Hydrogen-bonded supramolecular assembly of dyes at nanostructured solar cell interfaces
HHydrogen-bonded supramolecular assembly of dyesat nanostructured solar cell interfaces
Christopher E. Patrick and Feliciano Giustino
Department of Materials, University of Oxford,Parks Road, Oxford OX1 3PH, United Kingdom (Dated: October 28, 2018)
Abstract
We calculate from first principles the O1 s core-level shifts for a variety of atomistic modelsof the interface between TiO and the dye N3 found in dye-sensitized solar cells. A systematiccomparison between our calculations and published photoemission data shows that only interfacemodels incorporating hydrogen bonding between the dyes are compatible with experiment. Basedon our analysis we propose that at the TiO /N3 interface the dyes are arranged in supramolecularassemblies. Our work opens a new direction in the modeling of semiconductor/dye interfaces andbears on the design of more efficient nanostructured solar cells. a r X i v : . [ c ond - m a t . m t r l - s c i ] A p r mong promising low-cost photovoltaics, dye-sensitized solar cells (DSCs) [1] based onnanostructured TiO films sensitized with the dye Ru(dcbpyH ) (NCS) (N3) have gainedprominence over the past two decades due to their relatively high energy conversion effi-ciencies in excess of 10% [2–4]. In these devices the photocurrent is generated via ultrafastelectron transfer from the photoexcited dye to the nanostructured semiconductor [5]. Sincethe electron injection takes place within a sub-nanometer length scale, the atomistic na-ture of the TiO /N3 interface plays a critical role in the performance of DSCs [6]. Thedye N3 has four carboxylic acid groups [7]. It is generally agreed that the adsorption ofN3 onto the anatase TiO surface occurs through the anchoring of one or more of thesegroups via the formation of Ti-O bonds [8–12]. However the detailed atomic-scale structureof the TiO /N3 interface remains controversial, and questions such as how many and whichcarboxylic groups participate in the bonding to the substrate are being actively debated[8–12].In this work we propose a new atomic-scale model of the TiO /N3 interface by reverse-engineering measured X-ray photoemission spectra (XPS). We first calculate from first-principles the O1 s core-level shifts for a variety of atomistic models of the TiO /N3 interface.We then perform a quantitative comparison between our calculated core-level shifts and theXPS spectra of Ref. 9. Such comparison shows that only interface models which incorporatehydrogen bonding interactions between the dyes are compatible with the measured spectra.Based on our analysis we propose that at the TiO /N3 interface the dyes are arranged insupramolecular hydrogen-bonded assemblies.The existence of competing models of the atomic-scale structure of the TiO /N3 interfaceillustrates the complexity of the problem. Even in the case of an atomically perfect TiO surface there exists a plethora of possible adsorption geometries [13]. Previous computationalstudies have explored the potential energy landscape of one isolated N3 dye adsorbed onthe TiO surface, arguing in favor of specific models on the grounds of adsorption energycalculations [11, 12]. The difficulties with this approach are that (i) a thorough mappingof the total energy landscape is beyond current capabilities, and (ii) energetically favorableconfigurations may be kinetically inaccessible during the fabrication of DSCs. In order toavoid these difficulties from the outset we here follow a completely different strategy and askwhat is the atomic-scale model of the TiO /N3 interface which best reproduces measuredcore-level spectra. Our choice is motivated by the observation that core levels are sensitive to2 C a l c u l a t ed s h i ft ( e V ) Exp. sep. (eV) C a l c . s ep . ( e V ) FIG. 1. Comparison of calculated O1 s core-level shifts of test molecules [18] with the experimentaldata of Refs. 19 and 20. The shifts are referenced to that of the H O molecule, the bindingenergy increases towards negative energies. Blue and red disks indicate the core-level shifts ofmolecules containing carboxylic acid groups. The ball-and-stick representation of the acetic acidshows the carbonyl and the hydroxyl O atoms and the associated shifts (red and blue disks,respectively). Inset: