TheoremBase

The law of a unitary tuple satisfies the four axioms by direct operator computation (positivity from the Gram matrix of the vectors V(v)Omega)V^{(v)}Omega). Conversely, a unitary law induces a law alpha of 2d self-adjoint variables via uju_j = xjx_j + i xd+jx_{d+j}; its free product with the semicircular law is realised on the GNS space, cancellation forces the UjU_j to be unitary, and the relative Schwinger-Dyson equation gives the centring and covariance of the semicircular family.

Proof

Each result cited below is universally quantified over the data in its own statement.

Letters. We use the following consequences of Words in Unitary Letters: Generators, Signs, Inverse Letters, Lengths and Adjoint Words §letters. For j∈[d]j\in[d] the letter jj has g(j)=jg(j)=j, ε(j)=1\varepsilon(j)=1 and j−1=d+jj^{-1}=d+j, and the letter d+jd+j (which lies in [2d][2d] and exceeds dd, by Words in Unitary Letters: Generators, Signs, Inverse Letters, Lengths and Adjoint Words §letters) has g(d+j)=jg(d+j)=j, ε(d+j)=−1\varepsilon(d+j)=-1 and (d+j)−1=j(d+j)^{-1}=j, because jj is the only natural number mm with d+j=d+md+j=d+m. Consequently, for every letter l∈[2d]l\in[2d]: g(l−1)=g(l)g(l^{-1})=g(l), ε(l−1)=−ε(l)\varepsilon(l^{-1})=-\varepsilon(l) and (l−1)−1=l(l^{-1})^{-1}=l; and l=g(l)l=g(l) if ε(l)=1\varepsilon(l)=1, l=d+g(l)l=d+g(l) if ε(l)=−1\varepsilon(l)=-1; and these two cases are exhaustive, since by Words in Unitary Letters: Generators, Signs, Inverse Letters, Lengths and Adjoint Words §letters the sign ε(l)\varepsilon(l) takes only the values 11 and −1-1. For j∈[d]j\in[d] and s∈{1,−1}s\in\{1,-1\} we write ℓ(j,s)\ell(j,s) for the letter jj if s=1s=1 and d+jd+j if s=−1s=-1; thus g(ℓ(j,s))=jg(\ell(j,s))=j, ε(ℓ(j,s))=s\varepsilon(\ell(j,s))=s, and ℓ(g(l),ε(l))=l\ell(g(l),\varepsilon(l))=l for every letter ll. For a word ww of length kk and a letter ll, the word w lw\,l has length k+1k+1 and adjoint (w l)∗=l−1 w∗(w\,l)^{*}=l^{-1}\,w^{*}: by Words in Unitary Letters: Generators, Signs, Inverse Letters, Lengths and Adjoint Words §adjoint and reversal, ((wl)∗)1=((wl)k+1)−1=l−1((wl)^{*})_{1}=((wl)_{k+1})^{-1}=l^{-1} and ((wl)∗)i=(wk+2−i)−1=(w∗)i−1((wl)^{*})_{i}=(w_{k+2-i})^{-1}=(w^{*})_{i-1} for 2≤i≤k+12\le i\le k+1; also l∗=l−1l^{*}=l^{-1} and ∅∗=∅\varnothing^{*}=\varnothing. Every nonempty word is of the form w lw\,l or is a letter, by Basic Properties of Words: Associativity, Reversal, Finitely Many Factorisations, and Countability §last-letter.

Fact U (unitaries). Let HH be a complex Hilbert space and T:H→HT:H\to H linear. Then TT is unitary if and only if T∈L(H)T\in\mathcal{L}(H) and T∗T=TT∗=IT^{*}T=TT^{*}=I. Indeed, let TT be unitary. By claim 3 of Properties of Unitary Operators and Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §least-bound, T∈L(H)T\in\mathcal{L}(H), so TT has an adjoint T∗∈L(H)T^{*}\in\mathcal{L}(H) by Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint, and TT is the adjoint of T∗T^{*} by Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus; by Adjoint of a Linear Map between Complex Inner Product Spaces §adjoint this gives ⟨u,T∗Tv⟩=⟨Tu,Tv⟩=⟨u,v⟩\langle u,T^{*}Tv\rangle=\langle Tu,Tv\rangle=\langle u,v\rangle for all u,v∈Hu,v\in H, the second equality being the defining property of TT. Taking u=T∗Tv−vu=T^{*}Tv-v gives ∥T∗Tv−v∥2=0\lVert T^{*}Tv-v\rVert^{2}=0, so T∗Tv=vT^{*}Tv=v. Given v∈Hv\in H, write v=Tuv=Tu by surjectivity; then TT∗v=T(T∗Tu)=Tu=vTT^{*}v=T(T^{*}Tu)=Tu=v. Conversely, if T∈L(H)T\in\mathcal{L}(H) and T∗T=TT∗=IT^{*}T=TT^{*}=I, then every vv equals T(T∗v)T(T^{*}v), and ⟨Tu,Tv⟩=⟨u,T∗Tv⟩=⟨u,v⟩\langle Tu,Tv\rangle=\langle u,T^{*}Tv\rangle=\langle u,v\rangle as above; so TT is unitary.

