Dynamical Systems, Phase Space and Stability
CSIR NET advanced mechanics: hyperbolic fixed points, Lyapunov functions, limit cycles, bifurcations, Kapitza pendulum, logistic map and chaos.
- 23 min
On this page · 16 sections
- Prerequisites
- The Hook: The Pendulum That Stands on Its Head
- Flows, Maps and Phase Space
- Linearisation and When to Trust It
- Lyapunov’s Direct Method
- Closed Orbits in the Plane
- Bifurcations
- Stability of Periodic Orbits and Parametric Effects
- Maps and the Route to Chaos
- Memory Tricks
- Don’t Confuse
- Common Mistakes
- Practice
- Quick Revision Table
- Key Takeaways
- What’s Next
Prerequisites
- Newton’s Laws, Phase Space and Stability: phase portraits, the trace–determinant classification of 2D fixed points, the pendulum separatrix and Liouville’s theorem. This page assumes all of it and goes one level deeper.
- Small Oscillations and Normal Modes: linearising about an equilibrium.
- Linear Algebra, Matrices and Eigenvalues: eigenvalues of a \( 2\times2 \) and \( 3\times3 \) matrix in a few lines.
The Hook: The Pendulum That Stands on Its Head
Flows, Maps and Phase Space
A continuous dynamical system (a flow) is a set of first-order equations
$$\dot{\mathbf{x}} = \mathbf{f}(\mathbf{x}, t), \qquad \mathbf{x}\in\mathbb{R}^n$$It is autonomous if \( \mathbf{f} \) has no explicit \( t \). A non-autonomous system in \( n \) dimensions becomes autonomous in \( n + 1 \) dimensions by adding \( \dot\theta = 1 \) with \( \theta = t \) (the extended phase space). This is why a driven one-dimensional oscillator, with phase space \( (x, \dot x) \) plus time, can be chaotic.
A discrete dynamical system (a map) is an iteration
$$x_{n+1} = F(x_n)$$Maps arise directly (population models) or from flows by sampling: the Poincaré map records where a trajectory pierces a chosen surface each time round. A periodic orbit of the flow becomes a fixed point of the map.
| Object | Flow \( \dot x = f(x) \) | Map \( x_{n+1} = F(x_n) \) |
|---|---|---|
| Fixed point | \( f(x^*) = 0 \) | \( F(x^*) = x^* \) |
| Linear stability (1D) | stable if \( f'(x^*) < 0 \) | stable if \( \lvert F'(x^*)\rvert < 1 \) |
| Linear stability (nD) | all \( \operatorname{Re}\lambda < 0 \) of the Jacobian | all \( \lvert\lambda\rvert < 1 \) of the Jacobian |
| Volume contraction rate | \( \nabla\cdot\mathbf{f} \) | \( \lvert\det DF\rvert \) per step |
| Minimum dimension for chaos | 3 | 1 (non-invertible map), 2 (invertible map) |
Kinds of stability
For a fixed point \( \mathbf{x}^* \):
- Lyapunov stable: trajectories that start close stay close for all later time.
- Asymptotically stable: Lyapunov stable and nearby trajectories actually converge to \( \mathbf{x}^* \).
- Unstable: not Lyapunov stable.
A centre of the harmonic oscillator is Lyapunov stable but not asymptotically stable. The distinction matters in Part C statements.
Linearisation and When to Trust It
Near a fixed point, \( \dot{\boldsymbol\eta} = J\boldsymbol\eta \) with \( J_{ij} = \partial f_i/\partial x_j \) at \( \mathbf{x}^* \). The solution is a sum of terms \( \mathbf{v}e^{\lambda t} \) over eigenpairs of \( J \). The core page gives the trace–determinant table for two dimensions; here are the rules that decide whether that table tells the truth about the nonlinear system.
At a hyperbolic saddle, the eigenvectors of the linear system become curved stable and unstable manifolds: the set of points that flow into the saddle as \( t\to+\infty \), and the set that flow into it as \( t\to-\infty \). The pendulum separatrix is made of these manifolds joining the saddle at \( \theta = \pi \) to itself.
