betaFeedback

WorksMathematics

Expected amenable subalgebras preserving core commutants

Abstract

We prove that every inclusion N⊂MN\subset M of von Neumann algebras with separable preduals and a faithful normal conditional expectation contains an expected amenable subalgebra P⊂NP\subset N such that P′∩c(M)=N′∩c(M)P'\cap c(M)=N'\cap c(M), where c(M)c(M) denotes the continuous core of M. This resolves the relative bicentralizer conjecture in this setting.

Introduction

An inclusion N⊂MN \subset M is expected if it admits a faithful normal conditional expectation EN:M→NE_N : M \to N. Such an expectation makes the inclusion compatible with modular theory: a faithful normal state on NN extends through ENE_N to a state on MM whose modular action preserves NN. For a faithful normal state ψ\psi on MM, the continuous core is c(M)=M⋊σψRc(M) = M \mathbin{\rtimes_{\sigma^\psi}} \mathbb{R}, with its canonical copy of MM.

We ask whether an amenable expected subalgebra of NN can impose exactly the same commutation constraints as NN on the whole core. Equality of commutants inside MM alone does not answer this question: the core also contains the unitaries implementing the modular action. Our main theorem retains these additional constraints.

Theorem 1.1. Let N⊂MN \subset M be a unital inclusion of von Neumann algebras with separable preduals, and suppose that there is a faithful normal conditional expectation EN:M→NE_N : M \to N. There exist a unital amenable von Neumann subalgebra P⊂NP \subset N and a faithful normal conditional expectation F:N→PF : N \to P such that

P′∩c(M)=N′∩c(M).P' \cap c(M) = N' \cap c(M).

Here amenable means injective.

The canonical core isomorphism associated with a change of faithful normal state fixes MM, so the stated relative-commutant equality is independent of the state used to define the core.

History and the relative conjecture. Popa’s solution of Kadison’s problem for irreducible expected inclusions with separable preduals and type II bottom factor provides a model for finding amenable subalgebras that retain prescribed commutation properties [15]. For type III factors, modular asymptotic centralization supplies a related approach. The bicentralizer of a faithful normal state consists of those elements that commute asymptotically with every bounded sequence asymptotically centralizing the state in predual norm. Connes introduced the problem of its triviality in the classification of injective factors [3, 4]. Haagerup solved the amenable type III1_1 case, completing the uniqueness theorem for the injective type III1_1 factor [5]. For type III1_1 factors with separable predual, he characterized triviality by norm-density of faithful normal states whose centralizers have trivial relative commutant [5], Theorem 3.1. His proof uses Popa’s construction to find a maximal abelian subalgebra of such a centralizer that remains maximal abelian in the ambient factor; the expectations onto the centralizer and onto its abelian subalgebra compose [5], proof of Theorem 3.1.

Masuda introduced a relative bicentralizer in the finite-index setting [13], Definition 4.1. Ando, Haagerup, Houdayer, and Marrakchi constructed the canonical relative bicentralizer flow for expected inclusions with type III1{\rm III}_{1} bottom factor [2], Theorem A. This extends the absolute flow discovered by Haagerup and independently by Marrakchi, as described in that paper. The structural results of Houdayer and Isono [6], Proposition 3.4 and Theorem 3.5 and Marrakchi’s absolute flow-ergodicity theorem [9], Theorem D underlie the absolute reduction used here. Marrakchi formulated the relative bicentralizer conjecture and proved that, for separable preduals, it is equivalent to the conclusion of Theorem 1.1 [10], Conjecture C and Theorem D. Our theorem therefore resolves that conjecture positively, including inclusions with arbitrary centers and types.

Marrakchi’s recent Theorem B [11] gives an expected amenable subalgebra P⊂NP \subset N preserving N′∩MN' \cap M for general expected inclusions with separable preduals. Theorem 1.1 preserves the commutant in the continuous core. Applying the earlier result with upper algebra c(M)c(M) would require an expectation from c(M)c(M) onto NN, which the original inclusion does not supply. The binormal-state and zero-shift completely positive map of [11], Lemmas 2.1 and 2.2(1) are essential inputs to the proof below.

The absolute problem and the recent announcement. Houdayer and Marrakchi have announced that every type III1{\rm III}_{1} factor admitting a faithful normal state has trivial bicentralizer [7], Theorem G; their theorem statement has no separability restriction. Section 3 gives an alternative proof of its separable-predual instance as part of our relative argument. The multiplication maps and spectral resonance of [7], Section 8 are closely related to this absolute comparison. Their paper also develops relative bicentralizer theory for tracial group actions; the general expected-inclusion core-commutant conclusion of Theorem 1.1 is obtained here through the relative tensor and core argument.

The companion [14], Theorem 4.1 and Section 5.1 develops a complementary absolute argument through bounded recovery: for a state with scalar centralizer, a fixed multiplication defect whose modularly averaged squared norm has positive limsup along unit vectors with spectral supports shrinking to one point can be detected by uniformly bounded algebra elements. The binormal-state identity used here also admits the iteration that yields the relative estimate uniformly in tensor length and flow shifts.

An abelian consequence. A maximal abelian subalgebra, or MASA, A⊂NA \subset N satisfies A′∩N=AA' \cap N = A. Another part of Marrakchi’s characterization gives such a subalgebra with a precise core commutant. For an expected A⊂NA \subset N, we view c(A)c(A) inside c(N)⊂c(M)c(N) \subset c(M) using the expectation onto AA and its composition with ENE_N.

Corollary 1.2. Under the hypotheses of Theorem 1.1, there is a maximal abelian subalgebra A⊂NA \subset N with a faithful normal conditional expectation F:N→AF:N \to A such that

A′∩c(M)=c(A)∨(N′∩c(M)).A' \cap c(M) = c(A) \vee\left(N' \cap c(M)\right).

Proof. Apply [10], Theorem D, (4)⇒\Rightarrow(6) to Theorem 1.1. ∎

For an irreducible expected inclusion of factors, Kadison’s problem asks for a MASA in NN that is also maximal abelian in MM. The corollary gives such a MASA with expectation under the additional core condition N′∩c(M)⊂c(N)N' \cap c(M) \subset c(N): its commutant in MM then lies in c(N)∩M=Nc(N) \cap M = N, and hence equals AA. This is the expected formulation identified in [10], Conjecture B and the discussion after Theorem D.

Proof structure. The analytic work concerns the case where NN is a type III1_1 factor. Choose a faithful normal state φ\varphi on NN and extend it by ENE_N. The relative bicentralizer B⊂MB \subset M consists of elements that commute asymptotically with every bounded sequence in NN asymptotically centralizing φ\varphi. It carries a state-preserving flow γ\gamma. Section 2 recalls these objects and the binormal-state identity used below. Existing bicentralizer criteria reduce our task to showing that the extension of γ\gamma to B⋊σφRB \rtimes_{\sigma_\varphi} \mathbb{R} has fixed algebra N′∩c(M)N' \cap c(M).

The connection with multiplication is the central point. Set H=L2(N,φ)H = L^2(N,\varphi), K=L2(B,φ)K = L^2(B,\varphi), and Ω=φ1/2\Omega= \varphi^{1/2}. Consider maps from H⊗KH \otimes K to L2(M,φ)L^2(M,\varphi), initially given for x∈Nx \in N and a∈Ba \in B by

R(xΩ⊗aΩ)=xaΩ,L(xΩ⊗aΩ)=axΩ.R(x\Omega\otimes a\Omega) = xa\Omega,\qquad L(x\Omega\otimes a\Omega) = ax\Omega.

We prove that these maps are isometries and that

R=Lexp⁡(iX⊗Y).R = L\exp(iX \otimes Y).

where XX generates the modular action on HH and YY generates the bicentralizer flow on KK. After a change of coordinates in the core representation, NN acts on HH alone, whereas an element fixed by the bicentralizer flow acts independently of the HH coordinate. They therefore commute. This is the required fixed-point criterion.

The absolute theorem supplies the isometry assertion. If that theorem failed, the standard reduction would give a nontrivial factor equal to its own bicentralizer and with scalar modular fixed algebra. Its two multiplication maps become isometries after averaging the modular action. A binormal eigenstate identity gives their spectral comparison; reflection then forces a spectral lattice condition incompatible with modular ergodicity. Section 3 carries out this argument, and Section 4 applies the absolute theorem to obtain the unaveraged relative isometries.

To prove the relative comparison, Section 5 packages words of arbitrary tensor length into states on one fixed operator algebra. The binormal-state identity yields an estimate uniform in tensor length, all intervening flow parameters, and the location of the total modular energy. The packaging maps are completely positive but may be nonnormal; a spectral cutoff argument justifies the functional calculus they require. Section 6 removes the energy restriction by repeatedly testing a hypothetical discrepancy. Averaging the comparison contraction L∗RL^*R with its proposed unitary value loses norm wherever they differ. Random shifts capture such a difference with probabilities whose sum diverges, while the same probabilities bound the propagated error. The resulting expected norm loss exceeds the available norm, so the comparison is exact.

Finally, Section 7 proves the core fixed-point identity and derives the weak Dixmier property: every weak* closed convex hull of an NN-unitary orbit in c(M)c(M) meets N′∩c(M)N' \cap c(M). Marrakchi’s characterization then supplies the amenable subalgebra. Existing results handle the other factor types; a central decomposition and averaging over the center give the general inclusion.

Acknowledgment of related work. Houdayer and Marrakchi credit GPT-5.6 Sol with counterexample exploration and an initial strict-outerness proof, and GPT-6 Astra with a singular tracial-state ingredient in their resonance strategy [7], p. 7. They report checking and rewriting those proofs and retaining responsibility for their work.

