SMP Repository Description

We assemble a repository for sparse QP problems of different scales, most of which come from applications in animation, geometry processing, and simulation. Existing QP problem benchmarks are not large enough to stress-test large-scale QP solvers. To address this shortcoming, we gathered existing strictly-convex QP benchmark problems and also added a set of new QP problem instances mostly arising in computer graphics applications.

Our repository includes QP instances from shape deformation, contact simulation, model reconstruction, and cloth simulation from computer graphics; model predictive control (MPC) from robotics; and strictly-convex QP problems from the Maros-Mészáros repository. The number of variables ranges from 50–114309, and the number of constraints ranges from 20–10k.

We have also created a code repository, including the setup for running all other QP solvers that NASOQ is compared to so far. The code is available from here.

The instructions on how to work with SMP format and how to set the benchmark repository is available here and also in the github repo.

Our repository is growing. Share your most difficult mathematical programs with us through the submission form!

The table below shows a detailed description of different groups of QP problems in the SMP repository: