Each result cited is universally quantified over the data in its own statement.
Conventions. (C1) Evaluation. Let ( K , N , Ψ ) (K,N,\Psi) ( K , N , Ψ ) be a tracial W*-probability space and T T T a tuple in N N N . For a polynomial p p p , p ( T ) p(T) p ( T ) is the value of Evaluation of Noncommutative Polynomials at a Tuple of Bounded Operators §evaluation ; it is linear in p p p , ( p q ) ( T ) = p ( T ) q ( T ) (pq)(T)=p(T)q(T) ( pq ) ( T ) = p ( T ) q ( T ) , 1 ( T ) = I 1(T)=I 1 ( T ) = I , x j ( T ) = T j x_{j}(T)=T_{j} x j ( T ) = T j and T u v = T u T v T_{uv}=T_{u}T_{v} T uv = T u T v for words u , v u,v u , v , p ∗ ( T ) = p ( T ) ∗ p^{*}(T)=p(T)^{*} p ∗ ( T ) = p ( T ) ∗ if the entries of T T T are self-adjoint, and ( σ a ( p ) ) ( S ) = p ( a ( S ) ) (\sigma_{a}(p))(S)=p(a(S)) ( σ a ( p )) ( S ) = p ( a ( S )) for a tuple a a a of polynomials with substitution σ a \sigma_{a} σ a , by Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §values , Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §homomorphism , Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §adjoint and Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §substitution . Moreover p ( T ) ∈ N p(T)\in N p ( T ) ∈ N by Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §transport (with Φ \Phi Φ the identity map of N N N ), and p ( T ) p(T) p ( T ) is self-adjoint if the entries of T T T are self-adjoint and p p p is a self-adjoint polynomial. For a self-adjoint n n n -tuple z z z in N N N , λ z ( p ) = ⟨ Ψ , p ( z ) Ψ ⟩ \lambda_{z}(p)=\langle\Psi,p(z)\Psi\rangle λ z ( p ) = ⟨ Ψ , p ( z ) Ψ ⟩ by Self-Adjoint Tuples in a Tracial W*-Probability Space and Their Laws §law , so for q ∈ P n q\in\mathcal{P}_{n} q ∈ P n
∥ q ( z ) Ψ ∥ 2 = ⟨ Ψ , q ( z ) ∗ q ( z ) Ψ ⟩ = λ z ( q ∗ q ) . (E) \lVert q(z)\Psi\rVert^{2}=\langle\Psi,q(z)^{*}q(z)\Psi\rangle=\lambda_{z}(q^{*}q).\tag{E} ∥ q ( z ) Ψ ∥ 2 = ⟨ Ψ , q ( z ) ∗ q ( z ) Ψ ⟩ = λ z ( q ∗ q ) . ( E )
(C2) Law invariance. If Z , Z ′ Z,Z' Z , Z ′ are L 2 L^{2} L 2 k k k -tuples, possibly of different tracial W*-probability spaces, with l a w ( Z ) = l a w ( Z ′ ) \mathrm{law}(Z)=\mathrm{law}(Z') law ( Z ) = law ( Z ′ ) , then l a w ( T Z ) = T # l a w ( Z ) = l a w ( T Z ′ ) \mathrm{law}(TZ)=T_{\#}\mathrm{law}(Z)=\mathrm{law}(TZ') law ( TZ ) = T # law ( Z ) = law ( T Z ′ ) for every affine datum T T T from k k k variables, by Calculus of Laws of Square-Integrable Tuples: Bounded Tuples, the Lipschitz Bound, Moments, Affine Push-Forwards, Couplings and Embeddings §push-forward , and ⟨ Z i , Z l ⟩ = m i l ( l a w ( Z ) ) = ⟨ Z i ′ , Z l ′ ⟩ \langle Z_{i},Z_{l}\rangle=\mathrm{m}_{il}(\mathrm{law}(Z))=\langle Z'_{i},Z'_{l}\rangle ⟨ Z i , Z l ⟩ = m i l ( law ( Z )) = ⟨ Z i ′ , Z l ′ ⟩ for all i , l ∈ [ k ] i,l\in[k] i , l ∈ [ k ] , by Calculus of Laws of Square-Integrable Tuples: Bounded Tuples, the Lipschitz Bound, Moments, Affine Push-Forwards, Couplings and Embeddings §moments ; in particular ∥ Z ∥ 2 = ∥ Z ′ ∥ 2 \lVert Z\rVert_{2}=\lVert Z'\rVert_{2} ∥ Z ∥ 2 = ∥ Z ′ ∥ 2 (Square-Integrable Tuples in a Tracial W*-Probability Space: Their Norm, Affine Images, Pairs, Embedded Images and Laws §tuples ).
(C3) Bounded realisations. Let ( K , N , Ψ ) (K,N,\Psi) ( K , N , Ψ ) be a tracial W*-probability space, k ∈ N k\in\mathbb{N} k ∈ N , and Z Z Z an L 2 L^{2} L 2 k k k -tuple of it with l a w ( Z ) = κ k ( γ ) \mathrm{law}(Z)=\kappa_{k}(\gamma) law ( Z ) = κ k ( γ ) for some γ ∈ Σ k , ρ \gamma\in\Sigma_{k,\rho} γ ∈ Σ k , ρ , ρ > 0 \rho>0 ρ > 0 real. By Square-Integrable Tuples with a Bounded Law are Vacuum Tuples of Bounded Self-Adjoint Operators §operator , Square-Integrable Tuples with a Bounded Law are Vacuum Tuples of Bounded Self-Adjoint Operators §bound and Square-Integrable Tuples with a Bounded Law are Vacuum Tuples of Bounded Self-Adjoint Operators §law there is exactly one self-adjoint k k k -tuple z z z in N N N with z Ψ = Z z\Psi=Z z Ψ = Z , and it satisfies ∥ z j ∥ o p ≤ ρ \lVert z_{j}\rVert_{\mathrm{op}}\le\rho ∥ z j ∥ op ≤ ρ for all j j j and λ z = γ \lambda_{z}=\gamma λ z = γ . Every law in Σ k \Sigma_{k} Σ k lies in some Σ k , ρ \Sigma_{k,\rho} Σ k , ρ (Noncommutative Laws of Finitely Many Self-Adjoint Variables with a Norm Bound §law ). Conversely, for a self-adjoint k k k -tuple z z z in N N N , l a w ( z Ψ ) = κ k ( λ z ) \mathrm{law}(z\Psi)=\kappa_{k}(\lambda_{z}) law ( z Ψ ) = κ k ( λ z ) by Calculus of Laws of Square-Integrable Tuples: Bounded Tuples, the Lipschitz Bound, Moments, Affine Push-Forwards, Couplings and Embeddings §bounded ; and for self-adjoint d d d -tuples z , z ′ z,z' z , z ′ in N N N , the 2 d 2d 2 d -tuple ( z , z ′ ) (z,z') ( z , z ′ ) is self-adjoint with vacuum tuple ( z Ψ , z ′ Ψ ) (z\Psi,z'\Psi) ( z Ψ , z ′ Ψ ) and λ ( z , z ′ ) ∈ Π ( λ z , λ z ′ ) \lambda_{(z,z')}\in\Pi(\lambda_{z},\lambda_{z'}) λ ( z , z ′ ) ∈ Π ( λ z , λ z ′ ) by Laws of Self-Adjoint Tuples in a Tracial W*-Probability Space: Moments, Affine Images, Couplings, Embeddings and L^2 Approximation §coupling , so that λ ( z , z ′ ) ∘ ι 1 = λ z \lambda_{(z,z')}\circ\iota^{1}=\lambda_{z} λ ( z , z ′ ) ∘ ι 1 = λ z by Couplings of Two Noncommutative Laws and Their Quadratic Cost §coupling .
