Fourier Transform on R^n and the Plancherel Theorem

Intuition Beginner

A Fourier series writes a periodic function as a sum of pure sine and cosine waves at integer-multiple frequencies. The Fourier transform asks the same question for a non-periodic function: which pure frequencies do we need, and at what strengths, to reconstruct it? The catch is that a non-periodic function generally needs a continuous range of frequencies, not just integers. The discrete sum of harmonics becomes a continuous integral over the frequency axis, and the list of Fourier coefficients becomes a smooth function of frequency.

The physical motivation is a spectrogram. When a microphone records a voice, the waveform in time looks like a complicated wiggle. Run that waveform through the Fourier transform and a different picture appears: a smooth function of frequency that tells you the strength of each pure tone present in the voice. Vowels show clear peaks at certain frequencies called formants, while consonants spread their energy more broadly. The same picture works for any signal: a photograph has a Fourier transform that separates smooth gradients from sharp edges, and a radio signal has one that separates the carrier from the modulation.

The same engine solves linear differential equations. The heat equation is hard to solve directly, but each pure sine wave is an eigenfunction: it just decays exponentially in time, with high frequencies dying faster than low ones. The trick is to decompose the initial temperature into pure waves using the Fourier transform, solve each wave separately, and add the answers back together using the inverse Fourier transform. The same trick works for the wave equation, the Schrödinger equation of quantum mechanics, and Laplace's equation in electrostatics — each becomes a one-line solution in the frequency domain.

A central fact is that the Fourier transform conserves energy. The total energy of the signal, measured as the integral of its square over time, equals the total energy of its frequency content, measured as the integral of the squared transform over frequency. This is Plancherel's theorem. It says the Fourier transform is a perfect dictionary: nothing is lost when switching between the time-domain picture and the frequency-domain picture, and the round trip from time to frequency and back recovers the original signal exactly.

The one-sentence takeaway: the Fourier transform decomposes a non-periodic signal into a continuous spectrum of pure frequencies, the Plancherel theorem says the transform preserves total energy, and the inverse Fourier transform rebuilds the original signal from its spectrum.

Visual Beginner

Picture a Gaussian bump centered at the origin, a smooth bell-shaped curve. Now picture its Fourier transform: another Gaussian bump, centered at frequency zero. A wide bell in time corresponds to a narrow bell in frequency, and a narrow bell in time corresponds to a wide bell in frequency. Sharp localization in one domain forces spreading in the other. This is the picture behind the Heisenberg uncertainty principle.

Now picture a rectangular pulse: a function equal to one on a short interval and zero outside. Its Fourier transform is the sinc function, a smooth wave that oscillates and decays slowly as the frequency increases. The sharp edges of the rectangle generate slowly decaying high-frequency tails, while the smooth Gaussian generates rapidly decaying tails. The smoother the function, the faster its transform decays at infinity.

Six pictures showing the time-frequency duality: narrow becomes wide, smooth becomes rapidly decaying, sharp becomes slowly oscillating — that is the visual content of the Fourier transform.

Worked example Beginner

We compute the Fourier transform of the Gaussian function $f(x)=e^{-x^2/2}$ on the real line and observe its self-transforming property.

Step 1. The Fourier transform of a function $f$ on the real line is defined as the integral of $f(x)$ against the pure-frequency wave $e^{-i2\pi \xi x}$ over all of the real line. Different conventions place the factor of two pi in different positions, but the structural content is the same. For our calculation we use the convention without the factor of two pi: the Fourier transform of $f$ at frequency $\xi$ is the integral of $f(x)e^{-i\xi x}dx$ from minus infinity to plus infinity.

Step 2. Substituting $f(x)=e^{-x^2/2}$, we need to evaluate the integral of $e^{-x^2/2-i\xi x}$ over the real line. Complete the square in the exponent: $-x^2/2-i\xi x=-(x+i\xi {)}^2/2-{\xi }^2/2$. The first term is a shifted parabolic exponential; the second term is a constant in $x$ that can be pulled out of the integral.

Step 3. The remaining integral is $e^{-{\xi }^2/2}$ times the integral of $e^{-(x+i\xi {)}^2/2}$ over the real line. By contour shifting (the integrand is entire and decays as the real part of $x$ goes to plus or minus infinity), the integral over the shifted contour equals the integral over the real axis, which is the standard Gaussian integral equal to the square root of two pi.

Step 4. Putting it together, the Fourier transform of the Gaussian $f(x)=e^{-x^2/2}$ is $\sqrt{2\pi }\cdot e^{-{\xi }^2/2}$. Up to the multiplicative constant, the Gaussian is its own Fourier transform: the transform of a Gaussian is another Gaussian with the same shape. This self-transforming property is unique to the Gaussian (modulo dilation and translation) and is the source of many special properties — Gaussians are the unique minimizers of the Heisenberg uncertainty principle, the heat-equation kernel, and the ground-state wavefunction of the quantum harmonic oscillator.

Step 5. As a parallel example, the rectangular pulse $g(x)=1$ on the interval from minus one half to plus one half and zero outside has Fourier transform equal to the sinc function, $\sin (\xi /2)/(\xi /2)$, evaluated at the angular frequency $\xi$. The sinc function has slowly decaying oscillating tails, in contrast to the Gaussian's rapid decay. The sharp edges of the rectangle produce the slow decay; the smoothness of the Gaussian produces the fast decay. This decay-versus-smoothness duality is a recurring theme throughout the theory.

Check your understanding Beginner

Formal definition Intermediate+

Let $f\in L^1({\mathbb{R}}^n)$, the space of Lebesgue-integrable functions on ${\mathbb{R}}^n$ [from 02.07.06]. The Fourier transform of $f$ is the function $\hat{f}:{\mathbb{R}}^n\to \mathbb{C}$ defined by $$\hat{f}(\xi )={\int }_{{\mathbb{R}}^n}f(x)e^{-2\pi i\xi \cdot x}dx,\xi \in {\mathbb{R}}^n,$$ where $\xi \cdot x={\sum }_{j=1}^n{\xi }_j{x}_j$ is the Euclidean inner product on ${\mathbb{R}}^n$. The integral converges absolutely because $|f(x)e^{-2\pi i\xi \cdot x}|=|f(x)|$ is integrable. The $\pi$-convention used here makes Parseval's identity $\int |\hat{f}{|}^2=\int |f{|}^2$ hold with no $2\pi$ factor; alternative conventions place factors of $2\pi$ elsewhere with corresponding adjustments. We will note the convention-translation at key formulas.

The map $\mathcal{F}:L^1({\mathbb{R}}^n)\to L^{\infty }({\mathbb{R}}^n)$, $f\mapsto \hat{f}$, is linear and bounded with operator norm $\Vert \mathcal{F}\Vert \le 1$: for every $\xi$, $$|\hat{f}(\xi )|\le {\int }_{{\mathbb{R}}^n}|f(x)|dx=\Vert f{\Vert }_{L^1}.$$ Hence $\Vert \hat{f}{\Vert }_{L^{\infty }}\le \Vert f{\Vert }_{L^1}$.

Basic identities. For $f,g\in L^1({\mathbb{R}}^n)$ and constants $a\in {\mathbb{R}}^n$, $\lambda >0$, $\alpha ,\beta \in \mathbb{C}$:

• Linearity. $\widehat{(\alpha f+\beta g)}(\xi )=\alpha \hat{f}(\xi )+\beta \hat{g}(\xi )$.
• Translation. Let ${\tau }_{a}f(x)=f(x-a)$. Then $\widehat{{\tau }_{a}f}(\xi )=e^{-2\pi i\xi \cdot a}\hat{f}(\xi )$.
• Modulation. Let $M_{a}f(x)=e^{2\pi ia\cdot x}f(x)$. Then $\widehat{M_{a}f}(\xi )=\hat{f}(\xi -a)$.
• Dilation. Let $D_{\lambda }f(x)=f(\lambda x)$. Then $\widehat{D_{\lambda }f}(\xi )={\lambda }^{-n}\hat{f}({\lambda }^{-1}\xi )$.
• Conjugation. $\hat{\bar{f}}(\xi )=\overline{\hat{f}(-\xi )}$.

These identities follow from substitution in the defining integral.

Definition (Riemann-Lebesgue, on ${\mathbb{R}}^n$). The Fourier transform of an $L^1$-function vanishes at infinity: for $f\in L^1({\mathbb{R}}^n)$, $\hat{f}(\xi )\to 0$ as $|\xi |\to \infty$.

Definition (convolution). For $f,g\in L^1({\mathbb{R}}^n)$, the convolution $f\ast g$ is defined almost everywhere by $$(f\ast g)(x)={\int }_{{\mathbb{R}}^n}f(y)g(x-y)dy.$$ By Fubini-Tonelli [from 02.07.07] applied to the function $|f(y)g(x-y)|$ on ${\mathbb{R}}^n\times {\mathbb{R}}^n$, the convolution is well-defined for almost every $x$ and lies in $L^1$, with $\Vert f\ast g{\Vert }_{L^1}\le \Vert f{\Vert }_{L^1}\Vert g{\Vert }_{L^1}$.

Definition (Gaussian on ${\mathbb{R}}^n$). The Gaussian $G(x)=e^{-\pi |x{|}^2}$ with normalization ${\int }_{{\mathbb{R}}^n}G(x)dx=1$ satisfies the self-transforming property $\hat{G}(\xi )=G(\xi )$: the Gaussian is its own Fourier transform under the $\pi$-convention.

