The parameter c = 0.26 can stay within the familiar radius-two Mandelbrot test for 29 iterations and then escape at iteration 30. The nearby c = 0.25001 takes approximately 991 iterations before it escapes in an ordinary double-precision calculation. A picture using only 100 iterations can therefore paint the second point as apparently inside even though a longer run proves it is outside.
The Mandelbrot set is defined by a mathematical requirement on an infinite sequence, while a screen can carry out only finitely many operations. This guide separates a rigorous escape certificate from a finite non-escape observation, calculates representative orbits, explains the special status of the threshold two, and gives a reproducible method for interpreting coloured fractal images.
Find the question you need
Choose the proof, worked orbit, algorithm or checking method you need. All twelve teaching chapters remain open; only the chapter index is expandable.
Define boundedness
Chapters 1–2. Start here.
Prove and test radius two
Chapters 3–5. Start here.
Find the finite-iteration trap
Chapters 6–8. Start here.
Audit rendering choices
Chapters 9–10. Start here.
Practise and transfer
Chapters 11–12. Start here.
Complete chapter index · Practice problems · Fractals subject root
Chapter index
Define boundedness · Chapters 1–2
Prove and test radius two · Chapters 3–5
Find the finite-iteration trap · Chapters 6–8
Audit rendering choices · Chapters 9–10
Practise and transfer · Chapters 11–12
Define a complex parameter c and a sequence starting at z₀=0. At each step compute zₙ₊₁=zₙ²+c. The Mandelbrot set M consists precisely of those parameter values c for which this orbit remains bounded for all n. Its definition concerns indefinitely many iterations, not a palette, pixel count or software setting.
On the complex plane, a point c=x+iy contains real and imaginary parts. Complex squaring uses (x+iy)²=(x²−y²)+i(2xy). For a given parameter c, we iterate the critical point zero; we are not asking whether every possible starting z produces a bounded sequence.
A displayed Mandelbrot image assigns a pixel to a sample value of c and runs an algorithm to approximate the definition. A pixel with one colour may mean escape was detected quickly; another may mean no escape was observed within the chosen computation budget. Those statements are not logically symmetric.
Starting from the exact recurrence prevents a familiar teaching error: using the Julia-set question for a fixed parameter c and arbitrary starting point z as though it were the definition of the Mandelbrot parameter set. They are related, but the varied quantity is different.
A sequence is bounded when there is some finite B such that |zₙ|≤B for every iteration n. It is unbounded if no such finite upper bound exists. For the Mandelbrot recurrence, the practical escape test uses a radius of two, with the qualification that the test checks whether |zₙ| exceeds two.
The norm of a complex number z=x+iy is |z|=√(x²+y²). In computation it is often easier to compare x²+y² against four rather than evaluate a square root. Thus x²+y²>4 and |z|>2 are equivalent in exact arithmetic.
An observed |zₙ|>2 is a one-way certificate: the parameter is outside M. By contrast, the inequalities |z₁|≤2, |z₂|≤2 and so on through a fixed finite N only say that the parameter has not escaped by N. They do not establish the required statement for every future iteration.
This logical asymmetry is the central concept of the article. A failed search for escape is not proof of permanent boundedness, even though it is useful information for allocating computing resources and colouring a picture.
Contents · Previous · Next · Fractals root
If |c|>2, the first iterate is z₁=c, already outside the escape circle. Otherwise suppose |c|≤2 and that some iterate has magnitude r=|zₙ|>2. By the reverse triangle inequality, |zₙ₊₁|=|zₙ²+c|≥r²−|c|≥r²−2.
For r>2, r²−2>r because r²−r−2=(r−2)(r+1)>0. Once an orbit is outside radius two, the bound forces later magnitudes to increase, and the quadratic growth makes the orbit diverge. The criterion is mathematically justified, not a decorative number selected for plotting.