calculated splitting between the shifts of hydroxyl and the carbonyl O atomsvs. the experimental splitting. The r.m.s. deviation is 0.17 eV (note the change of scale). the local bonding environment, and therefore carry the signature of the atomistic interfacestructure.We here consider the O1 s core-level shifts of TiO /N3 interfaces reported in the XPSstudy of Ref 9. All our calculations are based on a generalized gradient approximation todensity-functional theory, and have been performed using the planewave pseudopotentialsoftware package quantum ESPRESSO [14]. Core-level shifts are calculated using the theorydeveloped in Refs. 15 and 16. A detailed description of our computational setup is given assupplementary material [17].Our ability to discriminate between candidate interface models relies critically on accuratecore-level shift calculations. In order to gauge the accuracy of the computational method weconsidered a number of test molecules containing C and O atoms whose structures are wellunderstood. In particular we included molecules which carry carboxylic acid groups COOHsimilarly to the N3 dye. In Fig. 1 we compare our calculated O1 s core-level shifts withexperiment [19, 20]. Our calculations exhibit very good agreement with experiment over awide energy range spanning 7 eV. In the inset of Fig. 1 we concentrate on the molecules3 A d s o r p t i on ene r g y ( e V ) I1 I2bI2aI2c I3bI3aH2aH2b -1.0 0.0 1.0 2.0 3.0 I n t. ( a . u . ) I1 -1.0 0.0 1.0 2.0 3.0O1s shift (eV) I2a -1.0 0.0 1.0 2.0 3.0
I2b (c)(a) (b)
FIG. 2. (a) Ball-and-stick representation of the interface models I1, I2a, and I2b. Color code: Ru(green), S (yellow), N (blue), C (cyan), O (red), H (white), Ti (silver). (b) Calculated adsorptionenergies for each interface model, with stability increasing towards negative energies. (c) CalculatedO1 s core-level spectra for the models I1, I2a, and I2b (black solid lines) and experimental spectrafrom Ref. 9 (red dashed lines). A Gaussian broadening of 1.6 eV, as estimated from the spectraof Ref. 9, has been applied to the calculated spectra in order to account for core-hole lifetimes,vibrational broadening, and configurational disorder. The peak arising from the TiO substrate isnot shown here but can be seen in Fig. 3(c). All the spectra have been aligned to the leftmost peak.The models I2c, I3a, and I3b and their calculated spectra are given as supplementary material [7],together with a quantitative analysis of all the spectral features. containing carboxylic acid groups. In these groups the two oxygen atoms are inequivalent,and the core electrons associated with the hydroxyl (COH) O atom are more tightly boundthan those associated with the carbonyl (CO) O atom. Our calculations describe veryaccurately the differences between the core-level shifts of the hydroxyl and of the carbonylO atoms, with an r.m.s. deviation from experiment below 0.2 eV.For our model substrate we have chosen the anatase (101) surface, which correspondsto the majority [21] of the total exposed surface of the TiO films used in DSCs and inRef 9. We considered eleven adsorption geometries of the N3 dye on this surface, includingpreviously proposed models [8, 10–12]. Schematic representations of these models can beseen in Figs. 2(a),3(a) and in the supplementary material [7]. Each model is labeled by thenumber of carboxylic groups which bind to the substrate. The models I2a and I2b have4een proposed in previous experimental work [8, 10], and the models I2c and I3a have beenintroduced in recent computational studies [11, 12].In order to make contact with previous studies we report in Fig. 2(b) the calculatedadsorption energies for each interface [18]. The calculated adsorption energies span a range of0.6 eV across all the models considered. This range is comparable to the energy of hydrogenbonds between carboxylic acid groups in related systems. Indeed, the energy of the H-bondin the formic acid dimer corresponds to 0.3 eV per monomer [22]. This observation suggeststhat hydrogen-bonding between the carboxylic acid groups of N3 cannot be neglected in theenergetics of N3 adsorption on TiO .In Fig. 2(c) we compare our calculated O1 s core-level shifts [18] with the XPS measure-ments of Ref. 9. The measured spectra exhibit peaks at 529.8 eV, 531.4 eV and 