\texttt{DR-DAQP}: An Hybrid Operator Splitting and Active-Set Solver for Affine Variational Inequalities
New hybrid solver combines operator splitting with active-set strategy to find exact solutions in finite time.
A team of researchers from academic institutions has released DR-DAQP, a high-performance solver for strongly monotone affine variational inequalities (AVIs). The solver innovatively merges first-order Douglas-Rachford operator splitting with a second-order active-set acceleration strategy. This hybrid approach allows the algorithm to estimate the active constraints during iterations and attempt a Newton-type correction. Crucially, when the active set is correctly identified, this step yields the exact solution to the AVI problem, overcoming the asymptotic convergence limitation typical of purely first-order methods. The implementation also leverages warm-starting and pre-factorization of key matrices to dramatically speed up individual iterations.
Numerical benchmarks demonstrate DR-DAQP's transformative performance. On randomly generated AVI problems, it was found to be up to two orders of magnitude (100x) faster than PATH, the current state-of-the-art solver. The impact is even more pronounced in applied domains: on a game-theoretic model predictive control (MPC) benchmark, DR-DAQP achieved solve times several orders of magnitude below those of the mixed-integer solver NashOpt. The researchers provide a high-performance C implementation with accessible interfaces for Julia, MATLAB, and Python, making this advanced optimization capability available to engineers and researchers working on complex systems in control, robotics, and multi-agent AI.
- Hybrid algorithm combines Douglas-Rachford splitting with active-set Newton steps for exact solutions.
- Proven to be up to 100x faster than PATH solver on standard AVI benchmarks.
- Provides open-source C code with Julia, MATLAB, and Python interfaces for immediate use.
Why It Matters
Drastically accelerates optimization for AI control systems, robotics, and game theory, enabling real-time solutions to previously intractable problems.