Documentation

Newton.MeasureTheory.Integral.Tonelli

Tonelli's Theorem for Nonnegative Functions #

These are the core statements of Tonelli's theorem that we need for the convolution case.

theorem Newton.tonelli_nonneg_prod_eq_iterated {G : Type u_1} [MeasurableSpace G] {μ : MeasureTheory.Measure G} [MeasureTheory.SFinite μ] (f : G × GENNReal) (hf_meas : Measurable f) :
∫⁻ (p : G × G), f p μ.prod μ = ∫⁻ (x : G), ∫⁻ (y : G), f (x, y) μ μ

Tonelli's Theorem: Integral Equality on Product Measures

For a measurable nonnegative function on a product space, the double integral equals the iterated integral (in any order).

This is the fundamental statement: If f : G × G → ℝ≥0∞ is measurable and nonnegative, then: ∫⁻ (x,y), f (x,y) ∂(μ.prod μ) = ∫⁻ x, ∫⁻ y, f (x,y) ∂μ ∂μ

theorem Newton.tonelli_ae_section_lt_top {G : Type u_1} [MeasurableSpace G] {μ : MeasureTheory.Measure G} [MeasureTheory.SFinite μ] {f : G × GENNReal} (hf_meas : Measurable f) (hf_int : ∫⁻ (p : G × G), f p μ.prod μ < ) :
∀ᵐ (x : G) μ, ∫⁻ (y : G), f (x, y) μ <

Section Finiteness from Product Finiteness (Tonelli Consequence)

If ∫⁻ (x,y), f (x,y) ∂(μ.prod μ) < ∞ where f is measurable and nonnegative, then for a.e. x, we have ∫⁻ y, f (x,y) ∂μ < ∞.

This is the key consequence of Tonelli for our application.

theorem Newton.tonelli_product_ae_section_lt_top {G : Type u_1} [MeasurableSpace G] {μ : MeasureTheory.Measure G} [MeasureTheory.SFinite μ] {f g : GENNReal} (hf_meas : Measurable f) (hg_meas : Measurable g) (hf_int : ∫⁻ (x : G), f x μ < ) (hg_int : ∫⁻ (y : G), g y μ < ) :
∀ᵐ (x : G) μ, ∫⁻ (y : G), f x * g y μ <

Tonelli for Norm Products (Nonnegative)

For measurable nonnegative functions f, g : G → ℝ≥0∞, if ∫⁻ (x, y), f x * g y ∂(μ.prod μ) < ∞, then for a.e. x, ∫⁻ y, f x * g y ∂μ < ∞.

Tonelli for Convolution Kernels #

These theorems apply Tonelli directly to the convolution kernel case.

theorem Newton.tonelli_norm_convolution_section_ae {G : Type u_1} [MeasurableSpace G] {μ : MeasureTheory.Measure G} [AddCommGroup G] {f g : G} [MeasureTheory.SFinite μ] (h_kernel_int : MeasureTheory.Integrable (fun (q : G × G) => f (q.1 - q.2) * g q.2) (μ.prod μ)) :
∀ᵐ (x : G) μ, MeasureTheory.Integrable (fun (y : G) => f (x - y) * g y) μ

Tonelli for Norm Convolution Kernels (Key Application)

For f, g : G → ℂ and measurable norms, if (x,y) ↦ f(x - y) * g(y) is integrable on μ.prod μ, then for a.e. x, y ↦ ‖f(x - y)‖ * ‖g(y)‖ is integrable on μ.

This theorem bridges from the product-level assumption to the section-level conclusion.

theorem Newton.tonelli_real_convolution_section_ae {G : Type u_1} [MeasurableSpace G] {μ : MeasureTheory.Measure G} [AddCommGroup G] {f g : G} [MeasurableSub₂ G] [MeasureTheory.SFinite μ] (hf_meas : Measurable f) (hg_meas : Measurable g) (h_kernel_int : ∫⁻ (p : G × G), ENNReal.ofReal (|f (p.1 - p.2)| * |g p.2|) μ.prod μ < ) :
∀ᵐ (x : G) μ, ∫⁻ (y : G), ENNReal.ofReal (|f (x - y)| * |g y|) μ <

Tonelli for Real-Valued Convolution Kernels

For f, g : G → ℝ and nonnegative values, if (x,y) ↦ f(x - y) * g(y) has finite double integral, then for a.e. x, the section has finite integral.

Tonelli's Theorem: Product Decomposition #

Finer-grained versions of Tonelli that separate the roles of f and g.

Tonelli: Separate Factors

For f, g : G → ℂ with appropriate boundedness and a translation-invariant measure μ, ∫⁻ (x,y), ‖f(x-y) * g(y)‖ ∂(μ.prod μ) ≤ (∫⁻ x, ‖f x‖ ∂μ) * (∫⁻ y, ‖g y‖ ∂μ)

when these integrals are finite.

