Double Linker Triphenylamine Dyes for Dye-Sensitized Solar Cells

: Most organic dyes synthesized for dye-sensitized solar cells (DSC) use a single linker group to bind to the metal oxide photo-anode. Here we describe the synthesis and testing of two new triphenylamine dyes containing either two carboxylic acids acid ( 10 ) or two cyanoacrylic acids (2Z, 2 (cid:48) Z)-3, 3 (cid:48) -(5-((E)-4-(diphenylamino) styryl)-1, 3-phenylene) bis (2-cyanoacrylic acid) ( 8 ) as linker groups. Full characterization data are reported for these dyes and their synthetic intermediates. DSC devices have been prepared from these new dyes either by passive or fast dyeing and the dyes have also been tested in co-sensitized DSC devices leading to a PCE ( η = 5.4%) for the double cyanoacrylate linker dye ( 8 ) co-sensitized with D149. The dye:TiO 2 surface interactions and dye excitations are interpreted using three modelling methods: density functional theory (at 0 K); molecular dynamics (at 298 K); time dependent density functional theory. The modelling results show the preferred orientation of both dyes on an anatase (1 0 1) TiO 2 surface to be horizontal, and both the simulated and experimental absorption spectra of the dye molecules indicate a red shifted band for ( 8 ) compared to ( 10 ). This is in line with broader light harvesting and J sc for ( 8 ) compared to ( 10 ). the synthesis of two new triarylamine dyes with double linker groups for adsorption onto TiO 2 photo-anodes. These new dyes have been fully characterized (including single crystal X-ray crystallography) and have been tested in DSC devices leading to power conversion e ﬃ ciencies of 1.2% for the double carboxylate dye ( 10 ) and 2.4% for the double cyanoacrylate linker dye ( 8 ). Ultra-fast co-sensitization has also been examined for these dyes which consistently shows improved light harvesting (i.e., higher J sc ) and PCE with the highest e ﬃ ciency achieved for a combination of ( 8 ) and D149 = 5.4%).


Introduction
O'Regan and Grätzel's breakthrough paper in dye-sensitized solar cells (DSC) used the Ru-bipyridyl dye (N3) adsorbed on a mesoporous TiO 2 photoanode leading to a power conversion efficiency (PCE) of 7.2% [1]. Given this step forward, it is not surprising that a great deal of subsequent work focused on Ru-based dyes including N719 [2] and C101 [3] and "black dye" [4]. Whilst the development of these new dyes raised the PCE of DSC devices to ca. 11%, further increases stalled. This was partly because the light harvesting of Ru-bipyridyl dyes is generally limited to <650 nm and partly due to the relatively average molar extinction coefficients of Ru-bipyridyl dyes (extinction coefficient = 12,000-18,000 M −1 cm −1 ) [5].
Following the Ru-bipyridyl dyes, two main approaches have been taken to further improve DSC device efficiencies. Firstly, organic dyes have been developed both for DSC and other photovoltaic solar cells [6]. For DSC, these dyes have been designed with higher extinction coefficient and,

Device Making and Testing
DSC active areas were prepared by doctor blading two layers of transparent TiO 2 paste (DSL18NRT, Dyesol) onto fluoride tin oxide-coated glass (TEC15, NSG) followed by one layer of scattering TiO 2 paste (WER4-O, Dyesol) to produce ca. 12 µm thick films. The TiO 2 films were shaped to 0.5 × 2 cm to give 1.0 cm 2 devices and each TiO 2 layer was sintered at 450 • C for 30 min. The TiO 2 films were then immersed in 50 mM TiCl 4 :THF 2(aq) at 70 • C for 30 min, rinsed with DI water, air-dried for 10 min and re-sintered at 450 • C for 10 min. The TiO 2 films were either dyed passively (18 h immersion in dye solution) or ultra-fast dyed (5 min) according to Holliman et al. [29]. The counter electrode (TEC 8, NSG) was coated with Pt paste (PT1, Dyesol) and heated to 400 • C. The two electrodes were sealed with a Surlyn gasket (DuPont) at 120 • C. Iodide/tri-iodide electrolyte (I 2 100 mM, imidazolium iodide 0.8 mol L −1 and LiI 100 mmol L −1 in distilled 3-methoxypropionitrile) was added through holes in the counter electrode which were then sealed using Surlyn (DuPont). Current-voltage characteristics were measured with an ABET Solar Simulator with a Xe arc lamp and a Keithley 2400 at 100 mW cm 2 (1 Sun). Spectral response was measured on a QEX10 Quantum Efficiency Measurement System in DC mode at a resolution of 10 nm. Lamps were calibrated using a certified (Oriel 91,150 V) monocrystalline Si reference cell traceable to the National Renewable Energy Laboratory (NREL).

