EPR and pulsed ENDOR study of intermediates from reactions of aromatic azides with group 13 metal trichlorides

Summary The reactions of group 13 metal trichlorides with aromatic azides were examined by CW EPR and pulsed ENDOR spectroscopies. Complex EPR spectra were obtained from reactions of aluminium, gallium and indium trichlorides with phenyl azides containing a variety of substituents. Analysis of the spectra showed that 4-methoxy-, 3-methoxy- and 2-methoxyphenyl azides all gave ‘dimer’ radical cations [ArNHC6H4NH2]+• and trimers [ArNHC6H4NHC6H4NH2]+• followed by polymers. 4-Azidobenzonitrile, with its electron-withdrawing substituent, did not react. In general the aromatic azides appeared to react most rapidly with AlCl3 but this reagent tended to generate much polymer. InCl3 was the least reactive group 13 halide. DFT computations of the radical cations provided corroborating evidence and suggested that the unpaired electrons were accommodated in extensive π-delocalised orbitals. A mechanism to account for the reductive conversion of aromatic azides to the corresponding anilines and thence to the dimers and trimers is proposed.

To help in elucidating the mechanisms of these reductions, we used CW EPR spectroscopy and attempted to characterise the reactive intermediates in selected reactions involving gallium trichloride. Surprisingly, we found that treatment of phenyl azide and 4-methoxyphenyl azide with gallium trichloride resulted in strong EPR spectra of long-lived paramagnetic species. By combining the results of product analyses with the results of EPR spectroscopy, we were able to show that persistent radical cations of 'dimers' (4-aminodiphenylamines) and 'trimers' (4′-phenylamino-4-aminodiphenylamines) were being formed [31]. We have now broadened the scope of this investigation to aromatic azides with a range of functionality. We report here our findings on the behaviour of aromatic azides when treated with the group 13 trichlorides of gallium, indium and aluminium.

Results and Discussion
Reaction of 4-methoxyphenyl azide (2) with group 13 metal chlorides A set of aromatic azides, each containing an electron-releasing or an electron-withdrawing substituent in the 4-position, was chosen for this study. The position of the substituent was also varied and several other azide types were included (Scheme 1). Each organic azide was reacted with the metal halide in dichloromethane/pentane or acetonitrile solution at rt, and an aliquot (~0.1 mL) was placed in a quartz capillary tube (diam 1 mm), purged with nitrogen for 15 min and transferred to the resonant cavity of an X-band EPR spectrometer. When either AlCl 3 , or GaCl 3 or InCl 3 was used, the reaction was accompanied by copious evolution of gas (probably nitrogen) and a deep blue or violet colour usually developed immediately or within a few minutes. In the case of AlCl 3 the reactions were very vigorous. Previously, we showed that the main product from the reaction of 4-methoxyphenyl azide 2 with GaCl 3 was 4-amino-4′-methoxydiphenylamine (11b, Variamine blue), together with traces of anisole, oxidised derivatives (including 4-(4-methoxyphenylamino)phenol, 4-((4-methoxyphenyl)imino)-cyclohexa-2,5-dienone) and much dark-coloured polymer [31]. The EPR spectrum showed the radical cation of Variamine blue (11b +• ) plus broad signals which we attributed to oligomer and/or polymer radical cations (Scheme 2). When anhydrous AlCl 3 in DCM -instead of the gallium halide -was added to a solution of azide 2, a vigorous reaction took place. The resulting deep-coloured solution was transferred to the EPR spectrometer and initially the spectrum, Figure 1a, was obtained at 300 K. The broad, poorly resolved signal suggested that the mixture was dominated by polymeric material. However, when the solution was cooled down to 220 K, the well-resolved spectrum, Figure 1b, was obtained. The resolution improvement may be due to the fact that most of the polymer separates from the solution at the lower temperature.
