Throughout, "the recursion" is the aggregate recursion of Existence, Uniqueness, Causality, and Measurability of the Open-Loop Aggregate Solution for (P(ω),a,x0), with the notation of that theorem (θk, x(k), κkc, Cc,(k), λkc, hkc, Jk, K), and c=(σ,γ) always denotes a transition label; there are 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 is a counting path by property 1 of Poisson Clock with a Horizon.
(D2). For fixed x∈GN the map u↦β(σ,γ,x,au) is measurable on [0,T]: the control path a has measurable components, and α↦β(σ,γ,x,α) is sequentially continuous on 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 Nxσ keeps measurability (Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions), and 0≤gc(u,x)≤NB because 0≤xσ≤1 and 0≤β≤B (condition 1 of Transition-Rate Family). Thus Λ=NB serves. The maps ϕc take values in GN by construction.
(D3). The recursion filtration is nested by definition. Let ι:Rl→GN be the map with ι(x)=x for x∈GN and ι(x)=x0 otherwise; since GN is finite, ι−1({x}) is a Borel subset of Rl (a singleton, or the complement of a finite set united with a singleton) for every x∈GN. As Σˉt=ι(Σt) and each coordinate Σtγ is a generator of Ft, the event {Σˉt=x}={Σt∈ι−1({x})} belongs to Ft: {Σt=x}=⋂γ{Σtγ=xγ} for each x, and {Σt∈/GN} is the complement of the finite union of the events {Σt=x}, x∈GN. 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) is B[0,T]⊗F-measurable for each γ, so {(u,ω):Σu(ω)=x} is product-measurable for each x∈Rl, hence so is {(u,ω):Σˉu(ω)=x} for x∈GN (equal to {Σu=x} for x=x0, and to {Σu=x0}∪{Σu∈/GN} for x=x0), and multiplying its indicator by 1Ω0(ω) (a rectangle indicator) preserves product measurability.
(D4). Each Ctc is a random variable with values in [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 ω∈Ω0. 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) 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; since Σ=Σˉ on Ω0 (the solution takes values in GN), this is ∫[0,t]gc(u,Σˉu)du. Hence (D4) holds with Ttc=Ctc, and the counters of the generic lemma are Pc(Ctc)=Ntc.
(H1). Fix ω∈Ω0, so that the recursion stops at K with θK=T and x(k)∈GN for all k≤K (the recursion stops at the first index where a point leaves GN, so conflict-freeness forces all points to lie in GN), and Σˉ=Σ. The path t↦Σt equals x(k) on [θk,θk+1) for k<K and x(K) at θK=T, with 0=θ0<θ1<⋯<θK=T (claim 1): this is the piecewise-constant right-continuous structure required in (H1), with the times θ1,…,θK (the count of (H1) coincides with the stopping index K of the recursion). Counters between and at recursion times. Fix k<K and a label c, and let j=Pc(κkc)+1, so that λkc=τj(Pc(ω)), the greatest lower bound of the times at which the counting path Pc(ω) has value at least j (+∞ if there are none). This path is at least j−1 on [κkc,∞) by monotonicity, and below j on [0,λkc) by the definition of λkc, so it equals j−1 on [κkc,λkc); and when λkc<+∞ it equals j at λkc (its value there is at least j by right-continuity, and at most (j−1)+1 since its left limit there is at most j−1 and jumps are of size at most 1; Counting Path and Its Jump Times). Only the case λkc<+∞ is used below, since it is forced for c∈Jk by Cθk+1c,(k)≥λkc. For t∈[θk,θk+1) one has κkc≤Ctc,(k)<λkc (the first inequality by monotonicity of Cc,(k) from its value κkc at θk; the second because Ctc,(k)≥λkc would give hkc≤t<θk+1≤hkc), so Ntc=Pc(Ctc,(k))=j−1 is constant there. At t=θk+1, Nθk+1c=Pc(κk+1c) with κk+1c=Cθk+1c,(k): if c∈Jk then κk+1c=λkc (claim 1) and Nθk+1c=j, while if c∈/Jk then κkc≤κk+1c<λkc and Nθk+1c=j−1. Hence every counter is constant on each [θk,θk+1), and at θk+1 exactly the counters with labels in Jk jump, each by 1. The jump rules. Let t∈(0,T] and let Nttot=∑cNtc be the grand total (written Nt in Forward Equation for a Finite-State Jump System Driven by Poisson Clocks with the Fresh-Start Property; here N is the number of agents). If t is not one of θ1,…,θK, then t lies in the interior of some [θk,θk+1), where all counters and Σ are constant, so Nttot=Nt−tot and Σt=Σt−. If t=θk+1 with k<K: the left limits are the constant values on [θk,θk+1), so Nttot−Nt−tot=∣Jk∣ and Σt−Σt−=x(k+1)−x(k)=N1∑c∈Jkvc. If Jk=∅ this gives Nttot=Nt−tot and Σt=Σt−; if Jk={c} then Nttot=Nt−tot+1, c is the unique label whose counter jumps at t, and Σt=Σt−+N1vc, which lies in GN (it is x(k+1)), so Σt=ϕc(Σt−); if ∣Jk∣≥2 nothing is required. This is (H1).
(H2). Fix r∈[0,T]. We apply Fresh-Start Property for Independent Poisson Clocks Read at Levels Satisfying a Clock-Reading Bound with the clock labels A= the transition labels, the initial data I={∅,Ω}, the clocks Yc=Pc (Poisson clocks with horizon R; the family consisting of the trivial σ-algebra I and the σ(Puc:u≥0) is independent, since adjoining {∅,Ω} to an independent family keeps every product formula true), the past F=Fr, the consumed levels Tc=Crc and cˉ=NBr: these levels are Fr-measurable with 0≤Crc≤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+uc−PCrcc consist of random variables, have independent Poisson increments (with the parameters of (H2)) over every finite time set in [0,R−) with R−=R−NBr, and the family consisting of Fr and the σ-algebras σ(Y^uc,r:0≤u<R−) is independent. Since ϱr=Λ(T−r)=NB(T−r)<R−NBr=R− (because NBT<R), every finite time set in [0,ϱr] lies in [0,R−) and σ(Y^uc,r:0≤u≤ϱr)⊆σ(Y^uc,r:0≤u<R−); independence of a family of σ-algebras passes to sub-σ-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=Σˉ and yields the displayed identity, the integrand there being 1Ω0∑cgc(u,Σˉu)(F(ϕc(Σˉu))−F(Σˉu)); when Σˉu+N1vc∈/GN one has ϕc(Σˉu)=Σˉu and the term vanishes, which matches the stated reading (and, for x=Σˉu∈GN (which holds at every ω, by the definition of Σˉ), the point x+N1vc=x+N1(δγ−δσ) fails to lie in GN exactly when xσ=0, its coordinates otherwise being nonnegative multiples of N1 summing to 1). Claim (b) of the generic lemma gives that (μuD)u∈[r,T] solves the forward equation on [r,T] for the rates qu(x,y)=∑c:ϕc(x)=ygc(u,x) with rate bound ∣A∣Λ=l(l−1)NB. Finally, for x=y: if y=x+N1vc for some label c then c is unique (vc=δγ−δσ determines (σ,γ)), ϕc(x)=y, no other label c′ has ϕc′(x)=y (as ϕc′(x)∈{x,x+N1vc′}), and qu(x,y)=gc(u,x)=Nxσβ(σ,γ,x,au); otherwise no label has ϕc(x)=y and qu(x,y)=0, as stated.