Differential Equations › Section E · Systems of differential equations › Topic E.6

Complex eigenvalues

A real matrix can have complex eigenvalues, and when it does the solutions rotate. This topic turns one complex solution into two real ones, then reads the rotation, its speed and its growth or decay straight from the eigenvalue.

Where rotation comes from

Suppose the real 2 × 2 matrix \(A\) has eigenvalues \(\alpha \pm \beta i\) with \(\beta \ne 0\), as happens whenever \((\operatorname{tr} A)^2 < 4\det A\). The recipe of "Solutions to systems" still works: \(e^{\lambda t}\mathbf{v}\) solves \(\mathbf{x}' = A\mathbf{x}\) for \(\lambda = \alpha + \beta i\) and its eigenvector \(\mathbf{v}\). But this solution is complex, and the quantities we model are real. The way out uses Euler's formula, familiar from "Complex roots":

e^{(\alpha + \beta i)t} = e^{\alpha t}(\cos \beta t + i\sin \beta t)

The cosine and sine produce rotation, and the factor \(e^{\alpha t}\) controls the size.

One complex solution gives two real ones

Write the complex solution as \(\mathbf{x}(t) = \mathbf{u}(t) + i\mathbf{w}(t)\), with \(\mathbf{u}\) and \(\mathbf{w}\) real. Substituting gives \(\mathbf{u}' + i\mathbf{w}' = A\mathbf{u} + iA\mathbf{w}\). Because \(A\) is real, the real parts and the imaginary parts must match separately, so \(\mathbf{u}' = A\mathbf{u}\) and \(\mathbf{w}' = A\mathbf{w}\). The real and imaginary parts are therefore real solutions, and they are linearly independent, so the general real solution is

\mathbf{x}(t) = c_1\mathbf{u}(t) + c_2\mathbf{w}(t)

Use only one eigenvalue of the conjugate pair. The other one produces the conjugate solution, whose real and imaginary parts are \(\mathbf{u}\) and \(-\mathbf{w}\): nothing new.

Worked example: a centre

Solve \(\mathbf{x}' = A\mathbf{x}\), \(\mathbf{x}(0) = (3, 1)\), for

A = \begin{bmatrix} 1 & -5 \\ 1 & -1 \end{bmatrix}

1. Trace 0 and determinant \(-1 + 5 = 4\) give \(\lambda^2 + 4 = 0\), so \(\lambda = \pm 2i\). 2. For \(2i\), the second row of \(A - 2iI\) is \((1, -1 - 2i)\), so \(v_1 = (1 + 2i)v_2\) and \(\mathbf{v} = (1 + 2i, 1)\). 3. Multiply out \((\cos 2t + i\sin 2t)\mathbf{v}\): the real part is \(\mathbf{u} = (\cos 2t - 2\sin 2t, \cos 2t)\) and the imaginary part is \(\mathbf{w} = (2\cos 2t + \sin 2t, \sin 2t)\). 4. At \(t = 0\): \(c_1(1, 1) + c_2(2, 0) = (3, 1)\), so \(c_1 = c_2 = 1\). 5. So \(x_1 = 3\cos 2t - \sin 2t\) and \(x_2 = \cos 2t + \sin 2t\).

Reading the centre

Every solution in the example is built from \(\cos 2t\) and \(\sin 2t\), so it repeats with period \(\frac{2\pi}{2} = \pi\). Trajectories are closed ellipses around the origin, which is a centre: stable, since solutions starting near it stay near, but not asymptotically stable, since they never approach it. To find the direction, compute \(A(1, 0) = (1, 1)\). The second entry is positive, so trajectories cross the positive \(x_1\)-axis moving upwards, which means anticlockwise rotation. A quick check on the solution: at \(t = 0\), \(\mathbf{x}'(0) = (-2, 2)\), which matches \(A(3, 1) = (3 - 5, 3 - 1)\).

