Exciton Delocalization Counteracts the Energy Gap: A New Pathway toward NIR-Emissive Dyes

Exciton coupling between the transition dipole moments of ordered dyes in supramolecular assemblies, so-called J/H-aggregates, leads to shifted electronic transitions. This can lower the excited state energy, allowing for emission well into the near-infrared regime. However, as we show here, it is not only the excited state energy modifications that J-aggregates can provide. A bay-alkylated quaterrylene was synthesized, which was found to form J-aggregates in 1,1,2,2-tetrachloroethane. A combination of superradiance and a decreased nonradiative relaxation rate made the J-aggregate four times more emissive than the monomeric counterpart. A reduced nonradiative relaxation rate is a nonintuitive consequence following the 180 nm (3300 cm–1) red-shift of the J-aggregate in comparison to the monomeric absorption. However, the energy gap law, which is commonly invoked to rationalize increased nonradiative relaxation rates with increasing emission wavelength, also contains a reorganization energy term. The reorganization energy is highly suppressed in J-aggregates due to exciton delocalization, and the framework of the energy gap law could therefore reproduce our experimental observations. J-Aggregates can thus circumvent the common belief that lowering the excited state energies results in large nonradiative relaxation rates and are thus a pathway toward highly emissive organic dyes in the NIR regime.


General Synthesis
All starting materials were purchased from Sigma-Aldrich Chemical Co. and used without further purification unless otherwise noticed. All moisture-and oxygen-sensitive reactions were carried out using Schlenk techniques in oven-dried glassware. Solvents used for moisture and oxygen-sensitive reactions were dried using a PS-MD-5/7 Inert technology purification system and, if necessary, degassed by freeze-pump-thaw cycles and stored over 4 Å molecular sieves under argon atmosphere. 1 H NMR (nuclear magnetic resonance) spectra were recorded on Bruker spectrometers at 900 MHz, 800 MHz or 600 MHz, 13 C NMR spectra were recorded at 201 MHz using CDCl 3 , CD 2 Cl 2 or 1,1,2,2tetrachloroethane-d 4 (C 2 D 2 Cl 4 ). All three magnets were equipped with a Bruker AVENCE III-HD console. The 900 MHz magnet had a 3 mm TCI H-C/N-D cryoprobe inserted while the 800 was equipped with a 5 mm TXO C/N-H-D, 13 C optimized cryoprobe. The 600 MHz magnet had a room temperature BBI probe inserted. The maximum current of the Z-gradient was 10 A in all the systems.
J-coupling values are given in Hertz, and chemical shifts are given in parts per million using tetramethylsilane, with 0.00 ppm as an internal standard. High-resolution MS was obtained from an Agilent 1290 infinity LC system equipped with an auto sampler in tandem with an Agilent 6520 Accurate Mass Q-TOF LC/MS.

General optical spectroscopy
The absorbance spectra were measured using a Perkin Elmer LAMBDA 950 spectrophotometer. Molar absorptivities were calculated as an average of three individual measurements. About 1 mg was weighed out in each measurement and then dissolved in either toluene (monomer) or TCE (J-aggregate).
Emission measurements were performed using an FLS1000 (Edinburgh Instruments) spectrofluorometer. Emission quantum yields were measured using the relative method using quinine sulfate (Sigma-Aldrich) in a 0.1-N solution of H 2 SO 4 (quantum yield of quinine sulfate, 0.59 1 ) as an internal standard for perylenes, Oxazine 1 (Sigma-Aldrich) in ethanol solution (quantum yield of Oxazine 1, 0.15 1 ) as internal standard for dihexylquaterrylene monomers and IR-140 (Sigma-Aldrich) in ethanol solution (quantum yield of IR-140, 0.167 2 ) as internal standard for dihexylquaterrylene Jaggregates. Emission quantum yields are reported as an average of three samples. The fluorescence lifetime of the monomer was determined using time-correlated single-photon counting using an Edinburgh Instruments LifeSpec II equipped with an MCP-PMT detector and a 660 nm pulsed diode laser (PicoQuant) as excitation source. The fluorescence lifetime of the J-aggregate was determined using a streak camera (C5680, Hamamatsu) equipped with a spectrometer (Acton SP2300, Princeton Instruments) and a synchroscan sweep unit (M5675, Hamamatsu). The excitation source was a picosecond pulsed, 80 MHz, Ti:sapphire laser (Tsunami, Spectra-Physics) pumped by a Millennia Pro X laser (Spectra-Physics). Emission decay traces at various wavelengths were obtained from the streak camera images by integrating over narrow wavelength ranges of 5-10 nm. The decay traces were fitted to single exponential decays using instrument response deconvolution. The instrument response function (IRF) was obtained by fitting the time profile of the scattered light from the laser pulse to a gaussian function, yielding an IRF with an FWHM of approximately 40 ps. All sample concentrations, for both emission and lifetime measurements, were set to a maximum of 0.1 absorbance unit to avoid inner filter effects.

