Original text: CC0 1.0. Prerequisite proofs and component terms.

Spatial comparison on an arbitrary representation

OA-MOD-SD-01. Spaces, weights and dependency contracts

Fix a concrete unital von Neumann algebra M⊆B(H)M\subseteq B(H), with commutant M′M'. The Hilbert space HH is arbitrary. Fix a normal semifinite faithful weight ψ′\psi' on M′M'. A numerator φ\varphi, when present, is normal and semifinite on MM; it need not be faithful.

Inner products are linear in the first variable. Use the semicyclic triple

(Hψ′,πψ′,Λψ′),nψ′={x′∈M′:ψ′(x′∗x′)<∞}. (H_{\psi'},\pi_{\psi'},\Lambda_{\psi'}),\qquad \mathfrak n_{\psi'}=\{x'\in M':\psi'(x'^*x')<\infty\}.

Thus Λψ′\Lambda_{\psi'} maps the finite left ideal into its Hilbert completion, and

πψ′(y′)Λψ′(x′)=Λψ′(y′x′),∥Λψ′(x′)∥2=ψ′(x′∗x′). \pi_{\psi'}(y')\Lambda_{\psi'}(x') =\Lambda_{\psi'}(y'x'),\qquad \|\Lambda_{\psi'}(x')\|^2=\psi'(x'^*x').

The following are exact dependencies, with their current status visible.

The bounded-vector proofs below use only SD-DEP-GNS and SD-DEP-BOUNDED. They do not assume SD-DEP-STANDARD or any spatial derivative already exists. References to OA-MOD-SC record downstream proofs of later contracts: SC uses SD-02, SD-04 and SD-05, while those three proofs have no SC premise. The direct construction does not use SD-DEP-STANDARD or SD-DEP-RELATIVE; those contracts govern the separate convention dictionary and relative modular identification.

OA-MOD-SD-02. Which vectors define bounded intertwiners?

Define

Dψ′(H)={ξ∈H: there is C≥0 such that∥x′ξ∥≤C ψ′(x′∗x′)1/2(x′∈nψ′)}. \begin{aligned} \mathcal D_{\psi'}(H)=\{\xi\in H:\ &\text{there is }C\geq0\text{ such that}\\ &\|x'\xi\|\leq C\,\psi'(x'^*x')^{1/2} \quad(x'\in\mathfrak n_{\psi'})\}. \end{aligned}

For ξ\xi in this set, define on Λψ′(nψ′)\Lambda_{\psi'}(\mathfrak n_{\psi'})

Rψ′(ξ)Λψ′(x′)=x′ξ. R_{\psi'}(\xi)\Lambda_{\psi'}(x')=x'\xi.

Proposition. This formula extends uniquely to a bounded linear map

Rψ′(ξ):Hψ′⟶H. R_{\psi'}(\xi):H_{\psi'}\longrightarrow H.

Its norm is the least permissible CC. The space Dψ′(H)\mathcal D_{\psi'}(H) is linear, ξ↦Rψ′(ξ)\xi\mapsto R_{\psi'}(\xi) is linear and injective, and

Rψ′(ξ)πψ′(y′)=y′Rψ′(ξ)(y′∈M′). R_{\psi'}(\xi)\pi_{\psi'}(y')=y'R_{\psi'}(\xi) \quad(y'\in M').

For a∈Ma\in M, the vector aξa\xi belongs to Dψ′(H)\mathcal D_{\psi'}(H), and

Rψ′(aξ)=aRψ′(ξ),∥Rψ′(aξ)∥≤∥a∥ ∥Rψ′(ξ)∥. R_{\psi'}(a\xi)=aR_{\psi'}(\xi),\qquad \|R_{\psi'}(a\xi)\|\leq\|a\|\,\|R_{\psi'}(\xi)\|.

Proof. The defining estimate says exactly that the displayed rule is bounded for the GNS norm. If two representatives have the same GNS vector, their difference has GNS norm zero, so the estimate makes their images equal. The rule is therefore well-defined, and extension from a dense subspace proves existence, uniqueness and the norm assertion. The triangle inequality proves linearity of the bounded-vector domain, and uniqueness of bounded extension proves linearity of Rψ′R_{\psi'}.

