TheoremBase

Iterate the two-fold gluing by dependent choice to get consistent laws, pass to their bounded resolvent-transform laws, which form a chain linked by marginal embeddings of GNS algebras, take the inductive limit, and reconstruct all square-integrable blocks there with one universal polynomial sequence.

Proof

Each result cited is universally quantified over the data in its own statement. Fix kk, π\pi, (mj)j∈N(m_{j})_{j\in\mathbb{N}} and (γj)j∈N(\gamma_{j})_{j\in\mathbb{N}} as in the statement. Laws λt\lambda_{t} of self-adjoint tuples tt are those of Self-Adjoint Tuples in a Tracial W*-Probability Space and Their Laws §law, resolvent transforms R(X)\mathbf{R}(X) of L2L^{2} tuples are those of The Resolvent of a Self-Adjoint Vector and the Resolvent Transform of a Square-Integrable Tuple §transform, and p(t)p(t) is the value of a polynomial at a tuple of operators, with the rules of Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators. For a law λ\lambda, Hλ\mathcal{H}_{\lambda}, Ωλ\Omega_{\lambda}, Mλ\mathcal{M}_{\lambda}, τλ\tau_{\lambda} and LpλL^{\lambda}_{p} are its complex GNS space, vacuum vector, tracial algebra, trace and left multiplication operators, as in Noncommutative Laws, Couplings and the Wasserstein Distance: Standing Notation §gns.

Notation. Put d0=kd_{0}=k and, for n∈Nn\in\mathbb{N}, ln=m1+⋯+mnl_{n}=m_{1}+\dots+m_{n} and dn=k+lnd_{n}=k+l_{n}; thus dn+1=dn+mn+1d_{n+1}=d_{n}+m_{n+1} and dn≥k+n>nd_{n}\ge k+n>n. For n∈Nn\in\mathbb{N} let An=(Gn,0)A^{n}=(G^{n},0) be the affine datum from dn+1d_{n+1} to dnd_{n} variables with Gijn=1G^{n}_{ij}=1 if j=ij=i and 00 otherwise, and let Bn=(Qn,0)B^{n}=(Q^{n},0) be the affine datum from dnd_{n} to k+mnk+m_{n} variables with Qijn=1Q^{n}_{ij}=1 if either i≤ki\le k and j=ij=i, or i>ki>k and j=dn−1+i−kj=d_{n-1}+i-k, and Qijn=0Q^{n}_{ij}=0 otherwise. By Square-Integrable Tuples in a Tracial W*-Probability Space: Their Norm, Affine Images, Pairs, Embedded Images and Laws §operations (the constant terms vanish and each row has exactly one entry 11), for every L2L^{2} dn+1d_{n+1}-tuple WW the tuple AnWA^{n}W consists of the first dnd_{n} entries of WW; for every L2L^{2} dnd_{n}-tuple WW the tuple BnWB^{n}W consists of the first kk entries of WW followed by its entries dn−1+1,…,dnd_{n-1}+1,\dots,d_{n}; and for every L2L^{2} (k+l)(k+l)-tuple WW the tuple FlWF^{l}W consists of its first kk entries. We use Calculus of Laws of Square-Integrable Tuples: Bounded Tuples, the Lipschitz Bound, Moments, Affine Push-Forwards, Couplings and Embeddings §push-forward in the form law(TW)=T#law(W)\mathrm{law}(TW)=T_{\#}\mathrm{law}(W) throughout.

Step 1 (consistent laws by dependent choice). Let SS be the set of pairs (n,Λ)(n,\Lambda) with n∈Nn\in\mathbb{N}, Λ∈Σdn2\Lambda\in\Sigma^{2}_{d_{n}} and F#lnΛ=πF^{l_{n}}_{\#}\Lambda=\pi, a subset of N×⋃d∈NΣd2\mathbb{N}\times\bigcup_{d\in\mathbb{N}}\Sigma^{2}_{d}. Let R\mathcal{R} be the relation on SS consisting of the pairs ((n,Λ),(n′,Λ′))((n,\Lambda),(n',\Lambda')) with n′=n+1n'=n+1, A#nΛ′=ΛA^{n}_{\#}\Lambda'=\Lambda and B#n+1Λ′=γn+1B^{n+1}_{\#}\Lambda'=\gamma_{n+1}. Since l1=m1l_{1}=m_{1} and F#m1γ1=πF^{m_{1}}_{\#}\gamma_{1}=\pi, the pair (1,γ1)(1,\gamma_{1}) lies in SS.

