Table of Contents
Fetching ...

Vibrational infrared and Raman spectra of the methanol molecule with equivariant neural-network property surfaces

Ayaki Sunaga, Albert P. Bartók, Edit Mátyus

Abstract

Electric dipole and polarizability surfaces are developed for the methanol molecule using {\it ab initio} electronic structure data, computed at the CCSD/aug-cc-pVTZ level of theory, and equivariant neural networks. These property surfaces are used to compute vibrational infrared and Raman intensities up to the OH stretching fundamental vibration. The intensity computations use the vibrational energies and wave functions obtained in continued variational vibrational computations from earlier work [J. Chem. Phys. 163, 064101 (2025)]. The vibrational representation accounts for the large-amplitude torsion and uses curvilinear normal coordinates for the small-amplitude modes, allowing truncation of the vibrational basis set and the integration grid.

Vibrational infrared and Raman spectra of the methanol molecule with equivariant neural-network property surfaces

Abstract

Electric dipole and polarizability surfaces are developed for the methanol molecule using {\it ab initio} electronic structure data, computed at the CCSD/aug-cc-pVTZ level of theory, and equivariant neural networks. These property surfaces are used to compute vibrational infrared and Raman intensities up to the OH stretching fundamental vibration. The intensity computations use the vibrational energies and wave functions obtained in continued variational vibrational computations from earlier work [J. Chem. Phys. 163, 064101 (2025)]. The vibrational representation accounts for the large-amplitude torsion and uses curvilinear normal coordinates for the small-amplitude modes, allowing truncation of the vibrational basis set and the integration grid.
Paper Structure (24 sections, 28 equations, 10 figures, 12 tables)

This paper contains 24 sections, 28 equations, 10 figures, 12 tables.

Figures (10)

  • Figure 1: Definition of the primitive internal coordinates and the primitive body-fixed (pBF) Cartesian frame of the methanol molecule used in this work. The pBF is shifted to the centre of mass.
  • Figure 2: Electric dipole moment of CH$_3$OH vs. $\tau$: components and $|\mu|$ length, as a function of the torsional angle; the other 11 internal coordinates were relaxed (MEP) on PES2025.Sunaga2025JCP_PES PFE: path-following Eckart frame, Eck: (single-reference) Eckart frame with the reference structure orientation according to pBF (Sec. ). The curves show the property surface components (and length), the points are the ab initio values (at the same CCSD/aug-cc-pVTZ level of theory as used for the training dataset).
  • Figure 3: Polarizability of CH$_3$OH vs. $\tau$: components and squared mean (isotropic) and anisotropic polarizabilities, as a function of the torsional angle with the other 11 internal coordinates relaxed (MEP) on PES2025.Sunaga2025JCP_PES PFE: path-following Eckart frame, Eck: (single-reference) Eckart frame with the reference structure orientation according to pBF (Sec. ). The curves show the property surface, the points are the ab initio values (at the same CCSD/aug-cc-pVTZ level of theory as the training dataset).
  • Figure 4: Electric dipole moment of CH$_3$OH vs. $r_4$: components and $|\mu|$ length, as a function of the $r_4$CH bond distance with all other coordinates fixed at one of the equilibrium structures of PES2025.Sunaga2025JCP_PES PFE: path-following Eckart frame. $\Delta \mu = \mu(\mathrm{pBF}) - \mu(\mathrm{PFE})$, where pBF refers to the primitive body-fixed frame (Sec. ). The curves show the property surface, the points are the ab initio values (at the same CCSD/aug-cc-pVTZ level of theory as the training dataset).
  • Figure 5: Polarizability of CH$_3$OH vs. $r_4$: components and squared mean (isotropic) and anisotropic polarizabilities, as a function of the $r_4$CH bond distance with all other coordinates fixed at one of the equilibrium structures of PES2025.Sunaga2025JCP_PES PFE: path-following Eckart frame, $\Delta \alpha = \alpha(\mathrm{pBF}) - \alpha(\mathrm{PFE})$ (and similar for $\Delta a^2$ and $\Delta \gamma^2$), where pBF refers to the primitive body-fixed frame (Sec. ). The curves show the property surface, the points are the ab initio values (at the same CCSD/aug-cc-pVTZ level of theory as the training dataset).
  • ...and 5 more figures