16 comments

[ 4.3 ms ] story [ 54.2 ms ] thread
Great writing!

Blog instantly added to my feedly.

This is beautiful. Can any shape be expressed as a mathematical function?
Yes! Well, as long as the shape is closed and continuous (even if it is 3-dimensional). I should've clarified that in the post.
Very well written article!

If you are interested in interactive explorations of the fourier transform you might also like this visualization I built [1], featuring the not so well known fractional fourier transform.

[1]: https://static.laszlokorte.de/frft-cube/

Thank you for posting my blog here!

I've improved my writing over time, so reading an older post makes it feel dated. I'll try to post more rigorous technical articles this year.

It is very well written. I've learned more about this topic in this article than in some classes in college. Thank you!
This is amazing. Very nice introduction to Fourier series (although i prefer Chebyshev!).
The latest incarnation of Fourier series is in AI, in the so called RoPe technique.
The FFT is present in some of the pre-Mamba state space models as well to optimize applying the convolution.
I recommend learning complex numbers and the problems they can solve before touching on the main topic, Fourier series.
I considered using the complex Fourier series for the epicycles, but also wanted the article to be self-contained and concise.

I might add that in a rewrite of the article if more people feel this way.

"...You may be wondering - the functions and aren't periodic, how come we can still decompose them into sine/cosine sums? One trick is to set the period to infinity, and compute the series at this limit."

Wouldn't this effectively make frequency 0?

Later on for the drawings you pick period T to be equal to 1, thus frequency to be 2*pi. This does make more sense both practically and mathematically too.

P.S. A small nit, a typo "f_t and y_t" should rather be "f_x and f_y". Very fun read, thanks!

(comment deleted)
> Wouldn't this effectively make frequency 0?

Yes! Letting the period length approach infinity, would make the lowest frequency (omega_0, also delta each other frequencies) approach 0, effectively turning the discrete sum of the fourier series into an integral, turning the fourier series into the continous fourier transform.

> ...turning the fourier series into the continous fourier transform

Yes, if talking about the Fourier transform. My understanding was that author proposed that "trick" in context of discrete series for bounded non-periodic functions x(t), y(t).

Instead, what followed was more like an extension to periodic function.

(comment deleted)