Every (n,Λ)∈S(n,\Lambda)\in S has an R\mathcal{R}-successor in SS. Indeed, F#lnΛ=π=F#mn+1γn+1F^{l_{n}}_{\#}\Lambda=\pi=F^{m_{n+1}}_{\#}\gamma_{n+1}, so Gluing Two Square-Integrable Noncommutative Laws along a Common Marginal §glue, applied with kk, lnl_{n}, mn+1m_{n+1}, Λ\Lambda and γn+1\gamma_{n+1} in place of its kk, mm, nn, π1\pi_{1} and π2\pi_{2}, gives a tracial W*-probability space with an L2L^{2} kk-tuple XX, an L2L^{2} lnl_{n}-tuple YY and an L2L^{2} mn+1m_{n+1}-tuple Z′Z' such that law(X,Y)=Λ\mathrm{law}(X,Y)=\Lambda and law(X,Z′)=γn+1\mathrm{law}(X,Z')=\gamma_{n+1}. Put W=(X,Y,Z′)W=(X,Y,Z'), an L2L^{2} dn+1d_{n+1}-tuple, and Λ′=law(W)∈Σdn+12\Lambda'=\mathrm{law}(W)\in\Sigma^{2}_{d_{n+1}}. By the Notation, AnW=(X,Y)A^{n}W=(X,Y), Bn+1W=(X,Z′)B^{n+1}W=(X,Z'), Fln+1W=XF^{l_{n+1}}W=X and Fln(X,Y)=XF^{l_{n}}(X,Y)=X; hence A#nΛ′=law(X,Y)=ΛA^{n}_{\#}\Lambda'=\mathrm{law}(X,Y)=\Lambda, B#n+1Λ′=law(X,Z′)=γn+1B^{n+1}_{\#}\Lambda'=\mathrm{law}(X,Z')=\gamma_{n+1}, and F#ln+1Λ′=law(X)=F#lnlaw(X,Y)=F#lnΛ=πF^{l_{n+1}}_{\#}\Lambda'=\mathrm{law}(X)=F^{l_{n}}_{\#}\mathrm{law}(X,Y)=F^{l_{n}}_{\#}\Lambda=\pi. So (n+1,Λ′)∈S(n+1,\Lambda')\in S and it is an R\mathcal{R}-successor of (n,Λ)(n,\Lambda).

By Axiom of Dependent Choice there is a sequence (an)n∈N(a_{n})_{n\in\mathbb{N}} in SS with a1=(1,γ1)a_{1}=(1,\gamma_{1}) and (an,an+1)∈R(a_{n},a_{n+1})\in\mathcal{R} for every nn. By induction the first component of ana_{n} is nn; write an=(n,Λn)a_{n}=(n,\Lambda_{n}). Then for every n∈Nn\in\mathbb{N}:

Λ1=γ1,Λn∈Σdn2,A#nΛn+1=Λn,B#n+1Λn+1=γn+1.(1.1)\Lambda_{1}=\gamma_{1},\qquad\Lambda_{n}\in\Sigma^{2}_{d_{n}},\qquad A^{n}_{\#}\Lambda_{n+1}=\Lambda_{n},\qquad B^{n+1}_{\#}\Lambda_{n+1}=\gamma_{n+1}.\qquad(1.1)

Step 2 (the bounded laws). Fix n∈Nn\in\mathbb{N}. By Every Square-Integrable Noncommutative Law is the Law of a Square-Integrable Tuple; Realisation of Couplings and of Almost Optimal Pairs §law there are a tracial W*-probability space (K,N,Ψ)(K,N,\Psi) and an L2L^{2} dnd_{n}-tuple X0X^{0} of it with law(X0)=Λn\mathrm{law}(X^{0})=\Lambda_{n}. By The Resolvent Transform of Square-Integrable Tuples: Consistency, Concatenation, the L^2 Lipschitz Bound and Universal Polynomial Recovery §consistent, R(X0)\mathbf{R}(X^{0}) is a self-adjoint 2dn2d_{n}-tuple in NN, so Γn=λR(X0)∈Σ2dn\Gamma_{n}=\lambda_{\mathbf{R}(X^{0})}\in\Sigma_{2d_{n}} by Laws of Self-Adjoint Tuples in a Tracial W*-Probability Space: Moments, Affine Images, Couplings, Embeddings and L^2 Approximation §law. By The Law of the Resolvent Transform is a Lipschitz Function of the Law, and Reconstruction of a Square-Integrable Tuple from a Bounded Tuple with the Law of Its Transform §law, λR(X)=Γn\lambda_{\mathbf{R}(X)}=\Gamma_{n} for every L2L^{2} dnd_{n}-tuple XX of every tracial W*-probability space with law(X)=Λn\mathrm{law}(X)=\Lambda_{n}; in particular Γn\Gamma_{n} does not depend on the choice of X0X^{0}, and is determined by Λn\Lambda_{n}.