Proof of clause 1 (law). Fix (H,M,Ω)(H,M,\Omega) and VV as in clause 1, and write τ=τM\tau=\tau_{M}; by Tracial W*-Probability Spaces §trace and Cyclic Tracial Operator Algebras and Their Traces §trace, τ(T)=⟨Ω,TΩ⟩\tau(T)=\langle\Omega,T\Omega\rangle for T∈MT\in M. Each VjV_{j} is unitary, so Vj∈L(H)V_{j}\in\mathcal{L}(H) by claim 3 of Properties of Unitary Operators and Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §least-bound, and hence VjV_{j} has an adjoint Vj∗∈L(H)V_{j}^{*}\in\mathcal{L}(H) by Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint, as in Fact U.

(i) V(uv)=V(u)V(v)V^{(uv)}=V^{(u)}V^{(v)} for all u,v∈W2du,v\in W_{2d}. If u=∅u=\varnothing or v=∅v=\varnothing this holds because V(∅)=IV^{(\varnothing)}=I and ∅v=v\varnothing v=v, u∅=uu\varnothing=u (Words over a Finite Alphabet: the Empty Word, Concatenation and Reversal §concatenation); otherwise the letters of uvuv are those of uu followed by those of vv (Words over a Finite Alphabet: the Empty Word, Concatenation and Reversal §concatenation), and composition of maps is associative.

(ii) (V(l))∗=V(l−1)(V^{(l)})^{*}=V^{(l^{-1})} for every letter ll. If ε(l)=1\varepsilon(l)=1, then V(l)=Vg(l)V^{(l)}=V_{g(l)} and, as ε(l−1)=−1\varepsilon(l^{-1})=-1 and g(l−1)=g(l)g(l^{-1})=g(l), V(l−1)=Vg(l)∗V^{(l^{-1})}=V_{g(l)}^{*}. If ε(l)=−1\varepsilon(l)=-1, then V(l)=Vg(l)∗V^{(l)}=V_{g(l)}^{*}, whose adjoint is Vg(l)=V(l−1)V_{g(l)}=V^{(l^{-1})} by Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus.

(iii) (V(w))∗=V(w∗)(V^{(w)})^{*}=V^{(w^{*})} for every w∈W2dw\in W_{2d}. For w=∅w=\varnothing both sides are II (Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus); for a letter this is (ii); and if it holds for ww of length kk, then for a letter ll, by (i), Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus and (ii), (V(wl))∗=(V(w)V(l))∗=V(l−1)V(w∗)=V(l−1w∗)=V((wl)∗)(V^{(wl)})^{*}=(V^{(w)}V^{(l)})^{*}=V^{(l^{-1})}V^{(w^{*})}=V^{(l^{-1}w^{*})}=V^{((wl)^{*})}. Induction on the length gives (iii).

(iv) V(l)V(l−1)=IV^{(l)}V^{(l^{-1})}=I for every letter ll: by Fact U, Vg(l)Vg(l)∗=IV_{g(l)}V_{g(l)}^{*}=I and Vg(l)∗Vg(l)=IV_{g(l)}^{*}V_{g(l)}=I, and by (ii) the product V(l)V(l−1)V^{(l)}V^{(l^{-1})} is the first of these if ε(l)=1\varepsilon(l)=1 and the second if ε(l)=−1\varepsilon(l)=-1.

Now we check the four properties of Laws of d-Tuples of Unitaries §law. Normalised: λV(∅)=τ(I)=⟨Ω,Ω⟩=∥Ω∥2=1\lambda_{V}(\varnothing)=\tau(I)=\langle\Omega,\Omega\rangle=\lVert\Omega\rVert^{2}=1 by Cyclic Tracial Operator Algebras and Their Traces §cyclic. Cancellation: by (i) and (iv), V(u l l−1 v)=V(u)V(l)V(l−1)V(v)=V(u)V(v)=V(uv)V^{(u\,l\,l^{-1}\,v)}=V^{(u)}V^{(l)}V^{(l^{-1})}V^{(v)}=V^{(u)}V^{(v)}=V^{(uv)}, so λV(u l l−1 v)=λV(uv)\lambda_{V}(u\,l\,l^{-1}\,v)=\lambda_{V}(uv). Cyclic: by (i) and Cyclic Tracial Operator Algebras and Their Traces §tracial, λV(uv)=τ(V(u)V(v))=τ(V(v)V(u))=λV(vu)\lambda_{V}(uv)=\tau(V^{(u)}V^{(v)})=\tau(V^{(v)}V^{(u)})=\lambda_{V}(vu). Positive: let F⊆W2dF\subseteq W_{2d} be nonempty and finite, and put ξv=V(v)Ω\xi_{v}=V^{(v)}\Omega for v∈Fv\in F. By (i), (iii), and since V(v)V^{(v)} is the adjoint of (V(v))∗(V^{(v)})^{*} (Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus, Adjoint of a Linear Map between Complex Inner Product Spaces §adjoint),

K(v,v′):=λV(v∗v′)=⟨Ω,(V(v))∗V(v′)Ω⟩=⟨ξv,ξv′⟩(v,v′∈F).K(v,v'):=\lambda_{V}(v^{*}v')=\langle\Omega,(V^{(v)})^{*}V^{(v')}\Omega\rangle=\langle\xi_{v},\xi_{v'}\rangle\qquad(v,v'\in F).

