Chaos

Law without long-range foresight. Traces follow the same deterministic equation, yet nearby starts can stretch apart while remaining confined to a strange attractor. The luminous butterfly or dust field is not a creature in ordinary space—it is a portrait of the states a system repeatedly visits.

A dynamical system specifies how a state changes. If the state is represented by variables ((x,y,z)), one instant is one point in a three-dimensional phase space. A continuous-time system supplies a velocity at every point,

Compact form: (d x)/(dt)= F( x),

while a discrete map supplies the next state,

Compact form: x_n+1= F( x_n).

The visual trails in Chaos are histories in this mathematical space. They are not necessarily positions, velocities or objects in the physical canvas. The screen is a projection of state trajectories.

The central distinction is between determinism and predictability. Determinism means that the current state and rule determine the next state. Predictability asks whether finite knowledge of the current state remains useful far into the future. In a chaotic regime, arbitrarily small initial uncertainty can grow rapidly enough that detailed long-range prediction fails even though the rule contains no random choice.

A standard local description is sensitive dependence on initial conditions. If two nearby states begin a distance (\delta_0) apart, a positive largest Lyapunov exponent (\lambda) would imply, over an appropriate range,

Compact form: \delta(t)≈ \delta_0 e^\lambda t.

This is a conceptual reference, not a telemetry claim. AL.Floys does not calculate Lyapunov exponents, shadowing time, entropy, invariant measure or fractal dimension. A visitor can see trajectories separate, but visual separation alone is not a formal diagnosis of chaos.

The system’s best-known mode is the Lorenz flow:

Compact form: \dot x=\sigma(y-x),\qquad \dot y=x(\rho-z)-y,\qquad \dot z=xy-\beta z.

Edward Lorenz derived this three-variable system as a severe reduction of equations for thermally driven fluid convection. At the familiar values (\sigma=10), (\rho=28) and (\beta=8/3), trajectories circulate around one lobe, cross a narrow region and circulate around the other with an irregular switching sequence.

The two lobes should not be presented as two literal atmospheric vortices or butterfly wings. They are regions of phase space visited by a reduced mathematical state. Their importance is that nearby trajectories can make different future lobe choices while remaining confined to the same global form.

The Rössler flow offers a complementary geometry:

Compact form: \dot x=-y-z,\qquad \dot y=x+ay,\qquad \dot z=b+z(x-c).

Its motion is often read as a spiral followed by an out-of-plane fold and reinjection. This makes the stretch-and-fold mechanism especially legible: local neighbourhoods are elongated, moved and returned to a bounded region, where the operation repeats.

AL.Floys integrates both continuous flows with a fixed-step midpoint method, commonly called second-order Runge–Kutta or RK2. For each requested step it evaluates the derivative at the current state, estimates a midpoint, evaluates again there and advances using the midpoint derivative. This is more accurate than a single explicit Euler step at the same size, but it remains a numerical approximation.

The dt parameter therefore has two meanings that must be kept separate. It controls how far the numerical integrator advances per flow step, and it also affects numerical error. Increasing it can make the attractor appear faster, but sufficiently large values may deform or destabilise the numerical trajectory rather than reveal new underlying physics.

The sphere also contains several two-dimensional maps: Clifford, De Jong, Ikeda, Tinkerbell, Hopalong and a Svensson-style trigonometric map. In these modes there is no continuous path between successive mathematical states. Each application of the map jumps from one point to the next.

For Clifford and De Jong, for example, AL.Floys computes target coordinates from trigonometric functions of the current coordinates. It then applies

Compact form: x_n+1 = x_n+\alpha\bigl( F( x_n)- x_n\bigr),

where the blend (\alpha) is derived from dt and frame scaling and is capped at one. When (\alpha=1), this is a direct map iteration. At lower values it becomes a relaxed interpolation toward the canonical next iterate. The public essay should therefore say Clifford-style or De Jong map dynamics as implemented, not imply that every slider setting is a textbook iteration.

The same caution applies to the engine’s Svensson mode. It is explicitly a Svensson-style coupled trigonometric map with implementation-specific offsets, not a claim to reproduce one canonical formula under every parameter label.

