Repeated Eigenvalues

If the matrix in x′ = Ax has a repeated eigenvalue λ with only one independent eigenvector v, the second solution is te^(λt)v + e^(λt)w, where w is a generalized eigenvector: (A − λI)w = v. The general solution is x = c₁e^(λt)v + c₂e^(λt)(tv + w). If A = λI, every vector is an eigenvector and x = e^(λt)x(0).

Before this lesson: Solving Linear Systems with Eigenvalues, Complex Eigenvalues and Repeated Roots of the Characteristic Equation

Two kinds of repeated eigenvalues

A 2 × 2 matrix has a repeated eigenvalue when its characteristic polynomial is a perfect square, (λ − r)². What happens next depends on how many independent eigenvectors that eigenvalue has.

  • Two eigenvectors (complete case): this happens only when A = λI. Every nonzero vector is an eigenvector, so x = e^(λt)x(0). Solutions move straight in or out along every line, a picture called a star (or proper node).
  • One eigenvector (defective case): this is the usual case. The eigenvalue method gives only one solution, e^(λt)v, so you need a second one.

Finding the second solution

For one equation with a repeated root, the second solution was te^(rt). For a system, te^(λt)v alone is not enough, because differentiating it leaves an extra term. Adding e^(λt)w fixes that term.

\mathbf{x} = te^{\lambda t}\mathbf{v} + e^{\lambda t}\mathbf{w}

Try this form, with v the eigenvector and w unknown.

\mathbf{x}' = e^{\lambda t}\mathbf{v} + \lambda te^{\lambda t}\mathbf{v} + \lambda e^{\lambda t}\mathbf{w}

Differentiate te^(λt) with the product rule.

A\mathbf{x} = \lambda te^{\lambda t}\mathbf{v} + e^{\lambda t}A\mathbf{w}

Multiply by A and use Av = λv.

e^{\lambda t}\mathbf{v} + \lambda e^{\lambda t}\mathbf{w} = e^{\lambda t}A\mathbf{w}

Set x′ = Ax and cancel λte^(λt)v from both sides.

(A - \lambda I)\mathbf{w} = \mathbf{v}

Divide by e^(λt) and subtract λw from both sides.

A vector w with (A − λI)w = v is called a generalized eigenvector. It is never unique: adding any multiple of v to w still works, and only changes the constant c₁.

\mathbf{x}(t) = c_1e^{\lambda t}\mathbf{v} + c_2e^{\lambda t}\left(t\mathbf{v} + \mathbf{w}\right)

The origin is then an improper node (or degenerate node). It is stable if λ is negative and unstable if λ is positive. As t → ∞, the tv term wins, so solutions become parallel to the eigenvector v.

How to solve x′ = Ax with a repeated eigenvalue

  1. Solve det(A − λI) = 0 and find the repeated root λ.
  2. If A = λI, write x = e^(λt)x(0) and stop.
  3. Find one eigenvector v from (A − λI)v = 0.
  4. Find a generalized eigenvector w from (A − λI)w = v; any solution works.
  5. Write x = c₁e^(λt)v + c₂e^(λt)(tv + w), and use c₁v + c₂w = x(0) for an initial-value problem.

Repeated eigenvalues examples

Example 1: A general solution

Find the general solution of x′ = [3 −1; 1 1]x.

\det(A - \lambda I) = (3 - \lambda)(1 - \lambda) + 1

Write the characteristic polynomial: (3 − λ)(1 − λ) − (−1)(1).

(3 - \lambda)(1 - \lambda) + 1 = \lambda^2 - 4\lambda + 4

Multiply out and combine like terms.

\lambda^2 - 4\lambda + 4 = (\lambda - 2)^2

Factor the perfect square trinomial; λ = 2 is repeated.

A - 2I = \begin{bmatrix} 1 & -1 \\ 1 & -1 \end{bmatrix} \Rightarrow v_1 - v_2 = 0

Subtract 2 on the diagonal; both rows give the same equation.

\mathbf{v} = \begin{bmatrix} 1 \\ 1 \end{bmatrix}

Choose v₁ = 1; there is only one independent eigenvector.

(A - 2I)\mathbf{w} = \mathbf{v}:\ w_1 - w_2 = 1

Set up the generalized eigenvector equation; both rows read w₁ − w₂ = 1.

\mathbf{w} = \begin{bmatrix} 1 \\ 0 \end{bmatrix}

Choose w₂ = 0, so w₁ = 1.

