Polaron-Adsorbate Coupling at the TiO2(110)-Carboxylate Interface

Understanding how adsorbates influence polaron behavior is of fundamental importance in describing the catalytic properties of TiO2. Carboxylic acids adsorb readily at TiO2 surfaces, yet their influence on polaronic states is unknown. Using UV photoemission spectroscopy (UPS), two-photon photoemission spectroscopy (2PPE), and density functional theory (DFT) we show that dissociative adsorption of formic and acetic acids has profound, yet different, effects on the surface density, crystal field, and photoexcitation of polarons in rutile TiO2(110). We also show that these variations are governed by the contrasting electrostatic properties of the acids, which impacts the extent of polaron–adsorbate coupling. The density of polarons in the surface region increases more in formate-terminated TiO2(110) relative to acetate. Consequently, increased coupling gives rise to new photoexcitation channels via states 3.83 eV above the Fermi level. The onset of this process is 3.45 eV, likely adding to the catalytic photoyield.

characterized by XPS, which produces characteristic C 1s signatures due to the ratio of COOand CH3 contributions. Partial ML coverages were characterized by XPS C 1s ratios, UPS OH 3 ratios or 2PPE workfunction changes, depending on the experiment.

Theoretical Considerations -DFT
Spin polarized DFT calculations were conducted using the open source package CP2K. 3,4 Normconserved Goedecker-Teter-Hutter (GTH) pseudopotentials 5 were used to describe the interactions between ion cores and valence electrons. A molecular optimized double zeta mixed Gaussian-planewave basis set was employed to represent the electronic wavefunctions with a plane wave energy cut-off of 280 Ry. To properly describe the BGS and resonance states of excess electrons, 6 the hybrid functional of Heyd, Scuseria, and Ernzerhof (HSE06) was used for all the calculations. 7,8 Only the Γ point was sampled in reciprocal space due to the use of large supercell models. All atoms in the slab were relaxed and the convergence threshold was ~ 0.02 eV/Å.
The rutile TiO2(110) surface was modeled using slabs of six TiO2 tri-layers and (4×2) surface supercell with dimensions (11.836 Å × 12.949 Å). As shown in Figure 1(a), Tiint was introduced into location L1, L2 or L3 of the 6 tri-layers slab, respectively. The inclusion of a Tiint defect results in 4 excess electrons in the system. To explore the effects of carboxylic acid adsorption, 2×1 monolayers consisting of dissociated bridging bidentate formate or acetate species were considered. Each bidentate species is bonded to two Ti5c sites, while the dissociated proton is adsorbed at an Obr site. Oscillator strengths were calculated according to: 6 = 2 Here, is the oscillator strength in the 5555⃗ polarization direction, ⟨ | and | ⟩ denote the Kohn-Sham orbitals corresponding to the BGS electrons and unoccupied MO's, respectively, % and & are the corresponding eigenvalues.
is the momentum operator along 5555⃗. The Fermi energy EF was defined as the conduction band minimum (CBM), in agreement with experimental observations.
- Figure S2 Example of normalization and fitting procedure of UPS and 2PPE spectra.
- Figure S3 Example of a continuous 2PPE (3.76 eV, 330 nm) experiment involving the heating of a formate saturated rutile TiO2(110) sample to remove feature 2.
- Figure S4 PDOS and oscillator strengths for Tiint located at L1, L2 and L3 at the clean, formate and acetate terminations.
- Table S2 Surface induced broadening of BGS.
- Figure S6 Auger features in 2PPE Spectra -Discussion and Extension of Figure 3(a), 2PPE measurements of the acetate terminated TiO2(110) surface with higher energy photons.

Normalization of 2PPE spectra
In previous work, the comparison of TiO2 2PPE spectra across experiments is achieved by normalizing at the workfunction cut-off energy. [9][10][11] Normalization is required principally because of fluctuations in laser power typical of femtosecond laser systems or when changing the photon energy. Taking this approach assumes that the 2PPE signal at the cut-off remains constant. However, this assumption becomes unreliable when the workfunction (and therefore cut off energy) changes. We address this issue in two ways. Firstly, in Figures 1(b,c) and Figure 2(a), where the workfunction changes significantly, we take measurements in-situ to reduce fluctuations in laser power. Secondly, in Figures 3(a,b) and Figures S5,6, where the workfunction is constant, we identify an Auger feature that acts as an ideal normalization point (see S6 for further details).

