Aslightly different version of this manuscript willappear in Mathematical Physics, Analysis and Geometry. GROUPOIDS, VON NEUMANN ALGEBRAS AND THE INTEGRATED DENSITY OF STATES DANIEL LENZ,NORBERTPEYERIMHOFF, ANDIVAN VESELIC´ 7 0 Abstract. Westudyspectralpropertiesofrandomoperatorsinthegeneral 0 2 settingofgroupoidsandvonNeumannalgebras. Inparticular,weestablish an explicit formula for the canonical trace of the von Neumann algebra of n random operators and definean abstract density of states. a Whilethetreatmentappliestoageneralframeworkwelayspecialempha- J sis on three particular examples: random Schr¨odinger operators on mani- 4 folds,quantumpercolationandquasicrystalHamiltonians. Fortheseexam- plesweshowthatthedistributionfunctionoftheabstract densityofstates 2 v coincides with the integrated density of states defined via an exhaustion 6 procedure. 2 0 3 1. Introduction 0 2 The aim of this paper is to review and present a unified treatment of basic 0 features of random (Schro¨dinger)operators usingtechniques from Connes’non- / h commutative integration theoryandvonNeumannalgebras[Con79]. Particular p emphasis will be laid on an application of the general setting to the example of - h • a group action on a manifold proposed by two of the authors [PV01]. t a This example merges and extends two situations, viz periodic operators on m manifolds as studied first by Adachi/Sunada [AS93] and random Schro¨dinger : operators on Rd or Zd as studied by various people (s. below) starting with the v i workofPastur[Pas71]. Inthefirstsituationakeyroleisplayedbythegeometry X of the underlying manifold. In the second situation, the crucial ingredient is r a the randomness of the corresponding potential. We also apply our discussion to two more examples: • random operators on tilings and Delone sets whose mathematically rig- orous study goes back to Hof [Hof93] and Kellendonk [Kel95]). • random operators on site-percolation graphs, see e.g. [dGLM59a, CCF+86, BK01, Ves05b]. As for the above three examples let us already point out the following differ- ences: inthefirstexampletheunderlyinggeometric spaceiscontinuous andthe group acting on it is discrete; in the second example the underlying geometric space is discrete and the group acting on it is continuous; finally, in the third Date: February 7, 2008, LPV-2007-01-04.tex. 1991 Mathematics Subject Classification. 46L10, 35J10; 46L51, 82B44. Key words and phrases. Groupoids, von Neumann algebras, integrated density of states, random operators, Schr¨odinger operators on manifolds, trace formula. 1 2 D.LENZ,N.PEYERIMHOFF,ANDI.VESELIC´ example both the underlying geometric space and the group acting on it are discrete. The use of von Neumann algebras in the treatment of special random oper- ators is not new. It goes back at least to the seminal work of Sˇubin on almost periodic operators [Sˇub79]. These points will be discussed in more detail next. Random Schro¨dinger operators arise in the quantum mechanical treatment of disordered solids. This includes, in particular, periodic operators, almost periodic operators and Anderson type operators on Zd or Rd (cf. the textbooks [CFSK87, Kir89, CL90, PF92, Sto01]). In all these cases one is given a family (H ) of selfadjoint operators H acting on a Hilbert space H , indexed by ω ω ω ω in a measure space (Ω,µ) and satisfying an equivariance condition with respect to a certain set of unitary operators (U ) . i i∈I While specific examples of these cases exhibit very special spectral features, there are certain characteristics shared by all models. These properties are as follows. (In parentheses we give a reference where the corresponding property is established.) (P1) Almost sure constancy of the spectral properties of H given some er- ω godicity condition. In particular, the spectrum Σ is nonrandom (Theo- rem 5.1). (P2) Absence of discrete spectrum (Corollary 5.9) and, in fact, a dichotomy (between zero and infinity) for the values of the dimensions of spectral projections. (P3) A naturally arising von Neumann algebra (Section 3) with a canonical trace τ, to which the random operators are affiliated (Section 4). (P4) A measure ρ, called the density of states, governing global features of the family (H ), in particular, having Σ as its support (Proposition ω 5.2). This measure is related to the trace of the von Neumann algebra. Let us furthermore single out the following point, which we show for the three abovementioned examples: (P5) Alocalproceduretocalculate ρviaanexhaustiongiven someamenabil- ity condition. This is known as Pastur-Sˇubin trace formula. It implies the self-averaging property of the density of states (discussed for the examples mentioned above in Sections 6, 7, 8). Let us now discuss these facts for earlier studied models. The interest in property (P5) arouse from the physics of disordered media. First mathemat- ically rigorous results on the (integrated) density of states are due to Pastur [Pas71, Pas72, Pas74], Fukushima, Nakao and Nagai [Fuk74, FNN75, FN77, Nak77, Fuk81], Kotani [Kot77], and Kirsch and Martinelli [KM82a, KM82c]. In these papers two different methods for constructing the integrated density of states (IDS) can be found (property (P5)). Either one uses the Laplace transform to conclude the convergence of certain normalized eigenvalue count- ing functions, or one analyzes the counting functions directly via the so called Dirichlet-Neumann bracketing. In our setting the Laplace transform method GROUPOIDS, VON NEUMANN ALGEBRAS AND THE IDS 3 seems to be of better use, since the pointwise superadditive ergodic theorem [AK81] used in the Dirichlet-Neumann bracketing approach [KM82a] has no counterpart in the (nonabelian) generality we are aiming at. For the more recent development in the study of the IDS of alloy type and relatedmodels,aswellastheresultsonitsregularityandasymptoticbehaviour, see [CL90, PF92, Sto01, Ves04] and the references cited there. For almost periodic differential operators on Rd and the associated von Neu- mann algebras, a thorough study of the above features (and many more) has been carried out in the seminal papers by Coburn, Moyer and Singer [CMS73] and Sˇubin [Sˇub79]. Almost periodic Schro¨dinger operators on Zd and Rd were then studied by Avron and Simon [AS82, AS83]. An abstract C∗-algebraic framework for the treatment of almost periodicop- erators was then proposed and studied by Bellissard [Bel86, Bel92] and Bellis- sard, Lima and Testard [BLT86]. While these works focus on K-theory and the socalled gap-labeling, theyalsoshow(P1)-(P5) foralmostperiodicSchro¨dinger type operators on Rd and Zd. Let us emphasize that large parts of this C∗- algebraic treatment are not confined to almost periodic operators. In fact, (P1)-(P4) are established there for crossed products arising from arbitrary ac- tions of locally compact abelian groups on locally compact spaces X. After the work of Aubry/Andr´e [AA] and the short announcement of Bel- lissard/Testard in [BT82], investigations in this framework, centered around so called spectral duality, were carried out by Kaminker and Xia [KX87] and Chojnacki [Cho92]. A special one-dimensional version of spectral duality based on [GJLS99] can also be found in [Len99]. An operator algebraic framework of crossed-products (involving von Neu- mann crossed products)can also beused in the study of general random opera- tors if one considers Rd actions together with operators on L2(Rd) (cf. [Len99]). However, certain of these models rather use actions of Zd together with opera- tors on L2(Rd), like the thoroughly studied alloy or continuous Anderson type models. This presents a difficulty which was overcome in a work by Kirsch [Kir85] introducing a so called suspension construction, see also [BLT86] for related material. This allows to “amplify“ these Zd actions to Rd actions and thus reduce the treatment of (P1)–(P4) in the Zd case to the Rd case. In recent years three more classes of examples have been considered. These are random operators on manifolds [Szn89, Szn90, PV01, LPV04, LPV03], discrete random operators on tilings [Hof93, Hof95, Kel95, Kel97, BHZ00, LS01, LS04], and random Hamiltonians generated by percolation processes [BK01, Ves05a, Ves05b, KM]. In these cases the algebraic framework devel- oped earlier could not be used to establish (P1)–(P5). Note, however, that partial results concerning, e.g., (P1) or restricted versions of (P5) arestill avail- able. Note also that continuous operators associated with tilings as discussed in [BHZ00, BBG01] fall within the C∗-algebraic framework of [Bel86, Bel92]. A more detailed analysis of the point spectrum of discrete operators associated to tilings and percolation graphs will be carried out in [LV]. 