Atomistic Modelling Simulations
The ab initio electronic structure calculations were carried out using the plane wave, pseudopotential code [51] within the formalism of density functional theory (DFT) [52][53][54]. The dye:TiO 2 models were built using Materials Studio [55] and the calculations were run on 192 processors per model system. Previous atomistic modelling work on six half-squaraine dye molecules adsorbed on anatase TiO 2 (1 0 1) [11] surface involved scanning a range of dye orientations in an attempt to identify the most probable mode of adsorption, i.e., the orientation of the dye on the surface that corresponded to the lowest energy. All six half-squaraine dyes preferred a horizontal orientation to the surface [12]. The carboxylic group of the dye:TiO 2 system was either directly bonded to the surface via O-Ti, and/or via a hydrogen bond; the cyano-dye was also bonded via N-Ti.
Whilst both double-linker dyes ( (8) and (10)) contain carboxylic linker groups, dye (8) contains two cyanoacrylate linker groups (i.e., both linkers have an additional nitrile unit). Based on the findings from the aforementioned (half-squaraine) work, that data suggests that attachment to the TiO 2 surface occurs via the nitrile moieties of vinyl dicyano units on derivatized half-squaraine dyes. Therefore, we focused the atomistic modeling for dyes (8) and (10) on vertical (the commonly hypothesized orientation in the dye sensitized research community), and horizontal dye-to-surface orientations only; the symmetry of the dyes along their carbon backbone excluded side orientations. Dye molecules (8) and (10) were placed horizontally and vertically oriented to the TiO 2 surface, and then allowed to relax to their lowest energy configurations at 0K; i.e., until the convergence criteria were met. Further details of the model system set-up and convergence criteria are in the Supplementary Material.
The molecular dynamics simulations were carried out using NAMD [56]. The models were built and visualized using VMD [57] and the calculations were run on 64 processors. The dyes were placed vertically on a fixed slab, before each dye:TiO 2 system underwent minimization, equilibration, and simulation at 298 K in an NVT ensemble. Further details of the simulation parameters and run can be found in the Supplementary Material.
A common method to study electronic excited states within the DFT framework is via time-dependent (TD) DFT [58][59][60][61]. The dyes were optimized at their ground state in gas-phase using DFT. The excited states were calculated with TDDFT and all calculations were done using the ORCA software package [62] and run on 32 processors. Further details of the TDDFT parameters can be found in the Supplementary Material.

Dye Synthesis
Dyes (8) and (10) were synthesized using the methods shown in Scheme 1.

Dye Synthesis
Dyes (8) and (10) were synthesized using the methods shown in Scheme 1.

