TOPOLOGY AND CONVEXITY IN THE SPACE OF ACTIONS MODULO WEAK EQUIVALENCE
2.3 Topology on the space of weak equivalence classes
Let Γ be a countable group and A∼(Γ,X, µ) be its space of actions modulo weak equivalence. We consider a metric on A∼(Γ,X, µ)which is implicit in [1].
Fix an enumeration (γi)∞
i=0 of Γ. If A = {A1, . . . ,Ak} is a partition of X into k pieces, a ∈ A(Γ,X, µ) and n ∈ N, let Mn,kA(a) ∈ [0,1]n×k×k be the point whose p,q,r coordinate is µ(γpaAq ∩ Ar), where p ≤ n and q,r ≤ k. Let Cn,k(a) = {Mn,kA(a):Ais a partition ofX intok pieces.} Then we can define a pseudometricdon A(Γ,X, µ)by the formula
d(a,b)=
∞
Õ
n,k=1
1
2n+kdH(Cn,k(a),Cn,k(b)),
where dH is the Hausdorff distance in the hyperspace of compact subsets of [0,1]n×k×k. It is easy to see thata∼ bif and only ifd(a,b)= 0, soddescends to a metric on A∼(Γ,X, µ), which we also denote by d. Letτ1 be the topology induced by d. We note that this definition extends to actions on countable spaces. We will write A∗∼(Γ)for the space of all actions ofΓon probability spaces.
We now describe a different construction of the topology on A∼(Γ,X, µ) due to Tucker-Drob [74] in order to show it agrees with the one we have just introduced.
(Tucker-Drob shows in [74] that his formulation agrees with the one from [1]).
Let S be a compact Polish space, and consider SΓ, which is also a compact Pol-
ish space. Γ acts on SΓ by the shift action s given by (γsf)(δ) = f(γ−1δ). Let Ms(SΓ) be the compact Polish space of shift-invariant probability measures on SΓ and letKS =K(Ms(SΓ))be the hyperspace of compact subsets of Ms(SΓ)equipped with the Hausdorff topology. Then KS is again compact and Polish. Now con- sider anS-valued random variable φ ∈ L(X, µ,S)onX, that is to say a measurable mapφ : X → S. For each measure-preserving actiona ∈ A(Γ,X, µ)we get a map Φφ,aS : X → SΓby lettingΦφ,aS (x)(γ)= φ((γ−1)ax)and consequently a shift-invariant measure(Φφ,a
S )∗µonSΓ. Then define a subsetE(a,S)ofMs(SΓ)by E(a,S)= {(Φφ,a
S )∗µ:φ: X → Sis measurable}.
LetΦS : A(Γ,X, µ) → KSbe given byΦS(a)= E(a,S). WhenS =K is the Cantor set, we omit the subscript S on the notations just introduced. By Proposition 3.5 in [74], we have a ∼ bif and only if Φ(a) = Φ(b) so we can consider the initial topology on A∼(Γ,X, µ)induced byΦ. Call thisτ2. We now work towards showing τ1agrees withτ2. There will be a series of preliminary steps. This entire argument can be regarded as a ‘perturbed’ version of Proposition 3.5 in [74].
We first fix a compatible metric on Ms(KΓ). Let AK be the collection of clopen subsets ofKΓ of the formπF−1
Îγ∈F Aγ
whereAγ ⊆ K an element of some fixed countable clopen basis for K, F ⊆ Γ is finite and π : KΓ → KF is the projection onto the F-coordinates. Since the elements ofAK generate the Borelσ-algebra of KΓ, for(νn)∞n=
1 ⊆ Ms(KΓ)we haveνn → νin Ms(KΓ)if and only ifνn(A) → ν(A) for every A∈ AK. So, enumerating the elements ofAK as(AiK)∞i=1,δK given by
δK(ν, ρ)=
∞
Õ
i=1
1
2i|ν(AiK) − ρ(AiK)|
is a compatible metric onMs(KΓ).