Definition (Schwartz space). The Schwartz space $\mathcal{S}({\mathbb{R}}^n)$ is the set of smooth functions $f:{\mathbb{R}}^n\to \mathbb{C}$ such that $$\Vert f{\Vert }_{\alpha ,\beta }:=\underset{x\in {\mathbb{R}}^n}{\sup }|x^{\alpha }({\partial }^{\beta }f)(x)|<\infty$$ for every pair of multi-indices $\alpha ,\beta \in {\mathbb{N}}^n$. The seminorms $\Vert \cdot {\Vert }_{\alpha ,\beta }$ make $\mathcal{S}({\mathbb{R}}^n)$ into a Fréchet space (complete metrizable locally convex topological vector space). Equivalently, $f\in \mathcal{S}({\mathbb{R}}^n)$ iff $f$ and all its derivatives decay faster than any polynomial at infinity. Examples: the Gaussian $e^{-\pi |x{|}^2}$, Hermite functions $H_n(x)e^{-x^2/2}$, smooth compactly supported functions $C_c^{\infty }({\mathbb{R}}^n)$.

Definition ($L^2$-Fourier transform via Plancherel). The Fourier transform $\mathcal{F}:L^1\cap L^2\to L^2$ defined by the integral formula extends uniquely to a continuous linear operator $\mathcal{F}:L^2({\mathbb{R}}^n)\to L^2({\mathbb{R}}^n)$ that is a unitary isomorphism. The extension is the content of the Plancherel theorem (Theorem 4 below).

Counterexamples to common slips Intermediate+

• $\hat{f}$ need not be in $L^1$. For $f\in L^1({\mathbb{R}}^n)$, the Fourier transform $\hat{f}$ is bounded and continuous and vanishes at infinity, but need not be in $L^1$. The rectangular pulse $f={\chi }_{[-1/2,1/2]}$ has Fourier transform $\hat{f}(\xi )=\sin (\pi \xi )/(\pi \xi )$ (sinc function), which decays like $1/|\xi |$ and is not in $L^1(\mathbb{R})$. Consequently, the Fourier inversion integral $\int \hat{f}(\xi )e^{2\pi i\xi x}d\xi$ does not converge absolutely; pointwise inversion requires additional hypotheses on $f$ or interpretation via summability methods (Cesàro, Abel) or in the principal-value sense.

• Pointwise inversion can fail. If $f\in L^1(\mathbb{R})$ but $\hat{f}\notin L^1(\mathbb{R})$, the Fourier inversion integral does not converge in the Lebesgue sense and may diverge pointwise. The cleanest framework where inversion holds pointwise is the Schwartz space $\mathcal{S}({\mathbb{R}}^n)$, where both $f$ and $\hat{f}$ are Schwartz and absolutely integrable; on $L^2$, inversion holds in the $L^2$-norm but not necessarily pointwise.

• Plancherel holds on $L^2$, not on $L^1$. The identity $\Vert \hat{f}{\Vert }_{L^2}=\Vert f{\Vert }_{L^2}$ is the central Plancherel statement on $L^2({\mathbb{R}}^n)$, but no analogous identity holds on $L^1$: the Fourier transform of an $L^1$-function is in $L^{\infty }$ (with $\Vert \hat{f}{\Vert }_{L^{\infty }}\le \Vert f{\Vert }_{L^1}$), and the only $L^p$-norm identity available across $L^1$ and $L^{\infty }$ is the strict inequality $\Vert \hat{f}{\Vert }_{L^{\infty }}\le \Vert f{\Vert }_{L^1}$.

• The Hausdorff-Young inequality fails for $p>2$. The inequality $\Vert \hat{f}{\Vert }_{L^{p^{\prime }}}\le \Vert f{\Vert }_{L^p}$ holds for $p\in [1,2]$ with conjugate exponent $p^{\prime }$, where $1/p+1/p^{\prime }=1$. For $p>2$, no $L^p\to L^q$ boundedness of the Fourier transform holds: the Fourier transform of a generic $L^p$-function with $p>2$ need not be a function at all (it is a tempered distribution), and the Hausdorff-Young inequality fails.

• Convolution and Fourier transform exchange. The identity $\widehat{f\ast g}=\hat{f}\cdot \hat{g}$ (Fourier transform of a convolution is the product of Fourier transforms) holds for $f,g\in L^1({\mathbb{R}}^n)$. The converse identity $\widehat{f\cdot g}=\hat{f}\ast \hat{g}$ (Fourier transform of a product is the convolution of Fourier transforms) requires more care: $f\cdot g$ need not be integrable even when $f,g\in L^1$, and the right-hand convolution may not converge. The cleanest formulation is on the Schwartz space, where both identities hold without integrability concerns.

Key theorem with proof Intermediate+

Theorem (Plancherel; Plancherel 1910 Rend. Circ. Mat. Palermo 30, 289). The Fourier transform, initially defined on $L^1\cap L^2({\mathbb{R}}^n)$ by the integral formula $$\hat{f}(\xi )={\int }_{{\mathbb{R}}^n}f(x)e^{-2\pi i\xi \cdot x}dx,$$ satisfies the Plancherel identity $\Vert \hat{f}{\Vert }_{L^2}=\Vert f{\Vert }_{L^2}$ for $f\in L^1\cap L^2$. Consequently $\mathcal{F}$ extends uniquely by continuity to a bounded linear operator $\mathcal{F}:L^2({\mathbb{R}}^n)\to L^2({\mathbb{R}}^n)$, and this extension is a unitary isomorphism: $\mathcal{F}$ is norm-preserving on $L^2$, and its inverse is the inverse Fourier transform ${\mathcal{F}}^{-1}g(x)=\int g(\xi )e^{2\pi i\xi \cdot x}d\xi$ (extended to $L^2$ by the same density argument).

Proof. The proof proceeds via the Schwartz space $\mathcal{S}({\mathbb{R}}^n)$ as a dense subspace of $L^2$.

Step 1 (Fourier inversion on Schwartz space). We first establish: for every $f\in \mathcal{S}({\mathbb{R}}^n)$, $\hat{f}\in \mathcal{S}({\mathbb{R}}^n)$, and the inversion formula $$f(x)={\int }_{{\mathbb{R}}^n}\hat{f}(\xi )e^{2\pi i\xi \cdot x}d\xi$$ holds pointwise for every $x\in {\mathbb{R}}^n$.

That $\hat{f}\in \mathcal{S}({\mathbb{R}}^n)$ when $f\in \mathcal{S}({\mathbb{R}}^n)$: differentiation under the integral sign (justified by the Schwartz-decay of $f$ and its derivatives) gives ${\partial }^{\alpha }\hat{f}(\xi )=\widehat{(-2\pi ix{)}^{\alpha }f}(\xi )$, and integration by parts gives $\widehat{{\partial }^{\beta }f}(\xi )=(2\pi i\xi {)}^{\beta }\hat{f}(\xi )$. Combining, ${\xi }^{\alpha }{\partial }^{\beta }\hat{f}(\xi )$ is the Fourier transform of a Schwartz function, hence bounded, so $\hat{f}$ satisfies the Schwartz seminorm bounds.

For the inversion formula, apply the Fourier transform to $\hat{f}$ and use the Gaussian approximation. Let $G_t(x)=t^{-n}{e}^{-\pi |x{|}^2/t^2}$ be the family of Gaussians with ${\hat{G}}_t(\xi )=e^{-\pi t^2|\xi {|}^2}$ (using the self-transforming property of the Gaussian and the dilation identity). The family $\{G_t{\}}_{t>0}$ is an approximate identity: $\int G_t=1$ for all $t$, and $G_t\to {\delta }_0$ as $t\to 0^{+}$ in the distributional sense.

For $f\in \mathcal{S}$, the integral $\int \hat{f}(\xi )e^{2\pi i\xi \cdot x}d\xi$ may not converge absolutely, but its Gaussian-regularized version does: $$I_t(x)={\int }_{{\mathbb{R}}^n}\hat{f}(\xi )e^{2\pi i\xi \cdot x}{e}^{-\pi t^2|\xi {|}^2}d\xi .$$

Substituting $\hat{f}(\xi )=\int f(y)e^{-2\pi i\xi \cdot y}dy$ and applying Fubini (legal because the double integrand is in $L^1({\mathbb{R}}^n\times {\mathbb{R}}^n)$ thanks to the Gaussian factor): $$I_t(x)={\int }_{{\mathbb{R}}^n}f(y)\left({\int }_{{\mathbb{R}}^n}{e}^{-\pi t^2|\xi {|}^2}{e}^{2\pi i\xi \cdot (x-y)}d\xi \right)dy={\int }_{{\mathbb{R}}^n}f(y)G_t(x-y)dy=(f\ast G_t)(x),$$ using the Fourier transform of the Gaussian $\int e^{-\pi t^2|\xi {|}^2}{e}^{2\pi i\xi \cdot z}d\xi =G_t(z)$.

As $t\to 0^{+}$, the approximate-identity property gives $(f\ast G_t)(x)\to f(x)$ pointwise for every $x$ (using continuity of $f\in \mathcal{S}$). Also, dominated convergence gives $I_t(x)\to \int \hat{f}(\xi )e^{2\pi i\xi \cdot x}d\xi$ pointwise (since $\hat{f}\in \mathcal{S}$ is Schwartz, hence in $L^1$, and the Gaussian factor $e^{-\pi t^2|\xi {|}^2}$ is bounded by $1$ uniformly and tends to $1$ pointwise). Equating limits: $$f(x)={\int }_{{\mathbb{R}}^n}\hat{f}(\xi )e^{2\pi i\xi \cdot x}d\xi .$$ This is Fourier inversion on $\mathcal{S}$.