Thousands of traces are used to reveal the geometry. They do not attract, repel, collide or communicate. Each follows the selected equation independently. The cloud becomes informative because many initial conditions sample different parts of the same state-space structure.

This is an important contrast with Flock and Particle Life. There, population-level form arises through agent–agent interaction. In Chaos, the visible collective is an observational ensemble: many trajectories reveal one dynamical rule, but the trajectories do not generate that rule or alter one another.

A chaotic attractor combines confinement with nonrepetition. Dissipation contracts some volumes of state space, while nonlinear stretching and folding prevents simple collapse to a point or ordinary cycle. The resulting invariant set can have geometry more intricate than a smooth curve or surface. The term strange attractor names that conjunction, but establishing strangeness rigorously requires more than a luminous image.

The engine provides two complementary visual instruments. Trail mode draws recent projected path segments, making local motion and switching visible. Density mode accumulates visits into a grid and applies logarithmic tonemapping, making frequently occupied regions legible even when individual paths overlap. Neither rendering changes the underlying equation. Trail persistence, exposure, colour by depth or speed, auto orbit and breathing zoom are observational choices.

Auto frame is also projection-only. It recentres and rescales recent projected bounds so the attractor fills the stage. Turn Y, the other simple visual control, rotates the projection. A shape that changes dramatically under rotation has not changed dynamically; the visitor has changed vantage point in phase space.

The public tool labels require special care. Force perturbs the state variables of traces whose current projection lies near the pointer. It is a genuine intervention on state, although selection occurs in screen space. Spawn does not increase the trace count: it reseeds affected existing traces with a positional bias. Erase does not delete them: it reseeds them in another biased region and clears their immediate trail anchors. These are useful perturbations, but their current names overstate their semantics.

Finally, divergence recovery is an explicit maintenance rule. When a trace becomes non-finite or leaves a large safe bound and Recover on blow-up is enabled, the engine replaces it with a new seeded state. The attractor display therefore includes a fail-safe that prevents numerical or parameter divergence from permanently emptying the scene. A respawned trace is not evidence that the mathematical system naturally returned.

The historical background reaches through nineteenth-century celestial mechanics and Henri Poincaré’s recognition that deterministic systems can possess extraordinarily complicated trajectories. The modern public lineage of chaos, however, is anchored by several twentieth-century interventions.

In 1963, Edward Lorenz published “Deterministic Nonperiodic Flow.” Working with a reduced model of thermal convection, he showed that a small deterministic system could produce bounded nonperiodic solutions that were unstable under small changes. The work transformed sensitivity from a numerical annoyance into a limit on long-range prediction.

David Ruelle and Floris Takens introduced the modern language of strange attractors in their 1971 account of turbulence. Their proposal helped shift attention from ever more complicated combinations of periodic motions toward geometrically intricate invariant sets generated by deterministic dynamics.

T.-Y. Li and James Yorke’s 1975 paper “Period Three Implies Chaos” gave the word chaos a precise role in the mathematics of one-dimensional maps. It established that simple discrete recurrence can contain an enormous hierarchy of periodic and nonperiodic behaviour.

Otto Rössler’s 1976 equation then demonstrated how a continuous chaotic flow could be built from a particularly spare three-variable system with one second-order nonlinearity. Its spiral-and-fold geometry became a second canonical visual archetype beside Lorenz’s two-lobed attractor.

The same year, Robert May’s review “Simple mathematical models with very complicated dynamics” showed a broad scientific audience that elementary deterministic difference equations could pass from fixed points through bifurcating cycles to apparently random fluctuations. This was crucial for ecology and other sciences accustomed to assuming that complicated observations required complicated or stochastic causes.

The later map families in AL.Floys belong to the culture of computational experimentation that followed: iterative formulas became laboratories for attractors, bifurcations and generative graphics. The sphere should preserve the eponymous names while distinguishing historically canonical equations from implementation-specific variants and renderings.

