Abstract. On many high-speed computers the dense matrix technique is preferable to sparse matrix technique when the matrices are not very large, because the high computational speed compensates fully the disadvantages of using more arithmetic operations and more storage. Dense matrix techniques can still be used if the computations are successively carried out in a sequence of large dense blocks. A method based on this idea will be discussed.
Statement of the problemConsider: Ax = b − r with A T r = 0, where A ∈ R m×n , m ≥ n, rank(A) = n, b ∈ R m×1 , r ∈ R m×1 and x ∈ R n×1 . It is difficult to use efficiently highspeed computers for solving the above problem if matrix A is general sparse. For dense matrices all difficulties disappear and the speed of the dense matrix computations is normally close to the peak performance of the computer used; see Table 1. A new method will be described. The method is based on a reordering algorithm LORA, [3], [4], which allows us to form easily a sequence of relatively large blocks. The blocks are handled as dense matrices. This is a trade off procedure: it is accepted to perform more computations, but with higher