TheoremBase

Each clause is obtained by applying the two one-sided estimates of the cross-level comparison theorem with the bounded coarse solutions wNw_N as intermediaries: a fine subsolution lies below wNw_N composed with the mode restriction up to theta, which in turn lies below a fine supersolution up to theta. Letting theta tend to zero gives comparison, applying it both ways gives uniqueness, and using a fine solution on both sides gives uniform convergence on penalty sublevel sets and hence pointwise convergence.

Proof

Each result cited is universally quantified over the data in its own statement.

Preliminaries. Let N∈NN\in\mathbb{N}. By hypothesis wNw_{N} is a viscosity solution of F(N)F^{(N)} relative to P(N)\mathcal{P}^{(N)}, hence, by Viscosity Subsolution, Supersolution and Solution of a First-Order Equation on the Noise Wasserstein Space Relative to a Noise Penalty Pair §solution taken at level NN, both a viscosity subsolution and a viscosity supersolution of F(N)F^{(N)} relative to P(N)\mathcal{P}^{(N)}. Since ∣wN(ν)∣≤b0|w_{N}(\nu)|\le b_{0} for every ν∈D(N)\nu\in\mathcal{D}^{(N)}, the two-sided bound Properties of the Absolute Value in an Ordered Field §two-sided gives −b0≤wN(ν)-b_{0}\le w_{N}(\nu) and wN(ν)≤b0w_{N}(\nu)\le b_{0} for every ν∈D(N)\nu\in\mathcal{D}^{(N)}. Likewise, by Viscosity Subsolution, Supersolution and Solution of a First-Order Equation on the Noise Wasserstein Space Relative to a Noise Penalty Pair §solution, a viscosity solution of FF relative to P\mathcal{P} is both a viscosity subsolution and a viscosity supersolution of FF relative to P\mathcal{P}. Finally, E\mathcal{E} is a function from D\mathcal{D} to R\mathbb{R} by Noise Penalty Pairs on the Noise Wasserstein Space: the Penalty, Its Score, and Their Domains §pair, so E(μ)\mathcal{E}(\mu) is a real number for every μ∈D\mu\in\mathcal{D}, and p#(N)μ∈D(N)p^{(N)}_{\#}\mu\in\mathcal{D}^{(N)} for μ∈D\mu\in\mathcal{D}, as recorded in the statement, so that wN(p#(N)μ)w_{N}(p^{(N)}_{\#}\mu) is defined.

Claim (comparison). Choose b,b′∈Rb,b'\in\mathbb{R} with u(μ)≤bu(\mu)\le b and b′≤v(μ)b'\le v(\mu) for every μ∈D\mu\in\mathcal{D}. Fix μ∈D\mu\in\mathcal{D}, put c=E(μ)c=\mathcal{E}(\mu), and let θ∈R\theta\in\mathbb{R} with 0<θ0<\theta. The choices are made in this order. First, A Cross-Level Comparison Principle for First-Order Equations on a Fine Noise Wasserstein Space and a Family of Coarse Levels Linked by Mode Restrictions §fine-below, applied with the numbers bb, −b0-b_{0}, θ\theta and cc in place of its bb, b′b', θ\theta and cc, provides N′∈NN'\in\mathbb{N}. Second, A Cross-Level Comparison Principle for First-Order Equations on a Fine Noise Wasserstein Space and a Family of Coarse Levels Linked by Mode Restrictions §coarse-below, applied with the numbers b0b_{0}, b′b', θ\theta and cc in place of its bb, b′b', θ\theta and cc, provides N′′∈NN''\in\mathbb{N}. Let NN be the larger of N′N' and N′′N''. Since N′≤NN'\le N, the first of these results, applied to the viscosity subsolution uu (bounded above by bb) and to the viscosity supersolution wNw_{N} of F(N)F^{(N)} (bounded below by −b0-b_{0} by the preliminaries), gives, because E(μ)≤c\mathcal{E}(\mu)\le c,

u(μ)≤wN(p#(N)μ)+θ.u(\mu)\le w_{N}(p^{(N)}_{\#}\mu)+\theta .

Since N′′≤NN''\le N, the second result, applied to the viscosity subsolution wNw_{N} of F(N)F^{(N)} (bounded above by b0b_{0}) and to the viscosity supersolution z=vz=v (bounded below by b′b'), gives

wN(p#(N)μ)≤v(μ)+θ.w_{N}(p^{(N)}_{\#}\mu)\le v(\mu)+\theta .

