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

Whole bounded-vector spaces between the graph theorem and weights

Complete local proof by GPT-6.1 Sol (OpenAI), Ultra, 2026-10-07. Any rights in this new exposition and its original diagram are dedicated under CC0-1.0. Self-checked by the writing AI.

Let AA be a left Hilbert algebra on an arbitrary complex Hilbert space HH. Inner products are linear in the first variable. Its multiplication operators are LaL_a, its closed involution is SS, and F=S∗F=S^*. Set M=L(A)′′M=L(A)''. The complete bounded-graph proof BG1–6 supplies its dual D=A′D=A', fullification C=A′′C=A'', the graph cores A2,C2A^2,C^2, and

R(D)′′=M′,C′=D,D=Br∩D(F),C=Bl∩D(S),S=JΔ1/2,F=JΔ−1/2.(BW0) R(D)''=M',\qquad C'=D,\qquad D=B_r\cap D(F),\quad C=B_l\cap D(S),\qquad S=J\Delta^{1/2},\quad F=J\Delta^{-1/2}. \tag{BW0}

Here Rηa=LaηR_\eta a=L_a\eta on AA. For the right-algebra product we use

ηζ=Rζη,Rηζ=RζRη. \eta\zeta=R_\zeta\eta, \qquad R_{\eta\zeta}=R_\zeta R_\eta.

Thus the ordinary left dual in BG becomes this right algebra by taking its opposite product; the operators and involution FF are unchanged. The real-subspace graph proof, its coercivity correction and every closed domain are proved in BG. Its freely accessible primary method is Rieffel–Van Daele, §5 and appendix. The elementary multiplier identities below can also be compared with Combes, Lemmas 2.2–2.5, printed pp.51–52. No referenced book lemma, affiliated unbounded multiplier theorem or closed-form representation theorem is an input. The other inputs are bounded continuous calculus and Hilbert projection and positive contraction density BD3.

The whole bounded-vector spaces and their multiplication ideals, with their exact duality and covariance maps

The diagram distinguishes the whole spaces Bl,BrB_l,B_r from their involution-domain intersections C,DC,D. The J arrow acts on the whole bounded-vector spaces. The lower identity tests arbitrary ξ∈Bl,η∈Br\xi\in B_l,\eta\in B_r; neither vector needs to lie in both graph domains. Proof locators are BW1–BW4. The SVG is the editable diagram source.

BW1. Both whole multiplication ideals and their vector maps

Call η\eta right bounded when a↦Laηa\mapsto L_a\eta on AA is bounded in ∥a∥H\|a\|_H, and write BrB_r for all such vectors. Associativity, tested on a,b∈Aa,b\in A, gives RηLb=LbRηR_\eta L_b=L_bR_\eta; testing b#b^\# as well gives Rη∈M′R_\eta\in M'. If Rη=0R_\eta=0, every Laη=0L_a\eta=0. Nondegeneracy of L(A)L(A), proved from density of A2A^2 in BG1, makes η=0\eta=0. For x′∈M′x'\in M', commutation gives

x′η∈Br,Rx′η=x′Rη.(BW1) x'\eta\in B_r,\qquad R_{x'\eta}=x'R_\eta. \tag{BW1}

Thus Ir={Rη:η∈Br}I_r=\{R_\eta:\eta\in B_r\} is a linear left ideal in M′M', with injective multiplier map and inverse vector map θr(Rη)=η\theta_r(R_\eta)=\eta.

Define BlB_l by boundedness of d↦Rdξd\mapsto R_d\xi on the entire dense algebra DD, and denote its extension by λξ\lambda_\xi. For d,e∈Dd,e\in D, the right product convention gives

λξRed=Rdeξ=ReRdξ=Reλξd. \lambda_\xi R_e d =R_{de}\xi=R_eR_d\xi=R_e\lambda_\xi d.

Density and boundedness put λξ∈R(D)′=M\lambda_\xi\in R(D)'=M. If λξ=0\lambda_\xi=0, then every Rdξ=0R_d\xi=0; the nondegenerate dual representation in BG5 makes ξ=0\xi=0. For x∈Mx\in M, commute xx through each RdR_d to obtain

xξ∈Bl,λxξ=xλξ.(BW2) x\xi\in B_l,\qquad\lambda_{x\xi}=x\lambda_\xi. \tag{BW2}

Consequently Il={λξ:ξ∈Bl}I_l=\{\lambda_\xi:\xi\in B_l\} is a linear left ideal in MM, with inverse vector map θl(λξ)=ξ\theta_l(\lambda_\xi)=\xi and θl(xa)=xθl(a)\theta_l(xa)=x\theta_l(a).

