ACCPM - Analytic Center Cutting Plane Method

ACCPM is a package for solving large scale convex optimization problems. The code is an implementation of the cutting plane method. Instead of solving every relaxed master problem to optimality (as is the case in classical decomposition approaches), ACCPM looks for an analytic center of the current localization set. The projective algorithm is successfully applied for this purpose.

The code can be licensed for academic purposes. Check what to do to get ACCPM library.

The theoretical development of the ACCPM started from the paper:

and was continued in:

An early (Matlab) implementation of the method was written by Olivier Bahn and Olivier du Merle. Later on, a prototype C code was developed by Olivier du Merle and successfully applied to solve nontrivial convex optimization problems:

The success of this prototype code, justified the effort to develop a state of the art general purpose implementation of the method. Its new variant has been written in C++ by Jacek Gondzio, Olivier du Merle and Robert Sarkissian. The new code has already been applied to solve different large scale optimization problems, see, e.g.:

It is our intention to make the ACCPM library available upon request to any research application. The reader intereseted in such a use should contact Jean-Philippe Vial: The library libaccpm.a was released in January 1996.

The following papers address practical use of the ACCPM library:

The research on ACCPM is being continued. The most recent progress is documented in the following papers:

Parallel version of ACCPM has been developed. For portability reasons it uses MPI for parallel communication; it has been run on SP2 machine of IBM and on a cluster of 10 Linux PC's: