Boundary unitary representations -- irreducibility and rigidity (Q550315)

The authors use functional analysis to adapt an ergodic theorem of \textit{G. A. Margulis} [Funct. Anal. Appl. 3, 335--336 (1969); translation from Funkts. Anal. Prilozh. 3, No. 4, 89--90 (1969; Zbl 0207.20305)], in order to construct a group representation in a Hilbert space for a dynamic flow of measure equivalence classes on the boundary of a Gromov hyperbolic space. Their group algebra has, as coefficients, operators on a Hilbert space, but the construction is problematic. The reviewer has changed some of the authors' notation. They consider a compact Riemannian manifold \((M,g)\) of dimension greater than 2, which is a \(\mathrm{CAT}(-1)\) space (Cartan-Aleksandrov-Toponogov comparison theorem for angles in a triangle); it has negative sectional curvature and is nowhere flat. It is hyperbolic in that, for instance, the sum of the angles at the vertices in a geodesic triangle will be less than 180 degrees and the triangle will be `thinner' than a corresponding triangle with the same vertices in Euclidean space. The \(\delta\) in Gromov's \(\delta\)-hyperbolic space relates to this thinning. The fundamental group, denoted \(\Gamma\), is discrete and acts as isometries on the universal cover \(X\) of \((M,g)\) where this is given the lifted Riemannian metric \(\tilde g\) and the associated metric \(d\). It is compactified by attaching a boundary so that \(\overline{X}\) is a topological sphere. The authors have chosen the radial limit set of all accumulation points of orbits of \(\Gamma\) starting at a base-point \(p \in X\); this is denoted by \(\partial X\). It is independent of the base-point chosen. There are no geodesics on the boundary nor is Ascoli's theorem valid. The radial set of accumulation points is provided with the Patterson-Sullivan measure \(\nu_{p}\) which is chosen such that \(\nu_{p}(\partial X) = 1\). We shall drop the suffix \(p\) when it is unnecessary. The radial limit set is the topological boundary as it seems that the authors have assumed the action of \(\Gamma\) is convex cocompact. Besides the boundary having been given topological-geometrical and measure-theoretical structures, metric structures can be constructed using suitable cocycles. When considering uniform boundedness in their \S4, the authors postulate that the diameter of \(M\) is bounded by \(R\). In order to prove Theorem 2.2 one needs to eliminate the dependence on the bound. It is known that \(\nu\) is (Ahlfors) \(d\)-regular in that there exists a \(c \geq 1\) such that for \(0 < t \leq \hbox{diam supp}(\nu)\) one has \(c^{-1}t^{d} \leq \nu B(p,t) \leq ct^{d}\), where \(B(p,t)\) denotes the ball with centre \(p\) and radius \(t\). This kind of sandwiching occurs throughout the paper and is called quasi-whatever. The critical exponent \(\eta = \eta(\Gamma)\) is such that the Poincaré series \(\sum_{\gamma \in \Gamma}{e^{-sd(p,\gamma q})}\), \(p,q \in X\), converges if \(s > \eta\) and diverges if \(s < \eta\). (Divergence at \(s = \eta\) relates to ergodicity of the group action.) The exponent can also be expressed in terms of orbits: Let \(n(r)\) denote the number of points of a \(\Gamma\)-orbit in a ball \(B(p,r) \subset X\). The cardinality \(n(r)\) increases exponentially with \(r\). Using concentric balls in \(X\), the exponent can also be expressed as \(\overline{\hbox{lim}}_{r} \hbox{log} {n_{r} \over r}\). \textit{D. Sullivan} [Publ. Math., Inst. Hautes Étud. Sci. 50, 171--202 (1979; Zbl 0439.30034)] demonstrated the equivalence of the definitions by associating normalised scale factors, using \(e^{- \eta s} d(p,\gamma q)\). The scale factors are weights of Dirac atomic particles as seen from \(p\). An equivalent interpretation of \(\eta\) is as the Hausdorff measure of \(\partial X\); cf. \textit{D. Sullivan} [Acta Math. 153, 259--277 (1984; Zbl 0566.58022), \S 6]. Denote by \(b(r)\) the number of closed geodesics in \(M\) with multiplicity one whose lengths do not exceed \(r\). \textit{Ya. G. Sinai} [Izv. Akad. Nauk SSSR, Ser. Mat. 30, 1275--1296 (1966; Zbl 0146.18103)] proved a theorem on asymptotics for \(b(r)\) conditioned on the upper and lower bounds of the sectional curvature of \((M,g)\). \textit{A. Avez} [C. R. Acad. Sci., Paris, Sér. A 270, 1512--1513 (1970; Zbl 0197.48002)] proved the theorem in a similar form, viz., assuming the sectional curvature between \( -\alpha^{2}\) and \(-\beta ^{2}\), and denoting the number of closed geodesics of multiplicity 1 and lengths \(r \geq 0\) by \(b(r)\), then \[ e^{( n-1)\beta} \leq \liminf_{r \to +\infty} b(r)^{1 \over r} \leq\limsup_{r \to +\infty} e^{(n-1)\alpha}\leq e^{(n-1)\alpha}, \] Margulis [loc.