Conventions. For ε∈{1,2} let Vε∈L(Hμ,Hγε) be the operator of Marginals of a Noncommutative Law: the Isometry of GNS Spaces, the Trace-Preserving Embedding of Tracial Algebras and the Conditional Expectation §isometry for γ=γε and a=aε (an operator, not an inner product space); by Two Noncommutative Laws with a Common Marginal: Standing Notation for Their Amalgamated Free Product §embeddings, πε and Eε are the embedding and the expectation of that lemma for the same data. By Marginals of a Noncommutative Law: the Isometry of GNS Spaces, the Trace-Preserving Embedding of Tracial Algebras and the Conditional Expectation §homomorphism and Marginals of a Noncommutative Law: the Isometry of GNS Spaces, the Trace-Preserving Embedding of Tracial Algebras and the Conditional Expectation §expectation, πε and Eε are linear and, for S,T∈N, c∈Aε and ζ,ζ′∈Hμ,
πε(I)=I,πε(ST)=πε(S)πε(T),πε(T∗)=πε(T)∗,Eε(c∗)=Eε(c)∗,Eε(πε(S)cπε(T))=SEε(c)T,⟨ζ,Eε(c)ζ′⟩=⟨Vεζ,cVεζ′⟩;
with T=I or S=I this gives Eε(πε(S)c)=SEε(c) and Eε(cπε(T))=Eε(c)T. The GNS spaces Hμ, Hγε are complex Hilbert spaces (The Complex GNS Space of a Tracial State on Noncommutative Polynomials §gns) and N, Aε consist of bounded operators on them (The Tracial Algebra of a Noncommutative Law and Its Trace §algebra), so all their elements have adjoints (Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint); N and Aε contain I and are closed under sums, scalar multiples, products and adjoints (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), hence are subspaces of the complex vector spaces of Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §operations, in which composition distributes over sums and commutes with scalar multiples. The rules (S+T)∗=S∗+T∗, (cT)∗=cT∗, (ST)∗=T∗S∗ and (T∗)∗=T of 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 (with Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-unique) are used freely.
As in The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §space, an alternating tuple t of length k also stands for the label (k,t); the labels (0,x), x∈N, are called N-labels. By The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §form:
(Z) h0(u,v)=0 unless u,v are both N-labels or are tuples of the same length and the same type; in particular h0(u,v)=0 if u,v are tuples whose types have different first entries.
Fix the label o=(0,I) and, for ξ∈F, put Fξ=suppξ∪{o}, a nonempty finite set by claim 1 of Peeling an Element off a Finite Set, and Unions of Finite Sets. A sum over a nonempty finite index set is a finite sum along one bijection from an initial segment (Sum over a Finite Index Set), so claims 2 and 3 of Properties of Finite Sums give
(S) ∑x∈F(f(x)+g(x))=∑x∈Ff(x)+∑x∈Fg(x) and ∑x∈Fcf(x)=c∑x∈Ff(x) for maps f,g:F→C and c∈C.
(C) Centred parts. Let c,d∈Aε. By The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §centred, (c∘)∗=c∗−πε(Eε(c))∗=c∗−πε(Eε(c)∗), so by the rules above
Eε((c∘)∗d)=Eε(c∗d)−Eε(c)∗Eε(d)=Eε(c∗d)−Eε(c∗πε(Eε(d)))=Eε(c∗d∘).
In particular, if c∈Aε∘ or d∈Aε∘, then Eε((c∘)∗d)=Eε(c∗d∘)=Eε(c∗d), as Eε(c)∗Eε(d)=0.
(X) Comparison of layers. Let s=(a1,…,ak), t=(b1,…,bk) be alternating tuples of length k and common type (e1,…,ek), let s′=(a1′,…,ak′′), t′=(b1′,…,bk′′) be alternating tuples of length k′ and common type (e1′,…,ek′′), and let d∈{0,1} with k+d≤k′. Suppose X1(s,t)=X1+d(s′,t′) and (ej,aj,bj)=(ej+d′,aj+d′,bj+d′) for every j∈[k] with j≥2. Then Xj(s,t)=Xj+d(s′,t′) for every j∈[k]. Indeed, let A be the set of j∈N with j>k or Xj(s,t)=Xj+d(s′,t′). Then 1∈A. If j∈A and j+1≤k, then 2≤j+1+d≤k′ and, by The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §nested,
Xj+1(s,t)=Eej+1(aj+1∗πej+1(Xj(s,t))bj+1)=Eej+1+d′(aj+1+d′∗πej+1+d′(Xj+d(s′,t′))bj+1+d′)=Xj+1+d(s′,t′);
if j+1>k, then j+1∈A trivially. So A=N by Principle of Induction for the Natural Numbers.
