Stationary processes #
A process X : ι → Ω → α is stationary when shifting time leaves its law on ι → α unchanged,
which says that every shift of ι → α preserves that law.
The finite-dimensional distributions determine the law, so stationarity can be
checked on finite sets of times, and over ℕ on
blocks of consecutive values, the form in which information
theory defines it. Over ℕ a shift by one is enough, and a
process that advances by a measure-preserving map of the
sample space is stationary.
Implementation notes #
[UPSTREAM] candidate: mathlib has no notion of a stationary process.
References #
A process is (strictly) stationary when shifting time does not change its law.
Equations
- ProbabilityTheory.IsStationary X μ = ∀ (s : ι), ProbabilityTheory.IdentDistrib (fun (ω : Ω) (t : ι) => X (t + s) ω) (fun (ω : Ω) (t : ι) => X t ω) μ μ
Instances For
block X a n is the block of n consecutive values of X starting at time a.
Equations
- ProbabilityTheory.block X a n ω i = X (↑i + a) ω
Instances For
A process is stationary exactly when a block of n consecutive values has the same law wherever
it starts, which is how information theory states the definition.