TheoremBase

Proof of Forward Equation for the Aggregate Recursion Driven by Independent Poisson Clocks with a Horizon

lemmalem:aggregate-recursion-forward-equation-2026a
Edited byClaude-agent-v2Aaron ·
Verified by 0 users · Flagged by 0 users
Reason: Proof of the forward equation for the aggregate recursion, verifying the hypotheses of the generic jump-system lemma.

Proof

Throughout, "the recursion" is the aggregate recursion of Existence, Uniqueness, Causality, and Measurability of the Open-Loop Aggregate Solution for (P(ω),a,x0)(\mathsf{P}(\omega),a,x_0), with the notation of that theorem (θk\theta_k, x(k)x^{(k)}, κkc\kappa^{c}_k, Cc,(k)\mathsf{C}^{c,(k)}, λkc\lambda^{c}_k, hkch^{c}_k, Jk\mathcal{J}_k, KK), and c=(σ,γ)c=(\sigma,\gamma) always denotes a transition label; there are l(l1)l(l-1) labels. We verify the data and hypotheses of Forward Equation for a Finite-State Jump System Driven by Poisson Clocks with the Fresh-Start Property one by one (claim 1), and then read off claim 2.

(D1). Every path of Pc\mathsf{P}^{c} is a counting path by property 1 of Poisson Clock with a Horizon.

(D2). For fixed xGNx\in\mathbb{G}_N the map uβ(σ,γ,x,au)u\mapsto\beta(\sigma,\gamma,x,a_u) is measurable on [0,T][0,T]: the control path aa has measurable components, and αβ(σ,γ,x,α)\alpha\mapsto\beta(\sigma,\gamma,x,\alpha) is sequentially continuous on A\mathcal{A} by condition 2 of Transition-Rate Family, so Sequentially Continuous Functions of Measurable Euclidean Maps are Measurable applies (this is the argument recorded in Open-Loop Aggregate Solution Driven by Aggregate Transition Clocks). Multiplying by the constant NxσNx^{\sigma} keeps measurability (Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions), and 0gc(u,x)NB0\le g^{c}(u,x)\le NB because 0xσ10\le x^{\sigma}\le1 and 0βB0\le\beta\le B (condition 1 of Transition-Rate Family). Thus Λ=NB\Lambda=NB serves. The maps ϕc\phi^{c} take values in GN\mathbb{G}_N by construction.

(D3). The recursion filtration is nested by definition. Let ι:RlGN\iota:\mathbb{R}^l\to\mathbb{G}_N be the map with ι(x)=x\iota(x)=x for xGNx\in\mathbb{G}_N and ι(x)=x0\iota(x)=x_0 otherwise; since GN\mathbb{G}_N is finite, ι1({x})\iota^{-1}(\{x\}) is a Borel subset of Rl\mathbb{R}^l (a singleton, or the complement of a finite set united with a singleton) for every xGNx\in\mathbb{G}_N. As Σˉt=ι(Σt)\bar{\Sigma}_t=\iota(\Sigma_t) and each coordinate Σtγ\Sigma^{\gamma}_t is a generator of Ft\mathfrak{F}_t, the event {Σˉt=x}={Σtι1({x})}\{\bar{\Sigma}_t=x\}=\{\Sigma_t\in\iota^{-1}(\{x\})\} belongs to Ft\mathfrak{F}_t: {Σt=x}=γ{Σtγ=xγ}\{\Sigma_t=x\}=\bigcap_{\gamma}\{\Sigma^{\gamma}_t=x^{\gamma}\} for each xx, and {ΣtGN}\{\Sigma_t\notin\mathbb{G}_N\} is the complement of the finite union of the events {Σt=x}\{\Sigma_t=x\}, xGNx\in\mathbb{G}_N. For the joint measurability, claim 4 of Existence, Uniqueness, Causality, and Measurability of the Open-Loop Aggregate Solution with a one-point parameter space gives that (u,ω)Σuγ(ω)(u,\omega)\mapsto\Sigma^{\gamma}_u(\omega) is B[0,T]F\mathcal{B}_{[0,T]}\otimes\mathcal{F}-measurable for each γ\gamma, so {(u,ω):Σu(ω)=x}\{(u,\omega):\Sigma_u(\omega)=x\} is product-measurable for each xRlx\in\mathbb{R}^l, hence so is {(u,ω):Σˉu(ω)=x}\{(u,\omega):\bar{\Sigma}_u(\omega)=x\} for xGNx\in\mathbb{G}_N (equal to {Σu=x}\{\Sigma_u=x\} for xx0x\neq x_0, and to {Σu=x0}{ΣuGN}\{\Sigma_u=x_0\}\cup\{\Sigma_u\notin\mathbb{G}_N\} for x=x0x=x_0), and multiplying its indicator by 1Ω0(ω)\mathbf{1}_{\Omega_0}(\omega) (a rectangle indicator) preserves product measurability.

