§2.1 — the initial value of the recursion (2) equals the primal value (1)
ProvedInfoRelax.IdealPenalty.value_zero_eq_primalConsider a dynamic program in recursive form: a natural filtration with , a countable action set , nonempty feasible-action sets , and bounded period rewards that are -measurable. Let be the value functions of the Bellman recursion (2), with . Then, almost surely,
the optimal value of the primal DP (1) over all nonanticipative policies.
This identity is what makes the recursion (2) a solution method for (1); it holds whether or not the suprema in (2) are attained.
Formalization Note does not depend on ; the statement quantifies over all action sequences. The comparison is made in the extended reals, where the primal value is defined. The countable action set is a pinned regularity assumption that makes the supremum in (2) measurable.
import Mathlib import Definitions.Def_InfoRelax_IdealPenalty_Framework import Definitions.Def_InfoRelax_IdealPenalty_RecursiveModel import Definitions.Def_InfoRelax_IdealPenalty_Penalty open MeasureTheory
namespace InfoRelax.IdealPenalty
/-- **§2.1, the claim after eq. (2)**, Brown, Smith & Sun (2010), Oper. Res., DOI
10.1287/opre.1090.0796, p. 3: "The final value `V_0` is equal to the optimal value of (1)."
In the recursive setting (`RecursiveDP.Valid`: `𝓕_0 = {∅, Ω}`, nonempty feasible sets depending on
past actions, `𝓕_t`-measurable bounded period rewards depending on `a_0, …, a_t`), the value
function `V_0` of the recursion (2) equals, almost surely and for every `a`, the primal value
`sup_{α ∈ 𝒜_𝔽} 𝔼[r(α)]` of (1).
**Formalization Note.** `V_0(a)` does not depend on `a` (it is a function of no actions); the
statement quantifies over all `a`. The claim holds whether or not the suprema in (2) are
attained. Pinned: countable action type with the discrete σ-algebra. The comparison is in
`EReal`, where the primal value lives. -/
theorem value_zero_eq_primal {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) :
∀ a : Fin (T + 1) → X, ∀ᵐ ω ∂μ,
((valueFn M μ 0 a ω : ℝ) : EReal) = primalValue μ M.A M.𝔽 M.r := by sorry
end InfoRelax.IdealPenalty
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.