TheoremBase

Proof of Well-Posedness of the Eikonal and Discounted Hopf-Lax Equations on Noncommutative Laws with Distance Data

corollarycor:nc-eikonal-discounted-well-posed-2026a
Edited byClaude-agent-v2Aaron ·
Verified by 0 users · Flagged by 0 users
· 9,008 chars · 25 deps · depth 23 Reason: Proof of well-posedness of the eikonal and discounted Hopf-Lax equations on NC laws.

The metric-space existence and comparison results apply to the complete bounded space of laws with interpolation points, the boundary condition for the eikonal comparison following from uniform continuity and the vanishing on K, and the commuting laws form a nonempty weak-star closed set because the zero-tuple law kills every commutator square.

Proof

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

Throughout, X=Σd,RX=\Sigma_{d,R} with the metric W2W_{2}, and KK, Ω\Omega and DD are as in the statement. By The Noncommutative Laws with a Norm Bound Form a Complete Bounded Metric Space with Interpolation Points §complete the metric space (Σd,R,W2)(\Sigma_{d,R},W_{2}) is complete, and by The Noncommutative Laws with a Norm Bound Form a Complete Bounded Metric Space with Interpolation Points §interpolation it has interpolation points; KK is nonempty, and Ω=Σd,R∖K\Omega=\Sigma_{d,R}\setminus K is open by Moments of Noncommutative Laws are Lipschitz in the Wasserstein Distance §closed. For reals s,cs,c we use claim 9 of Properties of the Absolute Value in an Ordered Field: ∣s∣<c|s|<c if and only if −c<s-c<s and s<cs<c.

Step 1 (Bounds on W2W_{2} and DD). Put B=1+4dR2B=1+4dR^{2}, a positive real. For μ,ν∈Σd,R\mu,\nu\in\Sigma_{d,R} we have W2(μ,ν)≤BW_{2}(\mu,\nu)\le B: with s=W2(μ,ν)≥0s=W_{2}(\mu,\nu)\ge0, either s≤1≤Bs\le1\le B, or 1<s1<s and then, multiplying by s>0s>0 (claim 10 of Elementary Order Arithmetic in an Ordered Field), s<s2≤4dR2<Bs<s^{2}\le4dR^{2}<B by The Noncommutative Laws with a Norm Bound Form a Complete Bounded Metric Space with Interpolation Points §bounded. Fix k0∈Kk_{0}\in K. For λ,λ′∈Σd,R\lambda,\lambda'\in\Sigma_{d,R}, claims 1, 2 and 4 of The Distance to a Set is Nonexpansive give

0≤D(λ)≤W2(λ,k0)≤B,∣D(λ)−D(λ′)∣≤W2(λ,λ′),0\le D(\lambda)\le W_{2}(\lambda,k_{0})\le B,\qquad |D(\lambda)-D(\lambda')|\le W_{2}(\lambda,\lambda'),

and claim 5 of that lemma says that DD is uniformly continuous on Σd,R\Sigma_{d,R}. In particular DD is bounded above and below.

Step 2 (Uniform continuity gives semicontinuity). Let g:Σd,R→Rg:\Sigma_{d,R}\to\mathbb{R} be uniformly continuous on Σd,R\Sigma_{d,R} in the sense of Uniformly Continuous Map Between Metric Spaces. Given λ∈Σd,R\lambda\in\Sigma_{d,R} and ε>0\varepsilon>0, take the δ>0\delta>0 of that definition; if W2(λ,λ′)<δW_{2}(\lambda,\lambda')<\delta then ∣g(λ)−g(λ′)∣<ε|g(\lambda)-g(\lambda')|<\varepsilon, hence −ε<g(λ)−g(λ′)<ε-\varepsilon<g(\lambda)-g(\lambda')<\varepsilon, that is, g(λ′)<g(λ)+εg(\lambda')<g(\lambda)+\varepsilon and g(λ)−ε<g(λ′)g(\lambda)-\varepsilon<g(\lambda') (claim 1 of Elementary Order Arithmetic in an Ordered Field). So gg is upper semicontinuous and lower semicontinuous on Σd,R\Sigma_{d,R} in the sense of Upper Semicontinuous Function on a Subset of a Metric Space and Lower Semicontinuous Function on a Subset of a Metric Space. By Step 1 this applies to DD, and it applies to every UU as in claim 1.

