Dynamical Systems: Bifurcations & Chaos¶

Job: develop bifurcation theory and chaos at frontier depth — the center manifold reduction and normal forms, the local Hopf bifurcation (first Lyapunov coefficient, super/subcritical), the global Silnikov / homoclinic bifurcations producing chaos, the Smale horseshoe and the chaos definitions (Devaney / Li–Yorke / positive entropy / horseshoe), and the topological / metric entropy with the variational principle and the Pesin formula.

Core¶

Local bifurcations and normal forms¶

For a parameter-dependent vector field $\dot x=f(x,\mu)$, a bifurcation at $\mu=\mu_0$ is a qualitative change in the orbit structure. The center manifold reduction: at a non-hyperbolic equilibrium (some eigenvalues with $\mathrm{Re}\,\lambda=0$), the dynamics reduces to the center manifold $W^c$ tangent to the center eigenspace, and the reduced dynamics on $W^c$ determines the bifurcation. The normal form is a simplified polynomial (after a near-identity change of coordinates) capturing the bifurcation's universal unfolding.

Bifurcation	Normal form (sketch)	Codim	Dynamic consequence
Saddle-node	$\dot x=\mu+x^2$	1	two equilibria collide and annihilate
Transcritical	$\dot x=\mu x\pm x^2$	1	exchange of stability between two equilibria
Pitchfork (symmetry)	$\dot x=\mu x\pm x^3$	1	symmetry-breaking; one $\to$ two stable equilibria
Hopf	$\dot z=(\alpha+i\beta)z+c	z	^2z+O(

The Hopf bifurcation in detail¶

For a pair of complex-conjugate eigenvalues crossing the imaginary axis at $\mu=0$ (the Hopf bifurcation), the normal form on the center manifold (a 2D disk) is

\[ \dot z = (\alpha+i\beta)z + c\,|z|^2 z + O(|z|^4),\qquad z\in\mathbb{C},\;\alpha=\alpha(\mu),\;\beta=\beta(\mu),\;c\in\mathbb{C}. \]

The first Lyapunov coefficient $\ell_1=\mathrm{Re}\,c$ determines the type:

Theorem (Hopf bifurcation). At a Hopf bifurcation with $\alpha(0)=0$, $\beta(0)\ne 0$, $\ell_1\ne 0$: - Supercritical ($\ell_1<0$): for $\alpha>0$ (small), a stable limit cycle is born with amplitude $r\sim\sqrt{\alpha/|\ell_1|}$ and period $T\sim 2\pi/\beta$. - Subcritical ($\ell_1>0$): for $\alpha<0$, an unstable limit cycle collapses into the equilibrium, and for $\alpha>0$ the equilibrium is unstable (no nearby stable cycle).

The amplitude grows like $\sqrt{\alpha}$ (a square-root scaling) and the period tends to $2\pi/\beta$ as $\alpha\to 0$. The bifurcation is generic (codim-1) when $\ell_1\ne 0$; the degenerate case $\ell_1=0$ is a higher-codimension (Bautin) bifurcation.

Global bifurcations and Silnikov chaos¶

The homoclinic / heteroclinic bifurcations involve an orbit connecting equilibria to themselves / each other; these are global (not reducible to a local normal form), and they produce chaotic dynamics.

Theorem (Silnikov criterion for chaos). Consider a 3D flow with a saddle-focus equilibrium at the origin (eigenvalues $\lambda<0<\mu\pm i\omega$ with $|\lambda|<|\mu|$, i.e. the expanding real part dominates the contracting one) and a homoclinic orbit $\gamma$ returning to the origin. Then in any $C^1$-neighborhood of the vector field, there is a system with a countable infinity of Smale horseshoes in any neighborhood of $\gamma$ — hence a countable infinity of periodic orbits and an uncountable scrambled set (Li–Yorke chaos). The Poincaré return map near $\gamma$ is a horseshoe map.