\mathbf{x} = c_1e^{2t}\begin{bmatrix} 1 \\ 1 \end{bmatrix} + c_2e^{2t}\left(t\begin{bmatrix} 1 \\ 1 \end{bmatrix} + \begin{bmatrix} 1 \\ 0 \end{bmatrix}\right)

Write the general solution.

Answer: x(t) = c₁e^(2t)(1, 1) + c₂e^(2t)(t + 1, t). In components, x₁ = e^(2t)(c₁ + c₂ + c₂t) and x₂ = e^(2t)(c₁ + c₂t). Since λ = 2 is positive, the origin is an unstable improper node.

Example 2: An initial-value problem with a stable node

Solve x′ = [−3 1; −1 −1]x with x(0) = (3, 5).

\lambda^2 + 4\lambda + 4 = (\lambda + 2)^2

Trace −4 and determinant 3 + 1 = 4, so det(A − λI) = λ² + 4λ + 4.

A + 2I = \begin{bmatrix} -1 & 1 \\ -1 & 1 \end{bmatrix} \Rightarrow \mathbf{v} = \begin{bmatrix} 1 \\ 1 \end{bmatrix}

For λ = −2: −v₁ + v₂ = 0.

-w_1 + w_2 = 1 \Rightarrow \mathbf{w} = \begin{bmatrix} 0 \\ 1 \end{bmatrix}

Solve (A + 2I)w = v; choose w₁ = 0.

c_1\begin{bmatrix} 1 \\ 1 \end{bmatrix} + c_2\begin{bmatrix} 0 \\ 1 \end{bmatrix} = \begin{bmatrix} 3 \\ 5 \end{bmatrix}

Set t = 0: c₁v + c₂w = x(0).

c_1 = 3,\quad c_1 + c_2 = 5 \Rightarrow c_2 = 2

Read c₁ from the first row, then solve the second.

\mathbf{x} = e^{-2t}\left(3\begin{bmatrix} 1 \\ 1 \end{bmatrix} + 2t\begin{bmatrix} 1 \\ 1 \end{bmatrix} + 2\begin{bmatrix} 0 \\ 1 \end{bmatrix}\right)

Substitute c₁ and c₂.

\mathbf{x} = e^{-2t}\begin{bmatrix} 3 + 2t \\ 5 + 2t \end{bmatrix}

Add the vectors.

Answer: x₁ = (3 + 2t)e^(−2t) and x₂ = (5 + 2t)e^(−2t). Check: x₁′ = (−4 − 4t)e^(−2t), and −3x₁ + x₂ = (−9 − 6t + 5 + 2t)e^(−2t) = (−4 − 4t)e^(−2t). The solution decays to the origin and comes in parallel to (1, 1).

Phase plane with axes x1 and x2 from -6 to 6. A dashed green line through the origin along the direction (1, 1) is the only eigenvector line. Six blue curves, including the solution starting at the marked point (3, 5), move toward the origin, bending so that they arrive parallel to the green line.
Example 2: a stable improper node. Every solution decays like e^(−2t) and arrives at the origin parallel to the eigenvector (1, 1) (dashed green).

Example 3: The complete case

Solve x′ = [−1 0; 0 −1]x with x(0) = (2, −3).

\det(A - \lambda I) = (-1 - \lambda)^2 \Rightarrow \lambda = -1

The eigenvalue −1 is repeated.

A + I = \begin{bmatrix} 0 & 0 \\ 0 & 0 \end{bmatrix}

Every nonzero vector solves (A + I)v = 0, so A = −I has two independent eigenvectors.

\mathbf{x} = e^{-t}\mathbf{x}(0) = e^{-t}\begin{bmatrix} 2 \\ -3 \end{bmatrix}

Each equation is xᵢ′ = −xᵢ on its own.

Answer: x = e^(−t)(2, −3). No t-term appears, and the solution moves straight toward the origin along the line through (2, −3). The origin is a stable star. For a 2 × 2 matrix, this complete case happens only when A is a multiple of I.

Example 4: Two tanks in a row (word problem)

Two 100 L tanks are connected in a row. Pure water flows into tank 1 at 5 L/min, the mixture flows from tank 1 into tank 2 at 5 L/min, and it leaves tank 2 at 5 L/min. Tank 1 starts with 10 kg of salt and tank 2 with none. Find x₁(t) and x₂(t), and find when tank 2 holds the most salt.

x_1' = -0.05x_1,\quad x_2' = 0.05x_1 - 0.05x_2

Salt leaves each tank at (amount/100)(5) kg/min, as in mixing problems.