(C4) GNS tuples. Let k ∈ { d , 2 d } k\in\{d,2d\} k ∈ { d , 2 d } and γ ∈ Σ k \gamma\in\Sigma_{k} γ ∈ Σ k . Each multiplication operator L x i L_{x_{i}} L x i (i ∈ [ k ] i\in[k] i ∈ [ k ] , Left and Right Multiplication Operators on the Complex GNS Space of a Noncommutative Law: Boundedness, Algebra Rules, Adjoints, Commutation, the Vacuum and the Conjugation §multiplication ) lies in A γ ⊆ A γ ′ ′ = M γ \mathcal{A}_{\gamma}\subseteq\mathcal{A}_{\gamma}''=\mathcal{M}_{\gamma} A γ ⊆ A γ ′′ = M γ (The Tracial Algebra of a Noncommutative Law is a Tracial W*-Probability Space: the W*-Closure of the Left Multiplications §w-star , The Commutant of a Set of Bounded Operators on a Complex Hilbert Space §commutant ) and 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 , since x i ∗ = x i x_{i}^{*}=x_{i} x i ∗ = x i (Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §adjoint ). So ℓ γ = ( L x 1 , … , L x k ) \ell^{\gamma}=(L_{x_{1}},\dots,L_{x_{k}}) ℓ γ = ( L x 1 , … , L x k ) is a self-adjoint k k k -tuple in M γ \mathcal{M}_{\gamma} M γ , and p ( ℓ γ ) = L p p(\ell^{\gamma})=L_{p} p ( ℓ γ ) = L p for p ∈ P k p\in\mathcal{P}_{k} p ∈ P k by Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §gns ; hence p ( ℓ γ ) Ω γ = p ^ p(\ell^{\gamma})\Omega_{\gamma}=\widehat{p} p ( ℓ γ ) Ω γ = p and ⟨ p ^ , q ^ ⟩ = γ ( p ∗ q ) \langle\widehat{p},\widehat{q}\rangle=\gamma(p^{*}q) ⟨ p , q ⟩ = γ ( p ∗ q ) for p , q ∈ P k p,q\in\mathcal{P}_{k} p , q ∈ P k (Left and Right Multiplication Operators on the Complex GNS Space of a Noncommutative Law: Boundedness, Algebra Rules, Adjoints, Commutation, the Vacuum and the Conjugation §vacuum ), and λ ℓ γ = γ \lambda_{\ell^{\gamma}}=\gamma λ ℓ γ = γ by The Tracial Algebra of a Noncommutative Law is a Tracial W*-Probability Space: the W*-Closure of the Left Multiplications §law . In particular ℓ λ Ω λ = X λ \ell^{\lambda}\Omega_{\lambda}=X_{\lambda} ℓ λ Ω λ = X λ , and ℓ π Ω π = ( X π , P π ) \ell^{\pi}\Omega_{\pi}=(X_{\pi},P_{\pi}) ℓ π Ω π = ( X π , P π ) for a bounded plan π \pi π at λ \lambda λ (The Shift of a Bounded Plan by a Self-Adjoint Field §tuples ).
Step 1 (the operator tuple of X X X ). Since l a w ( X ) = κ d ( λ ) \mathrm{law}(X)=\kappa_{d}(\lambda) law ( X ) = κ d ( λ ) with λ ∈ Σ d , r \lambda\in\Sigma_{d,r} λ ∈ Σ d , r , (C3) gives exactly one self-adjoint d d d -tuple s s s in M M M with s Ω = X s\Omega=X s Ω = X , and λ s = λ \lambda_{s}=\lambda λ s = λ . Put ℓ = ℓ λ \ell=\ell^{\lambda} ℓ = ℓ λ ; by (C4), ℓ Ω λ = X λ \ell\Omega_{\lambda}=X_{\lambda} ℓ Ω λ = X λ , λ ℓ = λ \lambda_{\ell}=\lambda λ ℓ = λ and q ( ℓ ) Ω λ = q ^ q(\ell)\Omega_{\lambda}=\widehat{q} q ( ℓ ) Ω λ = q for q ∈ P d q\in\mathcal{P}_{d} q ∈ P d .
Step 2 (mixed moments depend only on the joint law). We claim: let ( H i , M i , Ω i ) (H_{i},M_{i},\Omega_{i}) ( H i , M i , Ω i ) , i ∈ { 1 , 2 } i\in\{1,2\} i ∈ { 1 , 2 } , be tracial W*-probability spaces, s i s^{i} s i a self-adjoint d d d -tuple in M i M_{i} M i with λ s i = λ \lambda_{s^{i}}=\lambda λ s i = λ , and Y i Y^{i} Y i an L 2 L^{2} L 2 d d d -tuple of ( H i , M i , Ω i ) (H_{i},M_{i},\Omega_{i}) ( H i , M i , Ω i ) , and suppose l a w ( s 1 Ω 1 , Y 1 ) = l a w ( s 2 Ω 2 , Y 2 ) \mathrm{law}(s^{1}\Omega_{1},Y^{1})=\mathrm{law}(s^{2}\Omega_{2},Y^{2}) law ( s 1 Ω 1 , Y 1 ) = law ( s 2 Ω 2 , Y 2 ) . Then ⟨ Y j 1 , q ( s 1 ) Ω 1 ⟩ = ⟨ Y j 2 , q ( s 2 ) Ω 2 ⟩ \langle Y^{1}_{j},q(s^{1})\Omega_{1}\rangle=\langle Y^{2}_{j},q(s^{2})\Omega_{2}\rangle ⟨ Y j 1 , q ( s 1 ) Ω 1 ⟩ = ⟨ Y j 2 , q ( s 2 ) Ω 2 ⟩ for every q ∈ P d q\in\mathcal{P}_{d} q ∈ P d and j ∈ [ d ] j\in[d] j ∈ [ d ] .
