Turing Patterns

Morphology without a template. Two reacting, unequally diffusing fields can turn a seeded disturbance into spots, bands or labyrinths—or erase it back to uniformity. Pattern here is a sustained organisation of concentrations, held open by replenishment, removal and transport rather than by a drawn design.

Turing’s general mechanism

Alan Turing’s 1952 paper asked how an initially similar field of cells could acquire spatial differentiation without a pre-existing map. His mathematical proposal considered substances he called morphogens that react locally and diffuse through space. The counterintuitive insight was that diffusion, ordinarily associated with smoothing differences away, can under the right reaction kinetics destabilise a homogeneous state. A small fluctuation may be amplified at a preferred wavelength, producing stationary concentration differences.

A general two-field reaction–diffusion system has the form

Compact form: (∂ u)/(∂ t)=D_u∇^2u+f(u,v), \qquad (∂ v)/(∂ t)=D_v∇^2v+g(u,v).

The Laplacian terms spread each field spatially; the nonlinear reaction functions determine local production and removal. A classical Turing instability occurs when a homogeneous equilibrium that is stable without diffusion becomes unstable after unequal diffusion is introduced. That strict definition matters, because “reaction–diffusion pattern” and “Turing pattern” are often used more loosely in visual-computing culture.

What the Gray–Scott model adds

The actual AL.Floys engine implements the Gray–Scott model:

Compact form: (∂ A)/(∂ t) = D_A∇^2 A - AB^2 + f(1-A),

Compact form: (∂ B)/(∂ t) = D_B∇^2 B + AB^2 - (f+k)B.

Here (A) is supplied toward concentration one at feed rate (f). (B) is removed at the combined rate (f+k). The cubic term (AB^2) converts (A) into more (B): wherever enough (B) already exists, it promotes further local production. This autocatalysis creates positive feedback, while feed, removal and diffusion constrain it.

The equations represent an open system. Material is continually supplied and removed. Pattern persistence is therefore not a static equilibrium of a closed mixture; it is maintained by throughput. That is one reason the model supports long-lived islands, fronts and replicating-looking spots without implying organisms.

Why unequal diffusion matters

Diffusion couples local reaction sites. Fast diffusion can replenish depleted areas or spread inhibition; slow diffusion can preserve local peaks. In AL.Floys, the default diffusion coefficient for A is substantially larger than that for B. A therefore spreads broadly while B remains comparatively local. This separation of spatial scales allows a B-rich patch to sustain itself locally while drawing A from a wider neighbourhood.

The visible pattern scale is not determined by one parameter. It emerges from the relation among reaction rates, diffusion coefficients, grid spacing, time step and initial perturbation. Increasing diffusion does not simply “blur” an unchanged image; it changes which structures can exist and interact.

Finite-amplitude seeding is central in this implementation

Pearson’s 1993 study of Gray–Scott dynamics emphasised a surprising variety of patterns arising from finite-amplitude perturbations. The current AL.Floys wrapper follows that practical lineage. It deliberately lays down soft islands, maze fragments, bands or patches of B, then progressively advances the chemistry through a warm-up phase.

This has a major editorial consequence. The viewer is not currently watching a random microscopic fluctuation destabilise an exactly uniform field. The initial morphology already supplies spatial structure. The scientifically accurate question is:

Which seeded disturbances can the reaction–diffusion dynamics sustain, transform, divide or erase?

That is still a deep pattern-formation question, but it is not identical to the textbook demonstration of spontaneous wavelength selection from near-uniform noise.

Spots that divide are not reproducing organisms

One celebrated Gray–Scott regime contains spots that grow and split. The visual resemblance to cellular reproduction is powerful because a local concentration domain can enlarge, become unstable and divide into two domains. Nothing in the code, however, represents a cell, membrane, genome or reproductive programme. “Replication” is a morphological description of field-domain dynamics.

The correct conceptual lesson is more general: a distributed nonlinear field can create repeated units without an object-level instruction to copy an object.

Pattern regimes occupy parameter space

Feed and kill do not act like ordinary brightness or size sliders. Together they choose a dynamical regime. A small change may leave a pattern nearly unchanged, alter the preferred morphology, or cross a boundary into decay, expansion or irregular activity. The useful mental model is a map of regimes rather than a linear spectrum from “less pattern” to “more pattern.”