Claim 1 (vector space). By The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §space, sums and scalar multiples of elements of F lie in F, and the zero map lies in F (its support is empty). Two elements of F are equal when they agree at every label, and at a label s each of conditions 1--8 of Vector Space over a Field reduces to the corresponding field identity in C (associativity and commutativity of addition, ξ(s)+0=ξ(s), ξ(s)+(−1)ξ(s)=0, λ(μξ(s))=(λμ)ξ(s), 1ξ(s)=ξ(s) and the two distributive laws), with the zero map as zero vector and (−1)ξ as the vector w of condition 4. So F is a complex vector space with zero vector the zero map.
(E) For every ξ∈F and every nonempty finite F⊇suppξ, ξ=∑s∈Fξ(s)δs. At a label u the right side takes the value ∑s∈Fξ(s)δs(u) (The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §space). If u∈F, the terms with s=u vanish, so by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing and claim 1 of Peeling, Splitting, and Interchange for Sums over a Finite Index Set the value is ξ(u)δu(u)=ξ(u). If u∈/F, all terms vanish, so the value is 0 (Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing), which is ξ(u) because u∈/suppξ. For ξ=0 the set suppξ is nonempty and finite, and (E) with F=suppξ is the asserted expansion.
Claim 2 (symmetry of nested expectations). Let s=(a1,…,ak), t=(b1,…,bk) have type (e1,…,ek), and let A be the set of j∈N with j>k or Xj(t,s)=Xj(s,t)∗. By The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §nested and the rules of the conventions,
X1(t,s)=Ee1(b1∗a1)=Ee1((a1∗b1)∗)=Ee1(a1∗b1)∗=X1(s,t)∗,
so 1∈A. Let j∈A. If j+1>k, then j+1∈A. Otherwise Xj(t,s)=Xj(s,t)∗, and with e=ej+1
Xj+1(t,s)=Ee(bj+1∗πe(Xj(s,t)∗)aj+1)=Ee((aj+1∗πe(Xj(s,t))bj+1)∗)=Xj+1(s,t)∗.
So A=N by Principle of Induction for the Natural Numbers, which gives the claim for every j∈[k].
Claim 3 (positive Gram operators). HμM is a complex Hilbert space (Finite Direct Sums of a Complex Hilbert Space: the Hilbert Structure, Coordinate Inclusions, Block Entries of Bounded Operators and Commutation with Diagonal Operators §hilbert), so G∗∈L(HμM) exists (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 by Finite Direct Sums of a Complex Hilbert Space: the Hilbert Structure, Coordinate Inclusions, Block Entries of Bounded Operators and Commutation with Diagonal Operators §blocks and Claim 2 (with j=k)
(G∗)rq=(Gqr)∗=Xk(tq,tr)∗=Xk(tr,tq)=Grq(r,q∈[M]).
By the uniqueness in Finite Direct Sums of a Complex Hilbert Space: the Hilbert Structure, Coordinate Inclusions, Block Entries of Bounded Operators and Commutation with Diagonal Operators §blocks, G∗=G; so G is an adjoint of itself and hence self-adjoint (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 Complex Hilbert Spaces and Bounded Linear Maps: Standing Notation §maps it remains to show that G is positive semi-definite (Positive Semi-Definite Operator). We use:
(G) Let e∈{1,2} and D∈L(HγeM) be such that, for all r,q∈[M], (D∗D)rq∈Ae and Grq=Ee((D∗D)rq). Then ⟨ζ,Gζ⟩ is a real number ≥0 for every ζ∈HμM. Here D∗ exists as above. Put η=(Veζ1,…,VeζM)∈HγeM. By Finite Direct Sums of a Complex Hilbert Space: the Hilbert Structure, Coordinate Inclusions, Block Entries of Bounded Operators and Commutation with Diagonal Operators §blocks (twice), the identity for ⟨ζ,Ee(c)ζ′⟩ of the conventions, and 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,
⟨ζ,Gζ⟩=r=1∑Mq=1∑M⟨ζr,Ee((D∗D)rq)ζq⟩=r=1∑Mq=1∑M⟨Veζr,(D∗D)rqVeζq⟩=⟨η,D∗Dη⟩=∥Dη∥2≥0.
