Eigenvectors Are the Directions That Don't Turn
The final boss of linear algebra: eigenvalues and eigenvectors. The name is terrifying, but the core is simple. Lesson 5 said a matrix is a transformation that rotates and stretches all of space. So most vectors, after the transformation, get their direction yanked sideways. But a few special directions don't get yanked at all — they just stretch or shrink in place. Those un-yanked directions are the eigenvectors, and how much they stretched is the eigenvalue. That's all it is. And these two pull together everything you saw in lessons 5, 6, and 7.
First, let's see "most directions get yanked." I'll set a transformation A. Spin an arrow all the way around the circle. The original direction (faint arrow) and the transformed direction (bold arrow) are shown together. At most angles, the two point different ways. The transformation has pushed that vector off its original line — what pointed up now points at a slant, what pointed right now tilts upward. This is the norm. A transformation scrambles directions all over the place.
But keep spinning, and at certain special angles the bold arrow lands exactly on the same line as the faint one. The direction held; only the length changed. Right there — that direction is an eigenvector: the one that didn't get swept away in the current of the transformation, that kept its own line. How much it stretched along that line is the eigenvalue. Try to find them. In 2D you usually get two such directions. A transformation may look like chaos, but it secretly has its own axes — the directions it leaves alone.
Look at what the eigenvalue does on that direction, exactly. Ride along an eigenvector and the transformation is just "multiply by some number." Eigenvalue 2 doubles you along that direction; 0.5 halves you; 1 leaves you put. Negative? You flip to the opposite end of the same line — the line is kept, just front-to-back reversed. Build different transformations with the sliders and read each eigenvector's eigenvalue. The transformation that looked complicated shrinks, along its eigen-directions, to a single plain multiplication.
Here's the genuinely beautiful part. Redraw the grid not along the usual horizontal and vertical, but along the eigenvectors, and the transformation looks completely different. No rotating, no shearing — just each axis stretched by its own eigenvalue. That complicated matrix becomes a tidy "scale each axis by some number" in the coordinate system built from its eigenvectors. Toggle the grid to warp onto the eigen-axes. Eigenvectors are the most natural axes for that transformation. Seen along them, any transformation shows you its simplest face.
So how do you find these directions? An eigenvector v is one where applying A gives the same result as just multiplying by λ. As an equation, Av = λv. Rearrange it and you get (A − λI)v = 0. For a nonzero vector v to get sent to zero, the transformation (A − λI) must have crushed space flat. Remember lesson 7? Then its determinant has to be 0. So you solve det(A − λI) = 0 for λ, and out come the eigenvalues. That "determinant 0 = flattened = can't be reversed" from lessons 6 and 7 was the key to finding eigenvalues all along. As a bonus: the determinant equals the product of all the eigenvalues — the area scaling from lesson 6 is just the per-axis stretches multiplied together. It all connects.