Worked example: a stable spiral

Solve \(\mathbf{x}' = A\mathbf{x}\), \(\mathbf{x}(0) = (2, 0)\), for

A = \begin{bmatrix} -3 & -2 \\ 4 & 1 \end{bmatrix}

1. Trace -2 and determinant \(-3 + 8 = 5\) give \(\lambda^2 + 2\lambda + 5 = 0\), so \(\lambda = -1 \pm 2i\). 2. For \(-1 + 2i\), the first row of \(A - \lambda I\) is \((-2 - 2i, -2)\), so \(v_2 = -(1 + i)v_1\) and \(\mathbf{v} = (1, -1 - i)\). 3. The real and imaginary parts of \(e^{-t}(\cos 2t + i\sin 2t)\mathbf{v}\) are \(\mathbf{u} = e^{-t}(\cos 2t, \sin 2t - \cos 2t)\) and \(\mathbf{w} = e^{-t}(\sin 2t, -\cos 2t - \sin 2t)\). 4. At \(t = 0\): \(c_1(1, -1) + c_2(0, -1) = (2, 0)\), so \(c_1 = 2\) and \(c_2 = -2\). 5. So \(\mathbf{x} = e^{-t}(2\cos 2t - 2\sin 2t, 4\sin 2t)\).

Reading the spiral

The solution is \(e^{-t}\) times a vector that repeats every \(\pi\), so one full turn later the solution is exactly \(e^{-\pi} \approx 0.043\) times as large. Trajectories spiral into the origin, an asymptotically stable spiral point. The direction again comes from \(A(1, 0) = (-3, 4)\): the second entry is positive, so the rotation is anticlockwise. This rule always works for complex eigenvalues, because if the lower-left entry of \(A\) were 0, the matrix would be triangular and its eigenvalues would be real. In general, \(\alpha < 0\) spirals in, \(\alpha > 0\) spirals out (an unstable spiral point), and \(\alpha = 0\) gives a centre.

Growth rate and frequency

The two parts of \(\lambda = \alpha + \beta i\) do separate jobs. The real part \(\alpha\) is the growth or decay rate of the envelope \(e^{\alpha t}\). The imaginary part \(\beta\) is the angular frequency, so each turn takes \(\frac{2\pi}{|\beta|}\). This matches the second-order picture. Converting \(y'' + 2y' + 5y = 0\) gives the matrix with rows (0, 1) and (-5, -2), whose eigenvalues are \(-1 \pm 2i\), the same pair as the spiral example and the roots found by the method of "Complex roots". In "Mechanical vibrations" terms, this is an underdamped spring, and an undamped spring gives a centre.

Common mistakes to avoid

First, deciding stability from \(\beta\) or from the size of \(\lambda\); in continuous time only the sign of the real part \(\alpha\) matters. Second, forgetting the factor \(e^{\alpha t}\) when splitting into real and imaginary parts. Third, using both eigenvalues of the pair and ending up with four solutions, or with complex constants; one eigenvalue gives both real solutions. Fourth, slips with \(i^2 = -1\) when multiplying out, which often flip a sign; check the answer at \(t = 0\) and by substituting. Fifth, reading the rotation from the first entry of \(A(1, 0)\); the second entry decides up or down.

Key takeaways

For a real matrix with eigenvalues \(\alpha \pm \beta i\), take one eigenpair, expand \(e^{\lambda t}\mathbf{v}\) with Euler's formula, and use its real and imaginary parts as two independent real solutions. The real part \(\alpha\) sets growth or decay and \(\beta\) sets the angular frequency, with period \(\frac{2\pi}{|\beta|}\). Then \(\alpha < 0\) gives a stable spiral, \(\alpha > 0\) an unstable spiral, and \(\alpha = 0\) a centre. The sign of the second entry of \(A(1, 0)\) gives the direction: positive for anticlockwise, negative for clockwise.

Flashcards

Flashcard

Complex eigenvalues of a real 2 × 2 matrix

α ± βi, occurring when (tr A)² < 4 det A

They always come as a conjugate pair.

Test your understanding

Question 1 of 5

A real 2 × 2 matrix has eigenvalues −2 ± 3i. What does the phase portrait of x′ = Ax look like?

Question 2 of 5

For a real matrix A, x′ = Ax has the complex solution u(t) + i w(t) with u = (cos t, −sin t) and w = (sin t, cos t). What is the general real solution?

Question 3 of 5

A 2 × 2 system has eigenvalues ±4i. What is the period of every non-constant solution?

Question 4 of 5

A has rows (2, −3) and (5, −2), so its eigenvalues are ±i√11. Which way do the trajectories of x′ = Ax turn?

Question 5 of 5

A real matrix has eigenvalue 1 + 2i with eigenvector (1, i). Which is the real part of e^((1 + 2i)t)(1, i)?

← Previous topicNext topic →