(a) A Lipschitz bound. Let ( K , N , Ψ ) (K,N,\Psi) ( K , N , Ψ ) be a tracial W*-probability space with conjugation J N J_{N} J N . For b ∈ N b\in N b ∈ N and self-adjoint c ∈ N c\in N c ∈ N , The Commutator of a Square-Integrable Vector with a Bounded Self-Adjoint Operator: Linearity, the Norm Bound, Vacuum Vectors and Dependence on the Law Only §commutator and The Commutator of a Square-Integrable Vector with a Bounded Self-Adjoint Operator: Linearity, the Norm Bound, Vacuum Vectors and Dependence on the Law Only §vacuum give J N c J N ( b Ψ ) − c b Ψ = ( b c − c b ) Ψ J_{N}cJ_{N}(b\Psi)-cb\Psi=(bc-cb)\Psi J N c J N ( b Ψ ) − c b Ψ = ( b c − c b ) Ψ , that is, b c Ψ = J N c J N ( b Ψ ) bc\Psi=J_{N}cJ_{N}(b\Psi) b c Ψ = J N c J N ( b Ψ ) ; as J N J_{N} J N preserves norms (Conjugation of a Complex Hilbert Space §conjugation ), ∥ b c Ψ ∥ ≤ ∥ c ∥ o p ∥ b Ψ ∥ \lVert bc\Psi\rVert\le\lVert c\rVert_{\mathrm{op}}\lVert b\Psi\rVert ∥ b c Ψ ∥ ≤ ∥ c ∥ op ∥ b Ψ ∥ . By induction on m m m , applying this to b c 1 ⋯ c m − 1 ∈ N bc_{1}\cdots c_{m-1}\in N b c 1 ⋯ c m − 1 ∈ N and c m c_{m} c m , we get ∥ b c 1 ⋯ c m Ψ ∥ ≤ r m ∥ b Ψ ∥ \lVert bc_{1}\cdots c_{m}\Psi\rVert\le r^{m}\lVert b\Psi\rVert ∥ b c 1 ⋯ c m Ψ ∥ ≤ r m ∥ b Ψ ∥ for b ∈ N b\in N b ∈ N and self-adjoint c 1 , … , c m ∈ N c_{1},\dots,c_{m}\in N c 1 , … , c m ∈ N of operator norm at most r r r . Now let a , a ′ a,a' a , a ′ be self-adjoint d d d -tuples in N N N with ∥ a l ∥ o p ≤ r \lVert a_{l}\rVert_{\mathrm{op}}\le r ∥ a l ∥ op ≤ r and ∥ a l ′ ∥ o p ≤ r \lVert a'_{l}\rVert_{\mathrm{op}}\le r ∥ a l ′ ∥ op ≤ r for all l ∈ [ d ] l\in[d] l ∈ [ d ] , and let w = i 1 ⋯ i n w=i_{1}\cdots i_{n} w = i 1 ⋯ i n be a word of length n ≥ 1 n\ge1 n ≥ 1 . By (C1), a w − a w ′ a_{w}-a'_{w} a w − a w ′ is the sum over u ∈ [ n ] u\in[n] u ∈ [ n ] of a i 1 ⋯ a i u − 1 ( a i u − a i u ′ ) a i u + 1 ′ ⋯ a i n ′ a_{i_{1}}\cdots a_{i_{u-1}}(a_{i_{u}}-a'_{i_{u}})a'_{i_{u+1}}\cdots a'_{i_{n}} a i 1 ⋯ a i u − 1 ( a i u − a i u ′ ) a i u + 1 ′ ⋯ a i n ′ ; the left factor (I I I for u = 1 u=1 u = 1 , the value at the empty word) has operator norm at most r u − 1 r^{u-1} r u − 1 by Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §bound , and the bound just proved, with b = a i u − a i u ′ b=a_{i_{u}}-a'_{i_{u}} b = a i u − a i u ′ , bounds the norm of the rest applied to Ψ \Psi Ψ by r n − u ∥ a i u Ψ − a i u ′ Ψ ∥ r^{n-u}\lVert a_{i_{u}}\Psi-a'_{i_{u}}\Psi\rVert r n − u ∥ a i u Ψ − a i u ′ Ψ ∥ . Since ∥ a i Ψ − a i ′ Ψ ∥ ≤ ∥ a Ψ − a ′ Ψ ∥ 2 \lVert a_{i}\Psi-a'_{i}\Psi\rVert\le\lVert a\Psi-a'\Psi\rVert_{2} ∥ a i Ψ − a i ′ Ψ ∥ ≤ ∥ a Ψ − a ′ Ψ ∥ 2 (Square-Integrable Tuples in a Tracial W*-Probability Space: Their Norm, Affine Images, Pairs, Embedded Images and Laws §tuples ), ∥ a w Ψ − a w ′ Ψ ∥ ≤ n r n − 1 ∥ a Ψ − a ′ Ψ ∥ 2 \lVert a_{w}\Psi-a'_{w}\Psi\rVert\le nr^{n-1}\lVert a\Psi-a'\Psi\rVert_{2} ∥ a w Ψ − a w ′ Ψ ∥ ≤ n r n − 1 ∥ a Ψ − a ′ Ψ ∥ 2 , and the difference vanishes for the empty word. Writing q = ∑ w c w x w q=\sum_{w}c_{w}x_{w} q = ∑ w c w x w (a finite sum) and C q = ∑ w ≠ ∅ ∣ c w ∣ ∣ w ∣ r ∣ w ∣ − 1 C_{q}=\sum_{w\ne\varnothing}|c_{w}|\,|w|\,r^{|w|-1} C q = ∑ w = ∅ ∣ c w ∣ ∣ w ∣ r ∣ w ∣ − 1 , linearity of evaluation gives
∥ q ( a ) Ψ − q ( a ′ ) Ψ ∥ ≤ C q ∥ a Ψ − a ′ Ψ ∥ 2 . (A1) \lVert q(a)\Psi-q(a')\Psi\rVert\le C_{q}\lVert a\Psi-a'\Psi\rVert_{2}.\tag{A1} ∥ q ( a ) Ψ − q ( a ′ ) Ψ ∥ ≤ C q ∥ a Ψ − a ′ Ψ ∥ 2 . ( A1 )
(b) Approximation. The entries of Y i Y^{i} Y i are fixed by the conjugation of ( H i , M i , Ω i ) (H_{i},M_{i},\Omega_{i}) ( H i , M i , Ω i ) (Square-Integrable Tuples in a Tracial W*-Probability Space: Their Norm, Affine Images, Pairs, Embedded Images and Laws §tuples ), so by Laws of Self-Adjoint Tuples in a Tracial W*-Probability Space: Moments, Affine Images, Couplings, Embeddings and L^2 Approximation §limit we may choose, for i ∈ { 1 , 2 } i\in\{1,2\} i ∈ { 1 , 2 } and k ∈ N k\in\mathbb{N} k ∈ N , self-adjoint d d d -tuples y i , k y^{i,k} y i , k in M i M_{i} M i with t i , k = ∥ y i , k Ω i − Y i ∥ 2 → 0 t_{i,k}=\lVert y^{i,k}\Omega_{i}-Y^{i}\rVert_{2}\to0 t i , k = ∥ y i , k Ω i − Y i ∥ 2 → 0 as k → ∞ k\to\infty k → ∞ . Put γ i , k = λ ( s i , y i , k ) \gamma^{i,k}=\lambda_{(s^{i},y^{i,k})} γ i , k = λ ( s i , y i , k ) , a law in Σ 2 d \Sigma_{2d} Σ 2 d by Laws of Self-Adjoint Tuples in a Tracial W*-Probability Space: Moments, Affine Images, Couplings, Embeddings and L^2 Approximation §law ; by (C3), κ 2 d ( γ i , k ) = l a w ( s i Ω i , y i , k Ω i ) \kappa_{2d}(\gamma^{i,k})=\mathrm{law}(s^{i}\Omega_{i},y^{i,k}\Omega_{i}) κ 2 d ( γ i , k ) = law ( s i Ω i , y i , k Ω i ) and γ i , k ∘ ι 1 = λ s i = λ \gamma^{i,k}\circ\iota^{1}=\lambda_{s^{i}}=\lambda γ i , k ∘ ι 1 = λ s i = λ . By Calculus of Laws of Square-Integrable Tuples: Bounded Tuples, the Lipschitz Bound, Moments, Affine Push-Forwards, Couplings and Embeddings §lipschitz , W ^ 2 ( κ 2 d ( γ i , k ) , l a w ( s i Ω i , Y i ) ) ≤ t i , k \widehat{W}_{2}(\kappa_{2d}(\gamma^{i,k}),\mathrm{law}(s^{i}\Omega_{i},Y^{i}))\le t_{i,k} W 2 ( κ 2 d ( γ i , k ) , law ( s i Ω i , Y i )) ≤ t i , k ; as the laws l a w ( s 1 Ω 1 , Y 1 ) \mathrm{law}(s^{1}\Omega_{1},Y^{1}) law ( s 1 Ω 1 , Y 1 ) and l a w ( s 2 Ω 2 , Y 2 ) \mathrm{law}(s^{2}\Omega_{2},Y^{2}) law ( s 2 Ω 2 , Y 2 ) coincide, the triangle inequality and the symmetry of the metric W ^ 2 \widehat{W}_{2} W 2 give e k = W ^ 2 ( κ 2 d ( γ 1 , k ) , κ 2 d ( γ 2 , k ) ) ≤ t 1 , k + t 2 , k e_{k}=\widehat{W}_{2}(\kappa_{2d}(\gamma^{1,k}),\kappa_{2d}(\gamma^{2,k}))\le t_{1,k}+t_{2,k} e k = W 2 ( κ 2 d ( γ 1 , k ) , κ 2 d ( γ 2 , k )) ≤ t 1 , k + t 2 , k . Fix q ∈ P d q\in\mathcal{P}_{d} q ∈ P d and j ∈ [ d ] j\in[d] j ∈ [ d ] , and let f = x d + j ι 1 ( q ) ∈ P 2 d f=x_{d+j}\,\iota^{1}(q)\in\mathcal{P}_{2d} f = x d + j ι 1 ( q ) ∈ P 2 d . As ι 1 \iota^{1} ι 1 is the substitution of ( x 1 , … , x d ) (x_{1},\dots,x_{d}) ( x 1 , … , x d ) (Couplings of Two Noncommutative Laws and Their Quadratic Cost §marginals ), (C1) gives f ( s i , y i , k ) = y j i , k q ( s i ) f(s^{i},y^{i,k})=y^{i,k}_{j}\,q(s^{i}) f ( s i , y i , k ) = y j i , k q ( s i ) , so γ i , k ( f ) = ⟨ Ω i , y j i , k q ( s i ) Ω i ⟩ = ⟨ y j i , k Ω i , q ( s i ) Ω i ⟩ \gamma^{i,k}(f)=\langle\Omega_{i},y^{i,k}_{j}q(s^{i})\Omega_{i}\rangle=\langle y^{i,k}_{j}\Omega_{i},q(s^{i})\Omega_{i}\rangle γ i , k ( f ) = ⟨ Ω i , y j i , k q ( s i ) Ω i ⟩ = ⟨ y j i , k Ω i , q ( s i ) Ω i ⟩ , y j i , k y^{i,k}_{j} y j i , k being self-adjoint. By the Cauchy--Schwarz inequality,
∣ ⟨ Y j i , q ( s i ) Ω i ⟩ − γ i , k ( f ) ∣ ≤ t i , k ∥ q ( s i ) Ω i ∥ ( i ∈ { 1 , 2 } , k ∈ N ) . (A2) \bigl|\langle Y^{i}_{j},q(s^{i})\Omega_{i}\rangle-\gamma^{i,k}(f)\bigr|\le t_{i,k}\,\lVert q(s^{i})\Omega_{i}\rVert\qquad(i\in\{1,2\},\ k\in\mathbb{N}).\tag{A2} ⟨ Y j i , q ( s i ) Ω i ⟩ − γ i , k ( f ) ≤ t i , k ∥ q ( s i ) Ω i ∥ ( i ∈ { 1 , 2 } , k ∈ N ) . ( A2 )
(c) Comparison. Fix k k k . By Every Square-Integrable Noncommutative Law is the Law of a Square-Integrable Tuple; Realisation of Couplings and of Almost Optimal Pairs §distance , for 2 d 2d 2 d variables and ε = 1 / k \varepsilon=1/k ε = 1/ k , there are a tracial W*-probability space ( K , N , Ψ ) (K,N,\Psi) ( K , N , Ψ ) and L 2 L^{2} L 2 2 d 2d 2 d -tuples Z , Z ′ Z,Z' Z , Z ′ of it with l a w ( Z ) = κ 2 d ( γ 1 , k ) \mathrm{law}(Z)=\kappa_{2d}(\gamma^{1,k}) law ( Z ) = κ 2 d ( γ 1 , k ) , l a w ( Z ′ ) = κ 2 d ( γ 2 , k ) \mathrm{law}(Z')=\kappa_{2d}(\gamma^{2,k}) law ( Z ′ ) = κ 2 d ( γ 2 , k ) and ∥ Z − Z ′ ∥ 2 2 ≤ e k 2 + 1 / k \lVert Z-Z'\rVert_{2}^{2}\le e_{k}^{2}+1/k ∥ Z − Z ′ ∥ 2 2 ≤ e k 2 + 1/ k . By (C3) there are self-adjoint d d d -tuples a , b , a ′ , b ′ a,b,a',b' a , b , a ′ , b ′ in N N N with Z = ( a Ψ , b Ψ ) Z=(a\Psi,b\Psi) Z = ( a Ψ , b Ψ ) , Z ′ = ( a ′ Ψ , b ′ Ψ ) Z'=(a'\Psi,b'\Psi) Z ′ = ( a ′ Ψ , b ′ Ψ ) , λ ( a , b ) = γ 1 , k \lambda_{(a,b)}=\gamma^{1,k} λ ( a , b ) = γ 1 , k and λ ( a ′ , b ′ ) = γ 2 , k \lambda_{(a',b')}=\gamma^{2,k} λ ( a ′ , b ′ ) = γ 2 , k ; then λ a = γ 1 , k ∘ ι 1 = λ \lambda_{a}=\gamma^{1,k}\circ\iota^{1}=\lambda λ a = γ 1 , k ∘ ι 1 = λ and likewise λ a ′ = λ \lambda_{a'}=\lambda λ a ′ = λ . So l a w ( a Ψ ) = κ d ( λ ) \mathrm{law}(a\Psi)=\kappa_{d}(\lambda) law ( a Ψ ) = κ d ( λ ) with λ ∈ Σ d , r \lambda\in\Sigma_{d,r} λ ∈ Σ d , r , and the uniqueness in (C3) gives ∥ a l ∥ o p ≤ r \lVert a_{l}\rVert_{\mathrm{op}}\le r ∥ a l ∥ op ≤ r , and likewise ∥ a l ′ ∥ o p ≤ r \lVert a'_{l}\rVert_{\mathrm{op}}\le r ∥ a l ′ ∥ op ≤ r , for all l l l . As in (b), γ 1 , k ( f ) = ⟨ b j Ψ , q ( a ) Ψ ⟩ \gamma^{1,k}(f)=\langle b_{j}\Psi,q(a)\Psi\rangle γ 1 , k ( f ) = ⟨ b j Ψ , q ( a ) Ψ ⟩ and γ 2 , k ( f ) = ⟨ b j ′ Ψ , q ( a ′ ) Ψ ⟩ \gamma^{2,k}(f)=\langle b'_{j}\Psi,q(a')\Psi\rangle γ 2 , k ( f ) = ⟨ b j ′ Ψ , q ( a ′ ) Ψ ⟩ , and ∥ q ( a ) Ψ ∥ 2 = λ ( q ∗ q ) \lVert q(a)\Psi\rVert^{2}=\lambda(q^{*}q) ∥ q ( a ) Ψ ∥ 2 = λ ( q ∗ q ) by (E). Hence, by the Cauchy--Schwarz inequality and (A1),
