2.9 Proper proximality and deformation/rigidity theory
2.9.1 Proper proximality via malleable deformations
wherePtdenotes the rank-one projection ontoCδt⊂`2Γ. In particular, we see thatMandJMJboth commute withC⊗`∞Γ.
We then consider the u.c.p. mapΦ:B(L2M)→
B(L2M⊗`2Γ)]J∗
given byΦ(T) =A1/2(T⊗1)A1/2. SinceAi∈c0(Γ)it then follows thatΦis bothM andJMJ-bimodular. Also sinceAi is almostΓ-invariant it follows thatΦisΓ-equivariant, where the action on
B(L2M⊗`2Γ)]J∗
is given by Ad(ut). We also have that the range ofΦmaps into the subspace of invariant operators under the action of Ad(1⊗ρt). Hence, it follows that the restriction ofΦtoS(M)gives aM-bimodularΓ-equivariant u.c.p. map intoS(MoΓ).
Ifϕ(A)6=0, then consideringϕ◦Φwe obtain anM-centralΓ-invariant positive linear functional onS(M) that restricts toϕ(A)τonM. Moreover, sinceϕ|(
K∞,1(MoΓ)]J)∗=0 we then have thatϕ◦Φ|K∞,1(M)=0, which shows thatΓyMis not properly proximal.
Otherwise, ifϕ(A) =0, then we may consider theΓ-equivariant u.c.p. mapΨ:`∞Γ→
B(L2M⊗`2Γ)]J∗
given byΨ(f) =1⊗(1−An)1/2Mf(1−An)1/2. As above, we see that the restriction ofΨtoS(Γ)maps into S(MoΓ), and then the stateϕ◦Ψdefines aΓ-invariant state onS(Γ), which shows thatΓis not properly proximal.
Proof. We letVn denote the isometry fromL2M to L2M˜ given byVn(x) =ˆ α[n(x)for x∈M. We also let φn:B(L2M)˜ →B(L2M)be theΓ-equivariant u.c.p. map given byφn(T) =Vn∗TVn, forT ∈B(L2M).˜
If x∈M, then we have kxVn−Vnxk∞,2=supa∈(M)
1kxαn(a)−αn(xa)k2≤ kx−αn(x)k2. It therefore follows that for allx∈MandT∈B(L2M)˜ we have
kφn(T x)−φn(T)xk∞,1,kφn(xT)−xφn(T)k∞,1≤ kTkkx−αn(x)k2. (2.17)
We similarly have
kφn(T JxJ)−φn(T)JxJk∞,1,kφn(JxJT)−JxJφn(T)k∞,1≤ kTkkx−αn(x)k2. (2.18)
We view each φn as a u.c.p. map into the von Neumann algebra (B(L2M)]J)∗ and let φ :B(L2M)˜ → (B(L2M)]J)∗denote a point ultraweak limit point of{φn}n. From (2.17) and (2.18) we have thatφis bimodular with respect to bothMandJMJ. Moreover, sinceeM◦αn∈KLX(M,L2M)it follows thatφn(eM)∈K∞,1X (M) for eachn≥1 and henceφ(eM)∈(KX(M)]J)∗.
IfΓyM were not properly proximal relative toX, then by Lemma 2.8.6, there exists aΓ-invariantM- central stateϕoneSX(M)withϕ|M=τand such thatϕ vanishes on(KX(M)]J)∗. Sinceφ isJMJ-bimodular it follows that φ :JMJ0∩B(L2M)˜ →eSX(M). Since φ is M-bimodular and Γ-equivariant it follows that ϕ◦φisM-central andΓ-invariant, and we also see thatϕ◦φ|M=τ. And sinceϕvanishes on(KX(M)]J)∗and φ(eM)∈(KX(M)]J)∗it then follows that the stateϕ◦φ◦Ad(e⊥M)verifies the conclusion of the proposition.
As an application of the previous proposition, we give here examples of properly proximal actions stem- ming from Gaussian processes associated to an orthogonal representations. See, e.g., [KL16] for some details on Gaussian actions. LetH be a real Hilbert space, the Gaussian process gives a tracial abelian von Neumann algebraAH, together with an isometryS:H →L2R(AH)so that orthogonal vectors are sent to independent Gaussian random variables, and so that the spectral projections of vectors in the range ofSgenerateAH as a von Neumann algebra.
In this case, the complexification of the isometry S extends to a unitary operator from the symmetric Fock spaceS(H) =CΩ⊕L∞n=1(H ⊗C)ninto L2(AH). IfH =H1⊕H2, then conjugation by the unitary implementing the canonical isomorphismS(H1⊕H2)∼=S(H1)⊗S(H2)implements a canonical isomorphismAH1⊕H2∼=AH1⊗AH2.