For x′∈nψ′x'\in\mathfrak n_{\psi'} and y′∈M′y'\in M', the left-ideal property gives y′x′∈nψ′y'x'\in\mathfrak n_{\psi'}. On the dense GNS domain,

Rψ′(ξ)πψ′(y′)Λψ′(x′)=y′x′ξ=y′Rψ′(ξ)Λψ′(x′). R_{\psi'}(\xi)\pi_{\psi'}(y')\Lambda_{\psi'}(x') =y'x'\xi =y'R_{\psi'}(\xi)\Lambda_{\psi'}(x').

Both sides are bounded operators, proving the intertwining identity. If a∈Ma\in M, then x′aξ=ax′ξx'a\xi=ax'\xi, so the defining estimate holds for aξa\xi with constant ∥a∥∥Rψ′(ξ)∥\|a\|\|R_{\psi'}(\xi)\|. The same dense-domain computation proves the formula for Rψ′(aξ)R_{\psi'}(a\xi).

Finally, suppose Rψ′(ξ)=0R_{\psi'}(\xi)=0. The finite positive contractions ei′e_i' supplied by OA-MOD-WG-008 converge strongly to 11. They belong to nψ′\mathfrak n_{\psi'}, because (ei′)2≤ei′(e_i')^2\leq e_i' and ψ′(ei′)<∞\psi'(e_i')<\infty. Hence

ei′ξ=Rψ′(ξ)Λψ′(ei′)=0. e_i'\xi=R_{\psi'}(\xi)\Lambda_{\psi'}(e_i')=0.

Strong convergence gives ξ=0\xi=0. □\square

This proposition does not prove that Dψ′(H)\mathcal D_{\psi'}(H) is dense in HH. The finite cutoffs act on a bounded vector to detect it; the argument does not show that applying a cutoff to an arbitrary vector makes that vector bounded.

OA-MOD-SD-03. A useful complete norm on the bounded vectors

Proposition. The formula

∥ξ∥b=(∥ξ∥2+∥Rψ′(ξ)∥2)1/2 \|\xi\|_{\mathrm b} =\bigl(\|\xi\|^2+\|R_{\psi'}(\xi)\|^2\bigr)^{1/2}

makes Dψ′(H)\mathcal D_{\psi'}(H) a Banach space. For a∈Ma\in M, multiplication by aa has norm at most ∥a∥\|a\| on this space.

Proof. This is the norm induced by the linear graph embedding

ξ⟼(ξ,Rψ′(ξ))∈H⊕B(Hψ′,H), \xi\longmapsto (\xi,R_{\psi'}(\xi)) \in H\oplus B(H_{\psi'},H),

where the direct sum has the square-sum Banach norm. Let (ξn)(\xi_n) be Cauchy in that norm. There are ξ∈H\xi\in H and T∈B(Hψ′,H)T\in B(H_{\psi'},H) with ξn→ξ\xi_n\to\xi in Hilbert norm and Rψ′(ξn)→TR_{\psi'}(\xi_n)\to T in operator norm. For every x′∈nψ′x'\in\mathfrak n_{\psi'},

TΛψ′(x′)=lim⁡nx′ξn=x′ξ. T\Lambda_{\psi'}(x') =\lim_n x'\xi_n=x'\xi.

Thus ξ\xi satisfies the bounded-vector estimate with constant ∥T∥\|T\|, and uniqueness gives Rψ′(ξ)=TR_{\psi'}(\xi)=T. The graph is closed, proving completeness. The multiplication estimate follows by applying the two bounds in OA-MOD-SD-02 to the two terms of the norm. □\square

Completeness here concerns the bounded-vector norm, not the Hilbert norm. It gives no assertion that this domain is Hilbert-norm closed or dense.

OA-MOD-SD-04. Coefficients, matrices and the coefficient ideal

For bounded vectors define

θψ′(ξ,η)=Rψ′(ξ)Rψ′(η)∗∈B(H). \theta_{\psi'}(\xi,\eta) =R_{\psi'}(\xi)R_{\psi'}(\eta)^*\in B(H).

