Comparing weights through their finite-energy vectors
OA-MOD-DW-01 — Conventions and the exact comparison problem
Inner products are linear in the first variable. Fix a weight φ:M+→[0,∞], and write
nφ={x:φ(x∗x)<∞},mφ=span{y∗x:x,y∈nφ}.
The construction in OA-MOD-WG supplies a positive linear extension of φ to mφ, a Hilbert space Hφ, a dense-range map Λφ, and a unital representation πφ with
⟨Λφ(x),Λφ(y)⟩=φ(y∗x),πφ(a)Λφ(x)=Λφ(ax).
In particular mφ+=mφ∩M+={a∈M+:φ(a)<∞}, and every element of mφ is a complex linear combination of this cone. These are algebraic domains, not norm closures.
Let Dφ be the cone of positive linear forms ℓ:mφ→C such that, for some finite c≥0,
0≤ℓ(a)≤cφ(a)(a∈mφ+).
Positivity refers to the inherited cone above. It does not assert that ℓ is bounded for the operator norm, or that it has already been extended to a normal weight on all of M+. The order on these forms is positivity of their difference on this cone.
OA-MOD-DW-02 — A factorization inside the algebra
Lemma. If x,y∈M and y∗y≤x∗x, there is a unique contraction v∈M which vanishes on xH⊥ and satisfies y=vx, in any faithful concrete realization M⊆B(H).
Proof. On xH define xξ↦yξ. The inequality says both that this is well defined and that its norm is at most one. Extend it continuously to xH, and set it equal to zero on the orthogonal complement. This proves existence and uniqueness in B(H). Every unitary u∈M′ preserves xH; because x,y commute with u, both v and uvu∗ have the prescribed properties. Thus v commutes with every unitary of M′. Every element of a unital C*-algebra is a linear combination of unitaries: a self-adjoint contraction b is the real part of b+i(1−b2)1/2. Consequently v∈M′′=M. This uses bounded continuous functional calculus and the bicommutant theorem, recorded foundation contracts. □
For a,b≥0 put d=a+b and p=s(d), the projection onto dH. Applying the lemma to a1/2,d1/2 and b1/2,d1/2 gives contractions v,w∈M with
a1/2=vd1/2,b1/2=wd1/2,v=vp,w=wp,v∗v+w∗w=p.
To verify the last identity, its quadratic form agrees with that of p on d1/2H, by a+b=d; continuity extends the equality to pH. Both sides vanish on (1−p)H. This argument needs neither an inverse of d nor a positive lower bound for it.
OA-MOD-DW-03 — The commutant correspondence
Theorem. There is an additive, positively homogeneous order isomorphism
πφ(M)+′⟷Dφ,T⟼ℓT,
uniquely characterized by
ℓT(y∗x)=⟨TΛφ(x),Λφ(y)⟩(x,y∈nφ).
Moreover T≤cI if and only if ℓT≤cφ on mφ+. Thus the least admissible comparison constant is ∥T∥. The assertion includes the zero Hilbert space, where both cones consist only of zero.
Proof, from forms to operators. For ℓ∈Dφ, positivity applied to (x+zy)∗(x+zy), for every z∈C, gives
∣ℓ(y∗x)∣2≤ℓ(x∗x)ℓ(y∗y)≤c2∥Λφ(x)∥2∥Λφ(y)∥2.
For completeness, when ℓ(y∗y)>0, minimize this quadratic polynomial in z; when that diagonal term is zero, varying the magnitude and phase of z forces the cross term to vanish. The inequality shows that the expression is independent of null representatives and defines a bounded sesquilinear form on Λφ(nφ). It extends uniquely to Hφ. The Hilbert-space representation theorem for bounded sesquilinear forms gives a unique positive operator T with 0≤T≤cI.
The test vectors form a dense subspace, so Tπφ(a)=πφ(a)T. This proves membership in the commutant without any appeal to a modular group.
Proof, from operators to forms. Fix T∈πφ(M)+′. For finite-weight positive a, define
fT(a)=⟨TΛφ(a1/2),Λφ(a1/2)⟩.
We must prove additivity; merely writing a formula on products would not prove independence of their decompositions. Let a,b∈mφ+, and use d,p,v,w from OA-MOD-DW-02. Here d1/2∈nφ. Put ζ=Λφ(d1/2). Then πφ(p)ζ=ζ and
Homogeneity follows from the square root of a scalar. Thus fT extends to a real linear form on mφ,sa: assign fT(a)−fT(b) to a−b. If a−b=a′−b′, the equality a+b′=a′+b and additivity prove independence. Complexification gives a positive linear form ℓT on mφ. Also
0≤fT(a)≤∥T∥φ(a).
If x∈nφ, take its polar decomposition x=u∣x∣ in M. Then Λφ(x)=πφ(u)Λφ(∣x∣), and u∗u fixes ∣x∣. Commutation with T yields
⟨TΛφ(x),Λφ(x)⟩=fT(x∗x).
Polarization now proves the formula for y∗x. Uniqueness holds because these products span mφ, and their GNS vectors are dense. The two constructions are inverse. Addition and scalar multiplication follow from the defining pairings. Finally positivity of ℓT2−ℓT1 is equivalent to nonnegativity of the quadratic form of T2−T1 on a dense subspace, hence on all of Hφ. Apply this to cI−T to obtain the bound and the optimal constant. □
OA-MOD-DW-04 — Comparing two weights and the precise target space
Theorem. Let ψ be another weight on M, with ψ≤cφ on M+ for a finite c>0. There is a unique bounded map
Cψ∣φ:Hφ⟶Hψ,Cψ∣φΛφ(x)=Λψ(x)(x∈nφ),
with norm at most c. It intertwines the representations, and
The range closure is exactly
K=Λψ(nφ)⊆Hψ.
If C=UT1/2 is its polar decomposition, then U is a unitary from s(T)Hφ onto K, and is zero on kerT. These two subspaces reduce the corresponding representations, and U intertwines their restrictions.
Proof. Domination gives nφ⊆nψ and
∥Λψ(x)∥2≤c∥Λφ(x)∥2. Consequently the displayed assignment is well defined even for nonfaithful weights and extends from a dense domain. Its range closure is K by construction. On the dense GNS domain,
Cπφ(a)Λφ(x)=Λψ(ax)=πψ(a)CΛφ(x).
Boundedness extends this identity to all vectors. Taking adjoints and using the identity for a∗ shows that C∗ intertwines in the reverse direction, so C∗C commutes with πφ(M). The pairing formula is immediate. It agrees with OA-MOD-DW-03 applied to ψ∣mφ.
The closed subspaces kerC and K are invariant under the representations and their adjoints. Therefore their orthogonal projections commute with these representations. Since T1/2 also commutes with πφ(M), the equality
Uπφ(a)T1/2ξ=πψ(a)UT1/2ξ
holds first on ranT1/2, and then by continuity on its closure s(T)Hφ. Both sides vanish on its orthogonal complement. The initial and final subspaces of a polar decomposition give the claimed unitary. □
Do not replace K by Hψ in this level of generality. If φ is zero at zero and infinite at every nonzero positive element, then nφ={0}, while every weight ψ satisfies ψ≤φ. The comparison map is zero even when Hψ=0. This also shows why restriction to a finite domain need not determine the whole weight.