The inequality must be applied in its correct domain. The bound alone does not assert that an iterate sitting exactly at magnitude two must leave. A parameter on the boundary can produce an orbit that touches two and stays there.
The subtlety matters when software tests |z|≥2 instead of |z|>2. A greater-than-or-equal rule can mark a boundary point as an outsider before mathematical escape has occurred. Numerical implementations may deliberately use different bailout conventions for graphical rendering, but a rigorous claim needs the stated test.
Contents · Previous · Next · Fractals root
For c=0, every iterate is zero. This proves boundedness directly, so 0 belongs to M. For c=−1, the sequence goes 0, −1, 0, −1, and repeats. Exhibiting an exact periodic orbit is also a proof of boundedness.
For c=1, the orbit begins 0, 1, 2, 5. On the third iterate after starting at zero, |z₃|=5>2, giving a definite exclusion certificate. The computation does not need to continue to ten thousand iterations to establish this result.
For c=−2, the sequence is 0, −2, 2, 2 and remains there. This parameter belongs to M despite touching the radius-two circle. It provides an immediate reason to distinguish strict escape from a simplistic ≥2 rule.
These examples establish the difference between analytic membership certificates such as an exact period and finite numerical evidence of non-escape. A plotting tool often has no such analytic test for every parameter and therefore depends on a finite iteration budget.
Contents · Previous · Next · Fractals root
Start at z₀=0 with c=0.26. The sequence remains real because both the initial value and parameter are real. The first terms are z₁=0.26, z₂=0.26²+0.26=0.3276, and z₃≈0.36732176. Early growth looks modest, which can make the orbit appear harmless when only a few frames are plotted.
In a straightforward binary64 evaluation of z=z²+0.26, the magnitude first exceeds two at step 30, reaching approximately 3.99887 at that step. That outcome certifies the parameter is outside M in the exact dynamical sense; this parameter is not a near-zero noise effect.
By comparison, c=0.25 has a special bounded real orbit that rises toward the fixed point 0.5. Indeed, z²+0.25=z has a repeated fixed-point solution at z=0.5. Its slow approach is not an escape; it belongs to the boundary of the main cardioid.
The contrast shows why a tiny change in c can matter qualitatively. Near the boundary, an orbit can spend many iterations close to a nearly neutral pattern before it finally begins to move away.
Contents · Previous · Next · Fractals root
Take c=0.25001, just 0.00001 greater than the special real parameter 0.25. Under standard double-precision iteration, the first |z|>2 occurs at approximately iteration 991. At iteration 100 the computation has not found an escape. A conventional black or dark pixel after 100 steps therefore does not prove membership.
The argument does not rely on claiming that the pixel is truly inside for the first hundred iterations; it merely says the finite test has not yet supplied an exclusion certificate. Increasing the budget to 1000 discovers the escape in this example, while even 1000 iterations need not settle every nearby point.
Near-boundary orbits are particularly sensitive to precision, map scale and iteration count. An arbitrary cutoff can become part of the visible boundary of the image. An apparent smooth region may change shape when the budget or precision changes.
State the numerical parameters whenever the image is used as evidence: chosen pixel-to-c mapping, maximum iterations, escape threshold, arithmetic precision and whether analytic interior tests were used. Without these, a viewer may confuse rendering choices with mathematical properties.
| Parameter c | Orbit behaviour | First strict escape | Conclusion |
|---|---|---|---|
| 0 | All iterates 0 | Never | Certified bounded |
| −1 | 0, −1, 0, −1, … | Never | Certified periodic |
| 1 | 0, 1, 2, 5 | Iteration 3 | Outside |
| −2 | 0, −2, 2, 2, … | Never | Certified bounded |
| 0.26 | Initially slow real growth | ≈30 | Outside |
| 0.25001 | Near parabolic boundary | ≈991 | Outside |
Contents · Previous · Next · Fractals root
Write z=x+iy and c=a+ib. Expanding the square gives x_next=x²−y²+a and y_next=2xy+b. With x₀=y₀=0, repeated evaluation of these two real updates produces the same orbit as the complex recurrence in exact arithmetic.
