A function f on [-L, L] can be written as f(x) = a₀/2 + Σ(aₙcos(nπx/L) + bₙsin(nπx/L)) with coefficients given by integrals of f against the basis functions. Fourier series decompose a function into sines and cosines, revealing frequency content. The coefficients measure the contribution of each harmonic to the overall function.
You have been computing definite integrals and working with trigonometric identities. Fourier series bring these two tools together to answer a surprising question: can every periodic function be written as an infinite sum of sines and cosines? The answer is yes — under fairly mild conditions — and the formula for *how* to write it is what the Fourier series definition gives you.
The central idea is frequency decomposition. A complicated periodic signal — a square wave, a sawtooth, a human vowel sound — can be built by layering simple sinusoids of increasing frequency: the "fundamental" frequency, then twice that frequency (the first harmonic), then three times, and so on. The Fourier coefficients aₙ and bₙ tell you *how much* of each harmonic to include. A large a₂ means the function has significant content at the second harmonic frequency; a coefficient near zero means that harmonic contributes almost nothing.
Why do the integral formulas for the coefficients work? The reason is *orthogonality*. The integral of cos(mπx/L) × cos(nπx/L) over [-L, L] equals zero whenever m ≠ n, and equals L when m = n. (This is where your trigonometric identities earn their keep — the product-to-sum identities reduce these integrals to elementary ones.) This is directly analogous to perpendicular vectors: when you dot two perpendicular unit vectors you get zero. Because the basis functions are orthogonal, you can isolate each coefficient independently: multiply both sides of the Fourier series by cos(nπx/L), integrate, and every term except the aₙ term vanishes.
The a₀/2 convention for the constant term is worth a moment's attention. The formula for aₙ with n ≥ 1 gives a factor of 1/L from the normalization. For n = 0, cos(0) = 1 everywhere, and its self-integral is 2L instead of L, producing a factor of 1/(2L). Writing the constant term as a₀/2 in the series (rather than a₀) lets all coefficients share the same formula: aₙ = (1/L) ∫₋ₗᴸ f(x) cos(nπx/L) dx for n = 0, 1, 2, ... — a notational convenience worth recognizing when you encounter both conventions.
Fourier series do not always converge pointwise to f(x) at every point. At a jump discontinuity, the series converges to the midpoint of the jump, and the partial sums overshoot near the discontinuity (the Gibbs phenomenon). These convergence subtleties are studied in a follow-on topic, but for now the practical takeaway is: Fourier series are a powerful tool for smooth and piecewise-smooth periodic functions, and the coefficient integrals are the machinery that makes them computable.