Nonlinear Dynamics - Cheat Sheet 3.3.1 Supercritical Pitchfork Bifurcation (Forward bifu.

fork Bifurcation (Forward bifu., phase transition of 2nd order, soft bifu., safe 4.2 Nonuniform Oscillator
by Michael Garstka bifu.) • θ̇ = ω − a sin(θ) (here: a ≥ 0, ω ≥ 0) (Phase changes non-uniformally)
2 1D Problems • Normal form: ẋ = rx − x 3
(invariant: x ←→ −x, left-right-symmetry) • a < ω: Phase takes long through bottleneck π 2
, speeds up afterwards.
√ • a = ω: Systems stops oscillating at new half-stable f.p θ = π (S-N-B)
2.1 Flows on the line • Stable f.p. becomes unstable and two new stable f.p x∗ = ± r appear. 2
• Draw function ẋ = f (x). Get Fixed points by: f (x∗ ) = 0. • Cubic term is stabilizing.
• a > ω: Half-stable f.p. splits into a stable and unstable f.p.
• Complicated functions can be separately drawn: ẋ = x − cos(x).
• x − t-Plot: Pay attention to curvature!
2.2 Linear Stability Analysis (Linearization about stable fixed points)
• To gain quantitative information at a stable fixed point (e.g. rate of decay).
→
→ η(t) = x(t) − x? ⇒ η̇ = ẋ = f (x) = f (x? + η)
→ Taylor expansion: f (x∗ + η) = f (x∗ ) + ηf 0 (x∗ ) + O(η 2 ). → η̇ ≈ ηf 0 (x∗ )
⇒ If f 0 (x∗ ) < 0 the fixed point is stable. (If f 0 (x∗ ) = 0 nothing can be said)
• Measure of stability: characteristic time scale: 1 .
|f 0 (x∗ )| 3.3.2 Subcritical Pitchfork Bifurcation (Inverse bifu., backward bifu., phase transition of 1st order,
hard bifu., dangerous bifu.)
2.3 Existence and Uniqueness
3
• Theorem: If solution is smooth enough (f (x) and f 0 (x) continuous on open • Normal form: ẋ = rx + x
√
interval R of the x-axis and x0 ∈ R), then there exists a unique solution x(t) • Unstable f.p. becomes stable and two new unstable f.p x∗ = ± −r appear
for r ↓.
in time interval (−τ, τ ) around t0 . (C-Ex.: ẋ = x1/3 .)
• Cubic term is destabilizing. Helps r > 0 in driving the system to ±∞ → blow
• Blow Up: Solutions reach infinity in finite time. (e.g. x(t) = tan(t)) up.
2.4 Impossibility of Oscillations for ẋ = f (x)
• All traj. approach f.p. or ±∞ (phase point can’t reverse directions, no oscil-
dt = 02π dθdt dθ = R 2π dθ = ... = q 2π
RR
• Oscillation Period: T = 0 .
lations). ω−a sin(θ)
ω 2 −a2
 √ 
π 2 √ 1
2.5 Potentials • Square-root-scaling law: T ≈ √ .
ω ω−a
• Particle is sliding down the walls of a potential well
• Bottleneck: very slow passage
• ẋ = f (x) = − dV
dx • Stabilizing terms needed: ẋ = rx + x3 − x5 . (Jumps and Hysterisis) • Ghost: a former saddle-node (a slightly < ω) that leads to bottleneck → traj.
• V (t) decreases along trajectories → particle always moves to lower potential • rs < r < 0 : two different stable states coexist (origin and large-amplitude needs most of its time through this point.
dV = dV · dx = −ẋẋ = −|ẋ|2 < 0 fixed points)
dt dx dt
• local minima of V (x) = stable f.p., local maxima of V (x) = unstable f.p. • initial condition x0 determines which fixed point is approached as t → ∞
• origin is only locally stable
2.6 Examples
• Jumps and hysteresis: start at x∗ = 0 → increase r → state remains at 0 until
• RC-Circuit, Population Growth r = 0 → slightest pertub. will cause state to jump to high-amplitude branches
3 Bifurcations
