Suppose we have a fixed, real $m \times n$ matrix $A$, where $m < n$ and $\mathrm{rank}(A) = m$, e.g. $A$ has full row-rank. I find myself needing to repeatedly solve equations of the form $$ ADA^T \vec{x} = \vec{b} $$ for varying matrices $D$, though $D$ will always be diagonal and possess strictly positive entries. Thus, it makes sense to do some up-front work on $A$ to make subsequent solves faster.

Currently, the best thing to do seems to be the Cholesky decomposition for each product $ADA^T=L_DL_D^T$; though I was wondering if either

  • there was some nice way to parameterize the factorization in terms of $D$ explicitly (e.g. $L_D = L(D)$) and reuse prior work, or

  • if another matrix decomposition was available and made more sense?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.