Three dimensions and higher
The rule is the same: the fixed point is asymptotically stable if every eigenvalue has negative real part. With one real eigenvalue and a complex pair, a fixed point can attract along one direction and spiral out in the plane of the pair (a saddle-focus), a common ingredient of chaotic flows.
Hamiltonian fixed points can never attract
For a Hamiltonian system the Jacobian at a fixed point is \( J = \Omega\,\nabla^2H \), with \( \Omega = \begin{pmatrix}0 & I\\ -I & 0\end{pmatrix} \). Its eigenvalues come in pairs \( \lambda, -\lambda \) (and with complex conjugates, in quartets). So the only possibilities are saddles, centres, and combinations of them. A Hamiltonian fixed point is never asymptotically stable: that would shrink phase-space volume, which Liouville’s theorem forbids.
Lyapunov’s Direct Method
When linearisation fails, look for an energy-like function. A function \( V(\mathbf{x}) \) is a Lyapunov function for the fixed point \( \mathbf{x}^* \) if, in a neighbourhood,
- \( V(\mathbf{x}^*) = 0 \) and \( V(\mathbf{x}) > 0 \) elsewhere (positive definite), and
- \( \dot V = \nabla V\cdot\mathbf{f} \leq 0 \) along trajectories.
Then \( \mathbf{x}^* \) is Lyapunov stable. If \( \dot V < 0 \) away from \( \mathbf{x}^* \), it is asymptotically stable. No trajectories need to be computed; you only differentiate \( V \) along the flow.
Natural choices for \( V \): the energy for mechanical systems, \( x^2 + y^2 \) or \( ax^2 + by^2 \) with constants tuned so the cross terms in \( \dot V \) cancel.
Gradient systems \( \dot{\mathbf{x}} = -\nabla U \) have \( U \) as a Lyapunov function automatically (\( \dot U = -\lvert\nabla U\rvert^2 \)). They have no closed orbits and every minimum of \( U \) is asymptotically stable.
Closed Orbits in the Plane
Three tools decide whether a two-dimensional flow has periodic orbits.
| Tool | Statement | Use |
|---|---|---|
| Poincaré–Bendixson | A bounded trajectory in a closed, bounded region of the plane that contains no fixed point approaches a closed orbit | Prove a limit cycle exists (trap the flow in an annulus) |
| Bendixson–Dulac | If \( \nabla\cdot(g\mathbf{f}) \) has one strict sign throughout a simply connected region, for some smooth \( g \), there is no closed orbit entirely inside it | Prove closed orbits do not exist |
| Index theory | Nodes, spirals and centres have index \( +1 \), saddles \( -1 \); a closed orbit must enclose fixed points whose indices sum to \( +1 \) | Rule out orbits around a lone saddle |
Gradient systems and any system with a strict Lyapunov function over the whole region also have no closed orbits.
Limit cycles
A limit cycle is an isolated closed orbit. It needs energy input and loss that balance on average, so it never occurs in a conservative system (whose closed orbits come in continuous families of nested curves).
The cleanest example is written in polar coordinates:
$$\dot r = r(\mu - r^2), \qquad \dot\theta = \omega$$For \( \mu > 0 \), \( \dot r > 0 \) for \( r < \sqrt\mu \) and \( \dot r < 0 \) for \( r > \sqrt\mu \): every trajectory except the origin winds onto the circle \( r = \sqrt\mu \), a stable limit cycle of period \( 2\pi/\omega \). For \( \mu < 0 \) the origin is a stable spiral and there is no cycle.
The van der Pol oscillator \( \ddot x - \epsilon(1 - x^2)\dot x + x = 0 \) is the physical prototype. For small \( \epsilon > 0 \), averaging over one cycle gives a limit cycle of amplitude 2 (the energy pumped in at small \( \lvert x\rvert \) balances the energy removed at large \( \lvert x\rvert \)). For large \( \epsilon \) the motion becomes relaxation oscillations: slow drifts punctuated by fast jumps.
