From: Ian Scott <ian.scott@st...>  20020327 14:31:21

> This solves my (naive) test linear system but: > > 1) I have to set no. of interations to 2x size of A, otherwise I get > wrong solution, If the default value isn't big enough, it could indicate a bug in vnl_lsqr. Have a look at the documentation in the underlying netlib function, it may tell you something useful. > > 2) It takes 10 minutes for 10,000 x 10,000 while matlab takes only 1 > minute to give the same. > > 3) I have to solve ~ 10,000 x 10,000 maaaaaaany times in my > calculations > > Any ideas how to improve things? Find a better algorithm in the literature and add it to vnl. http://www.isbe.man.ac.uk/internal/software/c++/isbe_vxl_doc/books/vxl/book. html#SEC60 Ian. 