This is why controlled exploration matters. Moving feed and kill simultaneously makes it impossible to know which boundary was crossed. One parameter at a time reveals the local structure of the regime map.

Numerical implementation shapes the world

The inner engine stores two floating-point arrays on a toroidal grid and applies a weighted nine-point approximation of the Laplacian. It advances the equations explicitly for several internal iterations per rendered step and clamps both fields to the interval ([0,1]).

Each of those decisions modifies the mathematical system. Grid resolution limits the smallest feature. Explicit time stepping has stability constraints. Clamping prevents negative or above-one concentrations and can shift behaviour relative to an unclamped reference implementation. Toroidal boundaries remove walls but allow a pattern leaving one edge to influence the opposite edge.

The simulation therefore makes an important computational-science point: an equation family does not uniquely determine a visual world. Discretisation, initialisation, boundary conditions and parameter scaling are part of the model actually experienced.

Turing published “The Chemical Basis of Morphogenesis” in 1952. His target was biological form: how genes and chemistry might initiate spatial differentiation in an embryo. The enduring contribution was a general mechanism by which reaction and diffusion can generate patterned concentration fields, including stationary waves, from a homogeneous equilibrium under appropriate conditions.

Peter Gray and Stephen Scott studied a simple cubic autocatalytic reaction in open, continuously stirred reactors during the 1980s. Their kinetic system displayed multistability, ignition, extinction, oscillation and other nonlinear behaviour. When spatial diffusion is added, it becomes the Gray–Scott reaction–diffusion model now widely used in mathematical chemistry and visual simulation.

John E. Pearson’s 1993 paper “Complex Patterns in a Simple System” mapped a remarkable set of Gray–Scott behaviours: stationary irregular patterns, travelling structures and spots that grow and divide. Crucially for the present engine, Pearson described patterns arising in response to finite-amplitude disturbances.

The lineage should therefore be narrated accurately:

Turing’s general diffusion-driven morphogenesis → Gray–Scott cubic autocatalysis in open reactors → Pearson’s spatial pattern regimes → AL.Floys seeded, curated interactive implementation

Turing did not publish the Gray–Scott equations, and Gray–Scott patterns are not all interchangeable with the strict linear-instability case studied in introductory accounts of Turing patterns.

Aspect	Current implementation
Entities / field	Two continuous scalar arrays, A and B, on a two-dimensional grid.
Reaction	Cubic autocatalysis A·B², feed of A toward one, and removal of B through feed + kill.
Diffusion	Separate diffusion coefficients for A and B using a weighted nine-point discrete Laplacian.
Initial conditions	Curated finite B-rich seed morphologies: maze islands, islands, bands, distributed spots or classic patches.
Warm-up	A configured number of chemistry steps is consumed progressively over early frames, allowing a curated seed to develop without blocking reset.
Boundary conditions	Toroidal wrap.
Numerical method	Explicit Euler-style stepping with configurable dt and internal iterations; A and B are clamped to [0,1].
Wrapper architecture	Turing2Simulation translates its public parameter set into the older TuringSimulation, reusing the inner solver while supplying a different initialisation and rendering experience.
Rendering interpretation	Primarily blue-on-white concentration rendering with configurable contrast and visual variants. Colour is a field visualisation, not a literal chemical indicator.
Public simple controls	Seed style, feed, kill, brush radius and contrast.
Public tools	Brush, Spawn and Erase.

Current implementation defect: the public Brush radius control is inert

Turing2Params defines a live brushRadius control, and ecospheres.ai exposes it in the simple surface. The wrapper, however, never forwards that value to the inner solver. Turing2Simulation.toInnerParams() sets the inner seedRadius from seedSize, while TuringSimulation uses seedRadius for pointer painting. A repository search finds brushRadius only in the parameter schema and tests, not in the runtime mapping.

The result is a scientifically damaging UI contract: the visitor can change Brush radius without changing the painted radius. This should be treated as P0 because it invalidates an advertised intervention and would undermine the field experiments proposed below.

The correction should separate two concepts:

• seedSize → reset-time size of initial islands;
• brushRadius → live pointer radius forwarded to the inner paint parameter.

A regression test should compare affected-cell area for two distinct brushRadius values while holding the pointer state fixed. Existing tests verify that brushing changes B, but not that the control changes brush size.