Hence K(v′,v)=K(v,v′)‾K(v',v)=\overline{K(v,v')} by the conjugate symmetry of the inner product, and for z:F→Cz:F\to\mathbb{C}, with η=∑v∈Fz(v)ξv\eta=\sum_{v\in F}z(v)\xi_{v}, the sesquilinearity of the inner product (conjugate-linear in the first and linear in the second argument, Complex Hilbert Spaces and Bounded Linear Maps: Standing Notation §spaces) gives ∑(v,v′)∈F×Fz(v)‾ z(v′) K(v,v′)=⟨η,η⟩=∥η∥2\sum_{(v,v')\in F\times F}\overline{z(v)}\,z(v')\,K(v,v')=\langle\eta,\eta\rangle=\lVert\eta\rVert^{2}, a nonnegative real number. So KK is a positive semidefinite kernel on FF. Thus λV∈Ld\lambda_{V}\in\mathcal{L}_{d}.

Proof of clause 2 (realisation). Fix λ∈Ld\lambda\in\mathcal{L}_{d}. We use the notation of Noncommutative Laws, Couplings and the Wasserstein Distance: Standing Notation: Pn\mathcal{P}_{n}, monomials xvx_{v}, Σn,R\Sigma_{n,R}, Σn\Sigma_{n}, and, for a law, its complex GNS space, vacuum vector, classes p^\widehat{p}, multiplication operators LpL_{p} and tracial algebra (Noncommutative Laws, Couplings and the Wasserstein Distance: Standing Notation §polynomials, Noncommutative Laws, Couplings and the Wasserstein Distance: Standing Notation §laws, Noncommutative Laws, Couplings and the Wasserstein Distance: Standing Notation §gns). The monomials of P2d\mathcal{P}_{2d} are indexed by the words in the letters 1,…,2d1,\dots,2d, which are the elements of W2dW_{2d}: the words and their concatenation are defined by identical clauses in Words over a Finite Alphabet: the Empty Word, Concatenation and Reversal, used for W2dW_{2d}, and in the earlier version of the definition of words, used by the polynomial items, so this monomial indexing applies; so a variable index v∈[2d]v\in[2d] is also a letter and has a generator g(v)g(v) and a sign ε(v)\varepsilon(v). Every p∈Pnp\in\mathcal{P}_{n} satisfies

p=∑v∈supp⁡pp(v) xv(p≠0),(E)p=\sum_{v\in\operatorname{supp}p}p(v)\,x_{v}\qquad(p\neq0),\tag{E}

by the formula of Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §linear-extension(b) applied to the identity map of Pn\mathcal{P}_{n}, which is linear with value xvx_{v} at xvx_{v}; and xv=xv1xv2⋯xvkx_{v}=x_{v_{1}}x_{v_{2}}\cdots x_{v_{k}} for vv of length kk, by Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §monomials. The set {1,−1}k\{1,-1\}^{k} of kk-tuples of signs has 2k2^{k} elements (induction on kk), and for maps fm:{1,−1}→Cf_{m}:\{1,-1\}\to\mathbb{C} (m∈[k]m\in[k]), multiplying out by distributivity (induction on kk) gives

∑e∈{1,−1}k ∏m=1kfm(em)=∏m=1k(fm(1)+fm(−1));(D)\sum_{e\in\{1,-1\}^{k}}\ \prod_{m=1}^{k}f_{m}(e_{m})=\prod_{m=1}^{k}\bigl(f_{m}(1)+f_{m}(-1)\bigr);\tag{D}

the same multiplying out holds for products of kk sums of two terms in Pn\mathcal{P}_{n}, by the distributive and scalar rules of Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §algebra.

Step 1 (the polynomials blb_{l} and PωP_{\omega}). For a letter l∈[2d]l\in[2d] put bl=xg(l)+i ε(l) xd+g(l)∈P2db_{l}=x_{g(l)}+i\,\varepsilon(l)\,x_{d+g(l)}\in\mathcal{P}_{2d}, and for ω∈W2d\omega\in W_{2d} let Pω=bωP_{\omega}=b_{\omega} be the product along ω\omega of the 2d2d-tuple b=(b1,…,b2d)b=(b_{1},\dots,b_{2d}); thus P∅=1P_{\varnothing}=1, Pω=bω1⋯bωkP_{\omega}=b_{\omega_{1}}\cdots b_{\omega_{k}} for ω\omega of length kk, and Puv=PuPvP_{uv}=P_{u}P_{v} by Substitution is the Unique Unital Homomorphism with Prescribed Values on the Variables: Monomials, Products, Adjoints and Composition §values. By Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §adjoint, xj∗=xjx_{j}^{*}=x_{j} and iε(l)‾=−iε(l)\overline{i\varepsilon(l)}=-i\varepsilon(l), so bl∗=xg(l)−iε(l)xd+g(l)=bl−1b_{l}^{*}=x_{g(l)}-i\varepsilon(l)x_{d+g(l)}=b_{l^{-1}}, since g(l−1)=g(l)g(l^{-1})=g(l) and ε(l−1)=−ε(l)\varepsilon(l^{-1})=-\varepsilon(l). Hence Pω∗=Pω∗P_{\omega}^{*}=P_{\omega^{*}} for every ω\omega: this holds for ω=∅\omega=\varnothing (1∗=11^{*}=1) and for letters, and if it holds for ω\omega then (Pωl)∗=(Pωbl)∗=bl−1Pω∗=Pl−1ω∗=P(ωl)∗(P_{\omega l})^{*}=(P_{\omega}b_{l})^{*}=b_{l^{-1}}P_{\omega^{*}}=P_{l^{-1}\omega^{*}}=P_{(\omega l)^{*}} by (pq)∗=q∗p∗(pq)^{*}=q^{*}p^{*} (Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §adjoint); induct on the length.

