The Ionescu-Tulcea theorem
ProvedFamousTheorems.ionescu_tulcea_theoremThe Ionescu-Tulcea theorem. Let be measurable spaces and let be Markov kernels from to . Then for each starting time there is a kernel from to the infinite product whose projection to the first coordinates is, for every , the finite composition of the kernels .
This theorem, due to Ionescu Tulcea (1949), constructs the law of a discrete-time stochastic process from its transition kernels. Unlike the Kolmogorov extension theorem it needs no topological assumptions on the state spaces. It is the foundation for Markov chains and Markov decision processes on general state spaces.
Formalization note. Mathlib's ProbabilityTheory.Kernel.traj_map_frestrictLe, with witness Kernel.traj κ a. Preorder.frestrictLe b restricts a sequence to coordinates Finset.Iic b. Kernel.partialTraj κ a b is the kernel obtained by composing , which keeps the coordinates up to and samples the ones up to (or, for , simply forgets the later coordinates).
import Mathlib
namespace FamousTheorems
theorem ionescu_tulcea_theorem {X : ℕ → Type*} [∀ n, MeasurableSpace (X n)]
(κ : (n : ℕ) → ProbabilityTheory.Kernel ((i : Finset.Iic n) → X i) (X (n + 1)))
[∀ n, ProbabilityTheory.IsMarkovKernel (κ n)] (a : ℕ) :
∃ η : ProbabilityTheory.Kernel ((i : Finset.Iic a) → X i) ((n : ℕ) → X n),
∀ b : ℕ, η.map (Preorder.frestrictLe b) = ProbabilityTheory.Kernel.partialTraj κ a b := by sorry
end FamousTheorems