libkrylov, a Modular Open-Source Software Library for Extremely Large Eigenvalue and Linear Problems

Libkrylov is an open-source library which implements Krylov subspace methods for extremely large and dense linear and eigenvalue problems. Key specifications of libkrylov are:

• No matrix storage: Rather than attempting to manipulate extremly large matrix arrays, libkrylov provides an interface for user-defined "engines" for on-the-fly matrix-vector multiplication.
• Nonorthonormal methods: Maximize efficiency of matrix-vector multiplication
• "Block" methods: Entire subspaces instead of single vectors may be iterated and processed.
• Support for structured problems
• A-posteriori error bounds and condition checks for error control
• Compile-time polymorphism: Combines flexible choice of base types with high efficiency and generic program structure
• Extensible: Modular data structures using Fortran 2003 and newer features, encapsulation of functionality

Version 0.8.1-beta release notes:

• Initial implementation of core library functions: symmetric/Hermitian eigenvalue problems, linear systems, and Sylvester problems
• Support for real and complex single and double precision types with compile-time polymorphism
• Support of non-orthonormal algorithms, dynamic re-start
• Initial version of interfaces for user-defined functions and subroutines
• Script-based unit testing
• Automake-based configuration and compilation