A good computer simulation was achieved by utilising the hyperfine splitting constants (hfs) listed in Table 1. A wellresolved EPR spectrum of 11b +• , generated from 2 with GaCl 3 , is shown in Figure 1c for comparison. Figure 1d shows the experimental spectrum obtained from treatment of 2 with InCl 3 , together with the corresponding computer simulation. Although the three EPR spectra appear different at first sight, the hfs derived from the simulations (Table 1) are actually quite similar. The contrasts in the spectral appearances are mainly the result of different line widths with consequently different resolutions. It is evident that the main species in each case is the radical cation 11b +• . The acceptable agreement between the DFT-computed isotropic hfs of 11b +• (Table 1) and the experimental data provides additional support for this identification.  Table 1. (c) GaCl 3 in DCM/pentane at 300 K with the computer simulation below. (d) InCl 3 in DCM:CH 3 CN; 4:1 at 320 K; computer simulation below.
The small differences in the hfs obtained for the different group 13 metal chlorides can probably be attributed to the different counter ions, solvents and temperatures.
The 4-methoxy-tetradeuterio-azide 3 was also treated with AlCl 3 in DCM, and the resulting spectrum and simulation are shown in Figure 2.
The unpaired electron interacts with two non-equivalent N-atoms, a single comparatively large H-atom and a pair of equivalent H-atoms. The spectrum obtained previously on treatment of 3 with GaCl 3 was better resolved (Table 1) [31]. However, it is clear that the same 'dimer' species was formed with AlCl 3 , probably having picked up the NH and NH 2 hydrogen atoms from the solvent. The line width of the spectrum with AlCl 3 was ca. 0.7 G. Therefore, it is not surprising that hfs from aromatic ring D-atoms were not resolved. Again, differences in the hfs of the spectra from AlCl 3 and GaCl 3 can be attributed to the different counter ions.  Reactions of phenyl azide and 4-substitutedphenyl azides with group 13 metal chlorides Azides 1, 6 and 7 were chosen to vary the electronic properties and leaving group abilities of 4-substituents. We showed previously that treatment of phenyl azide 1 with GaCl 3 gave well-resolved spectra of 4-aminodiphenylamine radical cation (11a +• , the dimer) and of the trimer under different reaction conditions [31]. On treatment with InCl 3 , 1 gave little sign of reaction. No colour developed and no EPR spectra were obtained. However, a vigorous reaction took place between 1 and AlCl 3 with nitrogen evolution and development of a deep blue colour. The EPR spectrum, Figure 3a, was dominated by a broad component, probably due to polyaniline type material, together with some fine structure. The second derivative spectrum at low modulation amplitude discriminated against the broad signal, and spectrum, Figure 3b, was obtained after digitally removing the residual broad component. The hfs were similar to those of 11a +• ( Table 1, entry for 1 with GaCl 3 ) except that the smaller hfs were not resolved. Minor differences in the magnitudes of the hfs can be attributed to the different counter ions. The trimer radical cation was not observed, but clearly a contribution from this species could be hidden under the broad component.
No reaction of 4-azidobenzonitrile 7 with InCl 3 , GaCl 3 or AlCl 3 was observed and no paramagnetic species were detected by EPR spectroscopy. It appears the electron-accepting property of the CN group inhibited the coupling process at some stage. It is also worth mentioning that, as expected, aliphatic azides such as ethyl 5-azidopentanoate 9 did not react in the same way either. Treatment of 9 with InCl 3 or GaCl 3 led to gas evolution but no colour developed and no paramagnetic species could be detected.
Very interesting results were obtained from reactions of 1-azido-4-fluorobenzene (6). When 6 was treated with GaCl 3 in DCM, a deep blue-violet colour developed and the spectrum was dominated by a broad feature, Figure 4a, although underlying fine structure was evident. When dichlorogallium hydride, prepared from GaCl 3 and Et 3 SiH in CH 3 CN, was used to promote the reaction, a beautifully resolved spectrum resulted, Figure 4b.
The good simulation of this spectrum, Figure 4c, enabled the hfs shown in Table 1 to be determined. Comparison of these hfs  Table 2.
with those of the other species in Table 1 supports the identification of this intermediate as the corresponding dimer cation 11c +• , containing a single F-atom. The DFT computation on the dimer hfs gave satisfactory agreement (Table 1), with the possible exception of the para-F hfs. However, DFT-computed a(F) values varied from 7.2, to 7.1, to 9.5 and 4.3 G with 6-31G(d), 6-31+G(d,p), DGDZVP and 6-311++G(d,p) basis sets, respectively. This spread indicates the comparative unreliability of the DFT spin density computations for F-atoms in these cations.