Lemma 2.3.1. For any > 0 there exists k ∈ N such that every a and every φ ∈ L(X, µ,K) there is ψ ∈ L(X, µ,K) with δK((Φφ,a)∗µ,(Φψ,a)∗µ) < such that the range ofψhas size≤ k. Note that kdepends only on, not onaorφ.
Proof. Fix . Choose N large enough that Í∞ i=N 1
2i < . For each i ≤ N, write Ai = π−F1
i
Îγ∈Fi Aiγ
for Aiγ ⊆ K clopen and Fi ⊆ Γ finite. We have for all
a ∈A(Γ,X, µ)andφ, ψ ∈ L(X, µ,K),
|Φφ,a(Ai) −Φψ,a(Ai)| =
Φφ,a π−1Fi Ö
γ∈Fi
Aiγ
! !
−Φψ,a πF−1i Ö
γ∈Fi
Aiγ
! !
= |µ({x:Φφ,a(x)(γ) ∈ Aiγ for allγ ∈ Fi})
− µ({x:Φψ,a(x)(γ) ∈ Aiγ for allγ ∈ Fi})|
= |µ({x: φ((γ−1)ax) ∈ Aiγfor allγ ∈ Fi})
− µ({x:ψ((γ−1)ax) ∈ Aiγ for allγ ∈Fi})|
=
µ Ù
γ∈Fi
γaφ−1(Aiγ)
!
− Ù
γ∈Fi
γaψ−1(Aiγ)
!
. (2.1)
Now, fixφ∈ L(X, µ,K). Let(Bj)kj=
1be the finite partition ofK given by the atoms of the Boolean algebra generated by (Aiγ)i≤N,γ∈Fi. Note that k depends only on . For each j ≤ k, let yj be any point in Bj. Define a mapψ : X → K by letting ψ(x)= yj for the unique jsuch thatx ∈ φ−1(Bj). Thenψ−1(Bj)= φ−1(Bj)for each j, and hence φ−1(Aiγ)=ψ−1(Aiγ)for eachi ≤ N andγ ∈Fi. Therefore the value of the expression(1)is 0 andδK((Φφ,a)∗µ,(Φψ,a)∗µ)< . Lemma 2.3.2. If E(an,L) → E(a,L) in K(Ms(LΓ)) for every finite set L then E(an,K) → E(a,K)inK(Ms(KΓ)).
Proof. Fix > 0 in order to show that eventuallydK
E(an,K),E(a,K)
< , where dK is the Hausdorff distance in K(Ms(KΓ)) constructed from δK. For k ∈ N and b ∈A(Γ,X, µ)let
Ek(b,K)= {(Φφ,a)∗µ: φ:X →K is measurable and the range ofφhas size ≤ k}.
By Lemma 2.3.1 we can choosek ∈ Nsuch that E(b,K) ⊆ B
4(Ek(b,K))for every b ∈A(Γ,X, µ)whereBr(A)={ν ∈ Ms(KΓ):δK(ν, ρ)< rfor some ρ∈ A}. Notice thatEk(b,K)=Ð
L⊆K,
|L|=k
E(b,L). Fix a setLof sizekand chooseNlarge enough such that ifn≥ N thendKL
E(an,L),E(a,L)
< 4 wheredKL is the Hausdorff distance inK(Ms(LΓ)). Since the construction is independent of the set chosen to realize L, we have in fact dKL
E(an,L),E(a,L)
< 4 for every finite set L of size k. For a fixed finite L ⊆ K letEL(b,K) = {(Φb,φ)∗µ : φ : X → K measurable,φ(X) ⊆ L}. Then for anyb,c∈A(Γ,X, µ)we have
dK
EL(b,K),EL(c,K)
= dKL
E(b,L),E(c,L) ,
so that whenn ≥ N,
dK
Ek(an,K),Ek(a,K)
=dK
©
« Ø
L⊆K
|L|=k
E(an,L), Ø
L⊆K
|L|=k
E(a,L) ª
®
®
®
¬
≤ sup
L⊆K
|L|=k
dKL
E(an,L),E(a,L)
<
4. Therefore whenn≥ N,
dK
E(an,K),E(a,K)
≤ dK
E(an,K),Ek(an,K) +dK
Ek(an,K),Ek(a,K) +dK
Ek(a,K),E(a,K)
< 3 4 .