BW2. Adjoint products and the exact adjoint intersection

For arbitrary η,ζ∈Br\eta,\zeta\in B_r, (BW1) already makes Rη∗ζR_\eta^*\zeta right bounded, with multiplier Rη∗RζR_\eta^*R_\zeta. For a∈Aa\in A,

⟨Rη∗ζ,Sa⟩=⟨ζ,La#η⟩=⟨Laζ,η⟩=⟨a,Rζ∗η⟩.(BW3) \langle R_\eta^*\zeta,Sa\rangle =\langle\zeta,L_{a^\#}\eta\rangle =\langle L_a\zeta,\eta\rangle =\langle a,R_\zeta^*\eta\rangle. \tag{BW3}

Because AA is a whole graph core for SS, this extends to every a∈D(S)a\in D(S). The conjugate-linear adjoint criterion then proves

Rη∗ζ∈D,F(Rη∗ζ)=Rζ∗η,Ir∗Ir⊂R(D).(BW4) R_\eta^*\zeta\in D, \qquad F(R_\eta^*\zeta)=R_\zeta^*\eta, \qquad I_r^*I_r\subset R(D). \tag{BW4}

If η∈D\eta\in D, the dual definition in BG6 gives Rη∗=RFηR_\eta^*=R_{F\eta}. Conversely suppose Rη∗=RζR_\eta^*=R_\zeta for η,ζ∈Br\eta,\zeta\in B_r. For a,b∈Aa,b\in A,

⟨η,a#b⟩=⟨Rηa,b⟩=⟨a,Rζb⟩=⟨b#a,ζ⟩.(BW5) \langle\eta,a^\#b\rangle =\langle R_\eta a,b\rangle =\langle a,R_\zeta b\rangle =\langle b^\#a,\zeta\rangle. \tag{BW5}

Products span A2A^2, which is a whole graph core for SS by BG6. Taking conjugates in (BW5) and extending in graph norm is precisely the adjoint criterion for η∈D(F)\eta\in D(F), with Fη=ζF\eta=\zeta. Hence

Ir∩Ir∗=R(D),θr(Rη∗)=Fη(η∈D).(BW6) I_r\cap I_r^*=R(D),\qquad \theta_r(R_\eta^*)=F\eta\quad(\eta\in D). \tag{BW6}

Apply this same argument to the opposite of DD. BG6 identifies its closed involution as FF, its adjoint as SS, and its dual as CC. Its whole bounded-vector test is exactly the definition of BlB_l. Thus

Il∩Il∗=λ(C),θl(λξ∗)=Sξ(ξ∈C).(BW7) I_l\cap I_l^*=\lambda(C),\qquad \theta_l(\lambda_\xi^*)=S\xi\quad(\xi\in C). \tag{BW7}

These equalities concern the whole multiplication ideals, including their zero elements, and the complete closed domains. If AA is already full, C=AC=A.

BW3. Mixed multiplication without a graph-domain assumption

Choose positive contractions ci∈R(D)c_i\in R(D) tending strongly to 11, using BD3 and the nondegenerate representation proved in BG5. Each is Rdi=Rdi∗R_{d_i}=R_{d_i}^* for a unique di∈Dd_i\in D. For an arbitrary η∈Br\eta\in B_r, (BW4) puts

ηi=ciη=Rdi∗η∈D,Rηi=ciRη. \eta_i=c_i\eta=R_{d_i}^*\eta\in D, \qquad R_{\eta_i}=c_iR_\eta.

For every ξ∈Bl\xi\in B_l, the defining equality on DD is λξηi=Rηiξ\lambda_\xi\eta_i=R_{\eta_i}\xi. Both sides converge in HH, since λξ\lambda_\xi is bounded and ci→1c_i\to1 strongly. Therefore

λξη=Rηξ(ξ∈Bl, η∈Br).(BW8) \lambda_\xi\eta=R_\eta\xi \qquad(\xi\in B_l,\ \eta\in B_r). \tag{BW8}

In particular passing from AA to C=A′′C=A'' leaves BrB_r unchanged: if η\eta passes the original boundedness test, then for a∈Ca\in C (BW8) gives Laη=RηaL_a\eta=R_\eta a, with the same bound and operator. The converse restricts the test from CC to AA. The dual is also unchanged, since C′=DC'=D, so BlB_l is unchanged. Thus fullification preserves both complete bounded-vector spaces, their multiplier maps, and the closed graphs S,FS,F, not just their algebra intersections.

BW4. Modular maps on all bounded vectors

Set Ut=ΔitU_t=\Delta^{it}. BG6 proves UtC=CU_tC=C, UtD=DU_tD=D and

LUta=UtLaU−t,RUtd=UtRdU−t,RJa=JLaJ,LJd=JRdJ(a∈C,d∈D).(BW9) L_{U_ta}=U_tL_aU_{-t},\qquad R_{U_td}=U_tR_dU_{-t},\qquad R_{Ja}=JL_aJ,\quad L_{Jd}=JR_dJ \quad(a\in C,d\in D). \tag{BW9}

For ξ∈Bl\xi\in B_l and d∈Dd\in D, use the second equality to write

RdUtξ=UtRU−tdξ=UtλξU−td. R_dU_t\xi=U_tR_{U_{-t}d}\xi =U_t\lambda_\xi U_{-t}d.

This is the whole defining boundedness test for BlB_l, and its bound is ∥λξ∥\|\lambda_\xi\|. For η∈Br\eta\in B_r, test similarly on the entire algebra CC, which BW3 permits:

LaUtη=UtLU−taη=UtRηU−ta(a∈C). L_aU_t\eta=U_tL_{U_{-t}a}\eta =U_tR_\eta U_{-t}a\quad(a\in C).