533.2 eV.The peak at the lowest binding energy (529.8 eV) has been assigned to the O atoms of theTiO substrate. The other two peaks have been assigned to the inequivalent O atoms of thecarboxylic groups in the dye [9]. Our calculations correctly reproduce the three measuredpeaks. The uncertainty on the photoelectron escape depth [23], surface stoichiometry, andH-coverage makes the separation between the substrate peak at 529.8 eV and the two dyepeaks unreliable for a quantitative comparison. We therefore concentrate on the dye peaksat 531.4 eV and at 533.2 eV. First we consider the intensity ratios of these peaks. Theintensity of the peak at 533.2 eV scales with the number of protonated carboxylic groupson the dye. The best match between our calculated intensities and experiment is obtainedfor interface models where the dye has two protonated carboxylic groups (I2 and I3 mod-els). In model I1 the dye has three protonated COOH groups, leading to an intensity ratio(0.5) well off the experimental estimate (0.3) [9]. We therefore reject the candidate modelI1 on the grounds of intensity mismatch. Second we consider the binding energy of theadsorbate peaks [18]. As clearly shown in figure 2(c), the separations of adsorbate peaks inall the models of the I2 and I3 families fall within the range 2.3-2.6 eV, and overestimatethe measured peak separation of 1.8 eV [9]. This systematic deviation of 0.5-0.8 eV fromexperiment is well above our 0.2 eV error bar. We have carried out a number of tests inorder to confirm that such deviation is not a numerical artifact [17]. We therefore assignthe mismatch between theory and experiment to the inaccuracy of the models I2-I3.By carrying out a detailed analysis of our calculated core-level shifts we noted thatmoderate changes in the structural parameters of the interface models, such as dye twisting5r bond length variations, only lead to subtle changes in the shifts. We therefore concludethat structural variations across the models are not responsible for the observed 0.5-0.8 eVdeviation.These observations point us towards the possibility that supramolecular interactionswithin the dye monolayer may play a role in the measured XPS spectra. Our calcula-tions for different surface coverages [18] indicate that the separation between the dye peaksis not affected by long-range electrostatic effects. Hence sizeable changes in the calculatedpeak separation can only arise from short-range interactions of the free carboxylic groups inthe dye with other molecules. Such interactions can happen in two ways: either some of theN3 carboxylic groups form bonds with contaminant molecules, or the N3 dyes are bondedto each other within the monolayer.The ex-situ preparation of the TiO /N3 interface of Ref. 9 may lead to the presence ofcontaminant molecules in the system, such as water and hydrocarbons. It is unlikely thatlarge hydrocarbons systematically attach to N3, but H O molecules are small enough to formhydrogen bonds with the COOH groups and may alter the measured XPS spectra. However,our calculations of XPS spectra including water molecules exhibit heavily distorted peakintensities [18], and allow us to exclude this scenario on the grounds of intensity mismatch.The only remaining possibility is that of dye-dye interactions through the free COOHgroups. In order to test this hypothesis we considered two interface models, H2a and H2b,which probe the limiting regimes of strong and weak H-bonding respectively [Fig. 3(a)].Model H2a is derived from model I2a by forming N3 dimers (H bond length 1.51 ˚A). ModelH2b is a self-assembled dye monolayer derived from model I2b (H bond length 2.68 ˚A).Figure 3b shows the XPS spectra calculated for the H-bonded interface models. The agree-ment between theory and experiment is seen to improve dramatically upon inclusion ofsupramolecular interactions between the dyes. The calculated separation between dye peaksnow matches the measured value within our error bar. The modification of the XPS spec-trum arising from supramolecular interactions results from the lowering of the O1 s bindingenergy in the hydroxyl group participating in the hydrogen-bonding. This result is fullyconsistent with previous experiments [24] and calculations [25] on the formic acid dimer. Inmodel H2b this effect is less