For a variable index v∈[2d]v\in[2d] and s∈{1,−1}s\in\{1,-1\} put c(v,s)=12c(v,s)=\tfrac12 if ε(v)=1\varepsilon(v)=1 and c(v,s)=s2ic(v,s)=\tfrac{s}{2i} if ε(v)=−1\varepsilon(v)=-1. Then

xv=c(v,1) bℓ(g(v),1)+c(v,−1) bℓ(g(v),−1).(1)x_{v}=c(v,1)\,b_{\ell(g(v),1)}+c(v,-1)\,b_{\ell(g(v),-1)}.\tag{1}

Indeed, with j=g(v)j=g(v), bℓ(j,1)=xj+ixd+jb_{\ell(j,1)}=x_{j}+ix_{d+j} and bℓ(j,−1)=xj−ixd+jb_{\ell(j,-1)}=x_{j}-ix_{d+j}; if ε(v)=1\varepsilon(v)=1 then v=jv=j and the right side is 12(2xj)=xj\tfrac12(2x_{j})=x_{j}, and if ε(v)=−1\varepsilon(v)=-1 then v=d+jv=d+j and the right side is 12i(2i xd+j)=xd+j\tfrac{1}{2i}(2i\,x_{d+j})=x_{d+j}. For a word vv of length kk and s∈{1,−1}ks\in\{1,-1\}^{k} put c(v,s)=∏m=1kc(vm,sm)c(v,s)=\prod_{m=1}^{k}c(v_{m},s_{m}) and let ω(v,s)∈W2d\omega(v,s)\in W_{2d} be the word of length kk with mm-th letter ℓ(g(vm),sm)\ell(g(v_{m}),s_{m}); for v=∅v=\varnothing put c(∅,⋅)=1c(\varnothing,\cdot)=1 and ω(∅,⋅)=∅\omega(\varnothing,\cdot)=\varnothing, with a single (empty) sign tuple. Multiplying out xv=xv1⋯xvkx_{v}=x_{v_{1}}\cdots x_{v_{k}} with (1) gives

xv=∑s∈{1,−1}kc(v,s) Pω(v,s).(2)x_{v}=\sum_{s\in\{1,-1\}^{k}}c(v,s)\,P_{\omega(v,s)}.\tag{2}

Consequently, by (E), every p∈P2dp\in\mathcal{P}_{2d} with p≠0p\ne0 is a finite combination p=∑a∈Gz(a)Pωap=\sum_{a\in G}z(a)P_{\omega_{a}}, where GG is the nonempty finite set of pairs a=(v,s)a=(v,s) with v∈supp⁡pv\in\operatorname{supp}p and ss a sign tuple of the length of vv, z(a)=p(v)c(v,s)z(a)=p(v)c(v,s) and ωa=ω(v,s)\omega_{a}=\omega(v,s) (Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §pairs).

Step 2 (the law α\alpha). By Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §linear-extension(a) there is exactly one linear map α:P2d→C\alpha:\mathcal{P}_{2d}\to\mathbb{C} with α(1)=λ(∅)=1\alpha(1)=\lambda(\varnothing)=1 and

α(xv)=∑s∈{1,−1}kc(v,s) λ(ω(v,s))for every word v of length k∈N.\alpha(x_{v})=\sum_{s\in\{1,-1\}^{k}}c(v,s)\,\lambda\bigl(\omega(v,s)\bigr)\qquad\text{for every word }v\text{ of length }k\in\mathbb{N}.

Claim A: α(Pω)=λ(ω)\alpha(P_{\omega})=\lambda(\omega) for every ω∈W2d\omega\in W_{2d}. For ω=∅\omega=\varnothing this is α(1)=1=λ(∅)\alpha(1)=1=\lambda(\varnothing) (Laws of d-Tuples of Unitaries §normalised). Let ω\omega have length kk and jm=g(ωm)j_{m}=g(\omega_{m}). Since xℓ(j,1)=xjx_{\ell(j,1)}=x_{j} and xℓ(j,−1)=xd+jx_{\ell(j,-1)}=x_{d+j}, we have bωm=∑e∈{1,−1}βm(e) xℓ(jm,e)b_{\omega_{m}}=\sum_{e\in\{1,-1\}}\beta_{m}(e)\,x_{\ell(j_{m},e)} with βm(1)=1\beta_{m}(1)=1 and βm(−1)=iε(ωm)\beta_{m}(-1)=i\varepsilon(\omega_{m}); multiplying out,

Pω=∑e∈{1,−1}k(∏m=1kβm(em))xv(e),P_{\omega}=\sum_{e\in\{1,-1\}^{k}}\Bigl(\prod_{m=1}^{k}\beta_{m}(e_{m})\Bigr)x_{v(e)},

where v(e)v(e) is the word with mm-th letter ℓ(jm,em)\ell(j_{m},e_{m}). Now g(v(e)m)=jmg(v(e)_{m})=j_{m} and ε(v(e)m)=em\varepsilon(v(e)_{m})=e_{m}, so ω(v(e),s)\omega(v(e),s) is the word ωs\omega_{s} with mm-th letter ℓ(jm,sm)\ell(j_{m},s_{m}), independent of ee, and c(v(e)m,sm)c(v(e)_{m},s_{m}) is 12\tfrac12 if em=1e_{m}=1 and sm2i\tfrac{s_{m}}{2i} if em=−1e_{m}=-1. By linearity of α\alpha, exchanging the two finite sums, and (D),