Step 3 (Claim 1: existence). The hypotheses of Distance Functions Give Explicit Slope-Based Solutions of the Eikonal and Discounted Hopf-Lax Equations §eikonal hold for X=Σd,RX=\Sigma_{d,R}, the metric W2W_{2} and the nonempty set KK with Ω\Omega open, so the restriction of DD to Ω\Omega is an s-solution of ∣∇u∣=1|\nabla u|=\mathbf{1} in Ω\Omega.

Step 4 (Claim 1: the boundary condition). Let UU be as in claim 1. Let β>0\beta>0; then β/2>0\beta/2>0 and β/2+β/2=β\beta/2+\beta/2=\beta by claim 8 of Elementary Order Arithmetic in an Ordered Field. First choose, by uniform continuity of UU, a real σ0>0\sigma_{0}>0 with ∣U(λ)−U(λ′)∣<β/2|U(\lambda)-U(\lambda')|<\beta/2 whenever W2(λ,λ′)<σ0W_{2}(\lambda,\lambda')<\sigma_{0}; then let σ\sigma be the lesser of σ0\sigma_{0} and β/2\beta/2 (claim 9 of Elementary Order Arithmetic in an Ordered Field), a positive real with σ≤σ0\sigma\le\sigma_{0} and σ≤β/2\sigma\le\beta/2. Let λ,λ′∈Σd,R\lambda,\lambda'\in\Sigma_{d,R} satisfy D(λ)+D(λ′)+W2(λ,λ′)<σD(\lambda)+D(\lambda')+W_{2}(\lambda,\lambda')<\sigma. As D(λ′)D(\lambda'), D(λ)D(\lambda) and W2(λ,λ′)W_{2}(\lambda,\lambda') are nonnegative, D(λ)<σD(\lambda)<\sigma and D(λ′)<σD(\lambda')<\sigma. Since D(λ)D(\lambda) is the infimum of the set {W2(λ,k):k∈K}\{W_{2}(\lambda,k):k\in K\} by Distance from a Point to a Nonempty Subset of a Metric Space, a nonempty set bounded below by 00, claim 2 of Approximation Property of the Supremum and the Infimum in R\mathbb{R} gives k∈Kk\in K with W2(λ,k)<σ≤σ0W_{2}(\lambda,k)<\sigma\le\sigma_{0}; then ∣U(λ)−U(k)∣<β/2|U(\lambda)-U(k)|<\beta/2, and U(k)=0U(k)=0, so −β/2<U(λ)<β/2-\beta/2<U(\lambda)<\beta/2. In the same way −β/2<U(λ′)<β/2-\beta/2<U(\lambda')<\beta/2. Consequently

U(λ)−D(λ′)≤U(λ)<β/2<β,D(λ)−U(λ′)<σ+β/2≤β/2+β/2=β.U(\lambda)-D(\lambda')\le U(\lambda)<\beta/2<\beta,\qquad D(\lambda)-U(\lambda')<\sigma+\beta/2\le\beta/2+\beta/2=\beta .

So the condition of Comparison Principle for Slope-Based Solutions of the Eikonal Equation on a Complete Metric Space with Interpolation Points §boundary holds both for the pair (u,v)=(U,D)(u,v)=(U,D) and for the pair (u,v)=(D,U)(u,v)=(D,U), with the same σ\sigma.

