On normwise structured backward errors for the generalized saddle point systems (Q2359409): Difference between revisions

The author considers the problem of solving generalized saddle point systems. These are systems of linear equations whose coefficient matrix is in the form of a \(2 \times 2\) block matrix whose principal blocks are square and whose off-diagonal blocks are transposes of each other. When the trailing subblock is zero this is a saddle point system. If, in addition, the leading subblock is symmetric, one obtains what is often referred to as a Karush-Kuhn-Tucker system. The author obtains explicit and computable formula for the structured backward errors of the generalized saddle point system. Using numerical examples, it is shown that these expressions are handy in testing the stability of practical algorithms.