The escape comparison is x_next²+y_next²>4. It can be evaluated after each update to record the first iteration number on which the condition holds. That number is a useful rendering statistic, not itself a member of the Mandelbrot set.
Be precise about counting. Some programs check the current z before updating; others check after updating. With the same parameter, they may label escape counts one apart while agreeing on whether escape occurred during the computation budget. A written algorithm must specify this convention.
Because repeated squaring can magnify existing numerical perturbations, zoomed calculations may require more than standard machine precision. A faster implementation is not automatically a more trustworthy mathematical classifier near an intricate boundary.
Contents · Previous · Next · Fractals root
The simple escape-time algorithm colours a parameter according to how quickly escape was detected. For a parameter with no detected escape by N, software commonly assigns a default interior colour. In an ordinary visual rendering, that is a useful practical choice, not a theorem that its orbit is bounded forever.
Two parameter values that both fail to escape by N may have different mathematical statuses: one may be a proven interior point, while the other may eventually escape after N+1 or ten thousand steps. Treating both as established members loses the difference between evidence and uncertainty.
An analytically recognised region, such as part of the main cardioid or another certified periodic component, can provide stronger interior information. Exact periodic examples like c=−1 are easier to prove. Such methods are not equivalent to simply making N very large.
This is a broad scientific lesson: a finite negative test result can be informative without being logically conclusive. The same caution arises in numerical optimisation, simulation and algorithmic testing whenever the mathematical property quantifies over an infinite or unbounded collection.
Contents · Previous · Next · Fractals root
A screen pixel represents a finite area of the parameter plane, but the algorithm typically tests one parameter value such as the centre. If a mathematical boundary crosses the pixel, the colour of its centre cannot describe every parameter in the cell. Increasing the image width refines sampling, yet introduces more near-boundary points that can need higher iteration budgets.
Antialiasing may sample several parameters within a pixel and average their colours. That improves apparent visual smoothness, but does not certify that the underlying region is fully inside or outside. Smooth colour transitions may therefore combine numerical escape counts and deliberate graphics operations.
When zooming deeply, floating-point coordinates themselves may no longer distinguish nearby pixel centres reliably at the chosen precision. Before blaming an unexpected detail on chaotic dynamics, check whether the coordinate mapping has exhausted the arithmetic’s usable resolution.
Scientific image captions should separate the complex-plane domain, pixels per axis, escape test, iteration budget and colour map. Each is a different design decision with a different effect on the final picture.
Contents · Previous · Next · Fractals root
An escape count n is the number of algorithmic updates required before a threshold was exceeded. It is not elapsed clock time, a chemical reaction rate or distance travelled. Changes in processor speed do not alter the mathematical iteration number for the same arithmetic result.
Similarly, a colour palette can wrap, stretch or smooth the counts for visibility. Two images with different palettes may depict the same computed orbit data but look different. A choice of blue, orange or black does not add new mathematical evidence.
Some renderers compute smooth escape values from the magnitude of the orbit after threshold crossing. These can improve banding, but they inherit the assumptions of the escape-time calculation and are still not direct measurements of how ‘inside’ a point is.
For a classroom experiment, give learners two images of the same region with maximum iterations 50 and 1000 but identical coordinates and colour rules. Their task is to identify changes caused by additional evidence rather than by a different definition of M.
Contents · Previous · Next · Fractals root
Exercise A: c=1 gives 0→1→2→5. At which iterate is exclusion certified with strict radius-two escape? At n=3, because |5|>2. The value exactly two at n=2 does not yet meet that strict condition.
Exercise B: c=−1 produces a two-cycle after the starting point. Does checking that exact cycle prove membership? Yes. The orbit has only two values, 0 and −1, so its magnitude is bounded.
