Two-dimensional infrared spectroscopy of solute-solvent complexes from linear-scaling DFT and machine learning

Area of Science:

• Computational Chemistry
• Spectroscopy
• Molecular Dynamics

Background:

• Two-dimensional infrared (2DIR) spectroscopy reveals ultrafast molecular dynamics, particularly hydrogen-bonding.
• Accurate simulations of solute-solvent interactions are crucial for interpreting 2DIR spectra.
• Ab initio molecular dynamics (AIMD) provides rigorous simulations but is computationally expensive, limiting trajectory lengths.

Purpose of the Study:

• To develop a computationally efficient yet accurate method for simulating vibrational spectra using AIMD.
• To enable the study of molecular dynamics and hydrogen-bonding on nanosecond timescales.
• To bypass the need for empirical frequency maps in spectral simulations.

Main Methods:

• A hybrid approach combining linear-scaling density functional theory (LS-DFT) with a machine-learned (ML) interatomic potential (DeepMD).
• LS-DFT simulations generate reference data (energies, forces, dipole moments) for training the ML potential.
• Instantaneous vibrational frequencies are calculated from numerical Hessians, with dipoles learned directly from electron density.

Main Results:

• The ML potential enables nanosecond-scale molecular dynamics at significantly reduced computational cost compared to pure AIMD.
• Simulated linear and 2DIR spectra for N-methylacetamide in methanol show excellent agreement with experimental data.
• The characteristic doublet structure of the amide I band is accurately reproduced, validating the method's accuracy.

Conclusions:

• The developed hybrid LS-DFT/ML framework offers a practical and accurate route for simulating vibrational spectra.
• This method facilitates AIMD-level simulations for various IR-active solutes and hydrogen-bonded systems.
• It opens new possibilities for studying complex molecular interactions and dynamics with high fidelity.