LaGO (Lagrangian Global Optimizer) is a MINLP solver developed in the DFG-project "Optimization of a Complex Energy Conversion Plant".
Using the default parameter settings, LaGO is an implementation of a Branch-and-Cut solver for block-separable not necessarily convex MINLPs. This algorithm is illustrated in the recent paper "LaGO - a (heuristic) Branch and Cut algorithm for nonconvex MINLPs" and in the slides of the talk "LaGO - Branch and Cut for nonconvex block-separable MINLPs in the absence of algebraic formulations".
The current version of LaGO is still in a very early state. Integer variables other than binaries (i.e., discrete variables which are not restricted to be 0 or 1) are not supported yet. Nevertheless, testing LaGO on moderate size examples from the GAMS MINLP World shows that LaGO is able to find good solutions for many real-world problems (see the benchmarking results at the end of this page).

The development of LaGO continues in form of a new COIN-OR project. For latest news about LaGO, and information on how to obtain, install, and use the LaGO code, please refer to the webpage

Please refer to Ivo Nowak or Stefan Vigerske for any further questions or remarks.


We used examples from the MINLPLib and GlobalLib available from GAMS World for testing.
Benchmarks were made in comparision with BARON 7.8.1 on a Pentium IV 3GHz with 1 GB RAM and Linux 2.6.11. NLPs were solved with CONOPT, LPs were solved with CPLEX 10.0.
We considered 3 groups of models from the MINLPLib and GlobalLib, each model having at most 1000 variables and no functions of the type sin, cos, or errorf: