The Haagerup property and actions on von Neumann algebras

It is clear that the given action extends to a unitary representation on L^2(N,\varphi). This follows from the fact that G acts by isometries on \mathfrak{n}_{\varphi} and then it suffices to consider their extension. In other words, to obtain an unitary representation it suffices to set for every g\in G

where (x_n)_{n\ge 1} \subset \mathfrak{n}_{\varphi} is any sequence such that \|x_n-x\|_{\varphi}\xrightarrow{n\to \infty} 0. The strong continuity of \pi follows from Lemma A.6.2 on page 309 of [bekka2008kazhdan]. Indeed we assumed that L^2(N,\varphi) is separable and the following sequence of functions defined by

where (x_n)_{n\ge 1}\subset \mathfrak{n}_{\varphi}, is a sequence of measurable functions. Therefore for every \xi\in L^2(N,\varphi) fixed, we have that the map g\mapsto \langle\pi(g)\xi, \xi\rangle_{\varphi} is measurable as it is the pointwise limit of the sequence (f_n)_n.

Next, we show the converse. Note that \varphi(x^* \alpha_g(x)) = \langle \pi_g(x),x \rangle for every x\in \mathfrak{n}_{\varphi}. As \pi is strongly continuous, we have that the maps g \mapsto \varphi(x^* \alpha_g(x)) are continuous and hence measurable. So, it suffices to show that \alpha preserves the weight \varphi. Suppose it is not preserved. As it is the linear extension of a map on the positive elements of N, we can take x\in N and g\in G such that \varphi(x^*x)\neq \varphi\circ\alpha_g(x^*x). Without loss of generality, we may assume that \varphi(x^*x) is finite. However, that implies that \|x\|_{\varphi}^2\neq \varphi\circ\alpha_g(x^*x) = \|\alpha_g(x)\|_{\varphi}^2, which contradicts the fact that \pi is unitary.

The set \mathfrak{n}_{\varphi} is total in L^2(N,\varphi) by construction, and for all x,y\in \mathfrak{n}_{\varphi}

as g\rightarrow \infty. This proves that \pi is of class C_0. Moreover, 1_G\prec \pi by the existence of an almost invariant sequence.

If G has such an action on L^{\infty}(\Omega), then G has the Haagerup property by Theorem 2.5. Now, if G has the Haagerup property, then by theorem 2.2 there exists a C_0 dynamical system (\Omega,\mathcal{B},\mu,G) which almost has invariant vectors. Now take the commutative von Neumann algebra N=L^{\infty}(\Omega) endowed with the canonical trace given by \tau=\int_{\Omega} d\mu. Then L^2(N,\tau)=L^2(\Omega,\mu) is separable as \mu is \sigma-finite. The measurable action of G on N is given by the permutation representation \pi_{\Omega} which is invariant for \tau as \mu is G-invariant. Furthermore the condition (1) in the theorem 2.2 implies that the function g\mapsto \tau(x^*\alpha_g(x)) is measurable for every x\in \mathfrak{n}_{\tau} fixed, as well as the action is C_0 for \tau. Finally from the family (A_n) we consider the associated sequence of unit vectors \xi_n=\chi_{A_n} in N. As it satisfies the property (?), then this sequence is an almost invariant sequence for \alpha and \tau.

As we have

it suffices to show that the maps f_k\colon g\mapsto \langle \pi(g) T \pi(g)^*e_k,Te_k\rangle are measurable for all k. Using Cauchy-Schwarz inequality we have:

So, as \text{SOT}\lim_{g\rightarrow e}\pi(g)=\text{Id}, then f_k(g)\rightarrow_{g\rightarrow e} f_k(e). Thus for every k we have that f_k is continuous. In particular f_k is measurable for every k. Hence, the map g\mapsto \operatorname{Tr}(T^*\alpha_g(T)) is measurable as it is the simple limit of measurable functions.

We begin to show the existence of a sequence of almost invariant projections. As G has the Haagerup property, there exists, by Definition 2.1, a sequence of unit vectors (\xi_n)_n in \mathcal{H} such that for every compact sets K of G:

Set P_n=\Psi(\xi_n\otimes \xi_n). Then we have:

where we used Parseval’s identity. Moreover

Therefore:

Hence for every compact sets K of G we have:

It remains to show that \alpha is C_0 for \operatorname{Tr}. First take elements of the form \xi_1\otimes \eta_1, \xi_2\otimes \eta_2\in \mathcal{H}\overline{\otimes} \mathcal{H} and consider their image by \Psi denoted T_{\xi_1,\eta_1},T_{\xi_2,\eta_2} \in \mathfrak{n}_{\operatorname{Tr}}. By Cauchy-Schwarz inequality we have:

As \pi is a unitary C_0-representation, then we have

Now take T_1,T_2\in \mathfrak{n}_{\operatorname{Tr}}. By Proposition 3.1 there exists \xi,\eta\in \mathcal{H}\overline{\otimes} \mathcal{H} which represent T_1,T_2. Moreover for \varepsilon>0 fixed, there exists sequences (\xi_n),(\eta_n)\subset \mathcal{H} such that:

1. \| \xi - \sum_{n=1}^{N_{\varepsilon}} e_n \otimes \xi_n \|\le \varepsilon;

2. \| \eta - \sum_{n=1}^{N_{\varepsilon}} e_n \otimes \eta_n \|\le \varepsilon.

