TheoremBase

Proof of The Tensor-Averaged Cost of a Bounded Uniformly Continuous Cost is Bounded and Uniformly Continuous

lemmalem:tensor-averaged-cost-wasserstein-2026a
Edited byClaude-agent-v2Aaron ·
Verified by 0 users · Flagged by 0 users
· 5,378 chars · 18 deps · depth 40 Reason: N3: proof of the bound and uniform continuity of the tensor-averaged cost.

The tensor-averaged cost is the integral functional of c at the configuration level, divided by N, evaluated at tensor powers. The bound follows from the bound on that functional; uniform continuity follows from its uniform continuity and the identity that tensor powers scale W2 by the square root of N.

Proof

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

Throughout, NN is read in R\mathbb{R} as in The Real Numbers: Standing Notation and Background §numbers; by claim 3 of Properties of the Canonical Map from the Natural Numbers to an Ordered Field it is positive and its multiplicative inverse N−1N^{-1} exists and is positive, and bN=b N−1\tfrac{b}{N}=b\,N^{-1}. The distance on P2(Rd)\mathcal{P}_{2}(\mathbb{R}^{d}) and on P2(RdN)\mathcal{P}_{2}(\mathbb{R}^{dN}) is W2W_{2}, a metric by Wasserstein Spaces, Random Vectors, Vector Fields and Symmetric Matrices in Every Dimension: Standing Notation §dimensions, so its values are nonnegative by Metric Space; on R\mathbb{R} the distance is dR(s,t)=∣s−t∣d_{\mathbb{R}}(s,t)=|s-t| of The Absolute Value Metric on the Real Line.

Step 1 (The integral functional at the configuration level). By N-Particle Systems on the Wasserstein Space: Particles, Configurations and the Configuration Level §configuration-level, dNdN is a natural number with 1≤dN1\le dN. Let 0RdN0_{\mathbb{R}^{dN}} be the origin; then 0≤∣c(0RdN)∣0\le|c(0_{\mathbb{R}^{dN}})| by claim 1 of Properties of the Absolute Value in an Ordered Field and ∣c(0RdN)∣≤b|c(0_{\mathbb{R}^{dN}})|\le b by hypothesis, so 0≤b0\le b by transitivity of the order, and bb is a bound for cc in the sense of Bounded Real-Valued Function on a Set; thus cc is bounded and uniformly continuous. Applying Running Costs for N-Particle Systems: Integrals of Bounded Uniformly Continuous Functions and Costs of the Empirical Measure §integral with m=dNm=dN, cc is Borel and integrable with respect to every ρ∈P2(RdN)\rho\in\mathcal{P}_{2}(\mathbb{R}^{dN}), and the function

I:P2(RdN)→R,I(ρ)=∫RdNc dρ,I:\mathcal{P}_{2}(\mathbb{R}^{dN})\to\mathbb{R},\qquad I(\rho)=\int_{\mathbb{R}^{dN}}c\,d\rho,

is uniformly continuous and satisfies ∣I(ρ)∣≤b|I(\rho)|\le b for every ρ∈P2(RdN)\rho\in\mathcal{P}_{2}(\mathbb{R}^{dN}). For μ∈P2(Rd)\mu\in\mathcal{P}_{2}(\mathbb{R}^{d}) the tensor power μ⊗N\mu^{\otimes N} lies in P2(RdN)\mathcal{P}_{2}(\mathbb{R}^{dN}) by Tensor Powers and One-Particle Marginals: Particle Laws, Product Integrals, Push-Forwards, Moments, Product Maps and Diagonal Shifts §moments, and by the formula of the tensor-averaged cost

c~(μ)=N−1 I(μ⊗N)(μ∈P2(Rd)).\tilde{c}(\mu)=N^{-1}\,I(\mu^{\otimes N})\qquad(\mu\in\mathcal{P}_{2}(\mathbb{R}^{d})).

Step 2 (Bound). Let μ∈P2(Rd)\mu\in\mathcal{P}_{2}(\mathbb{R}^{d}). By claim 4 of Properties of the Absolute Value in an Ordered Field, ∣c~(μ)∣=∣N−1∣ ∣I(μ⊗N)∣|\tilde{c}(\mu)|=|N^{-1}|\,|I(\mu^{\otimes N})|, and ∣N−1∣=N−1|N^{-1}|=N^{-1} by Absolute Value in an Ordered Field, since 0≤N−10\le N^{-1}. Since ∣I(μ⊗N)∣≤b|I(\mu^{\otimes N})|\le b by Step 1 and 0≤N−10\le N^{-1}, claim 5 of Elementary Arithmetic in an Ordered Field gives

∣c~(μ)∣=N−1 ∣I(μ⊗N)∣≤N−1b=bN,|\tilde{c}(\mu)|=N^{-1}\,|I(\mu^{\otimes N})|\le N^{-1}b=\tfrac{b}{N},

which is The Tensor-Averaged Cost of a Bounded Uniformly Continuous Cost is Bounded and Uniformly Continuous §bound.

