Computational spectroscopy using MULTIMODE and machine-learned potentials

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Spectroscopy

Background:

• Advanced computational vibrational spectroscopy requires accurate molecular potentials.
• Machine-learned potentials (MLPs) offer highly precise fits to electronic energies.
• Integrating MLPs with post-harmonic quantum methods enables more accurate simulations.

Purpose of the Study:

• To review methods combining MLPs with post-harmonic quantum calculations.
• To present applications of vibrational self-consistent field and configuration interaction (VSCF + VCI) using MLPs.
• To introduce new interfaces for the MULTIMODE code to utilize MLPs.

Main Methods:

• Utilized vibrational self-consistent field and configuration interaction (VSCF + VCI) calculations.
• Developed Python-based GUI and PyFort (Fortran) interfaces for MULTIMODE.
• Employed Machine-Learned Potentials (MLPs) including PhysNet and MACE-OFF.

Main Results:

• VSCF + VCI calculations with MLPs accurately reproduced vibrational spectra for formic acid dimer and protonated oxalate anion.
• Demonstrated successful integration of MLPs via new MULTIMODE interfaces.
• Validated MLPs against permutationally invariant potentials and tested MACE-OFF with H2CO.

Conclusions:

• The developed interfaces facilitate the use of MLPs in advanced vibrational spectroscopy.
• Combining MLPs with VSCF + VCI methods provides accurate predictions for molecular spectra.
• This work enhances the capability of computational chemistry for studying complex molecular systems.