About - Overview

ML contains a variety of parallel multigrid schemes:

• smoothed aggregation
• FAS nonlinear multigrid
• a special algebraic multigrid for the eddy current approximations to Maxwell's equations. This eddy current solver is a unique capability of ML and utilizes the discrete nullspace of the operator in building the smoothers and grid hierarchy.

Extensibility

ML can also be used as a framework to generate new multigrid methods. Using ML's internal aggregation routines and Galerkin products, it is possible to focus on new types of inter-grid transfer operators without having to address the cumbersome aspects of generating an entirely new parallel algebraic multigrid code. We have used this flexibility to produce special multilevel methods using coarse grid finite element functions to serve as inter-grid transfers.

Our primary goal has been to provide state-of-the-art iterative methods that perform well on parallel computers (applications on over 3000 processors have been run) and that at the same time are easy to use for application engineers. In addition to providing algebraic multilevel methods to engineers, the ML library is also used in our research on preconditioners. At present, we are working closely with several specific Sandia applications in further extending our capabilities.

Capabilities

• Parallel: MPI-based
• Supported platforms: Linux clusters, Sun, SGI, Sandia's ASCI red machine (a.k.a. Janus), Sandia's RedStorm, IBM SP2, ...
• Data-Neutral Interface: Matrices are accessed via getrow() and matvec() functions. Wrappers for Aztec/MSR, Aztec/VBR and Epetra_RowMatrix matrices are furnished within the package.
• Primary Multilevel Schemes
• Smoothed aggregation
• Edge element eddy current AMG
• Smoothed aggregation domain decomposition preconditioners (additive, hybrid)
• Finite element based two-level schemes
• Refinement-based multigrid
• Classical AMG
• Parallel Smoother Methods
• Processor-based Gauss-Seidel type methods
• Processor-based block Jacobi, Gauss-Seidel and symmetric Gauss-Seidel
• Polynomial-based smoothers
• One step CG smoothing
• Two-stage distributed relaxation hybrid for eddy current equations
• Any AztecOO preconditioner
• Any IFPACK preconditioner
• Any Amesos solver
as coarse solver
• Sparse approximate inverse interface
• A set of efficient and user-friendly, MATLAB-like interface, called MLAPI.

ML is designed to interoperate with other Trilinos packages, and in particular with the AztecOO linear solver package, also developed at Sandia. However, ML can be run stand-alone. Alternatively, the user can call ML from his own application by supplying matrix-getrow functions and matrix-vector multiply functions. The user can optionally supply application-specific smoothers, or use ML's internal smoothers.

ML is used within several applications at Sandia and a few applications outside of Sandia. ML is released for external distribution and can be obtained as part of the Trilinos development environment.

Read More about ML, its project goals, functionalities, commented examples, more documentation.