Differential Equations · Undergraduate

Nonlinear Systems, Equilibria and Linearization

Quick answer

Most real systems are nonlinear: predators and prey, competing species, a swinging pendulum. Their equilibria are the points where x′ and y′ are both zero. Near an equilibrium, the system behaves almost like the linear system u′ = Ju, where J is the Jacobian matrix of partial derivatives there. The eigenvalues of J classify the equilibrium as a saddle, node or spiral, stable or unstable, as for linear systems. Purely imaginary eigenvalues are the one case the linearization cannot settle.

What you'll learn

  • Find the equilibria of a nonlinear system
  • Compute the Jacobian matrix at an equilibrium
  • Classify equilibria from the Jacobian's eigenvalues
  • Interpret predator–prey and competing-species models

Equilibria of a nonlinear system

A two-variable system

x′=f(x,y),y′=g(x,y)x' = f(x, y), \qquad y' = g(x, y)

defines a velocity at every point of the phase plane. An equilibrium is a point where f=g=0f = g = 0. Near an equilibrium (x∗,y∗)(x^*, y^*), write uu for the small displacement. Then, to first order,

u′=Ju,J=[fxfygxgy](x∗,y∗)u' = Ju, \qquad J = \begin{bmatrix} f_x & f_y\\ g_x & g_y \end{bmatrix}_{(x^*, y^*)}

The eigenvalues of J classify the equilibrium:

Eigenvalues of JJEquilibrium
real, opposite signssaddle (unstable)
real, both negativestable node
real, both positiveunstable node
complex, negative real partstable spiral
complex, positive real partunstable spiral
purely imaginaryundecided by linearization

Why the Jacobian governs the motion nearby

Near the equilibrium, f(x∗+u1,y∗+u2)≈f(x∗,y∗)+fxu1+fyu2f(x^* + u_1, y^* + u_2) \approx f(x^*, y^*) + f_xu_1 + f_yu_2 by the tangent plane approximation, and f(x∗,y∗)=0f(x^*, y^*) = 0. The same holds for gg. So close to an equilibrium, the nonlinear system is a linear system in disguise, and the leftover terms are too small to change a saddle, node or spiral. A center is different: its orbits close up exactly, and the smallest nonlinear push can make them spiral in or out.

Predator and prey in the phase plane Arrows of the velocity field for x′ = x(3 − y), y′ = y(x − 2), with prey x across and predators y up, both from 0 to 6. The arrows circle counterclockwise around the equilibrium (2, 3). At the origin, a second equilibrium, the arrows move away along the x-axis and in along the y-axis. 123456123456xy (2, 3)
Predator and prey in the phase plane

Worked examples

Common mistakes

Practice problems

  1. Find the equilibria of x′=x−yx' = x - y, y′=x2−4y' = x^2 - 4.

    Answer

    (2,2)(2, 2) and (−2,−2)(-2, -2)

    Full solution

    x2=4x^2 = 4 gives x=±2x = \pm 2, and x=yx = y.

  2. Classify the equilibria in Exercise 1.

    Answer

    (2,2)(2, 2): unstable spiral; (−2,−2)(-2, -2): saddle

    Full solution

    J=[1−12x0]J = \begin{bmatrix} 1 & -1\\ 2x & 0 \end{bmatrix}. At (2,2)(2, 2): trace 11, determinant 44, and 1−16<01 - 16 < 0, so complex with positive real part. At (−2,−2)(-2, -2): determinant −4-4.

  3. Classify the origin for x′=−x+xyx' = -x + xy, y′=−2y+x2y' = -2y + x^2.

    Answer

    Stable node

    Full solution

    J(0,0)=[−100−2]J(0, 0) = \begin{bmatrix} -1 & 0\\ 0 & -2 \end{bmatrix}, with eigenvalues −1-1 and −2-2.

  4. What does linearization say about the undamped pendulum θ′=ω\theta' = \omega, ω′=−sin⁡θ\omega' = -\sin\theta at (0,0)(0, 0)?

    Answer

    Eigenvalues ±i\pm i: a linear center, which linearization alone cannot confirm.

    Full solution

    Energy 12ω2−cos⁡θ\tfrac{1}{2}\omega^2 - \cos\theta is conserved, which shows the orbits near (0,0)(0, 0) are closed.

  5. In Example 1, what happens to the prey with no predators?

    Answer

    It grows exponentially: x′=3xx' = 3x.

    Full solution

    Setting y=0y = 0 leaves x′=3xx' = 3x, so x=x0e3tx = x_0e^{3t}.

  6. For the Lotka–Volterra model x′=ax−bxyx' = ax - bxy, y′=−cy+dxyy' = -cy + dxy, find the nonzero equilibrium.

    Answer

    (cd,ab)\left(\tfrac{c}{d}, \tfrac{a}{b}\right)

    Full solution

    x′=0x' = 0 with x≠0x \ne 0 gives y=aby = \tfrac{a}{b}; y′=0y' = 0 with y≠0y \ne 0 gives x=cdx = \tfrac{c}{d}.

  7. An equilibrium has JJ with trace −4-4 and determinant 33. Classify it.

    Answer

    Stable node

    Full solution

    λ2+4λ+3=0\lambda^2 + 4\lambda + 3 = 0 gives λ=−1,−3\lambda = -1, -3.

  8. A student classifies (2,3)(2, 3) in Example 1 using J(0,0)J(0, 0) and calls it a saddle. What went wrong?

    Answer

    The Jacobian must be evaluated at (2,3)(2, 3) itself.

    Full solution

    J(2,3)=[0−230]J(2, 3) = \begin{bmatrix} 0 & -2\\ 3 & 0 \end{bmatrix} has eigenvalues ±i6\pm i\sqrt{6}, not those of J(0,0)J(0, 0).

Frequently asked questions

What is an equilibrium of a system?

A point where x′ = 0 and y′ = 0 at once, so a solution starting there stays there.

How do you linearize at an equilibrium?

Compute the Jacobian J = [[f_x, f_y], [g_x, g_y]] at the point. Near it, the deviation u from the equilibrium satisfies approximately u′ = Ju.

How do the eigenvalues classify the equilibrium?

Real with opposite signs: saddle. Both negative: stable node. Both positive: unstable node. Complex with negative real part: stable spiral; positive real part: unstable spiral.

When does linearization fail?

When the eigenvalues are purely imaginary (or zero). The linear system has a center, but the nonlinear terms may turn it into a slow spiral either way. Other arguments, such as a conserved quantity, are needed.

What is the Lotka–Volterra model?

A predator–prey system x′ = ax − bxy, y′ = −cy + dxy. Prey grow alone, predators die alone, and meetings help predators and hurt prey. Populations cycle around the equilibrium (c/d, a/b).