_{0} τ 0 is the canonical trace on L ( Γ ) L(\Gamma) L ( Γ ) . For any x ∈ L ( Γ ) x \in L(\Gamma) x ∈ L ( Γ ) , we write supp ( x ) = { g ∈ Γ : τ 0 ( x λ ( g ) ∗ ) ≠ 0 } \operatorname{supp}(x)=\{g \in\Gamma:\tau_{0}(x\lambda(g)^{*})\ne0\} supp ( x ) = { g ∈ Γ : τ 0 ( x λ ( g ) ∗ ) = 0 } . Also, we shall use ∥ x ∥ 2 = τ ( x ∗ x ) 1 / 2 \lVert x\rVert_{2}=\tau(x^{*}x)^{1/2} ∥ x ∥ 2 = τ ( x ∗ x ) 1/2 , the ∥ ⋅ ∥ 2 \lVert\cdot\rVert_{2} ∥ ⋅ ∥ 2 -norm associated with the canonical trace. We also write τ = τ 0 \tau=\tau_{0} τ = τ 0 and use λ ( g ) \lambda(g) λ ( g ) for the canonical group unitaries. All subalgebras in this paper are unital von Neumann subalgebras.
Algebras
For a subalgebra A ≤ L ( Γ ) \mathcal{A}\le L(\Gamma) A ≤ L ( Γ ) , let E A : L ( Γ ) → A E_{\mathcal{A}}:L(\Gamma)\to\mathcal{A} E A : L ( Γ ) → A be the unique τ \tau τ -preserving conditional expectation [19 Chapter 1] . We equip SA ( L ( Γ ) ) \operatorname{SA}(L(\Gamma)) SA ( L ( Γ )) with the Effros–Maréchal topology. This topology is Polish [48 , 28 ] ; it is not compact when Γ \Gamma Γ is infinite [7 Theorem A] . In the finite tracial setting, A i → A \mathcal{A}_{i}\to\mathcal{A} A i → A in the Effros–Maréchal topology if and only if ∥ E A i ( x ) − E A ( x ) ∥ 2 → 0 \lVert E_{\mathcal{A}_{i}}(x)-E_{\mathcal{A}}(x)\rVert_{2}\to0 ∥ E A i ( x ) − E A ( x ) ∥ 2 → 0 for every x ∈ L ( Γ ) x\in L(\Gamma) x ∈ L ( Γ ) ; see [34 Corollary 2.12] and [10 Proposition 2.2] .
The following fact is well known to experts. A proof can be found in [10 Lemma 2.3] .
Lemma 2.1. Let A ≤ L ( Γ ) \mathcal{A}\le L(\Gamma) A ≤ L ( Γ ) be a subalgebra and t ∈ Γ t\in\Gamma t ∈ Γ . For every x ∈ L ( Γ ) x\in L(\Gamma) x ∈ L ( Γ ) ,
E λ ( t ) A λ ( t ) ∗ ( x ) = λ ( t ) E A ( λ ( t ) ∗ x λ ( t ) ) λ ( t ) ∗ . E_{\lambda(t)\mathcal{A}\lambda(t)^{*}}(x)=\lambda(t)E_{\mathcal{A}}\left(\lambda(t)^{*}x\lambda(t)\right)\lambda(t)^{*}. E λ ( t ) A λ ( t ) ∗ ( x ) = λ ( t ) E A ( λ ( t ) ∗ x λ ( t ) ) λ ( t ) ∗ .
Invariant random algebras and positive-definite functions
A function ϕ : Γ → C \phi:\Gamma\to\mathbb{C} ϕ : Γ → C is positive-definite if ∑ i , j = 1 n c i c j ‾ ϕ ( g j − 1 g i ) ≥ 0 \sum_{i,j=1}^{n}c_{i}\overline{c_{j}}\phi(g_{j}^{-1}g_{i})\ge0 ∑ i , j = 1 n c i c j ϕ ( g j − 1 g i ) ≥ 0 for every g 1 , … , g n ∈ Γ g_{1},\ldots,g_{n}\in\Gamma g 1 , … , g n ∈ Γ and c 1 , … , c n ∈ C c_{1},\ldots,c_{n}\in\mathbb{C} c 1 , … , c n ∈ C . It is normalized if ϕ ( e ) = 1 \phi(e)=1 ϕ ( e ) = 1 . A normalized positive-definite function which is conjugation invariant is called a character.
Given a von Neumann subalgebra A ≤ L ( Γ ) \mathcal{A}\le L(\Gamma) A ≤ L ( Γ ) , we associate to it the normalized positive-definite function defined by
ϕ A ( g ) = τ 0 ( E A ( λ ( g ) ) λ ( g ) ∗ ) = ∥ E A ( λ ( g ) ) ∥ 2 2 , g ∈ Γ . \phi_{\mathcal{A}}(g)=\tau_{0}\left(E_{\mathcal{A}}(\lambda(g))\lambda(g)^{*}\right)=\lVert E_{\mathcal{A}}(\lambda(g))\rVert_{2}^{2},\quad g\in\Gamma. ϕ A ( g ) = τ 0 ( E A ( λ ( g )) λ ( g ) ∗ ) = ∥ E A ( λ ( g )) ∥ 2 2 , g ∈ Γ.
The second equality follows because E A E_{\mathcal{A}} E A is the orthogonal projection onto L 2 ( A , τ 0 ) L^{2}(\mathcal{A},\tau_{0}) L 2 ( A , τ 0 ) . This function has been studied in [41 , 25 , 10 ] and, more recently, in [3 ] . For positive definiteness, see [41 Proposition 3.2] and [10 Remark 4.2(5)] . If A \mathcal{A} A is Γ \Gamma Γ -invariant, then ϕ A \phi_{\mathcal{A}} ϕ A is a character.
Recall that an invariant random von Neumann subalgebra is a Γ \Gamma Γ -invariant Borel probability measure on SA ( L ( Γ ) ) \operatorname{SA}(L(\Gamma)) SA ( L ( Γ )) . Given an IRA μ \mu μ , the associated function ϕ μ \phi_{\mu} ϕ μ defined by
ϕ μ ( g ) = ∫ SA ( L ( Γ ) ) ϕ A ( g ) d μ ( A ) , g ∈ Γ \phi_{\mu}(g)=\int_{\operatorname{SA}(L(\Gamma))}\phi_{\mathcal{A}}(g)\,d\mu(\mathcal{A}),\quad g\in\Gamma ϕ μ ( g ) = ∫ SA ( L ( Γ )) ϕ A ( g ) d μ ( A ) , g ∈ Γ
ંગ્રેસ
is a character. Indeed, normalization and positive definiteness are preserved by integration. For s , g ∈ Γ s,g \in\Gamma s , g ∈ Γ , Lemma 2.1 gives ϕ A ( s g s − 1 ) = ϕ ( s − 1 ) A ( g ) \phi_A(sgs^{-1})=\phi_{(s^{-1})\mathcal{A}}(g) ϕ A ( s g s − 1 ) = ϕ ( s − 1 ) A ( g ) . Integrating this equality and using the invariance of μ \mu μ gives ϕ μ ( s g s − 1 ) = ϕ μ ( g ) \phi_\mu(sgs^{-1})=\phi_\mu(g) ϕ μ ( s g s − 1 ) = ϕ μ ( g ) .
Positive-definite functions and UCP maps
We now establish a one-to-one correspondence between certain positive definite functions on the product group Γ × Γ \Gamma\times\Gamma Γ × Γ and trace preserving UCP maps on L ( Γ ) L(\Gamma) L ( Γ ) . Even though this correspondence might be well-known to the experts, we add the details here for completeness. We follow the bimodule construction in [13 Section 13.1.2 and Lemma 13.1.8] .
Recall that the compact metrizable space P D 1 ( Γ × Γ ) \mathrm{PD}_1(\Gamma\times\Gamma) PD 1 ( Γ × Γ ) consists of the normalized positive-definite functions on Γ × Γ \Gamma\times\Gamma Γ × Γ , with pointwise convergence and action ( s ⋅ ψ ) ( g , h ) = ψ ( s − 1 g s , s − 1 h s ) (s \cdot\psi)(g,h)=\psi(s^{-1}gs,s^{-1}hs) ( s ⋅ ψ ) ( g , h ) = ψ ( s − 1 g s , s − 1 h s ) .
Proposition 2.2. Let Γ \Gamma Γ be a countable group and let ψ ∈ P D 1 ( Γ × Γ ) \psi\in\mathrm{PD}_1(\Gamma\times\Gamma) ψ ∈ PD 1 ( Γ × Γ ) satisfy ψ ( g , e ) = ψ ( e , g ) = δ g , e \psi(g,e)=\psi(e,g)=\delta_{g,e} ψ ( g , e ) = ψ ( e , g ) = δ g , e for every g ∈ Γ g \in\Gamma g ∈ Γ . There is a unique normal trace-preserving UCP map T ψ : L ( Γ ) → L ( Γ ) T_\psi:L(\Gamma)\to L(\Gamma) T ψ : L ( Γ ) → L ( Γ ) such that τ 0 ( T ψ ( λ ( g ) ) λ ( h ) ∗ ) = ψ ( g , h ) \tau_0(T_\psi(\lambda(g))\lambda(h)^*)=\psi(g,h) τ 0 ( T ψ ( λ ( g )) λ ( h ) ∗ ) = ψ ( g , h ) for every g , h ∈ Γ g,h \in\Gamma g , h ∈ Γ . Conversely, every normal trace-preserving UCP map on L ( Γ ) L(\Gamma) L ( Γ ) arises in this way. The correspondence ψ ↦ T ψ \psi\mapsto T_\psi ψ ↦ T ψ is equivariant and is a homeomorphism for pointwise convergence on the functions and point ultraweak topology on the normal trace-preserving UCP maps.
Proof. Let ψ ∈ P D 1 ( Γ × Γ ) \psi\in\mathrm{PD}_1(\Gamma\times\Gamma) ψ ∈ PD 1 ( Γ × Γ ) be such that ψ ( g , e ) = ψ ( e , g ) = δ g , e \psi(g,e)=\psi(e,g)=\delta_{g,e} ψ ( g , e ) = ψ ( e , g ) = δ g , e for every g ∈ Γ g \in\Gamma g ∈ Γ . We equip C [ Γ × Γ ] \mathbb{C}[\Gamma\times\Gamma] C [ Γ × Γ ] with the positive semidefinite form ⟨ [ g , h ] , [ k , l ] ⟩ = ψ ( k − 1 g , l − 1 h ) \langle[g,h],[k,l]\rangle=\psi(k^{-1}g,l^{-1}h) ⟨[ g , h ] , [ k , l ]⟩ = ψ ( k − 1 g , l − 1 h ) . After the usual completion procedure, we obtain the associated Hilbert space H \mathcal{H} H . The operators π ( s , t ) [ g , h ] = [ s g , t h ] \pi(s,t)[g,h]=[sg,th] π ( s , t ) [ g , h ] = [ s g , t h ] define a unitary representation of Γ × Γ \Gamma\times\Gamma Γ × Γ . For ξ = [ e , e ] \xi=[e,e] ξ = [ e , e ] , one has ψ ( g , h ) = ⟨ π ( g , h ) ξ , ξ ⟩ \psi(g,h)=\langle\pi(g,h)\xi,\xi\rangle ψ ( g , h ) = ⟨ π ( g , h ) ξ , ξ ⟩ , and the vectors π ( g , h ) ξ \pi(g,h)\xi π ( g , h ) ξ span a dense subspace of H \mathcal{H} H . The two coordinate representations commute. Set A = { π ( g , e ) : g ∈ Γ } ′ ′ A=\{\pi(g,e):g\in\Gamma\}^{\prime\prime} A = { π ( g , e ) : g ∈ Γ } ′′ and φ ( a ) = ⟨ a ξ , ξ ⟩ \varphi(a)=\langle a\xi,\xi\rangle φ ( a ) = ⟨ a ξ , ξ ⟩ for a ∈ A a\in A a ∈ A . Since φ ( π ( g , e ) ) = δ g , e \varphi(\pi(g,e))=\delta_{g,e} φ ( π ( g , e )) = δ g , e , the state φ \varphi φ is tracial on the group algebra, hence on A A A by separate ultraweak continuity of multiplication.
We claim that φ \varphi φ is faithful. Indeed, if φ ( a ∗ a ) = 0 \varphi(a^*a)=0 φ ( a ∗ a ) = 0 , then
∥ a π ( g , e ) ξ ∥ 2 = φ ( π ( g , e ) ∗ a ∗ a π ( g , e ) ) = φ ( a ∗ a ) = 0. \lVert a\pi(g,e)\xi\rVert^2=\varphi(\pi(g,e)^*a^*a\pi(g,e))=\varphi(a^*a)=0. ∥ aπ ( g , e ) ξ ∥ 2 = φ ( π ( g , e ) ∗ a ∗ aπ ( g , e )) = φ ( a ∗ a ) = 0.
Since a a a commutes with π ( e , h ) \pi(e,h) π ( e , h ) , this gives a π ( g , h ) ξ = 0 a\pi(g,h)\xi=0 aπ ( g , h ) ξ = 0 for all g , h ∈ Γ g,h\in\Gamma g , h ∈ Γ , and therefore a = 0 a=0 a = 0 . Thus ( A , φ ) (A,\varphi) ( A , φ ) is a finite tracial von Neumann algebra. The map δ g ↦ π ( g , e ) \delta_g\mapsto\pi(g,e) δ g ↦ π ( g , e ) extends to a unitary ℓ 2 ( Γ ) → L 2 ( A , φ ) \ell^2(\Gamma)\to L^2(A,\varphi) ℓ 2 ( Γ ) → L 2 ( A , φ ) intertwining the left regular representation with left multiplication by π ( g , e ) \pi(g,e) π ( g , e ) . It induces a unital ∗ * ∗ -isomorphism π 1 : L ( Γ ) → A ⊆ B ( H ) \pi_1:L(\Gamma)\to A\subseteq B(\mathcal{H}) π 1 : L ( Γ ) → A ⊆ B ( H ) satisfying π 1 ( λ ( g ) ) = π ( g , e ) \pi_1(\lambda(g))=\pi(g,e) π 1 ( λ ( g )) = π ( g , e ) and φ ∘ π 1 = τ 0 \varphi\circ\pi_1=\tau_0 φ ∘ π 1 = τ 0 . Note that π 1 \pi_1 π 1 is normal. Indeed, if 0 ≤ x i ↑ x 0\leq x_i\uparrow x 0 ≤ x i ↑ x , then φ ( π 1 ( x ) − sup i π 1 ( x i ) ) = τ 0 ( x ) − sup i τ 0 ( x i ) = 0 \varphi(\pi_1(x)-\sup_i\pi_1(x_i))=\tau_0(x)-\sup_i\tau_0(x_i)=0 φ ( π 1 ( x ) − sup i π 1 ( x i )) = τ 0 ( x ) − sup i τ 0 ( x i ) = 0 , so that (using the faithfulness of φ \varphi φ ) π 1 ( x ) = sup i π 1 ( x i ) \pi_1(x)=\sup_i\pi_1(x_i) π 1 ( x ) = sup i π 1 ( x i ) .
Define V : ℓ 2 ( Γ ) → H V:\ell^2(\Gamma)\to\mathcal{H} V : ℓ 2 ( Γ ) → H by V δ g = π ( e , g − 1 ) ξ V\delta_g=\pi(e,g^{-1})\xi V δ g = π ( e , g − 1 ) ξ . Since ⟨ V δ g , V δ h ⟩ = ψ ( e , h g − 1 ) = δ g , h \langle V\delta_g,V\delta_h\rangle=\psi(e,hg^{-1})=\delta_{g,h} ⟨ V δ g , V δ h ⟩ = ψ ( e , h g − 1 ) = δ g , h , it follows that V V V is an isometry. Define, initially with values in B ( ℓ 2 ( Γ ) ) B(\ell^2(\Gamma)) B ( ℓ 2 ( Γ )) , T ψ ( x ) = V ∗ π 1 ( x ) V T_\psi(x)=V^*\pi_1(x)V T ψ ( x ) = V ∗ π 1 ( x ) V . This map is normal and UCP, since π 1 \pi_1 π 1 is normal and V V V is an isometry. Let R t δ g = δ g t − 1 R_t\delta_g=\delta_{gt^{-1}} R t δ g = δ g t − 1 be the right regular unitaries. For s , t , g , x ∈ Γ s,t,g,x\in\Gamma s , t , g , x ∈ Γ , we have
⟨ T ψ ( λ ( s ) ) R t δ g , δ x ⟩ = ⟨ V ∗ π ( s , t g − 1 ) ξ , δ x ⟩ = ⟨ π ( s , t g − 1 ) ξ , π ( e , x − 1 ) ξ ⟩ = ψ ( s , x t g − 1 ) , \begin{aligned}
\langle T_\psi(\lambda(s))R_t\delta_g,\delta_x\rangle
&=\langle V^*\pi(s,tg^{-1})\xi,\delta_x\rangle\\
&=\langle\pi(s,tg^{-1})\xi,\pi(e,x^{-1})\xi\rangle\\
&=\psi(s,xtg^{-1}),
\end{aligned} ⟨ T ψ ( λ ( s )) R t δ g , δ x ⟩ = ⟨ V ∗ π ( s , t g − 1 ) ξ , δ x ⟩ = ⟨ π ( s , t g − 1 ) ξ , π ( e , x − 1 ) ξ ⟩ = ψ ( s , x t g − 1 ) ,
⟨ R t T ψ ( λ ( s ) ) δ g , δ x ⟩ = ⟨ V ∗ π ( s , g − 1 ) ξ , R t ∗ δ x ⟩ = ⟨ π ( s , g − 1 ) ξ , π ( e , t − 1 x − 1 ) ξ ⟩ = ψ ( s , x t g − 1 ) . \begin{aligned}
\langle R_tT_\psi(\lambda(s))\delta_g,\delta_x\rangle
&=\langle V^*\pi(s,g^{-1})\xi,R_t^*\delta_x\rangle\\
&=\langle\pi(s,g^{-1})\xi,\pi(e,t^{-1}x^{-1})\xi\rangle\\
&=\psi(s,xtg^{-1}).
\end{aligned} ⟨ R t T ψ ( λ ( s )) δ g , δ x ⟩ = ⟨ V ∗ π ( s , g − 1 ) ξ , R t ∗ δ x ⟩ = ⟨ π ( s , g − 1 ) ξ , π ( e , t − 1 x − 1 ) ξ ⟩ = ψ ( s , x t g − 1 ) .
Thus T ψ ( λ ( s ) ) R t = R t T ψ ( λ ( s ) ) T_\psi(\lambda(s))R_t=R_tT_\psi(\lambda(s)) T ψ ( λ ( s )) R t = R t T ψ ( λ ( s )) . Normality of T ψ T_\psi T ψ and ultraweak density of C Γ \mathbb{C}\Gamma C Γ force T ψ ( a ) R t = R t T ψ ( a ) T_\psi(a)R_t=R_tT_\psi(a) T ψ ( a ) R t = R t T ψ ( a ) for every a ∈ L ( Γ ) a\in L(\Gamma) a ∈ L ( Γ ) . Hence T ψ ( a ) ∈ R ( Γ ) ′ = L ( Γ ) T_\psi(a)\in R(\Gamma)'=L(\Gamma) T ψ ( a ) ∈ R ( Γ ) ′ = L ( Γ ) , where R ( Γ ) = { R t : t ∈ Γ } ′ ′ R(\Gamma)=\{R_t:t\in\Gamma\}'' R ( Γ ) = { R t : t ∈ Γ } ′′ . Moreover, τ 0 ( T ψ ( x ) ) = ⟨ π 1 ( x ) ξ , ξ ⟩ = φ ( π 1 ( x ) ) = τ 0 ( x ) \tau_0(T_\psi(x))=\langle\pi_1(x)\xi,\xi\rangle=\varphi(\pi_1(x))=\tau_0(x) τ 0 ( T ψ ( x )) = ⟨ π 1 ( x ) ξ , ξ ⟩ = φ ( π 1 ( x )) = τ 0 ( x ) . Finally, τ 0 ( T ψ ( λ ( g ) ) λ ( h ) ∗ ) = ⟨ π ( g , e ) ξ , π ( e , h − 1 ) ξ ⟩ = ψ ( g , h ) \tau_0(T_\psi(\lambda(g))\lambda(h)^*)=\langle\pi(g,e)\xi,\pi(e,h^{-1})\xi\rangle=\psi(g,h) τ 0 ( T ψ ( λ ( g )) λ ( h ) ∗ ) = ⟨ π ( g , e ) ξ , π ( e , h − 1 ) ξ ⟩ = ψ ( g , h ) , which proves the claim. The Fourier coefficients determine T ψ T_\psi T ψ on the group algebra, hence everywhere by normality.
Conversely, let T : L ( Γ ) → L ( Γ ) T:L(\Gamma)\to L(\Gamma) T : L ( Γ ) → L ( Γ ) be a normal, trace-preserving UCP map. Define ψ ( g , h ) = τ 0 ( T ( λ ( g ) ) λ ( h ) ∗ ) \psi(g,h)=\tau_0(T(\lambda(g))\lambda(h)^*) ψ ( g , h ) = τ 0 ( T ( λ ( g )) λ ( h ) ∗ ) . For g i , h i ∈ Γ g_i,h_i\in\Gamma g i , h i ∈ Γ and c i ∈ C c_i\in\mathbb{C} c i ∈ C , complete positivity gives
∑ i , j c i ‾ c j ψ ( g i − 1 g j , h i − 1 h j ) = τ 0 ( ∑ i , j c i ‾ c j λ ( h i ) T ( λ ( g i ) ∗ λ ( g j ) ) λ ( h j ) ∗ ) ≥ 0. \sum_{i,j}\overline{c_i}c_j\psi(g_i^{-1}g_j,h_i^{-1}h_j)
=\tau_0\left(\sum_{i,j}\overline{c_i}c_j\lambda(h_i)T(\lambda(g_i)^*\lambda(g_j))\lambda(h_j)^*\right)\geq0. i , j ∑ c i c j ψ ( g i − 1 g j , h i − 1 h j ) = τ 0 ( i , j ∑ c i c j λ ( h i ) T ( λ ( g i ) ∗ λ ( g j )) λ ( h j ) ∗ ) ≥ 0.
Unitality and trace preservation give ψ ( g , e ) = ψ ( e , g ) = δ g , e \psi(g,e)=\psi(e,g)=\delta_{g,e} ψ ( g , e ) = ψ ( e , g ) = δ g , e . Thus ψ ∈ PD 1 ( Γ × Γ ) \psi\in\operatorname{PD}_1(\Gamma\times\Gamma) ψ ∈ PD 1 ( Γ × Γ ) , and uniqueness gives T = T ψ T=T_\psi T = T ψ .
Moreover, the inverse map T ⟼ ( ( g , h ) ⟼ τ 0 ( T ( λ ( g ) ) λ ( h ) ∗ ) ) T\longmapsto((g,h)\longmapsto\tau_0(T(\lambda(g))\lambda(h)^*)) T ⟼ (( g , h ) ⟼ τ 0 ( T ( λ ( g )) λ ( h ) ∗ )) is continuous. The space of normal trace-preserving UCP maps is compact for the point ultraweak topology [29 Remark 4.6] , so this continuous bijection has a continuous inverse. The coefficient identity and uniqueness give T s ⋅ ψ = Ad ( λ ( s ) ) ∘ T ψ ∘ Ad ( λ ( s ) ∗ ) T_{s\cdot\psi}=\operatorname{Ad}(\lambda(s))\circ T_\psi\circ\operatorname{Ad}(\lambda(s)^*) T s ⋅ ψ = Ad ( λ ( s )) ∘ T ψ ∘ Ad ( λ ( s ) ∗ ) . Thus the correspondence is equivariant. □
Thickness and vanishing
We recall the thickness and vanishing principle from [3 ] (which itself was motivated by [25 ] ). Let ϕ : Γ → C \phi:\Gamma\to\mathbb{C} ϕ : Γ → C be a normalized positive-definite function. A set K ⊆ Γ K\subseteq\Gamma K ⊆ Γ is ϕ \phi ϕ -thick if, for every g ∈ Γ ∖ { e } g\in\Gamma\setminus\{e\} g ∈ Γ ∖ { e } , there is a sequence ( h i ) ⊆ Γ (h_i)\subseteq\Gamma ( h i ) ⊆ Γ such that h j g h i − 1 h i g h j − 1 ∈ K h_jgh_i^{-1}h_igh_j^{-1}\in K h j g h i − 1 h i g h j − 1 ∈ K (i ≠ j i\ne j i = j ), and ϕ ( h i g h i − 1 ) ⟶ ϕ ( g ) \phi(h_igh_i^{-1})\longrightarrow\phi(g) ϕ ( h i g h i − 1 ) ⟶ ϕ ( g ) as i → ∞ i\to\infty i → ∞ .
The following is the GNS vanishing argument of [3 Lemma 3.6] . We include the proof for completeness.
Lemma 2.3. Let ϕ \phi ϕ be a normalized positive-definite function on Γ \Gamma Γ , and let K ⊆ Γ K\subseteq\Gamma K ⊆ Γ be ϕ \phi ϕ -thick. If ϕ ∣ K = 0 \left.\phi\right|_K=0 ϕ ∣ K = 0 , then ϕ = δ e \phi=\delta_e ϕ = δ e .
Proof. Let ( π , H , ξ ) (\pi,\mathcal{H},\xi) ( π , H , ξ ) be the GNS representation of ϕ \phi ϕ , so that ∥ ξ ∥ = 1 \lVert\xi\rVert=1 ∥ ξ ∥ = 1 and ϕ ( t ) = ⟨ π ( t ) ξ , ξ ⟩ \phi(t)=\langle\pi(t)\xi,\xi\rangle ϕ ( t ) = ⟨ π ( t ) ξ , ξ ⟩ for every t ∈ Γ t\in\Gamma t ∈ Γ . Fix g ≠ e g\ne e g = e , and choose ( h i ) (h_i) ( h i ) as in the definition of ϕ \phi ϕ -thickness. Set v i = π ( h i g h i − 1 ) ξ v_i=\pi(h_igh_i^{-1})\xi v i = π ( h i g h i − 1 ) ξ . Since π ( h i g h i − 1 ) \pi(h_igh_i^{-1}) π ( h i g h i − 1 ) is unitary, ∥ v i ∥ = 1 \lVert v_i\rVert=1 ∥ v i ∥ = 1 . If i ≠ j i\ne j i = j , then
⟨ v i , v j ⟩ = ⟨ π ( h i g h i − 1 ) ξ , π ( h j g h j − 1 ) ξ ⟩ = ⟨ π ( h j g − 1 h j − 1 h i g h i − 1 ) ξ , ξ ⟩ = ϕ ( h j g − 1 h j − 1 h i g h i − 1 ) = 0 , \begin{aligned}
\langle v_i,v_j\rangle
&=\langle\pi(h_igh_i^{-1})\xi,\pi(h_jgh_j^{-1})\xi\rangle\\
&=\langle\pi(h_jg^{-1}h_j^{-1}h_igh_i^{-1})\xi,\xi\rangle\\
&=\phi(h_jg^{-1}h_j^{-1}h_igh_i^{-1})=0,
\end{aligned} ⟨ v i , v j ⟩ = ⟨ π ( h i g h i − 1 ) ξ , π ( h j g h j − 1 ) ξ ⟩ = ⟨ π ( h j g − 1 h j − 1 h i g h i − 1 ) ξ , ξ ⟩ = ϕ ( h j g − 1 h j − 1 h i g h i − 1 ) = 0 ,
since the element in the last line belongs to K K K . Thus ( v i ) (v_i) ( v i ) is an orthonormal sequence. For every r ≥ 1 r \ge1 r ≥ 1 , Bessel’s inequality now gives ∑ i = 1 r ∣ ϕ ( h i g h i − 1 ) ∣ 2 = ∑ i = 1 r ∣ ⟨ v i , ξ ⟩ ∣ 2 ≤ ∥ ξ ∥ 2 = 1 \sum_{i=1}^{r} |\phi(h_i g h_i^{-1})|^2 = \sum_{i=1}^{r} |\langle v_i,\xi\rangle|^2 \le\|\xi\|^2 = 1 ∑ i = 1 r ∣ ϕ ( h i g h i − 1 ) ∣ 2 = ∑ i = 1 r ∣ ⟨ v i , ξ ⟩ ∣ 2 ≤ ∥ ξ ∥ 2 = 1 . If ϕ ( g ) ≠ 0 \phi(g) \ne0 ϕ ( g ) = 0 , choose i 0 i_0 i 0 such that ∣ ϕ ( h i g h i − 1 ) ∣ ≥ ∣ ϕ ( g ) ∣ / 2 |\phi(h_i g h_i^{-1})| \ge|\phi(g)|/2 ∣ ϕ ( h i g h i − 1 ) ∣ ≥ ∣ ϕ ( g ) ∣/2 for every i ≥ i 0 i \ge i_0 i ≥ i 0 . For r ≥ i 0 r \ge i_0 r ≥ i 0 , the preceding bound would imply ( r − i 0 + 1 ) ∣ ϕ ( g ) ∣ 2 / 4 ≤ 1 (r-i_0+1)|\phi(g)|^2/4 \le1 ( r − i 0 + 1 ) ∣ ϕ ( g ) ∣ 2 /4 ≤ 1 , which is impossible as r → ∞ r \to\infty r → ∞ . Therefore ϕ ( g ) = 0 \phi(g) = 0 ϕ ( g ) = 0 . Since g ≠ e g \ne e g = e was arbitrary and ϕ ( e ) = 1 \phi(e) = 1 ϕ ( e ) = 1 , we obtain ϕ = δ e \phi= \delta_e ϕ = δ e .
K K K -Split higher rank lattices
We will be interested in the case where Γ \Gamma Γ is a split, higher rank S S S -arithmetic group. To set the notation let K K K be a number field with ring of integers O \mathcal{O} O and S S S a finite set of places of K K K containing all of the Archimedean ones. Let G \mathbf{G} G be a connected absolutely simple adjoint algebraic group that is defined and split over K K K . Assume that G \mathbf{G} G is of higher rank in the sense that ∑ v ∈ S rk K v ( G ) ≥ 2 \sum_{v\in S}\operatorname{rk}_{K_v}(\mathbf{G}) \ge2 ∑ v ∈ S rk K v ( G ) ≥ 2 . We fix a faithful K K K -representation G → G L n \mathbf{G} \to\mathrm{GL}_n G → GL n and hence identify G \mathbf{G} G with its image under this representation. Let Γ < G ( K ) \Gamma< \mathbf{G}(K) Γ < G ( K ) be any subgroup that is commensurable to G ( O S ) = G ( K ) ∩ G L n ( O S ) \mathbf{G}(\mathcal{O}_S) = \mathbf{G}(K) \cap\mathrm{GL}_n(\mathcal{O}_S) G ( O S ) = G ( K ) ∩ GL n ( O S ) .
To fix notation, let B \mathbf{B} B be a Borel subgroup, T < B \mathbf{T} < \mathbf{B} T < B a maximal split torus and U \mathbf{U} U the unipotent radical of B \mathbf{B} B . We denote by Φ \Phi Φ the corresponding choice of root system, by Ψ ⊂ Φ \Psi\subset\Phi Ψ ⊂ Φ the positive basis corresponding to the chosen Borel subgroup and by θ ∈ Φ \theta\in\Phi θ ∈ Φ the highest root with respect to this basis. Let x α : G a → U α x_\alpha:\mathbf{G}_a \to U_\alpha x α : G a → U α be root parametrizations. In particular we obtain a map ι : S L 2 → G \iota:\mathrm{SL}_2 \to\mathbf{G} ι : SL 2 → G , ι ( ( 1 t 0 1 ) ) = x θ ( t ) \iota\left(\begin{pmatrix}1&t\\0&1\end{pmatrix}\right)=x_\theta(t) ι ( ( 1 0 t 1 ) ) = x θ ( t ) , ι ( ( 1 0 t 1 ) ) = x − θ ( t ) \iota\left(\begin{pmatrix}1&0\\t&1\end{pmatrix}\right)=x_{-\theta}(t) ι ( ( 1 t 0 1 ) ) = x − θ ( t ) . We denote the image of ι \iota ι by G θ \mathbf{G}_\theta G θ . The kernel of ι \iota ι is contained in { ± I } \{\pm I\} { ± I } , so G θ \mathbf{G}_\theta G θ is isomorphic to S L 2 \mathrm{SL}_2 SL 2 or P G L 2 \mathrm{PGL}_2 PGL 2 . A main feature which considerably simplifies our proof is that we often work inside this rank one group, rather than in the whole group. We denote Γ θ = Γ ∩ G θ ( K ) \Gamma_\theta= \Gamma\cap\mathbf{G}_\theta(K) Γ θ = Γ ∩ G θ ( K ) , Υ θ = Γ ∩ U θ ( K ) \Upsilon_\theta= \Gamma\cap U_\theta(K) Υ θ = Γ ∩ U θ ( K ) .
