TheoremBase

Noise Displacement and Cross Pairings Along Couplings: Bounds, Linearity, Displacement Couplings, Polarisation and a Vanishing Criterion

Along a coupling of finite noise cost the displacement is a square-integrable noise field whose squared norm is the noise cost; pairing it with a noise field gives a bounded linear functional, computed explicitly along displacement couplings. Also: the cross pairing of two noise fields along any coupling, the polarisation identity for their discrepancy, and a noise field whose pairings vanish to first order is zero.

Statement

In the setting of Probability Measures on a Hilbert Space Transported in the Noise Norm: Standing Notation, let μ∈P(X)\mu\in\mathcal{P}(X). For ν∈P(X)\nu\in\mathcal{P}(X), Π(ν,μ)\Pi(\nu,\mu) is the set of couplings of Probability Measures on a Hilbert Space Transported in the Noise Norm: Standing Notation §background, Πa(μ,ν)\Pi^{a}(\mu,\nu) is the set of Couplings of Finite Noise Cost and Their Noise Cost §couplings, and for π∈Πa(μ,ν)\pi\in\Pi^{a}(\mu,\nu) the noise cost Ia(π)=∫X×Xca dπI^{a}(\pi)=\int_{X\times X}c_{a}\,d\pi is a nonnegative real number by Couplings of Finite Noise Cost and Their Noise Cost §cost. Pρa\mathcal{P}^{a}_{\rho} is the set of The Measures Noise-Connected to the Reference Measure §space. The set DaD_{a} and the function cac_{a} are those of Probability Measures on a Hilbert Space Transported in the Noise Norm: Standing Notation §noise-space; the spaces L2(μ;Xa)L^{2}(\mu;X^{a}), L2(ν;Xa)L^{2}(\nu;X^{a}) and L2(π;Xa)L^{2}(\pi;X^{a}), with inner products ⟨⋅,⋅⟩μ\langle\cdot,\cdot\rangle_{\mu}, ⟨⋅,⋅⟩ν\langle\cdot,\cdot\rangle_{\nu}, ⟨⋅,⋅⟩π\langle\cdot,\cdot\rangle_{\pi} and norms ∥⋅∥μ\lVert\cdot\rVert_{\mu}, ∥⋅∥ν\lVert\cdot\rVert_{\nu}, ∥⋅∥π\lVert\cdot\rVert_{\pi}, and measurability of maps into XaX^{a} are those of Probability Measures on a Hilbert Space Transported in the Noise Norm: Standing Notation §fields; id\mathrm{id} is the identity map of XX; and square roots are those of Existence and Uniqueness of the Nonnegative Square Root. Then the following hold.

1. (The displacement field) Let ν∈P(X)\nu\in\mathcal{P}(X) and π∈Πa(μ,ν)\pi\in\Pi^{a}(\mu,\nu). The map δ:X×X→Xa\delta:X\times X\to X^{a} with δ(z)=y−x\delta(z)=y-x for z∈Daz\in D_{a} and δ(z)=0X\delta(z)=0_{X} for z∈(X×X)∖Daz\in(X\times X)\setminus D_{a} is measurable into XaX^{a} and square-integrable with respect to π\pi, and its class in L2(π;Xa)L^{2}(\pi;X^{a}), again written δ\delta, satisfies

∥δ∥π2=Ia(π).\lVert\delta\rVert_{\pi}^{2}=I^{a}(\pi).

2. (The noise displacement pairing) Let ν∈P(X)\nu\in\mathcal{P}(X), π∈Πa(μ,ν)\pi\in\Pi^{a}(\mu,\nu) and η∈L2(μ;Xa)\eta\in L^{2}(\mu;X^{a}), and let δ\delta be the displacement field of claim 1 for π\pi. For every representative of η\eta the function X×X→RX\times X\to\mathbb{R} with value ⟨η(x),δ(z)⟩a\langle\eta(x),\delta(z)\rangle_{a} at zz is Borel and integrable with respect to π\pi, and its integral does not depend on the representative chosen. That integral,

Ja(η,π)=∫X×X⟨η(x),δ(z)⟩a π(dz)∈R,\mathcal{J}^{a}(\eta,\pi)=\int_{X\times X}\langle\eta(x),\delta(z)\rangle_{a}\,\pi(dz)\in\mathbb{R},

whose integrand equals ⟨η(x),y−x⟩a\langle\eta(x),y-x\rangle_{a} at every z∈Daz\in D_{a}, is called the noise displacement pairing of η\eta along π\pi.

3. (Bound) For ν∈P(X)\nu\in\mathcal{P}(X), π∈Πa(μ,ν)\pi\in\Pi^{a}(\mu,\nu) and η∈L2(μ;Xa)\eta\in L^{2}(\mu;X^{a}),

∣Ja(η,π)∣≤∥η∥μ Ia(π).\bigl|\mathcal{J}^{a}(\eta,\pi)\bigr|\le\lVert\eta\rVert_{\mu}\,\sqrt{I^{a}(\pi)} .