∣ γ 1 , k ( f ) − γ 2 , k ( f ) ∣ ≤ ∥ b j Ψ − b j ′ Ψ ∥ ∥ q ( a ) Ψ ∥ + ∥ b j ′ Ψ ∥ ∥ q ( a ) Ψ − q ( a ′ ) Ψ ∥ ≤ ∥ Z − Z ′ ∥ 2 ( λ ( q ∗ q ) 1 / 2 + C q ∥ Z ′ ∥ 2 ) . \bigl|\gamma^{1,k}(f)-\gamma^{2,k}(f)\bigr|\le\lVert b_{j}\Psi-b'_{j}\Psi\rVert\,\lVert q(a)\Psi\rVert+\lVert b'_{j}\Psi\rVert\,\lVert q(a)\Psi-q(a')\Psi\rVert\le\lVert Z-Z'\rVert_{2}\bigl(\lambda(q^{*}q)^{1/2}+C_{q}\lVert Z'\rVert_{2}\bigr). γ 1 , k ( f ) − γ 2 , k ( f ) ≤ ∥ b j Ψ − b j ′ Ψ ∥ ∥ q ( a ) Ψ ∥ + ∥ b j ′ Ψ ∥ ∥ q ( a ) Ψ − q ( a ′ ) Ψ ∥ ≤ ∥ Z − Z ′ ∥ 2 ( λ ( q ∗ q ) 1/2 + C q ∥ Z ′ ∥ 2 ) .
By (C2), ∥ Z ′ ∥ 2 = ∥ ( s 2 Ω 2 , y 2 , k Ω 2 ) ∥ 2 ≤ ∥ s 2 Ω 2 ∥ 2 + ∥ Y 2 ∥ 2 + t 2 , k \lVert Z'\rVert_{2}=\lVert(s^{2}\Omega_{2},y^{2,k}\Omega_{2})\rVert_{2}\le\lVert s^{2}\Omega_{2}\rVert_{2}+\lVert Y^{2}\rVert_{2}+t_{2,k} ∥ Z ′ ∥ 2 = ∥( s 2 Ω 2 , y 2 , k Ω 2 ) ∥ 2 ≤ ∥ s 2 Ω 2 ∥ 2 + ∥ Y 2 ∥ 2 + t 2 , k , which is bounded in k k k , while ∥ Z − Z ′ ∥ 2 2 ≤ ( t 1 , k + t 2 , k ) 2 + 1 / k → 0 \lVert Z-Z'\rVert_{2}^{2}\le(t_{1,k}+t_{2,k})^{2}+1/k\to0 ∥ Z − Z ′ ∥ 2 2 ≤ ( t 1 , k + t 2 , k ) 2 + 1/ k → 0 . So γ 1 , k ( f ) − γ 2 , k ( f ) → 0 \gamma^{1,k}(f)-\gamma^{2,k}(f)\to0 γ 1 , k ( f ) − γ 2 , k ( f ) → 0 ; together with (A2), whose right sides tend to 0 0 0 as the vectors q ( s i ) Ω i q(s^{i})\Omega_{i} q ( s i ) Ω i do not depend on k k k , the two numbers of the claim differ by less than every positive real, so they are equal.
Step 3 (polynomial approximation of ζ \zeta ζ ). Each ζ l \zeta_{l} ζ l is fixed by J λ J_{\lambda} J λ , the conjugation of ( H λ , M λ , Ω λ ) (\mathcal{H}_{\lambda},\mathcal{M}_{\lambda},\Omega_{\lambda}) ( H λ , M λ , Ω λ ) (Square-Integrable Tuples in a Tracial W*-Probability Space: Their Norm, Affine Images, Pairs, Embedded Images and Laws §tuples , The Tracial Algebra of a Noncommutative Law is a Tracial W*-Probability Space: the W*-Closure of the Left Multiplications §w-star ). Let k ∈ N k\in\mathbb{N} k ∈ N and l ∈ [ d ] l\in[d] l ∈ [ d ] . By The Complex GNS Space of a Tracial State on Noncommutative Polynomials §gns , The Complex GNS Space of a Tracial State on Noncommutative Polynomials §classes and The Complex Hilbert Completion is a Complex Hilbert Space Containing a Dense Isometric Image, and Bounded Complex-Linear Maps Extend to It §dense there is p ∈ P d p\in\mathcal{P}_{d} p ∈ P d with ∥ p ^ − ζ l ∥ < 1 / k \lVert\widehat{p}-\zeta_{l}\rVert<1/k ∥ p − ζ l ∥ < 1/ k . Put q l k = 1 2 ( p + p ∗ ) q^{k}_{l}=\frac{1}{2}(p+p^{*}) q l k = 2 1 ( p + p ∗ ) , an element of P d , s a \mathcal{P}_{d,\mathrm{sa}} P d , sa by Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §self-adjoint (p + p ∗ p+p^{*} p + p ∗ is self-adjoint and P d , s a \mathcal{P}_{d,\mathrm{sa}} P d , sa is closed under real multiples). The class map is complex-linear (The Complex Hilbert Completion is a Complex Hilbert Space Containing a Dense Isometric Image, and Bounded Complex-Linear Maps Extend to It §isometry ), J λ p ^ = p ∗ ^ J_{\lambda}\widehat{p}=\widehat{p^{*}} J λ p = p ∗ , and J λ J_{\lambda} J λ is additive, conjugate-homogeneous and norm-preserving (Left and Right Multiplication Operators on the Complex GNS Space of a Noncommutative Law: Boundedness, Algebra Rules, Adjoints, Commutation, the Vacuum and the Conjugation §conjugation ); as J λ ζ l = ζ l J_{\lambda}\zeta_{l}=\zeta_{l} J λ ζ l = ζ l ,
q l k ^ − ζ l = 1 2 ( ( p ^ − ζ l ) + J λ ( p ^ − ζ l ) ) , ∥ q l k ^ − ζ l ∥ ≤ ∥ p ^ − ζ l ∥ < 1 k . \widehat{q^{k}_{l}}-\zeta_{l}=\tfrac{1}{2}\bigl((\widehat{p}-\zeta_{l})+J_{\lambda}(\widehat{p}-\zeta_{l})\bigr),\qquad\lVert\widehat{q^{k}_{l}}-\zeta_{l}\rVert\le\lVert\widehat{p}-\zeta_{l}\rVert<\tfrac{1}{k}. q l k − ζ l = 2 1 ( ( p − ζ l ) + J λ ( p − ζ l ) ) , ∥ q l k − ζ l ∥ ≤ ∥ p − ζ l ∥ < k 1 .
