Watch the bars jump on a music player's equalizer: one column for the deep bass, one for the mid-range, one for the shimmering treble, each rising and falling as the song plays. Somehow the machine has looked at a single wobbling voltage — the raw sound wave — and worked out exactly how much bass, how much treble, and how much of everything in between is hiding inside it. How?

For a periodic function, the answer is a precise recipe, not a guess. We want to write a function as a sum of sines and cosines. The orthogonality of those building blocks hands us each coefficient almost for free — by exactly the same projection that recovers a vector's components.

For a vector, the component along a unit axis \mathbf{e}_k is the dot product \mathbf{v}\cdot\mathbf{e}_k. For a function we project the same way: to find the amount of \sin\frac{n\pi x}{L} inside f, take the inner product of f with that mode and divide by the mode's own "length squared". That single idea — project onto each wave in turn — is the whole engine behind the equalizer's bouncing bars.

Deriving one coefficient

Suppose f(x) = \sum_k b_k \sin\frac{k\pi x}{L}. Take the inner product of both sides with a single mode \sin\frac{n\pi x}{L}:

Orthogonality kills every term in the sum except k = n, where the integral equals L. The entire infinite sum collapses to one term:

That is the whole idea: orthogonality turns "solve for infinitely many unknowns at once" into "read off each one with a single integral".

Worked example: a square wave, by hand

Let's actually turn the crank. Take the square wave of period 2\pi (so L = \pi):

First, a shortcut that saves real work: f is an odd function (f(-x) = -f(x)), and every cosine term \cos\frac{n\pi x}{L} is even. An odd function times an even one is odd, and the integral of an odd function over a symmetric interval [-L, L] is always zero. So without integrating a single cosine, we already know a_0 = 0 and a_n = 0 for every n.

Now the sine coefficients. Since f(x)\sin(nx) is even (odd times odd), we can integrate over half the interval and double it:

Since \cos(n\pi) = (-1)^n, this is b_n = \dfrac{2}{n\pi}\big(1 - (-1)^n\big) — which is 0 whenever n is even (because 1 - (-1)^n = 0) and \dfrac{4}{n\pi} whenever n is odd. Plugging in the first few:

A clean pattern falls out of the arithmetic: only the odd harmonics show up at all, and their size shrinks like 1/n. That is not a coincidence of this one example — it is exactly the kind of shortcut symmetry buys you, and it is the reason a square wave's full series (built from these very numbers) turns out to be a pure sum of odd sines.

It's worth double-checking the symmetry argument with a direct integral, just once, so you trust it. Take a_1 = \frac{1}{\pi}\int_{-\pi}^{\pi} f(x)\cos(x)\,dx. Split it at zero: \frac{1}{\pi}\left(\int_{-\pi}^{0} (-1)\cos(x)\,dx + \int_0^{\pi} (1)\cos(x)\,dx\right) = \frac{1}{\pi}\big(-[\sin x]_{-\pi}^{0} + [\sin x]_0^{\pi}\big) = \frac{1}{\pi}(0 + 0) = 0, exactly as the symmetry shortcut predicted — but it took a genuine integral (two of them, in fact) to confirm what the odd/even check gave us for free. That's the whole point of checking symmetry first: it lets you skip work you already know the answer to.

How an equalizer "hears" bass and treble

Back to the opening question. A spectrum analyzer or graphic equalizer chops a sound recording into short, roughly-periodic snippets, then computes something extremely close to a Fourier coefficient for a whole ladder of frequencies at once — one for deep bass, one a little higher, and so on up through the treble. The size of each coefficient is literally "how much of that frequency is present right now," which is exactly what each bar's height displays. Turn up the bass knob on a stereo and, under the hood, you're multiplying one of these coefficients before the sound is rebuilt.

Doing this fast enough to update thirty times a second, for thousands of frequencies, for every speaker in every phone, would be hopeless by hand-integrating each one — so engineers use a blisteringly efficient algorithm called the Fast Fourier Transform that computes an entire batch of coefficients at once. The formulas are exactly the ones above; the speed comes from clever bookkeeping, not new mathematics.

A concrete picture: a typical equalizer might report a "bin" for every few dozen hertz, from about 20 Hz (a deep rumble you feel more than hear) up past 15 kHz (a bright hiss). Each bin's height is |a_n| or |b_n|-style magnitude for the frequency that bin represents — feed in a pure bass drum thump and the low bins light up while the high ones stay flat; feed in a cymbal crash and the pattern flips. The whole visual is nothing more than a bar chart of Fourier coefficients, redrawn many times a second.

One mode at a time

Take f(x) = x on [-\pi, \pi]. Its sine coefficients work out to b_n = \dfrac{2(-1)^{n+1}}{n} — so the low modes carry most of the weight and the contributions taper off. The faint line is f; the bold curve is the single projected mode b_n\sin(nx). Step n to watch each ripple's amplitude — exactly what the integral above measures, and exactly what an equalizer's bars are doing thirty times a second.

See it explained