Eigenvalues and eigenvectors

Most vectors, hit by a matrix, come out pointing somewhere new: rotated, sheared, tangled up with the other dimensions. An eigenvector is the exception — a direction the matrix leaves alone, merely stretching or shrinking it, possibly flipping it end to end. In symbols, Av = λv: applying the matrix to v gives back the same vector times a single number, the eigenvalue λ. Along that one direction, a matrix multiplication has collapsed to ordinary multiplication.

Eigenvectors are found by solving det(A − λI) = 0 for λ, then solving (A − λI)v = 0 for v at each root — which is exactly where the determinant's meaning from earlier pays off: A − λI is singular precisely when it collapses some direction to zero, and that collapsed direction is the eigenvector. A zero eigenvalue is not a failure of the method; it says the matrix itself is not invertible, since some vector is being squashed to nothing.

The payoff is a change of perspective. Described in the standard axes, a transformation can look hopelessly tangled. Described along its eigenvectors — when there are enough independent ones to serve as a basis — the same transformation is just a set of independent stretches, one number per axis, with no coupling between them at all. This is why eigenvectors are the natural description of a system's own modes: a vibrating structure's normal modes, a quantum system's stationary states, and the stable directions of any linear process are found by exactly this calculation. Eigenvectors belonging to different eigenvalues also point in genuinely different directions — never the same one — which is what makes them useful as an alternative set of axes at all.