TheoremBase

The Relative Score with Respect to the Gibbs Measure of an Admissible Cylindrical Potential

A measure has a relative score with respect to the Gibbs measure of an admissible cylindrical potential if, coordinate by coordinate, integration by parts against bounded cylindrical functions holds with the Gibbs drift; the components are unique.

Statement

In the setting of A Diagonal Gaussian Reference Measure on the Noise Wasserstein Space, Rescaled Heads and Gaussian Tails: Standing Notation, let VV be an admissible cylindrical potential with head dimension dd, with the functions ∂kV\partial_{k}V of Admissible Cylindrical Potentials on a Hilbert Space in the Noise Geometry §gradient, let β∈R\beta\in\mathbb{R} be positive, and let γβV\gamma^{V}_{\beta} be the Gibbs measure of VV at temperature β\beta. FCb1(X)\mathcal{F}C^{1}_{b}(X) is the set of bounded C1C^{1} cylindrical functions and ∂kφ\partial_{k}\varphi are their partial derivatives, which exist by Bounded C^1 Cylindrical Functions: Linear Structure, the Gradient and the Partial Derivatives, Bounds and Integrability, and Density in the Square-Integrable Functions §gradient. Let μ∈P2(X)\mu\in\mathcal{P}_{2}(X), the set of The Second Moment of a Borel Probability Measure on a Hilbert Space and the Probability Measures with Finite Second Moment §space, be such that ∂kV\partial_{k}V is integrable with respect to μ\mu for every k∈[d]k\in[d]; the notion below is defined only for such μ\mu. Let L2(μ)=L2(X,B(X),μ)L^{2}(\mu)=L^{2}(X,\mathcal{B}(X),\mu) be the real Hilbert space of The Lebesgue Space of Square-Integrable Functions is a Real Hilbert Space §hilbert, with the inner product ⟨⋅,⋅⟩L2(μ)\langle\cdot,\cdot\rangle_{L^{2}(\mu)} and norm ∥⋅∥L2(μ)\lVert\cdot\rVert_{L^{2}(\mu)} of The Lebesgue Space of Square-Integrable Functions is a Real Hilbert Space §inner-product. For φ∈FCb1(X)\varphi\in\mathcal{F}C^{1}_{b}(X) and k∈Nk\in\mathbb{N}, the class of φ\varphi belongs to L2(μ)L^{2}(\mu) and is again written φ\varphi, and ∂kφ\partial_{k}\varphi is integrable with respect to μ\mu, by Bounded C^1 Cylindrical Functions: Linear Structure, the Gradient and the Partial Derivatives, Bounds and Integrability, and Density in the Square-Integrable Functions §square-integrable; the function x↦xkφ(x)x\mapsto x_{k}\varphi(x) is integrable with respect to μ\mu by Bounded C^1 Cylindrical Functions: Linear Structure, the Gradient and the Partial Derivatives, Bounds and Integrability, and Density in the Square-Integrable Functions §coordinate-integrable; hence, ckc_{k} being positive by Variance Sequences and Their Truncations §variances, the function x↦xkφ(x)/ck−∂kφ(x)x\mapsto x_{k}\varphi(x)/c_{k}-\partial_{k}\varphi(x) is integrable with respect to μ\mu by Linearity and Monotonicity of the Lebesgue Integral §integrable. Next, φ\varphi is Borel and bounded by Bounded C^1 Cylindrical Functions: Linear Structure, the Gradient and the Partial Derivatives, Bounds and Integrability, and Density in the Square-Integrable Functions §bounded-borel, so there is M∈RM\in\mathbb{R} with ∣φ(x)∣≤M|\varphi(x)|\le M for every x∈Xx\in X; and ∂kV\partial_{k}V is Borel by Basic Properties of an Admissible Cylindrical Potential: Continuity, Growth under Noise Translations, Integrability, the Tangent Inequality and Tangency of the Noise Gradient §continuity and integrable with respect to μ\mu, by hypothesis for k≤dk\le d and being 00 for k>dk>d. So ∂kV φ\partial_{k}V\,\varphi is Borel with ∣∂kV φ∣≤M∣∂kV∣|\partial_{k}V\,\varphi|\le M|\partial_{k}V| on XX, and is integrable with respect to μ\mu by the monotonicity of Linearity and Monotonicity of the Lebesgue Integral §nonnegative, applied with f=∣∂kV φ∣f=|\partial_{k}V\,\varphi| and g=M∣∂kV∣g=M|\partial_{k}V|. Hence the function

x↦(xkck+∂kV(x)β)φ(x)−∂kφ(x)x\mapsto\Bigl(\frac{x_{k}}{c_{k}}+\frac{\partial_{k}V(x)}{\beta}\Bigr)\varphi(x)-\partial_{k}\varphi(x)

is integrable with respect to μ\mu by Linearity and Monotonicity of the Lebesgue Integral §integrable.

(Relative score) The measure μ\mu has a relative score with respect to γβV\gamma^{V}_{\beta} if for every k∈Nk\in\mathbb{N} there is ζkV∈L2(μ)\zeta^{V}_{k}\in L^{2}(\mu) with

⟨ζkV,φ⟩L2(μ)=∫X((xkck+∂kV(x)β)φ(x)−∂kφ(x)) μ(dx)for every φ∈FCb1(X).\langle\zeta^{V}_{k},\varphi\rangle_{L^{2}(\mu)}=\int_{X}\Bigl(\Bigl(\frac{x_{k}}{c_{k}}+\frac{\partial_{k}V(x)}{\beta}\Bigr)\varphi(x)-\partial_{k}\varphi(x)\Bigr)\,\mu(dx)\qquad\text{for every }\varphi\in\mathcal{F}C^{1}_{b}(X).

For each kk there is at most one such ζkV\zeta^{V}_{k}: the difference of two of them is orthogonal to every φ∈FCb1(X)\varphi\in\mathcal{F}C^{1}_{b}(X), by linearity of the inner product, and so is the zero vector by Bounded C^1 Cylindrical Functions: Linear Structure, the Gradient and the Partial Derivatives, Bounds and Integrability, and Density in the Square-Integrable Functions §density. The relative score of μ\mu with respect to γβV\gamma^{V}_{\beta} is then the sequence (ζkV)k∈N(\zeta^{V}_{k})_{k\in\mathbb{N}}, and ζkV\zeta^{V}_{k} is its kk-th component.

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…