and stays there as r is increased
• Bifurcation: Qualitative change in dynamics as parameters change. (f.p. • if r is decreased the state stays on the branches, only if r < rs the state jumps
created/destroyed, stability changes) back to the origin 4.3 Example: Overdamped Pendulum
• Parameter values at which bifurcation occurs: bifurcation points. • lack of reversibility: Hysteresis 2
• mL θ̈ + bθ̇ + mgL sin(θ) = Γ (constant torque: Γ)
• Word Meaning: Splitting into two branches. • The bifurcation at rs is a saddle-node-bifurcation. b θ̇ =
• Consider large b to approximate: bθ̇ + mgL sin(θ) = Γ ⇔ mgL Γ −
mgL
• Bifurcation Diagram: Sometimes its easier to see x∗ as the independent va- mgL Γ dθ 0
riable for drawing. 3.4 Imperfect Bifurcation sin(θ) → Non-Dimensionalize: τ = b
t, γ = mgL ⇒ dτ = θ = γ − sin(θ).
• Slight imperfections lead to loss of symmetry.
– γ > 1: Torque can’t be balanced by gravity and pendulum overturns conti-
3.1 Saddle-Node-Bifurcation (Blue-sky-bifu., fold bifu., turning-point bifu.) • Example: ẋ = h+rx−x3 (h=0: normal superc. pitchfork), symmetry is broken nually (nonuniform).
• Fixed points are created/destroyed. for h 6= 0 (Imperfection parameter) – γ = 1: fixed point appears at π 2
.
– γ < 1: fixed points splits into two (lower one is stable).
• Ex: ẋ = r + x2 • plot y = rx − x3 and y = −h → intersection points=fixed points – γ = 0: unstable equilibrium at top and stable equilibrium at the bottom.
• r ≤ 0 : one fixed point r > 0 : 2,3 intersections possible, depending on value 4.4 Synchronization (Fireflies)
of h Model
• critical case if horizontal line is tangent → S-N-B. • θ(t): phase of firefly’s flashing rhythm (θ = 0 instant of the flash)
• Bifurcation diagram (h 6= 0): as r is increased from negative values → no sharp • θ̇ = ω: eigenfrequency of single fly, θ̇ = Ω: eigenfrequency of swarm.
transition at r = 0, instead fixed point glides smoothly along upper branch → • θ̇ = ω + A · sin(Θ − θ) with A : resetting strength
lower branch of f.p. only accessible via a large disturbance. → catastrophe
• For example, if Θ is ahead of θ the firefly speeds up (θ̇ > ω).
Analysis
• Phase difference: φ = Θ − θ, Velocity difference: φ̇ = Θ̇ − θ̇ = Ω − ω − A sin(φ)
• Normal forms: Prototypical examples for S-N-B. ẋ = a(r − rc ) + b(x − x∗ )2 • Nondimensionalize: τ = At, µ = Ω−ω (µ : ⇒ dφ = φ0 = µ − sin(φ).
(reason: we need 2 nearby roots of f (x), f (x) needs bowl-shape) A dτ
• Different cases for µ ≥ 0:
3.2 Transcritical Bifurcation – µ = 0: Synchronization without phase difference (s.f.p.: φ∗ = 0)
– 0 < µ < 1: Synchronization w/ phase difference (phase lock, s.f.p.: φ∗ > 0)
• Fixed point is not destroyed, but changes stability → Normal form: ẋ = rx−x2 – µ = 1 S-N-B and µ > 1: No synchronization (phase drift, φ increases cont.)
• Example: Laser Threshold: ṅ = (GN0 − k)n − (αG)n2
4 Flows on the Circle (1D)
• Vector field on the circle: θ̇ = f (θ) with θ : point on the circle., f (θ): real-
valued-2π-periodic function.
• Definition: A rule that assigns a unique velocity vector to every point on the
circle.
• New Property: Flowing particle can return to starting point (periodic solutions
/ oscillations are possible)

