Fluorinated azobenzenes as supramolecular halogen-bonding building blocks

ortho-Fluoroazobenzenes are a remarkable example of bistable photoswitches, addressable by visible light. Symmetrical, highly fluorinated azobenzenes bearing an iodine substituent in para-position were shown to be suitable supramolecular building blocks both in solution and in the solid state in combination with neutral halogen bonding acceptors, such as lutidines. Therefore, we investigate the photochemistry of a series of azobenzene photoswitches. Upon introduction of iodoethynyl groups, the halogen bonding donor properties are significantly strengthened in solution. However, the bathochromic shift of the π→π* band leads to a partial overlap with the n→π* band, making it slightly more difficult to address. The introduction of iodine substituents is furthermore accompanied with a diminishing thermal half-life. A series of three azobenzenes with different halogen bonding donor properties are discussed in relation to their changing photophysical properties, rationalized by DFT calculations.


S7
Irradiation of A2 and A3 was also performed in DCM and cyclohexane. Peak positons were determined using the OriginPro Peak Analyzer tool. The evolution of the absorbance at 345 nm (for A1) and 380 nm (for A2 and A3) was observed at 60 °C and fitted to a monoexponential fit.
S9 Figure S5. Thermal E → Z isomerization of A2 in MeCN at 60 °C. S10 Figure S6. Thermal E → Z isomerization of A3 in MeCN at 60 °C. The thermal stability of A3 was determined at different temperatures.

IV. Computation details
Quantum mechanical calculations were performed by applying density functional theory and the polarizable continuum model [4] (PCM) for the description of solvent effects. All geometry optimizations were performed with the Gaussian 16 program. [5] For the computation of absorption properties, TD-DFT methods were used. The exact procedures are described in the corresponding sections.

IV.1 Method evaluation and benchmark
To find a suitable method for the computation of thermal Z→E isomerization barriers and corresponding half-lives in implicit PCM solvent, we first performed a series of benchmark calculations on fluorinesubstituted azobenzenes (ABs). We tested a variety of functionals including pure, and hybrid approaches.
The computed data were compared with values for fluoro-ABs in MeCN, measured at 60°C. [6] Structures of the investigated systems are given in Figure S9. Note that we have slightly adapted the naming convention of Hecht and co-workers, [6] to match the naming of the fluorine/iodine target systems, thus F2 corresponds to F4 in the original paper, F3 is F6 etc. Figure S9. Azobenzene and fluorinated derivatives used for benchmarking. [6] The transition state (TS) was modeled by performing constrained geometry optimizations along the CNN angle starting from the Z-isomer of the corresponding azobenzene. The point of highest energy along this coordinate was used as the initial point for TS optimizations with the B3LYP method, the 6-311G* basis set and MeCN as implicit solvent. The stationary nature of TSs, educts and products was identified through vibrational analyses, yielding one imaginary frequency for TSs (380i-446i) and only positive eigenvalues for Zand E-structures. Intrinsic reaction coordinate (IRC) computations validated that the TS in fact leads to the corresponding Z-educt and E-product. The resulting geometries were then used as the starting point for structure optimizations with the tested functionals. The structures for Z-, E-isomers S14 and TS are very similar for the investigated systems. They are shown in Figure S10 by B3LYP/6-311G* optimized structures obtained for F2 as example. The optimized E-structures are not planar but significantly twisted among the CN axis and slightly distorted along CNNC dihedrals (20° and -177°). reported, e.g., by Knie et al., [6] Ritze et al., [7] and Liu et al. [8] Thermal rate constants and half-lives were calculated using the KistHelp software package. [9] Computations involved conventional transition state theory (TST) considering also the effect of one-dimensional Wigner tunneling. [9]   As the target molecules in this study include heavy and large iodine atoms, dispersion corrections were applied for all methods in the test set, except for one CAM-B3LYP reference calculation including dispersion (see Table S3). Comparison with the data from Knie et al., [6] who used the same set of molecules without including dispersion corrections, reveals that dispersion effectively increases the thermal barriers, thus leading to significantly longer half-lives (ca. 1-2 orders of magnitude). Long-range corrections, as, e.g., included in CAM-B3LYP or the wB97XD functionals tend to largely overshoot in our case. A similar study by Rietze et al. [7] indicates that this effect is likely an artifact caused by the PCM solvation, as it does not appear in gas-phase calculations. In further computations, we therefore did not include long-range corrections, neither for geometry optimizations nor for calculation of thermal barriers.
For the computation of rate constants and thus half-lives we shall expect an accuracy of ca. 1 order of magnitude. [10] The hybrid functionals B3LYP, TPSSh and PBE0 all lie within this range and reflect the trend also found in the experiments, namely F3 > F2 > F4 > F2-est > AB, considering half-lives. The same trend is also found for the pure functionals, but with significantly shorter half-lives as in the hybrids.
While the values for F2-F4 all lie within chemical accuracy, the times for AB and F2-est are underestimated in some cases. F4 half-lives are overestimated in the hybrids, and underestimated in all pure functionals. Apparently, none of the applied DFT methods yields a fully satisfying result that describes all systems on equal footing when PCM solvation is used. Overall, B3LYP is the only functional among the hybrids that delivers a suitable balance between F2 and F3, while TPSSh and PBE0 attribute a much larger relative half-life to F3. Among the pure functionals, M06L appears to be the most suitable in terms of balance, giving even the best match in absolute values for F2-F4. Based on these tests, we decided to use B3LYP-D3 and M06L-D3 approaches to compute the corresponding thermal data for the iodine/fluorine compounds in this study. S16

