Combining Deep Learning Neural Networks with Genetic Algorithms to Map Nanocluster Configuration Spaces with Quantum Accuracy at Low Computational Cost

The configuration spaces for bimetallic AuPd nanoclustersof varioussizes are explored efficiently and analyzed accurately by combininggenetic algorithms with neural networks trained on density functionaltheory. The methodology demonstrated herein provides an optimizablesolution to the problem of searching vast configuration spaces withquantum accuracy in a way that is computationally practical. We implementa machine learning algorithm which learns the density functional theorypotential with increasing performance while simultaneously generatingand relaxing structures within the system's global configurationspace unbiasedly. As a result, the algorithm naturally converges ontothe system's energy minima while mapping the configurationspace as a function of energy. The algorithm's simple designapplies not only to nanocluster configurations, as demonstrated, butto bulk, substrate, and adsorption sites as well, and it is designedto scale. To demonstrate its computational efficiency, we work withAuPd nanoclusters of sizes 15, 20, and 25 atoms. Results focus primarilyon evaluating the algorithm's performance; however, severalphysical insights into possible configurations for these nanoclustersnaturally emerge as well, such as geometric Au surface segregationand stoichiometric Au minimization as a function of stability.