Semi-conjugate direction methods for real positive definite systems (Q1826460)

The authors propose a new iterative scheme for solving non-symmetric systems of linear equations -- the left conjugate direction method (LCD). This method is based on the concept of conjugate directions and does not rely upon a minimization problem in a Krylov subspace. LCD becomes the conjugate gradient method for symmetric positive definite matrices. It is shown that LCD obtains the solution of systems with dimension \(n\) in at most \(n\) steps in the absence of roundoff errors. The efficiency of LCD depends heavily on the choice of the initial conjugate direction. Although the paper gives some guidance how to choose this direction, an optimal choice is an open problem. Each step of LCD requires two matrix-vector multiplications, one of them with the transposed matrix. Some numerical examples are presented in which LCD was quite efficient compared to established iterative schemes.