Machine learning for vibrational spectroscopy via divide-and-conquer semiclassical initial value representation...

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Machine Learning

Background:

• Accurate calculation of molecular power spectra is crucial for understanding molecular dynamics.
• Traditional methods face challenges in computational cost and accuracy for larger systems.
• Partitioning nuclear vibrational space is key to efficient semiclassical simulations.

Purpose of the Study:

• Introduce a novel machine learning algorithm for partitioning nuclear vibrational space.
• Improve the accuracy and efficiency of molecular power spectrum calculations.
• Integrate the algorithm with the divide-and-conquer semiclassical initial value representation (DCSIVR) method.

Main Methods:

• Developed a machine learning algorithm based on evolutionary selection and probability graphs.
• The partitioning criterion ensures the preservation of unitary via Liouville's theorem.
• Interfaced the algorithm with the DCSIVR method for power spectrum computation.

Main Results:

• Benchmarked the algorithm on model systems with known exact subspace divisions.
• Successfully calculated the vibrational power spectrum of methane, validating against existing data.
• Applied the method to compute the power spectrum of trans-N-methylacetamide.

Conclusions:

• The machine learning algorithm effectively partitions vibrational space for accurate semiclassical calculations.
• The approach offers a promising pathway for simulating larger and more complex molecular systems.
• This method enhances the reliability of molecular power spectrum predictions in computational chemistry.