Modular notation and bicentralizer inputs

The absolute proof and the relative tensor estimate use states on the bounded operators of the standard Hilbert space whose left and right marginals are prescribed and whose modular energy is concentrated at one point. We first state the resulting bicentralizer identity and the completely positive maps that shift its energy parameter.

We use standard modular theory [17] and the modular-invariance criterion for state-preserving expectations [16]. We may assume that the algebras are nonzero. Choose a faithful normal state on NN and extend it by ENE_N; denote both states by φ\varphi. Put

H=L2(M,φ),Ω=φ1/2,XM=log⁡Δφ.\mathcal{H}=L^2(M,\varphi),\qquad\Omega=\varphi^{1/2},\qquad X_M=\log\Delta_\varphi.

Scalar products are linear in the first variable. Our predual module conventions are

(xφ)(y)=φ(yx),(φx)(y)=φ(xy).(x\varphi)(y)=\varphi(yx),\qquad(\varphi x)(y)=\varphi(xy).

The left and right standard actions are denoted by λ\lambda and ρ\rho, with ρ(b)ξ=ξb\rho(b)\xi=\xi b; we usually omit λ\lambda. If A⊂MA\subset M is invariant under σφ\sigma^\varphi, its state-preserving expectation identifies L2(A,φ)L^2(A,\varphi) with a subspace of H\mathcal{H}. Write eAe_A for that projection.

Fix a free ultrafilter ω\omega on N\mathbb{N}; all state ultrapowers below are Ocneanu ultrapowers [1]. For now NN is a type III1\mathrm{III}_1 factor. Its relative analytic bicentralizer B=B(N⊂M,φ)B=B(N\subset M,\varphi) consists of those a∈Ma\in M for which [a,xn]→0[a,x_n]\to0 strongly whenever (xn)(x_n) is a bounded sequence in NN with ∥xnφ−φxn∥N∗→0\lVert x_n\varphi-\varphi x_n\rVert_{N_*}\to0. Equivalently, BB is the part of MM commuting in the state ultrapower with the centralizer of NωN^\omega. It is modularly invariant and has a φ\varphi-preserving faithful normal expectation EBE_B.

The flow convention. Write γ:R↷B\gamma:\mathbb{R}\curvearrowright B for the bicentralizer flow in the following logarithmic convention: an element v∈Nωv\in N^\omega of modular frequency ss, meaning σt(v)=eitsv\sigma_t(v)=e^{its}v, satisfies

va=γs(a)v(a∈B).(1)va=\gamma_s(a)v\qquad(a\in B). \tag*{(1)}

With the multiplicative parameter of [2], this means γs=βe−s\gamma_s=\beta_{e^{-s}}: modular frequency ss corresponds to vφω=e−sφωvv\varphi^\omega=e^{-s}\varphi^\omega v. The flow preserves φ∣B\varphi|_B and commutes with its modular action [2, 10].

Proposition 2.1 (Binormal states and spectral shifts). In this setting the following hold.

(i) If a state Φ\Phi on B(H)B(\mathcal{H}) satisfies

Φ∘λ=Φ∘ρ=φ,Φ(eN)=1,Φ(f(XM))=f(0)(f∈C0(R)),\Phi\circ\lambda=\Phi\circ\rho=\varphi,\qquad\Phi(e_N)=1,\qquad\Phi(f(X_M))=f(0)\quad(f\in C_0(\mathbb{R})),

then, for a,b∈Ba,b\in B,

Φ(λ(a)ρ(b))=⟨aΩb,Ω⟩.(2)\Phi(\lambda(a)\rho(b))=\langle a\Omega b,\Omega\rangle. \tag*{(2)}

(ii) For each s∈Rs\in\mathbb{R} there is a unital completely positive map θs:B(H)→B(H)\theta_s:B(\mathcal{H})\to B(\mathcal{H}) such that

θs(eN)=eN,θs∘ρ=ρ,(3)\theta_s(e_N)=e_N,\qquad\theta_s\circ\rho=\rho, \tag*{(3)}
θs∘λ=λ∘γs∘EB,θs(f(XM))=f(XM−s)(f∈C0(R)).(4)\theta_s\circ\lambda=\lambda\circ\gamma_s\circ E_B,\qquad\theta_s(f(X_M))=f(X_M-s)\quad(f\in C_0(\mathbb{R})). \tag*{(4)}

These maps need not be normal.

Part (i) is [11], Lemma 2.1, based on the binormal-state implementation [10], Theorem 2.9. For part (ii), we combine the zero-shift map of [11], Lemma 2.2(1) with the eigenoperator characterization of the flow in [2], Theorem A(ii). The construction below determines the spectral shift in convention (1).

The shift construction. Let ζ\zeta be the map from [11], Lemma 2.2(1), which has the properties in part (ii) for s=0s=0. Choose a finite row of partial isometries vi∈Nωv_i \in N^\omega of modular frequency ss with ∑ivivi∗=1\sum_i v_i v_i^*=1. Such rows exist by [2], Lemma 3.4 and the proof of Theorem A(ii). Starting with bounded representing sequences wi,nw_{i,n}, take the modularly smoothed lifts

vi,n=∫Rπ−1/2e−u2e−isuσu(wi,n) du.v_{i,n}=\int_{\mathbb{R}}\pi^{-1/2}e^{-u^2}e^{-isu}\sigma_u(w_{i,n})\,du.

Modular convolution commutes with passage to the ultrapower [1], Lemma 4.14; since viv_i has frequency ss, these are still lifts of viv_i. Translation of the convolution kernel shows that their modular orbits are uniformly norm-continuous in nn. Define

θs(T)=weaklim⁡n→ω∑iλ(vi,n)ζ(T)λ(vi,n∗).\theta_s(T)=\underset{n\to\omega}{\operatorname{weaklim}}\sum_i\lambda(v_{i,n})\zeta(T)\lambda(v_{i,n}^*).

The row normalization makes the limit unital. The right algebra and eNe_N commute with all row entries, so they are fixed. Relation (1) gives the left restriction γs∘EB\gamma_s\circ E_B.

For the spectral formula, the modular action passes to the ultrapower [1], Theorem 4.1. Hence, for each tt, ∑ivi,nσt(vi,n∗)→e−its1\sum_i v_{i,n}\sigma_t(v_{i,n}^*)\to e^{-its}1 strongly along ω\omega; the equicontinuity just obtained makes convergence uniform on compact tt-intervals. Now ζ(f(XM))=f(XM)\zeta(f(X_M))=f(X_M) for f∈C0(R)f\in C_0(\mathbb{R}). For a function of the form f(x)=∫eitxh(t) dtf(x)=\int e^{itx}h(t)\,dt, with h∈L1(R)h\in L^1(\mathbb{R}), we may therefore integrate the finite-row conjugations to obtain

θs(f(XM))=∫e−itsΔith(t) dt=f(XM−s).\theta_s(f(X_M))=\int e^{-its}\Delta^{it}h(t)\,dt=f(X_M-s).

Compact-uniform convergence of the row coefficients and the L1L^1 tail bound justify passing to the limit. Density gives the formula for all f∈C0(R)f\in C_0(\mathbb{R}). This calculation fixes the sign and uses no normality of ζ\zeta or θs\theta_s.

Binormal states at a nonzero frequency. If the last hypothesis of (i) is instead Φ(f(XM))=f(s)\Phi(f(X_M))=f(s), apply (i) to Φ∘θs\Phi\circ\theta_s. The fixed right algebra belongs to the multiplicative domain of θs\theta_s, so

Φ(λ(γs(a))ρ(b))=⟨aΩb,Ω⟩(a,b∈B).(5)\Phi(\lambda(\gamma_s(a))\rho(b))=\langle a\Omega b,\Omega\rangle\qquad(a,b\in B). \tag*{(5)}

The two marginals remain φ\varphi because γs\gamma_s and EBE_B preserve it.

The absolute bicentralizer

We first use the binormal identity to settle the absolute problem. This will make the ordinary multiplication maps for a relative bicentralizer isometric in Section 4. In the absolute argument itself, averaging the modular action provides the two isometries to be compared.

Theorem 3.1. For every type III1\mathrm{III}_1 factor NN with separable predual and every faithful normal state φ\varphi on NN, one has B(N,φ)=C1B(N,\varphi)=\mathbb{C}1.

Two established facts reduce the proof to a spectral contradiction. If the absolute conjecture fails, the self-bicentralizing reduction of [6], Proposition 3.4 and Theorem 3.5, recalled in [2], gives a nontrivial type III1\mathrm{III}_1 factor with separable predual and a faithful normal state satisfying

M=B(M,φ),Mφ=C1.(6)M=B(M,\varphi),\qquad M_\varphi=\mathbb{C}1. \tag*{(6)}

The absolute bicentralizer flow is ergodic [9], Theorem D. Neither fact assumes fullness. The averaged products, reflection, and spectral resonance below follow the architecture of [7], Propositions 8.8–8.10. Here the averaging uses the modular action, and the binormal identity supplies the comparison between the products.

Proof. Suppose that the theorem fails, and choose (M,φ)(M,\varphi) as in (6). Put H=L2(M,φ)H=L^2(M,\varphi) and X=log⁡ΔφX=\log\Delta_\varphi. Let eisYe^{isY} implement γs\gamma_s on HH, so that eisY(aΩ)=γs(a)Ωe^{isY}(a\Omega)=\gamma_s(a)\Omega. Since the flow preserves the state and commutes with its modular action, XX and YY strongly commute.

