The factorability of symmetric matrices and some implications for statistical linear models (Q5947458)

The paper is concerned with the Cholesky factorization for a symmetric indefinite matrix \(K\). The theoretical background that involves both the existence and the numerical stability of the factorization is presented but an applicability of the theoretical results is emphasized. The existence of the Cholesky factorization is not assured when the matrix \(K\) is not positive definite. Conditions are stated under which a symmetric permutation \(PKP^{T}\) of the matrix \(K\) is factorizable. The existence of the Cholesky factorization is shown for matrix structures that appear in statistical applications. The author demonstrates how the Cholesky factorization of an indefinite matrix relates to Kalman filtering, recursive least squares or to a likelihood function in linear statistical models.