4

The definition of Fourier series states that

It decomposes any periodic function or periodic signal into the weighted sum of a (possibly infinite) set of simple oscillating functions, namely sines and cosines (or, equivalently, complex exponentials)

I was a little confused as to how then, non-periodic functions like $f(x) = e^{\frac{-ax}{L}}$ defined over an interval $[0,L]$ can have a Fourier expansion ? We know that $$f(x) = \frac{a_0}{2} + \sum_{n=1}^{\infty}a_n\cos(nx) + \sum_{n=1}^{\infty}b_n\sin(nx)$$

Note $\rightarrow$ $a,L$ are constants $>0$

What I fail to understand is, how is this possible for a non-periodic function like an exponential ? Somewhere on another question in MSE, I learned that The Fourier series is described for the periodic extension of a non-periodic function, which failed to clarify my doubts.

The motivation to ask this question comes from my attempts to solve

Two fluids flowing perpendicular in thermal contact with a Wall [Help to mathematically model] and

Evaluating Coefficients for a Fourier Series when Exponential terms are present [Approach needed]

Avrana
  • 141
  • What are the "standard coefficient finding formulae"? The formula of which I am aware involves integrating over a single period of a periodic formula. How do you define the $n$-th Fourier coefficient of a non-periodic function? – Xander Henderson Feb 27 '19 at 05:03
  • If you write down the formula for the Fourier coefficients of a non-periodic function, you get a representation of the Fourier series of its periodic extension (which is a discontinuous function even if the original function was continuous). – Ian Feb 27 '19 at 05:31
  • @Ian My recollection is that the periodic extension of a function refers to the extension of a function which is defined on a finite interval to a function defined on $\mathbb{R}$. For example, the function $$ f : (-\frac{1}{2},\frac{1}{2})\to\mathbb{R} : x \mapsto x $$ is non-periodic, but has a sawtoolh function as its periodic extension. This is neither here nor there with respect to the question of a function defined on all of $\mathbb{R}$. – Xander Henderson Feb 27 '19 at 05:41
  • @XanderHenderson I have added the relations I used to find the coefficients – Avrana Feb 27 '19 at 05:41
  • @IndrasisMitra In those formulae, $L$ the the period of the function. How are you choosing $L$ when $f$ is a non-periodic function defined on $\mathbb{R}$? – Xander Henderson Feb 27 '19 at 05:42
  • @XanderHenderson Yes i will admit that I choose $L$ just because it is the region where I am trying to find the expansion. Can you elaborate a bit more on how then,I should go about with representing this as a Fourier expansion ? Or is it even possible ? – Avrana Feb 27 '19 at 05:44
  • The usual idea is to take a Fourier transform, which can be thought of as a continuous analog of the Fourier series. However, it might help to know why you want a Fourier "series" representation of this function. This is beginning to smell like an XY Problem. – Xander Henderson Feb 27 '19 at 05:47
  • @XanderHenderson You have guessd dexactly right. I am basically trying to find the Fourier coefficents at the last step of finding a Laplace equation solution where these terms crop up. You can check out this question which I posted https://math.stackexchange.com/questions/3125729/evaluating-coefficients-for-a-fourier-series-when-exponential-terms-are-present . Is there some way I could mitigate this term – Avrana Feb 27 '19 at 05:51
  • @XanderHenderson For a complete description of the physical phenomena check this https://math.stackexchange.com/questions/3117489/two-fluids-flowing-perpendicular-in-thermal-contact-with-a-wall-help-to-mathema – Avrana Feb 27 '19 at 05:53
  • @XanderHenderson The advantage of this perspective (that Fourier series really are defined on $\mathbb{R}$, with the coefficients being computed on a finite interval) is that it creates intuition for $L^2$ convergence rates and for pointwise behavior in the vicinity of discontinuities. – Ian Feb 27 '19 at 18:15
  • @Ian To be more specific can you provide some guiding points on how to find Fourier coefficients for this problem https://math.stackexchange.com/questions/3125729/evaluating-coefficients-for-a-fourier-series-when-exponential-terms-are-present Some guiding points is all I need. – Avrana Feb 27 '19 at 19:10

2 Answers2

1

Thanks to Ian and Xander Henderson for their suggestions.

We can consider any function defined on a finite interval $(a,b)$ or $[a,b]$ as a periodic function defined on $R$ by thinking that the function is extended to $R$ by repeating the values in $[a,b]$ to the remaining part of $R$.

