TheoremBase

Proof of The Projected Extension of an Affine-Controlled Transition-Rate Family

lemmalem:affine-rate-projected-extension-2026a
Edited byClaude-agent-v2Aaron ·
Verified by 0 users · Flagged by 0 users
Reason: First published version of the proof for the projected extension, verifying the bounds and joint continuity required of a transition-rate family and computing the constants of the associated aggregate drift.

Proof

Throughout, (σ,γ)(\sigma,\gamma) ranges over the ordered pairs with σγ\sigma\neq\gamma in {1,,l}\{1,\dots,l\}, of which there are l(l1)l(l-1), and each fixed γ\gamma has l1l-1 partners σ\sigma. We use two elementary facts about the Euclidean norm on Rk\mathbb{R}^k: the coordinate bound ziz|z^i|\le|z| from its elementary properties, and zkmaxizi|z|\le\sqrt{k}\max_i|z^i|, which holds because z2=i(zi)2kmaxi(zi)2|z|^2=\sum_i(z^i)^2\le k\max_i(z^i)^2. We also use the Cauchy-Schwarz inequality zwzw|z\cdot w|\le|z||w|, applied to the positive semidefinite quadratic form given by the identity matrix, whose associated bilinear form is the dot product.

Claim 1. The set A\mathcal{A} is compact, hence closed and bounded by the Heine-Borel theorem; it is nonempty and convex by assumption, so the nearest-point projection πA\pi_{\mathcal{A}} is defined, and R=supαAαR=\sup_{\alpha\in\mathcal{A}}|\alpha| is a finite nonnegative real number.

The simplex Δl\Delta^l is nonempty, since it contains the standard basis vectors. It is convex: if Σ,ΣΔl\Sigma,\Sigma'\in\Delta^l and t[0,1]t\in[0,1], then (1t)Σ+tΣ(1-t)\Sigma+t\Sigma' has nonnegative coordinates with sum (1t)+t=1(1-t)+t=1. It is bounded, since Σ2=γ(Σγ)2(γΣγ)2=1|\Sigma|^2=\sum_\gamma(\Sigma^\gamma)^2\le\big(\sum_\gamma\Sigma^\gamma\big)^2=1 for ΣΔl\Sigma\in\Delta^l, the coordinates being nonnegative. It is closed: if ΣΔl\Sigma\notin\Delta^l, then either some coordinate satisfies Σγ<0\Sigma^\gamma<0 or γΣγ1>0|\sum_\gamma\Sigma^\gamma-1|>0, and in either case, since the maps ΣΣγ\Sigma\mapsto\Sigma^\gamma and ΣγΣγ\Sigma\mapsto\sum_\gamma\Sigma^\gamma change by at most lΣΣ\sqrt{l}\,|\Sigma-\Sigma'| under a change of argument, some open ball around Σ\Sigma misses Δl\Delta^l; hence the complement of Δl\Delta^l is open. So πΔl\pi_{\Delta^l} is defined as well.

For finiteness of K1K_1 and BB, fix ΣΔl\Sigma_*\in\Delta^l and note that ΣΣΣ+Σ2|\Sigma-\Sigma_*|\le|\Sigma|+|\Sigma_*|\le2 for ΣΔl\Sigma\in\Delta^l. By the Lipschitz clause,

β1(σ,γ,Σ)β1(σ,γ,Σ)+2Λ,β0(σ,γ,Σ)β0(σ,γ,Σ)+2Λ,|\beta_1(\sigma,\gamma,\Sigma)|\le|\beta_1(\sigma,\gamma,\Sigma_*)|+2\Lambda,\qquad |\beta_0(\sigma,\gamma,\Sigma)|\le|\beta_0(\sigma,\gamma,\Sigma_*)|+2\Lambda ,

and there are finitely many pairs (σ,γ)(\sigma,\gamma), so K1K_1 is a finite nonnegative real number. Since πA(α)A\pi_{\mathcal{A}}(\alpha)\in\mathcal{A} for every αRm\alpha\in\mathbb{R}^m, Cauchy-Schwarz gives β1(σ,γ,Σ)πA(α)K1R|\beta_1(\sigma,\gamma,\Sigma)\cdot\pi_{\mathcal{A}}(\alpha)|\le K_1R, so β\beta is bounded above and BB is finite; B0B\ge0 because, as shown in Claim 2, β0\beta\ge0.

Claim 2. Let ΣΔl\Sigma\in\Delta^l and αRm\alpha\in\mathbb{R}^m. Then πA(α)A\pi_{\mathcal{A}}(\alpha)\in\mathcal{A}, so the nonnegativity clause of the definition gives β(σ,γ,Σ,α)0\beta(\sigma,\gamma,\Sigma,\alpha)\ge0, and β(σ,γ,Σ,α)B\beta(\sigma,\gamma,\Sigma,\alpha)\le B by the definition of BB. If αA\alpha\in\mathcal{A} then πA(α)=α\pi_{\mathcal{A}}(\alpha)=\alpha by the fixed-point clause of the projection lemma, which gives the stated identity.

