Let l and m be natural numbers with l≥2 and m≥1, let (L,G) be population cost data on l states with control dimension m, let Δl⊂Rl be the probability simplex, and let K≥0 be a real number. Points of Rl×Rm are written x=(Σ,α) and identified with points of Euclidean space Rl+m, with coordinates x1,…,xl+m, so that xγ=Σγ for γ≤l and xl+j=αj for j≤m; a product U′×V′ of sets U′⊆Rl and V′⊆Rm is regarded as a subset of Rl+m under this identification. For a real-valued function f on an open subset of a Euclidean space Rn and indices i,j∈{1,…,n} we write ∂if for the partial derivative of f with respect to the ith variable, which is unambiguous wherever it exists by Uniqueness of the Partial Derivative on a Euclidean Open Set, and ∂j∂if for the iterated partial derivative in the sense of clause 4 of that definition, namely ∂j applied to the function ∂if.
A triple (U,Lˉ,Gˉ) is a twice continuously differentiable extension of the population cost data (L,G) with second-derivative bound K if it consists of an open set U⊆Rl with Δl⊂U and functions Lˉ:U×Rm→R and Gˉ:U→R such that:
1. (Extension.) Lˉ(Σ,α)=L(Σ,α) for all (Σ,α)∈Δl×Rm, and Gˉ(Σ)=G(Σ) for all Σ∈Δl.
2. (Regularity.) Lˉ is of class C2 on U×Rm, which is an open subset of Rl+m by claims 1 and 2 of Products of Euclidean Open Sets are Open, U being open and Rm being open in Rm; and Gˉ is of class C2 on the open set U.
3. (Second-derivative bounds.) ∣∂j∂iLˉ(x)∣≤K for all i,j∈{1,…,l+m} and x∈U×Rm, and ∣∂γ′∂γGˉ(Σ)∣≤K for all γ,γ′∈{1,…,l} and Σ∈U.
4. (Uniform continuity of second derivatives.) For every real ε>0 there is a real δ>0 such that both of the following hold: ∣∂j∂iLˉ(x)−∂j∂iLˉ(y)∣≤ε for all i,j∈{1,…,l+m} and all x,y∈U×Rm whose Euclidean distance satisfies d(x,y)≤δ; and ∣∂γ′∂γGˉ(Σ)−∂γ′∂γGˉ(Σ′)∣≤ε for all γ,γ′∈{1,…,l} and all Σ,Σ′∈U with d(Σ,Σ′)≤δ.