Lemma 2.3.3. LetLbe a finite set of sizek. Then for each finite set(Ap)q
p=1of basic clopen setsAp⊆ LΓ and > 0there isδ >0such that ifd(a,b)< δthen for allφ∈ L(X, µ,L)there existsψ ∈ L(X, µ,L)such that |(Φa,φL )∗µ(Ap) − (ΦbL,ψ)∗µ(Ap)| <
for allp ≤ q.
Proof. Write Ap = Ñ
γ∈Fpπ−γ1(p(γ))for some Fp ⊆ Γ finite, : Fp → k and fix > 0. Choose a finite F ⊆ Γ with(Fp)2 ⊆ F for all p ≤ q. We may assume the identitye∈ F. Supposed(a,b) < δ
2|F|+k|F|; we will specify a value forδ later. Now fixφ : X → k and letBi = φ−1(i). Givenη : F → k, letBη = Ñ
γ∈FγaBη(γ). We can then find a partition{Dη}η∈kF such that
|µ(γaBη1 ∩Bη2) −µ(γbDη1∩Dη2)| < δ
for all η1, η2 ∈ kF and γ ∈ F. Define ψ : X → k, by ψ(y) = l if y ∈ Dη for someη with η(e) = l. Furthermore, for eachl ≤ k let Dl = Ã{Dη : η ∈ kF and η(e) = l} = ψ−1(l). For each J ⊆ F and σ ∈ kJ let Dσ = Ã{Dη : η ∈ kF and σ v η}, whereσ v ηmeansηextendsσand let ˜Dσ =Ñ
γ∈JγbDσ(γ). Furthermore ifγ ∈ Γ, J ⊆ Γandσ ∈ kJ letγ ·σ ∈ kγJ be given by(γ ·σ)(δ)= σ(γ−1δ). For σ ∈ KFp andγ ∈Fpwe have
|µ(γbDσ∩Dγ·σ) −µ(γaBσ∩Bγ·σ)|
≤ Õ
(η∈kF:σvη)
Õ
(η0∈kF:γ·σvη0)
|µ(γbDη∩Dη0) −µ(γaBη∩Bη0)|
≤ δ(k|F|)2.
In particular, settingγ =ewe see|µ(Bσ) −µ(Dσ)| < δk2|F| for everyσ : Fp→ k. SinceγaBσ = Bγ·σ = γaBσ∩Bγ·σ we have
|µ(Dσ) −µ(γbDσ∩Dγ·σ)| ≤ |µ(Dσ) − µ(γaBσ)|
+|µ(γaBσ∩Bγ·σ) − µ(γbDσ∩Dγ·σ)|
= |µ(Dσ) − µ(Bσ)|
+|µ(γaBσ∩Bγ·σ) − µ(γbDσ∩Dγ·σ)|
<2δk2|F|
and also
|µ(Dγ·σ) − µ(γbDσ∩Dγ·σ)| ≤ |µ(Dγ·σ) − µ(Bγ·σ)|
+|µ(γaBσ ∩Bγ·σ) − µ(γbDσ∩Dγ·σ)|
< 2δk2|F|. Therefore
µ((γbDσ)4(Dγ·σ))= µ(γbDσ)+ µ(Dγ·σ) −2µ(γbDσ∩Dγ·σ)
≤ |µ(Dγ·σ) −µ(γbDσ∩Dγ·σ)|+|µ(Dγ·σ) − µ(γbDσ∩Dγ·σ)|
< 4δk2|F|. (2.2)
Since(Dη)η∈kF is a partition ofX and(Fp)2 ⊆ F, we have
Dp = Ä
η∈kF
pvη
Dη = Ù
γ∈Fp
Ä
σ∈kγFp σ(γ)=p(γ)
Dσ = Ù
γ∈Fp
Ä
σ∈kFp σ(e)=p(γ)
Dγ·σ.
