Multiple linker Half-squarylium Dyes for Dye-Sensitized Solar Cells ; Are Two Linkers Better than One ?

The synthesis and full characterization of new half-squaraine dyes (Hf-SQ) containing two or three carboxylate-based linker units is reported and these dyes tested in dye-sensitized solar cell (DSC) devices. The data show improved device efficiency for a Hf-SQ dye with two linkers (η = 5.5%) compared to the highest efficiency Hf-SQ previously reported which had only a single linker (η = 5.0%); this is mainly due to improved Voc. To understand the effects of using multiple dye linker groups, device I–V data have been correlated with single crystal X-ray structural analysis and dye electrical properties (both in solution and adsorbed to TiO2) using UV-visible and ATR-IR spectroscopy along with cyclic voltammetry, and also theoretical studies using density functional theory (DFT) calculations. These data show that positioning the linkers near the dye LUMO and so that this enables complete linker chemisorption are key factors for device performance.

Dye-sensitized solar cells (DSCs) are a promising form of 3rd generation photovoltaic technology. O'Regan and Grätzel's breakthrough in DSC devices used Ru bipyridyl dye N3 to sensitize nanoparticulate TiO 2 . 1 Since 1991, Ru-bipy dyes have remained a widely used family of DSC dyes. 2 However, these dyes are expensive which is partly due to raw material costs but also due to lengthy purication procedures which hinder scaling. They also have relatively low molar extinction coefficients (3) and generally poor spectral response above 600 nm leading to DSC efficiencies of 11.1% for N719 (ref. 3) and 11.4% for C101. 4 Modication of the organic ligands to enhance the extinction coefficient and broaden the spectral response of Ru based dyes has been reported in the literature. [5][6][7][8] In addition, replacement of the thiocyanate ligands has been reported which can lead to improved dye properties. These improvements include extending the spectral response and enhancing the extinction coefficient 8 as well as increasing open circuit voltage and greater stability. [9][10][11][12] Furthermore, modication of the bipyridyl ligands has been reported to improve the hydrophobicity of Ru dyes. 13 Finally, it has been reported that the purity of N3 and N719 has a profound inuence on their performance as sensitizers in DSC devices. 14 This has led to the development of modied bipyridyl ligands which have been used to prepare sensitizers which do not require such extensive purication during the dye synthesis. 15 Whilst the aforementioned modications of the organic ligands enable the properties of Ru dyes to be tuned, these dyes are still an expensive component of DSC devices because ruthenium is an expensive raw material, the synthetic procedures are multi-step and dye purication is complicated.
These issues have led to the development of "Ru-free" organic dyes (e.g. triarylamines, [16][17][18] coumarins 19 indolines, 20,21 quinoxalines 22 and natural dyes 23,24 ). In general, all of these dyes have been designed with a donor-pi linkeracceptor (D-p-A) structural arrangement 25 to maximize electron injection efficiency. These organic dyes oen absorb in the same region as Ru-bipy dyes (450-600 nm) where AM1.5 solar intensity is highest but are generally simpler to purify than Ru-bipy complexes and have signicantly higher 3. This allows thinner photo-electrodes to be used which can reduce recombination losses and improve V oc . High 3 is also advantageous when co-sensitizing the TiO 2 electrode to broaden spectral response because, if fewer dye sorption sites are utilized for the dye harvesting light at 400-650 nm, this leaves more space for near infrared (NIR) dyes which absorb at l > 650 nm. 26 Furthermore, organic dyes have surpassed Ru dyes with the highest h liquid DSC reported (h > 12%) for a combination of porphyrin and triphenylamine dyes 27 or a push-pull porphyrin alone. 28 A relatively new organic chromophore is the half-squarylium (Hf-SQ) dye. These dyes are both synthetically versatile and are also used as an intermediate during the synthesis of unsymmetrical squaraine dyes. 29,30 Hf-SQ dyes are known to be uorescent 31 and early reports for ZnO-based DSC devices gave h ¼ 0.27% (ref. 32) and 0.53% (ref. 33) with h ¼ 3.54% (ref. 34) for TiO 2 photo-electrodes. Recently, we have reported the highest efficiency Hf-SQ dyes to date (h ¼ 5.0%). 35 In this report, the highest J sc was recorded for a Hf-SQ dye with a vinyl dicyano-modied squaric acid unit which desorbed when electrolyte was added so that the resulting device gave poor V oc and FF. In this paper, we report a study of multiple carboxylate linkers for Hf-SQ dyes and the effects these linkers have on DSC device performance when two or three linker groups positioned around the periphery of the Hf-SQ chromophore.

