Dynamical Systems
Bifurcations
1963. Cambridge, MIT. Edward Lorenz prints an intermediate result rounded to 0.506 instead of 0.506127. He reruns the simulation. Two months of computer time later: completely different weather. The butterfly effect gets its name. But well before that: a Hodgkin-Huxley neuron fires when the parameter crosses a threshold, the heart enters fibrillation, a laser starts lasing at pump threshold. All of these are bifurcation points. The mathematics is identical.
- **Neuroscience:** the Hodgkin-Huxley neuron model (1952) contains a Hopf bifurcation - the excitation threshold is literally the parameter at which Re(eigenvalue) changes sign
- **ML optimization:** an excessively high learning rate crosses the bifurcation point in the loss landscape; beyond it, no equilibria exist and loss diverges
- **Cardiology:** the transition from normal sinus rhythm to fibrillation is a Hopf bifurcation in models of excitable cells
- **Engineering:** buckling of a beam under load (Euler buckling) is a pitchfork bifurcation, used in structural design calculations
Предварительные знания
Saddle-Node Bifurcation
**A neural network training loss explodes.** It was stable for 10 epochs, then diverged to NaN in a single step. Not noise - a structural transition. An excessively large learning rate created exactly this: two quasi-equilibria (a stable minimum and an unstable saddle in the loss landscape) merged and annihilated. Saddle-node bifurcation is not a metaphor here; it is literally what happens in the loss landscape when a critical learning rate is exceeded.
**Saddle-node bifurcation**: as parameter μ changes, two equilibria (a stable "node" and an unstable "saddle") approach each other, merge, and disappear. Normal form: **dx/dt = μ + x^2**. For μ < 0 there are two equilibria x* = ±√(-μ); at μ = 0 they merge into x* = 0; for μ > 0 no equilibria exist.
| μ | Number of equilibria | Character |
|---|---|---|
| μ < 0 | 2 | x* = -√(-μ) stable, x* = +√(-μ) unstable |
| μ = 0 | 1 | Semi-stable (ghost point) |
| μ > 0 | 0 | No equilibria - system escapes |
Poincare and Bifurcation Theory
Henri Poincare introduced the term "bifurcation" in 1885, describing how the trajectories of celestial bodies change qualitatively as system parameters vary. The word itself means "forking" - and indeed many bifurcations involve the appearance or disappearance of solution branches.
**Hysteresis:** near a saddle-node bifurcation the system behaves differently depending on the direction of parameter change. If μ was decreasing from positive, the system "jumps" at μ = 0. This is critical for understanding tipping points in climate, ecology, and engineering: the system does not return by the same path it came.
A bifurcation is when the system "breaks" or becomes chaotic
A bifurcation is a qualitative change in the topology of the phase portrait as a parameter changes. The system can remain completely regular before and after a bifurcation.
The word itself means "branching". A saddle-node bifurcation destroys two equilibria; a transcritical one swaps their stability; a pitchfork creates two new ones. All of this is regular behavior - no chaos involved.
For the system dx/dt = μ - x^2, at what value of μ does the saddle-node bifurcation occur?
Hopf Bifurcation
**Hodgkin-Huxley 1952: four equations describe 10^11 neurons in the brain.** The excitation threshold of a neuron is a Hopf bifurcation point. Below threshold - equilibrium (rest). Above - a limit cycle (action potential). All of neuroscience is built on one mathematical point. The same mechanics operate in cardiology: the transition from sinus rhythm to fibrillation is Hopf bifurcation in a model of excitable cells.
**Hopf bifurcation** occurs when a pair of complex conjugate eigenvalues of the Jacobian crosses the imaginary axis: Re(λ) changes sign. Normal form in polar coordinates: **dr/dt = μr - r^3, dθ/dt = ω**. For μ < 0: stable equilibrium. For μ > 0: unstable equilibrium plus a limit cycle of radius r* = √μ.
| Hopf type | For μ < 0 | For μ > 0 | Example |
|---|---|---|---|
| Supercritical | Stable equilibrium | Unstable equilibrium + stable limit cycle | Van der Pol oscillator |
| Subcritical | Stable equilibrium + unstable limit cycle | Unstable equilibrium | Some neuron models |
Andronov-Hopf, 1942
German mathematician Eberhard Hopf rigorously proved the limit-cycle birth theorem in 1942. It later emerged that Poincare and Andronov had described the same phenomenon earlier, so in Russian literature the bifurcation is often called the Andronov-Hopf bifurcation. Today it explains oscillations ranging from cardiac cells to fluid turbulence.
What is born at a supercritical Hopf bifurcation as μ passes through 0?
Pitchfork Bifurcation
**A deterministic system makes a "choice".** The same differential equation: one equilibrium for μ < 0, three for μ > 0 - two stable ones on the sides and one unstable in the center. Which of the two stable states does the system land in? That is decided by infinitesimally small fluctuations. This is spontaneous symmetry breaking - from water freezing to iron becoming magnetized.