Scheme 1.
Synthetic pathway for dyes (8) and (10). Pd cat = trans-di(μ-acetato)bis[0 -(di-0tolylphosphoino)benzyl]dipalladium (II). Compounds (1) to (9) are described in the supplementary material (SM) Dye (8) was synthesized by first converting 4-(N, N-diphenylamino)-benzaldehyde to the known compound (4-vinyl-phenyl)-amine (1) by a Wittig reaction [19]. This was confirmed by NMR and mass spectrometry (see Supplementary Material). To make the double cyanoacrylate linker, bromo-isophthalic acid was esterified using methanol and H2SO4(aq) to make the dimethyl ester (6). Then, (6) was reduced to the di-alcohol (4) using LiAlH4 (see Supplementary Material). This reaction was followed by TLC before quenching with Na2SO4(aq). The product was confirmed by 1 H NMR with alcohol protons (2x-CH2OH) at 4.63 ppm as a singlet whilst 13 C NMR (DEPTQ) shows five carbon environments and a new signal at 63.01 ppm for (-CH2-OH). Mass spectrometry confirmed (4) with [M + NH4] + at 234.0126. The di-aldehyde (5) was synthesized by oxidizing (4) with pyridinium chlorochromate which was confirmed by the loss of CH2OH signals in the 1 H NMR data and the appearance a signal at 10.06 ppm for the aldehyde. 13 C NMR data also showed a signal at 189.48 ppm for the aldehyde carbons and mass spectrometry showed [M + H] + at 212.9546 for (5). Combining (1) and (5) in a Heck coupling then produced (7) which was confirmed by triphenylamine-vinyl (6.97-8.39 ppm) and aldehyde (10.13 ppm) signals in the 1 H NMR data and 15 C environments in the 13 C NMR data. Mass spectrometry also confirmed (7) with [M + H] + at 404.1645. Finally, (8) was synthesized by a Knoevenagel condensation reaction of (7) with cyanoacetic acid in the presence of piperidine. 1 H NMR data confirmed (8) with the loss of aldehyde signals. 13 C NMR data showed 18 carbon environments and a new signal at 115.99 ppm for (C≡N) which was also confirmed by FT-IR with a new peak at 2226 cm −1 . Mass spectrometry also confirmed (8)
Dye (8) was synthesized by first converting 4-(N, N-diphenylamino)-benzaldehyde to the known compound (4-vinyl-phenyl)-amine (1) by a Wittig reaction [19]. This was confirmed by NMR and mass spectrometry (see Supplementary Material). To make the double cyanoacrylate linker, bromo-isophthalic acid was esterified using methanol and H 2 SO 4(aq) to make the dimethyl ester (6). Then, (6) was reduced to the di-alcohol (4) using LiAlH 4 (see Supplementary Material). This reaction was followed by TLC before quenching with Na 2 SO 4(aq) . The product was confirmed by 1 H NMR with alcohol protons (2x-CH 2 OH) at 4.63 ppm as a singlet whilst 13 C NMR (DEPTQ) shows five carbon environments and a new signal at 63.01 ppm for (-CH 2 -OH). Mass spectrometry confirmed (4) with [M + NH 4 ] + at 234.0126. The di-aldehyde (5) was synthesized by oxidizing (4) with pyridinium chlorochromate which was confirmed by the loss of CH 2 OH signals in the 1 H NMR data and the appearance a signal at 10.06 ppm for the aldehyde. 13 C NMR data also showed a signal at 189.48 ppm for the aldehyde carbons and mass spectrometry showed [M + H] + at 212.9546 for (5). Combining (1) and (5) in a Heck coupling then produced (7) which was confirmed by triphenylamine-vinyl (6.97-8.39 ppm) and aldehyde (10.13 ppm) signals in the 1 H NMR data and 15 C environments in the 13 C NMR data. Mass spectrometry also confirmed (7) with [M + H] + at 404.1645. Finally, (8) was synthesized by a Knoevenagel condensation reaction of (7) with cyanoacetic acid in the presence of piperidine. 1 H NMR data confirmed (8) with the loss of aldehyde signals. 13 C NMR data showed 18 carbon environments and a new signal at 115.99 ppm for (C≡N) which was also confirmed by FT-IR with a new peak at 2226 cm −1 . Mass spectrometry also confirmed (8)  Dye (10) was synthesized in three steps. First, 5-bromo isophthalic acid was esterified to 5-bromo-isophthalic acid dimethyl ester (6) using methanol and a few drops of H 2 SO 4(aq) . The next step was to couple (1) with (6) to synthesize (9) which was confirmed by 1 H NMR data with a singlet at 3.91 ppm for methyl ester groups and triphenylamine signals (6.96 to 8.38 ppm) and an ester carbonyl at 165.34 ppm in the 13 C NMR data (see Supplementary Material). Mass spectrometry also confirmed [M + H] + for (9) at 464.1852 (see ESI). Finally, (10) was synthesized by hydrolytic de-esterification of (9) using LiOH before pH adjustment with HCl to yield (10) as the acid form. Both 1 H NMR and 13 C NMR confirm the loss of the methyl ester groups and a shift of the carbonyl carbon from 165.34 to 166.63 ppm for the carboxylic acid. Mass spectrometry also confirmed [M + H] + for (10) at 436.1542.
The UV-Vis spectra of dyes (8) and (10) are shown in Figure 1, the data in Table 1, and also in the Supplementary Material Figure S2. For the solution spectra (Figure 1a), the main two peaks on the spectra are at similar positions (ca. 295 and 375 nm) and are ascribed to a π-π* electron transition and intra-molecular charge transfer (i-MCT) between the triphenylamine donor and the acid acceptor of each dye. However, there is a difference in the relative intensities on these peaks between the two dyes. Thus, for (8) the π-π* transition is more intense (extinction coefficient = 35,421 M −1 cm −1 ) relative to the i-MCT (extinction coefficient = 21,077 M −1 cm −1 ). However, for (10) the opposite trend is observed; i.e., the π-π* transition is less intense (extinction coefficient = 25,718 M −1 cm −1 ) relative to the i-MCT (extinction coefficient = 35,196 M −1 cm −1 ). Importantly, the i-MCT peak for (8) is also much broader than for (10) and extends to ca. 500 nm which suggests a broader spectral response for (8). Figure 1b shows UV-Vis spectral data for a pristine, mesoporous TiO 2 film. The data show a strong absorbance band centered around 350 nm which continues up to approx. 390 nm as expected for anatase. Figure 1b also shows data for dyes (8) and (10) adsorbed onto TiO 2 . For clarity, the dyes were measured using a pristine TiO 2 film as the background so that only light absorption from the dyes is observed. For both of the adsorbed dyes (8) and (10), there is a slight red shift on the i-MCT band up to ca. 410 nm. For adsorbed dye (10), the absorption onset red shifts up to ca. 475 nm from 450 nm for pristine (10). For adsorbed dye (8), the absorption onset red shifts even further up to ca. 550 nm from 500 nm for pristine (8). These data correlate strongly with the EQE data for these dyes (Figure 2a).