General AFM and SEM
The diameter and size of aggregates was measured by atomic force microscopy (AFM). A solution of the aggregates was drop casted on a silicon wafer. After evaporation of the solvent, the AFM scan was conducted on a NT-MDT NTEGRA AFM in tapping mode using a silicon cantilever (Tap150Al-G from Budget Sensors). By the same sample preparation, the aggregates were observed under scanning electron microscopy (SEM) using a Zeiss Ultra 55 scanning electron microscope with an acceleration voltage of 15 kV.

General diffusion NMR and diffusion coefficient simulations
The diffusion coefficients were measured with the ledbpgp2s 3 (available in default Bruker library) and with the PROJECTED 4 exchange suppressed pulse program. Due to the shorter T 2 relaxation time of the aggregate the PROJECTED sequence can only be employed to measure the methyl group in dihexylquaterrylene. The gradient was set to the smoothed square shape (SMSQ10.100), which is included in the Bruker gradient shape library. The pulse program was extended to include multiple solvent suppression. The gradient calibration constant (GCC) was determined using a doped water sample. The big delta (d20) was set to 50 ms in the experiments recorded with the ledbpgp2s pulse sequence expect for the high concentration C 2 D 2 Cl 4 sample where it was set to 60 ms and to 20 ms (d2=40 ms) in the PROJECTED sequence with n=8. The little delta had to be set differently for 1hexylperylene, dihexylquaterrylene and the aggregate. The little delta for the dihexylquaterrylene in the PROJECTED sequence was set to 1.8 ms.
The diffusion data were fitted by a python script using the MINUIT -MINGRAD routine from the iminuit package. The error was assessed in two main ways. Firstly, the standard deviation of the noise of each slice of the pseudo 2D measurement was determined. This served as the error for each point in the integral region. The integral was formed as the sum of intensities within a defined region. The error of the integral was then determined by the square root of the sum of the squared errors. The quality of fit was assessed using the Variance Reduction method. 5 Secondly, the error in the fitted diffusion coefficients were assessed. If multiple peaks were available for the measurement of the same compound then the average was taken and the confidence interval was calculated from the separate measurement rather than fitting them all in the same time. This is only possible if the same diffusion coefficient can be determined using multiple sites in the molecule. This was not fulfilled for the aggregate where only one region can be used. In this case the Monte Carlo error estimation was employed. The fit was repeated 10000 times to "synthetic" data sets. These data sets were generated from the original using the error of the point determined as described above by randomly generating a point with normal distribution around the particular measured integral. The diffusion coefficient then is the average of the 10000 runs.
Diffusion coefficients were simulated using HYDRO++10, 6, 7 HYDROSUB 8 and HYDROPRO. 9 The simulation results were nearly identical for all three softwares, thus only one example (HYDRO++10) is given. Solvent viscosity was specified as 0.00413 Poise 10 for CD 2 Cl 2 and 0.0165 Poise 11 for C 2 D 2 Cl 4 at 25 °C when input files were generated.

Synthesis
a) 1-hexylperylene was synthesized according to the literature procedure developed earlier by our group. 12 NMR shifts and other spectral data was in line with those reported earlier. In addition to the previous study, a full spectral assignment has been carried out for 1-hexylperylene vide infra. b) 1,1'-dihexylquaterrylene Figure S1. The synthesis of 1,1'-dihexylquaterrylene.
After 10 min, DDQ (22.7 mg, 0.1 mmol) and triflic acid (9 µL, 0.1 mmol) were added and the microwave vial was sealed with a cap and sparged with nitrogen for 20 min. The reaction mixture was then stirred overnight at room temperature. The resulting mixture was quenched with diisopropylethylamine (1 mL) and passed through a column containing basic aluminium oxide eluting with (20/80 ratio of toluene/cyclohexene). The collected green fractions were concentrated and purified by a size exclusion chromatography using BioBeads SX-3 as a stationary phase and CH 2 Cl 2 as an eluent.
The collected blue fractions were concentrated under reduced pressure to afford the title compound as a dark purple solid 9.3 mg, 28% yield (85:15 isomeric ratio). The chemical structure of the minor isomer was not assigned due to insignificant amounts.