Bifurcations
A bifurcation is a qualitative change in the phase portrait as a parameter crosses a critical value. Four normal forms cover almost every exam question.
| Bifurcation | Normal form | What happens as \( r \) increases through 0 |
|---|---|---|
| Saddle-node | \( \dot x = r + x^2 \) | Two fixed points \( \pm\sqrt{-r} \) collide and vanish |
| Transcritical | \( \dot x = rx - x^2 \) | Fixed points \( 0 \) and \( r \) pass through each other and exchange stability |
| Supercritical pitchfork | \( \dot x = rx - x^3 \) | Origin loses stability; two stable points \( \pm\sqrt r \) appear |
| Subcritical pitchfork | \( \dot x = rx + x^3 \) | Two unstable points \( \pm\sqrt{-r} \) merge into the origin, which becomes unstable; trajectories jump far away |
| Supercritical Hopf | \( \dot r = \mu r - r^3,\ \dot\theta = \omega \) | Stable spiral becomes unstable; a stable limit cycle of radius \( \sqrt\mu \) grows |
The signature of a Hopf bifurcation in a 2D linearisation is a complex pair crossing the imaginary axis: the trace changes sign while the determinant stays positive. Saddle-node, transcritical and pitchfork bifurcations all have a real eigenvalue passing through zero.
graph TD
A[Parameter crosses a critical value] --> B{Which eigenvalue hits the imaginary axis?}
B -->|A real eigenvalue passes through 0| C{How do the fixed points change?}
B -->|A complex pair crosses with nonzero imaginary part| H[Hopf: limit cycle is born or dies]
C -->|Two appear or vanish together| SN[Saddle-node]
C -->|Two pass through each other| TC[Transcritical]
C -->|Symmetric pair splits off| PF[Pitchfork]Mechanical example: a bead on a hoop of radius \( R \) spinning at \( \Omega \) about a vertical diameter. The bottom is stable for \( \Omega^2 < g/R \); above that it becomes unstable and two symmetric off-axis equilibria at \( \cos\theta = g/(R\Omega^2) \) appear. That is a supercritical pitchfork, with \( \Omega^2 - g/R \) as the parameter.
Stability of Periodic Orbits and Parametric Effects
A periodic orbit is stable if the Poincaré map has a stable fixed point there. Its linearisation has eigenvalues called Floquet multipliers (one of them for a flow is always 1, the direction along the orbit). The orbit is stable if all the others lie inside the unit circle.
For linear equations with periodic coefficients, Floquet theory gives parametric resonance. For
$$\ddot x + \omega_0^2\left(1 + h\cos\gamma t\right)x = 0, \qquad h \ll 1$$the rest state \( x = 0 \) is unstable in narrow bands of the driving frequency near \( \gamma = 2\omega_0/n \), \( n = 1, 2, \ldots \). The widest and most important band is at \( \gamma \approx 2\omega_0 \): pumping a swing by standing and squatting twice per cycle. Unlike ordinary resonance, the amplitude grows exponentially, and only if the system is already slightly displaced.
The Kapitza pendulum
For a pendulum of length \( l \) whose pivot moves vertically as \( a\cos\gamma t \), with \( \gamma \gg \sqrt{g/l} \) and \( a \ll l \), averaging over the fast oscillation gives an effective potential for the slow angle \( \phi \) (measured from the downward vertical):
$$U_{\text{eff}}(\phi) = -mgl\cos\phi + \frac{ma^2\gamma^2}{4}\sin^2\phi$$At \( \phi = \pi \), \( U_{\text{eff}}'' = -mgl + \tfrac{1}{2}ma^2\gamma^2 \). The inverted position is stable when
$$\boxed{a^2\gamma^2 > 2gl}$$The fast term acts like a potential proportional to the mean square of the fast force; it pushes the system towards where that force is weakest, which for the pendulum is the vertical line.