Energies 2020, 13, x FOR PEER REVIEW 6 of 17 Dye (10) was synthesized in three steps. First, 5-bromo isophthalic acid was esterified to 5bromo-isophthalic acid dimethyl ester (6) using methanol and a few drops of H2SO4(aq). The next step was to couple (1) with (6) to synthesize (9) which was confirmed by 1 H NMR data with a singlet at 3.91 ppm for methyl ester groups and triphenylamine signals (6.96 to 8.38 ppm) and an ester carbonyl at 165.34 ppm in the 13 C NMR data (see Supplementary Material). Mass spectrometry also confirmed [M + H] + for (9) at 464.1852 (see ESI). Finally, (10) was synthesized by hydrolytic de-esterification of (9) using LiOH before pH adjustment with HCl to yield (10) as the acid form. Both 1 H NMR and 13 C NMR confirm the loss of the methyl ester groups and a shift of the carbonyl carbon from 165.34 to 166.63 ppm for the carboxylic acid. Mass spectrometry also confirmed [M + H] + for (10) at 436.1542.
The UV-Vis spectra of dyes (8) and (10) are shown in Figure 1, the data in Table 1, and also in the Supplementary Material Figure 2. For the solution spectra (Figure 1a), the main two peaks on the spectra are at similar positions (ca. 295 and 375 nm) and are ascribed to a π-π* electron transition and intra-molecular charge transfer (i-MCT) between the triphenylamine donor and the acid acceptor of each dye. However, there is a difference in the relative intensities on these peaks between the two dyes. Thus, for (8) the π-π* transition is more intense (extinction coefficient = 35,421 M −1 cm −1 ) relative to the i-MCT (extinction coefficient = 21,077 M −1 cm −1 ). However, for (10) the opposite trend is observed; i.e., the π-π* transition is less intense (extinction coefficient = 25,718 M −1 cm −1 ) relative to the i-MCT (extinction coefficient = 35,196 M −1 cm −1 ). Importantly, the i-MCT peak for (8) is also much broader than for (10) and extends to ca. 500 nm which suggests a broader spectral response for (8). Figure 1b shows UV-Vis spectral data for a pristine, mesoporous TiO2 film. The data show a strong absorbance band centered around 350 nm which continues up to approx. 390 nm as expected for anatase. Figure 1b also shows data for dyes (8) and (10) adsorbed onto TiO2. For clarity, the dyes were measured using a pristine TiO2 film as the background so that only light absorption from the dyes is observed. For both of the adsorbed dyes (8) and (10), there is a slight red shift on the i-MCT band up to ca. 410 nm. For adsorbed dye (10), the absorption onset red shifts up to ca. 475 nm from 450 nm for pristine (10). For adsorbed dye (8), the absorption onset red shifts even further up to ca. 550 nm from 500 nm for pristine (8). These data correlate strongly with the EQE data for these dyes (Figure 2a).       Table 2 and Supplementary Material Figure S3 show the data for DSC devices made using these new triphenylamine dyes. For 1 cm 2 DSC devices made with only single dyes which have been passively dyed onto the TiO 2 photoanodes, the data show power conversion efficiencies (PCE) of η = 2.4% and 1.2% for dyes (8) and (10), respectively. The higher PCE for (8) versus (10) is reflected in the higher photocurrent (J sc = 5.37 mA cm −2 versus 2.82 mA cm −2 ) and voltage (V oc = 0.71 V versus 0.63 V), respectively. The higher photocurrent for (8) can be explained by the broader spectral response for this dye compared to (10). For example, EQE data for (8) shows light harvesting from ca. 320-550 nm whereas (10) only harvests light from ca. 320-450 nm (Figure 2a). Previously, we observed a higher photocurrent for a single cyanoacrylate linker triphenylamine dye over the analogous carboxylate dye [19] which again was due to a broader spectral response for the cyanoacrylate linker dye. However, previously, this was accompanied by a lower V oc . The data here suggest that having two cyanoacrylate linkers improves the voltage. The previously published indoline dye D149 was used as a control dye for these devices and this gave PCE of 4.4% for a 1 cm 2 device. The J sc (13.50 mA cm −2 ) and V oc (0.60 V) for this dye are similar (but slightly lower) to published data [63] which might be due to a larger device area for the devices here. The different colors of the two dyes will also inevitably reflect different dye energy levels (i.e., highest occupied molecular orbital-HOMO; lowest unoccupied molecular orbital-LUMO). For example, the absorption onset of (8) is at ca. 500 nm which corresponds to a HOMO-LUMO gap of ca. 2.48 eV, whilst for (10) the onset is ca. 430 nm which gives EB of 2.88 eV. It is not clear from these data whether this variance of 400 mV reflects a shift in either the HOMO or LUMO levels of (8). However, if the LUMO did shift this might improve energy alignment with the TiO 2 conduction band.  Table 2 also shows data for ultra-fast dyed devices. When comparing between the passive and fast dying data there are no significant differences between (8) and (10) except that the J sc is decreased slightly for the fast-dyed devices which might suggest slightly lower dye loading. Devices H and K were prepared by co-sensitizing dyes (8+D149) and (10+D149), respectively. Device H shows improved data for co-sensitization compared with Device C with an increase in efficiency from 4.3% to 5.4% and J sc from 13.34 to 14.69 mA cm −2 . The EQE data for a combination of (8) and D149 (Device H and Figure 2b), shows a maximum EQE of 72% at 540 nm for D149 and of 62% at 410 nm for (8). Overall, these EQE data show higher spectral response and wider photon capture across solar spectrum for the co-sensitized devices leading to increase of J sc . In contrast, Device K did not show as good a PCE compared with Device H or C. Figure 2c shows a maximum EQE of 71% at 540 nm for D149 and of 55% at 400 nm for (10). The EQE data for Devices L and M which were made by co-sensitizing (8+N719) or (10+N719) (see Supplementary Material), were much lower and did not show as promising results which suggests that the lower extinction coefficient of N719 reduced photon harvesting meaning that more adsorbed dye was required to harvest the same amount of light.

