Release notes
Notes for GCG Release 3.0.0

General
================
We are proud to release GCG in version 3.0.0. The major version number
increase from 2.1.4 is justified, as we include major new features. Check
both the feature list as well as the CHANGELOG for the changes

Features
================

 * New detection scheme: Instead of calling a set of detectors in a 
   static fashion, now each detector is allowed to only fix a subest
   of constraints and variables. In the following these partial 
   decompositions can refined by other detectors. Detectors may 
   implement up to three callbacks: propagate, finish, and 
   postprocess. Constraint- and variable classification yield further
   problem knowledge resulting in decompositions for large problems
   for which no structure was known before. Furthermore there are 
   several new ways of user interaction (e.g., 'explore', 
   and 'decomposition_toolbox') and visualization.

 * New pricing scheme: The pricing scheme has been restructured. The
   pricing loop is now managed by a priority queue, which holds
   so-called "pricing jobs" (i.e., pricing problems that are to be
   solved with a certain solver, heuristically or exactly).
   In particular, the new pricing scheme now allows heuristic MIP
   pricing to be performed in multiple rounds, with increasing limits.
   The pricing scheme furthermore offers several possibilities to limit
   the pricing loop: These include a maximum number of columns to be
   generated, separately for each problem as well as for the overall
   loop, and a maximum percentage of pricing problems for which
   improving columns need to be found.

 * Benders' decomposition mode: The available decomposition algorithms has been
   extended to include Benders' decomposition. The Benders' decomposition mode
   uses the problem structure found through the detection and forms a master
   problem with a series of subproblems. The Benders' decomposition framework
   included in SCIP 6.0.0 is used to manage the subproblem solving and Benders'
   cut generation.
   The Benders' decomposition mode is activated by setting "relaxing/gcg/mode"
   to 1.
   Detection settings are provided to form different decompositions suitable for
   Benders' decomposition. These settings are: detect-benders,
   detect-benders-bin_master and detect-benders-cont_subpr. These settings use a
   new experimental scoring method that has been added for Benders'
   decomposition.

 * Other features: We revisited and improved several other parts of the
   code such as an improved dual variable stabilization, improved
   initialization, and aggregation of programs with continuous variables.
Notes for GCG Release 2.1.4

General
================
We are proud to release GCG in version 2.1.4. This is a bugfix
release, see the CHANGELOG for the changes.
Notes for GCG Release 2.1.3

General
================
We are proud to release GCG in version 2.1.3. This is a bugfix
release, see the CHANGELOG for the changes.
Notes for GCG Release 2.1.2

General
================
We are proud to release GCG in version 2.1.2. This is a bugfix
release, see the CHANGELOG for the changes.
Notes for GCG Release 2.1.1

General
================
We are proud to release GCG in version 2.1.1. This is a bugfix
release, see the CHANGELOG for the changes.
Notes for GCG Release 2.1.0

General
================

We are proud to release GCG in version 2.1.0. Check the CHANGELOG for
the changes.
Notes for GCG Release 2.0.1

General
================
We are proud to release GCG in version 2.0.1. This is a bugfix
release, see the CHANGELOG for the changes.
Notes for GCG Release 2.0.0

General
================
We are proud to release GCG in version 2.0.0. The major version number
increase from 1.1.1 is justified, as we include major new features. Check
both the feature list as well as the CHANGELOG for the changes

Features
================

 * parallelization of the pricing process: compile GCG with OPENMP=true and
   benefit from concurrently solving the pricing problems (you need to compile
   SCIP with PARASCIP=true) greatly reducing the time to solve difficult
   problems to optimality.

 * Generic branching: We can branch on arbitrary, bounded instances with
   discretization with and without aggregation. That enables us to solve
   cutting stock type instances much faster.

 * New detectors: We can now detect staircase structures as well as identical
   pricing problems that can be aggregated, thus removing some of the
   symmetry. Check out the staircase and isomorph detectors!

 * New heuristics: We can find even better solutions using a variety of diving
   and other heuristics in both the master and original space. See if we can
   find good solutions to your problem.
Notes for GCG Release 1.1.1

Features
========

- GCG can now automatically create a decomposition to solve the LP with
  empty pricing problems. See the cons/decomp/createbasicdecomp parameter


Makefile
========

- the version number is now part of the binary and library name
- fixed some Windows compiling issues
This is the first release of GCG. It may contain bugs. Have a lot of fun!

More information about this release can be found in the documentation, the
README file and the INSTALL file.