Every coordinate of the stationary chain has law
ProvedMarkovChainCLT.map_coord_chainMeasuremarkov-chainsprobability
Let be a Markov kernel with invariant probability distribution , and run the chain from . Then every coordinate has the same marginal law:
Equivalently, the pushforward of the trajectory law under the -th coordinate projection is . This is immediate from invariance by induction: the law of is the law of pushed through , and .
It is the fact used to identify with — in particular to check that the centred functional has mean zero, which is a standing hypothesis of the mixing central limit theorems of the mission — and to transfer moment conditions such as from to the path measure.
Preamble
import Definitions.Def_MarkovErgodicity import Definitions.Def_MarkovChainPathMeasure import Definitions.Def_MixingCoefficients open MeasureTheory ProbabilityTheory Filter open scoped ENNReal NNReal Topology ProbabilityTheory
Formal statement
theorem MarkovChainCLT.map_coord_chainMeasure {X : Type*} [MeasurableSpace X]
(P : Kernel X X) [IsMarkovKernel P] (π : Measure X) [IsProbabilityMeasure π]
(hinv : Kernel.Invariant P π) (n : ℕ) :
Measure.map (fun ω : ℕ → X => ω n) (chainMeasure P π) = π := by sorrySource
G. L. Jones, "On the Markov Chain Central Limit Theorem", Probability Surveys 1 (2004) 299-320, arXiv math/0409112v2, Section 1 (eq. (1)) and Section 4: E_pi f is the mean of f(X_n) under the stationary chain.