Frontier physics

Nonlinear dynamical systems, fixed points, and bifurcations

Nonlinear evolution equations can change stability or their long-term states as parameters vary; fixed points and bifurcations are central tools for analyzing these changes.

Linear models allow superposition and proportional response. Nonlinearity removes superposition: multiple equilibria may coexist, limit cycles can appear, and a small parameter change can qualitatively alter long-term behavior.

\dot{x}=f(x,\mu),\quad f(x^*,\mu)=0;\quad \dot{\delta x}\simeq D_xf(x^*,\mu)\,\delta x

Definition: Fixed point and linear stability

A fixed point satisfies f(x*,μ)=0. In one dimension, f_x<0 damps small perturbations and f_x>0 destabilizes them; in higher dimensions inspect the Jacobian eigenvalues' real parts. A zero eigenvalue calls for nonlinear analysis, often near a bifurcation.

See equilibria of dx/dt=μ−x² appear or disappear and change stability as μ crosses zero.

Saddle-node bifurcation

The normal form dx/dt=μ−x² has fixed points x*=±√μ for μ>0; the positive one is stable and the negative one unstable. At μ=0 they merge into a nonhyperbolic point; for μ<0 no real fixed point exists. This is a local universal structure, though a physical system may require coordinate changes and higher-order terms.

Example: Classify the equilibria

For dx/dt=μ−x² with μ=4, find the fixed points and their linear stability.

Solution

Setting 0=4−x² gives x*=±2. Since f_x=−2x, the slope is −4 at +2 (stable) and +4 at −2 (unstable).

Near a bifurcation, expansion in slow coordinates and a control parameter often reduces a vector field to a low-dimensional normal form. After rapidly decaying modes are removed, the normal form reveals universality without claiming that every system shares the same microscopic physics. In the logistic map $x_{n+1}=r x_n(1-x_n)$ , increasing $r$ produces period doublings before chaos; Feigenbaum identified an almost universal ratio for their accumulation. This illustrates how global structure can emerge from a simple equation, while specific predictions still require attention to noise, measured parameters, and the model's domain of validity.

A multidimensional system is often studied by continuing solutions as a parameter varies, tracking eigenvalues, and locating bifurcations. Numerical steps must be refined near degeneracies or a cycle may be missed or a stable branch misidentified. Local analysis should be checked against basins of attraction and long integrations; reduced models are reliable only over specified state and timescale ranges.

Quick check

What condition defines a fixed point of ẋ=f(x,μ)?

How many real fixed points does ẋ=μ−x² have for μ<0?

References

Steven H. Strogatz (2015). Nonlinear Dynamics and Chaos