The moment implies and survives centring
ProvedMarkovChainCLT.memLp_two_and_logMoment_sub_constLet be a probability distribution on and let be measurable with the Doukhan-Massart-Rio moment
Then two things follow, for every constant :
- , i.e. ;
- the shifted function satisfies the same log-moment, .
For (1), split on and : on the first region and is finite, while on the second , so . Hence the pointwise bound , whose right-hand side is integrable. Note that the log-moment alone genuinely carries this information: vanishes on , and it is finiteness of that controls the integral there.
For (2), use , the subadditivity , and to dominate by a linear combination of the integrable functions , , and ; here pointwise, because vanishes where .
This is the bookkeeping that lets Corollary 3 feed the centred functional into Theorem 6, which is stated for centred sequences, while the corollary's hypothesis is a moment condition on itself. The conclusion is what makes the asymptotic variance meaningful.
import Mathlib.Analysis.SpecialFunctions.Log.PosLog import Mathlib.MeasureTheory.Function.L2Space open MeasureTheory ProbabilityTheory Filter open scoped ENNReal NNReal Topology ProbabilityTheory
theorem MarkovChainCLT.memLp_two_and_logMoment_sub_const {X : Type*} [MeasurableSpace X]
(π : Measure X) [IsProbabilityMeasure π] (f : X → ℝ) (hf : Measurable f)
(hmom : Integrable (fun x => f x ^ 2 * Real.posLog |f x|) π) (c : ℝ) :
MemLp f 2 π ∧ Integrable (fun x => (f x - c) ^ 2 * Real.posLog |f x - c|) π := by sorry