Adopt the setting and notation of Measurable Reconstruction of the Controlled N-Agent Dynamics from Observation Records: natural numbers N≥1, l≥2, l~≥1, m≥1, a nonempty control set A⊆Rm in Euclidean space, a transition-rate family β with control set A and rate bound B, an observation-rate family β~ with rate bound B~, a horizon T>0, an N-agent driving system (Ω,F,P) with initial states ς0i, transition clocks Yi,σγ and observation clocks Y~i,υ, an A-valued observation-driven control policy h, the observation record space R with record σ-algebra R, the record-frozen control path ar and record-frozen policy h^r at r∈R, the σ-algebra T, the reconstructed states σsr,i, occupation indicators ηsr,i,γ and empirical measures Σsr, the good set G, and, for each r∈R, the event Ωr and the solution of the controlled N-agent dynamics for the policy h^r furnished by clause (d) of that lemma, whose state processes are the σr,i, whose control process is s↦ar(s), and whose regular event is Ωr. For this solution write Tti,σγ for its consumed transition-clock times, Nti,σγ=YTti,σγi,σγ for its transition counters, and (Ftsys,r)t∈[0,T] for its system filtration, all as in Solution of the Controlled N-Agent Dynamics (the dependence on the record r is suppressed in the first two symbols; from the second paragraph on, r is fixed). Let GN be the aggregate lattice (a nonempty finite set), let c=(σ,γ) range over the transition labels with the vectors vc=δγ−δσ of Open-Loop Aggregate Solution Driven by Aggregate Transition Clocks, write E for the expectation under P, B[0,T] for the trace Borel σ-algebra on [0,T], integrals over compact intervals for the Lebesgue integral over the compact interval (read as 0 over a degenerate interval), 1S for the indicator of a set S, and 1{⋯} for the indicator of the condition in the braces.
Fix r∈R. Let E={1,…,l}N, a nonempty finite set with lN elements; for x=(x1,…,xN)∈E let Σ(x)∈Rl be the point with coordinates Σ(x)γ=N1#{i∈{1,…,N}:xi=γ}, a point of GN. Put
Xt=(σtr,1,…,σtr,N)∈E,Σ^tr=Σ(Xt)∈GN(t∈[0,T]),
both defined at every ω∈Ω. Let A be the set of pairs e=(i,c) of an agent index i∈{1,…,N} and a transition label c=(σ,γ), a nonempty finite set with Nl(l−1) elements (the clock labels, written a in Forward Equation for a Finite-State Jump System Driven by Poisson Clocks with the Fresh-Start Property, are written e here, ar being reserved for the record-frozen control path), and for e=(i,(σ,γ))∈A, u∈[0,T] and x∈E put
ge(u,x)=1{xi=σ}β(σ,γ,Σ(x),ar(u))∈[0,B],ϕe(x)i=γ,ϕe(x)j=xj(j=i),
the last two formulas defining the point ϕe(x)∈E coordinatewise. Finally, for u∈[0,T] and y=y′ in GN put qur(y,y′)=Nyσβ(σ,γ,y,ar(u)) if y′=y+N1vc for the (then unique) transition label c=(σ,γ), and qur(y,y′)=0 if y′−y is not of the form N1vc. Then:
1. (The reconstructed solution is a Poisson-clock jump system.) On Ωr one has ηtr,i,γ=1{σtr,i=γ} for all t∈[0,T], i and γ, and Σ^tr=Σtr for all t∈[0,T]; moreover ar is a control path in the sense of that definition. The data consisting of (Ω,F,P), T, Λ=B, the state space E, the clock labels A, the clocks Y(i,(σ,γ))=Yi,σγ, the rate functions ge, the transition maps ϕe, the filtration (Ftsys,r)t∈[0,T], the event Ω0=Ωr, the state process X and the consumed clock times Tt(i,(σ,γ))=Tti,σγ satisfy (D1)--(D4), (H1) and (H2) of Forward Equation for a Finite-State Jump System Driven by Poisson Clocks with the Fresh-Start Property, and the counters of that lemma are the transition counters Nti,σγ. Consequently conclusions (a) and (b) of that lemma hold for these data (the time parameter written r in (H2) and in conclusions (a) and (b) of that lemma is here written s; the letter r is reserved for the record).
2. (Forward equation on the aggregate lattice.) For every function F:GN→R, all 0≤s≤t≤T and every event D∈Fssys,r,
E[(F(Σ^tr)−F(Σ^sr))1D]=E[1D∫[s,t]1Ωrc=(σ,γ)∑NΣ^ur,σβ(σ,γ,Σ^ur,ar(u))(F(Σ^ur+N1vc)−F(Σ^ur))du],
where Σ^ur,σ is the σ-th coordinate of Σ^ur, where F(Σ^ur+N1vc) is read as F(Σ^ur) when Σ^ur+N1vc∈/GN (in which case Σ^ur,σ=0, so the term vanishes anyway), and where the inner integral is defined for every ω and is F-measurable, its integrand being a bounded section of a B[0,T]⊗F-measurable map. Moreover, for all y=y′ in GN the map u↦qur(y,y′) is measurable on [0,T] with respect to B[0,T], ∑y′=yqur(y,y′)≤l(l−1)NB for all u∈[0,T] and y∈GN, and for every s∈[0,T) and D∈Fssys,r the family
νuD(y)=P(D∩{Σ^ur=y})=P(D∩{Σur=y})(u∈[s,T], y∈GN)
(both sets in the braces being events) is a solution of the forward equation on [s,T] for the rates qr in the sense of that lemma, applied with the rate bound there taken to be l(l−1)NB and with its pairing ν(F)=∑y∈GNν(y)F(y).