Adopt the setting, hypotheses (OC), (X), (W) and notation of Uniform Pair-Exponent Bound on the Synthetic Copy: Removed-Clock Intensities, the Insertion Response on the Clock-Good Event, and Bounds on the Pair Covariances (and hence of Pathwise Reduction of the Symmetrised Move Information on the Synthetic Copy: Removed-Clock Likelihoods, the Shifted Family, and the Averaging Bound and The Synthetic Copy: Independent Cell Structure, Deterministic-Count Clocks, the Copy Measure, and the Smoothed Joint Density of Parameter and Observation Record): the probability space (Ω,F,P) with expectation E, the cells indexed by L (d elements) with lengths μq (q∈L), the cell-count vector K=(Kq)q∈L, the move size m, the weights w=(wq)q∈L with ∥w∥1=∑q∣wq∣, the removal ratios ϱq and the random variables ϱq(Kq), the symmetrised kernel-weighted move information Jsym, the likelihoods ℓ♯,ω and ℓ−q,ω, the pathwise score map Ψ, the constants Γ, A0 and EˉN, the clock-good event GL,D, the tracked records Tω, the effective removed intensities λ−q,ω with likelihoods ℓλ−q,ω and likelihood ratios Lqω=ℓλ−q,ω/ℓ♯,ω, and the pair covariances Cqq′ω of Pair Expansion of the Weighted Likelihood-Ratio Square: Pair Covariances of Causal-Intensity Likelihoods, the Exact Expansion, and Its First-Order Form for the base λ♯,ω and the perturbed intensities (λ−q,ω)q. Let ϖμ be the Chernoff exponent of Series Formula, Exponential Moments, and Chernoff Tail Bounds for the Poisson Distribution, exp the real exponential function, t1/2 the nonnegative square root, and 1A the indicator of an event A. Put
ε0=NbΓA0,cN=exp(EˉN)−1,jq=E[ϱq(Kq)2]−1=∑i=1m(im)2i!μq−i(q∈L),
the last identity by claim 2 of Pathwise Reduction of the Symmetrised Move Information on the Synthetic Copy: Removed-Clock Likelihoods, the Shifted Family, and the Averaging Bound. Assume moreover:
(G) G∈F is an event with G⊆GL,D and ρ(R∖Tω)=0 for every ω∈G; put g=P(Ω∖G);
(P) δ is a real number with 0<δ≤1; ε0≤21; θ≥1 is an integer with 2θε0≤1; and for every q∈L, xq>0 is a real number with m(xq+m)≤δμq/2. Put
Eθch=8l~TNB~θ2ε02,Πˉ=∑q∈L(exp(−ϖμq(xq))+exp(−θδ/2+Eθch)).
For ω∈Ω and q∈L define the under-likelihood mass πqω=∫Rℓ♯,ω1{ϱq(Kq(ω))Lqω<1−δ}dρ∈[0,1] and πω=∑qπqω, and the weighted square
Qω=∫Rℓ♯,ω(∑q∈Lwq(1−ϱq(Kq(ω))Lqω))2dρ=(∑qwq(1−ϱq(Kq(ω))))2+∑q,q′∈Lwqwq′ϱq(Kq(ω))ϱq′(Kq′(ω))Cqq′ω
(the identity by claim 2 of Pair Expansion of the Weighted Likelihood-Ratio Square: Pair Covariances of Causal-Intensity Likelihoods, the Exact Expansion, and Its First-Order Form).
1. (Prior part) The family (Kq)q∈L is independent, hence so is (ϱq(Kq))q∈L, and
E[(∑q∈Lwq(1−ϱq(Kq)))2]=∑q∈Lwq2jq.
2. (Under-likelihood mass on the good event) The maps ω↦πqω (values in [0,1]), ω↦Qω (finite values), ω↦Cqq′ω (finite values) and ω↦∫RΨ(r,ω)ρ(dr) (values in [0,∞]) are F-measurable. (Throughout, q∈L is identified with its image in {1,…,d} under the fixed bijection when the cited lemmas index by {1,…,d}.) For every ω∈G and q∈L,
πqω≤1{Kq(ω)<μq−xq}+exp(−θδ/2+Eθch),henceE[1Gπω]≤Πˉ.
3. (Assembly) The record part C=E[1G(Qω−(∑qwq(1−ϱq(Kq)))2)] is a well-defined real number, equal to ∑q,q′wqwq′E[1Gϱq(Kq)ϱq′(Kq′)Cqq′ω], with ∣Cqq′ω∣≤cN on G, and
Jsym ≤ (1+δ)[∑q∈Lwq2jq+C]+2d∥w∥12B,
where the bad part is
B=Πˉ+∑q∈L(1+jq)1/2((dΠˉ)1/2+(cNΠˉ)1/2)+g+g1/2∑q∈L(1+jq)1/2.