Rationale for dye design
The series of dyes (4), (5), (7), (8), (10) and (11) have been synthesized following our recent report of the highest efficiency Hf-SQ DSC device (h ¼ 5.0%). 35 In the previous report, we studied the inuence of the linker position on device performance. Interestingly, the highest J sc was observed for a dye with vinyl dicyano-modied squaric acid unit; labelled compound (8) in paper; 35 here this dye will be labelled (8 0 ) (Fig. 1). However, the overall device efficiency was limited because (8 0 ) appeared to desorb when electrolyte was added resulting in poor V oc and FF. Here, we have studied the desorption process further by pumping electrolyte through the cavity of DSC device dyed with (8 0 ) and the data show that dye is desorbed by this (Fig. 2). The approach tested in this paper has been to synthesise Hf-SQ dyes which are related to the highly efficient dyes in 35 but with more than one carboxylate linker to test if this enables the dyes to adsorb more strongly to TiO 2 and to study how this affects device performance. For example, (4) is an analogous dye to (8 0 ) but with a second carboxylate linker attached to the N of the indole. To test the effect of having 2 linkers on dye desorption, a TiO 2 electrode was dyed with (4) and then electrolyte was pumped through the device cavity (Fig. 2b). This shows that dye desorption does not occur for (4). Instead the dye remains bonded to the surface and no discolouration is observed. By comparison, (5) is similar to (4) but without vinyl dicyano modication of the squaric acid unit. Hence, for both (4) and (5) the two linker groups are at the same end of the dye. To compare with this, dye (8) was designed with carboxylate linkers at opposite ends of the molecule and (11) with 3 carboxylate linkers; on the N atom and benzene of the indole and through the squaric acid unit. Dyes (7) and (10) are related to (8) and (11), respectively but without vinyl dicyano modication of the squaric acid unit.

Dye synthesis
The synthetic routes used to produce the Hf-SQ used in this work are shown in Scheme 1(a-c). The rst step is the esterication of squaric acid in ethanol to give (1) which was prepared according to the method described by Terpetschnig et al. in good yield with spectroscopic data in line with the literature. 36 Compounds (2) and (3) have been reported in our previous work. 35 To the best of our knowledge (4-11) have not been reported in the literature and were identied using several analytical techniques. The resonances identied in the 1 H NMR spectrum suggest that (4) has been isolated as the triethylamine salt. Six protons are identied in the aromatic region between 8.10 and 7.41 ppm with multiplicities including a doublet and multiplets. Resonances at 7.32, 4.25, 2.80 and 1.93 ppm have the correct multiplicities and integration to be identied as the methylene proton located near the squaric acid moiety, the ethyl group attached to the nitrogen of the indole and the methyl groups located at the 3 position of the indole unit, respectively. The quintet and triplet at 3.24 and 1.33 ppm, which integrate to six and nine protons respectively, are caused by the ethyl group of the triethylamine cation. Furthermore, the coupling constants calculated for the protons in each environment are similar to those reported for the previous half squaraine dyes. 35 13 C NMR has a resonance at 112 ppm which is associated with the nitrile functional group. In addition, resonances for the carbonyl group of the squaric acid and carboxylic acid moiety are also identied between 191 and 173 ppm. The ethyl group of the triethylamine cation can be identied between 38.39 and 7.93 ppm. Analysis using high resolution mass spectrometry identies the M + ion at 424.1294 which corresponds to the target molecule in the non-salted form. In addition, an isotope prole is detected which ts the theoretical isotope prole expected for this molecule. Attenuated total reectance infrared (ATR-IR) spectroscopy has been used to identify functional groups in the molecule including the nitrile and carbonyl groups at 2178, 1732 (s), 1626 (s), 1595 (s), 1537 (s), 1517 (s) cm À1 , respectively. (4) was isolated in 75% yield and has a melting point between 175-179 C, which is similar to a vinyl dicyano half squaraine reported in our previous work. 35 Finally, the structure of (4) was conrmed from single crystal X-ray crystallography and is show in Fig. 3. The crystal structure of (4) shows the expected molecular conguration with a propionic acid linker attached to the indole N and a second carboxylate linker being the squaric acid moiety. The crystal structure also conrms that the squaric acid unit has been modied by the addition of a vinyl dicyano group but interestingly that this group lies on the same side of the molecule to the propionic acid linker. Finally, this crystal structure also shows the presence of a triethylamine counterion as suggested by the NMR data. As expected, this is located closest to the oxygen atoms of the modied squaric acid unit where the carboxylate anionic charge is found.
