Ilya S. Steshin

Stu­dent, Depart­ment of Chem­istry, Lobachevsky State Uni­ver­si­ty of Nizh­ny Nov­gorod


Lab­o­ra­to­ry Assis­tant

Sci­en­tif­ic inter­ests:

Struc­ture and prop­er­ties of met­al clus­ters. Con­struc­tion of neur­al net­work poten­tials for met­al clus­ters.


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.

Комментарии запрещены.