Step 2 (Plancherel identity on Schwartz space). For $f,g\in \mathcal{S}({\mathbb{R}}^n)$, we show $⟨f,g{⟩}_{L^2}=⟨\hat{f},\hat{g}{⟩}_{L^2}$, where $⟨f,g⟩=\int f\bar{g}dx$.

Compute, using $\bar{g}(x)=\int \overline{\hat{g}(\xi )}{e}^{-2\pi i\xi \cdot x}d\xi$ (the conjugate of the Fourier inversion formula for $g$): $$⟨f,g⟩=\int f(x)\overline{g(x)}dx=\int f(x)\left(\int \overline{\hat{g}(\xi )}{e}^{-2\pi i\xi \cdot x}d\xi \right)dx.$$ By Fubini (both Schwartz, so both integrals converge absolutely): $$⟨f,g⟩=\int \overline{\hat{g}(\xi )}\left(\int f(x)e^{-2\pi i\xi \cdot x}dx\right)d\xi =\int \hat{f}(\xi )\overline{\hat{g}(\xi )}d\xi =⟨\hat{f},\hat{g}⟩.$$

Taking $g=f$ gives the Plancherel identity $\Vert f{\Vert }_{L^2}^2=\Vert \hat{f}{\Vert }_{L^2}^2$ on $\mathcal{S}$.

Step 3 (extension by density to $L^2$). The Schwartz space $\mathcal{S}({\mathbb{R}}^n)$ is dense in $L^2({\mathbb{R}}^n)$ (a consequence of density of $C_c^{\infty }\subset \mathcal{S}$ and density of $C_c^{\infty }$ in $L^2$, both standard). Given $f\in L^2({\mathbb{R}}^n)$, choose a Schwartz sequence $f_k\to f$ in $L^2$-norm. The Plancherel identity on $\mathcal{S}$ gives $\Vert {\hat{f}}_k-{\hat{f}}_j{\Vert }_{L^2}=\Vert f_k-f_j{\Vert }_{L^2}\to 0$, so $\{{\hat{f}}_k\}$ is Cauchy in $L^2$. By Riesz-Fischer completeness [from 02.07.06], the sequence has a limit, which we denote $\hat{f}\in L^2$. The limit is independent of the choice of approximating sequence (by another Cauchy argument), so $\hat{f}$ is well-defined.

By continuity, the Plancherel identity passes to the limit: $\Vert \hat{f}{\Vert }_{L^2}=\lim \Vert {\hat{f}}_k{\Vert }_{L^2}=\lim \Vert f_k{\Vert }_{L^2}=\Vert f{\Vert }_{L^2}$. So the extended operator $\mathcal{F}:L^2\to L^2$ is an isometry.

Step 4 (surjectivity and inverse). The inverse Fourier transform ${\mathcal{F}}^{-1}$, defined analogously by ${\mathcal{F}}^{-1}g(x)=\int g(\xi )e^{2\pi i\xi \cdot x}d\xi$ on $\mathcal{S}$ and extended by density to $L^2$, satisfies ${\mathcal{F}}^{-1}\mathcal{F}=\mathcal{F}{\mathcal{F}}^{-1}=\mathrm{id}$ on $\mathcal{S}$ by the inversion formula from Step 1. By density, the same identities hold on $L^2$. So $\mathcal{F}$ is an $L^2$-isometry with two-sided inverse ${\mathcal{F}}^{-1}$, hence a unitary isomorphism of $L^2({\mathbb{R}}^n)$.

$\square$

Bridge. The Plancherel theorem is the continuous-frequency analogue of the Parseval identity for Fourier series [from 02.10.01]. Both say the spectral decomposition is an isometry: discrete-frequency-amplitude space (${\ell }^2(\mathbb{Z})$) for the circle, continuous-frequency-density space ($L^2({\mathbb{R}}^n)$) for the line. The structural mechanism is the same in both cases: a dense subspace ($C^{\infty }(\mathbb{T})$ or $\mathcal{S}({\mathbb{R}}^n)$) on which the Fourier transform is well-behaved and the relevant identity is verifiable by direct calculation; the identity extends by continuity to the full $L^2$-completion using Riesz-Fischer. The Plancherel framework unifies the discrete and continuous spectral decompositions under the umbrella of Pontryagin duality on locally compact abelian groups: the Fourier transform on a group $G$ is always an $L^2$-isometry into $L^2$ of the Pontryagin dual $\hat{G}$, with $\mathbb{T}$ paired against $\mathbb{Z}$, $\mathbb{R}$ paired against $\mathbb{R}$, and finite cyclic groups paired against themselves. The Plancherel identity is the same theorem in each case, with the same proof template — density of a regular subspace plus continuity-extension via completeness.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: partial — Mathlib provides the central Fourier-transform infrastructure for ${\mathbb{R}}^n$. Real.fourierIntegral is the integral $\int f(x)e^{-2\pi i\xi x}dx$ for $f\in L^1$, Real.fourierInversion is the inverse-Fourier-transform integral, and Real.zero_at_infty_fourierIntegral is the Riemann-Lebesgue lemma stating $\hat{f}\in C_0(\mathbb{R})$ for $f\in L^1$. The Schwartz space is SchwartzMap, with the Fréchet-space structure as SchwartzMap.topology and the seminorms SchwartzMap.seminorm. The Fourier transform on Schwartz space is SchwartzMap.fourierTransformCLM as a continuous linear self-map. The Fourier-inversion theorem on Schwartz space is SchwartzMap.fourierInversion. The Plancherel theorem is MeasureTheory.fourier_isometryL2, with the unitary $L^2$-extension Real.fourierIntegralL2. The convolution theorem Real.fourierIntegral_convolution_eq_mul and the differentiation-multiplication duality Real.fourierIntegral_iteratedDeriv_eq_mul_pow are in Mathlib.Analysis.Fourier.FourierTransform. The Poisson summation formula is Real.tsum_eq_tsum_fourierIntegral, valid for Schwartz functions. The Heisenberg uncertainty principle is MeasureTheory.heisenberg_uncertainty. Tempered distributions are SchwartzMap.Dual, but the Fourier transform on tempered distributions as a continuous linear self-map of SchwartzMap.Dual is in the contributor branch of Mathlib (as of 2026, not in main). The Hausdorff-Young inequality Real.fourierIntegral_lp_le is partial: the $L^1\to L^{\infty }$ and $L^2\to L^2$ endpoints are in Mathlib via Plancherel, but the Riesz-Thorin interpolation to the intermediate $L^p\to L^{p^{\prime }}$ inequalities is not yet formalized. The Paley-Wiener support theorem is not in Mathlib as of 2026.

import Mathlib.Analysis.Fourier.FourierTransform
import Mathlib.Analysis.Distribution.SchwartzSpace
import Mathlib.MeasureTheory.Function.LpSpace

open MeasureTheory Real Complex SchwartzMap

variable (f : ℝ → ℂ)

-- Fourier transform of an L^1 function
noncomputable def fourierIntegral (f : ℝ → ℂ) (ξ : ℝ) : ℂ :=
  ∫ x : ℝ, f x * Complex.exp (-2 * π * I * ξ * x)

-- Riemann-Lebesgue lemma: Fourier integral of L^1 vanishes at infinity
theorem riemann_lebesgue_R (f : ℝ → ℂ) (hf : Integrable f) :
    Filter.Tendsto (fourierIntegral f) (Filter.cocompact ℝ) (nhds 0) := by
  sorry  -- proof via density of Schwartz functions + direct calculation

-- Plancherel identity for L^2 functions
theorem plancherel_identity (f : ℝ → ℂ) (hf : MemLp f 2) :
    ∫ ξ : ℝ, ‖fourierIntegral f ξ‖^2 = ∫ x : ℝ, ‖f x‖^2 := by
  sorry  -- proof via density of Schwartz + Plancherel on Schwartz

-- Convolution theorem
theorem fourierIntegral_convolution (f g : ℝ → ℂ) (hf : Integrable f) (hg : Integrable g) :
    fourierIntegral (f ⋆ g) = fun ξ => fourierIntegral f ξ * fourierIntegral g ξ := by
  sorry  -- proof via Fubini

-- Heisenberg uncertainty principle
theorem heisenberg_uncertainty (f : ℝ → ℂ) (hf : MemLp f 2) (h_norm : ∫ x, ‖f x‖^2 = 1) :
    Real.sqrt (∫ x : ℝ, x^2 * ‖f x‖^2) *
    Real.sqrt (∫ ξ : ℝ, ξ^2 * ‖fourierIntegral f ξ‖^2) ≥ 1 / (4 * π) := by
  sorry  -- proof via Cauchy-Schwarz + integration by parts

Advanced results Master

The advanced theory of the Fourier transform on ${\mathbb{R}}^n$ splits across nine strands: the Schwartz-space framework with its Fréchet-space topology, the full Fourier inversion theorem on Schwartz, the Plancherel theorem as a $L^2$-unitary, the Hausdorff-Young inequality interpolating between $L^1\to L^{\infty }$ and $L^2\to L^2$ endpoints, the extension to tempered distributions, the convolution theorem and its applications to constant-coefficient linear PDEs (heat, wave, Schrödinger), the Poisson summation formula, the Heisenberg uncertainty principle in its sharp form, and the Paley-Wiener-Schwartz support theorem relating compact support to entire-function exponential type.

Theorem 1 (Riemann-Lebesgue on ${\mathbb{R}}^n$). Let $f\in L^1({\mathbb{R}}^n)$. Then $\hat{f}$ is continuous and bounded on ${\mathbb{R}}^n$, and $\hat{f}(\xi )\to 0$ as $|\xi |\to \infty$.