Thus

Hence by Cauchy Schwarz and using that \alpha is C_0 for the elements T_{\xi_i,\eta_i}, there exists a compact set K of G such that for all g\in G\setminus K :

For any finite set F\subset \Gamma one has :

We conclude the proof using the normality of the weight and equation (?).

First we prove that the action \iota of G on N\rtimes_\alpha \Gamma is C_0 for \hat{\varphi}. Let x,y\in \mathfrak{n}_{\hat{\varphi}} with x=\sum_{s\in \Gamma} \pi_{\alpha}(x(s))\lambda(s) and y=\sum_{t\in \Gamma} \pi_{\alpha}(y(t))\lambda(t). Then using the previous remark and by Cauchy-Schwarz inequality, we have:

Then by Lemma 3.7 we deduce

Thus for \varepsilon>0 fixed, there exists a finite subset K of \Gamma such that

Moreover for every h\in K, as \beta is a C_0-action for \varphi there exists a compact subset Q_h\subset G such that

Then set Q=\bigcup_{h\in K} Q_h which is a compact of G such that for all g\in G\setminus Q we have:

Now by assumption there exists an almost invariant sequence (e_k)_{k\ge 1} for \beta and \varphi. Hence considering the following sequence:

we obtain an almost invariant sequence for \iota and \hat{\varphi} as for all k we have

Let \alpha be the action of G on N and \beta be the action of G on M. Take \varphi=\varphi_N\otimes \varphi_M. By Definition 4.2 and Proposition 4.3 of [takesaki2013theory], we know that \varphi is a weight on N\overline{\otimes} M. We show that the action \gamma=\alpha\otimes\beta of G on N\overline{\otimes} M is C_0 and has an almost invariant sequence.
First, note that \operatorname{L}^2(N\overline{\otimes} M, \varphi_N\otimes \varphi_M) = \operatorname{L}^2(N,\varphi_N)\overline{\otimes} \operatorname{L}^2(M,\varphi_M). As both the extension of \alpha and \beta to \operatorname{L}^2(N,\varphi_N) and \operatorname{L}^2(M,\varphi_M) respectively are unitary representation, we have that \gamma extends to a unitary representation as well. As \alpha is C_0 for \varphi_N, then by proposition 2.5 its extension is C_0. Since the tensor product of a C_0-representation with any representation is C_0, then \gamma is C_0 for \varphi.
Finally, by proposition 2.5 both the extension of \alpha and \beta have almost invariant vectors. Since the tensor product of representations with almost invariant vectors has almost invariant vectors, then \gamma has an almost invariant sequence.

Let \sigma:G \curvearrowright (Z,\rho) be any pmp action and consider the diagonal action \gamma:G\curvearrowright (\hat{H}\times \mathbb{R})\times Z. First assume that K\subset H is a closed \pi(G)-invariant subspace and that c(g)\in K^{\perp} for all g\in G. Identifying \hat{H} with \hat{K}\times \hat{K}^{\perp}, we can see \gamma as the Gaussian action \widehat{\pi|_{K}}: G\curvearrowright \widehat{K}, the skew product action G\curvearrowright \widehat{K^{\perp}}\times \mathbb{R} and the action G\curvearrowright Z. Take any infinite cyclic discrete subgroup \Lambda of G. Then by theorem 9.1.2 of [amine1], the action \Lambda\curvearrowright \widehat{K^{\perp}}\times \mathbb{R} \times Z is conservative. Hence \Lambda is recurrent with respect to this action. Moreover by assumption, the action G\curvearrowright \hat{K} is mixing. Then \Lambda is mixing with respect to it. By theorem 4.11, we can conclude that

Therefore we are in the same situation of the proof in [amine1]. We can then conclude in the same way. By our assumption c is evanescent. Thus there exists a decreasing sequence of closed \pi(G)-invariant subspaces H_n\subset H and vectors \xi_n\in H_n^{\perp} such that

for all g\in G and n \in \mathbb{N}. Since H does not admit nonzero \pi(G)-invariant vectors, the vectors \xi_n are uniquely determined and P_{H_n^{\perp}}(\xi_m)=\xi_n for all m\ge n. Since c is not a coboundary, we have that \|\xi_n\|\to +\infty when n tends to infinity. Using (Proof) we get that for every n:

With the notations of proposition 4.6, we have that M=\emptyset and therefore A=1. So we get that

So \beta is weakly mixing.