Aspect	Current implementation
Entities / field	Hundreds to thousands of independent mathematical traces; there is no interacting particle field.
State	Three floating-point coordinates per trace for flows; the third coordinate is held at zero for planar maps.
Continuous modes	Lorenz and Rössler ordinary differential equations.
Discrete modes	Clifford, De Jong, Ikeda, Tinkerbell, Hopalong and a Svensson-style trigonometric map.
Time / iteration update	Midpoint RK2 for flows; direct or relaxed map update depending on dt-derived blend.
Steps per frame	Repeats the flow advance or map iteration. It advances more simulated dynamics per display frame; it is not a subdivision of one fixed physical timestep.
Initial conditions	Seeded random states, distributed according to mode-specific ranges, Seed spread and Spawn jitter.
Divergence handling	Optional reseeding of non-finite or very large states.
Boundary conditions	None in the ordinary spatial sense; the system lives in unbounded mathematical state space with a software divergence threshold.
Randomness	Initial seeding, reseeding and pointer perturbation jitter. Ordinary equation updates are deterministic given state and parameters.
Rendering interpretation	Projected trails or accumulated visit density, with optional depth, speed, age or density colour. Auto framing, rotation and camera motion are visual only.
Public simple controls	Attractor, Traces, Steps / frame, Turn Y and Auto frame.
Public tools	Force plus two currently misleading reseeding tools labelled Spawn and Erase.
Formal diagnostics absent	No Lyapunov exponent, entropy, correlation dimension, recurrence plot or bifurcation analysis.

The most truthful one-line description is: an interactive phase-space observatory for deterministic nonlinear flows and maps.

Sensitive divergence — nearby beginnings lose detailed agreement

Apply a small local perturbation to part of the Lorenz ensemble. The disturbed traces initially remain inside the same broad geometry, yet their later lobe-switching sequence separates from their neighbours. The attractor can be stable while individual prediction is fragile.

Stretching and folding — complexity without random forcing

In the Rössler mode, trajectories spiral outward, leave the nearly planar region and are folded back. Repeated elongation and reinjection create intricate occupancy from a compact rule.

Bounded aperiodicity — confinement without a simple loop

A trace can remain inside a recognisable region indefinitely without closing into an ordinary periodic curve. The system is organised, but its organisation is not a repeating choreography.

Lobe switching — one rule, irregular symbolic history

In Lorenz mode, a trace circles one lobe and then crosses to the other. The sequence of left and right visits is a compact visual lesson in deterministic unpredictability.

Continuous flow versus discrete map — two forms of nonlinear time

Lorenz and Rössler produce connected trajectories because state changes continuously. Clifford, De Jong and the other maps produce successive iterates; apparent filaments are histories or densities of jumps, not continuous physical paths.

Parameter-induced regime change — an attractor is not guaranteed

Changing coefficients can shrink a cloud toward a fixed point, produce a cycle, create a different bounded set or drive states to divergence. Chaos names the sphere’s subject, not a promise that every possible setting is chaotic.

Density structure — where a trajectory spends time

Density rendering makes repeated occupancy visible. Brighter areas indicate more accumulated visits under the renderer’s sampling and tonemapping, not greater physical mass.

Projection dependence — a strange object is not a screen silhouette

Rotate a three-dimensional flow with Turn Y. Overlaps separate and hidden depth appears. The underlying trajectory is unchanged; only its two-dimensional projection differs.

First 30 seconds

1. Load Lorenz / The Double Vortex.
2. Leave Auto frame on and follow the switching between the two lobes. Do not interpret the full luminous form as one particle.
3. Move Turn Y slowly. Notice that the visible silhouette changes while the dynamics continue unchanged.
4. Open Steer, choose Force, and make a short, low-strength gesture over one part of a lobe.
5. Watch the perturbed traces return to the same global attractor while losing detailed agreement with nearby traces.
6. Load Rössler / Spiral Organ and compare its spiral-and-fold geometry with Lorenz’s lobe switching.

Three experiments

Experiment	Question	Do	Watch for	Why it matters
Perturbation without escape	Can detailed trajectories change while the global regime survives?	Lorenz / The Double Vortex → apply a brief, weak Force to a small projected region.	Local displacement, later separation and eventual reoccupation of the same broad attractor.	Distinguishes trajectory predictability from attractor stability.
Flow versus map	What changes when time is continuous rather than iterated?	Compare Lorenz / The Double Vortex with Clifford / Violet Dust Bloom while holding display style broadly similar.	Connected flow paths versus successive map iterates and dust-like occupancy.	Introduces the two principal forms of dynamical update in the sphere.
Dynamics versus projection	Is every visible change a change in the model?	Keep one Lorenz run active → vary Turn Y → toggle Auto frame.	Reframing, overlap and apparent shape changes without a new trajectory rule.	Teaches the difference between state evolution and observation.