Moreover, by Finite Direct Sums of a Complex Hilbert Space: the Hilbert Structure, Coordinate Inclusions, Block Entries of Bounded Operators and Commutation with Diagonal Operators §blocks, (D∗D)rq=∑j=1M(Djr)∗Djq for every D.
Let K be the set of k∈N such that, for every M∈N and all alternating tuples t1,…,tM of length k and a common type, the operator G of the claim is positive semi-definite.
1∈K. Let tr=((e,ar)) for r∈[M]. Let D∈L(HγeM) have blocks D1q=aq and Djq=0 for j=1 (Finite Direct Sums of a Complex Hilbert Space: the Hilbert Structure, Coordinate Inclusions, Block Entries of Bounded Operators and Commutation with Diagonal Operators §blocks). The terms with j=1 of ∑j(Djr)∗Djq are 0, so by claim 7 of Properties of Finite Sums of Vectors, (D∗D)rq=(ar)∗aq∈Ae, and Grq=X1(tr,tq)=Ee((ar)∗aq) (The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §nested). By (G), G is positive semi-definite.
k∈K implies k+1∈K. Let t1,…,tM have length k+1 and common type (e1,…,ek+1); put e=ek+1, let ar be the last entry of tr, and let sr be the alternating tuple of length k and type (e1,…,ek) formed by the first k entries of tr. By (X) with d=0, Xk(sr,sq)=Xk(tr,tq). Let P∈L(HμM) have blocks Prq=Xk(sr,sq); since k∈K and P is self-adjoint by the first paragraph, P≥0. Each Prq lies in N, so commutes with Rp for every p∈Pn (The Tracial Algebra of a Noncommutative Law and Its Trace §algebra); by Finite Direct Sums of a Complex Hilbert Space: the Hilbert Structure, Coordinate Inclusions, Block Entries of Bounded Operators and Commutation with Diagonal Operators §diagonal, Rp(M)∈L(HμM) commutes with P. Let C be given by Square Roots of Positive Bounded Operators on a Complex Hilbert Space, Commuting with Everything that Commutes with the Operator for P: C≥0, CC=P (Square Roots of Positive Bounded Operators on a Complex Hilbert Space, Commuting with Everything that Commutes with the Operator §square-root), and C commutes with every Rp(M) (Square Roots of Positive Bounded Operators on a Complex Hilbert Space, Commuting with Everything that Commutes with the Operator §commutation). By Finite Direct Sums of a Complex Hilbert Space: the Hilbert Structure, Coordinate Inclusions, Block Entries of Bounded Operators and Commutation with Diagonal Operators §diagonal, each block Crq∈L(Hμ) commutes with every Rp, i.e. Crq∈N. As C is self-adjoint, it is its own adjoint (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, Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-unique), so Crq=(C∗)rq=(Cqr)∗ and Prq=(CC)rq=∑j=1M(Cjr)∗Cjq (Finite Direct Sums of a Complex Hilbert Space: the Hilbert Structure, Coordinate Inclusions, Block Entries of Bounded Operators and Commutation with Diagonal Operators §blocks). Let D∈L(HγeM) have blocks Djq=πe(Cjq)aq∈Ae. Then, using the properties of πe and claim 4 of Properties of Finite Sums of Vectors for the linear map T↦(ar)∗πe(T)aq on N,
(D∗D)rq=j=1∑M(ar)∗πe(Cjr)∗πe(Cjq)aq=j=1∑M(ar)∗πe((Cjr)∗Cjq)aq=(ar)∗πe(Prq)aq∈Ae,
and Grq=Xk+1(tr,tq)=Ee((ar)∗πe(Xk(tr,tq))aq)=Ee((D∗D)rq). By (G), G is positive semi-definite, so k+1∈K.
By Principle of Induction for the Natural Numbers, K=N; with self-adjointness, G≥0.
