A note on a one-sided Jacobi algorithm (Q1262697)

A variant of Jacobi's diagonalization algorithm for positive definite matrices A is proposed that has the advantage of a better vectorization due to the fact that the matrix to be updated is accessed only by columns. In a first step the given matrix A is formally replaced by \(\hat A\) that is obtained by a step of the symmetric LR algorithm. In principle the matrix \(\hat A\) is diagonalized by a sequence of plane rotations in an implicit way by using one-sided rotations to the Cholesky factor L of \(A=LL^ t\). The high efficiency of the proposed algorithm in comparison with the QL algorithm is outlined.