478171 Genetic Algorithm Lead Discovery of Nanoparticle Structures

Monday, November 14, 2016
Grand Ballroom B (Hilton San Francisco Union Square)
Emma Overly1,2, Eliad Peretz3, Douglas Nevers3 and Tobias Hanrath4, (1)Bucknell University, Lewisburg, PA, (2)Cornell Center for Materials Research, Cornell University, Ithaca, NY, (3)Cornell University, Ithaca, NY, (4)Robert Frederick Smith School of Chemical and Biomolecular Engineering, Cornell University, Ithaca, NY

Nanoparticles are at the forefront of light and semiconductor technologies. In order to take advantage of a nanoparticle's tunable optic and electronic properties, it is important to know the 3-D structure of the nanoparticle. For example, nanoparticles of different sizes and shapes will emit varying wavelengths of light. But, because of nanoparticles' heterogenous structures, typical X-ray diffraction structure determination techniques are not able to represent the entire structure. Researchers have begun to use computational models to bridge the gap between the incomplete picture the experimental data presents about nanoparticles. These computational models do have challenges; the search space for all possible locations of each atom in the nanoparticle is on the order of 10139 – 10297 permutations, assuming a nanoparticle comprised of 100 atoms. This vast search space suggests the need for an intelligent algorithm to narrow the structure possiblities. A genetic algorithm was created to predict the 3-D structure of the nanoparticle. Genetic algorithms mimic the natural selection process in evolution. An initial population of candidates is created, and through many generations, the candidates are altered and crossed-over in order to produce a population of candidates with high fitness, or rather the best qualities. The genetic algorithm begins with bulk crystal structure predictions and alters the structure candidates through mutations and crossovers to fit Pair Distribution Function (PDF) X-ray diffraction data of a given nanoparticle. The algorithm guides each generation of candidates to have improved fits to the experimental PDF data. The dynamic computational model is used to create an advanced structural model of an experimentally observed Cadmium-Selenide nanoparticle. The algorithm is able to predict nanoparticle structures with low residuals of 0.18, meaning the model prediction is closely fitting the experimental data. With this genetic algorithm, the search space of possible structure predictions can be greatly reduced, allowing for intelligent and valuable structure guesses to be made. Modeling nanoparticle structures will lead to a deeper comprehension of structure-property relationships in nanoparticle technologies.

Extended Abstract: File Not Uploaded
See more of this Session: Undergraduate Student Poster Session: Computing and Process Control
See more of this Group/Topical: Student Poster Sessions