An advanced fourth experiment is to vary a single equation coefficient while keeping seed, timestep and rendering fixed. Record whether the system remains bounded, becomes periodic-looking or diverges. Do not label a regime chaotic solely from appearance.

Parameter	What it really controls	Increase / change it	Decrease / reverse it	What to watch
Attractor	Selects the differential equation or iterated map; structural.	Switches the entire dynamical rule.	Not scalar.	Continuous versus discrete time, geometry and boundedness.
Traces	Number of independent initial states sampling the rule; structural.	Denser coverage and smoother density estimates.	Easier individual tracking but sparser geometry.	Sampling quality, not stronger chaos.
Steps / frame	Number of integrations or map updates per display frame.	Faster simulated evolution and denser visual accumulation.	Slower, more legible motion.	It changes temporal throughput, not equation coefficients.
Δt (advanced)	Flow integration step and map-relaxation blend.	Faster advances and potentially more numerical distortion; maps approach direct iteration as blend reaches one.	Slower flow and more relaxed map interpolation.	Numerical stability versus true regime change.
a, b, c, d (advanced)	Equation or map coefficients with mode-specific meaning.	Entry-specific.	Entry-specific.	Bifurcation, collapse, boundedness and attractor geometry.
Seed spread (advanced)	Initial distribution scale.	Samples a wider part of state space.	Concentrates trajectories near one region.	Transient duration and basin selection.
Spawn jitter (advanced)	Random perturbation around seeded states.	Greater initial separation.	More tightly paired initial conditions.	Early divergence.
Recover on blow-up (advanced)	Whether divergent/non-finite states are reseeded.	Maintains a populated display.	Allows failed states to remain invalid or disappear from useful view.	Do not mistake reseeding for natural recurrence.
Turn Y (visual)	Rotation of the three-dimensional projection.	Reveals different overlaps and depth.	Reverses viewpoint.	Dynamics remain unchanged.
Auto frame (visual)	Recentres and rescales projected bounds.	Keeps the form prominent.	Preserves a fixed camera frame.	Apparent breathing or size is not necessarily state expansion.
Render mode (advanced visual)	Recent path history versus visit-density accumulation.	Not scalar.	Not scalar.	Motion versus invariant occupancy.
Trail hold / exposure (visual)	Persistence or tonemapping of drawn history.	More accumulated structure.	Cleaner instantaneous traces.	Rendering memory is not dynamical memory.

The best first public control is Attractor because it changes the rule in a conceptually legible way. Turn Y is the best control for teaching the distinction between dynamics and projection.

Lorenz / The Double Vortex

The canonical opening regime. It makes bounded nonperiodicity, lobe switching and sensitivity easiest to discuss. “Double vortex” is visual shorthand for phase-space lobes, not two literal fluid vortices.

Rössler / Spiral Organ

The best preset for seeing a spiral repeatedly leave a near-plane and fold back. It complements Lorenz rather than merely recolouring it.

Clifford / Violet Dust Bloom

A dense planar map. Use it after the visitor understands that map points are successive iterates rather than particles drifting through ordinary space.

De Jong / Fractal Weather

Another trigonometric map with soft filamentary occupancy. The “weather” phrase is aesthetic; it does not model atmosphere.

Ikeda / Aperture Scar

A folded planar map historically related to an optical-cavity model. In this sphere it should be treated as an iterative state-space specimen, not a laser simulation.

Tinkerbell / Winged Dust

A polynomial map with strong basin and divergence sensitivity. Useful for showing how a simple recurrence can create bounded wings beside escape.

Hopalong / Fossil Current

A square-root/sign map whose density rendering resembles mineral or current lines. The geological language is metaphorical.

Svensson / Sharp Field

An AL.Floys Svensson-style trigonometric map. Use this wording because the implementation contains custom offsets.

Lorenz / Depth Bloom and De Jong / Density Cloud

These presets repeat familiar rules through visit-density rendering. They are ideal for comparing trajectory history with occupancy.

