GLPK 3.0.2 (release date: Sep 24, 2001)

        A new implementation of the basis maintaining routines was
        included in the package. These routines, which are based on so
        called FHV-factorization (a variety of LU-factorization) of the
        basis matrix and Gustavson's data structures, allows performing
        the main operations faster at the expense of some worsening
        numerical accuracy.

        AFI (Advanced Form of the Inverse), which is the form of the
        basis matrix based on FHV-factorization, is available via the
        parameter form = 3 (on API level) or via the option --afi (in
        GLPSOL solver).

GLPK 3.0.1 (release date: Aug 01, 2001)

        Old GLPK API routines have been removed from the package.

        New GLPK API routines were added:

        - scaling routines;

        - a routine for writing problem data in MPS format;

        - a comprehensive driver to the simplex method;

        - basis maintaining routines.

        A description of the new API routines is given in the document
        "Additional GLPK API Routines". This document is included into
        the distribution in plain text format (see the file 'newapi.txt'
        in the subdirectory 'doc').

        Now the distribution includes a non-trivial example of using
        GLPK as a base LP solver for Concorde, a well known program that
        solves Traveling Salesman Problem (TSP). For further details see
        comments in the file 'sample/lpglpk30.c'.

GLPK 3.0 (release date: Jul 19, 2001)

        Now GLPK is provided with new API, which being more flexible
        can be used in more complex algorithmic schemes.

        New edition of the document "GLPK User's Guide" is included in
        the distribution. Now it completely corresponds to the new GLPK
        API routines.

        Old API routines are not removed yet from the package, however
        they became obsolete and therefore should not be used. Since now
        the header glpk.h corresponds to new API, in order to compile
        existing programs that use old GLPK API routines the statement

        #define GLP_OLD_API

        should be inserted before the statement

        #include "glpk.h"

GLPK 2.4.1 (release date: Jun 14, 2001)

        The document "Modeling language GLPK/L" is included into the
        distribution in texinfo format.

        New edition of the document "GLPK User's Guide" is included in
        the distribution. Now it describes all additional API routines
        which were recently added to the package.

GLPK 2.4 (release date: May 10, 2001)

        Now GLPK includes an implementation of a preliminary version
        of the GLPK/L modeling language. This language is intended for
        writing mathematcal programming models. The name GLPK/L is
        derived from GNU Linear Programming Kit Language.

        A brief description of the GLPK/L language is given in the
        document "GLPK/L Modeling Language: A Brief Description". This
        document is included into the distribution in plain text format
        (see the file 'language.txt' in the subdirectory 'doc').

        The language processor (which is a program that analyzes model
        description written in GLPK/L and translates it to internal data
        structures) is available as the GLPK API routine.

        The stand-alone solver GLPSOL now is able: a) to process model
        descriptions written in the GLPK/L language; b) to solve pure LP
        problems using the interior point method (therefore the program
        GLPIPM was removed from the package).

GLPK 2.3 (release date: Apr 09, 2001)

        New edition of the document "GLPK User's Guide" is included in
        the distribution. Now it describes all additional API routines
        which were recently added to the package.

        The MIP solver was fully re-programmed in order to improve its
        robustness and performance. In particular, a basis recovering
        procedure was implemented (this procedure allows switching to
        the primal simplex method in case when the dual simplex method
        fails).

GLPK 2.2 (release date: Mar 15, 2001)

        Now GLPK includes a tentative implementation of the
        branch-and-bound procedure based on the dual simplex method for
        mixed integer linear programming (MIP).

        Complete description of this new feature of the package is given
        in the preliminary document "Mixed Integer Linear Programming
        Using GLPK Version 2.2 (Supplement to GLPK User's Guide)". This
        document is included into the distribution in plain text format
        (see the file 'mip.txt' in the subdirectory 'doc').

        The MIP solver (glp_integer) can be used as GLPK API routine in
        the same way as the pure LP solver (glp_simplex).

        The stand-alone program 'glpsol' is now able to solve LP as well
        as MIP problems.

        Note that the current version of GLPK MIP solver is based on
        easiest heuristics for branching and backtrackng. Therefore the
        solver is fit mainly for MIP problems which are not very hard
        and have few integer variables.

GLPK 2.1 (release date: Feb 19, 2001)

        The document "GLPK Implementation of the Revised Simplex Method"
        is included into the distribution. This document describes most
        of routines related to the revised simplex method.

GLPK 2.0 (release date: Jan 25, 2001)

        Now GLPK includes a tentative implementation of the primal-dual
        interior point method for large-scale linear programming.

        The interior point solver can be used as GLPK API routine in the
        same manner as the simplex method solver (glp_simplex):

        ret = glp_interior();

        Note that currently the interior point solver implemented in
        GLPK doesn't include many important features, in particular:

        * it can't process dense columns; therefore if the problem has
          dense columns, the solving will be extremely inefficient;

        * it has no special features against numerical unstability;
          some problems may cause premature termination of the solving
          when the matrix A*D*A' becomes ill-conditioned;

        * it computes only values of primal (auxiliary and structural)
          variables and doesn't compute values of dual variables (i.e.
          reduced costs) which are just set to zero;

        * it doesn't identify optimal basis corresponding to the found
          interior point solution; all variables in the found solution
          are just marked as basic variables.

        GLPK also includes a stand-alone program 'glpipm' which is a
        demo based on the interior point method. It may be used in the
        same way as the program 'glpsol' that is based on the simplex
        method.
