The Bordering Method of the Cholesky Decomposition and its Backward Differentiation

This paper describes the backward differentiation of the Cholesky decomposition by the bordering method. The backward differentiation of the Cholesky decomposition by the inner product form and the outer product form have been described elsewhere. It is found that the resulting algorithm can be adapted to vector processing, as is also true of the algorithms developed from the inner product form and outer product form. The three approaches can also be fashioned to treat sparse matrices, but this is done by enforcing the same sparse structure found for the Cholesky decomposition on a secondary work space.