BanditRLlib
Lean gate passed before this site build; local proof declarations are shown as compiled.

Lean module · ETC

BanditRLProof.Algorithms.ETCCenteredDiffCanonicalTail

# ETC centered reward-difference canonical sub-Gaussian tail This module fixes the canonical exponential tail budget produced by the centered reward-difference independent sub-Gaussian route. It removes the need for downstream users to provide a separate tail-domination hypothesis when they choose the exact Mathlib sub-Gaussian bound as their tail function.

Module map

Teaching chapter
3. Explore-Then-Commit
Declarations
3
Placeholders
0

Imports

BanditRLProof.Algorithms.ETCCenteredDiffSubGaussianWitnesses

Imported by

BanditRLProof, BanditRLProof.Algorithms.ETCWrongCommitCanonicalTail

Declarations

Open an item to read its exact compact statement and source link. Detailed teaching notes are linked when registered.

def BanditRLProof.ETC.centeredDiffSubGaussianTail Compiled

Canonical pairwise tail budget for the centered reward-difference independent sub-Gaussian ETC route. The expression is exactly the `ENNReal.ofReal (Real.exp ...)` right-hand side used by the centered-diff producer over the ETC exploration horizon.

noncomputable def centeredDiffSubGaussianTail {K : Nat} (spec : ETC.Spec K) (model : FiniteBanditModel K) (c : Fin K -> Nat -> NNReal) (a : Fin K) : ENNReal
def BanditRLProof.ETC.centeredDiffSubGaussianWitnesses_of_indep_subG Compiled

Build the centered-diff witness package when the caller chooses the canonical sub-Gaussian exponential tail budget. This is the `ETC-CENTERED-DIFF-SUBGAUSSIAN-CANONICAL-TAIL` leaf. It still requires the actual reward-law independence and sub-Gaussian witnesses, but it discharges the tail-domination field definitionally.

noncomputable def centeredDiffSubGaussianWitnesses_of_indep_subG {Omega : Type u} {K : Nat} [MeasurableSpace Omega] (mu : MeasureTheory.Measure Omega) (spec : ETC.Spec K) (model : FiniteBanditModel K) (commitArm : Fin K) (reward : Omega -> RewardTrace Rat) (c : Fin K -> Nat -> NNReal) (h_indep : forall a : Fin K, (a = model.bestArm -> False) -> ProbabilityTheory.iIndepFun (fun t omega => ETC.centeredPairwiseRewardDiff spec model commitArm reward a t omega) mu) (h_subG : forall a : Fin K, (a = model.bestArm -> False) -> forall t, t ∈ Finset.range (spec.explorationPulls * K) -> ProbabilityTheory.HasSubgaussianMGF (fun omega => ETC.centeredPairwiseRewardDiff spec model commitArm reward a t omega) (c a t) mu) : ETC.CenteredDiffSubGaussianWitnesses mu spec model commitArm reward (ETC.centeredDiffSubGaussianTail spec model c)
theorem BanditRLProof.ETC.pairwiseEmpMeanTailContract_of_centeredDiff_indep_subG Compiled

Produce the fixed-commit ETC pairwise empirical-mean tail contract directly from concrete centered-diff independence and sub-Gaussian witnesses, using the canonical exponential tail budget. This theorem is the no-extra-tail-domination consumer for the independent sub-Gaussian centered-diff route.

theorem pairwiseEmpMeanTailContract_of_centeredDiff_indep_subG {Omega : Type u} {K : Nat} [MeasurableSpace Omega] (mu : MeasureTheory.Measure Omega) [MeasureTheory.IsFiniteMeasure mu] (spec : ETC.Spec K) (model : FiniteBanditModel K) (commitArm : Fin K) (reward : Omega -> RewardTrace Rat) (c : Fin K -> Nat -> NNReal) (hexplorationPulls_pos : 0 < spec.explorationPulls) (h_indep : forall a : Fin K, (a = model.bestArm -> False) -> ProbabilityTheory.iIndepFun (fun t omega => ETC.centeredPairwiseRewardDiff spec model commitArm reward a t omega) mu) (h_subG : forall a : Fin K, (a = model.bestArm -> False) -> forall t, t ∈ Finset.range (spec.explorationPulls * K) -> ProbabilityTheory.HasSubgaussianMGF (fun omega => ETC.centeredPairwiseRewardDiff spec model commitArm reward a t omega) (c a t) mu) : ETC.PairwiseEmpMeanTailContract mu spec model commitArm reward (ETC.centeredDiffSubGaussianTail spec model c)