Prediction of the Ground-State Electronic Structure from Core-Loss Spectra of Organic Molecules by Machine Learning

Abstract

The core-loss spectrum reflects the partial density of states (PDOS) of the unoccupied states at the excited state and is a powerful analytical technique to investigate local atomic and electronic structures of materials. However, various molecular properties governed by the ground-state electronic structure of the occupied orbital cannot be directly obtained from the core-loss spectra. Here, we constructed a machine learning model to predict the groundstate carbon s-and p-orbital PDOS in both occupied and unoccupied states from the C K-edge spectra. We also attempted an extrapolation prediction of the PDOS of larger molecules using a model trained by smaller molecules and found that the extrapolation prediction performance can be improved by excluding tiny molecules. Besides, we found that using smoothing preprocess and training by specific noise data can improve the PDOS prediction for noisecontained spectra, which pave a way for the application of the prediction model to the experimental data.

Toward the Atomic-Level Analysis of Ground-State Electronic Structures of Solid Materials via Prediction from Core-Loss Spectra: A Computational Study in Si

“…17−25 Furthermore, it has recently been reported that it is possible to predict the electronic structure of organic molecules in their ground state from C-K edge ELNES/XANES. 26 Building upon prior research employing ML methods, we aim to develop an ML-model to predict the atom-specific electronic structure of all orbitals in their ground state for solid materials from ELNES/XANES, toward the realization of atomic-level analysis of ground-state electronic structures. In this study, we focus on the Si-K edge of Si.…”

“…On the other hand, in recent years, the effectiveness of employing ML techniques in materials characterization has been demonstrated . ML has also increasingly been utilized in extracting information from ELNES/XANES spectra, with methods reported for directly predicting atomic structure and material properties from ELNES/XANES. − Furthermore, it has recently been reported that it is possible to predict the electronic structure of organic molecules in their ground state from C- K edge ELNES/XANES …”

Rational Design of Full-Color Fluorescent C3N Quantum Dots

“…developed a transition metal interconnected neural network (TMINN), effectively capturing the correlation between material geometric configuration and magnetic anisotropy energy for two-dimensional meta–organic framework (MOF) materials . A feedforward neural network was successfully utilized for predicting the electronic structure of organic molecules . Moreover, multilayer deep neural networks have shown promise in simultaneously predicting molecular properties, including energy gap, electronic spatial extent, and dipole moment .…”

“…26 A feedforward neural network was successfully utilized for predicting the electronic structure of organic molecules. 27 Moreover, multilayer deep neural networks have shown promise in simultaneously predicting molecular properties, including energy gap, electronic spatial extent, and dipole moment. 28 However, defining a universally applicable ML model for materials design poses significant challenges.…”

“…According to the charge density distribution of S 1 states, the large size of C 3 N QDs with a diameter of 1.66−1.82 nm (named C 36 N 12 , α-C 36 N 12 and C 30 N 10 ) shows significant electron−hole separation in real space involving the high excited state dipole moment (D Sd 1 = 3.48−4.95 D), which is not conducive to the occurrence of radiation transitions. In the case of smaller QDs (C 12 N 4 , C 18 N 6 , α-C 18 N 6 , C 24 N 8 , and α-C 24 N 8 ), the electron−hole pairs are not significantly separated and show uneven distribution, which results in asymmetric variations of the electron structure in QDs, and electron−hole pairs break the orbital transition symmetry barrier, ultimately leading to radiative recombination, which is well consistent with experimental report that the smaller C 3 N QDs have a large quantum yield (QY > 80%) 27. In experiments, carbonbased QDs are usually synthesized by the solvothermal method; we further explored the solvent effects on fluorescence performances of C 3 N QDs.…”