[2208.07612] Fast Discovery of Graphene Nanocrystals Utilizing DFT and Bayesian Optimization with Neural Community Kernel

View a PDF of the paper titled Fast Discovery of Graphene Nanocrystals Utilizing DFT and Bayesian Optimization with Neural Community Kernel, by c{S}ener “Oz”onder and H. Ok”ubra Ok”uc{c}”ukkartal

View PDF
HTML (experimental)

Summary:Density practical concept (DFT) is a robust computational methodology used to acquire bodily and chemical properties of supplies. Within the supplies discovery framework, it’s usually essential to just about display a big and high-dimensional chemical area to seek out supplies with desired properties. Nonetheless, grid looking out a big chemical area with DFT is inefficient as a result of its excessive computational price. We suggest an strategy using Bayesian optimization (BO) with a synthetic neural community kernel to allow sensible search. This methodology leverages the BO algorithm, the place the neural community, educated on a restricted variety of DFT outcomes, determines essentially the most promising areas of the chemical area to discover in subsequent iterations. This strategy goals to find supplies with goal properties whereas minimizing the variety of DFT calculations required. To show the effectiveness of this methodology, we investigated 63 doped graphene quantum dots (GQDs) with sizes starting from 1 to 2 nm to seek out the construction with the best gentle absorbance. Utilizing time-dependent DFT (TDDFT) solely 12 occasions, we achieved a major discount in computational price, roughly 20% of what could be required for a full grid search, by using the BO algorithm with a neural community kernel. Contemplating that TDDFT calculations for a single GQD require about half a day of wall time on high-performance computing nodes, this discount is substantial. Our strategy could be generalized to the invention of recent medicine, chemical compounds, crystals, and alloys with high-dimensional and huge chemical areas, providing a scalable resolution for varied functions in supplies science.