Infinite-dimensional operator ridgelet transform

4.2. Integral representation🔗

DefinitionAux.5
Statement uses 2
Statement dependency previews
Preview
Definition Aux.4
Loading preview
Statement dependency preview content is loaded from the rendered-fragment cache.
Used by 5
Reverse dependency previews
Preview
Theorem 4.5
Loading preview
Reverse dependency preview content is loaded from the rendered-fragment cache.
✓L∃∀N

This is the activation and distributional-pairing setup for Theorem 4.5(iii).

A tempered distribution \beta\in\mathcal S'(\mathbb R) that is a continuous function of polynomial growth is the pair of \beta and a continuous b:\mathbb R\to\mathbb R with |b(t)|\le C(1+|t|)^p and \langle\beta,\varphi\rangle=\int b\varphi. For a band-pass \rho the test filter \omega\mapsto\rho^\sharp(-\omega)|\omega|^{-\alpha} is a Schwartz function, and the distributional admissibility constant is C_{\beta,\rho}^{(\alpha)}=\frac1{2\pi}\langle\beta^\sharp,\rho^\sharp(-\,\cdot\,)|\cdot|^{-\alpha}\rangle.

Lean code for DefinitionAux.5●3 definitions
  • structure(3 fields)defined in OperatorRidgelet/Reconstruction/Defs.lean
    complete
    structure OperatorRidgelet.IsTemperedFunction (β : TemperedDistribution ℝ ℂ)
      (b : ℝ → ℝ) : Prop
    structure OperatorRidgelet.IsTemperedFunction
      (β : TemperedDistribution ℝ ℂ)
      (b : ℝ → ℝ) : Prop
    `β ∈ 𝒮'(ℝ)` is the continuous function `b` of polynomial growth: `b` is continuous,
    `|b(t)| ≤ C (1+|t|)^p`, and `β` acts on test functions by integration against `b`. 
    continuous : Continuous b
    `b` is continuous. 
    polynomialGrowth : OperatorRidgelet.HasPolynomialGrowth b
    `b` has polynomial growth. 
    apply_eq : ∀ (φ : SchwartzMap ℝ ℂ), β φ = ∫ (t : ℝ), ↑(b t) * φ t
    `β` is integration against `b`. 
  • complete
    def OperatorRidgelet.temperedTestFilter (α : ℝ) (ρ : SchwartzMap ℝ ℝ) :
      SchwartzMap ℝ ℂ
    def OperatorRidgelet.temperedTestFilter
      (α : ℝ) (ρ : SchwartzMap ℝ ℝ) :
      SchwartzMap ℝ ℂ
    The test filter `ω ↦ ρ♯(-ω) |ω|^{-α}` as a Schwartz function, when one with these values
    exists (in particular when `ρ` is band-pass); `0` otherwise. 
  • complete
    def OperatorRidgelet.temperedAdmissibilityConst (α : ℝ)
      (β : TemperedDistribution ℝ ℂ) (ρ : SchwartzMap ℝ ℝ) : ℂ
    def OperatorRidgelet.temperedAdmissibilityConst
      (α : ℝ) (β : TemperedDistribution ℝ ℂ)
      (ρ : SchwartzMap ℝ ℝ) : ℂ
    The distributional admissibility constant
    `C^{(α)}_{β,ρ} = (2π)⁻¹ ⟨β♯, ρ♯(-·) |·|^{-α}⟩` of a tempered activation `β` and a filter `ρ`. 
Theorem4.5
Statement uses 4
Statement dependency previews
Preview
Definition 2.2
Loading preview
Statement dependency preview content is loaded from the rendered-fragment cache.
Used by 7
Reverse dependency previews
Preview
Theorem 4.11
Loading preview
Reverse dependency preview content is loaded from the rendered-fragment cache.