4.1 Uniform Oscillator Conclusion

3.3 Pitchfork Bifurcation • θ̇ = ω with ω = const. and θ(t) = ωt + θ0 • Entrainment for ω − A ≤ Ω ≤ ω + A. (range of entrainment) → The single
• Fixed points appear/disappear in symmetrical pairs • Beat phenomenon: Two non-interacting oscillators with different ω will pe- oscillator can synchronize if its eigenfreq. is not too far off (by A) from the
• Occurs often in symmetrical problems, e.g. buckling problem riodically go in and out of phase. eigenfreq. of the swarm.

1
5 Linear Systems (2D) – Robust cases:
h i
• A two-dimensional linear system: ẋ = Ax with x = x
y ,A = a b 1. Sources / Repellers: Re(λ1 ) > 0, Re(λ2 ) > 0.
 
c d
• Linear system: If x1 and x2 are solutions, then x3 = c1 x1 +c2 x2 is a solution. 2. Sinks / Attractors: Re(λ1 ) < 0, Re(λ2 ) < 0.

• x∗ = 0 is always a fixed point for any choice of A. 3. Saddles: Re(λ1 ) > 0, Re(λ2 ) < 0.
• Phase portrait: Trajectories moving in the phase plane. – Limit cases (at least one eigenvalue has Re(λ) = 0):
h i 1. Centers (both eigenvalues purely imaginary)
• Example: ẋ = Ax, A = a 0
0 at
−1 ⇒ x(t) = x0 e , ẏ = y0 e
−t
2. Higher-order and non-isolated f.p.
• a < −1: x(t) decays faster than y(t) → the trajectories approach the origin
tangent to the slower (here: y-) direction. x∗ = 0 is a stable node. • Hartman-Grobman-Theorem: The local phase portrait near a hyperbolic
fixed point (Re(λi ) 6= 0) is topologically equivalent to the phase portrait of
y(t) y
• a = −1: x(t) = x0 = const.. x∗ is called a symmetrical node or star. the linearization.
0
• Basin of Attraction: Set of x0 , s.t. x(t) → x∗ as t → ∞. (Region in ph.
• −1 < a < 0 : trajectories approach x∗ from the slower x-direction.
plane)
• a = 0: → x(t) = x0 , one gets a line of fixed points along x-axis.
 p 
• λ1,2 = 1
2
τ ± τ 2 − 4∆ with τ = tr(A) = λ1 + λ2 and ∆ = det(A) =
• a > 0 : x∗ becomes unstable, due to exponential growth in x- λ1 · λ2 Drawing Rules
direction.Trajectories go to infinity. Exception: If they start on the y-axis. • Saddle point: ∆ < 0 ⇒ λ1 > 0, λ2 < 0. • Indicate f.p. and draw some traj. going in/out tangent to slow eigenvector.
x∗ is called a saddle point/hyperbolic fixed point.
• Fixed points: ∆ > 0, τ < 0 (attracting), ∆ > 0, τ > 0 (repelling) • Connect all lines, so that picture makes sense.
∗
– Stable manifold of the saddle point. (Set of x0 : x(t) → x as → +∞)(y)
– Unstable manifold of x∗ (Set of x0 : x(t) → x∗ as → −∞)(x-axis) – Nodes: τ 2 − 4∆ > 0, Real roots and λ1 , λ2 same sign • Nullclines (2D): Lines with ẋ = 0 or ẏ = 0. Use other eq. for vector direction.
– Spirals: τ 2 − 4∆ < 0, Complex conjugate λ’s.
• Centers: τ 2
− 4∆ < 0 and ∆ > 0, τ = 0 ⇒ λ = ±iω (WARNING n.l. sys.!) • Indicate Basin of Attraction (often stable/unstable manif. of saddle serves as
basin boundary)
• Stars and degenerate nodes: τ 2 − 4∆ > 0
• Non-isolated f.p.: ∆ = 0 (line or plane of f.p.)
6.2 Conservative Systems
6 Dynamics in the Phase Plane (Nonlinear 2D) • There exists a conserved quantity, i.e. a quantity that is const. along trajec-
• General form of a vector field in phase plane: ẋ = f (x). By flowing along tories. (ẋ = f (x), ∃ a function E(x) s.t. dE
dt
= 0 on all trajectories.)
the vector field, a phase point traces out a solution x(t), corresponding to a
trajectory. • mẍ = F (x) (F 6= F (t, ẋ), i.e. no damping or time-dependent driving force)