For n∈Nn\in\mathbb{N} let βn=σ(x1,…,x2dn):P2dn→P2dn+1\beta_{n}=\sigma_{(x_{1},\dots,x_{2d_{n}})}:\mathcal{P}_{2d_{n}}\to\mathcal{P}_{2d_{n+1}} be the substitution of the first 2dn2d_{n} variables. We claim

Γn+1∘βn=Γn.(2.1)\Gamma_{n+1}\circ\beta_{n}=\Gamma_{n}.\qquad(2.1)

Take, by Every Square-Integrable Noncommutative Law is the Law of a Square-Integrable Tuple; Realisation of Couplings and of Almost Optimal Pairs §law, an L2L^{2} dn+1d_{n+1}-tuple X0X^{0} of a tracial W*-probability space (K,N,Ψ)(K,N,\Psi) with law(X0)=Λn+1\mathrm{law}(X^{0})=\Lambda_{n+1}, and write X0=(X′,X′′)X^{0}=(X',X'') with X′=AnX0X'=A^{n}X^{0} its first dnd_{n} entries. By The Resolvent Transform of Square-Integrable Tuples: Consistency, Concatenation, the L^2 Lipschitz Bound and Universal Polynomial Recovery §consistent, the first 2dn2d_{n} entries of R(X0)\mathbf{R}(X^{0}) form R(X′)\mathbf{R}(X'). For p∈P2dnp\in\mathcal{P}_{2d_{n}}, Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §substitution and Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §values give (βnp)(R(X0))=p(R(X′))(\beta_{n}p)(\mathbf{R}(X^{0}))=p(\mathbf{R}(X')), so by Self-Adjoint Tuples in a Tracial W*-Probability Space and Their Laws §law, Γn+1(βnp)=⟨Ψ,p(R(X′))Ψ⟩=λR(X′)(p)\Gamma_{n+1}(\beta_{n}p)=\langle\Psi,p(\mathbf{R}(X'))\Psi\rangle=\lambda_{\mathbf{R}(X')}(p). Finally law(X′)=A#nΛn+1=Λn\mathrm{law}(X')=A^{n}_{\#}\Lambda_{n+1}=\Lambda_{n} by (1.1), so λR(X′)=Γn\lambda_{\mathbf{R}(X')}=\Gamma_{n} by the previous paragraph.

Step 3 (a chain of tracial W-probability spaces).* For n∈Nn\in\mathbb{N} write Hn=HΓnH_{n}=\mathcal{H}_{\Gamma_{n}}, Mn=MΓnM_{n}=\mathcal{M}_{\Gamma_{n}}, Ωn=ΩΓn\Omega_{n}=\Omega_{\Gamma_{n}} and Lp(n)=LpΓnL^{(n)}_{p}=L^{\Gamma_{n}}_{p} for p∈P2dnp\in\mathcal{P}_{2d_{n}}. By The Tracial Algebra of a Noncommutative Law is a Tracial W*-Probability Space: the W*-Closure of the Left Multiplications §w-star, (Hn,Mn,Ωn)(H_{n},M_{n},\Omega_{n}) is a tracial W*-probability space with trace τΓn\tau_{\Gamma_{n}}. Every variable xix_{i} is self-adjoint by Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §self-adjoint and Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §adjoint, so Lxi(n)L^{(n)}_{x_{i}} is self-adjoint by Left and Right Multiplication Operators on the Complex GNS Space of a Noncommutative Law: Boundedness, Algebra Rules, Adjoints, Commutation, the Vacuum and the Conjugation §adjoint, and it lies in MnM_{n} by The Tracial Algebra of a Noncommutative Law: a Norm-Closed Unital *-Algebra with a Faithful Positive Trace, Determined by Vacuum Vectors, Closed under Square Roots §star-algebra.