Write q k = ( q 1 k , … , q d k ) q^{k}=(q^{k}_{1},\dots,q^{k}_{d}) q k = ( q 1 k , … , q d k ) and q k ( s ) Ω = ( q 1 k ( s ) Ω , … , q d k ( s ) Ω ) q^{k}(s)\Omega=(q^{k}_{1}(s)\Omega,\dots,q^{k}_{d}(s)\Omega) q k ( s ) Ω = ( q 1 k ( s ) Ω , … , q d k ( s ) Ω ) .
Step 4 (the representation). Let Y Y Y be an L 2 L^{2} L 2 d d d -tuple of ( H , M , Ω ) (H,M,\Omega) ( H , M , Ω ) with l a w ( X , Y ) = l a w ( X λ , ζ ) \mathrm{law}(X,Y)=\mathrm{law}(X_{\lambda},\zeta) law ( X , Y ) = law ( X λ , ζ ) . We claim that
∥ Y l − q ( s ) Ω ∥ = ∥ ζ l − q ^ ∥ ( q ∈ P d , l ∈ [ d ] ) , (R) \lVert Y_{l}-q(s)\Omega\rVert=\lVert\zeta_{l}-\widehat{q}\rVert\qquad(q\in\mathcal{P}_{d},\ l\in[d]),\tag{R} ∥ Y l − q ( s ) Ω ∥ = ∥ ζ l − q ∥ ( q ∈ P d , l ∈ [ d ]) , ( R )
and hence, by Step 3, ∥ Y l − q l k ( s ) Ω ∥ < 1 / k \lVert Y_{l}-q^{k}_{l}(s)\Omega\rVert<1/k ∥ Y l − q l k ( s ) Ω ∥ < 1/ k for all k k k and l l l , so that Y l Y_{l} Y l is the limit in H H H of the sequence ( q l k ( s ) Ω ) k ∈ N (q^{k}_{l}(s)\Omega)_{k\in\mathbb{N}} ( q l k ( s ) Ω ) k ∈ N . In a complex Hilbert space ∥ u − v ∥ 2 = ∥ u ∥ 2 − 2 Re ⟨ u , v ⟩ + ∥ v ∥ 2 \lVert u-v\rVert^{2}=\lVert u\rVert^{2}-2\operatorname{Re}\langle u,v\rangle+\lVert v\rVert^{2} ∥ u − v ∥ 2 = ∥ u ∥ 2 − 2 Re ⟨ u , v ⟩ + ∥ v ∥ 2 ; we compare the three terms for ( u , v ) = ( Y l , q ( s ) Ω ) (u,v)=(Y_{l},q(s)\Omega) ( u , v ) = ( Y l , q ( s ) Ω ) and for ( u , v ) = ( ζ l , q ^ ) (u,v)=(\zeta_{l},\widehat{q}) ( u , v ) = ( ζ l , q ) . First, ∥ Y l ∥ = ∥ ζ l ∥ \lVert Y_{l}\rVert=\lVert\zeta_{l}\rVert ∥ Y l ∥ = ∥ ζ l ∥ by (C2), applied to ( X , Y ) (X,Y) ( X , Y ) and ( X λ , ζ ) (X_{\lambda},\zeta) ( X λ , ζ ) with both indices equal to d + l d+l d + l . Second, ⟨ Y l , q ( s ) Ω ⟩ = ⟨ ζ l , q ( ℓ ) Ω λ ⟩ = ⟨ ζ l , q ^ ⟩ \langle Y_{l},q(s)\Omega\rangle=\langle\zeta_{l},q(\ell)\Omega_{\lambda}\rangle=\langle\zeta_{l},\widehat{q}\rangle ⟨ Y l , q ( s ) Ω ⟩ = ⟨ ζ l , q ( ℓ ) Ω λ ⟩ = ⟨ ζ l , q ⟩ by Step 2, applied with s 1 = s s^{1}=s s 1 = s , Y 1 = Y Y^{1}=Y Y 1 = Y in ( H , M , Ω ) (H,M,\Omega) ( H , M , Ω ) and s 2 = ℓ s^{2}=\ell s 2 = ℓ , Y 2 = ζ Y^{2}=\zeta Y 2 = ζ in ( H λ , M λ , Ω λ ) (\mathcal{H}_{\lambda},\mathcal{M}_{\lambda},\Omega_{\lambda}) ( H λ , M λ , Ω λ ) (here λ s = λ ℓ = λ \lambda_{s}=\lambda_{\ell}=\lambda λ s = λ ℓ = λ and l a w ( s Ω , Y ) = l a w ( X , Y ) = l a w ( X λ , ζ ) = l a w ( ℓ Ω λ , ζ ) \mathrm{law}(s\Omega,Y)=\mathrm{law}(X,Y)=\mathrm{law}(X_{\lambda},\zeta)=\mathrm{law}(\ell\Omega_{\lambda},\zeta) law ( s Ω , Y ) = law ( X , Y ) = law ( X λ , ζ ) = law ( ℓ Ω λ , ζ ) by Step 1), and by Step 1. Third, ∥ q ( s ) Ω ∥ 2 = λ s ( q ∗ q ) = λ ( q ∗ q ) = ∥ q ^ ∥ 2 \lVert q(s)\Omega\rVert^{2}=\lambda_{s}(q^{*}q)=\lambda(q^{*}q)=\lVert\widehat{q}\rVert^{2} ∥ q ( s ) Ω ∥ 2 = λ s ( q ∗ q ) = λ ( q ∗ q ) = ∥ q ∥ 2 by (E) and (C4). This proves (R).
Step 5 (claim 1). If l a w ( X , Q ) = l a w ( X , Q ′ ) = l a w ( X λ , ζ ) \mathrm{law}(X,Q)=\mathrm{law}(X,Q')=\mathrm{law}(X_{\lambda},\zeta) law ( X , Q ) = law ( X , Q ′ ) = law ( X λ , ζ ) , Step 4 with Y = Q Y=Q Y = Q and with Y = Q ′ Y=Q' Y = Q ′ shows that Q l Q_{l} Q l and Q l ′ Q'_{l} Q l ′ are both limits of ( q l k ( s ) Ω ) k (q^{k}_{l}(s)\Omega)_{k} ( q l k ( s ) Ω ) k in the metric space H H H , so Q l = Q l ′ Q_{l}=Q'_{l} Q l = Q l ′ for every l ∈ [ d ] l\in[d] l ∈ [ d ] by Uniqueness of Limits in a Metric Space . Hence Q = Q ′ Q=Q' Q = Q ′ , which is claim 1.