For better clarity, we will, throughout, exemplify our main proof by following this special case. Here K = Q K = \mathbb{Q} K = Q , O = Z \mathcal{O} = \mathbb{Z} O = Z , S = { ∞ } S = \{\infty\} S = { ∞ } , G = P G L n \mathbf{G} = \mathrm{PGL}_n G = PGL n . The higher rank assumption here means that n ≥ 3 n \ge3 n ≥ 3 . The Borel group B \mathbf{B} B is the image of the upper triangular matrices, T \mathbf{T} T is the image of the diagonal matrices, and U \mathbf{U} U is the image of the unipotent upper triangular matrices. The roots are α i , j ( diag ( t 1 , … , t n ) ) = t i / t j \alpha_{i,j}(\operatorname{diag}(t_1,\ldots,t_n)) = t_i/t_j α i , j ( diag ( t 1 , … , t n )) = t i / t j for i ≠ j i \ne j i = j . The maps x i , j ( t ) x_{i,j}(t) x i , j ( t ) give the elementary unipotent matrices with 1 1 1 along the diagonal and t t t in the ( i , j ) (i,j) ( i , j ) -entry. The highest root here is θ = α 1 , n \theta= \alpha_{1,n} θ = α 1 , n and − θ = α n , 1 -\theta= \alpha_{n,1} − θ = α n , 1 . Here Γ \Gamma Γ is any group commensurable to the arithmetic lattice P S L n ( Z ) \mathrm{PSL}_n(\mathbb{Z}) PSL n ( Z ) .
When referring to this example we use the overline notation A ‾ \overline{A} A to denote the image of the matrix A ∈ S L n ( Z ) A \in\mathrm{SL}_n(\mathbb{Z}) A ∈ SL n ( Z ) modulo the center of S L n ( Z ) \mathrm{SL}_n(\mathbb{Z}) SL n ( Z ) . Similarly for a nonzero vector x ∈ R n ∖ { 0 } x \in\mathbb{R}^n \setminus\{0\} x ∈ R n ∖ { 0 } we denote by x ‾ \overline{x} x its image in the projective space P R n \mathbb{P}\mathbb{R}^n P R n .
Our proof relies heavily on strong rigidity results for these groups, which give a complete classification of their ergodic IRSs and extremal characters. We now recall these results.
Theorem 2.4 (Nevo–Stuck–Zimmer [53 , 50 ] ; [24 ] , Corollary 1.4 and the paragraph following it). *Let Γ \Gamma Γ be as in Subsection 2.5. Then every ergodic IRS on Sub ( Γ ) \operatorname{Sub}(\Gamma) Sub ( Γ )
is either ν = δ { e } \nu=\delta_{\{e\}} ν = δ { e } or ν = ν Σ \nu=\nu_{\Sigma} ν = ν Σ for some finite-index subgroup Σ < Γ \Sigma<\Gamma Σ < Γ , where ν Σ = 1 [ Γ : N Γ ( Σ ) ] ∑ g N Γ ( Σ ) ∈ Γ / N Γ ( Σ ) δ g Σ g − 1 \nu_{\Sigma}=\frac{1}{[\Gamma:N_{\Gamma}(\Sigma)]}\sum_{gN_{\Gamma}(\Sigma)\in\Gamma/N_{\Gamma}(\Sigma)}\delta_{g\Sigma g^{-1}} ν Σ = [ Γ : N Γ ( Σ )] 1 ∑ g N Γ ( Σ ) ∈ Γ/ N Γ ( Σ ) δ g Σ g − 1 .
Closely related is the following character rigidity theorem which gives a complete classification of all extremal characters. The theorem was first proved for PSL n ( Z ) \operatorname{PSL}_{n}(\mathbb{Z}) PSL n ( Z ) by Bekka in [16 ] and later generalized in [52 ] , [15 ] , [24 ] .
Theorem 2.5 ([24 Theorem 1.3 and Remark 6.6] , [15 Proposition 7.1] ). Let Γ \Gamma Γ be as in Subsection 2.5 and χ \chi χ be an extremal character of Γ \Gamma Γ . Then either χ = δ e \chi=\delta_e χ = δ e or χ \chi χ is the normalized character of an irreducible finite-dimensional representation. Moreover there are only countably many such unitary representations and they all factor via finite quotients of Γ \Gamma Γ .
For the next proposition, Γ \Gamma Γ is a K K K -split higher rank lattice as in Subsection 2.5. It relates the vanishing of the conditional expectations to the associated character. It essentially says that if an IRA is supported on non-trivial subalgebras, then the associated conditional expectation cannot vanish on a group element and all its powers.
Proposition 2.6. Let μ \mu μ be an IRA on SA ( L ( Γ ) ) \operatorname{SA}(L(\Gamma)) SA ( L ( Γ )) and let s ∈ Γ s\in\Gamma s ∈ Γ have infinite order. Assume that for μ \mu μ -a.e. M M M one has E M ( λ ( s k ) ) = 0 E_M(\lambda(s^k))=0 E M ( λ ( s k )) = 0 for all k ≠ 0 k\ne0 k = 0 . Then μ = δ C 1 \mu=\delta_{\mathbb{C}1} μ = δ C 1 , i.e. M = C 1 M=\mathbb{C}1 M = C 1 for μ \mu μ -a.e. M M M .
Proof. For g ∈ Γ g\in\Gamma g ∈ Γ set φ ( g ) = ∫ SA ( L ( Γ ) ) ∥ E M ( λ ( g ) ) ∥ 2 2 d μ ( M ) \varphi(g)=\int_{\operatorname{SA}(L(\Gamma))}\|E_M(\lambda(g))\|_2^2\,d\mu(M) φ ( g ) = ∫ SA ( L ( Γ )) ∥ E M ( λ ( g )) ∥ 2 2 d μ ( M ) . Then φ \varphi φ is a character of Γ \Gamma Γ (see Subsection 2.2, also see [39 Lemma 2.2] ). By Theorem 2.5 (since characters are traces on C ∗ ( Γ ) C^*(\Gamma) C ∗ ( Γ ) , they form a Choquet simplex, whose extreme points are countable) we may write φ = c 0 δ e + ∑ i ≥ 1 c i χ i \varphi=c_0\delta_e+\sum_{i\geq1}c_i\chi_i φ = c 0 δ e + ∑ i ≥ 1 c i χ i with c i ≥ 0 c_i\geq0 c i ≥ 0 , ∑ i ≥ 0 c i = 1 \sum_{i\geq0}c_i=1 ∑ i ≥ 0 c i = 1 , where { δ e , χ i : i ≥ 1 } \{\delta_e,\chi_i:i\geq1\} { δ e , χ i : i ≥ 1 } lists the countably many extremal characters of Γ \Gamma Γ ; in particular each χ i \chi_i χ i arises from an irreducible finite-dimensional representation π i \pi_i π i of a finite quotient of Γ \Gamma Γ .
Set H = ⟨ s ⟩ ≅ Z H=\langle s\rangle\cong\mathbb{Z} H = ⟨ s ⟩ ≅ Z . By the Herglotz–Bochner theorem, φ ∣ H ( s m ) = ∫ T z m d σ ( z ) \varphi|_H(s^m)=\int_{\mathbb{T}}z^m\,d\sigma(z) φ ∣ H ( s m ) = ∫ T z m d σ ( z ) for a probability measure σ \sigma σ on T \mathbb{T} T . Since E M ( λ ( s k ) ) = 0 E_M(\lambda(s^k))=0 E M ( λ ( s k )) = 0 for all k ≠ 0 k\ne0 k = 0 and μ \mu μ -a.e. M M M , we get φ ( s k ) = 0 \varphi(s^k)=0 φ ( s k ) = 0 for all k ≠ 0 k\ne0 k = 0 , so all nonzero Fourier coefficients of σ \sigma σ vanish and σ = m T \sigma=m_{\mathbb{T}} σ = m T , the Haar measure. Likewise δ e ∣ H \delta_e|_H δ e ∣ H has Herglotz measure m T m_{\mathbb{T}} m T , and writing χ i ( s m ) = ∫ T z m d σ i ( z ) \chi_i(s^m)=\int_{\mathbb{T}}z^m\,d\sigma_i(z) χ i ( s m ) = ∫ T z m d σ i ( z ) , uniqueness of Herglotz measures gives m T = c 0 m T + ∑ i ≥ 1 c i σ i m_{\mathbb{T}}=c_0m_{\mathbb{T}}+\sum_{i\geq1}c_i\sigma_i m T = c 0 m T + ∑ i ≥ 1 c i σ i on T \mathbb{T} T .
Fix i ≥ 1 i\geq1 i ≥ 1 . The unitary π i ( s ˉ ) \pi_i(\bar{s}) π i ( s ˉ ) has finite order, hence is diagonalizable with eigenvalues that are roots of unity, say ω 1 , … , ω d i \omega_1,\ldots,\omega_{d_i} ω 1 , … , ω d i ; then χ i ( s m ) = 1 d i ∑ j ω j m \chi_i(s^m)=\frac{1}{d_i}\sum_j\omega_j^m χ i ( s m ) = d i 1 ∑ j ω j m , so σ i = 1 d i ∑ j δ ω j \sigma_i=\frac{1}{d_i}\sum_j\delta_{\omega_j} σ i = d i 1 ∑ j δ ω j is atomic, supported on roots of unity. If c 0 < 1 c_0<1 c 0 < 1 , then ( 1 − c 0 ) m T = ∑ i c i σ i (1-c_0)m_{\mathbb{T}}=\sum_i c_i\sigma_i ( 1 − c 0 ) m T = ∑ i c i σ i would be a nonzero atomless measure equal to a countable sum of atomic measures, which is impossible. Hence c 0 = 1 c_0=1 c 0 = 1 and φ = δ e \varphi=\delta_e φ = δ e , i.e. for every g ≠ e g\ne e g = e , E M ( λ ( g ) ) = 0 E_M(\lambda(g))=0 E M ( λ ( g )) = 0 for μ \mu μ -a.e. M M M ; discarding countably many null sets, for μ \mu μ -a.e. M M M we have E M ( λ ( g ) ) = 0 E_M(\lambda(g))=0 E M ( λ ( g )) = 0 for all g ≠ e g\ne e g = e . For such M M M and any x ∈ M x\in M x ∈ M , the Fourier expansion gives x = E M ( x ) = τ ( x ) 1 x=E_M(x)=\tau(x)1 x = E M ( x ) = τ ( x ) 1 , so M = C 1 M=\mathbb{C}1 M = C 1 . Hence μ = δ C 1 \mu=\delta_{\mathbb{C}1} μ = δ C 1 .
We would like to remark that the above proposition is a higher rank phenomenon. For example, if we consider Γ = F 2 = ⟨ a , b ⟩ \Gamma=\mathbb{F}_2=\langle a,b\rangle Γ = F 2 = ⟨ a , b ⟩ , and N = ker ( Γ → Z ) N=\ker(\Gamma\to\mathbb{Z}) N = ker ( Γ → Z ) , where a ↦ 1 a\mapsto1 a ↦ 1 and b ↦ 0 b\mapsto0 b ↦ 0 . Then N N N is nontrivial and normal, but a k ∉ N a^k\notin N a k ∈ / N for every k ≠ 0 k\ne0 k = 0 . Consequently, E L ( N ) ( λ ( a k ) ) = 0 E_{L(N)}(\lambda(a^k))=0 E L ( N ) ( λ ( a k )) = 0 for every k ≠ 0 k\ne0 k = 0 , while μ = δ L ( N ) \mu=\delta_{L(N)} μ = δ L ( N ) is an IRA different from δ C 1 \delta_{\mathbb{C}1} δ C 1 .
Acylindrically hyperbolic groups
We recall the terminology used in [3 Subsection 2.1] (which was taken from [51 , 22 ] ). An isometric action Γ ↷ ( S , d ) \Gamma\curvearrowright(S,d) Γ ↷ ( S , d ) is acylindrical if, for every ϵ > 0 \epsilon> 0 ϵ > 0 , there are R , N > 0 R,N > 0 R , N > 0 such that ∣ { g ∈ Γ : d ( x , g x ) ≤ ϵ , d ( y , g y ) ≤ ϵ } ∣ ≤ N \left|\{g \in\Gamma: d(x,gx) \le\epsilon, d(y,gy) \le\epsilon\}\right| \le N ∣ { g ∈ Γ : d ( x , g x ) ≤ ϵ , d ( y , g y ) ≤ ϵ } ∣ ≤ N whenever d ( x , y ) ≥ R d(x,y) \ge R d ( x , y ) ≥ R . For an action on a hyperbolic space, s ∈ Γ s \in\Gamma s ∈ Γ is loxodromic if n ↦ s n x n \mapsto s^{n}x n ↦ s n x is a quasi-isometric embedding of Z \mathbb{Z} Z into S S S for some x ∈ S x \in S x ∈ S . Such an element has two fixed points x s + , x s − ∈ ∂ S x_s^{+},x_s^{-} \in\partial S x s + , x s − ∈ ∂ S . Two loxodromic elements are independent if their fixed-point pairs are disjoint. The action is non-elementary if it contains two independent loxodromic elements. A group is acylindrically hyperbolic if it admits a non-elementary acylindrical action on a hyperbolic space.
An element s s s satisfies weak proper discontinuity (WPD) if, for every x ∈ S x \in S x ∈ S and ϵ > 0 \epsilon> 0 ϵ > 0 , there is m ≥ 1 m \ge1 m ≥ 1 such that { g ∈ Γ : d ( x , g x ) ≤ ϵ , d ( s m x , g s m x ) ≤ ϵ } \{g \in\Gamma: d(x,gx) \le\epsilon, d(s^{m}x,gs^{m}x) \le\epsilon\} { g ∈ Γ : d ( x , g x ) ≤ ϵ , d ( s m x , g s m x ) ≤ ϵ } is finite. Every loxodromic element of an acylindrical action satisfies WPD. An element is generalized loxodromic if it is loxodromic for some acylindrical action of Γ \Gamma Γ on a hyperbolic space. Equivalently, it is loxodromic WPD for some action on a hyperbolic space [51 Theorem 1.4] . We denote the set of these elements by K l o x K_{\mathrm{lox}} K lox .
For s ∈ K l o x s \in K_{\mathrm{lox}} s ∈ K lox , there is a unique maximal virtually cyclic subgroup E Γ ( s ) E_{\Gamma}(s) E Γ ( s ) containing s s s , and E Γ ( s ) = { g ∈ Γ : g s m g − 1 = s ± m E_{\Gamma}(s) = \{g \in\Gamma: gs^{m}g^{-1} = s^{\pm m} E Γ ( s ) = { g ∈ Γ : g s m g − 1 = s ± m for some m ≥ 1 } m \ge1\} m ≥ 1 } . It is the stabilizer of { x s + , x s − } \{x_s^{+},x_s^{-}\} { x s + , x s − } in any action for which s s s is loxodromic WPD, and it is hyperbolically embedded in Γ \Gamma Γ ; see [22 Lemma 6.5, Corollaries 2.9 and 6.6] . In particular, C Γ ( s m ) ⊆ E Γ ( s ) C_{\Gamma}(s^{m}) \subseteq E_{\Gamma}(s) C Γ ( s m ) ⊆ E Γ ( s ) for every m ≠ 0 m \ne0 m = 0 .
For a fixed non-elementary acylindrical action Γ ↷ S \Gamma\curvearrowright S Γ ↷ S , the intersection ⋂ s ∈ K l o x E Γ ( s ) \bigcap_{s \in K_{\mathrm{lox}}} E_{\Gamma}(s) ⋂ s ∈ K lox E Γ ( s ) is the maximal finite normal subgroup of Γ \Gamma Γ (see [14 Lemma 5.6] ). Thus this intersection is trivial whenever Γ \Gamma Γ has trivial amenable radical.
Boomerang algebras, properties and associated characters
We begin with basic properties of boomerang subalgebras and the positive-definite functions associated with them. We then study finite dimensional and amenable boomerang subalgebras and give conditions under which they are trivial.
Boomerang subalgebras and their basic properties
We now formally define the notion of a boomerang subalgebra.
Definition 3.1 (Boomerang subalgebra). A von Neumann subalgebra A ≤ L ( Γ ) A \le L(\Gamma) A ≤ L ( Γ ) is called boomerang if for every s ∈ Γ s \in\Gamma s ∈ Γ there is a strictly increasing sequence ( n i ) ⊆ N (n_i) \subseteq\mathbb{N} ( n i ) ⊆ N such that λ ( s n i ) A λ ( s n i ) ∗ → A \lambda(s^{n_i})A\lambda(s^{n_i})^{*} \to A λ ( s n i ) A λ ( s n i ) ∗ → A in the Effros–Maréchal topology.
If s s s has finite order m m m , taking n i = i m n_i = im n i = im gives λ ( s n i ) A λ ( s − n i ) = A \lambda(s^{n_i})A\lambda(s^{-n_i}) = A λ ( s n i ) A λ ( s − n i ) = A for every i i i . Thus the condition in Definition 3.1 is automatic for finite-order elements. Also, note that every invariant subalgebra is boomerang.
We first relate boomerang subgroups to boomerang subalgebras through the maps L L L and J J J as defined in the introduction.
Proposition 3.2. The maps
Sub ( Γ ) → L SA ( L ( Γ ) ) SA ( L ( Γ ) ) → J Sub ( Γ ) Δ ↦ L ( Δ ) M ↦ Γ M : = { γ : λ ( γ ) ∈ M } \begin{aligned}
\operatorname{Sub}(\Gamma) &\xrightarrow{L} \operatorname{SA}(L(\Gamma)) &
\operatorname{SA}(L(\Gamma)) &\xrightarrow{J} \operatorname{Sub}(\Gamma) \\
\Delta&\mapsto L(\Delta) &
M &\mapsto\Gamma_M := \{\gamma: \lambda(\gamma) \in M\}
\end{aligned} Sub ( Γ ) Δ L SA ( L ( Γ )) ↦ L ( Δ ) SA ( L ( Γ )) M J Sub ( Γ ) ↦ Γ M := { γ : λ ( γ ) ∈ M } are measurable and Γ \Gamma Γ -equivariant, and satisfy J L ( Δ ) = Δ JL(\Delta)=\Delta J L ( Δ ) = Δ and L J ( M ) ≤ M LJ(M)\le M L J ( M ) ≤ M . Moreover, L L L is a topological embedding.
Proof. J J J is Borel, since { M : λ ( γ ) ∈ M } = { M : ∥ E M ( λ ( γ ) ) ∥ 2 = 1 } \{M:\lambda(\gamma)\in M\}=\{M:\lVert E_M(\lambda(\gamma))\rVert_2=1\} { M : λ ( γ ) ∈ M } = { M : ∥ E M ( λ ( γ )) ∥ 2 = 1 } and M ↦ E M ( x ) M\mapsto E_M(x) M ↦ E M ( x ) is Borel for the Effros–Maréchal structure [34 ] . It is Γ \Gamma Γ -equivariant since Γ λ ( γ ) M λ ( γ ) ∗ = γ Γ M γ − 1 \Gamma\lambda(\gamma)M\lambda(\gamma)^*=\gamma\Gamma_M\gamma^{-1} Γ λ ( γ ) M λ ( γ ) ∗ = γ Γ M γ − 1 . The map L L L is even continuous (see [8 ] , Proposition 4.1). Equivariance was also established in the same proposition. The identities J L ( Δ ) = Δ JL(\Delta)=\Delta J L ( Δ ) = Δ and L J ( M ) ≤ M LJ(M)\le M L J ( M ) ≤ M follow from the definitions. □
The topological embedding in Proposition 3.2 identifies boomerang subgroups with boomerang group von Neumann algebras.
Proposition 3.3. For a subgroup H ≤ Γ H\le\Gamma H ≤ Γ , the algebra L ( H ) L(H) L ( H ) is boomerang if and only if H H H is a boomerang subgroup.
Proof. Apply the equivariant topological embedding of Proposition 3.2 to the conjugates of H H H by powers of each s ∈ Γ s\in\Gamma s ∈ Γ . □
The next example contrasts with the subgroup case. Every finite-index subgroup is boomerang, since it has only finitely many conjugates. However, a finite-index subalgebra need not be boomerang.
Example 3.4. There exist finite-index subalgebras which are not boomerang.
Proof. Let F 2 = ⟨ a , b ⟩ \mathbb{F}_2=\langle a,b\rangle F 2 = ⟨ a , b ⟩ , and let σ \sigma σ be the automorphism interchanging a a a and b b b . We also write σ \sigma σ for the induced automorphism of M = L ( F 2 ) M=L(\mathbb{F}_2) M = L ( F 2 ) . Set N = M σ N=M^\sigma N = M σ . The automorphism σ \sigma σ is outer. Indeed, if σ = Ad ( v ) \sigma=\operatorname{Ad}(v) σ = Ad ( v ) for a unitary v ∈ M v\in M v ∈ M , then λ ( b − 1 ) v λ ( a ) = v \lambda(b^{-1})v\lambda(a)=v λ ( b − 1 ) v λ ( a ) = v . The Fourier coefficients of v v v would therefore be constant on every set { b − m h a m : m ∈ Z } \{b^{-m}ha^m:m\in\mathbb{Z}\} { b − m h a m : m ∈ Z } . For fixed h h h , these elements are pairwise distinct: their images in the abelianization Z 2 \mathbb{Z}^2 Z 2 differ by ( m , − m ) (m,-m) ( m , − m ) . Square summability would force all the Fourier coefficients of v v v to be zero, a contradiction. Thus the fixed-point inclusion for the outer order-two action ⟨ σ ⟩ \langle\sigma\rangle ⟨ σ ⟩ has [ M : N ] = 2 [M:N]=2 [ M : N ] = 2 .
We now show that N N N is not boomerang. Suppose otherwise. Applying the definition to b b b , choose a strictly increasing sequence ( n i ) ⊆ N (n_i)\subseteq\mathbb{N} ( n i ) ⊆ N such that N i = λ ( b n i ) N λ ( b − n i ) → N N_i=\lambda(b^{n_i})N\lambda(b^{-n_i})\to N N i = λ ( b n i ) N λ ( b − n i ) → N in the Effros–Maréchal topology. Since E N = ( id + σ ) / 2 E_N=(\operatorname{id}+\sigma)/2 E N = ( id + σ ) /2 , Lemma 2.1 gives E N i ( λ ( a ) ) = λ ( b n i ) E N ( λ ( b − n i a b n i ) ) λ ( b − n i ) = 1 2 ( λ ( a ) + λ ( b n i a − n i b a n i b − n i ) ) E_{N_i}(\lambda(a))=\lambda(b^{n_i})E_N(\lambda(b^{-n_i}ab^{n_i}))\lambda(b^{-n_i})=\frac{1}{2}\left(\lambda(a)+\lambda(b^{n_i}a^{-n_i}ba^{n_i}b^{-n_i})\right) E N i ( λ ( a )) = λ ( b n i ) E N ( λ ( b − n i a b n i )) λ ( b − n i ) = 2 1 ( λ ( a ) + λ ( b n i a − n i b a n i b − n i ) ) . For every n i ≥ 1 n_i\ge1 n i ≥ 1 , the reduced word b n i a − n i b a n i b − n i b^{n_i}a^{-n_i}ba^{n_i}b^{-n_i} b n i a − n i b a n i b − n i is different from b b b . Since distinct group unitaries are orthogonal in L 2 ( M , τ 0 ) L^2(M,\tau_0) L 2 ( M , τ 0 ) , ∥ E N i ( λ ( a ) ) − E N ( λ ( a ) ) ∥ 2 2 = 1 4 ∥ λ ( b n i a − n i b a n i b − n i ) − λ ( b ) ∥ 2 2 = 1 2 \lVert E_{N_i}(\lambda(a))-E_N(\lambda(a))\rVert_2^2=\frac{1}{4}\lVert\lambda(b^{n_i}a^{-n_i}ba^{n_i}b^{-n_i})-\lambda(b)\rVert_2^2=\frac{1}{2} ∥ E N i ( λ ( a )) − E N ( λ ( a )) ∥ 2 2 = 4 1 ∥ λ ( b n i a − n i b a n i b − n i ) − λ ( b ) ∥ 2 2 = 2 1 . This contradicts the pointwise ∥ ⋅ ∥ 2 \lVert\cdot\rVert_2 ∥ ⋅ ∥ 2 convergence of the conditional expectations, which follows from N i → N N_i\to N N i → N . □
We next study the Fourier support of E M ( λ ( s ) ) E_M(\lambda(s)) E M ( λ ( s )) when M M M is boomerang. Towards this end, we introduce a combinatorial set which locates the Fourier support of the image of a group element under the conditional expectation onto a boomerang subalgebra. For a countable group Γ \Gamma Γ and γ ∈ Γ \gamma\in\Gamma γ ∈ Γ , set R ( γ ) : = ⋂ t ∈ C Γ ( γ ) ⋃ k ≥ 1 C Γ ( t k ) R(\gamma):=\bigcap_{t\in C_\Gamma(\gamma)}\bigcup_{k\geq1}C_\Gamma(t^k) R ( γ ) := ⋂ t ∈ C Γ ( γ ) ⋃ k ≥ 1 C Γ ( t k ) . The following proposition shows that the support of E M ( λ ( s ) ) E_M(\lambda(s)) E M ( λ ( s )) is contained in R ( s ) R(s) R ( s ) .
Proposition 3.5. Let Γ \Gamma Γ be any countable discrete group and M ⊂ L ( Γ ) M\subset L(\Gamma) M ⊂ L ( Γ ) a boomerang von Neumann subalgebra. Then supp ( E M ( λ ( s ) ) ) ⊆ R ( s ) \operatorname{supp}(E_M(\lambda(s)))\subseteq R(s) supp ( E M ( λ ( s ))) ⊆ R ( s ) for every s ≠ e s\ne e s = e .
Proof. Write E M ( λ ( s ) ) = ∑ h ∈ Γ c h λ ( h ) E_M(\lambda(s))=\sum_{h\in\Gamma}c_h\lambda(h) E M ( λ ( s )) = ∑ h ∈ Γ c h λ ( h ) for its Fourier expansion. Fix g ∉ R ( s ) g\notin R(s) g ∈ / R ( s ) ; our goal is to show c g = 0 c_g=0 c g = 0 . We may assume g ≠ e g\ne e g = e , since c e = τ ( E M ( λ ( s ) ) ) = τ ( λ ( s ) ) = 0 c_e=\tau(E_M(\lambda(s)))=\tau(\lambda(s))=0 c e = τ ( E M ( λ ( s ))) = τ ( λ ( s )) = 0 . By the definition of R ( s ) R(s) R ( s ) there is t ∈ C Γ ( s ) t\in C_\Gamma(s) t ∈ C Γ ( s ) with g ∉ ⋃ k ≥ 1 C Γ ( t k ) g\notin\bigcup_{k\geq1}C_\Gamma(t^k) g ∈ / ⋃ k ≥ 1 C Γ ( t k ) , i.e. t k g t − k ≠ g t^kgt^{-k}\ne g t k g t − k = g for all k ≥ 1 k\geq1 k ≥ 1 ; note t ≠ e t\ne e t = e , since g ∈ Γ = ⋃ k C Γ ( e k ) g\in\Gamma=\bigcup_k C_\Gamma(e^k) g ∈ Γ = ⋃ k C Γ ( e k ) .
Since M M M is boomerang, there is n i → + ∞ n_i\to+\infty n i → + ∞ with λ ( t − n i ) M λ ( t n i ) → M \lambda(t^{-n_i})M\lambda(t^{n_i})\to M λ ( t − n i ) M λ ( t n i ) → M in the Effros–Maréchal topology; in the tracial setting this implies ∥ E λ ( t − n i ) M λ ( t n i ) ( x ) − E M ( x ) ∥ 2 → 0 \|E_{\lambda(t^{-n_i})M\lambda(t^{n_i})}(x)-E_M(x)\|_2\to0 ∥ E λ ( t − n i ) M λ ( t n i ) ( x ) − E M ( x ) ∥ 2 → 0 for every x x x (one of the standard reformulations of Effros–Maréchal convergence, see [34 ] ). Using E u M u ∗ ( x ) = u E M ( u ∗ x u ) u ∗ E_{uMu^*}(x)=uE_M(u^*xu)u^* E u M u ∗ ( x ) = u E M ( u ∗ xu ) u ∗ and t ∈ C Γ ( s ) t\in C_\Gamma(s) t ∈ C Γ ( s ) ,
∥ E λ ( t − n i ) M λ ( t n i ) ( λ ( s ) ) − E M ( λ ( s ) ) ∥ 2 2 = ∥ E M ( λ ( s ) ) − λ ( t n i ) E M ( λ ( s ) ) λ ( t − n i ) ∥ 2 2 = ∑ h ∈ Γ ∣ c t − n i h t n i − c h ∣ 2 ⟶ 0. \begin{aligned}
\left\|E_{\lambda(t^{-n_i})M\lambda(t^{n_i})}(\lambda(s))-E_M(\lambda(s))\right\|_2^2
&=\left\|E_M(\lambda(s))-\lambda(t^{n_i})E_M(\lambda(s))\lambda(t^{-n_i})\right\|_2^2\\
&=\sum_{h\in\Gamma}\left|c_{t^{-n_i}ht^{n_i}}-c_h\right|^2\longrightarrow0.
\end{aligned} E λ ( t − n i ) M λ ( t n i ) ( λ ( s )) − E M ( λ ( s )) 2 2 = E M ( λ ( s )) − λ ( t n i ) E M ( λ ( s )) λ ( t − n i ) 2 2 = h ∈ Γ ∑ ∣ c t − n i h t n i − c h ∣ 2 ⟶ 0.
In particular ∣ c t n i g t − n i − c g ∣ → 0 |c_{t^{n_i}gt^{-n_i}}-c_g|\to0 ∣ c t n i g t − n i − c g ∣ → 0 . The elements t n i g t − n i t^{n_i}gt^{-n_i} t n i g t − n i are pairwise distinct (if t a g t − a = t b g t − b t^agt^{-a}=t^bgt^{-b} t a g t − a = t b g t − b with a < b a<b a < b , then t b − a g t − ( b − a ) = g t^{b-a}gt^{-(b-a)}=g t b − a g t − ( b − a ) = g ), so t n i g t − n i → ∞ t^{n_i}gt^{-n_i}\to\infty t n i g t − n i → ∞ in Γ \Gamma Γ and c t n i g t − n i → 0 c_{t^{n_i}gt^{-n_i}}\to0 c t n i g t − n i → 0 , since ( c h ) h ∈ Γ ∈ ℓ 2 ( Γ ) (c_h)_{h\in\Gamma}\in\ell^2(\Gamma) ( c h ) h ∈ Γ ∈ ℓ 2 ( Γ ) . Hence c g = 0 c_g=0 c g = 0 .
Since s ∈ C Γ ( s ) s\in C_\Gamma(s) s ∈ C Γ ( s ) , taking t = s t=s t = s in the intersection defining R ( s ) R(s) R ( s ) and applying Proposition 3.5 gives supp ( E M ( λ ( s ) ) ) ⊆ R ( s ) ⊆ ⋃ k ≥ 1 C Γ ( s k ) \operatorname{supp}(E_M(\lambda(s)))\subseteq R(s)\subseteq\bigcup_{k\geq1}C_\Gamma(s^k) supp ( E M ( λ ( s ))) ⊆ R ( s ) ⊆ ⋃ k ≥ 1 C Γ ( s k ) . This observation will come in handy later.
We shall also use the following consequence of the boomerang property to preserve nonzero expectations under conjugation.
Lemma 3.6. Let M ≤ L ( Γ ) M\leq L(\Gamma) M ≤ L ( Γ ) be boomerang and suppose that E M ( λ ( t ) ) ≠ 0 E_M(\lambda(t))\ne0 E M ( λ ( t )) = 0 . For every s ∈ Γ s\in\Gamma s ∈ Γ , there are infinitely many n ≥ 1 n\geq1 n ≥ 1 such that E M ( λ ( s − n t s n ) ) ≠ 0 E_M(\lambda(s^{-n}ts^n))\ne0 E M ( λ ( s − n t s n )) = 0 .
