Convergence in Complex Systems: Resemblance, Constraint or Kinship?

A shell coils around an axis. A cyclone draws curved cloud bands. A galaxy displays arms around a bright center. Side by side, the images suggest that one law crosses biology, atmosphere and cosmos.

That intuition is powerful. It is also dangerous.

Similarity may indicate common ancestry, a shared constraint, a common class of equations or only visual resemblance. It may also be produced by projection, framing or selective attention. Science begins after wonder.

The useful question is not merely: "Do these objects look alike?" It is: do they share a history? a mechanism? a constraint? an equation? Do they respond similarly to perturbation? Is the similarity quantitative or only qualitative?

This article builds a method for answering those questions. The word convergence has several meanings — biological, mathematical, dynamical, statistical — and these meanings must not be merged. Comparing them, however, reveals a deep idea: some details can become irrelevant while constraints, symmetries and interactions dominate the final pattern.

[[figure:FigSameFormDifferentMechanisms]]

When two similar forms are observed, three questions must be asked in order.

1. What exactly is the resemblance?

Is it the silhouette? A metric relation? A growth law? A topology? A velocity field? A temporal sequence? A function? A perturbation response? A statistic? Two curves may look similar to the eye without sharing the same curvature, radial law or dynamics.

2. What is the generating cause?

The mechanisms that produce the form must be identified: differential growth, forces, transport, reaction-diffusion, natural selection, optimization, architectural constraints, local interaction, feedback, inheritance.

3. What shared testable prediction exists?

A comparison becomes stronger when one model predicts the same shape change under parameter variation, the same thresholds, the same scaling laws, the same invariants, the same response to perturbation. Without predictions, convergence often remains descriptive.

In evolutionary biology, convergence is independently evolved similarity. In mathematics, a sequence converges toward a limit. In numerical analysis, an algorithm converges toward a solution. In dynamical systems, a trajectory may converge toward a fixed point, cycle or attractor. In statistical physics, different systems may converge toward the same large-scale behavior near criticality.

The shared word hides different operations: becoming similar; approaching a limit; losing memory of microscopic details; reaching a stable solution; responding similarly to constraints. These distinctions must be preserved rather than merged.

Similarity may come from common ancestry. The human arm, bat wing and whale flipper share a related skeletal plan. Their functions diverged, but their history is connected. Homology concerns historical and developmental continuity, not perfect visual identity.

Testing kinship uses phylogeny, comparative anatomy, development, genetics, fossils and relative structural position. Kinship is therefore not a visual impression. It is a historical hypothesis supported by converging evidence.

Convergent evolution occurs when distinct lineages independently develop similar traits. Streamlined bodies in dolphins, ichthyosaurs and fish illustrate functional convergence. The aquatic environment imposes costs on drag, propulsion and stability.

Yet convergence does not make the entire organisms identical. Materials, internal structures, developmental paths and genetic histories remain different. Jonathan Losos emphasized an important nuance: convergence in similar environments is often evidence of adaptation, but it does not prove that selection is the only cause. Developmental constraints and accessible variation also channel evolution.

Complete convergence is rare. It is often mosaic: convergence of function, convergence of external form, divergence of internal structure, partial reuse of ancient genes, different solutions at certain scales.

Systems do not explore every imaginable form. Geometry, material, development, dynamics and history restrict accessible configurations. A morphospace represents possible forms in terms of parameters. ^[1]

David Raup showed that a large number of shells could be described using a small number of geometric parameters: tube expansion rate W, distance from axis D, translation along axis T, aperture shape. Varying these parameters generates a broad family of theoretical shells. But real organisms occupy only part of the morphospace.

Why? Some regions may be mechanically unstable, developmentally inaccessible, too costly, functionally poor or historically unreachable. A constraint usually reduces the space of solutions. It does not always select one unique solution.

[[figure:MorphospaceShell]]

The simple opposition between homology and convergence is incomplete. Octopus and vertebrate camera-like eyes evolved independently in their complex organization. Yet ancient developmental genes participate in eye formation across many groups. Shubin, Tabin and Carroll popularized the idea of "deep homology": apparently convergent structures may use deeply homologous genetic toolkits.