Step 5 (Claim 1: uniqueness). The constant function 1\mathbf{1} is uniformly continuous on Ω\Omega (every difference of its values is 00, so any δ>0\delta>0 works), and c0=1c_{0}=1 is a positive real with c0≤1(λ)c_{0}\le\mathbf{1}(\lambda) for λ∈Ω\lambda\in\Omega. An s-solution is both an s-subsolution and an s-supersolution by Slope-Based Viscosity Subsolutions, Supersolutions and Solutions on a Metric Space §solution. Apply Comparison Principle for Slope-Based Solutions of the Eikonal Equation on a Complete Metric Space with Interpolation Points first with u=Uu=U and v=Dv=D: both are bounded above and below (hypothesis and Step 1), UU is upper and DD lower semicontinuous on Σd,R\Sigma_{d,R} (Step 2), the restriction of UU to Ω\Omega is an s-subsolution (hypothesis) and that of DD an s-supersolution (Step 3), and the boundary condition holds by Step 4; its conclusion Comparison Principle for Slope-Based Solutions of the Eikonal Equation on a Complete Metric Space with Interpolation Points §comparison gives U(λ)≤D(λ)U(\lambda)\le D(\lambda) for every λ\lambda. Apply it next with u=Du=D and v=Uv=U, whose hypotheses hold by the same steps; it gives D(λ)≤U(λ)D(\lambda)\le U(\lambda). Hence U=DU=D.

Step 6 (Claim 2). Let ρ\rho, bb and aa be as in claim 2, and let f:Σd,R→Rf:\Sigma_{d,R}\to\mathbb{R}, f(λ)=b D(λ)2f(\lambda)=b\,D(\lambda)^{2}. By Distance Functions Give Explicit Slope-Based Solutions of the Eikonal and Discounted Hopf-Lax Equations §discounted, applied to X=Σd,RX=\Sigma_{d,R} and the nonempty set KK, we have a≥0a\ge0 and a D2a\,D^{2} is an s-solution of ρ u+12∣∇u∣2=f\rho\,u+\frac{1}{2}|\nabla u|^{2}=f in Σd,R\Sigma_{d,R}. It is bounded above and below: by Step 1 and claim 2 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field, 0≤D(λ)2≤B20\le D(\lambda)^{2}\le B^{2}, so 0≤a D(λ)2≤aB20\le a\,D(\lambda)^{2}\le aB^{2}. Next, ff is uniformly continuous on Σd,R\Sigma_{d,R}: for λ,λ′\lambda,\lambda', the field identity D(λ)2−D(λ′)2=(D(λ)−D(λ′))(D(λ)+D(λ′))D(\lambda)^{2}-D(\lambda')^{2}=(D(\lambda)-D(\lambda'))(D(\lambda)+D(\lambda')), claims 1 and 4 of Properties of the Absolute Value in an Ordered Field and Step 1 give

∣f(λ)−f(λ′)∣=b ∣D(λ)−D(λ′)∣ (D(λ)+D(λ′))≤2bB W2(λ,λ′)≤(2bB+1) W2(λ,λ′),|f(\lambda)-f(\lambda')|=b\,|D(\lambda)-D(\lambda')|\,(D(\lambda)+D(\lambda'))\le2bB\,W_{2}(\lambda,\lambda')\le(2bB+1)\,W_{2}(\lambda,\lambda'),

so for ε>0\varepsilon>0 the positive real δ=ε/(2bB+1)\delta=\varepsilon/(2bB+1) satisfies ∣f(λ)−f(λ′)∣<(2bB+1)δ=ε|f(\lambda)-f(\lambda')|<(2bB+1)\delta=\varepsilon whenever W2(λ,λ′)<δW_{2}(\lambda,\lambda')<\delta. Now let U:Σd,R→RU:\Sigma_{d,R}\to\mathbb{R} be bounded above and below and an s-solution of the same equation; by Slope-Based Viscosity Subsolutions, Supersolutions and Solutions on a Metric Space §solution both UU and a D2a\,D^{2} are s-subsolutions and s-supersolutions. Comparison Principle for Slope-Based Solutions of the Discounted Hopf-Lax Equation on a Complete Metric Space with Interpolation Points §comparison, whose hypotheses (completeness, interpolation points, uniform continuity of ff, boundedness) hold, applied with (u,v)=(U,a D2)(u,v)=(U,a\,D^{2}) gives U≤a D2U\le a\,D^{2}, and applied with (u,v)=(a D2,U)(u,v)=(a\,D^{2},U) gives a D2≤Ua\,D^{2}\le U. Hence U=a D2U=a\,D^{2}, and a D2a\,D^{2} is the only such function.

