Spotlight #57 – Chaos, Network Theory & Morphogenesis

Most mechanical systems are nonlinear: springs stiffen at high amplitude, pendulums slow near vertical equilibrium, and a metal beam under compression has two stable resting positions. The Duffing oscillator captures all of these phenomena in a single equation with a cubic spring term:

ẍ + δẋ + αx + βx³ = γ cos(ωt)

For β > 0 and α < 0 the potential energy V(x) = ½αx² + ¼βx⁴ forms a double well: two stable equilibria at x = ±√(|α|/β) separated by an unstable peak at x = 0. Without forcing (γ=0) the oscillator simply settles into whichever well it starts in. With moderate driving the oscillator can jump between wells unpredictably — the hallmark of chaos.

Period doubling and the road to chaos

As forcing amplitude γ increases from zero, the response passes through a period-doubling cascade: period-1 → period-2 → period-4 → period-8 → … → chaos, at successive parameter values following the Feigenbaum ratio δ ≈ 4.669. This same cascade appears in any smooth unimodal map and was one of the first quantitative bridges between different chaotic systems.

Reading Poincaré sections

A Poincaré section samples the phase-space state (x, ẋ) at a fixed phase of the forcing, reducing a 3D continuous flow to a 2D map.

• One dot — period-1 orbit: the system repeats exactly every forcing cycle T = 2π/ω.
• n dots — period-n orbit: the system repeats after n forcing cycles (n:1 subharmonic resonance).
• A fractal cloud — chaos: the Poincaré section itself has fractal structure, with self-similar layering at all scales.
• A closed curve — quasi-periodicity: two incommensurate frequencies, common near stability boundaries.

In the simulation you can watch the Poincaré section evolve in real time. Try increasing γ from 0.28 (two dots) to 0.38 (four dots) to 0.50 (chaotic cloud) to track the period-doubling cascade directly.

Explore in the library:

🌀 Duffing Oscillator 🌿 Bifurcation Diagram 🎲 Chaos Game ⚖️ Double Pendulum

By the 1990s it had become clear that many real-world networks — the Internet's router graph, protein interaction networks, co-authorship graphs, power grids — shared two striking properties: they had very short average path lengths (everyone is within "six degrees of separation") and very high local clustering (your friends tend to know each other). Classical random graphs had short paths but low clustering; regular lattices had high clustering but long paths. Albert-László Barabási and Réka Albert (1999) and Duncan Watts and Steven Strogatz (1998) each found a different piece of the puzzle.

Erdős–Rényi G(N,p) — the random baseline

The simplest model: each of the N(N−1)/2 possible edges is included independently with probability p. The degree distribution is binomial (Poisson-like for large N). A percolation phase transition occurs at the critical probability p_c = 1/N: below it, all components are small (O(log N)); above it, a single "giant component" of size O(N) appears abruptly. The Erdős–Rényi model predicts Poisson degree distributions — but real networks have many more high-degree hubs than this.

Barabási–Albert — preferential attachment and power laws

The BA model grows a network by adding nodes one at a time. Each new node attaches m edges to existing nodes, but not randomly: the probability of connecting to node i is proportional to its current degree k_i ("rich get richer"). This preferential attachment mechanism produces a scale-free degree distribution P(k) ~ k^−3, with a long power-law tail: a small number of very high-degree hubs (like Google or major airports) coexist with the majority of low-degree nodes. Scale-free networks are fault-tolerant (random failures rarely hit hubs) but fragile to targeted attacks on hubs.

Watts–Strogatz — the small-world regime

Starting from a regular ring lattice — each node connected to its k nearest neighbours — each edge is rewired to a random target with probability β.

• β = 0: regular lattice — high clustering coefficient C ≈ 3(k−2)/4(k−1), but long average path length L ~ N/2k.
• β = 1: random graph — low clustering C ≈ k/N, short path length L ~ ln(N)/ln(k).
• β ≈ 0.01–0.1: the small-world sweet spot — near-regular clustering but near-random path length. This regime matches the human cortex, C. elegans neural connectome, the Western US power grid and internet domain topology.

Explore in the library:

🌐 Network Science 🐜 Ant Colony 🐦 Birds Flock 🚌 Bus Bunching

In 1952, the same Alan Turing who cracked the Enigma code and founded computer science published what he considered his most important paper: "The Chemical Basis of Morphogenesis". The central question: how does a spherically symmetric fertilized egg become an asymmetric organism with distinct organs, a body axis, and surface patterns? Turing's answer was counterintuitive — diffusion can cause instability.

The Turing instability

Consider two chemicals mixing in a uniform medium. Normally diffusion smooths out any concentration gradient — it is a stabilizing force. Turing showed that if one chemical (activator) catalyzes its own production and the other (inhibitor) is both produced by the activator and diffuses faster, a homogeneous equilibrium becomes unstable to small spatial fluctuations. These fluctuations grow and saturate into patchy, periodic patterns — spots, stripes, hexagonal arrays — determined entirely by the ratio D_inhibitor / D_activator and the reaction kinetics.

No genetic program specifies "spot at position (x,y)". The pattern self-organizes from random noise through a physics instability. This is one of the most profound examples of spontaneous symmetry breaking in biology.

Gray-Scott dynamics and the (F, k) map

The Gray-Scott model is a clean realization of the Turing instability:

∂U/∂t = Dᵤ∇²U  −  UV²  +  F(1 − U)
∂V/∂t = Dᵥ∇²V  +  UV²  −  (F + k)V

The (F, k) parameter plane contains a rich atlas of pattern types, experimentally mapped by Pearson (1993). Five of the most striking regimes are available as presets:

• Spots (F=0.055, k=0.062) — Dense hexagonal packing of activator spots. Analogous to cheetah markings.
• Stripes (F=0.060, k=0.058) — Parallel bands of activator. Analogous to zebra stripes and reef-fish bars.
• Maze (F=0.029, k=0.057) — Connected labyrinthine channels, similar to patterns on some tropical shells.
• Coral (F=0.037, k=0.060) — Dendritic structures with crenellated edges, resembling coral polyps.
• Mitosis (F=0.028, k=0.053) — Spots that self-replicate by elongating and splitting, directly analogous to cell division kinematics.

Experimental and biological evidence

The Turing mechanism was controversial for 50 years because the specific morphogens weren't known. Evidence accumulated from the 1990s onward:

• Nodal–Lefty signalling in zebrafish pigmentation (Kondo & Asai, 1995) — first direct experimental confirmation.
• WNT–DKK signalling controls spacing of hair follicles (Sick et al., 2006).
• FGF–SPROUTY feedback produces digit spacing during limb development (Miura et al., 2006).
• Ridge patterns on the palate of mice are Turing-patterned (Economou et al., 2012).

Explore in the library:

🔬 Turing Diffusion ⬛ Cellular Automata 🌀 Belousov-Zhabotinsky 🌿 Barnsley Fern