**Pitchfork bifurcation** arises in systems with x -> -x symmetry. Normal form: **dx/dt = μx - x^3**. Equilibria: x* = 0 (always), x* = ±√μ (for μ > 0). For μ < 0: only x* = 0 is stable. For μ > 0: x* = 0 is unstable, x* = ±√μ are stable. The bifurcation diagram resembles a pitchfork - hence the name.
| Type | Normal form | Branches for μ>0 | Physical example |
|---|---|---|---|
| Supercritical | μx - x^3 | x* = ±√μ stable | Buckling of a loaded beam |
| Subcritical | μx + x^3 | x* = ±√(-μ) unstable | First-order transitions |
The pitchfork bifurcation is **symmetry breaking**. The system spontaneously chooses one of the two branches - which one is determined by infinitesimally small fluctuations. This mechanism is fundamental in physics: from water freezing to the spontaneous magnetization of a ferromagnet (Landau's model).
System dx/dt = μx - x^3. At μ = 4, which equilibria are stable?
Transcritical Bifurcation
**Sometimes equilibria neither appear nor disappear - they simply exchange stability.** Verhulst's logistic equation of 1838: dN/dt = rN(1 - N/K). When r < 0 (mortality exceeds birth rate) - population goes extinct: zero equilibrium is stable. When r > 0 - population stabilizes at K. The zero equilibrium becomes unstable; the non-zero one captures stability. This is exactly the transcritical bifurcation. The same mathematics governs epidemic models (SIR), ecosystems, and market dynamics.
**Transcritical bifurcation**: normal form: **dx/dt = μx - x^2**. Two equilibria: x* = 0 and x* = μ. For μ < 0: x* = 0 stable, x* = μ < 0 unstable. At μ = 0: they merge. For μ > 0: x* = 0 unstable, x* = μ > 0 stable. The equilibria do not disappear - they "pass through" each other and swap stability.
| Bifurcation | What happens | Symmetry | Number of equilibria |
|---|---|---|---|
| Saddle-node | 2 equilibria annihilate | None | 2 -> 0 |
| Transcritical | 2 equilibria swap stability | None (but x=0 is fixed) | 2 -> 2 |
| Pitchfork (supercritical) | 1 -> 3 equilibria | x -> -x | 1 -> 3 |
| Hopf | Equilibrium -> limit cycle | Rotational | Cycle born |
Verhulst's Logistic Equation
Pierre-Francois Verhulst proposed the logistic equation dN/dt = rN(1 - N/K) in 1838 to describe population growth with limited resources. The equation exhibits a transcritical bifurcation at r = 0: when r < 0 the population goes extinct; when r > 0 it stabilizes at K. Today the logistic equation is used in ecology, epidemiology, and demography.
All four bifurcation types are fundamentally different and unrelated
All codimension-1 equilibrium bifurcations are described by normal form theory: locally, any bifurcation reduces to one of the four normal forms via a coordinate change.
The center manifold theorem and normal form theory (Birkhoff, Arnold) show that near a bifurcation point the dynamics are determined by a small number of essential terms. The rest can be eliminated by a change of variables, leaving one of the canonical forms.
In the system dx/dt = μx - x^2 at μ = -2, which equilibrium is stable?
Key Takeaways
- **Saddle-node bifurcation:** two equilibria collide and disappear; normal form dx/dt = μ + x^2
- **Hopf bifurcation:** an equilibrium loses stability and gives birth to a limit cycle; the neuron threshold is one example
- **Pitchfork bifurcation:** symmetry breaking - one equilibrium splits into three; the system chooses a branch through fluctuations
- **Transcritical bifurcation:** two equilibria swap stability through a common point; the logistic population equation is the classic case
Related Topics
Bifurcations are the key to understanding both regular and chaotic behavior:
- Lyapunov Stability Theory — Stability of equilibria is the necessary foundation for bifurcation analysis
- Chaos and Strange Attractors — Chaos often arises through a cascade of period-doubling bifurcations
- Population Dynamics — Lotka-Volterra and logistic models are rich with bifurcations
Вопросы для размышления
- When designing a bridge: at what load will a bifurcation occur? How can it be detected in advance without pushing the structure to failure?
- Why does the pitchfork bifurcation require x -> -x symmetry? What happens when that symmetry is slightly broken?
- Climate tipping points are bifurcations. Can a bifurcation be predicted from characteristic warning signals that appear before the transition?
Связанные уроки
- dyn-03 — Stability of equilibria is the foundation for bifurcation analysis
- dyn-05 — Chaos often arises through a period-doubling bifurcation cascade
- dyn-10 — Lotka-Volterra and logistic models are rich with bifurcations
- calc-06-derivative-intro — Stability via sign of f'(x*) uses the same derivative language
- dg-04 — Sign of K classifies surface points - parallel to bifurcation types