pronounced due to the weaker H-bond [Fig. 3(b)]. Remarkably,if in the case of model H2a we include the substrate contribution to the spectrum [17],the agreement with experiment becomes excellent [Fig. 3(c)]. These findings indicate that6 I n t. ( a . u . ) H2a -1.0 0.0 1.0 2.0 3.0O1s shift (eV) I n t. ( a . u . ) H2b -2.0 -1.0 0.0 1.0 2.0 3.0 4.0O1s shift (eV) I n t. ( a . u . ) H2afinite λ (a) (b) (c) FIG. 3. (a) Ball-and-stick representations of the interface models H2a and H2b. (b) CalculatedO1 s core-level spectra for the models H2a and H2b (black solid lines) and experimental spectrafrom Ref. 9 (red dashed lines). (c) Calculated O1 s core-level spectrum for the interface model H2aincluding the contribution from the TiO substrate and finite escape depth effects compared to theexperimental spectrum. hydrogen-bonding between dyes is key to interpreting the photoemission data of Ref. 9.It is natural to ask whether additional H-bonded superstructures can exist at the TiO /N3DSC interface. Elementary geometric considerations show that, among all the model in-terfaces considered, models H2a and H2b are the only possible H-bonded homogeneoussupramolecular structures [18]. However, more complex heterogeneous assemblies of dyescannot be excluded.STM experiments could directly probe the proposed H-bonded assembly. Although thereare reports of STM studies on anatase TiO in the literature [26] to the best of our knowledgeno data exists on N3-sensitized (101) surfaces. However STM experiments of N3 on rutileTiO reveal distinctively elongated features (ovals) in the tunneling maps [27]. Figure 3(a)suggests that the dyes in our dimer model H2a would naturally lead to an elongated STMfootprint. Since the calculated Ru-Ru distance of 1.6 nm in our model H2a matches thelength of the ovals in the STM maps (1.8 nm), we speculate that the features observed inRef. 27 may correspond to hydrogen-bonded N3 dimers.It is worth asking whether our conclusions maintain their validity for other importantdyes. The dye (Bu N) [Ru(dcbpyH) (NCS) ] (N719) is structurally similar to N3, the onlydifference being that the protons on two carboxylic acid groups are replaced by counterions[8]. Since in our interface model H2a the H-bonding does not occur through the substituted7roups, model H2a is also a possible candidate for the TiO /N719 interface. Interestingly avery recent infrared and Raman study [28] of the TiO /N719 interface suggested that thedye may be involved in some form of hydrogen-bonding, possibly with the substrate. Ourmodel H2a provides a natural explanation of the data of Ref. 28 in terms of supramolecularhydrogen-bonding.In summary, we established that (i) models of the TiO /N3 interface based on isolateddyes are unable to explain the measured XPS spectra, and (ii) interface models wherethe N3 molecules form supramolecular hydrogen-bonded assemblies are in good agreementwith experiment. Since the lowest photoexcited electronic state of N3 is localized on thebipyridines and the carboxylic groups [29], we expect charge delocalization upon the for-mation of a supramolecular assembly, with potential implications on the light absorptionand the electron transfer mechanisms in DSCs. Our results are expected to hold also forother nanostructured solar cell concepts, such as for instance solid-state DSCs [30, 31].The present finding highlights the importance of supramolecular interactions at semicon-ductor/dye interfaces, and bears implications for the design of more efficient nanostructuredsolar cells.We thank F. De Angelis and H. Snaith for fruitful discussions. This work is supportedby the UK EPSRC and the ERC under the EU FP7 / ERC grant no. 239578. Calculationswere performed in part at the Oxford Supercomputing Centre. Figures rendered using VMD[32]. [1] B. O’Regan and M. Gr¨atzel, Nature , 737 (1991).[2] M. K. Nazeeruddin et al. , J. Am. Chem. Soc. , 6382 (1993).[3] M. K. Nazeeruddin et al. , Inorg. Chem. , 6298 (1999).[4] T. Bessho et al. , J. Am. Chem. Soc. , 5930 (2009).[5] W. R. Duncan and O. V. Prezhdo, Annu. Rev. Phys. Chem. , 143 (2007).[6] F. De Angelis et al. , Nano Lett. , 3189 (2007).[7] (See supplementary material for figures of all the models considered).[8] M. K. Nazeeruddin, R. Humphry-Baker, P. Liska, and M. Gr¨atzel, J. Phys. Chem. B ,8981 (2003).
