ML -- History of visible changes
================================
This file reports (some of) the visible changes to ML.
Please submit ML bugs at software.sandia.gov/buzgilla, or
report bugs to jhu@sandia.gov.
------------------------------------------------------------------------------
Jan 2005
* The first cut of the MLAPI is now available. MLAPI is the
MATLAB-like parallel programming module of ML. Right now only
a limited subset of ML functionalities is available.
* Improved the MultiLevelPreconditioner class (now derived from
Epetra_Operator and no longer from Epetra_RowMatrix). Updated the
filtering/GGB algorithm, which now supports only Anasazi and the
non-symmetric cycle. Some never-used methods have been deleted
(like scaling).
* Improved compilation on several machines; now ML *should* compile
with no warnings on Linux using `-g -ansi -pedantic -Wall
-Wno-format-extra-args'.
------------------------------------------------------------------------------
Dec 2004
* improved the organization of the examples, introduces more
detailed comments and some README files. Deleted some outdated
examples.
* bug fixes to the auxiliary matrix algorithm for problems defined
on stretched grids.
------------------------------------------------------------------------------
Nov 2004 (current version 3.1)
* ML now supports Zoltan as coarsening scheme (RCB only). Zoltan
can be used to create the aggregates, but also to redistribute
the next-level operator among the specified number of
processors (which can be less than the available number of
processoes)
* Several minor and major fixes to the example suite
* Added a test to verify the correctness of compiled examples
* Extended the test harness scripts and files
* a variable-block partiononing scheme (based on METIS) is
now available.
------------------------------------------------------------------------------
Oct 2004
* support for new IFPACK smoothers. ML now supports IFPACK's
variable block Jacobi, Gauss-Seidel and symmetric Gauss-Seidel
preconditioners, as well as the IFPACK-Amesos preconditioners
(subdomain-based exact LU factorizations)
* fixes support for Amesos solver (introducing LAPACK as coarse
solver)
------------------------------------------------------------------------------
May 2004 (current version 3.0, distributed with Trilinos 4.0)
* Update the ml user's guide (in the doc subdirectory)
* Improved interface to C++. Introduced two classes to use
ML as a preconditioner for Epetra_LinearProblem (both in the
ML_Epetra namespace): MultiLevelOperator and
MultiLevelPreconditioner.
* Basic feaures to solve Maxwell equations via MultiLevelPreconditioner
are now supported
* Improved interface to Amesos. Now ML can use Anasazi to estimate
the maximum eigenvalue for both symmetric and non-symmetric
problems.
* Improve interface to Anasazi
* Add power method to estimate lambda max for non-syymmetric
problems
* Improve the MLS preconditioner for block systems
* New options for configure. Drop --with-ml_epetra (and similar
for other Trilinos packages), and use --enable-epetra (and similar)
* Fixed various compiler warnings for LINUX (GNU and Intel),
SGI 64, DEC, west-cplant, ASCI Red
------------------------------------------------------------------------------
Apr 2004
* Change to support new Teuchos methods' name
* Change to support new Amesos methods' name
* Now support to ML native interface to SuperLU and SuperLUDIST
is deprecated in favor of Amesos interface
------------------------------------------------------------------------------
Jan 2004
* Add METIS and ParMETIS aggregation schemes
* Add Teuchos support
* Add interface to Amesos
* Add interface to IFPACK
* Add interface to Anasazi
* Add configure options --with-ml_epetra, --with-ml_parmetis2x,
--with-ml_parmetis3x, --with-ml_ifpack, --with-ml_anasazi
* Add two-level additive and hybrid domain decomposition
preconditioners (not officially supported)
------------------------------------------------------------------------------
Sep 2003 (version 2.0, distributed with Trilinos 3.1)
* ML has now been put "autotooled", i.e., autoconf and automake.
* Haim Waisman's GGB method has been implemented.
* Bill Cochran's cache optimizations for multiple RHSs have been
implemented.
* Bill also has added code so that ML will compile with c++ compilers.
------------------------------------------------------------------------------
ml2_1_0_0-branch-patch_1
* 1st patch for release ml2_1_0_0 (trilinos-release-3-01-branch).
* Bug fix for eigenvalue estimate that was causing Alegra
sinmag benchmark to fail.
* Added definitions for make macro CXX in architecture-dependent
makefiles in order for Alegra to be able to build epetra interfaces.
* Minor changes to block matrix (equivalent real form)
capabilities.
* Fixed compiler warnings.
------------------------------------------------------------------------------
mystring = ml2_1_0_0
* Many minor fixes to get the commuting relationship T_h P_n = P_e T_H
exactly right.
* Serial implementation of enriched prolongator.
* New norm for convergence of Maxwell solver. This norm measure
and scales properly the range space and null space components of
the (curl,curl) operator.
------------------------------------------------------------------------------
mystring = ml2_0_0_5
* Changes in aggregation post-processing to improve aggregate quality.
* Added wrappers so that ML can use Epetra objects & so that
AztecOO can use ML as a preconditioner.
* Added block Maxwell capabilities for real block systems that are
equivalent to the complex systems arising from Maxwell in the
frequency domain.
------------------------------------------------------------------------------
mystring = ml2_0_0_3
* Added in smoothing of Reitzinger/Schoeberl prolongator.
* Least squares and smoothed l.s. prolongators in serial for R/S AMG.
* Enhanced detection of degenerate coarse grid matrices in building
R/S AMG hierarchy.
------------------------------------------------------------------------------
mystring = ml2_0_0_2
* Symmetrize Hiptmair smoother and finished off MLS within Hiptmair.
------------------------------------------------------------------------------
ml2_0_0_1
* MLS polynomial smoother
* matrix does not need to be scaled.
* Added more print control:
- "ml_inputfile" now takes an additional argument in the general
- problem specs section:
- "Print Level = [an integer]"
- 0 => minimal output
- 1 => some output (default)
- 2,3,... => increasing amounts of output
------------------------------------------------------------------------------
ml2_0_0_0
* AMG for Maxwell's equations
* Reitzinger Schoeberl AMG.
* Hiptmair smoother.
