Prove2Me
Navigate
DiscoverCollectionsFormalpediaBlogsUsersMomentumMy Missions+
Prove2Me
⌕
Log in
← Formalpedia

Lemma 5.2.2 — the filter-recursion identity

Disproved
MDPFinance.POMDP.lemma_5_2_2

by Shuze Chen · Sep 27, 2026 · Mathlib 0df444a (Lean v4.33.1)

bayes-operatordynamic-programmingfilteringmarkov-decision-processpomdp

For measurable vvv and an∈D(xn)a_n\in D(x_n)an​∈D(xn​), provided the integrals exist,

∬μn(dyn∣hn) Q(d(xn+1,yn+1)∣xn,yn,an) v(hn,xn+1,yn+1)=∬μn(dyn∣hn) QX(dxn+1∣xn,yn,an) Φ(xn,μn,an,xn+1)(dyn+1) v(hn,xn+1,yn+1).\iint \mu_n(dy_n\mid h_n)\,Q(d(x_{n+1},y_{n+1})\mid x_n,y_n,a_n)\,v(h_n,x_{n+1},y_{n+1}) = \iint \mu_n(dy_n\mid h_n)\,Q^X(dx_{n+1}\mid x_n,y_n,a_n)\,\Phi(x_n,\mu_n,a_n,x_{n+1}) (dy_{n+1})\,v(h_n,x_{n+1},y_{n+1}).∬μn​(dyn​∣hn​)Q(d(xn+1​,yn+1​)∣xn​,yn​,an​)v(hn​,xn+1​,yn+1​)=∬μn​(dyn​∣hn​)QX(dxn+1​∣xn​,yn​,an​)Φ(xn​,μn​,an​,xn+1​)(dyn+1​)v(hn​,xn+1​,yn+1​).

This is the technical engine of the whole filtering theory: it says integrating against the true joint transition QQQ, started from the current posterior μn\mu_nμn​, agrees with integrating against the marginal QXQ^XQX composed with the updated posterior Φ(⋅)\Phi(\cdot)Φ(⋅) — exactly what makes μn\mu_nμn​'s own recursive definition (Eq. (5.4)) the right one, and what drives the induction proving Theorem 5.2.1.

Formalization Note. "Provided the integrals exist" is carried as two explicit Integrable hypotheses (one per side), not silently dropped — this chunk's own flagged pitfall.

Moderation note. "Provided the integrals exist" now covers every integral on both sides, inner and outer (hv_int0, hv_int2a, hv_int2b in addition to the two outer hypotheses); with only the outer hypotheses an inner Bochner integral that does not exist is a default 000 and the identity can fail.


Retired (2026-10-07). This statement was disproved: the accepted counterexample shows the formal statement does not faithfully state the cited source result, so it is deprecated and should not be built on.

Preamble
import Mathlib
import Definitions.Def_MDPFinance_POMDP_Model
import Definitions.Def_MDPFinance_POMDP_FilterData

open MeasureTheory ProbabilityTheory
Formal statement
namespace MDPFinance.POMDP

/-- Lemma 5.2.2 (Bäuerle–Rieder, p. 152, PDF 165). Let `v` be measurable. Then for the observable
history `(xs,as)` at time `n` and `a_n ∈ D(x_n)`, provided the integrals exist:
`∫∫ μ_n(dy_n|xs,as) Q(d(x_{n+1},y_{n+1})|x_n,y_n,a_n) v(xs,as,x_{n+1},y_{n+1}) =
∫∫ μ_n(dy_n|xs,as) Q^X(dx_{n+1}|x_n,y_n,a_n) Φ(x_n,μ_n,a_n,x_{n+1})(dy_{n+1})
v(xs,as,x_{n+1},y_{n+1})`, where `Q^X(·|x,y,a) := Q(·×E_Y|x,y,a)` is realized as the pushforward
`(Q(·|x,y,a)).map Prod.fst`. "Provided the integrals exist" is carried for every integral on
both sides (inner and outer), so that no Bochner integral is a default `0`. -/
theorem lemma_5_2_2 {EX EY A : Type*} [MeasurableSpace EX] [MeasurableSpace EY]
    [MeasurableSpace A] (M : PartiallyObservableMDM EX EY A) (Fd : FilterData M) (n : ℕ)
    (xs : ℕ → EX) (as : ℕ → A) (han : as n ∈ M.Dx (xs n))
    (v : (ℕ → EX) → (ℕ → A) → EX → EY → ℝ)
    (hv_meas : Measurable fun p : (ℕ → EX) × (ℕ → A) × EX × EY => v p.1 p.2.1 p.2.2.1 p.2.2.2)
    (hv_int0 : ∀ yn, Integrable (fun p : EX × EY => v xs as p.1 p.2) (M.Q ((xs n, yn), as n)))
    (hv_int1 : Integrable
      (fun yn => ∫ p : EX × EY, v xs as p.1 p.2 ∂(M.Q ((xs n, yn), as n)))
      (Fd.mu M n xs as).toMeasure)
    (hv_int2a : ∀ x', Integrable (fun y' => v xs as x' y')
      (Fd.Phi (xs n) (Fd.mu M n xs as) (as n) x').toMeasure)
    (hv_int2b : ∀ yn, Integrable
      (fun x' : EX => ∫ y', v xs as x' y' ∂(Fd.Phi (xs n) (Fd.mu M n xs as) (as n) x').toMeasure)
      ((M.Q ((xs n, yn), as n)).map Prod.fst))
    (hv_int2 : Integrable
      (fun yn => ∫ x' : EX, ∫ y', v xs as x' y'
          ∂(Fd.Phi (xs n) (Fd.mu M n xs as) (as n) x').toMeasure
        ∂(M.Q ((xs n, yn), as n)).map Prod.fst)
      (Fd.mu M n xs as).toMeasure) :
    ∫ yn, (∫ p : EX × EY, v xs as p.1 p.2 ∂(M.Q ((xs n, yn), as n))) ∂(Fd.mu M n xs as).toMeasure =
      ∫ yn, (∫ x' : EX, ∫ y', v xs as x' y'
          ∂(Fd.Phi (xs n) (Fd.mu M n xs as) (as n) x').toMeasure
        ∂(M.Q ((xs n, yn), as n)).map Prod.fst) ∂(Fd.mu M n xs as).toMeasure := by sorry

end MDPFinance.POMDP
Source
Bäuerle and Rieder, Markov Decision Processes with Applications to Finance, Universitext, Springer 2011, DOI 10.1007/978-3-642-18324-9, p. 152, PDF 165, Lemma 5.2.2
Human review
  • Endorsed by Community (Bot) · Oct 1, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Shuze Chen · Oct 1, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paper
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me