Treatment of 6 with InCl 3 in DCM led to the usual broad signal from oligomeric and polymeric species superimposed on a spectrum with much narrower lines. On recording the spectrum with a smaller modulation amplitude, and digitally removing the residual broad feature, the spectrum shown in Figure 4d resulted. This is obviously a different species from that of Figure 4b and, after many trials, a satisfactory simulation was obtained, Figure 4e. The derived hfs are presented in Table 2 and they clearly correspond to a trimer, also probably containing a single F-atom, i.e. 19b +• .
It seems clear that the MCl 3 reactions with aromatic azides entail a progression from the aniline XC 6  Reactions of 2-methoxy-and 3-methoxyphenyl azides with group 13 metal chlorides Aromatic azides 4 and 5 were chosen to investigate how the position of the MeO substituent influenced the reaction. Treatment of the 3-methoxy precursor 4 with InCl 3 or HInCl 2 gave only very weak and broad EPR spectra. However, reaction of 4 with GaCl 3 in CH 3 CN gave a strong EPR spectrum and the hfs derived from the computer simulation are presented in Table 1.
The comparatively large line width (~0.9 G) did not permit the resolution of small hfs from aromatic ring H-atoms. For the same reason, the hfs from the second N-atom were not resolved. However, it is clear that this species is probably a 'dimer' although the connectivity of the angular structure 17a +• is somewhat different from that of the 4-aminodiphenylamines derived from the 4-substituted phenyl azides (Scheme 3).

Scheme 3: Dimer and trimer radical cations.
Treatment of the 2-methoxy precursor 5 with GaCl 3 in DCM, or with HGaCl 2 in CH 3 CN, gave essentially the same strong spectrum, see Figure 5a and Supporting Information. The hfs derived from the simulations (Table 1) suggest that this is also a dimer type radical cation 17b +• . However, with the passage of time a central peak began to appear in this spectrum. When 5 was treated with HGaCl 2 , prepared from GaCl 3 and Et 3 SiH, the species with a central peak dominated the spectrum, Figure 5b. Treatment of 5 with InCl 3 in DCM or with HInCl 2 in THF (prepared from InCl 3 and DIBAL-H) also gave rise to a spectrum of this same species, Figure 5c. A well-resolved spectrum of this species was obtained by treatment of 5 with HGaCl 2 prepared with Et 3 SiH in CH 3 CN, Figure 5d. The hfs derived from the computer simulation, Figure 5e and Table 2 show the presence of three N-atoms and of four H-atoms with sizeable hfs that can probably be attributed to NH or NH 2 groups. Thus, this species is almost certainly a 'trimer' although this will necessarily have an angular structure 18 +• rather than the linear type structure of the trimers from 4-subsitituted azides such as 19 +• (Scheme 3).
The results from azides 4 and 5 showed that the position of the MeO substituent in the phenyl azides was not critical. The reactions with gallium and indium promoters proceeded along similar lines to that of phenyl and 4-substituted phenyl azides to give dimers, trimers and polymers.
The spectra obtained on treatment of 2-azidonaphthalene 8 with InCl 3 , GaCl 3 and AlCl 3 are shown in Figure 6a, Figure 6b and Figure 6c, respectively.
The broad signal in Figure 6c shows that polymerisation dominated the reaction with AlCl 3 . Similarly, the main broad feature in Figure 6b suggests that polymerisation was again dominant in the reaction with GaCl 3 . The comparatively well-resolved species observed in the InCl 3 -promoted reaction, Figure 6a, was well simulated on using the parameters shown in Table 1. The data show that the unpaired electron interacted with two

N-atoms, two (N)H 2 atoms, one (N)H-atom and one other
H-atom; other splittings were not resolved. The magnitudes of the hfs are somewhat smaller than those of analogous atoms in the dimer from 1. This is exactly as would be expected from the greater extent of aromatic delocalisation in a dimer from 8. Clearly, however, more than one isomer is possible.

