Data-Driven Reconstruction of Quasiparticle Band Dispersion in TaTe2 from ARPES Measurements

Tantalum telluride (TaTe2), a model two-dimensional (2D) semimetal, exhibits remarkable electronic properties that make it a potential candidate for energy applications. However, to enhance their electronic performance for successful adaptation such as electrocatalysts or supercapacitor, a deep understanding of their band structure is required. For this purpose, Density Functional Theory (DFT) calculations are often used, although discrepancies in fitting likely arise when compared with angle-resolved photoemission spectroscopy (ARPES) measurements. These errors are mainly manifested as the differences in quantifying the dispersiveness of bands and the relative spacing between bands. In this work, we employed a machine learning–based framework for reconstructing the electronic band dispersion of TaTe2 from ARPES data, using DFT-calculated bands as initialization. The framework employs probabilistic machine learning algorithms and optimization routines that iteratively refine quasiparticle dispersions. The model reconstructs multiple valence bands of TaTe2 and captures intricate momentum-dependent features not fully revealed by other conventional methods. This study demonstrates how integrating machine learning with theoretical and experimental approaches enables the successful reconstruction of electronic band structures, thereby offering new pathways for analysis of photoemission data.