Key Concepts of Fourier Series

Why This Matters

Fourier Series sit at the intersection of two major course themes: orthogonal decomposition from linear algebra and solving differential equations with boundary conditions. When you represent a periodic function as a sum of sines and cosines, you're projecting that function onto an orthonormal basis. It's the same concept you learned with vectors, just extended to infinite-dimensional function spaces. This technique transforms impossible-looking PDEs (heat equation, wave equation) into manageable systems of ODEs.

You're being tested on your ability to compute coefficients, recognize convergence behavior, and apply these series to solve boundary value problems. Don't just memorize the integral formulas. Each coefficient captures how much of a particular frequency "lives" in your function, and orthogonality is what makes the whole decomposition work cleanly.

---

The Foundation: Series Definition and Coefficients

Any reasonable periodic function can be written as a (possibly infinite) sum of sines and cosines. The coefficients tell you the "weight" of each frequency component.

Definition of Fourier Series

For a function with period $T$, the Fourier series takes the general form:

$f(x) = \frac{a_0}{2} + \sum_{n=1}^{\infty} \left(a_n \cos\!\left(\frac{2\pi nx}{T}\right) + b_n \sin\!\left(\frac{2\pi nx}{T}\right)\right)$

This expresses a periodic function as a superposition of harmonic oscillations. The term $\frac{a_0}{2}$ is the average value (the "DC component"), and every other term oscillates at an integer multiple (a harmonic) of the fundamental frequency $\frac{2\pi}{T}$.

Note: Many textbooks work on the interval $[-\pi, \pi]$ with $T = 2\pi$, which simplifies the formula to $f(x) = \frac{a_0}{2} + \sum_{n=1}^{\infty}(a_n \cos(nx) + b_n \sin(nx))$. Make sure you know which convention your course uses.

Fourier Coefficients ($a_n$ and $b_n$)

The coefficient formulas extract each frequency's amplitude by integrating against the corresponding basis function:

$a_n = \frac{2}{T} \int_{0}^{T} f(x) \cos\!\left(\frac{2\pi nx}{T}\right) dx, \qquad b_n = \frac{2}{T} \int_{0}^{T} f(x) \sin\!\left(\frac{2\pi nx}{T}\right) dx$

• $a_0$ deserves special attention: plugging $n=0$ into the $a_n$ formula gives $a_0 = \frac{2}{T}\int_0^T f(x)\,dx$, which is twice the average value. That's why the series has $\frac{a_0}{2}$ out front.
• Integration over one complete period ensures you capture the full behavior of the function. You can integrate over any interval of length $T$ (e.g., $[-T/2, T/2]$), not just $[0, T]$.

Euler's Formula and Complex Form

Euler's formula $e^{ix} = \cos(x) + i\sin(x)$ bridges trigonometric and exponential representations. Using it, you can rewrite the Fourier series in complex form:

$f(x) = \sum_{n=-\infty}^{\infty} c_n \, e^{i \cdot 2\pi nx/T}$

The complex coefficients are $c_n = \frac{1}{T}\int_0^T f(x)\,e^{-i \cdot 2\pi nx/T}\,dx$. Exponentials are often easier to differentiate and integrate, making this form preferred in advanced applications.

---

The Linear Algebra Connection: Orthogonality

Orthogonality of basis functions is what makes Fourier analysis work. Just as orthogonal vectors simplify projections in $\mathbb{R}^n$, orthogonal functions let you isolate each coefficient independently.

Orthogonality of Trigonometric Functions

The key orthogonality relations on $[0, T]$ (or any interval of length $T$) are:

• $\int_{0}^{T} \sin\!\left(\frac{2\pi nx}{T}\right) \cos\!\left(\frac{2\pi mx}{T}\right) dx = 0$ for all integers $n, m$
• $\int_{0}^{T} \cos\!\left(\frac{2\pi nx}{T}\right) \cos\!\left(\frac{2\pi mx}{T}\right) dx = 0$ when $n \neq m$
• $\int_{0}^{T} \sin\!\left(\frac{2\pi nx}{T}\right) \sin\!\left(\frac{2\pi mx}{T}\right) dx = 0$ when $n \neq m$

This independence of frequency components means changing one coefficient doesn't affect others. Each term in the series is decoupled.

The projection formula for coefficients follows directly: multiply both sides of the Fourier series by a basis function and integrate. Orthogonality kills every term except the one you want, just like $\text{proj}_{\vec{v}} \vec{u} = \frac{\vec{u} \cdot \vec{v}}{\vec{v} \cdot \vec{v}}\,\vec{v}$ in $\mathbb{R}^n$.

Parseval's Theorem

Parseval's theorem is an energy conservation statement:

$\frac{1}{T} \int_{0}^{T} |f(x)|^2 \, dx = \frac{a_0^2}{4} + \frac{1}{2}\sum_{n=1}^{\infty} (a_n^2 + b_n^2)$

The left side is the average of $|f|^2$ over one period (total "energy" in the time domain). The right side is the sum of energies in each frequency mode. This is useful for checking work: if your coefficients don't satisfy Parseval's identity, something went wrong in your calculation.

