Julia Set

If you start with 0.50.5 and keep squaring, you get 0.250.25, then 0.06250.0625, and the numbers keep shrinking toward zero. If you start with 22, you get 44, 1616, 256256, and they grow without limit. You can drag the ring along the line to try other numbers.

Keep squaring a number

0.5 → 0.25 → 0.0625 → 0.00391 → 0.0000153 → …

Heads toward 0

Everything between −1-1 and 11 gets drawn in toward 00, and everything outside grows without limit. The switch is abrupt: 0.990.99 heads to 00, while 1.011.01 flies off forever.

We can ask the same question about points on a plane. A point has a horizontal and a vertical coordinate, and together they form a complex number. To square one, you multiply its distance from zero by itself and double its angle from the line pointing right.

As you drag the ring below, the line joins the point to its square, then to the square of that, and so on. We call each step a jump and the whole path the point's orbit, and the chart shows the point's distance from zero after each jump.

Square a point on the plane

Stays for all 300 jumps

Start = (0.8, 0.55)

Distance from zero at each jump

Inside the circle, every orbit heads toward 00. Outside, the distance from zero grows without limit, and we say the point escapes. Close to the circle, the orbit takes longer to go either way, because 1.0011.001 squared is only about 1.0021.002. So the circle is the boundary between two groups of points, and that boundary is the simplest Julia set.1

One small change to the rule makes this much more interesting. Start at a point, z0z_0, then square it and add a fixed number cc to get z1z_1. Do the same to z1z_1, with the same cc, to get z2z_2, and keep going:

zn+1=zn2+cz_{n+1} = z_n^2 + c

In coordinates, if the current point is (x,y)(x, y) and cc is (a,b)(a, b), the next point is:

(x2−y2+a, 2xy+b)(x^2 - y^2 + a,\ 2xy + b)

In the demo below, the ring is the starting point and the cc dot is the number you add, which starts at about (−0.12,0.74)(-0.12, 0.74).

The old circle no longer separates the two groups. If you drag the ring to (0.45,0.05)(0.45, 0.05), it starts inside the circle and escapes, and at (1.1,−0.6)(1.1, -0.6), it starts outside the circle and stays. Points still split into two groups, but you can't tell where the boundary is yet. See if you can find it by dragging the ring around, and then turn on Show the edge.

Square, then add c

Stays for all 300 jumps

Start = (−0.3, 0.3)

c = (−0.12, 0.74)

Distance from zero at each jump

That edge is the Julia set for this cc. If you drag the cc dot onto 00, the rule is plain squaring again and the edge is the circle. As you drag cc away slowly, the circle at first only bends, then farther out it can pinch, and far enough out it breaks into dust. The buttons move cc to a few well-known shapes.

The next demo uses c=(−0.12256,0.74486)c=(-0.12256, 0.74486). The dark shape is every point that stays, with the Julia set as its edge. If you pick a pixel outside it, the point eventually crosses the dashed circle, two units from zero, and after that every jump takes it farther away.2

If you pick a pixel in the dark shape, the point ends up jumping around three places. A different pixel in the dark shape ends up on the same three places, which is something we'll come back to.

Follow one starting point

Pick a pixel to start there. Each jump squares the point, then adds c = (−0.12256, 0.74486).

JumpPoint
0(0.3, 0)
1(−0.033, 0.745)

Jump 1x² − y² + a2xy + b
Pointx = 0.3y = 0
Square0.3² − 0² = 0.092 × 0.3 × 0 = 0
Add c0.09 − 0.12256 = −0.032560 + 0.74486 = 0.74486
New pointnew x = −0.03256new y = 0.74486

Under the picture, the current jump is worked out by hand and rounded to five decimals, and Next jump feeds the new point back in as the next starting point.

You saw the edge break into dust as cc moved, so it's natural to ask which values of cc keep it in one piece. That's what the Mandelbrot set tells you. If you choose cc in the Mandelbrot set, the Julia set is connected, and if you choose it outside, the Julia set breaks into dust.3

In the demo below, each point of the left picture is one value of cc. To choose its color, the demo starts at 00 and follows the jumps, coloring the point black if it stays and blue if it crosses the dashed circle. The black shape is the Mandelbrot set.

The right picture shows the Julia set for the cc under the amber ring. The line is the orbit of one starting point, which ends when the point escapes or repeats, or after 96 jumps, and the repeating part is amber.