Auxiliary a.e.-finiteness lemmas for ℝ≥0∞-valued kernels assuming only AEMeasurable on the product space. Proofs will be supplied where used.

theorem Newton.tonelli_ae_section_lt_top_of_aemeasurable_left {G : Type u_1} [MeasurableSpace G] {μ : MeasureTheory.Measure G} [MeasureTheory.SFinite μ] {f : G × GENNReal} (hf_aemeas : AEMeasurable f (μ.prod μ)) (hf_int : ∫⁻ (p : G × G), f p μ.prod μ < ) :
∀ᵐ (x : G) μ, ∫⁻ (y : G), f (x, y) μ <

From product finiteness for an AEMeasurable nonnegative kernel on μ.prod μ, deduce that for a.e. x, the section y ↦ f (x, y) has finite lintegral.

theorem Newton.tonelli_ae_section_lt_top_of_aemeasurable_right {G : Type u_1} [MeasurableSpace G] {μ : MeasureTheory.Measure G} [MeasureTheory.SFinite μ] {f : G × GENNReal} (hf_aemeas : AEMeasurable f (μ.prod μ)) (hf_int : ∫⁻ (p : G × G), f p μ.prod μ < ) :
∀ᵐ (y : G) μ, ∫⁻ (x : G), f (x, y) μ <

Symmetric version: from product finiteness for an AEMeasurable nonnegative kernel, deduce that for a.e. y, the section x ↦ f (x, y) has finite lintegral.

theorem Newton.ae_integrable_left_of_lintegral_ofReal_lt_top {G : Type u_2} [MeasurableSpace G] (μ : MeasureTheory.Measure G) [MeasureTheory.SFinite μ] {H : G × G} (hH_nonneg : ∀ (p : G × G), 0 H p) (hH_aemeas : AEMeasurable H (μ.prod μ)) (hH_int : ∫⁻ (p : G × G), ENNReal.ofReal (H p) μ.prod μ < ) :
∀ᵐ (x : G) μ, MeasureTheory.Integrable (fun (y : G) => H (x, y)) μ

From product finiteness for an AEMeasurable nonnegative real-valued kernel on μ.prod μ, deduce that for a.e. x, the section y ↦ H (x, y) is Bochner integrable (as an ℝ-valued function).

theorem Newton.integrable_left_integral_of_lintegral_ofReal_lt_top {G : Type u_2} [MeasurableSpace G] (μ : MeasureTheory.Measure G) [MeasureTheory.SFinite μ] {H : G × G} (hH_nonneg : ∀ (p : G × G), 0 H p) (hH_aemeas : AEMeasurable H (μ.prod μ)) (hH_int : ∫⁻ (p : G × G), ENNReal.ofReal (H p) μ.prod μ < ) :
MeasureTheory.Integrable (fun (x : G) => (y : G), H (x, y) μ) μ

From product finiteness for an AEMeasurable nonnegative real-valued kernel on μ.prod μ, deduce that the outer integral x ↦ ∫ y, H (x, y) dμ is Bochner integrable on μ.

Fiberwise integrability for L²×L¹ convolution kernels.

theorem Newton.lintegral_ofReal_eq_ofReal_integral_prod {G : Type u_2} [MeasurableSpace G] (μ : MeasureTheory.Measure G) [MeasureTheory.SFinite μ] {H : G × G} (hH_nonneg : ∀ (p : G × G), 0 H p) (hH_int : MeasureTheory.Integrable H (μ.prod μ)) :
∫⁻ (p : G × G), ENNReal.ofReal (H p) μ.prod μ = ENNReal.ofReal ( (x : G), (y : G), H (x, y) μ μ)

Tonelli's theorem for nonnegative real-valued functions on a product space: the lintegral of ENNReal.ofReal ∘ H on μ.prod μ equals ENNReal.ofReal of the iterated Bochner integral when H is nonnegative and AEMeasurable.

This bridges the gap between ENNReal-valued lintegrals and real-valued iterated integrals.

theorem Newton.lintegral_ofReal_le_ofReal_integral_prod {G : Type u_2} [MeasurableSpace G] (μ : MeasureTheory.Measure G) [MeasureTheory.SFinite μ] {H : G × G} (hH_nonneg : ∀ (p : G × G), 0 H p) (hH_aemeas : AEMeasurable H (μ.prod μ)) (hH_lt_top : ∫⁻ (p : G × G), ENNReal.ofReal (H p) μ.prod μ < ) :
∫⁻ (p : G × G), ENNReal.ofReal (H p) μ.prod μ ENNReal.ofReal ( (x : G), (y : G), H (x, y) μ μ)

Upper bound version of Tonelli for nonnegative functions: the lintegral of ENNReal.ofReal ∘ H equals ENNReal.ofReal of the iterated integral when H is nonnegative, AEMeasurable, and has finite lintegral.