MCS Algorithm

The world's fastest implementation of the Multilevel Coordinate Search (MCS) algorithm for derivative-free optimization, originally developed by Huyer and Neumaier. This Rust implementation achieves a 500x+ performance improvement over the original MATLAB version and existing Python implementations.

Performance & Architecture

Built in Rust with const generics for compile-time dimension checking, the implementation supports arbitrary N-dimensional optimization problems with zero runtime overhead for dimension validation. The algorithm efficiently searches for global minima of functions without requiring gradient information.

Testing & Reliability

The codebase includes 400+ tests covering correctness, edge cases, and performance benchmarks. Every optimization path is validated against known analytical solutions to ensure numerical accuracy.

Collaboration

Developed in collaboration between SBER CIB (Sberbank Corporate & Investment Banking) and HSE (Higher School of Economics) for production use in quantitative finance and scientific research.

Ecosystem

The project includes a family of related repositories that extend its reach beyond Rust:

• Rust_MCS_web — Docker-based browser GUI for no-code users
• Rust_MCS_Python — Python connector for seamless integration with existing Python code
• Rust_MCS_Cpp — C++ connector for integration with existing C++ code
• Rust_MCS_Sharpe_Ratio — Example application optimizing Sharpe Ratio for portfolio allocation

Key Highlights

• >500x speedup — over MATLAB and Python implementations
• Const generics — compile-time dimension safety and speedup
• N-dimensional — supports optimization in arbitrary dimensions
• 400+ tests — comprehensive test suite with benchmark validation
• Cross-language — connectors for Python, C++, and a browser-based GUI