Diagonalisation
When a matrix has enough independent eigenvectors, it can be written as a diagonal matrix in disguise. Diagonalisation makes powers and long-term behaviour easy to compute.
The diagonalisation theorem
An n x n matrix A is diagonalisable if \(A = PDP^{-1}\) for an invertible P and diagonal D. This happens exactly when A has n linearly independent eigenvectors: the columns of P are the eigenvectors, and the diagonal entries of D are the corresponding eigenvalues, in the same order.
Why it works
If \(\mathbf{v}_1, \dots, \mathbf{v}_n\) are eigenvectors, then \(AP = [A\mathbf{v}_1 \cdots A\mathbf{v}_n] = [\lambda_1\mathbf{v}_1 \cdots \lambda_n\mathbf{v}_n] = PD\). If the eigenvectors are independent, P is invertible and \(A = PDP^{-1}\). In the eigenvector basis, A simply scales each coordinate by its eigenvalue.
When is A diagonalisable
An n x n matrix with n distinct eigenvalues is always diagonalisable, since eigenvectors for distinct eigenvalues are independent. With repeated eigenvalues, A is diagonalisable exactly when each eigenvalue's geometric multiplicity equals its algebraic multiplicity, so that the eigenspace dimensions add up to n. Matrices such as [[2, 1], [0, 2]] that fall short are called defective.
Powers of matrices
Diagonalisation turns powers into scalar powers: \(A^k = PD^kP^{-1}\), and \(D^k\) just raises each diagonal entry to the kth power. Computing \(A^{100}\) directly needs many multiplications; with diagonalisation it needs one power of each eigenvalue. This is how long-run behaviour of Markov chains and population models is found.
Long-term behaviour
For a discrete dynamical system \(\mathbf{x}_{k+1} = A\mathbf{x}_k\), writing \(\mathbf{x}_0\) in the eigenvector basis gives \(\mathbf{x}_k = c_1\lambda_1^k\mathbf{v}_1 + \dots + c_n\lambda_n^k\mathbf{v}_n\). As k grows, the term with the largest \(|\lambda|\) dominates, so the system's state aligns with that eigenvector and grows or decays at that rate, the basis of the power method for finding dominant eigenvalues.
Diagonalising symmetric matrices
Real symmetric matrices are always diagonalisable, with real eigenvalues and orthogonal eigenvectors, so P can be chosen orthogonal. This spectral theorem is developed in the decompositions unit and underlies principal component analysis and quadratic forms.
Beyond diagonalisation
Defective matrices cannot be diagonalised, but every square matrix has a Jordan canonical form, nearly diagonal with eigenvalues on the diagonal and some 1s just above it, introduced by Camille Jordan in 1870. In numerical work, the Schur decomposition, with an orthogonal change of basis to an upper triangular matrix, is preferred because it is stable.
Worked example
Diagonalise \(A = \begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix}\), with eigenvalues 5 and 2. For 5, an eigenvector is (1, 1); for 2, \(A - 2I\) has rows (2, 1), (2, 1), giving (1, -2). So \(P = \begin{bmatrix} 1 & 1 \\ 1 & -2 \end{bmatrix}\) and \(D = \begin{bmatrix} 5 & 0 \\ 0 & 2 \end{bmatrix}\). Then \(A^{10} = PD^{10}P^{-1}\), with \(D^{10}\) having diagonal entries 9,765,625 and 1,024.
Key takeaways
A is diagonalisable, A = P D P^{-1}, exactly when it has n independent eigenvectors, the columns of P, with eigenvalues on D's diagonal in matching order. AP = PD shows why. Distinct eigenvalues guarantee diagonalisability; repeated ones require geometric multiplicity to equal algebraic multiplicity, otherwise the matrix is defective. A^k = P D^k P^{-1} makes powers easy. In x_{k+1} = A x_k, the largest |lambda| dominates, underlying the power method. Real symmetric matrices diagonalise with orthogonal P. Jordan form (1870) handles defective matrices; Schur decomposition is the stable numerical alternative.
Flashcard
A = P D P^{-1}
Check your understanding
Why is a 3 x 3 matrix with eigenvalues 1, 2, and 7 guaranteed to be diagonalisable?
