Introduction

LASPack is a package for solving large sparse systems of linear equations like those which arise from discretization of partial differential equations.

Main features:

• The primary aim of LASPack is the implementation of efficient iterative methods for the solution of systems of linear equations. All routines and data structures are optimized for effective usage of resources especially with regard to large sparse matrices. The package can be accessed from an application through a straightforward interface defined in the form of procedure calls.
• Beside the obligatory Jacobi, succesive over-relaxation, Chebyshev, and conjugate gradient solvers, LASPack contains selected state-of-the-art algorithms which are commonly used for large sparse systems:
  • CG-like methods for non-symmetric systems: CGN, GMRES, BiCG, QMR, CGS, and BiCGStab,
  • multilevel methods such as multigrid and conjugate gradient method preconditioned by multigrid and BPX preconditioners.
  All above solvers are applicable not only to the positive definite or non-symmetric matrices, but are also adopted for singular systems (e.g. arising from discretization of Neumann boundary value problems).
• The implementation is based on an object-oriented approach (although it has been programmed in C). Vectors and matrices are defined as new data types in connection with the corresponding supporting routines. The basic operations are implemented so that they allow the programming of linear algebra algorithms in a natural way.
• LASPack is extensible in a simple manner. An access to the internal representation of vectors and matrices is not necessary and is, as required of the object-oriented programming, avoided. This allows an improvement of algorithms or a modification of data structures with no adjustment of application programs using the package.
• LASPack is written in ANSI C and is thus largely portable.