Student, Department of Chemistry, Lobachevsky State University of Nizhny Novgorod
Position:
Laboratory Assistant
Scientific interests:
Structure and properties of metal clusters. Construction of neural network potentials for metal clusters.
Languages:
Russian, English(Intermediate)
Key Ideas:
Metal clusters, even consisting of one element, have a large number of isomers. The number of isomers rapidly increases with increasing nuclearity. Therefore, the use of quantum chemical calculations becomes difficult even with a small number of atoms.
One of the ideas for a qualitative calculation of the structure and properties of metal clusters was the proposal to use neural network (NN) potentials.
Flexible interatomic potentials for calculating the structure and energy of atomic clusters can be developed based on a NN. However, the architecture of the NN and the methods of choosing descriptors for describing the structure continue to be a question that needs to be studied.
The choice of descriptors and architecture of the neural network affects:
- learning rates
- quality of predictions
- the possibility of using a neural network for a different number of atoms in the system
Comparison of the average binding energy per atom for Mg30 clusters calculated by the DFT method with HDNNP predictions.