Spectral assignment and structure determination a) 1-hexylperylene
Spectral assignment was carried our for 1-hexylperylene and it was based on HSQC (Fig S5), 2D anti-Z-COSY 13 (Fig. S6), TOCSY, and IMPACT-HMBC 14 spectra. The protons are labelled with letters and quaternary carbons are labelled with digits in the figures. The experiments were carried out at 25 °C in CD 2 Cl 2 . The chemical shifts and proton/carbon spectral assignment for 1hexylperylene are presented in Table S1. This information was used as an aid when making spectral assignment for 1,1'-dihexylquaterrylene vide infra. Figure S4. The nuclei labelling in 1-hexylperylene.

b) 1,1'-dihexylquaterrylene
In order to determine the main reaction product of the Scholl condensation, a detailed NMR evaluation of the reaction product was performed at 10 -5 M in CD 2 Cl 2 . As an aid in the analysis are the chemical structure of all plausible regiosisomers presented in Figure S7. The assignment was based on 2D TOCSY, 2D/1D NOESY and HSQC spectra. HMBC spectra could not be recorderd within a reasonable time given the very low concentration of the sample. For the 1D proton, 2D TOCSY and NOESY, the solvent and water peaks had to be suppressed in order to maximize the receiver gain. This was achieved by shaped pulse presaturation. The shaped pulse was set up by WAVEMAKER. 15 Figure S7. The chemical structures of all the plausible dihexylquaterrylene regioisomers that could be produced during the Schöll condensation of 1-hexylperylene.
From the proton spectrum it is evident that the reaction product is dominated by a single regioisomer and also contains minor isomers (Fig. S9-10). Through a series of NMR experiments the structure of the main regioisomer was derived as 1,1'-dihexylquaterrylene (QTRLN-A or QTRLN-C as they are magnetically equivalent and undistinguishable with NMR). Hereafter, 1,1'-dihexylquaterrylene will be referred to as the main product and QTRLN-A will be shown in schemes. The proton labelling is shown in Figure S8. The details of the spectral assignment are outlined below.   The HSQC spectrum shown in Figure S11 describes correlations between proton signals and corresponding carbons in dihexylquaterrylene. The aromatic part is represented by 9 protons that are coupled to the carbons and from this spectrum it is evident that all the signals are accounted for. Several signals were assigned with an aid of the HSQC spectrum of 1-hexylperylene (Fig. S5). Peaks u1 and u2 (Fig. S10) could not be identified but u2 is a CH 2 group according to the edited HSQC with no coupling partners or it only has very small couplings to other protons. It is interesting to note that u2 diffuses with a very similar diffusion coefficient to the dihexylquaterrelyene monomer if not identical ( Fig S20-21). At this low concentration, no HSQC peak could be seen for the peak u1 that appears as a proton coupled to a deuterium or another spin 1 nucleus.
The 2D-TOCSY spectrum describe correlations within spin systems and is shown in Figure S12. Green solid squares represent two spin systems, while dashed black squares represent 3 spin systems.   Figure S13 shows the aromatic protons o, k, l, m and n. It can be seen that o, l, n and m seemingly display extra coupling because they appear as doublet of doublets. This however cannot be seen in the TOCSY spectrum (Fig. S12), meaning no more than one coupling partners could be identified for o, l, n and m. It is thus highly probable that the two perylene rings within the quaterrylene scaffold are not completely equivalent. It is also interesting to note that the CH 3 protons appear as a "doublet of triplets"  It is interesting to note that the doubling of the peaks belonging to proton o disappears in CDCl 3 ( Figure   S15). Other peaks also show this effect, and this suggests that it is related to dynamics of the ring system. This is shown in Figure S15 where it is seen that possible exchange is affected by the solvent. The coupling constant between proton o and n is 8 Hz in both solvents.   Figure S16 shows the interaction between Hf (aliphatic CH 2 group next to the aromatic ring) and the ring protons Hg and Ho. Some of the integrals of the peaks are shown in Figure S17 and S18 for the aliphatic and aromatic regions respectively. Since the minor isomer´s methyl group is in overlap with the major isomer signals, it is not possible to integrate them separately. The assignment of the peaks which are not integrated are not shown. The integral of peak o is close to one only when both of doublets are integrated which also applies for the methyl groups. This also points out to them being in the same isomer rather than belonging to different regioisomers.
A very rough estimation of the ratio of the minor and major isomer is also possible to make, showing that the minor regiosomer is present in up to 15% amount. The chemical shifts and proton/carbon spectral assignment for 1,1'-dihexylquaterrylene are presented in Table S2.

