Fourier Series in the Language of Infinite Dimensional Vector Space

5 minute read

Preliminaries

Let $\{a_n: n \in\mathbb{Z}\}$ denote a sequence of complex numbers. We define $\ell_2$ norm of $\{a_n\}$ by

\[\begin{align*} \lVert \{a_n \}_{n\in\mathbb{Z}} \rVert_{\ell_2} = \sum_{n\in\mathbb{Z}}\lvert a_n \rvert^2 \end{align*}\]

and let $\ell^2(\mathbb{Z})$ denote the vector space of all sequences whose $\ell_2$ norm is finite.

Definition 1.2

For $A=\{a_n\}$ and $B=\{b_n\}$ in $\ell^2(\mathbb{Z})$, we define the inner product

\[\begin{align*} \langle A, B\rangle =\lim_{N\to\infty}\sum_{n=-N}^Na_n \overline{b_n} \end{align*}\]

and $\lVert A \rVert := \sqrt{\langle A, A \rangle}$ denote the induced norm.

Remark

Is $\ell_2(\mathbb{Z})$ closed under addition?

<Proof>

Let $(\mathbb{C}^n, +, \cdot)$ be vector space over $\mathbb{C}$ and define inner product $\langle (a_1,\ldots, a_n), (b_1, \ldots, b_n)\rangle = \sum_{i=1}^n a_i \overline{b_i}$. We define norm $\lVert \mathbf{x}\rVert = \sqrt{\langle \mathbf{x}, \mathbf{x}\rangle}$. Let $\mathbf{x}= (a_1, \ldots, a_n)$ and $\mathbf{y}=(b_1, \ldots, b_n)$.

\[\begin{align*} \lVert \mathbf{x}+\mathbf{y}\rVert^2 &= \lVert \mathbf{x}\rVert^2 + \lVert \mathbf{y}\rVert^2 + \langle \mathbf{x},\mathbf{y}\rangle + \overline{\langle \mathbf{x}, \mathbf{y}\rangle} \\ &= \lVert \mathbf{x}\rVert^2 + \lVert \mathbf{y}\rVert^2 + 2\text{Re}\langle \mathbf{x},\mathbf{y}\rangle \\ &\leq \lVert \mathbf{x}\rVert^2 + \lVert \mathbf{y}\rVert^2 + 2 \lvert \langle \mathbf{x}, \mathbf{y}\rangle \rvert \\ &\leq \lVert \mathbf{x}\rVert^2 + \lVert \mathbf{y}\rVert^2 + 2 \lVert \mathbf{x}\rVert^2 \lVert \mathbf{y}\rVert^2 \\ &\leq \left( \lVert \mathbf{x} \rVert + \lVert \mathbf{y}\rVert\right)^2. \end{align*}\]

It implies that $\lVert \mathbf{x}+\mathbf{y}\rVert \leq \lVert \mathbf{x}\rVert + \lVert \mathbf{y}\rVert$.

Thus, for all $N\in\mathbb{N}$,

\[\begin{align*} \sqrt{\sum_{n=-N}^N \lvert a_n+b_n\rvert^2} &\leq \sqrt{\sum_{n=-N}^N \lvert a_n\rvert^2}+ \sqrt{\sum_{n=-N}^N \lvert b_n\rvert^2} \\ &\leq \lim_{N\to\infty}\sqrt{\sum_{n=-N}^N \lvert a_n\rvert^2}+ \lim_{N\to\infty}\sqrt{\sum_{n=-N}^N \lvert b_n\rvert^2} \\ &<\infty \end{align*}\] \[\tag*{$\square$}\]

Since all the sequences are monotone increasing and bounded, $\lVert A +B \rVert \in \mathbb{R}$. Thus, $\ell_2(\mathbb{Z})$ is closed under the addition.

Definition 1.3

An inner product space with strictly positive definite inner product, which is complete with respect to the induced metric, is called a Hilbert space.

Remark

$\mathcal{R}=\{ f:\text{ Riemann integrable function on } 2\pi\mid \sqrt{\frac{1}{2\pi}\int_{-\pi}^\pi \lvert f(x)\rvert^2 dx} < \infty \}$ fails to be a Hilbert space. Since $\lVert f \rVert=0$ only implies $f=0$ almost everywhere.