Step 6 (claim 2: the joint law). Assume l a w ( X , Q ) = l a w ( X λ , ζ ) \mathrm{law}(X,Q)=\mathrm{law}(X_{\lambda},\zeta) law ( X , Q ) = law ( X λ , ζ ) , and let π \pi π and P P P be as in claim 2; thus π ∈ Σ 2 d \pi\in\Sigma_{2d} π ∈ Σ 2 d and π ∘ ι 1 = λ \pi\circ\iota^{1}=\lambda π ∘ ι 1 = λ (Marginal Isometries, Bounded Plans and Displacement Pairings for Noncommutative Laws §plans ). By (C3) there are self-adjoint d d d -tuples x , p x,p x , p in M M M with ( x Ω , p Ω ) = ( X , P ) (x\Omega,p\Omega)=(X,P) ( x Ω , p Ω ) = ( X , P ) and λ ( x , p ) = π \lambda_{(x,p)}=\pi λ ( x , p ) = π ; as x Ω = X = s Ω x\Omega=X=s\Omega x Ω = X = s Ω , the uniqueness in Step 1 gives x = s x=s x = s , so λ ( s , p ) = π \lambda_{(s,p)}=\pi λ ( s , p ) = π . Let k ∈ N k\in\mathbb{N} k ∈ N , and let ρ k \rho^{k} ρ k be the 3 d 3d 3 d -tuple in P 2 d \mathcal{P}_{2d} P 2 d with ρ i k = x i \rho^{k}_{i}=x_{i} ρ i k = x i for i ∈ [ 2 d ] i\in[2d] i ∈ [ 2 d ] and ρ 2 d + l k = ι 1 ( q l k ) \rho^{k}_{2d+l}=\iota^{1}(q^{k}_{l}) ρ 2 d + l k = ι 1 ( q l k ) for l ∈ [ d ] l\in[d] l ∈ [ d ] , with substitution σ ρ k : P 3 d → P 2 d \sigma_{\rho^{k}}:\mathcal{P}_{3d}\to\mathcal{P}_{2d} σ ρ k : P 3 d → P 2 d (Substitution of Noncommutative Polynomials into the Variables §substitution ); its entries are self-adjoint by Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §adjoint and Substitution is the Unique Unital Homomorphism with Prescribed Values on the Variables: Monomials, Products, Adjoints and Composition §adjoint . Let t k = ( s , p , q 1 k ( s ) , … , q d k ( s ) ) t^{k}=(s,p,q^{k}_{1}(s),\dots,q^{k}_{d}(s)) t k = ( s , p , q 1 k ( s ) , … , q d k ( s )) , a self-adjoint 3 d 3d 3 d -tuple in M M M by (C1), and u k = ( L ρ 1 k , … , L ρ 3 d k ) u^{k}=(L_{\rho^{k}_{1}},\dots,L_{\rho^{k}_{3d}}) u k = ( L ρ 1 k , … , L ρ 3 d k ) , a self-adjoint 3 d 3d 3 d -tuple in M π \mathcal{M}_{\pi} M π by (C4) and (C1). By (C1), t k t^{k} t k is the tuple of values of ρ k \rho^{k} ρ k at ( s , p ) (s,p) ( s , p ) (as ι 1 \iota^{1} ι 1 is the substitution of ( x 1 , … , x d ) (x_{1},\dots,x_{d}) ( x 1 , … , x d ) , Couplings of Two Noncommutative Laws and Their Quadratic Cost §marginals ), and by (C4) u k u^{k} u k is the tuple of values of ρ k \rho^{k} ρ k at ℓ π \ell^{\pi} ℓ π . Hence, for f ∈ P 3 d f\in\mathcal{P}_{3d} f ∈ P 3 d , f ( t k ) = ( σ ρ k f ) ( s , p ) f(t^{k})=(\sigma_{\rho^{k}}f)(s,p) f ( t k ) = ( σ ρ k f ) ( s , p ) and f ( u k ) = ( σ ρ k f ) ( ℓ π ) f(u^{k})=(\sigma_{\rho^{k}}f)(\ell^{\pi}) f ( u k ) = ( σ ρ k f ) ( ℓ π ) by (C1), and by Self-Adjoint Tuples in a Tracial W*-Probability Space and Their Laws §law and (C4)
λ t k ( f ) = λ ( s , p ) ( σ ρ k f ) = π ( σ ρ k f ) = λ ℓ π ( σ ρ k f ) = λ u k ( f ) . \lambda_{t^{k}}(f)=\lambda_{(s,p)}(\sigma_{\rho^{k}}f)=\pi(\sigma_{\rho^{k}}f)=\lambda_{\ell^{\pi}}(\sigma_{\rho^{k}}f)=\lambda_{u^{k}}(f). λ t k ( f ) = λ ( s , p ) ( σ ρ k f ) = π ( σ ρ k f ) = λ ℓ π ( σ ρ k f ) = λ u k ( f ) .
