Henon Map: Initial Conditions & Attractor Shape

⚡ Key Takeaways

The Hénon map $(x,y)\mapsto(1-ax^2+y,\,bx)$ is a two-dimensional invertible quadratic diffeomorphism whose classical parameters $a=1.4,\,b=0.3$ produce a strange attractor of fractal dimension $\approx 1.26$ . Key result: for almost every initial condition inside the basin of attraction, orbits converge to a single, geometrically fixed attractor whose shape is independent of the starting point; only points in the escaping set diverge to infinity. Applies to dissipative chaos, dimension theory, and dynamical systems pedagogy. Introduced by Michel Hénon in 1976.

💡 The Core Insight

There is a sharp dichotomy between the orbit and the attractor. The choice of initial condition $(x_0,y_0)$ determines which infinite sequence of points you visit and how fast you approach the limit set — but provided $(x_0,y_0)$ lies in the basin of attraction, it does not determine the shape, location, or fractal dimension of the limiting attractor. Initial conditions partition the plane into the basin (orbits captured by the attractor) and the escaping set (orbits diverging to $\infty$ ); the attractor shape is fixed by the parameters $(a,b)$ alone.

DefinitionThe Hénon Map

The Hénon map is the diffeomorphism $H_{a,b}:\mathbb{R}^2\to\mathbb{R}^2$ given by

1

\quad H_{a,b}(x,y)=\bigl(1-ax^2+y,\ bx\bigr)

Derivation hint: This is the standard Hénon (1976) form; the alternative $x_{n+1}=y_n+1-ax_n^2,\ y_{n+1}=bx_n$ is identical componentwise. Here $a>0$ controls the strength of the nonlinear (folding) term and $b$ with $0<|b|<1$ controls dissipation. The orbit of an initial condition $(x_0,y_0)$ is the sequence $(x_n,y_n)=H_{a,b}^n(x_0,y_0)$ , i.e.

2

\quad x_{n+1}=1-ax_n^2+y_n,\qquad y_{n+1}=bx_n.

Derivation hint: Apply [1] componentwise; $y_{n+1}=bx_n$ is the linear contraction, $x_{n+1}$ carries the quadratic nonlinearity. The Jacobian matrix is

3

\quad DH=\begin{pmatrix}-2ax & 1\\ b & 0\end{pmatrix},\qquad \det DH=-b.

Derivation hint: Partial-differentiate the components of [1]; the off-diagonal $b$ and $1$ give $\det = (-2ax)(0)-(1)(b)=-b$ . The constant determinant $-b$ means every iteration multiplies oriented area by $-b$ : contraction (since $|b|<1$ ) plus an orientation reversal.

DefinitionInverse and Invertibility

Because $\det DH=-b\neq0$ for $b\neq0$ , $H_{a,b}$ is a diffeomorphism with explicit inverse

4

\quad H_{a,b}^{-1}(x,y)=\left(\frac{y}{b},\ x-1+a\frac{y^2}{b^2}\right).

Derivation hint: From $y=bx$ solve $x=y/b$ ; substitute into $x_{\text{old new}}=1-ax^2+y_{\text{old}}$ and solve for the preimage's $y$ -coordinate. Invertibility distinguishes Hénon from the (non-invertible) logistic map and makes it a genuine 2D analog of a Poincaré return map.

TheoremExistence of the Attractor — Benedicks–Carleson

For a positive-measure set of parameters near $b=0$ and $a$ close to $2$ , the Hénon map possesses a topologically transitive attractor $\Lambda$ supporting an SRB (Sinai–Ruelle–Bowen) measure, with a positive Lyapunov exponent along the unstable direction. Consequently orbits exhibit sensitive dependence on initial conditions while statistically converging to a unique invariant measure on $\Lambda$ .

