From Fourier Series to the Fourier Transform

The wall a Fourier series cannot climb

The previous guides handed you a powerful machine: any well-behaved function that repeats with period 2L is a weighted sum of pure waves, its discrete spectrum of harmonics. But notice the fine print — it repeats. A Fourier series is built only from waves that fit a whole number of times into the period, so it can only ever produce something periodic. Feed it a single pulse, a lone Gaussian bump, a struck bell that rings once and fades, and the series will happily reproduce that shape inside one window of width 2L — and then copy-paste it forever, left and right, into ghost echoes you never asked for. For a one-off signal, that is wrong everywhere except in the original window.

Yet most of the signals we actually care about are not periodic. A spoken word, a hammer blow on a beam, a photon's wave packet, the temperature spike from a brief laser pulse — each happens once and is gone. We need a way to find the spectrum of a signal that lives on the whole line and never repeats. The trick is not to invent a new machine but to take a limit of the one we have: make the period bigger, and bigger, and bigger, until 2L becomes infinite and the one window swallows the entire line. Watching what happens to the harmonics as we do this is the whole story of this guide.

Start from the cleanest form: the complex exponential series

Taking the limit is far less painful if we begin with the complex exponential Fourier series rather than the two-formula sine-and-cosine version. Using Euler's identity e^{i theta} = cos(theta) + i sin(theta), a single complex wave e^{i n pi x / L} packs one cosine and one sine together, so the whole series collapses to one clean sum: f(x) = sum over all integers n of c_n e^{i n pi x / L}, where the single coefficient formula c_n = (1/2L) integral from -L to L of f(x) e^{-i n pi x / L} dx replaces the two Euler-Fourier formulas. The minus sign in the exponent is the projection 'reading off' the amount of the wave e^{+i n pi x / L} present in f — the same orthogonality move as before, now in one stroke.

Here is the key bookkeeping move before we stretch L. Each harmonic n corresponds to an angular frequency omega_n = n pi / L. As n steps through the integers, these frequencies sit on a ladder of evenly spaced rungs, and the gap between neighbours is delta-omega = pi / L. That spacing is the discrete spectrum's 'comb tooth width.' Watch it: as we grow L, the rungs slide closer together. The picket fence of allowed frequencies is about to get so dense it turns into a solid wall.

Let the period go to infinity: a sum becomes an integral

Now do the deed. Substitute the coefficient formula back into the series and group the terms so the spacing delta-omega = pi / L appears explicitly. As L grows toward infinity, two things happen at once. First, the frequency spacing delta-omega shrinks toward zero, so the allowed frequencies stop being a discrete list n pi / L and fill in a continuous variable omega that can be anything. Second — and this is the heart of it — a sum over n with a small spacing delta-omega multiplying each term is precisely the setup of a Riemann sum: in the limit, 'sum over n times delta-omega' turns into 'integral d omega.' The discrete sum melts into an integral over all frequencies.

Complex series, period 2L, with omega_n = n pi / L  and  delta-omega = pi / L:

  c_n = (1/2L) integral_{-L}^{L} f(x) e^{-i omega_n x} dx

  f(x) = sum_n  c_n e^{i omega_n x}
        = sum_n  [ (1/2L) integral_{-L}^{L} f(t) e^{-i omega_n t} dt ] e^{i omega_n x}
        = (1/2pi) sum_n  [ integral_{-L}^{L} f(t) e^{-i omega_n t} dt ] e^{i omega_n x} * delta-omega
                                                          \______ a Riemann sum in omega ______/

  L -> infinity   (so delta-omega -> 0,  omega_n -> continuous omega):

  define   F(omega) = integral_{-infinity}^{infinity} f(x) e^{-i omega x} dx     <-- FORWARD transform

  then     f(x)     = (1/2pi) integral_{-infinity}^{infinity} F(omega) e^{i omega x} d omega   <-- INVERSE

The transform pair: a function and its spectrum

What survives the limit is a matched pair of integrals — the Fourier transform pair. The forward transform, F(omega) = integral from -infinity to infinity of f(x) e^{-i omega x} dx, takes the signal f living over space or time and returns F over frequency: it answers 'how much of frequency omega is hiding inside f?' The inverse transform, f(x) = (1/2pi) integral from -infinity to infinity of F(omega) e^{i omega x} d omega, rebuilds the original by stacking up all those frequency pieces. The Fourier transform and its inverse are two views of the same object, the way a chord can be written as a waveform in time or as a list of notes — no information is lost passing between them.

Notice precisely what changed in crossing from series to transform. For a periodic signal, the spectrum was the discrete list of coefficients c_n — energy allowed only at the special harmonic frequencies, a row of isolated spikes. For a one-shot signal, the spectrum is the smooth function F(omega) — a continuous spectrum defined at every frequency, with no gaps between teeth. This is intuitive once you see why: a wave that never repeats cannot be made from a finite menu of exact harmonics; it needs an unbroken continuum of frequencies, each contributing an infinitesimal sliver, to build a shape that rises once and never comes back. Periodicity is exactly what quantizes a spectrum into discrete lines; lose it and the lines blur into a band.

Reading a spectrum: what F(omega) actually tells you

F(omega) is generally complex, and that is a feature, not a nuisance — it carries two real pieces of information at every frequency. Its magnitude |F(omega)|, the amplitude spectrum, says how strongly frequency omega is present; its angle, the phase spectrum, says where that wave's crests sit. The magnitude alone tells you the timbre — what a sound is made of, why a flute and a violin playing the same note sound different. The phase tells you the alignment — scramble the phases and you keep every frequency's strength yet smear a sharp click into a wash of noise. Both halves are needed to reconstruct the signal, which is why the inverse integral carries the full complex F.

One quantitative anchor makes the spectrum concrete: energy is conserved across the pair. The Parseval-Plancherel theorem says the integral of |f(x)|^2 dx equals (1/2pi) times the integral of |F(omega)|^2 d omega — the total energy you compute in space equals the total energy you compute in frequency. So |F(omega)|^2 is literally the energy density per unit frequency: the area under it over a band [a, b] is the energy carried by that band. This is the honest meaning of 'the spectrum of a signal' — not a vague metaphor but a measurable distribution of energy across frequency, the exact continuous heir of the per-harmonic energy you summed in Parseval's theorem for Fourier series.

What you now hold, and the family it opens

1. To find a signal's spectrum, compute the forward transform F(omega) = integral over the whole line of f(x) e^{-i omega x} dx — one improper integral that projects f onto the wave of each frequency omega.
2. Read the result: |F(omega)| is how much of each frequency is present (amplitude), its angle is the phase, and |F(omega)|^2 is the energy density per unit frequency — the true meaning of the spectrum.
3. To rebuild the signal, run the inverse: f(x) = (1/2pi) integral over the whole line of F(omega) e^{i omega x} d omega — stack the continuum of frequency pieces back into the original shape.

You have crossed the bridge from the periodic world to the whole real line, and the same projection idea that read off a single harmonic now reads off a continuum. Everything ahead in this rung builds on this pair. The transform turns differentiation into multiplication by i omega, so the Fourier transform cracks the heat equation and the wave equation the way the Laplace transform cracked initial-value ODEs — algebra in place of calculus. Computers sample the integral into a finite list and run it as the discrete and fast Fourier transform. And swap the kernel e^{-i omega x} for another and you get the rest of the family — the Mellin, Hankel, and Hilbert transforms — each the right lens for its own geometry. You have learned to hear a one-time signal as a continuous chord, and to name every frequency in it.