The demo starts on c=−1c = -1, where a point starting at 0.50.5 ends up jumping back and forth between −1-1 and 00. Moving to (0,0)(0, 0) on the left gives you the circle again, and moving outside the black shape breaks the Julia set into dust.

Each value of c has its own Julia set

Drag the ring in either picture.

Mandelbrot · choose c

c = (-1, 0)

Julia · follow a starting point

(0.5, 0)

Each arrow below starts at a value of cc on the Mandelbrot picture and points to that value's Julia set. Six of these values lie outside the Mandelbrot set, so their Julia sets are dust, even though you can't always tell from the picture. In some you can see the gaps, as with (0.45,0.1)(0.45, 0.1). In others the gaps are too small to show, as with (−0.8,0.156)(-0.8, 0.156), where a point starting at 00 takes 252 jumps to cross the dashed circle. The point (0,−1)(0, -1) looks outside too, but it sits on a thread too thin for this picture. Clicking a picture opens it at the top of this page.4

The Mandelbrot set, with an arrow from each marked value of c to its Julia set
Separate jagged bolts with wide gaps between themc = (−1.1, 0.5)Paired curls along a twisting threadc = (−0.8, 0.156)Black lobes, three at every jointc = (−0.12256, 0.74486)A solid black circlec = (0, 0)Thin golden branches forking at both endsc = (0.45, 0.35)Four rings of specks with gaps between themc = (0.45, 0.1)Pale golden lobes full of small spiralsc = (0.285, 0.01)Thin white branches with no black areac = (0, −1)A solid black shape with a jagged edgec = (−0.4, −0.45)Golden star-shaped clusters along a branching linec = (−0.70176, −0.3842)A chain of rounded black lobesc = (−1, 0)A thin line of tiny black lobes, the largest in the middlec = (−1.75488, 0)

The whole dark shape, edge and inside together, is called the filled Julia set: every starting point that never escapes. With c=0c=0 the calculation is plain squaring again, so the Julia set is the circle and the filled Julia set is the solid disk it encloses.

That makes these pictures slightly misleading, because none of them really shows the Julia set. Mathematicians split the starting points of a rule into two groups. A point belongs to the Fatou set if it and all the points close enough to it behave alike, such as a patch of points that all escape, or the points in the dark shape at c=(−0.12256,0.74486)c=(-0.12256, 0.74486), which all end up on the same three places. The Julia set is everything else: the points where, however closely you look, the neighbors around them do different things.5

To draw one of these pictures, the computer tests one point for each pixel. If the point escapes, it's in the Fatou set, and at this cc a point in the dark shape is in the Fatou set too. This Julia set has no area,6 so a test point almost never lands on it, which means what we paint is the Fatou set, and we only see where the Julia set is because the pixels on either side of it disagree. The bands of color aren't part of it either, since they only record how many jumps it took to cross the dashed circle, and a bigger circle would move every band while the edge stayed exactly where it is.

The computer can only test a limited number of steps, so dark pixels are only the ones that haven't escaped yet, and some may escape later. Take c=0.2501c = 0.2501. From 100 steps to 300 steps, this picture barely changes and shows a solid dark shape, so you'd probably assume it's finished. Try moving the slider to 320.

Give the computer more steps
Not escaped yet. It might be in the set.
Escaped, so outside the set. Color shows the step:
089 steps, then the colors repeat

Most of the shape changes color between 310 and 330 steps. Every colored pixel has escaped, so the pale area is outside the filled Julia set, and its pixels look alike because they escape at about the same step. Black and white drops the colors, so only the pixels left over stay black.

Why would so many points escape at nearly the same step? Almost all of them must pass close to 0.50.5, and that takes about 300 steps. With c=0.25c = 0.25, the point 0.50.5 would never move, because 0.52+0.25=0.50.5^2 + 0.25 = 0.5. With c=0.2501c = 0.2501, it moves by only 0.00010.0001, so jumps near 0.50.5 are very small. The point at the center of the picture spends 274 of its 315 steps within 0.050.05 of 0.50.5.

Some points land near 0.50.5 again and repeat the slow trip, which is why you can see more of them change color at about 650 and 1000 steps. The shape above is dust too, since c=0.2501c = 0.2501 lies just outside the Mandelbrot set.