✓L∃∀N

Let \alpha>0. Part (ii) assumes an \alpha-admissible Schwartz filter \rho; part (iii) assumes a band-pass filter. (i) For G\in L^1(\nu_\alpha), the function g_G is bounded with \|g_G\|_\infty\le\|G\|_{L^1(\nu_\alpha)} and continuous, and g_G=0 only if G=0 \nu_\alpha-almost everywhere. (ii) For G\in L^1(\nu_\alpha)\cap L^2(\nu_\alpha) and every x, the iterated integral \int_H[\int_{\mathbb R}\gamma_G(a,c)\rho(\langle a,x\rangle+c)\,\mathrm dc]\,\nu_\alpha(\mathrm da)=(\!(\rho,\rho)\!)_\alpha g_G(x) converges absolutely, and if \gamma_G\in L^1(\lambda_\alpha) its left side is the integral network S_\rho[\gamma_G\lambda_\alpha](x). (iii) For a tempered \beta that is a continuous function of polynomial growth and G satisfying Definition 4.2, the integrand \gamma_G(a,c)\beta(\langle a,x\rangle+c) is absolutely integrable on the product space. Its integral is the ordinary network with finite coefficient measure \gamma_G\lambda_\alpha and equals C_{\beta,\rho}^{(\alpha)}g_G(x). This identity allows a zero constant. Non-polynomiality is needed separately to choose a band-pass filter with a nonzero constant, then normalize it for reconstruction or universality. A polynomial activation has zero band-pass pairing.

Lean code for Theorem4.5●11 theorems
  • complete
    theorem OperatorRidgelet.Paper.thm_4_5_i_a.{u_1} {H : Type u_1}
      [NormedAddCommGroup H] [InnerProductSpace ℝ H] [MeasurableSpace H]
      (ν : MeasureTheory.Measure H) (G : H → ℂ) (hG : Measurable G)
      (hG₁ : MeasureTheory.Integrable G ν) (x : H) :
      ‖OperatorRidgelet.spectralTarget ν G x‖ ≤ ∫ (ξ : H), ‖G ξ‖ ∂ν
    theorem OperatorRidgelet.Paper.thm_4_5_i_a.{u_1}
      {H : Type u_1} [NormedAddCommGroup H]
      [InnerProductSpace ℝ H]
      [MeasurableSpace H]
      (ν : MeasureTheory.Measure H)
      (G : H → ℂ) (hG : Measurable G)
      (hG₁ : MeasureTheory.Integrable G ν)
      (x : H) :
      ‖OperatorRidgelet.spectralTarget ν G
            x‖ ≤
        ∫ (ξ : H), ‖G ξ‖ ∂ν
    **Theorem [thm:4.5]** Integral representation of targets with a spectral density.  For
    `G ∈ L¹(ν_α)` the target `g_G` is bounded by `‖G‖_{L¹(ν_α)}`. 
  • complete
    theorem OperatorRidgelet.Paper.thm_4_5_i_b.{u_1} {H : Type u_1}
      [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
      [SecondCountableTopology H] [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H) (G : H → ℂ) (hG : Measurable G)
      (hG₁ : MeasureTheory.Integrable G ν) :
      Continuous (OperatorRidgelet.spectralTarget ν G)
    theorem OperatorRidgelet.Paper.thm_4_5_i_b.{u_1}
      {H : Type u_1} [NormedAddCommGroup H]
      [InnerProductSpace ℝ H]
      [CompleteSpace H]
      [SecondCountableTopology H]
      [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H)
      (G : H → ℂ) (hG : Measurable G)
      (hG₁ : MeasureTheory.Integrable G ν) :
      Continuous
        (OperatorRidgelet.spectralTarget ν G)
    **Theorem [thm:4.5]** Integral representation of targets with a spectral density.  For
    `G ∈ L¹(ν_α)` the target `g_G` is continuous. 
  • complete
    theorem OperatorRidgelet.Paper.thm_4_5_i_c.{u_1} {H : Type u_1}
      [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
      [SecondCountableTopology H] [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H) (G : H → ℂ) (hG : Measurable G)
      (hG₁ : MeasureTheory.Integrable G ν)
      (h : OperatorRidgelet.spectralTarget ν G = 0) : G =ᵐ[ν] 0
    theorem OperatorRidgelet.Paper.thm_4_5_i_c.{u_1}
      {H : Type u_1} [NormedAddCommGroup H]
      [InnerProductSpace ℝ H]
      [CompleteSpace H]
      [SecondCountableTopology H]
      [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H)
      (G : H → ℂ) (hG : Measurable G)
      (hG₁ : MeasureTheory.Integrable G ν)
      (h :
        OperatorRidgelet.spectralTarget ν G =
          0) :
      G =ᵐ[ν] 0
    **Theorem [thm:4.5]** Integral representation of targets with a spectral density.  For
    `G ∈ L¹(ν_α)`, `g_G = 0` only if `G = 0` `ν_α`-almost everywhere. 