9] E. M. J. Johansson, M. Hedlund, H. Siegbahn, and H. Rensmo, J. Phys. Chem. B , 22256(2005).[10] H. Rensmo et al. , J. Chem. Phys. , 2744 (1999).[11] F. Schiffmann et al. , J. Phys. Chem. C , 8398 (2010).[12] F. De Angelis, S. Fantacci, A. Selloni, M. K. Nazeeruddin, and M. Gr¨atzel, J. Phys. Chem.C , 6054 (2010).[13] U. Diebold, Surf. Sci. Rep. , 53 (2003).[14] P. Giannozzi et al. , J. Phys.: Condens. Matter , 395502 (2009).[15] E. Pehlke and M. Scheffler, Phys. Rev. Lett. , 2338 (1993).[16] A. Pasquarello, M. S. Hybertsen, and R. Car, Phys. Rev. B , 10942 (1996).[17] (See supplementary material for computational methods).[18] (See supplementary material for discussion).[19] W. L. Jolly, K. D. Bomben, and C. J. Eyermann, At. Data Nucl. Data Tables , 433 (1984).[20] B. H. McQuaide and M. Banna, Can. J. Chem. , 1919 (1988).[21] M. Lazzeri, A. Vittadini, and A. Selloni, Phys. Rev. B , 155409 (2001).[22] S. Tsuzuki and H. P. L¨uthi, J. Chem. Phys. , 3949 (2001).[23] K. Shimada, in Very High Resolution Photoelectron Spectroscopy , edited by S. H¨ufner(Springer, Berlin Heidelberg, 2007).[24] R. W. Joyner and M. W. Roberts, P. Roy. Soc. Lond. A Mat. , 107 (1976).[25] P. Aplincourt, C. Bureau, J. Anthoine, and D. P. Chong, J. Phys. Chem. A , 7364 (2001).[26] X.-Q. Gong, A. Selloni, M. Batzill, and U. Diebold, Nat. Mater. , 665 (2006).[27] A. Sasahara et al. , Surf. Sci. , 106 (2010).[28] K. E. Lee, M. A. Gomez, S. Elouatik, and G. P. Demopoulos, Langmuir , 9575 (2010).[29] S. Fantacci, F. De Angelis, and A. Selloni, J. Am. Chem. Soc. , 4381 (2003).[30] U. Bach et al. , Nature , 583 (1998).[31] H. J. Snaith and L. Schmidt-Mende, Adv. Mater. , 3187 (2007).[32] W. Humphrey, A. Dalke, and K. Schulten, J. Mol. Graphics , 33 (1996). ydrogen-bonded supramolecular assembly of dyesat nanostructured solar cell interfaces:Supplementary Information Christopher E. Patrick and Feliciano Giustino
Department of Materials, University of Oxford,Parks Road, Oxford OX1 3PH, United Kingdom
Computational MethodsSupplementary Note 1
Discussion of the energetics of hydrogen bonds
Supplementary Note 2
Discussion of possibility of forming otherhomogeneous supramolecular structures
Supplementary Figure 1
Ball-and-stick model of the N3 dye
Supplementary Figure 2
Ball-and-stick and O1 s spectra of models I2c, I3a, I3b Supplementary Figure 3
Effect of increased dye coverage on O1 s spectra Supplementary Figure 4
Effect of H O on O1 s spectra Supplementary Table 1 O1 s shifts of test molecules Supplementary Table 2
Analysis of O1 s peak separations and intensities forinterface models omputational Methods The calculations were performed using density functional theory (DFT) within the gen-eralized gradient approximation of Ref. 1. We used periodic simulation cells and describedthe electronic wavefunctions and charge density using plane wave basis sets as implementedin the
Quantum ESPRESSO software distribution [2]. The core-valence interaction was takeninto account by means of ultrasoft pseudopotentials [3]. The structures were relaxed viadamped Car-Parrinello molecular dynamics by sampling the Brillouin zone at the Γ point[4, 5]. In order to generate the substrate model we optimized the bulk anatase TiO latticeparameters by sampling the Brillouin zone on six inequivalent Monkhorst-Pack points whilekeeping the I4 /amd symmetry fixed. We constructed a stoichiometric slab by taking a cutthrough the bulk TiO anatase such that the (101) surface was exposed. We described theTiO surface using a rectangular slab of area 20 . × . for the six interface modelsI1-I3. In the case of the H-bonded configurations we used a periodic rectangular slab witharea 31 . × . for model H2a, and an oblique cell with area 138 ˚A for model H2b.For all interface models the thickness of the TiO slabs was fixed to 12 layers of atoms (5.8˚A) and the interaction between periodic replicas along the direction perpendicular to thesurface was minimized by including a vacuum region of 10 ˚A. In order to compensate for theelectrostatic interactions between the interface replicas we calculated the core-level shiftsincluding self-consistently the dipole correction of Ref. 6. During the geometry optimiza-tions the bottom three layers of the slab were fixed in their bulk positions in order to mimicthe structure of the TiO nanoparticle far from the surface. Structural relaxations werecarried out until the force on each