Spectral Tuning and Photoisomerization Efficiency in Push–Pull Azobenzenes: Designing Principles

This work demonstrates how push–pull substitution can induce spectral tuning toward the visible range and improve the photoisomerization efficiency of azobenzene-based photoswitches, making them good candidates for technological and biological applications. The red-shifted bright ππ* state (S2) behaves like the lower and more productive dark nπ* (S1) state because less potential energy along the planar bending mode is available to reach higher energy unproductive nπ*/S0 crossing regions, which are responsible for the lower quantum yield of the parent compound. The stabilization of the bright ππ* state and the consequent increase in isomerization efficiency may be regulated via the strength of push–pull substituents. Finally, the torsional mechanism is recognized here as the unique productive route because structures with bending values attributable to the inversion mechanism were never detected, out of the 280 ππ* time-dependent density functional theory (RASPT2-validated) dynamics simulations.


■ INTRODUCTION
Azobenzene (AB) is a prototypical photoresponsive molecule undergoing a reversible photoinduced isomerization between its cis and trans configurations, which is strongly attractive for a widespread range of applications. The trans ↔ cis interconversion mechanism has been debated for a long time: 1−10 it could take place through rotation around the central double bond (torsion) or through an in-plane bending motion (Scheme 1). Eventually, hybrid torsion-bending processes were recently proposed. 11−14 Interestingly, the well-separated absorption wavelengths of the two isomers make this molecule suitable for optical switches in technological 15−17 or biological 18−20 devices and in the development of light-powered molecular machines. 2,3,21−27 Both isomers show two absorption bands in the UV−vis window: the more intense one is associated to a π → π* transition, peaking in the UV region (301/265 nm in the gas phase, trans/cis, respectively 28 ), while the much weaker band in the visible range (440/425 nm 28 ) is associated to a symmetry-forbidden n → π* transition. These ππ*/nπ* bands are separated enough to allow their selective irradiation: interestingly, excitation in the UV (ππ*) and in the visible region (nπ*) shows significantly different quantum yields (QYs), about 11% and 25%, respectively, in the trans case and 27% and 56% in the cis case in n-hexane. 29 The QY wavelength dependence, which is in contrast with Kasha's rule, suggests that different reaction mechanisms may take place starting from the ππ* or nπ* excited states (ESs), 12 an issue that is still under discussion in experimental 8,30 and theoretical 8,11,31−33 studies. Because of the reversibility of the isomerization, its speed, and the simplicity of incorporating azobenzene in complex structures, many studies are focused on red-shifting the intense ππ* bands, whose UV absorption is limiting technological and biological applications. For this purpose, push−pull substituents have demonstrated to be good candidates: 37−40 the simultaneous destabilization of the last π orbital (electron-donating substituent) and stabilization of the π* LUMO (electron-withdrawing substituent), results in red shift of the ππ* absorption, 8,18,33,41−46 which influences the ππ−nπ* energy gap and leads to a change in the photoisomerization properties. The aim of this work is to evaluate how push−pull substituents could control the capability of ABbased photoswitches, tuning the linear absorption energy and the isomerization efficiency, depending on the mechanism behind. For this purpose, we compare the behavior of the parent AB with two different push−pull-substituted systems with increasing electron-donating/withdrawing strength: 4methoxy-4′-cyanoazobenzene (NC−AB−OMe) and 4-(4nitrophenylazo)aniline (O 2 N−AB−NH 2 , also known as Disperse Orange 3 or DO3); see Figure 1. The comparison is made by means of time-dependent density functional theory (TD-DFT) semiclassical dynamics simulations (RASPT2validated at crossing points, see Table S11) accounting for multireference dynamically correlated energies. The results allow us to identify the control knobs of productive (i.e., photoisomerization) versus nonproductive (i.e., aborted photoisomerization) radiationless decays, thus paving the way to a rational design of AB derivatives with tuneable spectral properties and increased photoisomerization efficiency.