The vacuum tuple t k Ω = ( X , P , q k ( s ) Ω ) t^{k}\Omega=(X,P,q^{k}(s)\Omega) t k Ω = ( X , P , q k ( s ) Ω ) converges entrywise to ( X , P , Q ) (X,P,Q) ( X , P , Q ) by Step 4 with Y = Q Y=Q Y = Q . By (C4), u k Ω π = ( X π , P π , ι 1 ( q 1 k ) ^ , … , ι 1 ( q d k ) ^ ) u^{k}\Omega_{\pi}=(X_{\pi},P_{\pi},\widehat{\iota^{1}(q^{k}_{1})},\dots,\widehat{\iota^{1}(q^{k}_{d})}) u k Ω π = ( X π , P π , ι 1 ( q 1 k ) , … , ι 1 ( q d k ) ) , and ι 1 ( q l k ) ^ π = V π 1 q l k ^ λ \widehat{\iota^{1}(q^{k}_{l})}^{\,\pi}=V^{1}_{\pi}\widehat{q^{k}_{l}}^{\,\lambda} ι 1 ( q l k ) π = V π 1 q l k λ by Marginal Isometries, Bounded Plans and Displacement Pairings for Noncommutative Laws §isometries , as π ∘ ι 1 = λ \pi\circ\iota^{1}=\lambda π ∘ ι 1 = λ ; since V π 1 V^{1}_{\pi} V π 1 is linear with ( V π 1 ) ∗ V π 1 = I (V^{1}_{\pi})^{*}V^{1}_{\pi}=I ( V π 1 ) ∗ V π 1 = I (Marginals of a Noncommutative Law: the Isometry of GNS Spaces, the Trace-Preserving Embedding of Tracial Algebras and the Conditional Expectation §isometry ), ∥ V π 1 q l k ^ − V π 1 ζ l ∥ = ∥ q l k ^ − ζ l ∥ < 1 / k \lVert V^{1}_{\pi}\widehat{q^{k}_{l}}-V^{1}_{\pi}\zeta_{l}\rVert=\lVert\widehat{q^{k}_{l}}-\zeta_{l}\rVert<1/k ∥ V π 1 q l k − V π 1 ζ l ∥ = ∥ q l k − ζ l ∥ < 1/ k by Step 3, so u k Ω π u^{k}\Omega_{\pi} u k Ω π converges entrywise to ( X π , P π , V π 1 ζ ) (X_{\pi},P_{\pi},V^{1}_{\pi}\zeta) ( X π , P π , V π 1 ζ ) . By Square-Integrable Tuples in a Tracial W*-Probability Space: Their Norm, Affine Images, Pairs, Embedded Images and Laws §law , applied in ( H , M , Ω ) (H,M,\Omega) ( H , M , Ω ) to ( t k ) k (t^{k})_{k} ( t k ) k and in ( H π , M π , Ω π ) (\mathcal{H}_{\pi},\mathcal{M}_{\pi},\Omega_{\pi}) ( H π , M π , Ω π ) to ( u k ) k (u^{k})_{k} ( u k ) k , both l a w ( X , P , Q ) \mathrm{law}(X,P,Q) law ( X , P , Q ) and l a w ( X π , P π , V π 1 ζ ) \mathrm{law}(X_{\pi},P_{\pi},V^{1}_{\pi}\zeta) law ( X π , P π , V π 1 ζ ) are limits of the sequence ( κ 3 d ( λ t k ) ) k = ( κ 3 d ( λ u k ) ) k (\kappa_{3d}(\lambda_{t^{k}}))_{k}=(\kappa_{3d}(\lambda_{u^{k}}))_{k} ( κ 3 d ( λ t k ) ) k = ( κ 3 d ( λ u k ) ) k in ( Σ 3 d 2 , W ^ 2 ) (\Sigma^{2}_{3d},\widehat{W}_{2}) ( Σ 3 d 2 , W 2 ) , so they are equal by Uniqueness of Limits in a Metric Space .
Step 7 (claim 2: the shift law). Let π \pi π , P P P and Q Q Q be as in Step 6, let t t t be real, and let T T T be the affine datum from 3 d 3d 3 d to 2 d 2d 2 d variables with ( T Z ) i = Z i (TZ)_{i}=Z_{i} ( TZ ) i = Z i and ( T Z ) d + i = Z d + i + t Z 2 d + i (TZ)_{d+i}=Z_{d+i}+tZ_{2d+i} ( TZ ) d + i = Z d + i + t Z 2 d + i for i ∈ [ d ] i\in[d] i ∈ [ d ] (Square-Integrable Tuples in a Tracial W*-Probability Space: Their Norm, Affine Images, Pairs, Embedded Images and Laws §operations ). Then T ( X , P , Q ) = ( X , P + t Q ) T(X,P,Q)=(X,P+tQ) T ( X , P , Q ) = ( X , P + tQ ) and T ( X π , P π , V π 1 ζ ) = ( X π , P π + t V π 1 ζ ) T(X_{\pi},P_{\pi},V^{1}_{\pi}\zeta)=(X_{\pi},P_{\pi}+tV^{1}_{\pi}\zeta) T ( X π , P π , V π 1 ζ ) = ( X π , P π + t V π 1 ζ ) (Sums, Real Multiples and the Pairing of Square-Integrable Tuples in a Tracial W*-Probability Space §operations ), so (C2) and Step 6 give l a w ( X , P + t Q ) = l a w ( X π , P π + t V π 1 ζ ) = π ⊕ t ζ \mathrm{law}(X,P+tQ)=\mathrm{law}(X_{\pi},P_{\pi}+tV^{1}_{\pi}\zeta)=\pi\oplus t\zeta law ( X , P + tQ ) = law ( X π , P π + t V π 1 ζ ) = π ⊕ tζ by The Shift of a Bounded Plan by a Self-Adjoint Field §shift . This is the shift law of claim 2.
Step 8 (claim 2: the pairing). With π \pi π , P P P and Q Q Q as in Step 6, (C2) and Step 6, with the indices 2 d + j 2d+j 2 d + j and d + j d+j d + j , give ⟨ Q j , P j ⟩ = ⟨ V π 1 ζ j , x d + j ^ ⟩ \langle Q_{j},P_{j}\rangle=\langle V^{1}_{\pi}\zeta_{j},\widehat{x_{d+j}}\rangle ⟨ Q j , P j ⟩ = ⟨ V π 1 ζ j , x d + j ⟩ for j ∈ [ d ] j\in[d] j ∈ [ d ] (The Shift of a Bounded Plan by a Self-Adjoint Field §tuples ); summing, ⟨ Q , P ⟩ 2 = ∑ j = 1 d ⟨ V π 1 ζ j , x d + j ^ ⟩ \langle Q,P\rangle_{2}=\sum_{j=1}^{d}\langle V^{1}_{\pi}\zeta_{j},\widehat{x_{d+j}}\rangle ⟨ Q , P ⟩ 2 = ∑ j = 1 d ⟨ V π 1 ζ j , x d + j ⟩ by Sums, Real Multiples and the Pairing of Square-Integrable Tuples in a Tracial W*-Probability Space §pairing . The left side is real by the same clause, so it equals the real part of the right side, which is ∑ j Re ⟨ V π 1 ζ j , x d + j ^ ⟩ = J ( ζ , π ) \sum_{j}\operatorname{Re}\langle V^{1}_{\pi}\zeta_{j},\widehat{x_{d+j}}\rangle=\mathcal{J}(\zeta,\pi) ∑ j Re ⟨ V π 1 ζ j , x d + j ⟩ = J ( ζ , π ) by Marginal Isometries, Bounded Plans and Displacement Pairings for Noncommutative Laws §plan-pairing . This is the pairing of claim 2, which is now proved.
Step 9 (claim 3). Assume l a w ( X , Q ) = l a w ( X λ , ζ ) \mathrm{law}(X,Q)=\mathrm{law}(X_{\lambda},\zeta) law ( X , Q ) = law ( X λ , ζ ) . Then (C2), applied to ( X , Q ) (X,Q) ( X , Q ) and ( X λ , ζ ) (X_{\lambda},\zeta) ( X λ , ζ ) with both indices equal to d + j d+j d + j , gives ∥ Q j ∥ = ∥ ζ j ∥ \lVert Q_{j}\rVert=\lVert\zeta_{j}\rVert ∥ Q j ∥ = ∥ ζ j ∥ for every j ∈ [ d ] j\in[d] j ∈ [ d ] , hence ∥ Q ∥ 2 = ∥ ζ ∥ 2 \lVert Q\rVert_{2}=\lVert\zeta\rVert_{2} ∥ Q ∥ 2 = ∥ ζ ∥ 2 (Square-Integrable Tuples in a Tracial W*-Probability Space: Their Norm, Affine Images, Pairs, Embedded Images and Laws §tuples ). This proves claim 3.