TheoremBasin and Escape Dichotomy

For the classical parameters $a=1.4,\ b=0.3$ there exists a bounded trapping region $\mathcal{R}$ (a quadrilateral) with $H(\mathcal{R})\subset\mathrm{int}\,\mathcal{R}$ . Every initial condition in the basin $\mathcal{B}=\bigcup_{n\ge0}H^{-n}(\mathcal{R})$ has $\omega$ -limit set contained in the attractor $\Lambda=\bigcap_{n\ge0}H^n(\mathcal{R})$ ; every initial condition outside $\overline{\mathcal{B}}$ escapes to infinity. The attractor shape $\Lambda$ is independent of the chosen initial condition within $\mathcal{B}$ .

ProofWhy the Attractor Shape Is Initial-Condition Independent

Step 1 — Trapping. Construct a quadrilateral $\mathcal{R}$ whose image $H(\mathcal{R})$ is a horseshoe-like bent strip lying strictly inside $\mathcal{R}$ . Existence is verified by checking the four corner images numerically; this gives $H(\mathcal{R})\subset\mathrm{int}\,\mathcal{R}$ .

Step 2 — Nested intersection. Define $\mathcal{R}_n=H^n(\mathcal{R})$ . Trapping implies $\mathcal{R}_{n+1}\subset\mathcal{R}_n$ . The attractor is the decreasing intersection

5

\quad \Lambda=\bigcap_{n\ge0}H^n(\mathcal{R}).

Derivation hint: A nested family of nonempty compact sets has nonempty compact intersection; $H$ -invariance follows since $H(\Lambda)=\Lambda$ .

Step 3 — Area collapse. By [3], $\mathrm{area}(\mathcal{R}_n)=|b|^n\,\mathrm{area}(\mathcal{R})\to0$ . Thus $\Lambda$ has zero Lebesgue measure: it is a thin (Cantor-like) set transverse to the unstable foliation.

Step 4 — Attraction. Any $(x_0,y_0)\in\mathcal{B}$ eventually enters $\mathcal{R}$ and thereafter stays; its iterates lie in $\mathcal{R}_n$ for ever larger $n$ , so $\mathrm{dist}\bigl(H^n(x_0,y_0),\Lambda\bigr)\to0$ . The limit set is contained in $\Lambda$ for every such start.

Step 5 — Independence. $\Lambda$ in [5] is defined purely through $H_{a,b}$ and $\mathcal{R}$ ; no initial condition appears. Different starts merely yield different orbits dense in (essentially all of, for transitive $\Lambda$ ) the same fixed set. $\blacksquare$

∎

ProofFixed Points and Their Stability

Fixed points solve $x=1-ax^2+bx,\ y=bx$ , giving

6

\quad x_\pm=\frac{(b-1)\pm\sqrt{(1-b)^2+4a}}{2a},\qquad y_\pm=bx_\pm.

Derivation hint: Set $x_{n+1}=x_n=x$ , $y_{n+1}=y_n=y=bx$ ; substitute to get $ax^2+(1-b)x-1=0$ and apply the quadratic formula. Stability follows from the eigenvalues $\lambda$ of $DH$ at $(x_\pm,y_\pm)$ , roots of

7

\quad \lambda^2+2ax_\pm\,\lambda - b=0.

Derivation hint: Characteristic polynomial $\det(DH-\lambda I)=\lambda^2 -(\mathrm{tr})\lambda+\det=\lambda^2+2ax\,\lambda - b$ using $\mathrm{tr}=-2ax,\ \det=-b$ . For $a=1.4,b=0.3$ , $(x_+,y_+)\approx(0.6314,0.1894)$ is a saddle whose unstable manifold's closure is the visible attractor; $(x_-,y_-)$ is also a saddle lying off the attractor. The attractor is thus the closure of the unstable manifold of $(x_+,y_+)$ — again a construction with no reference to initial conditions.

∎

MH

Michel Hénon

1931–2013

A French astronomer and mathematician at the Nice Observatory, Hénon worked on stellar dynamics and the restricted three-body problem. While studying the Lorenz system he sought the simplest discrete map that would reproduce its stretching-and-folding chaos. By collapsing the Lorenz flow's Poincaré section into a two-line recurrence, he produced his 1976 map — deliberately minimal, analytically tractable, and explicitly invertible. His earlier work on the "Hénon–Heiles" Hamiltonian (1964) already revealed his taste for simple models exposing deep chaotic structure.