IV.2 Optimization of fluorine/iodine-azobenzenes
For structure optimization of the target systems A1-3, we adopted a similar strategy as described in the benchmark study. Geometries were optimized using B3LYP and M06L functionals with Grimme D3 dispersion corrections and the def2-TZVP basis set for H, C, N and F atoms. The all-electron basis DGDZVP was used to describe iodine. MeCN solvation was considered through the PCM model.

S17
Tables S5 and S6 report selected structural parameters for both methods. For the E-isomer, we find two structures in each model that are very close in energy. (Ediff < 0.04 kcal/mol), see Table S6) E1 corresponds to the optimized structure at the end of the IRC path. These structures are somewhat twisted along C-N bonds (ca. −10°, for M06L and up to −25° in B3LYP, see Table S5). The second minimum E2 is almost planar (C-C-N=N dihedral ≈1.4-2.6°). Z-isomers are slightly twisted along the N=N bond (ca 13-15°), the phenyl rings are tilted (C-C-N=N dihedral ca. 50-53°). Both Zand E-isomers have formal C2 symmetry, while the TS is asymmetric with almost straight connection along C-N=N (≈175°) on one side and almost perpendicular arrangement of the phenyl rings (C-C-N=N ≈70-100°). S18 Table S6. Selected structural parameter in B3LYP computed F/I-ABs. The TS structures are asymmetric; the corresponding values for differing structural parameters are given in italics.

IV.3 Computation of excited state absorption spectra
Absorption energies for B3LYP-D3 optimized geometries were computed using the time-dependent density functional approach (TDDFT) involving 20 roots, with the same basis set and solvent modelling used for optimization. The obtained vertical absorption energies and oscillator strengths were convoluted with a 0.2 eV Gaussian function (FWHM=0.2 eV). The resulting spectra are shown below. Table S9 reports excitation energies and oscillator strengths of the low-lying ππ* and nπ* states in comparison with experimentally obtained values in MeCN solution. Due to the mostly planar structures the nπ* vertical excitation in the E-isomers shows almost no oscillator strength, the corresponding excitation wavelengths are indicated in the UV spectra. The overall trend follows the experimental observations; we note, however, a red shift for all computed excitation energies wrt. to experimental values. Especially the nπ* excitation energies are largely underestimated. Considering long-range interactions with the CAM-B3LYP functional leads to better energies for the ππ* states in the E-isomer, but now these states are overestimated in the Z-isomer (Table S10).
S20 Figure S12. UV-vis absorption spectra computed with TD-B3LYP, obtained by convolution with a Gaussian function of 0.2 eV full-width half maximum (FWHM). The E-isomer is shown in violet, Z in magenta.

IV.4 Computation of electrostatic potentials
To obtain information on the halogen-bonding properties of the systems we computed electrostatic potential (ESP) maps at electron densities of 0.001 and 0.0001 with B3LYP/def2-TZVP and Grimme D3 dispersion corrections. The MoleCoolQt program [11] was used to visualize the results. ESP maps are shown in Figure S13, revealing strong positive potential values at the iodine atoms in the order A3 > A2 > A1.

IV.5 Calculation of thermodynamic data
The KistHelp program [9] was used to obtain thermodynamic properties of F/I-substituted ABs. A reaction path was modelled using the vibrational analyses of Z-isomers and the TS at experimental temperature (333.15 K), employing classical transition state theory (TST) and including Wigner tunneling. [9] Tables S11 and S12 show the resulting data for M06L and B3LYP methods in comparison with the experimental data. The computed half-lives follow the experimental trend, namely A1 > A2 > A3, with M06L obtained values underestimating the lifetimes by ca 1-2 orders of magnitude. B3LYP data are within chemical accuracy.

V. Crystallographic details
Single-crystals were mounted using a microfabricated polymer film crystal-mounting tool (dual-thickness MicroMount, MiTeGen) using low viscosity oil (perfluoropolyalkylether; viscosity 1800 cSt, ABCR) to reduce the X-ray absorption and scattering. A Bruker D8 Venture single-crystal X-ray diffractometer with area detector using Cu Kα (λ = 1.54178 Å) radiation was used for data collection at the temperature stated for each compound. Multiscan absorption corrections implemented in SADABS [12] were applied to the data. The structures were solved by intrinsic phasing (SHELXT-2013) [13] and refined by full-matrix leastsquares methods on F 2 (SHELXL-2014). [14] The hydrogen atoms were placed at calculated positions and refined by using a riding model. CCDC 1936418 (U1⋯A2) contains the supplementary crystallographic data for this paper. These data can be obtained free of charge from The Cambridge Crystallographic Data Centre.