\,cit.] denoted by \(a_{x,y}(r)\), with \( x,y \in X\), essentially the number of orbits from \(x\) passing within a distance \(r\) from \(y\); he announced his results in his doctoral thesis giving asymptotic estimates for these functions in terms of the bounds on the sectional curvatures; he used Sinai's theorem [loc.\,cit.] and also a theorem on the existence of the measure dealt with in his paper [\textit{G. A. Margulis}, Funct. Anal. Appl. 4, 55--67 (1970); translation from Funkts. Anal. Prilozh. 4, No. 1, 62--76 (1970; Zbl 0245.58003)]. This measure, now called the Margulis-Bowen measure, splits the tangent bundle into three Anosov-flow-invariant tangent bundles, each the direct sum of a \(2n\)-dimensional tangent space and an interval of the line with Lebesgue measure. The proof of his theorem involves the trick, due to \textit{D. V. Anosov} and \textit{Ya. G. Sinai} [Usp. Mat. Nauk 22, No. 5(137), 107--172 (1967; Zbl 0177.42002)], of using ergodic theory (which does not involve points) to prove geometric results by approximating points by letting small neighbourhoods asymptotically contract. The authors assume these neighbourhoods are open, presumably because this eliminates measures of their boundaries. One needs then to count the (finite) number of flows passing through a neighbourhood close the boundary. The authors start by connecting flows of measure classes to quasiregular representations. It turns out that they need quasiregularity to construct a flow of measures on the boundary. The boundary representation is a cocycle representation, a particular projective representation (cf. [\textit{C. C. Moore}, Contemp. Math. 449, 263--300 (2008; Zbl 1158.22004)]). The boundary representation, denoted \(\rho\), of \(\Gamma\) as operators on the Hilbert space \(\mathcal{H} =L^{2}(\partial X,\nu)\) is such that \[ \rho(\gamma)\lambda(\xi) = e^{-{1 \over 2}{\eta \beta_{\xi}(p,\gamma p)}}\lambda(\rho(\gamma^{-1} \xi)), \] where \(\xi \in \partial X\) and \(\lambda \in \mathcal{H}\), is indeed a valid quasiregular representation. A regular representation involves the action of \(\Gamma\) on itself by (left) multiplication but there is now also a multiplier function, a character of \(\Gamma\) which is a power of the Radon-Nikodým derivative \( \xi \mapsto {d\nu_{\gamma* \xi} \over d\nu_\xi}\) (where \(\gamma*\nu(A) = \nu(\gamma A)\) for all measurable \(A\)) and all \(\gamma \in \Gamma\) [Sullivan (1984), loc.\,cit.] equates \(\gamma* \nu\) with the \(\eta\)'th power of `distortion' of \(\nu\).) The Radon-Nikodým density function is measurable but generally not defined everywhere on \(\partial X\). The Gromov product \((x|y)_{p}\) (based at \(p\)) for a \(\delta\)-hyperbolic metric space is defined to be \({1\over 2}[d(p,x) + d(p,y) - d(x, y)]\), for \( x, y \in X\). It connects the topology and the geometry; heuristically, it measures the distance along which two geodesic rays from \(p\) remain within a finite Hausdorff distance 2\(\delta\) of each other. The product can be extended to \(\partial X\), but there is usually no continuous extension of the Gromov product to \(\overline{X}\). The Busemann function \(b_{\gamma}\) associated with a geodesic ray \(r_{t}\) from a \(p\) to \(\gamma p\) and continuing to \(\xi \in \partial X\) is defined as \(\lim_{t}{|d(p,r_{t}) - t |}\). The level sets of \(b_{\lambda}\) are called horospheres or boundary spheres (see N. I. Lobachevski's imaginary geometry (1835) and E. Beltrami's work from 1868); an orbit may leave from a boundary point \(\zeta\) traversing `expanding' horospheres. and it will converge, traversing `contracting' horospheres, to \(\xi\). The Busemann function is defined only up to a constant dependent on the base point chosen, so in its stead \textit{V. A. Kajmanovich} [Ann. Inst. Henri Poincaré, Phys. Théor. 