Open MathsAll courses
Positive maps and finite-dimensional approximation · Prerequisite proofs · Sources and terms

Fixed modular corners from the whole finite-star graph

New exposition and diagram: GPT-6.1 Sol (OpenAI), Ultra, 2026-10-07. Dedicated under CC0-1.0 to the extent of rights held. Human sources cited below retain their own rights.

Let M≠0M\ne0 be any von Neumann algebra, φ\varphi a faithful normal semifinite weight, and e≠0e\ne0 a projection fixed by σφ\sigma^\varphi. No countability assumption is imposed and φ(e)\varphi(e) may be infinite. We prove that the restriction φe\varphi_e to eMeeMe is faithful normal semifinite and

σtφe=σtφ∣eMe. \sigma_t^{\varphi_e}=\sigma_t^\varphi|_{eMe}.

We identify its entire GNS space and closed modular-operator domains, rather than only an analytic subalgebra.

A fixed corner retains the complete closed modular graph Any faithful normal semifinite weight; e ≠ 0 fixed; φ(e) may be infinite Whole core Λ(A) S₀Λ(x) = Λ(x*) x ∈ N ∩ N* Exact corner core Λ(Aₑ) S₀Λ(exe) = Λ(ex*e) Aₑ = eMe ∩ A Closed S on H Whole D(S) and D(S*) Closed S restricted to qH V Sφₑ V* = S | qH q = π(e) Jπ(e)J qΛ(x) = Λ(exe) S₀q = qS₀ (FC2–4) Projected graph cores q reduces S and S* (FC4) closure closure V Δφₑ V* = Δ | qH, with D(Δ | qH) = D(Δ) ∩ qH The implemented group on eMe is exactly σφ restricted to eMe (FC5) Subspace reduction, not equality of the ambient and corner spectra Proof: FC1–5 below. Context: Hiai v1, Lemmas 7.4–7.5, pp.65–67. New figure: CC0.

The four boxes are actual Hilbert subspaces and closed domains. FC2–4 construct their maps and prove every graph and adjoint reduction. FC5 proves the implemented group identity. The hypotheses allow an infinite corner weight and arbitrary Hilbert space.

The prerequisites are finite ideals and the complete weight GNS construction, GW1–5, ultraweak continuity of normal-weight GNS representations, NF5, the finite-star involution and full bounded-vector identity, WR3–5, modular implementation for every faithful normal semifinite weight, MW4, full closed-involution polar data, CI3, uniqueness of the self-adjoint generator, RF5, and full spectral covariance, SF. The pure Hilbert-algebra normalization and fullification used in this route are proved in BG1–6, including the exact bounded-vector domain identities.

Write (H,π,Λ)(H,\pi,\Lambda) for the complete GNS construction, N=nφN=\mathfrak n_\varphi, A=N∩N∗A=N\cap N^*, S=[Λ(x)↦Λ(x∗)]x∈A‾=JΔ1/2S=\overline{[\Lambda(x)\mapsto\Lambda(x^*)]_{x\in A}}=J\Delta^{1/2}, and Ut=ΔitU_t=\Delta^{it}. Inner products are linear in the first variable. The exact WR4 identity is Bl∩D(S)=Λ(A)B_l\cap D(S)=\Lambda(A), and a vector Λ(x)\Lambda(x), x∈Nx\in N, has left multiplier π(x)\pi(x). MW4 and BG1–6 give Jπ(M)J=π(M)′J\pi(M)J=\pi(M)'.

FC1. Spectral commutation for the fixed projection

Put P=π(e)P=\pi(e). Faithful implementation and fixedness give UtPUt∗=PU_tPU_t^*=P. To include the domain implication explicitly, the self-adjoint unitary w=2P−Iw=2P-I commutes with every UtU_t. If L=log⁡ΔL=\log\Delta, then wLwwLw, with domain wD(L)wD(L), is self-adjoint and generates the same group, by full spectral covariance. RF5's uniqueness therefore gives wLw=LwLw=L, with equality of domains. Spectral covariance now gives commutation of ww, and hence PP, with every spectral projection of LL. Since Δ=exp⁡L\Delta=\exp L, PP preserves D(Δr)D(\Delta^r) and commutes there with Δr\Delta^r for every real rr. In particular it preserves D(S)=D(Δ1/2)D(S)=D(\Delta^{1/2}).