Compounds (5)(6)(7)(8)(9)(10)(11) were isolated in good yield and analysed using the same techniques used to identify (4). Focussing on the major differences between these molecules, (5) is essentially the same as (4) but without a vinyl dicyano group on the squaric acid unit. The 13 C NMR and ATR data conrm the absence of nitrile groups in (5) through the absence of signals at 112 ppm and 2178 cm À1 , respectively. The other NMR and ATR data are similar to (4) as expected and mass spectrometry also conrms (5) with M + at 270.1126. Single crystal X-ray crystallography also shows the expected molecular structure of (5) and it does mirror the structure of (4) as expected but without the vinyl dicyano modication to the squaric acid group.
The synthesis of (8) begins with alkylation of a carboxylate indole to produce (6) which is conrmed by mass spectrometry (M + ¼ 372.887). This was converted to the half squaraine dye (7) by reaction with (1) as evidenced by the additional peaks between 0.95 and 1.71 ppm for the alkyl chains on the indole nitrogen and in the 1 H NMR and 3 signals for C]O between 185 and 190 ppm. (7) was also conrmed by mass spectrometry (M + ¼ 494.2929). Single crystal X-ray crystallography (Fig. 3) veries the molecular structure of (7) with a carboxylate linker attached to the benzene ring of the indole and an ester group attached to the squaraine moiety. In the context of this study, this is important because it means that (7) can only link to TiO 2 through a single carboxylate linker.
The squaraine ester of (7) is then de-esteried and a vinyl dicyano group added by reaction with CH 2 (CN) 2 to produce (8). This is conrmed by additional signals for nitrile at 117-118 ppm in the 13 C NMR and at 2182 cm À1 in the infrared data. Single crystal X-ray crystallography again proves the expected structure (Fig. 3d). Importantly, the data show the presence of two potential linker groups for this dye. These are the carboxylate linker on the benzene ring of the indole and the vinyl dicyano-modied squaric acid unit. These are oriented on one side of the molecule in the solid state which suggests it might be possible for both linkers to adsorb to TiO 2 at the same time. Interestingly, we have found that (8) also crystallises into a second polymorph; (8) (polymorph 2) (see ESI †). This varies from (8) (polymorph 1) shown in Fig. 3 mainly through the H-bonding which inuences the both the crystal packing and the relative positions of the Hf-SQ anions and triethylamine cations.
In the rst step towards the synthesis of (11), a propionic group is rst added to the N atom of a carboxy indole to make (9) which is conrmed by mass spectrometry (M + ¼ 276.1227). This is then converted to the half-squaraine (10) which is evidenced by the 1 H NMR signals at between 3.2 and 4.4 ppm for the C-H of the propionic acid and squaraine ester groups, the C]O signals at 190-200 ppm in the 13 C NMR and the mass spectrometry (M + ¼ 398.1235). This molecular structure of (10) is important for this study because the molecule contains two potential carboxylate linkers attached to the benzene ring and the N atom of the indole, respectively. The nal dye (11) is produced from (10) by simultaneous modication and de-esterication of the squaraine ester to produce the vinyl dicyano modied half-squaraine. The production of (11) is conrmed by the additional nitrile signals in the 13 C NMR between 117-119 ppm and in the IR at 2198 cm À1 and the mass spectrometry (M + ¼ 418.1036). This molecule is important for this study because it contains 3 potential carboxylate linkers; the squaric  7) and (d) (8) (polymorph 1). Displacement ellipsoids -50% probability. For clarity counteranions as well as molecular disorder components have been omitted for (4), (5) and (8) (polymorph 1). Asymmetric part of the unit cell of (7) contains two independent molecules (Z 0 ¼ 2). Only one has been shown.
