13

Consider two types of parametrized curves $\gamma:[0,2\pi]\rightarrow \mathbb{R}^2$

  1. open curves $\gamma_\sim(t) = (t,a(t) + b(t))$

  2. closed curves $\gamma_\bigcirc(t) = (a(t),b(t)) = a(t) + ib(t)$

with $a(t)$, $b(t)$ being $2\pi$-periodic functions, i.e. $a(0) = a(2\pi)$ and $b(0) = b(2\pi)$.

These curves – considered as geometric shapes – have some properties, for example:

  • symmetries (rotational and others)

  • number of singularities (where the curve goes to infinity)

  • number of vertices (where the curve is not smooth)

  • number of straight edges

  • being convex

  • number of self-intersection points (only for closed curves)

As a periodic function, $a(t) + b(t)$ has Fourier series

$$a_k = \int_0^{2\pi}(a(t)+b(t))\cos(kt)\mathrm{d}t$$

$$b_k = \int_0^{2\pi}(a(t)+b(t))\sin(kt)\mathrm{d}t$$

For the sake of simplicity consider only such $a(t)$, $b(t)$ with

$$a(t) = \sum_{k=0}^\infty a_k\cos(kt)$$

$$b(t) = \sum_{k=0}^\infty b_k\sin(kt)$$

(constant factors omitted), i.e. $a(t)$ being an even function and $b(t)$ being an odd function.

The Fourier series $a_k$, $b_k$ have some properties, too, for example

  • having an envelope of shape $k^{-c}$ or $c^{-k}$ for some $c\geq 1$

  • a periodic pattern by which the signs of $a_k$, $b_k$ change between $+1$, $0$ and $-1$

  • $a_k$ and $b_k$ having the same envelope

  • $a_k$ and $b_k$ having equal or shifted periodic patterns

Sticking to these sets of properties (of shapes and Fourier series) one may ask:

How do these properties relate, and what can be told about a shape knowing the properties of its Fourier series and vice versa?

Here is just one relation that I already have found:

The Fourier series

$$a_k = \begin{cases} +k^{-2} & \text{ for } k \equiv 1 \pmod n \\ +k^{-2} & \text{ for } k \equiv (n-1) \pmod n\\ 0 & \text{ otherwise } \end{cases}$$

$$b_k = \begin{cases} +k^{-2} & \text{ for } k \equiv 1 \pmod n\\ -k^{-2} & \text{ for } k \equiv (n-1) \pmod n\\ 0 & \text{ otherwise } \end{cases}$$

give rise to a regular $n$-gon (with all its symmetries, vertices, and straight edges).

My question (which is a big-list question of sort) asks for more examples, preferrably "simple" and otherwise "astonishing" ones.


Going one step further: How are relations between the Fourier series of two different shapes reflected in relations between the two shapes (the one being a rotation, homeomorphic distortion or other transformation of the other) and vice versa?

Again I have found a specific example:

The Fourier series

$$a_k = \begin{cases} +k^{-2} & \text{ for } k \equiv 1 \pmod {2n} \\ -k^{-2} & \text{ for } k \equiv (n-1) \pmod {2n}\\ -k^{-2} & \text{ for } k \equiv (n+1) \pmod {2n}\\ +k^{-2} & \text{ for } k \equiv (2n-1) \pmod {2n}\\ 0 & \text{ otherwise } \end{cases}$$

$$b_k = \begin{cases} +k^{-2} & \text{ for } k \equiv 1 \pmod {2n} \\ +k^{-2} & \text{ for } k \equiv (n-1) \pmod {2n}\\ -k^{-2} & \text{ for } k \equiv (n+1) \pmod {2n}\\ -k^{-2} & \text{ for } k \equiv (2n-1) \pmod {2n} \\ 0 & \text{ otherwise } \end{cases}$$

give rise to a regular $n$-gon rotated by $\pi/n$.

