Programme

O.12 -- Oral, VCTP-48

Date: Tuesday, 1 August 2023

Time: 17:05 - 17:30

Machine-Learning-Based Prediction of Material Properties from Energy Spectra of Magnetoexcitons in Monolayer Transition Metal Dichalcogenides

Duong-D. Hoang-Trong1, Khang Tran2, Ngoc-Loan Phan1, Duy-Nhat Ly1, and Van-Hoang Le1

1 Ho Chi Minh City University of Education, Ho Chi Minh City, Vietnam 2 New Jersey Institute of Technology, Newark, NJ 07102, USA

Extracting material properties such as the exciton-reduced mass µ and 2D static polarizability χ2D of monolayer transition metal dichalcogenides (TMDCs) from measured exciton energy spectra is essential because these parameters cannot be retrieved directly from experiments [1, 2]. Recently, a method based on the fitting procedure was proposed to recover µ and χ2D from s-state exciton energies of WSe2 and WS2 [3]. This method is promising for applying to other monolayer TMDCs; however, a huge amount of data needs to be analyzed. Besides, each TMDC semiconduction material necessitates a unique set of parameters in the fitting procedure while repeating the same process, which is time-consuming and expensive. Nevertheless, recent advanced artificial models with machine learning (ML) capabilities are emerging to help solve similar problems [4]. We report some achievements in applying ML to build a simple model for retrieval of material parameters (µ and χ2D) of any monolayer TMDC from experimental exciton energies of 1s, 2s, 3s, and 4s states. To do it, we have trained the ML model with more than 300 thousand theoretical data sets of exciton energies calculated by the FORTRAN codes constructed by us previously. A prediction is carried out for WSe2 with the experimental energy data in Ref. [5]. The results demonstrate a very high accuracy of our ML model. First, the predicted parameters are highly compatible with the currently available data. Second, the theoretical energies calculated by the Keldysh model potential with these material parameters coincide well with the experimental data (with errors less than 1% in most cases). Therefore, the constructed ML model are applicable for more other TMDC materials. [1] A.V. Stier, et al., Magnetooptics of exciton Rydberg states in a monolayer semiconductor, Phys. Rev. Lett. 120 (2018) 057405. [2] M. Goryca, et al., Revealing exciton masses and dielectric properties of monolayer semiconductors with high magnetic fields, Nature Comm. 10 (2019) 4172. [3] Duy-Nhat Ly, et al., Retrieval of material properties of monolayer transition metal dichalcogenides from magnetoexciton energy spectra, Phys. Rev. B 107 (2023) 205304. [4] A. M. R. Gherman, K. Kov ́acs, M. V. Cristea, V. Tosa, Artificial neural network trained to predict high-harmonic flux, Applied Sciences 8 (2018) 2106. [5] E. Liu, et al., Magnetophotoluminescence of exciton Rydberg states in monolayer WSe2, Phys. Rev. B 99 (2019).

Presenter: Hoàng Trọng Đại Dương