α(Pω)=∑s∈{1,−1}kλ(ωs)∏m=1k(12+iε(ωm)sm2i)=∑s∈{1,−1}kλ(ωs)∏m=1k1+ε(ωm)sm2.\alpha(P_{\omega})=\sum_{s\in\{1,-1\}^{k}}\lambda(\omega_{s})\prod_{m=1}^{k}\Bigl(\tfrac12+i\varepsilon(\omega_{m})\tfrac{s_{m}}{2i}\Bigr)=\sum_{s\in\{1,-1\}^{k}}\lambda(\omega_{s})\prod_{m=1}^{k}\tfrac{1+\varepsilon(\omega_{m})s_{m}}{2}.

Each factor 1+ε(ωm)sm2\tfrac{1+\varepsilon(\omega_{m})s_{m}}{2} is 11 if sm=ε(ωm)s_{m}=\varepsilon(\omega_{m}) and 00 otherwise. So only the tuple s=(ε(ω1),…,ε(ωk))s=(\varepsilon(\omega_{1}),\dots,\varepsilon(\omega_{k})) contributes, with product 11 (Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing), and for it ωs=ω\omega_{s}=\omega because ℓ(g(l),ε(l))=l\ell(g(l),\varepsilon(l))=l. Hence α(Pω)=λ(ω)\alpha(P_{\omega})=\lambda(\omega).

Claim B: α∈Σ2d,2\alpha\in\Sigma_{2d,2}. α\alpha is linear and α(1)=1\alpha(1)=1. Let p,q∈P2dp,q\in\mathcal{P}_{2d}; if p=0p=0 or q=0q=0 then pq=qp=0pq=qp=0 and p∗p=0p^{*}p=0 (Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §algebra), so the conditions below hold trivially. Otherwise write p=∑a∈Gz(a)Pωap=\sum_{a\in G}z(a)P_{\omega_{a}} and q=∑b∈G′z′(b)Pωb′q=\sum_{b\in G'}z'(b)P_{\omega'_{b}} as in Step 1. By Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §algebra, PuPv=PuvP_{u}P_{v}=P_{uv}, Claim A and Laws of d-Tuples of Unitaries §cyclic,

α(pq)=∑(a,b)∈G×G′z(a)z′(b)λ(ωaωb′)=∑(a,b)∈G×G′z(a)z′(b)λ(ωb′ωa)=α(qp).\alpha(pq)=\sum_{(a,b)\in G\times G'}z(a)z'(b)\lambda(\omega_{a}\omega'_{b})=\sum_{(a,b)\in G\times G'}z(a)z'(b)\lambda(\omega'_{b}\omega_{a})=\alpha(qp).

By Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §adjoint and Pω∗=Pω∗P_{\omega}^{*}=P_{\omega^{*}}, p∗=∑a∈Gz(a)‾Pωa∗p^{*}=\sum_{a\in G}\overline{z(a)}P_{\omega_{a}^{*}}, so by Claim A α(p∗p)=∑(a,b)∈G×Gz(a)‾z(b)λ(ωa∗ωb)\alpha(p^{*}p)=\sum_{(a,b)\in G\times G}\overline{z(a)}z(b)\lambda(\omega_{a}^{*}\omega_{b}). Let F={ωa:a∈G}F=\{\omega_{a}:a\in G\}, nonempty and finite, and Z(ω)=∑a∈G, ωa=ωz(a)Z(\omega)=\sum_{a\in G,\ \omega_{a}=\omega}z(a) for ω∈F\omega\in F. Grouping the terms according to (ωa,ωb)(\omega_{a},\omega_{b}), that is, reindexing the finite sum over G×GG\times G as the iterated sum over (ω,ω′)∈F×F(\omega,\omega')\in F\times F and over the nonempty set of pairs (a,b)∈G×G(a,b)\in G\times G with ωa=ω\omega_{a}=\omega and ωb=ω′\omega_{b}=\omega' (Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §pairs, together with reindexing along a bijection, claim 2 of Properties of a Sum over a Finite Index Set), and using distributivity and Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §conjugate, gives α(p∗p)=∑(ω,ω′)∈F×FZ(ω)‾Z(ω′)λ(ω∗ω′)\alpha(p^{*}p)=\sum_{(\omega,\omega')\in F\times F}\overline{Z(\omega)}Z(\omega')\lambda(\omega^{*}\omega'), which is real and nonnegative by Laws of d-Tuples of Unitaries §positive and Positive Semidefinite Kernel on a Finite Set §kernel. So α\alpha is a tracial state on P2d\mathcal{P}_{2d}. For vv of length kk, ∣c(v,s)∣=(12)k|c(v,s)|=(\tfrac12)^{k} by claim 4 of Properties of Complex Conjugation and Modulus (as ∣s2i∣=12|\tfrac{s}{2i}|=\tfrac12), and ∣λ(ω(v,s))∣≤1|\lambda(\omega(v,s))|\le1 by Unitary Laws: Adjoints, the Bound One, and Reduction to Cyclically Reduced Words §bound; by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §modulus and the count 2k2^{k} of sign tuples, ∣α(xv)∣≤2k(12)k=1≤2k|\alpha(x_{v})|\le2^{k}(\tfrac12)^{k}=1\le2^{k}. So α\alpha has norm bound 22, i.e. α∈Σ2d,2⊆Σ2d\alpha\in\Sigma_{2d,2}\subseteq\Sigma_{2d}.

