A finite measure on a path space is determined by its finite-dimensional marginals
ProvedPathSpace.ext_of_map_frestrictLeLet be a family of measurable spaces and let be finite measures on the path space . If for every the two measures have the same image under the restriction map
then .
Why it is needed. The product -algebra on a path space is generated by cylinder sets, so a measure on is determined by its finite-dimensional marginals. This is used constantly and informally — "two processes with the same finite-dimensional distributions have the same law" — but in a formalization it is not a one-liner: the marginals live in different types as varies, and the argument has to route through the projective-limit machinery.
Concretely, any statement about a process built by the Ionescu–Tulcea theorem (Kernel.traj) that is not already in the library's API has to be proved by comparing finite-dimensional marginals, and this lemma is what licenses that reduction. Typical uses: identifying the law of a shifted trajectory, proving stationarity of a chain started from an invariant measure, or showing that two different constructions of the same process agree.
Why finiteness matters. Agreement of all finite-dimensional marginals does not determine a general (possibly non--finite) measure on a path space, so some finiteness hypothesis is genuinely needed; IsFiniteMeasure on both sides is the form in which it is always available in probability, where the measures are probability measures.
Restriction to intervals suffices. Only the initial segments are tested, not all finite subsets. This is a real simplification for the user — checking one marginal per natural number rather than one per finite subset — and is legitimate because every finite subset of is contained in some initial segment, so the initial-segment marginals already determine the whole projective family.
Proof. The family is projective: restricting the -th marginal to coordinates gives the -th, because the restriction maps compose. Hence the induced family indexed by finite subsets is a projective measure family, and is its projective limit — by the criterion that it suffices to check the defining property on intervals . By hypothesis is the projective limit of the same family. Uniqueness of projective limits of finite measures gives .
import Mathlib.Probability.Kernel.IonescuTulcea.Traj open Filter Finset Function MeasurableEquiv MeasurableSpace MeasureTheory Preorder ProbabilityTheory open scoped ENNReal NNReal Topology
theorem PathSpace.ext_of_map_frestrictLe {X : ℕ → Type*} [∀ n, MeasurableSpace (X n)]
{μ ν : Measure (Π n, X n)} [IsFiniteMeasure μ] [IsFiniteMeasure ν]
(h : ∀ n, μ.map (frestrictLe n) = ν.map (frestrictLe n)) : μ = ν := by sorry