Two products after averaging. The scalar fixed algebra MφM_\varphi implies that the weak* averages of σt(x)\sigma_t(x) over [−T,T][-T,T] converge to φ(x)1\varphi(x)1. Indeed, any cluster point is modularly invariant and has state value φ(x)\varphi(x). We use this factorization to turn the two orders of multiplication into isometries. Let mm be the symmetric translation-invariant mean obtained from these interval averages as T→∞T\to\infty along an ultrafilter. Quotient bounded measurable HH-valued functions by the null space of m(∥f(t)∥2)m(\lVert f(t)\rVert^2), and complete to a Hilbert space Hm\mathcal{H}_m. Define on elementary tensors

R(xΩ⊗aΩ)(t)=σt(x)aΩ,R(x\Omega\otimes a\Omega)(t)=\sigma_t(x)a\Omega,
L(xΩ⊗aΩ)(t)=aσt(x)Ω.L(x\Omega\otimes a\Omega)(t)=a\sigma_t(x)\Omega.

The inner products of the images of xΩ⊗aΩx\Omega\otimes a\Omega and yΩ⊗bΩy\Omega\otimes b\Omega are, respectively,

m(φ(b∗σt(y∗x)a)),m(\varphi(b^*\sigma_t(y^*x)a)),
m(φ(y∗σ−t(b∗a)x)).m(\varphi(y^*\sigma_{-t}(b^*a)x)).

Both equal φ(y∗x)φ(b∗a)\varphi(y^*x)\varphi(b^*a). By linearity and completion, R,L:H⊗H→HmR,L:H\otimes H\to\mathcal{H}_m are isometries. Translation (τuf)(t)=f(t+u)(\tau_u f)(t)=f(t+u) is unitary on Hm\mathcal{H}_m and satisfies

τuR=R(eiuX⊗1),τuL=L(eiuX⊗1).\tau_uR=R(e^{iuX}\otimes1),\qquad\tau_uL=L(e^{iuX}\otimes1).

From eigenstates to the product identity. We claim that

R=Lexp⁡(iX⊗Y).(7)R=L\exp(iX\otimes Y). \tag*{(7)}

The binormal identity first compares the two sides on vectors whose XX-spectrum lies in a short interval. Translation covariance then lets us combine these comparisons without accumulating an error over the intervals.

For a unit vector h∈Hh\in H, define the state

Φh(T)=m(⟨TeitXh,eitXh⟩).\Phi_h(T)=m(\langle Te^{itX}h,e^{itX}h\rangle).

The modular averages just computed give Φh∘λ=Φh∘ρ=φ\Phi_h\circ\lambda=\Phi_h\circ\rho=\varphi; here eM=1e_M=1. Moreover, Φh(f(X))=⟨f(X)h,h⟩\Phi_h(f(X))=\langle f(X)h,h\rangle. For modular-analytic a∈Ma\in M, put a+=σi/2(a)a_+=\sigma_{i/2}(a), so that Ωa+=aΩ\Omega a_+=a\Omega. The definitions of RR and LL give

R(h⊗aΩ)(t)=ρ(a+)eitXh,L(h⊗aΩ)(t)=aeitXh.R(h\otimes a\Omega)(t)=\rho(a_+)e^{itX}h,\qquad L(h\otimes a\Omega)(t)=ae^{itX}h.

These identities extend from h=xΩh=x\Omega to every h∈Hh\in H by continuity. Suppose the spectral support of a sequence of unit vectors hh is contained in intervals shrinking to ss. Every weak* accumulation state of Φh\Phi_h then has spectral evaluation at ss. By (5),

⟨R(h⊗aΩ),L(h⊗γs(a)Ω)⟩=Φh(λ(γs(a∗))ρ(a+))⟶⟨a∗Ωa+,Ω⟩=∥aΩ∥2.\langle R(h\otimes a\Omega),L(h\otimes\gamma_s(a)\Omega)\rangle=\Phi_h(\lambda(\gamma_s(a^*))\rho(a_+)) \longrightarrow\langle a^*\Omega a_+,\Omega\rangle=\lVert a\Omega\rVert^2.

Both vectors on the left have squared norm ∥aΩ∥2\lVert a\Omega\rVert^2, so their difference tends to zero.

This convergence is uniform when the centers ss range over a fixed compact set and the interval lengths tend to zero. Otherwise choose violating vectors and a convergent subsequence of centers, and take a weak* accumulation state. The preceding argument gives a contradiction. In this step the varying test γs(a∗)\gamma_s(a^*) may be replaced by its limit: the common left marginal φ\varphi and Cauchy–Schwarz bound the change by

∥a+∥∥γs(a)Ω−γs0(a)Ω∥.\lVert a_+\rVert\lVert\gamma_s(a)\Omega-\gamma_{s_0}(a)\Omega\rVert.

which tends to zero. Thus no operator-norm continuity of γ\gamma is needed.

Analytic vectors aΩa\Omega are dense in HH, and all maps being compared are isometric. Hence the same local conclusion holds for a fixed η∈H\eta\in H, with γs(a)Ω\gamma_s(a)\Omega replaced by eisYηe^{isY}\eta. Spectral calculus gives

∥exp⁡(iX⊗Y)(h⊗η)−h⊗eisYη∥≤sup⁡∣r−s∣≤δ∥eirYη−eisYη∥.\lVert\exp(iX\otimes Y)(h\otimes\eta)-h\otimes e^{isY}\eta\rVert\le\sup_{\lvert r-s\rvert\le\delta}\lVert e^{irY}\eta-e^{isY}\eta\rVert.

whenever supp⁡X(h)⊂[s−δ,s+δ]\operatorname{supp}_X(h)\subset[s-\delta,s+\delta] and ∥h∥=1\lVert h\rVert=1. Consequently, for T=R−Lexp⁡(iX⊗Y)T=R-L\exp(iX\otimes Y), a fixed compact interval II, and fixed η\eta, there is a function αI,η(δ)→0\alpha_{I,\eta}(\delta)\to0 such that

∥T(h⊗η)∥≤αI,η(δ)∥h∥\lVert T(h\otimes\eta)\rVert\le\alpha_{I,\eta}(\delta)\lVert h\rVert

when hh has spectral support in an interval of length at most δ\delta contained in II.

Translation covariance shows that T∗TT^*T commutes with the spectral projections of X⊗1X\otimes1. For supp⁡X(h)⊂I\operatorname{supp}_X(h)\subset I, partition II into finitely many disjoint Borel intervals IjI_j of length at most δ\delta and write hj=1Ij(X)hh_j=1_{I_j}(X)h. The vectors T(hj⊗η)T(h_j\otimes\eta) are orthogonal, so

∥T(h⊗η)∥2=∑j∥T(hj⊗η)∥2≤αI,η(δ)2∑j∥hj∥2=αI,η(δ)2∥h∥2.\lVert T(h\otimes\eta)\rVert^2=\sum_j\lVert T(h_j\otimes\eta)\rVert^2\le\alpha_{I,\eta}(\delta)^2\sum_j\lVert h_j\rVert^2=\alpha_{I,\eta}(\delta)^2\lVert h\rVert^2.

Letting δ↓0\delta\downarrow0 and then using density proves (7).

Reflection forces a lattice. Define (J0f)(t)=Δitf(−t)(J_0f)(t)=\Delta^{it}f(-t). Symmetry of mm makes J0J_0 a unitary involution on Hm\mathcal{H}_m. On elementary tensors,

J0R(xΩ⊗aΩ)(t)=Δitσ−t(x)aΩ=xσt(a)Ω=L(aΩ⊗xΩ)(t).J_0R(x\Omega\otimes a\Omega)(t)=\Delta^{it}\sigma_{-t}(x)a\Omega=x\sigma_t(a)\Omega=L(a\Omega\otimes x\Omega)(t).

Similarly J0L=RΣJ_0L=R\Sigma, where Σ\Sigma flips the input factors. Applying J0J_0 to (7) and composing with Σ\Sigma gives L=Rexp⁡(iY⊗X)L=R\exp(iY\otimes X). Substitute this back into (7) and cancel the isometry RR to obtain

exp⁡(iY⊗X)exp⁡(iX⊗Y)=1.(8)\exp(iY\otimes X)\exp(iX\otimes Y)=1. \tag*{(8)}

Let S⊂R2S\subset\mathbb{R}^2 be the support of the joint spectral measure of (X,Y)(X,Y). The tensor-product spectral measure has support S×SS\times S, so the continuous phase in (8) satisfies

rs′+sr′∈2πZfor all (r,s),(r′,s′)∈S.(9)rs'+sr'\in2\pi\mathbb{Z}\quad\text{for all }(r,s),(r',s')\in S. \tag*{(9)}

If SS contains two linearly independent points, pairing with them confines SS to the inverse image of (2πZ)2(2\pi\mathbb{Z})^2 under an invertible linear map. Thus SS is countable. If SS spans a non-axis line, choose a nonzero (r0,s0)∈S(r_0,s_0)\in S; then r0s0≠0r_0s_0\ne0, and any point t(r0,s0)∈St(r_0,s_0)\in S satisfies 2tr0s0∈2πZ2tr_0s_0\in2\pi\mathbb{Z}. Again SS is countable. In both cases the joint spectral measure is supported on countably many points, so XX has pure point spectrum. The remaining possibilities put SS on a coordinate axis and give X=0X=0 or Y=0Y=0.