Let θn:Mn→Mn+1\theta_{n}:M_{n}\to M_{n+1} be the map π\pi of Marginals of a Noncommutative Law: the Isometry of GNS Spaces, the Trace-Preserving Embedding of Tracial Algebras and the Conditional Expectation §embedding, with 2dn+12d_{n+1}, 2dn2d_{n}, Γn+1\Gamma_{n+1} and the 2dn2d_{n}-tuple (x1,…,x2dn)(x_{1},\dots,x_{2d_{n}}) in P2dn+1,sa\mathcal{P}_{2d_{n+1},\mathrm{sa}} in place of its mm, nn, γ\gamma and aa; its substitution σa\sigma_{a} is βn\beta_{n}, so its marginal law γ∘σa\gamma\circ\sigma_{a} is Γn\Gamma_{n} by (2.1), and θn\theta_{n} maps MnM_{n} into Mn+1M_{n+1}. By Marginals of a Noncommutative Law: the Isometry of GNS Spaces, the Trace-Preserving Embedding of Tracial Algebras and the Conditional Expectation §homomorphism, θn\theta_{n} is linear, unital, multiplicative, ∗*-preserving and satisfies τΓn+1∘θn=τΓn\tau_{\Gamma_{n+1}}\circ\theta_{n}=\tau_{\Gamma_{n}}, so it is a trace-preserving embedding of (Hn,Mn,Ωn)(H_{n},M_{n},\Omega_{n}) into (Hn+1,Mn+1,Ωn+1)(H_{n+1},M_{n+1},\Omega_{n+1}) in the sense of Trace-Preserving Embeddings of Tracial W*-Probability Spaces, Their Implementing Isometries and Conditional Expectations §embedding; and the same clause with Substitution is the Unique Unital Homomorphism with Prescribed Values on the Variables: Monomials, Products, Adjoints and Composition §values gives

θn(Lxi(n))=Lβn(xi)(n+1)=Lxi(n+1)(i∈[2dn]).(3.1)\theta_{n}(L^{(n)}_{x_{i}})=L^{(n+1)}_{\beta_{n}(x_{i})}=L^{(n+1)}_{x_{i}}\qquad(i\in[2d_{n}]).\qquad(3.1)

Step 4 (the inductive limit and one sequence of operators). Apply Inductive Limits of Tracial W*-Probability Spaces along Trace-Preserving Embeddings with θn\theta_{n} in place of its πk\pi_{k}: there are a tracial W*-probability space (H,M,Ω)(H,M,\Omega) and trace-preserving embeddings ρn\rho_{n} of (Hn,Mn,Ωn)(H_{n},M_{n},\Omega_{n}) into it with ρn+1(θn(T))=ρn(T)\rho_{n+1}(\theta_{n}(T))=\rho_{n}(T) for all nn and T∈MnT\in M_{n}, by Inductive Limits of Tracial W*-Probability Spaces along Trace-Preserving Embeddings §compatible. Since (dn)(d_{n}) is increasing, if i≤2dni\le2d_{n} then i≤2dn′i\le2d_{n'} for all n′≥nn'\ge n, and induction on n′n' using (3.1) and compatibility gives ρn′(Lxi(n′))=ρn(Lxi(n))\rho_{n'}(L^{(n')}_{x_{i}})=\rho_{n}(L^{(n)}_{x_{i}}) for all n′≥nn'\ge n. Hence, for i∈Ni\in\mathbb{N}, the element

Si=ρn(Lxi(n))∈MS_{i}=\rho_{n}(L^{(n)}_{x_{i}})\in M

is the same for every n∈Nn\in\mathbb{N} with i≤2dni\le2d_{n} (compare two such indices through the smaller one); such nn exist since di>id_{i}>i.

Fix n∈Nn\in\mathbb{N} and let un=(Lx1(n),…,Lx2dn(n))u^{n}=(L^{(n)}_{x_{1}},\dots,L^{(n)}_{x_{2d_{n}}}), a self-adjoint 2dn2d_{n}-tuple in MnM_{n} by Step 3. By The Tracial Algebra of a Noncommutative Law is a Tracial W*-Probability Space: the W*-Closure of the Left Multiplications §law and Self-Adjoint Tuples in a Tracial W*-Probability Space and Their Laws §law, λun=Γn\lambda_{u^{n}}=\Gamma_{n}. By Laws of Self-Adjoint Tuples in a Tracial W*-Probability Space: Moments, Affine Images, Couplings, Embeddings and L^2 Approximation §embedding, sn=(ρn(u1n),…,ρn(u2dnn))=(S1,…,S2dn)s^{n}=(\rho_{n}(u^{n}_{1}),\dots,\rho_{n}(u^{n}_{2d_{n}}))=(S_{1},\dots,S_{2d_{n}}) is a self-adjoint 2dn2d_{n}-tuple in MM with