Experimental
Interestingly, a comparison of the UV-vis ( Figure 1) and EQE data ( Figure 2) shows that the light harvesting of both (8) and (10) is red shifted when the dyes are absorbed on TiO 2 compared to in solution. Not only is this advantageous in terms of harvesting more light in the part of the AM1.5 spectrum where more sunlight is available, but it is also indicative of the bonding interactions between the dye and the TiO 2 . The gas phase spectra of dyes (8) and (10) were calculated using TDDFT (see Section 3.2.2. below). Although these calculations do not include solvent, they show similar trends in light absorption with (8) exhibiting red-shifted bands relative to (10).

Simulation of the Dye Gas-Phase Spectra
The TDDFT calculations show that the orbitals involved in the electronic excitations are π-π * in nature ( Figure 3). The most intense peak (2 in (8) and 1 in (10)) for both the dyes involves similar π−π * orbitals and the transition occurs at 350.1 nm for (8) and 329.7 nm for (10). The major difference between the dyes' spectra is seen in the peak at 379.9 nm (peak 1 in Figure 3) in (8). This is also a π−π * transition. However, the π * of the LUMO is localized on the cyanoacrylate part of the molecule. This is important in the context of often quoted donor-linker-acceptor design for DSC dyes because the excited state LUMO is located on the linker part of the dye which should make electron injection into the TiO 2 more favorable. Similar transitions involving π-π * orbitals, where the π * is localized at the carboxylate part, is largely blue shifted (296.7 nm, peak 2 in figure) in (10). Whilst again the location of the LUMO on the linker part of (10) should favor electron injection into TiO 2 , for (10), the blue shift means that this transition is driven by the harvesting of lower solar intensity, higher energy photons which ultimately limits device photocurrent. Overall, we can say that the spectral transitions are red shifted for (8)-a trend similar to that observed experimentally.