Proof. Since M M M is boomerang, choose a strictly increasing sequence ( n i ) ⊆ N (n_i)\subseteq\mathbb{N} ( n i ) ⊆ N such that M i = λ ( s n i ) M λ ( s − n i ) → M M_i=\lambda(s^{n_i})M\lambda(s^{-n_i})\to M M i = λ ( s n i ) M λ ( s − n i ) → M in the Effros–Maréchal topology. By Lemma 2.1 , E M i ( λ ( t ) ) = λ ( s n i ) E M ( λ ( s − n i t s n i ) ) λ ( s − n i ) ⟶ E M ( λ ( t ) ) E_{M_i}(\lambda(t))=\lambda(s^{n_i})E_M(\lambda(s^{-n_i}ts^{n_i}))\lambda(s^{-n_i})\longrightarrow E_M(\lambda(t)) E M i ( λ ( t )) = λ ( s n i ) E M ( λ ( s − n i t s n i )) λ ( s − n i ) ⟶ E M ( λ ( t )) in ∥ ⋅ ∥ 2 \|\cdot\|_2 ∥ ⋅ ∥ 2 . Choose i 0 i_0 i 0 such that ∥ E M i ( λ ( t ) ) − E M ( λ ( t ) ) ∥ 2 < 1 2 ∥ E M ( λ ( t ) ) ∥ 2 \|E_{M_i}(\lambda(t))-E_M(\lambda(t))\|_2<\frac12\|E_M(\lambda(t))\|_2 ∥ E M i ( λ ( t )) − E M ( λ ( t )) ∥ 2 < 2 1 ∥ E M ( λ ( t )) ∥ 2 for every i ≥ i 0 i\geq i_0 i ≥ i 0 . Then ∥ E M i ( λ ( t ) ) ∥ 2 ≥ 1 2 ∥ E M ( λ ( t ) ) ∥ 2 > 0 \|E_{M_i}(\lambda(t))\|_2\geq\frac12\|E_M(\lambda(t))\|_2>0 ∥ E M i ( λ ( t )) ∥ 2 ≥ 2 1 ∥ E M ( λ ( t )) ∥ 2 > 0 , so E M ( λ ( s − n i t s n i ) ) ≠ 0 E_M(\lambda(s^{-n_i}ts^{n_i}))\ne0 E M ( λ ( s − n i t s n i )) = 0 for every i ≥ i 0 i\geq i_0 i ≥ i 0 .
Boomerang characters
We now consider the corresponding recurrence property for positive-definite functions. We show that the function associated with a boomerang subalgebra is boomerang, and that the converse fails.
Definition 3.7. A normalized positive-definite function ϕ \phi ϕ on Γ \Gamma Γ is called boomerang if for every s ∈ Γ s\in\Gamma s ∈ Γ there is a strictly increasing sequence ( n i ) ⊆ N (n_i)\subseteq\mathbb{N} ( n i ) ⊆ N such that ϕ ( s n i g s − n i ) → ϕ ( g ) \phi(s^{n_i}gs^{-n_i})\to\phi(g) ϕ ( s n i g s − n i ) → ϕ ( g ) for every g ∈ Γ g\in\Gamma g ∈ Γ . We will refer to these functions as Boomerang characters.
The assignment A ↦ ϕ A \mathcal{A} \mapsto\phi_{\mathcal{A}} A ↦ ϕ A preserves the boomerang property.
Lemma 3.8. If A \mathcal{A} A is a boomerang subalgebra, then ϕ A \phi_{\mathcal{A}} ϕ A is a boomerang positive-definite function.
Proof. Fix s ∈ Γ s \in\Gamma s ∈ Γ . Since A \mathcal{A} A is boomerang, choose a strictly increasing sequence ( n i ) ⊆ N (n_i) \subseteq\mathbb{N} ( n i ) ⊆ N such that A i = λ ( s − n i ) A λ ( s n i ) → A \mathcal{A}_i=\lambda(s^{-n_i})\mathcal{A}\lambda(s^{n_i}) \to\mathcal{A} A i = λ ( s − n i ) A λ ( s n i ) → A in the Effros–Maréchal topology. By Lemma 2.1 , for every g ∈ Γ g \in\Gamma g ∈ Γ , ϕ A ( s n i g s − n i ) = ∥ E A i ( λ ( g ) ) ∥ 2 2 ⟶ ∥ E A ( λ ( g ) ) ∥ 2 2 = ϕ A ( g ) \phi_{\mathcal{A}}(s^{n_i}gs^{-n_i})=\lVert E_{\mathcal{A}_i}(\lambda(g))\rVert_2^2 \longrightarrow\lVert E_{\mathcal{A}}(\lambda(g))\rVert_2^2=\phi_{\mathcal{A}}(g) ϕ A ( s n i g s − n i ) = ∥ E A i ( λ ( g )) ∥ 2 2 ⟶ ∥ E A ( λ ( g )) ∥ 2 2 = ϕ A ( g ) . □
The converse to Lemma 3.8 fails even for torsion-free hyperbolic groups.
Example 3.9. Let Γ = F 2 = ⟨ a , b ⟩ \Gamma=\mathbb{F}_2=\langle a,b\rangle Γ = F 2 = ⟨ a , b ⟩ , and let α ∈ Aut ( Γ ) \alpha\in\operatorname{Aut}(\Gamma) α ∈ Aut ( Γ ) be determined by α ( a ) = a − 1 \alpha(a)=a^{-1} α ( a ) = a − 1 and α ( b ) = b − 1 \alpha(b)=b^{-1} α ( b ) = b − 1 . We also denote by α \alpha α the induced trace-preserving automorphism of L ( Γ ) L(\Gamma) L ( Γ ) . Set M = L ( Γ ) α \mathcal{M}=L(\Gamma)^\alpha M = L ( Γ ) α . Since α 2 = id \alpha^2=\operatorname{id} α 2 = id , E M = ( id + α ) / 2 E_{\mathcal{M}}=(\operatorname{id}+\alpha)/2 E M = ( id + α ) /2 .
For every g ≠ e g\ne e g = e , we have α ( g ) ≠ g \alpha(g)\ne g α ( g ) = g . Hence, for g ≠ e g\ne e g = e , we have
ϕ M ( g ) = 1 2 τ 0 ( ( λ ( g ) + λ ( α ( g ) ) ) λ ( g ) ∗ ) = 1 2 . \phi_{\mathcal{M}}(g)=\frac{1}{2}\tau_0\bigl((\lambda(g)+\lambda(\alpha(g)))\lambda(g)^*\bigr)=\frac{1}{2}. ϕ M ( g ) = 2 1 τ 0 ( ( λ ( g ) + λ ( α ( g ))) λ ( g ) ∗ ) = 2 1 . Moreover, ϕ M ( e ) = 1 \phi_{\mathcal{M}}(e)=1 ϕ M ( e ) = 1 . Thus ϕ M = ( 1 + δ e ) / 2 \phi_{\mathcal{M}}=(1+\delta_e)/2 ϕ M = ( 1 + δ e ) /2 is conjugation invariant, and in particular is boomerang.
But the same argument used in Example 3.4 shows that M \mathcal{M} M is not boomerang. In particular, M \mathcal{M} M is not invariant. Thus conjugation invariance of ϕ M \phi_{\mathcal{M}} ϕ M does not imply that M \mathcal{M} M is boomerang or invariant.
Amenable boomerang subalgebras
In [8 Theorem B] , we showed that every amenable IRA of Γ \Gamma Γ is supported on subalgebras of L ( Rad ( Γ ) ) L(\operatorname{Rad}(\Gamma)) L ( Rad ( Γ )) . In particular, when the amenable radical is trivial, every amenable IRA is trivial. Here we study individual amenable boomerang subalgebras and give conditions under which they are trivial.
Before that, let us fix some notation. For a compact Γ \Gamma Γ -space X X X , fix x 0 ∈ X x_0 \in X x 0 ∈ X and represent C ( X ) C(X) C ( X ) as multiplication operators on ℓ 2 ( Γ ) \ell^2(\Gamma) ℓ 2 ( Γ ) by f δ g = f ( g x 0 ) δ g f\delta_g=f(gx_0)\delta_g f δ g = f ( g x 0 ) δ g . When X X X is minimal, this representation is faithful, and we identify C ( X ) C(X) C ( X ) with its image in B ( ℓ 2 ( Γ ) ) B(\ell^2(\Gamma)) B ( ℓ 2 ( Γ )) ; see [8 Section 3] . We use this convention throughout, with λ ( g ) f λ ( g ) ∗ = g ⋅ f \lambda(g)f\lambda(g)^*=g\cdot f λ ( g ) f λ ( g ) ∗ = g ⋅ f , where ( g ⋅ f ) ( x ) = f ( g − 1 x ) (g\cdot f)(x)=f(g^{-1}x) ( g ⋅ f ) ( x ) = f ( g − 1 x ) .
Finite dimensional boomerang algebras
We begin by exploring finite dimensional boomerang subalgebras, which are natural examples of amenable subalgebras.
To deal with finite-dimensional invariant subalgebras inside L ( Γ ) L(\Gamma) L ( Γ ) for i.c.c. groups Γ \Gamma Γ , Chifan-Das-Sun [21 ] used that every finite subset F ⊂ Γ ∖ { e } F\subset\Gamma\setminus\{e\} F ⊂ Γ ∖ { e } admits t ∈ Γ t\in\Gamma t ∈ Γ such that t F t − 1 ∩ F = ∅ tFt^{-1}\cap F=\varnothing tF t − 1 ∩ F = ∅ . For boomerang subalgebras, recurrence is available along powers of a fixed element. We therefore need separation for all sufficiently large powers of one element, so that it holds along any recurrence sequence. This motivates the definition below.
Definition 3.10. Let Γ \Gamma Γ be a countable discrete group. We say that Γ \Gamma Γ has property ESD if, given finitely many infinite index subgroups Λ i < Γ \Lambda_i<\Gamma Λ i < Γ and finite subsets F i ⊂ Γ F_i\subset\Gamma F i ⊂ Γ , there are s ∈ Γ s\in\Gamma s ∈ Γ and N ≥ 1 N\ge1 N ≥ 1 such that s n ∉ ⋃ t ∈ F i t Λ i s^n\notin\bigcup_{t\in F_i}t\Lambda_i s n ∈ / ⋃ t ∈ F i t Λ i for all n ≥ N n\ge N n ≥ N . We say that Γ \Gamma Γ has property ECI if, for every finite subset F ⊂ Γ ∖ { e } F\subset\Gamma\setminus\{e\} F ⊂ Γ ∖ { e } , there are s ∈ Γ s\in\Gamma s ∈ Γ and N ≥ 1 N\ge1 N ≥ 1 such that s n F s − n ∩ F = ∅ s^nFs^{-n}\cap F=\varnothing s n F s − n ∩ F = ∅ for all n ≥ N n\ge N n ≥ N .
See also [6 Definition 2.10] for a formulation of property E C I ECI E C I requiring escape under every fixed nontrivial conjugating element, for finite sets outside its centralizer.
It turns out we only require powers of one element to escape finite unions of cosets of arbitrary infinite index subgroups and gives a sufficient condition for E C I ECI E C I .
Proposition 3.11. Let Γ \Gamma Γ be an icc group with property E S D ESD E S D . Then Γ \Gamma Γ has property E C I ECI E C I .
Proof. Fix a finite subset F ⊂ Γ ∖ { e } F \subset\Gamma\setminus\{e\} F ⊂ Γ ∖ { e } . For each conjugate pair g , h ∈ F g,h \in F g , h ∈ F , choose t g , h ∈ Γ t_{g,h} \in\Gamma t g , h ∈ Γ with t g , h g t g , h − 1 = h t_{g,h}gt_{g,h}^{-1}=h t g , h g t g , h − 1 = h . Then
{ t ∈ Γ : t F t − 1 ∩ F ≠ ∅ } = ⋃ g , h ∈ F g conjugate to h t g , h C Γ ( g ) . \{t\in\Gamma:tFt^{-1}\cap F\ne\varnothing\}=
\bigcup_{\substack{g,h\in F\\g\text{ conjugate to }h}}t_{g,h}C_\Gamma(g). { t ∈ Γ : tF t − 1 ∩ F = ∅ } = g , h ∈ F g conjugate to h ⋃ t g , h C Γ ( g ) .
Since Γ \Gamma Γ is icc, every C Γ ( g ) C_\Gamma(g) C Γ ( g ) in this union has infinite index. Apply property E S D ESD E S D to these finitely many cosets.
Lemma 3.12. A countable discrete group Γ \Gamma Γ has property E C I ECI E C I if and only if, for every finite subset F ⊂ Γ ∖ { e } F\subset\Gamma\setminus\{e\} F ⊂ Γ ∖ { e } , there exists t ∈ Γ t\in\Gamma t ∈ Γ such that t n f ≠ f t n t^nf\ne ft^n t n f = f t n for every f ∈ F f\in F f ∈ F and every n ≥ 1 n\ge1 n ≥ 1 .
Proof. Suppose that Γ \Gamma Γ has property E C I ECI E C I , and choose t t t for F F F as in the definition. If t n t^n t n commuted with some f ∈ F f\in F f ∈ F , then t k n f t − k n = f t^{kn}ft^{-kn}=f t k n f t − k n = f for every k ≥ 1 k\ge1 k ≥ 1 , a contradiction.
Conversely, choose t t t satisfying the stated condition. For each f , h ∈ F f,h\in F f , h ∈ F , there is at most one n ≥ 1 n\ge1 n ≥ 1 with t n f t − n = h t^nft^{-n}=h t n f t − n = h . Indeed, two such exponents would give a nonzero power of t t t commuting with f f f . There are therefore only finitely many n ≥ 1 n\ge1 n ≥ 1 with t n F t − n ∩ F ≠ ∅ t^nFt^{-n}\cap F\ne\varnothing t n F t − n ∩ F = ∅ . This proves property E C I ECI E C I .
To obtain examples of groups with property E C I ECI E C I , we use the properties P a n a P_{\mathrm{ana}} P ana , P g e o P_{\mathrm{geo}} P geo and P n a i P_{\mathrm{nai}} P nai introduced in [17 ] .
A group Γ \Gamma Γ is said to have property P a n a P_{\mathrm{ana}} P ana (see [17 Definition 1] ) if, for every finite subset F ⊂ Γ ∖ { e } F\subset\Gamma\setminus\{e\} F ⊂ Γ ∖ { e } , there exist s ∈ Γ s\in\Gamma s ∈ Γ and C > 0 C>0 C > 0 such that
∥ ∑ j ≥ 1 a j λ ( s − j x s j ) ∥ ≤ C ∥ a ∥ 2 \left\|\sum_{j\ge1}a_j\lambda(s^{-j}xs^j)\right\|\le C\|a\|_2 j ≥ 1 ∑ a j λ ( s − j x s j ) ≤ C ∥ a ∥ 2
for every x ∈ F x\in F x ∈ F and every finitely supported a ∈ ℓ 2 ( N ) a\in\ell^2(\mathbb{N}) a ∈ ℓ 2 ( N ) .
Proposition 3.13. Every group with property P a n a P_{\mathrm{ana}} P ana has property E C I ECI E C I .
Proof. Fix a finite subset F ⊂ Γ ∖ { e } F\subset\Gamma\setminus\{e\} F ⊂ Γ ∖ { e } , and choose s , C s,C s , C as above. Suppose that s n x = x s n s^nx=xs^n s n x = x s n for some x ∈ F x\in F x ∈ F and n ≥ 1 n\ge1 n ≥ 1 . For M ≥ 1 M\ge1 M ≥ 1 , set a j = M − 1 a_j=M^{-1} a j = M − 1 when j ∈ { n , 2 n , … , M 2 n } j\in\{n,2n,\ldots,M^2n\} j ∈ { n , 2 n , … , M 2 n } , and a j = 0 a_j=0 a j = 0 otherwise. Then ∥ a ∥ 2 = 1 \|a\|_2=1 ∥ a ∥ 2 = 1 , whereas the displayed inequality gives M ∥ λ ( x ) ∥ = M ≤ C M\|\lambda(x)\|=M\le C M ∥ λ ( x ) ∥ = M ≤ C , a contradiction. Apply Lemma 3.12.
Definition 3.14 ([17 Definition 3] ). Let Γ \Gamma Γ act continuously on a nonempty compact Hausdorff space B B B . The pair ( Γ , B ) (\Gamma,B) ( Γ , B ) has property P g e o P_{\mathrm{geo}} P geo if, for every finite subset $ F ⊂ Γ ∖ { e } F\subset\Gamma\setminus\{e\} F ⊂ Γ ∖ { e } , there exist t ∈ Γ t\in\Gamma t ∈ Γ , a finite set { b s : s ∈ S } = Fix B ( t ) \{b_s:s\in S\}=\operatorname{Fix}_B(t) { b s : s ∈ S } = Fix B ( t ) , and open neighbourhoods V s V_s V s of b s b_s b s such that:
for every b ∈ B b\in B b ∈ B , the sequence t n b t^nb t n b converges to some b s b_s b s ;
(2) g V s ∩ V s ′ = ∅ gV_s \cap V_{s'} = \varnothing g V s ∩ V s ′ = ∅ for every g ∈ F g \in F g ∈ F and s , s ′ ∈ S s,s' \in S s , s ′ ∈ S ; (3) if b ∈ V s b \in V_s b ∈ V s and t n b ∉ V s t^n b \notin V_s t n b ∈ / V s for some n ≥ 1 n \ge1 n ≥ 1 , then t n + 1 b ∉ V s t^{n+1}b \notin V_s t n + 1 b ∈ / V s .
Let G G G be a connected real semisimple Lie group without compact factors. If Γ ≤ G \Gamma\leq G Γ ≤ G has trivial centre and Zariski-dense image in the adjoint algebraic group of G G G , then its action on the Furstenberg boundary of G G G has property P g e o P_{\mathrm{geo}} P geo [17 Theorem 2] . In particular, this applies to lattices in such groups G G G with trivial centre. For a concrete example, ( P S L d ( Z ) , R P d − 1 ) (\mathrm{PSL}_d(\mathbb{Z}),\mathbb{R}\mathbb{P}^{d-1}) ( PSL d ( Z ) , R P d − 1 ) has property P g e o P_{\mathrm{geo}} P geo for every d ≥ 2 d \geq2 d ≥ 2 [17 Example 1] . Property P g e o P_{\mathrm{geo}} P geo implies P a n a P_{\mathrm{ana}} P ana by [17 Lemmas 2.3 and 2.4] . Thus these groups also have property E C I ECI E C I by Proposition 3.13 .
Definition 3.15 ([17], Definition 4). A countable discrete group Γ \Gamma Γ has property P n a i P_{\mathrm{nai}} P nai if, for every finite subset F ⊂ Γ ∖ { e } F \subset\Gamma\setminus\{e\} F ⊂ Γ ∖ { e } , there exists an infinite order element s ∈ Γ s \in\Gamma s ∈ Γ such that ⟨ s , t ⟩ ≅ ⟨ s ⟩ ∗ ⟨ t ⟩ \langle s,t\rangle\cong\langle s\rangle* \langle t\rangle ⟨ s , t ⟩ ≅ ⟨ s ⟩ ∗ ⟨ t ⟩ for every t ∈ F t \in F t ∈ F , under the canonical homomorphism.
Every acylindrically hyperbolic group with no nontrivial finite normal subgroup has property P n a i P_{\mathrm{nai}} P nai by [1 Theorem 2.3] . There are also examples beyond this class. By [43 Corollary 1.3] , a group acting properly and cocompactly on a finite dimensional essential CAT(0) cube complex with no Euclidean factor has property P n a i P_{\mathrm{nai}} P nai , provided it has no nontrivial finite normal subgroup. This includes the infinite simple Burger–Mozes lattices in products of trees; see [43 p. 545] . These groups are not acylindrically hyperbolic, since a countable acylindrically hyperbolic group has uncountably many normal subgroups [51 Theorem 8.1(a) and the following paragraph] .
Proposition 3.16. Every group with property P n a i P_{\mathrm{nai}} P nai has property E C I ECI E C I .
Proof. Fix a finite subset F ⊂ Γ ∖ { e } F \subset\Gamma\setminus\{e\} F ⊂ Γ ∖ { e } , and choose s s s as in Definition 3.15. The free product normal form shows that s n t ≠ t s n s^n t \ne t s^n s n t = t s n for every t ∈ F t \in F t ∈ F and n ≥ 1 n \geq1 n ≥ 1 . Apply Lemma 3.12. □
Chifan–Das–Sun proved that every finite dimensional Γ \Gamma Γ -invariant von Neumann subalgebra of L ( Γ ) L(\Gamma) L ( Γ ) is C1 when Γ \Gamma Γ is icc [21 Proposition 3.2] . We prove the analogous statement for boomerang subalgebras under property E C I ECI E C I .
Proposition 3.17. Assume that Γ \Gamma Γ has property E C I ECI E C I . Then every finite dimensional boomerang von Neumann subalgebra M ≤ L ( Γ ) \mathcal{M} \leq L(\Gamma) M ≤ L ( Γ ) is C1.
Proof. Let M ≤ L ( Γ ) \mathcal{M} \leq L(\Gamma) M ≤ L ( Γ ) be a finite dimensional boomerang subalgebra. Suppose that M ≠ C 1 \mathcal{M} \ne\mathrm{C}1 M = C 1 . Choose { b 1 , … , b k } \{b_1,\ldots,b_k\} { b 1 , … , b k } to be an orthonormal basis of M ⊖ C 1 \mathcal{M} \ominus\mathrm{C}1 M ⊖ C 1 with respect to τ 0 \tau_0 τ 0 . Thus ∥ b j ∥ 2 = 1 \lVert b_j\rVert_2 = 1 ∥ b j ∥ 2 = 1 and τ 0 ( b j ) = 0 \tau_0(b_j)=0 τ 0 ( b j ) = 0 for every 1 ≤ j ≤ k 1 \leq j \leq k 1 ≤ j ≤ k . Let ϵ > 0 \epsilon> 0 ϵ > 0 . By truncating the Fourier expansions of the b j b_j b j , choose a finite subset F ϵ ⊂ Γ ∖ { e } F_\epsilon\subset\Gamma\setminus\{e\} F ϵ ⊂ Γ ∖ { e } and elements b j ϵ ∈ C Γ b_j^\epsilon\in\mathbb{C}\Gamma b j ϵ ∈ C Γ such that ∥ b j − b j ϵ ∥ 2 < ϵ \lVert b_j-b_j^\epsilon\rVert_2 < \epsilon ∥ b j − b j ϵ ∥ 2 < ϵ , supp ( b j ϵ ) ⊆ F ϵ \operatorname{supp}(b_j^\epsilon) \subseteq F_\epsilon supp ( b j ϵ ) ⊆ F ϵ ( 1 ≤ j ≤ k ) (1 \leq j \leq k) ( 1 ≤ j ≤ k ) . In particular, ∥ b j ϵ ∥ 2 2 ≤ 1 + ϵ \lVert b_j^\epsilon\rVert_2^2 \leq1+\epsilon ∥ b j ϵ ∥ 2 2 ≤ 1 + ϵ . By property E C I ECI E C I , choose s ∈ Γ s \in\Gamma s ∈ Γ and N ≥ 1 N \geq1 N ≥ 1 such that s n F ϵ s − n ∩ F ϵ = ∅ s^nF_\epsilon s^{-n}\cap F_\epsilon=\varnothing s n F ϵ s − n ∩ F ϵ = ∅ for every n ≥ N n \geq N n ≥ N .
For each t ∈ Γ t \in\Gamma t ∈ Γ , the element E M ( λ ( t ) ∗ b 1 λ ( t ) ) \mathbb{E}_{\mathcal{M}}(\lambda(t)^*b_1\lambda(t)) E M ( λ ( t ) ∗ b 1 λ ( t )) has trace zero. We may therefore write E M ( λ ( t ) ∗ b 1 λ ( t ) ) = ∑ j = 1 k α ( t , j ) b j \mathbb{E}_{\mathcal{M}}(\lambda(t)^*b_1\lambda(t))=\sum_{j=1}^{k}\alpha(t,j)b_j E M ( λ ( t ) ∗ b 1 λ ( t )) = ∑ j = 1 k α ( t , j ) b j . Since conditional expectations are ∥ ⋅ ∥ 2 \lVert\cdot\rVert_2 ∥ ⋅ ∥ 2 -contractive, ∑ j = 1 k ∣ α ( t , j ) ∣ 2 = ∥ E M ( λ ( t ) ∗ b 1 λ ( t ) ) ∥ 2 2 ≤ ∥ b 1 ∥ 2 2 = 1 \sum_{j=1}^{k}|\alpha(t,j)|^2=\lVert\mathbb{E}_{\mathcal{M}}(\lambda(t)^*b_1\lambda(t))\rVert_2^2\leq\lVert b_1\rVert_2^2=1 ∑ j = 1 k ∣ α ( t , j ) ∣ 2 = ∥ E M ( λ ( t ) ∗ b 1 λ ( t )) ∥ 2 2 ≤ ∥ b 1 ∥ 2 2 = 1 . Thus ∣ α ( t , j ) ∣ ≤ 1 |\alpha(t,j)|\leq1 ∣ α ( t , j ) ∣ ≤ 1 for every t t t and j j j . By Lemma 2.1 , E λ ( t ) M λ ( t ) ∗ ( b 1 ) = ∑ j = 1 k α ( t , j ) λ ( t ) b j λ ( t ) ∗ \mathbb{E}_{\lambda(t)\mathcal{M}\lambda(t)^*}(b_1)=\sum_{j=1}^{k}\alpha(t,j)\lambda(t)b_j\lambda(t)^* E λ ( t ) M λ ( t ) ∗ ( b 1 ) = ∑ j = 1 k α ( t , j ) λ ( t ) b j λ ( t ) ∗ . Since M \mathcal{M} M is boomerang, choose a strictly increasing sequence ( n i ) ⊆ N (n_i) \subseteq\mathbb{N} ( n i ) ⊆ N such that M i = λ ( s n i ) M λ ( s − n i ) → M \mathcal{M}_i = \lambda(s^{n_i})\mathcal{M}\lambda(s^{-n_i}) \to\mathcal{M} M i = λ ( s n i ) M λ ( s − n i ) → M in the Effros–Maréchal topology. Since E M ( b 1 ) = b 1 \mathbb{E}_{\mathcal{M}}(b_1)=b_1 E M ( b 1 ) = b 1 , choose i i i large enough that n i ≥ N n_i \geq N n i ≥ N and
∥ ∑ j = 1 k α ( s n i , j ) λ ( s n i ) b j λ ( s − n i ) − b 1 ∥ 2 = ∥ E M i ( b 1 ) − b 1 ∥ 2 < ϵ . \left\|\sum_{j=1}^{k}\alpha(s^{n_i},j)\lambda(s^{n_i})b_j\lambda(s^{-n_i})-b_1\right\|_2=\|\mathbb{E}_{\mathcal{M}_i}(b_1)-b_1\|_2<\epsilon. j = 1 ∑ k α ( s n i , j ) λ ( s n i ) b j λ ( s − n i ) − b 1 2 = ∥ E M i ( b 1 ) − b 1 ∥ 2 < ϵ .
For this i i i , the choice of the Fourier truncations gives
∥ E M i ( b 1 ) − ∑ j = 1 k α ( s n i , j ) λ ( s n i ) b j ϵ λ ( s − n i ) ∥ 2 ≤ ∑ j = 1 k ∣ α ( s n i , j ) ∣ ∥ b j − b j ϵ ∥ 2 ≤ k ϵ . \left\|\mathbb{E}_{\mathcal{M}_i}(b_1)-\sum_{j=1}^{k}\alpha(s^{n_i},j)\lambda(s^{n_i})b_j^\epsilon\lambda(s^{-n_i})\right\|_2
\leq\sum_{j=1}^{k}|\alpha(s^{n_i},j)|\|b_j-b_j^\epsilon\|_2\leq k\epsilon. E M i ( b 1 ) − j = 1 ∑ k α ( s n i , j ) λ ( s n i ) b j ϵ λ ( s − n i ) 2 ≤ j = 1 ∑ k ∣ α ( s n i , j ) ∣∥ b j − b j ϵ ∥ 2 ≤ k ϵ .
and
∥ ∑ j = 1 k α ( s n i , j ) λ ( s n i ) b j ϵ λ ( s − n i ) ∥ 2 ≤ ∑ j = 1 k ∣ α ( s n i , j ) ∣ ∥ b j ϵ ∥ 2 ≤ k ( 1 + ϵ ) . \left\|\sum_{j=1}^{k}\alpha(s^{n_i},j)\lambda(s^{n_i})b_j^\epsilon\lambda(s^{-n_i})\right\|_2
\leq\sum_{j=1}^{k}|\alpha(s^{n_i},j)|\|b_j^\epsilon\|_2\leq k(1+\epsilon). j = 1 ∑ k α ( s n i , j ) λ ( s n i ) b j ϵ λ ( s − n i ) 2 ≤ j = 1 ∑ k ∣ α ( s n i , j ) ∣∥ b j ϵ ∥ 2 ≤ k ( 1 + ϵ ) .
The Fourier support of this last sum is contained in s n i F ϵ s − n i s^{n_i}F_\epsilon s^{-n_i} s n i F ϵ s − n i , whereas the Fourier support of b 1 ϵ b_1^\epsilon b 1 ϵ is contained in F ϵ F_\epsilon F ϵ . By the choice of s s s and N N N , these supports are disjoint, so
⟨ b 1 ϵ , ∑ j = 1 k α ( s n i , j ) λ ( s n i ) b j ϵ λ ( s − n i ) ⟩ = 0. \left\langle b_1^\epsilon,\sum_{j=1}^{k}\alpha(s^{n_i},j)\lambda(s^{n_i})b_j^\epsilon\lambda(s^{-n_i})\right\rangle=0. ⟨ b 1 ϵ , j = 1 ∑ k α ( s n i , j ) λ ( s n i ) b j ϵ λ ( s − n i ) ⟩ = 0.
Consequently, by Cauchy–Schwarz and the preceding estimates,
1 = ⟨ b 1 , b 1 ⟩ ≤ ∣ ⟨ b 1 , b 1 − E M i ( b 1 ) ⟩ ∣ + ∣ ⟨ b 1 , E M i ( b 1 ) − ∑ j = 1 k α ( s n i , j ) λ ( s n i ) b j ϵ λ ( s − n i ) ⟩ ∣ + ∣ ⟨ b 1 − b 1 ϵ , ∑ j = 1 k α ( s n i , j ) λ ( s n i ) b j ϵ λ ( s − n i ) ⟩ ∣ ≤ ϵ + k ϵ + ϵ k ( 1 + ϵ ) . \begin{aligned}
1=\langle b_1,b_1\rangle
&\leq|\langle b_1,b_1-\mathbb{E}_{\mathcal{M}_i}(b_1)\rangle|\\
&\quad+\left|\left\langle b_1,\mathbb{E}_{\mathcal{M}_i}(b_1)-\sum_{j=1}^{k}\alpha(s^{n_i},j)\lambda(s^{n_i})b_j^\epsilon\lambda(s^{-n_i})\right\rangle\right|\\
&\quad+\left|\left\langle b_1-b_1^\epsilon,\sum_{j=1}^{k}\alpha(s^{n_i},j)\lambda(s^{n_i})b_j^\epsilon\lambda(s^{-n_i})\right\rangle\right|\\
&\leq\epsilon+k\epsilon+\epsilon k(1+\epsilon).
\end{aligned} 1 = ⟨ b 1 , b 1 ⟩ ≤ ∣ ⟨ b 1 , b 1 − E M i ( b 1 )⟩ ∣ + ⟨ b 1 , E M i ( b 1 ) − j = 1 ∑ k α ( s n i , j ) λ ( s n i ) b j ϵ λ ( s − n i ) ⟩ + ⟨ b 1 − b 1 ϵ , j = 1 ∑ k α ( s n i , j ) λ ( s n i ) b j ϵ λ ( s − n i ) ⟩ ≤ ϵ + k ϵ + ϵ k ( 1 + ϵ ) .
Since ϵ > 0 \epsilon>0 ϵ > 0 was arbitrary, this is a contradiction. □
Triviality of amenable boomerang subalgebras
We now pass from finite dimensional to arbitrary amenable boomerang subalgebras. Under the following hypotheses, the conclusion holds for every amenable boomerang subalgebra, strengthening the corresponding conclusion for amenable IRAs in [8 Theorem B] .
For an amenable von Neumann subalgebra M ≤ L ( Γ ) \mathcal{M}\leq L(\Gamma) M ≤ L ( Γ ) , let Hype τ 0 ( M ) \operatorname{Hype}_{\tau_0}(\mathcal{M}) Hype τ 0 ( M ) denote the states φ \varphi φ on B ( ℓ 2 ( Γ ) ) \mathbb{B}(\ell^2(\Gamma)) B ( ℓ 2 ( Γ )) such that φ ∣ L ( Γ ) = τ 0 \varphi|_{L(\Gamma)}=\tau_0 φ ∣ L ( Γ ) = τ 0 and φ ( x T ) = φ ( T x ) \varphi(xT)=\varphi(Tx) φ ( x T ) = φ ( T x ) for every x ∈ M x\in\mathcal{M} x ∈ M and T ∈ B ( ℓ 2 ( Γ ) ) T\in\mathbb{B}(\ell^2(\Gamma)) T ∈ B ( ℓ 2 ( Γ )) . This set is nonempty by [8 Proposition 2.4] .
