Efficiently updating a covariance matrix and its LDL decomposition

Abstract

Equations are presented which efficiently update or downdate the covariance matrix of a large number of $m$-dimensional observations. Updates and downdates to the covariance matrix, as well as mixed updates/downdates, are shown to be rank-$k$ modifications, where $k$ is the number of new observations added plus the number of old observations removed. As a result, the update and downdate equations decrease the required number of multiplications for a modification to $\Theta((k+1)m^2)$ instead of $\Theta((n+k+1)m^2)$ or $\Theta((n-k+1)m^2)$, where $n$ is the number of initial observations. Having the rank-$k$ formulas for the updates also allows a number of other known identities to be applied, providing a way of applying updates and downdates directly to the inverse and decompositions of the covariance matrix. To illustrate, we provide an efficient algorithm for applying the rank-$k$ update to the LDL decomposition of a covariance matrix.