Waves & Oscillations · Upper Division

Fourier Analysis

Any periodic function — however complicated — can be decomposed into a sum of sines and cosines. This single idea underlies signal processing, quantum mechanics, optics, and partial differential equations.

PrerequisitesWave properties (Ch. 9) · Calculus · Linear algebra helps but is not required
Learning Goals
  • Compute Fourier series coefficients using orthogonality integrals.
  • Identify the Gibbs phenomenon and explain why the overshoot persists at discontinuities.
  • Define the Fourier transform and apply it to compute spectra of simple functions.
  • State and apply Parseval's theorem to relate energy in position and frequency space.
  • Derive the Heisenberg uncertainty relation from the Fourier uncertainty principle.

F.1 The Fourier Series

Let f(x) be a periodic function with period 2L. The Fourier series of f is the decomposition into harmonics.Fourier introduced this1822 · Heat flow invents a universal languageJoseph Fourier submitted his heat-flow paper to the French Academy of Sciences in 1807. The judges — Lagrange, Laplace, and Legendre among them — rejected it. The mathematics was not rigorous enough, they said, and the claim that any function could be written as a sum of sines seemed almost certainly wrong. Fourier spent fifteen years refining the work before it appeared as a book in 1822. The Fourier transform is now the most widely used mathematical tool in applied science — signal processing, quantum mechanics, image compression, medical imaging — all descended from a rejected paper about how iron rods cool.Source: Joseph Fourier, Theorie analytique de la chaleur.Further reading: Joseph Fourier

f(x)=a02+∑n=1∞[ancos⁡(nπxL)+bnsin⁡(nπxL)]f(x)=\frac{a_0}{2}+\sum_{n=1}^{\infty}\left[a_n\cos\left(\frac{n\pi x}{L}\right)+b_n\sin\left(\frac{n\pi x}{L}\right)\right](F.1)

The coefficients aₙ and bₙ are found by exploiting the orthogonality of sines and cosines: ∫−LLsin⁡(mπx/L) sin⁡(nπx/L) dx=0\int_{-L}^{L} \sin(m\pi x/L)\,\sin(n\pi x/L)\,dx = 0 unless m = n. Multiplying both sides by the appropriate trig function and integrating isolates each coefficient:

an=1L∫−LLf(x)cos⁡(nπxL) dxbn=1L∫−LLf(x)sin⁡(nπxL) dxa_n=\frac{1}{L}\int_{-L}^{L}f(x)\cos\left(\frac{n\pi x}{L}\right)\,dx \qquad b_n=\frac{1}{L}\int_{-L}^{L}f(x)\sin\left(\frac{n\pi x}{L}\right)\,dx(F.2)
Example F.1 — Fourier Series of a Square Wave

f(x)=+1f(x) = +1 for 0<x<L,−10 < x < L, -1 for −L<x<0-L < x < 0. Find the Fourier coefficients.

Symmetry:f is odd →allan=0\to all a_{n} = 0 (cosine terms vanish)
bnb_{n}:bn=1L∫−LLf(x)sin⁡(nπx/L) dx=2L∫0Lsin⁡(nπx/L) dxb_n = \frac{1}{L}\int_{-L}^{L} f(x)\sin(n\pi x/L)\,dx = \frac{2}{L}\int_{0}^{L} \sin(n\pi x/L)\,dx
Integrate:=2L[−Lnπcos⁡(nπx/L)]0L=2nπ(1−cos⁡nπ)= \frac{2}{L}\left[-\frac{L}{n\pi}\cos(n\pi x/L)\right]_0^L = \frac{2}{n\pi}(1 - \cos n\pi)
Result:bn=4/(nπ)b_{n} = 4/(n\pi) for odd n, 0 for even n
Series:f(x)=(4/π)[sin⁡(πx/L)+sin⁡(3πx/L)/3+sin⁡(5πx/L)/5+⋯ ]f(x) = (4/\pi)[\sin(\pi x/L) + \sin(3\pi x/L)/3 + \sin(5\pi x/L)/5 + \cdots]
Figure F.1. Fourier series approximation. Each slider adds more harmonics to the sum (blue curve). Individual harmonics are shown as faint colored lines. The target function is the dim gray curve. Notice the Gibbs phenomenon — the overshoot near discontinuities never disappears, it just moves closer to the jump.

F.2 The Gibbs Phenomenon

When a Fourier series approximates a function with a jump discontinuity, the partial sums overshoot by approximately 9% of the jump height near the discontinuity — and this overshoot persists no matter how many terms are added; it just becomes narrower. This is theGibbs phenomenon, discovered in 1899.

It reflects a deep truth: a uniformly convergent series cannot converge to a discontinuous function. The Fourier series converges in the mean square sense (minimum total error), not pointwise at the discontinuity. This distinction matters in signal processing (ringing artifacts in audio and image compression) and in solving PDEs with discontinuous boundary conditions.

