Twist angle, interlayer coupling, and the dielectric environment affect electron-hole interactions in moiré transition metal dichalcogenide heterojunctions, thereby shaping exciton properties such as binding energy, peak position, and line shape. As a result, isolating the individual contributions of these factors to the excitonic response remains difficult. To address this problem, we combine first-principles calculations with physics-aware machine learning. We use monolayer ${\mathrm{MoS}}_2$ as a reference system to test whether screening corrections can be reliably transferred to excitonic spectra and then investigate how the dielectric environment and interaction scale affect exciton properties in the ${\mathrm{MoS}}_2/{\mathrm{MoSe}}_2$ moiré heterojunction. In monolayer ${\mathrm{MoS}}_2$, the screening-related intermediate quantity is predicted with an ${\mathrm{R}}^2$ of 0.998. Under a low-fidelity Bethe-Salpeter equation setting, the deviation of the main exciton peak from the reference is reduced to 9 meV. Prediction time for a single spectrum is reduced from 9 h to 1.5 s, corresponding to a speedup of approximately $2.1\times {10}^4$. We then apply the framework to the ${\mathrm{MoS}}_2/{\mathrm{MoSe}}_2$ moiré heterojunction. The mean absolute error of exciton binding energy prediction is ${10}^{-3}$ eV. A physics-based hybrid model reduces this error by 35% compared with a purely physical power-law baseline. These results clarify the roles of dielectric screening, interlayer coupling, and moiré potential in exciton spectra and give guidance for parameter exploration and structural design in moiré optoelectronic devices.

Close

PrevNext

View figure in article

• Bethe-Salpeter equation
• Excitons
• Optical properties of clusters
• Dielectrics
• Heterostructures
• Moire materials
• Transition metal dichalcogenides
• Two-dimensional electron system
• First-principles calculations
• Machine learning