Recall that an action Γ ↷ X \Gamma\curvearrowright X Γ ↷ X on a compact Hausdorff space is called extremely proximal if, for every proper closed subset C ⊊ X C\subsetneq X C ⊊ X and every nonempty open subset U ⊆ X U\subseteq X U ⊆ X , there is g ∈ Γ g\in\Gamma g ∈ Γ such that g C ⊆ U gC\subseteq U g C ⊆ U .
Theorem 3.18. Let Γ \Gamma Γ be a countable group satisfying one of the following conditions:
Γ \Gamma Γ has property P n a i \mathrm{P}_{\mathrm{nai}} P nai ;
(2) Γ \Gamma Γ admits a faithful extremely proximal action on a compact Hausdorff space with more than two points;
(3) Γ \Gamma Γ acts continuously on a nonempty compact Hausdorff space B B B such that ( Γ , B ) (\Gamma, B) ( Γ , B ) has property P g e o P_{\mathrm{geo}} P geo .
Then every amenable boomerang von Neumann subalgebra M ≤ L ( Γ ) \mathcal{M} \leq L(\Gamma) M ≤ L ( Γ ) is C 1 \mathbb{C}1 C 1 .
Proof. Let M \mathcal{M} M be an amenable boomerang subalgebra of L ( Γ ) L(\Gamma) L ( Γ ) . Let us first assume that Γ \Gamma Γ has property P n a i P_{\mathrm{nai}} P nai . We shall show that for every a ∈ M a \in\mathcal{M} a ∈ M , τ 0 ( a λ ( h ) ∗ ) = 0 \tau_0(a\lambda(h)^*) = 0 τ 0 ( aλ ( h ) ∗ ) = 0 for every h ≠ e h \ne e h = e . Let h ≠ e h \ne e h = e . Since Γ \Gamma Γ has property P n a i P_{\mathrm{nai}} P nai , we can find t ∈ Γ t \in\Gamma t ∈ Γ of infinite order such that Λ = ⟨ h , t ⟩ ≅ ⟨ h ⟩ ∗ ⟨ t ⟩ \Lambda= \langle h,t\rangle\cong\langle h\rangle* \langle t\rangle Λ = ⟨ h , t ⟩ ≅ ⟨ h ⟩ ∗ ⟨ t ⟩ . In particular, Λ \Lambda Λ is a non-elementary hyperbolic group. Set X = ∂ Λ X = \partial\Lambda X = ∂ Λ . Since ⟨ h , t ⟩ ≅ ⟨ h ⟩ ∗ ⟨ t ⟩ \langle h,t\rangle\cong\langle h\rangle* \langle t\rangle ⟨ h , t ⟩ ≅ ⟨ h ⟩ ∗ ⟨ t ⟩ , it follows that h { x t + , x t − } ∩ { x t + , x t − } = ∅ h\{x_t^+,x_t^-\} \cap\{x_t^+,x_t^-\} = \varnothing h { x t + , x t − } ∩ { x t + , x t − } = ∅ , where x t + , x t − x_t^+,x_t^- x t + , x t − are the fixed points of t t t . Since M \mathcal{M} M is a boomerang subalgebra of L ( Γ ) L(\Gamma) L ( Γ ) , there exists an increasing sequence of distinct elements { n i } ⊆ N \{n_i\} \subseteq\mathbb{N} { n i } ⊆ N such that λ ( t ) n i M λ ( t ) − n i → M \lambda(t)^{n_i}\mathcal{M}\lambda(t)^{-n_i} \to\mathcal{M} λ ( t ) n i M λ ( t ) − n i → M in EM-topology.
Since M \mathcal{M} M is amenable, it follows from [8 Proposition 2.4] that Hype τ 0 ( M ) ≠ ∅ \operatorname{Hype}_{\tau_0}(\mathcal{M}) \ne\varnothing Hype τ 0 ( M ) = ∅ . Let φ ∈ Hype τ 0 ( M ) \varphi\in\operatorname{Hype}_{\tau_0}(\mathcal{M}) φ ∈ Hype τ 0 ( M ) . We now observe that
φ i ( ⋅ ) = φ ( λ ( t ) − n i ⋅ λ ( t ) n i ) ∈ Hype τ 0 ( λ ( t ) n i M λ ( t ) − n i ) . \varphi_i(\cdot)=\varphi\left(\lambda(t)^{-n_i}\cdot\lambda(t)^{n_i}\right)\in\operatorname{Hype}_{\tau_0}\left(\lambda(t)^{n_i}\mathcal{M}\lambda(t)^{-n_i}\right). φ i ( ⋅ ) = φ ( λ ( t ) − n i ⋅ λ ( t ) n i ) ∈ Hype τ 0 ( λ ( t ) n i M λ ( t ) − n i ) .
Therefore, after passing to a subnet and using the proof of [8 Proposition 4.7] , we see that ψ : = weak ∗ − lim i φ i ∈ Hype τ 0 ( M ) \psi:= \operatorname{weak}^*-\lim_i\varphi_i \in\operatorname{Hype}_{\tau_0}(\mathcal{M}) ψ := weak ∗ − lim i φ i ∈ Hype τ 0 ( M ) . Moreover, since t t t has north–south dynamics on X X X , it follows that ψ ∣ C ( X ) = c δ x t + + ( 1 − c ) δ x t − \psi|_{C(X)}=c\delta_{x_t^+}+(1-c)\delta_{x_t^-} ψ ∣ C ( X ) = c δ x t + + ( 1 − c ) δ x t − for some c ∈ [ 0 , 1 ] c\in[0,1] c ∈ [ 0 , 1 ] . Since h { x t + , x t − } ∩ { x t + , x t − } = ∅ h\{x_t^+,x_t^-\}\cap\{x_t^+,x_t^-\}=\varnothing h { x t + , x t − } ∩ { x t + , x t − } = ∅ , we can find a non-negative continuous function f ∈ C ( X ) f\in C(X) f ∈ C ( X ) with f ∣ { x t + , x t − } = 1 f|_{\{x_t^+,x_t^-\}}=1 f ∣ { x t + , x t − } = 1 and f ∣ h { x t + , x t − } = 0 f|_{h\{x_t^+,x_t^-\}}=0 f ∣ h { x t + , x t − } = 0 . Since ψ ( ( 1 − f ) 2 ) = 0 \psi((1-f)^2)=0 ψ (( 1 − f ) 2 ) = 0 , Cauchy–Schwarz gives
τ 0 ( a λ ( h ) ∗ ) = ψ ( a λ ( h ) ∗ ) = ψ ( a λ ( h ) ∗ f ) = ψ ( a ( h − 1 ⋅ f ) λ ( h ) ∗ ) = ψ ( a h − 1 ⋅ f h − 1 ⋅ f λ ( h ) ∗ ) . \begin{aligned}
\tau_0(a\lambda(h)^*)&=\psi(a\lambda(h)^*)\\
&=\psi(a\lambda(h)^*f)\\
&=\psi\left(a(h^{-1}\cdot f)\lambda(h)^*\right)\\
&=\psi\left(a\sqrt{h^{-1}\cdot f}\sqrt{h^{-1}\cdot f}\lambda(h)^*\right).
\end{aligned} τ 0 ( aλ ( h ) ∗ ) = ψ ( aλ ( h ) ∗ ) = ψ ( aλ ( h ) ∗ f ) = ψ ( a ( h − 1 ⋅ f ) λ ( h ) ∗ ) = ψ ( a h − 1 ⋅ f h − 1 ⋅ f λ ( h ) ∗ ) .
Using the Cauchy–Schwarz inequality, we see that
∣ ψ ( a λ ( h ) ∗ ) ∣ ≤ ψ ( a ( h − 1 ⋅ f ) a ∗ ) ψ ( λ ( h ) ( h − 1 ⋅ f ) λ ( h ) ∗ ) = ψ ( ( h − 1 ⋅ f ) a ∗ a ) ψ ( f ) , \begin{aligned}
|\psi(a\lambda(h)^*)|
&\leq\sqrt{\psi\left(a(h^{-1}\cdot f)a^*\right)}
\sqrt{\psi\left(\lambda(h)(h^{-1}\cdot f)\lambda(h)^*\right)}\\
&=\sqrt{\psi\left((h^{-1}\cdot f)a^*a\right)}\sqrt{\psi(f)},
\end{aligned} ∣ ψ ( aλ ( h ) ∗ ) ∣ ≤ ψ ( a ( h − 1 ⋅ f ) a ∗ ) ψ ( λ ( h ) ( h − 1 ⋅ f ) λ ( h ) ∗ ) = ψ ( ( h − 1 ⋅ f ) a ∗ a ) ψ ( f ) ,
where the last equality uses the fact that ψ \psi ψ is a M \mathcal{M} M -hypertrace. Observe that ( h − 1 ⋅ f ) ( x t ± ) = f ( h x t ± ) = 0 (h^{-1}\cdot f)(x_t^\pm)=f(hx_t^\pm)=0 ( h − 1 ⋅ f ) ( x t ± ) = f ( h x t ± ) = 0 , hence, another application of the Cauchy–Schwarz inequality gives
∣ ψ ( ( h − 1 ⋅ f ) a ∗ ) ∣ 2 ≤ ψ ( ( h − 1 ⋅ f ) 2 ) ψ ( ( a ∗ a ) 2 ) = 0. |\psi((h^{-1}\cdot f)a^*)|^2\leq\psi((h^{-1}\cdot f)^2)\psi((a^*a)^2)=0. ∣ ψ (( h − 1 ⋅ f ) a ∗ ) ∣ 2 ≤ ψ (( h − 1 ⋅ f ) 2 ) ψ (( a ∗ a ) 2 ) = 0.
Thus τ 0 ( a λ ( h ) ∗ ) = 0 \tau_0(a\lambda(h)^*)=0 τ 0 ( aλ ( h ) ∗ ) = 0 . Since h ≠ e h\ne e h = e was arbitrary, M = C 1 \mathcal{M}=\mathbb{C}1 M = C 1 .
Assume now that Γ \Gamma Γ admits a faithful extremely proximal action on a space X X X with more than two points. Fix g ∈ Γ ∖ { e } g\in\Gamma\setminus\{e\} g ∈ Γ ∖ { e } . As before, we show that τ 0 ( a λ ( g ) ∗ ) = 0 \tau_0(a\lambda(g)^*)=0 τ 0 ( aλ ( g ) ∗ ) = 0 for every a ∈ M a\in\mathcal{M} a ∈ M . First, X X X has no isolated points. Indeed, if z z z were isolated, extreme proximality would send the proper closed set X ∖ { z } X\setminus\{z\} X ∖ { z } into { z } \{z\} { z } , contradicting injectivity since ∣ X ∣ > 2 |X|>2 ∣ X ∣ > 2 . By faithfulness, X ∖ Fix ( g ) X\setminus\operatorname{Fix}(g) X ∖ Fix ( g ) is a nonempty open set. Since X X X has no isolated points, this set is infinite. Choose x 1 ∉ Fix ( g ) x_1\notin\operatorname{Fix}(g) x 1 ∈ / Fix ( g ) and x 2 ∉ Fix ( g ) ∪ { x 1 , g x 1 , g − 1 x 1 } x_2\notin\operatorname{Fix}(g)\cup\{x_1,gx_1,g^{-1}x_1\} x 2 ∈ / Fix ( g ) ∪ { x 1 , g x 1 , g − 1 x 1 } . Thus g { x 1 , x 2 } ∩ { x 1 , x 2 } = ∅ g\{x_1,x_2\}\cap\{x_1,x_2\}=\varnothing g { x 1 , x 2 } ∩ { x 1 , x 2 } = ∅ . Choose open neighbourhoods U ∋ x 1 U\ni x_1 U ∋ x 1 and V ∋ x 2 V\ni x_2 V ∋ x 2 such that U ‾ ∩ V ‾ = ∅ \overline{U}\cap\overline{V}=\varnothing U ∩ V = ∅ , g ( U ‾ ∪ V ‾ ) ∩ ( U ‾ ∪ V ‾ ) = ∅ g(\overline{U}\cup\overline{V})\cap(\overline{U}\cup\overline{V})=\varnothing g ( U ∪ V ) ∩ ( U ∪ V ) = ∅ . By extreme proximality, choose t ∈ Γ t\in\Gamma t ∈ Γ such that t ( X ∖ U ) ⊆ V t(X\setminus U)\subseteq V t ( X ∖ U ) ⊆ V . Equivalently, t − 1 ( X ∖ V ) ⊆ U t^{-1}(X\setminus V)\subseteq U t − 1 ( X ∖ V ) ⊆ U . Since U ∩ V = ∅ U\cap V=\varnothing U ∩ V = ∅ , one has t ( V ) ⊆ V t(V)\subseteq V t ( V ) ⊆ V . Consequently, t n ( X ∖ U ) ⊆ V t^n(X\setminus U)\subseteq V t n ( X ∖ U ) ⊆ V for every n ≥ 1 n\geq1 n ≥ 1 . Choose φ ∈ Hype τ 0 ( M ) \varphi\in\operatorname{Hype}_{\tau_0}(\mathcal{M}) φ ∈ Hype τ 0 ( M ) . By the boomerang property, choose a strictly increasing sequence ( n j ) (n_j) ( n j ) such that M j = λ ( t n j ) M λ ( t − n j ) → M \mathcal{M}_j=\lambda(t^{n_j})\mathcal{M}\lambda(t^{-n_j})\to\mathcal{M} M j = λ ( t n j ) M λ ( t − n j ) → M in the Effros–Maréchal topology. Set φ j = φ ∘ Ad ( λ ( t − n j ) ) \varphi_j=\varphi\circ\operatorname{Ad}(\lambda(t^{-n_j})) φ j = φ ∘ Ad ( λ ( t − n j )) , and take a weak∗ ^* ∗ convergent subnet with limit ψ \psi ψ . As in case (1), it follows that ψ ∈ Hype τ 0 ( M ) \psi\in\operatorname{Hype}_{\tau_0}(\mathcal{M}) ψ ∈ Hype τ 0 ( M ) . Let ν = φ ∣ C ( X ) \nu=\varphi|_{C(X)} ν = φ ∣ C ( X ) , regarded as a probability measure on X X X . If f ∈ C ( X ) f\in C(X) f ∈ C ( X ) vanishes on U ‾ ∪ V ‾ \overline{U}\cup\overline{V} U ∪ V , then f ( t n y ) → 0 f(t^{n}y)\to0 f ( t n y ) → 0 for every y ∈ X y\in X y ∈ X . Dominated convergence gives φ j ( f ) = ∫ X f ( t n j y ) d ν ( y ) ⟶ 0 \varphi_j(f)=\int_X f(t^{n_j}y)\,d\nu(y)\longrightarrow0 φ j ( f ) = ∫ X f ( t n j y ) d ν ( y ) ⟶ 0 . Thus ψ ( f ) = 0 \psi(f)=0 ψ ( f ) = 0 for every such f f f . Choose 0 ≤ f ≤ 1 0\leq f\leq1 0 ≤ f ≤ 1 in C ( X ) C(X) C ( X ) with f ∣ U ‾ ∪ V ‾ = 1 f|_{\overline{U}\cup\overline{V}}=1 f ∣ U ∪ V = 1 and f ∣ g ( U ‾ ∪ V ‾ ) = 0 f|_{g(\overline{U}\cup\overline{V})}=0 f ∣ g ( U ∪ V ) = 0 . It follows that ψ ( ( 1 − f ) 2 ) = ψ ( ( g − 1 ⋅ f ) 2 ) = 0 \psi((1-f)^2)=\psi((g^{-1}\cdot f)^2)=0 ψ (( 1 − f ) 2 ) = ψ (( g − 1 ⋅ f ) 2 ) = 0 . Arguing as in case (1), using the Cauchy–Schwarz inequality and the M \mathcal{M} M -centrality of ψ \psi ψ , we obtain that τ 0 ( a λ ( g ) ∗ ) = 0 \tau_0(a\lambda(g)^*)=0 τ 0 ( aλ ( g ) ∗ ) = 0 for every a ∈ M a\in\mathcal{M} a ∈ M . Since g ≠ e g\neq e g = e was arbitrary, M = C 1 \mathcal{M}=\mathbb{C}1 M = C 1 .
Finally, assume that ( Γ , B ) (\Gamma,B) ( Γ , B ) has property P g e o P_{\mathrm{geo}} P geo . Fix g ∈ Γ ∖ { e } g\in\Gamma\setminus\{e\} g ∈ Γ ∖ { e } . Apply [17 ] with F = { g } F=\{g\} F = { g } . There are t ∈ Γ t\in\Gamma t ∈ Γ , a finite set { b s : s ∈ S } ⊆ B \{b_s:s\in S\}\subseteq B { b s : s ∈ S } ⊆ B of fixed points of t t t , and open neighbourhoods V s ∋ b s V_s\ni b_s V s ∋ b s such that every t n b t^n b t n b converges to some b s b_s b s , and g V s ∩ V s ′ = ∅ gV_s\cap V_{s'}=\varnothing g V s ∩ V s ′ = ∅ for all s , s ′ ∈ S s,s'\in S s , s ′ ∈ S . In particular, g { b s : s ∈ S } ∩ { b s : s ∈ S } = ∅ g\{b_s:s\in S\}\cap\{b_s:s\in S\}=\varnothing g { b s : s ∈ S } ∩ { b s : s ∈ S } = ∅ . Set K s = { b ∈ B : lim n → ∞ t n b = b s } K_s=\{b\in B:\lim_{n\to\infty}t^n b=b_s\} K s = { b ∈ B : lim n → ∞ t n b = b s } . These sets form a Borel partition of B B B . Choose φ ∈ Hype τ 0 ( M ) \varphi\in\operatorname{Hype}_{\tau_0}(\mathcal{M}) φ ∈ Hype τ 0 ( M ) and a recurrence sequence ( n j ) (n_j) ( n j ) for M \mathcal{M} M under conjugation by λ ( t ) \lambda(t) λ ( t ) . As in case (1), a weak∗ ^* ∗ convergent subnet of φ ∘ Ad ( λ ( t − n j ) ) \varphi\circ\operatorname{Ad}(\lambda(t^{-n_j})) φ ∘ Ad ( λ ( t − n j )) has limit ψ ∈ Hype τ 0 ( M ) \psi\in\operatorname{Hype}_{\tau_0}(\mathcal{M}) ψ ∈ Hype τ 0 ( M ) . Let ν = φ ∣ C ( B ) \nu=\varphi|_{C(B)} ν = φ ∣ C ( B ) , regarded as a probability measure on B B B . For f ∈ C ( B ) f\in C(B) f ∈ C ( B ) , dominated convergence gives lim j ∫ B f ( t n j b ) d ν ( b ) = ∑ s ∈ S ν ( K s ) f ( b s ) \lim_j\int_B f(t^{n_j}b)\,d\nu(b)=\sum_{s\in S}\nu(K_s)f(b_s) lim j ∫ B f ( t n j b ) d ν ( b ) = ∑ s ∈ S ν ( K s ) f ( b s ) , hence,
ψ ∣ C ( B ) = ∑ s ∈ S ν ( K s ) δ b s . \psi|_{C(B)}=\sum_{s\in S}\nu(K_s)\delta_{b_s}. ψ ∣ C ( B ) = s ∈ S ∑ ν ( K s ) δ b s .
Choose 0 ≤ f ≤ 1 0\leq f\leq1 0 ≤ f ≤ 1 in C ( B ) C(B) C ( B ) with f ( b s ) = 1 f(b_s)=1 f ( b s ) = 1 and f ( g b s ) = 0 f(gb_s)=0 f ( g b s ) = 0 for every s ∈ S s\in S s ∈ S . Then ψ ( ( 1 − f ) 2 ) = ψ ( ( g − 1 ⋅ f ) 2 ) = 0 \psi((1-f)^2)=\psi((g^{-1}\cdot f)^2)=0 ψ (( 1 − f ) 2 ) = ψ (( g − 1 ⋅ f ) 2 ) = 0 . Arguing as in case (1), using the Cauchy–Schwarz inequality and the M \mathcal{M} M -centrality of ψ \psi ψ , we obtain τ 0 ( a λ ( g ) ∗ ) = 0 \tau_0(a\lambda(g)^*)=0 τ 0 ( aλ ( g ) ∗ ) = 0 for every a ∈ M a\in\mathcal{M} a ∈ M . Since g ≠ e g\neq e g = e was arbitrary, M = C 1 \mathcal{M}=\mathbb{C}1 M = C 1 . □
Of course, there are many natural classes of examples which admit extremely proximal faithful actions that are not topologically free. Let us consider the class of Thompson’s groups first.
Example 3.19. Every amenable boomerang von Neumann subalgebra of L ( T ) L(T) L ( T ) is C 1 \mathbb{C}1 C 1 , where T T T is Thompson’s group. The natural action of Thompson’s group T T T on S 1 S^1 S 1 is faithful but not topologically free; see [20 ] , Sections 1 and 5. Indeed, T T T contains nonidentity elements supported in a proper closed arc, which therefore fix the complementary open arc pointwise. Let C ⊆ S 1 C\subseteq S^1 C ⊆ S 1 be closed and let U ⊆ S 1 U\subseteq S^1 U ⊆ S 1 be nonempty and open. Choose proper closed arcs I , J I,J I , J with dyadic endpoints such that C ⊆ I C\subseteq I C ⊆ I and J ⊆ U J \subseteq U J ⊆ U . There is t ∈ T t \in T t ∈ T mapping I I I onto J J J , so t C ⊆ U tC \subseteq U tC ⊆ U . Thus the action is extremely proximal, and Theorem 3.18 applies.
Example 3.20. Let C C C be the binary Cantor set. The natural action of Thompson’s group V V V on C C C is faithful, extremely proximal and not topologically free; see the proof of [46 Theorem 4.6(iii)] . The same holds for every countable subgroup Γ ≤ Homeo ( C ) \Gamma\leq\operatorname{Homeo}(C) Γ ≤ Homeo ( C ) containing V V V . Consequently, every amenable boomerang von Neumann subalgebra of L ( Γ ) L(\Gamma) L ( Γ ) is C 1 \mathbb{C}1 C 1 by Theorem 3.18 .
The examples above may suggest that our result applies only to C ∗ C^* C ∗ -simple groups. The following example of Le Boudec [45 ] dispels this impression.
Example 3.21. Let Ω \Omega Ω be a finite set with ∣ Ω ∣ ≥ 3 |\Omega| \geq3 ∣Ω∣ ≥ 3 , and let F < F ′ ≤ Sym ( Ω ) F < F' \leq\operatorname{Sym}(\Omega) F < F ′ ≤ Sym ( Ω ) , where F F F acts freely and transitively on Ω \Omega Ω . Consider Le Boudec’s group Γ = G ( F , F ′ ) \Gamma= G(F,F') Γ = G ( F , F ′ ) from [45 ] . Its action on the boundary of the ∣ Ω ∣ |\Omega| ∣Ω∣ -regular tree is faithful and not topologically free; see the proof of [45 Theorem C] . It is extremely proximal by [46 Proposition 4.26] . Thus every amenable boomerang von Neumann subalgebra of L ( Γ ) L(\Gamma) L ( Γ ) is C 1 \mathbb{C}1 C 1 by Theorem 3.18 . These groups have trivial amenable radical but are not C ∗ C^* C ∗ -simple [45 Theorem C] .
Split higher rank lattices
We now prove Theorem 1.2 for the groups from Subsection 2.5. We first use Proposition 3.5 to locate the Fourier support of the conditional expectation of a highest root element.
Lemma 4.1. Let Γ \Gamma Γ be as in Subsection 2.5 and M ≤ L ( Γ ) M \leq L(\Gamma) M ≤ L ( Γ ) be a boomerang subalgebra. Then
E M ( λ ( x θ ( t ) ) ) ∈ L ( U θ ( K ) ∩ Γ ) , whenever x θ ( t ) ∈ Γ . E_M(\lambda(x_\theta(t))) \in L(U_\theta(K) \cap\Gamma), \qquad\text{whenever } x_\theta(t) \in\Gamma. E M ( λ ( x θ ( t ))) ∈ L ( U θ ( K ) ∩ Γ ) , whenever x θ ( t ) ∈ Γ. Proof. Since U θ = Z ( U ) U_\theta= Z(U) U θ = Z ( U ) , every a ∈ Γ ∩ U ( K ) a \in\Gamma\cap U(K) a ∈ Γ ∩ U ( K ) commutes with x θ ( t ) x_\theta(t) x θ ( t ) . Moreover, C Γ ( a m ) = C Γ ( a ) C_\Gamma(a^m) = C_\Gamma(a) C Γ ( a m ) = C Γ ( a ) for every m ≥ 1 m \geq1 m ≥ 1 . Indeed, in a faithful linear representation, commuting with a m a^m a m implies commuting with log ( a m ) = m log a \log(a^m) = m\log a log ( a m ) = m log a , and hence with a a a . Proposition 3.5 therefore gives supp E M ( λ ( x θ ( t ) ) ) ⊆ R ( x θ ( t ) ) ⊆ C Γ ( Γ ∩ U ( K ) ) \operatorname{supp} E_M(\lambda(x_\theta(t))) \subseteq R(x_\theta(t)) \subseteq C_\Gamma(\Gamma\cap U(K)) supp E M ( λ ( x θ ( t ))) ⊆ R ( x θ ( t )) ⊆ C Γ ( Γ ∩ U ( K )) . The group Γ ∩ U ( K ) \Gamma\cap U(K) Γ ∩ U ( K ) is Zariski dense in U U U . Indeed, commensurability supplies a nontrivial root element in Γ \Gamma Γ for every positive root; its powers are Zariski dense in that root group, and the positive root groups generate U U U .
It remains to show that C G ( U ) = U θ C_G(U) = U_\theta C G ( U ) = U θ . To see this, an element centralizing U U U belongs to N G ( U ) = B N_G(U) = B N G ( U ) = B . Its torus component acts trivially on U / [ U , U ] U/[U,U] U / [ U , U ] , hence all simple roots take the value 1 1 1 on it. Since G G G is adjoint, that component is trivial. The element therefore belongs to Z ( U ) = U θ Z(U) = U_\theta Z ( U ) = U θ .
We next construct suitable conjugates of the highest root group inside G θ G_\theta G θ . The following properties will allow us to compare the Fourier coefficients of products of conditional expectations.
Proposition 4.2. Let Γ θ = Γ ∩ G θ ( K ) \Gamma_\theta= \Gamma\cap G_\theta(K) Γ θ = Γ ∩ G θ ( K ) and Υ θ = Γ ∩ U θ ( K ) \Upsilon_\theta= \Gamma\cap U_\theta(K) Υ θ = Γ ∩ U θ ( K ) . Then there is some element s ∈ Υ θ s \in\Upsilon_\theta s ∈ Υ θ and an infinite sequence γ i ∈ Γ θ \gamma_i \in\Gamma_\theta γ i ∈ Γ θ . Write w i : = γ i w γ i − 1 w_i := \gamma_i w\gamma_i^{-1} w i := γ i w γ i − 1 and define the product map ψ i , j : Υ θ × Υ θ → Γ θ \psi_{i,j}:\Upsilon_\theta\times\Upsilon_\theta\to\Gamma_\theta ψ i , j : Υ θ × Υ θ → Γ θ , ψ i , j ( u , v ) = u i v j = γ i u γ i − 1 γ j v γ j − 1 \psi_{i,j}(u,v)=u_i v_j=\gamma_i u\gamma_i^{-1}\gamma_jv\gamma_j^{-1} ψ i , j ( u , v ) = u i v j = γ i u γ i − 1 γ j v γ j − 1 . The following properties hold for every i ≠ j i\ne j i = j :
(1) ψ i , j \psi_{i,j} ψ i , j is injective.
(2) If for k , n ∈ Z ∖ { 0 } k,n\in\mathbb{Z}\setminus\{0\} k , n ∈ Z ∖ { 0 } and v , w ∈ Υ θ v,w\in\Upsilon_\theta v , w ∈ Υ θ we have
[ s i k v j , ( w i s j k ) n ] = [ γ i s i k γ i − 1 γ j v γ j − 1 , ( γ i w γ i − 1 γ j s j k γ j − 1 ) n ] = 1 , [s_i^k v_j,(w_i s_j^k)^n]=[\gamma_i s_i^k\gamma_i^{-1}\gamma_jv\gamma_j^{-1},(\gamma_iw\gamma_i^{-1}\gamma_j s_j^k\gamma_j^{-1})^n]=1, [ s i k v j , ( w i s j k ) n ] = [ γ i s i k γ i − 1 γ j v γ j − 1 , ( γ i w γ i − 1 γ j s j k γ j − 1 ) n ] = 1 , then v = w = s k v=w=s^k v = w = s k .
Proof. In this proof we make the matrix calculations in SL 2 ( K ) \operatorname{SL}_2(K) SL 2 ( K ) and apply ι \iota ι . Since ker ι ⊆ { ± I 2 } \ker\iota\subseteq\{\pm I_2\} ker ι ⊆ { ± I 2 } , an equality between the images gives the corresponding equality in PGL 2 ( K ) \operatorname{PGL}_2(K) PGL 2 ( K ) . The projective image of a root element determines its parameter. In particular,
x θ ( k ) = ι ( 1 k 0 1 ) , x − θ ( k ) = ι ( 1 0 k 1 ) , U θ ( K ) = { x θ ( k ) ∣ k ∈ K } ≅ K . x_\theta(k)=\iota\begin{pmatrix}1&k\\0&1\end{pmatrix},\qquad x_{-\theta}(k)=\iota\begin{pmatrix}1&0\\k&1\end{pmatrix},\qquad U_\theta(K)=\{x_\theta(k)\mid k\in K\}\cong K. x θ ( k ) = ι ( 1 0 k 1 ) , x − θ ( k ) = ι ( 1 k 0 1 ) , U θ ( K ) = { x θ ( k ) ∣ k ∈ K } ≅ K .
The groups Υ θ \Upsilon_\theta Υ θ and U θ ( O S ) U_\theta(\mathcal{O}_S) U θ ( O S ) are commensurable, and the same holds for the negative root groups. Hence we can find d ∈ N d\in\mathbb{N} d ∈ N such that x θ ( d ) , x − θ ( d ) ∈ Γ x_\theta(d),x_{-\theta}(d)\in\Gamma x θ ( d ) , x − θ ( d ) ∈ Γ . Set J = { t ∈ K : x θ ( t ) ∈ Υ θ } J=\{t\in K:x_\theta(t)\in\Upsilon_\theta\} J = { t ∈ K : x θ ( t ) ∈ Υ θ } , so that J J J is commensurable with O S \mathcal{O}_S O S .
Consider the set T K = { ζ + ζ − 1 − 2 ∈ K ∖ { 0 } : ζ is a root of unity } T_K=\{\zeta+\zeta^{-1}-2\in K\setminus\{0\}:\zeta\text{ is a root of unity}\} T K = { ζ + ζ − 1 − 2 ∈ K ∖ { 0 } : ζ is a root of unity } . Each such root of unity ζ \zeta ζ satisfies a quadratic equation over K K K , so that [ K ( ζ ) : K ] ≤ 2 [K(\zeta):K]\le2 [ K ( ζ ) : K ] ≤ 2 and hence [ Q ( ζ ) : Q ] ≤ 2 [ K : Q ] [\mathbb{Q}(\zeta):\mathbb{Q}]\le2[K:\mathbb{Q}] [ Q ( ζ ) : Q ] ≤ 2 [ K : Q ] . Thus the set T K T_K T K is finite.
Choose a finite place p ∉ S \mathfrak{p}\notin S p ∈ / S such that v p ( a ) = 0 v_{\mathfrak{p}}(a)=0 v p ( a ) = 0 for every a ∈ T K a\in T_K a ∈ T K and v p ( r ) ≥ 0 v_{\mathfrak{p}}(r)\ge0 v p ( r ) ≥ 0 for every r ∈ J r\in J r ∈ J . Such a place exists: T K T_K T K is a finite set of nonzero algebraic integers, and m J ⊆ O S mJ\subseteq\mathcal{O}_S m J ⊆ O S for some positive integer m m m , so we also take p ∤ m \mathfrak{p}\nmid m p ∤ m .