Bifurcation	Type	Dynamic consequence
Hopf	local (codim 1)	limit cycle born (super/subcritical by $\ell_1$)
Saddle-loop (homoclinic to a saddle)	global	a limit cycle is born / destroyed (homoclinic orbit at parameter)
Silnikov (homoclinic to saddle-focus, $	\lambda	<
Silnikov (homoclinic to saddle-focus, $	\lambda	>
Heteroclinic (saddle-to-saddle)	global	connection of basins; can produce chaos under genericity
Homoclinic to a periodic orbit	global	infinite period (the orbit slows down); bifurcation of cycles

The Smale horseshoe and chaos definitions¶

The Smale horseshoe $\Sigma=\{0,1\}^{\mathbb{Z}$ is the universal model of chaotic dynamics: a diffeomorphism $f$ restricted to a hyperbolic set $\Lambda$ that is conjugate (or semi-conjugate) to the full two-sided shift $\sigma:\Sigma\to\Sigma$. The horseshoe is the symbolic-dynamics prototype (see Hyperbolic Dynamics).

Chaos definition	Statement	Relation
Devaney chaos	(i) topologically transitive, (ii) periodic points dense, (iii) sensitive dependence on initial conditions	(i)+(ii) $\Rightarrow$ (iii) on compact metric spaces (Banks–Brooks–Cairns–Davis–Stacey); overkill — sensitive dependence is automatic
Li–Yorke chaos	there is an uncountable scrambled set $S$: for $x,y\in S$, $\liminf d(f^n x,f^n y)=0$ and $\limsup d(f^n x,f^n y)>0$	weaker than Devaney; period-3 implies Li–Yorke
Period-3-implies-chaos (Li–Yorke)	an interval map with a period-3 orbit has a periodic point of every period and a scrambled set	"Period three implies chaos" (Li–Yorke 1975)
Positive topological entropy	$h_{\mathrm{top}}(f)>0$	implies Li–Yorke chaos on compact intervals (Blanchard–Glasner–Sergiescu)
Horseshoe	$f$ has an invariant hyperbolic set conjugate to a subshift with positive entropy	implies positive entropy, Devaney, Li–Yorke

Topological and metric entropy¶

The topological entropy $h_{\mathrm{top}}(f)$ measures the exponential growth rate of distinguishable orbits: $h_{\mathrm{top}}(f)=\lim_{n\to\infty}\frac1n\log r(n,\varepsilon)$, where $r(n,\varepsilon)$ is the minimal number of $\varepsilon$-balls covering $n$-step orbits (the spanning / separated sets count).

The metric entropy $h_\mu(f)$ (Kolmogorov–Sinai): for an invariant measure $\mu$, $h_\mu(f)=\sup_\mathcal{P}h_\mu(f,\mathcal{P})$ over finite partitions $\mathcal{P}$, with $h_\mu(f,\mathcal{P})=\lim_{n\to\infty}\frac1n H_\mu\bigl(\bigvee_{k=0}^{n-1}f^{-k}\mathcal{P}\bigr)$ (the Shannon entropy of the refinement).

Theorem (variational principle). $h_{\mathrm{top}}(f)=\sup_\mu h_\mu(f)$, the supremum over all $f$-invariant probability measures. The measures attaining the supremum are the measures of maximal entropy (Bowen measure for hyperbolic systems).

Theorem (Ruelle inequality; Pesin formula). For a $C^{1+\alpha}$ diffeomorphism $f$ and an invariant measure $\mu$, $$ h_\mu(f)\;\le\;\int\sum_{\chi_i(x)>0}\chi_i(x)\,d\mu(x), $$ the sum over the positive Lyapunov exponents (counted with multiplicity). Equality (the Pesin formula) holds iff $\mu$ is an SRB measure (sinai–Ruelle–Bowen, see Ergodic & Physical Measures). For the measure of maximal entropy of an Anosov diffeomorphism, the Pesin equality holds, giving $h_\mu=\sum_{\chi_i>0}\chi_i$.

