Machine Learning of Quasiparticle Energies in Molecules and Clusters

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Machine Learning

Background:

• Predicting electronic properties like quasiparticle energies and photoelectron spectra is crucial for understanding molecular behavior.
• Traditional methods (e.g., GW approximation, Bethe-Salpeter equation) are computationally expensive, limiting their application to large systems or complex simulations.

Purpose of the Study:

• To introduce a novel machine learning approach, Delta-MLQP, for accurate and efficient prediction of GW quasiparticle energies and photoelectron spectra.
• To develop orbital-sensitive representations (OSRs) that capture essential electronic information from density functional theory (DFT) calculations.

Main Methods:

• Utilized kernel ridge regression with orbital-sensitive representations (OSRs) derived from molecular Cartesian coordinates.
• Augmented Coulomb matrix, bag-of-bond, and bond-angle-torsion representations with atom-centered orbital charges and Kohn-Sham orbital energies.
• Trained and validated the Delta-MLQP model on the QM8 dataset for frontier orbital energies and photoelectron spectra prediction.

Main Results:

• Achieved high accuracy in predicting frontier orbital energies across 22,000 molecules.
• Successfully predicted full photoelectron spectra with a mean absolute error below 0.1 eV.
• Demonstrated the model's ability to capture conformational effects in water monomers and dimers.
• Showcased integration within multiscale simulations for solvatochromic shifts of excited states.

Conclusions:

• The Delta-MLQP model provides GW-level accuracy for quasiparticle energies and photoelectron spectra at DFT computational cost.
• This approach significantly accelerates predictions for molecules and clusters, enabling broader applications in computational chemistry.
• The OSR-based Delta-MLQP is a powerful tool for studying electronic properties and integrating quantum mechanics with molecular dynamics.