• Define V (x) as potential energy and set F (x) = − dV →: mẍ + dV = 0 ⇔

Stability language 1. Fixed Points: f (x∗ ) = 0. (like d ( 1 mẋ2 + V (X)) = 0
dx dx
A,B,C) mẋẍ + dV ẋ = 0 ⇔
• F.P. is locally attractive, if x(t) → x∗ , as t → ∞ for x0 close to x∗ . dx dt 2
1 mẋ2 + V (x) = const.
• Total Energy E = 2
• F.P. is globally attractive, if x(t) → x∗ , as t → ∞ for all x0 . 2. Closed Orbits: Periodic solutions
x(t + T ) = x(t)∀t (like D)
∗ • To prevent trivial sol. E(x) ≡ 0, E has to be nonconstant on every open set.
• F.P. is Liapunov stable, if all trajectories that start sufficiently close to x
remain close to it for all time. (if for each  > 0∃ a δ > 0 : ||x(t) − x∗ || < 3. Draw flow close to fixed
points/close to orbits. • Attractive f.p. can’t appear in conservative sys.,since then E(x) would be
 whenever t ≥ 0 and ||x(0) − x∗ || < δ. const. on traj. flowing towards f.p.. Instead saddles and centers.
• F.P. is neutrally stable, if it’s Lyapunov stable but not attractive. (Ex: (d)) 4. Complete the picture.
6.1 Existence and Uniqueness Theorem 1 x2 +
• Example: Particle m = 1 moves in a double-well-potential: V (x) = − 2
• F.P. is asymptotically stable, if it’s attractive and Lyapunov stable. • Consider the initial value problem ẋ = f (x), x(0) = x0 . Suppose that f and its 1 x4 . Classify f.p. / sketch phase portrait!
derivatives are continuous and smooth. Then for x0 the initial value problem 4
• F.P. is unstable, if it’s neither attracting nor Lyapunov stable. Components
grow without bound. has a unique solution x(t) on some time interval (−τ, τ ) about t = 0.
⇒ Trajectories never intersect. Otherwise at the intersection point two solu- – Force F = −dVdx
= x − x3 → mẍ = F → ẋ = y, ẏ = x − x3 .
5.1 Classification of Linear Systems
tions would start (not unique). – F.P.: saddle at (0, 0), centers at (±1, 0) determined by Jacobian.
• Poincaré-Bendixson theorem: If a trajectory is confined to a closed, boun- – Constant Energy: E = 2 1 y 2 − 1 x2 + 1 x4 = const.
• Use Eigenanalysis to create picture of phase portrait. 2 4
ded region and there are no fixed points in the region, then the trajectory
• Find eigensolutions of the form: x(t) = veλt →: x(t) = c1 v1 eλ1 t + c2 v2 eλ2 t . must eventually approach a closed orbit.
h i h i h i h i
• Example: ẋẏ = 4
1 1 x
−2 y → (...) → x(t) = c1 1 2t
1 e + c2 −4 e
1 −3t 6.3 Fixed Points and Linearization
• To approximate the phase portrait near a fixed point by a corresponding linear
• More cases: system .
– λ2 < λ1 < 0: Fixed point is stable, trajectories approach fixed point tan- • Small
h idisturbances
h u = x − x∗ i
, v = y − y ∗ → Taylor expansion
∂f ∂f ∂g ∂g ∗ ∗
gent to slow eigendirection, for t → ∞ trajectories become parallel to fast → u̇v̇ = ∂x
;
∂y ∂x ∂y [u
v ] (Jacobian with x , y plugged-in)
eigendirection. If you reverse the arrows you get an unstable node. (x∗ ,y ∗ )
– Complex eigenvalues: Either yield a center (Re(λ) = 0,neutrally stable) or • Neglecting h.o.t. to give qualitatively correct picture works for saddle, nodes,
a spiral (Re(λ) 6= 0, stable/unstable) x(t) = c1 v1 eλ1 t + c2 v2 eλ2 t with spirals, but not for borderline cases.
eα±ωt → x(t) involves a combination of eαt , cos ωt, sin ωt. Determine di- • Example: ẋ = −x + x3 , ẏ = −2y • Most solutions are periodic, except for 2 trajectories.
