Solving large numerical optimization problems in HPC with Python

Numerical optimization is a complex problem in which many different algorithms can be used. Distributed metaheuristics have received attention but they normally focus on small problems. Many large scientific problems can take advantage of these techniques to find optimal solutions for the problems. However, solving large scientific problems presents specific issues that traditional implementations of metaheuristics do not tackle. This research presents a large parallel optimization solver that uses Python to follow a generic model that can be easily extended with new algorithms. It also makes extensive use of NumPy for an efficient utilization of the computational resources and MPI4py for communication in HPC environments. The presented model has proven to be an excellent approach for solving very large problems in an efficient manner while using the computational resources in different HPC environments adequately.