Theorem (Cauchy Schwarz Inequality)

<Proof>

Using the fact $2AB \leq (A^2 + B^2)$, we see that for any $\lambda >0$,

\[\begin{align*} \lvert f(x) \overline{g(x)}\rvert \leq \frac{1}{2}( \lambda \lvert f(x)\rvert^2 + \lambda^{-1}\lvert g(x)\rvert^2). \end{align*}\]

Then

\[\begin{align*} \lvert \langle f, g \rangle \rvert &\leq \frac{1}{2}\int_0^{2\pi} \lvert f(x) \overline{g(x)}\rvert dx \\ &\leq \frac{1}{2}( \lambda \lVert f\rVert^2 + \lambda^{-1}\lVert g\rVert^2). \end{align*}\]

Taking $\lambda=\lVert g\rVert / \lVert f\rVert$ yields the Cauchy-Schwarz inequality.

\[\tag*{$\square$}\]

Remark

Let $\mathcal{R}$ be integrable functions on the circle and we define an inner product

\[\begin{align*} \langle f, g\rangle := \frac{1}{2\pi}\int_0^{2\pi}f(\theta)\overline{g(\theta)}d\theta \end{align*}\]

with induced norm $\lVert f\rVert^2_2 =\langle f, f\rangle$. Let $e_n(\theta) = e^{in\theta}$. Since $\langle e^{in\theta},e^{im\theta}\rangle=\mathbf{1}\{m=n \}$, $\{e_n: n\in\mathbb{Z}\}$ is an orthonormal set.

Remark that the Fourier coefficients are precisely the coefficient of $f$ in terms of $\{e_n: n\in\mathbb{Z}\}$:

\[\begin{align*} \hat{f}(n) = \langle f, e_n \rangle. \end{align*}\]

For example, we can express the partial sum of Fourier series as

\[\begin{align*} S_N(f) = \sum_{n=-N}^N\langle f, e_n\rangle e_n. \end{align*}\]

Observation

  1. $f -\sum_{\lvert n \rvert \leq N}\langle f , e_n\rangle e_n$ is orthonormal to $e_k$ for all $\lvert k \rvert \leq N$.

  2. Using the above, we can show the Pythagorean theorem.

Since, $f-S_N(f)$ is orthogonal to $S_N(f)$,

\[\begin{align*} \langle f- S_N(f), S_N(f) \rangle &= \langle f, S_N(f)\rangle - \langle S_N(f) ,S_N(f)\rangle\\ &= \sum_{\lvert n \rvert \leq N}\langle f, e_n \rangle \overline{\langle f, e_n \rangle} - \sum_{\lvert n \rvert \leq N} \sum_{\lvert m \rvert \leq N} \langle f, e_n\rangle \overline{\langle f, e_m \rangle} \langle e_m, e_n\rangle\\ &= \sum_{\lvert n \rvert \leq N}\langle f, e_n \rangle \overline{\langle f, e_n \rangle}- \sum_{\lvert n \rvert \leq N}\langle f, e_n \rangle \overline{\langle f, e_n \rangle} \\ &=0 \end{align*}\]

Thus, we get the Pythagorean theorem as

\[\begin{align*} \lVert f \rVert^2 & = \lVert f - S_N(f) \rVert^2 + \lVert S_N(f)\rVert^2 \\ &=\lVert f - S_N(f) \rVert^2 + \sum_{\lvert n \rvert \leq N}\ \langle f, e_n\rangle \overline{\langle f, e_n \rangle} \\ &=\lVert f - S_N(f) \rVert^2 + \sum_{\lvert n \rvert \leq N}\ \lvert\langle f, e_n\rangle \rvert^2. \end{align*}\]

Best Approximation Lemma

For any $\sum_{\lvert n \rvert \leq N}c_ne_n$, we have

\[\begin{align*} \lVert f-S_N(f) \rVert \leq \lVert f - \sum_{\lvert n \rvert \leq N} c_n e_n \rVert. \end{align*}\]