rection of rotation by computing some vectors in the vector field. ⇒ Homoclinic Orbits: Traj. that appear to start and end at the origin,
1. Fixed points: (0,0), (1,0), (-1,0) by ẋ = 0, ẏ = 0 which they approach at t → ∞.
2
h i
2. Jacobian: A = −1 + 3x 0
• Heteroclinic trajectories/saddle connections. Traj. that connect two
0 −2
3. Evaluate A for the fixed points: (0,0) is stable node, (1,0), (-1,0) are saddle points.
saddle points. (see. 5.1)
• Theorem: Nonlinear centers for cons. sys.: Consider conservative system
4. Phase portrait (sketch with eigenvectors, values):
ẋ = f (x), x ∈ R2 and f smooth. If x∗ is a local minima of E, then all traj. in
the neighborhood are close orbits ⇒ x∗ is a center.

– λ1 = λ2 = λ: 6.3 Reversible Systems (Dynamics looks the same whether time runs forward or bachward)
• Definition: Any 2nd-order system that is invariant under t → −t and y → −y.
∗ Case 1: Distinct eigenvectors, that span the whole plane, so every vec-
(e.g. ẋ = f (x, y), ẏ = g(x, y) with f odd in y and g even in y). More general:
tor can be composed by them → all trajectories are straight lines to the
origin, f.p. is a star node. ẋ = f (x) is reversible if it’s invariant under t → −t, x → R(x), where the
∗ Case 2: λ = 0 → The whole plane is filled with fixed points. mapping satisfies R(x)2 = x.
∗ Case 3: Only one eigenvector. The fixed point is a degenerate node.
∗ Intuition: Ordinary node → Two eigendirections get scissored together. • Structural Stabilty (Topology of phase-portrait is robust against small per- • Reversible sys. have many of the same properties as conservative sys., e.g.
→ Borderline between spiral and node. turb.) centers are robust in reversible sys. as well.

2
6.4 Pendulum Lyapunov Functions 8.3 Hopf Bifurcations
2
• Equation of motion: d 2θ + L g p
sin θ = 0 Set ω = g/L and τ = ωt → Nondi- • Theorem: If for a given system ẋ = f (x) with fixed point at x∗ a Lyapunov • Fixed point changes stability, b/c both complex conjugate eigenvalues cross
dt function exists, then x∗ is gl. asympt. stable and closed orbits aren’t possible. the Im-Axis into the right half plane. (for dimensions n ≥ 2)
mensionalize: θ̈ + sin θ = 0 → θ̇ = ν, ν̇ = − sin θ. • A Lyapunov function has the following properties: • You cant use linearization to distinguish between subcr. and supercr. bifu.
• F.P.: (θ ∗ , ν ∗ = (kπ, 0) (centers), (θ ∗ , ν ∗ = (0, 0) (saddle) 1. V (x) > 0 for all x 6= x∗ , and V (x∗ ) = 0 (V is positive definite.).
8.3.1 Supercritical Hopf Bifurcation (soft, continuous, safe)
• A stable spiral changes into an unstable spiral surrounded by a small nearly
• Energy: E = 1 θ̇ 2 − cos θ. from:(θ̇(θ̈ + sin θ) = 0) ∗ ∗
2 2. V̇ < 0 for all x 6= x (All trajectories flow „downhill“ toward x ). elliptical limit cycle when control-parameter µ is changed. (or vice versa)
• Intuition: All trajectories move monotonically down the graph of V (x) to- • Example: ṙ = µr − r 3 , θ̇ = ω + br 2
wards x∗ . The solution can’t get stuck anywhere else, b/c then V wouldn’t
change.
7.2 Poincaré-Bendixson Theorem (To proof Existence of Closed Orbits)
• In 2D systems, you can’t have something more complicated then a fixed-point
or a closed orbit (chaos is not possible in 2D).