Step 3 (the joint law γ\gamma). Let scd\mathrm{sc}_{d} be the semicircular law of dd variables. It satisfies the Schwinger-Dyson equation, so it equals the vacuum law λS\lambda_{S} of The Vacuum Law of the Semicircular Operators is the Unique Noncommutative Law Satisfying the Schwinger-Dyson Equation §unique, which belongs to Σd,2\Sigma_{d,2} by The Vacuum Law of the Semicircular Operators is the Unique Noncommutative Law Satisfying the Schwinger-Dyson Equation §vacuum-law. Let γ=α⋆scd∈Σ3d\gamma=\alpha\star\mathrm{sc}_{d}\in\Sigma_{3d} be the free product for the split of 3d=2d+d3d=2d+d variables into the first 2d2d and the last dd, and let ι1:P2d→P3d\iota^{1}:\mathcal{P}_{2d}\to\mathcal{P}_{3d} be the substitution of (x1,…,x2d)(x_{1},\dots,x_{2d}) (Freeness of Two Groups of Variables under a Noncommutative Law §free). Then γ∘ι1=α\gamma\circ\iota^{1}=\alpha by The Free Product of Two Noncommutative Laws: Existence, Uniqueness, Norm Bound, Substitutions and Weak-Star Continuity §product, γ∈Σ3d,2\gamma\in\Sigma_{3d,2} by The Free Product of Two Noncommutative Laws: Existence, Uniqueness, Norm Bound, Substitutions and Weak-Star Continuity §bound with R=2R=2, and, by Free Semicircular Variables are Characterised by the Schwinger-Dyson Equation Relative to the Other Variables; Semicircular Systems and Rotations §characterisation (with m=2dm=2d, n=dn=d) and The Schwinger-Dyson Equation for a Noncommutative Law §relative,

γ(x2d+j q)=∂2d+jγ(q)(j∈[d], q∈P3d),(SD)\gamma(x_{2d+j}\,q)=\partial^{\gamma}_{2d+j}(q)\qquad(j\in[d],\ q\in\mathcal{P}_{3d}),\tag{SD}

with ∂2d+jγ\partial^{\gamma}_{2d+j} the free difference quotient. By Substitution is the Unique Unital Homomorphism with Prescribed Values on the Variables: Monomials, Products, Adjoints and Composition §values, Substitution is the Unique Unital Homomorphism with Prescribed Values on the Variables: Monomials, Products, Adjoints and Composition §homomorphism and Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §monomials, ι1\iota^{1} is linear and multiplicative with ι1(1)=1\iota^{1}(1)=1 and ι1(xv)=xv\iota^{1}(x_{v})=x_{v} for every v∈W2dv\in W_{2d}, the word vv being read as a word in the letters 1,…,3d1,\dots,3d all of whose letters are at most 2d2d; and ι1(p∗)=ι1(p)∗\iota^{1}(p^{*})=\iota^{1}(p)^{*} by Substitution is the Unique Unital Homomorphism with Prescribed Values on the Variables: Monomials, Products, Adjoints and Composition §adjoint, the variables being self-adjoint. Put yl=ι1(bl)=xg(l)+iε(l)xd+g(l)∈P3dy_{l}=\iota^{1}(b_{l})=x_{g(l)}+i\varepsilon(l)x_{d+g(l)}\in\mathcal{P}_{3d} and Qω=ι1(Pω)Q_{\omega}=\iota^{1}(P_{\omega}); thus Q∅=1Q_{\varnothing}=1, Qω=yω1⋯yωkQ_{\omega}=y_{\omega_{1}}\cdots y_{\omega_{k}}, QuQv=QuvQ_{u}Q_{v}=Q_{uv}, Qω∗=Qω∗Q_{\omega}^{*}=Q_{\omega^{*}}, and, by Claim A,

γ(Qω)=α(Pω)=λ(ω)(ω∈W2d).(3)\gamma(Q_{\omega})=\alpha(P_{\omega})=\lambda(\omega)\qquad(\omega\in W_{2d}).\tag{3}