🧠 Intuition

Picture the trapping quadrilateral as a sheet of dough. Each iteration does three things encoded in [1]: the quadratic term $-ax^2$ folds the sheet over (a parabola-shaped bend), the $+y$ term shears, and the $y'=bx$ rule squashes it to $0.3$ of its width. Fold–shear–squash, repeated forever, lays down infinitely many parallel sheets nested at every scale — the Cantor cross-section that gives fractal dimension $\approx1.26$ . Now drop two nearby drops of ink (two nearby initial conditions). They both get pulled onto the dough's accumulating layers (same shape!) but their individual paths diverge exponentially because the folding repeatedly separates neighbors. That is the whole story: same canvas, unpredictable brushstrokes.

Figure

Hénon attractor: orbit converges to a shape fixed by (a,b), not by the start

ExampleLocating the Fixed Point and Verifying the Saddle

Given: $a=1.4,\ b=0.3$ . Find the "physical" fixed point $(x_+,y_+)$ and confirm it is a saddle.

Step 1 — Apply [6]. Discriminant $=(1-b)^2+4a=(0.7)^2+5.6=0.49+5.6=6.09$ , $\sqrt{6.09}=2.4678$ .

x_+=\frac{(0.3-1)+2.4678}{2(1.4)}=\frac{-0.7+2.4678}{2.8}=\frac{1.7678}{2.8}=0.6313.

y_+=bx_+=0.3(0.6313)=0.1894.

Step 2 — Eigenvalues via [7]. $2ax_+=2(1.4)(0.6313)=1.7676$ . Solve $\lambda^2+1.7676\lambda-0.3=0$ :

\lambda=\frac{-1.7676\pm\sqrt{1.7676^2+1.2}}{2}=\frac{-1.7676\pm\sqrt{4.3244}}{2}=\frac{-1.7676\pm2.0795}{2}.

$\lambda_u=0.1560,\ \lambda_s=-1.9236.$

Step 3 — Classify. Wait: check magnitudes — $|\lambda_s|=1.9236>1$ and $|\lambda_u|=0.156<1$ , so one expanding, one contracting direction.

\boxed{(x_+,y_+)=(0.6313,\,0.1894),\ \ \lambda\approx\{-1.924,\ 0.156\}\Rightarrow \text{saddle}}

ExampleTwo Initial Conditions, Same Limit Set

Given: Iterate [2] from $(x_0,y_0)=(0,0)$ and from $(0.1,0.1)$ , $a=1.4,b=0.3$ .

Step 1 — Start A $(0,0)$ : $(x_1,y_1)=(1-0+0,\ 0)=(1,0)$ ; $(x_2,y_2)=(1-1.4+0,\ 0.3)=(-0.4,0.3)$ ; $(x_3,y_3)=(1-1.4(0.16)+0.3,\ -0.12)=(1.076,-0.12)$ .

Step 2 — Start B $(0.1,0.1)$ : $(x_1,y_1)=(1-1.4(0.01)+0.1,\ 0.03)=(1.086,0.03)$ ; $(x_2,y_2)=(1-1.4(1.179)+0.03,\ 0.3258)=(-0.621,0.326)$ .

Step 3 — Observe divergence then capture. After $\sim$ 5 steps the two orbits are macroscopically different (sensitive dependence), yet after discarding $\sim$ 1000 transient iterations, the point clouds coincide — both fill the same crescent attractor bounded roughly by $x\in[-1.28,1.27],\ y\in[-0.38,0.38]$ .

\boxed{\text{Different transients} \Rightarrow \text{identical limiting attractor shape};\ D_0\approx1.26}

⚠️ Common mistake

"Changing the initial condition changes the attractor." False inside the basin. The attractor's geometry, box-counting dimension ( $\approx1.26$ ), and SRB measure are functions of $(a,b)$ only. What the initial condition changes is the transient and the specific point sequence. The one case where the start matters qualitatively is whether it lies inside the basin at all — start outside and the orbit diverges to $\infty$ (no attractor reached).