See it¶

The Silnikov bifurcation: a 3D flow with a saddle-focus and a homoclinic orbit produces a Smale horseshoe via the Poincaré return map. Consider the system (in a neighborhood of the origin)

\[ \dot x=\lambda x,\qquad\dot y=-\mu y-\omega z,\qquad\dot z=\omega y-\mu z, \]

with eigenvalues $\lambda<0<\mu\pm i\omega$ (a saddle-focus: one stable direction $\lambda$, two unstable with oscillation $\mu\pm i\omega$), $|\lambda|<|\mu|$. Assume a homoclinic orbit $\gamma$ returning to the origin (i.e. a trajectory $\gamma(t)$ with $\gamma(t)\to 0$ as $t\to\pm\infty$). The Poincaré return map $P$ on a cross-section $\Sigma$ transverse to $\gamma$ near the origin: a point near $\gamma$ returns to $\Sigma$ after a long time $T\to\infty$, and the return map $P$ stretches in one direction (the unstable, by $e^{\mu T}$) and contracts in the other (the stable, by $e^{\lambda T}$), with a rotation by $\omega T$ in the unstable plane. For $|\lambda|<|\mu|$, the expansion dominates the contraction, and $P$ acts as a Smale horseshoe map on a strip of $\Sigma$ — the strip is stretched, folded (by the $\omega T$ rotation), and reinserted into $\Sigma$, producing a horseshoe. Hence: - a countable infinity of periodic orbits (one for each periodic word in the horseshoe shift); - an uncountable scrambled set (the Li–Yorke chaos: the set of points in the horseshoe whose forward orbits shadow two distinct sequences); - positive topological entropy $h_{\mathrm{top}}>0$ (the horseshoe has entropy $\log 2$).

The criterion $|\lambda|<|\mu|$ is sharp: for $|\lambda|>|\mu|$ (the stable direction dominates), the return map contracts more than it expands and no horseshoe forms; the Silnikov chaos requires the eigenvalue ratio. This single computation — the Poincaré return map of a saddle-focus homoclinic orbit is a horseshoe when $|\lambda|<|\mu|$ — is the operational content of the Silnikov theorem: a local eigenvalue ratio (the saddle-focus spectrum) controls a global chaotic consequence (the horseshoe).

Check¶

Compute the amplitude and period of the limit cycle born in a supercritical Hopf bifurcation, and identify the role of the first Lyapunov coefficient.

For the normal form $\dot z=(\alpha+i\beta)z+c|z|^2z+O(|z|^4)$ with $\ell_1=\mathrm{Re}\,c<0$ (supercritical), write $z=re^{i\theta}$; the amplitude equation is $\dot r=\alpha r+\ell_1 r^3+O(r^5)$ and the phase equation is $\dot\theta=\beta+\mathrm{Im}\,c\,r^2+O(r^4)$. The nontrivial equilibrium of the amplitude equation (the limit cycle) is $r^\ast=\sqrt{-\alpha/\ell_1}$ (positive for $\alpha>0$, $\ell_1<0$); the cycle is stable (the linearization of $\dot r$ at $r^\ast$ is $-2|\ell_1|r^{\ast 2}<0$). The period is $T=2\pi/(\beta+\mathrm{Im}\,c\,r^{\ast 2})\to 2\pi/\beta$ as $\alpha\to 0$. The first Lyapunov coefficient $\ell_1$ controls the type (super if $<0$, sub if $>0$) and the amplitude scaling $r\sim\sqrt{\alpha/|\ell_1|}$; the degenerate $\ell_1=0$ is a higher-codim bifurcation (Bautin). The Hopf theorem is the rigorous content: the asymptotic formula $r^\ast\sim\sqrt{\alpha/|\ell_1|}$, $T\sim 2\pi/\beta$ holds for small $\alpha$.

State precisely why the Silnikov criterion requires $|\lambda|<|\mu|$ (the eigenvalue ratio) for the horseshoe, and what happens in the opposite case.