The proof mirrors the Fourier-series version (Theorem 1 of 02.10.01): density of step functions in $L^1({\mathbb{R}}^n)$ plus direct computation of the Fourier transform on indicators of rectangles. The continuity of $\hat{f}$ follows from dominated convergence: $|f(x)e^{-2\pi i\xi \cdot x}|$ is bounded by $|f(x)|\in L^1$ uniformly in $\xi$, so $\xi \mapsto \hat{f}(\xi )$ is continuous by DCT. The vanishing at infinity is the structural Riemann-Lebesgue statement ^{[Riemann 1854; Lebesgue 1903]} — Fourier-transform values inherit the cancellation behaviour from simple test functions at high frequency. The image $\mathcal{F}(L^1({\mathbb{R}}^n))$ is a strict subset of $C_0({\mathbb{R}}^n)$, the space of continuous functions vanishing at infinity (a deep fact: not every $g\in C_0$ is the Fourier transform of an $L^1$-function).

Theorem 2 (Schwartz space as a Fréchet space). The Schwartz space $\mathcal{S}({\mathbb{R}}^n)$ with the seminorm family $\{\Vert \cdot {\Vert }_{\alpha ,\beta }\}$ is a Fréchet space (complete metrizable locally convex topological vector space). It is dense in every $L^p({\mathbb{R}}^n)$ for $p\in [1,\infty )$, and dense in $C_0({\mathbb{R}}^n)$.

Schwartz 1950 ^{[Schwartz 1950]} introduced $\mathcal{S}({\mathbb{R}}^n)$ in his foundational Théorie des distributions as the natural test-function space for the Fourier transform. The Fréchet structure: the countable family $\{\Vert \cdot {\Vert }_{N,M}{\}}_{N,M\ge 0}$ defined by $\Vert f{\Vert }_{N,M}={\sup }_x(1+|x|{)}^N{\sum }_{|\beta |\le M}|{\partial }^{\beta }f(x)|$ generates the topology, equivalent to the multi-index formulation. Completeness follows from completeness of $C^{\infty }({\mathbb{R}}^n)$ in each polynomial-weighted $C^M$-norm. Density in $L^p$ follows from $C_c^{\infty }\subset \mathcal{S}$ and density of $C_c^{\infty }$ in $L^p$.

Theorem 3 (Fourier inversion on Schwartz space). The Fourier transform $\mathcal{F}:\mathcal{S}({\mathbb{R}}^n)\to \mathcal{S}({\mathbb{R}}^n)$ is a continuous linear bijection with inverse ${\mathcal{F}}^{-1}f(x)=\int \hat{f}(\xi )e^{2\pi i\xi \cdot x}d\xi$. Both $\mathcal{F}$ and ${\mathcal{F}}^{-1}$ are continuous in the Fréchet topology.

The inversion formula on $\mathcal{S}$ is the heart of the Fourier-analytic edifice. Proof via Gaussian regularization (as in the Key Theorem proof, Step 1): approximate the inverse-Fourier integral by a Gaussian-damped version, apply Fubini, recognize the result as a convolution with the heat kernel $G_t$, and pass to the limit $t\to 0^{+}$ using the approximate-identity property of $\{G_t\}$. The continuity of $\mathcal{F}$ on $\mathcal{S}$ follows from the seminorm bound $\Vert \hat{f}{\Vert }_{N,M}\le C_{N,M}\Vert f{\Vert }_{M+n+1,N}$ (proved by integration-by-parts in the integral defining $\hat{f}$, with the polynomial-decay-of-$x^{\alpha }f$ traded against the high-derivative-bound on $f$).

Theorem 4 (Plancherel theorem; Plancherel 1910 Rend. Circ. Mat. Palermo 30, 289). The Fourier transform extends from $L^1\cap L^2({\mathbb{R}}^n)$ to a unitary isomorphism $\mathcal{F}:L^2({\mathbb{R}}^n)\to L^2({\mathbb{R}}^n)$ with $\Vert \hat{f}{\Vert }_{L^2}=\Vert f{\Vert }_{L^2}$ for all $f\in L^2$.

Plancherel's 1910 paper ^{[Plancherel 1910]} extended the Parseval identity for Fourier series to the Fourier-transform setting, establishing the $L^2$-isometry that bears his name. The original proof used a direct truncation argument (approximate $f$ by its restrictions to bounded intervals and pass to the limit); the modern proof via the Schwartz-space density (Key Theorem) is cleaner and more conceptual. The Plancherel theorem is the structural origin of all $L^2$-based Fourier analysis: the Fourier transform diagonalizes the translation operators ${\tau }_a$ in the $L^2$-spectral sense (with each translation becoming multiplication by $e^{-2\pi i\xi \cdot a}$ in the Fourier picture), and the simultaneous diagonalization of the abelian translation group is the spectral decomposition of $L^2({\mathbb{R}}^n)$ as a representation of ${\mathbb{R}}^n$.

Theorem 5 (Hausdorff-Young inequality; Hausdorff 1923 Math. Z. 16, 163; Young 1912 Proc. Roy. Soc. A 87, 331). For $p\in [1,2]$ with conjugate exponent $p^{\prime }$ ($1/p+1/p^{\prime }=1$), the Fourier transform extends to a bounded linear map $\mathcal{F}:L^p({\mathbb{R}}^n)\to L^{p^{\prime }}({\mathbb{R}}^n)$ with $\Vert \hat{f}{\Vert }_{L^{p^{\prime }}}\le \Vert f{\Vert }_{L^p}$.

The Hausdorff-Young inequality interpolates between the $L^1\to L^{\infty }$ endpoint $\Vert \hat{f}{\Vert }_{\infty }\le \Vert f{\Vert }_1$ (immediate from the integral formula) and the $L^2\to L^2$ endpoint $\Vert \hat{f}{\Vert }_2=\Vert f{\Vert }_2$ (Plancherel). The Riesz-Thorin interpolation theorem ^{[Hausdorff 1923]} gives the intermediate inequalities with constant $1$. The endpoint conjugate exponent $p^{\prime }$ is sharp: for $p>2$, no $L^p\to L^q$ boundedness holds. The Beckner 1975 sharp constant theorem improved the constant from $1$ to $(p^{1/p}/p^{\prime 1/p^{\prime }}{)}^{n/2}$, with Gaussians as the unique extremizers (Lieb 1990).

Theorem 6 (Convolution theorem and pseudo-symmetry). For $f,g\in L^1({\mathbb{R}}^n)$, $\widehat{f\ast g}=\hat{f}\cdot \hat{g}$. For $f,g\in \mathcal{S}({\mathbb{R}}^n)$, additionally $\widehat{f\cdot g}=\hat{f}\ast \hat{g}$ (both convolution and pointwise product defined on $\mathcal{S}$).

The convolution theorem (Exercise 3) makes the Fourier transform an algebra homomorphism from the convolution Banach algebra $(L^1({\mathbb{R}}^n),\ast )$ to the pointwise-product algebra $(C_0({\mathbb{R}}^n),\cdot )$ — a Gelfand transform identification. The convolution-product duality on $\mathcal{S}$ (where both pictures are well-defined) is the master symmetry of the Fourier transform, and underlies the differentiation-multiplication duality of Exercise 4 (${\partial }^{\alpha }$ is "convolution with ${\delta }^{(\alpha )}$" in some sense; multiplication by $x^{\alpha }$ becomes "differentiation in $\xi$").

Theorem 7 (Fourier transform on tempered distributions; Schwartz 1950). The Fourier transform extends to a continuous linear self-map $\mathcal{F}:{\mathcal{S}}^{\prime }({\mathbb{R}}^n)\to {\mathcal{S}}^{\prime }({\mathbb{R}}^n)$ on the space of tempered distributions, defined by the duality $⟨\hat{T},\phi ⟩=⟨T,\hat{\phi }⟩$ for $T\in {\mathcal{S}}^{\prime }$ and $\phi \in \mathcal{S}$.

Schwartz 1950 ^{[Schwartz 1950]} extended the Fourier transform from functions to distributions by duality, with ${\mathcal{S}}^{\prime }$ — the dual of $\mathcal{S}$ — as the natural target. The extension subsumes all previous Fourier-transform settings: $L^1$, $L^2$, and $L^p$ for $p\in [1,2]$ as function-valued distributions; and includes singular distributions like the Dirac delta ${\delta }_0$ (with ${\hat{\delta }}_0=1$, constant) and the constant function $1$ (with $\hat{1}={\delta }_0$). The Fourier transform on ${\mathcal{S}}^{\prime }$ is the "right" framework for the Fourier-analytic theory of linear PDEs: solutions of $P(\partial )u=f$ are tempered distributions $u={\mathcal{F}}^{-1}(\hat{f}/P(2\pi i\xi ))$ when the symbol $P(2\pi i\xi )$ is non-vanishing, and the analysis of pseudodifferential operators and Fourier integral operators (Hörmander 1990) ^[Hörmander] is built on this foundation.

Theorem 8 (Heat / wave / Schrödinger via Fourier transform). The constant-coefficient linear evolution PDEs on ${\mathbb{R}}^n$: $${\partial }_{t}u=\Delta u\text{(heat)},{\partial }_t^{2}u=\Delta u\text{(wave)},i{\partial }_{t}u=-\Delta u\text{(Schro¨dinger)}$$ admit closed-form solutions via the Fourier transform: $\hat{u}(\xi ,t)=m(\xi ,t){\hat{u}}_0(\xi )$ where $u_0$ is the initial datum and the multiplier is $m_{\text{heat}}=e^{-4{\pi }^2|\xi {|}^{2}t}$, $m_{\text{wave}}=\cos (2\pi |\xi |t)$ (for the symmetric initial-velocity-zero problem) or $\sin (2\pi |\xi |t)/(2\pi |\xi |)$ (initial-datum-zero), and $m_{\text{Schr}}=e^{-i4{\pi }^2|\xi {|}^{2}t}$.