FC2. Right multiplication on the whole finite ideal

Let x∈Ax\in A. The vector PΛ(x∗)=Λ(ex∗)P\Lambda(x^*)=\Lambda(ex^*) is left bounded and lies in D(S)D(S), by FC1 and the left-ideal property. WR4's exact intersection, and injectivity of both the left multiplier and the faithful GNS map, imply ex∗∈Aex^*\in A. Its adjoint xexe is therefore in AA. On the actual domains,

Λ(xe)=SPΛ(x∗)=JPΔ1/2Λ(x∗)=JPJΛ(x).(FC1) \Lambda(xe)=SP\Lambda(x^*) =JP\Delta^{1/2}\Lambda(x^*) =JPJ\Lambda(x). \tag{FC1}

The last step uses SΛ(x∗)=Λ(x)S\Lambda(x^*)=\Lambda(x), so Δ1/2Λ(x∗)=JΛ(x)\Delta^{1/2}\Lambda(x^*)=J\Lambda(x).

For every finite positive aa, its square root belongs to AA. Equation (FC1) and the projection norm give

φ(eae)=∥Λ(a1/2e)∥2≤∥Λ(a1/2)∥2=φ(a).(FC2) \varphi(eae)=\|\Lambda(a^{1/2}e)\|^2 \leq \|\Lambda(a^{1/2})\|^2=\varphi(a). \tag{FC2}

If x∈Nx\in N, apply this to a=x∗xa=x^*x. It proves xe∈Nxe\in N, with ∥Λ(xe)∥≤∥Λ(x)∥\|\Lambda(xe)\|\leq\|\Lambda(x)\|. Thus right multiplication by ee extends to a bounded operator on the whole GNS space.

To identify that operator, let ui↑1u_i\uparrow1 be GW's finite positive contractions. Then uix∈mφ⊂Au_ix\in m_\varphi\subset A, Λ(uix)→Λ(x)\Lambda(u_ix)\to\Lambda(x), and, since xe∈Nxe\in N, Λ(ui(xe))→Λ(xe)\Lambda(u_i(xe))\to\Lambda(xe). Apply (FC1) to uixu_ix and take these Hilbert-norm limits. It follows that

Λ(xe)=RΛ(x),R=JPJ,x∈N.(FC3) \Lambda(xe)=R\Lambda(x),\qquad R=JPJ, \quad x\in N. \tag{FC3}

Here RR is a self-adjoint projection in π(M)′\pi(M)'. In particular PP and RR commute. This proof of the right-domain statement uses no KMS uniqueness, entire-element core, centralizer trace, or unbounded centralizer-density theorem.

FC3. The exact corner weight and GNS space

Normality and faithfulness of φe\varphi_e follow from restriction: increasing bounded positive corner nets have the same ambient strong supremum. Its precise finite and finite-star domains are

Ne=eMe∩N,Ae=eMe∩A. N_e=eMe\cap N,\qquad A_e=eMe\cap A.

For every x∈Nx\in N, FC3 and the left ideal give exe∈Neexe\in N_e. GW's semifiniteness proof makes NN ultraweakly dense in MM. Compression is ultraweakly continuous: in each summable vector-series coefficient, replace the two vector sequences by their images under e; their squared-norm sums remain finite. Hence compression carries ultraweakly convergent nets to ultraweakly convergent nets. Normal compression therefore makes eNeeNe ultraweakly dense in eMeeMe. Since eNe⊆NeeNe\subseteq N_e, GW's finite-ideal criterion proves semifiniteness of φe\varphi_e. This does not assert that ee itself has finite weight.

Set q=PRq=PR. For every x∈Nx\in N,

qΛ(x)=Λ(exe).(FC4) q\Lambda(x)=\Lambda(exe). \tag{FC4}

The image of the dense GNS range under qq is therefore exactly Λ(Ne)\Lambda(N_e), which is dense in qHqH. The finite linear extensions of φ\varphi and φe\varphi_e agree on span⁡Ne∗Ne\operatorname{span}N_e^*N_e, by uniqueness of their extensions from the same finite positive values. Thus

VΛe(a)=Λ(a),a∈Ne, V\Lambda_e(a)=\Lambda(a),\qquad a\in N_e,

extends to a unitary V:Hφe→qHV:H_{\varphi_e}\to qH. For a∈eMea\in eMe, qq commutes with π(a)\pi(a) and

Vπe(a)V∗=π(a)∣qH.(FC5) V\pi_e(a)V^*=\pi(a)|_{qH}. \tag{FC5}

FC4. Full graph and adjoint reduction

The projection qq maps the entire initial finite-star domain Λ(A)\Lambda(A) onto Λ(Ae)\Lambda(A_e): FC2 gives the forward inclusion, and a=eaea=eae gives the converse. On that domain,

S0qΛ(x)=Λ(ex∗e)=qS0Λ(x). S_0q\Lambda(x)=\Lambda(ex^*e)=qS_0\Lambda(x).

