Strict stationarity is preserved by a measurable functional
ProvedMarkovChainCLT.isStrictlyStationary_comp_of_measurableLet be a sequence of measurable random elements of which is strictly stationary, and let be measurable. Then the functional process is strictly stationary.
Strict stationarity says the law of the shifted path on the sequence space does not depend on . The coordinatewise map is measurable, and the path of the functional process is applied to the path of . Functoriality of the pushforward, , then turns the equality of the shifted path laws for into the corresponding equality for .
Why measurability of is needed. The pushforward of a non-measurable map is the zero measure by convention, so for a non-measurable the stationarity hypothesis degenerates to and carries no information, while the conclusion can still be a substantive claim — may well be measurable even when is not (take , with non-measurable and measurable, and ). One can then choose non-stationary and the statement fails. Assuming each measurable — which holds in every application — makes the pushforward identities available and the result true.
This is the bookkeeping step that lets the functional process of a stationary Markov chain be fed into the central limit theorems of the mission, which are stated for general strictly stationary sequences.
import Definitions.Def_MixingCoefficients open MeasureTheory ProbabilityTheory Filter open scoped ENNReal NNReal Topology ProbabilityTheory
theorem MarkovChainCLT.isStrictlyStationary_comp_of_measurable {Ω X E : Type*}
[MeasurableSpace Ω] [MeasurableSpace X] [MeasurableSpace E]
(P : Measure Ω) (Y : ℕ → Ω → X) (hY : ∀ n, Measurable (Y n))
(hstat : IsStrictlyStationary P Y) (g : X → E) (hg : Measurable g) :
IsStrictlyStationary P (fun i ω => g (Y i ω)) := by sorry