There is no bilaterally-symmetrical, nor eccentrically-periodic curve used in any branch of astrophysics or observational astronomy which could not be smoothly plotted as the resultant motion of a point turning within a constellation of epicycles, finite in number, revolving around a fixed deferent.
A friend recently showed me this image…
…and thus I was drawn into the world of epicycles and fractals.
Epicycles were first used by the Greeks to reconcile observational data of the motions of the planets with the theory that all bodies orbit the Earth in perfect circles. It was found that epicycles allowed astronomers to retain their belief in perfectly circular orbits, as well as the centrality of Earth. The cost of this, however, was a system with many adjustable parameters (as many parameters as there were epicycles).
There’s a somewhat common trope about adding on endless epicycles to a theory, the idea being that by being overly flexible and accommodating of data you lose epistemic credibility. This happens to fit perfectly with my most recent posts on model selection and overfitting! The epicycle view of the solar system is one that is able to explain virtually any observational data. (There’s a pretty cool reason for this that has to do with the properties of Fourier series, but I won’t go into it.) The cost of this is a massive model with many parameters. The heliocentric model of the solar system, coupled with the Newtonian theory of gravity, turns out to be able to match all the same data with far fewer adjustable parameters. So by all of the model selection criteria we went over, it makes sense to switch over from one to the other.
Of course, it is not the case that we should have been able to tell a priori that an epicycle model of the planets’ motions was a bad idea. “Every planet orbits Earth on at most one epicycle”, for instance, is a perfectly reasonable scientific hypothesis… it just so happened that it didn’t fit the data. And adding epicycles to improve the fit to data is also not bad scientific practice, so long as you aren’t ignoring other equally good models with fewer parameters.)
Okay, enough blabbing. On to the pretty pictures! I was fascinated by the Hilbert curve drawn above, so I decided to write up a program of my own that generates custom fractal images from epicycles. Here are some gifs I created for your enjoyment:
Negative doubling of angular velocity
(Each arm rotates in the opposite direction of the previous arm, and at twice its angular velocity. The length of each arm is half that of the previous.)
Trebling of angular velocity
Negative trebling
Here’s a still frame of the final product for N = 20 epicycles:
Quadrupling
ωn ~ (n+1) 2n
(or, the Fractal Frog)
ωn ~ n, rn ~ 1/n
ωn ~ n, constant rn
ωn ~ 2n, rn ~ 1/n2
And here’s a still frame of N = 20:
(All animations were built using Processing.py, which I highly recommend for quick and easy construction of visualizations.)
Is it possible to calculate the converging volumes of these epicycle fractals? Are they interesting?
That’s a great question! I don’t know how to calculate the volume that these shapes converge to, but I’m sure it’s possible. I’ll think about it. Thanks for commenting!