Matrix decompositions

Matrix factorisations play a key role in the solution of problems of the type $A x = b$. Often (e.g. ODE solvers), you have a fixed matrix $A$ that must be solved with many different $b$ vectors. A matrix factorisation is effectivly a pre-processing step that allows you to partition $A$ into multiple factors (e.g. $A = LU$ in the case of $LU$ decomposition), so that the actual solve is as quick as possible. Different decompositions have other uses besides solving $A x = b$, for example:

  • the $LU$, $QR$ and Cholesky decomposition can be used to quickly find the determinant of a large matrix, since $\det(AB) = \det(A) \det(B)$ and the determinant of a triangular matrix is simply the product of its diagonal entries.
  • The Cholesky decomposition can be used to sample from a multivariate normal distribution, and is a very efficient technique to solve $A x = b$ for the specific case of a positive definite matrix.
  • The $QR$ decomposition can be used to solve a minimum least squares problem, to find the eigenvalues and eigenvectors of a matrix, and to calulcate the Singular Value Decomposition (SVD), which is itself another very useful decomposition!