53, No. 4, 361--393 (1990; Zbl 0725.58026)] constructed the (measurable) Busemann 2-cocycle as \(\beta_{\gamma}(p,q) =b_{\gamma} (p) - b{_\gamma}(q)\). Since the definition involves going to \(\xi\), and all rays within a finite distance of each other may be considered as an equivalence class, we may denote it as \(\beta_{\xi}(p,q)\). There will be a unique flow in \(\overline{X}\) from \(\zeta\) to \(\xi\) as an indirect flow of mass on the boundary. In [\textit{V. Kaimanovich} (ed.) et al., Representation theory, dynamical systems, and asymptotic combinatorics. Based on the conference, St. Petersburg, Russia, June 8--13, 2004 on the occasion of the 70th birthday of Anatoly Vershik. Providence, RI: American Mathematical Society (AMS) (2006; Zbl 1098.22001)], the cocycle was extended to the boundary, and it may be written as \(\beta_\xi(\lambda.\zeta)\). `Looking' along geodesic rays from \(p\) to points on the boundary one gets local `visual metrics at' these points on the boundary. These metrics are different but are conformally equivalent in that they share the same angular structure for isomorphic neighbourhoods, differing only by a conformal scale factor; the conformal function on the boundary can be thought of as scaling infinitesimal patches of the boundary. Any isometry that takes the one interior point to another is a conformal mapping on the boundary sphere such that the derivatives at these points are isometries of the tangent spaces, times the conformal factor. Another postulate of the authors is that there is a quasi-conformal density on \(\partial X\). Quasi-conformal densities were introduced by \textit{M. Coornaert} [Sur les groupes proprement discontinus d'isométries des espaces hyperboliques au sens de Gromov. Dissertation, Université Louis Pasteur, Strasbourg (1990)]. To achieve his results, he used a theorem of \textit{M. Gromov} [Publ., Math. Sci. Res. Inst. 8, 75--263 (1987; Zbl 0634.20015)] that any \(\delta\)-hyperbolic space can be approximated arbitrarily closely by a (metric) tree (i.e., any subset homeomorphic to an interval of the line is a geodesic; also \(\delta = 0\)). The Radon-Nikodým density function is \(e^{-\eta \beta_{\xi}(p,q)}\). A conformal density is called quasi-conformal if it is sandwiched between \(C^{\pm 1}e^{-\eta \beta_{\gamma}(p,q)}\), for large enough \( C \geq 1\), \(\nu\)-almost everywhere on \(\partial X\). A tree has a boundary made up of its endpoints (so of Dirac measures). The Patterson-Sullivan boundary can be thought of as Dirac measures at all radial limit points, \(\nu\) being the weak limit of these Dirac measures. I found the authors' attempt at a proof that the Radon-Nikodým density is positive with \(L^{1}\) bounds above and below unconvincing, in that they change the geometry and topology of \(X\) but seem not to show that the changed topologies converge uniformly to the topology of the boundary. They define the vector, \(\lambda^{\gamma p}\), a strange notation which varies with the boundary point as it involves \(\beta_{\xi}\). It is defined to be the function \(e^{-{ 1 \over 2} \eta \beta_{\xi}(p, q)}\), a vector in \(\mathcal{H}\). They have then replaced the Gromov product by \(\overline{(q|\xi)} = \min[(z^q_p|\xi)_pd(p,q)]\), where a geodesic ray from \(p\) is cut off at \(z^q_p\), a point on the ray before it reaches the boundary. Presumably this is what the authors mean by `chopping'. The chopped vector \(\overline{\lambda}^{\gamma p} = e^{-{1 \over 2}\eta} \beta_{\gamma}(p.z^{\gamma p}_{p})\) needs to be clearly defined in \(\mathcal{H}\). For what follows, the authors fix a base point \(p \in X\) and proceed to approximate \(\partial X\) by sets of small balls, doing a Shannnon-type sampling of enough points on the boundary to get optimally uniform bounds for the vectors \(\lambda^{\gamma p}\). It seems that they should be doing this in neighbourhoods of the boundary in \(\overline{X}\), so not divorcing the boundary from \(X\). The technique comes from image processing where images are dealt with as Riemannian manifolds, cf. [\textit{N. Sochen} and \textit{Y. Y. Ze'evi}, in: Mathematical methods for curves and surfaces II. 2nd international conference, Lillehammer, Norway, July 3--8, 1997. Nashville, TN: Vanderbilt Univ. Press, 469--476 (1998; Zbl 0901.68217)]. Take \(S\) to be a finite sampling set of elements in \(\Gamma\) such that the points \(\gamma p\) are in the neighbourhood of \(\partial X\) so that the region is covered by a union of balls of fixed radius about points close to the \(z^{\gamma p}_p\) and also such that