Fourier Series

Fourier series represent any reasonable periodic function as an infinite sum of sines and cosines (or complex exponentials). Key result: a function of period

2 π

that is piecewise smooth equals

\frac{a _{0}}{2} + \sum_{n = 1}^{\infty} (a_{n} cos n x + b_{n} sin n x)

, with coefficients given by integrals against the trig functions. Applies to: signal processing, heat conduction, vibrations, electrical engineering, quantum mechanics, number theory. Who/when: Joseph Fourier (1807, in his study of heat), building on Euler, Bernoulli, and d'Alembert.

A Fourier series is the expansion of a periodic function $f$ into a (possibly infinite) linear combination of harmonically related sinusoids, whose frequencies are integer multiples of a fundamental frequency. Its importance is foundational: it asserts that the trigonometric system forms a complete orthogonal basis for square-integrable periodic functions, so that nearly any periodic signal — including discontinuous ones like square waves — can be reconstructed from elementary oscillations. This matters because differentiation and convolution become algebraic operations on the coefficients, transforming differential equations into solvable arithmetic. The surprising fact, fiercely contested in the early 19th century, is that a sum of perfectly smooth, infinitely differentiable sines can converge to a function with sharp corners or jumps. Even more surprising: the partial sums near a jump always overshoot by about 9% no matter how many terms you take — the Gibbs Phenomenon.

The Core Insight

The sines and cosines ${1, cos x, sin x, cos 2 x, sin 2 x, \dots}$ are mutually orthogonal over one period: the integral of any two distinct members' product vanishes. This means we can extract each coefficient independently by projecting $f$ onto that basis function — exactly like finding the components of a vector along orthogonal axes. The Fourier coefficient $a_{n}$ is simply "how much of $cos n x$ is inside $f$ ," measured by an inner product. The whole theory is linear algebra in infinite dimensions.

Formal Statement

Fourier Series of a $2\pi$-Periodic Function

Let $f : R \to R$ be periodic with period $2 π$ and integrable on $[- π, π]$ . Its Fourier series is

f (x) \sim \frac{a _{0}}{2} + n = 1 \sum \infty (a_{n} cos n x + b_{n} sin n x)

⟳postulate a trig expansion, then determine coefficients by exploiting orthogonality (see proof below). The symbol

\sim

denotes "has Fourier series," deferring the question of convergence.

with coefficients

a_{n} = \frac{1}{π} \int_{- π}^{π} f (x) cos n x d x, n = 0, 1, 2, \dots

⟳multiply [1] by

cos m x

, integrate over

[- π, π]

, and use

\int_{- π}^{π} cos n x cos m x d x = π δ_{nm}

for

n, m \geq 1

b_{n} = \frac{1}{π} \int_{- π}^{π} f (x) sin n x d x, n = 1, 2, \dots

⟳multiply [1] by

sin m x

and integrate; use

\int_{- π}^{π} sin n x sin m x d x = π δ_{nm}

Here $a_{0} /2$ is the mean value (DC component) of $f$ , $n$ is the harmonic index, $a_{n}, b_{n}$ are the Fourier coefficients, and $δ_{nm}$ is the Kronecker delta. The fundamental angular frequency is $1$ (period $2 π$ ); harmonic $n$ oscillates $n$ times faster.

Complex (Exponential) Form

Using Euler's formula $e^{in x} = cos n x + i sin n x$ , the series compresses to

f (x) \sim n = - \infty \sum \infty c_{n} e^{in x}, c_{n} = \frac{1}{2 π} \int_{- π}^{π} f (x) e^{- in x} d x

⟳substitute

cos n x = \frac{1}{2} (e^{in x} + e^{- in x})

sin n x = \frac{1}{2 i} (e^{in x} - e^{- in x})

into [1]; collect terms to get

c_{n} = \frac{1}{2} (a_{n} - i b_{n})

c_{- n} = \frac{1}{2} (a_{n} + i b_{n})

c_{0} = a_{0} /2

For a general period $T = 2 L$ , the angular frequency becomes $ω_{0} = π / L$ and

f (x) \sim n = - \infty \sum \infty c_{n} e^{inπ x / L}, c_{n} = \frac{1}{2 L} \int_{- L}^{L} f (x) e^{- inπ x / L} d x

