Modal Backflow Neural Quantum States for Anharmonic Vibrational Calculations.

Area of Science:

• Quantum mechanics
• Computational physics
• Quantum chemistry

Background:

• Neural quantum states (NQS) offer expressiveness for many-body quantum problems.
• Backflow determinants are effective for electronic structure but backflow permanents for bosons are computationally impractical.
• Existing methods struggle with particle conservation in bosonic systems.

Purpose of the Study:

• To introduce a new NQS design, modal backflow (MBF), for bosonic systems.
• To address the computational cost and particle conservation issues of previous bosonic NQS.
• To achieve high accuracy in spectroscopic calculations for anharmonic vibrational problems.

Main Methods:

• Developed a modal backflow (MBF) neural quantum state (NQS) ansatz.
• Implemented a selected-configuration scheme for accurate evaluation of observables and gradients.
• Utilized a vibrational self-consistent field calculation as a pretraining step within the MBF network.

Main Results:

• The MBF NQS design overcomes limitations of previous bosonic approaches.
• Spectroscopic calculations achieved high accuracy across all anharmonic regimes.
• Demonstrated efficacy on both artificial and ab initio Hamiltonians.

Conclusions:

• The MBF NQS is a powerful and efficient ansatz for bosonic quantum problems.
• This method enables accurate prediction of zero-point energies and vibrational transitions.
• MBF NQS provides a viable path for advanced spectroscopic calculations.