Differential Equations › Section E · Systems of differential equations › Topic E.11

Nonlinear systems: equilibria and linearization

Most nonlinear systems have no solution formula, yet close to each equilibrium they behave almost exactly like a linear system. Finding the equilibria and linearising there gives a reliable local picture of the whole phase plane.

Beyond linear systems

A two-dimensional autonomous system has the form \(x' = f(x, y)\), \(y' = g(x, y)\), where \(f\) and \(g\) do not depend on \(t\). In the earlier topics of this section \(f\) and \(g\) were linear, so superposition held and eigenvectors gave every solution. Once terms like \(xy\), \(x^2\) or \(\sin x\) appear, superposition fails, several equilibria can exist, and exact formulas are rare. The approach changes from solving to describing: locate the equilibria, find what nearby trajectories do, and join the local pictures into a phase portrait.

Equilibria and nullclines

An equilibrium, or critical point, is a point where \(f(x, y) = 0\) and \(g(x, y) = 0\) at the same time, so a solution starting there stays there. The \(x\)-nullcline is the curve \(f = 0\), where trajectories move vertically, and the \(y\)-nullcline is \(g = 0\), where they move horizontally. Equilibria are exactly the points where an \(x\)-nullcline meets a \(y\)-nullcline. When \(f\) factors, as in \(f = x(4 - 2x - y)\), every factor gives a separate nullcline, here the lines \(x = 0\) and \(2x + y = 4\). Pair each \(x\)-nullcline with each \(y\)-nullcline in turn so that no equilibrium is missed.

The Jacobian and linearisation

Near an equilibrium \((x^*, y^*)\), write \(u = x - x^*\) and \(v = y - y^*\). Expanding \(f\) and \(g\) in Taylor series, their values at the equilibrium are zero, so the first order terms dominate when \(u\) and \(v\) are small. Dropping the higher terms gives the linearisation

\begin{bmatrix} u \\ v \end{bmatrix}' = J\begin{bmatrix} u \\ v \end{bmatrix}, \quad J = \begin{bmatrix} f_x & f_y \\ g_x & g_y \end{bmatrix}

where every partial derivative is evaluated at \((x^*, y^*)\). The matrix \(J\) is the Jacobian matrix. Each equilibrium has its own Jacobian, so a system with four equilibria needs four separate evaluations.

Classifying with trace and determinant

The eigenvalues of \(J\) classify the linearisation, as in the earlier topics on real, complex and repeated eigenvalues. With trace \(\tau\) and determinant \(\Delta\), the eigenvalues solve \(\lambda^2 - \tau\lambda + \Delta = 0\). If \(\Delta < 0\), the eigenvalues are real with opposite signs: a saddle. If \(\Delta > 0\) and \(\tau^2 > 4\Delta\), they are real with the same sign: a node. If \(\tau^2 = 4\Delta > 0\), the eigenvalue is repeated: a star or degenerate node. If \(\Delta > 0\) and \(\tau^2 < 4\Delta\), they are complex: a spiral, or a centre when \(\tau = 0\). When \(\Delta > 0\), the sign of \(\tau\) gives stability: \(\tau < 0\) is stable and \(\tau > 0\) is unstable.

When linearisation can be trusted

An equilibrium is hyperbolic when no eigenvalue of \(J\) has zero real part. The Hartman-Grobman theorem says that near one, the nonlinear phase portrait is a continuous deformation of the linear one, so stability carries over. For smooth \(f\) and \(g\), a further result adds that with real distinct eigenvalues, or complex eigenvalues with nonzero real part, the type carries over too: saddles stay saddles, nodes stay nodes and spirals stay spirals. Equal real eigenvalues keep their stability, but the node may become a spiral. Two cases are inconclusive. A zero eigenvalue can hide almost anything. And a linear centre, with purely imaginary eigenvalues, can become a centre, a stable spiral or an unstable spiral in the nonlinear system.

Worked example: competing species

Assume two species each grow logistically alone, and competition removes each at a rate proportional to the product of the populations: \(x' = x(4 - 2x - y)\), \(y' = y(5 - x - 2y)\). 1. Equilibria: \((0, 0)\), \((2, 0)\), \((0, 2.5)\), and coexistence at \((1, 2)\), solving \(2x + y = 4\), \(x + 2y = 5\). 2. \(f_x = 4 - 4x - y\), \(f_y = -x\), \(g_x = -y\), \(g_y = 5 - x - 4y\). 3. At \((1, 2)\), \(J\) has rows \((-2, -1)\) and \((-2, -4)\): \(\tau = -6\), \(\Delta = 6\), and \(\tau^2 = 36 > 4\Delta\), a stable node. 4. At \((2, 0)\), \(\Delta = -12\), and at \((0, 2.5)\), \(\Delta = -7.5\): saddles. \((0, 0)\), with eigenvalues 4 and 5, is an unstable node. Coexistence is the stable outcome.