Excluding the spectral alternatives. A modular action with scalar fixed algebra has no Hilbert-space eigenvectors other than multiples of Ω\Omega; we recall the argument of [12], Lemma 2.1. Let PrP_r be the spectral projection of XX at rr. For x∈Mx \in M, the bounded character averages

xT=12T∫−TTe−irtσt(x) dtx_T=\frac{1}{2T}\int_{-T}^{T}e^{-irt}\sigma_t(x)\,dt

satisfy ∥xT∥≤∥x∥\lVert x_T\rVert\le\lVert x\rVert and xTΩ→Pr(xΩ)x_T\Omega\to P_r(x\Omega). A weak* cluster point y∈My \in M represents this vector, since φ(z∗xT)\varphi(z^*x_T) converges to ⟨Pr(xΩ),zΩ⟩\langle P_r(x\Omega),z\Omega\rangle for every z∈Mz \in M. Thus yΩ=Pr(xΩ)y\Omega=P_r(x\Omega), and the separating property of Ω\Omega gives σt(y)=eirty\sigma_t(y)=e^{irt}y. Density of MΩM\Omega shows that any nonzero eigenspace is detected by a nonzero such eigenoperator.

For this yy, both y∗yy^*y and yy∗yy^* lie in Mφ=C1M_\varphi=\mathbb{C}1. Hence yy is a nonzero scalar multiple of a unitary uu. Since σ−i(u)=eru\sigma_{-i}(u)=e^r u, the KMS identity gives

1=φ(uu∗)=φ(u∗σ−i(u))=erφ(u∗u)=er.1=\varphi(uu^*)=\varphi(u^*\sigma_{-i}(u))=e^r\varphi(u^*u)=e^r.

Therefore r=0r=0, and then yy itself is scalar. It follows that the only eigenvectors of XX are multiples of Ω\Omega. They cannot span HH because MM is nontrivial. This excludes pure point spectrum, including X=0X=0. Finally, Y=0Y=0 would make the bicentralizer flow trivial on MM, contradicting its ergodicity. All alternatives are excluded, proving the theorem. □\square

The two relative products

Return to an expected inclusion with NN a type III1_1 factor; MM remains arbitrary. Keep B=B(N⊂M,φ)B=\mathrm{B}(N \subset M,\varphi) and put

H=L2(N,φ),K=L2(B,φ),X=log⁡Δφ∣N.H=L^2(N,\varphi),\qquad K=L^2(B,\varphi),\qquad X=\log\Delta_\varphi|_N.

Let Ws=eisYW_s=e^{isY} implement γs\gamma_s on KK. Since the flow preserves φ∣B\varphi|_B, these unitaries commute with the modular implementation on KK. The absolute theorem makes multiplication in either order isometric. This will let us compare the two products on the fixed Hilbert space H⊗KH\otimes K.

Lemma 4.1. The expectations satisfy

EN∣B=φ(⋅)1,EB∣N=φ(⋅)1.E_N|_B=\varphi(\mathord{\cdot})1,\qquad E_B|_N=\varphi(\mathord{\cdot})1.

Consequently the maps

R(xΩ⊗aΩ)=xaΩ,L(xΩ⊗aΩ)=axΩ(x∈N, a∈B)R(x\Omega\otimes a\Omega)=xa\Omega,\qquad L(x\Omega\otimes a\Omega)=ax\Omega\qquad(x\in N,\ a\in B)

extend to isometries from H⊗KH\otimes K to H\mathcal H.

Proof. Applying ENE_N to the asymptotic commutation condition gives EN(B)⊂B(N,φ)=C1E_N(B)\subset B(N,\varphi)=\mathbb{C}1, by Theorem 3.1. State preservation determines the scalar. For x∈Nx\in N and b∈Bb\in B,

φ(b∗x)=φ(EN(b)∗x)=φ(b∗)φ(x),\varphi(b^*x)=\varphi(E_N(b)^*x)=\varphi(b^*)\varphi(x),

which characterizes EB(x)=φ(x)1E_B(x)=\varphi(x)1. Using EBE_B for the first product and ENE_N for the second gives

⟨xaΩ,ybΩ⟩=φ(b∗y∗xa)=φ(y∗x)φ(b∗a),\langle xa\Omega,yb\Omega\rangle=\varphi(b^*y^*xa)=\varphi(y^*x)\varphi(b^*a),
⟨axΩ,byΩ⟩=φ(y∗b∗ax)=φ(y∗x)φ(b∗a).\langle ax\Omega,by\Omega\rangle=\varphi(y^*b^*ax)=\varphi(y^*x)\varphi(b^*a).

Polarization and completion prove the assertion. □\square

Define the contraction UU and the unitary DD on H⊗KH \otimes K by

U=L∗R,D=exp⁡(iX⊗Y).(10)U = L^*R,\qquad D = \exp(iX \otimes Y). \tag*{(10)}

Thus

⟨U(xΩ⊗aΩ),yΩ⊗bΩ⟩=φ(y∗b∗xa).(11)\langle U(x\Omega\otimes a\Omega),y\Omega\otimes b\Omega\rangle= \varphi(y^*b^*xa). \tag*{(11)}

Our goal is U=DU=D. The next section compares arbitrarily long words in these operators on vectors with narrowly supported total modular energy. Section 6 will turn that uniform estimate into the exact identity.

An estimate uniform in total modular energy

For n≥1n \ge1, work on H⊗n⊗KH^{\otimes n} \otimes K. Let XjX_j act on the jjth copy of HH, and let Uj,DjU_j,D_j act on that copy together with the common last factor KK. Write Sn=∑j=1nXjS_n=\sum_{j=1}^{n}X_j for the total modular generator on H⊗nH^{\otimes n}. For arbitrary real numbers d0,…,dnd_0,\ldots,d_n, set

C=WdnUnWdn−1⋯Wd1U1Wd0,(12)C=W_{d_n}U_nW_{d_{n-1}}\cdots W_{d_1}U_1W_{d_0}, \tag*{(12)}
Dall=Wdexp⁡(iSn⊗Y),d=∑j=0ndj.(13)D_{\mathrm{all}}=W_d\exp(iS_n\otimes Y),\qquad d=\sum_{j=0}^{n}d_j. \tag*{(13)}

Here every WW acts on KK alone. Replacing each UjU_j in CC by DjD_j gives DallD_{\mathrm{all}}, because those unitaries commute with all the shifts.

Proposition 5.1 (Uniform energy estimate). For every η∈K\eta\in K there is a function ϵη:(0,∞)→[0,∞)\epsilon_\eta:(0,\infty)\to[0,\infty) with ϵη(δ)→0\epsilon_\eta(\delta)\to0 as δ↓0\delta\downarrow0 such that

∥(C−Dall)(ξ⊗η)∥≤ϵη(δ)(14)\|(C-D_{\mathrm{all}})(\xi\otimes\eta)\|\le\epsilon_\eta(\delta) \tag*{(14)}

for every n≥1n\ge1, every d0,…,dn∈Rd_0,\ldots,d_n\in\mathbb{R}, and every unit vector ξ∈H⊗n\xi\in H^{\otimes n} whose SnS_n-spectral support lies in an interval of length δ\delta. The same function works for all these choices and for all positions of the interval.

The estimate must remain uniform even when both the tensor length and the shifts vary. To achieve this, we express a coefficient of CC through a state on the single algebra B(H)\mathbf{B}(\mathcal H). The state will have fixed left and right marginals, while its modular distribution records the sum Sn+dS_n+d. Proposition 2.1(i) can then be applied in one fixed state space.

For xj,yj∈Nx_j,y_j\in N, define a kernel by

⟨Z(T)⨂j=1nxjΩ,⨂j=1nyjΩ⟩=⟨T0Ω,Ω⟩,(15)\left\langle Z(T)\bigotimes_{j=1}^{n}x_j\Omega,\bigotimes_{j=1}^{n}y_j\Omega\right\rangle=\langle T_0\Omega,\Omega\rangle, \tag*{(15)}

where the operators Tj∈B(H)T_j\in\mathbf{B}(\mathcal H) are obtained backwards:

Tn=θ−dn(T),Tj−1=θ−dj−1(yj∗Tjxj)(j=n,…,1).T_n=\theta_{-d_n}(T),\qquad T_{j-1}=\theta_{-d_{j-1}}(y_j^*T_jx_j)\quad(j=n,\ldots,1).

The operators xj,yjx_j,y_j here act on H\mathcal H by left multiplication.

Lemma 5.2 (The iterated kernel). Formula (15) defines a unital completely positive map Z:B(H)→B(H⊗n)Z : \mathbf{B}(\mathcal{H}) \to\mathbf{B}(H^{\otimes n}), possibly nonnormal. It satisfies

Z(λ(a))=Z(ρ(a))=φ(a)1(a∈M),Z(eN)=1,(16)Z(\lambda(a)) = Z(\rho(a)) = \varphi(a)1 \quad(a \in M), \qquad Z(e_N) = 1, \tag*{(16)}
Z(f(XM))=f(Sn+d)(f∈C0(R)).(17)Z(f(X_M)) = f(S_n+d) \quad(f \in C_0(\mathbb{R})). \tag*{(17)}

For modular-analytic a∈Ba \in B, put a+=σi/2(a)a_+ = \sigma_{i/2}(a), so that Ωa+=aΩ\Omega a_+ = a\Omega. Then, for b∈Bb \in B and ξ,ξ′∈H⊗n\xi,\xi' \in H^{\otimes n},

⟨Z(b∗ρ(a+))ξ,ξ′⟩=⟨C(ξ⊗aΩ),ξ′⊗bΩ⟩.(18)\left\langle Z\bigl(b^*\rho(a_+)\bigr)\xi,\xi'\right\rangle = \left\langle C(\xi\otimes a\Omega),\xi'\otimes b\Omega\right\rangle. \tag*{(18)}

Proof.