4 D.LENZ,N.PEYERIMHOFF,ANDI.VESELIC´ The model considered in [PV01] includes periodic operators on manifolds. In fact, it was motivated by work of Adachi and Sunada [AS93], who establish an exhaustion construction for the IDS as well as a representation as a Γ-trace in the periodic case. For further investigations related to the IDS of periodic operators in both discrete and continuous geometric settings, see e.g. [DM97, DM98, MY02, MSY03, DLMSY03, LV]. More precisely, our first example concerned with Random Schro¨dinger oper- ators on Manifolds (RSM) can be described as follows,see [PV01, LPV04]: Example. (RSM) Let (X,g ) be the Riemannian covering of a compact Rie- 0 mannian manifold M = X/Γ. We assume that there exists a family (g ) of ω ω∈Ω Riemannian metrics on X which are parameterized by the elements of a proba- bility space (Ω,B ,P) and which are uniformly boundedby g , i.e., there exists Ω 0 a constant A ≥ 1 such that 1 g (v,v) ≤ g (v,v) ≤ Ag (v,v) for all v ∈ TX and ω ∈ Ω. 0 ω 0 A Let λω denote the Riemannian volume form corresponding to the metric g . ω WeassumethatΓacts ergodically onΩbymeasurepreservingtransformations. The metrics are compatible in the sense that for all γ ∈ Γ the corresponding deck transformations γ : (X,g )→ (X,g ) ω γω are isometries. Then the induced maps U : L2(X,λγ−1ω) → L2(X,λω), (ω,γ) (U f)(x)= f(γ−1x) are unitary operators. Based on this geometric setting, (ω,γ) we consider a family (H : ω ∈ Ω), H = ∆ +V , of Schro¨dinger operators ω ω ω ω satisfying the following equivariance condition (1) Hω = U(ω,γ)Hγ−1ωU(∗ω,γ), for all γ ∈ Γ and ω ∈ Ω. We also assume some kind of weak measurability in ω, namely, we will assume that (2) ω 7→ hf(ω,·),F(H )f(ω,·)i := f¯(ω,x)[F(H )f](ω,x)dλω(x) ω ω ω ZX is measurableforevery measurablef onΩ×X withf(ω,·) ∈ L2(X,λω), ω ∈ Ω, andeveryfunctionF onRwhichisuniformlyboundedonthespectraoftheH . ω Note that L2(X,λω) considered as a set of functions (disregarding the scalar product) is independent of ω. The expectation with respect to the measure P will be denoted by E. This example covers the following two particular cases: (I) A family of Schro¨dinger operators (∆+V ) on a fixed Riemannian ω ω∈Ω manifold (X,g ) with random potentials, see [PV01]. In this case the 0 equivariance condition (1) transforms into the following property of the potentials V (x) = V (γ−1x). γω ω GROUPOIDS, VON NEUMANN ALGEBRAS AND THE IDS 5 (II) Afamily ofLaplacians ∆ onamanifoldX withrandommetrics(g ) ω ω ω∈Ω satisfying some additional assumptions [LPV04]. By the properties of X and M in (RSM), the group Γ is discrete, finitely generated and acts cocompactly, freely and properly discontinuously on X. In the physical literature the equivariance condition (1) is denoted ei- ther as equivariance condition, see e.g. [Bel86], or as ergodicity of operators [Pas80, Kir89], where it is assumed that the measure preserving transforma- tions are ergodic. From the probabilistic point of view this property is simply the stationarity of an operator valued stochastic process. Itisouraimheretopresentagroupoidbasedapproachto(P1)–(P4)covering all examples studied so far. This includes, in particular, the case of random operators on manifolds, the tiling case and the percolation case. Our framework applies also to Schro¨dinger operators on hyperbolic space (e.g. the Poissonian model considered in [Szn89]). However, our proof of (P5) does not apply to this setting becauseof the lack of amenability of theisometry group. For the example (RSM) with amenable group action Γ, we will prove (P5) in Section 6. Note that case (I) of (RSM) includes the models treated earlier by the suspension construction. Thus, as a by product of our approach, we get an algebraic treatment of (P5) for these models. As mentioned already, our results can also be applied to further examples. Application to tilings is dis- cussed in [LS01, LS04]. There, a uniform ergodic type theorem for tilings along with a strong version of Pastur-Sˇubin-formula (P5) is given. The results also apply to random operators on percolation graphs. In particular they