Relation (19) — F̂(x, s, M) = φᵢ(xᵢ, sᵢ, M) Pᵢ(x, s, M) + ∫_M^∞ φᵢ(xᵢ, sᵢ, m) d_m Pᵢ(x, s, m)
ProvedWhittleGittins.Retirement.relation19_partial_integrationIn Whittle's bandit process with per-project horizons , let
and let as in (17). Then, for every project and every , partial integration of (13) gives
where the last integral is the Lebesgue–Stieltjes integral over against the distribution function .
Relation (19) expresses as a mixture of single-project values, which is how the one-project dynamic programming equation (14) transfers to .
Formalization Note The Stieltjes measure is supplied as the measure of any StieltjesFunction (a monotone right-continuous function) that agrees with everywhere; such a exists because right derivatives of convex functions are non-decreasing and right-continuous. The integral is taken over the open half-line , so a jump of at is accounted for by the first term. is the right derivative.
import Mathlib import Definitions.Def_WhittleGittins_Retirement_BanditProcess
namespace WhittleGittins.Retirement
open MeasureTheory ProbabilityTheory
/-- Relation (19) (p. 147), partial integration of (13):
`F̂(x, s, M) = φᵢ(xᵢ, sᵢ, M) Pᵢ(x, s, M) + ∫_M^∞ φᵢ(xᵢ, sᵢ, m) d_m Pᵢ(x, s, m)`, the Stieltjes
integral being taken against any Stieltjes function `G` that agrees with `Pᵢ(x, s, ·)`, over
`m > M`. -/
theorem relation19_partial_integration {N : ℕ} {X : Fin N → Type*} [∀ i, MeasurableSpace (X i)]
(B : BanditProcess N X) (x : ∀ j, X j) (s : Fin N → ℕ) (i : Fin N) (M : ℝ)
(G : StieltjesFunction ℝ) (hG : ∀ m, G m = Pdist B x s i m) :
Fhat B x s M = phi B i (x i) (s i) M * Pdist B x s i M +
∫ m in Set.Ioi M, phi B i (x i) (s i) m ∂G.measure := by sorry
end WhittleGittins.Retirement
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.