(11), proof of Theorem 2.1, pp. 5–6 — an adapted feasible x̃ with ‖E[x̄_t|F̃_t] − x̃_t‖ ≤ L_t(…)
ProvedMultistageStab.Value.display_11Work in the setting of program (1) with a randomness pattern and its exponents . Let be an admissible input and assume complete fixed recourse (A1). Then there are constants with the following property. For every admissible input , with generated filtration , and every feasible policy , there is a feasible policy (so is -measurable, in , , and ) with and, almost surely for every ,
If only the costs are random, the first sum is absent.
The constants do not depend on or . This estimate transfers a policy that is adapted to the filtration of to one adapted to the filtration of , and is the core of the stability proof.
Formalization Note The paper constructs for in an -level set; the statement here takes any feasible , which the construction allows and which is stronger. The case "only costs random" is the pattern costs, in which the first sum is replaced by .
import Mathlib import Definitions.Def_MultistageStab_Value_Setting open MeasureTheory ProbabilityTheory open scoped ENNReal
namespace MultistageStab.Value
theorem display_11 (D : Data) (pat : Pattern) (hpat : pat.Holds D)
{Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
(ξ : ℕ → Ω → E D.d) (hA1 : CompleteRecourse D) (hA3 : Admissible P D pat ξ) :
∃ L : ℕ → ℝ, (∀ t, 0 < L t) ∧
∀ ξ' : ℕ → Ω → E D.d, Admissible P D pat ξ' →
∀ xbar ∈ feasible P D pat ξ, ∃ xtil ∈ feasible P D pat ξ', xtil 1 = xbar 1 ∧
∀ t ∈ Finset.Icc 2 D.T, ∀ᵐ ω ∂P,
‖P[xbar t | sigmaGen ξ' t] ω - xtil t ω‖ ≤
L t * ((if pat = Pattern.costs then 0 else
∑ τ ∈ Finset.Icc 2 t,
P[fun ω => ‖ξ τ ω - ξ' τ ω‖ | sigmaGen ξ' τ] ω) +
∑ τ ∈ Finset.Icc 2 (t - 1),
P[fun ω => ‖xbar τ ω - P[xbar τ | sigmaGen ξ' τ] ω‖ | sigmaGen ξ' (τ + 1)] ω) := by sorry
end MultistageStab.Value
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.