Pulse ENDOR spectrum of the intermediate from 4-fluorophenyl azide 6
Pulsed ENDOR experiments, based on the ESE effect, were carried out on the frozen solution from azide 6 at 50 K. The echo signal was created by the microwave pulse sequence, and an rf pulse was applied during the 'mixing period', which corresponded to the time T in the Davies ENDOR sequence [32]. The rf pulse drove the nuclear spin transitions, which led to a change in the ESE intensity. The ENDOR signal was therefore measured by monitoring the ESE intensity while the rf frequency was varied. In the case of an S = ½ system coupled with a nucleus with nuclear spin I = ½, the Davies ENDOR spectrum consists of two lines at the nuclear resonance frequencies ν α and ν β , which correspond to the transitions associated with the electron spin manifolds M s = +½ and M s = −½, respectively. If the Larmor frequency (ν n ) of the nucleus in question is larger than the hyperfine interaction, then the resonance frequencies are given by: If ν n is less than ½a iso , the frequencies are then given by: An additional complication arises if the nuclear spin is >½, which adds another term describing the nuclear quadrupole interaction in the above equations [33]. In frozen solution all orientations of the paramagnetic species are observed and therefore an anisotropic ENDOR spectrum is expected. The latter is more complex and requires a detailed understanding of the anisotropy of the system. The above equations are not suitable for such a situation and a more complete resonance condition that considers all the orientations is needed. In the case of a system with I = ½ the parameter a iso in Equation 1 and Equation 2 is replaced by A i (i.e. one of the principal components of the hyperfine tensor).
The Davies ENDOR spectrum from the species derived from the 4-fluoroazide 6 sample at 50 K is shown in Figure 7. The inset shows the ESE-EPR spectrum, with an arrow indicating the magnetic field position at which the ENDOR experiment was performed. The ENDOR spectrum shows powder pattern lineshapes, as expected for frozen solutions, due to the anisotropic hyperfine interactions. Two main features cover the whole spectrum; a powder pattern centred about the 1 H Larmor frequency and a second broad signal located at lower frequency and spread over 8 MHz width. The lack of resolution encountered in the ENDOR spectrum makes an unequivocal analysis difficult. Therefore, our ENDOR simulation was based mainly on the CW EPR results. A simulated spectrum is displayed in Figure 7, and a deconvoluted version is in the Supporting Information together with one chosen set of ENDOR hyperfine tensor parameters. Almost all the 1 H hyperfine splittings fit well within the ENDOR spectra, but it is worth noting, as mentioned above, that a satisfactory simulation was only achieved by assuming an anisotropic lineshape of the hyperfine couplings. Extra weak hyperfine couplings, unresolved in the CW EPR, were also included in the simulation. These probably correspond to hyperfine coupling in polymer which was undoubtedly present. The broad feature at low frequency is related to a mixture of fluorine ( 19 F) and nitrogen ( 14 N) contributions. The anisotropy and the unresolved nuclear quadrupole of the nitrogen couplings make the spectra difficult to interpret. Each 1 H contributes three sets of peaks to the spectrum times the number of 1 H's present. This represents an enormous number of lines in one spectrum. Obviously, they cannot all be assigned from this broad unresolved powder pattern. Almost axial tensors were assumed (see Supporting Information). However, it should be noted that it may well be possible to simulate these spectra with other parameter sets. The experimental Davies ENDOR data support the CW EPR data in confirming the magnitudes of the hyperfine couplings and the nitrogen interactions. Further pulse techniques such as electron spin echo envelope modulation (ESEEM) and its multidimensional extension Hyscore would be required to get more insight into the nitrogen contribution. Regarding the 19 F contribution; only the low-frequency part of the 19 F coupling fits well with the experimental data. The 19 F high-frequency line in our simulation is not consistent with the experimental spectrum, which suggests that the latter might be highly asymmetric. Such situations have been previously reported in other studies where it was shown that this could be related to the relaxation time. Sometimes relaxation processes can lead to a partial saturation in the nuclear transitions such that the observed signal is the result of a transition in one manifold only [34]. Partial saturation may explain the absence of the 19 F high-frequency line in our spectrum.

DFT computations of radical cation properties
Quantum chemical calculations were carried out with the Gaussian 03 programme package [35,36]. Density functional theory with the UB3LYP functional was employed. The equilibrium geometries were fully optimised with respect to all geometric variables, no symmetry being assumed either with the 6-31+G(d,p) basis set (dimers) or with the 6-31G(d) basis set (trimers). Isotropic EPR hfs were derived from computed Fermi contact integrals evaluated at the H-and N-nuclei. The hfs were taken directly from the Gaussian output files and are shown in Table 1 and Table 2.