■ RESULTS AND DISCUSSION
Comparison between AB and the two push−pull-substituted systems NC−AB−OMe and O 2 N−AB−NH 2 (Figure 1) was done by conducting mixed quantum−classical dynamics simulations at the TD-DFT/CAM-B3LYP/6-31G* level, following 40 gas-phase trajectories on both the cis and trans isomer of the three systems (240 trajectories in all), initiated on the bright ππ* state. Nonadiabatic events were treated with a simplified hopping scheme, relying on the energy gap as a criterion for changing the electronic state, fixed lower than 3 kcal/mol. Back hopping was always allowed between ESs, while it was not permitted once the trajectory decayed on the ground state (GS). We ran 40 additional trajectories on the trans-AB nπ* state that were employed as a reference for the photobehavior of the more productive dark state in the parent compound. TD-DFT/CAM-B3LYP/6-31G* was validated by benchmarking against RASPT2 static calculations at the S 1 /S 0 crossing points, using an accurate setup that was previously tested for the parent system (MS-RASPT2/RASSCF/ANO-L-VDZP), 12,47 where the active space (including the valence πorbitals and the nitrogen lone pairs) was appropriately enlarged for the push−pull-substituted systems (Figures S1− S3 in the Supporting Information). However, because TD-DFT fails to produce correctly shaped potential energy surfaces (PESs) in the region surrounding intersections with S 0 , we limit our analysis to the ES dynamics until the S 1 /S 0 gap is lower than 3 kcal/mol (S 1 /S 0 crossing seam). The O 2 N− AB−NH 2 and NC−AB−OMe derivatives were selected after a preliminary study (at the CAM-B3LYP/6-31G* level) of eight systems with increasing push−pull strength: Figure 1 clearly shows how the substituents red-shift the ππ* state, leaving the dark nπ* roughly unchanged. Increasing the push−pull strength reduces the ππ*/nπ* energy gap, until inversion of the ππ*/nπ* energy order ( Figure 1 and Table S1). Because of their small size, the selected systems are good candidates to make accurate predictions about these promising push−pull derivatives. The quantitative accuracy of the employed method is supported by the good agreement between the experimental and computed vertical energies for trans/cis-AB, trans-NC− AB−OMe, and trans-O 2 N−AB−NH 2 (see Table 1). This also validates the prediction for the absorption values (ππ* and nπ*), which are not available in the literature, in particular for the push−pull cis-conformers, which are thermally unstable and therefore difficult to isolate and characterize. 40 Besides stabilizing the ππ* state, the growing strength of the push− pull substituents also affects the charge distribution on the two phenyl rings and, consequently, the molecular dipole moment (see Figure S4 and Table S1 in the Supporting Information). The charge separation is proportional to the electrondonating/withdrawing strength, as shown in Figure 1: −NH 2 is a better push group than −OMe because of the lower electronegativity of the nitrogen atom; at the same time, the −NO 2 substituent "pulls" more than the −CN. The charge excess on the two halves is notably larger on the bright ππ* ES, with a consequent increase in the dipole moment value: 0.162/ 0.247 D and 0.241/0.327 D on ππ* for trans/cis NC−AB− OMe and the O 2 N−AB−NH 2 , respectively, compared with 0.074/0.071 D and 0.115/0.106 D of the GS. The larger dipole moment of the cis conformer could be referred to the nonplanar geometry that hinders the orbital delocalization, leading to a larger charge separation between the two halves.
We show how the push−pull derivatives behave dynamically different, compared to the parent system, when they are excited to the bright ππ* state in the following. Because photoexcited trans-and cis-isomers lead to quite different paths 11 (as demonstrated by the experimental lifetimes in Table 1), we will discuss them separately.
trans-AB Systems. Looking at the S 2 dynamics leading to the initial S 2 → S 1 decay, we notice that the CNNC dihedral angle stays close to 180°in all the systems, while the CNN bending angles close and then oscillate around a value that is a bit smaller than that in the FC geometry, in agreement with recent studies on the AB photoisomerization. 12 Figure 2 shows the normalized distribution of the CNNC torsion (top) and CNN bending (bottom) trajectories along the trans-dynamics of the three systems, including the nπ* trans-dynamics of AB. The left panels refer to the dynamics on S 2 [panels (a−c) and (h−j)], while the right plots refer to the dynamics on S 1 after decay from S 2 [panels (e−g) and (l−n)] or after direct excitation [for trans-AB, panels (d,k)]. The most significant effect of push−pull substitution is a drastically shorter ππ* lifetime with respect to the parent compound, where S 2 is living two times longer than in the substituted trans-systems (168 fs for AB against 70 and 86 fs for NC−AB−OMe and O 2 N−AB−NH 2 , respectively; see vertical dashed lines in Figure 2, left part).
