(2.4), p.5 — the stopped reward and
ProvedRiskStopTC.eq_2_4dynamic-programmingoptimal-stoppingp2o-gran-per-chapterp2o-plan-paperp2o-strengthenedp2o-v1risk-measuresstochastic-programming
Throughout, is a probability space with a filtration , , , and for a fixed . Relations between random variables hold -almost surely.
Let for and let be a stopping time with values in . Then for every and with ,
and the stopped reward belongs to .
This is the representation (2.4) through which the risk functionals act on stopped rewards.
Differences from the source:
- encoding: Z_τ (the value Z_{τ(ω)}(ω)) = Σ_t 1{τ=t} Z_t → Z_τ is defined as the sum (stoppedRV); the identity is stated pointwise: the sum equals Z_t(ω) when τ(ω) = t ≤ T; reason: Z_{τ(ω)}(ω) is not a Lean term of the right shape; the pointwise statement is the same identity
- generalization: (2.4) for (Z_0, …, Z_T) ∈ Z_{0,T} → also concludes Z_τ ∈ Z_T (IsLp (ℱ T) P p); reason: the paper presupposes Z_τ ∈ Z_T when it applies ρ_{0,T} to Z_τ (p.17); dropping the second conjunct gives (2.4)
- encoding: Z_t = L^p(Ω, F_t, P) as classes of a.e.-equal functions; ⪯ and = almost surely → functions Ω → ℝ with an F_t-measurable representative in L^p (IsLp (ℱ t) P p); relations ≤ᵐ[P] and =ᵐ[P]; reason: every class has such a representative; Lean convention; same instances
- encoding: stopping times τ : Ω → {0, …, T} with {τ = t} ∈ F_t; 𝔗_m = {τ ∈ 𝔗 : τ ≥ m} → IsBddStoppingTime ℱ T τ (Mathlib IsStoppingTime, values in WithTop ℕ, τ ≤ T everywhere) and m ≤ τ everywhere (IsStopTimeFrom); reason: in discrete time {τ ≤ t} ∈ F_t for all t is equivalent to {τ = t} ∈ F_t for all t; same instances
Preamble
import Mathlib import Definitions.Def_RiskFunctionals_NestedAVaR_Compositions import Definitions.Def_MultiperiodRisk_Bellman_TestSet import Definitions.Def_RiskStopTC_stoppedRV open MeasureTheory RiskFunctionals.NestedAVaR
Formal statement
namespace RiskStopTC
/-- **The stopped reward (2.4).** Pichler, Liu & Shapiro, arXiv:1808.10807v3, (2.4), p.5: "Moreover
1{τ=t} Zτ = 1{τ=t} Zt and thus for (Z0, . . . , ZT) ∈ Z0,T it follows that
Zτ = ∑_{t=0}^T 1{τ=t} Zτ = ∑_{t=0}^T 1{τ=t} Zt."
For `(Z_0, …, Z_T) ∈ Z_{0,T}` and a stopping time `τ` with values in `{0, …, T}`:
`∑_{t=0}^T 1{τ=t}(ω) Z_t(ω) = Z_t(ω)` whenever `τ(ω) = t ≤ T`, and `Z_τ ∈ Z_T`.
**Formalization Note.** `Z_t ∈ Z_t = L^p(Ω, F_t, P)` is `IsLp (ℱ t) P p (Z t)`; stopping times are
`IsBddStoppingTime ℱ T τ` (values in `WithTop ℕ`, `τ ≤ T` everywhere). -/
theorem eq_2_4 {Ω : Type*} {mΩ : MeasurableSpace Ω} (ℱ : Filtration ℕ mΩ) (P : Measure[mΩ] Ω)
[IsProbabilityMeasure P] (p : ENNReal) (hp : 1 ≤ p) (T : ℕ) (hℱ : IsStandardFiltration ℱ T)
(Z : ℕ → Ω → ℝ) (hZ : ∀ t ≤ T, IsLp (ℱ t) P p (Z t))
(τ : Ω → WithTop ℕ) (hτ : MultiperiodRisk.Bellman.IsBddStoppingTime ℱ T τ) :
(∀ ω, ∀ t ≤ T, τ ω = (t : WithTop ℕ) → stoppedRV Z τ T ω = Z t ω) ∧
IsLp (ℱ T) P p (stoppedRV Z τ T) := by sorry
end RiskStopTC
Source
Pichler, Liu & Shapiro, Risk averse stochastic programming: time consistency and optimal stopping, arXiv:1808.10807v3, equation (2.4), p.5
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.