Computing Excited States of Molecules Using Normalizing Flows.

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Molecular Spectroscopy

Background:

• Calculating highly excited molecular vibrational states is computationally intensive.
• The accuracy of these calculations depends heavily on the chosen coordinate system.
• Existing methods face challenges with delocalized states and basis set convergence.

Purpose of the Study:

• To develop a new method for learning optimal vibrational coordinates.
• To improve the accuracy and efficiency of calculating molecular vibrational energy spectra.
• To address the computational challenges in describing highly excited and delocalized molecular vibrations.

Main Methods:

• Utilizing normalizing flows, which are parametrized invertible functions.
• Learning optimal vibrational coordinates that satisfy the variational principle.
• Applying the method to calculate the 100 lowest excited vibrational states of H2S, H2CO, and HCN/HNC.

Main Results:

• The new method significantly increases accuracy and enhances basis set convergence.
• Optimized coordinates improve the separability of the molecular Hamiltonian.
• The approach allows for effective assignment of approximate quantum numbers.
• Demonstrated transferability of optimized coordinates across different basis set truncations.

Conclusions:

• The normalizing flow method provides a cost-efficient protocol for computing vibrational spectra.
• This approach offers tailored coordinates for specific molecular vibrational problems.
• Enables more accurate and efficient calculations for high-dimensional systems.