Photophysical data
In order to verify that no non-linear processes affected the emission lifetime measurements of the Jaggregate, the fluorescence lifetime were measured at various excitation intensities. From Figure 19 it can be seen that the fluorescence lifetime is independent on excitation intensity, and hence any effect of non-linear processes or multi-excitation of the J-aggregates can be excluded at these low pulse energies (< 1.2 nJ/pulse). Figure 19. Normalized emission decay of J-aggregate at different excitation intensities. Measured using streak camera detection. Excitation wavelength 800 nm. Table S3. Values for the energy difference between S1 and S0, the energy separation between a promoting vibrational mode, and the reorganization energy for both the Monomer and J-aggregate.

Experimental diffusion coefficients
The diffusion NMR experiments were conducted in order to estimate the physical size of the aggregates.
It should be noted that the NMR experiments were performed at 10 -5 M and 10 -4 M solutions, which is at the limit of sensitivity of the instrument for a reasonable timeframe of analysis. Thus, results presented below may provide an over-estimation of the physical size of 1,1'-dihexylquaterrylene Jaggregates. The diffusion coefficient for 1-hexylperylene could be determined even at concentrations >10 -3 M in CD 2 Cl 2 and C 2 D 2 Cl 4 since no signs of aggregation were observed. However, the diffusion coefficient for the 1,1'-dihexylquaterrylene monomer had to be determined at 10 -5 M in CD 2 Cl 2 since concentrations above that induce aggregation (Fig. 1). It should be noted that at just one order of magnitude higher concentration (10 -4 M in CD 2 Cl 2 ), 1,1'-dihexylquaterrylene seems to form aggregates that are in dynamic exchange with monomers. However, by using ledbpgp2s 1 and PROJECTED 2 sequences it was possible to suppress the exchange and therefore identify two separate diffusion coefficients that would allow to distinguish between the faster diffusing units (monomers) and the slowly diffusing units (aggregates). In case of C 2 D 2 Cl 4 the PROJECTED sequence was used to check that the diffusion coefficient of the monomer is the same no matter what sequence is used.
Noteworthy, at both 10 -5 M and 10 -4 M in CD 2 Cl 2 the diffusion coefficient for the monomeric specie is the same within the error of the measurement. This confirms that the exchange between the aggregate and the monomer is slow enough that it is possible to measure the separate diffusion coefficients for the monomer and for the aggregate in the same sample ( Fig. S20-21). The experimental diffusion coefficients for 1-hexylperylene and 1,1'-dihexylquaterrylene at different concentrations are shown in the Table S4.  The aggregation pattern is even more prominent in C 2 D 2 Cl 4 . The dynamic exchange seems to be present in this case as well as can be seen in Figure S23 where first and 12 th slice of the DOSY pseudo 2D experiment is shown. As before, at the 12 th slice where the gradient strength is 34.44 Gauss/cm only the slow diffusing components are present. The measurement of the diffusion coefficient was carried out using the exchange suppressed pulse sequence (PROJECTED 2 ). This sequence requires that the transverse relaxation rates of the peaks in question is slow enough to survive the CPMG train. Only the methyl peak fulfils this requirement in at this concentration. The diffusion experiment shows that at larger concentrations (>10 -3 M in C 2 D 2 Cl 4 ) 1,1-dihexylquaterrylene forms larger aggregates than at lower concentrations (10 -4

Simulated diffusion coefficients
The simulation of diffusion coefficients were performed using the HYDRO++10 software. The advantage of such simulation is that the software accounts for the ellipsoidal geometry of the constructed aggregate, whereas the Einstein-Stokes equation can only describe objects that can be approximated as spheres. The structure of the quaterrylene was optimized using MMFF94 force field from Open Babel within the Avogadro software. A simple python script was written to remove hydrogen from the xyz files and add a line to all the carbon atoms describing the radius of the sphere replacing the carbon atom in the HYDRO++10 simulation. A value of 0.7 Å was applied to all the carbons 4-5 .