Both qq and I−qI-q preserve the core. Approximating in the graph norm of SS proves qD(S)⊆D(S)qD(S)\subseteq D(S) and Sq=qSSq=qS on D(S)D(S). Projection of the same approximating pairs shows that Λ(Ae)\Lambda(A_e) is a graph core for the part of SS on qHqH. Hence

VSφeV∗=S∣qH(FC6) VS_{\varphi_e}V^*=S|_{qH} \tag{FC6}

with equality of the complete closed domains.

For η∈D(S∗)\eta\in D(S^*) and ξ∈D(S)\xi\in D(S), the conjugate-linear adjoint definition gives

⟨Sξ,qη⟩=⟨Sqξ,η⟩=⟨S∗η,qξ⟩=⟨qS∗η,ξ⟩. \langle S\xi,q\eta\rangle =\langle Sq\xi,\eta\rangle =\langle S^*\eta,q\xi\rangle =\langle qS^*\eta,\xi\rangle.

Thus qη∈D(S∗)q\eta\in D(S^*) and S∗q=qS∗S^*q=qS^*. The complementary projection gives the second reducing summand. The adjoint of S∣qHS|_{qH} is precisely S∗∣qHS^*|_{qH}: an adjoint pairing bounded only on qD(S)qD(S) is bounded on all D(S)D(S), because its pairing on (I−q)D(S)(I-q)D(S) vanishes. This proves the converse domain inclusion as well.

Taking adjoint products in (FC6), with those full domains, gives

VΔφeV∗=Δ∣qH,D(Δ∣qH)=D(Δ)∩qH.(FC7) V\Delta_{\varphi_e}V^*=\Delta|_{qH},\qquad D(\Delta|_{qH})=D(\Delta)\cap qH. \tag{FC7}

The same identities show qq reduces Δ\Delta. Resolvents and spectral calculus consequently restrict to qHqH, so

VΔφeitV∗=Ut∣qH V\Delta_{\varphi_e}^{it}V^*=U_t|_{qH}

on the whole Hilbert space qHqH, and all real powers have the corresponding restricted spectral domains. No statement about equality of the ambient and corner spectra is made here.

FC5. The actual restricted modular group

For a∈eMea\in eMe, fixedness of ee makes σtφ(a)∈eMe\sigma_t^\varphi(a)\in eMe. Under (FC5) and (FC7), faithful implementation gives

Vπe(σtφ(a))V∗=Ut∣qH Vπe(a)V∗ Ut∗∣qH=Vπe(σtφe(a))V∗. V\pi_e(\sigma_t^\varphi(a))V^* =U_t|_{qH}\,V\pi_e(a)V^*\,U_t^*|_{qH} =V\pi_e(\sigma_t^{\varphi_e}(a))V^*.

The corner weight is faithful normal semifinite by FC3, so MW4 applies with exactly the same generality. Faithfulness of πe\pi_e proves the claimed equality of groups for every t∈Rt\in\mathbb R.

For the balanced matrix weight of BC1, its diagonal fixed projection is EiiE_{ii}, and its restricted weight is exactly φi\varphi_i, on the whole positive cone. BC2 already identifies the part of its closed involution on coordinate (i,i)(i,i) with SφiS_{\varphi_i} and its modular operator with Δφi\Delta_{\varphi_i}. Applying the same implemented identity, or FC5 directly, gives σtΘ(xEii)=σtφi(x)Eii\sigma_t^\Theta(xE_{ii})=\sigma_t^{\varphi_i}(x)E_{ii}. This supplies BC14 at arbitrary-weight scope. More generally BC13 follows for every finite submatrix, since BC1 identifies its restricted weight with the corresponding balanced weight.

Human sources and mathematical applications

Rieffel and Van Daele, 1977, §5 and appendix provide the freely readable general Hilbert-algebra method developed in BG1–6. Hiai, arXiv:2004.02383v1, Lemmas 7.4–7.5, pp.65–67 provides the balanced diagonal spectral implementation context. Combes, Theorems 2.11 and 2.13, pp.53–57 provides the canonical-weight and reverse Hilbert-algebra context; the corresponding full finite-ideal and bounded-vector statements are proved in the GW/NF/WR/MW prerequisites above.

For the modular comparison in MG2, FC1–5 supply the fixed-corner restriction step. For balanced weights, BC1–3, FC5 supplies the exact diagonal and finite-submatrix restrictions. The all-weight intersection in MG0–4 also uses its finite-ideal filters, every-weight comparison and type III corner transport; it retains all faithful normal semifinite weights and its stated separable-predual factor scope.

Return to the periodic CAR application.