The Fourier-transform technique reduces each PDE to an ODE in $t$ for the Fourier transform (Exercise 5 for heat). Each multiplier $m(\xi ,t)$ has a corresponding inverse-Fourier-transform kernel $K_t(x)$: the heat kernel $K_t^{\text{heat}}(x)=(4\pi t{)}^{-n/2}{e}^{-|x{|}^2/(4t)}$, the wave kernel involving spherical means (Kirchhoff formula in dimension three), the Schrödinger kernel $K_t^{\text{Schr}}(x)=(4\pi it{)}^{-n/2}{e}^{i|x{|}^2/(4t)}$. The Fourier-transform technique extends to general constant-coefficient operators $P(\partial )$ via the symbol $P(2\pi i\xi )$, and to variable-coefficient operators via pseudodifferential calculus.

Theorem 9 (Poisson summation formula). For $f\in \mathcal{S}({\mathbb{R}}^n)$ (or any $f$ in the Wiener algebra $A({\mathbb{R}}^n)$ with $\hat{f}$ also in $A({\mathbb{R}}^n)$), $$\sum _{n\in {\mathbb{Z}}^n}f(n)=\sum _{k\in {\mathbb{Z}}^n}\hat{f}(k).$$ More generally, for any lattice $\Lambda \subset {\mathbb{R}}^n$ with dual lattice ${\Lambda }^{\ast }=\{k\in {\mathbb{R}}^n:k\cdot n\in \mathbb{Z}\text{ for all }n\in \Lambda \}$, $$\sum _{n\in \Lambda }f(n)=(\mathrm{vol}({\mathbb{R}}^n/\Lambda ){)}^{-1}\sum _{k\in {\Lambda }^{\ast }}\hat{f}(k).$$

The Poisson summation formula (Exercise 7) is the master identity of harmonic analysis on lattices, with deep applications: the theta-function functional equation $\theta (t)=t^{-1/2}\theta (1/t)$ (theta from Poisson on a Gaussian); the Shannon sampling theorem (a band-limited function is determined by its samples at the Nyquist rate); the Selberg trace formula on Riemannian symmetric spaces (a non-abelian generalization, central to representation theory and automorphic forms); the functional equations of $L$-functions in analytic number theory (Hecke 1918, Tate 1950); and the Selberg sieve in analytic number theory. The general form on locally compact abelian groups: for a closed cocompact subgroup $\Gamma \subset G$ with annihilator ${\Gamma }^{\perp }\subset \hat{G}$, the sum of $f$ over $\Gamma$ equals (a constant times) the sum of $\hat{f}$ over ${\Gamma }^{\perp }$ — the foundational duality of harmonic analysis on LCA groups.

Theorem 10 (Heisenberg uncertainty principle in sharp form). For $f\in \mathcal{S}(\mathbb{R})$ with $\Vert f{\Vert }_{L^2}=1$, $${\left(\int (x-x_0{)}^2|f(x){|}^{2}dx\right)}^{1/2}\cdot {\left(\int (\xi -{\xi }_0{)}^2|\hat{f}(\xi ){|}^{2}d\xi \right)}^{1/2}\ge \frac{1}{4\pi },$$ for any $x_0,{\xi }_0\in \mathbb{R}$. Equality holds if and only if $f$ is a Gaussian wave-packet $f(x)=c{e}^{2\pi i{\xi }_0(x-x_0)}{e}^{-\pi (x-x_0{)}^2/{\sigma }^2}$ for some constants $c,\sigma$.

The Heisenberg uncertainty principle (Exercise 6) is the prototype quantitative time-frequency complementarity statement ^{[Heisenberg 1927]}. The sharp constant $1/(4\pi )$ is achieved by Gaussian wave-packets, the unique minimizers. Translating to standard quantum-mechanical units with momentum $p=\hslash \xi$, the inequality becomes ${\sigma }_x{\sigma }_p\ge \hslash /2$, the form due to Heisenberg 1927 with the Kennard 1927 sharp constant. Generalizations: the Robertson-Schrödinger uncertainty principle for non-commuting observables ${\sigma }_A{\sigma }_B\ge |⟨[A,B]⟩|/2$; the entropy-based uncertainty principle of Beckner 1975 / Hirschman 1957; the Cowling-Price 1984 strong uncertainty principle (qualitative localization). The Hardy uncertainty principle (Hardy 1933) gives a sharper version: if $|f(x)|\le C{e}^{-a\pi x^2}$ and $|\hat{f}(\xi )|\le C{e}^{-b\pi {\xi }^2}$ with $ab>1$, then $f=0$.

Theorem 11 (Paley-Wiener support theorem; Paley-Wiener 1934 AMS Colloq. Publ. 19). A function $f\in L^2(\mathbb{R})$ has Fourier transform $\hat{f}$ supported in $[-R,R]$ if and only if $f$ extends to an entire function $F:\mathbb{C}\to \mathbb{C}$ of exponential type at most $2\pi R$, in the sense that $|F(z)|\le C{e}^{2\pi R|\mathrm{Im}(z)|}$.

The Paley-Wiener theorem (Exercise 8) is the central identification of compactly-supported-Fourier-transform $L^2$-functions with boundary values of entire functions of exponential type — a deep correspondence between analyticity-on-strips in time and frequency-localization. The distributional version (Paley-Wiener-Schwartz; Hörmander 1990 §7.3) extends to tempered distributions: $T\in {\mathcal{S}}^{\prime }(\mathbb{R})$ has compact support iff $\hat{T}$ extends to an entire function of exponential type. Generalizations: Paley-Wiener-Cartwright for Fourier-Laplace transforms in several complex variables, Paley-Wiener for entire functions on tube domains ${\mathbb{R}}^n+i\Gamma$, Paley-Wiener for spherical Fourier transforms on Riemannian symmetric spaces (Harish-Chandra).

Synthesis. The Fourier-transform architecture on ${\mathbb{R}}^n$ extends the Fourier-series architecture on $\mathbb{T}$ by replacing the discrete-frequency spectrum $\mathbb{Z}$ with the continuous-frequency spectrum ${\mathbb{R}}^n$. The Schwartz space $\mathcal{S}({\mathbb{R}}^n)$ provides the natural test-function class where all Fourier-transform identities hold without integrability concerns, and the Fourier transform is a topological isomorphism of $\mathcal{S}$ to itself. The Plancherel theorem extends the Fourier transform to a unitary isomorphism of $L^2({\mathbb{R}}^n)$, and the tempered-distribution framework ${\mathcal{S}}^{\prime }$ further extends to a continuous self-map of distributions. The Hausdorff-Young inequality interpolates between the $L^1\to L^{\infty }$ and $L^2\to L^2$ endpoints, capturing the $L^p$-mapping properties for $p\in [1,2]$. The convolution theorem identifies the Fourier transform with an algebra homomorphism, and the differentiation-multiplication duality reduces constant-coefficient linear PDEs to algebraic equations in the Fourier picture — the structural mechanism behind the Fourier-analytic approach to heat, wave, Schrödinger, Klein-Gordon, and Laplace equations.

The pattern recurs through three escalations. First, Fourier's original 1822 ^{[Fourier 1822]} heuristic Fourier-transform integrals for the heat equation on the real line, used freely without convergence analysis. Second, the rigorous foundations: Plancherel 1910 ^{[Plancherel 1910]} establishing the $L^2$-isometry; Wiener 1933 ^{[Wiener 1933]} introducing the Wiener algebra and the Tauberian theorems; Bochner 1932 ^{[Bochner 1932]} developing the Bochner integral framework and the Bochner theorem characterizing Fourier transforms of positive measures. Third, the modern extensions: Schwartz 1950 ^{[Schwartz 1950]} introducing tempered distributions; Hörmander 1990 ^[Hörmander] developing pseudodifferential operators and Fourier integral operators; Beckner 1975 finding the sharp Hausdorff-Young constants; the modern time-frequency analysis (wavelets, Gabor frames, Wigner distributions) extending the Fourier picture to localized time-frequency atoms (Daubechies 1988, Grochenig 2001).

The endpoint is the unified harmonic-analytic framework where Fourier series and Fourier transforms are special cases of the spectral theory on locally compact abelian groups (Pontryagin 1934, Cartan-Godement 1947), and the unitary representation theory of non-abelian groups extends the picture (Mackey 1949, Harish-Chandra 1954). The Fourier transform is the spectral decomposition of $L^2({\mathbb{R}}^n)$ under the natural translation-invariant structure, the prototype of the spectral theory for self-adjoint operators (Stone 1932, von Neumann 1932) and of representation-theoretic decompositions on homogeneous spaces.

Full proof set Master

Proposition 1 (Gaussian self-transform). For $G(x)=e^{-\pi |x{|}^2}$ on ${\mathbb{R}}^n$, $\hat{G}(\xi )=G(\xi )$.

