The disintegration theorem (existence of conditional kernels)
ProvedFamousTheorems.disintegration_theorem_6bThe disintegration theorem (existence of conditional kernels). Let be a finite measure on a product , where is a nonempty standard Borel space. Then there is a Markov kernel from to such that
for all measurable . Here is the first marginal of . Equivalently, .
The kernel is the conditional distribution of the second coordinate given that the first equals . The theorem gives regular conditional probabilities, which are needed to condition on events of probability zero. It is used throughout probability theory, for example in the construction of Markov processes and in Bayesian statistics.
Formalization note. Mathlib's instance MeasureTheory.Measure.condKernel.instIsCondKernel, which states that ρ.condKernel disintegrates , together with the instance that ρ.condKernel is a Markov kernel. ρ.fst is the first marginal and Measure.compProd is the composition–product .
import Mathlib
namespace FamousTheorems
theorem disintegration_theorem_6b {α Ω : Type*} {mα : MeasurableSpace α} {mΩ : MeasurableSpace Ω} [StandardBorelSpace Ω] [Nonempty Ω]
(ρ : MeasureTheory.Measure (α × Ω)) [MeasureTheory.IsFiniteMeasure ρ] :
∃ κ : ProbabilityTheory.Kernel α Ω, ProbabilityTheory.IsMarkovKernel κ ∧ ρ.fst.compProd κ = ρ := by sorry
end FamousTheorems