acid moiety in addition to carboxylates on the benzene ring and N atom of the indole. Fig. 4 shows UV-vis spectra of dyes in solution and aer sorption onto TiO 2 photo-electrodes. The spectra in solution (Fig. 4a) show that modication of the central squaric acid moiety by replacing a carbonyl group with a vinyl dicyano group causes red shis in the spectra of (4), (8) and (11) by comparison to the unmodied dyes (5), (7) and (10), respectively. In addition, the highest molar extinction coefficient (3) is observed for the esteried squaraine (7) with generally higher 3 for vinyl dicyano modied dyes apart from (11) which has the lowest 3 of all the dyes synthesized. The spectra obtained for (4) and (5) are similar to dyes (7b 0 ) and (8 0 ) reported in our previous work. 35 This is interesting because the main structural difference associated with the new dyes reported in this paper is the addition of a propionic acid linker group to the N of the indole. This suggests that this new group plays little role in the HOMO or LUMO of these dyes in solution. In turn, this suggests that the main role of these additional groups is predominantly as linkers onto TiO 2 . Fig. 4b shows the transmission UV-vis data when the dyes are adsorbed onto transparent, mesoporous titania lms. The data show that the absorption is broadened as might be expected for molecules which are chemisorbed on a surface. Interestingly, as for the solution data, the spectra for adsorbed (4) and (5) are similar to (7b 0 ) and (8 0 ) reported previously. 35 This supports the notion that, because the additional propionic linkers of (4) and (5) are attached to the indole N atom, they only act as additional linkers to the TiO 2 surface and that they do not contribute to the HOMO or LUMO of the adsorbed dyes. This is important because (7b 0 ) from 35 can only chemisorb to TiO 2 through covalent ester bonds between the squaric acid moiety and surface hydroxyls whilst the adsorption mode of and (8 0 ) is less clear. By comparison, (4) and (5) can chemisorb both through the squaric acid and/or from the propionic acid linker. Fig. 2 illustrates our observations that (5) adsorbs much more strongly to TiO 2 than (8 0 ) producing stable devices that can easily be measured over periods of days. By comparison, devices made from (8 0 ) last minutes at most. Hence, we can assume that (4) does attach to the TiO 2 surface via the propionic linker group because the dye is not desorbed upon infusion of electrolyte into the cell. In addition, because the UV-vis spectra is similar to (8 0 ), 35 this suggests that the vinyl dicyano-modied squaric acid moiety may also interact with the TiO 2 . For (5), one could argue that this can attach to the surface via either the central squaric acid moiety and/or from the carboxylic acid linker attached to the nitrogen of the indole. Given that our previous observations 35 are that unmodied squaric acid interacts strongly with TiO 2 , it seems most likely that (5) may bonds to TiO 2 in a similar orientation to (4); i.e. that both the squaric and propionic acid units are involved. In further support of this assertion, the UV-vis spectrum for (5) is similar to those of (7b 0 ) and (10 0 ) reported in our previous work 35 because (7) can only attach through the squaric acid unit and (10 0 ) can only attach through the indole propionate because it's squaraine unit is esteried.