Worked example: a damped pendulum

Assume a point mass on a rigid, massless rod 2.45 m long, so \(g/L = 4\ \mathrm{s^{-2}}\), friction proportional to angular velocity with coefficient \(1\ \mathrm{s^{-1}}\), and no driving force: \(\theta'' + \theta' + 4\sin\theta = 0\). 1. Set \(x = \theta\), \(y = \theta'\): \(x' = y\), \(y' = -4\sin x - y\). 2. Equilibria need \(y = 0\) and \(\sin x = 0\): the points \((n\pi, 0)\). 3. The Jacobian has rows \((0, 1)\) and \((-4\cos x, -1)\). 4. Hanging down, at \((0, 0)\): \(\tau = -1\), \(\Delta = 4\), and \(\tau^2 < 4\Delta\), a stable spiral, so swings die away. 5. Upright, at \((\pi, 0)\): \(\Delta = -4 < 0\), a saddle, so the balanced position is unstable. Both are hyperbolic, so the conclusions hold for the nonlinear pendulum.

A centre that does not survive

Without friction the pendulum's Jacobian at \((0, 0)\) has \(\tau = 0\) and \(\Delta = 4\): a linear centre, the inconclusive case. Here the closed orbits do survive, as an energy argument in the next topic, Limit cycles, energy, and Lyapunov functions, shows. But they need not. The system \(x' = -y - x(x^2 + y^2)\), \(y' = x - y(x^2 + y^2)\) has the same linearisation as a centre at the origin. In polar coordinates it becomes \(r' = -r^3\) and \(\theta' = 1\), so trajectories wind inwards: the origin is a stable spiral. Flipping both cubic signs gives \(r' = r^3\), an unstable spiral. The small terms decide.

Basins of attraction and separatrices

When a system has more than one stable equilibrium, each one attracts a region of starting points, its basin of attraction. The boundaries between basins are usually separatrices: the special trajectories that flow into a saddle. In the damped pendulum, each downward position \((2n\pi, 0)\) is stable, and the trajectories entering the saddles at odd multiples of \(\pi\) separate starts that come to rest at one downward position from starts that swing over the top and come to rest at the next one. In a competition model whose coexistence point is a saddle, its separatrix divides starts where one species wins from starts where the other does.

Common mistakes to avoid

First, losing equilibria by dividing by a variable, such as cancelling \(x\) from \(x(4 - 2x - y) = 0\); the factor \(x = 0\) gives equilibria too. Second, writing the Jacobian with variables still in it, or evaluating it at the wrong point. Third, swapping its rows and columns: the first row holds the derivatives of \(f\). Fourth, trusting a linear centre or a zero eigenvalue; these are exactly the cases the linearisation theorem excludes. Fifth, assuming a spiral when \(\tau < 0\) without checking \(\tau^2\) against \(4\Delta\).

Key takeaways

Equilibria of \(x' = f\), \(y' = g\) are the intersections of the nullclines \(f = 0\) and \(g = 0\). At each one, the Jacobian of partial derivatives gives a linearisation, classified by its trace and determinant. At hyperbolic equilibria the nonlinear stability matches the linear prediction; linear centres and zero eigenvalues need further analysis. A damped pendulum has stable spirals hanging down and saddles upright. Saddles and their separatrices divide the plane into basins of attraction.

Flashcards

Flashcard

Autonomous system

x′ = f(x, y), y′ = g(x, y), with no t on the right

The phase plane picture does not change with time.

Test your understanding

Question 1 of 5

What are all the equilibria of x′ = x(3 − y), y′ = y(x − 2)?

Question 2 of 5

For the frictionless pendulum system x′ = y, y′ = −9 sin x, what kind of equilibrium is (π, 0)?

Question 3 of 5

At an equilibrium of a nonlinear system, the Jacobian has eigenvalues 2i and −2i. What can you conclude from the linearisation alone?

Question 4 of 5

The Jacobian at an equilibrium has trace −5 and determinant 4. What type is the equilibrium?

Question 5 of 5

What is the Jacobian of x′ = x² − y, y′ = x − y at the equilibrium (1, 1)?

← Previous topicNext topic →