ANALYSIS OF THE PHONON DENSITY OF STATES IN IRON SULFIDE MATERIALS USING A DEEP LEARNING APPROACH WITH EUCLIDEAN NEURAL NETWORKS

• Main Author: Dwi Fitriani, Nita
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/81532
• Institution: Institut Teknologi Bandung

The author uses deep learning with Euclidean Neural Networks to predict the phonon density of states (PhDOS) of hexagonal FeS material. The aim is to understand phonon dynamics and the influence of structural and compositional factors on the thermal and mechanical properties of the material. Testing with unit cell expansions from 1×1×1 to 4×4×4 demonstrated very low mean square error (MSE) values and consistent PhDOS peak predictions at 320 cm-1 . Vacancy defects modifications in FeS showed that increasing vacancies, whether close or distant, significantly changed the intensity and peak shifts, affecting phonon stability and thermal conductivity. Substitutional defects with tellurium (Te) lowered the phonon peak frequency to 180 cm-1 , while substitutional defects with magnesium (Mg) and silicon (Si) showed minimal changes. In contrast, antimony (Sb) and titanium (Ti) exhibited more moderate changes in MSE and enhanced thermal and mechanical stability. Substitutional defects with copper (Cu) maintained the peak at 320 cm-1 , indicating the stability of the main vibrational mode. This approach allows for accurate predictions and is crucial for developing materials with desired properties for various technological applications. This research opens new opportunities in material design and optimization for industrial needs, particularly in thermal conductivity, mechanical stability, and electron-phonon interactions.