spectral theorem

The spectral theorem is the clean, no-surprises promise for symmetric matrices, the ones that equal their own transpose (A = A^T). It guarantees two lovely things: all of A's eigenvalues are real numbers (no awkward complex values), and you can always find a full set of eigenvectors that are mutually perpendicular and unit length.

Stacking those orthonormal eigenvectors into a matrix Q lets you write A = Q*D*Q^T, where D is diagonal holding the eigenvalues. This is diagonalization at its most pristine: Q is orthogonal, so Q^T is also its inverse, and the whole map becomes 'rotate to the eigen-directions, scale along each axis, rotate back'.

This gold case matters in practice because symmetric matrices appear everywhere worth caring about: covariance matrices in statistics, energy and stiffness matrices in physics, and the kernels behind many algorithms. When your matrix is symmetric, the spectral theorem says the nicest possible structure is guaranteed.