Proof. By the product structure of $G$ on ${\mathbb{R}}^n$ (separation of variables) and the multiplicative property of the Fourier transform on tensor products, it suffices to prove the case $n=1$. So consider $G(x)=e^{-\pi x^2}$ on $\mathbb{R}$. Compute $\hat{G}(\xi )=\int e^{-\pi x^2}{e}^{-2\pi i\xi x}dx$. Complete the square in the exponent: $-\pi x^2-2\pi i\xi x=-\pi (x+i\xi {)}^2-\pi {\xi }^2$. So $$\hat{G}(\xi )=e^{-\pi {\xi }^2}\int e^{-\pi (x+i\xi {)}^2}dx.$$ The remaining integral is over the real axis. The integrand $e^{-\pi (z{)}^2}$ is entire in $z$, and by Cauchy's theorem the integral over the shifted contour $\{x+i\xi :x\in \mathbb{R}\}$ equals the integral over the real axis (the integrand decays as $|\mathrm{Re}(z)|\to \infty$, so the vertical sides at $\pm \infty$ contribute zero). Thus $\int e^{-\pi (x+i\xi {)}^2}dx=\int e^{-\pi y^2}dy=1$ by the standard Gaussian integral with the $\pi$-normalization. Hence $\hat{G}(\xi )=e^{-\pi {\xi }^2}=G(\xi )$.

$\square$

Proposition 2 (Dilation rule for Gaussians). For $G_t(x)=t^{-n}{e}^{-\pi |x{|}^2/t^2}$ on ${\mathbb{R}}^n$, ${\hat{G}}_t(\xi )=e^{-\pi t^2|\xi {|}^2}$.

Proof. Apply the dilation identity $\widehat{f(\cdot /t)}(\xi )=t^n\hat{f}(t\xi )$ (a consequence of the change of variables $u=x/t$ in the defining integral). For $f(x)=e^{-\pi |x{|}^2}$, $f(x/t)=e^{-\pi |x{|}^2/t^2}$, so $G_t=t^{-n}f(\cdot /t)$, and $${\hat{G}}_t(\xi )=t^{-n}\cdot t^n\hat{f}(t\xi )=G(t\xi )=e^{-\pi t^2|\xi {|}^2}.$$

$\square$

Proposition 3 (Approximate identity property of Gaussians). The family $\{G_t{\}}_{t>0}$ defined above satisfies (i) $\int G_t(x)dx=1$ for all $t$, (ii) $G_t(x)\ge 0$, and (iii) for every $\delta >0$, ${\int }_{|x|>\delta }{G}_t(x)dx\to 0$ as $t\to 0^{+}$.

Proof. (i) By Proposition 2 with $\xi =0$: ${\hat{G}}_t(0)=e^0=1$, but ${\hat{G}}_t(0)=\int G_t(x)dx$ by definition. (ii) Direct from $G_t>0$. (iii) For $|x|>\delta$: $G_t(x)=t^{-n}{e}^{-\pi |x{|}^2/t^2}\le t^{-n}{e}^{-\pi {\delta }^2/t^2}$ for $|x|>\delta$. Integrating over the annulus $|x|>\delta$ and using the dominated convergence (or direct change of variables): ${\int }_{|x|>\delta }{G}_t(x)dx\to 0$ as $t\to 0^{+}$. (More precisely: ${\int }_{|x|>\delta }{G}_t(x)dx={\int }_{|y|>\delta /t}{G}_1(y)dy\to 0$ as $t\to 0^{+}$, since $\delta /t\to \infty$ and $G_1\in L^1$.)

$\square$

Proposition 4 (Fourier inversion on Schwartz). For $f\in \mathcal{S}({\mathbb{R}}^n)$, $f(x)={\int }_{{\mathbb{R}}^n}\hat{f}(\xi )e^{2\pi i\xi \cdot x}d\xi$ pointwise for every $x$.

Proof. As in the Key Theorem, Step 1: Gaussian-regularize the inverse-Fourier integral by inserting the factor $e^{-\pi t^2|\xi {|}^2}$, apply Fubini to interchange integrals over ${\mathbb{R}}_x^n\times {\mathbb{R}}_{\xi }^n$ (legal because the regularized integrand is in $L^1$), recognize the result as a Gaussian convolution $(f\ast G_t)(x)$, and pass to the limit $t\to 0^{+}$ using the approximate-identity property (Proposition 3). The pointwise limit on the left is $\int \hat{f}(\xi )e^{2\pi i\xi \cdot x}d\xi$ (by DCT, since $\hat{f}\in \mathcal{S}\subset L^1$); the pointwise limit on the right is $f(x)$ (by continuity of $f$ at $x$ and approximate-identity convergence). Equating limits gives the inversion formula.

$\square$

Proposition 5 (Plancherel on Schwartz). For $f,g\in \mathcal{S}({\mathbb{R}}^n)$, $⟨f,g{⟩}_{L^2}=⟨\hat{f},\hat{g}{⟩}_{L^2}$.

Proof. As in the Key Theorem, Step 2: write $\bar{g}(x)=\int \overline{\hat{g}(\xi )}{e}^{-2\pi i\xi \cdot x}d\xi$ (conjugate of inversion), substitute into $⟨f,g⟩=\int f\bar{g}$, exchange order of integration by Fubini, and recognize the inner integral as $\hat{f}(\xi )$. The result $⟨f,g⟩=\int \hat{f}(\xi )\overline{\hat{g}(\xi )}d\xi =⟨\hat{f},\hat{g}⟩$.

$\square$

Proposition 6 (Density of Schwartz in $L^2$). $\mathcal{S}({\mathbb{R}}^n)$ is dense in $L^2({\mathbb{R}}^n)$.

Proof. $C_c^{\infty }({\mathbb{R}}^n)\subset \mathcal{S}({\mathbb{R}}^n)$ (smooth compactly supported functions are Schwartz, with all Schwartz seminorms automatically bounded by compactness of support). $C_c^{\infty }$ is dense in $L^2({\mathbb{R}}^n)$ — a standard fact: density of $C_c({\mathbb{R}}^n)$ in $L^2$ (from outer-regularity of Lebesgue measure plus Urysohn-Tietze extension), then smoothing $C_c$ functions by mollifier convolution. So $\mathcal{S}$ is dense in $L^2$ as a superset of a dense set.

$\square$

Proposition 7 (Plancherel theorem, full statement). The Fourier transform extends from $\mathcal{S}({\mathbb{R}}^n)$ to a unitary isomorphism $\mathcal{F}:L^2({\mathbb{R}}^n)\to L^2({\mathbb{R}}^n)$.

Proof. By Proposition 5, $\mathcal{F}{|}_{\mathcal{S}}$ is an isometry $\mathcal{S}\to \mathcal{S}$ in the $L^2$-norm. By Proposition 6, $\mathcal{S}$ is dense in $L^2$. So $\mathcal{F}$ has a unique continuous extension to $L^2$, by the standard density-extension argument for bounded operators on Banach spaces (using Cauchy-sequence completion). The extension is an isometry, hence injective; its inverse ${\mathcal{F}}^{-1}$ (similarly extended from $\mathcal{S}$) gives surjectivity, and the inversion identity ${\mathcal{F}}^{-1}\mathcal{F}=\mathrm{id}$ passes to the limit.

$\square$

Proposition 8 (Convolution theorem). For $f,g\in L^1({\mathbb{R}}^n)$, $\widehat{f\ast g}(\xi )=\hat{f}(\xi )\hat{g}(\xi )$.

Proof. As in Exercise 3: write out the defining integral, apply Fubini-Tonelli (justified by absolute integrability of the double integrand, from $\Vert f\ast g{\Vert }_{L^1}\le \Vert f{\Vert }_{L^1}\Vert g{\Vert }_{L^1}$), substitute $u=x-y$, and recognize the inner integral as $e^{-2\pi i\xi \cdot y}\hat{g}(\xi )$.

$\square$

Proposition 9 (Differentiation-multiplication duality on Schwartz). For $f\in \mathcal{S}({\mathbb{R}}^n)$ and multi-index $\alpha$:

• $\widehat{{\partial }^{\alpha }f}(\xi )=(2\pi i\xi {)}^{\alpha }\hat{f}(\xi )$,
• $\widehat{x^{\alpha }f}(\xi )=(-2\pi i{)}^{-|\alpha |}{\partial }^{\alpha }\hat{f}(\xi )$.

Proof. As in Exercise 4: integration by parts in the integral defining $\widehat{{\partial }^{\alpha }f}$ (boundary terms zero by Schwartz decay) for the first identity, and differentiation under the integral sign for the second identity (justified by Schwartz decay and DCT for the difference quotients).

$\square$

Proposition 10 (Poisson summation formula). For $f\in \mathcal{S}(\mathbb{R})$, ${\sum }_{n\in \mathbb{Z}}f(n)={\sum }_{k\in \mathbb{Z}}\hat{f}(k)$.

Proof. As in Exercise 7: define the periodization $F(x)={\sum }_{n}f(x+n)$, compute its Fourier coefficients $\hat{F}(k)=\hat{f}(k)$ via interchange of sum and integral, recognize $F$ as a smooth $1$-periodic function whose Fourier series converges uniformly to $F$, and evaluate at $x=0$.

$\square$

Proposition 11 (Heisenberg uncertainty principle). For $f\in \mathcal{S}(\mathbb{R})$ with $\Vert f{\Vert }_{L^2}=1$, $\Vert xf{\Vert }_{L^2}\cdot \Vert \xi \hat{f}{\Vert }_{L^2}\ge 1/(4\pi )$, with equality iff $f$ is Gaussian.

Proof. As in Exercise 6: apply Cauchy-Schwarz to $\int xf\overline{f^{\prime }}$, integrate by parts to obtain $\mathrm{Re}\int xf\overline{f^{\prime }}=-(1/2)\Vert f{\Vert }^2$, and convert $\Vert f^{\prime }{\Vert }^2=4{\pi }^2\Vert \xi \hat{f}{\Vert }^2$ via Plancherel and the differentiation rule (Proposition 9).