Where the orbits go

Earlier, different pixels in the dark shape ended up on the same three places. With c=(−0.12256,0.74486)c=(-0.12256, 0.74486), a point can do two things: it can escape, or it can join one repeating orbit, which goes (0,0)(0, 0), then cc, then (−0.66236,0.56228)(-0.66236, 0.56228), then (0,0)(0, 0) again. The point goes round that orbit forever and never stops at any of the three places. This repeating orbit is called the attractor. The one exception is a point exactly on the edge of the dark shape, which stays on the edge forever.5

The demo below colors each starting point by its turn on the orbit, so points of one color reach the same place on the same jump and then move on together, while gray points escape. When you play it, each color bunches together and joins the orbit. It stops once every dot has joined or escaped, but the dots would go round forever.

Where each point's orbit goes

Mandelbrot · choose c

c = (−0.123, 0.745)

Where each point starts

Where the points are now

Orbit: (0, 0) → (−0.12256, 0.74486) → (−0.66236, 0.56228) → (0, 0) → …

On the orbit0
Escaped0
Still on the way4117

You can see why this happens in the dark shape, which is made of rounded pieces called lobes, with one color for each lobe. A jump moves a whole lobe onto another lobe, so its points stay together, and after enough jumps every lobe has landed on one of the three lobes that hold the orbit's places.5

Notice that the orbit passes through 00: if you start at 00, you land on cc, then on (−0.66236,0.56228)(-0.66236, 0.56228), then back on 00.7 This is why points settle so quickly, because squaring makes small numbers much smaller, as when 0.10.1 becomes 0.010.01. Each time a point comes back near 00, its distance from 00 shrinks fast, so a point 0.10.1 from 00 is 0.0260.026 away three jumps later, then 0.00180.0018, then 0.0000090.000009. The gap keeps shrinking forever, though after about 30 jumps it is too small for the computer to store, and the numbers repeat exactly.

You might ask whether the same cc could have a second attractor. It can't, because Pierre Fatou and Gaston Julia proved that if a cc has an attractor, 00 falls into it.5 The point 00 has only one orbit, so it can fall into only one attractor. This also gives you a direct way to find the attractor: start at 00 and follow the jumps, which is exactly what the Mandelbrot test does.

That means the Mandelbrot picture also tells you how long the orbit is, which you can check by dragging the ring in the demo. Inside the heart-shaped body, the orbit has one place, called a fixed point: one jump lands it on itself, so here the points really do end up in one spot. Inside the round bulb on the left, it has two places, as c=−1c = -1 sends 00 to −1-1 and back, and inside the big bulbs above and below the heart, it has three.8 The numbers on the Mandelbrot picture mark more bulbs. Outside the black shape, 00 escapes, so there is no attractor and all the dots escape.

You can also read that number straight off the picture of the Julia set. Count the colors that meet at a joint of the dark shape, and for a bulb that touches the heart, that count is the number of places. The lines that meet at the main fork of the bulb's largest antenna give the same count.8

Not every cc in the Mandelbrot set has an attractor, though. With c=(0,1)c = (0, 1), 00 lands on a two-place orbit that pushes nearby points away, and the picture of its Julia set has no dark area, because every point that doesn't escape lies on the Julia set itself.5 And close to a bulb's rim, the points need far more jumps, so the demo may not find the attractor.

Further exploring

Videos

Reading

Sources and notes

  1. Introduction to Complex Dynamics.
  2. Calculation: if ∣z∣>2|z|>2 and ∣z∣≥∣c∣|z|\ge|c|, then ∣z2+c∣≥∣z∣2−∣c∣≥∣z∣(∣z∣−1)>∣z∣|z^2+c|\ge|z|^2-|c|\ge|z|(|z|-1)>|z|. Here ∣c∣≈0.75|c|\approx0.75.
  3. Local Connectivity of Julia Sets and Bifurcation Loci: Three Theorems of J.-C. Yoccoz.
  4. Diagram idea: Julia Set Fractal (2D).
  5. Dynamics in One Complex Variable: Introductory Lectures. Section 3 defines the two sets, Lemma 14.2 covers the two futures and the lobes, Lemma 3.1 the edge, Lemmas 10.2 and 10.4 and Remark 10.5 the attractor and 00, and Figure 1c the case c=(0,1)c = (0, 1).
  6. Quadratic Julia Sets with Positive Area. Values of cc with positive area do exist, but this cc is not one of them.
  7. Calculation: this cc is rounded to five decimals, so the third jump lands about 0.0000040.000004 from 00, and the attractor's place sits that close to 00. Distances are measured from that place.
  8. Rotation Numbers and Internal Angles of the Mandelbrot Bulbs.