⟳rescale the variable

x \mapsto π x / L

to map period

2 L

onto period

2 π

Dirichlet Convergence Theorem

If $f$ is periodic with period $2 π$ and piecewise smooth (piecewise continuous with a piecewise continuous derivative), then for every $x$ the Fourier series converges, and

\frac{a _{0}}{2} + n = 1 \sum \infty (a_{n} cos n x + b_{n} sin n x) = \frac{f ( x ^{+} ) + f ( x ^{-} )}{2}

⟳write the partial sum

S_{N}

as a convolution with the Dirichlet kernel

D_{N}

, then apply the Riemann–Lebesgue lemma to the remainder; the kernel's mass-1 property gives the average of one-sided limits.

At points of continuity this equals $f (x)$ ; at a jump it converges to the midpoint of the jump.

Parseval's Identity / Completeness

For $f \in L^{2} [- π, π]$ the trigonometric system is complete, and energy is conserved:

\frac{1}{π} \int_{- π}^{π} ∣ f (x) ∣^{2} d x = \frac{a _{0}^{2}}{2} + n = 1 \sum \infty (a_{n}^{2} + b_{n}^{2})

⟳expand

\int ∣ f ∣^{2} = \int f \cdot f

, substitute the series for one factor, integrate term-by-term using orthogonality. In complex form:

\frac{1}{2 π} \int ∣ f ∣^{2} = \sum ∣ c_{n} ∣^{2}

This identity is the infinite-dimensional Pythagorean theorem.

Why It Holds / Derivation

Coefficient Formulas via Orthogonality

Step 1 — Orthogonality relations. Direct integration using product-to-sum formulas gives, for integers

n, m \geq 1

\int_{- π}^{π} cos n x cos m x d x = π δ_{nm}, \int_{- π}^{π} sin n x sin m x d x = π δ_{nm}, \int_{- π}^{π} cos n x sin m x d x = 0.

Also

\int_{- π}^{π} cos n x d x = \int_{- π}^{π} sin n x d x = 0

for

n \geq 1

, while

\int_{- π}^{π} 1 d x = 2 π

. Step 2 — Assume the expansion holds and may be integrated term-by-term:

f (x) = \frac{a _{0}}{2} + k = 1 \sum \infty (a_{k} cos k x + b_{k} sin k x) .

Step 3 — Project onto $cos m x$ . Multiply both sides by

cos m x

and integrate over

[- π, π]

. Every term vanishes by Step 1 except the one with

k = m

\int_{- π}^{π} f (x) cos m x d x = a_{m} \cdot π \Rightarrow a_{m} = \frac{1}{π} \int_{- π}^{π} f (x) cos m x d x .

Step 4 — The mean. Projecting onto

1

(i.e.

cos 0 x

\int f d x = \frac{a _{0}}{2} \cdot 2 π

, recovering

a_{0} = \frac{1}{π} \int f d x

. This is why we wrote

a_{0} /2

— so formula [2] holds uniformly for all

n \geq 0

. Step 5 — Project onto $sin m x$ identically yields

b_{m} = \frac{1}{π} \int_{- π}^{π} f sin m x d x

. Step 6 — Justification of term-by-term integration. For

f \in L^{2}

, the partial sums converge in the

L^{2}

norm, and the inner product is continuous in that norm, so interchanging sum and integral is valid.

■

Dirichlet Kernel Sketch of Pointwise Convergence

Step 1. The

N

-th partial sum is

S_{N} (x) = \sum_{∣ n ∣ \leq N} c_{n} e^{in x} = \frac{1}{2 π} \int_{- π}^{π} f (t) D_{N} (x - t) d t

, where the Dirichlet kernel is

D_{N} (u) = n = - N \sum N e^{in u} = \frac{sin ( ( N + \frac{1}{2} ) u )}{sin ( u /2 )} .

Step 2.

D_{N}

has total integral

2 π

and concentrates near

u = 0

N \to \infty

, but with oscillating sidelobes (the source of Gibbs overshoot). Step 3. Writing

S_{N} (x) - \frac{f ( x ^{+} ) + f ( x ^{-} )}{2}

as an integral of

D_{N}

against the smooth difference quotient of

f

, the Riemann–Lebesgue lemma (rapidly oscillating integrals of integrable functions vanish) forces this remainder to

0

for piecewise-smooth

f

■

Jean-Baptiste Joseph Fourier | 1768–1830

A French mathematician and physicist who accompanied Napoleon to Egypt and later served as prefect of Isère. In his 1807 memoir and the 1822 treatise Théorie analytique de la chaleur, Fourier solved the heat equation by boldly asserting that any function could be expanded in trigonometric series. The claim scandalized Lagrange, Laplace, and Legendre, whose committee criticized its lack of rigor — yet the method worked spectacularly. Fourier also discovered the greenhouse effect and the dimensional analysis of physical equations. His insistence on the universality of trigonometric expansion forced mathematicians to clarify the very notions of function, integral, and convergence, seeding a century of analysis.

Building Intuition

Think of $f$ as a musical chord. A pure tone is a single sine; a real instrument's sound is a sum of harmonics whose relative strengths — the coefficients $a_{n}, b_{n}$ — define the timbre. Fourier analysis is the ear's job in reverse: given the chord, recover the individual notes.

Geometrically, each basis function is an axis in an infinite-dimensional space, and $f$ is a vector. The coefficient $c_{n}$ is the shadow $f$ casts on the $e^{in x}$ axis. Because the axes are perpendicular (orthogonal), each shadow can be measured without interference — that's the whole trick.

Symmetry shortcuts: if $f$ is even, all $b_{n} = 0$ (only cosines survive). If $f$ is odd, all $a_{n} = 0$ (only sines). Smoother functions have coefficients decaying faster: a jump gives $\sim 1/ n$ decay, a corner $\sim 1/ n^{2}$ , and an infinitely smooth periodic function decays faster than any power — the spectrum reads off the regularity of $f$ .

Square wave (gold) and its partial Fourier sums approaching it

<svg viewBox="0 0 520 300" xmlns="http://www.w3.org/2000/svg">
<rect width="520" height="300" fill="#0a0e17"/>
<line x1="30" y1="150" x2="500" y2="150" stroke="#94a3b8" stroke-width="1"/>
<line x1="30" y1="40" x2="30" y2="260" stroke="#94a3b8" stroke-width="1"/>
<text x="505" y="154" fill="#e2e8f0" font-size="12">x</text>
<text x="14" y="44" fill="#e2e8f0" font-size="12">f</text>
<!-- target square wave -->
<polyline points="30,90 265,90 265,210 500,210" fill="none" stroke="#d4a853" stroke-width="2.5"/>
<text x="120" y="80" fill="#d4a853" font-size="12">square wave</text>
<!-- N=1 sine -->
<path d="M30,150 Q147,55 265,150 Q382,245 500,150" fill="none" stroke="#64748b" stroke-width="1.4" stroke-dasharray="4 3"/>
<text x="60" y="125" fill="#64748b" font-size="11">N=1</text>
<!-- N=5 wiggly approximation -->
<path d="M30,150 C60,95 80,95 100,98 C120,100 140,95 160,95 C200,95 230,95 265,95
         C300,205 330,205 370,205 C400,205 440,205 470,202 C485,200 495,180 500,150"
      fill="none" stroke="#e2e8f0" stroke-width="1.6"/>
<text x="300" y="190" fill="#e2e8f0" font-size="11">N=5 (note overshoot)</text>
<!-- Gibbs overshoot markers -->
<circle cx="262" cy="78" r="3" fill="#f87171"/>
<text x="270" y="74" fill="#f87171" font-size="10">Gibbs ~9%</text>
</svg>

Worked Examples

Square Wave

Given: the odd,

2 π

-periodic square wave

f (x) = {+ 1 - 1 0 < x < π - π < x < 0

. Step 1 — Symmetry.

f

is odd, so all

a_{n} = 0

. Step 2 — Sine coefficients.

b_{n} = \frac{1}{π} \int_{- π}^{π} f sin n x d x = \frac{2}{π} \int_{0}^{π} sin n x d x = \frac{2}{π} \cdot \frac{1 - cos nπ}{n} .

