Proposition 2.2(ii) — the penalty components are -adapted and nonanticipative in the actions
ProvedInfoRelax.IdealPenalty.proposition_2_2_iiconditional-expectationdynamic-programminginformation-relaxationp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be a relaxation of and let be generating functions as in Proposition 2.2(i): each depends only on the first actions, is almost everywhere strongly measurable, and the family is bounded almost everywhere by one constant. Let . Then
- for every action sequence , the sequence is adapted to ;
- each depends only on the first actions of .
Consequently the penalized objective decomposes into period- terms that depend only on what is known at time under and the actions chosen up to , which is what allows the dual problem to be solved by the recursion (10).
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
/-- **Proposition 2.2(ii) (Constructing Good Penalties)**, Brown, Smith & Sun (2010), Oper. Res.,
DOI 10.1287/opre.1090.0796, p. 5.
Let `𝔾` be a relaxation of `𝔽` and `(w_0, …, w_T)` generating functions, each `w_t` depending only
on `a_0, …, a_t`, and `z_t(a) = 𝔼[w_t(a) | 𝒢_t] − 𝔼[w_t(a) | 𝓕_t]`. Then `(z_0(a), …, z_T(a))` is
adapted to `𝔾` and `z_t` depends only on the first `t + 1` actions `(a_0, …, a_t)` of `a`.
**Formalization Note.** Same hypotheses as Proposition 2.2(i) (`GeneratingFns`); the countable
action type is not needed here and is not assumed. "Adapted" is Mathlib's `Adapted 𝔾` (each
`z_t(a)` is `𝒢_t`-measurable). The statement is close to definitional; it is the paper's
numbered result. -/
theorem proposition_2_2_ii {Ω X : Type*} [mΩ : MeasurableSpace Ω] {T : ℕ}
(μ : Measure Ω) [IsProbabilityMeasure μ] (𝔽 𝔾 : Filtration (Fin (T + 1)) mΩ)
(h𝔾 : IsRelaxation 𝔽 𝔾) (w : Fin (T + 1) → (Fin (T + 1) → X) → Ω → ℝ)
(hw : GeneratingFns μ w) :
(∀ a : Fin (T + 1) → X, Adapted 𝔾 (fun t => penaltyComp μ 𝔽 𝔾 w t a)) ∧
∀ (t : Fin (T + 1)) (a a' : Fin (T + 1) → X), (∀ s, s ≤ t → a s = a' s) →
penaltyComp μ 𝔽 𝔾 w t a = penaltyComp μ 𝔽 𝔾 w t a' := 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, Proposition 2.2(ii)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.