Maps and the Route to Chaos
The logistic map \( x_{n+1} = rx_n(1 - x_n) \) on \( [0, 1] \), \( 0 < r \leq 4 \), is the standard exam map.
| Range of \( r \) | Long-time behaviour |
|---|---|
| \( 0 < r < 1 \) | \( x_n \to 0 \) |
| \( 1 < r < 3 \) | \( x_n \to x^* = 1 - 1/r \); multiplier \( F'(x^*) = 2 - r \) |
| \( 3 < r < 1 + \sqrt6 \approx 3.449 \) | stable 2-cycle |
| up to about 3.57 | period doublings 4, 8, 16, … accumulating |
| beyond about 3.57 | chaos, interrupted by periodic windows (a period-3 window opens at \( 1 + \sqrt8 \approx 3.83 \)) |
| \( r = 4 \) | fully chaotic on \( [0, 1] \), Lyapunov exponent \( \ln 2 \) |
The 2-cycle is found from \( F(F(x)) = x \) after dividing out the fixed points:
$$x_\pm = \frac{(r + 1)\pm\sqrt{(r - 3)(r + 1)}}{2r}$$and its multiplier is \( F'(x_+)F'(x_-) = -r^2 + 2r + 4 \). It equals \( +1 \) at \( r = 3 \) (birth) and \( -1 \) at \( r = 1 + \sqrt6 \) (the next doubling).
The ratio of successive intervals between period doublings tends to the Feigenbaum constant \( \delta \approx 4.669 \), the same for every smooth one-humped map. A multiplier passing through \( -1 \) signals a period doubling; through \( +1 \), a saddle-node (or the birth of a pair).
Lyapunov exponents
Sensitive dependence on initial conditions is measured by
$$\lambda = \lim_{N\to\infty}\frac{1}{N}\sum_{n=0}^{N-1}\ln\lvert F'(x_n)\rvert$$for a 1D map (for a flow, the growth rate of a small separation, \( \lvert\delta(t)\rvert \sim \lvert\delta_0\rvert e^{\lambda t} \)). \( \lambda < 0 \) on a stable cycle, \( \lambda = 0 \) at a bifurcation point, \( \lambda > 0 \) for chaos. For a flow, one exponent is always zero (along the trajectory), and the sum of all exponents equals the average of \( \nabla\cdot\mathbf{f} \), so it is zero for a Hamiltonian flow and negative for a dissipative one.
Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| Lyapunov stable (stays close) | Asymptotically stable (converges) | |
| Linear centre (eigenvalues \( \pm i\omega \)) | Nonlinear centre (needs a conserved quantity) | |
| Limit cycle: isolated, attracts or repels | Conservative closed orbits: continuous family, neutral | |
| Saddle-node: fixed points created from nothing | Pitchfork: one point splits into three | |
| Supercritical: small, continuous onset | Subcritical: sudden jump, hysteresis | |
| Ordinary resonance: driving at \( \omega_0 \), linear growth without damping | Parametric resonance: modulating at \( 2\omega_0 \), exponential growth | |
| Lyapunov function (proves stability) | Lyapunov exponent (measures divergence of nearby trajectories) | |
| Flow fixed point stable: \( f'(x^*) < 0 \) | Map fixed point stable: \( \lvert F'(x^*)\rvert < 1 \) |
Common Mistakes
Practice
All questions below are original practice questions written for this page. They are not from any CSIR NET paper.
Level 1: Foundation
- Find the fixed points of \( \dot x = x - x^3 \) and their stability.
- State the condition under which linearisation correctly predicts the type of a fixed point.
- For the logistic map with \( r = 2.5 \), find the non-zero fixed point and decide whether it is stable.
Answers
- \( x^* = 0, \pm1 \). \( f'(x) = 1 - 3x^2 \): \( f'(0) = 1 > 0 \), unstable; \( f'(\pm1) = -2 < 0 \), both stable. (A supercritical pitchfork at parameter value 1.)
- The fixed point must be hyperbolic: no eigenvalue of the Jacobian with zero real part (Hartman–Grobman).
- \( x^* = 1 - 1/r = 0.6 \). \( F'(x^*) = r(1 - 2x^*) = 2 - r = -0.5 \), and \( \lvert -0.5\rvert < 1 \), so it is stable; iterates approach it alternating from either side.