provide complementary information to the results in [Ves05a, Ves05b, KM], where the integrated density of states was defined rigorously for site and edge percolation Hamiltonians. A basic discussion of these examples and the connection to our studyhereis given in Sections 7and 8 respectively. Thiswill inparticular show that (P5) remains valid for these examples. For further details we refer to the cited literature. The results also apply to random operators on foliations (see [Kor95] for related results). Our approach is based on groupoids and Connes theory of noncommutative integration [Con79]. Thus, let us conclude this section by sketching the main aspects of the groupoid framework used in this article. The work [Con79] on noncommutative integration theory consists of three parts. In the first part an abstract version of integration on quotients is presented. This is then used to introduce certain von Neumann algebras (viz. von Neumann algebras of random operators) and to classify their semifinite normal weights. Finally, Connes studies an index type formula for foliations. We will be only concerned with the first two parts of [Con79]. The starting point of the noncommutative integration is the fact that cer- tain quotients spaces (e.g., those coming from ergodic actions) do not admit a nontrivial measure theory, i.e., there do not exist many invariant measurable 6 D.LENZ,N.PEYERIMHOFF,ANDI.VESELIC´ functions. To overcome this difficulty the inaccessible quotient is replaced by a nicerobject, agroupoid. Groupoidsadmitmanytransversefunctions,replacing the invariant functions on the quotient. In fact, thenotion of invariant function can be further generalized yielding the notion of random variable in the sense of [Con79]. Such a random variable consists of a suitable bundle together with a family of measures admitting an equivariant action of the groupoid. This situation gives rise to the so called von Neumann algebra of random operators and it turns out that the random operators of the form (H ) introduced above ω are naturally affiliated to this von Neumann algebra. Moreover, each family (H ) of random operators gives naturally rise to many random variables in the ω sense of Connes. Integration of these random variables in the sense of Connes yieldsquitegeneralproofsformainfeaturesof randomoperators. Inparticular, an abstract version of the integrated density of states is induced by the trace on the von Neumann algebra. Acknowledgments: ItisapleasuretothankW.KirschandP.Stollmannforvar- ious stimulating discussions on random operators and hospitality at Ruhr-Universit¨at Bochum, respectively TU Chemnitz. This work was supported in part by the DFG through the SFB 237, the Schwerpunktprogramm 1033, and grants Ve 253/1 & Ve 253/2 within the Emmy-Noether Programme. 2. Abstract setting for basic geometric objects: groupoids and random variables In this section we discuss an abstract generalization for the geometric situa- tion given in example (RSM). The motivation for this generalization is that it covers many different settings at once, such as tilings, percolation, foliations, equivalence relations and our concrete situation, a group acting on a metric space. Let us first introduce some basic general notations which are frequently used inthispaper. Foragivenmeasurablespace(S,B)wedenotethesetofmeasures by M(S) and the corresponding set of measurable functions by F(S). The symbol F+(S) stands for the subset of nonnegative measurable functions. M f denotes the operator of multiplication with a function f. Webeginourabstractsettingwithageneralization of theaction ofthegroup Γ on the measurable space (Ω,B ). This generalization, given by G = Ω×Γ in Ω thecase athand,is called agroupoid. Themain reason toconsider itis thefact that it serves as a useful substitute of the quotient space Ω/Γ, which often is a very unpleasant space (e.g., in the case when Γ acts ergodically). The general definition of a groupoid is as follows [Ren80]. Definition 2.1. A triple (G,·,−1) consisting of a set G, a partially defined associative multiplication ·, and an inverse operation −1 : G → G is called a groupoid if the following conditions are satisfied: • (g−1)−1 = g for all g ∈ G, • If g ·g and g ·g exist, then g ·g ·g exists as well, 1 2 2 3 