Restricting a composition-product to a cylinder over the first coordinate
ProvedProbabilityTheory.compProd_restrict_prod_univRestricting a joint law to an event that depends only on the first coordinate is the same as restricting the first marginal.
Let be an s-finite measure on and an s-finite kernel from to , and let be measurable. Then
The event constrains only the first coordinate, so cutting the joint law down to it and cutting the first marginal down to before composing produce the same measure — the kernel is untouched either way.
This is the form in which one localises a two-stage experiment to an event decided at the first stage: for instance, restricting attention to the histories on which a stopping time has not yet fired before analysing the next round.
Formalization Note Mathlib has the analogous statement for Kernel.compProd; this is the version for a measure composed with a kernel. Only s-finiteness is needed, so it applies to the sub-probability measures produced by restriction.
import Mathlib.Probability.Kernel.Composition.MeasureCompProd open MeasureTheory ProbabilityTheory Set open scoped ENNReal
theorem ProbabilityTheory.compProd_restrict_prod_univ {α β : Type*}
{mα : MeasurableSpace α} {mβ : MeasurableSpace β}
(μ : Measure α) [SFinite μ] (κ : Kernel α β) [IsSFiniteKernel κ]
{s : Set α} (hs : MeasurableSet s) :
(μ ⊗ₘ κ).restrict (s ×ˢ (Set.univ : Set β)) = (μ.restrict s) ⊗ₘ κ := by
sorry