Spectroscopic analysis
To further study how Hf-SQ dyes adsorb on TiO 2 , new Hf-SQ dyes were prepared with an esteried squaraine moiety to prevent this bonding to the surface. In addition, the carboxylate linker was positioned on the benzene ring of the indole so that the squaraine ester should be orientated away from the TiO 2 surface. For the rst example, (7), the UV-vis spectrum of TiO 2sorbed (7) exhibits a smaller bathochromic shi relative to solvated (7) than that observed for dyes where it is believed that the squaric acid unit can interact with the surface. These dyes include (7b 0 ), (8 0 ) and (10 0 ) from, 35 and (4) and (5) reported here. By comparison, when (7) is modied to form the vinyl dicyano analogue (8), this also de-esteries the squaric acid unit which would be expected to enable dye-TiO 2 chemisorption through the squaraine moiety. In line with this assertion, a larger bathochromic shi is observed for (8) which is in line with other vinyl dicyano modied dyes (60 to 80 nm). Whilst the UV-vis data cannot prove the orientation of the dyes on the TiO 2 surface, these data do suggest that larger bathochromic shis are observed for sorbed dyes when adsorbed through the squaraine moiety and that (8) may bind preferentially through the vinyl dicyano-modied squaraine compared to the indole benzene carboxylate. The peak broadening and red-shis observed when the dyes chemisorb to TiO 2 are common in DSC dyes and has been observed previously for Hf-SQ dyes. 33,35 To rationalise this, Cicero et al. used DFT and related the phenomena to the formation of ester bonds between squaric acid oxygen and surface metal atoms which lowers the HOMO-LUMO gap. Interestingly, whilst these workers found that the HOMO changes little on sorption, they reported that the LUMO of the TiO 2 -dye system does shi. They also suggested that electron excitation into the new LUMO, which includes character from the O and C 2p orbitals of the squaric acid moiety and surface Ti 3d orbitals, should enhance electron injection from the dye into the TiO 2 .
The next dye (10) was designed with an esteried squaraine and two carboxylate linkers; on the benzene ring and the N of the indole. The UV-vis spectrum of solvated (10) shows three absorption bands; at 420 nm and weaker bands at 480 and 515 nm. For TiO 2 -adsorbed (10), these bands all broaden considerably. Whilst this makes precise assignment of peak positions difficult, there is little evidence of any major shis on these peaks and certainly not to the extent observed for the squaraine-bonding dyes. In line with our previous data, modi-cation of (10) to form the de-esteried vinyl dicyano squaraine analogue (11) does show a bathochromic shi for the adsorbed versus solvated data. This is interesting because (11) has three potential carboxylate linkers; on the benzene ring and N atom of the indole and through the modied squaraine unit. Whilst these data do not rule out some interaction through the rst two linkers, they do suggest that that the modied squaraine moiety does interact with the TiO 2 surface.
To further investigate dye-TiO 2 interactions, ATR-infrared spectroscopy has been used to compare adsorbed dye and dye powders (see ESI †). Firstly, the molecular structure of dyes (4) and (5) should make it possible for these dyes to covalently link to TiO 2 through two carboxylate linker groups; the squaraine moiety and the propionic acid on the indole N. For (4), the sharp, intense carbonyl (1650 cm À1 ) and nitrile (2225 cm À1 ) peaks in the powder broaden and shi in frequency (to 2000-2100 cm À1 for nitrile). The shi in the nitrile peak suggests an interaction with the TiO 2 surface which, although it cannot clarify any surface interaction from the propionic acid, does suggest that (4) may interact with TiO 2 through the nitrile groups of the vinyl dicyano-modied squaraine unit. For (5), which has no nitrile groups, the data for the dyed lm show a broadening and downward shi of the sharp carbonyl signal of the powder. This dye can only chemisorb through carboxylate linkers but the broadness of the sorbed carbonyl signals does not allow any differentiation between sorption modes. This may be due to the poor resolution of the peaks or it may reect a mixture of surface binding modes (e.g. mono-and bi-dentate). Interestingly, the spectra for (4) and (5) show much greater carbonyl peak shis between neat powder and TiO 2 -sorbed dyes than the single linker dyes reported previously 35 (see ESI †); thus, (7b 0 ) or (8 0 ) can only bond to the surface from the squaraine unit and (10 0 ) which can only link to TiO 2 via an indole N linker. For instance, the carbonyl signals of (7b 0 ) and nitrile and carbonyl peaks of (8 0 ) do broaden on sorption to TiO 2 but only shi ca. 10 cm À1 in frequency. These data suggest that, for dyes (4) and (5) which contain both propionic and squaric acid groups, both are involved in surface binding to TiO 2 .
