• March 07, 2021, 09:43:28 AM

Login with username, password and session length

Author Topic: (Question) How to tell if you are in a cardioid or a bulb in the Mandelbrot set.  (Read 710 times)

0 Members and 1 Guest are viewing this topic.

Offline pauldelbrot

  • 3f
  • ******
  • Posts: 2537
(Question) Re: How to tell if you are in a cardioid or a bulb in the Mandelbrot set.
« Reply #15 on: January 26, 2021, 06:19:32 PM »
Does anyone know of a period for which there are more bulbs than cardioids?

2 (one bulb, no cardioids) and 3 (two bulbs, one cardioid). 4 is a tie (three bulbs, three cardioids), 1 has a cardioid and no bulbs, and at 5 there are four bulbs and 11 cardioids. The cardioids explode from 6 on upward.

On one side, there are countable infinitely many bulbs and as well cardioids. So in a sense, they are "equal in numbers". On the other hand, every minibrot consists of one cardioid and infinitely many attached bulbs, so in that sense, there are vastly more bulbs.

Bulbs occur in arithmetic progressions. Start with a cardioid or a bulb and its bond point cusp. About midway between the cusp and itself will be a single bulb the sum of the relevant periods (so, 2x the period of the parent in this case). That bulb's bond point becomes a second cusp. Between this cusp and the parent bond cusp, going in either direction, there will be a single bulb the sum of the relevant periods (so this time, 3x the period of the parent). That gives two more cusps, and four more points midway between pairs of cusps, and so forth.

So if you have a bulb, its period is a sum of adjacent larger bulbs or parent-component cusps that all have smaller periods. Any larger bulbs have the same property. So the period of a bulb is just the sum of the parent component period with itself some number of times. The way that sum is parenthesized describes the specific bulb, relative to its parent. So, for a bulb of period 5 directly on the main cardioid that sum is 1 + 1 + 1 + 1 + 1. If we parenthesize it as 1 + (1 + 1 + 1 + 1) then it's between a period 1 bond point and a period 4 one. So, between elephant valley and a period 4 bulb. If we parenthesize it as (1 + 1) + (1 + 1 + 1) it's between bulbs of periods 2 and 3, so the starfish bulbs above and below the seahorse valleys. Those are the only two places to split the sum into two smaller parts, plus their symmetric counterparts, so there's just the four bulbs.

For a generic period p, there will be a number of bulbs about equal to p itself on the main cardioid. (Actually, equal to the number of invertible numbers mod p, in practice; so, never more than p - 1 and usually not much less than that.) For bulbs of period p elsewhere, for each divisor k of p, there will be about p/k bulbs of period p directly on each component of period k, the latter numbering a bit fewer than 2k-1. Out of the 2p-1 components of period p, then, the number of bulbs is bounded above by \( \sum_{k:k|p}\frac{p}{k}2^{k-1} \). Since divisors are fairly sparse, and get sparser the bigger a number is (scaling as the logarithm of that number), this will tend to be well below the sum over every k < p of 2k-1, which is 2p-1 - 1, which is the number of components of period dividing p in total, less one. The "missing" components all have to be cardioids. In particular, p - 1 doesn't divide p when p > 2 and contributes half of the latter sum, and p - 2 doesn't divide p when p > 4 and contributes another quarter, so for p > 4 at least half and typically over 3/4 of the components are going to be cardioids.

So, there is a sense (counting components of a period) where cardioids vastly outnumber bulbs.

This all goes to show how counterintuitive infinite sets can be. Whether there's more of one infinite subset or another usually depends on the order they get counted in (always, when the subsets have the same infinite cardinal, and in particular when the subsets and parent are all countable and infinite).

What would happen to the ratio of bulbs up to a certain period divided by the number of cardioids? Will that value have a limit, grow unboundedly or simply jumping around?

Long run? Asymptotic to zero. The largest proper divisor of a number is no more than half of it, so the bulbs are bounded above (usually very loosely) by p(1 + 2p/2 - 1), which is order of sqrt(number of components of period p) for large enough p, which makes the bulb to cardioid ratio stay below 1/sqrt(number of components of period p) past a fairly small p, empirically determinable to be about 5.