Phase Storm / Collapse Front

A high-count, high-throughput Lorenz presentation. It is primarily a dramatic rendering regime and a performance stress case, not a separate scientific model.

Recommended learning order: Lorenz / The Double Vortex → Rössler / Spiral Organ → Clifford / Violet Dust Bloom → Lorenz / Depth Bloom → one of Ikeda, Tinkerbell or Hopalong.

• Force perturbs the internal ((x,y,z)) state of traces whose current projection lies within the pointer radius. It is the strongest scientific interaction because it tests how local state perturbation changes later trajectories.
• Spawn does not add traces. It reseeds affected existing traces with a bias derived from pointer position. The population count remains unchanged.
• Erase does not delete traces. It reseeds affected traces in an opposite-biased region and clears their immediate screen-history anchors.
• Pointer selection occurs in the two-dimensional projection. A gesture can therefore affect phase-space points that happen to overlap on screen even if they are separated in depth.

The public tool contract should eventually become Perturb · Reseed · Scatter or implement literal spawn/delete semantics. Until then, visitor copy should describe what the tools do rather than repeat their labels as if they were exact.

Chaos is not randomness. The ordinary equation updates are deterministic. Randomness enters through initial seeding, reseeding and pointer jitter, but it is not the source of the canonical Lorenz or Rössler mechanism.

A complicated shape is not sufficient evidence of chaos. Periodic trajectories can look elaborate; random noise can look irregular; a plotted attractor can be distorted by timestep, projection and trail persistence. The engine does not calculate a Lyapunov exponent or other formal diagnostic.

The traces are not interacting particles, insects, weather parcels or celestial bodies. They are samples of mathematical state.

The Lorenz mode is not a weather forecast and does not simulate the atmosphere in operational detail. It descends from a highly reduced convection model.

The Rössler, Clifford, De Jong, Ikeda, Tinkerbell and Hopalong names identify mathematical lineages, but AL.Floys adds common stepping, seeding, divergence recovery and rendering choices. Low map blend values are not canonical direct iterations.

A trace that disappears and returns may have been software-reseeded after divergence. A breathing or rotating attractor may reflect auto framing or camera motion. These interventions must not be narrated as spontaneous properties of the equation.

The sphere supports a strong, bounded conclusion:

Deterministic nonlinear rules can generate trajectories that are globally structured yet locally prediction-limiting.

Chaos adds the distinction between law and prediction. Gravity already showed that a law and its initial conditions jointly determine a trajectory. Chaos sharpens that lesson: even when the law is known and the system is low-dimensional, finite precision in initial state can destroy detailed long-range foresight. It also introduces phase space as an atlas-wide way of seeing a system—not as a picture of objects, but as a picture of possible states.

Foundational and primary sources

1. Edward N. Lorenz, “Deterministic Nonperiodic Flow,” Journal of the Atmospheric Sciences 20, 130–141 (1963). DOI: 10.1175/1520-0469(1963)020<0130:DNF>2.0.CO;2.
2. David Ruelle and Floris Takens, “On the Nature of Turbulence,” Communications in Mathematical Physics 20, 167–192 (1971). DOI: 10.1007/BF01646553.
3. Tien-Yien Li and James A. Yorke, “Period Three Implies Chaos,” The American Mathematical Monthly 82, 985–992 (1975). DOI: 10.1080/00029890.1975.11994008.
4. Otto E. Rössler, “An Equation for Continuous Chaos,” Physics Letters A 57, 397–398 (1976). DOI: 10.1016/0375-9601(76)90101-8.
5. Robert M. May, “Simple Mathematical Models with Very Complicated Dynamics,” Nature 261, 459–467 (1976). DOI: 10.1038/261459a0.
6. Kensuke Ikeda, “Multiple-Valued Stationary State and Its Instability of the Transmitted Light by a Ring Cavity System,” Optics Communications 30, 257–261 (1979). DOI: 10.1016/0030-4018(79)90090-7.

Modern scientific context

7. Steven H. Strogatz, Nonlinear Dynamics and Chaos, 2nd ed. (Westview Press, 2015). Useful modern treatment of phase space, flows, maps, bifurcation, Lyapunov exponents and strange attractors.
8. Colin Sparrow, The Lorenz Equations: Bifurcations, Chaos, and Strange Attractors (Springer, 1982).