IfV :K →H is an isometry, then we obtain an isometryVS:S(K)→S(H)on the level of the symmetric Fock spaces, and conjugation by this isometry gives an embedding of von Neumann algebras Ad(VS):AK →AH. IfV were a co-isometry the conjugation byVSimplements instead a conditional
expectation from AH toAK. In particular, ifU ∈O(H) is an orthogonal operator, then we obtain a trace-preserving∗-isomorphismαU =Ad(US)∈Aut(AH). Ifπ:Γ→O(H)is an orthogonal represen- tation, then the Gaussian action associated to π is given by Γ3t 7→απ(t)∈Aut(AH). The correspond- ing Koopman representation is then canonically isomorphic to the symmetric Fock space representation πS:Γ→U(S(H)).
Every contractionA:K →H can be written as the composition of an isometryV:K →L followed by a co-isometryW :L →H so that conjugation by AS:S(K)→S(H)implements a u.c.p. map φA:B(L2AH)→B(L2AK), which mapsAH intoAK. Note that the associationA7→φAis continuous as a map from the space of contractions endowed with the strong operator topology into the space of u.c.p. maps endowed with the topology of point-ultraweak convergence.
IfAis self-adjoint this can be realized explicitly by considering the orthogonal matrix onH ⊕H given by ˜A=
A −√
1−A2
√
1−A2 A
, so thatA=V∗AV, where˜ Vξ =ξ⊕0, forξ∈H.
We recall that an orthogonal representationπ:Γ→O(H)is amenable in the sense of Bekka [Bek90]
if there exists a stateϕ onB(H ⊗C)that is invariant under the action given by conjugation byπ(t). If π is a nonamenable representation, then neither is π⊗ρ for any representation ρ [Bek90, Corollary 5.4].
In particular, ifπ is a nonamenable representation, then neither is the restriction ofπStoS(H) CΩ∼= H ⊗S(H)as it is isomorphic toπ⊗πS.
Theorem 2.9.2. Letπ:Γ→O(H)be a nonamenable representation on a separable real Hilbert space, then the associated Gaussian actionΓyαπAH is properly proximal.
Proof. Note that sinceπis a nonamenable representation we have thatΓyAH is ergodic (see, e.g., [PS12]).
Denote byPkthe orthogonal projection fromL2AH ontoCΩ⊕Lkn=1(H ⊗C)nfor eachk≥1 and setP0 to be the projection ontoCΩ. Consider the hereditaryC∗-algebraXF⊂B(L2AH)generated by{Pk}k≥0in B(L2AH), i.e.,T ∈XFif and only if limk→∞kT−T Pkk=limk→∞kT−PkTk=0.
To see hatXFis a boundary piece it suffices to show that the multiplier algebra contains any von Neu- mann subalgebra of the formARξ whereξ ∈H \ {0}. Considering the canonical isomorphismsS(H)∼= S(Rξ)⊗S(H Rξ)andAH ∼=ARξ⊗AH Rξ we may decomposePkasPk=∑kj=0Qj⊗Rk−j≤Qk⊗1 whereQjdenotes the projection ontoCΩ⊕Ln=1j (Rξ⊗C)nandRk−jdenotes the projection onto((H Rξ)⊗C)k−j.
Since{Qj}j≥0gives an increasing family of finite rank projections such that∨jQj=idS(Rξ)we see that for eacha∈ARξ,j≥0, andε>0 there existsk>jso thatkQjaQ⊥k−jk<ε. Ifi≤jwe have(1⊗Rj−i)Pk⊥=
Q⊥k−j+i⊗Rj−iand hence
kPj(a⊗1)Pk⊥k= max
0≤i≤jkQiaQ⊥k−j+ik ≤ kQjaQ⊥k−jk<ε.
From this fact it is then easy to see that(a⊗1)∈M(XF), and so we indeed haveARξ ⊂M(XF)for each ξ ∈H \ {0}.
Fort≥0 we consider the orthogonal matrix
cos(πt/2) −sin(πt/2) sin(πt/2) cos(πt/2)
∈O(H ⊕H)and letαt ∈ Aut(AH⊕H)be the associated automorphism.
