TheoremBase

Perturbed Maximisers of the Doubled Difference Penalised by the Wall-Confined Free Energy

For bounded u and v the penalised doubled difference has the usual bounds and monotonicity; when u minus delta times the free energy is Wasserstein upper semicontinuous and v plus delta times the free energy lower semicontinuous on the domain, after a small series of squared-distance perturbations centred at laws of controlled energy it attains a strict maximum at an almost maximising pair.

Statement

In the setting of The Discounted Hamilton-Jacobi-Bellman Equation with Free Langevin Noise in a Wall on Noncommutative Laws: Standing Notation, let RR be the wall radius of The Discounted Hamilton-Jacobi-Bellman Equation with Free Langevin Noise in a Wall on Noncommutative Laws: Standing Notation §parameters, let E:D→R\mathcal{E}:\mathcal{D}\to\mathbb{R} be as in The Discounted Hamilton-Jacobi-Bellman Equation with Free Langevin Noise in a Wall on Noncommutative Laws: Standing Notation §free-energy, and assume D≠∅\mathcal{D}\ne\varnothing. Semicontinuity on D\mathcal{D} refers to the metric space (Σd,R,W2)(\Sigma_{d,R},W_{2}) of The Noncommutative Laws with a Norm Bound Form a Complete Bounded Metric Space with Interpolation Points, which contains D\mathcal{D} by The Wall-Confined Free Energy: Norm Bound, Lower Bound, Weak-Star Compact Sublevel Sets and Displacement Monotonicity of the Score §bounds. Fix a real ee with e≤E(λ)e\le\mathcal{E}(\lambda) for every λ∈D\lambda\in\mathcal{D}, as provided by The Wall-Confined Free Energy: Norm Bound, Lower Bound, Weak-Star Compact Sublevel Sets and Displacement Monotonicity of the Score §bounds. Let u,v:Σd2→Ru,v:\Sigma^{2}_{d}\to\mathbb{R}, let b≥0b\ge0 be real with ∣u(λ)∣≤b|u(\lambda)|\le b and ∣v(λ)∣≤b|v(\lambda)|\le b for every λ∈Σd2\lambda\in\Sigma^{2}_{d}. For reals δ≥0\delta\ge0 and α>0\alpha>0 let Ψδ,α:D×D→R\Psi_{\delta,\alpha}:\mathcal{D}\times\mathcal{D}\to\mathbb{R} be

Ψδ,α(μ,ν)=u(κd(μ))−v(κd(ν))−α2W2(μ,ν)2−δ E(μ)−δ E(ν).\Psi_{\delta,\alpha}(\mu,\nu)=u\bigl(\kappa_{d}(\mu)\bigr)-v\bigl(\kappa_{d}(\nu)\bigr)-\tfrac{\alpha}{2}W_{2}(\mu,\nu)^{2}-\delta\,\mathcal{E}(\mu)-\delta\,\mathcal{E}(\nu).

1. (Bounds) For all reals δ≥0\delta\ge0 and α>0\alpha>0, every value of Ψδ,α\Psi_{\delta,\alpha} is at most 2b−2δe2b-2\delta e, and the supremum S(δ,α)S(\delta,\alpha) of the values of Ψδ,α\Psi_{\delta,\alpha} is a real number; moreover S(δ,α)≤S(0,α)−2δeS(\delta,\alpha)\le S(0,\alpha)-2\delta e.

2. (Diagonal) For every real α>0\alpha>0, S(0,α)≥u(κd(λ))−v(κd(λ))S(0,\alpha)\ge u(\kappa_{d}(\lambda))-v(\kappa_{d}(\lambda)) for every λ∈D\lambda\in\mathcal{D}.

3. (Monotone in the strength) S(0,α′)≥S(0,α)S(0,\alpha')\ge S(0,\alpha) for all reals α,α′\alpha,\alpha' with 0<α′≤α0<\alpha'\le\alpha.

4. (Doubling strength) Let δ≥0\delta\ge0, θ≥0\theta\ge0 and α,α′\alpha,\alpha' be reals with 0<α′<α0<\alpha'<\alpha, and let (μ,ν)∈D×D(\mu,\nu)\in\mathcal{D}\times\mathcal{D} satisfy Ψδ,α(μ,ν)≥S(δ,α)−θ\Psi_{\delta,\alpha}(\mu,\nu)\ge S(\delta,\alpha)-\theta. Then

S(δ,α)−θ+α−α′2 W2(μ,ν)2≤S(δ,α′).S(\delta,\alpha)-\theta+\tfrac{\alpha-\alpha'}{2}\,W_{2}(\mu,\nu)^{2}\le S(\delta,\alpha').

5. (Vanishing weight) For all reals α>0\alpha>0 and ε>0\varepsilon>0 there is a real δ1>0\delta_{1}>0 with S(δ,α)≥S(0,α)−εS(\delta,\alpha)\ge S(0,\alpha)-\varepsilon for every real δ\delta with 0<δ≤δ10<\delta\le\delta_{1}.

6. (Perturbed maximiser) Let δ>0\delta>0 be real, and assume that the function μ↦u(κd(μ))−δ E(μ)\mu\mapsto u(\kappa_{d}(\mu))-\delta\,\mathcal{E}(\mu) on D\mathcal{D} is upper semicontinuous on D\mathcal{D} and the function ν↦v(κd(ν))+δ E(ν)\nu\mapsto v(\kappa_{d}(\nu))+\delta\,\mathcal{E}(\nu) on D\mathcal{D} is lower semicontinuous on D\mathcal{D}. Then there is a real cc with the following property. For all reals α>0\alpha>0 and θ\theta with 0<θ<10<\theta<1 there are (μ^,ν^)∈D×D(\hat{\mu},\hat{\nu})\in\mathcal{D}\times\mathcal{D}, sequences (μk)k∈N(\mu_{k})_{k\in\mathbb{N}} and (νk)k∈N(\nu_{k})_{k\in\mathbb{N}} in D\mathcal{D} and positive reals εk\varepsilon_{k} (k∈N)(k\in\mathbb{N}) such that the series ∑k=1∞εk\sum_{k=1}^{\infty}\varepsilon_{k} converges with ∑k=1∞εk≤θ\sum_{k=1}^{\infty}\varepsilon_{k}\le\theta; E(μ^)≤c\mathcal{E}(\hat{\mu})\le c, E(ν^)≤c\mathcal{E}(\hat{\nu})\le c, and E(μk)≤c\mathcal{E}(\mu_{k})\le c and E(νk)≤c\mathcal{E}(\nu_{k})\le c for every k∈Nk\in\mathbb{N}; Ψδ,α(μ^,ν^)≥S(δ,α)−θ\Psi_{\delta,\alpha}(\hat{\mu},\hat{\nu})\ge S(\delta,\alpha)-\theta; for every (μ,ν)∈D×D(\mu,\nu)\in\mathcal{D}\times\mathcal{D} the series ∑k=1∞εk(W2(μ,μk)2+W2(ν,νk)2)\sum_{k=1}^{\infty}\varepsilon_{k}\bigl(W_{2}(\mu,\mu_{k})^{2}+W_{2}(\nu,\nu_{k})^{2}\bigr) converges; and, with Φ(μ,ν)\Phi(\mu,\nu) denoting Ψδ,α(μ,ν)\Psi_{\delta,\alpha}(\mu,\nu) minus the sum of that series,

Φ(μ,ν)<Φ(μ^,ν^)for every (μ,ν)∈D×D with (μ,ν)≠(μ^,ν^).\Phi(\mu,\nu)<\Phi(\hat{\mu},\hat{\nu})\qquad\text{for every }(\mu,\nu)\in\mathcal{D}\times\mathcal{D}\text{ with }(\mu,\nu)\ne(\hat{\mu},\hat{\nu}).

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…