Documentation

Atlas.DifferentialAnalysis.code.FourierInversion

Existence form of the Schwartz Fourier isomorphism (Theorem 9.1 of Melrose): the Fourier transform on the Schwartz space is a continuous -linear equivalence whose inverse is the inverse Fourier transform.

The Fourier transform as a continuous -linear equivalence on the Schwartz space 𝓢(V, E).

Instances For
    @[simp]

    The continuous linear equivalence fourierSchwartzCLE acts as the Fourier transform 𝓕 on Schwartz functions.

    @[simp]

    The inverse of fourierSchwartzCLE acts as the inverse Fourier transform 𝓕⁻ on Schwartz functions.

    Parseval's identity (Lemma 9.2 of Melrose): for two Schwartz functions φ, ψ : V → ℂ, the integral of φ(x) · conj(ψ(x)) equals the integral of 𝓕φ(ξ) · conj(𝓕ψ(ξ)).

    The Fourier transform on tempered distributions packaged as a -linear equivalence 𝓢'(E, F) ≃ₗ[ℂ] 𝓢'(E, F).

    Instances For
      @[simp]

      Fourier inversion on tempered distributions: 𝓕⁻(𝓕 u) = u.

      The k-fold iterated distributional derivative in the j-th coordinate direction, applied to a tempered distribution u.

      Instances For

        The mixed iterated distributional derivative ∂^α u indexed by a multi-index α : Fin n → ℕ, defined by composing the per-coordinate iterated derivatives over all coordinates.

        Instances For

          The k-fold iterated operation of multiplication by 2πi·ξ_j on a tempered distribution, which is the Fourier-side counterpart of the iterated derivative in the j-th coordinate.

          Instances For

            The multi-index iterated multiplication operation (2πi)^|α| ξ^α u on a tempered distribution, Fourier-dual to the multi-index derivative ∂^α.

            Instances For

              The k-fold iterated operation of multiplication by -2πi·x_j on a tempered distribution, used to express the iterated coordinate derivative on the Fourier side after inversion.

              Instances For

                The multi-index iterated multiplication (-2πi)^|α| x^α u on a tempered distribution, packaged via a fold over coordinates.

                Instances For

                  The Fourier transform exchanges iterated coordinate derivatives with iterated coordinate multiplication by 2πi·ξ_j: 𝓕 (∂_j^k u) = (2πi ξ_j)^k · 𝓕 u.

                  Multi-index version of the Fourier-derivative exchange: 𝓕 (∂^α u) = (2πi ξ)^α · 𝓕 u.

                  Dual identity: the iterated coordinate derivative on 𝓕 u equals the Fourier transform of (-2πi x_j)^k u.

                  noncomputable def SobolevSpace.sobolevWeight (n : ) (s : ) (ξ : EuclideanSpace (Fin n)) :

                  The Sobolev weight ⟨ξ⟩^s = (1 + ‖ξ‖²)^{s/2}, written using the Japanese bracket. This is the standard symbol whose -decay determines the Sobolev exponent s.

                  Instances For
                    theorem SobolevSpace.sobolevWeight_pos (n : ) (s : ) (ξ : EuclideanSpace (Fin n)) :
                    0 < sobolevWeight n s ξ

                    The Sobolev weight ⟨ξ⟩^s is strictly positive for every ξ and every real exponent s.

                    The Sobolev weight is nonzero, which is needed when inverting it inside integrals.

                    Membership in the Sobolev space H^s(ℝⁿ): a tempered distribution u belongs to H^s if there exists g ∈ L² such that, for every Schwartz function φ, (𝓕 u)(φ) = ∫ ⟨ξ⟩^{-s} g(ξ) φ(ξ) dξ. Equivalently, 𝓕 u is given by integration against the function ⟨ξ⟩^{-s} g.

                    Instances For
                      theorem SobolevSpace.memHs_of_le {n : } {s t : } (hst : t s) (u : TemperedDistribution (EuclideanSpace (Fin n)) ) (hu : MemHs n s u) :
                      MemHs n t u

                      Monotonicity of Sobolev spaces: H^s ⊆ H^t whenever t ≤ s. The witness function for H^t is obtained by multiplying the witness for H^s by ⟨ξ⟩^{t-s} ≤ 1.

                      def SobolevSpace.Hs (n : ) (s : ) :

                      The Sobolev space H^s(ℝⁿ) realised as the subtype of tempered distributions satisfying MemHs n s.

                      Instances For
                        noncomputable def SobolevSpace.Hs.witnessFunc (n : ) (s : ) (u : Hs n s) :

                        The "witness" function attached to a Sobolev distribution: it is the function g produced by the existential in MemHs. Concretely, 𝓕 u = ⟨ξ⟩^{-s} · witnessFunc s u as distributions.

                        Instances For

                          The witness function attached to a Sobolev distribution is in .

                          noncomputable def SobolevSpace.Hs.witnessL2 (n : ) (s : ) (u : Hs n s) :

                          The witness function packaged as an honest element of the Hilbert space L²(ℝⁿ; ℂ).

                          Instances For
                            theorem SobolevSpace.Hs.witnessFunc_spec (n : ) (s : ) (u : Hs n s) (φ : SchwartzMap (EuclideanSpace (Fin n)) ) :
                            (FourierTransform.fourier u) φ = (ξ : EuclideanSpace (Fin n)), (↑(sobolevWeight n s ξ))⁻¹ * witnessFunc n s u ξ * φ ξ

                            The defining property of Hs.witnessFunc: pairing 𝓕 u against a Schwartz function recovers the integral of ⟨ξ⟩^{-s} · witnessFunc · φ.

                            noncomputable def SobolevSpace.Hs.fromL2 (n : ) (s : ) (f : (MeasureTheory.Lp 2 MeasureTheory.volume)) :
                            Hs n s

                            Given an function f, construct the element of H^s whose witness function is f: namely, 𝓕⁻¹ (⟨ξ⟩^{-s} f) as a tempered distribution.

                            Instances For

                              Right inverse: extracting the witness from Hs.fromL2 s f recovers f. Uses Lebesgue uniqueness of integration against smooth, compactly supported test functions to conclude the underlying functions are equal a.e.

                              theorem SobolevSpace.Hs.witnessL2_fromL2 (n : ) (s : ) (u : Hs n s) :
                              fromL2 n s (witnessL2 n s u) = u

                              Left inverse: starting from a Sobolev distribution u, taking the witness and then applying Hs.fromL2 returns u. Uses Fourier inversion on tempered distributions.

                              noncomputable def SobolevSpace.Hs.equivLp (n : ) (s : ) :

                              The set-level bijection Hs n s ≃ L²(ℝⁿ; ℂ) provided by the witness function construction.

                              Instances For
                                @[reducible]
                                noncomputable instance SobolevSpace.Hs.instNormedAddCommGroup (n : ) (s : ) :

                                Transport the NormedAddCommGroup structure from L²(ℝⁿ; ℂ) to Hs n s along the bijection Hs.equivLp.

                                @[reducible]
                                noncomputable instance SobolevSpace.Hs.instInnerProductSpace (n : ) (s : ) :

                                Transport the complex inner-product-space structure from L²(ℝⁿ; ℂ) to Hs n s along the bijection Hs.equivLp. The inner product on H^s is ⟨u, v⟩_{H^s} := ⟨witnessL2 u, witnessL2 v⟩_{L²}.

                                The Sobolev space H^s is complete: it inherits completeness from L²(ℝⁿ; ℂ) via the isometric bijection Hs.equivLp.

                                theorem SobolevSpace.Hs.norm_witnessL2 (n : ) (s : ) (u : Hs n s) :

                                The norm on H^s is defined to equal the norm of the witness: ‖u‖_{H^s} = ‖witnessL2 u‖_{L²}.

                                theorem SobolevSpace.Hs.witnessL2_add (n : ) (s : ) (u v : Hs n s) :
                                witnessL2 n s (u + v) = witnessL2 n s u + witnessL2 n s v

                                The witness map Hs n s → L² is additive.

                                theorem SobolevSpace.Hs.witnessL2_smul (n : ) (s : ) (c : ) (u : Hs n s) :
                                witnessL2 n s (c u) = c witnessL2 n s u

                                The witness map Hs n s → L² is -linear.

                                The "Fourier-weight" linear isometric equivalence between H^s(ℝⁿ) and L²(ℝⁿ; ℂ) sending each u to its witness ⟨ξ⟩^s · 𝓕u. This is the fundamental Hilbert-space identification underlying the definition of H^s.

                                Instances For

                                  Schwartz functions are dense in H^s: for any linear isometric identification Φ : H^s ≃ₗᵢ L², the composition Φ⁻¹ ∘ toLp of the canonical Schwartz-to- map has dense range.

                                  noncomputable def SobolevSpace.dualIdentification (n : ) (s : ) :

                                  The Sobolev duality identification: H^{-s} ≃ (H^s)' as anti-linear isometric equivalences, obtained by composing the two Fourier-weight isometries with the Riesz representation H^s ≃ (H^s)'. This is part of Proposition 9.8 of Melrose.

                                  Instances For

                                    The duality identification is norm-preserving: ‖dualIdentification s u'‖ = ‖u'‖.

                                    The duality identification is a bijection between H^{-s} and the continuous dual (H^s)'.

                                    Proposition 9.8 of Melrose: for each s ∈ ℝ, the Sobolev space H^s is a Hilbert space (linearly isometric to with Schwartz functions dense), and there is an anti-linear isometric identification of H^{-s} with the continuous dual of H^s.

                                    The complex-valued Sobolev weight ξ ↦ ⟨ξ⟩^s has temperate growth, so multiplication by it acts on Schwartz functions.

                                    Every Schwartz function belongs to H^s for every real s: the witness function is ⟨ξ⟩^s · 𝓕f, which is again Schwartz and in particular in .

                                    Fourier exchange between derivatives and monomial multiplication on Schwartz functions: ∂^α (𝓕 φ) = 𝓕 (x^α · φ).

                                    Combining Fourier inversion with iterSchwartzDeriv_fourier_exchange: 𝓕⁻ (∂^α 𝓕 φ)(ξ) = ξ^α · φ(ξ), evaluated pointwise.

                                    /Fourier pairing identity: for f ∈ L² and a Schwartz function φ, the integral of f · ∂^α(𝓕 φ) equals the integral of 𝓕 f · ξ^α · φ, using the Fourier transform on the right.

                                    The Japanese bracket ⟨ξ⟩ = √(1 + ‖ξ‖²) is at least 1.

                                    Each coordinate |ξ i| is bounded by the Japanese bracket ⟨ξ⟩.

                                    Pointwise norm bound: for any multi-index α of order at most m, the product of the monomial ξ^α and the negative-Sobolev weight ⟨ξ⟩^{-m} has complex norm at most 1. This is the key bound that makes the construction of witnesses for H^{-m} work.

                                    Existence form: for each function f and multi-index α, there exists an function fhat (concretely the Fourier transform of f) such that pairing f with ∂^α (𝓕 ψ) equals pairing ξ^α · fhat with ψ.

                                    theorem SobolevSpace.monomial_sobolev_weight_bound (n m : ) (α : Fin n) ( : α SchwartzRepresentation.multiIndicesBall n m) (ξ : EuclideanSpace (Fin n)) :
                                    (∏ i : Fin n, (ξ.ofLp i) ^ α i) * (sobolevWeight n (-m) ξ) 1

                                    Restatement of monomial_sobolevWeight_norm_le_one using the explicit product ∏ ξ_i^{α_i} instead of monomial α.

                                    If h is bounded by 1 almost everywhere and g ∈ L², then the pointwise product h · g lies in . This is the order in which the measurability argument is most convenient.

                                    theorem SobolevSpace.hasTemperateGrowth_prod' {E : Type u_1} [NormedAddCommGroup E] [NormedSpace E] {R : Type u_2} [NormedCommRing R] [NormedAlgebra R] {ι : Type u_3} [DecidableEq ι] (s : Finset ι) (f : ιER) (hf : is, Function.HasTemperateGrowth (f i)) :
                                    Function.HasTemperateGrowth fun (x : E) => is, f i x

                                    A finite product of functions with temperate growth has temperate growth. The proof is a Finset induction using closure of temperate growth under multiplication.

                                    theorem SobolevSpace.monomial_hasTemperateGrowth (n : ) (α : Fin n) :
                                    Function.HasTemperateGrowth fun (ξ : EuclideanSpace (Fin n)) => i : Fin n, (ξ.ofLp i) ^ α i

                                    The monomial ξ ↦ ∏_i (ξ_i)^{α_i} (as a complex-valued function) has temperate growth.

                                    theorem SobolevSpace.memLp_monomial_mul (n m : ) (α : Fin n) ( : α SchwartzRepresentation.multiIndicesBall n m) (f : EuclideanSpace (Fin n)) (hf : MeasureTheory.MemLp f 2 MeasureTheory.volume) (φ : SchwartzMap (EuclideanSpace (Fin n)) ) :
                                    MeasureTheory.Integrable (fun (ξ : EuclideanSpace (Fin n)) => (∏ i : Fin n, (ξ.ofLp i) ^ α i) * f ξ * φ ξ) MeasureTheory.volume

                                    The integrand ξ^α · f · φ is integrable when f ∈ L² and φ is a Schwartz function (so ξ^α · φ is still Schwartz, hence in ).

                                    One direction of Proposition 9.7 of Melrose: if a tempered distribution u can be written as a finite sum u = ∑_{|α| ≤ m} ∂^α (v_α) with each v_α ∈ L², then u ∈ H^{-m}. The witness function for H^{-m} is built from the Fourier transforms of the v_α.