The optimum structures of the radical cations 17a +• and 17b +• their associated SOMOs are shown in Figure 8.
The C-NH bond lengths in radical cation 17a +• (1.40 and 1.38 Å) and 17b +• (1.38, 1.39 Å) indicated significant double bond character. The CNC angles in 17a +• and 17b +• were 128.9 and 131.6°, showing significant widening from trigonal. The aromatic rings in all the structures were twisted significantly out of co-planarity. As might be expected on steric grounds, this increased in the dimers as the substitution site moved from the 4-to the 3-to 2-position. For example, the computed dihedral angles between the rings increased from 26.8 to 30.7 to 39.0 in the series 4-MeO-11b +• , 3-MeO-17a +• and 2-MeO-17b +• , respectively. It seems that a compromise was reached in which the repulsive steric interaction between substituents of neighbouring rings was balanced against the stabilising effect from conjugation of the π-systems. The SOMOs depicted in Figure 8 show that there was still sufficient orbital overlap in the linear and angular dimer and trimer radical cations to support lengthy π-systems extending over all the rings and N-atoms. This is in accordance with the EPR spectroscopic data, that show extensive delocalisation of the unpaired electron in dimer and trimer radical cations. The computed hfs in Table 1 and Table 2 show reasonable correspondence with the experimentally observed values.

Conclusion
Literature reports show that anilines can easily be oxidised to the corresponding resonance-stabilised radical cations, which can couple with more aniline to afford very persistent radical cation dimers [37,38]. The generation of these radical cations depends critically on the reaction conditions, in particular on the degree of protonation, which can facilitate electron transfer (ET) [39,40]. It has also been reported that electrochemical oxidation of aromatic amines can generate the same radical cations which can polymerise giving oligo-and poly-anilines [41]. In view of the fact that product analyses [31] identified aniline amongst the products from 1 and anisole amongst the products from 2, it seems probable that the aromatic amines are the precursors of the dimer and trimer species.
A possible mechanism for production of anilines from the aromatic azides is set out in Scheme 4. Coordination of the metal halide to the starting azide should produce the Lewis base-acid adduct 12 that could undergo reduction by ET from more azide to afford, after nitrogen loss, the metal-coordinated aminyl radical 13 together with the ArN 3 +• radical cation.
Aminyl radical 13 could then abstract an H-atom from solvent RH (or from HMCl 2 when the metal hydrides were used) with the production of metal-coordinated amine 14. The latter can then pick up a proton to produce an aromatic amine and regenerate the metal halide. The reason 4-azidobenzonitrile 7 did not react with any of the group 13 metal chlorides may well be that the ET step 12 → 13 was inhibited by the presence of the electron acceptor CN group. course, proton transfer could occur earlier in the reaction, such that coupling takes place with the anilinium cation instead. Trimer 18 +• could be produced by coupling of 17 +• with more 13a followed by a similar sequence of steps. The trimer could then grow into oligomer and polymer by a succession of such coupling reactions.
In general, the aromatic azides appeared to react most rapidly with AlCl 3 but this reagent tended to generate much polymer. InCl 3 was the least reactive group 13 halide such that no reac-tion was observed with PhN 3 and very little reaction occurred with the 3-methoxyazide 4. GaCl 3 and HGaCl 2 were the best promoters. The dimers were the main products from reactions of excess 1 and 2 with GaCl 3 such that the process could have synthetic potential. In general, the GaCl 3 -and HGaCl 2promoted reactions were also best for spectroscopic studies because they gave the most intense and well-resolved spectra of [ArNHArNH 2 ] +• and/or {[ArNH] 2 ArNH 2 } +• with the narrowest line widths.