And then this one:

The Fourier series

$$a_k = \begin{cases} +\gamma^{-k} & \text{ for } k \equiv 1 \pmod 4 \\ +\gamma^{-k} & \text{ for } k \equiv 3 \pmod 4\\ 0 & \text{ otherwise } \end{cases}$$

$$b_k = \begin{cases} +\gamma^{-k} & \text{ for } k \equiv 1 \pmod 4\\ -\gamma^{-k} & \text{ for } k \equiv 3 \pmod 4\\ 0 & \text{ otherwise } \end{cases}$$

with $\gamma = \sqrt{2+2\sqrt{3}+\sqrt{15+8\sqrt{3}}}=3.291795...$ gives rise to a square with rounded corners.

Further examples are welcome.


I wrote a little tool to play around with Fourier series and compare them with the shapes they give rise to. The following gallery shows by some simple examples what can be done and observed with it. If you are interested in using it by yourself, just drop me an email.


enter image description here


enter image description here


enter image description here


enter image description here


enter image description here


enter image description here


[There is a follow-up question to this one: The number $\pi$ in an unexpected context.]


Edit: I would guess the following is true.

The shape $\gamma_\bigcirc(t)=(a(t),b(t))$ defined by the series $a_k$, $b_k$ is symmetric with respect to the y-axis iff $a_k, b_k = 0$ for even $k$.

In this case $\gamma_\sim(t) = (t,a(t)+b(t))$ is an odd function, i.e. invariant under rotation by $\pi$ around the point $(\frac{\pi}{2},0)$.

  • 2
    For the big-list you desire, dig into the abundant literature on "Fourier descriptors" (see, among many papers in the 1990s-2000s, for example https://cis.temple.edu/~lakaemper/courses/cis595_2004/papers/fourierShape.pdf) which has had its years of glory in shape recognition. – Jean Marie Apr 05 '19 at 13:00
  • @JeanMarie: My question obviously has to do with Fourier descriptors (see this funny video by Mathologer) - but it goes into another direction: I'm mainly concerned not with arbitrary shapes but with highly regular ones (with lots of symmetries), related to highly regular Fourier series (describable by a small handful of parameters). Are you interested in having a look at the tool I mentioned? – Hans-Peter Stricker Apr 05 '19 at 14:28
  • It's not quiet the same thing, but ... Part of my personal research has been in "spectral realizations of graphs". This answer of mine discusses the basic idea. In a way, it might be considered a discrete form of the stuff you are doing. – Blue Apr 09 '19 at 09:11
  • $+100$ points ought to be merely a moderate price for just reading your post. – uniquesolution Apr 16 '19 at 07:25
  • @uniquesolution: I did not understand your comment, but anyway, I just started a bounty worth +200 points;-) – Hans-Peter Stricker Apr 20 '20 at 09:15
  • wow, I didn't know this stuff. thanks a lot for the question! – mathworker21 Apr 20 '20 at 09:39
  • If you replace the $a(t),b(t)$ with any monotonic $a(f(t)),b(f(t))$ that has $f(0)=0;f(2\pi)=2\pi$ you will have the same shape but different Fourier components. – user619894 Apr 27 '20 at 08:50
  • @user619894: Is this so, resp. why is this so? (I wonder.) – Hans-Peter Stricker Apr 27 '20 at 08:58
  • Are you asking why the shape is the same or why the Fourier components are different? – user619894 Apr 27 '20 at 09:20
  • @user619894: First why the shape is the same. (Maybe it's obvious and I'm only stubborn.) – Hans-Peter Stricker Apr 27 '20 at 09:21
  • @user619894: Supposing you are right, the Fourier components of the same shape may be different but still might share the same general features. If not so, my question becomes obsolete, right? – Hans-Peter Stricker Apr 27 '20 at 09:23
  • suppose you have a shape, $a(t),b(t)$ you can reparametrize the "speed" $t$ to go faster or slower along the shape, and still have the same shape. As to whether the same general features are shared, I guess that depends on "feature", so your question is still valid, and interesting. – user619894 Apr 27 '20 at 10:12
  • @user619894: Ah, I see, that's perfectly true, I'll have to digest this. And maybe the question becomes even more interesting? – Hans-Peter Stricker Apr 27 '20 at 12:13
  • in fact there is a "canonical" speed for curves, the speed at which "$dt$" induces a fixed length change "$dl$", e.g. https://en.wikipedia.org/wiki/Frenet%E2%80%93Serret_formulas – user619894 Apr 27 '20 at 12:50

0 Answers0