Note thatαtcommutes with the action ofΓonAH, and that limt→0kx−αt(x)k2=0 for eachx∈AH. We may compute E◦αt explicitly as E◦αt =P0+∑∞n=1cosn(πt/2)(Pn−Pn−1), so that E◦αt∈XF for each 0<t<1. Using the isomorphismL2(AH⊕H)∼=L2(AH)⊗L2(AH)∼=S(H)⊗S(H)we obtain an isomorphism ofAH-modulesL2(AH⊕H) L2(AH⊕0)∼=L2(AH⊕0)⊗(S(H) CΩ). So that we have an isomorphism
A0H ∩B(L2(AH⊕H) L2(AH⊕0))∼=AH⊗B(S(H) CΩ),
whereΓacts on the latter space byαπ(t)⊗Ad(π(t)S). If we had aΓ-invariant state onAH⊗B(S(H) CΩ), then restricting it to B(S(H) CΩ) would show that the restriction of the representation πS to S(H) CΩis amenable, which would then imply thatπis an amenable representation, giving a contradic- tion. Thus, we conclude that there is noΓ-invariant state onAH ⊗B(S(H) CΩ)and it then follows from Proposition 2.9.1 thatΓyAH is properly proximal relative toXF.
We now take aΓ-almost invariant approximate identityAn∈K(H)withkAnk<1 for eachn≥0, and letαA˜n be the automorphisms ofAH⊕H as given above. As in the proof of Proposition 2.9.1 we letVn: L2(AH)→L2(AH⊕H)be given byVn(x) =ˆ α\A˜n(x), forx∈AH. We also denote byφn:B(L2(AH⊕H))→ B(L2(AH))the u.c.p. mapφn(T) =Vn∗TVn, forT∈B(L2(AH⊕H)).
We view eachφnas a u.c.p. map into the von Neumann algebra(XF]
J)∗, which we view as a corner of (B(L2(AH))]J)∗. We then letφ:B(L2(AH⊕H))→(XF]
J)∗be a point ultraweak cluster point of these maps.
SinceAn→1 in the strong operator topology it follows thatkαA˜
n(x)−xk2→0 for eachx∈AH⊕H. Also, since eachAnis compact, and sincekAnk<1, it follows thatASn ∈K(S(H))and the same computation as above then shows thatEAH⊕0◦αA˜n ∈K(L2M). The proof of Proposition 2.9.1 then shows thatφ isAH- bimodular and satisfiesφ(eAH⊕0)∈(K(M)]J)∗.
If we taket∈Γ, then askπ(t)Anπ(t−1)−Ank →0 we have that limn→∞k(απ(t)⊕π(t)Vn−Vnαπ(t))Pkk=0
for eachk≥1, and hence for eachT∈B(L2(AH⊕H))we have
n→∞limkPk(απ(t)φn(T)απ(t−1)−φn(απ(t)⊕π(t)Tαπ(t−1)⊕π(t−1)))Pkk=0.
Since{Pk}kgives an approximate identity forXFit then follows by passing to a limit thatφisΓ-equivariant.
Suppose now that ΓyAH is not properly proximal, then by Lemma 2.8.5 there exists anAH-central andΓ-invariant stateϕon ˜S(M). We may further assume it vanishes on((K∞,1AH)])∗. AsΓyAH is properly proximal relative to XF we must then have thatϕ is supported on((XF)]J)∗, but then by considering the composition ofϕwith the restriction ofφtoA0H∩B(L2(AH⊕H) L2(AH⊕0))we obtain aΓ-invariant state on this space, which would give a contradiction, as noted above.
Proof of Theorem 2.1.2. From the proof of Theorems 2.9.1 and 2.9.2 we show, in fact, that ifπ:Γ→O(H) is an orthogonal representation, then there are AH-bimodular Γ-equivariant maps ψ :AH ⊗B(S(H) CΩ)→S˜XF(AH) and φ :AH⊗B(S(H) CΩ)→((XF)]J)∗∩S˜(AH). If R is a subequivalence re- lation of the orbit equivalence relation associated to the action ΓyAH, then just as above, if the ac- tion NLR(AH)yAH is not properly proximal, then we would obtain aNLR(AH)-invariant state ϕ on
AH ⊗B(S(H) CΩ)such thatϕ|AH is normal with supportp∈AΓH. Under the isomorphismNLR(AH)/U(AH)∼= [R], where[R]denotes the full group of the equivalence relation we then see thatϕis[R]-invariant for the
natural action of[R]onAH ⊗B(S(H) CΩ).
By a standard argument using Day’s trick (e.g., as in [Bek90]) this would then give a net of vectors ξi∈L2(AH)⊗HS(S(H) CΩ)such thathaξi,ξii=τ(p)1 τ(ap)fora∈AH and such that{ξi}iis asymp- totically[R]-invariant. Ifπ≺λ, then this, in turn, gives an asymptotically[R]-invariant net of vectors in L2(AH)⊗`2Γ, which would show thatRpis an amenable equivalence relation [ADR00, Theorem 6.1.4]. A simple maximality argument then gives the result.
More generally, if some tensor powerπ⊗ksatisfiesπ⊗k≺λ, then by considering self-tensor powers of the net{ξi}iit would again follow thatRpis an amenable equivalence relation.