The directions that don't turn

Apply a transformation and almost every vector gets knocked off its line — stretched and turned. But for most matrices a few special directions refuse to turn. Find them and you've found the transformation's skeleton.

The directions that don't turn

An eigenvector of A is a nonzero vector that A only scales:

$Av = \lambda v$

The scaling factor λ is its eigenvalue. The vector may stretch (λ > 1), shrink (λ between 0 and 1), or flip backwards (λ negative) — but its line is sacred: A maps that line onto itself.

Why they're the skeleton

For A = [2 1 / 1 2] the eigen-directions are (1, 1) with λ = 3 and (1, −1) with λ = 1. Read the whole motion from just that: stretch ×3 along the (1, 1) diagonal, leave the (1, −1) diagonal alone. Every other vector's fate is a mix of those two behaviors — decompose it along the two eigen lines and each piece scales by its own λ. Two numbers and two directions replace the whole grid picture.

Try it live in the visualizer: the dashed invariant lines are drawn for you — put a test vector on one and watch it refuse to turn while the morph runs.

Not every matrix cooperates fully

• A rotation turns every direction — no real eigenvectors at all (its eigenvalues go complex; module 3 of the algebra course would call that "no real roots").
• A shear has just one invariant line, not two.
• A uniform scaling is the opposite extreme: every direction is an eigenvector.

The visualizer reports each case — worth five minutes of experiments.

Why this matters

Eigen-analysis is how the huge systems of the real world get understood: the natural vibration modes of a bridge, the principal components of a dataset, Google's original PageRank — all "find the directions the transformation treats simply." Next lesson: how to actually compute them.