there is a finite number of \(\gamma p \) contained each ball. To each \(t > 0\) they associate a set \(S_{t} \subset S\). It can be shown, cf. Corollary 4.3, that there are suitable sampling sets \(S_t\) with balls centred at the \(z^{\gamma p}_p\) and radii \(r(t)\) decreasing as \(t\) increases. They could then allocate the sets \( S_t\) in such a manner that the centres of the balls at these points approach the boundary as \(t \to \infty\). In their presentation of Theorem 2.2, there are two limiting procedures, taking \(t \to \infty\) extends the operators to \(\partial X\); once there, they need to compute the limits, in the von Neumann topology, of operator-valued coefficients of the group algebra. The essential part of Theorem 2.2 is to prove that the ranges of operators \(T_t\) in \(\mathcal{L}(\mathcal{H})\) are in the positive cone of the von Neumann algebra generated by the span of \(\rho(\Gamma)\) in \(\mathcal{L}(\mathcal{H})\). This assertion is a consequence of Proposition 3.5. The operator-valued coefficients of the group algebra are the limits as \(t \to \infty\) of \({1\over {|S_{t}|}} \sum_{\gamma \in S_t} f(z^{\gamma p}_p).\) Thus \(f\) seems to be a function on \(X\), not \(\partial X\). Using the quasi-conformal bounds, Proposition 4.5 proves that the limits of the \({1 \over{|S_t|}} \sum_{\gamma \in S_t} {{\rho(\gamma)1} \over {\langle \rho(\gamma) 1 | 1\rangle}}\) are uniformly \(L^\infty\)-bounded as \(t \to\infty\). In their proof, the authors seem to be approaching the \(\beta_\xi\) by a \(\beta\) defined at the visual image on \(\partial X\) of \(z^{\gamma p}_p\), as seen from \(p\). These bounds are conditioned, as in Lemma A.6, which considers the sampling process, so that the Borel functions \(f\) on \(\partial X\) are almost continuous. This means that, for a small enough arc on the boundary between \(\xi\) and \(\zeta\), the difference \(|\log f(\xi) - \log f(\zeta)|\) is bounded. The \(T_{t}\) operate as \(T_{t}\rho(\gamma)\) on the chopped vectors. The authors illustrate their Theorem 2.2, using an \(f\) which is identically 1 on the boundary, to get a multiplication operator \(\lambda \mapsto \int_{\partial X}{\lambda(\xi) d\nu(\xi)}\cdot \lambda\), where \(\lambda \in \mathcal{H}\). They illustrate the more general case involving \(\sum {{f(z^{\gamma p}_p)} \over {\langle\rho(\gamma)1 | 1\rangle}}\), where the limiting operators are well-defined on \(L^\infty(\partial X)\). The authors' Theorem 2.1 is unnecessary. They seem confused about the difference and similarity between a von Neumann and a W*-algebra. A W*-algebra is a C*-algebra which has a (unique) Banach predual. Concrete von Neumann and W*-algebras are essentially the same thing. The proof of Theorem 2.2 is unnecessarily complicated. The authors' Corollary B.2 is a direct consequence of a well-known theorem in that their predual has dual \(CB(A,{\mathcal {L}}{(\mathcal{H}})\), where \(A\) an operator algebra. In the article under review, \(A\) is the commutative operator algebra \(L^{\infty}(\partial X)\) with predual \(\mathcal{L}^1(\partial X)\) and \(\mathcal{L}(\mathcal{H})\) has as predual the Banach space of trace-class operators. This predual is relevant in that its elements can be identified with \(\mathcal{L}(\mathcal{H})\)-valued measures on \(\partial X\). The theorem was proved by \textit{E. G. Effros} and \textit{R. Exel} [Lond. Math. Soc. Lect. Note Ser. 135, 81--94 (1988; Zbl 0698.46049)], and \textit{E. G. Effros} and \textit{Z.-J. Ruan} [Operator spaces. London Mathematical Society Monographs. New Series. 23. Oxford: Clarendon Press (2000; Zbl 0969.46002)] using completed projective tensor products of the Banach spaces. The authors' notation is confusing as they use \(H\) to denote both the Hilbert space and a Banach space, Theorem 1.1 follows from Theorem 2.2. A length spectrum, cf. [\textit{C. Croke}, Topology 31, No. 4, 847--855 (1992; Zbl 0779.53025)], as opposed to an eigenvalue spectrum, comprises lengths of closed geodesic flows, multiplicities being the number of flows with the same length. The marked length spectrum counts homotopy classes of flows and/or conjugacy classes in \(\Gamma\). Theorem 6.1, and hence also Theorem 1.2, is a straightforward statement of known results on equivalence of manifolds. However, this is not a proper issue in an article concerning a fixed manifold.