Local-MIP

Local Search Solver for Mixed Integer Programming

Quick Start Software Source Code Tutorials Examples

Project Leader Shaowei Cai Developer Peng Lin

🏆 CP 2024 Best Paper Award Peng Lin, Mengchuan Zou, Shaowei Cai 📄 Artificial Intelligence 2025 Peng Lin, Shaowei Cai, Mengchuan Zou, Jinkun Lin

Customizable Modules

Initialization Restart Strategies Neighbor Operators Weighting Schemes Scoring Functions

Highlights

Rapid Optimization

Local-MIP delivers top-tier performance among open-source solvers on the general MIPLIB benchmark, capable of finding high-quality solutions within seconds.

Highly Customizable

Lightweight CLI plus C++/Python APIs and callback hooks let you customize starts, restarts, weights, scoring, and neighborhood behavior.

Seamless Integration

Zero third-party dependencies. Local-MIP natively supports standard MPS and LP formats and can be easily integrated into other solvers via C++ or Python interfaces.

Quick Start

Start from the project root directory (after extracting or cloning the repository).

Software: Grab the latest release from the software page.

Build: Compile using the provided script:

 ./build.sh release
 # or build everything
 ./build.sh all

Run: Solve your first instance:

 cd build
 ./Local-MIP --model_file ../test-set/2club200v15p5scn.mps --time_limit 300

Full Quick Start →

Software using Local-MIP

Local-MIP is integrated into several state-of-the-art optimization tools:

1. NVIDIA cuOpt

NVIDIA cuOpt is a GPU-accelerated solver.

Official Site: NVIDIA cuOpt Product Page
Reference: For technical details, refer to the preprint arXiv:2510.20499.

2. Exact

Exact is a leading Pseudo-Boolean (PB) solver.

Source Code: Exact on GitLab
Reference: See the solver description in Exact_PBComp_2025.pdf.

Documentation & Resources

Getting Started

Quick Start — Get up and running in 3 steps: download, build, and solve.
Software — Access the latest source code and release versions.

Technical Documentation

Tutorials — Comprehensive guides and technical references.
- Command-Line Usage — Installation, CLI usage, and library integration.
- Modeling API — Build MIP models via C++/Python API (no .mps/.lp required).
- Parameters Reference — A complete list of command-line flags and options.
- Callback System — Detailed guide on customizing solver behavior.
Examples — Practical code examples and callback implementation walkthroughs.

Performance & Research

MIPLIB Records — New best-found solutions for the MIPLIB open instances.
Papers — Academic publications and research background.

Latest Release: v2.1

License: MIT License