                                    Polar-form identity: for any complex number z, the product z · (conj(z) / ‖z‖) equals ‖z‖ (with the convention that the formula holds even for z = 0).

                                    Convenience reordering of memLp_mul_of_bound_one_backward: if h is a.e. bounded by 1 (and strongly measurable) and g ∈ L², then h · g is in .

                                    theorem SobolevSpace.sgn_denom_weight_norm_le_one (n m : ) (α : Fin n) (ξ : EuclideanSpace (Fin n)) :
                                    (starRingEnd ) (∏ i : Fin n, (ξ.ofLp i) ^ α i) / i : Fin n, (ξ.ofLp i) ^ α i * (↑(∑ βSchwartzRepresentation.multiIndicesBall n m, i : Fin n, (ξ.ofLp i) ^ β i))⁻¹ * (↑(sobolevWeight n (-m) ξ))⁻¹ 1

                                    Pointwise norm bound for the prefactor used in the explicit Sobolev representation: the product sign(ξ^α) · (∑_β ‖ξ^β‖)⁻¹ · ⟨ξ⟩^m has complex norm at most 1.

                                    Dual /Fourier pairing identity: for f ∈ L² and a Schwartz function φ, the integral of ξ^α · f · 𝓕⁻¹ φ equals the integral of f · ∂^α φ.

                                    theorem SobolevSpace.distrib_identity_from_fourier_witness (n m : ) (u : TemperedDistribution (EuclideanSpace (Fin n)) ) (g : EuclideanSpace (Fin n)) (hg_mem : MeasureTheory.MemLp g 2 MeasureTheory.volume) (hg_eq : ∀ (φ : SchwartzMap (EuclideanSpace (Fin n)) ), (FourierTransform.fourier u) φ = (ξ : EuclideanSpace (Fin n)), (↑(sobolevWeight n (-m) ξ))⁻¹ * g ξ * φ ξ) (v : (Fin n)EuclideanSpace (Fin n)) (hv_mem : αSchwartzRepresentation.multiIndicesBall n m, MeasureTheory.MemLp (v α) 2 MeasureTheory.volume) (hv_sum : ∀ (ξ : EuclideanSpace (Fin n)), αSchwartzRepresentation.multiIndicesBall n m, (∏ i : Fin n, (ξ.ofLp i) ^ α i) * v α ξ = (↑(sobolevWeight n (-m) ξ))⁻¹ * g ξ) (φ : SchwartzMap (EuclideanSpace (Fin n)) ) :

                                    Bridge step in the proof of the Sobolev derivative decomposition: given the Fourier witness g for u ∈ H^{-m} and functions v_α summing (after multiplication by ξ^α) to ⟨ξ⟩^m · g, the distribution u equals ∑_{|α| ≤ m} ∂^α v_α. The proof uses Fourier inversion together with l2_monomial_fourierInv_schwartz_eq.

                                    theorem SobolevSpace.sgn_denom_algebraic_identity (n m : ) (g : EuclideanSpace (Fin n)) (v : (Fin n)EuclideanSpace (Fin n)) (hv_def : ∀ (α : Fin n) (ξ : EuclideanSpace (Fin n)), v α ξ = (starRingEnd ) (∏ i : Fin n, (ξ.ofLp i) ^ α i) / i : Fin n, (ξ.ofLp i) ^ α i * (↑(∑ βSchwartzRepresentation.multiIndicesBall n m, i : Fin n, (ξ.ofLp i) ^ β i))⁻¹ * (↑(sobolevWeight n (-m) ξ))⁻¹ * g ξ) (ξ : EuclideanSpace (Fin n)) :
                                    αSchwartzRepresentation.multiIndicesBall n m, (∏ i : Fin n, (ξ.ofLp i) ^ α i) * v α ξ = (↑(sobolevWeight n (-m) ξ))⁻¹ * g ξ

                                    Pointwise algebraic identity for the explicit v_α used in the Sobolev decomposition: if v_α(ξ) = sign(ξ^α) · (∑_β ‖ξ^β‖)⁻¹ · ⟨ξ⟩^m · g(ξ) for every α of order at most m, then ∑_α ξ^α · v_α(ξ) = ⟨ξ⟩^m · g(ξ).

                                    Converse direction of Proposition 9.7 of Melrose: every distribution u ∈ H^{-m} admits a representation as a finite sum u = ∑_{|α| ≤ m} ∂^α v_α with each v_α ∈ L². The proof constructs an explicit v_α from the Fourier witness using the polar-decomposition identity.

                                    Proposition 9.7 of Melrose: a tempered distribution u belongs to H^{-m} if and only if it can be written as a finite sum u = ∑_{|α| ≤ m} ∂^α v_α with all v_α ∈ L².