1. R is a closed, bounded subset of the plane.

• Rules
√ of thumb: 1. Size of limit cycle grows continuously from zero with
2. ẋ = f (x) is a continuously differentiable vector µ − µc , 2. The frequency of the limit cycle is approximately ω = Im(λ)|µc .
• Physical interpretation: field on an open set containing R;
8.3.2 Subcritical Hopf Bifurcation (hard, discontinuous, dangerous)
3. R does not contain any fixed points; (a repelling 3 5 2
1. Center: stable equilibrium, pendulum hangs down • Example:ṙ = µr + r − r , θ̇ = ω + br
2. Small oscillations about equilibrium (E = −1) (librations) (E < 1) f.p. might be excluded from R) • Unstable cycle surrounds a stable spiral. Cycle shrinks and at µ = 0 f.p. beco-
mes unstable. µ > 0 Large limit cycle is suddenly the only attractor → Small
3. Critial case (E = 1) Heteroclinic trajectories: Motion in which the pen- 4. There exists a closed orbit C that is „confined“ in oscillations now grow to large oscillations.
dulum slows to a halt precisely as it approaches the inverted position R (it starts in R and stays in R) • System exhibits hysteresis: Large oscillations cant turned off by decreasing µ.
(saddle node) ⇒ Then either C is a closed orbit or it spirals to- They remain until µ < µc = − 4 1 , when stable and unstable cycle annihilate.
ward a closed orbit as t → ∞.
4. E > 1: Pendulum whirls repeatedly over the top → periodic solution 7.3 Liénard Systems
• The natural phase space for a pendulum is the cylindrical phase space (peri- • 2nd-order Liénard equation: ẍ + f (x)ẋ + g(x) = 0.
odic whirling motion looks periodic) • Equivalent system: ẋ = y, ẏ = −g(x) − f (x)y.
dE 2 • Theorem: Suppose f (x) and f (x) fulfill the following conditions: 1.f and g are
• Add damping: θ̈ + bθ̇ + sin θ = 0, b < 0, dτ
= −bθ̇ ≤0
continuously differentiable; 2.g(−x) = −g(x) g is an odd function; 3. R g(x) > 0
b>0
b>0 b<0 for x > 0; 4.f (x) is an even function; 5. The odd function F (x) = 0x f (u)du
has exactly one positive zero at x = a, is negative for 0 < x < a, is positive
and nondecreasing for x > a and F (x) → ∞ as x → ∞.
⇒ Then the system has a unique, stable limit cycle surrounding the origin. 8.3.3 Degenerate Hopf Bifurcation
• Ex.: ẍ + µẋ + sin x = 0. At µ = 0 you get an origin that changes from stable to
• Intuition: Assumptions on g(x) mean that the restoring force acts like an
unstable spiral and a cont. band of closed orbits around it (no limit cycles).
ordinary spring. Assumptions on f (x) imply that the damping is negative for
small |x| and positive at large |x| → Results in self-sustained oscillation. 8.4 Global Bifurcations (Creation/Destruction of Limit Cycles)
• 4 Ways: Hopf Bifu. (local), Saddle-Node, ∞-Period, Homoclinic Bifu. (global)
7 Limit Cycles (→ Closed Orbit)
8 Bifurcation Revisited (2D) • Involves large regions of the phase plane rather than just nbh. of single f.p..
• Definition: A limit cycle is an isolated closed trajectory. Isolated means that • Bifurcation: The phase portrait changes its topological structure as a para- 8.4.1 Saddle-Node-Bifurcation of Cycles (Coalesce / Annihilation of Limit Cycles)
meter is varied, e.g. change in number and stability of f.p.’s, closed orbits or 3 5 2
neighboring trajectories are not closed. They spiral either toward or away from saddle connections. • Example: ṙ = µr + r − r , θ̇ = ω + br
the l.c.. • Classification: • At µc = − 1 a new half-stable cycle is born and splits into a pair of limit
4
• Types: stable, unstable, half-stable(traj. from one side towards, from other cycles (one stable, one unstable) as µ ↑ (see. Bifurcation Diagram in 8.3.2).
1. Bifurcation of Fixed Points
side away).
a) λ = 0 Bifurcations (Saddle-Node,Transcritical,Pitchfork)
• System has a standard oscillation and if perturbed returns to standard cycle
(e.g. heartbeat, body temp., feedback controlled system, airplane wings). b) λ = ±iω (Hopf Bifurcation) → creation of closed orbits.
• L.C. can’t occur in linear systems (if x(t) is periodic, so is cx(t)). 2. Bifurcation of Closed Orbits
a) Coalescense of Cycles (Saddle-Node-Bifurcation of Cycles)
Example 1: A Simple Limit Cycle b) Sniper, Snic (Saddle-Node-Infinite-Period Bifurcation)
• System: ṙ = r(1 − r 2 ), θ̇ = 1 → Uncoupled. Look at r first. r ∗ = 0 is an
∗ c) Homoclinic Bifurcation
unstable, r = 1 is a stable fixed point. Traj. will approach unit circle.
8.1 Saddle-Node Bifurcation
Example 2: Van der Pol Oscillator √
• Prototypical Example: ẋ = µ − x2 , ẏ = −y with 2 f.p.:(x∗ , y ∗ ) = (± µ, 0)
2 8.4.2 Infinite-Period Bifurcation
• System: ẍ + µ(x − 1)ẋ + x = 0 (van der Pol equation) with parameter µ ≥ 0. and ev:λ1 = −2x∗ , λ2 = −1 2
• Harmonic oscillator with nonlinear damping term. Damps oscillations for • Example: ṙ = r(1 − r ), θ̇ = µ − sin θ, µ ≥ 0 (Combination of Ch. 3 and 4)
• As µ decreases, the saddle and node approach and collide at µ = 0 ( λ1 = 0)
|x| > 1 and pumps up for |x| < 1. System settles into a self-sustained oscilla- and disappear. It remains a ghost that slows down incoming traj. • µ > 1: Angular motion is everywhere clockwise. µ → 1+ : oscillation period
tion, where the dissipated energy balances the energy pumped in. lengthens and becomes infinite at µc = 1, when a fixed points appears on the
circle at π/2. µ < 1: Fixed points split into a saddle and a node. You get two
invariant rays (sin θ = µ) ⇒ Limit cycle is destroyed via infinite period.
8.4.3 Homoclinic Bifurcation
• Example System: ẋ = y, ẏ = µy + x − x2 + xy (numerical solved)
• Part of the limit cycle moves closer to a saddle point → at the bifurcation
µ = µc the cycle touches the saddle point and becomes a homoclinic orbit.
• Key: Behavior of the unstable manifold of the saddle. After it loops around it
either hits the origin (c) or veers off to one side or the other (a,d)
• Nullclines: (to graph vector field)
7.1 Ruling Out Closed Orbits – Graph implicit lines: ẋ = f (x, y) = 0, ẏ = g(x, y) = 0
Gradient Systems – Intersection points are fixed points that move by changing µ
• Gradient systems with potential function V (x): ẋ = −∇V . – The vector directions on the nullclines are either vertical or horizontal.
• Theorem: Close orbits are impossible in gradient systems! 8.2 Transcritical and Pitchfork Bifurcation
2
• Proof: Assume there is a limit cycle. Then going ones around would mean • Transcritical: ẋ = µx − x , ẏ = −y (f.p. changes stability)
∆V = 0. But on the other hand: ∆V = 0T dV dt = 0T (∇V · ẋ)dt = • Supercritical: ẋ = µx − x3 ,
R R
dt ẏ = −y (2 new stable. f.p.)
RT 2
− 0 ||ẋ|| dt < 0 (unless ẋ = 0, in which the trajectory is a fixed point) • Subcritical: ẋ = µx + x3 , ẏ = −y (2 new unstable f.p.)