Proposition. These coefficients belong to MM, are linear in ξ\xi and conjugate-linear in η\eta, and satisfy

θ(ξ,η)∗=θ(η,ξ),θ(aξ,bη)=aθ(ξ,η)b∗,θ(ξ,ξ)≥0,∥θ(ξ,η)∥≤∥R(ξ)∥ ∥R(η)∥. \begin{gathered} \theta(\xi,\eta)^*=\theta(\eta,\xi),\qquad \theta(a\xi,b\eta)=a\theta(\xi,\eta)b^*,\\ \theta(\xi,\xi)\geq0,\qquad \|\theta(\xi,\eta)\|\leq\|R(\xi)\|\,\|R(\eta)\|. \end{gathered}

For every finite family ξ1,…,ξn\xi_1,\ldots,\xi_n, the matrix

[θ(ξi,ξj)]i,j=1n [\theta(\xi_i,\xi_j)]_{i,j=1}^n

is positive in Mn(M)M_n(M). Consequently

Jψ′=span⁡{θ(ξ,η):ξ,η∈Dψ′(H)} \mathcal J_{\psi'}= \operatorname{span}\{\theta(\xi,\eta):\xi,\eta\in\mathcal D_{\psi'}(H)\}

is an algebraic two-sided *-ideal of MM, and its positive cone is exactly

Jψ′∩M+={∑j=1nθ(ξj,ξj):n≥1, ξj∈Dψ′(H)}. \mathcal J_{\psi'}\cap M_+ =\left\{\sum_{j=1}^n\theta(\xi_j,\xi_j): n\geq1,\ \xi_j\in\mathcal D_{\psi'}(H)\right\}.

The zero operator is included by taking a zero vector.

Proof. Taking adjoints of the intertwining formula with y′∗y'^* gives

πψ′(y′)R(η)∗=R(η)∗y′. \pi_{\psi'}(y')R(\eta)^*=R(\eta)^*y'.

Therefore y′R(ξ)R(η)∗=R(ξ)R(η)∗y′y'R(\xi)R(\eta)^*=R(\xi)R(\eta)^*y' for every y′∈M′y'\in M', and the bicommutant theorem puts the coefficient in MM. All scalar, adjoint and covariance identities follow from the corresponding identities for RR.

For v1,…,vn∈Hv_1,\ldots,v_n\in H,

∑i,j⟨θ(ξi,ξj)vj,vi⟩=∥∑jR(ξj)∗vj∥2≥0. \sum_{i,j}\langle\theta(\xi_i,\xi_j)v_j,v_i\rangle =\left\|\sum_j R(\xi_j)^*v_j\right\|^2\geq0.

This proves matrix positivity. Covariance under a,b∈Ma,b\in M and the adjoint identity make Jψ′\mathcal J_{\psi'} a two-sided *-ideal.

It remains to prove the assertion about its positive cone; polarization alone does not prove positivity of the coefficients in a chosen linear expansion. Let z∈Jψ′∩M+z\in\mathcal J_{\psi'}\cap M_+. Absorb scalar coefficients into the first vectors and write z=∑j=1nθ(ξj,ηj)z=\sum_{j=1}^n\theta(\xi_j,\eta_j). Since z=z∗z=z^*, positivity of θ(ξj−ηj,ξj−ηj)\theta(\xi_j-\eta_j,\xi_j-\eta_j) gives

0≤z≤P:=12∑j=1n(θ(ξj,ξj)+θ(ηj,ηj)). 0\leq z\leq P:=\frac12\sum_{j=1}^n \bigl(\theta(\xi_j,\xi_j)+\theta(\eta_j,\eta_j)\bigr).

Define a contraction on ran⁡P1/2\operatorname{ran}P^{1/2} by

T(P1/2v)=z1/2v. T(P^{1/2}v)=z^{1/2}v.

The inequality z≤Pz\leq P proves that this is well-defined and contractive. Extend it continuously to ran⁡P‾\overline{\operatorname{ran}P} and set it to zero on ker⁡P\ker P. For every unitary u′∈M′u'\in M', both P1/2P^{1/2} and z1/2z^{1/2} commute with u′u'. On ran⁡P1/2\operatorname{ran}P^{1/2}, this implies Tu′=u′TTu'=u'T; the range closure and kernel of PP are also u′u'-invariant. Thus the identity holds on all of HH. Every element of M′M' is a linear combination of unitaries, so T∈MT\in M.