\lambda = -0.05 \text{ (repeated)}

A = [−0.05 0; 0.05 −0.05] is triangular, so read the diagonal.

A + 0.05I = \begin{bmatrix} 0 & 0 \\ 0.05 & 0 \end{bmatrix} \Rightarrow \mathbf{v} = \begin{bmatrix} 0 \\ 1 \end{bmatrix}

The equation 0.05v₁ = 0 forces v₁ = 0.

0.05w_1 = 1 \Rightarrow \mathbf{w} = \begin{bmatrix} 20 \\ 0 \end{bmatrix}

Solve (A + 0.05I)w = v; choose w₂ = 0.

c_1\begin{bmatrix} 0 \\ 1 \end{bmatrix} + c_2\begin{bmatrix} 20 \\ 0 \end{bmatrix} = \begin{bmatrix} 10 \\ 0 \end{bmatrix} \Rightarrow c_1 = 0,\ c_2 = 0.5

Set t = 0 and match x(0) = (10, 0).

\mathbf{x} = 0.5e^{-0.05t}\left(t\begin{bmatrix} 0 \\ 1 \end{bmatrix} + \begin{bmatrix} 20 \\ 0 \end{bmatrix}\right) = e^{-0.05t}\begin{bmatrix} 10 \\ 0.5t \end{bmatrix}

Substitute and multiply by 0.5.

x_2' = 0.5e^{-0.05t}(1 - 0.05t)

Differentiate x₂ = 0.5te^(−0.05t) with the product rule and factor.

1 - 0.05t = 0 \Rightarrow t = 20

Set x₂′ = 0; e^(−0.05t) is never 0.

x_2(20) = 0.5(20)e^{-1} = 10e^{-1} \approx 3.68

Substitute t = 20.

Answer: x₁ = 10e^(−0.05t) and x₂ = 0.5te^(−0.05t) kilograms. Tank 2 holds the most salt, about 3.68 kg, at t = 20 minutes. Equal outflow rates give equal diagonal entries, which is why the eigenvalue repeats and t appears in the answer.

Example 5: Critical damping as a system

Write y″ + 6y′ + 9y = 0 with y(0) = 1 and y′(0) = 0 as a system, solve it, and compare with the scalar method.

\mathbf{x}' = \begin{bmatrix} 0 & 1 \\ -9 & -6 \end{bmatrix}\mathbf{x}

Let x₁ = y and x₂ = y′, as in systems of differential equations.

\lambda^2 + 6\lambda + 9 = (\lambda + 3)^2

det(A − λI) = −λ(−6 − λ) + 9; λ = −3 is repeated.

A + 3I = \begin{bmatrix} 3 & 1 \\ -9 & -3 \end{bmatrix} \Rightarrow \mathbf{v} = \begin{bmatrix} 1 \\ -3 \end{bmatrix}

The first row gives 3v₁ + v₂ = 0.

3w_1 + w_2 = 1 \Rightarrow \mathbf{w} = \begin{bmatrix} 0 \\ 1 \end{bmatrix}

Solve (A + 3I)w = v; choose w₁ = 0.

c_1\begin{bmatrix} 1 \\ -3 \end{bmatrix} + c_2\begin{bmatrix} 0 \\ 1 \end{bmatrix} = \begin{bmatrix} 1 \\ 0 \end{bmatrix} \Rightarrow c_1 = 1,\ c_2 = 3

The first row gives c₁ = 1; the second gives −3 + c₂ = 0.

\mathbf{x} = e^{-3t}\left(\begin{bmatrix} 1 \\ -3 \end{bmatrix} + 3t\begin{bmatrix} 1 \\ -3 \end{bmatrix} + 3\begin{bmatrix} 0 \\ 1 \end{bmatrix}\right)

Substitute c₁ and c₂.

\mathbf{x} = e^{-3t}\begin{bmatrix} 1 + 3t \\ -9t \end{bmatrix}

Add the vectors.

Answer: y = x₁ = (1 + 3t)e^(−3t), and the velocity is y′ = x₂ = −9te^(−3t). This is the same answer the scalar method gives in repeated roots of the characteristic equation: the te^(rt) term there is the tv term here.