Current editorial defect: the generated description is implementation copy

“Gray–Scott patch mode: reuses the Turing engine with a simple blue-on-white view and scale-based grid” is useful developer documentation, not public atlas copy. It foregrounds reuse architecture and rendering rather than the scientific world. Replace it with a causal description of reaction, differential diffusion and pattern regime.

Survival or extinction of a seed

Some B-rich disturbances shrink and disappear; others persist. This is the first evidence that the field is selecting a dynamical regime rather than merely displaying the initial paint.

Spot growth and division

A spot may expand, narrow at its centre and separate into two domains. Watch the concentration field, not an imagined organism. Division is an instability of shape under local reaction and diffusion.

Labyrinth formation

Neighbouring domains elongate and join into connected bands. The network-like maze is not traced by moving agents; it is a continuous concentration domain maintained across the field.

Exclusion zones

A B-rich region consumes A and changes the conditions around it. Nearby spots may fail to grow or may organise at a characteristic spacing because each pattern modifies its chemical-like neighbourhood.

Regime boundaries

A small feed or kill adjustment can turn stable spots into expanding fronts, shrink patterns to nothing or replace islands with bands. The sharpness of the change is more informative than any one attractive image.

Initial-condition memory

Different seed styles under the same chemistry may converge toward related morphologies or preserve different large-scale organisations. The comparison reveals how much of the final pattern is selected by dynamics and how much is inherited from the initial disturbance.

Toroidal continuity

A band reaching one edge reappears at the opposite edge. This is a periodic world, not a dish with walls.

First 30 seconds

1. Load Prototype Blue Maze and let the progressive warm-up finish. Watch whether the original islands merely blur or actively change topology.
2. Move Feed by one small step while leaving Kill fixed. Wait for the field to respond; reaction–diffusion regimes can have delayed consequences.
3. Restore the preset and move Kill by one small step in the opposite direction. Compare survival, spacing and edge motion.
4. Use Spawn once in an empty region and follow the disturbance. Do not use Brush radius as an experimental variable until the implementation defect is fixed.

Three experiments

Experiment	Question	Do	Watch for	Why it matters
Local regime map	How sharply do pattern types depend on reaction parameters?	Prototype Blue Maze; change Feed in small increments while holding Kill fixed; reset between trials.	Seed extinction, persistence, banding, spot division or uncontrolled expansion.	Teaches that feed selects a regime, not simply an intensity.
History versus law	How much does final form depend on the initial disturbance?	Hold feed/kill fixed; compare Maze islands, Distributed spots and Field bands seed styles.	Convergence toward common local scales versus persistence of large-scale seed geometry.	Separates dynamical selection from inherited morphology.
Intervention and recovery	Can a patterned field incorporate a new local disturbance?	In a mature pattern, use Spawn once to add B; later use Erase to restore A in a nearby patch.	Growth, absorption, splitting, local extinction and reorganisation.	Shows that the pattern is a live concentration regime rather than a static texture.

After brushRadius is repaired, add a fourth controlled experiment comparing the fate of equal-strength disturbances with different radii. Finite-amplitude and finite-size thresholds are scientifically central to Gray–Scott dynamics.

Parameter	What it really controls	Increase it	Decrease it	What to watch
Feed f	Rate at which A is replenished and, in the standard formulation, part of B removal.	Changes resource supply and shifts the pattern regime.	Can starve B-rich domains or select another morphology.	Regime boundaries rather than monotonic “more pattern.”
Kill k	Additional removal rate for B.	Makes B harder to sustain.	Allows B-rich regions to persist or spread more readily.	Extinction, spot size and front behaviour.
Diffusion A	Spatial spreading of the supplied field.	Broader replenishment and altered wavelength.	More local depletion.	Domain spacing and edge smoothness.
Diffusion B	Spatial spreading of the autocatalytic field.	Wider B domains and stronger spatial coupling.	More localised peaks.	Spot width and band continuity.
dt	Numerical time increment per internal step.	Faster apparent evolution but lower stability margin.	Slower, usually safer integration.	Numerical artefacts, not only “chemistry speed.”
Iterations per step	Model time advanced per rendered update.	Faster visible evolution and higher CPU cost.	Slower evolution.	Distinguish pacing from regime.
Seed style	Geometry of finite initial B disturbance.	Categorical.	Categorical.	Initial-condition memory.
Seed count / size	Number and scale of reset-time disturbances.	Larger/more interacting seeds.	Sparse, isolated tests.	Critical disturbance size and collision.
Warm-up steps	Hidden early development before/while the visitor sees the curated world settle.	More mature startup morphology.	More visible initial transient.	What portion of formation the visitor is allowed to witness.
Scale	Grid-cell size relative to viewport.	Coarser grid when pixel scale grows.	Finer grid and greater cost.	Numerical resolution, not physical molecule size.
Brush radius	Intended live intervention size.	Should paint a larger region.	Should paint a smaller region.	Currently inert due to wrapper-mapping defect.
Contrast	Rendering only.	Stronger visual separation of concentrations.	Flatter field appearance.	Must not be interpreted as changed chemistry.

