Question: Can we improve on the accuracy of this fractal graphic?

I am trying to reproduce this fractal graphic by Daniel Geisler, using maple 18:

I suspect the images were produced by fractint - with a floating point hardware accelerator, but I am not sure. In any case, they were produced by a numerical program, not a symbolic one and probably with 16-18 digits of accuracy.

The Maple code I use makes for a valiant effort, but is nowhere close in accuracy to the two images above. Here's the Maple 18 code:

Several questions here:

1) Geisler's image colors according to period (right graph). Period 1 is red. 2 is yellow (although not all yellow regions are period 2) 3 is green, 4 is cyan, 5 is blue, etc.

I have some period checking code in my doc, which stores the last 20 iterates of the maxIter iterations of the orbit in orb[n]  then extracts a period p, based on some backwards comparisons.

I am only interested in graphing max period up to N=20, hence the orb[1..N] array, stores only the last N elements of the iterates. This is relatively fast, but extracting the period (below) from the orb array, puts an additional strain on the overall calculations, probably of the order of O((N+p)*n) for each n in the iteration, where p is the period returned. That's why I avoid storing in orb unless the global bound "bail" has not been exceeded.

The period finding algorithm works. It rerurns p correctly, and that's the value the graphic proc returns for plot3d. The problem is that the range of values is small (p\in[1..N]) and therefore the palette in plot3d is poor. I.e., it doesn't differentiate very much between colors of different period regions. How can I spread out the color pallete with plot3d with such results? ideally, I'd like to assign colors roughly from red to violet, with period 1 red, 2 orange, 3 green, 4 blue, 5 purple, etc.

I tried returning log(p) and this spreads the zhue some, but not enough (see attached document). Further, I'd like to assign black to points which escape the calculations, such as when the bail bound is exceeded in my iteration loop. By default, these show up as "purple", which is the same as the color for period 1 with the latest implementation. How can I except these escape points from the palette, i.e. what does the fractal proc have to return to color these black? Note that for these points, orb[n] defaults to 0, so the palette colors this close to p=1 region (these are the black "hairs"/threads of the Cantor Bouquets in Geisler's second image. That's the first question.


Second question is, why are there "white" regions interlaced under this scheme? is the calculation failing at these points because of low resolution? (epsilon~0.01 in this doc). (If points fail "bound"=1e10 then the orbit is orb[1..N]=0, so p=1 and zhue=1?) White seems to decrease as I increase epsilon and maxIter, but i can't test further, because the calculation limit is already prohibitive for this version of Maple. The graph with a plot3d grid [400,400] and maxIter=100 and epsilon=0.01, takes around 25 minutes. Testing anything finer than than, is time uneasonable.

Would increasing Digits help with any of the above?

Does anyone have any ideas on how to improve the code performance for this graph in maple 18, as per the above problems?.

Many thanks.

Please Wait...