  • complete
    theorem OperatorRidgelet.Paper.thm_4_5_ii_a.{u_1} {H : Type u_1}
      [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
      [SecondCountableTopology H] [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H) [MeasureTheory.SigmaFinite ν]
      [ν.IsOpenPosMeasure] {α : ℝ} (hα : 0 < α)
      (hν : OperatorRidgelet.IsHomogeneous α ν) (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsAdmissible α ρ) (G : H → ℂ)
      (hG : Measurable G) (hG₁ : MeasureTheory.Integrable G ν)
      (hG₂ : MeasureTheory.MemLp G 2 ν) (x : H) :
      (∀ᵐ (a : H) ∂ν,
          MeasureTheory.Integrable
            (fun c =>
              OperatorRidgelet.coefficientFormula (⇑ρ) G (a, c) *
                ↑(ρ (inner ℝ a x + c)))
            MeasureTheory.volume) ∧
        MeasureTheory.Integrable
          (fun a =>
            ∫ (c : ℝ),
              OperatorRidgelet.coefficientFormula (⇑ρ) G (a, c) *
                ↑(ρ (inner ℝ a x + c)))
          ν
    theorem OperatorRidgelet.Paper.thm_4_5_ii_a.{u_1}
      {H : Type u_1} [NormedAddCommGroup H]
      [InnerProductSpace ℝ H]
      [CompleteSpace H]
      [SecondCountableTopology H]
      [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H)
      [MeasureTheory.SigmaFinite ν]
      [ν.IsOpenPosMeasure] {α : ℝ}
      (hα : 0 < α)
      (hν :
        OperatorRidgelet.IsHomogeneous α ν)
      (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsAdmissible α ρ)
      (G : H → ℂ) (hG : Measurable G)
      (hG₁ : MeasureTheory.Integrable G ν)
      (hG₂ : MeasureTheory.MemLp G 2 ν)
      (x : H) :
      (∀ᵐ (a : H) ∂ν,
          MeasureTheory.Integrable
            (fun c =>
              OperatorRidgelet.coefficientFormula
                  (⇑ρ) G (a, c) *
                ↑(ρ (inner ℝ a x + c)))
            MeasureTheory.volume) ∧
        MeasureTheory.Integrable
          (fun a =>
            ∫ (c : ℝ),
              OperatorRidgelet.coefficientFormula
                  (⇑ρ) G (a, c) *
                ↑(ρ (inner ℝ a x + c)))
          ν
    **Theorem [thm:4.5]** Integral representation of targets with a spectral density.  For
    `G ∈ L¹(ν_α) ∩ L²(ν_α)` and every `x`, the iterated integral
    `∫ [∫ γ_G(a,c) ρ(⟨a,x⟩+c) dc] ν_α(da)` converges absolutely: the inner integral converges
    absolutely for `ν_α`-almost every `a`, and the outer integrand is `ν_α`-integrable. 