Claim 4 (the form). (H) For ξ,η∈F and nonempty finite sets F⊇suppξ, F′⊇suppη,
h(ξ,η)=s∈F∑t∈F′∑ξ(s)η(t)h0(s,t).
If ξ=0 or η=0, every term vanishes and both sides are 0 (Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing, The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §form). Otherwise, for each s the terms with t∈/suppη vanish, and for s∈/suppξ the whole inner sum vanishes; two applications of Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing reduce the right side to the defining double sum over suppξ×suppη.
Linearity in the second argument. With F=Fξ and F′=Fη∪Fζ, which contains suppη, suppζ and supp(η+ζ) (The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §space) and is finite by claim 3 of Peeling an Element off a Finite Set, and Unions of Finite Sets, (H) and (S) give
h(ξ,η+ζ)=s∈F∑t∈F′∑ξ(s)(η(t)+ζ(t))h0(s,t)=h(ξ,η)+h(ξ,ζ).
Likewise, with F′=Fη⊇supp(cη), (H) and (S) give h(ξ,cη)=ch(ξ,η).
Hermitian symmetry. First, h0(t,s)=h0(s,t) for all labels s,t. For N-labels s=(0,x), t=(0,y), 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 §trace, h0(t,s)=τμ(y∗x)=τμ((x∗y)∗)=τμ(x∗y). For tuples of the same length k and type, by Claim 2 and 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 §trace, h0(t,s)=τμ(Xk(s,t)∗)=τμ(Xk(s,t)). For all other pairs both values are 0, as the condition in (Z) is symmetric. Now take F=Fξ, F′=Fη. By (H), claim 5 of Peeling, Splitting, and Interchange for Sums over a Finite Index Set (interchange), and Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §conjugate with the rules for conjugates of products,
h(η,ξ)=s∈F∑t∈F′∑η(t)ξ(s)h0(s,t)=s∈F∑t∈F′∑ξ(s)η(t)h0(s,t)=h(ξ,η).
Basis values. δs,δt=0 with supports {s}, {t}, so by The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §form and claim 1 of Peeling, Splitting, and Interchange for Sums over a Finite Index Set, h(δs,δt)=1⋅1⋅h0(s,t)=h0(s,t).
Positivity. If ξ=0, then h(ξ,ξ)=0. Let ξ=0, F=suppξ, and f(s,t)=ξ(s)ξ(t)h0(s,t). Let κ((0,x))=0 and κ(t)=(k,type of t) for a tuple t of length k; by (Z), h0(s,t)=0 if κ(s)=κ(t). The set Γ={κ(s):s∈F} is nonempty and finite (claim 4 of Basic Properties of Finite Sets), and each Fγ={s∈F:κ(s)=γ}, γ∈Γ, is nonempty and finite. For s∈Fγ, ∑t∈Ff(s,t)=∑t∈Fγf(s,t) by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing. The map (γ,s)↦s is a bijection from the set of pairs (γ,s) with γ∈Γ, s∈Fγ onto F (inverse s↦(κ(s),s)), so by Sum over a Finite Index Set (composing bijections) and Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §pairs,
h(ξ,ξ)=γ∈Γ∑Qγ,Qγ=s∈Fγ∑t∈Fγ∑f(s,t).
Each Qγ is a real number ≥0. Enumerate Fγ by a bijection from [M] and compute the sums along it (Sum over a Finite Index Set). If γ=0, write F0={(0,x1),…,(0,xM)} and cr=ξ((0,xr)); by The Tracial Algebra of a Noncommutative Law and Its Trace §trace, the defining property of the adjoint, and claim 2 of Elementary Properties of a Complex Inner Product, crcqτμ(xr∗xq)=crcq⟨xrΩμ,xqΩμ⟩=⟨yr,yq⟩ with yr=crxrΩμ, so by claim 5 of Properties of Finite Sums of Vectors (twice) Q0=⟨y,y⟩=∥y∥2≥0 with y=∑r=1Myr. If γ=(k,e), write Fγ={t1,…,tM}, tuples of length k and common type e, and cr=ξ(tr); with G as in Claim 3 and ζ=(c1Ωμ,…,cMΩμ)∈HμM, The Tracial Algebra of a Noncommutative Law and Its Trace §trace, claim 2 of Elementary Properties of a Complex Inner Product and Finite Direct Sums of a Complex Hilbert Space: the Hilbert Structure, Coordinate Inclusions, Block Entries of Bounded Operators and Commutation with Diagonal Operators §blocks give
Qγ=r=1∑Mq=1∑Mcrcqτμ(Xk(tr,tq))=r=1∑Mq=1∑M⟨crΩμ,Grq(cqΩμ)⟩=⟨ζ,Gζ⟩,
a real number ≥0 by Claim 3 and Positive Semi-Definite Operator. Hence h(ξ,ξ) is a sum of nonnegative real numbers, a real number ≥0 by Nonnegativity and Monotonicity of a Sum over a Finite Index Set §nonnegative.
