All lessons

Module 3: Classical mechanics · Lesson 3.4 · about 55 min

Nonlinear dynamics and chaos

The logistic map from fixed points to chaos: cycles, period doubling, superstable orbits, Feigenbaum's constants and periodic windows; Lyapunov exponents and the limits of prediction; the Lorenz system and the Hénon map; strange attractors and fractal dimensions; Poincaré sections and the KAM picture of chaos in Hamiltonian systems.

Key points, meanings and formulas on one page: choose "Save as PDF" in the print box.

Module 3 · about 55 minutes. By the end you will be able to follow the logistic map from a stable fixed point through period doubling into chaos, compute Lyapunov exponents and prediction horizons, find the fixed points and volume contraction of the Lorenz system and the Hénon map, compute fractal dimensions, and explain what KAM theory says about Hamiltonian chaos.

This lesson continues lesson 3.3, phase space, fixed points and stability, which explains why a flow in the plane cannot be chaotic.

A simple deterministic rule can produce motion that never repeats and that magnifies every tiny uncertainty exponentially. The way order breaks down, by repeated period doubling, is the same for a huge class of systems, down to the numbers.

1. The logistic map

xn+1=f(xn)=rxn(1−xn),0≤x≤1,0<r≤4

A fixed point of a map satisfies x∗=f(x∗); it is stable when the multiplier |f′(x∗)|<1 (not f′<0 as for flows). For the logistic map, x∗=0 (stable for r<1) and x∗=1−1/r, with f′(x∗)=2−r, stable for 1<r<3. At r=2 the multiplier is zero (superstable): errors are squared each step.

At r=3 the multiplier passes through −1 and a stable 2-cycle is born (a flip, or period-doubling, bifurcation). Its points solve f(f(x))=x without being fixed points:

x±=(r+1)±(r−3)(r+1)2r,f′(x+)f′(x−)=4+2r−r2

The 2-cycle’s multiplier reaches −1 at r=1+6≈3.449, giving period 4, then 8 at 3.544, and so on.

Worked example. At r=3.2: x±=(4.2±0.9165)/6.4=0.7995,0.5130, multiplier 4+6.4−10.24=0.16: stable. The superstable 2-cycle (through x=1/2) is at r=1+5≈3.236.

2. Feigenbaum universality and windows

The period-doubling points rn crowd together geometrically:

δ=limn→∞⁡rn−rn−1rn+1−rn=4.6692…,r∞≈3.5699

The same δ (and a second constant α=2.5029 for the shrinking widths) appears for every smooth map with a single quadratic hump, and in experiments on convecting fluids and electronic circuits. Using it: from r2=3.4495 and r3=3.5441, the remaining intervals sum to (r3−r2)/(δ−1)=0.0258, so r∞≈3.570.

Beyond r∞ chaos is interrupted by periodic windows. The biggest, period 3, opens at r=1+8≈3.828 by a tangent (saddle-node) bifurcation of f3; just below it the motion shows intermittency, long nearly periodic stretches broken by chaotic bursts. By Sharkovskii’s theorem, a period-3 orbit in a continuous map of an interval implies orbits of every period.

Top: the long-term values of x for each r (the bifurcation diagram), showing period doubling into chaos and the periodic windows. Bottom: the Lyapunov exponent λ(r): negative on stable cycles, zero at each bifurcation, positive in chaos, and ln 2 at r = 4. Move the slider to read the attractor and λ at any r.

3. Lyapunov exponents and prediction

Two orbits starting δ0 apart separate as δn≈δ0eλn. For a one-dimensional map,

λ=limn→∞⁡1n∑i=0n−1ln⁡|f′(xi)|
λ > 0
sensitive dependence: chaos
λ < 0
orbits converge: a stable fixed point or cycle (λ = ln|multiplier|/period)
λ = 0
a bifurcation point, or neutral motion along a flow
1/λ
the time for errors to grow by a factor e: the Lyapunov time
Physical meaning
A positive exponent means each step reveals new digits of the initial condition: the doubling map 2x mod 1 shifts the binary expansion of x one place per step. Prediction then costs exponentially more precision for each extra step.

Exact values: the doubling map and the tent map with slope 2 have λ=ln⁡2; a tent map with slope μ has ln⁡μ; the logistic map at r=4 also has ln⁡2 (it becomes the tent map under x=sin2⁡(πθ/2), with invariant density 1/πx(1−x)). At a stable fixed point λ=ln⁡|f′(x∗)|, e.g. ln⁡0.5=−0.69 at r=2.5.

The prediction horizon for an initial error δ0 and tolerance a is

tpred≈1λln⁡aδ0

It grows only logarithmically with precision: a million times better data buys 13.8/λ more.

