Documentation

Mathlib.Probability.Kernel.Disintegration.MeasurableStieltjes

Measurable parametric Stieltjes functions #

We provide tools to build a measurable function α → StieltjesFunction with limits 0 at -∞ and 1 at +∞ for all a : α from a measurable function f : α → ℚ → ℝ. These measurable parametric Stieltjes functions are cumulative distribution functions (CDF) of transition kernels. The reason for going through instead of defining directly a Stieltjes function is that since is countable, building a measurable function is easier and we can obtain properties of the form ∀ᵐ (a : α) ∂μ, ∀ (q : ℚ), ... (for some measure μ on α) by proving the weaker ∀ (q : ℚ), ∀ᵐ (a : α) ∂μ, ....

This construction will be possible if f a : ℚ → ℝ satisfies a package of properties for all a: monotonicity, limits at +-∞ and a continuity property. We define IsRatStieltjesPoint f a to state that this is the case at a and define the property IsMeasurableRatCDF f that f is measurable and IsRatStieltjesPoint f a for all a. The function α → StieltjesFunction obtained by extending f by continuity from the right is then called IsMeasurableRatCDF.stieltjesFunction.

In applications, we will often only have IsRatStieltjesPoint f a almost surely with respect to some measure. In order to turn that almost everywhere property into an everywhere property we define toRatCDF (f : α → ℚ → ℝ) := fun a q ↦ if IsRatStieltjesPoint f a then f a q else defaultRatCDF q, which satisfies the property IsMeasurableRatCDF (toRatCDF f).

Finally, we define stieltjesOfMeasurableRat, composition of toRatCDF and IsMeasurableRatCDF.stieltjesFunction.

Main definitions #

theorem StieltjesFunction.measurable_measure {α : Type u_1} {x✝ : MeasurableSpace α} {f : αStieltjesFunction} (hf : ∀ (q : ), Measurable fun (a : α) => (f a) q) (hf_bot : ∀ (a : α), Filter.Tendsto (↑(f a)) Filter.atBot (nhds 0)) (hf_top : ∀ (a : α), Filter.Tendsto (↑(f a)) Filter.atTop (nhds 1)) :
Measurable fun (a : α) => (f a).measure

A measurable function α → StieltjesFunction with limits 0 at -∞ and 1 at +∞ gives a measurable function α → Measure ℝ by taking StieltjesFunction.measure at each point.

structure ProbabilityTheory.IsRatStieltjesPoint {α : Type u_1} (f : α) (a : α) :

a : α is a Stieltjes point for f : α → ℚ → ℝ if f a is monotone with limit 0 at -∞ and 1 at +∞ and satisfies a continuity property.

theorem ProbabilityTheory.isRatStieltjesPoint_unit_prod_iff {α : Type u_1} (f : α) (a : α) :
IsRatStieltjesPoint (fun (p : Unit × α) => f p.2) ((), a) IsRatStieltjesPoint f a
theorem ProbabilityTheory.IsRatStieltjesPoint.ite {α : Type u_1} {f g : α} {a : α} (p : αProp) [DecidablePred p] (hf : p aIsRatStieltjesPoint f a) (hg : ¬p aIsRatStieltjesPoint g a) :
IsRatStieltjesPoint (fun (a : α) => if p a then f a else g a) a
structure ProbabilityTheory.IsMeasurableRatCDF {α : Type u_1} [MeasurableSpace α] (f : α) :

A function f : α → ℚ → ℝ is a (kernel) rational cumulative distribution function if it is measurable in the first argument and if f a satisfies a list of properties for all a : α: monotonicity between 0 at -∞ and 1 at +∞ and a form of continuity.

A function with these properties can be extended to a measurable function α → StieltjesFunction. See ProbabilityTheory.IsMeasurableRatCDF.stieltjesFunction.

theorem ProbabilityTheory.IsMeasurableRatCDF.nonneg {α : Type u_1} [MeasurableSpace α] {f : α} (hf : IsMeasurableRatCDF f) (a : α) (q : ) :
0 f a q
theorem ProbabilityTheory.IsMeasurableRatCDF.le_one {α : Type u_1} [MeasurableSpace α] {f : α} (hf : IsMeasurableRatCDF f) (a : α) (q : ) :
f a q 1
theorem ProbabilityTheory.IsMeasurableRatCDF.iInf_rat_gt_eq {α : Type u_1} [MeasurableSpace α] {f : α} (hf : IsMeasurableRatCDF f) (a : α) (q : ) :
⨅ (r : (Set.Ioi q)), f a r = f a q

A function with the property IsMeasurableRatCDF. Used in a piecewise construction to convert a function which only satisfies the properties defining IsMeasurableRatCDF on some set into a true IsMeasurableRatCDF.