Thus for $f(x)$ defined on $[a,b]$ where $(\frac{b-a}{2}) = l$, we have

$$f(x) = \frac{a_0}{2} + \sum_{n=1}^{\infty} \bigg[ a_n \cos\bigg(\frac{n\pi x}{l}\bigg) + b_n \sin\bigg(\frac{m\pi x}{l}\bigg)\bigg]$$

where,

$$a_0 = \frac{1}{l} \int_a^b f(x) \mathrm{d}x$$

$$a_n = \frac{1}{l} \int_a^b f(x)\cos\bigg(\frac{n\pi x}{l}\bigg)\mathrm{d}x$$

$$b_n = \frac{1}{l} \int_a^b f(x)\sin\bigg(\frac{n\pi x}{l}\bigg)\mathrm{d}x$$

Applying these for our function $f(x) = e^{-\frac{ax}{L}}$ defined on $x \in [0,L]$.

Hence, $a = 0$,$b = L$ and $l=\frac{L}{2}$, leads us to:

$$e^{\frac{-a x}{L}} = \frac{(1-e^{-a})}{a} + \sum_{n=1}^{\infty}\bigg[\frac{2a(1-e^{-a})}{(a)^2 + (2n\pi)^2}\cos\bigg(\frac{2n\pi x}{L}\bigg) + \frac{4n\pi(1-e^{-a})}{(a)^2 + (2n\pi)^2}\sin\bigg(\frac{2n\pi x}{L}\bigg)\bigg]$$

Avrana
  • 141
1

For representing a non-periodic function on the interval $[0,L]$, we use \begin{equation*} f(x)=k_1+\frac{k_2x}{L}+\sum_{n=1}^{\infty} \left(A_n\cos\frac{2n\pi x}{L}+B_n\sin\frac{2n\pi x}{L}\right). \end{equation*} By evaluating the above series at $x=0,L$, we get \begin{equation*} k_2=f(L)-f(0). \end{equation*} Let \begin{equation*} \tilde{f}(x)=f(x)-\frac{k_2x}{L}. \end{equation*} Then using the orthogonality of the sine and cosine functions, we get \begin{align*} k_1&=\frac{1}{L}\int_0^L \tilde{f}(\hat{x})\,d\hat{x}, \\ A_n&=\frac{2}{L}\int_0^L \tilde{f}(\hat{x})\cos\frac{2n\pi\hat{x}}{L}\,d\hat{x}, \\ B_n&=\frac{2}{L}\int_0^L \tilde{f}(\hat{x})\sin\frac{2n\pi\hat{x}}{L}\,d\hat{x}. \end{align*} For the function $e^{-ax/L}$, we get \begin{equation*} f(x)=\frac{(a+2)(1-e^{-a})}{2a}+\frac{(e^{-a}-1)x}{L}+a(1-e^{-a})\sum_{n=1}^{\infty} \frac{2n\pi\cos\frac{2n\pi x}{L}-a\sin\frac{2n\pi x}{L}}{n\pi[a^2+(2n\pi)^2]}. \end{equation*} For why the $x$ term should be present while representing non-periodic functions, see https://www.awesomemath.org/wp-pdf-files/math-reflections/mr-2022-03/mr_3_2022_fourier.pdf

With the $x$ term present, you will get a much faster rate of convergence compared to when it is not present.

Jog
  • 584
  • Thankyou for the answer and the link to the reference. Much appreciated. – Avrana Apr 04 '25 at 22:54
  • Thanks for accepting! Glad to know that you benefited from my article. Took me quite a while to write that article since it breaks away from conventional wisdom, and so had to be doubly sure that I was not making a mistake. – Jog Apr 06 '25 at 09:06
  • It is indeed a well-written article. I am still unpacking it and might come back to you with doubts. I had posed this (and related) questions back during my PhD, where this problem came up as a natural consequence of a cross-flow heat exchanger analysis I was dealing with. My thesis is done, but the interest and use of PDEs still remain. – Avrana Apr 07 '25 at 21:58
  • 1
    Yes sure. I would also enjoy discussing this topic. My e-mail is jogc@iisc.ac.in in case you need to get in touch via e-mail. This Fourier stuff has major implications in heat conduction problems. For example, see my answer at https://math.stackexchange.com/questions/4724382/laplace-equation-with-given-boundary-conditions/5035485#5035485 – Jog Apr 09 '25 at 03:03