A few shapes of matrix are so well-behaved they get their own names. Two will matter most for what's ahead.

A diagonal matrix has non-zero entries only on the main diagonal: \begin{bmatrix} a & 0 \\ 0 & d \end{bmatrix}. It does the simplest possible thing to a vector — scales the first component by a and the second by d, each axis stretched on its own, with no mixing.

A symmetric matrix equals its own transpose, A = A^{\mathsf{T}} — it's a mirror image across the diagonal, so a_{ij} = a_{ji}. For example \begin{bmatrix} 3 & 2 \\ 2 & 5 \end{bmatrix}.

A diagonal matrix stretches the axes

Watch a diagonal matrix act on the unit square. Slider a stretches it horizontally, slider d vertically — and the square stays a neat rectangle, because the axes never mix. That clean, independent stretching is exactly why diagonal matrices are the easiest of all to work with.

Why we hunt for these

Diagonal matrices are trivial to compute with — multiplying, powering and inverting all come almost for free. So a recurring dream in linear algebra is to make a matrix look diagonal by choosing the right basis; that's diagonalization. Symmetric matrices are the heroes there: they can always be diagonalized, by a set of perpendicular axes — the fact that makes principal component analysis work.