Claim 5 (formal actions). Write ρb=ρε,b, ℓ(b)=ℓε(b), π=πε, E=Eε. (L) For ξ∈F and nonempty finite F⊇suppξ, ℓ(b)ξ=∑u∈Fξ(u)ρb(u): at each label the terms with u∈/suppξ vanish, so by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §vanishing the value agrees with The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §actions (both are 0 if ξ=0). With F=Fξ∪Fη, (L) and (S) evaluated at each label give ℓ(b)(ξ+η)=ℓ(b)ξ+ℓ(b)η; with F=Fξ they give ℓ(b)(cξ)=cℓ(b)ξ. So ℓ(b) is linear (Linear Map, Claim 1). By (L) with F={u} and claim 1 of Peeling, Splitting, and Interchange for Sums over a Finite Index Set, ℓ(b)δu=ρb(u).
(B) Let U,W be nonempty finite sets, c:U→C, d:W→C, φ:U→F, ψ:W→F. Then
h(u∈U∑c(u)φ(u),w∈W∑d(w)ψ(w))=u∈U∑w∈W∑c(u)d(w)h(φ(u),ψ(w)).
Let S be the union of {o}, of the sets suppφ(u) and of the sets suppψ(w); it is finite by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §finite-union and claims 1 and 3 of Peeling an Element off a Finite Set, and Unions of Finite Sets, and it contains the supports of both combinations (The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §space). Apply (H) with F=F′=S to the left side; expand the conjugate of the first combination at s by Sums over Finite Index Sets: Finite Unions, Disjoint Unions, Vanishing Terms, Dependent Pairs, Conjugation and the Modulus §conjugate, multiply out by (S), interchange the sums over S with those over U and W (claim 5 of Peeling, Splitting, and Interchange for Sums over a Finite Index Set), and pull out c(u)d(w) by (S); the remaining double sum over S×S is h(φ(u),ψ(w)) by (H).
(D2) For labels u1,u2,v: h(δu1+δu2,δv)=h0(u1,v)+h0(u2,v) and h(δv,δu1+δu2)=h0(v,u1)+h0(v,u2). The second is additivity and the basis values of Claim 4; the first follows from it by Hermitian symmetry and h0(u,v)=h0(v,u) (Claim 4).
Reduction. Let ξ,η∈F. By (E), (L), (B) (with U=Fξ, W=Fη) and b∗∈Aε,
h(ℓ(b)ξ,η)=u∈Fξ∑v∈Fη∑ξ(u)η(v)h(ρb(u),δv),h(ξ,ℓ(b∗)η)=u∈Fξ∑v∈Fη∑ξ(u)η(v)h(δu,ρb∗(v)),
so it suffices to prove, for all labels u,v, the statement (⋆)(u,v): h(ρb(u),δv)=h(δu,ρb∗(v)) for every b∈Aε. If (⋆)(u,v) holds, so does (⋆)(v,u): for b∈Aε, (⋆)(u,v) for b∗ gives h(ρb∗(u),δv)=h(δu,ρb(v)) (as b∗∗=b), and conjugating with Hermitian symmetry gives h(δv,ρb∗(u))=h(ρb(v),δu). Every label is an N-label, a tuple whose type starts with ε, or one whose type starts with εˉ; so it suffices to treat the six cases below. Each side is evaluated by The Formal Amalgamated Free Product over a Common Marginal: Alternating Tuples, the Free Vector Space, the Nested-Expectation Form and the Formal Actions §actions and (D2), and terms are discarded by (Z).
