🚀 go-pugleaf

  I'm trying to convert a mainframe program to run on a PC, and need to find
a way to solve a large sparse system of linear equations in limited space.
The format of the system is "Ax=y" where A is a square matrix, and y is a
known column vector, x is unknown.  The matrix A has the format

	(  I  P  )
	(  Q  0  )

 where I is the identity, 0 is a zero matrix, and P,Q are both sparse (they
will usually have less than 10 non-zero entries in any row).  I and 0 are
square matrices of comparable size (generally dim(I) > dim(0) ).

  My problem arises in that the dimension of the system can be as high as
2000 - which makes a full representation impractical on a PC. Given the
special format, does anyone know a useful algorithm (or better yet, have
some working C code) I can use?
-------------------------------------------------------------------------------
    /   \   /                 |    /      Alan Young @ uk.ac.ox.vax
   /____/  /  ___    __       |___/
  /    /  /  /   |  /  |         /           "The bigger they are,
 /     \_/\__\__/|_/  /_________/               the harder they fall on you."