atom was below 0.07 eV/˚A. Calculations were found tobe converged with kinetic energy cutoffs of 35 Ry and 200 Ry for the electron wavefunctionsand charge density, respectively.O1 s core-level shifts were calculated by following the method of Refs. 7, 8 which takes intoaccount final state effects. In the calculations with a core hole charge neutrality was restoredby using a positive jellium background. For the gas phase molecules we calculated the core-level shifts for various cell sizes and used the Makov-Payne expansion [9] for the extrapolationto infinite simulation cells. Due to error cancellation the difference between the O1 s shifts oftwo O atoms in the same computational cell converges much faster with increasing cell sizeas compared to the individual shifts. For example, in the case of formic acid on changing the3ell size from 9 × × to 27 × ×
27 ˚A the individual shifts of the O atoms change by0.25 eV. However, the difference between the shifts of the carbonyl O atom and the hydroxylO atom change by less than 0.02 eV. For this reason in the interface calculations we usedsimulation cells identical to those adopted for the geometry optimizations. Core-level shiftcalculations for the interface models using thicker slabs of 24 atomic layers did not yield anysignificant differences with respect to the 12-layer calculations.There is an uncertainty in the calculation of the substrate O1 s peak at 529.8 eV. Infact the number of TiO layers which contribute to this peak depends on the photoelectronescape depth. In addition surface dipoles associated with possible surface defects may affectthe energy separation between the substrate and the two adsorbate peaks at 531.4 eV and533.2 eV. In the calculation of the full spectrum in Figure 3(c) we only included the topmostlayer of O atoms, and we used an escape depth of 10 ˚A. The latter value has been estimatedfrom the inelastic mean free paths reported in Ref. 10 for a photon energy of 758 eV [11].Since DFT might not describe hydrogen bonds accurately, we conducted extensive testson the geometry and core-level shifts of H-bonded systems. For example we calculated thehydrogen-bond energy in the formic acid dimer to be 0.38 eV per molecule. This valuecompares favorably with the value of 0.3 eV from the MP2 calculations and spectroscopicdata reported in Ref. 12. The difference between the core-level shifts of the carbonyl O atomand the hydroxyl O atom was calculated to be 1.19 eV. This value compares favorably withexperiment, yielding 1.3 eV [13].The broadening of the photoemission data of Ref. 11 arises from the finite lifetimes of thecore-holes, from vibrational broadening, and from the averaging over all the possible adsorp-tion configurations. In the present study we do not address these aspects. In particular ourbest candidate interface models are meant to describe only the dye adsorption configurationwith the highest yield.While the present study focuses on XPS experiments performed on dry interfaces (i.e.without the redox electrolyte), our conclusions are expected to remain valid even for completeDSC devices because the electrolyte is introduced after the sensitization step, when theTiO /N3 interface has already formed. The present study also bear relevance to solid-stateDSCs where the electrolyte is replaced by a molecular hole-transporter [14, 15].4 upplementary Note 1: The hydrogen-bonded adsorption models considered in the main text are derived from thesingle molecule adsorption modes I2a and I2b. Model H2a is obtained from model I2a byforming hydrogen bonds between two dyes (bond length 1.51 ˚A). Model H2b is a hydrogen-bonded chain of dyes derived from model I2b (bond length 2.68 ˚A). The formation of hy-drogen bonds stabilizes the interface in both models, by 270 meV per dye in model H2a,and by 20 meV per dye in model H2b [Figure 2(b)].H-bonded dye chains were also proposed in Ref. 16 but in that work one of the twoH-bonded carboxylic groups is deprotonated, while the same carboxylic groups are fullyprotonated in our model H2b. We tested the interface model proposed in Ref. 16 and foundthat the loss of a proton halves the intensity of the dye peak at 533.2 eV, resulting in acalculated core-level spectrum in sharp disagreement with experiment. Incidentally we notethat we calculate the model proposed in Ref. 16 to be considerably less stable than all theother models considered here.
