Further to the above, the change in behavior is because the period-"doubling" divides ln(
z - P) by the power
n from
zn +
c, where P is the center of the minibrot. That makes angular features repeat
n times, so with the quadratic M-set they double, then double again, etc. approaching the minibrot, but in
z3 they triple, then triple again, and etc. Meanwhile, the logarithm of magnification shrinks similarly: depth to first doubling in quadratic is halved to get further depth to second doubling. With the cubic M-set it's reduced to 1/3.
The actual depth of the minibrot can be calculated from the depth of the final eccentric zoom by noting that the above is a geometric series with ratio 1/
n, which converges on
which is, as plynch27 said,
n/(
n - 1).
So if the last eccentric zoom is at some depth, the further depth to the minibrot is equal to that for
z2, only half that for
z3, a third for
z4, and so forth.
Put another way, if you want the section from the final off-center zoom to the minibrot to have a given length, the length of the zoom
to that off-center zoom has to be larger by a factor of
n - 1. In other words by how many lobes the minibrot has! The two-lobed
z3 requires double, for example. So if you want the zoom from doubling on to be a factor of 10
50 you need to go off-center from time to time all the way down to 10
100. With three-lobed
z4 it would have to be 10
150. And so forth.
This makes getting those lengthy zooms to minibrots
much slower with higher powers, too, because compute time contributed by multiplies goes up quadratically with log mag and the number of multiplies per iteration goes up as log power -- for example,
z8 requires three successive squarings. So if you want the final segment from last off-center zoom to minibrot to be of a fixed length, the time needed for the whole zoom sequence scales as
n(log
2 n)
2. The first few such numbers, divided by that for
n = 2, are:
n | | time |
2 | | 1 |
3 | | 3.768159193038392 |
4 | | 8 |
5 | | 13.478375194568137 |
6 | | 20.04609339040372 |
7 | | 27.584345804403696 |
8 | | 36 |
9 | | 45.2179103164607 |
10 | | 55.176031338009906 |
11 | | 65.8221691821511 |
12 | | 77.11173678946133 |
It's growing slower than quadratic, but substantially faster than linear. In reality it's not quite as tidy; the actual number of multiplies is never fractional, so the log
2 ceiling should really be taken, which gives this corrected table:
n | | time |
2 | | 1 |
3 | | 6 |
4 | | 8 |
5 | | 22.5 |
6 | | 27 |
7 | | 31.5 |
8 | | 36 |
9 | | 72 |
10 | | 80 |
11 | | 88 |
12 | | 96 |
For
z12, this predicts a nearly 100-fold slowdown for a fixed depth from final off-center zoom to minibrot compared to for the plain Mandelbrot set, with a 12-times deeper zoom overall.
So, basically what everyone else said, but putting concrete numbers to "minibrots approach much quicker".