Experimental
EPR and ENDOR spectroscopy. EPR spectra were obtained with a Bruker EMX X-Band 10/12 spectrometer fitted with a rectangular ER4122 SP resonant cavity and operating at 9.4 GHz with 100 kHz modulation. An aliquot (~0.1 mL) of the reaction mixture from each aromatic azide and the metal chloride in CH 2 Cl 2 /pentane or CH 3 CN solution was placed in a 1 mm o.d. quartz capillary tube, de-aerated by bubbling nitrogen for 20 min and transferred to the resonant cavity. Spectra were examined at several temperatures but generally best resolution and signal intensity were obtained at around 300 K. Most of the EPR spectra were recorded with 2.0 mW power, 1.0-0.2 G pp modulation intensity and a gain of ~10 6 . In all cases where spectra were obtained, hfs were assigned with the aid of computer simulations using the Bruker SimFonia and NIEHS Winsim2002 software packages.
Pulsed EPR and ENDOR were performed using a pulsed EPR X-band spectrometer (Bruker Elexsys E580) equipped with a Dice-ENDOR accessory, a radio frequency (rf) amplifier and a dielectric-ring ENDOR resonator (Bruker EN4118X-MD-4-W1). Samples were maintained at 50 K using liquid helium in an Oxford CF-935 cryostat. Field-swept electron spin echo (ESE) spectra were recorded using a two-pulse ESE sequence while ESE-ENDOR experiments were carried out using Davies three-pulse sequence π-T-π/2-τ-π-echo with a selective rf pulse of variable frequency applied during time T. The pulse lengths used were 128 and 256 ns for π/2 and π respectively, and 10 μs for the π-rf pulse. ENDOR data were processed and simulated using the EasySpin package (freeware from http:// www.easyspin.org/).
General procedure for the reaction of aryl azides with indium trichloride. The starting azide (1 mmol) was added at 0 °C to an acetonitrile solution of indium trichloride (1.1 mmol) in DCM (4 mL) and stirred for 5 min at 0 °C. Gas was evolved and the solutions took on a dark blue or violet colour. The resulting solutions were rapidly transferred into a quartz capillary tube and purged with nitrogen for few minutes. The tube was sealed and placed in the EPR resonant cavity. Spectra were recorded at several different temperatures. Some samples were photolysed with a 500 W super pressure Hg arc.
General procedure for the reaction of aryl azides with dichloroindium hydride. The starting azide (1 mmol) was added at 0 °C to an acetonitrile solution of dichloroindium hydride (1.1 mmol), generated in situ by stirring under an argon atmosphere anhydrous indium trichloride (243 mg, 1.1 mmol, previously dried by heating at 130 °C under argon for 1 h) and triethylsilane (177 μL, 1.1 mmol) in ACN (4 mL) for 5 min at 0 °C [47]. The resulting solution was rapidly transferred into a quartz capillary tube and nitrogen was bubbled inside for few minutes. The tube was sealed and placed in the EPR cavity. Spectra were recorded at several different temperatures. Some samples were photolysed with a 500 W super pressure Hg arc. Selected samples were given an aqueous work-up with NaHCO 3 followed by extraction with diethylether. In each case the corresponding aromatic amine was identified by comparison with literature data.
General procedure for the reaction of aryl azides with AlCl 3 . Aluminium trichloride (1.1 mmol) was dried under reduced pressure at 25 °C for 1 h. Then DCM (3 mL) was added and a DCM solution of the azide (1 mmol in 1 mL) was introduced at rt. Gas was evolved, sometimes violently, and dark blue or violet colours developed. The resulting solution was then transferred to a capillary quartz tube and purged with nitrogen. The capillary was sealed and several EPR spectra were run at different temperatures. Product analysis was performed by quenching the reaction with an aqueous solution of NaOH and extracting with DCM. The mixtures were analysed by GC-MS and, when possible, by 1 H NMR and 13 C NMR spectroscopy.
General procedure for the reaction of aryl azides with GaCl 3 . A pentane solution of gallium trichloride (0.55 mL of 0.5 M; 0.28 mmol) was added under a nitrogen atmosphere to a DCM solution of the azide (0.25 mmol in 4 mL) at rt. Gas was evolved and an intense blue or violet colour developed. The resulting solution was then transferred into a capillary quartz tube and purged with nitrogen. The capillary was sealed, and the sample was analysed by EPR spectroscopy at several different temperatures. Products analysis was performed as above.

Supporting Information
Supporting information features general procedures, EPR spectra from azides 4 and 5, deconvolution of ENDOR spectrum from azide 6, Cartesian coordinates for DFT-computed structures of dimer and trimer radical cations.