Now, by(2),
µ
©
«
©
« Ù
γ∈Fp
Ä
σ∈kFp σ(e)=(γ)
Dγ·σ
ª
®
®
®
¬ 4
©
« Ù
γ∈Fp
Ä
σ∈kFp σ(e)=p(γ)
γbDσ
ª
®
®
®
®
¬ ª
®
®
®
®
¬
< (|Fp|k|Fp|)(4δk2|F|). (2.3)
Note that
Ù
γ∈Fp
Ä
σ∈kFp σ(e)=p(γ)
γbDσ = Ù
γ∈Fp
γbDp(γ) = D˜p,
so(3)reads
|µ(Dp) − µ(D˜p)| < (|Fp|k|Fp|)(4δk2|F|).
Moreover,
(Φb,ψL )∗µ(Ap)= µ({x:ΦbL,ψ(x) ∈ Ap})
= µ({x:ΦbL,ψ(x)(γ)= p(γ)for allγ ∈Fp})
= µ({x:ψ((γ−1)bx)= p(γ)for allγ ∈Fp})
= µ({x: x ∈γbψ−1(p(γ))for allγ ∈Fp})
= µ©
« Ù
γ∈Fp
γbDp(γ)ª
®
¬
= µ(D˜p). Similarly,(Φa,φ
L )∗µ(Ap)= µ(Bp). So we finally have
|(Φb,ψL )∗µ(Ap) − (Φa,φL )∗µ(Ap)| = |µ(D˜p) −µ(Bp)|
≤ |µ(D˜p) −µ(Dp)|+|µ(Dp) − µ(Bp)|
<(|Fp|k|Fp|)(4δk2|F|)+2δk2|F|. Since k is fixed in advance, |Fp| ≤ |F| and F depends only on (Ap)qp=
1, it is clear thatδcan be chosen so(|Fp|k|Fp|)(4δk2|F|)+2δk2|F| < for allp ≤ q.
We can now prove the main result of this section.
Theorem 2.3.1. τ1=τ2.
Proof. Suppose thatan→ ainτ1. We need to proveΦ(an) →Φ(a)inK(Ms(KΓ)).
By Lemma 2.3.2 it suffices to fix a finite set L and show E(an,L) → E(a,L) in K(Ms(LΓ)). Letk = |L|. WriteEn= E(an,L)andE = E(a,L). As before, if we let AL = (AiL)∞
i=1 be the collection of clopen subsets of LΓ of the formÑ
γ∈Fπγ−1(jγ) for a finiteF ⊆ Γand jγ ≤ k, then
δL(ν, ρ)=
∞
Õ
i=1
1
2i|ν(AiL) − ρ(ALi)|
is a compatible metric on Ms(LΓ). Fix > 0 in order to show that eventually dL(En,E) < , wheredL is the Hausdorff distance inK(Ms(LΓ))constructed from δL. ChooseN sufficiently large thatÍ∞
i=N 1
2i < 2. By Lemma 2.3.3 there isδ > 0 such that if d(a,b) < δ then for each i ≤ N and all φ ∈ L(X, µ,L) there exists ψ ∈ L(X, µ,L) such that |(Φa,φ
L )∗µ(AiL) − (Φb,ψ
L )∗µ(AiL)| < 2. Thus if M is large
enough thatd(an,a)< δforn ≥ M, we havedL(En,E)< .
