Local Environments and Phonon Spectra in Twisted Bilayer Graphene: A Machine- Learning Perspective

İçkecan D., Okyayli Y. E., Bleda E. A., Erbahar D.

Carbon 2025, Saint-Lo, Fransa, 29 Haziran - 04 Temmuz 2025, ss.1-713, (Özet Bildiri)

  • Yayın Türü: Bildiri / Özet Bildiri
  • Basıldığı Şehir: Saint-Lo
  • Basıldığı Ülke: Fransa
  • Sayfa Sayıları: ss.1-713
  • İstanbul Ticaret Üniversitesi Adresli: Evet

Özet

The number of atomic environments that arise in twisted bilayer graphene superlattices is not analytically addressed, even though Bernal stacked bilayer graphene has two different atom types in its lattice. In this study, we used descriptor functions to computationally analyze 140 twisted bilayer superlattices to investigate the emergent local environments. We find that the number of atoms having distinct local surroundings is linearly related to the size of the superlattice. Furthermore, according to their respective space groups, this linear dependence appears on two different lines. A new classification scheme based on the local environment is automatically suggested by unique local properties of lattice structures that arise on these lines. As a potential application, the use of local environments in studying vibrational properties is presented and discussed in light of the body of current work. The phonon density of states of the 140 structures and the local phonon density of states of their individual atoms are computed using molecular dynamics simulations. It is shown that the local density of state contributions from atoms with the same local environment are similar. A machine learning model is then trained using the local phonon density of states of the atoms with distinct local surroundings. The phonon spectra of twisted bilayer structures are predicted using this approach. It is demonstrated that the model works well for forecasting the vibrational characteristics of any twisted bilayer structure. Since the general technique described goes well beyond twisted bilayer graphene, potential uses are also discussed, including non-periodic structures and strain-induced moiré superlattices.

 

This work is based upon work from COST Action CA21126 - Carbon molecular nanostructures in space (NanoSpace), supported by COST (European Cooperation in Science and Technology).