4. Flows and maps in more dimensions

The Lorenz system (a model of convection):

x˙=σ(y−x),y˙=rx−y−xz,z˙=xy−bz
  • The divergence is −(σ+1+b) everywhere, so volumes shrink as e−(σ+1+b)t (e−13.67t for σ=10, b=8/3): the attractor has zero volume.
  • The origin is stable for r<1; at r=1 a pitchfork creates C±=(±b(r−1),±b(r−1),r−1), which lose stability at r≈24.74.
  • At r=28 the Lyapunov exponents are about (0.91,0,−14.57): one positive (chaos), one zero (along the flow), and a sum equal to the divergence.

The Hénon map x′=1−ax2+y, y′=bx is invertible, with Jacobian determinant −b: areas shrink by |b| per step (0.3 at the classic a=1.4). Its fixed points solve ax2+(1−b)x−1=0.

A Poincaré section turns a flow into a map by recording each crossing of a surface: a periodic orbit gives a few points, a quasi-periodic orbit a closed curve, a chaotic one a scattered fractal set.

Worked example. For r=28, b=8/3: C± sit at z=27, x=y=±72=±8.49.

5. Strange attractors and fractal dimensions

A strange attractor combines stretching (which separates orbits) with folding (which keeps them bounded), building a layered, Cantor-like structure of non-integer dimension.

Measure Formula Example
similarity D=ln⁡N/ln⁡s for N copies scaled by 1/s Cantor set ln⁡2/ln⁡3=0.631; Koch ln⁡4/ln⁡3=1.262; Sierpinski ln⁡3/ln⁡2=1.585
box counting N(ϵ)∝ϵ−D, D=ln⁡(N2/N1)/ln⁡(ϵ1/ϵ2) coastlines, attractors
Kaplan-Yorke D=j+∑i≤jλi/|λj+1| Lorenz: 2+0.91/14.57=2.06

Hamiltonian chaos. Without dissipation there are no attractors (Liouville). An integrable system moves on invariant tori. The KAM theorem says that under a small perturbation most tori survive (those with sufficiently irrational frequency ratios), while resonant ones break into island chains and thin chaotic layers. In the standard map the last barrier breaks at K≈0.97, after which chaotic orbits wander through all of phase space.

“Chaos means randomness.” Chaotic systems are fully deterministic: the same initial condition always gives the same orbit. What makes them unpredictable is the exponential growth of the unavoidable error in the initial condition. Equally wrong: “chaos needs many variables”. Three ODEs are enough for a flow, and a one-variable map such as x→4x(1−x) is already chaotic, because it is not invertible.

  • Map fixed point stable if |f′(x∗)|<1. Logistic: x∗=1−1/r, f′=2−r, stable for 1<r<3; superstable at r=2.
  • Period doubling at r=3, 1+6=3.449, 3.544, …; δ=4.669, r∞=3.5699; period-3 window at 1+8.
  • 2-cycle: x±=[(r+1)±(r−3)(r+1)]/2r, multiplier 4+2r−r2.
  • λ=⟨ln⁡|f′|⟩; doubling, tent(2), logistic(4): ln⁡2. Horizon t≈λ−1ln⁡(a/δ0).
  • Lorenz: contraction σ+1+b; C± at z=r−1, x=y=±b(r−1); exponents (+,0,−).
  • Hénon: area factor |b|. Chaos in flows needs 3 dimensions; maps can be chaotic in 1.
  • D=ln⁡N/ln⁡s; Kaplan-Yorke for attractors. KAM: irrational tori survive small perturbations.

Practise: exam questions on this lesson

They start easy and climb to top-rank level. More on the practice page.

Before this: lesson 3.3, phase space, fixed points and stability. Related: Hamilton-Jacobi theory and action-angle variables for the integrable systems that KAM perturbs, and numerical methods for integrating chaotic flows.

Next: module 4 begins with the electromagnetic theory roadmap.

Check yourself

1For the logistic map x → rx(1 − x), the non-zero fixed point is stable for
2A system has Lyapunov exponent 1 per day. Improving the initial data 1000 times extends the useful forecast by about
3The Koch curve is made of 4 copies of itself, each one-third the size. Its fractal dimension is

Sources

  • S. H. Strogatz, Nonlinear Dynamics and Chaos, 2nd ed. (2015), chapters 9 to 12
  • E. Ott, Chaos in Dynamical Systems, 2nd ed. (2002), chapters 1 to 4 and 7
  • M. J. Feigenbaum, "Quantitative universality for a class of nonlinear transformations", J. Stat. Phys. 19, 25 (1978)
  • CSIR-UGC NET Physical Sciences syllabus, Classical Mechanics (non-linear dynamics and chaos)

Your ticks stay in this browser. Save them as a file on the course home.

Updated 2026-11-22. Not affiliated with CSIR, UGC or the National Testing Agency. Check the official CSIR NET website for the current pattern. All questions are original.