Now suppose Φ(an) → Φ(a) inK(Ms(KΓ)). Fixr,q and > 0 in order to show that eventuallydH(Cr,q(an),Cr,q(a)) < . Chooseqdistinct points(xp)qp=
1 ∈ K and let(Dp)qp=
1 be a family of disjoint clopen subsets of K with xp ∈ Dp. Now let M be large enough that all sets of the form πγ−1s(Dp) ∩πe−1(Dt)for s ≤ r and p,t ≤ q appear as some AKi for i ≤ M in our previously chosen clopen basis AK. Then choose N large enough that when n ≥ N, dK(Φ(an),Φ(a)) < 2M. Then for each φ ∈ L(X, µ,K)we haveψ ∈ L(X, µ,K)such thatδK((Φan,φ)∗µ,(Φa,ψ)∗µ)< 2M. So in particular, ifn≥ Nthen for eachφ ∈L(X, µ,K)there existsψ ∈ L(X, µ,K)such that
|(Φan,φ)∗µ(πγ−1s(Dp) ∩π−1e (Dt)) − (Φa,ψ)∗µ(π−1γs(Dp) ∩πe−1(Dt))| <
for allp,t ≤ qands ≤r.
Now suppose n ≥ N and let (Bp)qp=1 be a partition of X. Define φ : X → K by takingφ(x)= xpfor the uniquep ≤ qwith x ∈Bpso by the previous paragraph we have a correspondingψ. Observe that for allγ ∈Γwe have
µ(γanBp∩Bt)= µ(γanφ−1(Dp) ∩φ−1(Dt))
= µ({x: φ((γan)−1x) ∈ Dpandφ(x) ∈ Dt})
= µ({x:Φφ,an(x)(γ) ∈ DpandΦφ,an(x)(e) ∈ Dt})
= µ({x:Φφ,an(x) ∈π−γ1(Dp)andΦφ,an(x) ∈πe−1(Dt)})
= µ({x:Φφ,an(x) ∈π−1γ (Dp) ∩π1−1(Dt)})
= (Φφ,an)∗µ(π−1γ (Dp) ∩π1−1(Dt)). Similarly lettingHp =ψ−1(Dp)we have
µ(γaHp∩Ht)= (Φψ,an)∗µ(π−1γ (Dp) ∩π1−1(Dt)). Thus for allp,t ≤ qands ≤ r,
|µ(γasnBp∩Bt) − µ(γsaHp∩Ht)|
= |(Φφ,an)∗µ(π−1γ+s(Dp) ∩πe−1(Dt)) − (Φψ,an)∗µ(π−1γs(Dp) ∩πe−1(Dt))|
< .
We have shown that when n ≥ N, Cr.q(an) ⊆ B(Cr,q(a)). The argument that eventuallyCr,q(a) ⊆ B(Cr,q(an))is identical.
Topology on the space of stable weak equivalence classes
Let A∼s(Γ,X, µ) be the space of stable weak equivalence classes and let ι be the trivial action of Γ on an standard probability space. By Lemma 3.7 in [74], we have a ≺s b if and only if a ≺ ι× b. Moreover, Theorem 1.1 in [74] says that E(a×ι,K) = cch(E(a,K)), where Ms(KΓ) carries its natural topological convex structure as a compact convex subset of a Banach space. LettingΨ : A(Γ,X, µ) → K(Ms(KΓ)) be the map a 7→ cch(E(a,K)) we have Ψ(a) = Ψ(b) if and only if a∼s b. Tucker-Drob gives A∼s(Γ,X, µ)the initial topology induced byΨ, in which it is a compact Polish space. Thus we havean →ain the topology of A∼s(Γ,X, µ)if and only ifan×ι→ a×ιin the topology of A∼(Γ,X, µ). Therefore we can introduce a metricdson A∼s(Γ,X, µ)by settingds(a,b)= d(a×ι,b×ι).