F.3 The Fourier Transform

For non-periodic functions, we take the period L → ∞ and the discrete sum becomes an integral. The Fourier transform decomposes f(x) into a continuous spectrum of frequencies:

F^(k)=∫−∞∞f(x)e−ikx dx(forward transform)\hat F(k)=\int_{-\infty}^{\infty}f(x)e^{-ikx}\,dx \qquad \text{(forward transform)}(F.3)
f(x)=12π∫−∞∞F^(k)eikx dk(inverse transform)f(x)=\frac{1}{2\pi}\int_{-\infty}^{\infty}\hat F(k)e^{ikx}\,dk \qquad \text{(inverse transform)}(F.4)

The variable k is the spatial frequency (wavenumber). |F̂(k)|² is the power spectrum— how much of each frequency is present. The Fourier transform turns convolutions into multiplications, differential equations into algebraic equations, and is the mathematical backbone of every filter, antenna, and spectrometer.

Theorem F.1 — Parseval's Theorem
The total energy is the same whether computed in position space or frequency space:∫−∞∞∣f(x)∣2 dx=12π∫−∞∞∣F^(k)∣2 dk\int_{-\infty}^{\infty} |f(x)|^2\,dx = \frac{1}{2\pi}\int_{-\infty}^{\infty} |\hat F(k)|^2\,dkThis means no information is lost in the transform — it is a unitary operation.

F.4 Connection to Physics

The Fourier transform is everywhere in physics, often wearing disguises:

Quantum mechanics. The position wavefunction ψ(x) and momentum wavefunction φ(p) are Fourier transforms of each other: ϕ(p)=12πℏ∫ψ(x) e−ipx/ℏ dx\phi(p) = \frac{1}{\sqrt{2\pi\hbar}}\int \psi(x)\,e^{-ipx/\hbar}\,dx. The Heisenberg uncertainty relation Δx·Δp ≥ ℏ/2 is a direct consequence of the Fourier uncertainty principle: a narrow ψ(x) requires a broad φ(p) and vice versa.

Diffraction.The far-field diffraction pattern of an aperture is exactly the Fourier transform of the aperture's transmission function. The single-slit sinc² pattern from Chapter 20 is the squared magnitude of the Fourier transform of a rectangular function.

Signal processing. The discrete Fourier transform (DFT), computed efficiently by the Fast Fourier Transform (FFT) algorithm (Cooley–Tukey, 1965), underlies MP3 audio compression, JPEG image compression, radar, MRI, and Wi-Fi (OFDM).

Definition F.2 — Common Traps
  • Orthogonality sets coefficients: guessing harmonic amplitudes usually misses normalization factors.
  • Gibbs overshoot does not vanish: adding terms narrows the ringing but leaves the overshoot height near a jump.
  • Transform conventions vary: factors of 2π2\pi move between the forward and inverse transforms by convention.
  • Narrow in x means broad in k: localization and bandwidth are reciprocal, not independently adjustable.
Exercises — F.1–F.4 Fourier Analysis
0.
Compute the RMS value of the sawtooth wave f(x)=x/πf(x) = x/\pi on −π<x<π-\pi < x < \pi using the definition (1/2π)∫f(x)2dx1/2\pi)\int f(x)^{2}dx, and verify using Parseval's theorem.
Straightforward
1.Find the Fourier sine series of f(x)=x/πf(x) = x/\pi on −π<x<π-\pi < x < \pi (the sawtooth wave).
Straightforward
2.Find the Fourier transform F̂(k) of f(x)=e−a∣x∣f(x) = e^{-a|x|} for a > 0. How does the width of f in x relate to the width of F̂ in k?
Intermediate
3.State the convolution theorem for Fourier transforms. Show how it can be used to solve the ODEy′′+y=f(x)ODE y'' + y = f(x).
Intermediate
4.Derive the Heisenberg uncertainty relation ΔxΔp≥ℏ/2\Delta x\Delta p \ge \hbar/2 from the Fourier uncertainty principle. Which wavefunction shape achieves the minimum uncertainty?
Challenging
Key Takeaways
  • Fourier series: any periodic f(x)=∑[ancos⁡+bnsin⁡]f(x) = \sum[a_{n}\cos + b_{n}\sin]; coefficients found by orthogonality integrals.
  • Square wave: only odd harmonics, bn=4/(nπ)b_{n} = 4/(n\pi). Adding terms reduces error but Gibbs overshoot persists.
  • Fourier transform: extends series to non-periodic f;F^(k)=∫f(x)e−ikxdxf; \hat{F}(k) = \int f(x)e^{-ikx}dx.
  • Parseval's theorem: energy is preserved by the transform (it's unitary).
  • Narrow in x ↔ broad in k: the Fourier uncertainty principle, which implies Heisenberg.
  • Diffraction pattern = |FT of aperture|². Convolution in space = multiplication in frequency.