Choose a positive integer q ∈ d Z q\in d\mathbb{Z} q ∈ d Z with v p ( q ) > 0 v_{\mathfrak{p}}(q)>0 v p ( q ) > 0 , and set s = x θ ( d ) s=x_\theta(d) s = x θ ( d ) and γ i = x − θ ( q i ) \gamma_i=x_{-\theta}(qi) γ i = x − θ ( q i ) . Thus if v = x θ ( t ) ∈ Υ θ v=x_\theta(t)\in\Upsilon_\theta v = x θ ( t ) ∈ Υ θ then
v i : = γ i v γ i − 1 = ι ( I 2 + t A i ) , where A i = ( − q i 1 − q 2 i 2 q i ) . v_i:=\gamma_i v\gamma_i^{-1}=\iota(I_2+tA_i),\qquad\text{where }A_i=\begin{pmatrix}-qi&1\\-q^2i^2&qi\end{pmatrix}. v i := γ i v γ i − 1 = ι ( I 2 + t A i ) , where A i = ( − q i − q 2 i 2 1 q i ) .
Note that A i A_i A i are nilpotent of order 2 and that ( I 2 + t 1 A i ) ( I 2 + t 2 A i ) = I 2 + ( t 1 + t 2 ) A i (I_2+t_1A_i)(I_2+t_2A_i)=I_2+(t_1+t_2)A_i ( I 2 + t 1 A i ) ( I 2 + t 2 A i ) = I 2 + ( t 1 + t 2 ) A i for all t 1 , t 2 ∈ J t_1,t_2\in J t 1 , t 2 ∈ J . We proceed to show that these choices give us the desired properties.
Note that the nilpotent matrices A i , A j A_i,A_j A i , A j have distinct kernels, namely K ( 1 , q i ) T K(1,qi)^{\mathsf T} K ( 1 , q i ) T and K ( 1 , q j ) T K(1,qj)^{\mathsf T} K ( 1 , q j ) T . Therefore the one-parameter groups I 2 + K A i I_2+KA_i I 2 + K A i and I 2 + K A j I_2+KA_j I 2 + K A j have trivial intersection even after passage to PGL 2 ( K ) \operatorname{PGL}_2(K) PGL 2 ( K ) . Rearranging an equality of two products proves (1).
For (2) take v = x θ ( m ) v=x_\theta(m) v = x θ ( m ) , w = x θ ( r ) w=x_\theta(r) w = x θ ( r ) for some m , r ∈ J m,r\in J m , r ∈ J and put
X = ( I 2 + d k A i ) ( I 2 + m A j ) , T = ( I 2 + r A i ) ( I 2 + d k A j ) . X=(I_2+dkA_i)(I_2+mA_j),\qquad T=(I_2+rA_i)(I_2+dkA_j). X = ( I 2 + d k A i ) ( I 2 + m A j ) , T = ( I 2 + r A i ) ( I 2 + d k A j ) .
The hypothesis gives [ X , T n ] = 1 [X,T^n]=1 [ X , T n ] = 1 in PGL 2 ( K ) \operatorname{PGL}_2(K) PGL 2 ( K ) . Our goal is to show that r = m = d k r=m=dk r = m = d k . If r = 0 r=0 r = 0 , then T n T^n T n has the unique fixed point ker A j \ker A_j ker A j , whereas X X X does not fix ker A j \ker A_j ker A j , a contradiction. Hence r ≠ 0 r\ne0 r = 0 .
Since Tr ( A i A j ) = − q 2 ( i − j ) 2 \operatorname{Tr}(A_iA_j)=-q^2(i-j)^2 Tr ( A i A j ) = − q 2 ( i − j ) 2 , Tr ( T ) − 2 = − r k d q 2 ( i − j ) 2 ≠ 0 \operatorname{Tr}(T)-2=-rkdq^2(i-j)^2\ne0 Tr ( T ) − 2 = − r k d q 2 ( i − j ) 2 = 0 , v p ( Tr ( T ) − 2 ) > 0 v_{\mathfrak{p}}(\operatorname{Tr}(T)-2)>0 v p ( Tr ( T ) − 2 ) > 0 . It follows that no eigenvalue of T T T is a root of unity. Otherwise, writing its eigenvalues as ζ , ζ − 1 \zeta,\zeta^{-1} ζ , ζ − 1 would give Tr ( T ) − 2 ∈ T K \operatorname{Tr}(T)-2\in T_K Tr ( T ) − 2 ∈ T K , a contradiction. In particular, its eigenvalues λ , λ − 1 \lambda,\lambda^{-1} λ , λ − 1 are distinct and λ 2 \lambda^2 λ 2 has infinite order. For every n ≥ 1 n\ge1 n ≥ 1 , the centralizer of T ‾ n \overline{T}^{n} T n in PGL 2 ( K ) \operatorname{PGL}_2(K) PGL 2 ( K ) fixes each of the two eigenlines of T T T .
Choose a basis of K 2 K^2 K 2 consisting of vectors in the kernels of A i A_i A i and A j A_j A j . In this basis, for some a , b ∈ K × a,b\in K^\times a , b ∈ K × ,
I 2 + t A i = ( 1 a t 0 1 ) , I 2 + z A j = ( 1 0 b z 1 ) . I_2+tA_i=\begin{pmatrix}1&at\\0&1\end{pmatrix},\qquad I_2+zA_j=\begin{pmatrix}1&0\\bz&1\end{pmatrix}. I 2 + t A i = ( 1 0 a t 1 ) , I 2 + z A j = ( 1 b z 0 1 ) .
For t , z ≠ 0 t,z\ne0 t , z = 0 , the two projective fixed points ω ± \omega_\pm ω ± of their product are the roots of the polynomial: Q ( Z ) = Z 2 − a t Z − a t b z = 0 Q(Z)=Z^2-atZ-\frac{at}{bz}=0 Q ( Z ) = Z 2 − a tZ − b z a t = 0 . In particular ω + + ω − = a t \omega_++\omega_-=at ω + + ω − = a t and ω + ω − = − a t b z \omega_+\omega_-=-\frac{at}{bz} ω + ω − = − b z a t . Solving, we see that the unordered pair { ω ± } \{\omega_\pm\} { ω ± } determines t t t and z z z .
In particular, since X ‾ \overline{X} X commutes with T ‾ n \overline{T}^{n} T n , it fixes both eigenlines of T T T individually. If m = 0 m=0 m = 0 , then X = I 2 + d k A i X=I_2+dkA_i X = I 2 + d k A i is nontrivial unipotent and has only one fixed point, a contradiction. Thus m ≠ 0 m\ne0 m = 0 . Moreover, X ‾ ≠ 1 \overline{X}\ne1 X = 1 , by the trivial intersection of the two unipotent groups. Hence X ‾ \overline{X} X has precisely these two fixed points. Comparing the polynomial Q ( Z ) Q(Z) Q ( Z ) for ( t , z ) = ( k d , m ) (t,z)=(kd,m) ( t , z ) = ( k d , m ) and ( t , z ) = ( r , k d ) (t,z)=(r,kd) ( t , z ) = ( r , k d ) gives the desired equalities r = k d r=kd r = k d and m = k d m=kd m = k d . □
As promised, we now return to our general K K K -split lattice Γ \Gamma Γ from Subsection 2.5 and apply Proposition 4.2 to Γ θ = G θ ( K ) ∩ Γ \Gamma_\theta=G_\theta(K)\cap\Gamma Γ θ = G θ ( K ) ∩ Γ . We show that, when Γ M = { e } \Gamma_M=\{e\} Γ M = { e } , at most one of the conditional expectations E M ( λ ( s i k ) ) E_M(\lambda(s_i^k)) E M ( λ ( s i k )) can be nonzero for each fixed k ≠ 0 k\ne0 k = 0 .
Proposition 4.3. Let Γ \Gamma Γ be as in Subsection 2.5 and apply Proposition 4.2 to the subgroup Γ θ \Gamma_\theta Γ θ , retaining all the notation there. Let M ∈ SA ( L ( Γ ) ) M\in\operatorname{SA}(L(\Gamma)) M ∈ SA ( L ( Γ )) be a boomerang algebra with the property that Γ M = { γ ∣ λ ( γ ) ∈ M } = ⟨ 1 ⟩ \Gamma_M=\{\gamma\mid\lambda(\gamma)\in M\}=\langle1\rangle Γ M = { γ ∣ λ ( γ ) ∈ M } = ⟨ 1 ⟩ . Then for every 0 ≠ k ∈ Z 0\ne k\in\mathbb{Z} 0 = k ∈ Z and every i ≠ j ∈ N i\ne j\in\mathbb{N} i = j ∈ N either E M ( λ ( s i k ) ) = 0 E_M(\lambda(s_i^k))=0 E M ( λ ( s i k )) = 0 or E M ( λ ( s j k ) ) = 0 E_M(\lambda(s_j^k))=0 E M ( λ ( s j k )) = 0 .
Proof. Suppose both expectations are nonzero. Applying Lemma 4.1 to the boomerang subalgebras λ ( γ i ) ∗ M λ ( γ i ) \lambda(\gamma_i)^*M\lambda(\gamma_i) λ ( γ i ) ∗ M λ ( γ i ) and λ ( γ j ) ∗ M λ ( γ j ) \lambda(\gamma_j)^*M\lambda(\gamma_j) λ ( γ j ) ∗ M λ ( γ j ) , we write
A : = E M ( λ ( s i k ) ) = ∑ v ∈ Υ θ c v λ ( v i ) , B : = E M ( λ ( s j k ) ) = ∑ w ∈ Υ θ d w λ ( w j ) . A:=E_M(\lambda(s_i^k))=\sum_{v\in\Upsilon_\theta}c_v\lambda(v_i),\qquad
B:=E_M(\lambda(s_j^k))=\sum_{w\in\Upsilon_\theta}d_w\lambda(w_j). A := E M ( λ ( s i k )) = v ∈ Υ θ ∑ c v λ ( v i ) , B := E M ( λ ( s j k )) = w ∈ Υ θ ∑ d w λ ( w j ) .
Since E M E_M E M is the orthogonal projection onto L 2 ( M ) L^2(M) L 2 ( M ) , we have c s k = ∥ A ∥ 2 2 > 0 c_{s_k}=\|A\|_2^2>0 c s k = ∥ A ∥ 2 2 > 0 and d s k = ∥ B ∥ 2 2 > 0 d_{s_k}=\|B\|_2^2>0 d s k = ∥ B ∥ 2 2 > 0 . Also, c e = τ 0 ( A ) = 0 c_e=\tau_0(A)=0 c e = τ 0 ( A ) = 0 and d e = τ 0 ( B ) = 0 d_e=\tau_0(B)=0 d e = τ 0 ( B ) = 0 .
By the injectivity in Proposition 4.2 (1), the Fourier coefficient of A B AB A B at v i w j v_iw_j v i w j is c v d w c_vd_w c v d w . Indeed, this holds for finite partial sums of A A A , and the general statement follows by L 2 L^2 L 2 -approximation, since multiplication on the right by B B B is continuous on L 2 L^2 L 2 . Since A ∈ M A\in M A ∈ M , bimodularity gives A B = E M ( A λ ( s j k ) ) = ∑ v ∈ Υ θ c v E M ( λ ( v i s j k ) ) AB=E_M(A\lambda(s_j^k))=\sum_{v\in\Upsilon_\theta}c_vE_M(\lambda(v_is_j^k)) A B = E M ( A λ ( s j k )) = ∑ v ∈ Υ θ c v E M ( λ ( v i s j k )) , where the series converges in L 2 L^2 L 2 . Suppose that d w ≠ 0 d_w\ne0 d w = 0 for some w ≠ s k w\ne s_k w = s k . Since d e = 0 d_e=0 d e = 0 , w ≠ e w\ne e w = e . The Fourier coefficient of A B AB A B at s i k w j s_i^kw_j s i k w j is c s k d w ≠ 0 c_{s_k}d_w\ne0 c s k d w = 0 . Thus there is some v ∈ Υ θ v\in\Upsilon_\theta v ∈ Υ θ with c v ≠ 0 c_v\ne0 c v = 0 such that s i k w j ∈ supp E M ( λ ( v i s j k ) ) s_i^kw_j\in\operatorname{supp}E_M(\lambda(v_is_j^k)) s i k w j ∈ supp E M ( λ ( v i s j k )) . Since c e = 0 c_e=0 c e = 0 , v ≠ e v\ne e v = e . Also, v i s j k ≠ e v_is_j^k\ne e v i s j k = e by Proposition 4.2 (1). Proposition 3.5 therefore gives n ≥ 1 n\ge1 n ≥ 1 such that [ s i k w j , ( v i s j k ) n ] = 1 [s_i^kw_j,(v_is_j^k)^n]=1 [ s i k w j , ( v i s j k ) n ] = 1 . By Proposition 4.2 (2), w = s k w=s^k w = s k , a contradiction. Therefore B = d s k λ ( s j k ) B=d_{s_k}\lambda(s_j^k) B = d s k λ ( s j k ) . Applying E M E_M E M again gives d s k = d s k 2 d_{s_k}=d_{s_k}^2 d s k = d s k 2 . Since d s k > 0 d_{s_k}>0 d s k > 0 , d s k = 1 d_{s_k}=1 d s k = 1 , so λ ( s j k ) ∈ M \lambda(s_j^k)\in M λ ( s j k ) ∈ M . But s j k ≠ e s_j^k\ne e s j k = e , contradicting Γ M = { e } \Gamma_M=\{e\} Γ M = { e } . □
We now combine the IRS classification in Theorem 2.4 with Proposition 4.3 . In the case where J ∗ μ = δ { e } J_*\mu=\delta_{\{e\}} J ∗ μ = δ { e } , averaging over the conjugates of s k s^k s k gives the vanishing required in Proposition 2.6 .
Proof of Theorem 1.2 . Let μ \mu μ be an ergodic IRA on SA ( L ( Γ ) ) \operatorname{SA}(L(\Gamma)) SA ( L ( Γ )) and push it forward to an ergodic IRS ν = J ∗ μ \nu=J_*\mu ν = J ∗ μ on Γ \Gamma Γ . We may split the proof into two cases by Theorem 2.4 .
Case 1: ν = ν H \nu=\nu_H ν = ν H for a finite-index subgroup H < Γ H<\Gamma H < Γ . In this case, μ ( { M : Γ M ∈ { γ H γ − 1 : γ ∈ Γ } } ) = 1 \mu(\{M:\Gamma_M\in\{\gamma H\gamma^{-1}:\gamma\in\Gamma\}\})=1 μ ({ M : Γ M ∈ { γ H γ − 1 : γ ∈ Γ }}) = 1 . Let H 0 : = ⋂ γ ∈ Γ γ H γ − 1 H_0:=\bigcap_{\gamma\in\Gamma}\gamma H\gamma^{-1} H 0 := ⋂ γ ∈ Γ γ H γ − 1 be the normal core of H H H ; since [ Γ : H ] < ∞ [\Gamma:H]<\infty [ Γ : H ] < ∞ , H H H has finitely many conjugates, so H 0 ◃ Γ H_0\triangleleft\Gamma H 0 ◃ Γ and [ Γ : H 0 ] < ∞ [\Gamma:H_0]<\infty [ Γ : H 0 ] < ∞ . For μ \mu μ -a.e. M M M we then have L ( H 0 ) ⊆ L ( Γ M ) ⊆ M ⊆ L ( Γ ) L(H_0)\subseteq L(\Gamma_M)\subseteq M\subseteq L(\Gamma) L ( H 0 ) ⊆ L ( Γ M ) ⊆ M ⊆ L ( Γ ) . By Borel density, every finite-index subgroup of Γ \Gamma Γ is Zariski dense in G G G . Hence an element of Γ \Gamma Γ with finite conjugacy class belongs to Z ( G ) = { e } Z(G)=\{e\} Z ( G ) = { e } , so Γ \Gamma Γ is icc. Since Γ \Gamma Γ is icc and [ Γ : H 0 ] < ∞ [\Gamma:H_0]<\infty [ Γ : H 0 ] < ∞ , the subgroup H 0 H_0 H 0 is relatively icc in Γ \Gamma Γ : for γ ≠ e \gamma\ne e γ = e , [ H 0 : C H 0 ( γ ) ] ≥ [ Γ : C Γ ( γ ) ] [ Γ : H 0 ] = ∞ [H_0:C_{H_0}(\gamma)]\ge\frac{[\Gamma:C_\Gamma(\gamma)]}{[\Gamma:H_0]}=\infty [ H 0 : C H 0 ( γ )] ≥ [ Γ : H 0 ] [ Γ : C Γ ( γ )] = ∞ . By [38 Lemma 3.5] , every intermediate von Neumann algebra L ( H 0 ) ⊆ M ⊆ L ( Γ ) L(H_0)\subseteq M\subseteq L(\Gamma) L ( H 0 ) ⊆ M ⊆ L ( Γ ) is of the form M = L ( K ) M=L(K) M = L ( K ) for a subgroup H 0 < K < Γ H_0<K<\Gamma H 0 < K < Γ . By maximality, K = Γ M K=\Gamma_M K = Γ M , so M = L ( Γ M ) = L ( J ( M ) ) M=L(\Gamma_M)=L(J(M)) M = L ( Γ M ) = L ( J ( M )) for μ \mu μ -a.e. M M M , whence μ = ( L ∘ J ) ∗ μ = L ∗ ν \mu=(L\circ J)_*\mu=L_*\nu μ = ( L ∘ J ) ∗ μ = L ∗ ν .
Case 2: ν = δ { e } \nu=\delta_{\{e\}} ν = δ { e } . Our goal is to prove that μ = δ C \mu=\delta_C μ = δ C . By Proposition 2.6 it suffices to find s ∈ Γ s\in\Gamma s ∈ Γ of infinite order with E M ( λ ( s k ) ) = 0 \mathbb{E}_M(\lambda(s^k))=0 E M ( λ ( s k )) = 0 for all k ≠ 0 k\ne0 k = 0 and μ \mu μ -a.e. M M M . Apply Proposition 4.2 to the group Γ θ = Γ ∩ G θ ( K ) \Gamma_\theta=\Gamma\cap G_\theta(K) Γ θ = Γ ∩ G θ ( K ) and Υ θ = Γ ∩ U θ ( K ) \Upsilon_\theta=\Gamma\cap U_\theta(K) Υ θ = Γ ∩ U θ ( K ) . We keep all the notation of that proposition and proceed to show that the element s ∈ Υ θ s\in\Upsilon_\theta s ∈ Υ θ provided there satisfies this vanishing property. For this let us fix 0 ≠ k ∈ Z 0\ne k\in\mathbb{Z} 0 = k ∈ Z .
The two assumptions of Proposition 4.3 hold for μ \mu μ -almost every M M M : by our standing Case 2 assumption, Γ M = { e } \Gamma_M=\{e\} Γ M = { e } , and M M M is boomerang by Poincaré recurrence, as noted in the introduction. Hence, by Proposition 4.3 , almost surely, at most one of the N N N terms { E M ( λ ( s i k ) ) } i ≤ N \{\mathbb{E}_M(\lambda(s_i^k))\}_{i\le N} { E M ( λ ( s i k )) } i ≤ N is nonzero, and since ∥ E M ( λ ( s i k ) ) ∥ 2 ≤ 1 \|\mathbb{E}_M(\lambda(s_i^k))\|_2\le1 ∥ E M ( λ ( s i k )) ∥ 2 ≤ 1 , ∑ i = 1 N ∥ E M ( λ ( s i k ) ) ∥ 2 2 ≤ 1 \sum_{i=1}^N\|\mathbb{E}_M(\lambda(s_i^k))\|_2^2\le1 ∑ i = 1 N ∥ E M ( λ ( s i k )) ∥ 2 2 ≤ 1 for μ \mu μ -a.e. M M M . Let s i = γ i s γ i − 1 s_i=\gamma_i s\gamma_i^{-1} s i = γ i s γ i − 1 . Integrating and using conjugation-invariance of μ \mu μ together with
∥ E M ( λ ( γ ) x λ ( γ ) ∗ ) ∥ 2 = ∥ E λ ( γ ) − 1 M λ ( γ ) ( x ) ∥ 2 , \|\mathbb{E}_M(\lambda(\gamma)x\lambda(\gamma)^*)\|_2=\|\mathbb{E}_{\lambda(\gamma)^{-1}M\lambda(\gamma)}(x)\|_2, ∥ E M ( λ ( γ ) x λ ( γ ) ∗ ) ∥ 2 = ∥ E λ ( γ ) − 1 M λ ( γ ) ( x ) ∥ 2 ,
we get that
1 ≥ ∫ ∑ i = 1 N ∥ E M ( λ ( s i k ) ) ∥ 2 2 d μ ( M ) = ∑ i = 1 N ∫ ∥ E M ( λ ( s k ) ) ∥ 2 2 d μ ( M ) = N ∫ ∥ E M ( λ ( s k ) ) ∥ 2 2 d μ ( M ) . \begin{aligned}
1\ge\int\sum_{i=1}^{N}\|\mathbb{E}_M(\lambda(s_i^k))\|_2^2\,d\mu(M)=\sum_{i=1}^{N}\int\|\mathbb{E}_M(\lambda(s^k))\|_2^2\,d\mu(M) \\
&=N\int\|\mathbb{E}_M(\lambda(s^k))\|_2^2\,d\mu(M).
\end{aligned} 1 ≥ ∫ i = 1 ∑ N ∥ E M ( λ ( s i k )) ∥ 2 2 d μ ( M ) = i = 1 ∑ N ∫ ∥ E M ( λ ( s k )) ∥ 2 2 d μ ( M ) = N ∫ ∥ E M ( λ ( s k )) ∥ 2 2 d μ ( M ) .
As N N N is arbitrary, ∫ ∥ E M ( λ ( s k ) ) ∥ 2 2 d μ ( M ) = 0 \int\|\mathbb{E}_M(\lambda(s^k))\|_2^2\,d\mu(M)=0 ∫ ∥ E M ( λ ( s k )) ∥ 2 2 d μ ( M ) = 0 , i.e. E M ( λ ( s k ) ) = 0 \mathbb{E}_M(\lambda(s^k))=0 E M ( λ ( s k )) = 0 for μ \mu μ -a.e. M M M . Discarding a countable union of null sets, for μ \mu μ -a.e. M M M we have E M ( λ ( s k ) ) = 0 \mathbb{E}_M(\lambda(s^k))=0 E M ( λ ( s k )) = 0 for all k ≥ 1 k\ge1 k ≥ 1 , and taking adjoints, E M ( λ ( s − k ) ) = E M ( λ ( s k ) ∗ ) = E M ( λ ( s k ) ) ∗ = 0 \mathbb{E}_M(\lambda(s^{-k}))=\mathbb{E}_M(\lambda(s^k)^*)=\mathbb{E}_M(\lambda(s^k))^*=0 E M ( λ ( s − k )) = E M ( λ ( s k ) ∗ ) = E M ( λ ( s k ) ) ∗ = 0 as well; hence E M ( λ ( s k ) ) = 0 \mathbb{E}_M(\lambda(s^k))=0 E M ( λ ( s k )) = 0 for all k ≠ 0 k\ne0 k = 0 . Proposition 2.6 now gives μ = δ C = L ∗ δ { e } = L ∗ ν \mu=\delta_C=L_*\delta_{\{e\}}=L_*\nu μ = δ C = L ∗ δ { e } = L ∗ ν . □ \square □
Boomerang subalgebras of acylindrically hyperbolic groups
We prove that every boomerang subalgebra of an acylindrically hyperbolic group with trivial amenable radical is a group von Neumann algebra. We then extend the argument to finite products and deduce the corresponding statement for invariant random algebras.
We use the terminology and notation from Subsection 2.6. Our first step is to find a generalized loxodromic element on which the conditional expectation is nonzero. We do this by showing that K l o x K_{\mathrm{lox}} K lox is thick for every boomerang character.
Lemma 5.1. Let Γ \Gamma Γ be acylindrically hyperbolic with trivial amenable radical. For every Boomerang character φ \varphi φ on Γ \Gamma Γ , the set K l o x K_{\mathrm{lox}} K lox is φ \varphi φ -thick.
Proof. Fix a non-elementary acylindrical action Γ ↷ S \Gamma\curvearrowright S Γ ↷ S and g ≠ e g \ne e g = e . For a loxodromic element h h h , write E Γ ( h ) E_{\Gamma}(h) E Γ ( h ) for its maximal virtually cyclic subgroup. Every loxodromic element of this action is WPD. By [14 Lemma 5.6] , ⋂ h ∈ K l o x E Γ ( h ) \bigcap_{h\in K_{\mathrm{lox}}} E_{\Gamma}(h) ⋂ h ∈ K lox E Γ ( h ) is the maximal finite normal subgroup of Γ \Gamma Γ . This subgroup is trivial, since Γ \Gamma Γ has trivial amenable radical. Choose a loxodromic element h h h with g ∉ E Γ ( h ) g \notin E_{\Gamma}(h) g ∈ / E Γ ( h ) .
The fixed-point pairs of h h h and g − 1 h g g^{-1}hg g − 1 h g are disjoint. Indeed, if they shared a point, then ⟨ h , g − 1 h g ⟩ \langle h,g^{-1}hg\rangle ⟨ h , g − 1 h g ⟩ would fix that point and hence contain no independent loxodromic elements. By [51 Theorem 1.1] , this subgroup would be virtually cyclic. The infinite cyclic subgroups generated by h h h and g − 1 h g g^{-1}hg g − 1 h g would then have an infinite intersection, so these elements would have the same fixed-point pair. Since the stabilizer of the fixed-point pair of h h h is E Γ ( h ) E_{\Gamma}(h) E Γ ( h ) , this would give g ∈ E Γ ( h ) g\in E_{\Gamma}(h) g ∈ E Γ ( h ) , a contradiction; see [22 Lemma 6.5] .
Apply [1 Lemma 1.2] to h − 1 h^{-1} h − 1 and g − 1 h g g^{-1}hg g − 1 h g . There is N ≥ 1 N\ge1 N ≥ 1 such that h − k g − 1 h k g = ( h − 1 ) k ( g − 1 h g ) k ∈ K l o x h^{-k}g^{-1}h^{k}g=(h^{-1})^{k}(g^{-1}hg)^{k}\in K_{\mathrm{lox}} h − k g − 1 h k g = ( h − 1 ) k ( g − 1 h g ) k ∈ K lox for k ≥ N k\ge N k ≥ N . Since φ \varphi φ is boomerang, there is a strictly increasing sequence ( n i ) ⊆ N ≥ 1 (n_i)\subseteq\mathbb{N}_{\ge1} ( n i ) ⊆ N ≥ 1 such that φ ( h − n i t h n i ) ⟶ φ ( t ) \varphi(h^{-n_i}th^{n_i})\longrightarrow\varphi(t) φ ( h − n i t h n i ) ⟶ φ ( t ) for every t ∈ Γ t\in\Gamma t ∈ Γ . Passing to a subsequence, assume that n i + 1 − n i ≥ N n_{i+1}-n_i\ge N n i + 1 − n i ≥ N for every i i i . Set h i = h − n i h_i=h^{-n_i} h i = h − n i . Then φ ( h i g h i − 1 ) ⟶ φ ( g ) \varphi(h_i g h_i^{-1})\longrightarrow\varphi(g) φ ( h i g h i − 1 ) ⟶ φ ( g ) . If j > i j>i j > i , put k = n j − n i ≥ N k=n_j-n_i\ge N k = n j − n i ≥ N . By the choice of N N N , h j g − 1 h j − 1 h i g h i − 1 = h − n i ( h − k g − 1 h k g ) h n i ∈ K l o x h_jg^{-1}h_j^{-1}h_igh_i^{-1}=h^{-n_i}(h^{-k}g^{-1}h^kg)h^{n_i}\in K_{\mathrm{lox}} h j g − 1 h j − 1 h i g h i − 1 = h − n i ( h − k g − 1 h k g ) h n i ∈ K lox . For i > j i>j i > j , the expression is the inverse of the one with the indices reversed, so it also belongs to K l o x K_{\mathrm{lox}} K lox . Thus K l o x K_{\mathrm{lox}} K lox is φ \varphi φ -thick. □
:::
Applying this to the associated boomerang character φ M \varphi_M φ M gives the following consequence.
Corollary 5.2. Let Γ \Gamma Γ be acylindrically hyperbolic with trivial amenable radical, and let M ≤ L ( Γ ) \mathcal{M}\le L(\Gamma) M ≤ L ( Γ ) be boomerang. If E M ( λ ( s ) ) = 0 E_{\mathcal{M}}(\lambda(s))=0 E M ( λ ( s )) = 0 for every s ∈ K l o x s\in K_{\mathrm{lox}} s ∈ K lox , then M = C 1 \mathcal{M}=\mathbb{C}1 M = C 1 .
Proof. By Lemma 3.8 , φ M \varphi_M φ M is boomerang. Apply Lemmas 5.1 and 2.3 to obtain φ M = δ e \varphi_M=\delta_e φ M = δ e . Thus E M ( λ ( g ) ) = 0 E_{\mathcal{M}}(\lambda(g))=0 E M ( λ ( g )) = 0 for every g ≠ e g\ne e g = e . For x ∈ M x\in\mathcal{M} x ∈ M and g ≠ e g\ne e g = e , τ 0 ( x λ ( g ) ∗ ) = τ 0 ( x E M ( λ ( g ) ) ∗ ) = 0 \tau_0(x\lambda(g)^*)=\tau_0(xE_{\mathcal{M}}(\lambda(g))^*)=0 τ 0 ( x λ ( g ) ∗ ) = τ 0 ( x E M ( λ ( g ) ) ∗ ) = 0 . Hence x = τ 0 ( x ) 1 x=\tau_0(x)1 x = τ 0 ( x ) 1 . □
We now show that the conditional expectation of a generalized loxodromic element is either zero or the element itself. We first locate its Fourier support using Proposition 3.5 . For every s ∈ K l o x s\in K_{\mathrm{lox}} s ∈ K lox , the subgroup E Γ ( s ) E_{\Gamma}(s) E Γ ( s ) is virtually cyclic and hyperbolically embedded in Γ \Gamma Γ by [22 Corollary 2.9] . Moreover, C Γ ( s k ) ⊆ E Γ ( s ) C_{\Gamma}(s^{k}) \subseteq E_{\Gamma}(s) C Γ ( s k ) ⊆ E Γ ( s ) for every k ≠ 0 k \ne0 k = 0 . Consequently, Proposition 3.5 gives E M ( λ ( s ) ) ∈ L ( E Γ ( s ) ) \mathbb{E}_{\mathcal M}(\lambda(s)) \in L(E_{\Gamma}(s)) E M ( λ ( s )) ∈ L ( E Γ ( s )) . In particular, when Γ \Gamma Γ is torsion-free, E Γ ( s ) = ⟨ a ⟩ E_{\Gamma}(s)=\langle a\rangle E Γ ( s ) = ⟨ a ⟩ for a primitive element a a a with s ∈ ⟨ a ⟩ s \in\langle a\rangle s ∈ ⟨ a ⟩ .
The following proof adapts Jiang’s Fourier coefficient comparison argument [38 ] , as used in [9 Section 2 and Proposition 2.2] , to boomerang subalgebras; see also [40 ] .
Proposition 5.3. Let Γ \Gamma Γ be acylindrically hyperbolic with trivial amenable radical, and let M ≤ L ( Γ ) \mathcal M \le L(\Gamma) M ≤ L ( Γ ) be boomerang. For every s ∈ K l o x s \in K_{\mathrm{lox}} s ∈ K lox , E M ( λ ( s ) ) ∈ { 0 , λ ( s ) } \mathbb{E}_{\mathcal M}(\lambda(s)) \in\{0,\lambda(s)\} E M ( λ ( s )) ∈ { 0 , λ ( s )} .
Proof. Suppose that E M ( λ ( s ) ) ≠ 0 \mathbb{E}_{\mathcal M}(\lambda(s)) \ne0 E M ( λ ( s )) = 0 , and put A = E Γ ( s ) A=E_{\Gamma}(s) A = E Γ ( s ) . The subgroup A A A is hyperbolically embedded in Γ \Gamma Γ , and Γ \Gamma Γ has no nontrivial finite normal subgroup. By the proof of [1 Theorem 1] , choose an infinite-order element t ∈ Γ t \in\Gamma t ∈ Γ such that ⟨ A , t ⟩ ≅ A ∗ ⟨ t ⟩ \langle A,t\rangle\cong A * \langle t\rangle ⟨ A , t ⟩ ≅ A ∗ ⟨ t ⟩ , under the canonical homomorphism.
By Lemma 3.6 , choose n ≥ 1 n \ge1 n ≥ 1 such that E M ( λ ( t − n s t n ) ) ≠ 0 \mathbb{E}_{\mathcal M}(\lambda(t^{-n}st^{n})) \ne0 E M ( λ ( t − n s t n )) = 0 . Set b = t − n s t n b=t^{-n}st^{n} b = t − n s t n and B = t − n A t n B=t^{-n}At^{n} B = t − n A t n . Conjugating the unique maximal virtually cyclic subgroup containing s s s gives B = E Γ ( b ) B=E_{\Gamma}(b) B = E Γ ( b ) .