Level 2: Part B Style
Practice Q1. The fixed points of \( \dot x = y \), \( \dot y = -x + x^3 \) are:
(a) a centre at the origin and saddles at \( (\pm1, 0) \) (b) a saddle at the origin and centres at \( (\pm1, 0) \) (c) a stable spiral at the origin and saddles at \( (\pm1, 0) \) (d) three centres
Answer: (a)
The system is conservative, \( \ddot x = -V'(x) \) with \( V = \tfrac{1}{2}x^2 - \tfrac{1}{4}x^4 \). \( V''(x) = 1 - 3x^2 \): at \( x = 0 \), \( V'' = 1 > 0 \) (a minimum, so a true centre, with eigenvalues \( \pm i \)); at \( x = \pm1 \), \( V'' = -2 < 0 \) (maxima, saddles with eigenvalues \( \pm\sqrt2 \)). Option (c) is impossible for a conservative system.
Practice Q2. For the linear system \( \dot x = -x + 4y \), \( \dot y = -x - y \), the origin is:
(a) a stable node (b) a stable spiral (c) a saddle (d) a centre
Answer: (b)
\( J = \begin{pmatrix}-1 & 4\\ -1 & -1\end{pmatrix} \): \( \tau = -2 \), \( \Delta = 1 + 4 = 5 \), \( \tau^2 - 4\Delta = -16 < 0 \). Eigenvalues \( -1 \pm 2i \): complex with negative real part, a stable spiral that decays as \( e^{-t} \) and turns with angular frequency 2.
Practice Q3. For \( \dot x = -y - x^3 \), \( \dot y = x - y^3 \), the origin is:
(a) a centre, because the Jacobian has eigenvalues \( \pm i \) (b) unstable (c) asymptotically stable (d) a saddle
Answer: (c)
The linear centre is not hyperbolic, so linearisation is inconclusive and (a) is the trap. With \( V = x^2 + y^2 \), \( \dot V = -2(x^4 + y^4) < 0 \) away from the origin, so the origin is asymptotically stable.
Practice Q4. In polar form, \( \dot r = r(0.25 - r^2) \), \( \dot\theta = 3 \). The long-time motion of a trajectory starting at \( r = 2 \) is:
(a) a spiral into the origin (b) a circle of radius 0.5 traversed with period \( 2\pi/3 \) (c) a circle of radius 0.25 traversed with period \( 2\pi/3 \) (d) escape to infinity
Answer: (b)
\( \dot r = 0 \) at \( r = \sqrt{0.25} = 0.5 \). For \( r > 0.5 \), \( \dot r < 0 \), so the trajectory shrinks onto \( r = 0.5 \); it is a stable limit cycle. The angular speed is 3, so the period is \( 2\pi/3 \). Option (c) confuses \( \mu \) with \( \sqrt\mu \).
Level 3: Part C Style
Practice Q5. Consider \( \dot x = y \), \( \dot y = -x - y(1 + x^2) \). Which of the following statements are correct?
- The origin is the only fixed point.
- The origin is a stable spiral.
- The system has no closed orbit anywhere in the plane.
- The phase-space area of a set of initial conditions is conserved.
(a) 1 and 2 only (b) 1, 2 and 3 only (c) 1 and 3 only (d) 1, 2, 3 and 4
Answer: (b) 1, 2 and 3 only
- \( y = 0 \) and then \( -x = 0 \). True.
- \( J = \begin{pmatrix}0 & 1\\ -1 & -1\end{pmatrix} \) at the origin: \( \tau = -1 \), \( \Delta = 1 \), \( \tau^2 - 4\Delta = -3 < 0 \). Hyperbolic, so the stable spiral of the linearisation is the true behaviour. True.
- \( \nabla\cdot\mathbf{f} = 0 + \partial_y[-x - y(1 + x^2)] = -(1 + x^2) < 0 \) everywhere, and the plane is simply connected. By Bendixson–Dulac (with \( g = 1 \)) there is no closed orbit. True.
- Area shrinks at the local rate \( -(1 + x^2) \). False.