Supplementary Note 2:
While a dimer of model I2b could be formed by mixing different enantiomers of the N3dye, there would be a mismatch between the positions of the anchor carboxylic groups andthe Ti chemisorption sites on the TiO surface. The formation of strong H-bonds within amonolayer of dyes adsorbed as in the interface models I3a, I3b and I2c is prevented by theunadsorbed carboxylic group pointing out of the layer. The interface model I1 is rejectedon grounds of intensity mismatch. 5 IG. 1. The N3 moleculeA central Ru atom is sixfold coordinated to the N atoms of two bipyridines and two thiocyanateligands, with each bipyridine carrying two carboxylic acid groups (highlighted). The carboxylicacid groups anchor the molecule to the substrate via Ti-O bonds. IG. 2. Additional single-molecule adsorption models considered (a) Ball-and-stick representations of interface models I2c, I3a, and I3b. In interface models I2a and I2b[Fig. 2(a) in main text] the dye binds to the TiO substrate via two carboxylic groups, both in bridgingbidentate modes. In both models the H atoms from the binding carboxylic groups bind to the substrate Oatoms [17]. In model I2a the bridging carboxylic groups belong to the same bipyridine [18], while in modelI2b the carboxylic groups belong to different bipyridines [19]. In model I2c, the two carboxylic groupswhich participate in binding belong to different bipyridines, with one in a bridging mode and the other ina unidentate mode. This configuration was proposed in Ref. 16. In model I3a the N3 dye binds to thesubstrate via one bridging carboxylic group and two unidentate groups. This configuration was proposedin Ref. 20 for the related N719 dye. In model I3b the three carboxylic groups bind to the substrate inunidentate modes. In the three interface models I2c, I3a and I3b the interaction of the dye with the protonson the TiO surface stabilizes the structure, consistent with studies of formic acid on the same surface [17].Model I3b is the most energetically favourable owing to the minimum strain exerted on the dye molecule.(b) Calculated O1 s core-level spectra for the interface models I2c, I3a, and I3b (black solid line), compared tothe experimental spectrum of Ref. 11 (red dashed line). The calculated spectra systematically overestimatethe separation of the peaks as obtained from experiment, as indicated by the vertical dashed lines. Thepeak energies and intensities are reported in Supplementary Table 2. We note from both here and Figure2(b) in the main text that substantial changes in structural parameters lead to subtle changes in shifts.For instance, the spectrum of model I2c is remarkably similar to that of I2a and I2b, even though the dyebinds to the substrate through 4 Ti-O bonds in models I2a and I2b and only 3 Ti-O bonds in I2c. It isalso interesting to note that the disagreement between theory and experiment is most severe for model I3b,which is calculated to be the most energetically stable configuration [Figure 2(b) in the main text]. IG. 3. Effect of increased dye coverage on O1 s spectra (a) Ball-and-stick representation of the interface model I2c with two dyes per simulation cell. In this newconfiguration the areal density is 0.5 nm − [compared to 0.25 nm − for the data presented in SupplementaryFigure 2(b)]. The shortest distance between the carboxylic groups of neighbouring dyes is 4.4 ˚A. Conse-quently no hydrogen bonds are formed between neighbouring dyes.(b) Calculated O1 s core-level spectrum for the model interface with higher surface coverage (blue dottedline), compared to the calculated spectrum of model I2c (black line). The peak energies in the two spectradiffer by less than 0.03 eV. We conclude that the observed discrepancy in peak separation in the experimentaland calculated spectra is not explained by long-range electrostatic effects. IG. 4. Effect of H O on O1 s spectra (a) Ball-and-stick representation of the interface model I2c with additional water molecules forming H-bondswith each carboxylic acid group in the dye. The bond lengths of the H-bonds formed with the hydroxylgroup and with the carbonyl group are 1.80 ˚A and 2.05 ˚A respectively.