Construction and marginals. Take finitely many elementary tensors. For T≥0T \ge0, apply the recurrence simultaneously to their two indices. At each stage the operator matrix is multiplied on the left and right by the adjoint and the original of the diagonal matrix of that position’s labels. Its positivity is preserved, and complete positivity of θ−dj\theta_{-d_j} preserves it at the next stage. Taking the final Ω\Omega coefficients therefore gives a positive scalar kernel. For T=1T=1, the identity EB∣N=φ(⋅)1E_B|_N=\varphi(\mathord{\cdot})1 reduces this kernel to

∏j=1nφ(yj∗xj),\prod_{j=1}^{n}\varphi(y_j^*x_j),

the Gram kernel of the elementary tensors. Applying positivity to 0≤T≤∥T∥10\le T\le\lVert T\rVert1 now bounds the quadratic form by ∥T∥\lVert T\rVert times the tensor norm squared. Thus the form depends only on the tensor vector, extends to the completion, and defines a positive unital map. The same argument with an additional matrix index proves complete positivity.

For T=λ(a)T=\lambda(a), the first map produces γ−dn(EB(a))∈B\gamma_{-d_n}(E_B(a))\in B with state value φ(a)\varphi(a). At a subsequent stage, an element b∈Bb\in B is replaced by γ−dj−1(EB(yj∗bxj))∈B\gamma_{-d_{j-1}}(E_B(y_j^* b x_j))\in B. Its state value is

φ(yj∗bxj)=φ(b)φ(yj∗xj),\varphi(y_j^*b x_j)=\varphi(b)\varphi(y_j^*x_j),

using EN∣B=φ(⋅)1E_N|_B=\varphi(\mathord{\cdot})1. Iterating this scalar identity gives the left marginal in (16). For the right marginal, the full right algebra is fixed by each θ\theta, hence belongs to its multiplicative domain. Right multipliers commute with the labels and pass through every stage, while EB∣N=φ(⋅)1E_B|_N=\varphi(\mathord{\cdot})1 factors the labels as above. For eNe_N, use [eN,N]=0[e_N,N]=0 and the fact that the fixed projection eNe_N also belongs to every map’s multiplicative domain. Since eNΩ=Ωe_N\Omega=\Omega, the recurrence gives the same Gram kernel as for T=1T=1. These computations establish all of (16).

The modular unitary group. First extend the spectral formula for each individual θ−q\theta_{-q} from C0(R)C_0(\mathbb{R}) to the functions s↦eitss\mapsto e^{its}. Its restriction to C0(XM)C_0(X_M) is a homomorphism, so that algebra belongs to its multiplicative domain. For f∈C0(R)f\in C_0(\mathbb{R}), it follows that

f(XM+q)θ−q(eitXM)=θ−q(f(XM)eitXM)=f(XM+q)eit(XM+q).f(X_M+q)\theta_{-q}(e^{itX_M}) = \theta_{-q}(f(X_M)e^{itX_M}) = f(X_M+q)e^{it(X_M+q)}.

The ranges of these cutoffs span a dense subspace of H\mathcal{H}. Consequently θ−q(eitXM)=eit(XM+q)\theta_{-q}(e^{itX_M})=e^{it(X_M+q)}; these unitaries also belong to the multiplicative domain. No limiting operation has been passed through θ−q\theta_{-q}.

In the kernel recurrence, commute eitXMe^{itX_M} to the right across each xjx_j, using eitXMxj=σt(xj)eitXMe^{itX_M}x_j=\sigma_t(x_j)e^{itX_M}. The next map replaces the left factor yj∗σt(xj)∈Ny_j^*\sigma_t(x_j)\in N by its state value. The resulting coefficient is

eitd∏j=1nφ(yj∗σt(xj)),e^{itd}\prod_{j=1}^{n}\varphi(y_j^*\sigma_t(x_j)),

which proves

Z(eitXM)=eit(Sn+d).(19)Z(e^{itX_M})=e^{it(S_n+d)}. \tag*{(19)}

A nonnormal map need not preserve weak operator integrals, so this identity alone does not yet justify (17). We next show that the coefficient states needed to recover continuous functions are supported inside a compact spectral interval.

Spectral cutoffs and continuous functions. Call x∈Nx\in N bandlimited if it admits a finite bound L≥0L\geq0 such that, for every interval JJ,

x1J(XM)H⊂1J+[−L,L](XM)H.x1_J(X_M)\mathcal{H}\subset1_{J+[-L,L]}(X_M)\mathcal{H}.

Such elements are strongly dense. Indeed, convolve the modular action with an approximate identity having smooth, compactly supported Fourier transforms. If the Fourier support of a smoothing kernel lies in [−L,L][-L,L], the resulting element has the displayed transfer property: testing eitXMxe−itXM=σt(x)e^{itX_M}xe^{-itX_M}=\sigma_t(x) against scalar Fourier kernels shows that spectral pieces separated by more than LL have zero matrix coefficient.

Two operations in the kernel preserve a suitable interval on which a positive operator vanishes. Suppose that A≥0A\geq0 annihilates the spectral subspace associated with an open interval II. For f∈Cc(I)f\in C_c(I), f(XM)A=0f(X_M)A=0, so the multiplicative-domain identity gives

f(XM+q)θ−q(A)=0.f(X_M+q)\theta_{-q}(A)=0.

The cutoff ranges span the spectral subspace for I−qI-q; thus θ−q(A)\theta_{-q}(A) annihilates that translated interval. If xx has transfer in [−L,L][-L,L], the preceding transfer rule also shows that x∗Axx^*Ax annihilates every interval JJ with J+[−L,L]⊂IJ+[-L,L]\subset I.

Apply these rules to a diagonal kernel with bandlimited labels xjx_j, of transfer bounds LjL_j, starting from A=1{∣XM∣>R}A=1_{\{|X_M|>R\}}. Translating by each −dj-d_j and shrinking by each LjL_j leaves an open interval about zero whenever

R>∑j=0n∣dj∣+∑j=1nLj.R>\sum_{j=0}^{n}|d_j|+\sum_{j=1}^{n}L_j.

The final operator then annihilates Ω\Omega, whose modular energy is zero. Hence its diagonal kernel coefficient vanishes. For a finite sum of bandlimited elementary tensors, choose one RR that works for every term; positivity and Cauchy–Schwarz make the off-diagonal coefficients vanish as well. These tensor sums form a dense subspace of H⊗nH^{\otimes n}.

For a unit vector ξ\xi in this subspace, set Φξ(T)=⟨Z(T)ξ,ξ⟩\Phi_\xi(T)=\langle Z(T)\xi,\xi\rangle and PR=1[−R,R](XM)P_R=1_{[-R,R]}(X_M). We have proved Φξ(1−PR)=0\Phi_\xi(1-P_R)=0. Cauchy–Schwarz therefore yields

Φξ(T)=Φξ(PRTPR).\Phi_\xi(T)=\Phi_\xi(P_RTP_R).

The transfer property applied to Ω\Omega also puts xjΩx_j\Omega in the XX-spectral subspace for [−Lj,Lj][-L_j,L_j]. Our choice of RR therefore places the spectral distribution of Sn+dS_n+d on ξ\xi in [−R,R][-R,R] as well. Both sides of (19), evaluated on ξ\xi, are thus determined by functions restricted to this same compact interval. Finite linear combinations of the exponentials form a uniformly dense self-adjoint algebra in C([−R,R])C([-R,R]). Uniform approximation now proves

⟨Z(f(XM))ξ,ξ⟩=⟨f(Sn+d)ξ,ξ⟩(f∈C0(R)).\langle Z(f(X_M))\xi,\xi\rangle=\langle f(S_n+d)\xi,\xi\rangle\qquad(f\in C_0(\mathbb{R})).

Continuity in ξ\xi and polarization prove (17) on the whole tensor product. The radius RR may depend on the labels, shifts, and tensor length: only the resulting exact identity will enter the uniform estimate.

The word coefficient. First, (11) and state preservation of EBE_B give

⟨U(xΩ⊗ζ),yΩ⊗bΩ⟩=⟨ζ,EB(x∗by)Ω⟩(x,y∈N, b∈B, ζ∈K).\langle U(x\Omega\otimes\zeta),y\Omega\otimes b\Omega\rangle=\langle\zeta,E_B(x^*by)\Omega\rangle\qquad(x,y\in N,\ b\in B,\ \zeta\in K).

The identity follows first for ζ=aΩ\zeta=a\Omega and then by density. It shows that moving an output test backwards through UU keeps its last factor in BΩB\Omega. To compute a word coefficient, take elementary tensors and record the adjoints of the successive test labels as

cn=γ−dn(b∗),cj−1=γ−dj−1(EB(yj∗cjxj))(j=n,…,1).c_n=\gamma_{-d_n}(b^*),\qquad c_{j-1}=\gamma_{-d_{j-1}}\bigl(E_B(y_j^*c_jx_j)\bigr)\qquad(j=n,\ldots,1).

Each cjc_j belongs to BB. Because the right multiplier is fixed and lies in the multiplicative domain of every θ\theta, the recurrence for T=b∗ρ(a+)T=b^*\rho(a_+) gives Tj=cjρ(a+)T_j=c_j\rho(a_+). Its final coefficient is φ(c0a)\varphi(c_0a).

