The portmanteau theorem (closed and open set characterisations of weak convergence)
ProvedFamousTheorems.portmanteau_theoremThe portmanteau theorem. Let be a topological space with its Borel σ-algebra in which closed sets can be approximated from outside by bounded continuous functions (for example, any pseudo-metrizable space). For probability measures along a countably generated filter (for example, a sequence), the following are equivalent:
- weakly, i.e. for every bounded continuous ;
- for every closed set ;
- for every open set .
This is the standard toolkit for checking weak convergence of probability measures, used throughout the theory of limit theorems (central limit theorem, Donsker's invariance principle).
Formalization note. Assembled from Mathlib's MeasureTheory.ProbabilityMeasure.limsup_measure_closed_le_of_tendsto, MeasureTheory.tendsto_of_forall_isClosed_limsup_le', MeasureTheory.ProbabilityMeasure.le_liminf_measure_open_of_tendsto and MeasureTheory.tendsto_of_forall_isOpen_le_liminf'. Convergence in MeasureTheory.ProbabilityMeasure Ω is weak convergence (the topology is defined by integrals of bounded continuous functions), and measures are evaluated as ℝ≥0∞-valued. HasOuterApproxClosed Ω is the approximation hypothesis on the space.
import Mathlib
namespace FamousTheorems
theorem portmanteau_theorem {Ω ι : Type*} [MeasurableSpace Ω] [TopologicalSpace Ω] [OpensMeasurableSpace Ω] [HasOuterApproxClosed Ω]
{L : Filter ι} [L.IsCountablyGenerated] (μ : MeasureTheory.ProbabilityMeasure Ω)
(μs : ι → MeasureTheory.ProbabilityMeasure Ω) :
(Filter.Tendsto μs L (nhds μ) ↔ ∀ F : Set Ω, IsClosed F →
Filter.limsup (fun i => (μs i : MeasureTheory.Measure Ω) F) L ≤ (μ : MeasureTheory.Measure Ω) F) ∧
(Filter.Tendsto μs L (nhds μ) ↔ ∀ G : Set Ω, IsOpen G →
(μ : MeasureTheory.Measure Ω) G ≤ Filter.liminf (fun i => (μs i : MeasureTheory.Measure Ω) G) L) := by sorry
end FamousTheorems