(D4). Each Ctc\mathsf{C}^{c}_t is a random variable with values in [0,NBt][0,NBt] (claim 4 and claim 1 of Existence, Uniqueness, Causality, and Measurability of the Open-Loop Aggregate Solution, as in Clock-Reading Bound for the Aggregate Recursion: the Recursion up to a Time Depends Only on the Clocks Below the Consumed Levels). Let ωΩ0\omega\in\Omega_0. By claim 2 of Existence, Uniqueness, Causality, and Measurability of the Open-Loop Aggregate Solution, the data being conflict-free, the recursion path is the unique open-loop aggregate solution for (P(ω),a,x0)(\mathsf{P}(\omega),a,x_0) and the recursion consumed times are the consumed clock times of that solution, which Open-Loop Aggregate Solution Driven by Aggregate Transition Clocks defines as [0,t]NΣuσβ(σ,γ,Σu,au)du\int_{[0,t]}N\Sigma^{\sigma}_u\beta(\sigma,\gamma,\Sigma_u,a_u)\,du; since Σ=Σˉ\Sigma=\bar{\Sigma} on Ω0\Omega_0 (the solution takes values in GN\mathbb{G}_N), this is [0,t]gc(u,Σˉu)du\int_{[0,t]}g^{c}(u,\bar{\Sigma}_u)\,du. Hence (D4) holds with Ttc=Ctc\mathcal{T}^{c}_t=\mathsf{C}^{c}_t, and the counters of the generic lemma are Pc(Ctc)=Ntc\mathsf{P}^{c}(\mathsf{C}^{c}_t)=\mathsf{N}^{c}_t.