3
8.5 Coupled Oscillators and Quasiperiodicity 9.3 Chaos on a Strange Attractor 11.2.2 Similarity Dimension (for self-similar fractals)
• New 2D-Phase-Space: Torus. Natural phase space for systems: θ̇1 = • Plots for system and r > rH
f1 (θ1 , θ2 ), θ̇2 = f2 (θ1 , θ2 ), where f is a periodic function.
• Ex.: Coupled oscillators: θ̇1 = ω1 +K1 sin(θ2 −θ1 ), θ̇2 = ω2 +K2 sin(θ1 −θ2 )
• Visualize on torus or on a square with periodic boundary conditions:

• Similarity dimension d is the exponent m = r d or equivalently: d = ln m.

ln r
• y(t)-plot: Solution settles into an irregular,nonrepeating (aperiodic) motion.
11.3 Box Dimension (for not self-similar fractals)
• 2D: Trajectory spirals on both sides. Number of circuits are unpredictably. • Cover the set with boxes of size .
• 3D: Infintite sets of surfaces, extremely close to one or the other two merging • For a curve of Length L: number of boxes N ()˜L
Uncoupled System (K1 = K2 = 0; θ̇1 = ω1 , θ̇2 = ω2 ) surfaces.→ Fractal. 
• Trajectories on square are straight lines with slope ω2 .
ω • For a planar region of Area A: number of boxes N ()˜ A
2
1 Exponential Divergence of Nearby Trajectories 
ω • Motion in the attractor exhibits sensitive dependence on initial condi- ln N ()
• Case 1: Slope ω2 = p e.g. = 3 is rational → All traj. are closed orbits on • Box dimension: d = lim→0 (Lorenz-Attractor: d ≈ 2.05).
1 q 2 tions. Two traj. starting very close together will exponentially fast diverge ln( 1 )

torus (e.g. Trefoil knot) from e.o.. → Long-term predictions become impossible.
11.4 Pointwise and Correlation Dimension (To determine dimension of strange attractor)
• Case 2: Slope is irrational → flow is quasiperiodic, every traj. winds around • Suppose x(t) is a point on the attractor at time and consider a nearby point
endlessly on torus, never intersecting, never closing. • Fix a point x on attractor A. Let Nx () denote number of points on A in a
x(t) + δ(t) with seperation ||δ(t)|| = 10−15 . ball of radius  about x. Vary  and determine pointwise dimension d at
Coupled System • Numerical studies find||δ(t)|| ∼ ||δ0 ||eλt with Liapunov exponent λ (λ ≈ x: Nx () ∝ d . Average Nx () over many x to get correlation dimension
• Consider Phase-Difference: φ = θ1 − θ2 . → φ̇ = ω1 − ω2 − (K1 + K2 ) sin φ (2 0.9 for Lorenz) C() ∝ d .
nonuniform oscillator) • Liapunov exponent is a quantity that characterizes the rate of seperation of in-
finitesimal close trajectories. The seperation can be different in diff. direction, 12 Strange Attractors
thus ∃ n λ’s. Liapunov exponent is chosen as the biggest one. 12.1 Stretching and Folding property of Dynamic Systems
• 2 f.p. if |ω1 − ω2 | < K1 + K2 and no fixed points else • How can attractors separate from their neighbors exponentially fast, but stay
→ Saddle-Node-Bifurcation at transition (→ 8.4.1). bounded.→ Basic mechanism: Stretching and Folding.
∗ ω1 −ω2 • Consider a blob (Tropfen) of initial conditions in phase space.
• f.p. defined by: sin φ = K1 +K2
the trajectories will
approach a phase-locked solution, where the oscilla-
tors are separated by stable f.p. φ∗ .
• The phase-locked solution is periodic. Both oscilla- • A strange attractor typically arises when flow contracts the blob in some di-
tors run at a constant compromise frequency given • For systems with a positive Liapunov exponent there is a time horizon,
beyond which prediction breaks down. Assume initial conditions are measured rection and stretches in others. The stretching can’t go on forever → distorted
by: ω ∗ = θ̇1 = θ̇2 = ω2 + K2 sin φ∗ very accurately (with some error ||δ0 ||. After a time t the discrepancy grows
blob must be folded back on itself to remain in bounded region.
• Analog: Flattening and folding on dough to make croissant. Small drop of
8.6 Poincaré Maps to ||δ(t)|| ∼ ||δ0 ||eλt . Let a be our tolerance, then our prediction
 fails,
 when food coloring (initial condition) will be spread throughout the dough after
||δ(t)|| ≥ a and this occurs after a time: thorizon ∼ O 1 ln a . some iterations.
• N-dimensional system: ẋ = f (x). Let S be an n-1 dimen- λ ||δ0 ||
sional surface of section (transverse to the flow, i.e. all 12.2 Attractor Reconstruction
• The logarithmic dependence on ||δ0 || hurts, since you need an extreme reduc-
trajectories starting on S flow through it) 1. • For systems governed by an attractor, the dynamics in the full phase space
• Poincaré map P : mapping from S to itself, by followi- tion of ||δ0 || to achieve a longer prediction of λ can be reconstructed from measurements of just a single time series B(t).
ng trajectories from one intersection with S to the next: • Method is based on delays:Define two-dimensional vector x(t) = (B(t), B(t +
xk+1 = P (xk ). Definitions
τ ))
• If x∗ is a fixed point: P (x∗ ) = x∗ → x∗ returns to x∗ → Chaos is aperiodic long-term behavior in a deterministic system that exhibits
• Plotting the two dimensional vector shows the shape of an attractor.
closed orbit. sensitive dependence on initial conditions.
1. „Aperiodic long-term behavior“ means that there are trajectories which • To obtain the Poincaré section, one can compute intersections of the orbits
• Looking at behaviour of P near fixed point we can determine stability of clo- x(t) with a fixed plane approximately normal to the orbits.
se orbit.⇒ Poincaré Map converts problems about closed orbits (which are dont settle down to f.p., periodic orbits or quasiperiodic orbits as t → ∞.
2. „Deterministic“ means that the system has no noise or randomness. Same
difficult) to problems about fixed points of a mapping. initial conditions always lead to same behavior.
Linear Stability of Periodic Orbits
3. „Sensitive dependence on initial conditions“ means that nearby trajectories
separate exponentially fast (i.e. positive Liapunov exponent)
• Is closed orbit stable? → Is corresponding fixed point x∗ on Poincaré map An attractor is a closed set A with the following properties:
stable? Choose infinitesimal perturbation v0 .
1. Invariant set: Any trajectory x(t), that starts in A stays in A for all time.
• x∗ + v1 = P (x∗ + v0 ) = P (x∗ ) + [DP (x∗ )]v0 + O(||v0 ||2 ) with DP (x∗ ) 2. A attracts an open set of initial conditions: A attracts all trajectories
linearized Poincare map at x∗ . Since x∗ = P (x∗ ): v1 = [DP (x∗ )]v0 . that start sufficiently close to it.
• Close orbit is linearly stable iff eigenvalues |λj | < 1 of DP (x∗ ) (also called 3. A is minimal: There is no proper subset of A that satisfies conditions 1,2.
Important Equations
4. Sensitive to initial conditions (for strange attractor)
the characteristic or Floquet multiplier ). 
N