For joint continuity, let ΣkΣ\Sigma_k\to\Sigma in Δl\Delta^l and αkα\alpha_k\to\alpha in Rm\mathbb{R}^m in Euclidean distance. The Lipschitz clause gives β0(σ,γ,Σk)β0(σ,γ,Σ)ΛΣkΣ0|\beta_0(\sigma,\gamma,\Sigma_k)-\beta_0(\sigma,\gamma,\Sigma)|\le\Lambda|\Sigma_k-\Sigma|\to0 and β1(σ,γ,Σk)β1(σ,γ,Σ)ΛΣkΣ0|\beta_1(\sigma,\gamma,\Sigma_k)-\beta_1(\sigma,\gamma,\Sigma)|\le\Lambda|\Sigma_k-\Sigma|\to0, while nonexpansiveness of πA\pi_{\mathcal{A}} gives πA(αk)πA(α)αkα0|\pi_{\mathcal{A}}(\alpha_k)-\pi_{\mathcal{A}}(\alpha)|\le|\alpha_k-\alpha|\to0. Hence, by Cauchy-Schwarz,

β1(σ,γ,Σk)πA(αk)β1(σ,γ,Σ)πA(α)ΛRΣkΣ+K1αkα0,\big|\beta_1(\sigma,\gamma,\Sigma_k)\cdot\pi_{\mathcal{A}}(\alpha_k)-\beta_1(\sigma,\gamma,\Sigma)\cdot\pi_{\mathcal{A}}(\alpha)\big|\le\Lambda R\,|\Sigma_k-\Sigma|+K_1|\alpha_k-\alpha|\longrightarrow0 ,

after inserting the intermediate term β1(σ,γ,Σ)πA(αk)\beta_1(\sigma,\gamma,\Sigma)\cdot\pi_{\mathcal{A}}(\alpha_k) and using πA(αk)R|\pi_{\mathcal{A}}(\alpha_k)|\le R. Therefore β(σ,γ,Σk,αk)β(σ,γ,Σ,α)\beta(\sigma,\gamma,\Sigma_k,\alpha_k)\to\beta(\sigma,\gamma,\Sigma,\alpha). Both requirements of a transition-rate family with rate bound BB are met.

Finally, for Σ,ΣΔl\Sigma,\Sigma'\in\Delta^l and αRm\alpha\in\mathbb{R}^m,

β(σ,γ,Σ,α)β(σ,γ,Σ,α)ΛΣΣ+(β1(σ,γ,Σ)β1(σ,γ,Σ))πA(α)Λ(1+R)ΣΣ.|\beta(\sigma,\gamma,\Sigma,\alpha)-\beta(\sigma,\gamma,\Sigma',\alpha)|\le\Lambda|\Sigma-\Sigma'|+\big|\big(\beta_1(\sigma,\gamma,\Sigma)-\beta_1(\sigma,\gamma,\Sigma')\big)\cdot\pi_{\mathcal{A}}(\alpha)\big|\le\Lambda(1+R)|\Sigma-\Sigma'| .

Claim 3. For αA\alpha\in\mathcal{A} we may substitute β(σ,γ,Σ,α)=β0(σ,γ,Σ)+β1(σ,γ,Σ)α\beta(\sigma,\gamma,\Sigma,\alpha)=\beta_0(\sigma,\gamma,\Sigma)+\beta_1(\sigma,\gamma,\Sigma)\cdot\alpha, from Claim 2, into the formula defining the aggregate state drift,

bγ(Σ,α)=σγ(Σσβ(σ,γ,Σ,α)Σγβ(γ,σ,Σ,α)),b^\gamma(\Sigma,\alpha)=\sum_{\sigma\neq\gamma}\big(\Sigma^\sigma\beta(\sigma,\gamma,\Sigma,\alpha)-\Sigma^\gamma\beta(\gamma,\sigma,\Sigma,\alpha)\big),

and collect the terms without α\alpha and the terms linear in α\alpha; this gives exactly b0γ(Σ)+b1γ(Σ)αb^\gamma_0(\Sigma)+b^\gamma_1(\Sigma)\cdot\alpha.

Claim 4. Let Σ,ΣΔl\Sigma,\Sigma'\in\Delta^l and α,αRm\alpha,\alpha'\in\mathbb{R}^m. Since 0Σσ10\le\Sigma^\sigma\le1 and 0βB0\le\beta\le B,

bγ(Σ,α)σγ(ΣσB+ΣγB)2(l1)B,|b^\gamma(\Sigma,\alpha)|\le\sum_{\sigma\neq\gamma}\big(\Sigma^\sigma B+\Sigma^\gamma B\big)\le2(l-1)B ,

and hence b(Σ,α)l2(l1)B|b(\Sigma,\alpha)|\le\sqrt{l}\cdot2(l-1)B.

