Documentation

TauCeti.Analysis.Fourier.L2Multiplier

Fourier multipliers on L² #

Let E be a finite-dimensional real inner product space and F a complex Hilbert space. For an essentially bounded symbol m ∈ L^∞(E), the Fourier multiplier with symbol m is the operator

T_m f = 𝓕⁻ (m • 𝓕 f)

on L²(E, F), where 𝓕 = MeasureTheory.Lp.fourierTransformₗᵢ is the Fourier transform on L². By Plancherel's theorem it is bounded, with ‖T_m‖ ≤ ‖m‖_∞.

Mathlib defines Fourier multipliers on Schwartz functions and on tempered distributions (SchwartzMap.fourierMultiplierCLM, TemperedDistribution.fourierMultiplierCLM), but they are meaningful only for symbols of temperate growth, which are in particular smooth. The symbols of the classical singular integral operators are bounded but not smooth at the origin: the Riesz transforms have symbols -i ξⱼ / ‖ξ‖, and ∂ⱼ∂ₖ Δ⁻¹ has symbol ξⱼ ξₖ / ‖ξ‖². The L² multiplier accepts any essentially bounded symbol, and it is the L²-bounded operator that the Calderón–Zygmund theory and the Mihlin–Hörmander multiplier theorem start from. Where both are defined, it agrees with Mathlib's multipliers.

Main declarations #

Kernels of multipliers #

A bounded symbol m is in general not the Fourier transform of an integrable function, so T_m is not a convolution with an integrable kernel. The Calderón–Zygmund theory instead asks for a kernel representation off the support of the argument, and the proof of the Mihlin–Hörmander multiplier theorem obtains it by approximation: m is cut into Schwartz symbols σ, for which T_σ is convolution with 𝓕⁻ σ almost everywhere, the convolution integral converging at every point, and T_m is recovered as the strong limit of the operators T_σ.

References #

The Fourier multiplier on L²(E, F) with symbol m ∈ L^∞(E): the bounded operator f ↦ 𝓕⁻ (m • 𝓕 f), where 𝓕 is the Fourier transform on L².

Instances For
    @[simp]

    On the Fourier side, the multiplier with symbol m is multiplication by m.

    A bounded operator on L² is the multiplier with symbol m if and only if it acts as multiplication by m on the Fourier side.

    The multiplier with symbol m changes the L² norm by at most the factor ‖m‖_∞.

    Plancherel bound for Fourier multipliers: ‖T_m‖ ≤ ‖m‖_∞.

    Dependence on the symbol #

    The multiplier of a finite sum of symbols is the sum of the multipliers.

    A multiplier whose symbol is almost everywhere the constant c is multiplication by c.

    The composition of two multipliers is the multiplier with the product symbol.

    The adjoint of the multiplier with symbol m is the multiplier with the complex conjugate symbol star m. In particular a real symbol gives a self-adjoint multiplier.

    The pairing of g with the multiplier T_m f, computed on the Fourier side: ⟪g, T_m f⟫ = ∫ m ξ ⟪𝓕 g ξ, 𝓕 f ξ⟫ dξ.

    Comparison with multipliers on Schwartz functions and tempered distributions #

    A Schwartz function g whose Fourier transform is m • 𝓕 f is, in L², the multiplier with symbol m applied to f. Here m need not have temperate growth: any bounded symbol works, so this also covers symbols that are discontinuous at the origin.

    For a bounded symbol g of temperate growth, the L² multiplier agrees with the multiplier TemperedDistribution.fourierMultiplierCLM F g on tempered distributions: for every f ∈ L², both send f to the same tempered distribution.

    Schwartz symbols and convolution kernels #

    The L² multiplier with a Schwartz symbol σ is convolution with the Schwartz kernel 𝓕⁻ σ: for every f ∈ L², T_σ f x = ∫ 𝓕⁻ σ (x - y) • f y for almost every x. The integral converges for every x, since y ↦ 𝓕⁻ σ (x - y) lies in L².

    With k x y = 𝓕⁻ σ (x - y) • id, this is the kernel representation of T_σ that the Calderón–Zygmund theorems take as hypothesis, valid at almost every point and not only off the support of f.

    theorem MeasureTheory.Lp.tendsto_fourierMultiplierCLM_apply {E : Type u_1} {F : Type u_2} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E] [NormedAddCommGroup F] [InnerProductSpace ℂ F] [CompleteSpace F] {ι : Type u_3} {l : Filter ι} [l.IsCountablyGenerated] {m : ι → ↥(Lp ℂ ⊤ volume)} {m₀ : ↥(Lp ℂ ⊤ volume)} {C : ℝ} (hC : ∀ᶠ (i : ι) in l, ‖m i‖ ≤ C) (hm : ∀ᵐ (ξ : E), Filter.Tendsto (fun (i : ι) => ↑↑(m i) ξ) l (nhds (↑↑m₀ ξ))) (f : ↥(Lp F 2 volume)) :
    Filter.Tendsto (fun (i : ι) => (fourierMultiplierCLM F (m i)) f) l (nhds ((fourierMultiplierCLM F m₀) f))

    If the symbols m i are eventually bounded in L^∞ uniformly in i along a countably generated filter l and converge to m₀ almost everywhere along l, then T_{m i} f → T_{m₀} f in L² along l for every f ∈ L².