4. (Linearity in the field) For ν∈P(X)\nu\in\mathcal{P}(X), π∈Πa(μ,ν)\pi\in\Pi^{a}(\mu,\nu), η,ξ∈L2(μ;Xa)\eta,\xi\in L^{2}(\mu;X^{a}) and s,t∈Rs,t\in\mathbb{R},

Ja(s η+t ξ,π)=s Ja(η,π)+t Ja(ξ,π).\mathcal{J}^{a}(s\,\eta+t\,\xi,\pi)=s\,\mathcal{J}^{a}(\eta,\pi)+t\,\mathcal{J}^{a}(\xi,\pi).

5. (Displacement couplings) Let h∈L2(μ;Xa)h\in L^{2}(\mu;X^{a}), let h:X→Xah:X\to X^{a} also denote any representative of it, and let t∈Rt\in\mathbb{R}; the following conclusions hold for every such representative. Then the map St:X→XS_{t}:X\to X, St(x)=x+t h(x)S_{t}(x)=x+t\,h(x), is Borel, the coupling πt=(id,St)#μ\pi_{t}=(\mathrm{id},S_{t})_{\#}\mu belongs to Πa(μ,(St)#μ)\Pi^{a}(\mu,(S_{t})_{\#}\mu), and

Ia(πt)=t2 ∥h∥μ2,Ja(η,πt)=t ⟨η,h⟩μfor every η∈L2(μ;Xa).I^{a}(\pi_{t})=t^{2}\,\lVert h\rVert_{\mu}^{2},\qquad\mathcal{J}^{a}(\eta,\pi_{t})=t\,\langle\eta,h\rangle_{\mu}\quad\text{for every }\eta\in L^{2}(\mu;X^{a}).

If moreover μ∈Pρa\mu\in\mathcal{P}^{a}_{\rho}, then (St)#μ∈Pρa(S_{t})_{\#}\mu\in\mathcal{P}^{a}_{\rho}.

6. (The cross pairing) Let ν∈P(X)\nu\in\mathcal{P}(X), π∈Π(ν,μ)\pi\in\Pi(\nu,\mu), q∈L2(ν;Xa)q\in L^{2}(\nu;X^{a}) and η∈L2(μ;Xa)\eta\in L^{2}(\mu;X^{a}). For all representatives of qq and of η\eta the function X×X→RX\times X\to\mathbb{R} with value ⟨q(x),η(y)⟩a\langle q(x),\eta(y)\rangle_{a} at zz is Borel and integrable with respect to π\pi, and its integral does not depend on the representatives chosen. That integral,

Ka(q,η,π)=∫X×X⟨q(x),η(y)⟩a π(dz)∈R,\mathcal{K}^{a}(q,\eta,\pi)=\int_{X\times X}\langle q(x),\eta(y)\rangle_{a}\,\pi(dz)\in\mathbb{R},

is called the cross pairing of qq and η\eta along π\pi. It satisfies ∣Ka(q,η,π)∣≤∥q∥ν ∥η∥μ|\mathcal{K}^{a}(q,\eta,\pi)|\le\lVert q\rVert_{\nu}\,\lVert\eta\rVert_{\mu}. Moreover, for all representatives the function X×X→RX\times X\to\mathbb{R} with value ∣q(x)−η(y)∣a2|q(x)-\eta(y)|_{a}^{2} at zz is Borel, nonnegative and integrable with respect to π\pi, its integral, called the discrepancy of qq and η\eta along π\pi, does not depend on the representatives chosen, and

∫X×X∣q(x)−η(y)∣a2 π(dz)=∥q∥ν2−2 Ka(q,η,π)+∥η∥μ2.\int_{X\times X}|q(x)-\eta(y)|_{a}^{2}\,\pi(dz)=\lVert q\rVert_{\nu}^{2}-2\,\mathcal{K}^{a}(q,\eta,\pi)+\lVert\eta\rVert_{\mu}^{2}.

7. (Vanishing) Suppose moreover μ∈Pρa\mu\in\mathcal{P}^{a}_{\rho}, let η∈L2(μ;Xa)\eta\in L^{2}(\mu;X^{a}), and suppose that for every ε∈R\varepsilon\in\mathbb{R} with 0<ε0<\varepsilon there is θ∈R\theta\in\mathbb{R} with 0<θ0<\theta such that

∣Ja(η,π)∣≤ε Ia(π)\bigl|\mathcal{J}^{a}(\eta,\pi)\bigr|\le\varepsilon\,\sqrt{I^{a}(\pi)}

for every ν∈Pρa\nu\in\mathcal{P}^{a}_{\rho} and every π∈Πa(μ,ν)\pi\in\Pi^{a}(\mu,\nu) with Ia(π)<θ2I^{a}(\pi)<\theta^{2}. Then η\eta is the zero element of L2(μ;Xa)L^{2}(\mu;X^{a}).

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…