Skip to content
Tutorials Center
My learning

Topics

Lesson 20 of 39 · Systems and Phase Planes

Linear Systems and Eigenvalues

Solve u' = Au with eigenvalues and eigenvectors, handle complex and repeated eigenvalues, use the matrix exponential, and check everything in Python.

A linear system with constant coefficients, u' = A u, is the one family of systems you can solve completely by hand. The key is a guess that worked for the scalar equation y' = a y, whose solutions are c e^(at). For a system you try the same shape with a constant vector in front: u(t) = v e^(λt).

Substituting that guess turns the differential equation into plain linear algebra, namely the eigenvalue problem A v = λ v. So every fact about eigenvalues and eigenvectors becomes a fact about the motion of the system. This lesson builds the general solution, deals with the awkward cases, and connects it to the matrix exponential.

Why the guess works

Put u = v e^(λt) into u' = A u. The left side is λ v e^(λt) and the right side is A v e^(λt). Cancel the exponential and you get A v = λ v. So v must be an eigenvector of A and λ the matching eigenvalue. Each eigenpair gives one solution.

If A is n x n with n independent eigenvectors, the general solution is

TEXT
u(t) = c1 v1 e^(λ1 t) + c2 v2 e^(λ2 t) + ... + cn vn e^(λn t)

The constants come from the initial condition: solve V c = u(0), where the columns of V are the eigenvectors. Superposition is allowed because the system is linear.

Python General solution from eigenvalues

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

Program input · 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 eigenvalues are 4 and -1. One mode grows like e^(4t) and the other decays like e^(-t), so for large t the growing mode wins and the solution lines up with its eigenvector.

The three cases for 2 by 2 systems

The characteristic equation λ^2 - tr(A) λ + det(A) = 0 has roots of three kinds.

Real and distinct. The formula above applies. Signs of the eigenvalues decide the behavior: both negative means every solution decays, both positive means growth, opposite signs means a saddle.

Complex pair λ = α ± iβ. The eigenvectors are complex conjugates too. Real and imaginary parts of one complex solution are two independent real solutions:

TEXT
u(t) = e^(αt) [ c1 (a cos βt - b sin βt) + c2 (a sin βt + b cos βt) ]

where v = a + i b is the eigenvector for α + iβ. The factor e^(αt) sets growth or decay and β sets the rotation speed, so solutions spiral.

Repeated λ. If A has only one eigenvector for the double root, take v an eigenvector, solve (A - λI) p = v for a generalized eigenvector p, and the second solution is (t v + p) e^(λt). The extra factor t bends the straight-line solutions.

Python Classify a 2x2 system by trace and determinant

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

Program input · 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 matrix exponential

For a single equation y' = a y the solution is y(0) e^(at). The same formula holds for systems if you define e^(At) by the power series I + At + (At)^2/2! + .... Then u(t) = e^(At) u(0). When A = V D V^-1 with D diagonal, e^(At) = V e^(Dt) V^-1, which is exactly the eigenvector formula in compact form.

The matrix exponential is the right tool when A is not diagonalizable, and SciPy provides scipy.linalg.expm. It is also the bridge to numerical thinking: if you step a distance h, u(t+h) = e^(Ah) u(t) exactly, so a numerical method is just an approximation of e^(Ah).

Python Compare expm with the eigenvector formula

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

Program input · 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 eigenvalues are -0.2 ± 1.99i: real part -0.2 is the decay rate, imaginary part about 1.99 is the angular frequency.

Stability in one sentence

The origin of u' = A u is asymptotically stable (all solutions tend to 0) when every eigenvalue has a negative real part. If any eigenvalue has a positive real part, almost every solution blows up. Purely imaginary eigenvalues give neutral oscillation. You will see this geometrically in Phase Portraits.

Common mistakes

  • Using eigenvectors as columns versus rows. NumPy returns them as columns of V.
  • Forgetting to take the real part when working with complex eigenvalues of a real matrix.
  • Treating a repeated eigenvalue as if it always gave two eigenvectors.
  • Judging decay by the eigenvalue magnitude: the real part decides decay, the imaginary part only decides oscillation speed.

Check your understanding

Quick check · 4 questions

of 4 answered You scored / 4 · perfect!

  1. 1. If A has eigenvalues -1 and -3, what happens to solutions of u' = Au as t grows?

    Both exponentials e^(-t) and e^(-3t) tend to zero, so every combination does.
  2. 2. Eigenvalues -0.2 + 3i and -0.2 - 3i produce:

    The negative real part -0.2 gives decay and the imaginary part 3 gives rotation.
  3. 3. The solution of u' = Au with initial vector u0 can be written as:

    The matrix exponential e^(At) is the solution operator for constant-coefficient linear systems.
  4. 4. A 2x2 matrix has det(A) < 0. What is the origin?

    A negative determinant means the eigenvalues are real with opposite signs, which is a saddle.

Summary

  • Try u = v e^(λt): it works exactly when A v = λ v.
  • With enough eigenvectors, u(t) = Σ c_k v_k e^(λ_k t), with constants from u(0).
  • Complex eigenvalues α ± iβ give spirals; real parts decide growth or decay.
  • Repeated eigenvalues may need a generalized eigenvector and a factor t.
  • u(t) = expm(At) u(0) always works, and stability depends on the signs of the real parts.

Next: Phase Portraits- heading: Solution recipe cards: - icon: "1️⃣" title: Eigenpairs text: Solve det(A - lambda I) = 0. - icon: "2️⃣" title: Combine text: Sum c v e to the lambda t. - icon: "3️⃣" title: Fit text: Match u at time zero. note: Or simply use expm(A t).