Hence u(μ)≤v(μ)+2θu(\mu)\le v(\mu)+2\theta for every positive θ\theta. If u(μ)≤v(μ)u(\mu)\le v(\mu) failed, then θ=(u(μ)−v(μ))/4\theta=(u(\mu)-v(\mu))/4 would be positive and the last inequality would give u(μ)−v(μ)≤(u(μ)−v(μ))/2u(\mu)-v(\mu)\le(u(\mu)-v(\mu))/2, which is false for the positive number u(μ)−v(μ)u(\mu)-v(\mu). Therefore u(μ)≤v(μ)u(\mu)\le v(\mu), and μ∈D\mu\in\mathcal{D} was arbitrary.

Claim (uniqueness). By the preliminaries, ww is a viscosity subsolution of FF relative to P\mathcal{P} that is bounded above and w′w' is a viscosity supersolution of FF relative to P\mathcal{P} that is bounded below, so claim 1 gives w(μ)≤w′(μ)w(\mu)\le w'(\mu) for every μ∈D\mu\in\mathcal{D}. Exchanging the roles of ww and w′w', claim 1 gives w′(μ)≤w(μ)w'(\mu)\le w(\mu). Hence w(μ)=w′(μ)w(\mu)=w'(\mu) for every μ∈D\mu\in\mathcal{D}.

Claim (convergence). Choose bw,bw′∈Rb_{w},b'_{w}\in\mathbb{R} with bw′≤w(μ)≤bwb'_{w}\le w(\mu)\le b_{w} for every μ∈D\mu\in\mathcal{D}, and let θ,c∈R\theta,c\in\mathbb{R} with 0<θ0<\theta. First, A Cross-Level Comparison Principle for First-Order Equations on a Fine Noise Wasserstein Space and a Family of Coarse Levels Linked by Mode Restrictions §fine-below, applied with bwb_{w}, −b0-b_{0}, θ\theta and cc in place of its bb, b′b', θ\theta and cc, provides N′∈NN'\in\mathbb{N}; second, A Cross-Level Comparison Principle for First-Order Equations on a Fine Noise Wasserstein Space and a Family of Coarse Levels Linked by Mode Restrictions §coarse-below, applied with b0b_{0}, bw′b'_{w}, θ\theta and cc, provides N′′∈NN''\in\mathbb{N}. Let N1N_{1} be the larger of N′N' and N′′N'', and let N∈NN\in\mathbb{N} with N1≤NN_{1}\le N and μ∈D\mu\in\mathcal{D} with E(μ)≤c\mathcal{E}(\mu)\le c. The first result, applied to the viscosity subsolution u=wu=w of FF (bounded above by bwb_{w}) and the viscosity supersolution v=wNv=w_{N} of F(N)F^{(N)} (bounded below by −b0-b_{0}), gives w(μ)≤wN(p#(N)μ)+θw(\mu)\le w_{N}(p^{(N)}_{\#}\mu)+\theta. The second result, applied to the viscosity subsolution wNw_{N} of F(N)F^{(N)} (bounded above by b0b_{0}) and the viscosity supersolution z=wz=w of FF (bounded below by bw′b'_{w}), gives wN(p#(N)μ)≤w(μ)+θw_{N}(p^{(N)}_{\#}\mu)\le w(\mu)+\theta. Thus −θ≤wN(p#(N)μ)−w(μ)≤θ-\theta\le w_{N}(p^{(N)}_{\#}\mu)-w(\mu)\le\theta, and the two-sided bound Properties of the Absolute Value in an Ordered Field §two-sided gives

∣wN(p#(N)μ)−w(μ)∣≤θ.\bigl|w_{N}(p^{(N)}_{\#}\mu)-w(\mu)\bigr|\le\theta .

Claim (pointwise). Let μ∈D\mu\in\mathcal{D} and let ε∈R\varepsilon\in\mathbb{R} with 0<ε0<\varepsilon. Apply claim 3 with θ=ε/2\theta=\varepsilon/2, which is positive, and c=E(μ)c=\mathcal{E}(\mu), to obtain N1∈NN_{1}\in\mathbb{N}. For every N∈NN\in\mathbb{N} with N1≤NN_{1}\le N, μ\mu satisfies E(μ)≤c\mathcal{E}(\mu)\le c, so ∣wN(p#(N)μ)−w(μ)∣≤ε/2<ε|w_{N}(p^{(N)}_{\#}\mu)-w(\mu)|\le\varepsilon/2<\varepsilon. By Limit of a Sequence of Real Numbers, the sequence (wN(p#(N)μ))N∈N(w_{N}(p^{(N)}_{\#}\mu))_{N\in\mathbb{N}} converges to w(μ)w(\mu).

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…