Fluctuating Solvent Fields and Vibrational Shifts Revealed by Machine Learned Force Fields

ORAL

Abstract

Fluctuating solvent fields is one of the driving mechanism behind vibrational shifts observed in solvatochromic experiments [1]. Previously, experimental infrared spectra have been correlated with mean electric fields obtained from empirical force field molecular dynamics simulations with fixed partial charges or multipoles to gauge molecular vibrational probes and to infer local electric fields inside proteins [2]. However, vibrational dynamics of these force fields are insufficient to reproduce the experimental line shape properties.

Here, machine learned force fields (MLFFs) offer multiple opportunities: they are highly accurate, efficient, and can be interpretable if based on physical models. We explore the capabilities of modeling vibrational solvatochromism from molecular dynamics simulations of vibrational probes in solvent and protein environments employing the SO3LR [3] MLFF. SO3LR combines the SO3krates architecture for short-range interactions with pairwise analytical electrostatic and dispersion potentials in the long range, and was pretrained on a diverse data set of molecular complexes. We show how fluctuating solvent electrostatic and dispersion fields can be related to vibrational bands in a self consistent manner due to the versatility of MLFFs.

[1] Brünig, FN, Netz, RR, et al. JPCB 126, 1579–1589 (2022).

[2] Weaver, JB, Kozuch, J, Kirsh, JM & Boxer, SG. JACS 144, 7562–7567 (2022).

[3] Kabylda, A, Frank, JT, Müller, KR, Tkatchenko A, et al. JACS 147, 33723–33734 (2025).

Presenters

  • Florian Brünig

    • University of Luxembourg

Authors

  • Florian Brünig

    • University of Luxembourg
  • Alexandre Tkatchenko

    • University of Luxembourg