One case may therefore combine: convergent final structures; homologous developmental components; distinct developmental pathways; similar selective function. The answer to "convergence or kinship?" may be "both, at different levels."

Statistical physics offers a particularly strong form of convergence. Near continuous phase transitions, systems with different microscopic components may share the same critical exponents. A magnet, a fluid near its critical point and certain lattice models may show common scaling laws. ^[2]

An order parameter may scale as M ~ |t|^β, where t measures the reduced distance to the critical point and β is a critical exponent. The same β can occur in different materials with different particles and interactions.

Why? At large scales, many microscopic details become "irrelevant" in the renormalization group sense. Symmetry, dimension and interaction range dominate. This is stronger than visual similarity. Two systems belong to the same universality class when they share a measurable asymptotic structure: exponents, scaling functions, data collapse.

Complex systems often allow several causal paths to the same outcome. This is equifinality. An identical pattern may be produced by different equations, different parameters, different histories, different architectures.

Equifinality creates an inverse problem. Observing only the final form y, can we recover the mechanism m? Write: $y=F(m,\theta,x_0)$, where m is a model, θ is its parameter set, $x_{0}$ is the initial state and F is the generating process. Different combinations can produce similar y. The inverse problem is then non-unique.

This non-identifiability explains why a final image alone cannot identify the mechanism. Additional evidence is needed: time series, perturbations, multi-scale measurements, independent constraints, model comparison.

[[figure:FigEquifinality]]

Diffusion equations appear in heat, molecules, populations and information models. The shared form reveals a common mathematical structure of spreading and smoothing. The transported entities and physical mechanisms remain different.

A common reduced equation may indicate: (1) a common physical process; (2) a common approximation; (3) universal mathematics; (4) formal analogy only. Variables, units, assumptions, boundary conditions and validity domains must be compared. The mere presence of a Laplacian or exponential is not sufficient.

A logarithmic spiral satisfies $r(\theta)=a\exp(b\theta)$. Its most celebrated geometric property is the constant angle between the tangent and the radius. Many "spiral" forms do not follow this law.

A form may be an Archimedean spiral, a Fermat spiral, a helix, a spiral wave, a precessing trajectory, a rosette, an irregular galactic arm, a curved cloud band or a rotating pattern. Geometry must be identified before mechanism, and mechanism before parameter testing.

Shell

A shell may preserve shape while growing by proportional increments. If each new increment is proportional to existing size, a logarithmic geometry can emerge. The mechanism involves growth at the aperture, material secretion, edge displacement and expansion, biological constraints.

Cyclone

A cyclone is a rotating fluid structure driven by pressure gradients, convection, moisture, planetary rotation and angular momentum. The spiral bands are precipitation and flux structures, not growth lines of a shell.

Galaxy

Galactic arms may involve density waves, disk instabilities, gravitational interaction and star formation. Stars often pass through the arms.

Their visual resemblance is real, but the level of shared mechanism is weak. They share at most abstract themes: rotation, radial variation, broken symmetry, transport, scale evolution. They do not share one universal cause.

[[figure:FigLogFit]]

Cardiac tissue and oscillating chemistry are materially different. Both can behave as excitable media. A cardiac cell activates its neighbors then enters a refractory period. Certain chemical reactions have comparable dynamics: excitation, propagation and recovery.

A generic reaction-diffusion system is:

The molecules of a reaction and cardiac cells are not related. Yet both systems may belong to the same dynamical family of excitable media. The kinship concerns front propagation, refractory period, wave break, rotation around a core, perturbation response. A premature stimulus can break a front and create a rotating wave in both systems. This is mathematical mechanism kinship, not material kinship.

[[figure:ExcitableMedia]]

Spiral arrangements of leaves, seeds or florets are often associated with Fibonacci. The number sequence alone does not explain the mechanism. Physical and developmental models show that local interactions can produce divergence angles near the golden angle. Ingredients may include meristem growth, successive primordium initiation, local inhibition, auxin transport, spatial packing and radial displacement.