Step 4 (the operators). Let (H,M,Ω)=(Hγ,Mγ,Ωγ)(H,M,\Omega)=(\mathcal{H}_{\gamma},\mathcal{M}_{\gamma},\Omega_{\gamma}), a tracial W*-probability space by The Tracial Algebra of a Noncommutative Law is a Tracial W*-Probability Space: the W*-Closure of the Left Multiplications §w-star, with Mγ=Aγ′′\mathcal{M}_{\gamma}=\mathcal{A}_{\gamma}'' where Aγ={Lp:p∈P3d}\mathcal{A}_{\gamma}=\{L_{p}:p\in\mathcal{P}_{3d}\}. Every LpL_{p} belongs to MM: it commutes with every element of Aγ′\mathcal{A}_{\gamma}' by The Commutant of a Set of Bounded Operators on a Complex Hilbert Space §commutant, so Lp∈Aγ′′L_{p}\in\mathcal{A}_{\gamma}''. The trace is τM(T)=⟨Ω,TΩ⟩\tau_{M}(T)=\langle\Omega,T\Omega\rangle (Tracial W*-Probability Spaces §trace, Cyclic Tracial Operator Algebras and Their Traces §trace). Since γ∈Σ3d,2\gamma\in\Sigma_{3d,2}, Left and Right Multiplication Operators on the Complex GNS Space of a Noncommutative Law: Boundedness, Algebra Rules, Adjoints, Commutation, the Vacuum and the Conjugation applies with r=2r=2: for p,q∈P3dp,q\in\mathcal{P}_{3d} and c∈Cc\in\mathbb{C}, Lp+q=Lp+LqL_{p+q}=L_{p}+L_{q}, Lcp=cLpL_{cp}=cL_{p}, Lpq=LpLqL_{pq}=L_{p}L_{q}, L1=IL_{1}=I (Left and Right Multiplication Operators on the Complex GNS Space of a Noncommutative Law: Boundedness, Algebra Rules, Adjoints, Commutation, the Vacuum and the Conjugation §algebra), Lp∗=Lp∗L_{p}^{*}=L_{p^{*}} (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), ∥Lxj∥op≤2\lVert L_{x_{j}}\rVert_{\mathrm{op}}\le2 (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), and LpΩ=p^L_{p}\Omega=\widehat{p}, ⟨p^,q^⟩=γ(p∗q)\langle\widehat{p},\widehat{q}\rangle=\gamma(p^{*}q), τM(Lp)=γ(p)\tau_{M}(L_{p})=\gamma(p) (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).

For j∈[d]j\in[d] put Uj=LyjU_{j}=L_{y_{j}}, where yj=xj+ixd+jy_{j}=x_{j}+ix_{d+j}, and Sj=Lx2d+jS_{j}=L_{x_{2d+j}}; both belong to MM. Each SjS_{j} is self-adjoint (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, x2d+jx_{2d+j} being self-adjoint) with ∥Sj∥op≤2\lVert S_{j}\rVert_{\mathrm{op}}\le2. For a letter ll we have U(l)=LylU^{(l)}=L_{y_{l}} in the notation of the statement: if ε(l)=1\varepsilon(l)=1 then l=g(l)l=g(l) and this is the definition; if ε(l)=−1\varepsilon(l)=-1 then l=d+jl=d+j with j=g(l)j=g(l) and U(l)=Uj∗=Lyj∗U^{(l)}=U_{j}^{*}=L_{y_{j}^{*}} with yj∗=xj−ixd+j=yd+jy_{j}^{*}=x_{j}-ix_{d+j}=y_{d+j}. By Lpq=LpLqL_{pq}=L_{p}L_{q} and L1=IL_{1}=I it follows that

U(ω)=LQω(ω∈W2d).(4)U^{(\omega)}=L_{Q_{\omega}}\qquad(\omega\in W_{2d}).\tag{4}

Unitarity. Let ll be a letter and r=Ql l−1−1r=Q_{l\,l^{-1}}-1. Since (l l−1)∗=(l−1)−1l−1=l l−1(l\,l^{-1})^{*}=(l^{-1})^{-1}l^{-1}=l\,l^{-1}, we get r∗=rr^{*}=r and r∗r=Ql l−1 l l−1−2Ql l−1+1r^{*}r=Q_{l\,l^{-1}\,l\,l^{-1}}-2Q_{l\,l^{-1}}+1 (Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §algebra). By (3) and Laws of d-Tuples of Unitaries §cancellation (with u=∅u=\varnothing and v=l l−1v=l\,l^{-1}, resp. u=v=∅u=v=\varnothing) and Laws of d-Tuples of Unitaries §normalised, γ(Ql l−1 l l−1)=λ(l l−1)=λ(∅)=1\gamma(Q_{l\,l^{-1}\,l\,l^{-1}})=\lambda(l\,l^{-1})=\lambda(\varnothing)=1, so ∥LrΩ∥2=⟨r^,r^⟩=γ(r∗r)=1−2+1=0\lVert L_{r}\Omega\rVert^{2}=\langle\widehat{r},\widehat{r}\rangle=\gamma(r^{*}r)=1-2+1=0. As Lr∈ML_{r}\in M and (H,M,Ω)(H,M,\Omega) is a cyclic tracial operator algebra, Lr=0L_{r}=0 by The Trace, the Conjugation and the Right Action of a Cyclic Tracial Operator Algebra §separating. By (4), U(l)U(l−1)=LQl l−1=Lr+L1=IU^{(l)}U^{(l^{-1})}=L_{Q_{l\,l^{-1}}}=L_{r}+L_{1}=I. For l=jl=j this reads UjUj∗=IU_{j}U_{j}^{*}=I, and for l=d+jl=d+j it reads Uj∗Uj=IU_{j}^{*}U_{j}=I. By Fact U, every UjU_{j} is unitary. By (4) and (3), λU(ω)=τM(LQω)=γ(Qω)=λ(ω)\lambda_{U}(\omega)=\tau_{M}(L_{Q_{\omega}})=\gamma(Q_{\omega})=\lambda(\omega) for every ω\omega, i.e. λU=λ\lambda_{U}=\lambda.