  • complete
    theorem OperatorRidgelet.Paper.thm_4_5_ii_b.{u_1} {H : Type u_1}
      [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
      [SecondCountableTopology H] [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H) [MeasureTheory.SigmaFinite ν]
      [ν.IsOpenPosMeasure] {α : ℝ} (hα : 0 < α)
      (hν : OperatorRidgelet.IsHomogeneous α ν) (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsAdmissible α ρ) (G : H → ℂ)
      (hG : Measurable G) (hG₁ : MeasureTheory.Integrable G ν)
      (hG₂ : MeasureTheory.MemLp G 2 ν) (x : H) :
      ∫ (a : H),
          ∫ (c : ℝ),
            OperatorRidgelet.coefficientFormula (⇑ρ) G (a, c) *
              ↑(ρ (inner ℝ a x + c)) ∂ν =
        ↑(OperatorRidgelet.admissibilityConst α ⇑ρ) *
          OperatorRidgelet.spectralTarget ν G x
    theorem OperatorRidgelet.Paper.thm_4_5_ii_b.{u_1}
      {H : Type u_1} [NormedAddCommGroup H]
      [InnerProductSpace ℝ H]
      [CompleteSpace H]
      [SecondCountableTopology H]
      [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H)
      [MeasureTheory.SigmaFinite ν]
      [ν.IsOpenPosMeasure] {α : ℝ}
      (hα : 0 < α)
      (hν :
        OperatorRidgelet.IsHomogeneous α ν)
      (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsAdmissible α ρ)
      (G : H → ℂ) (hG : Measurable G)
      (hG₁ : MeasureTheory.Integrable G ν)
      (hG₂ : MeasureTheory.MemLp G 2 ν)
      (x : H) :
      ∫ (a : H),
          ∫ (c : ℝ),
            OperatorRidgelet.coefficientFormula
                (⇑ρ) G (a, c) *
              ↑(ρ (inner ℝ a x + c)) ∂ν =
        ↑(OperatorRidgelet.admissibilityConst
              α ⇑ρ) *
          OperatorRidgelet.spectralTarget ν G
            x
    **Theorem [thm:4.5]** Integral representation of targets with a spectral density.  For
    `G ∈ L¹(ν_α) ∩ L²(ν_α)` the spectral synthesis identity
    `∫ [∫ γ_G(a,c) ρ(⟨a,x⟩+c) dc] ν_α(da) = C^{(α)}_ρ g_G(x)` holds for every `x`. 
  • complete
    theorem OperatorRidgelet.Paper.thm_4_5_ii_c.{u_1} {H : Type u_1}
      [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
      [SecondCountableTopology H] [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H) [MeasureTheory.SigmaFinite ν]
      [ν.IsOpenPosMeasure] {α : ℝ} (hα : 0 < α)
      (hν : OperatorRidgelet.IsHomogeneous α ν) (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsAdmissible α ρ) (G : H → ℂ)
      (hG : Measurable G) (hG₁ : MeasureTheory.Integrable G ν)
      (hG₂ : MeasureTheory.MemLp G 2 ν)
      (hγ :
        MeasureTheory.Integrable
          (OperatorRidgelet.coefficientFormula (⇑ρ) G)
          (OperatorRidgelet.parameterMeasure ν))
      (x : H) :
      ∫ (a : H),
          ∫ (c : ℝ),
            OperatorRidgelet.coefficientFormula (⇑ρ) G (a, c) *
              ↑(ρ (inner ℝ a x + c)) ∂ν =
        OperatorRidgelet.integralNetworkDensity (fun t => ↑(ρ t))
          (OperatorRidgelet.parameterMeasure ν)
          (OperatorRidgelet.coefficientFormula (⇑ρ) G) x
    theorem OperatorRidgelet.Paper.thm_4_5_ii_c.{u_1}
      {H : Type u_1} [NormedAddCommGroup H]
      [InnerProductSpace ℝ H]
      [CompleteSpace H]
      [SecondCountableTopology H]
      [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H)
      [MeasureTheory.SigmaFinite ν]
      [ν.IsOpenPosMeasure] {α : ℝ}
      (hα : 0 < α)
      (hν :
        OperatorRidgelet.IsHomogeneous α ν)
      (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsAdmissible α ρ)
      (G : H → ℂ) (hG : Measurable G)
      (hG₁ : MeasureTheory.Integrable G ν)
      (hG₂ : MeasureTheory.MemLp G 2 ν)
      (hγ :
        MeasureTheory.Integrable
          (OperatorRidgelet.coefficientFormula
            (⇑ρ) G)
          (OperatorRidgelet.parameterMeasure
            ν))
      (x : H) :
      ∫ (a : H),
          ∫ (c : ℝ),
            OperatorRidgelet.coefficientFormula
                (⇑ρ) G (a, c) *
              ↑(ρ (inner ℝ a x + c)) ∂ν =
        OperatorRidgelet.integralNetworkDensity
          (fun t => ↑(ρ t))
          (OperatorRidgelet.parameterMeasure
            ν)
          (OperatorRidgelet.coefficientFormula
            (⇑ρ) G)
          x
    **Theorem [thm:4.5]** Integral representation of targets with a spectral density.  If moreover
    `γ_G ∈ L¹(λ_α)`, the left side of the spectral synthesis identity is the integral network
    `S_ρ[γ_G λ_α](x)`. 