1 2 3 GROUPOIDS, VON NEUMANN ALGEBRAS AND THE IDS 7 • g−1·g exists always and g−1·g·h= h, whenever g·h exists, • h·h−1 exists always and g·h·h−1 = g, whenever g·h exists. A given groupoid G comes along with the following standard objects. The subset G0 = {g·g−1 | g ∈ G} is called the set of units. For g ∈ G we define its range r(g) by r(g) = g·g−1 and its source by s(g) = g−1 ·g. Moreover, we set Gω = r−1({ω}) for any unit ω ∈ G0. One easily checks that g ·h exists if and only if r(h) = s(g). Thegroupoidsunderconsiderationwillalwaysbemeasurable,i.e.,theyposses a σ-algebra B such that all relevant maps are measurable. More precisely, we require that · :G(2) → G, −1 :G → G, s,r : G → G0 are measurable, where G(2) := {(g ,g ) |s(g )= r(g )} ⊂ G2 1 2 1 2 and G0 ⊂ G are equipped with the induced σ-algebras. Analogously, Gω ⊂ G are measurable spaces with the induced σ-algebras. Asmentionedabove,thegroupoidassociatedwith(RSM)issimplyG = Ω×Γ and the corresponding operations are defined as (3) (ω,γ)−1 = (γ−1ω,γ−1), (4) (ω ,γ )·(ω ,γ )= (ω ,γ γ ), 1 1 2 2 1 1 2 where the left hand side of (4) is only defined if ω = γ ω . It is very useful to 1 1 2 consider the elements (ω,γ) of this groupoid as the set of arrows γ−1ω 7−γ→ ω. This yields a nice visualization of the operation · as concatenation of arrows and of the operation −1 as reversing the arrow. The units G0 = {(ω,ǫ) | ω ∈ Ω} can canonically be identified with the elements of the probability space Ω. Via this identification, the maps s and r assign to each arrow its origin and its destination. Our groupoid can be seen as a bundle over the base space Ω of units with the fibers Gω = {(ω,γ) | γ ∈ Γ} ∼= Γ. For simplicity, we henceforth refer to the set of units as Ω also in the setting of an abstract groupoid. The notions associated with the groupoid are illustrated in Figure 1 for both the abstract case and the concrete case of (RSM). Next, we introduce an appropriate abstract object which corresponds to the Riemannian manifold X in (RSM). Definition2.2. LetG beameasurablegroupoidwiththepreviouslyintroduced notations. A triple (X,π,J) is called a (measurable) G-space if the following properties are satisfied: X is a measurable space with associated σ-algebra B . X The map π : X → Ω is measurable. Moreover, with Xω = π−1({ω}), the map J assigns, to every g ∈ G, an isomorphism J(g) : Xs(g) → Xr(g) of measurable spaces with the properties J(g−1) = J(g)−1 and J(g ·g ) = J(g )◦J(g ) if 1 2 1 2 s(g )= r(g ). 1 2 Note that a picture similar to Figure 1 exists for a G-space X. An easy observation is that every groupoid G itself is a G-space with π = r and J(g)h = g·h. 8 D.LENZ,N.PEYERIMHOFF,ANDI.VESELIC´ The G-space in (RSM) is given by X = Ω × X together with the maps π(ω,x) = ω and J(ω,γ) :Xγ−1ω → Xω, J(ω,γ)(γ−1ω,x) = (ω,γx). Similarly to the groupoid G, an arbitrary G-space can be viewed as a bundle over the base Ω with fibers Xω. Our next aim is to exhibit natural measures on these objects. We first intro- duce families of measures on the fibers Gω. In the case of (RSM), this can be viewed as an appropriate generalization of the Haar measure on Γ. Definition 2.3. Let G be a measurable groupoid and the notation be given as above. (a) A kernel of G is a map ν : Ω → M(G) with the following properties: • the map ω 7→ νω(f) is measurable for every f ∈ F+(G), • νω is supported on Gω, i.e., νω(G −Gω) = 0. (b)Atransverse functionν of G is akernel satisfyingthefollowing invariance condition f(g·h)dνs(g)(h) = f(k)dνr(g)(k) ZGs(g) ZGr(g) for all g ∈ G and f ∈ F+(Gr(g)). In (RSM) the discreteness of Γ implies that any kernel ν can be identified with a function L ∈ F+(Ω×Γ) via νω = L(ω,γ)δ . For an arbitrary γ∈Γ (ω,γ) unimodular group Γ, the Haar measure m induces a transverse function ν by Γ νω = m for all ω ∈ Ω on the groupoid Ω×PΓ via the identification Gω ∼= Γ. Γ In the next definition we introduce appropriate measures on the base space Ω of an abstract groupoid G. Definition 2.4. Let G be a measurable groupoid with a transverse function ν. A measure µ on the base space (Ω,B ) of units is called ν-invariant (or simply Ω invariant, if there is no ambiguity in the choice of ν) if µ◦ν = (µ◦ν)∼, where (µ ◦ν)(f) = νω(f)dµ(ω) and (µ ◦ν)∼(f) = (µ ◦ν)(f˜) with f˜(g) = Ω f(g−1). R In (RSM) it can