Comparing classical approaches with empirical or quantum-mechanically derived force fields for the simulation electronic lineshapes: application to coumarin dyes

The simulation of optical properties in complex and heterogeneous systems, as those involved in photovoltaic devices, requires efficient protocols able to account for the complexity of the environment. An usual procedure is based on molecular dynamics (MD) sampling and subsequent computation of the transition properties, at quantum mechanics/molecular mechanics (QM/MM) level, adopting, e.g., time-dependent density functional theory for the QM part. In this work, we adopt such protocol to simulate the spectra of two structurally related coumarin dyes employed as sensitizers in solar cells, c343 and nkx-2586, and we focus on gas-phase results with the scope to investigate in detail the impact of two key aspects of the protocol: the choice of the force field (FF) and the classical approximation of the spectra. We start from the selection of a FF able to accurately describe the intramolecular interactions in the system, comparing the predictions of a generic empirical force field (namely, GAFF) and a molecule-specific quantum-mechanically derived FF. The MD simulations generated by adopting each FF are analysed in terms of the sampled normal mode distributions, providing a rationale to understand the features (position and width) of the absorption spectra simulated from the distribution of vertical energies. By comparing those simulations with classical, semiclassical and fully quantum formulations of the vibronic lineshapes based on harmonic models, we demonstrate the necessity of high-quality FFs. Notwithstanding this, our results highlight that irrespective of the choice between the two FFs, the largest errors arise from the classical description of the fast nuclear motions (e.g. bond stretchings), thus emphasizing the need to account for quantum effects in order to achieve an accurate simulation of the spectrum. We further analyse the spectra simulated using a number of harmonic models of the initial and final states at either QM or FF levels, pointing out the occurrence of remarkable inaccuracies due to the large differences in the QM and FF normal modes. This problem is inherent to the fixed functional forms associated to any MM FF, and our results indicate a convenient route to avoid it. This knowledge is a prerequisite towards the goal of developing hybrid quantum/classical QM/MM approaches for the calculations of spectra in complex environments.