  • complete
    theorem OperatorRidgelet.Paper.thm_4_5_iii_a.{u_1} {H : Type u_1}
      [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
      [SecondCountableTopology H] [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H) [MeasureTheory.SigmaFinite ν]
      [ν.IsOpenPosMeasure] {α : ℝ} (hα : 0 < α)
      (hν : OperatorRidgelet.IsHomogeneous α ν) (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsBandPass ρ) (I : Set ℝ)
      (hI : OperatorRidgelet.IsFrequencyWindow (⇑ρ) I)
      (β : TemperedDistribution ℝ ℂ) (b : ℝ → ℝ)
      (hβ : OperatorRidgelet.IsTemperedFunction β b) (G : H → ℂ)
      (hG : OperatorRidgelet.IsRegularAlongRays ν I G) (x : H) :
      ∀ᵐ (a : H) ∂ν,
        MeasureTheory.Integrable
          (fun c =>
            OperatorRidgelet.coefficientFormula (⇑ρ) G (a, c) *
              ↑(b (inner ℝ a x + c)))
          MeasureTheory.volume
    theorem OperatorRidgelet.Paper.thm_4_5_iii_a.{u_1}
      {H : Type u_1} [NormedAddCommGroup H]
      [InnerProductSpace ℝ H]
      [CompleteSpace H]
      [SecondCountableTopology H]
      [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H)
      [MeasureTheory.SigmaFinite ν]
      [ν.IsOpenPosMeasure] {α : ℝ}
      (hα : 0 < α)
      (hν :
        OperatorRidgelet.IsHomogeneous α ν)
      (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsBandPass ρ)
      (I : Set ℝ)
      (hI :
        OperatorRidgelet.IsFrequencyWindow
          (⇑ρ) I)
      (β : TemperedDistribution ℝ ℂ)
      (b : ℝ → ℝ)
      (hβ :
        OperatorRidgelet.IsTemperedFunction β
          b)
      (G : H → ℂ)
      (hG :
        OperatorRidgelet.IsRegularAlongRays ν
          I G)
      (x : H) :
      ∀ᵐ (a : H) ∂ν,
        MeasureTheory.Integrable
          (fun c =>
            OperatorRidgelet.coefficientFormula
                (⇑ρ) G (a, c) *
              ↑(b (inner ℝ a x + c)))
          MeasureTheory.volume
    **Theorem [thm:4.5]** Integral representation of targets with a spectral density.  For a
    tempered `β` that is a continuous function `b` of polynomial growth, and
    `G` regular along rays, the inner integral `∫ γ_G(a,c) β(⟨a,x⟩+c) dc` converges absolutely for
    `ν_α`-almost every `a`. 
  • complete
    theorem OperatorRidgelet.Paper.thm_4_5_iii_b.{u_1} {H : Type u_1}
      [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
      [SecondCountableTopology H] [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H) [MeasureTheory.SigmaFinite ν]
      [ν.IsOpenPosMeasure] {α : ℝ} (hα : 0 < α)
      (hν : OperatorRidgelet.IsHomogeneous α ν) (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsBandPass ρ) (I : Set ℝ)
      (hI : OperatorRidgelet.IsFrequencyWindow (⇑ρ) I)
      (β : TemperedDistribution ℝ ℂ) (b : ℝ → ℝ)
      (hβ : OperatorRidgelet.IsTemperedFunction β b) (G : H → ℂ)
      (hG : OperatorRidgelet.IsRegularAlongRays ν I G) (x : H) :
      MeasureTheory.Integrable
        (fun a =>
          ∫ (c : ℝ),
            OperatorRidgelet.coefficientFormula (⇑ρ) G (a, c) *
              ↑(b (inner ℝ a x + c)))
        ν
    theorem OperatorRidgelet.Paper.thm_4_5_iii_b.{u_1}
      {H : Type u_1} [NormedAddCommGroup H]
      [InnerProductSpace ℝ H]
      [CompleteSpace H]
      [SecondCountableTopology H]
      [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H)
      [MeasureTheory.SigmaFinite ν]
      [ν.IsOpenPosMeasure] {α : ℝ}
      (hα : 0 < α)
      (hν :
        OperatorRidgelet.IsHomogeneous α ν)
      (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsBandPass ρ)
      (I : Set ℝ)
      (hI :
        OperatorRidgelet.IsFrequencyWindow
          (⇑ρ) I)
      (β : TemperedDistribution ℝ ℂ)
      (b : ℝ → ℝ)
      (hβ :
        OperatorRidgelet.IsTemperedFunction β
          b)
      (G : H → ℂ)
      (hG :
        OperatorRidgelet.IsRegularAlongRays ν
          I G)
      (x : H) :
      MeasureTheory.Integrable
        (fun a =>
          ∫ (c : ℝ),
            OperatorRidgelet.coefficientFormula
                (⇑ρ) G (a, c) *
              ↑(b (inner ℝ a x + c)))
        ν
    **Theorem [thm:4.5]** Integral representation of targets with a spectral density.  For a
    tempered `β` that is a continuous function `b` of polynomial growth, and
    `G` regular along rays, the `ν_α`-integral of the inner integral converges absolutely. 