Bonding Interactions between the Dye and TiO 2 Surface-Atomistic Modelling
The final configurations of the dyes on the (1 0 1) anatase surface of TiO 2 calculated using DFT are shown in Figure 4. The horizontal orientation of both dyes resulted in lower total energies than the vertical orientations. These findings agree with those we reported for the smaller, half-squaraine dyes [12] implying that the length of the dye molecule has no effect on the tendency of a dye to lie horizontally to the surface at 0 K.  Dye (8) interacts with the TiO2 surface via its cyano nitrogen, carboxylic oxygen, and hydrogen, in both horizontal and vertical orientations (see Table 3). The horizontally oriented dye has an additional interaction between the benzene carbon ( Figure 4) and surface oxygen, which, judging by its relatively short distance of 1.53 Å suggests a covalent bond. The previously planar benzene ring distorts by approximately 40° towards a surface apical oxygen, which results in the conversion of a Dye (8) interacts with the TiO 2 surface via its cyano nitrogen, carboxylic oxygen, and hydrogen, in both horizontal and vertical orientations (see Table 3). The horizontally oriented dye has an additional interaction between the benzene carbon ( Figure 4) and surface oxygen, which, judging by its relatively short distance of 1.53 Å suggests a covalent bond. The previously planar benzene ring distorts by approximately 40 • towards a surface apical oxygen, which results in the conversion of a double to single C-C bond. Table 3. The dyes; dye moieties and their interaction distances with the surface; approximate range of the angle of inclination of the dyes' carbon backbone to the surface; the estimated projected surface area of the molecules on the surface based on their van der Waals outline.  Figure 4 for dye and Ti labels. b Angle was measured using the two vectors: 1. triphenyl nitrogen and carbon on the benzene (C b ) joining the two linkers; 2. normal vector to the TiO 2 surface.

Dye and Orientation
Comparing the electron density of the horizontal and vertical orientations represented by Mulliken charges [64][65][66][67] reveals an increase in electron density around the distorted benzene ring, with an accompanying decrease in electron density dissipated across the closest surface titania. Dye (10) lacks cyano groups and interacts with the surface via its carboxylic oxygen and hydrogen, in both horizontal and vertical orientations (see Table 3), with the former exhibiting shorter interaction distances between the carboxylic oxygen and five-coordinated Ti, than are seen in horizontally orientated dye (8). The N-Ti and C=O-Ti distances exhibited by both dyes indicate covalent to ionic interactions, and the C-OH-O distances indicate hydrogen bonding and electrostatic interactions.
During the MD equilibration both molecules (8) and (10) became horizontal against the surface-indicating the influence of van der Waals and electrostatic interactions between the dye and surface-and remained so throughout the simulation of 40 ns (example snapshot in Supplementary Material Figure S1). As can be seen in Table 3, the MD interaction distances are generally longer than those measured from the DFT simulations, which is not surprising as MD simulations do not include electronic interactions. Such interactions are modeled via the force field parameters; hence the results of a molecular dynamics simulation depend on the suitability of the classical force field used. In this study, the TiO 2 slab was kept rigid limiting closer interaction with the dye molecules. However, the reasonably short carboxylic hydrogen-to-surface distances and their nearly constant distributions ( Figure 5) show that, at 298 K, the molecules are stable in their horizontal configurations in close proximity to the surface. Moreover, for both dye molecules, the angles between the molecular plane and the slab surface agree very well with the DFT calculations (Table 3), and the dyes exhibit a number of dominant interactions with the TiO 2 surface: covalent, hydrogen-bonds, van der Waals and electrostatic.
From our atomistic simulations we infer that in experimental devices, both dyes adsorb to the surface in horizontal orientations, which might change with an increase in dye concentration, i.e., if space becomes limited for horizontal, mono-layer dye packing, the molecules might be forced to pack closer together in more vertical than horizontal orientations. Multiple dye packing is the focus of our next study.
The atomistic modelling work carried out using density functional theory, time dependent density functional theory, and molecular dynamics provides a deeper level of insight to the experimental work, with the proviso that the models are simplifications of the experimental systems and as such are subject to limitations, notwithstanding which the intention is to capture the essential interactions prevalent in a physical system. From our atomistic simulations we infer that in experimental devices, both dyes adsorb to the surface in horizontal orientations, which might change with an increase in dye concentration, i.e., if space becomes limited for horizontal, mono-layer dye packing, the molecules might be forced to pack closer together in more vertical than horizontal orientations. Multiple dye packing is the focus of our next study.
The atomistic modelling work carried out using density functional theory, time dependent density functional theory, and molecular dynamics provides a deeper level of insight to the experimental work, with the proviso that the models are simplifications of the experimental systems and as such are subject to limitations, notwithstanding which the intention is to capture the essential interactions prevalent in a physical system.