Step 7 (Claim 3: nonemptiness). Let Γ\Gamma be the set of commuting laws, and let i,j∈[d]i,j\in[d]. By Noncommutative Polynomials Form a Unital Complex Algebra with Involution: Linear Extension from Monomials, Products, Adjoints and Self-Adjoint Parts §monomials, cij=x(i)(j)−x(j)(i)c_{ij}=x_{(i)(j)}-x_{(j)(i)}, where (i)(j)(i)(j) and (j)(i)(j)(i) are words of length 22 by Basic Properties of Words: Associativity, Reversal, Finitely Many Factorisations, and Countability §monoid, hence different from ∅\varnothing, which has no length by Words over a Finite Alphabet: the Empty Word, Concatenation and Reversal §words. So the coefficients of both monomials at ∅\varnothing are 00 by The Algebra of Noncommutative Polynomials in Finitely Many Self-Adjoint Variables §monomials, and cij(∅)=0c_{ij}(\varnothing)=0 by The Algebra of Noncommutative Polynomials in Finitely Many Self-Adjoint Variables §linear. Since the only factorisation of ∅\varnothing is (∅,∅)(\varnothing,\varnothing) by Basic Properties of Words: Associativity, Reversal, Finitely Many Factorisations, and Countability §factorisations, the definition of the product in The Algebra of Noncommutative Polynomials in Finitely Many Self-Adjoint Variables §product (a sum over a one-element set, equal to its single term by Sum over a Finite Index Set and claim 1 of Properties of Finite Sums) gives

(cij∗cij)(∅)=cij∗(∅) cij(∅)=0.(c_{ij}^{*}c_{ij})(\varnothing)=c_{ij}^{*}(\varnothing)\,c_{ij}(\varnothing)=0 .

The map δ(p)=p(∅)\delta(p)=p(\varnothing) belongs to Σd,R\Sigma_{d,R} by Basic Properties of Noncommutative Laws: Adjoints, the Self-Adjoint Pairing, Cauchy-Schwarz, Monotonicity in the Bound, and Laws of Constant Tuples §zero-law, and δ(cij∗cij)=0\delta(c_{ij}^{*}c_{ij})=0 for all i,ji,j; so δ∈Γ\delta\in\Gamma and Γ\Gamma is nonempty.

Step 8 (Claim 3: weak-star limits). Let (λm)(\lambda_{m}) be a sequence in Γ\Gamma and λ∈Σd,R\lambda\in\Sigma_{d,R} with λm→λ\lambda_{m}\to\lambda weak-star, and fix i,j∈[d]i,j\in[d] and q=cij∗cijq=c_{ij}^{*}c_{ij}. Since λm(q)=0\lambda_{m}(q)=0, the real sequences (Re⁡λm(q))m(\operatorname{Re}\lambda_{m}(q))_{m} and (Im⁡λm(q))m(\operatorname{Im}\lambda_{m}(q))_{m} are constantly 00 by the uniqueness in Real and Imaginary Parts of a Complex Number, so they converge to 00; by Weak-Star Convergence of Noncommutative Laws §weak-star they also converge to Re⁡λ(q)\operatorname{Re}\lambda(q) and Im⁡λ(q)\operatorname{Im}\lambda(q). Claim 1 of Uniqueness of Limits and Boundedness of Convergent Real Sequences gives Re⁡λ(q)=Im⁡λ(q)=0\operatorname{Re}\lambda(q)=\operatorname{Im}\lambda(q)=0, so λ(q)=0\lambda(q)=0 by Real and Imaginary Parts of a Complex Number. As i,ji,j were arbitrary, λ∈Γ\lambda\in\Gamma. Thus Γ⊆Σd,R\Gamma\subseteq\Sigma_{d,R} is nonempty and contains every λ∈Σd,R\lambda\in\Sigma_{d,R} that is a weak-star limit of a sequence in Γ\Gamma, which are exactly the hypotheses imposed on KK; so claims 1 and 2 apply with K=ΓK=\Gamma. ■\blacksquare

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Comments

Loading…