Practice Q6. A pendulum of length 0.20 m has its pivot driven vertically as \( a\cos\gamma t \) with \( a = 1.0 \) cm. Take \( g = 9.8 \) m s\( ^{-2} \). The minimum driving frequency \( \gamma/2\pi \) that stabilises the inverted position is closest to:
(a) 5 Hz (b) 14 Hz (c) 32 Hz (d) 198 Hz
Answer: (c)
Stability of the top requires \( a^2\gamma^2 > 2gl \), so
$$\gamma_{\min} = \frac{\sqrt{2gl}}{a} = \frac{\sqrt{2\times9.8\times0.20}}{0.010} = \frac{1.98}{0.010} \approx 198\ \text{rad s}^{-1}$$and \( \gamma_{\min}/2\pi \approx 31.5 \) Hz. Option (d) is the angular frequency mislabelled as hertz. Check the assumptions: \( a/l = 0.05 \ll 1 \) and \( \gamma \approx 198 \) rad s\( ^{-1} \gg \sqrt{g/l} = 7 \) rad s\( ^{-1} \), so the averaging is justified.
Practice Q7. For the logistic map \( x_{n+1} = 3.2\,x_n(1 - x_n) \), which statements are correct?
- The fixed point \( x^* = 0.6875 \) is unstable.
- The map has a 2-cycle with points approximately 0.513 and 0.799.
- The 2-cycle is stable, with multiplier 0.16.
- The Lyapunov exponent on the attractor is positive.
(a) 1 and 2 only (b) 1, 2 and 3 only (c) 2 and 3 only (d) 1, 2, 3 and 4
Answer: (b) 1, 2 and 3 only
- \( x^* = 1 - 1/3.2 = 0.6875 \); \( F'(x^*) = 2 - r = -1.2 \), and \( \lvert -1.2\rvert > 1 \). Unstable. True.
- \( x_\pm = \dfrac{4.2 \pm\sqrt{0.2\times4.2}}{6.4} = \dfrac{4.2 \pm 0.9165}{6.4} = 0.7995,\ 0.5130 \). Check: \( 3.2\times0.5130\times0.4870 = 0.7995 \). True.
- Multiplier \( = r^2(1 - 2x_+)(1 - 2x_-) = -r^2 + 2r + 4 = -10.24 + 6.4 + 4 = 0.16 \), and \( \lvert 0.16\rvert < 1 \). True; \( r = 3.2 \) lies between 3 and 3.449.
- On a stable cycle the Lyapunov exponent is \( \tfrac{1}{2}\ln 0.16 \approx -0.92 < 0 \). False.
Quick Revision Table
| Idea | Result |
|---|---|
| Hyperbolic fixed point | no eigenvalue with zero real part; linearisation is faithful |
| Flow stability | all \( \operatorname{Re}\lambda < 0 \) |
| Map stability | all \( \lvert\lambda\rvert < 1 \) |
| Lyapunov function | \( V > 0 \), \( \dot V \leq 0 \): stable; \( \dot V < 0 \): asymptotically stable |
| Hamiltonian eigenvalues | \( \pm\lambda \) pairs: saddles and centres only |
| No closed orbits | Bendixson–Dulac (one-signed divergence, simply connected), gradient systems, index \( \neq +1 \) |
| Limit cycle example | \( \dot r = r(\mu - r^2) \): radius \( \sqrt\mu \) |
| Bifurcations | saddle-node \( r + x^2 \); transcritical \( rx - x^2 \); pitchfork \( rx \mp x^3 \); Hopf |
| Parametric resonance | strongest at \( \gamma \approx 2\omega_0 \) |
| Kapitza pendulum | top stable if \( a^2\gamma^2 > 2gl \) |
| Logistic map | \( x^* = 1 - 1/r \), stable for \( 1 < r < 3 \); 2-cycle stable to \( 1 + \sqrt6 \) |
| Feigenbaum | \( \delta \approx 4.669 \) |
| Lorenz | \( C_\pm \) exist for \( r > 1 \); divergence \( -(\sigma + 1 + b) \) |
Key Takeaways
What’s Next
- Poisson Brackets and Canonical Transformations: the structure that makes Hamiltonian flows volume-preserving.
- Phase Space, Microstates and the Microcanonical Ensemble: Liouville’s theorem as the foundation of statistical mechanics.
- Singular Solutions and Systems of First-Order ODEs: the matrix-exponential view of linear systems, as Mathematical Sciences tests it.
- Classical Mechanics (Advanced): Quick Revision.