<Proof>

Since

\[\begin{align*} \left\langle f-S_N(f), S_N(f) - \sum_{\lvert n \rvert \leq N} c_n e_n\right\rangle &= \left\langle f-S_N(f), -\sum_{\lvert n \rvert \leq N} c_ne_n\right\rangle \\ &=-\left \langle f, \sum_{\lvert n \rvert \leq N}c_ne_n \right\rangle + \left\langle S_N(f), \sum_{\lvert n \rvert \leq N}c_n e_n \right\rangle \\ &=-\sum_{\lvert n\rvert \leq N} \overline{c_n}\langle f, e_n\rangle+ \sum_{\lvert n \rvert \leq N}\overline{c_n}\left\langle \sum_{\lvert m \rvert \leq N} \langle f,e_m\rangle e_m, e_n \right\rangle \\ &=-\sum_{\lvert n\rvert \leq N} \overline{c_n}\langle f, e_n\rangle+ \sum_{\lvert n \rvert \leq N}\overline{c_n}\left\langle f, e_n \right\rangle \\ &= 0, \end{align*}\]

we can conclude that

\[\begin{align*} \lVert f- \sum_{\lvert n\rvert \leq N}c_ne_n\rVert^2 &=\lVert f - S_N(f) + S_N(f) - \sum_{\lvert n\rvert \leq N}c_ne_n\rVert^2 \\ &= \lVert f -S_N(f) \rVert^2 + \lVert S_N(f) - \sum_{\lvert n\rvert \leq N}c_ne_n\rVert^2 \\ &\geq \lVert f-S_N(f) \rVert^2. \end{align*}\] \[\tag*{$\square$}\]

Theorem 3.1

Let $f$ be an integrable function on the circle. Then

\[\begin{align*} \left(\frac{1}{2\pi}\int_0^{2\pi} \lvert f(\theta) - S_N(f)\vert d\theta\right)^{1/2} \to 0 \text{ as } N\to\infty, \end{align*}\]

i.e., the Fourier series converges to $f$ in the $L^2$ sense.

<Proof>

Suppose $f$ is continuous. Using Weierstrass approximation theorem, we can choose a trigometric polynomial $p_M$ of degree $M$ such that $p_M \rightrightarrows f$. In other words, for all $\epsilon>0$ there is $N_1\in\mathbb{N}$ such that if $M \geq N_1$,

\[\begin{align*} \lvert f(x) - p_M(x) \rvert < \epsilon. \end{align*}\]

It implies that

\[\begin{align*} \lVert f - P_M\rVert_{2} = \left( \frac{1}{2\pi}\int_{-\pi}^\pi \lvert f(x) - p_M(x)\rvert^2\right)^{1/2} < \epsilon. \end{align*}\]

By the best approximation lemma,

\[\begin{align*} \lVert f - S_N(f)\rVert <\epsilon \end{align*}\]

for all $N\geq M$.

Now, suppose that $f$ is integrable. Applying the approximation lemma, choose a continuous function $g$ on the circle such that

\[\begin{align*} \int_{-\pi}^\pi \lvert f(\theta) -g(\theta)\rvert d\theta < \epsilon^2. \end{align*}\]

Since we $f-g$ in integrable, $\sup_{\theta \in [0,2\pi]} \lvert f(\theta) -g(\theta)\rvert < B$.

Then we get

\[\begin{align*} \lVert f - g\rVert^2_2 &= \frac{1}{2\pi}\int_{-\pi}^\pi \lvert f(\theta) - g(\theta)\rvert^2 d\theta \\ &= \frac{1}{2\pi}\int_{-\pi}^\pi \lvert f(\theta) - g(\theta)\rvert \cdot \lvert f(\theta) - g(\theta)\rvert d\theta \\ &\leq \frac{B}{\pi} \int_{-\pi}^\pi \lvert f(\theta) - g(\theta)\rvert d\theta \\ &\leq C\epsilon^2, \end{align*}\]

where $C >0$ is some constant. Now we approximate the continuous function $g$ by a trigometric polynomial $p_M$ so that

\[\lvert p_M(x) - g(x) \rvert < \epsilon\]

for all $x\in[0,2\pi]$. As shown previously, $\lVert g- p_M\rVert_2 < \epsilon$. By the triangular ineqaulity, $\lVert f - p_M\rVert_2 < C^\prime \epsilon$. Using the best approximation lemma,

\[\begin{align*} \lVert f - S_N(f) \rVert_2 < C^\prime \epsilon \end{align*}\]

for all $N\geq M$.

$\therefore \lim_{N\to\infty}S_N(f) = f$.

\[\tag*{$\square$}\]

Reference