TheoremBase

Continuity of the Coordinate Projections and of the Quadratic Cost Function, and Passage of a Cost Bound to a Weak Limit

The coordinate projections of a Euclidean product are nonexpansive and the quadratic cost function is continuous; consequently a bound on the quadratic cost that holds along a weakly convergent sequence of probability measures holds in the limit.

Statement

In the setting of Probability Measures on Euclidean Space and Random Vectors: Standing Notation, let m∈Nm\in\mathbb{N} satisfy 1≤m1\le m and let dEd_{E} denote the Euclidean distance, on Rm\mathbb{R}^{m} and on Rm+m\mathbb{R}^{m+m} as the dimension of its arguments requires, so that dE(x,y)=∥x−y∥d_{E}(x,y)=\lVert x-y\rVert by claim 2 of Properties of the Euclidean Norm on Rn\mathbb{R}^n: Square, Distance, Vanishing, Coordinate Bounds, Homogeneity, Triangle Inequalities and Expansion of a Square. Let R\mathbb{R} carry the absolute-value metric, and let continuity and the Lipschitz property be those of the metric spaces so named. As in Probability Measures on Euclidean Space and Random Vectors: Standing Notation (v), pr1,pr2:Rm+m→Rm\mathrm{pr}_{1},\mathrm{pr}_{2}:\mathbb{R}^{m+m}\to\mathbb{R}^{m} are the coordinate projections for the splitting m+mm+m.

Let φ:Rm+m→R\varphi:\mathbb{R}^{m+m}\to\mathbb{R} be the map with φ(z)=∥pr1(z)−pr2(z)∥2\varphi(z)=\lVert\mathrm{pr}_{1}(z)-\mathrm{pr}_{2}(z)\rVert^{2}, which is Borel by Pairs of Euclidean Points: Coordinate Projections, Pairings, the Product Measure on a Euclidean Space, Borel Norm Functions and Finite Sets (iv) and nonnegative by Nonnegativity of Squares in an Ordered Field, being a square, and for π∈P(Rm+m)\pi\in\mathcal{P}(\mathbb{R}^{m+m}) put

J(π)=∫Rm+mφ dπ∈[0,∞],J(\pi)=\int_{\mathbb{R}^{m+m}}\varphi\,d\pi\in[0,\infty],

the integral of Probability Measures on Euclidean Space and Random Vectors: Standing Notation (iii) of φ\varphi read as a map into [0,∞][0,\infty]. For μ,ν∈P(Rm)\mu,\nu\in\mathcal{P}(\mathbb{R}^{m}) and a coupling π∈Π(μ,ν)\pi\in\Pi(\mu,\nu), J(π)J(\pi) is the quadratic cost I(π)I(\pi).

Then the following hold.

1. (The coordinate projections are nonexpansive) Let q,p∈Nq,p\in\mathbb{N} satisfy 1≤q1\le q and 1≤p1\le p. Then the coordinate projections pr1q,p:Rq+p→Rq\mathrm{pr}^{q,p}_{1}:\mathbb{R}^{q+p}\to\mathbb{R}^{q} and pr2q,p:Rq+p→Rp\mathrm{pr}^{q,p}_{2}:\mathbb{R}^{q+p}\to\mathbb{R}^{p} of Pairs of Euclidean Points: Coordinate Projections, Pairings, the Product Measure on a Euclidean Space, Borel Norm Functions and Finite Sets (i) are Lipschitz with constant 11, and continuous on Rq+p\mathbb{R}^{q+p}.

2. (The cost function is continuous) The map Rm+m→Rm\mathbb{R}^{m+m}\to\mathbb{R}^{m} with value pr1(z)−pr2(z)\mathrm{pr}_{1}(z)-\mathrm{pr}_{2}(z) at zz is Lipschitz with constant 22, and φ\varphi is continuous on Rm+m\mathbb{R}^{m+m}.

3. (A cost bound passes to a weak limit) Let (πn)n∈N(\pi_{n})_{n\in\mathbb{N}} be a sequence in P(Rm+m)\mathcal{P}(\mathbb{R}^{m+m}) that converges weakly to π∈P(Rm+m)\pi\in\mathcal{P}(\mathbb{R}^{m+m}) on the metric space (Rm+m,dE)(\mathbb{R}^{m+m},d_{E}), and let c∈Rc\in\mathbb{R} be such that for every ε∈R\varepsilon\in\mathbb{R} with 0<ε0<\varepsilon there is N∈NN\in\mathbb{N} with

J(πn)≤c+εfor every n∈N with N≤n.J(\pi_{n})\le c+\varepsilon\qquad\text{for every }n\in\mathbb{N}\text{ with }N\le n .

Then J(π)≤cJ(\pi)\le c; in particular J(π)J(\pi) is a real number.

Proofs

Log in to submit a proof.

Loading...

Citations

Loading…

Dependencies

Loading…

Related

0 relations

Curated associations between results. These are editable and subjective — they do not replace the dependency graph, which is derived from the references in the text.

No relations recorded yet.

Comments

Log in to comment.

Loading…