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

A faithful weight on the GNS commutant

OA-MOD-OW-01. Setting, domains, and existing inputs

Let MM be an arbitrary von Neumann algebra and let φ:M+→[0,∞]\varphi:M_+\to[0,\infty] be a normal semifinite weight, not necessarily faithful. We use the density convention for semifiniteness: nφ={x∈M:φ(x∗x)<∞},mφ=span⁡{y∗x:x,y∈nφ}, \mathfrak n_\varphi=\{x\in M:\varphi(x^*x)<\infty\},\qquad \mathfrak m_\varphi=\operatorname{span}\{y^*x:x,y\in\mathfrak n_\varphi\}, and mφ\mathfrak m_\varphi is ultraweakly dense in MM. Inner products are linear in the first variable. Write (H,π,Λ)=(Hφ,πφ,Λφ),N=π(M)′⊆B(H). (H,\pi,\Lambda)=(H_\varphi,\pi_\varphi,\Lambda_\varphi), \qquad N=\pi(M)'\subseteq B(H). The map Λ\Lambda has domain exactly nφ\mathfrak n_\varphi. It need not be injective. Every bounded operator on HH has domain all of HH; membership in NN means commutation with every π(a)\pi(a), a∈Ma\in M.

The unit uses these already written course results with their own foundation status retained:

The predual norm fact used below can also be read directly from positivity: the bounded-functional Cauchy–Schwarz inequality gives ∣ω(a)∣2≤ω(a∗a)ω(1)≤∥a∥2ω(1)2|\omega(a)|^2\leq\omega(a^*a)\omega(1)\leq\|a\|^2\omega(1)^2, while evaluation at 11 gives the reverse norm inequality. Consequently norms add on positive functionals.

OA-MOD-OW-02. What a comparison operator remembers

Put Eφ={ω∈M∗+:ω≤cφ on M+ for some finite c≥0},Φφ={ω∈M∗+:ω≤φ}. \mathcal E_\varphi =\{\omega\in M_*^+:\omega\leq c\varphi \text{ on }M_+\text{ for some finite }c\geq0\}, \qquad \Phi_\varphi=\{\omega\in M_*^+:\omega\leq\varphi\}. The extended-value convention is 0⋅∞=00\cdot\infty=0; hence comparison with c=0c=0 requires ω=0\omega=0. Equivalently Eφ\mathcal E_\varphi is the union of the positive scalar multiples of Φφ\Phi_\varphi, together with zero. It is an additive cone.

By DW-03, every ω∈Eφ\omega\in\mathcal E_\varphi determines a unique hω∈N+h_\omega\in N_+ such that ω(y∗x)=⟨hωΛ(x),Λ(y)⟩(x,y∈nφ).(OW1) \omega(y^*x)=\langle h_\omega\Lambda(x),\Lambda(y)\rangle \qquad(x,y\in\mathfrak n_\varphi). \tag{OW1} This is initially a correspondence with the restriction of ω\omega to mφ\mathfrak m_\varphi. The next argument proves the stronger uniqueness needed here.

Proposition. The map ω↦hω\omega\mapsto h_\omega is additive, positively homogeneous, injective, and reflects order on Eφ\mathcal E_\varphi. Furthermore ω≤cφ⟺hω≤cIH(c≥0),(OW2) \omega\leq c\varphi\quad\Longleftrightarrow\quad h_\omega\leq cI_H \qquad(c\geq0), \tag{OW2} so the least possible comparison constant is ∥hω∥\|h_\omega\|.

Proof. Additivity and homogeneity follow from (OW1) and density of Λ(nφ)\Lambda(\mathfrak n_\varphi). Suppose hω≤hρh_\omega\leq h_\rho. For every a∈mφ+a\in\mathfrak m_\varphi^+, substitute x=y=a1/2x=y=a^{1/2} into (OW1) to obtain ω(a)≤ρ(a)\omega(a)\leq\rho(a).

To extend this inequality, let a∈M+a\in M_+ be arbitrary. The cutoffs of WG-008 satisfy 0≤eαaeα≤∥a∥eα2≤∥a∥eα. 0\leq e_\alpha a e_\alpha\leq\|a\|e_\alpha^2 \leq\|a\|e_\alpha. Thus eαaeα∈mφ+e_\alpha a e_\alpha\in\mathfrak m_\varphi^+. This norm-bounded net converges strongly and ultraweakly to aa, although it need not be increasing. Normality of the bounded functionals gives ω(a)=lim⁡αω(eαaeα)≤lim⁡αρ(eαaeα)=ρ(a). \omega(a)=\lim_\alpha\omega(e_\alpha a e_\alpha) \leq\lim_\alpha\rho(e_\alpha a e_\alpha)=\rho(a). Hence the map reflects order. The forward order implication follows directly from quadratic forms in (OW1). Applying order reflection in both directions proves injectivity.

