HDSDP is a C library and command-line solver for sparse semidefinite programs.
It implements a homogeneous dual-scaling interior-point method. The sdpasolve
program reads SDPA .dat-s and MPS .mps input files. The repository also
contains example instances, a user manual in doc/, and the separate ADPCG
and SPEIGS components in derivative/.
Project page: https://github.com/COPT-Public/HDSDP Project manager and contact: Wenzhi Gao gwz@stanford.edu
For support, feature requests, and bug reports, open an issue at https://github.com/COPT-Public/HDSDP/issues. Include the input format, build platform, the command run, and the relevant solver output when reporting a bug. The COIN-OR code of conduct is at https://www.coin-or.org/code-of-conduct/.
The default build requires CMake 3.21 or later, a C compiler, and BLAS and
LAPACK libraries. On macOS, the system Accelerate framework can provide BLAS
and LAPACK. Optional Pardiso support requires Intel MKL and local configuration
of its library paths; it is disabled by default. See INSTALL for the tested
build, install, and run procedure and TESTING for the installation test.
The source bundle includes adaptations of SuiteSparse CSparse/CXSparse and LDL
(LGPL-2.1-or-later) and OSQP QDLDL (Apache-2.0). SPEIGS has its own MIT
license. These parts retain their own licenses; the top-level LICENSE records
the project's MIT notice. See THIRD_PARTY_NOTICES and the
license files under external/lics/ and derivative/SPEIGS/.
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build --parallel
ctest --test-dir build --output-on-failure
cmake --install build --prefix "$PWD/install"
./install/bin/sdpasolve examples/truss1.dat-s
The example should report SDP Status: Primal dual optimal. For the solver
API and other examples, see doc/hdsdp_doc.pdf and interface/hdsdp.h.
Wenzhi Gao, Dongdong Ge, and Yinyu Ye, “HDSDP: Software for Semidefinite Programming,” arXiv:2207.13862 (2022).