Presentation Title

Improving the Genetic Algorithm on the Atomic Simulation Environment (ASE) Through Intelligent Starting Population Creation and Clustering

Faculty Mentor

Dr. Michael Groves

Start Date

18-11-2017 9:30 AM

End Date

18-11-2017 9:45 AM

Location

9-285

Session

Physical Sciences 2

Type of Presentation

Oral Talk

Subject Area

physical_mathematical_sciences

Abstract

A chemical genetic algorithm (GA) provides computational chemists a means to find the lowest energy conformation molecule of a given stoichiometry. Our goal is to improve how the ASE GA creates its starting population, leading to faster, more efficient convergence to the global minimum energy molecule. Our strategy was to alter the population creator to form molecules based on the possible hybridized orbitals of each atom. We will also be testing the effectiveness of a clustering algorithm which uses intermolecular distances to differentiate molecular structures. Selecting molecules with varied structures provides increased variety to the starting population which should lead to faster convergence. C9H7N was used as the molecular stoichiometry because the global minimum is known (quinoline), the potential energy surface is well explored, and because it will test the algorithms ability to form rings.

Summary of research results to be presented

The project is showing strong positive results. The success rate of the genetic algorithm using my starting population creation method is nearly double that of the original going from 35% to 64% and that number is likely to be even higher by November. These results are compiled without any mutation step and success is defined by whether or not the global minimum molecule (quinoline) is found within 5000 iterations of the genetic algorithm.

This document is currently not available here.

Share

COinS
 
Nov 18th, 9:30 AM Nov 18th, 9:45 AM

Improving the Genetic Algorithm on the Atomic Simulation Environment (ASE) Through Intelligent Starting Population Creation and Clustering

9-285

A chemical genetic algorithm (GA) provides computational chemists a means to find the lowest energy conformation molecule of a given stoichiometry. Our goal is to improve how the ASE GA creates its starting population, leading to faster, more efficient convergence to the global minimum energy molecule. Our strategy was to alter the population creator to form molecules based on the possible hybridized orbitals of each atom. We will also be testing the effectiveness of a clustering algorithm which uses intermolecular distances to differentiate molecular structures. Selecting molecules with varied structures provides increased variety to the starting population which should lead to faster convergence. C9H7N was used as the molecular stoichiometry because the global minimum is known (quinoline), the potential energy surface is well explored, and because it will test the algorithms ability to form rings.