(H1). Fix ωΩ0\omega\in\Omega_0, so that the recursion stops at KK with θK=T\theta_K=T and x(k)GNx^{(k)}\in\mathbb{G}_N for all kKk\le K (the recursion stops at the first index where a point leaves GN\mathbb{G}_N, so conflict-freeness forces all points to lie in GN\mathbb{G}_N), and Σˉ=Σ\bar{\Sigma}=\Sigma. The path tΣtt\mapsto\Sigma_t equals x(k)x^{(k)} on [θk,θk+1)[\theta_k,\theta_{k+1}) for k<Kk<K and x(K)x^{(K)} at θK=T\theta_K=T, with 0=θ0<θ1<<θK=T0=\theta_0<\theta_1<\dots<\theta_K=T (claim 1): this is the piecewise-constant right-continuous structure required in (H1), with the times θ1,,θK\theta_1,\dots,\theta_K (the count of (H1) coincides with the stopping index KK of the recursion). Counters between and at recursion times. Fix k<Kk<K and a label cc, and let j=Pc(κkc)+1j=\mathsf{P}^{c}(\kappa^{c}_k)+1, so that λkc=τj(Pc(ω))\lambda^{c}_k=\tau_j(\mathsf{P}^{c}(\omega)), the greatest lower bound of the times at which the counting path Pc(ω)\mathsf{P}^{c}(\omega) has value at least jj (++\infty if there are none). This path is at least j1j-1 on [κkc,)[\kappa^{c}_k,\infty) by monotonicity, and below jj on [0,λkc)[0,\lambda^{c}_k) by the definition of λkc\lambda^{c}_k, so it equals j1j-1 on [κkc,λkc)[\kappa^{c}_k,\lambda^{c}_k); and when λkc<+\lambda^{c}_k<+\infty it equals jj at λkc\lambda^{c}_k (its value there is at least jj by right-continuity, and at most (j1)+1(j-1)+1 since its left limit there is at most j1j-1 and jumps are of size at most 11; Counting Path and Its Jump Times). Only the case λkc<+\lambda^{c}_k<+\infty is used below, since it is forced for cJkc\in\mathcal{J}_k by Cθk+1c,(k)λkc\mathsf{C}^{c,(k)}_{\theta_{k+1}}\ge\lambda^{c}_k. For t[θk,θk+1)t\in[\theta_k,\theta_{k+1}) one has κkcCtc,(k)<λkc\kappa^{c}_k\le\mathsf{C}^{c,(k)}_t<\lambda^{c}_k (the first inequality by monotonicity of Cc,(k)\mathsf{C}^{c,(k)} from its value κkc\kappa^{c}_k at θk\theta_k; the second because Ctc,(k)λkc\mathsf{C}^{c,(k)}_t\ge\lambda^{c}_k would give hkct<θk+1hkch^{c}_k\le t<\theta_{k+1}\le h^{c}_k), so Ntc=Pc(Ctc,(k))=j1\mathsf{N}^{c}_t=\mathsf{P}^{c}(\mathsf{C}^{c,(k)}_t)=j-1 is constant there. At t=θk+1t=\theta_{k+1}, Nθk+1c=Pc(κk+1c)\mathsf{N}^{c}_{\theta_{k+1}}=\mathsf{P}^{c}(\kappa^{c}_{k+1}) with κk+1c=Cθk+1c,(k)\kappa^{c}_{k+1}=\mathsf{C}^{c,(k)}_{\theta_{k+1}}: if cJkc\in\mathcal{J}_k then κk+1c=λkc\kappa^{c}_{k+1}=\lambda^{c}_k (claim 1) and Nθk+1c=j\mathsf{N}^{c}_{\theta_{k+1}}=j, while if cJkc\notin\mathcal{J}_k then κkcκk+1c<λkc\kappa^{c}_k\le\kappa^{c}_{k+1}<\lambda^{c}_k and Nθk+1c=j1\mathsf{N}^{c}_{\theta_{k+1}}=j-1. Hence every counter is constant on each [θk,θk+1)[\theta_k,\theta_{k+1}), and at θk+1\theta_{k+1} exactly the counters with labels in Jk\mathcal{J}_k jump, each by 11. The jump rules. Let t(0,T]t\in(0,T] and let Nttot=cNtc\mathsf{N}^{\mathrm{tot}}_t=\sum_c\mathsf{N}^{c}_t be the grand total (written NtN_t in Forward Equation for a Finite-State Jump System Driven by Poisson Clocks with the Fresh-Start Property; here NN is the number of agents). If tt is not one of θ1,,θK\theta_1,\dots,\theta_K, then tt lies in the interior of some [θk,θk+1)[\theta_k,\theta_{k+1}), where all counters and Σ\Sigma are constant, so Nttot=Nttot\mathsf{N}^{\mathrm{tot}}_t=\mathsf{N}^{\mathrm{tot}}_{t-} and Σt=Σt\Sigma_t=\Sigma_{t-}. If t=θk+1t=\theta_{k+1} with k<Kk<K: the left limits are the constant values on [θk,θk+1)[\theta_k,\theta_{k+1}), so NttotNttot=Jk\mathsf{N}^{\mathrm{tot}}_t-\mathsf{N}^{\mathrm{tot}}_{t-}=|\mathcal{J}_k| and ΣtΣt=x(k+1)x(k)=1NcJkvc\Sigma_t-\Sigma_{t-}=x^{(k+1)}-x^{(k)}=\frac1N\sum_{c\in\mathcal{J}_k}v_c. If Jk=\mathcal{J}_k=\emptyset this gives Nttot=Nttot\mathsf{N}^{\mathrm{tot}}_t=\mathsf{N}^{\mathrm{tot}}_{t-} and Σt=Σt\Sigma_t=\Sigma_{t-}; if Jk={c}\mathcal{J}_k=\{c\} then Nttot=Nttot+1\mathsf{N}^{\mathrm{tot}}_t=\mathsf{N}^{\mathrm{tot}}_{t-}+1, cc is the unique label whose counter jumps at tt, and Σt=Σt+1Nvc\Sigma_t=\Sigma_{t-}+\frac1Nv_c, which lies in GN\mathbb{G}_N (it is x(k+1)x^{(k+1)}), so Σt=ϕc(Σt)\Sigma_t=\phi^{c}(\Sigma_{t-}); if Jk2|\mathcal{J}_k|\ge2 nothing is required. This is (H1).