(b) Calculated O1 s core-level spectrum of this interface model. The H-bonding reduces the separationbetween the two dye peaks to 2.1 eV, leading to a good agreement with the experimental separation of 1.9eV measured in Ref. 11 (red dashed line). However, the O atom of the H O molecule contributes to theleftmost peak. The resulting intensity ratio of 0.5 is in disagreement with the experimental value of 0.3.On taking into account the finite escape depth of the photoelectrons the disagreement becomes even morepronounced. In general, if other contaminant molecules were to alter the measured XPS spectra, their arealdensity would have to be comparable to the dye surface coverage, and additional features would appear inthe measured XPS spectra. This scenario is in contrast with the findings of Ref. 11. ABLE I. O1 s shifts of test molecules. All values are referenced to the shift of the H O molecule.For molecules containing carboxylic acid groups the shifts of the carbonyl O atoms and of thehydroxyl O atoms are reported separately. In the other cases where a molecule has more than oneO atom, the relevant atom is indicated in boldface. The experimental data are from Refs. 21, 22.We have also calculated the separation between carbonyl and hydroxyl O1 s core-level shifts in N3(2.5 eV), isonicotinic acid (2.1 eV) and bi-isonicotinic acid (2.2 eV) but we are unaware of anypublished gas phase data for these low volatility molecules. Molecule Theory Experiment(eV) (eV)H O 0.00 0.00O -4.18 -3.8N O -1.82 -1.5F C O O O CF (average) -1.83 -1.6F CO O OCF -2.94 -2.8N-(OH)-2-Pyridone (CO) 2.85 2.9N-(OH)-2-Pyridone (OH) 0.09 -0.2CF NO -2.12 -2.4HCOOH (Formic acid) CO 0.99 0.9HCOOH OH -0.80 -0.7CH COOH (Acetic acid) CO 1.64 1.6CH COOH OH -0.25 -0.2CH CH COOH (Propionic acid) CO 1.90 1.6CH CH COOH OH 0.01 -0.1C H OH (Phenol) 0.76 1.0F CHCOOH (Difluoroacetic acid) CO 0.78 0.6F CHCOOH OH -1.12 -1.1CF COOH (Trifluoroacetic acid) CO 0.42 0.3CF COOH OH -1.45 -1.4C H COOH (Benzoic acid) CO 2.42 2.2C H COOH OH 0.30 0.1C H (COOH) (Phthalic acid) CO 2.51 1.8C H (COOH) OH 0.42 -0.1C H (COOH) (Isophthalic acid) CO 2.28 1.9C H (COOH) OH 0.10 -0.2C H (COOH) (Terephthalic acid) CO 2.28 1.8C H (COOH) OH 0.02 -0.2 ABLE II. Calculated energy separation and intensity ratio between the two dye peaks for eachinterface model considered, compared to the experimental data of Ref. 11.The calculations were performed in the limit of infinite escape depth.
Model separation intensity(eV) ratioI1 2.45 0.50I2a 2.25 0.25I2b 2.41 0.25I2c 2.41 0.25I3a 2.48 0.25I3b 2.59 0.25H2a 1.92 0.21H2b 2.24 0.25Experiment 1.8 0.31
1] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. , 3865 (1996).[2] P. Giannozzi et al. , J. Phys.: Condens. Matter , 395502 (2009).[3] D. Vanderbilt, Phys. Rev. B , 7892 (1990).[4] R. Car and M. Parrinello, Phys. Rev. Lett. , 2471 (1985).[5] K. Laasonen et al. , Phys. Rev. B , 10142 (1993).[6] L. Bengtsson, Phys. Rev. B , 12301 (1999).[7] E. Pehlke and M. Scheffler, Phys. Rev. Lett. , 2338 (1993).[8] A. Pasquarello, M. S. Hybertsen, and R. Car, Phys. Rev. B , 10942 (1996).[9] G. Makov and M. C. Payne, Phys. Rev. B , 4014 (1995).[10] K. Shimada, in Very High Resolution Photoelectron Spectroscopy (H¨ufner S. (ed); Springer,Berlin Heidelberg, 2007) pp. 85–112.[11] E. M. J. Johansson, M. Hedlund, H. Siegbahn, and H. Rensmo, J. Phys. Chem. B , 22256(2005).[12] S. Tsuzuki and H. P. L¨uthi, J. Chem. Phys. , 3949 (2001).[13] R. W. Joyner and M. W. Roberts, P. Roy. Soc. Lond. A Mat. , 107 (1976).[14] U. Bach et al. , Nature , 583 (1998).[15] H. J. Snaith and L. Schmidt-Mende, Adv. Mater. , 3187 (2007).[16] F. Schiffmann et al. , J. Phys. Chem. C , 8398 (2010).[17] A. Vittadini, A. Selloni, F. P. Rotzinger, and M. Gr¨atzel, J. Phys. Chem. B , 1300 (2000).[18] H. Rensmo et al. , J. Chem. Phys. , 2744 (1999).[19] M. K. Nazeeruddin, R. Humphry-Baker, P. Liska, and M. Gr¨atzel, J. Phys. Chem. B ,8981 (2003).[20] F. De Angelis, S. Fantacci, A. Selloni, M. K. Nazeeruddin, and M. Gr¨atzel, J. Phys. Chem.C , 6054 (2010).[21] W. L. Jolly, K. D. Bomben, and C. J. Eyermann, At. Data Nucl. Data Tables , 433 (1984).[22] B. H. McQuaide and M. Banna, Can. J. Chem. , 1919 (1988)., 1919 (1988).