Since I did not found it anywhere here (even though I thought this would be a known approach), here is a method to compute Julia-sets for continous functions:
Known definitions from the Julia- and Fatou-set (taken from the german wiki-page and translated):
Fatou-SetThe initial values from the Fatou-set iterated over a function lead to a static dynamic behaviour. So if the initial value changes a bit, the dynamic flow of the iterated function shows a similar behaviour.
Julia-SetThe points from this set lead to instable dynamics: Every inifinitesimal small change of the initial value leads to a completely different dynamic.
To further explain this behaviour, I am going to show this on an example later on.
To make a numerical approach we have to balance the complex plane, since functions can also have this static behaviour when converging to ∞, but we can't describe this using the normal complex numbers.
A possible way of doing so is using the Riemann Sphere. We use following moebius-transformation to get the height of every point z = a+bi on the Riemann Sphere:
Zriem = (abs(z)^2 - 1)/(abs(z)^2 + 1) = (a^2+b^2 - 1) / (a^2+b^2 + 1)After transformation we have the Complex plane + ∞ as the unit sphere. All further definitions are applied on this sphere.
Numerical approach to qualify a pointAfter a certain number of iterations n0 the following behaviours occur for points Z of the Fatou- and Julia-set.
For every number of iterations n greater then n0 in relation to its neighbourpoint Zp we have for the Fatou-set:
And for the Julia-set:
d(Zn,Zpn) is not beeing calculated on the complex plane. We have a sphere so we would have to use spherical distances.
Numerical approach:
We calculate the distance by the subtraction of the angles of the z-values. Therefore we have a maximum distance for j of pi (from d(∞,0)).
Through this method we get two escape values, one for each set, and therefore can use two different color gradients.
It is quite slow, but produces a lot of new patterns and the possibility to extend it to quaternion numbers.
This approach produces nice Julia-fractals for all functions, especially the continous ones. (Polynom with positive and negative exponent, sin cos tan (sinh cosh tanh) and exp).
For multivalued functions we get non-exact results.
Comparison to known examples from the quadratic Julia-setJulia-set for z^2 + c with c = -0.8 +0.2i (in red, Fatou-set in blue)
Julia-set for z^2 + c with c = -0.2 +0.3i (in red, Fatou-set in blue)
Some complex and beautiful examples from my gallery:
Painted Julia fractalRed ButterflySmooth rational JuliaCredits to
Ijon (german site with mathematical explanation of the properties) for the mathematical backround.