Equations
noncomputable def ProbabilityTheory.toRatCDF {α : Type u_1} (f : α) :
α

Turn a function f : α → ℚ → ℝ into another with the property IsRatStieltjesPoint f a everywhere. At a that does not satisfy that property, f a is replaced by an arbitrary suitable function. Mainly useful when f satisfies the property IsRatStieltjesPoint f a almost everywhere with respect to some measure.

Equations
theorem ProbabilityTheory.toRatCDF_of_isRatStieltjesPoint {α : Type u_1} {f : α} {a : α} (h : IsRatStieltjesPoint f a) (q : ) :
toRatCDF f a q = f a q
theorem ProbabilityTheory.toRatCDF_unit_prod {α : Type u_1} {f : α} (a : α) :
toRatCDF (fun (p : Unit × α) => f p.2) ((), a) = toRatCDF f a
@[irreducible]
noncomputable def ProbabilityTheory.IsMeasurableRatCDF.stieltjesFunctionAux {α : Type u_2} (f : α) :
α

Auxiliary definition for IsMeasurableRatCDF.stieltjesFunction: turn f : α → ℚ → ℝ into a function α → ℝ → ℝ by assigning to f a x the infimum of f a q over q : ℚ with x < q.

Equations
theorem ProbabilityTheory.IsMeasurableRatCDF.stieltjesFunctionAux_def {α : Type u_2} (f : α) (a : α) (x : ) :
stieltjesFunctionAux f a x = ⨅ (q : { q' : // x < q' }), f a q
theorem ProbabilityTheory.IsMeasurableRatCDF.stieltjesFunctionAux_def' {α : Type u_1} (f : α) (a : α) :
stieltjesFunctionAux f a = fun (t : ) => ⨅ (r : { r' : // t < r' }), f a r
theorem ProbabilityTheory.IsMeasurableRatCDF.stieltjesFunctionAux_eq {α : Type u_1} {f : α} [MeasurableSpace α] (hf : IsMeasurableRatCDF f) (a : α) (r : ) :
stieltjesFunctionAux f a r = f a r
noncomputable def ProbabilityTheory.IsMeasurableRatCDF.stieltjesFunction {α : Type u_1} {f : α} [MeasurableSpace α] (hf : IsMeasurableRatCDF f) (a : α) :

Extend a function f : α → ℚ → ℝ with property IsMeasurableRatCDF from to , to a function α → StieltjesFunction.

Equations
theorem ProbabilityTheory.IsMeasurableRatCDF.stieltjesFunction_eq {α : Type u_1} {f : α} [MeasurableSpace α] (hf : IsMeasurableRatCDF f) (a : α) (r : ) :
(hf.stieltjesFunction a) r = f a r
noncomputable def ProbabilityTheory.stieltjesOfMeasurableRat {α : Type u_1} [MeasurableSpace α] (f : α) (hf : Measurable f) :

Turn a measurable function f : α → ℚ → ℝ into a measurable function α → StieltjesFunction. Composition of toRatCDF and IsMeasurableRatCDF.stieltjesFunction.

Equations
theorem ProbabilityTheory.stieltjesOfMeasurableRat_eq {α : Type u_1} {f : α} [MeasurableSpace α] (hf : Measurable f) (a : α) (r : ) :
(stieltjesOfMeasurableRat f hf a) r = toRatCDF f a r
theorem ProbabilityTheory.stieltjesOfMeasurableRat_unit_prod {α : Type u_1} {f : α} [MeasurableSpace α] (hf : Measurable f) (a : α) :
stieltjesOfMeasurableRat (fun (p : Unit × α) => f p.2) ((), a) = stieltjesOfMeasurableRat f hf a
theorem ProbabilityTheory.stieltjesOfMeasurableRat_nonneg {α : Type u_1} {f : α} [MeasurableSpace α] (hf : Measurable f) (a : α) (r : ) :
theorem ProbabilityTheory.stieltjesOfMeasurableRat_le_one {α : Type u_1} {f : α} [MeasurableSpace α] (hf : Measurable f) (a : α) (x : ) :
theorem ProbabilityTheory.measurable_stieltjesOfMeasurableRat {α : Type u_1} {f : α} [MeasurableSpace α] (hf : Measurable f) (x : ) :
Measurable fun (a : α) => (stieltjesOfMeasurableRat f hf a) x