Elements of AU\mathcal{A}_{U}. Let Z∈AUZ\in\mathcal{A}_{U}, say Z=∑r=1NcrU(ωr)Z=\sum_{r=1}^{N}c_{r}U^{(\omega_{r})}. By (4) and the algebra rules, Z=LqZ=L_{q} with q=∑rcrQωr=ι1(q~)q=\sum_{r}c_{r}Q_{\omega_{r}}=\iota^{1}(\tilde q), q~=∑rcrPωr∈P2d\tilde q=\sum_{r}c_{r}P_{\omega_{r}}\in\mathcal{P}_{2d}. If q~≠0\tilde q\neq0, then by (E) and the properties of ι1\iota^{1},

q=∑v∈supp⁡q~q~(v) xv,τM(Z)=γ(q)=∑v∈supp⁡q~q~(v) γ(xv),(5)q=\sum_{v\in\operatorname{supp}\tilde q}\tilde q(v)\,x_{v},\qquad\tau_{M}(Z)=\gamma(q)=\sum_{v\in\operatorname{supp}\tilde q}\tilde q(v)\,\gamma(x_{v}),\tag{5}

where every vv is a word all of whose letters are at most 2d2d; if q~=0\tilde q=0 then Z=L0=0Z=L_{0}=0.

(a) Centred. Let i∈[d]i\in[d] and Z=LqZ=L_{q} as above. Then τM(SiZ)=τM(Lx2d+iq)=γ(x2d+iq)=∂2d+iγ(q)\tau_{M}(S_{i}Z)=\tau_{M}(L_{x_{2d+i}q})=\gamma(x_{2d+i}q)=\partial^{\gamma}_{2d+i}(q) by (SD). If q~=0\tilde q=0 this is 00. Otherwise, by (5) and linearity, it is ∑vq~(v) ∂2d+iγ(xv)\sum_{v}\tilde q(v)\,\partial^{\gamma}_{2d+i}(x_{v}); here ∂2d+iγ(x∅)=∂2d+iγ(1)=0\partial^{\gamma}_{2d+i}(x_{\varnothing})=\partial^{\gamma}_{2d+i}(1)=0, and for vv of length kk every term of the sum defining ∂2d+iγ(xv)\partial^{\gamma}_{2d+i}(x_{v}) in The Free Difference Quotient and the Conjugate Variables of a Noncommutative Law §difference-quotient carries the factor δvm,2d+i=0\delta_{v_{m},2d+i}=0, since vm≤2d<2d+iv_{m}\le2d<2d+i. So τM(SiZ)=0\tau_{M}(S_{i}Z)=0.

(b) Covariance. Let i,l∈[d]i,l\in[d] and Z=LqZ=L_{q}, Z′=Lq′Z'=L_{q'} as above. Then τM(SiZSlZ′)=γ(x2d+i q x2d+l q′)=∂2d+iγ(q x2d+l q′)\tau_{M}(S_{i}ZS_{l}Z')=\gamma(x_{2d+i}\,q\,x_{2d+l}\,q')=\partial^{\gamma}_{2d+i}(q\,x_{2d+l}\,q') by (SD). If q~=0\tilde q=0 or q~′=0\tilde q'=0, both this and δilτM(Z)τM(Z′)\delta_{il}\tau_{M}(Z)\tau_{M}(Z') vanish. Otherwise, by (5), the algebra rules and xvx2d+lxv′=xv (2d+l) v′x_{v}x_{2d+l}x_{v'}=x_{v\,(2d+l)\,v'} (Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §monomials),

∂2d+iγ(q x2d+l q′)=∑v∈supp⁡q~ ∑v′∈supp⁡q~′q~(v)q~′(v′) ∂2d+iγ(xv (2d+l) v′).\partial^{\gamma}_{2d+i}(q\,x_{2d+l}\,q')=\sum_{v\in\operatorname{supp}\tilde q}\ \sum_{v'\in\operatorname{supp}\tilde q'}\tilde q(v)\tilde q'(v')\,\partial^{\gamma}_{2d+i}\bigl(x_{v\,(2d+l)\,v'}\bigr).

The word w=v (2d+l) v′w=v\,(2d+l)\,v' has the letter 2d+l2d+l at the position n0n_{0} following the letters of vv, while all its other letters are at most 2d2d and so differ from 2d+i2d+i. Hence in The Free Difference Quotient and the Conjugate Variables of a Noncommutative Law §difference-quotient only the term at n0n_{0} can be nonzero; it carries δ2d+l,2d+i=δil\delta_{2d+l,2d+i}=\delta_{il}, and there w<n0=vw_{<n_{0}}=v and w>n0=v′w_{>n_{0}}=v'. So ∂2d+iγ(xv(2d+l)v′)=δilγ(xv)γ(xv′)\partial^{\gamma}_{2d+i}(x_{v(2d+l)v'})=\delta_{il}\gamma(x_{v})\gamma(x_{v'}), and summing with (5),

τM(SiZSlZ′)=δil(∑vq~(v)γ(xv))(∑v′q~′(v′)γ(xv′))=δil τM(Z) τM(Z′).\tau_{M}(S_{i}ZS_{l}Z')=\delta_{il}\Bigl(\sum_{v}\tilde q(v)\gamma(x_{v})\Bigr)\Bigl(\sum_{v'}\tilde q'(v')\gamma(x_{v'})\Bigr)=\delta_{il}\,\tau_{M}(Z)\,\tau_{M}(Z').

This proves clause 2 with (H,M,Ω)(H,M,\Omega), U=(U1,…,Ud)U=(U_{1},\dots,U_{d}) and S1,…,SdS_{1},\dots,S_{d}. ■\blacksquare

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…