
26
D. C. Sorensen
9 for j = 1, 2, ..., b,
z ~ z + MT(Mjv);
1.
end.
Moreover, on a distributed memory parallel machine or on a network or cluster of work-
stations these blocks may be physically distributed to the local memories and the memory
traffic associated with reading the matrix M every time a matrix vector product is required
could be completely avoided.
4 LARGE SCALE APPLICATIONS OF THE SVD
In this section we discuss two important large scale applications of the SVD. The first of
these is the 3-D image reconstruction of biological macromolecules from 2-D projections
obtained through ...