λsn=λun=Γn.(4.1)\lambda_{s^{n}}=\lambda_{u^{n}}=\Gamma_{n}.\qquad(4.1)

Step 5 (reconstruction of all blocks at once). Fix, once and for all, a sequence (Pr)r∈N(P_{r})_{r\in\mathbb{N}} as in The Resolvent Transform of Square-Integrable Tuples: Consistency, Concatenation, the L^2 Lipschitz Bound and Universal Polynomial Recovery §recovery. Fix n∈Nn\in\mathbb{N}, and take by Every Square-Integrable Noncommutative Law is the Law of a Square-Integrable Tuple; Realisation of Couplings and of Almost Optimal Pairs §law an L2L^{2} dnd_{n}-tuple X0X^{0} of some tracial W*-probability space with law(X0)=Λn\mathrm{law}(X^{0})=\Lambda_{n}; by Step 2 and (4.1), λsn=Γn=λR(X0)\lambda_{s^{n}}=\Gamma_{n}=\lambda_{\mathbf{R}(X^{0})}. So The Law of the Resolvent Transform is a Lipschitz Function of the Law, and Reconstruction of a Square-Integrable Tuple from a Bounded Tuple with the Law of Its Transform §reconstruction, applied with dnd_{n}, X0X^{0}, the space (H,M,Ω)(H,M,\Omega) in place of its (K,N,Ψ)(K,N,\Psi), and sns^{n} in place of its ss, shows: for every j∈[dn]j\in[d_{n}] the sequence (Pr(S2j−1,S2j)Ω)r∈N(P_{r}(S_{2j-1},S_{2j})\Omega)_{r\in\mathbb{N}} converges in HH to a vector fixed by the conjugation of (H,M,Ω)(H,M,\Omega), and the dnd_{n}-tuple of these limits is an L2L^{2} dnd_{n}-tuple of (H,M,Ω)(H,M,\Omega) with law Λn\Lambda_{n}.

The sequence (Pr(S2j−1,S2j)Ω)r(P_{r}(S_{2j-1},S_{2j})\Omega)_{r} does not involve nn. For j∈Nj\in\mathbb{N} let Vj∈HV_{j}\in H be its limit, which exists by the preceding paragraph applied with n=jn=j (as j≤djj\le d_{j}) and is unique by Uniqueness of Limits in a Metric Space. Then, for every n∈Nn\in\mathbb{N}, W^n=(V1,…,Vdn)\widehat{W}^{n}=(V_{1},\dots,V_{d_{n}}) is an L2L^{2} dnd_{n}-tuple of (H,M,Ω)(H,M,\Omega) with

law(W^n)=Λn.(5.1)\mathrm{law}(\widehat{W}^{n})=\Lambda_{n}.\qquad(5.1)

Step 6 (the tuples). Let Z=(V1,…,Vk)Z=(V_{1},\dots,V_{k}), an L2L^{2} kk-tuple of (H,M,Ω)(H,M,\Omega), and for j∈Nj\in\mathbb{N} let Yj=(Vdj−1+1,…,Vdj)Y_{j}=(V_{d_{j-1}+1},\dots,V_{d_{j}}), an L2L^{2} mjm_{j}-tuple of (H,M,Ω)(H,M,\Omega), since dj−dj−1=mjd_{j}-d_{j-1}=m_{j}. For j=1j=1 we have d0=kd_{0}=k, so (Z,Y1)=W^1(Z,Y_{1})=\widehat{W}^{1} and law(Z,Y1)=Λ1=γ1\mathrm{law}(Z,Y_{1})=\Lambda_{1}=\gamma_{1} by (5.1) and (1.1). For j≥2j\ge2, write j=n+1j=n+1 with n∈Nn\in\mathbb{N}; by the Notation (Z,Yj)=BjW^j(Z,Y_{j})=B^{j}\widehat{W}^{j}, so by Calculus of Laws of Square-Integrable Tuples: Bounded Tuples, the Lipschitz Bound, Moments, Affine Push-Forwards, Couplings and Embeddings §push-forward, (5.1) and (1.1),

law(Z,Yj)=B#jlaw(W^j)=B#n+1Λn+1=γj.\mathrm{law}(Z,Y_{j})=B^{j}_{\#}\mathrm{law}(\widehat{W}^{j})=B^{n+1}_{\#}\Lambda_{n+1}=\gamma_{j}.

Thus (H,M,Ω)(H,M,\Omega), ZZ and (Yj)j∈N(Y_{j})_{j\in\mathbb{N}} have the required properties.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…