Exercise C: A 100-iteration image shows c=0.25001 as a dark pixel. Is the correct description ‘inside M’? No. It is ‘no escape detected within 100 iterations’; a longer run finds escape at about 991 for ordinary double precision.
Exercise D: c=−2 reaches −2 and 2 but stays there. Which code condition would incorrectly mark it as escaped? Testing |z|≥2 as a definitive exclusion rule. Using strict >2 avoids that specific mathematical error.
Contents · Previous · Next · Fractals root
Begin by defining the recurrence and z₀ precisely. For each parameter sample, record the real and imaginary coordinates, iterate with a documented arithmetic type and check x²+y²>4 after every update. Stop when escape is certified or the maximum iteration count is reached.
Store an explicit status, not just one integer: escaped at iteration n, or not escaped by iteration N. If a pixel needs to support an inference about membership, use analytic arguments or rigorous numerical bounds where feasible rather than equating a finite cutoff with the set definition.
Include tests with c=0, c=−1, c=1 and c=−2. They check bounded fixed/periodic patterns, fast escape and the strict escape-threshold boundary case. Add c=0.26 and c=0.25001 to test how the algorithm handles delayed escape.
The Gold Standard of Fractals subject root connects this dynamical example with geometric self-similarity. Compare the Sierpiński and Koch constructions to see why fractal questions may involve recurrences, area, boundary length or escape rather than one universal formula.
Applied learning lab
A safe minimal escape-time algorithm
An educational implementation can use real numbers: initialise x=y=0; for n from one to N, set xnew=x²−y²+a, ynew=2xy+b, then update x and y. If x²+y²>4, return escaped together with n. If N iterations finish without meeting the test, return ‘not escaped by N’, not ‘member of M’.
A structured return value matters because the later page renderer should never confuse the two outcomes. A graphing program may use a default dark fill for all undecided pixels, but the underlying computation can still preserve honest statuses for later refinement.
Give the implementation known cases before mapping thousands of pixels. Correct output for c=0, c=−1, c=1 and c=−2 checks the fundamental recurrence and strict escape convention. If it fails an elementary case, a more colourful picture cannot make it mathematically reliable.
Why the 0.25 example approaches a fixed point
For c=1/4, a real fixed point solves z=z²+1/4 or (z−1/2)²=0. Start at z₀=0. For values from zero through 1/2, the recurrence increases z without jumping past the fixed point, and the sequence remains bounded above by 1/2.
The increments become small near the repeated fixed point, explaining why a finite orbit can seem to settle. Moving the real parameter slightly above 1/4 removes the real fixed point; the orbit can still linger for many steps in the former fixed-point region before eventually escaping.
That is why c=0.25001 requires far more iterations to detect exclusion than c=0.26. The delay is a mathematical feature of the parameter choice, although the precise recorded escape count in a computer is still conditioned by the arithmetic and counting convention.
A three-level reasoning exercise for students
At the first level, ask a student to compute z₁, z₂ and z₃ for c=1 and c=−1. Correct arithmetic and the ability to identify a repeated state establish mastery of the recurrence.
At the second level, ask whether 100 observed bounded terms establish permanent boundedness. The correct answer is no; the set definition quantifies over the whole infinite sequence. A counterexample such as c=0.25001 illustrates why the distinction matters in practice.
At the third level, ask the student to prove the escape test using the reverse triangle inequality. This moves the explanation from a memorised radius-two rule to a justified implication. A polished answer states both the assumption |c|≤2 and the fact that |z|>2 makes subsequent magnitudes grow.
The same mathematical habit is valuable beyond fractal software: state the proposition to be proved, identify what the computation really shows, and avoid converting a finite diagnostic into an unsupported universal conclusion.
Document uncertainty rather than hiding it in a palette
For a reproducible image, write an explicit caption: parameter window in complex coordinates, image resolution, sample location within pixels, maximum iterations, strict escape radius, precision type and colour scale. If an analytic interior shortcut or adaptive iteration count is used, name it.