The same recurrence computes the coefficient of CC backwards from its output vector. Moving the output test across WdjW_{d_j} applies γ−dj\gamma_{-d_j}, and (11) moves it across UjU_j by replacing cjc_j with EB(yj∗cjxj)E_B(y_j^*c_jx_j). The last coefficient is again φ(c0a)\varphi(c_0a). This proves (18) for elementary tensors; linearity and continuity give the general case. □

Proof of Proposition 5.1. First restrict to data for which the spectral support of Sn+dS_n+d on ξ\xi lies in [−δ,δ][-\delta,\delta]. The state

Φ(T)=⟨Z(T)ξ,ξ⟩\Phi(T)=\langle Z(T)\xi,\xi\rangle

has both marginals equal to φ\varphi and satisfies Φ(eN)=1\Phi(e_N)=1. By (17), along any sequence of such data with δ→0\delta\to0, every weak* accumulation state satisfies Φ(f(XM))=f(0)\Phi(f(X_M))=f(0) for f∈C0(R)f\in C_0(\mathbb{R}). Proposition 2.1(i), applied to the fixed operator a∗ρ(a+)a^*\rho(a_+), and (18) give

⟨C(ξ⊗aΩ),ξ⊗aΩ⟩⟶⟨a∗Ωa+,Ω⟩=∥aΩ∥2.\langle C(\xi\otimes a\Omega),\xi\otimes a\Omega\rangle\longrightarrow\langle a^*\Omega a_+,\Omega\rangle=\|a\Omega\|^2.

for every modular-analytic a∈Ba\in B. Since CC is a contraction,

∥C(ξ⊗aΩ)−ξ⊗aΩ∥2≤2∥aΩ∥2−2Re⁡⟨C(ξ⊗aΩ),ξ⊗aΩ⟩,\|C(\xi\otimes a\Omega)-\xi\otimes a\Omega\|^2\le2\|a\Omega\|^2-2\operatorname{Re}\langle C(\xi\otimes a\Omega),\xi\otimes a\Omega\rangle,

so this difference tends to zero. Density of analytic vectors aΩa\Omega and contractivity give the same conclusion for each fixed η∈K\eta\in K. The conclusion is uniform in nn, the shifts, and ξ\xi: otherwise a sequence of violating choices would have an accumulation state in the state space of the same algebra B(H)B(\mathcal{H}) and contradict the argument just given.

For the ideal word, the same spectral localization gives

∥Dall(ξ⊗η)−ξ⊗η∥≤sup⁡∣s∣≤δ∥Wsη−η∥⟶0.\|D_{\mathrm{all}}(\xi\otimes\eta)-\xi\otimes\eta\|\le\sup_{|s|\le\delta}\|W_s\eta-\eta\|\longrightarrow0.

Combining these bounds proves the centered estimate. Finally suppose the SnS_n-support of ξ\xi lies in any interval of length δ\delta. Choose qq so that the containing interval for Sn+d+qS_n+d+q is centered at zero, and replace the final shift dnd_n by dn+qd_n+q. This replaces CC and DallD_{\mathrm{all}} by WqCW_qC and WqDallW_qD_{\mathrm{all}}. Their difference has the same norm, and the input vector η\eta is unchanged. The centered estimate therefore proves (14) with all the asserted uniformity. □

Repeated tensor tests

We now prove that the comparison is exact, using only the uniform estimate and the operator structure in (12)–(13). A discrepancy between UU and DD makes their average (U+D)/2(U+D)/2 lose squared norm. Yet arbitrarily long words in these averages remain uniformly close, after time averaging, to the ideal unitary words. Random recentering will capture the discrepancy with a nonsummable sequence of probabilities, forcing more total norm loss than is available. The sampling must control the approximation error as well as detect the discrepancy; independent sampling provides the same probability factor in both estimates.

Lemma 6.1 (Hybrid words and time averaging). The estimate (14) holds with the same modulus if any subset of the UjU_j is replaced by DjD_j, or if all UjU_j are replaced by Gj=(Uj+Dj)/2G_j=(U_j+D_j)/2. For every δ>0\delta>0 there is a probability density pδp_\delta on R\mathbb{R} such that, for every unit ξ∈H⊗n\xi\in H^{\otimes n} without a spectral support restriction,

∫R∥(C−Dall)(eitSnξ⊗η)∥2pδ(t) dt≤ϵη(δ)2.(20)\int_{\mathbb{R}}\left\lVert(C-D_{\mathrm{all}})(e^{itS_n}\xi\otimes\eta)\right\rVert^2p_\delta(t)\,\mathrm{d}t\leq\epsilon_\eta(\delta)^2. \tag*{(20)}

The statement includes the hybrid and averaged words.

Proof. If no UjU_j remains, the word equals DallD_{\mathrm{all}}. Otherwise disintegrate the replaced positions over their XX-spectral variables. On a fiber with values (xj)(x_j), each replaced DjD_j acts as the shift WxjW_{x_j} on KK, and the sum of the remaining energies lies in the original interval translated by −∑xj-\sum x_j. The interval’s length is unchanged. The fiber vector in the remaining positions need not be a simple tensor: Proposition 5.1 applies to every such vector, with the same last factor η\eta. If a fiber has spectral multiplicity, expand in an orthonormal basis of that multiplicity space and sum the homogeneous squared bounds. Integrating proves the hybrid assertion. Convex expansion proves the assertion for GjG_j; the coefficients of the expansion sum to one, independently of nn.

Choose an even real L2L^2-normalized smooth function gg supported in an interval of length δ\delta, and put pδ(t)=∣g^(t)∣2p_\delta(t)=|\widehat{g}(t)|^2, with unitarily normalized Fourier transform. For the bounded operator T=C−DallT=C-D_{\mathrm{all}}, vector-valued Plancherel gives

∫R∥T(eitSnξ⊗η)∥2∣g^(t)∣2 dt=∫R∥T(g(Sn−z)ξ⊗η)∥2 dz.\int_{\mathbb{R}}\left\lVert T(e^{itS_n}\xi\otimes\eta)\right\rVert^2|\widehat{g}(t)|^2\,\mathrm{d}t=\int_{\mathbb{R}}\left\lVert T(g(S_n-z)\xi\otimes\eta)\right\rVert^2\,\mathrm{d}z.

Here Plancherel is applied to z↦g(Sn−z)ξ⊗ηz\mapsto g(S_n-z)\xi\otimes\eta and then to its image under the bounded operator TT; no commutation of TT with SnS_n is required. The Fourier sign does not matter with this choice of gg. Each vector on the right is spectrally localized as required by (14), and

∫R∥g(Sn−z)ξ∥2 dz=∥ξ∥2=1.\int_{\mathbb{R}}\left\lVert g(S_n-z)\xi\right\rVert^2\,\mathrm{d}z=\left\lVert\xi\right\rVert^2=1.

Apply the homogeneous form of that estimate and integrate.

Theorem 6.2. For the operators in (10), one has U=DU=D.

Proof. Suppose otherwise. Density of compact-spectrum vectors gives unit vectors h∈Hh\in H and η∈K\eta\in K with supp⁡X(h)⊂[−l,l]\operatorname{supp}_X(h)\subset[-l,l] and ∥(U−D)(h⊗η)∥>0\left\lVert(U-D)(h\otimes\eta)\right\rVert>0. By strong continuity there are c>0c>0 and 0<b≤10<b\leq1 such that

∥(U−D)(eit′Xh⊗Ws′η)∥≥c(∣t′∣,∣s′∣≤b).(21)\left\lVert(U-D)(e^{it'X}h\otimes W_{s'}\eta)\right\rVert\geq c\qquad(|t'|,|s'|\leq b). \tag*{(21)}

Use this fixed η\eta in Lemma 6.1. Choose δ\delta so that ϵ:=ϵη(δ)<c/8\epsilon:=\epsilon_\eta(\delta)<c/8, and choose A≥1A\geq1 with

∫−AApδ(t) dt≥3/4.\int_{-A}^{A}p_\delta(t)\,\mathrm{d}t\geq3/4.

The repeated experiment. Fix a finite terminal length mm. On one product probability space choose tt with density pδp_\delta and mutually independent parameters

rj∼Unif⁡[−2A,2A],sj∼Unif⁡[−Bj,Bj],Bj=b+1+jl(1≤j≤m),r_j \sim\operatorname{Unif}[-2A,2A], \qquad s_j \sim\operatorname{Unif}[-B_j,B_j], \qquad B_j=b+1+jl \quad(1\le j\le m),

all independent of tt. At position jj put the fresh input ei(t−rj)Xhe^{i(t-r_j)X}h. Starting with η\eta on KK, propagate using WsjGjW−sjW_{s_j}G_jW_{-s_j}. Let ζk\zeta_k denote the result after kk positions, and let χk\chi_k be the ideal result with DjD_j in place of GjG_j. Since DjD_j commutes with the shifts on KK,

χk=exp⁡(iSk⊗Y)(⨂j=1kei(t−rj)Xh⊗η).\chi_k=\exp(iS_k\otimes Y)\left(\bigotimes_{j=1}^{k}e^{i(t-r_j)X}h\otimes\eta\right).

Condition on every parameter other than tt. Then ξk=⨂j=1ke−irjXh\xi_k=\bigotimes_{j=1}^{k}e^{-ir_jX}h is a fixed unit vector, the input is eitSkξk⊗ηe^{itS_k}\xi_k\otimes\eta, and the successive conjugations WsjGjW−sjW_{s_j}G_jW_{-s_j} form an averaged word with fixed shifts summing to zero. Lemma 6.1 therefore gives

∫∥ζk−χk∥2pδ(t) dt≤ϵ2.(22)\int\|\zeta_k-\chi_k\|^2p_\delta(t)\,dt\le\epsilon^2. \tag*{(22)}

For k=0k=0, both vectors are η\eta.