Common mistakes

MistakeExample of the errorFix
Using te^(λt)v as the second solutionx = c₁e^(2t)(1, 1) + c₂te^(2t)(1, 1)Add e^(λt)w: the second solution is e^(λt)(tv + w).
Solving (A − λI)w = 0 for wGetting w = (1, 1) again, the same as vSolve (A − λI)w = v, with v on the right side.
Putting t on the wrong vectore^(λt)(v + tw)The t goes with the eigenvector: e^(λt)(tv + w).
Assuming every repeated eigenvalue is defectiveAdding a t-term for A = −IIf A − λI is the zero matrix, every vector is an eigenvector and x = e^(λt)x(0).
Using t = 0 incorrectly in the initial valuesc₁v + c₂(v + w) = x(0)At t = 0, tv is the zero vector: c₁v + c₂w = x(0).
Thinking w must be uniqueCalling w = (2, 1) wrong when the answer key has (1, 0)Any w with (A − λI)w = v works; only c₁ changes.

Related ideas

Where repeated eigenvalues sit among the cases

For a 2 × 2 matrix, the eigenvalues repeat exactly when (tr A)² − 4 det A = 0. This is the border between two real eigenvalues, from solving linear systems with eigenvalues, and a complex pair, from complex eigenvalues. An improper node looks like a spiral that does not quite turn. Phase Portraits compares all of these pictures.

Larger systems

In an n × n system, an eigenvalue can repeat more than twice. Then you may need a chain of generalized eigenvectors, (A − λI)w₁ = v and (A − λI)w₂ = w₁, which give solutions with t²/2 terms. These chains are the idea behind the Jordan form in linear algebra.

Practice problems

  1. Find the eigenvalues of A = [5 −4; 1 1].
  2. For A = [5 −4; 1 1], find an eigenvector v and a generalized eigenvector w.
  3. Write the general solution of x′ = [5 −4; 1 1]x.
  4. Solve x′ = [−2 0; 0 −2]x with x(0) = (4, 1).
  5. Solve x′ = [1 1; 0 1]x with x(0) = (2, 3).
  6. For A = [3 −1; 1 1], λ = 2 and v = (1, 1), show that te^(2t)v alone is not a solution of x′ = Ax.
  7. Write y″ − 4y′ + 4y = 0 as a system, find v and w, and give y(t).
  8. Classify the origin for x′ = [3 −1; 1 1]x and for x′ = [−1 0; 0 −1]x.
  9. Two 200 L tanks are in a row with water flowing at 10 L/min, pure water entering tank 1. Tank 1 starts with 20 kg of salt and tank 2 with none. Find x₂(t) and its maximum.
  10. In Example 1, show that w = (2, 1) is also a generalized eigenvector, and explain why it gives the same solutions.
Show answers

1.

\lambda^2 - 6\lambda + 9 = 0

Trace 6 and determinant 5 + 4 = 9.

(\lambda - 3)^2 = 0 \Rightarrow \lambda = 3

Factor the perfect square.

Answer: λ = 3, repeated

2.

A - 3I = \begin{bmatrix} 2 & -4 \\ 1 & -2 \end{bmatrix} \Rightarrow \mathbf{v} = \begin{bmatrix} 2 \\ 1 \end{bmatrix}

The second row gives v₁ − 2v₂ = 0.

w_1 - 2w_2 = 1 \Rightarrow \mathbf{w} = \begin{bmatrix} 1 \\ 0 \end{bmatrix}

Solve (A − 3I)w = v using the second row; row 1 is twice row 2.

Answer: v = (2, 1) and w = (1, 0)

3.

\mathbf{x} = c_1e^{3t}\begin{bmatrix} 2 \\ 1 \end{bmatrix} + c_2e^{3t}\left(t\begin{bmatrix} 2 \\ 1 \end{bmatrix} + \begin{bmatrix} 1 \\ 0 \end{bmatrix}\right)

Use v and w from Problem 2.

Answer: x = c₁e^(3t)(2, 1) + c₂e^(3t)(2t + 1, t)

4.

A = -2I \Rightarrow \mathbf{x} = e^{-2t}\mathbf{x}(0)

The complete case: every vector is an eigenvector.

Answer: x = e^(−2t)(4, 1)

5.

A - I = \begin{bmatrix} 0 & 1 \\ 0 & 0 \end{bmatrix} \Rightarrow \mathbf{v} = \begin{bmatrix} 1 \\ 0 \end{bmatrix},\ \mathbf{w} = \begin{bmatrix} 0 \\ 1 \end{bmatrix}

λ = 1 is repeated; v₂ = 0 for v, and w₂ = 1 for w.

c_1 = 2,\quad c_2 = 3

Match c₁(1, 0) + c₂(0, 1) = (2, 3).

\mathbf{x} = e^{t}\left(2\begin{bmatrix} 1 \\ 0 \end{bmatrix} + 3t\begin{bmatrix} 1 \\ 0 \end{bmatrix} + 3\begin{bmatrix} 0 \\ 1 \end{bmatrix}\right)

Substitute the constants.