We have TP1/2=z1/2TP^{1/2}=z^{1/2}, hence TPT∗=zTPT^*=z. Substituting the definition of PP and using covariance gives

z=∑j=1n[θ(Tξj2,Tξj2)+θ(Tηj2,Tηj2)]. z=\sum_{j=1}^n \left[ \theta\left(\frac{T\xi_j}{\sqrt2},\frac{T\xi_j}{\sqrt2}\right) +\theta\left(\frac{T\eta_j}{\sqrt2},\frac{T\eta_j}{\sqrt2}\right) \right].

All new vectors are bounded by OA-MOD-SD-02. This proves the nontrivial inclusion; the other inclusion follows from positivity. □\square

The proof has not asserted ultraweak density of Jψ′\mathcal J_{\psi'}. OA-MOD-SC-04 proves that density and supplies increasing positive approximants, using this algebraic ideal result.

OA-MOD-SD-05. The energy formula before closure

Let φ\varphi be the numerator weight. Define the finite initial domain

Eφ,ψ′={ξ∈Dψ′(H):φ(θψ′(ξ,ξ))<∞}. \mathcal E_{\varphi,\psi'} =\{\xi\in\mathcal D_{\psi'}(H): \varphi(\theta_{\psi'}(\xi,\xi))<\infty\}.

Proposition. This is a linear subspace. For ξ,η\xi,\eta in it, θψ′(ξ,η)\theta_{\psi'}(\xi,\eta) belongs to mφ\mathfrak m_\varphi, and

q0(ξ,η)=φ~(θψ′(ξ,η)) q_0(\xi,\eta) =\widetilde\varphi(\theta_{\psi'}(\xi,\eta))

is a nonnegative sesquilinear form, with

q0[ξ]=φ(θψ′(ξ,ξ)). q_0[\xi]=\varphi(\theta_{\psi'}(\xi,\xi)).

Here φ~\widetilde\varphi is the finite linear extension proved in OA-MOD-WG-004. No infinite subtraction occurs.

Proof. Expanding the positive operator (R(ξ)−R(η))(R(ξ)−R(η))∗(R(\xi)-R(\eta))(R(\xi)-R(\eta))^* gives

θ(ξ+η,ξ+η)≤2θ(ξ,ξ)+2θ(η,η). \theta(\xi+\eta,\xi+\eta) \leq2\theta(\xi,\xi)+2\theta(\eta,\eta).

Monotonicity and additivity of the weight prove closure of the finite domain under sums; scalar closure follows from homogeneity. Polarization, with the first-variable-linear convention, gives

θ(ξ,η)=14∑k=03ikθ(ξ+ikη,ξ+ikη). \theta(\xi,\eta) =\frac14\sum_{k=0}^3 i^k \theta(\xi+i^k\eta,\xi+i^k\eta).

Every diagonal on the right is finite for φ\varphi. It therefore belongs to the finite positive cone of mφ\mathfrak m_\varphi. Its linear span is mφ\mathfrak m_\varphi, proving membership of the cross coefficient. Composing its sesquilinearity with the linear extension of φ\varphi proves sesquilinearity of q0q_0; the diagonal is visibly nonnegative. □\square

The preceding algebraic argument alone does not prove density of Eφ,ψ′\mathcal E_{\varphi,\psi'}, closability of q0q_0, or completeness in its form norm. OA-MOD-SC-05 through OA-MOD-SC-07 prove density and closability and construct the closed completion. The representation theorem applies to that completion; the initial form need not itself be closed.

Although Dψ′(H)\mathcal D_{\psi'}(H) is invariant under MM, no such invariance of Eφ,ψ′\mathcal E_{\varphi,\psi'} is asserted. A general weight does not satisfy φ(aza∗)≤∥a∥2φ(z)\varphi(a z a^*)\leq\|a\|^2\varphi(z).

Editable source · Proof dependencies and component terms