We claim that ⟨ A , B ⟩ ≅ A ∗ B \langle A,B\rangle\cong A * B ⟨ A , B ⟩ ≅ A ∗ B under the canonical homomorphism. Indeed, take a nonempty reduced word with alternating nonidentity syllables from A A A and B B B . Write every B B B -syllable as t − n a t n t^{-n}at^{n} t − n a t n , where a ∈ A ∖ { e } a \in A \setminus\{e\} a ∈ A ∖ { e } . After these substitutions, the resulting word is reduced in A ∗ ⟨ t ⟩ A * \langle t\rangle A ∗ ⟨ t ⟩ : every two successive powers t n t^{n} t n and t − n t^{-n} t − n are separated by a nonidentity A A A -syllable. Thus the original word is nontrivial. In particular, A ∩ B = { e } A \cap B=\{e\} A ∩ B = { e } .
Since b b b is conjugate to s s s , it also belongs to K l o x K_{\mathrm{lox}} K lox . By the support inclusion above, write E M ( λ ( s ) ) = ∑ x ∈ A c x λ ( x ) \mathbb{E}_{\mathcal M}(\lambda(s))=\sum_{x\in A}c_x\lambda(x) E M ( λ ( s )) = ∑ x ∈ A c x λ ( x ) , and E M ( λ ( b ) ) = ∑ y ∈ B d y λ ( y ) \mathbb{E}_{\mathcal M}(\lambda(b))=\sum_{y\in B}d_y\lambda(y) E M ( λ ( b )) = ∑ y ∈ B d y λ ( y ) . Trace preservation gives c e = d e = 0 c_e=d_e=0 c e = d e = 0 , since s , b ≠ e s,b \ne e s , b = e . Bimodularity gives
∑ x ∈ A , y ∈ B c x d y λ ( x y ) = E M ( λ ( s ) ) E M ( λ ( b ) ) = E M ( λ ( s ) E M ( λ ( b ) ) ) = ∑ y ′ ∈ B d y ′ E M ( λ ( s y ′ ) ) . (1) \begin{aligned}
\sum_{x\in A,\,y\in B}c_xd_y\lambda(xy)
&=\mathbb{E}_{\mathcal M}(\lambda(s))\mathbb{E}_{\mathcal M}(\lambda(b))\\
&=\mathbb{E}_{\mathcal M}\bigl(\lambda(s)\mathbb{E}_{\mathcal M}(\lambda(b))\bigr)\\
&=\sum_{y'\in B}d_{y'}\mathbb{E}_{\mathcal M}(\lambda(sy')). \tag*{(1)}
\end{aligned} x ∈ A , y ∈ B ∑ c x d y λ ( x y ) = E M ( λ ( s )) E M ( λ ( b )) = E M ( λ ( s ) E M ( λ ( b )) ) = y ′ ∈ B ∑ d y ′ E M ( λ ( s y ′ )) . (1)
Fix x ∈ A ∖ { e , s } x \in A \setminus\{e,s\} x ∈ A ∖ { e , s } and y ∈ B ∖ { e } y \in B \setminus\{e\} y ∈ B ∖ { e } . We show that the Fourier coefficient at x y xy x y of every term on the right of (1) is zero. The term y ′ = e y'=e y ′ = e vanishes because d e = 0 d_e=0 d e = 0 . For y ′ ∈ B ∖ { e } y' \in B \setminus\{e\} y ′ ∈ B ∖ { e } , the word s y ′ sy' s y ′ is reduced and nontrivial. If x y ∈ supp ( E M ( λ ( s y ′ ) ) ) xy \in\operatorname{supp}(\mathbb{E}_{\mathcal M}(\lambda(sy'))) x y ∈ supp ( E M ( λ ( s y ′ ))) , then Proposition 3.5 gives k ≥ 1 k \ge1 k ≥ 1 such that ( x y ) ( s y ′ ) k = ( s y ′ ) k ( x y ) (xy)(sy')^{k}=(sy')^{k}(xy) ( x y ) ( s y ′ ) k = ( s y ′ ) k ( x y ) . Both sides are reduced words in A ∗ B A * B A ∗ B , since x , s x,s x , s are nonidentity elements of A A A and y , y ′ y,y' y , y ′ are nonidentity elements of B B B . The first syllable on the left is x x x , whereas the first syllable on the right is s s s . Uniqueness of the reduced normal form therefore gives x = s x=s x = s , a contradiction.
Fourier coefficients are ∥ ⋅ ∥ 2 \lVert\cdot\rVert_{2} ∥ ⋅ ∥ 2 -continuous, so the coefficient at x y xy x y of the right-hand series is zero. Comparing coefficients in (1) gives c x d y = 0 c_xd_y=0 c x d y = 0 whenever x ∈ A ∖ { e , s } x \in A \setminus\{e,s\} x ∈ A ∖ { e , s } and y ∈ B ∖ { e } y \in B \setminus\{e\} y ∈ B ∖ { e } . Since E M ( λ ( b ) ) ≠ 0 \mathbb{E}_{\mathcal M}(\lambda(b)) \ne0 E M ( λ ( b )) = 0 and d e = 0 d_e=0 d e = 0 , there is y ∈ B ∖ { e } y \in B \setminus\{e\} y ∈ B ∖ { e } with d y ≠ 0. H e n c e d_y \ne0. Hence d y = 0. H e n ce c_x=0$ for every x ∈ A ∖ { e , s } x \in A \setminus\{e,s\} x ∈ A ∖ { e , s } . Together with c e = 0 c_e=0 c e = 0 , this gives E M ( λ ( s ) ) = c s λ ( s ) \mathbb{E}_{\mathcal M}(\lambda(s))=c_s\lambda(s) E M ( λ ( s )) = c s λ ( s ) . Applying E M \mathbb{E}_{\mathcal M} E M once more yields c s λ ( s ) = c s 2 λ ( s ) c_s\lambda(s)=c_s^2\lambda(s) c s λ ( s ) = c s 2 λ ( s ) . Since the expectation is nonzero, c s ≠ 0 c_s \ne0 c s = 0 , and therefore c s = 1 c_s=1 c s = 1 .
We now pass from generalized loxodromic elements to arbitrary group elements. The preceding proposition and bimodularity will show that the image of a group element under the canonical conditional expectation is either zero or that unitary.
Theorem 5.4. Let Γ \Gamma Γ be a countable acylindrically hyperbolic group with a trivial amenable radical. Every boomerang von Neumann subalgebra M ≤ L ( Γ ) \mathcal{M} \leq L(\Gamma) M ≤ L ( Γ ) is L ( H ) L(H) L ( H ) for a unique boomerang subgroup H ≤ Γ H \leq\Gamma H ≤ Γ .
Proof. The case M = C 1 \mathcal{M} = \mathbb{C}1 M = C 1 is immediate. Otherwise, Corollary 5.2 gives s ∈ K l o x s \in K_{\mathrm{lox}} s ∈ K lox with E M ( λ ( s ) ) ≠ 0 E_{\mathcal{M}}(\lambda(s)) \ne0 E M ( λ ( s )) = 0 . Proposition 5.3 gives λ ( s ) ∈ M \lambda(s) \in\mathcal{M} λ ( s ) ∈ M . Choose t t t and n ≥ 1 n \geq1 n ≥ 1 as in its proof and set b = t − n s t n b = t^{-n}st^{n} b = t − n s t n . Thus b ∈ K l o x b \in K_{\mathrm{lox}} b ∈ K lox and E M ( λ ( b ) ) ≠ 0 E_{\mathcal{M}}(\lambda(b)) \ne0 E M ( λ ( b )) = 0 . Applying the proposition to b b b gives λ ( b ) ∈ M \lambda(b) \in\mathcal{M} λ ( b ) ∈ M . The same proof gives ⟨ E Γ ( s ) , E Γ ( b ) ⟩ ≅ E Γ ( s ) ∗ E Γ ( b ) \langle E_{\Gamma}(s), E_{\Gamma}(b)\rangle\cong E_{\Gamma}(s)*E_{\Gamma}(b) ⟨ E Γ ( s ) , E Γ ( b )⟩ ≅ E Γ ( s ) ∗ E Γ ( b ) , so E Γ ( s ) ∩ E Γ ( b ) = { e } E_{\Gamma}(s) \cap E_{\Gamma}(b) = \{e\} E Γ ( s ) ∩ E Γ ( b ) = { e } .
Fix g ≠ e g \ne e g = e . Since E Γ ( s ) ∩ E Γ ( b ) = { e } E_{\Gamma}(s) \cap E_{\Gamma}(b) = \{e\} E Γ ( s ) ∩ E Γ ( b ) = { e } , g g g cannot belong to both subgroups. Choose r ∈ { s , b } r \in\{s,b\} r ∈ { s , b } such that g ∉ E Γ ( r ) g \notin E_{\Gamma}(r) g ∈ / E Γ ( r ) . The virtually cyclic subgroup E Γ ( r ) E_{\Gamma}(r) E Γ ( r ) is infinite, finitely generated and hyperbolically embedded in Γ \Gamma Γ . Choose a relative generating set Y Y Y for which it is hyperbolically embedded. Since r r r has infinite order, the last assertion of Corollary 6.12 gives k ∈ Z k \in\mathbb{Z} k ∈ Z such that g r k gr^{k} g r k is loxodromic WPD for the action on the relative Cayley graph with generating set Y ⊔ E Γ ( r ) Y \sqcup E_{\Gamma}(r) Y ⊔ E Γ ( r ) . Thus g r k ∈ K l o x gr^{k} \in K_{\mathrm{lox}} g r k ∈ K lox by Theorem 1.4 [51 ] .
Since λ ( r ) ∈ M \lambda(r) \in\mathcal{M} λ ( r ) ∈ M , we have λ ( r k ) ∈ M \lambda(r^{k}) \in\mathcal{M} λ ( r k ) ∈ M . Bimodularity and Proposition 5.3 therefore give E M ( λ ( g ) ) λ ( r k ) = E M ( λ ( g r k ) ) ∈ { 0 , λ ( g r k ) } E_{\mathcal{M}}(\lambda(g))\lambda(r^{k}) = E_{\mathcal{M}}(\lambda(gr^{k})) \in\{0,\lambda(gr^{k})\} E M ( λ ( g )) λ ( r k ) = E M ( λ ( g r k )) ∈ { 0 , λ ( g r k )} . Thus E M ( λ ( g ) ) ∈ { 0 , λ ( g ) } E_{\mathcal{M}}(\lambda(g)) \in\{0,\lambda(g)\} E M ( λ ( g )) ∈ { 0 , λ ( g )} for every g ∈ Γ g \in\Gamma g ∈ Γ . Set H = { g ∈ Γ : λ ( g ) ∈ M } H = \{g \in\Gamma: \lambda(g) \in\mathcal{M}\} H = { g ∈ Γ : λ ( g ) ∈ M } . This is a subgroup and L ( H ) ⊆ M L(H) \subseteq\mathcal{M} L ( H ) ⊆ M . If x ∈ M x \in\mathcal{M} x ∈ M and g ∉ H g \notin H g ∈ / H , then τ 0 ( x λ ( g ) ∗ ) = τ 0 ( x E M ( λ ( g ) ) ∗ ) = 0 \tau_{0}(x\lambda(g)^{*}) = \tau_{0}(xE_{\mathcal{M}}(\lambda(g))^{*}) = 0 τ 0 ( x λ ( g ) ∗ ) = τ 0 ( x E M ( λ ( g ) ) ∗ ) = 0 . Thus all Fourier coefficients of x x x outside H H H vanish, so E L ( H ) ( x ) = x E_{L(H)}(x) = x E L ( H ) ( x ) = x . Hence x ∈ L ( H ) x \in L(H) x ∈ L ( H ) , proving M ⊆ L ( H ) \mathcal{M} \subseteq L(H) M ⊆ L ( H ) and therefore M = L ( H ) \mathcal{M} = L(H) M = L ( H ) . Finally, Proposition 3.3 shows that H H H is boomerang. □
We now extend the argument to finite products. We follow [11 Theorem 5.4 and Remark 5.5] , using the preceding thickness and Fourier support arguments in each coordinate. By recurrence, it suffices to prove the corresponding statement for boomerang subalgebras.
Theorem 5.5. Let n ≥ 2 n \geq2 n ≥ 2 and let Γ i \Gamma_i Γ i be countable acylindrically hyperbolic groups with trivial amenable radical. Set Γ = ⨁ 1 ≤ i ≤ n Γ i \Gamma= \bigoplus_{1 \leq i \leq n}\Gamma_i Γ = ⨁ 1 ≤ i ≤ n Γ i . Then every boomerang von Neumann subalgebra M ≤ L ( Γ ) \mathcal{M} \leq L(\Gamma) M ≤ L ( Γ ) is L ( H ) L(H) L ( H ) for a unique boomerang subgroup H ≤ Γ H \leq\Gamma H ≤ Γ .
Proof. Write [ n ] = { 1 , 2 , … , n } [n] = \{1,2,\ldots,n\} [ n ] = { 1 , 2 , … , n } . For i ∈ [ n ] i \in[n] i ∈ [ n ] , write K l o x i K^{i}_{\mathrm{lox}} K lox i for the generalized loxodromic elements of Γ i \Gamma_i Γ i . For ∅ ≠ I ⊆ [ n ] \varnothing\ne I \subseteq[n] ∅ = I ⊆ [ n ] , set K l o x I = { ( s i ) ∈ Γ : s i ∈ K l o x i for i ∈ I , s i = e for i ∉ I } K^{I}_{\mathrm{lox}} = \{(s_i) \in\Gamma: s_i \in K^{i}_{\mathrm{lox}} \text{ for } i \in I,\ s_i = e \text{ for } i \notin I\} K lox I = {( s i ) ∈ Γ : s i ∈ K lox i for i ∈ I , s i = e for i ∈ / I } . The case M = C 1 \mathcal{M} = \mathbb{C}1 M = C 1 is immediate. From now on, assume that M ≠ C 1 \mathcal{M} \ne\mathbb{C}1 M = C 1 . Set ϕ M ( s ) = τ 0 ( E M ( λ ( s ) ) λ ( s ) ∗ ) \phi_{\mathcal{M}}(s) = \tau_0(E_{\mathcal{M}}(\lambda(s))\lambda(s)^*) ϕ M ( s ) = τ 0 ( E M ( λ ( s )) λ ( s ) ∗ ) . By Lemma 3.8 , ϕ M \phi_{\mathcal{M}} ϕ M is a boomerang character. We first observe that, for every J ⊆ [ n ] J \subseteq[n] J ⊆ [ n ] , E M ( L ( ⨁ i ∈ J Γ i ) ) ⊆ L ( ⨁ i ∈ J Γ i ) E_{\mathcal{M}}(L(\bigoplus_{i \in J}\Gamma_i)) \subseteq L(\bigoplus_{i \in J}\Gamma_i) E M ( L ( ⨁ i ∈ J Γ i )) ⊆ L ( ⨁ i ∈ J Γ i ) . Indeed, for every i i i , Lemma 5.6 [14 ] gives
⋂ t ∈ Γ i ⋂ k ≥ 1 C Γ i ( t k ) ⊆ ⋂ h ∈ K l o x i E Γ i ( h ) = { e } , \bigcap_{t\in\Gamma_i}\bigcap_{k\geq1} C_{\Gamma_i}(t^k)\subseteq\bigcap_{h\in K^i_{\mathrm{lox}}} E_{\Gamma_i}(h)=\{e\}, t ∈ Γ i ⋂ k ≥ 1 ⋂ C Γ i ( t k ) ⊆ h ∈ K lox i ⋂ E Γ i ( h ) = { e } ,
where the last intersection is taken over the loxodromic elements of a fixed non-elementary acylindrical action of Γ i \Gamma_i Γ i . If s ∈ ⨁ i ∈ J Γ i s \in\bigoplus_{i\in J}\Gamma_i s ∈ ⨁ i ∈ J Γ i and i ∉ J i \notin J i ∈ / J , every element of Γ i \Gamma_i Γ i centralizes s s s . Proposition 3.5 therefore forces the i i i th coordinate of every element of supp ( E M ( λ ( s ) ) ) \operatorname{supp}(\mathcal{E}_{\mathcal{M}}(\lambda(s))) supp ( E M ( λ ( s ))) to be e e e . This proves the inclusion on group unitaries, hence everywhere by normality. Consequently, E M ∣ L ( ⨁ i ∈ J Γ i ) = E M ∩ L ( ⨁ i ∈ J Γ i ) \mathcal{E}_{\mathcal{M}}|_{L(\bigoplus_{i\in J}\Gamma_i)}=\mathcal{E}_{\mathcal{M}\cap L(\bigoplus_{i\in J}\Gamma_i)} E M ∣ L ( ⨁ i ∈ J Γ i ) = E M ∩ L ( ⨁ i ∈ J Γ i ) . Moreover, M ∩ L ( ⨁ i ∈ J Γ i ) \mathcal{M}\cap L(\bigoplus_{i\in J}\Gamma_i) M ∩ L ( ⨁ i ∈ J Γ i ) is boomerang in L ( ⨁ i ∈ J Γ i ) L(\bigoplus_{i\in J}\Gamma_i) L ( ⨁ i ∈ J Γ i ) . Indeed, for conjugation by any element of ⨁ i ∈ J Γ i \bigoplus_{i\in J}\Gamma_i ⨁ i ∈ J Γ i , restrict the converging conditional expectations supplied by recurrence of M \mathcal{M} M to this coordinate subalgebra.
Claim 1. If E M ( λ ( s ) ) = 0 \mathcal{E}_{\mathcal{M}}(\lambda(s))=0 E M ( λ ( s )) = 0 for every s ∈ ⋃ ∅ ≠ I ⊆ [ n ] K l o x I s\in\bigcup_{\varnothing\ne I\subseteq[n]}K_{\mathrm{lox}}^I s ∈ ⋃ ∅ = I ⊆ [ n ] K lox I , then M = C 1 \mathcal{M}=\mathbb{C}1 M = C 1 .
Fix e ≠ g = ( g i ) ∈ Γ e\ne g=(g_i)\in\Gamma e = g = ( g i ) ∈ Γ , and set J = { i : g i ≠ e } J=\{i:g_i\ne e\} J = { i : g i = e } . For each i ∈ J i\in J i ∈ J , choose a loxodromic h i ∈ Γ i h_i\in\Gamma_i h i ∈ Γ i such that g i ∉ E Γ i ( h i ) g_i\notin E_{\Gamma_i}(h_i) g i ∈ / E Γ i ( h i ) . As in the proof of Lemma 5.1, there is N ≥ 1 N\ge1 N ≥ 1 such that h i − k g i − 1 h i k g i ∈ K l o x i h_i^{-k}g_i^{-1}h_i^kg_i\in K_{\mathrm{lox}}^i h i − k g i − 1 h i k g i ∈ K lox i for every i ∈ J i\in J i ∈ J and k ≥ N k\ge N k ≥ N . Set h = ( h i ) h=(h_i) h = ( h i ) , with h i = e h_i=e h i = e for i ∉ J i\notin J i ∈ / J . Choose a recurrence sequence ( n k ) (n_k) ( n k ) for φ M \varphi_{\mathcal{M}} φ M under conjugation by h − 1 h^{-1} h − 1 , and pass to a subsequence with n k + 1 − n k ≥ N n_{k+1}-n_k\ge N n k + 1 − n k ≥ N . Then φ M ( h − n k g h n k ) → φ M ( g ) \varphi_{\mathcal{M}}(h^{-n_k}gh^{n_k})\to\varphi_{\mathcal{M}}(g) φ M ( h − n k g h n k ) → φ M ( g ) . For ℓ > k \ell>k ℓ > k , put m = n ℓ − n k ≥ N m=n_\ell-n_k\ge N m = n ℓ − n k ≥ N . The i i i th coordinate of h − n ℓ g − 1 h n ℓ h − n k g h n k h^{-n_\ell}g^{-1}h^{n_\ell}h^{-n_k}gh^{n_k} h − n ℓ g − 1 h n ℓ h − n k g h n k is h i − n k ( h i − m g i − 1 h i m g i ) h i n k h_i^{-n_k}(h_i^{-m}g_i^{-1}h_i^mg_i)h_i^{n_k} h i − n k ( h i − m g i − 1 h i m g i ) h i n k for i ∈ J i\in J i ∈ J , and is e e e otherwise. Thus this element belongs to K l o x J K_{\mathrm{lox}}^J K lox J . For k > ℓ k>\ell k > ℓ , use the inverse of the element with the indices reversed. Hence ⋃ ∅ ≠ I ⊆ [ n ] K l o x I \bigcup_{\varnothing\ne I\subseteq[n]}K_{\mathrm{lox}}^I ⋃ ∅ = I ⊆ [ n ] K lox I is φ M \varphi_{\mathcal{M}} φ M -thick. Lemma 2.3 gives φ M = δ e \varphi_{\mathcal{M}}=\delta_e φ M = δ e , and hence M = C 1 \mathcal{M}=\mathbb{C}1 M = C 1 .
Claim 2. For every ∅ ≠ I ⊆ [ n ] \varnothing\ne I\subseteq[n] ∅ = I ⊆ [ n ] and s ∈ K l o x I s\in K_{\mathrm{lox}}^I s ∈ K lox I , one has E M ( λ ( s ) ) ∈ { 0 , λ ( s ) } \mathcal{E}_{\mathcal{M}}(\lambda(s))\in\{0,\lambda(s)\} E M ( λ ( s )) ∈ { 0 , λ ( s )} .
Assume that E M ( λ ( s ) ) ≠ 0 \mathcal{E}_{\mathcal{M}}(\lambda(s))\ne0 E M ( λ ( s )) = 0 . Put A i = E Γ i ( s i ) A_i=E_{\Gamma_i}(s_i) A i = E Γ i ( s i ) for i ∈ I i\in I i ∈ I . Proposition 3.5 and preservation of the coordinate subalgebras give E M ( λ ( s ) ) ∈ L ( ⨁ i ∈ I A i ) \mathcal{E}_{\mathcal{M}}(\lambda(s))\in L(\bigoplus_{i\in I}A_i) E M ( λ ( s )) ∈ L ( ⨁ i ∈ I A i ) . By the proof of [1 Theorem 1] , choose infinite-order elements t i ∈ Γ i t_i\in\Gamma_i t i ∈ Γ i such that ⟨ A i , t i ⟩ ≅ A i ∗ ⟨ t i ⟩ \langle A_i,t_i\rangle\cong A_i*\langle t_i\rangle ⟨ A i , t i ⟩ ≅ A i ∗ ⟨ t i ⟩ under the canonical homomorphism. Set t = ( t i ) t=(t_i) t = ( t i ) , with t i = e t_i=e t i = e for i ∉ I i\notin I i ∈ / I . By Lemma 3.6, choose m ≥ 1 m\ge1 m ≥ 1 such that E M ( λ ( t − m s t m ) ) ≠ 0 \mathcal{E}_{\mathcal{M}}(\lambda(t^{-m}st^m))\ne0 E M ( λ ( t − m s t m )) = 0 . Set b i = t i − m s i t i m b_i=t_i^{-m}s_it_i^m b i = t i − m s i t i m and B i = t i − m A i t i m B_i=t_i^{-m}A_it_i^m B i = t i − m A i t i m for i ∈ I i\in I i ∈ I , and set b = t − m s t m b=t^{-m}st^m b = t − m s t m . Then B i = E Γ i ( b i ) B_i=E_{\Gamma_i}(b_i) B i = E Γ i ( b i ) and ⟨ A i , B i ⟩ ≅ A i ∗ B i \langle A_i,B_i\rangle\cong A_i*B_i ⟨ A i , B i ⟩ ≅ A i ∗ B i , as in the proof of Proposition 5.3. In particular, A i ∩ B i = { e } A_i\cap B_i=\{e\} A i ∩ B i = { e } . Write E M ( λ ( s ) ) = ∑ x ∈ ⨁ i ∈ I A i c x λ ( x ) \mathcal{E}_{\mathcal{M}}(\lambda(s))=\sum_{x\in\bigoplus_{i\in I}A_i}c_x\lambda(x) E M ( λ ( s )) = ∑ x ∈ ⨁ i ∈ I A i c x λ ( x ) , E M ( λ ( b ) ) = ∑ y ∈ ⨁ i ∈ I B i d y λ ( y ) \mathcal{E}_{\mathcal{M}}(\lambda(b))=\sum_{y\in\bigoplus_{i\in I}B_i}d_y\lambda(y) E M ( λ ( b )) = ∑ y ∈ ⨁ i ∈ I B i d y λ ( y ) . If x i = e x_i=e x i = e for some i ∈ I i\in I i ∈ I , then preservation of the coordinate subalgebras gives c x = τ 0 ( λ ( s ) E M ( λ ( x ) ∗ ) ) = 0 c_x=\tau_0(\lambda(s)\mathcal{E}_{\mathcal{M}}(\lambda(x)^*))=0 c x = τ 0 ( λ ( s ) E M ( λ ( x ) ∗ )) = 0 , since s i ≠ e s_i\ne e s i = e . Similarly, d y = 0 d_y=0 d y = 0 if y i = e y_i=e y i = e for some i ∈ I i\in I i ∈ I . By bimodularity,
∑ x , y c x d y λ ( x y ) = E M ( λ ( s ) ) E M ( λ ( b ) ) = E M ( λ ( s ) E M ( λ ( b ) ) ) = ∑ y ′ d y ′ E M ( λ ( s y ′ ) ) . (2) \begin{aligned}
\sum_{x,y}c_xd_y\lambda(xy)&=\mathcal{E}_{\mathcal{M}}(\lambda(s))\mathcal{E}_{\mathcal{M}}(\lambda(b))\\
&=\mathcal{E}_{\mathcal{M}}\bigl(\lambda(s)\mathcal{E}_{\mathcal{M}}(\lambda(b))\bigr)\\
&=\sum_{y'}d_{y'}\mathcal{E}_{\mathcal{M}}(\lambda(sy')). \tag*{(2)}
\end{aligned} x , y ∑ c x d y λ ( x y ) = E M ( λ ( s )) E M ( λ ( b )) = E M ( λ ( s ) E M ( λ ( b )) ) = y ′ ∑ d y ′ E M ( λ ( s y ′ )) . (2) Fix x , y x,y x , y with x i , y i ≠ e x_i,y_i\ne e x i , y i = e for every i ∈ I i\in I i ∈ I , and suppose that x i ≠ s i x_i\ne s_i x i = s i for some i ∈ I i\in I i ∈ I . For a term on the right with d y ′ ≠ 0 d_{y'}\ne0 d y ′ = 0 , all coordinates y j ′ y'_j y j ′ are nonidentity, so s j y j ′ s_jy'_j s j y j ′ is reduced and nontrivial in A j ∗ B j A_j*B_j A j ∗ B j . If x y ∈ supp ( E M ( λ ( s y ′ ) ) ) xy\in\operatorname{supp}(\mathcal{E}_{\mathcal{M}}(\lambda(sy'))) x y ∈ supp ( E M ( λ ( s y ′ ))) , Proposition 3.5 gives k ≥ 1 k\ge1 k ≥ 1 such that ( x i y i ) ( s i y i ′ ) k = ( s i y i ′ ) k ( x i y i ) (x_iy_i)(s_iy'_i)^k=(s_iy'_i)^k(x_iy_i) ( x i y i ) ( s i y i ′ ) k = ( s i y i ′ ) k ( x i y i ) . Both sides are reduced words in A i ∗ B i A_i*B_i A i ∗ B i . Their first syllables are x i x_i x i and s i s_i s i , respectively, a contradiction. Comparing Fourier coefficients in (2) gives c x d y = 0 c_xd_y=0 c x d y = 0 whenever x ≠ s x\ne s x = s and all coordinates of x , y x,y x , y are nonidentity. Since the expectation of λ ( b ) \lambda(b) λ ( b ) is nonzero, choose y y y with d y ≠ 0 d_y\ne0 d y = 0 . It follows that c x = 0 c_x=0 c x = 0 for
every x ≠ s x \ne s x = s . Thus 0 ≠ E M ( λ ( s ) ) = c s λ ( s ) 0 \ne\mathbb{E}_{\mathcal{M}}(\lambda(s)) = c_s\lambda(s) 0 = E M ( λ ( s )) = c s λ ( s ) . Applying E M \mathbb{E}_{\mathcal{M}} E M once more gives c s = c s 2 c_s = c_s^2 c s = c s 2 , hence c s = 1 c_s = 1 c s = 1 . This proves Claim 2.
By Claims 1 and 2 , there is some s ∈ K l o x I s \in K_{\mathrm{lox}}^I s ∈ K lox I with λ ( s ) ∈ M \lambda(s) \in\mathcal{M} λ ( s ) ∈ M . Write I \mathcal{I} I for the set of all nonempty I ⊆ [ n ] I \subseteq[n] I ⊆ [ n ] for which such an s s s exists.
Claim 3. For each I ∈ I I \in\mathcal{I} I ∈ I and i ∈ I i \in I i ∈ I , there is a i ∈ K l o x i a_i \in K_{\mathrm{lox}}^i a i ∈ K lox i with λ ( a i ) ∈ M ∩ L ( Γ i ) \lambda(a_i) \in\mathcal{M} \cap L(\Gamma_i) λ ( a i ) ∈ M ∩ L ( Γ i ) . Moreover, I \mathcal{I} I contains a unique maximal element.
Fix s ∈ K l o x I s \in K_{\mathrm{lox}}^I s ∈ K lox I with λ ( s ) ∈ M \lambda(s) \in\mathcal{M} λ ( s ) ∈ M , and fix i ∈ I i \in I i ∈ I . Choose a loxodromic t i ∈ Γ i t_i \in\Gamma_i t i ∈ Γ i with s i ∉ E Γ i ( t i ) s_i \notin E_{\Gamma_i}(t_i) s i ∈ / E Γ i ( t i ) . As in the proof of Lemma 5.1 , t i − m s i − 1 t i m s i t_i^{-m}s_i^{-1}t_i^m s_i t i − m s i − 1 t i m s i is loxodromic for every sufficiently large m m m . Regard t i t_i t i as an element of Γ \Gamma Γ supported in coordinate i i i . By Lemma 3.6 , choose such an m m m with E M ( λ ( t i − m s i − 1 t i m ) ) ≠ 0 \mathbb{E}_{\mathcal{M}}(\lambda(t_i^{-m}s_i^{-1}t_i^m)) \ne0 E M ( λ ( t i − m s i − 1 t i m )) = 0 . Since t i − m s i − 1 t i m ∈ K l o x I t_i^{-m}s_i^{-1}t_i^m \in K_{\mathrm{lox}}^I t i − m s i − 1 t i m ∈ K lox I , Claim 2 gives λ ( t i − m s i − 1 t i m ) ∈ M \lambda(t_i^{-m}s_i^{-1}t_i^m) \in\mathcal{M} λ ( t i − m s i − 1 t i m ) ∈ M . Multiplying by λ ( s ) \lambda(s) λ ( s ) gives λ ( t i − m s i − 1 t i m s ) = λ ( t i − m s i − 1 t i m s i ) ∈ M ∩ L ( Γ i ) \lambda(t_i^{-m}s_i^{-1}t_i^m s) = \lambda(t_i^{-m}s_i^{-1}t_i^m s_i) \in\mathcal{M} \cap L(\Gamma_i) λ ( t i − m s i − 1 t i m s ) = λ ( t i − m s i − 1 t i m s i ) ∈ M ∩ L ( Γ i ) . Set a i = t i − m s i − 1 t i m s i a_i = t_i^{-m}s_i^{-1}t_i^m s_i a i = t i − m s i − 1 t i m s i . This proves the first assertion. If I 1 , I 2 ∈ I I_1,I_2 \in\mathcal{I} I 1 , I 2 ∈ I , choose such an a i a_i a i for every i ∈ I 1 ∪ I 2 i \in I_1 \cup I_2 i ∈ I 1 ∪ I 2 . Their product belongs to K l o x I 1 ∪ I 2 K_{\mathrm{lox}}^{I_1\cup I_2} K lox I 1 ∪ I 2 , and its unitary belongs to M \mathcal{M} M . Thus I 1 ∪ I 2 ∈ I I_1 \cup I_2 \in\mathcal{I} I 1 ∪ I 2 ∈ I . Since [ n ] [n] [ n ] is finite, I \mathcal{I} I has a unique maximal element. From now on, denote it by I I I .
Claim 4. M ⊆ L ( ⨁ i ∈ I Γ i ) \mathcal{M} \subseteq L\left(\bigoplus_{i\in I}\Gamma_i\right) M ⊆ L ( ⨁ i ∈ I Γ i ) .
