README

This package can solve large sparse linear system and sparse least squares 
problems and allows positive "damping".
The core algorithms (lsqr.c lsqr.h) are provided by SOL (System Optimisation 
Laboratory), from Stanford University. The source code from C. C. Paige and
M. A. Saunders is avalaible here :
	http://www.stanford.edu/group/SOL/software.html

We have added and wrapped around a sparse matrix implementation 
using linked list (row and column) of non nul elements, to speed up things.
We use these tools for sesimic travel time tomography purpose.

lsqrsolve code is very specific tools (works with ray2mesh) but 
we provide also some samples code (rse, rse2, rse3) which implement generic
lsqr solver Ax=B (A could be sparse or not)  in command line.

Sparse 
------
file format for sparse matrix (A) (only value!=0 should be set)

ni nj
i j value
...

file format for vector (B)

nj
value[0]
...
value[j-1]


all indices i and j start with '0' and not '1'.


Example
-------

	rse3 sparseA.txt B.txt x.txt nbiter_max damping [dump_sol_iter]

where
sparseA.txt   : sparse matrix in input
B.txt         : standard vector in input
x.txt         : solution in output
nbiter_max    : nb maximun iteration to perform
damping       : damping value in input
dump_sol_iter : dump solution for each dump_sol_iter ieration in iter-%d.txt file (optional)


When using sparse matrix please sort your data first ! It speeds up things !!
You can use the sort unix command line utility :

	sort -n -k1 -k2.1n sparseA.txt > sparseA-sorted.txt



Marc Grunberg <marc.grunberg at unistra.fr>