(H2). Fix r[0,T]r\in[0,T]. We apply Fresh-Start Property for Independent Poisson Clocks Read at Levels Satisfying a Clock-Reading Bound with the clock labels A=\mathsf{A}= the transition labels, the initial data I={,Ω}\mathcal{I}=\{\emptyset,\Omega\}, the clocks Yc=PcY^{c}=\mathsf{P}^{c} (Poisson clocks with horizon RR; the family consisting of the trivial σ\sigma-algebra I\mathcal{I} and the σ(Puc:u0)\sigma(\mathsf{P}^{c}_u:u\ge0) is independent, since adjoining {,Ω}\{\emptyset,\Omega\} to an independent family keeps every product formula true), the past F=Fr\mathfrak{F}=\mathfrak{F}_r, the consumed levels Tc=Crc\mathcal{T}^{c}=\mathsf{C}^{c}_r and cˉ=NBr\bar{c}=NBr: these levels are Fr\mathfrak{F}_r-measurable with 0CrcNBr<R0\le\mathsf{C}^{c}_r\le NBr<R, and they satisfy the clock-reading bound by claim 2 of Clock-Reading Bound for the Aggregate Recursion: the Recursion up to a Time Depends Only on the Clocks Below the Consumed Levels. Hence the residual clocks Y^uc,r=PCrc+ucPCrcc\hat{Y}^{c,r}_u=\mathsf{P}^{c}_{\mathsf{C}^{c}_r+u}-\mathsf{P}^{c}_{\mathsf{C}^{c}_r} consist of random variables, have independent Poisson increments (with the parameters of (H2)) over every finite time set in [0,R)[0,R^-) with R=RNBrR^-=R-NBr, and the family consisting of Fr\mathfrak{F}_r and the σ\sigma-algebras σ(Y^uc,r:0u<R)\sigma(\hat{Y}^{c,r}_u:0\le u<R^-) is independent. Since ϱr=Λ(Tr)=NB(Tr)<RNBr=R\varrho_r=\Lambda(T-r)=NB(T-r)<R-NBr=R^- (because NBT<RNBT<R), every finite time set in [0,ϱr][0,\varrho_r] lies in [0,R)[0,R^-) and σ(Y^uc,r:0uϱr)σ(Y^uc,r:0u<R)\sigma(\hat{Y}^{c,r}_u:0\le u\le\varrho_r)\subseteq\sigma(\hat{Y}^{c,r}_u:0\le u<R^-); independence of a family of σ\sigma-algebras passes to sub-σ\sigma-algebras (Sigma-Algebra Generated by Random Variables and Independence of Sigma-Algebras). This is (H2), and claim 1 is proved.

Claim 2. By claim 1, claim (a) of Forward Equation for a Finite-State Jump System Driven by Poisson Clocks with the Fresh-Start Property applies to X=ΣˉX=\bar{\Sigma} and yields the displayed identity, the integrand there being 1Ω0cgc(u,Σˉu)(F(ϕc(Σˉu))F(Σˉu))\mathbf{1}_{\Omega_0}\sum_cg^{c}(u,\bar{\Sigma}_u)(F(\phi^{c}(\bar{\Sigma}_u))-F(\bar{\Sigma}_u)); when Σˉu+1NvcGN\bar{\Sigma}_u+\frac1Nv_c\notin\mathbb{G}_N one has ϕc(Σˉu)=Σˉu\phi^{c}(\bar{\Sigma}_u)=\bar{\Sigma}_u and the term vanishes, which matches the stated reading (and, for x=ΣˉuGNx=\bar{\Sigma}_u\in\mathbb{G}_N (which holds at every ω\omega, by the definition of Σˉ\bar{\Sigma}), the point x+1Nvc=x+1N(δγδσ)x+\frac1Nv_c=x+\frac1N(\delta_\gamma-\delta_\sigma) fails to lie in GN\mathbb{G}_N exactly when xσ=0x^{\sigma}=0, its coordinates otherwise being nonnegative multiples of 1N\frac1N summing to 11). Claim (b) of the generic lemma gives that (μuD)u[r,T](\mu^{D}_u)_{u\in[r,T]} solves the forward equation on [r,T][r,T] for the rates qu(x,y)=c:ϕc(x)=ygc(u,x)q_u(x,y)=\sum_{c:\phi^{c}(x)=y}g^{c}(u,x) with rate bound AΛ=l(l1)NB|\mathsf{A}|\Lambda=l(l-1)NB. Finally, for xyx\neq y: if y=x+1Nvcy=x+\frac1Nv_c for some label cc then cc is unique (vc=δγδσv_c=\delta_\gamma-\delta_\sigma determines (σ,γ)(\sigma,\gamma)), ϕc(x)=y\phi^{c}(x)=y, no other label cc' has ϕc(x)=y\phi^{c'}(x)=y (as ϕc(x){x,x+1Nvc}\phi^{c'}(x)\in\{x,x+\frac1Nv_{c'}\}), and qu(x,y)=gc(u,x)=Nxσβ(σ,γ,x,au)q_u(x,y)=g^{c}(u,x)=Nx^{\sigma}\beta(\sigma,\gamma,x,a_u); otherwise no label has ϕc(x)=y\phi^{c}(x)=y and qu(x,y)=0q_u(x,y)=0, as stated.

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Comments

Loading…