The paper reports on the development of two types of interatomic interaction potentials for the atomistic modelling of carbon nanomaterial complexes with lithium. The first potential is constructed using the Gaussian approximation method and the second using the deep learning approach. These potentials have been trained using the results of density functional modelling and provide an accuracy close to that of this method with significantly lower computational requirements. The datasets contained more than 8000 structures of about 100 atoms. A given structure was placed in the training set with 90 % probability, otherwise it was placed in the validation set. The resulting potentials allow accurate reproduction of the energies of the complexes and the forces acting on the atoms. The computation time increases linearly with the number of atoms in the model and can vary by several orders of magnitude depending on the type of potential and the hardware used. The potential obtained by the deep learning method seems promising for realistic and accurate modelling of lithium on the surface of carbon nanotubes and various graphene-like structures at temperatures up to 450 K. Keywords: molecular dynamics, deep learning, regression, lithium, carbon nanomaterials.

Подсчитывается количество просмотров абстрактов ("html" на диаграммах) и полных версий статей ("pdf"). Просмотры с одинаковых IP-адресов засчитываются, если происходят с интервалом не менее 2-х часов.

Дата начала обработки статистических данных - 27 января 2016 г.