Large scale Raman spectrum calculations in defective 2D materials using deep learning.

Area of Science:

• Computational Materials Science
• Solid-State Physics
• Spectroscopy

Background:

• The Raman response of 2D materials is sensitive to structural defects.
• Accurate simulation of large-scale defect structures is computationally challenging.
• Understanding defect impacts is crucial for tailoring material properties.

Purpose of the Study:

• To develop a scalable machine learning (ML) workflow for predicting the Raman response of 2D materials.
• To investigate the influence of defects on the Raman spectra of materials like graphene and hexagonal boron nitride.
• To validate the ML predictions against experimental Raman spectroscopy data.

Main Methods:

• Implementation of machine-learned interatomic potentials for accurate atomic interactions.
• Utilizing the Raman-active Γ-weighted density of states (DOS) method.
• Employing a patch-based configuration splitting approach to enable large-scale simulations (tens of thousands of atoms).

Main Results:

• The ML workflow successfully simulated large systems, with diagonalization identified as the primary computational bottleneck.
• Predictions for isotopic graphene and defective hexagonal boron nitride showed good agreement with experimental Raman response data.
• The study demonstrated the feasibility of using ML for defect impact analysis on Raman spectra.

Conclusions:

• The developed ML workflow provides an efficient approach to study defect effects on 2D material Raman spectra.
• This methodology significantly advances the scale of simulations possible for such investigations.
• The approach holds promise for future research in solid-state physics and materials science, particularly in understanding defect-related phenomena.