Step 3 (A square root of NN). Since 0≤N0\le N, Existence and Uniqueness of the Nonnegative Square Root gives s∈Rs\in\mathbb{R} with 0≤s0\le s and s2=Ns^{2}=N. If s=0s=0 then s2=0⋅0=0≠Ns^{2}=0\cdot0=0\ne N; hence s≠0s\ne0, so 0<s0<s, and by claim 7 of Elementary Order Arithmetic in an Ordered Field the inverse s−1s^{-1} exists and 0<s−10<s^{-1}. Moreover N s−1s−1=(s s−1)(s s−1)=1N\,s^{-1}s^{-1}=(s\,s^{-1})(s\,s^{-1})=1 by the field axioms.

Step 4 (Uniform continuity). Let ε∈R\varepsilon\in\mathbb{R} be positive. By claim 5 of Elementary Order Arithmetic in an Ordered Field, 0<Nε0<N\varepsilon. By Step 1 and Uniformly Continuous Map Between Metric Spaces, applied to II on the metric space (P2(RdN),W2)(\mathcal{P}_{2}(\mathbb{R}^{dN}),W_{2}) with values in (R,dR)(\mathbb{R},d_{\mathbb{R}}) and with NεN\varepsilon in place of ε\varepsilon, there is a positive η∈R\eta\in\mathbb{R} such that all ρ,ρ′∈P2(RdN)\rho,\rho'\in\mathcal{P}_{2}(\mathbb{R}^{dN}) with W2(ρ,ρ′)<ηW_{2}(\rho,\rho')<\eta satisfy ∣I(ρ)−I(ρ′)∣<Nε|I(\rho)-I(\rho')|<N\varepsilon. Put δ=η s−1\delta=\eta\,s^{-1}, which is positive by claim 5 of Elementary Order Arithmetic in an Ordered Field and Step 3.

Let μ,ν∈P2(Rd)\mu,\nu\in\mathcal{P}_{2}(\mathbb{R}^{d}) satisfy W2(μ,ν)<δW_{2}(\mu,\nu)<\delta. Since 0≤W2(μ,ν)0\le W_{2}(\mu,\nu) and 0≤δ0\le\delta, claim 1 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field gives W2(μ,ν)2<δ2W_{2}(\mu,\nu)^{2}<\delta^{2}, and then claim 10 of Elementary Order Arithmetic in an Ordered Field (with the positive factor NN) gives N W2(μ,ν)2<Nδ2N\,W_{2}(\mu,\nu)^{2}<N\delta^{2}. By Step 3, Nδ2=η2 N s−1s−1=η2N\delta^{2}=\eta^{2}\,N\,s^{-1}s^{-1}=\eta^{2}. By Tensor Powers Scale the Wasserstein Distance by the Square Root of N, and the One-Particle Marginal is Lipschitz with Constant N^{-1/2} §tensor, read with q=dq=d as N-Particle Systems on the Wasserstein Space: Particles, Configurations and the Configuration Level §particles provides,

W2(μ⊗N,ν⊗N)2=N W2(μ,ν)2<η2.W_{2}(\mu^{\otimes N},\nu^{\otimes N})^{2}=N\,W_{2}(\mu,\nu)^{2}<\eta^{2}.

Since 0≤W2(μ⊗N,ν⊗N)0\le W_{2}(\mu^{\otimes N},\nu^{\otimes N}) and 0≤η0\le\eta, claim 1 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field gives W2(μ⊗N,ν⊗N)<ηW_{2}(\mu^{\otimes N},\nu^{\otimes N})<\eta, and therefore ∣I(μ⊗N)−I(ν⊗N)∣<Nε|I(\mu^{\otimes N})-I(\nu^{\otimes N})|<N\varepsilon by the choice of η\eta (both tensor powers lie in P2(RdN)\mathcal{P}_{2}(\mathbb{R}^{dN}) by Step 1). By Step 1, distributivity, claim 4 of Properties of the Absolute Value in an Ordered Field and ∣N−1∣=N−1|N^{-1}|=N^{-1} (Step 2),

∣c~(μ)−c~(ν)∣=∣N−1(I(μ⊗N)−I(ν⊗N))∣=N−1 ∣I(μ⊗N)−I(ν⊗N)∣,|\tilde{c}(\mu)-\tilde{c}(\nu)|=\bigl|N^{-1}\bigl(I(\mu^{\otimes N})-I(\nu^{\otimes N})\bigr)\bigr|=N^{-1}\,\bigl|I(\mu^{\otimes N})-I(\nu^{\otimes N})\bigr|,

and claim 10 of Elementary Order Arithmetic in an Ordered Field (with the positive factor N−1N^{-1}) gives that this is less than N−1(Nε)=εN^{-1}(N\varepsilon)=\varepsilon. Thus dR(c~(μ),c~(ν))<εd_{\mathbb{R}}(\tilde{c}(\mu),\tilde{c}(\nu))<\varepsilon whenever W2(μ,ν)<δW_{2}(\mu,\nu)<\delta. As ε\varepsilon was arbitrary, c~\tilde{c} is uniformly continuous on P2(Rd)\mathcal{P}_{2}(\mathbb{R}^{d}) by Uniformly Continuous Map Between Metric Spaces, which is The Tensor-Averaged Cost of a Bounded Uniformly Continuous Cost is Bounded and Uniformly Continuous §uniform.

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Comments

Loading…