The divergence angle near α = 2π(1 − 1/φ) reduces repeated alignments and favors efficient distribution. But different mechanisms can produce similar patterns. The observed convergence between mechanical, biological and geometric models must be analyzed through quantitative predictions: parastichy transitions, initiation order, response to meristem size, defects, growth dynamics.

A scientific comparison defines a distance. For two forms A and B, one can measure Hausdorff distance, curvature difference, fitted parameters, moments, symmetry, topology, spectra, scaling exponents or dynamic response.

In evolutionary studies, methods like Stayton's ask whether lineages became more similar than their ancestors. A simplified condition is $d_{\mathrm{tip}}$ < $d_{\mathrm{ancestor}}$. The inference depends on phylogeny, ancestral reconstruction, trait choice and evolutionary model.

In physics, data collapse is the stronger test. If several curves become one function after normalization, the convergence is quantitative. In dynamics, one compares attractors, spectra, bifurcations, exponents and response functions.

This ladder is pedagogical, not an absolute law. It makes the level of evidence explicit for a given comparison.

1. Level 0 — Visual resemblance: two images look similar.
2. Level 1 — Geometric similarity: a measurement confirms a comparable shape or relation.
3. Level 2 — Functional similarity: the structures serve a similar function.
4. Level 3 — Shared constraint: a shared limitation reduces the solution space.
5. Level 4 — Shared generative model: the same equations or local rules reproduce both systems.
6. Level 5 — Shared quantitative predictions: the model predicts the same thresholds, transitions or scaling laws.
7. Level 6 — Shared perturbation response: the systems respond as the model predicts.
8. Level 7 — Invariants or universality class: microscopic details change, but certain measured properties remain identical.

[[figure:EvidenceLadder]]

Static form is ambiguous. Perturbation reveals dynamics. To distinguish mechanisms, one can modify a parameter, an initial condition, a boundary, a scale, an interaction, a delay or an energy source. Systems sharing a mechanism should often transform in comparable normalized ways.

Examples: modifying excitability changes spiral wave speed and stability; modifying meristem growth changes parastichy numbers; modifying network interactions changes collective transitions; modifying viscosity or rotation changes a fluid regime. Perturbation moves the comparison from image toward causality.

A helix viewed from the front may look circular. A three-dimensional path projected into two dimensions may look spiral. A form may be spiral locally but not globally. A galaxy may have arm segments without following a single logarithmic spiral.

Selection bias is powerful. Searching for spirals everywhere can preserve positive examples and discard counterexamples. Good analysis requires explicit selection rules, negative examples, blind measurement, alternative models and uncertainty intervals.

The comparator below allows selecting two systems and examining their characteristics side by side: geometry, rotating variable, energy source, generating mechanism, equations, key parameters, scale, perturbation response, invariants and convergence evidence level. The score is a teaching tool, not a universal scientific truth.

[[figure:SystemComparator]]

[[figure:FigDecisionTree]]

[[figure:FigSummaryMatrix]]

The project title should not imply one cause behind all sciences. A stronger interpretation is that similar problems restrict possible solutions; some forms are stable or efficient; scaling laws reappear; different systems share reduced models; microscopic details may disappear at large scales; mathematical tools become transferable.

Convergence does not erase differences. It allows asking which details matter and which become secondary. The scientific claim is not: "Everything is the same." It is: "Different systems can become comparable at a precise, measurable and limited level."

Resemblances between disciplines are powerful engines of discovery. They allow equations, methods and intuitions to travel. But analogy becomes dangerous when it replaces explanation.

The method proposed in this article is simple: define similarity; identify the generator; measure parameters; test predictions; perturb the system; search for invariants; state the limits.

The spiral perfectly illustrates this intellectual discipline. A shell, cyclone, galaxy and cardiac wave may all be called spiral. Their scientific kinships are nevertheless radically different. That is not a weakness of Spirals Everywhere. It is the project's deepest question.

The beauty of resemblances does not lie in claiming they all have the same origin. It lies in discovering exactly what they share, what they do not share, and why certain forms reappear.