---

Exploiting Symmetry: Even and Odd Functions

Recognizing symmetry cuts your work in half. Even and odd functions have simplified Fourier representations that eliminate entire families of coefficients.

Even and Odd Functions in Fourier Series

• Even functions ($f(-x) = f(x)$) have only cosine terms: all $b_n = 0$. Why? The product of an even function with $\sin(nx)$ (which is odd) is odd, and odd functions integrate to zero over symmetric intervals.
• Odd functions ($f(-x) = -f(x)$) have only sine terms: all $a_n = 0$. The product of an odd function with $\cos(nx)$ (which is even) is odd, so again the integral vanishes.
• Symmetry detection should be your first step before computing anything. It reduces computation and helps verify your final answer.

Half-Range Expansions

When your function is defined only on $[0, L]$ (not a full period), you can extend it to create a periodic function in two ways:

• Even extension (reflect across the y-axis) produces a cosine series
• Odd extension (reflect with a sign flip) produces a sine series

Boundary value problems often dictate which extension to use. Dirichlet conditions (fixed values at endpoints, like $f(0) = f(L) = 0$) call for a sine series because $\sin(n\pi x/L)$ vanishes at both endpoints. Neumann conditions (zero derivative at endpoints) call for a cosine series because the derivative of $\cos(n\pi x/L)$ vanishes at both endpoints.

---

Convergence Behavior and Limitations

Not all functions behave equally well under Fourier expansion. Understanding convergence tells you when to trust your series and where to expect trouble.

Convergence of Fourier Series

• Pointwise convergence: at points where $f$ is continuous, the series converges to $f(x)$ exactly as you add more terms
• At discontinuities, the series converges to the average of left and right limits: $\frac{1}{2}[f(x^-) + f(x^+)]$
• Dirichlet conditions guarantee convergence. A function satisfies these if it's piecewise continuous, has finitely many discontinuities per period, and has bounded variation. Most functions you'll encounter in this course qualify.

Gibbs Phenomenon

Near a jump discontinuity, the partial sums of the Fourier series overshoot by about 9% of the jump size. This overshoot persists no matter how many terms you include. As $n \to \infty$, the oscillations get narrower and concentrate closer to the discontinuity, but the peak overshoot never goes away.

If an exam asks why a Fourier approximation looks "wrong" near a jump, Gibbs phenomenon is your answer. It's a fundamental limitation of representing discontinuities with continuous sinusoidal functions, not a computational error.

---

Applications to Differential Equations

This is where Fourier Series earn their keep. The technique transforms PDEs with periodic or boundary conditions into algebraic problems.

Solving Differential Equations with Fourier Series

The standard approach combines separation of variables with Fourier Series. Here's how it works for a PDE like the heat equation $u_t = k\,u_{xx}$:

1. Separate variables: assume $u(x,t) = X(x)\,T(t)$ and substitute into the PDE. This splits the PDE into two ODEs, one in $x$ and one in $t$.
2. Solve the spatial ODE with the given boundary conditions. This produces eigenfunctions (typically sines or cosines) and eigenvalues.
3. Write the general solution as a superposition: $u(x,t) = \sum_{n=1}^{\infty} B_n \sin\!\left(\frac{n\pi x}{L}\right) e^{-k(n\pi/L)^2 t}$ (for Dirichlet boundary conditions).
4. Apply the initial condition $u(x,0) = f(x)$. This gives $f(x) = \sum B_n \sin(n\pi x/L)$, which is exactly a Fourier sine series. Compute the $B_n$ using the coefficient formula.

The superposition principle is what makes this work: because the PDE is linear, you can solve for each Fourier mode separately and sum the results.

Fourier Transforms and Their Relationship

• Non-periodic generalization: Fourier transforms extend the series concept to functions on $(-\infty, \infty)$ by letting the period $T \to \infty$
• Continuous spectrum replaces discrete frequencies; instead of coefficients $c_n$, you get a continuous function $\hat{f}(\omega)$

---

Quick Reference Table

---

Self-Check Questions

1. Why does orthogonality of sine and cosine functions allow you to compute each Fourier coefficient independently? How is this analogous to projecting onto orthogonal vectors?

2. Given a function with a jump discontinuity, what value does the Fourier series converge to at that point, and what phenomenon affects the approximation nearby?

3. When would you use a half-range cosine expansion versus a half-range sine expansion? What boundary conditions does each naturally satisfy?

4. If $f(x)$ is an odd function, which coefficients are automatically zero? Explain why using the properties of odd and even functions under integration.

5. A function $f(x)$ on $[0, \pi]$ satisfies $f(0) = f(\pi) = 0$. You need to solve the heat equation with these boundary conditions. Would you use a sine series or cosine series expansion? Justify your choice and outline how Fourier's method transforms the PDE into solvable components.