9.1 Lorenz Map • Logistic Equation: Ṅ = rN 1− K (pop. growth)
9 Lorenz Equation • Idea: zn should predict zn+1 (z is the local maximum of z(t)).
• Lotka-Volterra-Model: Ṅ1 = N1 (1 − γ1 N2 ); Ṅ2 = −N2 (2 − γ2 N1 ) (Räuber-
• ẋ = σ(y − x), ẏ = rx − y − xz, ż = xy − bz, σ, r, b > 0. (often:σ = 10, b = 8
3
) • Map: Plot zn+1 vs. zn Beute, : Reprodukt.rate, 2 : Sterber., γ1 : Fressr. R/B, γ2 : Reprod.r. R/B)
• Determinisitc system, that exhibits strange dynamics. Over a wide range of • Extract a function zn+1 = f (zn ). It shows: |f (z ∗ )| > 1∀z.
Useful Stuff
parameters, the solutions oscillate irregularly, never exactly repeating, but
11 Fractals • ex =
P∞ xn x2 3 x5 ± . . . ,
+ . . . , sin(x) = x − x6 + 120
always in a bounded region of the phase space (strange attractor). n=0 n! = 1 + x + 2
Fractals are complex geometric shapes with fine structure at arbitrarily small 2 4
• r < 1: Gl. stable f.p. at origin. At r = 1 Supercritical Pitchfork Bifu. cos(x) = 1 − x2 + x ∓ ...
scales. Usually they have some degree of self-similarity. (Magnify a tiny part of 24
• 1 < r < rH Origin turns unstable, two new linearly stable f.p. x∗ = y ∗ = the fractal and you see features reminiscent of the whole). Their dimension is P∞ (−1)n+1 (x−1)2 (x−1)3
σ(σ+b+3)
± b(r − 1), z ∗ = r − 1 → C + , C − for 1 < r < rH = σ−b−1 . Stable f.p.
p
not an integer. • ln(x) = n=1 n
(x − 1)n = (x − 1) − 2
+ 3
− ...
11.1 Cantor Set df dg
are surrounded by a saddle cycle. • Find tangent-point of two functions: f (x) = g(x) and dx
= dx .
• Start with closed interval S0 = [0, 1] and remove open middle third → S1 .
• r = rH Subcritical Hopf-Bifurcation,cycle shrinks and is absorbed by fixed • ln(x, y) = (≤ 0, N aN ), (0.1, −2.3); (0.5, −0.7); (1, 0); (2, 0.7); (3, 1.1); (10, 2.3)
point, that changes into a saddle point. • The limiting case is the cantor set: C = S∞
• exp(x, y) = (−3, 0.05); (−1, 0.37); (0, 1); (1, 2.7); (2, 7.4); (3, 20.1)
• r > rH No attractors in the neighborhood remain. • More than points (0D) less than a line (1D) • tan(x, y) = (−π/2, −∞); (−π/4, −1); (0, 0); (π/4, 1); (π/2, +∞)
• d = ln m = ln 2 ≈ 0.63. Arc Length: L = ( 2 )n → 0. √ √ √ √ √ √ √
ln r ln 3 n 3 • 2 = 1.41; 3 = 1.73; 5 = 2.24; 7 = 2.65; 8 = 2 2; 10 = 3.16
ln(N ()) ln(2n ) ln(2)
• Box dimension: d = dim→0 ln(1/) = ln(3n ) = ln(3) Definitions
• Phase Space: Abstract space in which the states form a basis.
11.2 Dimension of Self-Similar Fractals • Trajectory: Give the evolution of the states in phase space.
11.2.1 Van Koch curve • Phase Portrait: Shows all qualitatively different trajectories.
• More than a line (1D) less than a surface (2D) • Open set: Sets, which contain an open ball around each of their points.
m = ln 4 ≈ 1.26. Arc Length L = ( 4 )n L → ∞ • Trapping Region: A cl. connected set, s.t. the v. f. points inward everywhere.
• d = ln
ln r ln 3 n 3 0