Predicting ELNES/XANES spectra by machine learning with an atomic coordinate-independent descriptor and its...

Area of Science:

• Materials Science
• Computational Chemistry
• Spectroscopy

Background:

• Electron Energy Loss Near Edge Structure (ELNES) and X-ray Absorption Near Edge Structure (XANES) spectra reveal unoccupied electronic states.
• Calculating these spectra requires significant computational resources for structure and electronic structure computations.
• Existing methods are computationally intensive, limiting high-throughput analysis.

Purpose of the Study:

• To develop a machine learning approach for direct prediction of ELNES/XANES spectra.
• To utilize an atomic-coordinate-independent descriptor (SMILES) for spectral prediction.
• To extend the method for predicting ground-state electronic structures, such as Partial Density of States (PDOS).

Main Methods:

• Employed a machine learning technique using Simplified Molecular Input Line Entry System (SMILES) strings as input.
• Developed an atomic-coordinate-independent descriptor for direct spectral prediction.
• Trained the model on a dataset incorporating long-SMILES molecules to improve accuracy.

Main Results:

• Achieved direct and precise prediction of ELNES/XANES spectra from SMILES strings.
• Successfully extended the methodology to predict ground-state electronic structures (PDOS).
• Demonstrated enhanced prediction accuracy with the inclusion of longer SMILES molecules in the training data.

Conclusions:

• The machine learning approach offers a computationally efficient alternative for ELNES/XANES spectral prediction.
• The method's ability to predict ground-state electronic structures broadens its applicability.
• Direct derivation of spectroscopy from SMILES strings accelerates spectroscopic investigations and materials discovery.