Conclusions
Whilst organic DSC dyes are often designed using a donor-spacer-linker motif, this is usually hypothesized to be oriented vertically to the TiO2 surface. From our atomistic simulations we infer that in experimental devices, both dyes adsorb to the TiO2 surface in horizontal orientations. However, we modelled single dye molecules and so the orientation might change with increase in dye loading; i.e., if space becomes limited for horizontal, mono-layer dye packing, the molecules might be forced to pack closer together in more vertical than horizontal orientations. A similar effect has been reported previously for terephthalic acid on rutile [68]. Whilst this does not affect the donorspacer-linker approach, it does affect the surface area that the dye molecules occupy, which dye moieties interact with the TiO2 surface, and how dyes should be designed moving forward. For example, for dye (8), the DFT unexpectedly shows dye:TiO2 distances short enough to suggest a bonding interaction between the benzene ring and the TiO2 surface which could influence where electron injection takes place from this dye into the TiO2. DSC device data do show better PCE for dye (8) which has a common triphenyl donor to (10) but two cyanoacrylate linker groups rather than two carboxylates. This is a similar trend to devices made from the structurally related single linker dyes [19]. However, we ascribe this mainly to broadened spectral response in line with the EQE data and the red-shifted light absorption observed both experimentally and by TDDFT. MD simulations emphasize that dye-surface interactions are not static which again agrees with experimental data for DSC devices which are known to improve with ageing in the first week after manufacture. Finally, photovoltaic device lifetime is an important parameter for technology commercialization. Whilst we have not reported this for these dyes in this paper, our previous work on related triphenylamine dyes [18] showed that they exhibited excellent stability for 1800 h of light soaking so we expect these dyes to also exhibit similar device stability.

Conclusions
Whilst organic DSC dyes are often designed using a donor-spacer-linker motif, this is usually hypothesized to be oriented vertically to the TiO 2 surface. From our atomistic simulations we infer that in experimental devices, both dyes adsorb to the TiO 2 surface in horizontal orientations. However, we modelled single dye molecules and so the orientation might change with increase in dye loading; i.e., if space becomes limited for horizontal, mono-layer dye packing, the molecules might be forced to pack closer together in more vertical than horizontal orientations. A similar effect has been reported previously for terephthalic acid on rutile [68]. Whilst this does not affect the donor-spacer-linker approach, it does affect the surface area that the dye molecules occupy, which dye moieties interact with the TiO 2 surface, and how dyes should be designed moving forward. For example, for dye (8), the DFT unexpectedly shows dye:TiO 2 distances short enough to suggest a bonding interaction between the benzene ring and the TiO 2 surface which could influence where electron injection takes place from this dye into the TiO 2 . DSC device data do show better PCE for dye (8) which has a common triphenyl donor to (10) but two cyanoacrylate linker groups rather than two carboxylates. This is a similar trend to devices made from the structurally related single linker dyes [19]. However, we ascribe this mainly to broadened spectral response in line with the EQE data and the red-shifted light absorption observed both experimentally and by TDDFT. MD simulations emphasize that dye-surface interactions are not static which again agrees with experimental data for DSC devices which are known to improve with ageing in the first week after manufacture. Finally, photovoltaic device lifetime is an important parameter for technology commercialization. Whilst we have not reported this for these dyes in this paper, our previous work on related triphenylamine dyes [18] showed that they exhibited excellent stability for 1800 h of light soaking so we expect these dyes to also exhibit similar device stability.