The Geometric Intuition Behind Eigenvalues (Finally)

A linear transformation maps vectors to vectors. Most vectors change direction under a linear transformation — a vector pointing north-east might, after transformation, point south-west or at an entirely different angle. Eigenvectors are the exceptions: vectors whose direction is unchanged by the transformation (they may be scaled, but they stay on their own line through the origin).

The eigenvalue is the scale factor: how much longer or shorter the eigenvector gets after the transformation.

A concrete example

Consider the transformation T that multiplies the x-component by 3 and leaves the y-component unchanged:

T = [[3, 0], [0, 1]]

If we apply T to the vector v = (1, 0) (pointing along the x-axis), we get T·v = (3, 0). The vector tripled in length but stayed on the x-axis. Direction unchanged. This is an eigenvector with eigenvalue 3.

If we apply T to the vector w = (0, 1) (pointing along the y-axis), we get T·w = (0, 1). The vector is unchanged. This is an eigenvector with eigenvalue 1.

Any other vector — say (1, 1) — maps to (3, 1), which points in a different direction. It is not an eigenvector of this transformation.

What the characteristic polynomial is doing

To find eigenvectors algebraically, we need vectors v such that T·v = λ·v (the transformation just scales v by some number λ). Rearranging: (T − λI)·v = 0. For this to have a non-zero solution v, the matrix (T − λI) must be singular — its determinant must be zero.

det(T − λI) = 0 is the characteristic polynomial. For our example:

det([[3−λ, 0], [0, 1−λ]]) = (3−λ)(1−λ) = 0

Solutions: λ = 3 and λ = 1 — exactly the eigenvalues we found geometrically.

Why eigenvalues matter beyond the definition

Diagonalisation: if a matrix has n independent eigenvectors, it can be written as T = PDP^{−1} where D is a diagonal matrix of eigenvalues and P is the matrix of eigenvectors. Diagonal matrices are easy to raise to powers — T^n = PD^nP^{−1}, and D^n just raises each diagonal entry to the nth power. This is how Google's PageRank algorithm, Markov chain steady states, and many physics problems are solved efficiently.

When eigenvectors do not exist (over the reals)

A rotation by 90° has no real eigenvectors — every non-zero vector changes direction after a 90° rotation. The characteristic polynomial has complex roots. Over the complex numbers, every matrix has eigenvalues (Fundamental Theorem of Algebra). The complex eigenvalues of a rotation encode the angle of rotation.