Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

§2.3, eq. (10) — the dual recursion solves the dual problem and bounds the primal DP

Proved
InfoRelax.IdealPenalty.eq_10_dual_recursion

by mikedeng1 · Oct 8, 2026 · Mathlib 0df444a (Lean v4.33.1)

bellman-equationdualitydynamic-programminginformation-relaxationp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1

Consider a recursive DP (natural filtration with F0={∅,Ω}\mathcal F_0=\{\emptyset,\Omega\}F0​={∅,Ω}, countable action set, nonempty feasible sets At(a0,…,at−1)A_t(a_0,\dots,a_{t-1})At​(a0​,…,at−1​), bounded Ft\mathcal F_tFt​-measurable period rewards rt(a0,…,at)r_t(a_0,\dots,a_t)rt​(a0​,…,at​)). Let G\mathbb GG be a relaxation of F\mathbb FF, let (wt)(w_t)(wt​) be generating functions as in Proposition 2.2 and z=∑tztz=\sum_t z_tz=∑t​zt​ the penalty they generate. Let VtGV^{\mathbb G}_tVtG​ be the dual value functions of (10): VT+1G=0V^{\mathbb G}_{T+1}=0VT+1G​=0 and

VtG(a0,…,at−1)=sup⁡at∈At(a0,…,at−1){rt(a0,…,at)−zt(a0,…,at)+E[Vt+1G(a0,…,at)∣Gt]}.V^{\mathbb G}_t(a_0,\dots,a_{t-1})=\sup_{a_t\in A_t(a_0,\dots,a_{t-1})}\big\{r_t(a_0,\dots,a_t)-z_t(a_0,\dots,a_t)+\mathbb E[V^{\mathbb G}_{t+1}(a_0,\dots,a_t)\mid\mathcal G_t]\big\}.VtG​(a0​,…,at−1​)=at​∈At​(a0​,…,at−1​)sup​{rt​(a0​,…,at​)−zt​(a0​,…,at​)+E[Vt+1G​(a0​,…,at​)∣Gt​]}.

Then the recursion solves the dual problem for (G,z)(\mathbb G,z)(G,z) and its initial value bounds the primal DP:

E[V0G]=sup⁡αG∈AGE[r(αG)−z(αG)],sup⁡αF∈AFE[r(αF)]≤E[V0G].\mathbb E[V^{\mathbb G}_0]=\sup_{\alpha_G\in\mathcal A_{\mathbb G}}\mathbb E[r(\alpha_G)-z(\alpha_G)],\qquad \sup_{\alpha_F\in\mathcal A_{\mathbb F}}\mathbb E[r(\alpha_F)]\le\mathbb E[V^{\mathbb G}_0].E[V0G​]=αG​∈AG​sup​E[r(αG​)−z(αG​)],αF​∈AF​sup​E[r(αF​)]≤E[V0G​].

Formalization Note V0GV^{\mathbb G}_0V0G​ is G0\mathcal G_0G0​-measurable but in general not constant; the page's "initial value" is read as its expectation. V0G(a)V^{\mathbb G}_0(a)V0G​(a) does not depend on aaa, and the statement quantifies over all aaa. Values are extended reals.

Preamble
import Mathlib
import Definitions.Def_InfoRelax_IdealPenalty_Framework
import Definitions.Def_InfoRelax_IdealPenalty_RecursiveModel
import Definitions.Def_InfoRelax_IdealPenalty_Penalty

open MeasureTheory
Formal statement
namespace InfoRelax.IdealPenalty

/-- **§2.3, eq. (10)**, Brown, Smith & Sun (2010), Oper. Res., DOI 10.1287/opre.1090.0796, p. 5.

In the recursive setting, let `𝔾` be a relaxation of `𝔽`, `w` generating functions as in
Proposition 2.2 and `z = ∑_t z_t` the penalty it constructs. The dual value functions `V^𝔾_t` of
(10) (the recursion (2) with rewards `r_t − z_t` and filtration `𝔾`, `V^𝔾_{T+1} = 0`) solve the
dual problem, `𝔼[V^𝔾_0] = sup_{α_G ∈ 𝒜_𝔾} 𝔼[r(α_G) − z(α_G)]`, and the initial value provides an
upper bound on the primal DP (1): `sup_{α_F ∈ 𝒜_𝔽} 𝔼[r(α_F)] ≤ 𝔼[V^𝔾_0]`.

**Formalization Note.** `V^𝔾_0` is `𝒢_0`-measurable but not constant in general; the page's
"initial value" is read as its expectation. `V^𝔾_0(a)` does not depend on `a`; the statement
quantifies over all `a`. Pinned: countable action type with the discrete σ-algebra, generating
functions as in `GeneratingFns`. -/
theorem eq_10_dual_recursion {Ω X : Type*} [mΩ : MeasurableSpace Ω] [MeasurableSpace X]
    [DiscreteMeasurableSpace X] [Countable X] [DecidableEq X] {T : ℕ}
    (μ : Measure Ω) [IsProbabilityMeasure μ] (M : RecursiveDP Ω X T) (hM : M.Valid)
    (𝔾 : Filtration (Fin (T + 1)) mΩ) (h𝔾 : IsRelaxation M.𝔽 𝔾)
    (w : Fin (T + 1) → (Fin (T + 1) → X) → Ω → ℝ) (hw : GeneratingFns μ w) :
    ∀ a : Fin (T + 1) → X,
      ((∫ ω, dualValueFn M μ 𝔾 w 0 a ω ∂μ : ℝ) : EReal) =
          dualBound μ M.A 𝔾 M.r (penalty μ M.𝔽 𝔾 w) ∧
        primalValue μ M.A M.𝔽 M.r ≤ ((∫ ω, dualValueFn M μ 𝔾 w 0 a ω ∂μ : ℝ) : EReal) := by sorry

end InfoRelax.IdealPenalty
Source
Brown, Smith & Sun, Information Relaxations and Duality in Stochastic Dynamic Programs, Oper. Res. (Articles in Advance, 2010), DOI 10.1287/opre.1090.0796, p. 5, §2.3, eq. (10) and the sentence after it
Human review
  • Endorsed by Shuze Chen · Oct 8, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 8, 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 paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me