⚠️ Common mistake

"Sensitive dependence means the long-term statistics are unpredictable." The pointwise future is unpredictable, but the statistical future is not: histograms of long orbits converge to the SRB measure regardless of initial condition (for a.e. start). Predictability of averages coexists with unpredictability of trajectories.

CounterexampleAn Initial Condition That Does NOT Reach the Attractor

Take $(x_0,y_0)=(10,10)$ with $a=1.4,b=0.3$ . Then $x_1=1-1.4(100)+10=-129$ , $x_2=1-1.4(129^2)+3=-23297$ , and $|x_n|\to\infty$ super-exponentially. This orbit never approaches $\Lambda$ — it lies in the escaping set. Hence the claim "every initial condition converges to the Hénon attractor" is false; the correct statement restricts to the basin $\mathcal{B}$ .

CounterexamplePeriodic Windows Hide the Strange Attractor

At $a=1.4,b=0.3$ the attractor is strange, but at, e.g., $a=1.0,b=0.3$ the attractor is a stable period-7 cycle — a finite point set, dimension $0$ . Picking initial conditions there shows orbits converging to seven points, not a fractal. Thus "Hénon $\Rightarrow$ strange attractor" depends entirely on parameters, not on initial conditions.

connection

The Hénon map is the discrete sibling of the Lorenz System: Hénon distilled the Lorenz Poincaré-section folding into two lines. Its fold-and-stretch mechanism is a smooth realization of the Smale Horseshoe, which provides the symbolic dynamics underlying chaos. The cross-sectional Cantor structure ties it to Fractal Dimension and the Box-Counting Dimension; its exponential separation of orbits is quantified by Lyapunov Exponents. The statistical convergence to a fixed measure is governed by SRB Measures, and the basin/escape split exemplifies Basins of Attraction. The 1D shadow of Hénon as $b\to0$ is the Logistic Map.

🔬 Where this appears

⏳ History

- 1963 — Edward Lorenz publishes "Deterministic Nonperiodic Flow," exhibiting the first strange attractor in a flow and motivating discrete analogs.

- 1964 — Hénon & Heiles study a 2-DOF Hamiltonian, revealing chaotic Poincaré sections in celestial mechanics.

- 1967 — Stephen Smale formalizes the horseshoe, the topological skeleton of folding chaos.

- 1976 — Michel Hénon publishes "A two-dimensional mapping with a strange attractor" (Commun. Math. Phys.), introducing [1] with

a=1.4,b=0.3

.

- 1980 — Numerical estimates pin the attractor's fractal dimension near

1.26

and Lyapunov exponent

\lambda_1\approx0.42

.

- 1985 — Benedicks–Carleson begin the program proving abundance of chaos in 1D unimodal maps, the technical precursor to the 2D case.

- 1991 — Benedicks & Carleson prove, for a positive-measure parameter set near

b=0

, that the Hénon map has a genuine strange attractor with positive Lyapunov exponent.

- 1993 — Benedicks & Young construct the SRB measure for these Hénon attractors, establishing statistical regularity.

- 2000s — Rigorous computer-assisted dimension and topological-entropy bounds (interval arithmetic) refine the classical-parameter picture.

open-problem

At the classical parameters $a=1.4,b=0.3$ it is not rigorously proven that the observed object is a strange attractor in the Benedicks–Carleson sense — those theorems hold for small $b$ near a positive-measure parameter set, which may not include $(1.4,0.3)$ . The set could in principle contain a long-period stable sink with a microscopically thin basin. Open questions: (i) Determine the exact Hausdorff (vs. box-counting) dimension of the classical attractor. (ii) Characterize the full measure-theoretic structure of the basin boundary, which appears fractal. (iii) Map the precise parameter regions where the attractor is strange versus periodic — the "Hénon parameter zoo" is only partially understood, with intricate interleaving of chaotic and periodic windows.

Henon Map: Initial Conditions & Attractor Shape

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also