1. Vectors That Stay on Their Span

Consider (, ).

  • Most vectors get knocked off their span (the line through the origin and their tip).
  • Some vectors stay on their span — the transformation just stretches/squishes them, like a scalar.
VectorsWhat happensFactor
On the x-axis (e.g. )Stay on x-axis×3
On the line through Stay on that diagonal×2
All othersRotated off their span—

Check: ✓

Definition 1.1 – Eigenvector & eigenvalue

An eigenvector of a transformation is a nonzero vector that stays on its own span. Its eigenvalue is the factor by which it is scaled.

  • may be negative (e.g. : flipped and halved) — what matters is staying on the line.

2. Why Care?

3D rotations

An eigenvector of a 3D rotation is its axis of rotation, with eigenvalue (rotations don’t stretch). Thinking “axis + angle” is far easier than reading a 3×3 matrix.

The columns of a matrix depend on your coordinate system; eigenvectors/eigenvalues capture what the transformation does independent of coordinates.


3. Computing Eigenvalues

Derivation

where is the matrix that scales everything by ( on the diagonal, 0 elsewhere).

We need a nonzero sent to . That’s only possible if squishes space to lower dimension (Ch. 6–7):

Theorem 3.1 – Characteristic equation

Picture: subtract a variable from the diagonal and “turn the knob” until the determinant hits 0.

Example. :

At , squishes the plane onto a line; there’s a nonzero with (fixed in place). The other eigenvalue is .

Procedure

  1. Solve for .
  2. For each , solve (a linear system) — its null space is the set of eigenvectors.

Worked example. :

  • .
  • : .
  • : .

4. Special Cases

TransformationMatrixEigenvaluesEigenvectors
90° rotation (none real)None (every vector rotates off its span)
Shear onlyOnly the x-axis
Scale by 2 onlyEvery vector

⚠️ Pitfalls:

  • Complex eigenvalues ⇒ no real eigenvectors (often a sign of rotation).
  • A single eigenvalue can come with just a line of eigenvectors (shear) or a whole plane (uniform scaling).
  • is never counted as an eigenvector.

5. Eigenbasis

Diagonal matrices

If both basis vectors are eigenvectors, e.g. scaled by and by :

Definition 5.1 – Diagonal matrix

Zeros everywhere off the diagonal. Interpretation: every basis vector is an eigenvector, with the diagonal entries as eigenvalues.

Big advantage — powers are easy:

(Computing the 100th power of a non-diagonal matrix directly is a nightmare.)

Definition 5.2 – Eigenbasis

A basis made of eigenvectors. If enough eigenvectors exist to span the space, change to that basis:

because in that coordinate system the transformation just scales each basis vector.

Worked example. , , :

⚠️ Pitfall: Not every matrix has an eigenbasis — a shear doesn’t have enough eigenvectors to span the plane.


Exercises: For , find its eigenvalues and an eigenbasis, use it to compute a formula for , and see what it reveals about the sequence (lesson page).