For the Lipschitz bound in the state, each of the 2(l1)2(l-1) terms constituting bγb^\gamma obeys

Σσβ(σ,γ,Σ,α)(Σ)σβ(σ,γ,Σ,α)Σσ(Σ)σB+(Σ)σβ(σ,γ,Σ,α)β(σ,γ,Σ,α)(B+Λβ)ΣΣ,\big|\Sigma^\sigma\beta(\sigma,\gamma,\Sigma,\alpha)-(\Sigma')^\sigma\beta(\sigma,\gamma,\Sigma',\alpha)\big|\le|\Sigma^\sigma-(\Sigma')^\sigma|\,B+(\Sigma')^\sigma\big|\beta(\sigma,\gamma,\Sigma,\alpha)-\beta(\sigma,\gamma,\Sigma',\alpha)\big|\le(B+\Lambda_\beta)|\Sigma-\Sigma'| ,

using the coordinate bound, (Σ)σ1(\Sigma')^\sigma\le1 and Claim 2, and similarly for the terms with Σγ\Sigma^\gamma. Summing, bγ(Σ,α)bγ(Σ,α)2(l1)(B+Λβ)ΣΣ|b^\gamma(\Sigma,\alpha)-b^\gamma(\Sigma',\alpha)|\le2(l-1)(B+\Lambda_\beta)|\Sigma-\Sigma'|, whence b(Σ,α)b(Σ,α)ΛbΣΣ|b(\Sigma,\alpha)-b(\Sigma',\alpha)|\le\Lambda_b|\Sigma-\Sigma'|.

For the Lipschitz bound in the control, Claim 2 and nonexpansiveness of πA\pi_{\mathcal{A}} give

β(σ,γ,Σ,α)β(σ,γ,Σ,α)=β1(σ,γ,Σ)(πA(α)πA(α))K1αα,|\beta(\sigma,\gamma,\Sigma,\alpha)-\beta(\sigma,\gamma,\Sigma,\alpha')|=\big|\beta_1(\sigma,\gamma,\Sigma)\cdot\big(\pi_{\mathcal{A}}(\alpha)-\pi_{\mathcal{A}}(\alpha')\big)\big|\le K_1|\alpha-\alpha'| ,

and summing the 2(l1)2(l-1) terms, each weighted by a coordinate of Σ\Sigma which is at most 11, gives bγ(Σ,α)bγ(Σ,α)2(l1)K1αα|b^\gamma(\Sigma,\alpha)-b^\gamma(\Sigma,\alpha')|\le2(l-1)K_1|\alpha-\alpha'| and then the stated bound.

Claim 5. Summing the drift over γ\gamma,

γ=1lbγ(Σ,α)=γσγΣσβ(σ,γ,Σ,α)γσγΣγβ(γ,σ,Σ,α).\sum_{\gamma=1}^lb^\gamma(\Sigma,\alpha)=\sum_{\gamma}\sum_{\sigma\neq\gamma}\Sigma^\sigma\beta(\sigma,\gamma,\Sigma,\alpha)-\sum_{\gamma}\sum_{\sigma\neq\gamma}\Sigma^\gamma\beta(\gamma,\sigma,\Sigma,\alpha) .

Both double sums run over all ordered pairs with distinct entries: the first is the sum of Σσβ(σ,γ,Σ,α)\Sigma^\sigma\beta(\sigma,\gamma,\Sigma,\alpha) over such pairs (σ,γ)(\sigma,\gamma), and the second becomes the same sum after interchanging the names of the two summation indices. Hence the difference is 00.

For the inflow bound, the first group of terms is nonnegative because Σσ0\Sigma^\sigma\ge0 and β0\beta\ge0, so

bγ(Σ,α)Σγσγβ(γ,σ,Σ,α)(l1)BΣγ.b^\gamma(\Sigma,\alpha)\ge-\Sigma^\gamma\sum_{\sigma\neq\gamma}\beta(\gamma,\sigma,\Sigma,\alpha)\ge-(l-1)B\,\Sigma^\gamma .

Claim 6. For xRlx\in\mathbb{R}^l we have πΔl(x)Δl\pi_{\Delta^l}(x)\in\Delta^l, so b^(x,α)=b(πΔl(x),α)\hat{b}(x,\alpha)=b(\pi_{\Delta^l}(x),\alpha) is defined, and it satisfies the same bound as bb by Claim 4. For xΔlx\in\Delta^l, the fixed-point clause of the projection lemma gives πΔl(x)=x\pi_{\Delta^l}(x)=x and hence b^(x,α)=b(x,α)\hat{b}(x,\alpha)=b(x,\alpha). Finally, by Claim 4 and nonexpansiveness of πΔl\pi_{\Delta^l},

b^(x,α)b^(x,α)ΛbπΔl(x)πΔl(x)Λbxx.|\hat{b}(x,\alpha)-\hat{b}(x',\alpha)|\le\Lambda_b\big|\pi_{\Delta^l}(x)-\pi_{\Delta^l}(x')\big|\le\Lambda_b|x-x'| . \qquad\blacksquare
Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Prerequisites

Loading...

Comments

Loading…