By comparison with (4) and (5), dye (7) possesses only one carboxylate linker which is located on the benzene ring of the indole. The ATR-IR data show that, although there is a broadening of the peaks for the TiO 2 -sorbed (7) compared to the neat powder (ESI Fig. 4 †), the carbonyl peaks do not shi although there is a signicant reduction in the O-H signal (>3000 cm À1 ) on sorption. This presumably reects the formation of an ester linkage between the carboxylate group and the TiO 2 surface. A similar scenario is observed for (8) with a reduction in the O-H signal, and a slight broadening but very little shi on the nitrile and carbonyl peaks. This suggests that bigger peak shis are observed when the dye sorbs to TiO 2 through linker groups which are more associated with increased electron density of the LUMO. 35 In support of this assertion, dye (10) shows two carbonyl signals in the powder which corresponds to the two linkers in the structure. These broaden and shi very slightly to lower frequency on sorption which is in line with the molecule sorbing to TiO 2 at the opposite end to the squaraine moiety. By comparison, (11) is interesting because it possesses 3 linker groups and the position of these groups around the outside of the molecule suggest it would be difficult for all 3 linkers to chemisorb to TiO 2 simultaneously (Scheme 1). In this context, the absence of a signicant shi in the nitrile signal (ESI Fig. 7 †) suggests that this dye does not appear to sorb strongly through the squaraine unit but rather through the other carboxylate linkers. Table 1 shows I-V data for Hf-SQ devices with and without 5 mM chenodeoxycholic acid (CDCA). As stated above, one aim when designing these Hf-SQ dyes was to retain the optical properties of (8 0 ) from our previous work 35 but to modify the molecular structure to enable the new dye to bind more strongly to a TiO 2 electrode so that it might remain bonded aer the addition of electrolyte solution.

Device testing
Thus, dye (4) was designed and synthesized with a vinyl dicyano modied squaric acid unit but with a second carboxylate linker attached the indole N. The power conversion efficiency (h) of devices made using (4) is 4.7% and 4.5%, with and without CDCA, respectively. By comparison, the previous dye 8% as a result of a lower ll factor (FF) and open circuit voltage (V oc ) which was ascribed to rapid dye desorption in liquid electrolyte during measurement. The data for (4) show much higher V oc and FF and the stability of the device measurements conrm that the two linker approach has been successful. However, in this study, the highest performing dye was actually (5) which gives h ¼ 5.5%. This dye actually performs slightly better than (7b 0 ) from our recent report 35 which gave h ¼ 5.0% and, at that time, was the most efficient Hf-SQ dye. Thus, (5) is currently the most efficient Hf-SQ dye mainly due to an increase in V oc which here is 0.79 V compared to 0.71 V in the presence of CDCA. This is further evidence that attaching the Hf-SQ dyes to the surface from two anchoring points is advantageous to device performance. However, for dyes which can link to the surface from a carboxylic acid attached to the benzene ring of the indole an increase of V oc was not observed when an additional linker group was added to the nitrogen of the indole. For example, dyes (7) and (8) gave devices with V oc of 0.71 and 0.68 V, respectively. By comparison, (10) and (11) also have a carboxylic acid linker which is attached to the nitrogen of the indole but the V oc is lower i.e. 0.63 and 0.64 V, respectively. In addition, modication of the central unit to a vinyl dicyano group results in a lower V oc in analogous dyes and a comparable J sc , despite a broadening of the spectral response. This, allied to the ATR-IR data, suggests that the close proximity of the vinyl dicyano unit to the squaric acid linker may have a negative inuence on electron injection and/or dye regeneration processes.
