Ergodic theory of affine isometric actions on Hilbert spaces (Q2126275): Difference between revisions

Let \(\sigma\) be a measure-preserving action of a locally compact abelian polish group \(G\) on a probability space \((X, \mu)\). Then, by the spectral theorem, there is an orthogonal representation of \(G\) on \(L^2(X, \mu)\) which is the sum of the unit representation and a countable sum of representations equivalent to the regular representation (that is, the representation of \(G\) obtained in \(L^2(G)\) by translations). Conversely, given an orthogonal representation \(\pi\) of \(G\) on a Hilbert separable space \(H\), there is a measure-preserving action of a probabilty space of \(G\) called the Gaussian action of \(\pi\). Here the authors consider the similar problem for nonsingular group actions. Let \(H\) be a real affine Hilbert separable space, that is, a Hilbert space without an origin. By rescaling the metric of \(H\) by a parameter \(t > 0\), we obtain a new affine Hilbert space \(H^t\). Moreover, if \(\alpha\) is an affine isometric action of \(G\) of a locally compact polish group \(G\) on \(H\) (that is, a continuous group homomorphism \(\alpha : G \mapsto \mathrm {Isom}(H)\)), then we can construct a new affine isometric action \(\alpha^t\) on \(H^t\) defined by \(\alpha_g^t(tx)=t\alpha_g (x).\) We thus have \((\alpha^t)^s=\alpha^{ts}\), for all \(t,s \geq 0\). The (quadratic) Poincaré exponent \(\delta(\alpha)\) is defined as the infinimum of all \(s > 0\) such that \[ \int_G e^{-s\|gx-y\|^2} dg <+\infty, ~~~x,y \in H.\] This number does not depend on the choice of \(x, y \in H\). If \(\alpha\) is the affine isometric action of a discrete automorphism group \(\Gamma<\mathrm{Aut}(T)\) of a locally finite tree \(T\) (which can be viewed as a metric space where the distance \(d(x, y)\) between two vertices \(x, y \in T\) is simply the length of the segment \([x,y]\)), then \(\delta(\alpha)=\delta(\Gamma)\) where \(\delta(\Gamma)\) the Poincaré exponent of \(\Gamma\). This exponent is defined, for a given discrete isometry groups \(\Gamma\) of the group \(\mathrm{Isom}(X)\), where \(X\) is a metric space, as the infinimum of all \(s > 0\) such that \[ \sum_{g \in \Gamma} e^{-sd(gx,y)} <+\infty,\] where \(x,y\) is any given pair of points of \(X.\) In this setting, the authors prove that the Krieger type of the Gaussian actions \({\widehat{\alpha}}^t\) for \(t >\) 0 is ergodic of type \(\textrm{III}_1\) if \(t<2\sqrt{2\delta}\) and not ergodic and of type \(I\) (it has a fundamental domain) if \(t>2\sqrt{2\delta}\). The study of the critical point \(t=2\sqrt{2\delta}\) is left as an open problem. Moreover, if \(\delta<+\infty\) and \(t<2\sqrt{2\delta}\), then (i) the actions \({\widehat{\alpha}}^t\) are pairwise non-conjugate, and (ii) for \(t\) small enough, \({\widehat{\alpha}}^t\) is strongly ergodic. We recall that the nonsingular action \(\alpha\) is strongly ergodic if every almost invariant bounded sequence \((a_n)_{n \in \mathbb{N}} \subset L^\infty(X)\) is trivial, i.e., there exists a bounded complex sequence \((\lambda_n)_{n \in \mathbb{N}}\) such that \(a_n-\lambda_n 1 \) converges to \(0\) in the measure topology. Let us also recall that any orthogonal representation \(\pi : G \rightarrow \mathcal{O}(H)\) has spectral gap if it has no almost invariant vectors. This property plays a fundamental rule in the proof of (ii). It is well known that the strong ergodicity implies that the action is not amenable in the sense of Zimmer. We recall that a nonsingular action \(\alpha\) is amenable in Zimmer's sense, if there exists a \(G\)-equivariant conditional expectation \(\mathbb{E}:L^\infty(G \otimes X) \rightarrow L^\infty(X)\) where \(G\) acts diagonally on \(G \otimes X\). They further establish that for any affine isometric action of a locally compact polish group, there exists \(t_{\mathrm{diss}(\alpha)} \in [\sqrt{2 \delta(\alpha)},2\sqrt{2 \delta(\alpha)}]\) such that the Gaussian action \(\widehat{\alpha}\) the Gaussian action is recurrent for all \(t< t_{\mathrm{diss}(\alpha)}\) and dissipative for all \(t > t_{\mathrm{diss}(\alpha)}.\) The computation of the exact value of \(t_{\mathrm{diss}(\alpha)}\) seems to be difficult and the study of the case \(t=t_{\mathrm{diss}(\alpha)}\) is left as an open problem. For actions without property Kazhdan's property (T), they prove that for \(t>0\), the Gaussian action is free, weakly mixing of stable type \(\mathrm{III}_1\), has almost vanishing entropy with an invariant mean and is nonamenable if \(G\) is nonamenable. Let us further recall that a nonsingular action \(\sigma\) is said to have almost vanishing entropy if for every \(\varepsilon > 0\) and every compact set \(K \subset G\) there exists a faithful probability measure \(\eta\) on \(X\) such that \(\forall g \in K\), \(H(g_{*}\eta |\eta) \leq \varepsilon\), where \(H(\mu|\nu)\) is the relative entropy given by \[H(\mu|\nu)=-\int_X \log\Big(\frac{d\mu}{d\nu}\Big) d\nu.\] When \(\sigma\) is the left-translation action on \(G\), we say that \(G\) has almost vanishing entropy.