The DSDP software is a free open source implementation of an interior-point
method for semidefinite programming.
It provides primal and dual solutions,
exploits low-rank structure and sparsity in the data,
and has relatively low memory requirements for an interior-point method.
It allows feasible and infeasible starting points and provides approximate
certificates of infeasibility when no feasible solution exists.
The dual-scaling algorithm implemented in this package
has a convergence proof and worst-case
polynomial complexity under mild assumptions on the data.
The software can be used as a set of subroutines, through
Matlab, or by reading and writing to data files.
Furthermore, the solver offers scalable parallel
performance for large problems and a well documented interface.
Some of the most popular applications of semidefinite programming and
linear matrix inequalities (LMI) are model control, truss topology design,
and semidefinite relaxations of combinatorial and global optimization problems.
DSDP5 was released January, 2005. The interface
of the most recent version
is compatible to DSDP versions 5.0 and later, but it includes
new features and improvements in performance.
The package has been used in many applications and
tested for efficiency, robustness, and ease of use.
We welcome and encourage further use under the terms
of the license included in the distribution.
View independent benchmarks of DSDP and other solvers on
SDPLIB, the
DIMACS problem set, and some
large SDPs
to measure its robustness, speed, and efficiency.
Download:
Source code, precompiled binaries, and Matlab utilities can be
downloaded free of charge.
Full source code
(DSDP5.8.zip,
.tar.gz),
written in C, contains a subroutine library, utilities,
a Matlab mex interface, documentation, and examples.
Precompiled DSDP executables using
Linux (GCC-3.2) on a Pentium 4
and statically linked to Atlas BLAS.
Examples and documentation are included.
Precompiled
dsdp.dll
for Matlab release 14 (version 7.1) on Microsoft Windows.
Examples and documentation are included.
Source code for previous version DSDP 2.1 is available, but not supported. This version is specialized for rank-one constraints and performs
very well on relaxations of the bisection-cut and other graph problems.