TheoremBase

Relative Entropy with Respect to a Diagonal Gaussian Measure on a Hilbert Space: the Moment Bound, the Cutoff Projections, and Bounded, Tight, Weakly Closed, Wasserstein-Closed and Weakly Sequentially Compact Sublevel Sets

Relative to a diagonal Gaussian measure on a Hilbert space, finite relative entropy implies finite second moment with a linear bound, the entropy is the limit of the entropies of the finite-dimensional cutoffs, and entropy sublevel sets are bounded, tight, weakly closed, Wasserstein-closed and weakly sequentially compact.

Statement

In the setting of Borel Probability Measures on a Real Hilbert Space with an Orthonormal Basis: Standing Notation, with the coordinate maps pnp_{n}, push-forwards, weak convergence ⇒\Rightarrow and tightness as in Borel Probability Measures on a Real Hilbert Space with an Orthonormal Basis: Standing Notation §coordinates, Borel Probability Measures on a Real Hilbert Space with an Orthonormal Basis: Standing Notation §pushforward and Borel Probability Measures on a Real Hilbert Space with an Orthonormal Basis: Standing Notation §weak, let cc be a variance sequence with truncations c(n)c^{(n)} and diagonal Gaussian measures γc(n)\gamma_{c^{(n)}} on Rn\mathbb{R}^{n}, and let γc\gamma_{c} be the diagonal Gaussian measure on XX with variances cc, so that (pn)#γc=γc(n)(p_{n})_{\#}\gamma_{c}=\gamma_{c^{(n)}} for every n∈Nn\in\mathbb{N} by that definition. Let cˉ\bar{c} be the sum of the series ∑k=1∞ck\sum_{k=1}^{\infty}c_{k}, which converges by Variance Sequences and Their Truncations §variances; cˉ\bar{c} is a positive real number, since 0<c1≤cˉ0<c_{1}\le\bar{c} by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §dominates. Finite relative entropy and H(⋅ ∣ ⋅)H(\cdot\,|\,\cdot) are those of that definition on the measurable spaces (X,B(X))(X,\mathcal{B}(X)) and (Rn,B(Rn))(\mathbb{R}^{n},\mathcal{B}(\mathbb{R}^{n})); P2(X)\mathcal{P}_{2}(X) and M2M_{2} are the set and the second moment of The Second Moment of a Borel Probability Measure on a Hilbert Space and the Probability Measures with Finite Second Moment §space and The Second Moment of a Borel Probability Measure on a Hilbert Space and the Probability Measures with Finite Second Moment §moment, and W2W_{2} is the Wasserstein distance on P2(X)\mathcal{P}_{2}(X). For C∈RC\in\mathbb{R} let KC\mathcal{K}_{C} be the set of the μ∈P(X)\mu\in\mathcal{P}(X) that have finite relative entropy with respect to γc\gamma_{c} and satisfy H(μ ∣ γc)≤CH(\mu\,|\,\gamma_{c})\le C.

1. (Moment bound) Let μ∈P(X)\mu\in\mathcal{P}(X) have finite relative entropy with respect to γc\gamma_{c}. Then μ∈P2(X)\mu\in\mathcal{P}_{2}(X) and

M2(μ)≤4cˉ H(μ ∣ γc)+2cˉ.M_{2}(\mu)\le4\bar{c}\,H(\mu\,|\,\gamma_{c})+2\bar{c}.

2. (Bounded sublevel sets) Let C∈RC\in\mathbb{R}. Then KC⊆P2(X)\mathcal{K}_{C}\subseteq\mathcal{P}_{2}(X), and M2(μ)≤4cˉ C+2cˉM_{2}(\mu)\le4\bar{c}\,C+2\bar{c} for every μ∈KC\mu\in\mathcal{K}_{C}.

3. (Cutoffs) Let μ∈P(X)\mu\in\mathcal{P}(X) and μn=(pn)#μ\mu_{n}=(p_{n})_{\#}\mu for n∈Nn\in\mathbb{N}. Then μ\mu has finite relative entropy with respect to γc\gamma_{c} if and only if there is C∈RC\in\mathbb{R} such that for every n∈Nn\in\mathbb{N} the measure μn\mu_{n} has finite relative entropy with respect to γc(n)\gamma_{c^{(n)}} and H(μn ∣ γc(n))≤CH(\mu_{n}\,|\,\gamma_{c^{(n)}})\le C; in that case the sequence (H(μn ∣ γc(n)))n∈N\bigl(H(\mu_{n}\,|\,\gamma_{c^{(n)}})\bigr)_{n\in\mathbb{N}} is nondecreasing and converges to H(μ ∣ γc)H(\mu\,|\,\gamma_{c}).

4. (Tight sublevel sets) For every C∈RC\in\mathbb{R} the set KC\mathcal{K}_{C} is tight.

5. (Weakly closed sublevel sets) Let C∈RC\in\mathbb{R}, let (μj)j∈N(\mu_{j})_{j\in\mathbb{N}} be a sequence in KC\mathcal{K}_{C}, and let μ∈P(X)\mu\in\mathcal{P}(X) satisfy μj⇒μ\mu_{j}\Rightarrow\mu. Then μ∈KC\mu\in\mathcal{K}_{C}.

6. (Wasserstein-closed sublevel sets) Let C∈RC\in\mathbb{R}, let (μj)j∈N(\mu_{j})_{j\in\mathbb{N}} be a sequence in KC\mathcal{K}_{C}, a subset of P2(X)\mathcal{P}_{2}(X) by claim 2, and let μ∈P2(X)\mu\in\mathcal{P}_{2}(X) satisfy W2(μj,μ)→0W_{2}(\mu_{j},\mu)\to0 as j→∞j\to\infty. Then μ∈KC\mu\in\mathcal{K}_{C}.

7. (Weak sequential compactness) Let C∈RC\in\mathbb{R}. Every sequence in KC\mathcal{K}_{C} has a subsequence that converges weakly to an element of KC\mathcal{K}_{C}.

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…