easily be checked that, with the above choice ν ≡ m , a Γ measureµon(Ω,B )isν-invariant ifandonlyifµis Γ-invariantintheclassical Ω sense, see[Con79, Cor. II.7] as well. Thus acanonical choice for am -invariant Γ measure on Ω is P. Analogously to transverse functions on the groupoid, we introduce a corre- spondingfiberwiseconsistent family α of measures on the G-space, see the next definition. We refer to the resulting object (X,α) as a random variable in the sense of Connes. These random variables are useful substitutes for measurable functions on the quotient space Ω/Γ with values in X. Measurable functions on Ω/Γ can be identified with Γ-invariant measurable functions on Ω. Note that, GROUPOIDS, VON NEUMANN ALGEBRAS AND THE IDS 9 Gω′ = Gs(g) Gω = Gr(g) g· k = g·h G = Ω×Γ h r Ω ω′ =s(g) (Ω,µ) ω = r(g) = r(h) = r(k) g (Gω,νω)≡ (Γ,m ) Γ g = (ω,γ) ∼= ω′ 7−γ→ ω (cid:16) (cid:17) Figure 1. Notations of the groupoid G =Ω×Γ in (RSM) in the case of an ergodic action of Γ on Ω, there are no nontrivial Γ-invariant measurable functions, whereas there are usually lots of random variables in the sense of Connes (see below for examples). Definition 2.5. LetG beameasurablegroupoidandX beaG-space. Achoice of measures α : Ω→ M(X) is called a random variable (in the senseof Connes) with values in X if it has the following properties • the map ω 7→ αω(f) is measurable for every f ∈F+(X), • αω is supported on Xω, i.e., αω(X −Xω) = 0, • α satisfies the following invariance condition f(J(g)p)dαs(g)(p) = f(q)dαr(g)(q) ZXs(g) ZXr(g) for all g ∈ G and f ∈ F+(Xr(g)). To simplify notation, we write gh respectively gp for g·h respectively J(g)p. The general setting for the sequel consists of a groupoid G equipped with a fixed transverse function ν and an ν-invariant measure µ on Ω, and a fixed random variable (X,α). We use the follwing notation for the “averaging” of a u∈ F+(X) with respcet to ν ν ∗u (p):= u (g−1p)dνπ(p)(g) for p ∈ X. 0 0 ZGπ(p) We will need the following further assumptions in order to apply the integra- tion theory developed in [Con79]. Definition 2.6. Let(G,ν,µ)beameasurablegroupoidand(X,α)bearandom variable on the associated G-space X satisfying the following two conditions (6) The σ-algebras B and B are generated by a countable family of sets, X Ω all of which have finite measure, w.r.t. µ◦α (respectively w.r.t. µ). 10 D.LENZ,N.PEYERIMHOFF,ANDI.VESELIC´ (7) There exists a strictly positive function u ∈ F+(X) satisfying 0 ν ∗u (p)= 1 for all p ∈X. 0 Then we call the tupel (G,ν,µ,X,α,u ) an admissible setting. 0 Before continuing our investigation let us shortly discuss the relevance of the above conditions: Condition (6) is a strong type of separability condition for the Hilbert space L2(X,µ◦α). It enables us to use the techniques from direct integral theory discussed in AppendixA which are crucial to the considerations in Section 3. Condition (7) is important to apply Connes’ non-commutative integration theory. Namely, it says that (X,J) is proper in the sense of Lemma III.2 and Definition III.3of[Con79]. Therefore(X,J) issquareintegrable byProposition IV.12 of [Con79]. This square integrability in turn is a key condition for the applications of [Con79] we give in Sections 3 to 5. On an intuitive level, (7) can be understood as providing an “embedding” of G into X. Namely, every u ∈ F(X) with ν ∗u ≡ 1 gives rise to the fibrewise defined map q = q : F(G) → F(X) by u (8) q(f)(p) := u(g−1p)f(g)dνπ(p)(g) ZGπ(p) for all p ∈ X. Note that the convolution property of u implies that the map (8) satisfies q(1 ) = 1 . Moreover, q can be used to obtain new functions G X w ∈ F(X) satisfying ν ∗w ≡ 1. This is the statement of the next proposition. Proposition 2.7. Let u ∈ F+(X) with ν ∗u ≡ 1 be given, and q be as above. For any function f ∈ F(G) with ν(f˜) ≡ 1 on Ω we have ν ∗q(f)≡ 1. Proof. Theproofis given bythefollowing directcalculation withω = π(p) ∈ Ω: (ν ∗q(f))(p) = q(f)(g−1p)dνω(g) ZGω = u(h−1g−1p)f(h)dνs(g)(h)dνω(g) ZGω ZGs(g) (ν transverse function) = u(k−1p)f(g−1k)dνω(k)dνω(g) ZGω ZGω (Fubini) = u(k−1p) f(g−1k)dνω(g)dνω(k) ZGω ZGω = u(k−1p) f˜(k−1g)dνω(g)dνω(k) ZGω ZGω (ν(f˜)≡ 1) = u(k−1p)dνω(k) ZGω = 1. Note that in the above calculation the integration variable g has the property r(g) = π(p) = ω. This finishes the proof. 2