A definite loss at each captured test. At step j=k+1j=k+1, insert the fresh position into ζk\zeta_k and apply W−sjW_{-s_j} on KK; call this input vv. Let v0v_0 be the analogous input formed from χk\chi_k. The parallelogram identity and contractivity give

∥v∥2−∥Gjv∥2≥14∥(Uj−Dj)v∥2.(23)\|v\|^2-\|G_jv\|^2\ge\frac{1}{4}\|(U_j-D_j)v\|^2. \tag*{(23)}

Define the random projection

Q=1{∣t−rj∣≤b}1{∣Sk−sj∣≤b},Q=1_{\{|t-r_j|\le b\}}1_{\{|S_k-s_j|\le b\}},

where the first factor is a scalar indicator and the second acts on the previous kk positions. For k=0k=0, interpret S0S_0 as zero. The projection QQ commutes with UjU_j, DjD_j and the last-position shifts. Put

aj=b2AbBj.a_j=\frac{b}{2A}\frac{b}{B_j}.

We claim

E∥(Uj−Dj)Qv0∥2≥34c2aj,(24)\mathbb{E}\|(U_j-D_j)Qv_0\|^2\ge\frac{3}{4}c^2a_j, \tag*{(24)}
E∥(Uj−Dj)Q(v−v0)∥2≤4ϵ2aj.(25)\mathbb{E}\|(U_j-D_j)Q(v-v_0)\|^2\le4\epsilon^2a_j. \tag*{(25)}

Here E\mathbb{E} integrates all parameters involved, including tt.

For (24), disintegrate the previous ideal vector over Sk=eS_k=e. Its last factor is WeηW_e\eta, and ∣e∣≤kl|e|\le kl. The interval [e−b,e+b][e-b,e+b] lies in [−Bj,Bj][-B_j,B_j], so the probability over sjs_j of capturing it is exactly b/Bjb/B_j, as illustrated in Figure 1. For ∣t∣≤A|t|\le A, the probability over rjr_j of ∣t−rj∣≤b|t-r_j|\le b is b/(2A)b/(2A). On these fibers the current last factor is We−sjηW_{e-s_j}\eta, so (21) applies. Integrating, and using the mass at least 3/43/4 of [−A,A][-A,A], proves (24).

Capture at step $j=k+1$, shown schematically

Figure 1. Capture at step j=k+1j=k+1, shown schematically. For an ideal spectral value ∣e∣≤kl|e|\le kl, the entire capture interval [e−b,e+b][e-b,e+b] lies in the sampling interval, giving probability 2b/(2Bj)=b/Bj2b/(2B_j)=b/B_j. The actual propagated error has no assumed spectral bound; for any of its spectral values the capture probability is at most b/Bjb/B_j.

For (25), fix tt and all parameters from the previous kk steps, so that ζk−χk\zeta_k-\chi_k is fixed. No spectral support bound on this error is available. For every real ee,

Pr⁡sj(∣e−sj∣≤b)=∣[e−b,e+b]∩[−Bj,Bj]∣2Bj≤bBj.\Pr_{s_j}(|e-s_j|\le b)=\frac{|[e-b,e+b]\cap[-B_j,B_j]|}{2B_j}\le\frac{b}{B_j}.

The spectral theorem therefore gives the quadratic-form inequality

12Bj∫−BjBj1{∣Sk−s∣≤b} ds≤bBj.\frac{1}{2B_j}\int_{-B_j}^{B_j}\mathbf{1}_{\{|S_k-s|\le b\}}\,ds\le\frac{b}{B_j}.

Also Pr⁡rj(∣t−rj∣≤b)≤b/(2A)\Pr_{r_j}(|t-r_j|\le b)\le b/(2A) for every fixed tt. The two current sampling parameters are independent of each other and of the past. The fresh vector has norm one, while W−sjW_{-s_j} commutes with QQ and does not affect its norm. Consequently

Erj,sj∥Q(v−v0)∥2≤aj∥ζk−χk∥2.\mathbb{E}_{r_j,s_j}\|Q(v-v_0)\|^2\le a_j\|\zeta_k-\chi_k\|^2.

Now integrate over tt and the past parameters to obtain

E∥Q(v−v0)∥2≤ajE∥ζk−χk∥2≤ajϵ2\mathbb{E}\|Q(v-v_0)\|^2\le a_j\mathbb{E}\|\zeta_k-\chi_k\|^2\le a_j\epsilon^2

by (22). Since ∥Uj−Dj∥≤2\|U_j-D_j\|\le2, this proves the claim.

The contradiction. Since QQ commutes with the step operators, ∥(Uj−Dj)v∥2≥∥(Uj−Dj)Qv∥2\|(U_j-D_j)v\|^2\ge\|(U_j-D_j)Qv\|^2. The triangle inequality in the mean-square Hilbert space, followed by (23), shows that the expected squared-norm loss at step jj is at least

aj4(32c−2ϵ)2.(26)\frac{a_j}{4}\left(\frac{\sqrt{3}}{2}c-2\epsilon\right)^2. \tag*{(26)}

The constant in parentheses is positive. On the other hand, inserting a unit fresh vector and applying the outer shift do not change norm. The losses therefore telescope from one and have total at most one. Since ∑aj=∞\sum a_j=\infty, choosing sufficiently many steps contradicts (26). Hence U=DU=D.

The continuous core and the general inclusion

The multiplication identity now determines the fixed points of the bicentralizer flow in the core. We recall why the core fixed-point equality yields an amenable subalgebra, then prove that equality in coordinates adapted to NN. The final reduction passes from factors to an arbitrary center and invokes the existing amenable-subalgebra theorem.

The core criterion. For a type III1_1 bottom factor, the bicentralizer in the core and its flow have the following description. Here B(N⊂c(M),φ)\mathrm{B}(N \subset c(M),\varphi) uses the same asymptotic commutation condition on bounded sequences in NN, with the state φ\varphi on NN. In the realization c(M)=M⋊σφRc(M)=M\rtimes_{\sigma_\varphi}\mathbb{R}, with group unitaries l(t)l(t), one has

B(N⊂c(M),φ)=B⋊σφR,γs(l(t))=e−itsl(t).(27)\mathrm{B}(N\subset c(M),\varphi)=B\rtimes_{\sigma_\varphi}\mathbb{R},\qquad\gamma_s(l(t))=e^{-its}l(t). \tag*{(27)}

The algebra identity is [10], Proposition 7.2; the action extends the flow on BB as in [10], Proposition 8.6, proof. One inclusion in the desired fixed-point equality is automatic:

N′∩c(M)⊂B(N⊂c(M),φ)γ.(28)N'\cap c(M)\subset\mathrm{B}(N\subset c(M),\varphi)^\gamma. \tag*{(28)}

This inclusion follows by combining [10], Proposition 8.1 and its proof, Theorem 6.20(5), and Proposition 8.4. We will prove the reverse inclusion. Equality is equivalent to the relative bicentralizer conjecture by [10], Proposition 8.6. For an expected inclusion, that conjecture is equivalent to the weak Dixmier property for N⊂c(M)N\subset c(M):

conv⁡‾ w∗{uxu∗:u∈U(N)}∩(N′∩c(M))≠∅(x∈c(M)).(29)\overline{\operatorname{conv}}^{\,w^*}\{uxu^*:u\in\mathcal{U}(N)\}\cap(N'\cap c(M))\ne\varnothing\qquad(x\in c(M)). \tag*{(29)}

With separable preduals, (29) is equivalent to the conclusion of Theorem 1.1, by [10], Theorem D.

Proposition 7.1. If NN is a type III1_1 factor, then

B(N⊂c(M),φ)γ=N′∩c(M).\mathrm{B}(N\subset c(M),\varphi)^\gamma=N'\cap c(M).

Proof.

The algebra generated by NN and BB. By Theorem 6.2, L∗R=DL^*R=D is unitary. Since L,RL,R are isometries, this implies R=LDR=LD, so their ranges agree. Put A=B∨NA=B\vee N. The common range is L2(A,φ)L^2(A,\varphi): it contains Ω\Omega, is invariant under left NN by its RR description and under left BB by its LL description, and is contained in L2(A,φ)L^2(A,\varphi). Applying adjoints gives invariance under the generated von Neumann algebra. Both NN and BB are modularly invariant, so the expected algebra AA has a canonically embedded core inside c(M)c(M).

Coordinates adapted to NN. Use LL to identify L2(A,φ)L^2(A,\varphi) with H⊗KH\otimes K. In these coordinates left BB acts as 1⊗B1\otimes B, left NN acts as D(N⊗1)D∗D(N\otimes1)D^*, and the modular group is

eitX⊗Δφ∣Bite^{itX}\otimes\Delta_{\varphi\mid B}^{it}

Represent c(A)c(A) faithfully on H⊗K⊗L2(R)H\otimes K\otimes L^2(\mathbb{R}) with constant base algebra and group unitaries

eitX⊗Δφ∣Bit⊗lR(t),e^{itX}\otimes\Delta_{\varphi\mid B}^{it}\otimes l_{\mathbb{R}}(t),

where lRl_{\mathbb{R}} is the translation representation. This is the usual regular crossed-product representation after conjugating its variable base algebra by the modular implementation. Conjugate by D∗D^* on the first two positions. This removes the twist from the action of NN, which now acts on HH alone, and transfers it to the core of BB. Since WsW_s commutes with Δφ∣Bit\Delta_{\varphi\mid B}^{it}, the algebra B⋊RB\rtimes\mathbb{R} is represented, at spectral value X=hX=h, by

a⟼γ−h(a)⊗1,l(t)⟼eithΔφ∣Bit⊗lR(t).(30)a\longmapsto\gamma_{-h}(a)\otimes1,\qquad l(t)\longmapsto e^{ith}\Delta_{\varphi\mid B}^{it}\otimes l_{\mathbb{R}}(t). \tag*{(30)}

The two transformations in (30) are the same flow: on BB it is γ−h\gamma_{-h}, and on l(t)l(t) it multiplies by eithe^{ith}. Thus these fibers are a fixed normal representation of B⋊RB \rtimes\mathbb{R} composed with the core bicentralizer flow at −h-h, by (27). More explicitly, if H=∫⊕Hh dν(h)H = \int^{\oplus} H_h\,d\nu(h) and π0\pi_0 is the fixed regular representation on K⊗L2(R)K \otimes L^2(\mathbb{R}), the representation of an element z∈B⋊Rz \in B \rtimes\mathbb{R} is

∫⊕IHh⊗π0(γ−h(z)) dν(h).\int^{\oplus} I_{H_h} \otimes\pi_0(\gamma_{-h}(z))\,d\nu(h).