Offline marcm200

  • 3c
  • ***
  • Posts: 927
Re: How to tell if you are in a cardioid or a bulb in the Mandelbrot set.
« Reply #16 on: January 28, 2021, 09:45:24 PM »
@pauldelbrot: Sorry for the delayed reply, I had trouble setting up my new hardware.

Yes, infinite sets are weird - but fascinating.

I was thinking about a method to count bulbs and cardioids so that the ratio reflects what I see when looking at a Mandelbrot set image: bulbs everywhere, but the eye catches the (few) minibrots. I turnt to an area-driven method (as opposed to the period-driven one from my last post).

Suppose we compute the Mset at a finite resolution, a finite maxit with finite precision in a point-sampled manner. We judge the components first by some filter (not too small in size, interior DE Koebe radius around the hyperbolic center not too small, only points where a period can be found). The components after this filtering are then judged with the w-method I talked about earlier. Then we calculate the ratio bulbs / cardioids.

This process is repeated in a higher image size, with a (slightly) increasing max it and precision. That way one catches more and more components and judges them more and more correctly.

I did a quick numerical test for 2^10 to 2^16 pixel image (max it constant 5000, double precision) which led to the values:

Code: [Select]
level   # bulbs   # cardioids   ratio b/c
10      3         1             3.0
11      4         1             4.0
12      7         2             3.5
13      28        3             9.3
14      87        5             14.5
15      194       9             21.6
16      465       22            21.1

The values are above 1, they grow and I took it as a slight hint to what I was expecting - unbounded ratio, but just at the computational limit for the current naive implementation, there's a slight drop. This would call for a larger precision and max it. Or maybe only now the minibrots' cadioids start to appear and be of a decent size so that the ratio in the lower levels are incorrect per se.

Offline claude

  • 3f
  • ******
  • Posts: 1793
    • mathr.co.uk
Bulbs and Cardioids
« Reply #17 on: January 28, 2021, 10:13:53 PM »
What is area(bulbs) / area(cardioids) ?

Using size estimates I get 0.27645 but those size estimates would give a Feigenbaum constant of 4 instead of 4.66 so take it with salt...
« Last Edit: January 28, 2021, 11:28:56 PM by claude »

Offline marcm200

  • 3c
  • ***
  • Posts: 927
Re: Bulbs and Cardioids
« Reply #18 on: January 29, 2021, 12:11:23 PM »
What is area(bulbs) / area(cardioids) ?

I'm working on that forum link - the values there using a visual classification as bulb or cardioid, but am currently implementing a reliable version of the w-method to be used there.

The numerical value for L15 is 0.2878 computed on 7 cardioids and 170 bulbs (long double precision, classification uses 25000 maxit, components whose smallest multiplier is above 0.01 were excluded from analysis - the hyperbolic center is then not approximated well enough and bulbs appear as cardioids. In these small numbers of components, the area ratio changes quite strongly with inclusion or exclusion of components.

(But I stop adding new filters now, it's too much of fine-tuning to the result I want to see)

"Time Span"

Started by cricke49 on Fractal Image Gallery

0 Replies
Last post August 02, 2018, 07:05:21 AM
by cricke49
Mandelbrot ... but with no main cardioid!

Started by quadralienne on Share a fractal

6 Replies
Last post August 02, 2019, 11:21:43 PM
by quadralienne
Neural Style Transfer enhanced Mandelbrot zoom

Started by schizo on Fractal movie gallery

6 Replies
Last post December 22, 2020, 09:31:14 PM
by schizo
A new style of fractal imagery using fractal neural style transfer

Started by iRyanBell on Fractal Image Gallery

3 Replies
Last post October 03, 2020, 10:50:39 PM
by Jimw338
How to color 3D Mandelbrot height map images using DE colors without "spilling"

Started by Duncan C on Programming

1 Replies
Last post February 24, 2020, 04:37:54 AM
by C0ryMcG