If i ∉ I i \notin I i ∈ / I , then M ∩ L ( Γ i ) = C 1 \mathcal{M} \cap L(\Gamma_i) = \mathbb{C}1 M ∩ L ( Γ i ) = C 1 . Indeed, this intersection is boomerang, as shown above. If it were nontrivial, Corollary 5.2 and Claim 2 would give a i ∈ K l o x i a_i \in K_{\mathrm{lox}}^i a i ∈ K lox i with λ ( a i ) ∈ M \lambda(a_i) \in\mathcal{M} λ ( a i ) ∈ M , contradicting maximality of I I I . It follows that φ M ∣ Γ i = δ e \varphi_{\mathcal{M}|_{\Gamma_i}} = \delta_e φ M ∣ Γ i = δ e for every i ∉ I i \notin I i ∈ / I .
Fix g = ( g j ) ∈ Γ g=(g_j) \in\Gamma g = ( g j ) ∈ Γ with g i ≠ e g_i \ne e g i = e for some i ∉ I i \notin I i ∈ / I . Choose a loxodromic h ∈ Γ i h \in\Gamma_i h ∈ Γ i with g i ∉ E Γ i ( h ) g_i \notin E_{\Gamma_i}(h) g i ∈ / E Γ i ( h ) , and choose N ≥ 1 N \ge1 N ≥ 1 such that h − m g i − 1 h m g i ∈ K l o x i h^{-m}g_i^{-1}h^m g_i \in K_{\mathrm{lox}}^i h − m g i − 1 h m g i ∈ K lox i for every m ≥ N m \ge N m ≥ N . Regard h h h as supported in coordinate i i i . Choose a recurrence sequence ( n k ) (n_k) ( n k ) for φ M \varphi_{\mathcal{M}} φ M under conjugation by h − 1 h^{-1} h − 1 with n k + 1 − n k ≥ N n_{k+1}-n_k \ge N n k + 1 − n k ≥ N . Then φ M ( h − n k g h n k ) → φ M ( g ) \varphi_{\mathcal{M}}(h^{-n_k}gh^{n_k}) \to\varphi_{\mathcal{M}}(g) φ M ( h − n k g h n k ) → φ M ( g ) , and, for k ≠ ℓ k \ne\ell k = ℓ , h − n ℓ g − 1 h n ℓ h − n k g h n k ∈ K l o x i ⊆ Γ i h^{-n_\ell}g^{-1}h^{n_\ell}h^{-n_k}gh^{n_k} \in K_{\mathrm{lox}}^i \subseteq\Gamma_i h − n ℓ g − 1 h n ℓ h − n k g h n k ∈ K lox i ⊆ Γ i . The pairwise differences therefore have φ M \varphi_{\mathcal{M}} φ M -value zero. The Bessel argument in Subsection 2.4 gives φ M ( g ) = 0 \varphi_{\mathcal{M}}(g)=0 φ M ( g ) = 0 . Thus E M ( λ ( g ) ) = 0 \mathbb{E}_{\mathcal{M}}(\lambda(g))=0 E M ( λ ( g )) = 0 whenever g ∉ ⨁ i ∈ I Γ i g \notin\bigoplus_{i\in I}\Gamma_i g ∈ / ⨁ i ∈ I Γ i . For x ∈ M x \in\mathcal{M} x ∈ M , all Fourier coefficients outside this subgroup consequently vanish. This proves Claim 4.
For every i ∈ I i \in I i ∈ I , Claim 3 gives a i ∈ K l o x i a_i \in K_{\mathrm{lox}}^i a i ∈ K lox i with λ ( a i ) ∈ M ∩ L ( Γ i ) \lambda(a_i) \in\mathcal{M} \cap L(\Gamma_i) λ ( a i ) ∈ M ∩ L ( Γ i ) . Apply the proof of Proposition 5.3 to the boomerang subalgebra M ∩ L ( Γ i ) \mathcal{M}\cap L(\Gamma_i) M ∩ L ( Γ i ) and a i a_i a i . Together with Claim 2, it gives b i ∈ K l o x i b_i \in K_{\mathrm{lox}}^i b i ∈ K lox i such that λ ( b i ) ∈ M ∩ L ( Γ i ) \lambda(b_i) \in\mathcal{M} \cap L(\Gamma_i) λ ( b i ) ∈ M ∩ L ( Γ i ) , E Γ i ( a i ) ∩ E Γ i ( b i ) = { e } E_{\Gamma_i}(a_i) \cap E_{\Gamma_i}(b_i) = \{e\} E Γ i ( a i ) ∩ E Γ i ( b i ) = { e } . Fix g = ( g i ) ∈ ⨁ i ∈ I Γ i g=(g_i) \in\bigoplus_{i\in I}\Gamma_i g = ( g i ) ∈ ⨁ i ∈ I Γ i . For every i ∈ I i \in I i ∈ I with g i ≠ e g_i \ne e g i = e , choose r i ∈ { a i , b i } r_i \in\{a_i,b_i\} r i ∈ { a i , b i } such that g i ∉ E Γ i ( r i ) g_i \notin E_{\Gamma_i}(r_i) g i ∈ / E Γ i ( r i ) . By [22 Corollary 6.12] , choose k i ∈ Z k_i \in\mathbb{Z} k i ∈ Z such that g i r i k i ∈ K l o x i g_i r_i^{k_i} \in K_{\mathrm{lox}}^i g i r i k i ∈ K lox i . If g i = e g_i=e g i = e , take r i = a i r_i=a_i r i = a i and k i = 1 k_i=1 k i = 1 . Set r = ( r i k i ) i ∈ I r=(r_i^{k_i})_{i\in I} r = ( r i k i ) i ∈ I , with identity coordinates outside I I I . Since every λ ( r i ) \lambda(r_i) λ ( r i ) belongs to M \mathcal{M} M , one has λ ( r ) ∈ M \lambda(r) \in\mathcal{M} λ ( r ) ∈ M . Moreover, g r ∈ K l o x I gr \in K_{\mathrm{lox}}^I g r ∈ K lox I . Bimodularity and Claim 2 give E M ( λ ( g ) ) λ ( r ) = E M ( λ ( g r ) ) ∈ { 0 , λ ( g r ) } \mathbb{E}_{\mathcal{M}}(\lambda(g))\lambda(r)=\mathbb{E}_{\mathcal{M}}(\lambda(gr))\in\{0,\lambda(gr)\} E M ( λ ( g )) λ ( r ) = E M ( λ ( g r )) ∈ { 0 , λ ( g r )} . Hence E M ( λ ( g ) ) ∈ { 0 , λ ( g ) } \mathbb{E}_{\mathcal{M}}(\lambda(g))\in\{0,\lambda(g)\} E M ( λ ( g )) ∈ { 0 , λ ( g )} . Together with Claim 4, this holds for every g ∈ Γ g\in\Gamma g ∈ Γ . Set H = { g ∈ Γ : λ ( g ) ∈ M } H=\{g\in\Gamma:\lambda(g)\in\mathcal{M}\} H = { g ∈ Γ : λ ( g ) ∈ M } . Then H H H is a subgroup and L ( H ) ⊆ M L(H)\subseteq\mathcal{M} L ( H ) ⊆ M . For x ∈ M x\in\mathcal{M} x ∈ M and g ∉ H g\notin H g ∈ / H , τ 0 ( x λ ( g ) ∗ ) = τ 0 ( x E M ( λ ( g ) ) ∗ ) = 0 \tau_0(x\lambda(g)^*)=\tau_0(x\mathbb{E}_{\mathcal{M}}(\lambda(g))^*)=0 τ 0 ( x λ ( g ) ∗ ) = τ 0 ( x E M ( λ ( g ) ) ∗ ) = 0 . Thus x ∈ L ( H ) x\in L(H) x ∈ L ( H ) , so M = L ( H ) \mathcal{M}=L(H) M = L ( H ) . The definition of H H H gives uniqueness, and Proposition 3.3 shows that H H H is boomerang. ∎
Proof of Theorem 1.1 . By the recurrence argument in the introduction, μ \mu μ -almost every subalgebra is boomerang. For n = 1 n=1 n = 1 , apply Theorem 5.4 ; for n ≥ 2 n\ge2 n ≥ 2 , apply Theorem 5.5 .
Thus μ ( L ( Sub ( Γ ) ) ) = 1 \mu(L(\operatorname{Sub}(\Gamma)))=1 μ ( L ( Sub ( Γ ))) = 1 . Set ν = J ∗ μ \nu=J_*\mu ν = J ∗ μ . By Proposition 3.2, ν \nu ν is an IRS of Γ \Gamma Γ and L ∗ ν = μ L_*\nu=\mu L ∗ ν = μ . Since J L = id JL=\operatorname{id} J L = id , ν \nu ν is unique. If μ \mu μ is ergodic, then ν \nu ν is ergodic by equivariance of J J J .
Invariant random algebras of mixed identity-free groups
We prove IRA rigidity for mixed identity-free groups and their finite products. We use homomorphisms fixing the original group pointwise to construct an IRA of a larger group, and then apply the results of the preceding section. We denote by UCP τ 0 ( L ( Γ ) ) \operatorname{UCP}_{\tau_0}(L(\Gamma)) UCP τ 0 ( L ( Γ )) the collection of all trace preserving UCP maps on L ( Γ ) L(\Gamma) L ( Γ ) .
We first record the measurability and equivariance properties of the maps used below.
Lemma 6.1. Let Γ \Gamma Γ be a countable group.
For M ∈ SA ( L ( Γ ) ) M\in\operatorname{SA}(L(\Gamma)) M ∈ SA ( L ( Γ )) , let ψ M ( g , h ) = τ 0 ( E M ( λ ( g ) ) λ ( h ) ∗ ) \psi_M(g,h)=\tau_0(E_M(\lambda(g))\lambda(h)^*) ψ M ( g , h ) = τ 0 ( E M ( λ ( g )) λ ( h ) ∗ ) . Then ψ M ∈ PD 1 ( Γ × Γ ) \psi_M\in\operatorname{PD}_1(\Gamma\times\Gamma) ψ M ∈ PD 1 ( Γ × Γ ) , and M ↦ ψ M M\mapsto\psi_M M ↦ ψ M is continuous and injective.
The map f : UCP τ 0 ( L ( Γ ) ) → SA ( L ( Γ ) ) f:\operatorname{UCP}_{\tau_0}(L(\Gamma))\to\operatorname{SA}(L(\Gamma)) f : UCP τ 0 ( L ( Γ )) → SA ( L ( Γ )) that sends Φ ↦ Fix ( Φ ) \Phi\mapsto\operatorname{Fix}(\Phi) Φ ↦ Fix ( Φ ) is Borel. Here, UCP τ 0 ( L ( Γ ) ) \operatorname{UCP}_{\tau_0}(L(\Gamma)) UCP τ 0 ( L ( Γ )) is equipped with point ultraweak topology.
Let Λ \Lambda Λ be a countable group containing Γ \Gamma Γ , and let ρ : Λ → Γ \rho:\Lambda\to\Gamma ρ : Λ → Γ be a homomorphism with ρ ∣ Γ = id Γ \rho|_\Gamma=\operatorname{id}_\Gamma ρ ∣ Γ = id Γ . Define ψ ρ , M ( g , h ) : = ψ M ( ρ ( g ) , ρ ( h ) ) \psi_{\rho,M}(g,h):=\psi_M(\rho(g),\rho(h)) ψ ρ , M ( g , h ) := ψ M ( ρ ( g ) , ρ ( h )) . Equip SA ( L ( Γ ) ) \operatorname{SA}(L(\Gamma)) SA ( L ( Γ )) with the Λ \Lambda Λ -action s ⋅ M = λ ( ρ ( s ) ) M λ ( ρ ( s ) ) ∗ s\cdot M=\lambda(\rho(s))M\lambda(\rho(s))^* s ⋅ M = λ ( ρ ( s )) M λ ( ρ ( s ) ) ∗ . Then ψ ρ , M ∣ Γ × Γ = ψ M \psi_{\rho,M}|_{\Gamma\times\Gamma}=\psi_M ψ ρ , M ∣ Γ × Γ = ψ M . Moreover, pd : SA ( L ( Γ ) ) → PD 1 ( Λ × Λ ) \operatorname{pd}:\operatorname{SA}(L(\Gamma))\to\operatorname{PD}_1(\Lambda\times\Lambda) pd : SA ( L ( Γ )) → PD 1 ( Λ × Λ ) that sends M ↦ ψ ρ , M M\mapsto\psi_{\rho,M} M ↦ ψ ρ , M is measurable and equivariant, measurable and Λ \Lambda Λ -equivariant, where the action on PD 1 ( Λ × Λ ) \operatorname{PD}_1(\Lambda\times\Lambda) PD 1 ( Λ × Λ ) is ( s ⋅ ψ ) ( g , h ) = ψ ( s − 1 g s , s − 1 h s ) (s\cdot\psi)(g,h)=\psi(s^{-1}gs,s^{-1}hs) ( s ⋅ ψ ) ( g , h ) = ψ ( s − 1 g s , s − 1 h s ) . In particular, for every IRA μ \mu μ of Γ \Gamma Γ , one has pd ∗ μ ∈ Prob Λ ( PD 1 ( Λ × Λ ) ) \operatorname{pd}_*\mu\in\operatorname{Prob}_\Lambda(\operatorname{PD}_1(\Lambda\times\Lambda)) pd ∗ μ ∈ Prob Λ ( PD 1 ( Λ × Λ )) .
Proof. The left-right representation on L 2 ( L ( Γ ) ) ⊗ M L 2 ( L ( Γ ) ) L^2(L(\Gamma))\otimes_M L^2(L(\Gamma)) L 2 ( L ( Γ )) ⊗ M L 2 ( L ( Γ )) has coefficient ψ M \psi_M ψ M at the unit vector 1 ⊗ 1 1\otimes1 1 ⊗ 1 ; see [13 Section 13.1.2] . Thus ψ M \psi_M ψ M is normalized and positive definite. If M i → M M_i\to M M i → M , then ∣ ψ M i ( g , h ) − ψ M ( g , h ) ∣ ≤ ∥ E M i ( λ ( g ) ) − E M ( λ ( g ) ) ∥ 2 → 0 |\psi_{M_i}(g,h)-\psi_M(g,h)|\leq\|E_{M_i}(\lambda(g))-E_M(\lambda(g))\|_2\to0 ∣ ψ M i ( g , h ) − ψ M ( g , h ) ∣ ≤ ∥ E M i ( λ ( g )) − E M ( λ ( g )) ∥ 2 → 0 . Moreover, ψ M = ψ N \psi_M=\psi_N ψ M = ψ N implies E M ( λ ( g ) ) = E N ( λ ( g ) ) E_M(\lambda(g))=E_N(\lambda(g)) E M ( λ ( g )) = E N ( λ ( g )) for every g g g , hence M = N M=N M = N . For the second assertion, Fix ( Φ ) \operatorname{Fix}(\Phi) Fix ( Φ ) is a von Neumann subalgebra by [29 Lemma 4.1] . The Effros–Maréchal topology on SA ( L ( Γ ) ) \operatorname{SA}(L(\Gamma)) SA ( L ( Γ )) is the smallest topology for which the maps M ↦ ∥ E M ( x ) ∥ 2 M\mapsto\|E_M(x)\|_2 M ↦ ∥ E M ( x ) ∥ 2 , x ∈ L ( Γ ) x\in L(\Gamma) x ∈ L ( Γ ) are continuous. Choose a countable L 2 L^2 L 2 -dense family of elements x x x . Thus we need to show that { ϕ : c 1 < ∥ E Fix ( ϕ ) ( x ) ∥ 2 < c 2 } \{\phi:c_1<\|E_{\operatorname{Fix}(\phi)}(x)\|_2<c_2\} { ϕ : c 1 < ∥ E Fix ( ϕ ) ( x ) ∥ 2 < c 2 } is Borel for every x ∈ L ( Γ ) x\in L(\Gamma) x ∈ L ( Γ ) and c 1 < c 2 c_1<c_2 c 1 < c 2 . We first work with pointwise L 2 L^2 L 2 convergence. By the mean ergodic theorem, we see that ∥ E Fix ( ϕ ) ( x ) ∥ 2 = inf n ≥ 1 a n , ϕ ( x ) \|E_{\operatorname{Fix}(\phi)}(x)\|_2=\inf_{n\geq1}a_{n,\phi}(x) ∥ E Fix ( ϕ ) ( x ) ∥ 2 = inf n ≥ 1 a n , ϕ ( x ) , where a n , ϕ ( x ) : = ∥ 1 n ∑ k = 1 n ϕ k ( x ) ∥ 2 a_{n,\phi}(x):=\left\|\frac{1}{n}\sum_{k=1}^n\phi^k(x)\right\|_2 a n , ϕ ( x ) := n 1 ∑ k = 1 n ϕ k ( x ) 2 . Note that inf n ≥ 1 a n , ϕ ( x ) < c 2 ⟺ ∃ N ≥ 1 ∃ n ≥ 1 : a n , ϕ ( x ) < c 2 − 1 N \inf_{n\geq1}a_{n,\phi}(x)<c_2\Longleftrightarrow\exists N\geq1\ \exists n\geq1:a_{n,\phi}(x)<c_2-\frac{1}{N} inf n ≥ 1 a n , ϕ ( x ) < c 2 ⟺ ∃ N ≥ 1 ∃ n ≥ 1 : a n , ϕ ( x ) < c 2 − N 1 , and inf n ≥ 1 a n , ϕ ( x ) > c 1 ⟺ ∃ N ≥ 1 ∀ n ≥ 1 : a n , ϕ ( x ) ≥ c 1 + 1 N \inf_{n\geq1}a_{n,\phi}(x)>c_1\Longleftrightarrow\exists N\geq1\ \forall n\geq1:a_{n,\phi}(x)\geq c_1+\frac{1}{N} inf n ≥ 1 a n , ϕ ( x ) > c 1 ⟺ ∃ N ≥ 1 ∀ n ≥ 1 : a n , ϕ ( x ) ≥ c 1 + N 1 . Consequently,
{ ϕ : c 1 < ∥ E Fix ( ϕ ) ( x ) ∥ 2 < c 2 } = { ϕ : c 1 < inf n ≥ 1 a n , ϕ ( x ) } ∩ { ϕ : inf n ≥ 1 a n , ϕ ( x ) < c 2 } = [ ⋃ N ≥ 1 ⋂ n ≥ 1 { ϕ : a n , ϕ ( x ) ≥ c 1 + 1 N } ] ∩ [ ⋃ N ≥ 1 ⋃ n ≥ 1 { ϕ : a n , ϕ ( x ) < c 2 − 1 N } ] . \begin{aligned}
\left\{\phi:c_1<\|E_{\operatorname{Fix}(\phi)}(x)\|_2<c_2\right\} \\
&=\left\{\phi:c_1<\inf_{n\geq1}a_{n,\phi}(x)\right\}\cap\left\{\phi:\inf_{n\geq1}a_{n,\phi}(x)<c_2\right\} \\
&=\left[\bigcup_{N\geq1}\bigcap_{n\geq1}\left\{\phi:a_{n,\phi}(x)\geq c_1+\frac{1}{N}\right\}\right]\cap\left[\bigcup_{N\geq1}\bigcup_{n\geq1}\left\{\phi:a_{n,\phi}(x)<c_2-\frac{1}{N}\right\}\right].
\end{aligned} { ϕ : c 1 < ∥ E Fix ( ϕ ) ( x ) ∥ 2 < c 2 } = { ϕ : c 1 < n ≥ 1 inf a n , ϕ ( x ) } ∩ { ϕ : n ≥ 1 inf a n , ϕ ( x ) < c 2 } = [ N ≥ 1 ⋃ n ≥ 1 ⋂ { ϕ : a n , ϕ ( x ) ≥ c 1 + N 1 } ] ∩ [ N ≥ 1 ⋃ n ≥ 1 ⋃ { ϕ : a n , ϕ ( x ) < c 2 − N 1 } ] .
The sets inside the first curly bracket are closed, and those inside the second curly bracket are open, for pointwise L 2 L^2 L 2 convergence. Indeed, fix ψ \psi ψ , n ≥ 1 n \ge1 n ≥ 1 and ε > 0 \varepsilon> 0 ε > 0 , and put F = { ψ i ( x ) : 0 ≤ i < n } F = \{\psi^i(x) : 0 \le i < n\} F = { ψ i ( x ) : 0 ≤ i < n } . Suppose that ∥ ( ϕ − ψ ) ( z ) ∥ 2 < ε \|(\phi-\psi)(z)\|_2 < \varepsilon ∥ ( ϕ − ψ ) ( z ) ∥ 2 < ε for every z ∈ F z \in F z ∈ F . Using L 2 L^2 L 2 -contractivity and ϕ k ( x ) − ψ k ( x ) = ∑ i = 0 k − 1 ϕ k − 1 − i ( ( ϕ − ψ ) ( ψ i ( x ) ) ) \phi^k(x)-\psi^k(x) = \sum_{i=0}^{k-1}\phi^{k-1-i}((\phi-\psi)(\psi^i(x))) ϕ k ( x ) − ψ k ( x ) = ∑ i = 0 k − 1 ϕ k − 1 − i (( ϕ − ψ ) ( ψ i ( x ))) , we obtain that
∣ a n , ϕ ( x ) − a n , ψ ( x ) ∣ = 1 n ∣ ∥ ∑ k = 1 n ϕ k ( x ) ∥ 2 − ∥ ∑ k = 1 n ψ k ( x ) ∥ 2 ∣ ≤ 1 n ∥ ∑ k = 1 n ( ϕ k ( x ) − ψ k ( x ) ) ∥ 2 ≤ 1 n ∑ k = 1 n ∑ i = 0 k − 1 ∥ ( ϕ − ψ ) ( ψ i ( x ) ) ∥ 2 ≤ ε n ∑ k = 1 n k = ε ( n + 1 ) 2 . \begin{aligned}
|a_{n,\phi}(x)-a_{n,\psi}(x)| &= \frac{1}{n}\left|\left\|\sum_{k=1}^{n}\phi^k(x)\right\|_2-\left\|\sum_{k=1}^{n}\psi^k(x)\right\|_2\right| \\
&\le\frac{1}{n}\left\|\sum_{k=1}^{n}(\phi^k(x)-\psi^k(x))\right\|_2 \\
&\le\frac{1}{n}\sum_{k=1}^{n}\sum_{i=0}^{k-1}\|(\phi-\psi)(\psi^i(x))\|_2 \\
&\le\frac{\varepsilon}{n}\sum_{k=1}^{n} k = \frac{\varepsilon(n+1)}{2}.
\end{aligned} ∣ a n , ϕ ( x ) − a n , ψ ( x ) ∣ = n 1 k = 1 ∑ n ϕ k ( x ) 2 − k = 1 ∑ n ψ k ( x ) 2 ≤ n 1 k = 1 ∑ n ( ϕ k ( x ) − ψ k ( x )) 2 ≤ n 1 k = 1 ∑ n i = 0 ∑ k − 1 ∥ ( ϕ − ψ ) ( ψ i ( x )) ∥ 2 ≤ n ε k = 1 ∑ n k = 2 ε ( n + 1 ) .
Thus ϕ ↦ a n , ϕ ( x ) \phi\mapsto a_{n,\phi}(x) ϕ ↦ a n , ϕ ( x ) is continuous, proving the asserted closedness and openness. Hence Φ ↦ Fix ( Φ ) \Phi\mapsto\operatorname{Fix}(\Phi) Φ ↦ Fix ( Φ ) is Borel for pointwise L 2 L^2 L 2 convergence. Finally, the pointwise L 2 L^2 L 2 topology has a countable basis formed from finite intersections of sets { ϕ : ∥ ϕ ( z ) − y ∥ 2 < r } \{\phi: \|\phi(z)-y\|_2 < r\} { ϕ : ∥ ϕ ( z ) − y ∥ 2 < r } . Each such set is Borel for pointwise ultraweak convergence, since { ϕ : ∥ ϕ ( z ) − y ∥ 2 < r } = ⋃ m ≥ 1 { ϕ : ∥ ϕ ( z ) − y ∥ 2 ≤ r − 1 m } \{\phi: \|\phi(z)-y\|_2 < r\} = \bigcup_{m \ge1}\{\phi: \|\phi(z)-y\|_2 \le r-\frac{1}{m}\} { ϕ : ∥ ϕ ( z ) − y ∥ 2 < r } = ⋃ m ≥ 1 { ϕ : ∥ ϕ ( z ) − y ∥ 2 ≤ r − m 1 } , and the L 2 L^2 L 2 norm is ultraweakly lower semicontinuous. Therefore Φ ↦ Fix ( Φ ) \Phi\mapsto\operatorname{Fix}(\Phi) Φ ↦ Fix ( Φ ) is also Borel in point ultraweak topology (also see [7 Lemma 6.3] and [29 Section 4] ).
For the third assertion, observe that pd \operatorname{pd} pd is continuous by the first assertion. The restriction identity follows from ρ ∣ Γ = id Γ \rho|_{\Gamma} = \operatorname{id}_{\Gamma} ρ ∣ Γ = id Γ . Let μ \mu μ be an IRA of Γ \Gamma Γ . The action s ⋅ M = λ ( ρ ( s ) ) M λ ( ρ ( s ) ) ∗ s \cdot M = \lambda(\rho(s))M\lambda(\rho(s))^* s ⋅ M = λ ( ρ ( s )) M λ ( ρ ( s ) ) ∗ of Λ \Lambda Λ preserves μ \mu μ . It follows from Lemma 2.1 that
ψ ρ , λ ( ρ ( s ) ) M λ ( ρ ( s ) ) ∗ ( g , h ) = ψ ρ , M ( s − 1 g s , s − 1 h s ) . \psi_{\rho,\lambda(\rho(s))M\lambda(\rho(s))^*}(g,h) = \psi_{\rho,M}(s^{-1}gs,s^{-1}hs). ψ ρ , λ ( ρ ( s )) M λ ( ρ ( s ) ) ∗ ( g , h ) = ψ ρ , M ( s − 1 g s , s − 1 h s ) .
Thus, for every s ∈ Λ s \in\Lambda s ∈ Λ and bounded Borel function f f f ,
∫ f ( s ⋅ pd ( M ) ) d μ ( M ) = ∫ f ( pd ( λ ( ρ ( s ) ) M λ ( ρ ( s ) ) ∗ ) ) d μ ( M ) = ∫ f ( pd ( M ) ) d μ ( M ) . \int f(s \cdot\operatorname{pd}(M))\,d\mu(M) = \int f(\operatorname{pd}(\lambda(\rho(s))M\lambda(\rho(s))^*))\,d\mu(M) = \int f(\operatorname{pd}(M))\,d\mu(M). ∫ f ( s ⋅ pd ( M )) d μ ( M ) = ∫ f ( pd ( λ ( ρ ( s )) M λ ( ρ ( s ) ) ∗ )) d μ ( M ) = ∫ f ( pd ( M )) d μ ( M ) .
Hence pd ∗ μ \operatorname{pd}_*\mu pd ∗ μ is Λ \Lambda Λ -invariant. □
Recall that a nontrivial group Γ \Gamma Γ is mixed identity-free if every nonidentity h ∈ Γ ∗ F r h \in\Gamma* F_r h ∈ Γ ∗ F r , for every r ≥ 1 r \ge1 r ≥ 1 , survives under some homomorphism Γ ∗ F r → Γ \Gamma* F_r \to\Gamma Γ ∗ F r → Γ fixing Γ \Gamma Γ pointwise. Here F r F_r F r is the free group of rank r r r ; it suffices to take r = 1 r = 1 r = 1 by [35 Remark 5.1] . We now apply this construction to the homomorphisms supplied by mixed identity-freeness.
Theorem 6.2. Every countable nontrivial mixed identity-free group is IRA rigid.
Proof. Let μ \mu μ be an IRA of Γ \Gamma Γ , and enumerate the nonidentity elements of Λ = Γ ∗ ⟨ t ⟩ \Lambda= \Gamma* \langle t\rangle Λ = Γ ∗ ⟨ t ⟩ , with ⟨ t ⟩ ≅ Z \langle t\rangle\cong\mathbb{Z} ⟨ t ⟩ ≅ Z , by h 1 , h 2 , … h_1,h_2,\ldots h 1 , h 2 , … . By [35 Proposition 5.3] , choose homomorphisms ρ j : Λ → Γ \rho_j : \Lambda\to\Gamma ρ j : Λ → Γ fixing Γ \Gamma Γ pointwise such that ρ j ( h i ) ≠ e \rho_j(h_i) \ne e ρ j ( h i ) = e for i ≤ j i \le j i ≤ j . Let pd j \operatorname{pd}_j pd j be the map in Lemma 6.1 (3) associated with ρ j \rho_j ρ j , and let η j = ( pd j ) ∗ μ \eta_j = (\operatorname{pd}_j)_*\mu η j = ( pd j ) ∗ μ . Each η j \eta_j η j is a Λ \Lambda Λ -invariant probability measure on PD 1 ( Λ × Λ ) \operatorname{PD}_{1}(\Lambda\times\Lambda) PD 1 ( Λ × Λ ) . Since this space is compact metrizable, pass to a subsequence such that η j → η \eta_j \to\eta η j → η weakly. Note that η \eta η is Λ \Lambda Λ -invariant. We next verify that every ψ ∈ supp η \psi\in\operatorname{supp}\eta ψ ∈ supp η satisfies ψ ( g , e ) = ψ ( e , g ) = δ g , e \psi(g,e)=\psi(e,g)=\delta_{g,e} ψ ( g , e ) = ψ ( e , g ) = δ g , e for every g ∈ Γ ∗ ⟨ t ⟩ g \in\Gamma* \langle t\rangle g ∈ Γ ∗ ⟨ t ⟩ . This allows us to apply Proposition 2.2 . Fix e ≠ g = h i ∈ Λ e \ne g=h_i \in\Lambda e = g = h i ∈ Λ . For every j ≥ i j \ge i j ≥ i and every M ∈ SA ( L ( Γ ) ) M \in\operatorname{SA}(L(\Gamma)) M ∈ SA ( L ( Γ )) , we have ψ ρ j , M ( g , e ) = ψ ρ j , M ( e , g ) = τ 0 ( λ ( ρ j ( g ) ) ) = 0 \psi_{\rho_j,M}(g,e)=\psi_{\rho_j,M}(e,g)=\tau_0(\lambda(\rho_j(g)))=0 ψ ρ j , M ( g , e ) = ψ ρ j , M ( e , g ) = τ 0 ( λ ( ρ j ( g ))) = 0 . Observe that the set C g = { ψ ∈ PD 1 ( Λ × Λ ) : ψ ( g , e ) = ψ ( e , g ) = 0 } C_g=\{\psi\in\operatorname{PD}_{1}(\Lambda\times\Lambda):\psi(g,e)=\psi(e,g)=0\} C g = { ψ ∈ PD 1 ( Λ × Λ ) : ψ ( g , e ) = ψ ( e , g ) = 0 } is closed. Since pd j ( M ) ∈ C g \operatorname{pd}_j(M) \in C_g pd j ( M ) ∈ C g for every M M M and every j ≥ i j \ge i j ≥ i , one has η j ( C g ) = 1 \eta_j(C_g)=1 η j ( C g ) = 1 for j ≥ i j \ge i j ≥ i . Therefore, it follows that 1 = lim sup j η j ( C g ) ≤ η ( C g ) ≤ 1 1=\limsup_j\eta_j(C_g)\le\eta(C_g)\le1 1 = lim sup j η j ( C g ) ≤ η ( C g ) ≤ 1 . Hence supp η ⊆ C g \operatorname{supp}\eta\subseteq C_g supp η ⊆ C g . Since g ≠ e g\ne e g = e was arbitrary and ψ ( e , e ) = 1 \psi(e,e)=1 ψ ( e , e ) = 1 , every ψ ∈ supp η \psi\in\operatorname{supp}\eta ψ ∈ supp η satisfies ψ ( g , e ) = ψ ( e , g ) = δ g , e \psi(g,e)=\psi(e,g)=\delta_{g,e} ψ ( g , e ) = ψ ( e , g ) = δ g , e for every g ∈ Λ g\in\Lambda g ∈ Λ . Apply the equivariant Borel identification ψ ↦ T ψ \psi\mapsto T_\psi ψ ↦ T ψ from Proposition 2.2 . By Lemma 6.1 (2), the map ψ ↦ Fix ( T ψ ) \psi\mapsto\operatorname{Fix}(T_\psi) ψ ↦ Fix ( T ψ ) is Borel; it is also Λ \Lambda Λ -equivariant. Thus μ ~ = ( F ~ ) ∗ η ∈ IRA ( L ( Λ ) ) \widetilde{\mu}=(\widetilde{\mathcal{F}})_*\eta\in\operatorname{IRA}(L(\Lambda)) μ = ( F ) ∗ η ∈ IRA ( L ( Λ )) .
