Skip to content
Tutorials Center
My learning

Topics

Lesson 22 of 39 ยท Systems and Phase Planes

Nonlinear Systems and Linearization

Find equilibria of nonlinear systems, linearize with the Jacobian matrix, classify stability, and test the predictions on a pendulum and predator-prey model.

Most interesting systems are nonlinear: a pendulum has sin(ฮธ), predators meet prey through a product x y, and chemical reactions multiply concentrations. Exact formulas for solutions usually do not exist. But you can still learn a lot by asking a local question: what happens near an equilibrium?

Near an equilibrium a smooth system looks almost linear, and you already know how to read linear systems from eigenvalues and phase portraits. This technique, linearization, is the standard first analysis of any nonlinear model.

Step 1: find the equilibria

For an autonomous system x' = f(x, y), y' = g(x, y), an equilibrium is a point where both right-hand sides vanish: f = 0 and g = 0. Solving these algebraic equations gives all the places where a solution can sit still.

Take the damped pendulum with angle ฮธ and angular velocity ฯ‰:

TEXT
ฮธ' = ฯ‰
ฯ‰' = -sin(ฮธ) - 0.4 ฯ‰

Equilibria need ฯ‰ = 0 and sin ฮธ = 0, so ฮธ = 0, ยฑฯ€, ยฑ2ฯ€, .... Physically, hanging straight down (ฮธ = 0) and balanced upright (ฮธ = ฯ€).

Step 2: the Jacobian matrix

Write the system as u' = F(u) with u = (x, y). Near an equilibrium u*, put u = u* + w with w small. Taylor's theorem gives F(u* + w) โ‰ˆ F(u*) + J w = J w, where J is the Jacobian matrix of first partial derivatives:

TEXT
J = [ โˆ‚f/โˆ‚x  โˆ‚f/โˆ‚y ]
    [ โˆ‚g/โˆ‚x  โˆ‚g/โˆ‚y ]    evaluated at (x*, y*)

So the small deviation w satisfies the linear system w' = J w. The eigenvalues of J classify the equilibrium exactly as in the linear case.

Python Jacobian of the pendulum with SymPy

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

At ฮธ = 0 the eigenvalues are -0.2 ยฑ 0.98i: a stable spiral, the pendulum swinging down and settling. At ฮธ = ฯ€ one eigenvalue is positive and one negative: a saddle, the upright position that is balanced but unstable.

Step 3: classify, with a warning

The Hartman-Grobman theorem says: if no eigenvalue of J has zero real part (a hyperbolic equilibrium), the nonlinear system near the equilibrium has the same type of portrait as the linearization. Saddles stay saddles, stable nodes and spirals stay stable.

Two caveats:

  • A center in the linearization (purely imaginary eigenvalues) is not decisive. The nonlinear terms can turn it into a slow spiral in or out.
  • Nodes that are degenerate or borderline can change shape slightly, though stability is kept.

So the linearization is trusted for hyperbolic points and only a first hint otherwise. Lyapunov functions, covered in the projects, handle the doubtful cases.

Linearization recipe

Equilibria Solve f = 0 and g = 0.
Jacobian Differentiate and plug in the point.
Eigenvalues Classify like a linear system.
Check Zero real part needs extra care.
1 / 2

Example: predator and prey

The classic Lotka-Volterra model has prey x and predators y:

TEXT
x' = x (1 - 0.5 y)      (prey grow, get eaten)
y' = y (-0.75 + 0.25 x) (predators starve without prey)

Equilibria: (0, 0) and (3, 2). The Jacobian is

TEXT
J = [ 1 - 0.5y      -0.5x      ]
    [ 0.25y     -0.75 + 0.25x  ]

At (0, 0) this is diag(1, -0.75): a saddle. At (3, 2) it is [[0, -1.5], [0.5, 0]] with eigenvalues ยฑ i sqrt(0.75), purely imaginary. The linearization predicts a center, and for this particular model, the true system really has closed orbits (it has a conserved quantity). It is a case where the doubtful center happens to be correct.

Python Lotka-Volterra: eigenvalues and a conserved quantity

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

H stays constant to many digits, which proves the orbit is closed and also confirms the numerical solver is accurate.

Linearization fails to see the whole picture

Linearization is local. Far from an equilibrium, new behavior can appear: limit cycles (isolated closed orbits, as in the Van der Pol oscillator), separatrices between basins of attraction, or finite-time blow-up. The damped pendulum shows this nicely: for small angles it spirals down to ฮธ = 0, but start it with enough energy and it swings over the top several times first. Both behaviors are in the same system.

Python Small versus large pendulum swings

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The second start has enough speed to pass over the upright position, so ฮธ ends up near a multiple of 2ฯ€ rather than near 0: the pendulum has wound around once or more before settling. The angle is not wrapped, which is why it can be around 6.28 at the end.

Common mistakes

  • Evaluating the Jacobian before substituting the equilibrium. You need the point first, the matrix second.
  • Trusting a linear center in a nonlinear system without further checks.
  • Forgetting equilibria: factor completely, as in x (1 - 0.5y) = 0 which has two branches.
  • Treating the linearization as valid far from the equilibrium.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. What are the equilibria of x' = x(2 - y), y' = y(x - 3)?

    x' = 0 gives x = 0 or y = 2; y' = 0 gives y = 0 or x = 3. Combining the branches that agree gives (0,0) and (3,2).
  2. 2. The Jacobian at an equilibrium has eigenvalues 2 and -1. The equilibrium is:

    Real eigenvalues of opposite sign give a saddle, and since it is hyperbolic, the nonlinear system is a saddle too.
  3. 3. When is the linearization NOT reliable on its own?

    Hartman-Grobman needs a hyperbolic equilibrium. Zero real parts leave the nonlinear terms in charge.
  4. 4. The matrix of first partial derivatives of F is called:

    The Jacobian linearizes a vector function; the Hessian collects second derivatives.

Summary

  • Equilibria solve F(u) = 0.
  • The Jacobian at an equilibrium gives the linearized system w' = J w.
  • Hyperbolic equilibria keep the type (and stability) of their linearization.
  • Centers in the linearization need extra tools such as conserved quantities or Lyapunov functions.
  • Nonlinear systems have global features, like limit cycles and basins, that linearization cannot see.

Next: Power Series Solutions