The spectral response of the dyes has also been tested (Fig. 5) and the data show that dye (4) responds most strongly (EQE ca. 60%) between 500 and 575 nm which is where solar intensity is highest. This correlates with this dye having good efficiency (h ¼ 4.7%) and also the vinyl dicyano modication of the squaraine unit which re-shis the light harvesting relative to (5). Thus, by comparison, the spectral response of (5) is highest between 430 and 540 nm. However, the EQE of (5) is signicantly higher ca. 80% which is reected in the higher efficiency of (5) with h ¼ 5.5%. The spectral response data are very much in line with the DRUV data of the adsorbed dyes (Fig. 4b) and also with the 3 measured for dissolved dyes which are 47 000 M À1 cm À1 for (4) and 18 400 M À1 cm À1 for (5) suggesting that, for devices with similar TiO 2 lms and electrolyte, light harvesting most strongly inuences EQE and device performance. A similar pattern is observed for the other dyes tested in that the EQE of (7), (8), (10) and (11) all strongly resemble the DRUV of the adsorbed dyes (Fig. 4b). Thus, as expected, the EQE of dyes (8) and (11) are red-shied relative to (7) or (10) due to the presence of the vinyl dicyano modications of the squaraine unit. However, the solution 3 of these dyes seems less important as all four dyes have very similar device efficiencies (h is ca. 3.0%) whilst 3 varies from 2500 to 78 000 M À1 cm À1 .

HOMO-LUMO characteristics
Cyclic voltammetry (CV) measurements of selected dyes (10 mM) have been carried out in degassed DMF to compare dye oxidation and reduction processes with spectral and DSC device data along with theoretical DFT calculations to try to further examine any structure-activity relationships arising from changes to the dye linker position.
As an example, the CV data for (10) are shown in Fig. 6 with the onsets of the oxidation and reduction peaks which have been used to calculate the dye band gap labelled. Furthermore, the CV data show no changes following ten cycles scanned at a rate of 50 mV s À1 . However, the reduction signals for all the dyes tested are very broad (see ESI Fig. 11-19 †). To clarify the reduction voltages, each dye was held at 2.0 V for 60 s to oxidise all the dye molecules in solution. By doing this, the shape of the reduction peak changed and a more dened peak was obtained. By using these peaks to calculate the band gap of the dyes, good correlations with the bang gaps calculated using the UV-vis onsets of the dyes in solution were obtained (Table 2). In addition, the UV-vis data (Fig. 4) have shown that modication of the central squaric acid moiety to a vinyl dicyano group  resulted in a bathochromic shi in solution when the dye attaches to a titania surface compared to in solution (e.g. for dyes (4), (8) and (11)). This resulted in a decrease of the dye band gap when calculated from the absorption onset which was also observed in the CV data (Table 2). For example, shis in the anodic and cathodic waves were observed for (4), (8) and (11) by comparison to (5), (7) and (10), respectively. In addition, both the onset and mid-wave CV data have been used to calculate HOMO-LUMO potentials and band gaps versus the NHE. As expected, the band gap energies obtained from onset CV data correlate with the onset of the absorption measured from the solution UV-visible spectroscopy whilst the mid-wave CV values correlate well with the DFT calculations. Furthermore, the CV data ( Fig. 6 and Table 2) suggest that the LUMO levels of the dyes reported here should be above the conduction band of TiO 2 to facilitate successful electron injection from the dye into the TiO 2 .
In order to compare the band gaps calculated from the HOMO-LUMO levels ( Table 2) with device V oc values measured from the I-V data (Table 1), the V oc data from CDCA devices have been used based on the assumption that CDCA treatment should mitigate dye aggregation and recombination losses. The data show the highest V oc values measured are for (5) and (7) and these dyes have the highest dye band gaps (2.60 and 2.67 eV). By comparison, the lowest V oc is recorded is for (11) which also has the lowest band bap; V oc ¼ 0.64 V and E B ¼ 2.14 eV. This suggests that higher dye band gaps do result in higher V oc . However, the dye energy levels will set the maximum V oc and, as shown with the data from the CDCA treatment, voltage losses must be minimised to achieve this.