Case u=(0,x), v=(0,y). Left side: h0((0,E(bπ(x))),(0,y))=τμ(E(bπ(x))∗y)=τμ(x∗E(b∗)y), since E(bπ(x))∗=E(π(x∗)b∗)=x∗E(b∗). Right side: h0((0,x),(0,E(b∗π(y))))=τμ(x∗E(b∗)y). The pairings involving the 1-tuples vanish.
Case u=(0,x), v=t=(a1,…,ak), type starting with ε. If k≥2, both sides vanish: ρb(u) consists of a 1-tuple and an N-label, and ρb∗(t) of two tuples. If k=1, by (C) (as a1∈Aε∘) the left side is
h0(((ε,(bπ(x))∘)),t)=τμ(E(((bπ(x))∘)∗a1))=τμ(E(π(x∗)b∗a1))=τμ(x∗E(b∗a1)),
and the right side is h0((0,x),(0,E(b∗a1)))=τμ(x∗E(b∗a1)).
Case u=(0,x), v=t with type starting with εˉ. The 1-tuple in ρb(u) has type (ε), and ρb∗(t) consists of two tuples; both sides are 0.
Case u=s=((e1,a1),…,(ek,ak)), v=t=((e1′,a1′),…,(el′,al′)), e1=e1′=ε. By clause (c), ρb(s)=δs′+δw with s′=((ε,(ba1)∘),(e2,a2),…,(ek,ak)) of the type of s, and w an N-label (k=1) or a tuple of type starting with e2=εˉ; likewise ρb∗(t)=δt′+δw′ with t′=((ε,(b∗a1′)∘),(e2′,a2′),…) of the type of t. So the left side is h0(s′,t) and the right side h0(s,t′), both 0 unless l=k and s,t have the same type. In that case, by (C) (as a1,a1′∈Aε∘), X1(s′,t)=E(a1∗b∗a1′)=X1(s,t′), and the pairs (s′,t), (s,t′) have the same entries in positions j≥2; so Xk(s′,t)=Xk(s,t′) by (X) with d=0, and the sides agree.
Case u=s as before (e1=ε), v=t=((e1′,a1′),…,(el′,al′)), e1′=εˉ. With ρb(s)=δs′+δw as before, h0(s′,t)=0. By clause (b), ρb∗(t)=δt++δt− with t+=((ε,(b∗)∘),(e1′,a1′),…,(el′,al′)) and t− of the type of t, so h0(s,t−)=0. The left side is h0(w,t), the right side h0(s,t+). If k≥2, w=((e2,πe2(E(ba1))a2),(e3,a3),…,(ek,ak)) has length k−1 and type (e2,…,ek); so both sides vanish unless (†) k=l+1 and (e2,…,ek)=(e1′,…,el′) (if k=1, w is an N-label and (†) fails as l≥1). Under (†), by (C) (as a1∈Aε∘) X1(s,t+)=E(a1∗b∗), so, as E(ba1)∗=E(a1∗b∗),
X2(s,t+)=Ee2(a2∗πe2(E(a1∗b∗))a1′)=Ee2((πe2(E(ba1))a2)∗a1′)=X1(w,t),
and for 2≤j≤k−1 the j-th entries of w,t are (ej+1,aj+1), (ej+1,aj′), the (j+1)-th entries of s,t+. By (X) with d=1, Xk−1(w,t)=Xk(s,t+), and the sides agree.
Case u=s=((e1,a1),…), v=t=((e1′,a1′),…), e1=e1′=εˉ. By clause (b), ρb(s)=δs++δs− and ρb∗(t)=δt++δt−, where s+,t+ have types starting with ε and s−=((e1,πe1(E(b))a1),(e2,a2),…), t−=((e1′,πe1(E(b∗))a1′),(e2′,a2′),…) have the types of s, t. So the left side is h0(s−,t) and the right side h0(s,t−), both 0 unless s,t have the same length k and type. In that case, as E(b)∗=E(b∗),
X1(s−,t)=Ee1(a1∗πe1(E(b))∗a1′)=Ee1(a1∗πe1(E(b∗))a1′)=X1(s,t−),
the later entries coincide, and (X) with d=0 gives Xk(s−,t)=Xk(s,t−).
In all cases (⋆) holds, which proves h(ℓε(b)ξ,η)=h(ξ,ℓε(b∗)η).