Let N be the set of natural numbers, C the field of complex numbers and R⊆C the real numbers. Let d∈N and for n∈N let Pn=C⟨x1,…,xn⟩ be the noncommutative polynomials, with unit 1, monomials xw indexed by the words w∈Wn, and substitutions σa. Let Σn,R be the set of tracial states on Pn with norm bound R and Σn the set of noncommutative laws of n variables, weak-star convergence as in Weak-Star Convergence of Noncommutative Laws §weak-star, and let ι1,ι2, Π(μ,ν) and I be the marginal substitutions, the sets of couplings and the cost. For q in Pn write q2=qq, and for real s write s2=ss. For a tracial state λ on Pd write M(λ)=∑j=1dλ(xj2). Let μ,ν∈Σd and let R>0 be real.
1. (Norm bound)¶ If μ,ν∈Σd,R, then Π(μ,ν)⊆Σ2d,R.
2. (Cost)¶ M(μ) and M(ν) are real and nonnegative, and for every γ∈Π(μ,ν) the numbers γ(xjxd+j) are real and
I(γ)=M(μ)+M(ν)−2j=1∑dγ(xjxd+j),0≤I(γ)≤2M(μ)+2M(ν).
3. (Tensor coupling)¶ Let σ1 and σ2 be the substitutions, from P2d to Pd, of the 2d-tuples (x1,…,xd,1,…,1) and (1,…,1,x1,…,xd) in Pd. There is exactly one linear map μ⊗ν:P2d→C with (μ⊗ν)(xw)=μ(σ1(xw))ν(σ2(xw)) for every w∈W2d (by Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §linear-extension), and μ⊗ν∈Π(μ,ν). In particular Π(μ,ν) is nonempty.
4. (Diagonal coupling)¶ Let δ be the substitution, from P2d to Pd, of the 2d-tuple (x1,…,xd,x1,…,xd). Then μ∘δ∈Π(μ,μ) and I(μ∘δ)=0.
5. (Swapped coupling)¶ Let s be the substitution, from P2d to P2d, of the 2d-tuple (xd+1,…,x2d,x1,…,xd). If γ∈Π(μ,ν), then γ∘s∈Π(ν,μ) and I(γ∘s)=I(γ).
6. (Weak-star closedness)¶ Let (μm) and (νm) be sequences in Σd,R converging weak-star to μ and ν, and let γm∈Π(μm,νm) for every m∈N. Then there are a strictly increasing sequence (mi)i∈N in N and γ∈Π(μ,ν) such that γmi→γ weak-star and the real sequence (I(γmi))i∈N converges to I(γ).
7. (Displacement interpolants)¶ Let μ,ν∈Σd,R, γ∈Π(μ,ν) and t∈R with 0≤t≤1. Let at be the d-tuple in P2d with ajt=(1−t)xj+txd+j, and put γt=γ∘σat. Then γt∈Σd,R, γ0=μ and γ1=ν. Moreover, γ∘σ(x1,…,xd,a1t,…,adt)∈Π(μ,γt) with cost t2I(γ), and γ∘σ(a1t,…,adt,xd+1,…,x2d)∈Π(γt,ν) with cost (1−t)2I(γ).