Rendering quality and mathematical certainty are distinct objectives. Increasing the iteration limit can improve boundary classification, while increasing pixel count can improve spatial sampling. Improving one without the other can reveal different artefacts rather than one straightforward improvement.
Readers should know whether a dark pixel means ‘analytically certified interior’ or only ‘no escape detected’. That language prevents the article itself from misleading the learner while still allowing a visually compelling exploration.
Frequently asked questions
Why does radius two work?
For |c|≤2, the inequality |z²+c|≥|z|²−2 implies that any |z|>2 subsequently grows. For |c|>2, the first iterate already escapes.
Does a dark pixel prove membership?
No. Unless an analytic membership test was used, a dark pixel can simply mean that escape was not detected within the program’s finite iteration budget.
Why not use ≥2?
The strict condition matters for mathematical exclusion: c=−2 produces a bounded orbit touching magnitude two, so equality alone is not a certificate of divergence.
Sources and connected reading
The displayed orbit values are calculated teaching examples. Their meaning is conditional on the defined recurrence and the specified iteration convention.
- Complex Analysis — The Mandelbrot Set and escape-time algorithm
- The Escape-Time Algorithm — Visual Mathematics
- The Gold Standard of Fractals — subject root
Further worked lab: delayed escape, proof and rendering tests
A rigorous argument for real parameters just beyond one quarter
For real c=1/4+δ with δ>0, the recurrence z_next=z²+1/4+δ satisfies z_next−z=(z−1/2)²+δ. The square is nonnegative, so every real iterate grows by at least δ. Starting from zero, zₙ≥nδ by induction.
For δ=0.00001, eventually nδ exceeds two; any such iteration must therefore have zₙ>2, certifying exclusion mathematically. This is an intentionally loose bound. A double-precision computation finds escape near iteration 991, but the inequality proves eventual escape independently of the displayed iteration count.
At δ=0 the same identity becomes z_next−z=(z−1/2)². Beginning at zero the orbit is nondecreasing yet remains below the fixed point one-half, so it is bounded. The transition at c=1/4 has a structural explanation rather than merely a change of colour.
The distinction between an analytical argument and a computed count is valuable. One proves that escape must eventually occur for the specified real parameter; the other records how soon a particular program noticed it.
Why iteration count cannot simply be read as distance
A parameter that escapes on step 30 is not necessarily ten times farther from the Mandelbrot boundary than one that escapes on step 300. Escape time describes the orbit’s dynamics and is not a universal geometric distance metric.
Near c=1/4, the completed-square identity explains slow growth. Elsewhere in the complex plane, periodic structures and different orbit geometries create their own delays. Two parameters with the same escape count can be located in very different parts of a plotted region.
A contour coloured according to a fixed escape count is therefore a contour of algorithmic behaviour, not automatically a constant-distance curve from the true boundary. Distance estimation requires a separate method with its own assumptions.
This matters when readers interpret a decorative colour band as evidence of membership certainty or geometric thickness. A display can represent iteration counts truthfully while remaining limited as a spatial measurement.
A small regression test for the renderer
Create a table with parameter c, maximum iteration budget N, first strict escape if seen, and a status that distinguishes escaped from not yet escaped. Begin with c=1 and c=−1 to check rapid exclusion and exactly periodic boundedness.
Then test c=0.26 with N=20 and N=40. The shorter run does not find escape, whereas the longer run does. For c=0.25001 use N=100, N=500 and N=1000; only the last budget sees escape near step 991 in the specified arithmetic.
Repeat selected points using greater precision if studying delicate boundary detail. A changed numerical escape count is information about the numerical computation, not evidence that the exact Mandelbrot set itself changed.
Retain statuses for certified escape, analytically certified boundedness where known, and unresolved at the current budget. The user-facing image can have a compact palette without falsely collapsing these into one logical category.