You can think of eigenvalues and their associated eigenvectors as a decomposition of a matrix. Eigenvalues and eigenvectors are needed for PCA. But computing eigenvalues and eigenvectors directly is ...