Grouping Search Regions for Evolutionary Algorithm of Quantum Chemical Systems

Michael Groves, California State University-Fullerton

Usage Details

The intern will develop protocols which will enhance the success rate and speed of the evolutionary algorithm, a global optimisation search scheme, used to determine the best structure for molecules. The student will use existing machine learning algorithms to write protocols in Python to cluster starting populations into groups and then search exclusively within those groups. This strategy has two benefits: 1) It takes advantage of the evolutionary algorithm’s tendency to quickly evolve given close links in structures and 2) Puts a limit on the overall duration of the search. The nature of this strategy means that regions of the potential energy surface can be surveyed quickly and then left alone. Once all the initially created groups have been searched the GA search stops so that a new one, with a new starting population can be initialized. The intern will use quantum mechanics based calculations to relax trial structures and use clustering algorithms to organise existing structures into groups. Then, to search within groups, a covariance matrix adaptation evolution strategy will be used. This strategy will be tested on a variety of trial systems to determine their capabilities on a broad set of molecules. To quantify the effectiveness of global optimisation search schemes, hundreds of runs comprised of thousands of trial structures must be calculated to get the statistics necessary to quantify improvements with statistical accuracy to benchmark improvements.