Fourier series

A Fourier series is a representation of a periodic function as a sum of sinusoidal basis functions — sines and cosines, or equivalently complex exponentials. Named after Joseph Fourier, who introduced the idea in his 1822 study of heat conduction, the Fourier series asserts that any sufficiently well-behaved periodic function can be decomposed into a superposition of harmonics, each oscillating at an integer multiple of a fundamental frequency. Formally, for a function f with period 2π, the Fourier series is:

f(x) ∼ a₀/2 + Σ_{n=1}^∞ (aₙ cos(nx) + bₙ sin(nx))

where the coefficients aₙ and bₙ are determined by integrating f against the corresponding basis function over one period. In complex form, the series becomes Σ_{n=-∞}^∞ cₙ e^{inx}, where cₙ = (1/2π) ∫_{-π}^{π} f(x) e^{-inx} dx.

The Fourier series is not merely a computational technique; it is a change of basis. The functions {e^{inx}} form an orthonormal set in the Hilbert space L²[-π, π], and the Fourier coefficients cₙ are simply the coordinates of f in this basis. This perspective reveals the deep structural analogy between Fourier analysis and linear algebra: the Fourier transform diagonalizes the differentiation operator, just as eigenvalues diagonalize a matrix. The nth derivative of e^{inx} is (in)ⁿ e^{inx}; in the Fourier basis, differentiation becomes multiplication. This is why differential equations with constant coefficients become algebraic equations when Fourier-transformed — the system's dynamics are separated into independent, non-interacting modes.

Convergence, Kernels, and the Limits of Representation

The question of whether a Fourier series actually converges to its original function occupied mathematicians for over a century and drove the development of modern analysis. The Dirichlet kernel D_N(x) = Σ_{n=-N}^N e^{inx} governs the partial sums of the series, and its properties determine convergence behavior. For continuous functions, the Fejér means — averages of the partial sums — converge uniformly, a result that motivated the study of summability methods. For piecewise smooth functions, the series converges pointwise except at discontinuities, where the Gibbs phenomenon produces persistent overshoot. The full resolution of convergence questions required the development of measure theory, functional analysis, and the theory of distributions.

The Convergence of Fourier series is not a single theorem but a landscape of results: Dirichlet's theorem for piecewise monotonic functions, Carleson's theorem (1966) establishing almost-everywhere convergence for L² functions, and the delicate boundary cases that separate convergence from divergence. This history illustrates a general pattern in mathematics: a practical technique (Fourier's heat equation solution) generates theoretical questions that reshape the foundations of the discipline.

Fourier Series in Systems and Signals

In signal processing and control theory, the Fourier series provides the frequency-domain representation of periodic signals. Where the Laplace transform handles transient behavior and the Discrete Fourier Transform handles sampled signals, the Fourier series occupies the middle ground: continuous, periodic, and analytically tractable. The series decomposition reveals which frequency components are present in a signal and in what proportions, enabling filter design, spectral analysis, and the interpretation of physical oscillations.

The connection to Pontryagin duality places the Fourier series in its most general context: the circle group 𝕋 = ℝ/2πℤ is dual to the integers ℤ, and the Fourier series is precisely the Pontryagin transform on this pair. The Riemann mapping theorem employs Fourier-like decompositions on the disk to solve boundary value problems. In each case, the same principle recurs: a system's behavior is simplified when viewed in a basis adapted to its symmetries.

The Fourier series is often taught as a tool for solving differential equations, but this framing misses its deeper significance. The series is a claim about the architecture of reality: that periodic phenomena are composed of simple harmonics, that complexity is built from elementary oscillations, that the inhomogeneous can be understood through the homogeneous. This claim is not universally true — it fails for non-periodic signals, nonlinear systems, and processes without translational symmetry. But where it holds, it holds with extraordinary power. The danger is not in using Fourier analysis but in forgetting that it is a symmetry assumption. The world is full of phenomena that are not periodic, not linear, and not decomposable. To assume otherwise is not mathematics; it is wishful thinking dressed in trigonometric notation.