Fast accurate eigenvalue methods for graded positive definite matrices (Q1923282)

If \(A\) is a positive definite matrix, and its condition number is much less than that of a diagonal matrix \(D\), then \(H=DAD\) is called a graded matrix. The Cholesky factorization of \(H\) is used to propose fast algorithms for the computation of the eigenvalues of \(H\). The eigenvalues can be computed to high relative accuracy and more efficiently than by Jacobi's method. In some cases the new methods can be four times faster than tridiagonalization methods.