Accurate Spectral Predictions of Medium-to-large Molecular Systems: Balancing Performance and Reliability

Computational spectroscopy is being commonly used to understand the observed spectra of molecular systems, like biological molecules or photosensitive molecules that attract significant interest for solar energy conversion, organic display, or bio-probes. The correct prediction of the physicochemical properties of these systems can provide guidelines to understand their activities and optimize their applications. However, the methods commonly used for such studies have become insufficiently accurate to match the most advanced spectroscopic techniques or those of higher sensitivity, like chiroptical ones. While more advanced models have been proposed in the literature, they have been mostly confined to small or simple molecules. The scale-up to larger and more complex structures presents several challenges, which were at the core of this thesis. For vibrational spectra, the double harmonic approximation remains the standard, despite its systematic overestimation of the energy levels and its inability to predict overtones and combinations. Anharmonic simulations do not have these deficiencies, but have a steep increase in the computational cost. Among them, variational methods provide a systematic improvement, but become quickly prohibitive, making them only feasible on very small molecules without significant approximations. While second-order vibrational perturbation theory (VPT2) can be used on systems with dozens of atoms, the resonances present in the system can induce significant errors in the simulations. To deal with the problem of resonances, deperturbed (DVPT2) and generalized (GVPT2) VPT2 have been proposed. Using these two models as reference, a series of benchmarks have been designed during this thesis targeting medium-to-large molecular systems, and new protocols have been proposed to identify the potential pitfalls and maximize the accuracy and performance of anharmonic simulations. 15 molecules ranging from small to large have thus been studied, classified into three groups. The first group is composed of 8 small molecules, with less than 7 atoms. These small molecules have been used for a systematic benchmark of DVPT2 and GVPT2, including the intensity of vibrational spectra. With the aim of determining the achievable accuracy and reliability of DFT-based VPT2 calculations, several DFT functionals with different basis sets have been tested, as well as hybridization schemes combining different levels for the harmonic base and the anharmonic correction. The best performing electronic structure methods (ESCMs) are then used to treat medium-sized molecular systems (the second group). The second group comprises furan, fluorobenzene, methyloxirane, and glycine, with more complex topologies and resonance patterns. The definition of the most suitable ESCMs has been further refined with this second group, especially the hybrid schemes. However, the presence of new resonance patterns and large amplitude motions (LAMs) in methyloxirane and glycine are two other important factors that can influence the accuracy of VPT2 in general. Methods dealing with resonance patterns and LAMs have been tested on the molecules of the second group, which paved the way to the simulation of large-sized molecules in group three. For intensity, a more complete version of DVPT2, called IDVPT2, was implemented to correctly remove resonances of all types. Three molecules have been chosen to represent large molecules, naphthalene, -pinene and artemisinin. More in-depth studies on resonance have been carried out on these three molecules, especially on -pinene. With pinene, a new class of Fermi resonances (FRs), specific to intensity, was identified, and new correction schemes were devised, implemented, and tested. Artemisinin was chosen in the study as an example of a large and complex system, with 42 atoms and 7 chiral centers. It is an important pharmaceutical molecule, which is known for its high anti-malaria and anticancer bioactivity. The final computational protocol has been used to simulate the vibrational circular dichroism (VCD) and Raman optical activity (ROA) spectra of artemisinin at a very accurate level. Thanks to this, we can proceed to describe qualitatively the coupling between the reactive and recognition sites. This result is also an interesting contribution for the bioactivity and structural studies of pharmaceutical molecules. For electronic spectra, pure electronic transitions are often used, completely ignoring the vibronic effects. This has been shown to be insufficient when used for high-resolution UV-visible spectra or many chiroptical spectra. Several models have been developed to properly consider the vibrational effects in electronic spectra, which can help balance accuracy and feasibility in terms of computational cost. There are generally two groups according to the extrapolation of the potential energy surface, which are the adiabatic models (Adiabatic Hessian, AH, and Adiabatic Shift, AS) and vertical models (Vertical Hessian, VH, and Vertical Gradient, VG). The Duschinsky transformation describes the evolution of the vibrational modes in relation to the electronic transition. Then Franck-Condon (FC) and its extension, Herzberg-Teller (FCHT), are used to represent the dependence of the transition properties on the nuclear positions. The final spectra can be generated through two algorithms, the time-independent (TI) and the timedependent (TD) methods. TI is based on an explicit sum of all transitions, while TD describes the evolution of a wave packet after the transition. Because of the many possible combinations offered by these models, vibronic simulations require specific knowledge and experience to select the proper modal, depending on the targeted accuracy and characteristics of the system. A significant hurdle to the definition of reliable protocols is that the strengths and weaknesses of these models have not yet been extensively studied for large and complex systems, which may exhibit some degree of flexibility, especially for chiroptical spectra. Hence, a protocol has been built for the study of a family of Ir-complexes, which emit deep blue phosphorescence. Through the step-up study of the molecules in this family, from prototype to more complex molecules, the available methods were systematically assessed and compared. Because of the size and complexity of the system, reduced-dimensionality schemes on top of internal coordinates were used and defined to achieve good reliability and accuracy. Based on the vibronic simulations, the most relevant transitions and the associated vibrational modes were identified, and possible improvements on the experimental side have been proposed. This extensive study served as a basis for the investigation of a chiral boron dipyrromethene dye (BODIPY) and a porphyrin derivative (zinc octaethylporphyrin, ZnOEP). All these examples have shown the importance of including the vibrational effects, not only to improve the accuracy of the results, but also in some cases to get a correct understanding of the underlying physicochemical properties. They are also good targets to study the importance of balancing the computational cost and accuracy. From the study of these molecules, a complete protocol for vibronic simulation has been built for organic and organometallic systems. In these studies, we also investigate the value of visualization tools, in particular the electronic transition current density, to gain further insights on the origin of the chiroptical signal at a local level.