Applying the same arguments with −t-t proves the onto equalities

UtBl=Bl,UtBr=Br,λUtξ=UtλξU−t,RUtη=UtRηU−t.(BW10) U_tB_l=B_l,\quad U_tB_r=B_r,\qquad \lambda_{U_t\xi}=U_t\lambda_\xi U_{-t},\quad R_{U_t\eta}=U_tR_\eta U_{-t}. \tag{BW10}

The extension of JJ also uses only whole bounded tests. For a∈Ca\in C and ξ∈Bl\xi\in B_l, (BW8)–(BW9) give

LaJξ=JRJaξ=Jλξ(Ja)=(JλξJ)a. L_aJ\xi=JR_{Ja}\xi =J\lambda_\xi(Ja)=(J\lambda_\xi J)a.

For d∈Dd\in D and η∈Br\eta\in B_r, they give

RdJη=JLJdη=JRη(Jd)=(JRηJ)d. R_dJ\eta=JL_{Jd}\eta =JR_\eta(Jd)=(JR_\eta J)d.

Consequently

JBl=Br,JBr=Bl,RJξ=JλξJ,λJη=JRηJ.(BW11) JB_l=B_r,\quad JB_r=B_l,\qquad R_{J\xi}=J\lambda_\xi J,\quad \lambda_{J\eta}=JR_\eta J. \tag{BW11}

Here the two compositions containing JJ are bounded linear operators. No vector in BW10–BW11 is required to be in D(S)∩D(F)D(S)\cap D(F). Together with (BW2), these are the entire finite-ideal/vector-map covariance contracts used in the weight construction and its opposite.

BW5. The bounded polar column used for additivity

Let a,b∈M+a,b\in M_+, X:H→H⊕HX:H\to H\oplus H be Xv=(a1/2v,b1/2v)Xv=(a^{1/2}v,b^{1/2}v), and c=(a+b)1/2c=(a+b)^{1/2}. The rule cv↦Xvcv\mapsto Xv is a well-defined isometry, since ∥Xv∥2=∥cv∥2\|Xv\|^2=\|cv\|^2. Extend it on cH‾\overline{cH} and set it to zero on ker⁡c\ker c. This gives a bounded partial isometry V:H→H⊕HV:H\to H\oplus H. For every unitary u∈M′u\in M', (u⊕u)X=Xu(u\oplus u)X=Xu and uc=cuuc=cu; the range-isometry rule, its continuous extension and its kernel summand therefore imply (u⊕u)V=Vu(u\oplus u)V=Vu. Writing V=(v1,v2)TV=(v_1,v_2)^T, each vjv_j commutes with M′M' and thus belongs to M=M′′M=M''. We have proved directly

X=Vc,V∗X=c,V∗V=s(c),VV∗=Q=[XH],QX=X.(BW12) X=Vc,\quad V^*X=c,\quad V^*V=s(c),\quad VV^*=Q=[XH],\quad QX=X. \tag{BW12}

The same range-isometry construction for any bounded x∈Mx\in M, using h=(x∗x)1/2h=(x^*x)^{1/2}, proves x=uhx=uh with u∈Mu\in M, u∗u=[x∗H]u^*u=[x^*H] and uu∗=[xH]uu^*=[xH]. It covers zero and noninjective xx. Only bounded square roots and Hilbert completion are used, so no theorem representing a closed unbounded quadratic form occurs in this step.

For the covariant map θl:Il→H\theta_l:I_l\to H in BW1, suppose a1/2,b1/2∈Ila^{1/2},b^{1/2}\in I_l and put α=θl(a1/2),β=θl(b1/2)\alpha=\theta_l(a^{1/2}),\beta=\theta_l(b^{1/2}). Equation (BW12) and the left-ideal property give

c=v1∗a1/2+v2∗b1/2∈Il,θl(c)=V∗(α,β). c=v_1^*a^{1/2}+v_2^*b^{1/2}\in I_l,\qquad \theta_l(c)=V^*(\alpha,\beta).

Every entry of QQ lies in MM. Applying the module identity (BW2) componentwise to QX=XQX=X gives Q(α,β)=(α,β)Q(\alpha,\beta)=(\alpha,\beta). The isometry V∗V^* on Q(H⊕H)Q(H\oplus H) now gives

∥θl((a+b)1/2)∥2=∥θl(a1/2)∥2+∥θl(b1/2)∥2.(BW13) \|\theta_l((a+b)^{1/2})\|^2 =\|\theta_l(a^{1/2})\|^2+\|\theta_l(b^{1/2})\|^2. \tag{BW13}

This is the finite-value additivity mechanism. The complete weight, including its infinite values, arbitrary-net normality, semifiniteness and GNS graphs, is constructed in WF1–6. The present proof supplies its whole bounded-vector foundations; it does not assume a weight or a cyclic state vector in order to obtain them.

Bounded graph foundation · Given-weight recovery · Modular covariance.