This normal direct-integral representation agrees with (30) on generators, so the formula applies to the entire crossed product.

The fixed algebra. If zz is fixed by the core bicentralizer flow, then π0(γ−h(z))=π0(z)\pi_0(\gamma_{-h}(z)) = \pi_0(z) in every fiber. It therefore acts on K⊗L2(R)K \otimes L^2(\mathbb{R}) alone, while NN acts on HH alone. Hence it commutes with NN, proving

B(N⊂c(M),φ)γ⊂N′∩c(M).B(N \subset c(M),\varphi)^\gamma\subset N' \cap c(M).

The opposite inclusion is (28), so the fixed algebras agree.

Proof of Theorem 1.1.

Factor bottom algebras. For a type III1_1 bottom factor, Proposition 7.1 and [10], Proposition 8.6 and Theorem D give the weak Dixmier property (29). The state on NN is a strictly semifinite weight, as required there. For a bottom factor of any other type, the same weak Dixmier property is [10], Proposition 4.11. These factor cases allow arbitrary upper algebras.

Disintegration over the center. For a general NN, put J=Z(N)J = \mathcal{Z}(N) and M0=J′∩MM_0 = J' \cap M. The passage to M0M_0 makes JJ central in both algebras, so that we can apply the factor cases fiber by fiber. The modular action of φ=φ∘EN\varphi= \varphi\circ E_N fixes JJ pointwise. Consequently M0M_0 is modularly invariant, and ENE_N restricts to a faithful normal expectation M0→NM_0 \to N. Moreover,

J′∩c(M)=c(M0)(31)J' \cap c(M) = c(M_0) \tag*{(31)}

inside the chosen core, by [10], Proposition 4.7.

Disintegrate N⊂M0N \subset M_0 over the common central algebra JJ, using the probability measure μ\mu determined by φ∣J\varphi|_J. The bottom fibers NzN_z are factors because J=Z(N)J = \mathcal{Z}(N); the upper fibers (M0)z(M_0)_z need not be factors. The simultaneous disintegration theorem for operator-valued weights gives faithful normal fiber expectations Ez:(M0)z→NzE_z : (M_0)_z \to N_z and compatible faithful state fibers φz\varphi_z [8], Propositions 2.1–2.2 and Theorem 2.3. Indeed, ENE_N is JJ-bimodular, and its fiber weights satisfy Ez(1)=1E_z(1) = 1, so they are conditional expectations. The state identity φ=φ∘EN\varphi= \varphi\circ E_N and the modular actions also disintegrate, as in the proof of that theorem.

To see the resulting core compatibility, write H0=L2(M0,φ)=∫⊕Hz dμ(z)H_0 = L^2(M_0,\varphi) = \int^{\oplus} H_z\,d\mu(z), where Hz=L2((M0)z,φz)H_z = L^2((M_0)_z,\varphi_z). The regular representation of the core on L2(R,H0)=∫⊕L2(R,Hz) dμ(z)L^2(\mathbb{R},H_0) = \int^{\oplus} L^2(\mathbb{R},H_z)\,d\mu(z) is given by

(π(a)ξ)(t)=σ−tφ(a)ξ(t),(l(s)ξ)(t)=ξ(t−s).(\pi(a)\xi)(t) = \sigma_{-t}^\varphi(a)\xi(t), \qquad(l(s)\xi)(t) = \xi(t-s).

Both formulas act fiberwise. A countable family of algebra generators and the rational translation parameters generate the fiber cores; the common central algebra JJ is already present in the representation. Thus c(M0)=∫⊕c((M0)z) dμ(z)c(M_0) = \int^{\oplus} c((M_0)_z)\,d\mu(z) with the corresponding copies of NzN_z. Separability supplies these standard measurable fields. The factor cases and the direct-integral criterion [10], Proposition 4.12 now give the weak Dixmier property for N⊂c(M0)N \subset c(M_0).

Averaging into the central commutant. We have proved the weak Dixmier property inside c(M0)c(M_0). To apply it to an arbitrary x∈c(M)x \in c(M), we first average xx into this smaller core. The weak* compact convex hull of its orbit under U(J)\mathcal{U}(J) is preserved by the commuting affine maps Ad⁡(u)\operatorname{Ad}(u), u∈U(J)u \in\mathcal{U}(J). The compact-convex fixed-point theorem for commuting affine maps gives a common fixed point y∈J′∩c(M)=c(M0)y \in J^{\prime} \cap c(M) = c(M_0). One can obtain it by taking rectangular Cesàro averages for each finite family of commuting maps, then using compactness and the finite-intersection property. Since J⊂NJ \subset N, this point belongs to the weak* closed convex NN-orbit hull of xx. That hull is NN-invariant. Applying weak Dixmier averaging inside c(M0)c(M_0) therefore produces a point of N′∩c(M)N^{\prime} \cap c(M) still in the original hull. This proves (29) for the original inclusion.

The amenable subalgebra. Finally [10], Theorem D, (3)⇒\Rightarrow(4) supplies an amenable subalgebra P⊂NP \subset N with a faithful normal conditional expectation and P′∩c(M)=N′∩c(M)P^{\prime} \cap c(M) = N^{\prime} \cap c(M). The expectation is faithful and normal by the meaning of “with expectation” in that theorem. Unitality is automatic: if p=1P<1Np = 1_P < 1_N, bimodularity would imply F(1−p)=0F(1-p) = 0, contradicting faithfulness. This proves the result.

References

  1. [1]H. Ando and U. Haagerup, Ultraproducts of von Neumann algebras, J. Funct. Anal. 266 (2014), no. 12, 6842–6913. arXiv:1212.5457v3.arxiv.org/abs/1212.5457
  2. [2]H. Ando, U. Haagerup, C. Houdayer, and A. Marrakchi, Structure of bicentralizer algebras and inclusions of type III factors, Math. Ann. 376 (2020), 1145–1194. arXiv:1804.05706v2.DOI
  3. [3]A. Connes, Classification des facteurs, in Operator Algebras and Applications, Part 2 (Kingston, Ontario, 1980), Proc. Sympos. Pure Math. 38, American Mathematical Society, Providence, RI, 1982, 43–109.
  4. [4]A. Connes, Factors of type III₁, property L′λ, and closure of inner automorphisms, J. Operator Theory 14 (1985), 189–211.
  5. [5]U. Haagerup, Connes’ bicentralizer problem and uniqueness of the injective factor of type III₁, Acta Math. 158 (1987), 95–148. doi:10.1007/BF02392257.DOI
  6. [6]C. Houdayer and Y. Isono, Unique prime factorization and bicentralizer problem for a class of type III factors, Adv. Math. 305 (2017), 402–455. arXiv:1503.01388v3.DOI
  7. [7]C. Houdayer and A. Marrakchi, The classification of flows on II₁ factors and Connes’ bicentralizer problem, 2026, version 1, September 10, 2026. arXiv:2609.11462v1.
  8. [8]R. Ichihara and S. Kawakami, Disintegration of operator valued weights, Scientiae Mathematicae Japonicae Online e-2007 (2007), 693–699. Publisher PDF.
  9. [9]A. Marrakchi, Full factors, bicentralizer flow and approximately inner automorphisms, Invent. Math. 222 (2020), 375–398. arXiv:1811.10253v2.arxiv.org/abs/1811.10253
  10. [10]A. Marrakchi, Kadison’s problem for type III subfactors and the bicentralizer conjecture, Invent. Math. 239 (2025), 79–163. arXiv:2308.15163v3.arxiv.org/abs/2308.15163
  11. [11]A. Marrakchi, Kadison’s problem and ergodicity of the bicentralizer flow, 2026, version 2, 11 August 2026. arXiv:2606.26366v2.
  12. [12]A. Marrakchi and S. Vaes, Ergodic states on type III₁ factors and ergodic actions, J. Reine Angew. Math. 809 (2024), 247–260. arXiv:2305.14217v3.
  13. [13]T. Masuda, An analogue of Connes–Haagerup approach for classification of subfactors of type III₁, J. Math. Soc. Japan 57 (2005), 959–1001. doi:10.2969/jmsj/1150287301.
  14. [14]OpenAI, Bounded recovery for modular spectral averages, OpenAI Math Release preprint OAI:Bounded-recovery-for-modular-spectral-averages-September-23-2026, 2026.
  15. [15]S. Popa, On a problem of R. V. Kadison on maximal abelian ∗-subalgebras in factors, Invent. Math. 65 (1981), 269–281. doi:10.1007/BF01389015.DOI
  16. [16]M. Takesaki, Conditional expectations in von Neumann algebras, J. Funct. Anal. 9 (1972), 306–321. doi:10.1016/0022-1236(72)90004-3.DOI
  17. [17]M. Takesaki, Theory of Operator Algebras II, Encyclopaedia of Mathematical Sciences 125, Springer-Verlag, Berlin, 2003. doi:10.1007/978-3-662-10451-4.DOI

Paper details

vlnc:W_evnmrnht
Version
v1 · latest
Published
License
LicenseRef-OpenAI-public
Usage

views · 8

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

unreleased internal OpenAI model

View statement
Looking for related papers…
Contents