Step 3 — Evaluate. Since

cos nπ = (- 1)^{n}

b_{n} = \frac{2 ( 1 - ( - 1 ) ^{n} )}{π n}

, which is

0

for even

n

and

\frac{4}{π n}

for odd

n

. Step 4 — Series.

f (x) = \frac{4}{π} k = 0 \sum \infty \frac{sin ( ( 2 k + 1 ) x )}{2 k + 1} = \frac{4}{π} (sin x + \frac{1}{3} sin 3 x + \frac{1}{5} sin 5 x + \dots)

Bonus: evaluating at

x = π /2

where

f = 1

gives the Leibniz series

1 - \frac{1}{3} + \frac{1}{5} - \dots = \frac{π}{4}

Parseval Yields $\zeta(2)$

Given:

f (x) = x

(- π, π)

, extended

2 π

-periodically (a sawtooth). Find

\sum 1/ n^{2}

. Step 1 — Coefficients.

f

is odd, so

a_{n} = 0

. Integrate by parts:

b_{n} = \frac{2}{π} \int_{0}^{π} x sin n x d x = \frac{2}{π} [- \frac{x cos n x}{n} + \frac{sin n x}{n ^{2}}]_{0}^{π} = \frac{2}{π} \cdot \frac{- π ( - 1 ) ^{n}}{n} = \frac{2 ( - 1 ) ^{n + 1}}{n} .

Step 2 — Apply Parseval [7].

\frac{1}{π} \int_{- π}^{π} x^{2} d x = \sum_{n = 1}^{\infty} b_{n}^{2} .

The left side is

\frac{1}{π} \cdot \frac{2 π ^{3}}{3} = \frac{2 π ^{2}}{3}

. Step 3 — Right side.

\sum b_{n}^{2} = \sum_{n = 1}^{\infty} \frac{4}{n ^{2}} = 4 \sum \frac{1}{n ^{2}}

. Step 4 — Solve.

\frac{2 π ^{2}}{3} = 4 \sum \frac{1}{n ^{2}} \Rightarrow

n = 1 \sum \infty \frac{1}{n ^{2}} = \frac{π ^{2}}{6} \approx 1.6449

This is the famous Basel problem, solved by Fourier methods in a few lines.

Common Misconceptions

The $\sim$ is not always $=$ . A Fourier series need not converge to

f

at every point. At jump discontinuities it converges to the midpoint, not to either one-sided value. There even exist continuous functions whose Fourier series diverge at some points (Du Bois-Reymond, 1873). Pointwise convergence everywhere requires extra hypotheses; $L^{2}$ convergence (in the energy norm) is what always holds for

f \in L^{2}

. Increasing $N$ does not eliminate overshoot. Near a jump the partial sums overshoot by a fixed ~8.95% of the jump, no matter how large

N

is — the spike merely narrows. This is the Gibbs Phenomenon, not a numerical error.

Coefficient decay does not guarantee uniform convergence

The series $\sum_{n = 1}^{\infty} \frac{s i n n x}{n}$ has coefficients $b_{n} = 1/ n \to 0$ and converges pointwise (to the sawtooth $(π - x) /2$ on $(0, 2 π)$ ), yet it does not converge uniformly: the limit is discontinuous at $x = 0$ , while every partial sum is continuous, so uniform convergence would force a continuous limit. The Gibbs spikes near $0$ never shrink in height. Mere $1/ n$ decay is the signature of a jump discontinuity, and uniform convergence is impossible there.