  • complete
    theorem OperatorRidgelet.Paper.thm_4_5_iii_c.{u_1} {H : Type u_1}
      [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
      [SecondCountableTopology H] [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H) [MeasureTheory.SigmaFinite ν]
      [ν.IsOpenPosMeasure] {α : ℝ} (hα : 0 < α)
      (hν : OperatorRidgelet.IsHomogeneous α ν) (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsBandPass ρ) (I : Set ℝ)
      (hI : OperatorRidgelet.IsFrequencyWindow (⇑ρ) I)
      (β : TemperedDistribution ℝ ℂ) (b : ℝ → ℝ)
      (hβ : OperatorRidgelet.IsTemperedFunction β b) (G : H → ℂ)
      (hG : OperatorRidgelet.IsRegularAlongRays ν I G) (x : H) :
      ∫ (a : H),
          ∫ (c : ℝ),
            OperatorRidgelet.coefficientFormula (⇑ρ) G (a, c) *
              ↑(b (inner ℝ a x + c)) ∂ν =
        OperatorRidgelet.temperedAdmissibilityConst α β ρ *
          OperatorRidgelet.spectralTarget ν G x
    theorem OperatorRidgelet.Paper.thm_4_5_iii_c.{u_1}
      {H : Type u_1} [NormedAddCommGroup H]
      [InnerProductSpace ℝ H]
      [CompleteSpace H]
      [SecondCountableTopology H]
      [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H)
      [MeasureTheory.SigmaFinite ν]
      [ν.IsOpenPosMeasure] {α : ℝ}
      (hα : 0 < α)
      (hν :
        OperatorRidgelet.IsHomogeneous α ν)
      (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsBandPass ρ)
      (I : Set ℝ)
      (hI :
        OperatorRidgelet.IsFrequencyWindow
          (⇑ρ) I)
      (β : TemperedDistribution ℝ ℂ)
      (b : ℝ → ℝ)
      (hβ :
        OperatorRidgelet.IsTemperedFunction β
          b)
      (G : H → ℂ)
      (hG :
        OperatorRidgelet.IsRegularAlongRays ν
          I G)
      (x : H) :
      ∫ (a : H),
          ∫ (c : ℝ),
            OperatorRidgelet.coefficientFormula
                (⇑ρ) G (a, c) *
              ↑(b (inner ℝ a x + c)) ∂ν =
        OperatorRidgelet.temperedAdmissibilityConst
            α β ρ *
          OperatorRidgelet.spectralTarget ν G
            x
    **Theorem [thm:4.5]** Integral representation of targets with a spectral density.  For a
    tempered `β` that is a continuous function `b` of polynomial growth, and
    `G` regular along rays, the tempered spectral synthesis identity
    `∫ [∫ γ_G(a,c) β(⟨a,x⟩+c) dc] ν_α(da) = C^{(α)}_{β,ρ} g_G(x)` holds. 
