PCx is an interior-point predictor-corrector linear programming package. The code has been developed at the Optimization Technology Center, which is a collaboration between Argonne National Laboratory and Northwestern University. PCx has nothing to do with the .pcx file format; it cannot be used to open files in this format.
The current release is 1.1. Numerous small changes have been made since the last major release on 6/10/98.
Version 1.1 contains no new algorithmic features; it differs from version 1.0 in that it provides hooks in the code that enable alternative linear solvers to be provided by the user.
PCx is designed as a stand-alone solver. Because of its modular structure and fairly transparent data structures, it is not too difficult to integrate into your application.Together with some of our users, we have recently investigated new features such as a callable library and a MATLAB interface. Users who are interested in contributing other enhancements to PCx are encouraged to contact the authors.
PCx is freely available, but is not public domain software. Commercial users should obtain a license. For conditions of use of PCx, see the copyright statement.
To download PCx or its auxiliary tools (such as the AMPL interface and PCxGUI), go to the PCx Download Page (You must first fill out the request form for PCx to get a password.)
The main distribution file contains the full Unix distribution, containing source code, Makefiles, User Guide, benchmarks, and copyright statement. Executable files for the following Unix architectures are also available:
A sample specifications file and some test problems are included in the source distribution. Sample problems can be executed by typing
PCx afiro
or
PCx 25fv47
An AMPL interface to PCx and a Java-coded graphical interface called PCxGUI are now available. These can also be obtained from the PCx Download Page. (Again, you need to fill out a request form first for your password.)
The Download Page also contains
A parallel implementation of PCx was prepared by some collaborators at Cornell. This version is based on release 1.0 of the Unix version.
Users with multiprocessor shared-memory IBM RS6000 platforms can use the WSSMP version of PCx; parallelism can be achieved by simply linking to the shared-memory version of WSSMP.
Postscript, dvi and pdf versions of the User Guide are available.
Here's a NEOS web page that explains MPS format, the incredibly ugly but standard input format for LPs.
See here for some results obtained with the WSSMP and Ng-Peyton versions of PCx. Note that the nested-dissection algorithm as implemented in WSSMP is often much faster on the larger problems. Hans Mittelmann has compiled some comparative benchmarks in which various interior-point codes are compared. He reports mostly large problems that show up differences between the codes.
See our log of changes to PCx since inception, including
new release announcements, details of the latest bug fixes and new executables, and so on.
PCx uses the sparse Cholesky linear algebra routines of Esmond Ng and Barry Peyton.
Marc Wenzel programmed the dense-column-handling and conjugate gradient refinement features that were added for the beta-2.0 release of PCx. Doug Moore gave valuable advice and in particular pointed out and repaired many memory leaks in the beta-1.0 release. Hans Mittelmann prepared the executables for numerous architectures and ran many of the benchmark tests. Jean-Pierre Goux set up the request form and download system.
We welcome suggestions and comments on PCx.
[ PCx Home Page | Optimization Technology Center | NEOS Server | NEOS Guide | NEOS Tools ]