The forward implication of (OW2) is the operator bound in DW-03. Conversely, hω≤cIHh_\omega\leq cI_H gives ω(a)≤cφ(a)\omega(a)\leq c\varphi(a) whenever φ(a)<∞\varphi(a)<\infty, by taking x=a1/2x=a^{1/2}. For c>0c>0, the remaining infinite-weight values impose no restriction. For c=0c=0, one has hω=0=h0h_\omega=0=h_0, so injectivity gives ω=0\omega=0; the required global inequality then holds with the stated convention. This proves (OW2), including the zero constant. The norm of a positive operator is its least scalar upper bound. ∎

There are two different norms in this construction. The operator norm ∥hω∥\|h_\omega\| measures how strongly ω\omega is dominated by φ\varphi. The functional norm ∥ω∥=ω(1)\|\omega\|=\omega(1) will be the value of the opposite weight at hωh_\omega. They are generally unequal.

OA-MOD-OW-03. The vector implementing a finite observation

Let (Hω,πω,Ωω)(H_\omega,\pi_\omega,\Omega_\omega) be the bounded-functional GNS construction for ω∈Eφ\omega\in\mathcal E_\varphi. The vector has norm ∥Ωω∥2=ω(1)\|\Omega_\omega\|^2=\omega(1), including H0={0}H_0=\{0\}. DW-04 supplies the bounded map Cω:H→Hω,CωΛ(x)=πω(x)Ωω,x∈nφ, C_\omega:H\to H_\omega, \qquad C_\omega\Lambda(x)=\pi_\omega(x)\Omega_\omega, \quad x\in\mathfrak n_\varphi, with Cω∗Cω=hωC_\omega^*C_\omega=h_\omega and Cωπ(a)=πω(a)CωC_\omega\pi(a)=\pi_\omega(a)C_\omega for a∈Ma\in M.

Proposition. The range of CωC_\omega is dense in HωH_\omega. If Cω=Uωhω1/2,ηω=Uω∗Ωω, C_\omega=U_\omega h_\omega^{1/2},\qquad \eta_\omega=U_\omega^*\Omega_\omega, then ∥ηω∥2=∥ω∥,hω1/2Λ(x)=π(x)ηω(x∈nφ).(OW3) \|\eta_\omega\|^2=\|\omega\|,\qquad h_\omega^{1/2}\Lambda(x)=\pi(x)\eta_\omega \quad(x\in\mathfrak n_\varphi). \tag{OW3} The vector ηω\eta_\omega is the unique vector satisfying the second identity. It also implements the whole functional: ω(a)=⟨π(a)ηω,ηω⟩(a∈M).(OW4) \omega(a)=\langle\pi(a)\eta_\omega,\eta_\omega\rangle \qquad(a\in M). \tag{OW4}

Proof. Every cutoff eαe_\alpha belongs to nφ\mathfrak n_\varphi. Since ω\omega is normal, ∥πω(eα)Ωω−Ωω∥2=ω((1−eα)2)≤ω(1−eα)⟶0. \|\pi_\omega(e_\alpha)\Omega_\omega-\Omega_\omega\|^2 =\omega((1-e_\alpha)^2) \leq\omega(1-e_\alpha)\longrightarrow0. Thus Ωω\Omega_\omega belongs to the closure of the range of CωC_\omega. That closed subspace is invariant under πω(M)\pi_\omega(M) by the intertwining identity. Since Ωω\Omega_\omega is cyclic, the subspace is all of HωH_\omega.