  • complete
    theorem OperatorRidgelet.Paper.thm_4_5_iii_d {α : ℝ} (hα : 0 < α)
      (β : TemperedDistribution ℝ ℂ) (b : ℝ → ℝ)
      (hβ : OperatorRidgelet.IsTemperedFunction β b)
      (hb : ¬OperatorRidgelet.IsPolynomialFun b) :
      ∃ ρ,
        OperatorRidgelet.IsBandPass ρ ∧
          OperatorRidgelet.temperedAdmissibilityConst α β ρ ≠ 0
    theorem OperatorRidgelet.Paper.thm_4_5_iii_d
      {α : ℝ} (hα : 0 < α)
      (β : TemperedDistribution ℝ ℂ)
      (b : ℝ → ℝ)
      (hβ :
        OperatorRidgelet.IsTemperedFunction β
          b)
      (hb :
        ¬OperatorRidgelet.IsPolynomialFun b) :
      ∃ ρ,
        OperatorRidgelet.IsBandPass ρ ∧
          OperatorRidgelet.temperedAdmissibilityConst
              α β ρ ≠
            0
    **Theorem [thm:4.5]** Integral representation of targets with a spectral density.  For every
    tempered `β` that is a continuous function of polynomial growth and not a polynomial, there is a
    band-pass filter `ρ` with `C^{(α)}_{β,ρ} ≠ 0`. 
  • complete
    theorem OperatorRidgelet.Paper.thm_4_5_iii_e.{u_1} {H : Type u_1}
      [NormedAddCommGroup H] [InnerProductSpace ℝ H] [MeasurableSpace H]
      [BorelSpace H] (ν : MeasureTheory.Measure H)
      [MeasureTheory.SigmaFinite ν] (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsBandPass ρ) (I : Set ℝ)
      (hI : OperatorRidgelet.IsFrequencyWindow (⇑ρ) I)
      (β : TemperedDistribution ℝ ℂ) (b : ℝ → ℝ)
      (hβ : OperatorRidgelet.IsTemperedFunction β b) (G : H → ℂ)
      (hG : OperatorRidgelet.IsRegularAlongRays ν I G) (x : H) :
      MeasureTheory.Integrable
        (fun θ =>
          OperatorRidgelet.coefficientFormula (⇑ρ) G θ *
            ↑(b (inner ℝ θ.1 x + θ.2)))
        (OperatorRidgelet.parameterMeasure ν)
    theorem OperatorRidgelet.Paper.thm_4_5_iii_e.{u_1}
      {H : Type u_1} [NormedAddCommGroup H]
      [InnerProductSpace ℝ H]
      [MeasurableSpace H] [BorelSpace H]
      (ν : MeasureTheory.Measure H)
      [MeasureTheory.SigmaFinite ν]
      (ρ : SchwartzMap ℝ ℝ)
      (hρ : OperatorRidgelet.IsBandPass ρ)
      (I : Set ℝ)
      (hI :
        OperatorRidgelet.IsFrequencyWindow
          (⇑ρ) I)
      (β : TemperedDistribution ℝ ℂ)
      (b : ℝ → ℝ)
      (hβ :
        OperatorRidgelet.IsTemperedFunction β
          b)
      (G : H → ℂ)
      (hG :
        OperatorRidgelet.IsRegularAlongRays ν
          I G)
      (x : H) :
      MeasureTheory.Integrable
        (fun θ =>
          OperatorRidgelet.coefficientFormula
              (⇑ρ) G θ *
            ↑(b (inner ℝ θ.1 x + θ.2)))
        (OperatorRidgelet.parameterMeasure ν)
    **Theorem [thm:4.5]** The network integrand is jointly absolutely integrable. 
Proof for Theorem 4.5
Proof uses 4
Proof dependency previews
Preview
Lemma 3.2
Loading preview
Proof dependency preview content is loaded from the rendered-fragment cache.

Part (i) is Lemma 4.1. Parseval in the bias turns the inner integral into a frequency integral of \rho^\sharp(\omega)G(-\omega a) against \rho^\sharp(-\omega)e^{-i\omega\langle a,x\rangle}, and the homogeneous substitution \xi=-\omega a separates the admissibility constant from g_G(x). For a tempered \beta the bias integral is a distributional pairing with a test function supported in -\operatorname{supp}\rho^\sharp; Definition 4.2 makes a\mapsto (test function) Bochner integrable in a C^m norm, so the pairing commutes with the direction integral. Joint absolute integrability follows independently from Lemma 4.3, choosing a moment at least as large as the activation's polynomial growth order. If every band-pass test function paired to zero with \beta^\sharp, its support would be \{0\} and \beta a polynomial.