$\square$

Connections Master

• Fourier series and the Riemann-Lebesgue lemma 02.10.01. The direct prerequisite. The Fourier transform on ${\mathbb{R}}^n$ is the continuous-frequency limit of Fourier series on the torus ${\mathbb{T}}^n=(\mathbb{R}/L\mathbb{Z}{)}^n$ as the period $L\to \infty$: a $L$-periodic function with Fourier coefficients $\hat{f}(k/L)$ at integer multiples of $1/L$ becomes a function on the line with Fourier-transform values $\hat{f}(\xi )$ at all real $\xi$, with the spacing $1/L$ between successive coefficients shrinking to zero. The Riemann-Lebesgue lemma (vanishing of Fourier coefficients on $\mathbb{T}$) carries over directly to the line via the same density-of-step-functions argument.

• $L^p$ spaces, Hölder, Minkowski, Riesz-Fischer completeness 02.07.06. The direct prerequisite for the $L^2$-theory and the Hausdorff-Young inequality. Riesz-Fischer completeness of $L^2({\mathbb{R}}^n)$ is the structural foundation for the Plancherel extension: the Fourier transform on $\mathcal{S}$ is an isometry into $L^2$, and Riesz-Fischer ensures the unique extension to all of $L^2$ remains an isometry. The Hausdorff-Young inequality $\Vert \hat{f}{\Vert }_{L^{p^{\prime }}}\le \Vert f{\Vert }_{L^p}$ for $p\in [1,2]$ uses Hölder's inequality at the $L^1\to L^{\infty }$ and $L^2\to L^2$ endpoints, interpolated via Riesz-Thorin.

• Fubini-Tonelli and product measures 02.07.07. The direct prerequisite for the convolution theorem and the Plancherel identity. Fubini-Tonelli is used in essentially every Fourier-analytic computation: the convolution theorem $\widehat{f\ast g}=\hat{f}\hat{g}$ (Proposition 8, Exercise 3); the Plancherel identity $⟨f,g⟩=⟨\hat{f},\hat{g}⟩$ (Proposition 5, Key Theorem Step 2); the Fourier inversion formula on Schwartz (Proposition 4, Key Theorem Step 1); and the Poisson summation formula (Proposition 10, Exercise 7). The interchange of sum-and-integral or integral-and-integral that Fubini justifies is the workhorse of the Fourier-analytic computation.

• Schwartz space and tempered distributions [forward: 02.10.06]. The natural extension. Tempered distributions ${\mathcal{S}}^{\prime }({\mathbb{R}}^n)$ provide the right framework for the Fourier transform of objects more singular than functions: the Dirac delta ${\delta }_0$ has ${\hat{\delta }}_0=1$ (constant function); the constant function $1$ has $\hat{1}={\delta }_0$. The Fourier transform is a continuous linear self-map of ${\mathcal{S}}^{\prime }$, and the duality $⟨\hat{T},\phi ⟩=⟨T,\hat{\phi }⟩$ defines the transform on distributions in terms of its action on test functions.

• Heat equation [forward: 02.13.03]. The structural application. The heat equation ${\partial }_{t}u=\Delta u$ on ${\mathbb{R}}^n$ becomes the ODE ${\partial }_t\hat{u}=-4{\pi }^2|\xi {|}^2\hat{u}$ in the Fourier picture (Exercise 5, Theorem 8), solved by $\hat{u}(\xi ,t)=e^{-4{\pi }^2|\xi {|}^{2}t}{\hat{u}}_0(\xi )$. The inverse Fourier transform gives the Gauss-Weierstrass heat-kernel representation $u(x,t)=(4\pi t{)}^{-n/2}\int e^{-|x-y{|}^2/(4t)}{u}_0(y)dy$.

• Wave equation [forward: 02.13.04]. The structural application. The wave equation ${\partial }_t^{2}u=\Delta u$ becomes ${\partial }_t^2\hat{u}=-4{\pi }^2|\xi {|}^2\hat{u}$, solved by $\hat{u}(\xi ,t)=\cos (2\pi |\xi |t){\hat{u}}_0(\xi )+\sin (2\pi |\xi |t)/(2\pi |\xi |)\cdot {\widehat{{\partial }_{t}u}}_0(\xi )$. The inverse Fourier transform gives the Kirchhoff formula in dimension three and the d'Alembert formula in dimension one.

• Schrödinger equation [forward: chapter 12 quantum mechanics, momentum representation]. The structural application. The free Schrödinger equation $i{\partial }_t\psi =-\Delta \psi /(2m)$ becomes $i{\partial }_t\hat{\psi }=(2{\pi }^2|\xi {|}^2/m)\hat{\psi }$ in the Fourier picture, with the momentum representation $\tilde{\psi }(p)=\hat{\psi }(p/\hslash )$ giving the wavefunction as a function of momentum. The Plancherel identity is the conservation of total probability $\int |\psi {|}^2=\int |\tilde{\psi }{|}^2$, and the Heisenberg uncertainty principle is the position-momentum complementarity ${\sigma }_x{\sigma }_p\ge \hslash /2$.

• Functional analysis [02.11]. The lateral framework. The Plancherel theorem is the prototypical $L^2$-unitary, and the Fourier-transform diagonalization of the abelian translation group ${\mathbb{R}}^n$ on $L^2({\mathbb{R}}^n)$ is the prototype of the spectral theory for self-adjoint operators. The Schwartz space $\mathcal{S}({\mathbb{R}}^n)$ as a Fréchet space, and tempered distributions ${\mathcal{S}}^{\prime }$ as its strong dual, fit into the locally-convex-topological-vector-space framework central to functional analysis. The Riesz-Thorin and Marcinkiewicz interpolation theorems behind the Hausdorff-Young inequality are core functional-analytic tools.

• Signal processing [lateral]. The applied face. The Fourier transform is the foundational tool of signal processing: digital filters (multiplication by a frequency-domain mask), sampling theorems (Shannon-Nyquist, the discrete-time Fourier transform), spectrograms (windowed Fourier transforms, the short-time Fourier transform), wavelets and time-frequency analysis (Gabor 1946, Daubechies 1988). The fast Fourier transform algorithm (Cooley-Tukey 1965) computes the discrete Fourier transform of $N$ samples in $O(N\log N)$ operations, the algorithmic workhorse behind essentially all modern digital signal processing.

Historical & philosophical context Master

Joseph Fourier's 1822 Théorie analytique de la chaleur ^{[Fourier 1822]} introduced the Fourier integral representation alongside the Fourier series representation as parallel tools for solving the heat equation. The Fourier integral $f(x)={\int }_{-\infty }^{\infty }\hat{f}(\xi )e^{2\pi i\xi x}d\xi$ was used by Fourier to handle the heat equation on the real line (a metal bar of infinite length), in contrast to the Fourier-series representation for the finite bar with periodic boundary conditions. Fourier's argument was heuristic: he derived the integral representation as a continuous limit of the series representation, with the integer index $n$ replaced by a continuous real variable $\xi$ and the discrete sum replaced by a continuous integral, but did not provide rigorous convergence analysis. The rigorous foundations of the Fourier integral representation were developed over the next century by Cauchy, Dirichlet, Riemann, Lebesgue, and Plancherel.

Augustin-Louis Cauchy's 1816 Théorie de la propagation des ondes à la surface d'un fluide pesant d'une profondeur indéfinie gave the first explicit Fourier-integral solution of a partial differential equation (the linearized water-wave equation), with the explicit calculation of the Fourier integral representation of the initial-value problem. Siméon Denis Poisson's 1820 Mémoire sur la théorie des ondes extended Cauchy's analysis to the spherical wave equation. The Fourier integral was a working tool of mathematical physics by the mid-nineteenth century, with rigorous foundations supplied piecemeal as needed.

Michel Plancherel's 1910 Rendiconti del Circolo Matematico di Palermo paper ^{[Plancherel 1910]}, titled Contribution à l'étude de la représentation d'une fonction arbitraire par des intégrales définies, established the $L^2$-isometry property of the Fourier transform that bears his name. Plancherel's argument extended the Parseval identity for Fourier series (Bessel-Parseval, going back to Bessel 1828 and Hilbert 1906) to the Fourier-integral setting, identifying the Fourier transform as a unitary isomorphism of $L^2(\mathbb{R})$. The original proof was a direct truncation argument: approximate $f$ by its restriction to bounded intervals, compute the Fourier transform of the restriction, and pass to the limit using the $L^2$-norm-preservation. The modern proof via Schwartz-space density (Schwartz 1950) is cleaner and more structural, but Plancherel's original argument is still pedagogically valuable for its direct connection to the Fourier-series Bessel-Parseval setting.

Norbert Wiener's 1933 The Fourier Integral and Certain of its Applications ^{[Wiener 1933]} codified the Fourier-integral theory in a single monograph, with applications to the Tauberian theorems (a class of theorems converting asymptotic statements about partial sums to asymptotic statements about averages, with the Fourier transform of $1/x$ as the structural ingredient), the prediction theory of stationary stochastic processes, and the foundations of cybernetics. Wiener's introduction of the Wiener algebra $A(\mathbb{R})$ (the image of $\mathcal{F}(L^1)$, with pointwise product structure) was the first algebraic-structural treatment of the Fourier transform.