The Bass–Serre action of Λ \Lambda Λ is non-elementary and acylindrical, with trivial edge stabilizers. A finite normal subgroup fixes a vertex, and its fixed subtree is Λ \Lambda Λ -invariant. Minimality makes this fixed subtree the whole tree, so triviality of the edge stabilizers makes the subgroup trivial. By [51 Corollary 7.3(a)] , the amenable radical of Λ \Lambda Λ is finite, hence trivial. By the Poincaré recurrence argument in the introduction, μ ~ \widetilde{\mu} μ -almost every subalgebra is boomerang. Theorem 5.4 therefore gives μ ~ ( L ( Sub ( Λ ) ) ) = 1 \widetilde{\mu}(L(\operatorname{Sub}(\Lambda)))=1 μ ( L ( Sub ( Λ ))) = 1 . Since μ ~ = F ~ ∗ η \widetilde{\mu}=\widetilde{\mathcal{F}}_*\eta μ = F ∗ η , for η \eta η -almost every ψ \psi ψ there is a subgroup H ≤ Λ H\le\Lambda H ≤ Λ such that Fix ( T ψ ) = L ( H ) \operatorname{Fix}(T_\psi)=L(H) Fix ( T ψ ) = L ( H ) .
We claim that, for η \eta η -almost every ψ ∈ PD 1 ( Λ × Λ ) \psi\in\operatorname{PD}_{1}(\Lambda\times\Lambda) ψ ∈ PD 1 ( Λ × Λ ) , there exists a unique M ∈ SA ( L ( Γ ) ) M\in\operatorname{SA}(L(\Gamma)) M ∈ SA ( L ( Γ )) such that ψ ∣ Γ × Γ = pd ( M ) \psi|_{\Gamma\times\Gamma}=\operatorname{pd}(M) ψ ∣ Γ × Γ = pd ( M ) . Let Res Γ : PD 1 ( Λ × Λ ) → PD 1 ( Γ × Γ ) \operatorname{Res}_{\Gamma}:\operatorname{PD}_{1}(\Lambda\times\Lambda)\to\operatorname{PD}_{1}(\Gamma\times\Gamma) Res Γ : PD 1 ( Λ × Λ ) → PD 1 ( Γ × Γ ) be the continuous map given by Res Γ ( ψ ) = ψ ∣ Γ × Γ \operatorname{Res}_{\Gamma}(\psi)=\psi|_{\Gamma\times\Gamma} Res Γ ( ψ ) = ψ ∣ Γ × Γ . Since ρ j ∣ Γ = id Γ \rho_j|_{\Gamma}=\operatorname{id}_{\Gamma} ρ j ∣ Γ = id Γ , one has Res Γ ∘ pd j = pd \operatorname{Res}_{\Gamma}\circ\operatorname{pd}_j=\operatorname{pd} Res Γ ∘ pd j = pd . By the definition of η j \eta_j η j , ( Res Γ ) ∗ η j = ( Res Γ ∘ pd j ) ∗ μ = pd ∗ μ (\operatorname{Res}_{\Gamma})_*\eta_j=(\operatorname{Res}_{\Gamma}\circ\operatorname{pd}_j)_*\mu=\operatorname{pd}_*\mu ( Res Γ ) ∗ η j = ( Res Γ ∘ pd j ) ∗ μ = pd ∗ μ . By the continuity of Res Γ \operatorname{Res}_{\Gamma} Res Γ and weak convergence of η j \eta_j η j to η \eta η , it follows that ( Res Γ ) ∗ η = pd ∗ μ (\operatorname{Res}_{\Gamma})_*\eta=\operatorname{pd}_*\mu ( Res Γ ) ∗ η = pd ∗ μ . By Lemma 6.1 (1), pd \operatorname{pd} pd is continuous and injective. Thus pd ( SA ( L ( Γ ) ) ) \operatorname{pd}(\operatorname{SA}(L(\Gamma))) pd ( SA ( L ( Γ ))) is Borel by the Lusin–Souslin theorem [44 Theorem 15.1] . Since ( Res Γ ) ∗ η = pd ∗ μ (\operatorname{Res}_{\Gamma})_*\eta=\operatorname{pd}_*\mu ( Res Γ ) ∗ η = pd ∗ μ , one has Res Γ ( ψ ) ∈ pd ( SA ( L ( Γ ) ) ) \operatorname{Res}_{\Gamma}(\psi)\in\operatorname{pd}(\operatorname{SA}(L(\Gamma))) Res Γ ( ψ ) ∈ pd ( SA ( L ( Γ ))) for η \eta η -almost every ψ \psi ψ . Injectivity of pd \operatorname{pd} pd gives a unique M ∈ SA ( L ( Γ ) ) M\in\operatorname{SA}(L(\Gamma)) M ∈ SA ( L ( Γ )) such that Res Γ ( ψ ) = pd ( M ) \operatorname{Res}_{\Gamma}(\psi)=\operatorname{pd}(M) Res Γ ( ψ ) = pd ( M ) . This proves the claim.
Fix such a ψ \psi ψ with Fix ( T ψ ) = L ( H ) \operatorname{Fix}(T_\psi)=L(H) Fix ( T ψ ) = L ( H ) , and let M M M satisfy Res Γ ( ψ ) = pd ( M ) \operatorname{Res}_{\Gamma}(\psi)=\operatorname{pd}(M) Res Γ ( ψ ) = pd ( M ) . For g , h ∈ Γ g,h\in\Gamma g , h ∈ Γ , we have τ 0 ( T ψ ( λ ( g ) ) λ ( h ) ∗ ) = ψ ( g , h ) = ψ M ( g , h ) = τ 0 ( E M ( λ ( g ) ) λ ( h ) ∗ ) \tau_0(T_\psi(\lambda(g))\lambda(h)^*)=\psi(g,h)=\psi_M(g,h)=\tau_0(E_M(\lambda(g))\lambda(h)^*) τ 0 ( T ψ ( λ ( g )) λ ( h ) ∗ ) = ψ ( g , h ) = ψ M ( g , h ) = τ 0 ( E M ( λ ( g )) λ ( h ) ∗ ) . Comparing Fourier coefficients gives E L ( Γ ) ( T ψ ( λ ( g ) ) ) = E M ( λ ( g ) ) E_{L(\Gamma)}(T_\psi(\lambda(g)))=E_M(\lambda(g)) E L ( Γ ) ( T ψ ( λ ( g ))) = E M ( λ ( g )) . Consequently, it follows that E L ( Γ ) ( T ψ ( x ) ) = E M ( x ) E_{L(\Gamma)}(T_\psi(x))=E_M(x) E L ( Γ ) ( T ψ ( x )) = E M ( x ) for every x ∈ L ( Γ ) x\in L(\Gamma) x ∈ L ( Γ ) . If x ∈ M x\in M x ∈ M , then E L ( Γ ) ( T ψ ( x ) ) = x E_{L(\Gamma)}(T_\psi(x))=x E L ( Γ ) ( T ψ ( x )) = x . Using the Schwarz inequality along with the fact that T ψ T_\psi T ψ is trace preserving, we get that
∥ T ψ ( x ) − x ∥ 2 2 = ∥ T ψ ( x ) ∥ 2 2 − ∥ x ∥ 2 2 = τ 0 ( T ψ ( x ) ∗ T ψ ( x ) ) − ∥ x ∥ 2 2 ≤ τ 0 ( T ψ ( x ∗ x ) ) − τ 0 ( x ∗ x ) = 0. \begin{aligned}
\lVert T_\psi(x)-x\rVert_2^2=\lVert T_\psi(x)\rVert_2^2-\lVert x\rVert_2^2 \\
&=\tau_0(T_\psi(x)^*T_\psi(x))-\lVert x\rVert_2^2\le\tau_0(T_\psi(x^*x))-\tau_0(x^*x)=0.
\end{aligned} ∥ T ψ ( x ) − x ∥ 2 2 = ∥ T ψ ( x ) ∥ 2 2 − ∥ x ∥ 2 2 = τ 0 ( T ψ ( x ) ∗ T ψ ( x )) − ∥ x ∥ 2 2 ≤ τ 0 ( T ψ ( x ∗ x )) − τ 0 ( x ∗ x ) = 0.
Thus x ∈ Fix ( T ψ ) x\in\operatorname{Fix}(T_\psi) x ∈ Fix ( T ψ ) . Since x = E L ( Γ ) ( x ) x=E_{L(\Gamma)}(x) x = E L ( Γ ) ( x ) , this proves M ⊆ E L ( Γ ) ( Fix ( T ψ ) ) M\subseteq E_{L(\Gamma)}(\operatorname{Fix}(T_\psi)) M ⊆ E L ( Γ ) ( Fix ( T ψ )) . Conversely, if x ∈ E L ( Γ ) ( Fix ( T ψ ) ) = E L ( Γ ) ( L ( H ) ) = L ( H ∩ Γ ) x\in E_{L(\Gamma)}(\operatorname{Fix}(T_\psi))=E_{L(\Gamma)}(L(H))=L(H\cap\Gamma) x ∈ E L ( Γ ) ( Fix ( T ψ )) = E L ( Γ ) ( L ( H )) = L ( H ∩ Γ ) , then T ψ ( x ) = x T_\psi(x)=x T ψ ( x ) = x , so that E M ( x ) = E L ( Γ ) ( T ψ ( x ) ) = x E_M(x)=E_{L(\Gamma)}(T_\psi(x))=x E M ( x ) = E L ( Γ ) ( T ψ ( x )) = x . Therefore M = E L ( Γ ) ( Fix ( T ψ ) ) = L ( H ∩ Γ ) M=E_{L(\Gamma)}(\operatorname{Fix}(T_\psi))=L(H\cap\Gamma) M = E L ( Γ ) ( Fix ( T ψ )) = L ( H ∩ Γ ) . By Proposition 3.2 , L ( Sub ( Γ ) ) L(\operatorname{Sub}(\Gamma)) L ( Sub ( Γ )) is compact, and so is pd ( L ( Sub ( Γ ) ) ) \operatorname{pd}(L(\operatorname{Sub}(\Gamma))) pd ( L ( Sub ( Γ ))) . The preceding argument and injectivity of pd \operatorname{pd} pd give μ ( L ( Sub ( Γ ) ) ) = ( pd ∗ μ ) ( pd ( L ( Sub ( Γ ) ) ) ) = η ( Res Γ − 1 ( pd ( L ( Sub ( Γ ) ) ) ) ) = 1 \mu(L(\operatorname{Sub}(\Gamma)))=(\operatorname{pd}_*\mu)(\operatorname{pd}(L(\operatorname{Sub}(\Gamma))))=\eta(\operatorname{Res}_{\Gamma}^{-1}(\operatorname{pd}(L(\operatorname{Sub}(\Gamma)))))=1 μ ( L ( Sub ( Γ ))) = ( pd ∗ μ ) ( pd ( L ( Sub ( Γ )))) = η ( Res Γ − 1 ( pd ( L ( Sub ( Γ ))))) = 1 . Hence ν = J ∗ μ \nu=J_*\mu ν = J ∗ μ is an IRS of Γ \Gamma Γ and L ∗ ν = μ L_*\nu=\mu L ∗ ν = μ . Finally, J L = id JL=\operatorname{id} J L = id gives uniqueness. □ \square □
We now apply the same construction in each coordinate. The larger group is a finite product of the free products Γ i ∗ ⟨ t i ⟩ \Gamma_i * \langle t_i\rangle Γ i ∗ ⟨ t i ⟩ , so we can apply Theorem 1.1 . The proof is similar to that of Theorem 6.2 ; however, we spell out the details. Together with Theorem 6.2 , the following theorem proves Theorem 1.4 .
Theorem 6.3. Let n ≥ 2 n \ge2 n ≥ 2 , and let Γ i \Gamma_i Γ i be countable nontrivial mixed identity-free groups for 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n . Set Γ = ⨁ i = 1 n Γ i \Gamma= \bigoplus_{i=1}^{n} \Gamma_i Γ = ⨁ i = 1 n Γ i . Then every IRA μ \mu μ of Γ \Gamma Γ is L ∗ ν L_* \nu L ∗ ν for a unique IRS ν \nu ν of Γ \Gamma Γ .
Proof. Set Λ i = Γ i ∗ ⟨ t i ⟩ \Lambda_i = \Gamma_i * \langle t_i\rangle Λ i = Γ i ∗ ⟨ t i ⟩ , where ⟨ t i ⟩ ≅ Z \langle t_i\rangle\cong\mathbb{Z} ⟨ t i ⟩ ≅ Z , and Λ = ⨁ i = 1 n Λ i \Lambda= \bigoplus_{i=1}^{n} \Lambda_i Λ = ⨁ i = 1 n Λ i . For each i i i , enumerate Λ i ∖ { e } \Lambda_i \setminus\{e\} Λ i ∖ { e } by h 1 ( i ) , h 2 ( i ) , … h_1^{(i)}, h_2^{(i)}, \ldots h 1 ( i ) , h 2 ( i ) , … . By [35 Proposition 5.3] , choose homomorphisms ρ j ( i ) : Λ i → Γ i \rho_j^{(i)} : \Lambda_i \to\Gamma_i ρ j ( i ) : Λ i → Γ i fixing Γ i \Gamma_i Γ i pointwise such that ρ j ( i ) ( h k ( i ) ) ≠ e \rho_j^{(i)}(h_k^{(i)}) \ne e ρ j ( i ) ( h k ( i ) ) = e for k ≤ j k \le j k ≤ j . Define ρ j = ⨁ i = 1 n ρ j ( i ) : Λ → Γ \rho_j = \bigoplus_{i=1}^{n} \rho_j^{(i)} : \Lambda\to\Gamma ρ j = ⨁ i = 1 n ρ j ( i ) : Λ → Γ . Then ρ j ∣ Γ = id Γ \rho_j|_\Gamma= \operatorname{id}_\Gamma ρ j ∣ Γ = id Γ , and, for every g ∈ Λ ∖ { e } g \in\Lambda\setminus\{e\} g ∈ Λ ∖ { e } , one has ρ j ( g ) ≠ e \rho_j(g) \ne e ρ j ( g ) = e for all sufficiently large j j j . Indeed, if g i = h k ( i ) ≠ e g_i = h_k^{(i)} \ne e g i = h k ( i ) = e , then the i i i th coordinate of ρ j ( g ) \rho_j(g) ρ j ( g ) is nontrivial for j ≥ k j \ge k j ≥ k .
Let p d ( M ) = ψ M \mathfrak{pd}(M) = \psi_M pd ( M ) = ψ M be the map in Lemma 6.1 (1). Let p d j \mathfrak{pd}_j pd j be the map in Lemma 6.1 (3) associated with ρ j \rho_j ρ j , and set η j = ( p d j ) ∗ μ \eta_j = (\mathfrak{pd}_j)_*\mu η j = ( pd j ) ∗ μ . By that lemma, each η j \eta_j η j is a Λ \Lambda Λ -invariant probability measure on PD 1 ( Λ × Λ ) \operatorname{PD}_1(\Lambda\times\Lambda) PD 1 ( Λ × Λ ) . The space PD 1 ( Λ × Λ ) \operatorname{PD}_1(\Lambda\times\Lambda) PD 1 ( Λ × Λ ) is compact metrizable, so pass to a subsequence such that η j → η \eta_j \to\eta η j → η weakly. Continuity of the action makes η \eta η Λ \Lambda Λ -invariant.
Fix g ∈ Λ ∖ { e } g \in\Lambda\setminus\{e\} g ∈ Λ ∖ { e } . For all sufficiently large j j j and every M ∈ SA ( L ( Γ ) ) M \in\operatorname{SA}(L(\Gamma)) M ∈ SA ( L ( Γ )) , trace preservation gives p d j ( M ) ( g , e ) = p d j ( M ) ( e , g ) = τ 0 ( λ ( ρ j ( g ) ) ) = 0 \mathfrak{pd}_j(M)(g,e) = \mathfrak{pd}_j(M)(e,g) = \tau_0(\lambda(\rho_j(g))) = 0 pd j ( M ) ( g , e ) = pd j ( M ) ( e , g ) = τ 0 ( λ ( ρ j ( g ))) = 0 . Thus
∫ ( ∣ ψ ( g , e ) ∣ 2 + ∣ ψ ( e , g ) ∣ 2 ) d η ( ψ ) = lim j ∫ ( ∣ ψ ( g , e ) ∣ 2 + ∣ ψ ( e , g ) ∣ 2 ) d η j ( ψ ) = 0. \int\left(\lvert\psi(g,e)\rvert^2 + \lvert\psi(e,g)\rvert^2\right)\,d\eta(\psi)
= \lim_{j} \int\left(\lvert\psi(g,e)\rvert^2 + \lvert\psi(e,g)\rvert^2\right)\,d\eta_j(\psi) = 0. ∫ ( ∣ ψ ( g , e ) ∣ 2 + ∣ ψ ( e , g ) ∣ 2 ) d η ( ψ ) = j lim ∫ ( ∣ ψ ( g , e ) ∣ 2 + ∣ ψ ( e , g ) ∣ 2 ) d η j ( ψ ) = 0.
By countability of Λ \Lambda Λ , for η \eta η -almost every ψ \psi ψ , one has ψ ( g , e ) = ψ ( e , g ) = δ g , e \psi(g,e) = \psi(e,g) = \delta_{g,e} ψ ( g , e ) = ψ ( e , g ) = δ g , e for every g ∈ Λ g \in\Lambda g ∈ Λ . Apply the equivariant homeomorphism ψ ↦ T ψ \psi\mapsto T_\psi ψ ↦ T ψ from Proposition 2.2 . By Lemma 6.1 (2), the map F ( ψ ) = Fix ( T ψ ) F(\psi) = \operatorname{Fix}(T_\psi) F ( ψ ) = Fix ( T ψ ) is Borel; it is also Λ \Lambda Λ -equivariant. Thus μ ~ = F ∗ η \widetilde{\mu} = F_*\eta μ = F ∗ η is an IRA of Λ \Lambda Λ .
The Bass–Serre action of each Λ i \Lambda_i Λ i is non-elementary and acylindrical, with trivial edge stabilizers. A finite normal subgroup fixes a vertex, and its fixed subtree is Λ i \Lambda_i Λ i -invariant. Minimality makes this fixed subtree the whole tree, so triviality of the edge stabilizers makes the subgroup trivial. By [51 Corollary 7.3(a)] , the amenable radical of Λ i \Lambda_i Λ i is finite, hence trivial. By the recurrence argument in the introduction, μ ~ \widetilde{\mu} μ -almost every subalgebra is boomerang. Theorem 5.5 therefore gives Fix ( T ψ ) = L ( H ) \operatorname{Fix}(T_\psi) = L(H) Fix ( T ψ ) = L ( H ) for some H ≤ Λ H \le\Lambda H ≤ Λ , for η \eta η -almost every ψ \psi ψ . Let Res Γ : PD 1 ( Λ × Λ ) → PD 1 ( Γ × Γ ) \operatorname{Res}_\Gamma: \operatorname{PD}_1(\Lambda\times\Lambda) \to\operatorname{PD}_1(\Gamma\times\Gamma) Res Γ : PD 1 ( Λ × Λ ) → PD 1 ( Γ × Γ ) be the continuous map given by Res Γ ( ψ ) = ψ ∣ Γ × Γ \operatorname{Res}_\Gamma(\psi) = \psi|_{\Gamma\times\Gamma} Res Γ ( ψ ) = ψ ∣ Γ × Γ . Since ρ j ∣ Γ = id Γ \rho_j|_\Gamma= \operatorname{id}_\Gamma ρ j ∣ Γ = id Γ , one has Res Γ ∘ p d j = p d \operatorname{Res}_\Gamma\circ\mathfrak{pd}_j = \mathfrak{pd} Res Γ ∘ pd j = pd . Consequently, ( Res Γ ) ∗ η j = ( Res Γ ∘ p d j ) ∗ μ = p d ∗ μ (\operatorname{Res}_\Gamma)_*\eta_j = (\operatorname{Res}_\Gamma\circ\mathfrak{pd}_j)_*\mu= \mathfrak{pd}_*\mu ( Res Γ ) ∗ η j = ( Res Γ ∘ pd j ) ∗ μ = pd ∗ μ . Continuity of Res Γ \operatorname{Res}_\Gamma Res Γ and weak convergence give ( Res Γ ) ∗ η = p d ∗ μ (\operatorname{Res}_\Gamma)_*\eta= \mathfrak{pd}_*\mu ( Res Γ ) ∗ η = pd ∗ μ . By Lemma 6.1 (1), p d \mathfrak{pd} pd is continuous and injective. Thus p d ( SA ( L ( Γ ) ) ) \mathfrak{pd}(\operatorname{SA}(L(\Gamma))) pd ( SA ( L ( Γ ))) is Borel by the Lusin–Souslin theorem [44 Theorem 15.1] . It follows that Res Γ ( ψ ) ∈ p d ( SA ( L ( Γ ) ) ) \operatorname{Res}_\Gamma(\psi) \in\mathfrak{pd}(\operatorname{SA}(L(\Gamma))) Res Γ ( ψ ) ∈ pd ( SA ( L ( Γ ))) for η \eta η -almost every ψ \psi ψ . For such a ψ \psi ψ , there is a unique M ∈ SA ( L ( Γ ) ) \mathcal{M} \in\operatorname{SA}(L(\Gamma)) M ∈ SA ( L ( Γ )) such that Res Γ ( ψ ) = p d ( M ) \operatorname{Res}_\Gamma(\psi) = \mathfrak{pd}(\mathcal{M}) Res Γ ( ψ ) = pd ( M ) . Fix such a ψ \psi ψ with Fix ( T ψ ) = L ( H ) \operatorname{Fix}(T_\psi) = L(H) Fix ( T ψ ) = L ( H ) . For g , h ∈ Γ g,h \in\Gamma g , h ∈ Γ , τ 0 ( T ψ ( λ ( g ) ) λ ( h ) ∗ ) = ψ ( g , h ) = ψ M ( g , h ) = τ 0 ( E M ( λ ( g ) ) λ ( h ) ∗ ) \tau_0(T_\psi(\lambda(g))\lambda(h)^*) = \psi(g,h) = \psi_{\mathcal{M}}(g,h) = \tau_0(E_{\mathcal{M}}(\lambda(g))\lambda(h)^*) τ 0 ( T ψ ( λ ( g )) λ ( h ) ∗ ) = ψ ( g , h ) = ψ M ( g , h ) = τ 0 ( E M ( λ ( g )) λ ( h ) ∗ ) . Comparing Fourier coefficients gives E L ( Γ ) ( T ψ ( λ ( g ) ) ) = E M ( λ ( g ) ) E_{L(\Gamma)}(T_\psi(\lambda(g))) = E_{\mathcal{M}}(\lambda(g)) E L ( Γ ) ( T ψ ( λ ( g ))) = E M ( λ ( g )) . Both maps are normal, so E L ( Γ ) ( T ψ ( x ) ) = E M ( x ) \mathbb{E}_{L(\Gamma)}(T_{\psi}(x))=\mathbb{E}_{M}(x) E L ( Γ ) ( T ψ ( x )) = E M ( x ) for every x ∈ L ( Γ ) x\in L(\Gamma) x ∈ L ( Γ ) . The Schwarz inequality and trace preservation give
∥ T ψ ( x ) ∥ 2 2 = τ 0 ( T ψ ( x ) ∗ T ψ ( x ) ) ≤ τ 0 ( T ψ ( x ∗ x ) ) = ∥ x ∥ 2 2 . \lVert T_{\psi}(x)\rVert_{2}^{2}=\tau_{0}(T_{\psi}(x)^{*}T_{\psi}(x))\leq\tau_{0}(T_{\psi}(x^{*}x))=\lVert x\rVert_{2}^{2}. ∥ T ψ ( x ) ∥ 2 2 = τ 0 ( T ψ ( x ) ∗ T ψ ( x )) ≤ τ 0 ( T ψ ( x ∗ x )) = ∥ x ∥ 2 2 .
If x ∈ M x\in M x ∈ M , then E L ( Γ ) ( T ψ ( x ) ) = x \mathbb{E}_{L(\Gamma)}(T_{\psi}(x))=x E L ( Γ ) ( T ψ ( x )) = x . Orthogonality gives ∥ T ψ ( x ) − x ∥ 2 2 = ∥ T ψ ( x ) ∥ 2 2 − ∥ x ∥ 2 2 ≤ 0 \lVert T_{\psi}(x)-x\rVert_{2}^{2}=\lVert T_{\psi}(x)\rVert_{2}^{2}-\lVert x\rVert_{2}^{2}\leq0 ∥ T ψ ( x ) − x ∥ 2 2 = ∥ T ψ ( x ) ∥ 2 2 − ∥ x ∥ 2 2 ≤ 0 . Thus x ∈ Fix ( T ψ ) x\in\operatorname{Fix}(T_{\psi}) x ∈ Fix ( T ψ ) , and M ⊆ E L ( Γ ) ( Fix ( T ψ ) ) M\subseteq\mathbb{E}_{L(\Gamma)}(\operatorname{Fix}(T_{\psi})) M ⊆ E L ( Γ ) ( Fix ( T ψ )) .
Conversely, if x ∈ E L ( Γ ) ( Fix ( T ψ ) ) = E L ( Γ ) ( L ( H ) ) = L ( H ∩ Γ ) x\in\mathbb{E}_{L(\Gamma)}(\operatorname{Fix}(T_{\psi}))=\mathbb{E}_{L(\Gamma)}(L(H))=L(H\cap\Gamma) x ∈ E L ( Γ ) ( Fix ( T ψ )) = E L ( Γ ) ( L ( H )) = L ( H ∩ Γ ) , then T ψ ( x ) = x T_{\psi}(x)=x T ψ ( x ) = x , so that E M ( x ) = E L ( Γ ) ( T ψ ( x ) ) = x \mathbb{E}_{M}(x)=\mathbb{E}_{L(\Gamma)}(T_{\psi}(x))=x E M ( x ) = E L ( Γ ) ( T ψ ( x )) = x . Therefore M = E L ( Γ ) ( Fix ( T ψ ) ) = L ( H ∩ Γ ) M=\mathbb{E}_{L(\Gamma)}(\operatorname{Fix}(T_{\psi}))=L(H\cap\Gamma) M = E L ( Γ ) ( Fix ( T ψ )) = L ( H ∩ Γ ) . By Proposition 3.2 , L ( Sub ( Γ ) ) L(\operatorname{Sub}(\Gamma)) L ( Sub ( Γ )) is compact, and so is pd ( L ( Sub ( Γ ) ) ) \operatorname{pd}(L(\operatorname{Sub}(\Gamma))) pd ( L ( Sub ( Γ ))) . The preceding argument and injectivity of pd \operatorname{pd} pd give μ ( L ( Sub ( Γ ) ) ) = ( pd ∗ μ ) ( pd ( L ( Sub ( Γ ) ) ) ) = η ( Res Γ − 1 ( pd ( L ( Sub ( Γ ) ) ) ) = 1 \mu(L(\operatorname{Sub}(\Gamma)))=(\operatorname{pd}_{*}\mu)(\operatorname{pd}(L(\operatorname{Sub}(\Gamma))))=\eta(\operatorname{Res}_{\Gamma}^{-1}(\operatorname{pd}(L(\operatorname{Sub}(\Gamma))))=1 μ ( L ( Sub ( Γ ))) = ( pd ∗ μ ) ( pd ( L ( Sub ( Γ )))) = η ( Res Γ − 1 ( pd ( L ( Sub ( Γ )))) = 1 . Hence ν = J ∗ μ \nu=J_{*}\mu ν = J ∗ μ is an IRS of Γ \Gamma Γ and L ∗ ν = μ L_{*}\nu=\mu L ∗ ν = μ . Finally, J L = id JL=\operatorname{id} J L = id gives uniqueness. □
It is not true in general that all mixed identity-free groups are Boomerang rigid as the following example demonstrates.
Example 6.4. By [37 Theorem 1.3] , there exists a countable nontrivial locally finite 3 3 3 -group Γ \Gamma Γ which is mixed identity-free. Thus Γ \Gamma Γ is IRA rigid by Theorem 6.2 . Every von Neumann subalgebra of L ( Γ ) L(\Gamma) L ( Γ ) is boomerang, since conjugation by each γ ∈ Γ \gamma\in\Gamma γ ∈ Γ has finite order. Choose g ∈ Γ g\in\Gamma g ∈ Γ of order 3 3 3 and set p = ( 1 + λ ( g ) + λ ( g 2 ) ) / 3 p=(1+\lambda(g)+\lambda(g^{2}))/3 p = ( 1 + λ ( g ) + λ ( g 2 )) /3 . Then p p p is a projection with τ ( p ) = 1 / 3 \tau(p)=1/3 τ ( p ) = 1/3 , so M = C p ⊕ C ( 1 − p ) ⊆ L ( ⟨ g ⟩ ) M=\mathbb{C}p\oplus\mathbb{C}(1-p)\subseteq L(\langle g\rangle) M = C p ⊕ C ( 1 − p ) ⊆ L (⟨ g ⟩) is a two-dimensional unital von Neumann subalgebra. It cannot equal L ( Δ ) L(\Delta) L ( Δ ) for any subgroup Δ ≤ Γ \Delta\leq\Gamma Δ ≤ Γ , since this would imply ∣ Δ ∣ = 2 |\Delta|=2 ∣Δ∣ = 2 .
We end this section by showing that a big class of groups or their finite products are IRA rigid, thanks to Theorem 6.3 .
Corollary 6.5. Let n ≥ 1 n\geq1 n ≥ 1 , and let Γ i \Gamma_i Γ i be countable groups for 1 ≤ i ≤ n 1\leq i\leq n 1 ≤ i ≤ n . Suppose that each Γ i \Gamma_i Γ i satisfies one of the following conditions:
(1) Γ i \Gamma_i Γ i admits a topologically free minimal extremely proximal action on a compact Hausdorff space with more than two points.
(2) Γ i \Gamma_i Γ i admits a faithful minimal action of general type on a tree, and the induced action on the boundary of the tree is topologically free.
Then Γ = ⨁ i = 1 n Γ i \Gamma=\bigoplus_{i=1}^{n}\Gamma_i Γ = ⨁ i = 1 n Γ i is IRA rigid. In particular, every finite product of Baumslag–Solitar groups BS ( m , n ) \operatorname{BS}(m,n) BS ( m , n ) with ∣ m ∣ , ∣ n ∣ ≥ 2 |m|,|n|\geq2 ∣ m ∣ , ∣ n ∣ ≥ 2 and ∣ m ∣ ≠ ∣ n ∣ |m|\neq|n| ∣ m ∣ = ∣ n ∣ is IRA rigid.
Proof. Each Γ i \Gamma_i Γ i is mixed identity-free by [4 Proposition 3.4 and Theorem 3.6] in case (1) and [27 Theorem B] in case (2). Apply Theorems 6.2 and 6.3 . The Baumslag–Solitar case follows from [27 Theorem B and Lemma 8.6] . □
Amenable groups with the ISR property were first obtained by Jiang–Zhou for the group of finitely supported permutations of N \mathbb{N} N [41 ] . More examples of such amenable groups can be found in [25 ] and [5 ] . We can now obtain the following examples of amenable IRA rigid groups.
Corollary 6.6. The following hold true.
There exists a two-generated infinite elementary amenable group which is IRA rigid.
For each prime p p p , there exists an infinite locally finite p p p -group which is IRA rigid.
There exist continuum many pairwise nonisomorphic finitely generated nonamenable IRA rigid groups which contain no nonabelian free subgroup and have nontrivial amenable radical.
Proof. (i) and (ii) follow from [37 Theorems 1.2 and 1.3] and Theorem 6.2 . For (iii), [36 Corollary 1.3] put together with Theorem 6.2 gives the conclusion. □
We can also embed any amenable group into an IRA rigid amenable group that is finitely generated.
Corollary 6.7. Let Γ \Gamma Γ be a countable amenable group. Then, it can embed into a finitely generated amenable IRA rigid group. If the given group is elementary amenable, the overgroup can be chosen elementary amenable.
Proof. It follows from [36 Corollary 1.2(1)] that every countable (elementary) amenable group embeds into a finitely generated (elementary) amenable MIF group. The proof finishes by applying Theorem 6.2 . □
References
Institute of Mathematics of the Polish Academy of Sciences, ul. Sniadeckich 8, 00-656, Warszawa, Poland
Email address: tattwamasiamrutam@gmail.com
Ben Gurion University of the Negev. Department of Mathematics. Be’er Sheva, 8410501, Israel.
Email address: yairgl@bgu.ac.il
Ben Gurion University of the Negev. Department of Mathematics. Be’er Sheva, 8410501, Israel.
Email address: hartmany@bgu.ac.il
School of Mathematical Sciences, Dalian University of Technology, Dalian, 116024, China
Email address: yonglejiang@dlut.edu.cn