On the Use of Structural Zeros in Orthogonal Factorization
Jesse L. Barlow · SIAM Journal on Scientific and Statistical Computing · 1990
If the orthogonal factorization of a sparse matrix A is the result of column updates, the numerical upper triangular factor R may have numerical nonzeros where there are structural zeros. This problem arises in the solution of sparse inequality-constrained least squares problems by active set methods. In order to fit R into a static data structure, these nonzeros have to be neglected, even though, theoretically, they may be quite large. In this communication, it is shown that neglecting these nonzeros does not have an adverse effect.