On the other hand, in the subsequent dynamics on the nπ* state (S 1 ), bending oscillations accompany the torsional motion ( Figure 2, right part), leading to a S 1 → S 0 crossing region spanning from planar to fully rotated CNNC values (Tables S4−S6 in the Supporting Information). This is due to an extended S 0 /S 1 crossing seam, that has been extensively documented in previous studies, 11,47,48 covering both bending and torsional modes, where the fully (∼90°) rotated structures are the lowest in energy, but also, higher energy, less-rotated structures could be accessible through the bending mode, provided that enough kinetic energy is available in the dynamics. Based on the characteristics of the S 1 → S 0 seam, we have grouped the trajectories in two different sets, labelled torsional and bending paths, based on the CNNC torsional value at the S 1 → S 0 decay: the former includes trajectories decaying on S 0 at CNNC < 135°(half between 180 and 90°), the latter includes trajectories which, to a great extent, preserve the planarity of azobenzene until decaying to the GS (CNNC > 135°). Most trajectories for all the three trans-systems follow the bending path (82.5/65/65% for AB/NC−AB−OMe/  The Journal of Physical Chemistry A pubs.acs.org/JPCA Article O 2 N−AB−NH 2 , respectively; see Table 2), but none of them reach bending values that could justify a possible inversiondriven isomerization process (i.e., close to 180°; see Scheme 1, bottom part of Figure 2, and Tables S4−S6). This explains the smaller QY of trans-AB from ππ* (11% vs 25% from nπ* 29 ): the most populated bending paths are reaching S 0 /S 1 CIs with neither bending nor torsion values large enough to allow the trans−cis isomerization. Moreover, the bending motions are mainly symmetric (see values in Tables S4−S6), and even a hypothetical concerted bending mechanism would lead back to the reagent. Consequently, on the basis of the large number of dynamics on the three trans-systems (120, 40 for each system), we conclude that the only productive process follows the torsion mechanism. However, because our analysis is limited to the ES dynamics until decay to the GS, we can only have an upper bound estimate of the ππ* QY, which is given by the number of torsional paths populated for each system: we obtained 17.5%, 35%, and 35% for trans-AB, NC−AB−OMe, and O 2 N−AB−NH 2 , respectively, envisaging a larger QY in the push−pull systems than that in the parent AB (see Table  2). To prove that the isomerization QY correlates well with the population of the torsion mode, we ran 40 dynamics for trans-AB (using the same initial conditions as for the ππ* state) starting from the more productive nπ* state (experimental QY = 25% 29 ): in this case, 32.4% of the trajectories belong to the torsional path ( To further support and rationalize that the push−pull systems could be more productive than the parent trans-AB because of the larger population of torsional paths, Figure 3 shows the S 2 → S 1 (red) and the S 1 → S 0 (dark blue) hopping point distribution, along the bending/torsional coordinates, for the ππ* trajectories of the three different systems. Interestingly, for the push−pull systems, the S 1 → S 0 hopping point distribution obtained starting from the bright ππ* state is matching with the trans-AB distribution obtained starting from the more efficient nπ* state [light-blue points in Figure 3 panel (a) versus blue points in panels (b,c)], envisaging that the push−pull derivatives excited to ππ* behave exactly as AB excited to nπ*, for which a larger isomerization productivity is experimentally documented. Instead, the S 1 → S 0 decay points for trans-AB when excited to S 2 show a clearly different distribution, largely concentrated in the bending region. Additionally, the average bending values at the S 1 → S 0 hopping points are a bit smaller for the torsional trajectories (and in the nπ* dynamics) than for the bending ones (see Table 2 and Figure 3), which is perfectly in line with the shape of the S 1 /S 0 crossing region depicted in our earlier studies, 11 showing that fully rotated CIs (∼90°) display smaller bending values than less-rotated (and therefore less productive) ones. 11,47 It is thus apparent that by calibrating the strength of push−pull substituents, one could red-shift the absorption maximum of the bright ππ* state, bringing it closer to that of the productive nπ* and concurrently increase the photoisomerization efficiency, two main achievements in the design of photoactive AB-based systems.