Consequently the polar partial isometry satisfies UωUω∗=IHωU_\omega U_\omega^*=I_{H_\omega}; it is not asserted that Uω∗Uω=IHU_\omega^*U_\omega=I_H. The latter is s(hω)s(h_\omega). Polar intertwining from DW-04 gives Uω∗πω(a)=π(a)Uω∗. U_\omega^*\pi_\omega(a)=\pi(a)U_\omega^*. It follows that π(x)ηω=Uω∗πω(x)Ωω=Uω∗CωΛ(x)=s(hω)hω1/2Λ(x)=hω1/2Λ(x). \begin{aligned} \pi(x)\eta_\omega &=U_\omega^*\pi_\omega(x)\Omega_\omega =U_\omega^*C_\omega\Lambda(x)\\ &=s(h_\omega)h_\omega^{1/2}\Lambda(x) =h_\omega^{1/2}\Lambda(x). \end{aligned} Also ∥ηω∥2=⟨UωUω∗Ωω,Ωω⟩=∥Ωω∥2\|\eta_\omega\|^2 =\langle U_\omega U_\omega^*\Omega_\omega,\Omega_\omega\rangle =\|\Omega_\omega\|^2. The same coisometry and intertwining identities give (OW4): ⟨π(a)Uω∗Ωω,Uω∗Ωω⟩=⟨πω(a)Ωω,Ωω⟩. \langle\pi(a)U_\omega^*\Omega_\omega,U_\omega^*\Omega_\omega\rangle =\langle\pi_\omega(a)\Omega_\omega,\Omega_\omega\rangle.

For uniqueness, if π(x)η=π(x)ζ\pi(x)\eta=\pi(x)\zeta for all x∈nφx\in\mathfrak n_\varphi, apply it to eαe_\alpha. Since π\pi is normal, π(eα)↑IH\pi(e_\alpha)\uparrow I_H strongly. Passing to this strong limit gives η=ζ\eta=\zeta. ∎

The density of CωC_\omega's range is where semifiniteness repairs a potential loss of norm. Without that density, the polar factor only gives ∥Uω∗Ωω∥≤∥Ωω∥\|U_\omega^*\Omega_\omega\|\leq\|\Omega_\omega\|, which is insufficient to define a weight by ∥ω∥\|\omega\|.

OA-MOD-OW-04. A hereditary cone in the commutant

Define Pφ={hω:ω∈Eφ}⊆N+. P_\varphi=\{h_\omega:\omega\in\mathcal E_\varphi\}\subseteq N_+.

Proposition. This is an additive hereditary cone: if 0≤h≤k0\leq h\leq k and k∈Pφk\in P_\varphi, then h∈Pφh\in P_\varphi. The assignment f(hω)=∥ω∥ f(h_\omega)=\|\omega\| is well-defined, additive, positively homogeneous, and increasing on PφP_\varphi.

Proof. Injectivity in OW-02 makes ff well-defined. Additivity and positive homogeneity of the correspondence, together with ∥ω∥=ω(1)\|\omega\|=\omega(1), prove the analogous properties of ff.

Suppose 0≤h≤hω0\leq h\leq h_\omega. Apply the bounded factorization DW-02 inside the von Neumann algebra NN to obtain a contraction s∈Ns\in N such that h1/2=shω1/2. h^{1/2}=s h_\omega^{1/2}. Define the positive functional ρ(a)=⟨π(a)sηω,sηω⟩(a∈M). \rho(a)=\langle\pi(a)s\eta_\omega,s\eta_\omega\rangle \qquad(a\in M). It is bounded, and normality of π\pi makes it normal. For x∈nφx\in\mathfrak n_\varphi, commutation with ss and (OW3) give ρ(x∗x)=∥π(x)sηω∥2=∥shω1/2Λ(x)∥2=∥h1/2Λ(x)∥2≤∥h∥φ(x∗x). \begin{aligned} \rho(x^*x) &=\|\pi(x)s\eta_\omega\|^2 =\|s h_\omega^{1/2}\Lambda(x)\|^2\\ &=\|h^{1/2}\Lambda(x)\|^2 \leq\|h\|\varphi(x^*x). \end{aligned} Testing finite-weight a≥0a\geq0 with x=a1/2x=a^{1/2}, and taking any strictly positive constant at least ∥h∥\|h\| for the remaining infinite values, proves ρ∈Eφ\rho\in\mathcal E_\varphi. Polarizing the equality of quadratic forms proves hρ=hh_\rho=h. Thus PφP_\varphi is hereditary.

Finally OW-02 reflects order, so hρ≤hωh_\rho\leq h_\omega implies ρ≤ω\rho\leq\omega, and evaluation at 11 gives f(hρ)≤f(hω)f(h_\rho)\leq f(h_\omega). Alternatively the construction gives ∥ρ∥=∥sηω∥2≤∥ω∥\|\rho\|=\|s\eta_\omega\|^2\leq\|\omega\|. ∎

This proof constructs the functional for a smaller commutant operator. An assertion that positive operators have square roots would not by itself prove that the finite cone is hereditary.

Editable source · Proof dependencies and component terms