Prototype Blue Maze

The best general introduction. It gives connected and disconnected structures, making topology change easy to see.

Soft Labyrinth

A gentler banded regime useful for following interface motion and for comparing small feed/kill adjustments.

Prototype Field Bands

Foregrounds elongated structures and makes band breakup or reconnection more legible than a spot-dominated field.

Dense Maze Islands

Increases interaction among nearby structures. It is useful for observing competition for A and local exclusion.

Distributed Spots

The clearest preset for spot survival, growth, division and spacing.

Prototype Patches Classic

Connects the wrapper to the older patch-style initialisation and is useful for comparing curated new seeds with the engine’s prior pattern vocabulary.

Recommended learning order:

Distributed Spots → Prototype Blue Maze → Soft Labyrinth → Dense Maze Islands

Brush

Adds B and reduces A over a local area with a softer application than Spawn. In causal terms, it creates a finite chemical-like perturbation. Its intended radius control is currently disconnected.

Spawn

Injects a stronger B-rich disturbance and removes more A. It is the best current tool for testing whether a local seed survives, expands or disappears.

Erase

Sets the region back toward the background state, approximately A = 1 and B = 0. It does not delete a graphical object; it resets local concentrations.

These tools are scientifically valuable because Gray–Scott patterning can depend on disturbance amplitude and size. That value will only be fully realised when Brush radius works and the UI distinguishes reset-time Seed size from live Brush radius.

The blue shapes are not cells, tissues, animal markings or organisms. They are concentration domains in a two-field numerical model.

The system does not contain genes, development, growth of physical tissue or a measured biochemical pathway. Turing’s biological proposal is the intellectual lineage, not a claim of biological fidelity.

The Gray–Scott variables are generic reactant concentrations. The public palette is not the colour of a specified laboratory reagent.

Not every visible Gray–Scott pattern should be called a strict Turing instability. The current implementation deliberately supplies finite B-rich seeds, and Pearson’s canonical Gray–Scott patterns include finite-amplitude phenomena. Publication copy should use “Turing reaction–diffusion lineage” and then identify the exact Gray–Scott model.

The field also does not run a continuous mathematical PDE exactly. It is a finite periodic grid using explicit updates, a chosen Laplacian stencil, clamping and curated warm-up.

Turing Patterns contributes spatial differentiation without moving agents or a drawn template. Physarum builds networks through mobile traffic and environmental memory. Excitable Waves transmits transient pulses through a recovering medium. Turing Patterns shows a third possibility: continuous fields can settle into persistent spatial domains because local reaction and diffusion favour different scales. It gives the atlas a foundational model of how homogeneous ingredients can support heterogeneous form.

Foundational and primary sources

1. Alan M. Turing, “The Chemical Basis of Morphogenesis”, Philosophical Transactions of the Royal Society B 237, 37–72 (1952). DOI: 10.1098/rstb.1952.0012.
2. Peter Gray and Stephen K. Scott, “Autocatalytic reactions in the isothermal, continuous stirred tank reactor: Oscillations and instabilities in the system A + 2B → 3B; B → C”, Chemical Engineering Science 39, 1087–1097 (1984). DOI: 10.1016/0009-2509(84)87017-7.
3. John E. Pearson, “Complex Patterns in a Simple System”, Science 261, 189–192 (1993). DOI: 10.1126/science.261.5118.189.

Editorially important distinction

Turing’s paper defines a broad reaction–diffusion morphogenesis programme and a strict homogeneous-instability mechanism. Pearson’s Gray–Scott study shows rich spatial patterns generated by finite-amplitude perturbations. The public name can honour the broader lineage only if the Field Note preserves that distinction.