0

My first question regarding the Fourier Series is: What are the "certain conditions" necessary for a function to be expressible as a Fourier Series? Most treatments I've come across are either very involved, or omit any further discussion of these necessary conditions.

Here is my understanding of what a Fourier Series is:

(I'm following Georg Joos's development in http://store.doverpublications.com/0486652270.html his Theoretical Physics.)

A periodic function $f:\mathbb{R\to\mathbb{R}}$ is one that may be written as

$$ f\left[t\right]=f\left[\frac{1}{\nu}+t\right]=f\left[t+T\right]=f\left[t+\frac{2\pi}{\omega}\right]; $$

where $\nu$ is the periodic frequency, $T=\frac{1}{\nu}$ is the period of one cycle, and $\omega=2\pi\nu$ is the angular frequency.

Under certain conditions, which are not usually hard to satisfy the periodic function $f$ may be written as an infinite sum known as a Fourier Series

$$ f\left[t\right]=\sum_{n=-\infty}^{+\infty}a_{n}e^{in{\omega}t}. $$

The origin of the Fourier Series lies in the orthogonality properties of $e^{in\omega t}$, where $n$ is an integer.

$$ n\ne0\implies\int_{0}^{\frac{2\pi}{\omega}}a_{n}e^{in\omega t}dt=\frac{a_{n}}{in\omega}\left(e^{in2\pi}-e^{0}\right)=0. $$

$$ n=0\implies\int_{0}^{\frac{2\pi}{\omega}}a_{0}e^{i0\omega t}dt=\frac{2\pi}{\omega}a_{0}. $$

So

$$ \int_{0}^{\frac{2\pi}{\omega}}f\left[t\right]dt=\sum_{n=-\infty}^{+\infty}\int_{0}^{\frac{2\pi}{\omega}}a_{n}e^{ni{\omega}t}dt=\frac{2\pi}{\omega}a_{0}. $$

Thus $a_{0}$ becomes

$$ a_{0}=\frac{\omega}{2\pi}\int_{0}^{\frac{2\pi}{\omega}}f\left[t\right]dt=\overline{f\left[t\right]}. $$

Multiplying $f\left[t\right]$ by $e^{in\omega t}$ ($n\ne0$) and changing the name of the summation index leads to

$$ \int_{0}^{\frac{2\pi}{\omega}}f\left[t\right]e^{-i\color{blue}{n}\omega t}dt $$

$$ =\sum_{\color{red}{m}=-\infty}^{+\infty}\int_{0}^{\frac{2\pi}{\omega}}a_{\color{red}{m}}e^{i\omega(\color{red}{m}-\color{blue}{n})t}dt $$

$$ =\int_{0}^{\frac{2\pi}{\omega}}a_{\color{blue}{n}}dt=\frac{2\pi}{\omega}a_{\color{blue}{n}}. $$

Thus

$$ \frac{\omega}{2\pi}\int_{0}^{\frac{2\pi}{\omega}}a_{n}dt=a_{n}. $$ Replacing the symbols $a_{n}$ with these values produces

$$ f\left[\color{blue}{t}\right]=\sum_{n=-\infty}^{+\infty}\frac{\omega}{2\pi}e^{in\omega\color{blue}{t}}\int_{0}^{\frac{2\pi}{\omega}}f\left[\color{red}{\alpha}\right]e^{-in\omega\color{red}{\alpha}}d\color{red}{\alpha}. $$

Since the integrals do not involve the variable $\color{blue}{t}$, the expression $e^{in\omega\color{blue}{t}}$ may be moved under the integral

$$ f\left[\color{blue}{t}\right]=\sum_{n=-\infty}^{+\infty}\frac{\omega}{2\pi}\int_{0}^{\frac{2\pi}{\omega}}f\left[\color{red}{\alpha}\right]e^{in\omega\left(\color{blue}{t}-\color{red}{\alpha}\right)}d\color{red}{\alpha}. $$

Using $T=\frac{2\pi}{\omega}$ this may be written as

$$ f\left[t\right]=\sum_{n=-\infty}^{+\infty}\frac{1}{T}\int_{0}^{T}f\left[\alpha\right]e^{in\omega\left(t-\alpha\right)}d\alpha. $$

  • The answer depends on the meaning of "expressible". I take it that it means the partial sums of Fourier series converge to $f$ in some sense, but in what sense? Pointwise convergence everywhere, convergence almost everywhere, convergence in $L^2$ norm, convergence in $L^1$ norm... –  Dec 18 '17 at 20:11

1 Answers1

1

In most general cases that I am familiar with, the functions that are in $L^1(\Omega)=\{f:\int_{\Omega}|f|< \infty \}$ have a Fourier transform. For $f:\mathbb{R}\rightarrow \mathbb{R}$ , if $f\in L^1_{loc}(\mathbb{R})$(that is locally integrable) then $f$ usually(not always) has a convergent Fourier series representation. You need more conditions than being in $f\in L^1_{loc}(\mathbb{R})$. But I think if $f\in L^2(\Omega)=\{f:\int_{\Omega}|f|^2< \infty \}$ then you get almost everwhere convergence of the Fourier series of $f$.

For more detail see

https://en.wikipedia.org/wiki/Dirichlet_conditions

BR Pahari
  • 2,694