Concerning the lifetimes, we see a nice agreement between experiments and theory: time-resolved photoelectron spectroscopy experiments 5 Table 1); the longer one (420 fs) is close to the computed S 2 + S 1 average decay time of 323 fs of the Trans-system dynamics: torsional path = CNNC < 135°at the S 1 /S 0 decay and bending path = 135°< CNNC < 180°at the S 1 /S 0 decay. The geometrical parameters are averaged over all the set of trajectories belonging to each torsional/bending group. Cis-system dynamics: all trajectories are ascribable to the torsional path (>99%), for which CNNC > 45°at the S 1 /S 0 decay.   Table 1 and details on average lifetimes in Table S4). Even though the original work 5 attributed the longer experimental lifetime of 420 fs to two higher lying ππ* states (S 3 −S 4 ), the low oscillator strength reported for them 5,11 suggests that the population of S 2 is by far more probable and that the 420 fs time constant could instead be associated to the S 2 + S 1 deactivation following the CNNC torsional motion toward the twisted S 1 /S 0 crossing region. This hypothesis was already proposed by Granucci et al., 49 and it is also supported by the following theoretical 50−52 and experimental 34 studies reporting a S 1 lifetime of about 0.4 ps. An insight into the behavior of the dynamics following S 2 → S 1 decay clearly shows an average nπ* S 1 lifetime that is almost doubled in the push−pull derivatives than in AB (155 fs, 141 fs, and 63 fs for NC−AB−OMe, O 2 N−AB−NH 2 , and AB, respectively, black dashed lines in Figure 2e−g). Interestingly, the S 1 average lifetimes of the push−pull systems resemble those of the more productive dark nπ* state of the parent AB when it is directly excited, (130 fs, see black dashed line in Figure 2d,k), once again showing that the dynamics of the push−pull systems excited to the ππ* resembles that of the nπ* state of AB. Eventually, we observe that the S 1 torsional path average lifetime in the trans push−pull systems (red dashed line, Figure 2f,g) is about three times longer than that in the bending paths (green dashed lines Figure 2f,g), which is again similar to the dynamics of trans-AB from the nπ* state (62 vs 270 fs, Figure 2d,k). The longer lifetime of the torsional versus bending path could be simply referred to the time needed for internal vibrational energy redistribution from the bending to torsional mode, which is necessary to populate the MEP leading to the nπ* decay process to the GS. 47 This is in line with the recently published AB ππ* CASPT2 dynamics, 12 indicating that the productive CNNC torsional mechanism is slower than the unproductive route characterized by symmetric bending modes.
To explain the opposite trend in the S 2 and S 1 lifetimes observed in push−pull AB as compared to the parent compound, we propose a simple model, which rationalizes entirely the differences documented in both ESs for the three systems. Because the push−pull substituents stabilize only the bright state, while keeping the nπ* energy unaffected, we imagine a simple shift of the ππ* PES, as shown in Scheme 2. By lowering the ππ* state, the crossing with nπ* becomes The Journal of Physical Chemistry A pubs.acs.org/JPCA Article more accessible (i.e., lower activation energy), thus leading to a shorter S 2 lifetime for the push−pull derivatives ( Figure 2). Additionally, less energy becomes available along the initially populated bending modes on S 1 to eventually access the higher energy S 1 /S 0 crossing region at roughly planar structures (torsional angle CNNC around 180°). Eventually, vibrational energy redistribution takes place, triggering population of the nπ* (torsional) minimum energy path and populating the slower, but more productive, torsional paths leading to rotated S 1 /S 0 CI structures. cis-AB Systems. Cis-isomers behave similarly to the trans ones: the push−pull substituents red-shift the ππ* intense band according to their electron-donating/withdrawing strength, leaving the nπ* state energy roughly unchanged ( Table 1). The main difference with respect to the trans-conformers is that except for few outliers, more than 99% of the 120 cisdynamics reach S 1 → S 0 regions, which is always attributed to the CNNC torsional decay mechanism (CNNC > 45°; see Tables 2 and S7−S9 in the Supporting Information), as clearly shown in Figure 3. This is in line with the larger experimental QY observed in cis-AB (Φ = 0.27 vs 0.11 of the trans 29 ). Moreover, torsion is activated already on S 2 (reaching torsion values up to 50°; see Figure 3) and becomes notably larger on the S 1 , as shown by the torsion panels of Figure 4, because of the nonplanar FC starting structure. The earlier activation of the torsional motion, compared to the trans analogues, impedes the early decay to the nπ* state through the bending funnel, resulting in longer S 2 lifetimes of the cis-isomers, compared to the trans ones, in agreement with previous dynamics simulations of AB from the ππ* state. 52 The bending motions are more asymmetric than those in the trans-systems (Table  2), and to be more specific, the larger bendings are mainly attributed to the fragment bearing the electron donor group (−OMe or −NH 2 ; see Tables S8 and S9 in the Supporting Information). Anyways, none of the cis-dynamics reach bending angles close to 180°(see Tables S7−S9), suggesting that the inversion path is not populated, as already noted for trans-systems. The S 2 lifetime is shortening with the increasing push−pull strength (Figure 4a−c and g−i), supporting the previously explained hypothesis that the ππ* red shift speeds up the access to the ππ*/nπ* crossing seam (Scheme 2). Instead, the nπ* lifetime in the cis-isomers is not affected by the push−pull substituents (Figure 4d−f and j−l) because the steeper gradient along the torsional coordinate drives the system straight to the rotated nπ*/S 0 peaked CIs, as documented previously. 11,53 These differences in the S 1 PES shape (compared to the flat trans-nπ* surface) correlate with a larger amount of kinetic energy along the torsional mode, inevitably leading to an increased photoisomerization QY with respect to the trans analogues.
That said, looking at the S 2 /S 1 and S 1 /S 0 CI distribution along the torsion/bending coordinates in Figure 3, we see that the parent and push−pull-derivatives behave similarly, populating the same photoisomerization processes and thus suggesting similar photoisomerization QYs (which is expected to remain higher than that in the trans analogues).

■ CONCLUSIONS
Kasha's rule violation in AB systems was often attributed to two different decay channels that are populated when exciting directly the ππ* (S 2 ) or nπ* (S 1 ) state. The present work supports and extends this hypothesis by proposing a unified mechanistic model, which can be applied to both azobenzene and its push−pull derivatives, foreseeing a higher QY for the latter, with respect to the parent compound. By analyzing a large number of TD-DFT (RASPT2 validated) ππ* molecular dynamics on AB and two push−pull AB derivatives, we see that S 2 trajectories in the parent compound are mainly characterized by CNN/NNC bending motions preserving the system planarity and eventually leading to S 2 /S 1 and, subsequently, to S 1 /S 0 crossing regions, which are unproductive and drive the systems back to reactant repopulation. Indeed, only a small number of trajectories redistributes the vibrational energy along the torsional mode that could drive the system to the fully rotated S 1 /S 0 CI (∼90°), triggering the isomerization. We demonstrate that push−pull substituents mitigate this situation, leading to a behavior from the ππ* state (bright) that is similar to that of the productive nπ* state (dark). Indeed, the substituents induce a ππ* red shift, bringing the bright state  The Journal of Physical Chemistry A pubs.acs.org/JPCA Article applications. Moreover, the large number of trajectories is a strong statistical support to finally assign the photoisomerization process exclusively to the torsion mechanism, even if it is assisted by large CNN/NNC bending motions. 11,12,14,48,53 Indeed, structures distorted enough to support a photoisomerization driven by the inversion route are never reached (Scheme 1). Therefore, only the torsion is the productive path, while the pure bending is an unproductive reaction coordinate, justifying the lower QY observed in AB when exciting the ππ* state (bending-dominated) as compared to direct nπ* excitation (torsion-driven). Because of the importance of the embedding on the excited-state dynamics, 54−56 QM/MM studies are currently undergoing to disclose effects of solvent polarity and viscosity on the photoactivity of these systems.
Computational details, vertical excitation energies and dipole moments of AB push−pull derivatives, nature of TD-DFT and RASPT2 ESs, charge distribution in the FC geometry, details on TD-DFT dynamics, RASPT2 energies at S 1 /S 0 crossings (0 K dynamics), and Cartesian coordinates of the B3LYP/DFT/6-31G*optimized GS minima (PDF)