Lemma 5.4.1 — the posterior's explicit -density
ProvedMDPFinance.BayesianModels.lemma_5_4_1Lemma 5.4.1. The posterior distribution of a Bayesian Model has an explicit closed form as a -density: for every history and ,
This is Bayes' rule made explicit: the posterior density (relative to the prior ) is proportional to the likelihood of the observed disturbances given .
Proof idea (not part of the formalization). Unrolling Posterior's one-step Bayes update
times from telescopes the ratio of one-step likelihoods into the single
-fold product shown above.
Moderation note. Stated with Lebesgue integrals, for histories along which the book's quotients are defined: the normalizing constants of the first updates are finite and the -th is positive (an observed history of positive likelihood). The draft's real Bochner quotient was a default when the product is not integrable, and undefined () on histories of zero likelihood, where is a probability measure nonetheless.
import Mathlib import Definitions.Def_MDPFinance_BayesianModels_Model import Definitions.Def_MDPFinance_BayesianModels_Posterior open MeasureTheory ProbabilityTheory
namespace MDPFinance.BayesianModels
/-- Lemma 5.4.1 (Bäuerle–Rieder, p. 160, PDF 173). The posterior distribution `μ_n` has a
`Q_0`-density, i.e. `μ_n(C|h̃_n) = [∫_C ∏_{k=0}^{n-1} q_Z(x_k,θ,a_k,z_{k+1}) Q_0(dθ)] /
[∫_Θ ∏_{k=0}^{n-1} q_Z(x_k,θ,a_k,z_{k+1}) Q_0(dθ)]`, `C ∈ B(Θ)`. Follows from `Posterior`'s
one-step Bayes update `hmu_rec` (this chunk's `Def_..._Posterior`) by induction on `n`: unrolling
`n` applications of the ratio-of-integrals update starting from `μ_0 = Q_0` (`hmu0`) telescopes
into the single `n`-fold-product ratio against `Q_0` stated here. Stated for histories along
which the book's quotients are defined: the normalizing constants of the first `n` updates are
finite and the `n`-th is positive (an observed history of positive likelihood), with Lebesgue
integrals. -/
theorem lemma_5_4_1 {EX Θ A Z : Type*} [MeasurableSpace EX] [MeasurableSpace Θ]
[MeasurableSpace A] [MeasurableSpace Z] (M : BayesModel EX Θ A Z) (Pf : M.Posterior) (n : ℕ)
(xs : ℕ → EX) (as : ℕ → A) (zs : ℕ → Z)
(hfin : ∀ k ≤ n, (∫⁻ θ, ∏ j ∈ Finset.range k,
ENNReal.ofReal (M.qZ (xs j) θ (as j) (zs (j + 1))) ∂M.Q0) < ⊤)
(hpos : 0 < ∫⁻ θ, ∏ j ∈ Finset.range n,
ENNReal.ofReal (M.qZ (xs j) θ (as j) (zs (j + 1))) ∂M.Q0)
(C : Set Θ) (hC : MeasurableSet C) :
Pf.mu n xs as zs C =
(∫⁻ θ in C, ∏ k ∈ Finset.range n, ENNReal.ofReal (M.qZ (xs k) θ (as k) (zs (k + 1))) ∂M.Q0) /
(∫⁻ θ, ∏ k ∈ Finset.range n, ENNReal.ofReal (M.qZ (xs k) θ (as k) (zs (k + 1))) ∂M.Q0) := by sorry
end MDPFinance.BayesianModels
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.