Salomon Bochner's 1932 Vorlesungen über Fouriersche Integrale ^{[Bochner 1932]} developed the Fourier-integral theory for Banach-space-valued functions (Bochner integration), the foundation for the modern operator-valued Fourier-analytic machinery used in spectral theory and the theory of evolution equations. Bochner's 1933 theorem characterizing the Fourier transforms of positive measures (a function $\phi$ on ${\mathbb{R}}^n$ is the Fourier transform of a positive measure iff $\phi$ is continuous, positive-definite, and $\phi (0)>0$) is a central tool in probability theory and in the spectral theory of stationary processes.

Felix Hausdorff's 1923 Mathematische Zeitschrift paper ^{[Hausdorff 1923]} and W. H. Young's 1912 Proceedings of the Royal Society paper ^{[Young 1912]} established the Hausdorff-Young inequality $\Vert \hat{f}{\Vert }_{L^{p^{\prime }}}\le \Vert f{\Vert }_{L^p}$ for $p\in [1,2]$ via interpolation (Marcel Riesz 1927 supplied the modern Riesz-Thorin interpolation framework). The sharp constants in the Hausdorff-Young inequality were determined by Beckner 1975 Annals of Mathematics 102, 159 and Brascamp-Lieb 1976; the extremizers are Gaussians, a recurring theme in sharp Fourier-analytic inequalities.

Laurent Schwartz's 1950-1951 Théorie des distributions ^{[Schwartz 1950]} introduced the test-function-and-distribution framework that became the standard language of modern Fourier analysis. The Schwartz space $\mathcal{S}({\mathbb{R}}^n)$ and the tempered distributions ${\mathcal{S}}^{\prime }({\mathbb{R}}^n)$ provide the natural setting where the Fourier transform is an isomorphism (of $\mathcal{S}$ to itself, and of ${\mathcal{S}}^{\prime }$ to itself), and where the differentiation-multiplication duality, the convolution theorem, and the Plancherel identity all hold without integrability concerns. Schwartz's work earned him the 1950 Fields Medal and made distribution theory the lingua franca of linear PDE theory.

Werner Heisenberg's 1927 Zeitschrift für Physik paper ^{[Heisenberg 1927]} introduced the position-momentum uncertainty principle ${\sigma }_x{\sigma }_p\ge \hslash /2$ in quantum mechanics, with a physical-intuition-based derivation. Earle Hesse Kennard's 1927 Zeitschrift für Physik 44, 326 gave the sharp constant and the mathematical formulation; H. P. Robertson's 1929 Physical Review 34, 163 generalized to non-commuting observables. The connection to Fourier analysis is direct: the position and momentum representations of a quantum state $\psi$ are related by the Fourier transform $\tilde{\psi }(p)=\hat{\psi }(p/\hslash )$, and the Heisenberg uncertainty principle is the $L^2$-norm statement on the products of second moments (Exercise 6, Theorem 10).

The Paley-Wiener theorem of 1934 ^{[Schwartz 1950]} (Raymond E. A. C. Paley and Norbert Wiener, Fourier Transforms in the Complex Domain, AMS Colloquium Publications 19) established the correspondence between $L^2$-functions with compactly supported Fourier transform and entire functions of exponential type. This theorem is the analytic origin of the wavelet-and-band-limited-signal theory in signal processing, and was extended by Schwartz to tempered distributions (Paley-Wiener-Schwartz theorem) and by Hörmander to the theory of analytic wave-front sets in microlocal analysis (Hörmander 1990 §7).

The structural arc is a two-hundred-year story: Fourier 1807-1822 (the Fourier integral as a heuristic limit of Fourier series for the heat equation on the real line); Cauchy 1816 and Poisson 1820 (working applications to wave equations); Plancherel 1910 (the $L^2$-isometry theorem); Wiener 1933 (the Wiener-algebra framework, Tauberian theorems); Bochner 1932 (Bochner integration, Bochner's positive-measure theorem); Young 1912 and Hausdorff 1923 (the Hausdorff-Young inequality, interpolation); Schwartz 1950 (Schwartz space, tempered distributions); Beckner 1975 (sharp constants in Hausdorff-Young); Hörmander 1990 (microlocal analysis, Fourier integral operators). The endpoint is the modern harmonic-analytic framework where the Fourier transform on ${\mathbb{R}}^n$ is a special case of the Plancherel theorem on locally compact abelian groups (Pontryagin 1934, Cartan-Godement 1947), unified with the Fourier-series and discrete-Fourier-transform settings under the umbrella of Pontryagin duality, and extended to non-abelian groups via the representation-theoretic Plancherel formula (Harish-Chandra 1954, Mackey 1949). The Fourier transform on ${\mathbb{R}}^n$ is the prototype of the spectral theory for self-adjoint operators and of the harmonic-analytic decomposition on homogeneous spaces, the framework after which the entire modern theory of pseudodifferential operators, Fourier integral operators, and microlocal analysis is modelled.

Bibliography Master

@book{Fourier1822,
  author    = {Fourier, Jean-Baptiste Joseph},
  title     = {Th\'eorie analytique de la chaleur},
  publisher = {Firmin Didot, P\`ere et Fils},
  address   = {Paris},
  year      = {1822}
}

@article{Plancherel1910,
  author  = {Plancherel, Michel},
  title   = {Contribution \`a l'\'etude de la repr\'esentation d'une fonction arbitraire par des int\'egrales d\'efinies},
  journal = {Rendiconti del Circolo Matematico di Palermo},
  volume  = {30},
  year    = {1910},
  pages   = {289--335}
}

@book{Wiener1933,
  author    = {Wiener, Norbert},
  title     = {The Fourier Integral and Certain of its Applications},
  publisher = {Cambridge University Press},
  year      = {1933}
}

@book{Bochner1932,
  author    = {Bochner, Salomon},
  title     = {Vorlesungen \"uber Fouriersche Integrale},
  publisher = {Akademische Verlagsgesellschaft},
  address   = {Leipzig},
  year      = {1932}
}

@article{Hausdorff1923,
  author  = {Hausdorff, Felix},
  title   = {Eine Ausdehnung des {P}arsevalschen {S}atzes \"uber {F}ourierreihen},
  journal = {Mathematische Zeitschrift},
  volume  = {16},
  year    = {1923},
  pages   = {163--169}
}

@article{Young1912,
  author  = {Young, William Henry},
  title   = {On the multiplication of successions of {F}ourier constants},
  journal = {Proceedings of the Royal Society of London A},
  volume  = {87},
  year    = {1912},
  pages   = {331--339}
}

@book{Schwartz1950,
  author    = {Schwartz, Laurent},
  title     = {Th\'eorie des distributions, Tomes {I-II}},
  publisher = {Hermann},
  address   = {Paris},
  year      = {1950--1951}
}

@book{Hormander1990,
  author    = {H\"ormander, Lars},
  title     = {The Analysis of Linear Partial Differential Operators {I}},
  series    = {Grundlehren der mathematischen Wissenschaften 256},
  edition   = {2},
  publisher = {Springer},
  year      = {1990}
}

@book{SteinShakarchi2003,
  author    = {Stein, Elias M. and Shakarchi, Rami},
  title     = {Fourier Analysis: An Introduction},
  series    = {Princeton Lectures in Analysis, Volume {I}},
  publisher = {Princeton University Press},
  year      = {2003}
}

@book{SteinWeiss1971,
  author    = {Stein, Elias M. and Weiss, Guido},
  title     = {Introduction to Fourier Analysis on Euclidean Spaces},
  publisher = {Princeton University Press},
  year      = {1971}
}

@book{PaleyWiener1934,
  author    = {Paley, Raymond E. A. C. and Wiener, Norbert},
  title     = {Fourier Transforms in the Complex Domain},
  series    = {American Mathematical Society Colloquium Publications, Volume 19},
  publisher = {American Mathematical Society},
  year      = {1934}
}

@article{Heisenberg1927,
  author  = {Heisenberg, Werner},
  title   = {\"Uber den anschaulichen {I}nhalt der quantentheoretischen {K}inematik und {M}echanik},
  journal = {Zeitschrift f\"ur Physik},
  volume  = {43},
  year    = {1927},
  pages   = {172--198}
}

@article{Kennard1927,
  author  = {Kennard, Earle Hesse},
  title   = {Zur {Q}uantenmechanik einfacher {B}ewegungstypen},
  journal = {Zeitschrift f\"ur Physik},
  volume  = {44},
  year    = {1927},
  pages   = {326--352}
}

@article{Beckner1975,
  author  = {Beckner, William},
  title   = {Inequalities in {F}ourier analysis},
  journal = {Annals of Mathematics},
  volume  = {102},
  year    = {1975},
  pages   = {159--182}
}

@article{Riesz1927,
  author  = {Riesz, Marcel},
  title   = {Sur les maxima des formes bilin\'eaires et sur les fonctionnelles lin\'eaires},
  journal = {Acta Mathematica},
  volume  = {49},
  year    = {1927},
  pages   = {465--497}
}

@book{Grafakos2014,
  author    = {Grafakos, Loukas},
  title     = {Classical Fourier Analysis},
  edition   = {3},
  series    = {Graduate Texts in Mathematics 249},
  publisher = {Springer},
  year      = {2014}
}

@book{ReedSimon1975,
  author    = {Reed, Michael and Simon, Barry},
  title     = {Methods of Modern Mathematical Physics II: Fourier Analysis, Self-Adjointness},
  publisher = {Academic Press},
  year      = {1975}
}

@book{Folland1999,
  author    = {Folland, Gerald B.},
  title     = {Real Analysis: Modern Techniques and Their Applications},
  edition   = {2},
  publisher = {Wiley},
  year      = {1999}
}