The Poincaré return map $P$ on the cross-section $\Sigma$ near the homoclinic orbit $\gamma$ stretches by $e^{\mu T}$ in the unstable direction (where $T\to\infty$ is the return time near the origin) and contracts by $e^{\lambda T}$ in the stable direction, with a rotation by $\omega T$ in the unstable plane. For $P$ to be a Smale horseshoe, the expansion must dominate the contraction: $e^{\mu T}\cdot e^{-\lambda T}=e^{(\mu-|\lambda|)T}\to\infty$ as $T\to\infty$, requiring $\mu>|\lambda|$ (the unstable real part exceeds the stable contraction). This is the $|\lambda|<|\mu|$ criterion. In the opposite case $|\lambda|>\mu$, the contraction dominates, $e^{(\mu-|\lambda|)T}\to 0$, the return map is contracting (no horseshoe), and the Silnikov chaos fails — the system may still have a homoclinic orbit but not the horseshoe / scrambled set. Hence the eigenvalue ratio is sharp: it controls whether the local dynamics near the saddle-focus produces global chaos via the return map.

Pitfalls¶

The Hopf first Lyapunov coefficient $\ell_1$ determines super vs subcritical: $\ell_1<0$ gives a stable cycle (supercritical), $\ell_1>0$ an unstable one (subcritical); the degenerate $\ell_1=0$ is a higher-codim Bautin bifurcation, not a generic Hopf. Confusing the sign is the standard error.
Silnikov chaos requires $|\lambda|<|\mu|$ (saddle-focus eigenvalue ratio): the opposite ratio gives no horseshoe; the criterion is sharp. The Silnikov theorem is for the saddle-focus with $|\lambda|<|\mu|$, not for an arbitrary homoclinic orbit.
Devaney chaos is overkill: the Banks et al. result shows sensitive dependence is automatic from (i)+(ii); the operative content of Devaney is transitivity + density of periodic points. Do not conflate the three conditions as independent.
Positive entropy $\neq$ Li–Yorke in general (but they coincide on compact intervals): the Blanchard–Glasner–Sergiescu theorem equates them on interval maps, but not on general compact spaces; the implication is one-way in general.
The variational principle gives the supremum, not the maximum for all systems: the measure of maximal entropy exists for hyperbolic systems (Bowen) but not for all topological dynamical systems; the sup is always attained on compact metric spaces by upper-semicontinuity of $\mu\mapsto h_\mu$.
Pesin's formula is an equality for SRB measures; the Ruelle inequality is strict in general: $h_\mu\le\int\sum_{\chi_i>0}\chi_i\,d\mu$ (Ruelle), equality iff $\mu$ is SRB. For non-SRB measures, the inequality is strict — the formula is not an identity for all invariant measures.
The horseshoe has entropy $\log 2$ (full 2-shift), but subshifts of finite type have entropy $\log\rho(A)$ (spectral radius of the transition matrix): the full shift's $\log 2$ is the special case $\rho(A)=2$; do not equate entropy with $\log 2$ for general horseshoes.

Refs¶

Kuznetsov, Elements of Applied Bifurcation Theory; Guckenheimer–Holmes, Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields
Silnikov, A case of the existence of a countable number of periodic motions (Sov. Math. Dokl. 1965)
Smale, Differentiable dynamical systems (BAMS 1967, horseshoe); Devaney, An Introduction to Chaotic Dynamical Systems
Li–Yorke, Period three implies chaos (Amer. Math. Monthly 1975); Banks–Brooks–Cairns–Davis–Stacey, On Devaney's definition of chaos
Walters, An Introduction to Ergodic Theory (variational principle); Pesin, Lyapunov characteristic exponents and smooth ergodic theory (Pesin formula)

Next¶

Hyperbolic Dynamics · Hamiltonian & KAM · Ergodic & Physical Measures · DS index · Foundations · Ergodic theory · Topology · Mathematical physics