Answer: x = e^t(2 + 3t, 3)

6.

\left(te^{2t}\mathbf{v}\right)' = (1 + 2t)e^{2t}\mathbf{v}

Product rule.

A\left(te^{2t}\mathbf{v}\right) = te^{2t}A\mathbf{v} = 2te^{2t}\mathbf{v}

Use Av = 2v.

Answer: the two sides differ by e^(2t)v = e^(2t)(1, 1), so it is not a solution

7.

\mathbf{x}' = \begin{bmatrix} 0 & 1 \\ -4 & 4 \end{bmatrix}\mathbf{x},\quad (\lambda - 2)^2 = 0

Let x₁ = y, x₂ = y′; det(A − λI) = λ² − 4λ + 4.

\mathbf{v} = \begin{bmatrix} 1 \\ 2 \end{bmatrix},\quad \mathbf{w} = \begin{bmatrix} 0 \\ 1 \end{bmatrix}

Row 1 of A − 2I: −2v₁ + v₂ = 0, and −2w₁ + w₂ = 1.

y = x_1 = c_1e^{2t} + c_2te^{2t}

First entries of c₁e^(2t)v + c₂e^(2t)(tv + w).

Answer: y = (c₁ + c₂t)e^(2t)

8.

\lambda = 2 \gt 0 \text{ with one eigenvector}

First matrix, from Example 1.

\lambda = -1 \lt 0,\ A = -I

Second matrix.

Answer: unstable improper node; stable star (proper node)

9.

x_1' = -0.05x_1,\quad x_2' = 0.05x_1 - 0.05x_2

10/200 = 0.05; same matrix as Example 4.

c_1 = 0,\quad 20c_2 = 20 \Rightarrow c_2 = 1

With v = (0, 1) and w = (20, 0), match x(0) = (20, 0).

\mathbf{x} = e^{-0.05t}\begin{bmatrix} 20 \\ t \end{bmatrix}

e^(−0.05t)(t(0, 1) + (20, 0)).

x_2' = e^{-0.05t}(1 - 0.05t) = 0 \Rightarrow t = 20

Product rule; set the derivative to 0.

Answer: x₂ = te^(−0.05t) kg, with a maximum of 20e^(−1) ≈ 7.36 kg at t = 20 min

10.

(A - 2I)\begin{bmatrix} 2 \\ 1 \end{bmatrix} = \begin{bmatrix} 2 - 1 \\ 2 - 1 \end{bmatrix} = \begin{bmatrix} 1 \\ 1 \end{bmatrix}

Multiply [1 −1; 1 −1] by (2, 1).

(2, 1) = (1, 0) + (1, 1)

The new w is the old w plus v.

Answer: (2, 1) works; c₂e^(2t)(tv + w + v) = c₂e^(2t)(tv + w) + c₂e^(2t)v, and the extra c₂e^(2t)v joins the c₁ term

Repeated eigenvalues FAQ

What is a generalized eigenvector?

It is a vector w with (A − λI)w = v, where v is an eigenvector for λ. It is not an eigenvector itself, but A − λI sends it to one. It supplies the missing second solution when an eigenvalue repeats.

Why is the second solution te^(λt)v + e^(λt)w?

Substituting te^(λt)v into x′ = Ax leaves an extra e^(λt)v. Adding e^(λt)w cancels it exactly when (A − λI)w = v.

What if a repeated eigenvalue has two eigenvectors?

Then there is no t-term: x = c₁e^(λt)v₁ + c₂e^(λt)v₂. For a 2 × 2 matrix this happens only when A = λI, and every solution is e^(λt)x(0).

What does the phase portrait look like for a repeated eigenvalue?

With one eigenvector, the origin is an improper node: solutions approach (λ < 0) or leave (λ > 0) along curves that become parallel to the eigenvector. With A = λI, it is a star, with straight-line solutions in every direction.

Summary

  • A repeated eigenvalue λ usually has only one eigenvector v.
  • Find a generalized eigenvector w from (A − λI)w = v.
  • The general solution is x = c₁e^(λt)v + c₂e^(λt)(tv + w); at t = 0, c₁v + c₂w = x(0).
  • If A = λI, the solution is simply x = e^(λt)x(0).
  • The origin is an improper node (one eigenvector) or a star (A = λI), stable when λ is negative.

Previous lesson: Complex Eigenvalues

Next lesson: Phase Portraits

Back to the Differential Equations Index