Connections

The Fourier series is the period-restricted ancestor of the Fourier Transform: as the period $T \to \infty$ , the discrete spectrum ${c_{n}}$ blurs into a continuous spectrum $\hat{f} (ω)$ , and the sum [5] becomes an integral. The orthogonality at its heart is a special case of the spectral theory of self-adjoint operators — the $e^{in x}$ are eigenfunctions of $- d^{2} / d x^{2}$ on the circle, linking it to Sturm–Liouville Theory and general Orthogonal Polynomials. Parseval's identity makes the trig system an orthonormal basis of Hilbert Space $L^{2}$ , placing Fourier analysis squarely inside Functional Analysis. The discrete computational cousin is the Discrete Fourier Transform, evaluated rapidly by the Fast Fourier Transform. Termwise differentiation connects it to solving the Heat Equation and Wave Equation by separation of variables.

Real-World Applications

Audio compression (MP3, AAC): sounds are decomposed into frequency components; inaudible coefficients are discarded, slashing file size.
Image compression (JPEG): uses the closely related discrete cosine transform on 8×8 blocks to concentrate energy in few coefficients.
Heat conduction & PDEs: Fourier's original use — solving the heat equation in a rod or plate by expanding initial temperature profiles in eigenmodes that decay at rate $e^{- n^{2} t}$ .
Electrical engineering: AC circuit analysis and harmonic distortion measurement; any periodic voltage waveform is a sum of harmonics driving impedances frequency-by-frequency.
Vibration & structural analysis: modal analysis of beams, membranes, and bridges identifies resonant harmonics to avoid catastrophic resonance.
Crystallography: X-ray diffraction patterns are the Fourier coefficients of electron density; inverting them reconstructs molecular structure (how DNA's helix was found).
Magnetic resonance imaging (MRI): the scanner samples the Fourier (k-space) representation of the body; inverse transform yields the image.

Historical Development

1748 — Euler expands functions in trigonometric series while studying vibrating strings, computing some coefficients.
1753 — Daniel Bernoulli argues a vibrating string's motion is a superposition of harmonic modes; d'Alembert and Euler object that this cannot represent arbitrary shapes.
1777 — Euler derives the coefficient integral formulas (the orthogonality trick) in a special context.
1807 — Fourier submits his memoir on heat to the Paris Academy, asserting universal trigonometric expansion; the committee (Lagrange, Laplace, Legendre) withholds the prize over rigor.
1822 — Fourier publishes Théorie analytique de la chaleur, the definitive treatise.
1829 — Dirichlet gives the first rigorous convergence theorem for piecewise-monotonic functions, founding modern analysis of the subject.
1854 — Riemann develops his integral specifically to handle Fourier coefficients of badly behaved functions, in his Habilitation.
1873 — Du Bois-Reymond constructs a continuous function whose Fourier series diverges at a point.
1899 — Gibbs (after Wilbraham, 1848) explains the persistent overshoot at discontinuities; 1907 — Riesz–Fischer establish $L^{2}$ completeness. 1966 — Carleson proves Fourier series of $L^{2}$ functions converge almost everywhere.

Subtleties & Limits

Convergence theory contains deep and partly open territory. Carleson's theorem (1966) guarantees almost-everywhere convergence for $L^{2}$ functions, and Hunt extended it to $L^{p}$ for $p > 1$ ; but Kolmogorov (1923) built an $L^{1}$ function whose Fourier series diverges everywhere, so $L^{1}$ is hopeless. The sharp characterization of which continuous functions have everywhere-convergent series, and the precise interplay between coefficient decay rates and pointwise regularity, remains subtle. In higher dimensions, the convergence of multidimensional Fourier series depends delicately on the summation method (spherical vs. rectangular partial sums) — the ball multiplier problem (Fefferman, 1971) shows the disc multiplier is unbounded on $L^{p}$ for $p \neq = 2$ , revealing that naive radial summation fails. Questions about optimal convergence under minimal smoothness, and connections to additive combinatorics via exponential sums, remain active research.