DFT molecular modelling studies of (4), (5), (8), (10) and (11) have also been carried out to study both the optimized molecular geometry and HOMO and LUMO maps of these dyes (Fig. 7). Whilst it is important to note that these are effectively gas phase structures (i.e. solvent free), for vinyl dicyano modi-ed dyes ((4), (8) and (11)) the nitrile groups are consistently located on the same side of the molecule as the propionic acid linker. This implies that, if these dyes chemisorb through the indole propionate, the vinyl dicyano group is located closer to the TiO 2 surface than the acid unit of the squaraine moiety. This is supported by the ATR-IR data for these dyes which show signicant nitrile shis on sorption to TiO 2 . In addition, the HOMO-LUMO calculations show electron density spread through the p-framework of each dye but not on the propionic acid linker. This suggests that the latter group acts mainly as a dye-TiO 2 anchoring point and does not play a major role in electron injection from the dye excited state. This correlates well with the UV-vis data which suggests that the indole propionate is not strongly involved with the dye HOMO or LUMO levels. Analysis of the HOMO-LUMO energy level calculations shows that the band gaps are in good agreement with the mid-wave CV data apart from for (5) where there is closer agreement to the UV-vis onset value. This may reect that the DFT relates to gas phase structures. Interestingly, when considering the relative energies of the HOMO and LUMO levels, the dyes are divided into 2 groups with the vinyl dicyano modied dyes much higher in energy. If these differences occurred for sorbed dyes, this might result in poorer overlap between dye and TiO 2 orbitals which might result in poorer electron injection. This might explain why these dyes performed less well than expected in DSC devices. This is further supported when comparing the HOMO and LUMO maps of (4) and (5). For (4), the LUMO map shows little electron density near the vinyl dicyano-modied squaraine or propionic acid groups which are expected to be closest to TiO 2 . By comparison, (5) is the best performing dye and the LUMO map shows electron density on the squaraine moiety which would be expected to enhance electron injection for this dye.

Instrumentation and chemicals
All chemicals were purchased from Aldrich and used as supplied unless otherwise stated. Anhydrous solvents were used as supplied except tetrahydrofuran (THF) which was dried using Na wire. NMR spectra were recorded on a Bruker AC500 at 500 MHz for 1 H and 125 MHz for 13 C. Chemical shis (d) are given in ppm relative to (CH 3 ) 4 Si and J values (in Hz) refer to J H,H unless otherwise stated. Mass spectra were recorded at the EPSRC National Mass Spectrometry Service at the University of Swansea. Infrared spectra were recorded on a PE1600 series FTIR spectrometer using an ATR attachment. UV-visible spectroscopy was measured on a Perkin Elmer spectrometer. Cyclic voltammetry was measured on 10 mM dyes in N 2 -saturated DMF at RT using 0.1 M Bu 4 NPF 6 as supporting electrolyte on an Autolab PGSTAT 30 computer-controlled electrochemical measurement system (Eco Chemie, Holland). Sweep rate was 50 mV s À1 in a 3-electrode cell using a glassy carbon (1 cm 2 ) working electrode, a Ag/AgCl reference electrode and a Pt counter electrode (1 cm 2 ). All voltammetric potentials were re-calculated and are reported versus NHE.

X-ray crystallography
Single-crystal X-ray diffraction data of (4), (5), (7), (8) (polymorph 1) and (8) (polymorph 2) were collected at 100 K on Rigaku AFC12 goniometer equipped with an enhanced sensitivity (HG) Saturn 724 + detector mounted at the window of an  37 Unit cell parameters were rened against all data. An empirical absorption correction was carried out using Crystal-Clear soware. 38 All crystal structures were solved by charge ipping methods 39 and rened on F o 2 by full-matrix leastsquares renements using programs of the SHELX-2013 soware. 40 All non-hydrogen atoms were rened with anisotropic displacement parameters. All hydrogen atoms were added at calculated positions and rened using a riding model with isotropic displacement parameters based on the equivalent isotropic displacement parameter (U eq ) of the parent atom.
In the crystal structure of (4), the HNEt 3 + cation is disordered and modelled over two sites with a 78 : 22 ratio, whereas in (7) disordered ethoxy group on one of two independent molecules present in the asymmetric unit was modelled over two sites with 68 : 32 ratio. In (8) (polymorph 2), the aliphatic chain is badly disordered and modelled and constrained over two sites with a 72 : 28 ratio. Vibrational restraints (SIMU/DELU) as well as distance/angle restraints DFIX/DANG were used to maintain sensible molecular geometry and atomic displacement ellipsoids. Some disordered atoms required EADP and ISOR restraints to be used.