Fitting of 2PPE spectra
The fitting of TiO2 2PPE spectra has typically been achieved with Gaussian or Voigt profiles with Tougaard or linear backgrounds. 9,12,13 When using peak fitting to isolate peaks (Figures 1(b) and 2(c)) we apply Tougaard, linear and exponentially modified Gaussian backgrounds to determine the energy and intensities of our peaks. We restrict peak fitting to features far removed from the background low energy cut-off, where the background signal is minimal.

Background removal in UPS spectra
The background removal for UPS spectra employed a Tougaard function. This is a well-established method that describes the secondary electron contribution by a calculated loss-function.   [14], Linear [13] and exponentially modified Gaussian (shown in (d)) to accurately determine the peak location.

S3 -Formate decomposition on rutile TiO2 (110)-monitored by 2PPE
To confirm the dependence of feature 2 on the formate overlayer, 2PPE measurements were recorded from the formate saturated surface as it was heated. At ∼340 K, an increase in the workfunction indicates the disruption of the formate overlayer, which is accompanied by the sharp disappearance of feature 2. By 385 K, the workfunction value reaches a plateau at ∼5.1 eV, which is close to the value of the freshly prepared reduced surface, indicating that both the majority and minority species are disrupted by the heating. Upon cooling, the workfunction decreases slightly by 0.1 eV and feature 2 does not reappear.  High Low

Effects of Tiint and adsorbed carboxylates on the crystal field
At rutile TiO2(110) surfaces, the calculated oscillator strength of the feature 1 transition is enhanced when the surface contains hydroxyls. This is due to the occupied t2g state acquiring more pronounced dxy character, and thus coupling more effectively to the Ti 3+ resonant excited states in the conduction band (CB). 6,15 This trend is maintained on formate and acetate covered rutile TiO2(110) surfaces. However, the presence of carboxylates has a distinct impact on excess electrons at Tiint. As can be seen from the PDOS, occupied states of dz2 character appear in carboxylate terminated models. The spin density contour shows that these dz2-like states are solely located at the Tiint site, indicating that those electrons are more appropriately described according to a trigonal prismatic crystal field rather than the usual octahedral field of Ti ions in TiO2. On the other hand, electrons at a larger distance from the interstitial remain in an octahedral field and occupy orbitals of t2g-like symmetry. This gives rise to a complex mixture of states induced by the adsorbate. The proportion of electrons present in each crystal field is dependent on the properties of the specific carboxylate. Formate attracts excess electrons towards it, meaning there is a higher proportion of electrons away from Tiint, and therefore more electrons in an octahedral field. This is evidenced in the PDOS by an increased density of states with dxz character. Oscillator strength calculations suggest that these states couple with dz2 orbitals and give rise to feature 2 in the 2PPE spectra. This also explains the absence of feature 2 in the acetate terminated models, where there is an increased density of occupied dz2 states. Table S2 Energy difference between the highest and lowest BGS peak in the PDOS of Tiint in layers 1-3 of pristine TiO2(110) (C-R110), the formate/OH covered surface (FA-R110) and acetate/OH covered surface (AA-R110). (Unit: eV)

Effects on the distribution of gap states
The adsorbate induced surface structural distortions affect not only the excitations but also the distribution of the occupied states. As shown in Table S2, for Tiint at L3, the energy distribution of the BGS is rather narrow, with all 4 states in a range 0.2-0.3 eV. For Tiint at L2, all terminations experience broadening relative to L3, with BGS at the clean and formate termination separated by 0.28 and 0.32 eV, and those at the acetate termination by 0.59 eV. For Tiint at both L2 and L3, carboxylate adsorption also induces a shift of the BGS peaks to slightly lower energies. Interestingly, for Tiint at L1, the acetate termination shows an energy separation between the BGS of only 0.26 eV, smaller than for Tiint at L3 and L2. The different behavior of this termination is likely related to the electron donating effects of the methyl group.

S5 -Further 2PPE spectra of FA-R110 (p-[001])
This figure shows further 2PPE spectra (p-[001], 3.35 -3.87 eV, 370 -320 nm) of FA-R110. Four features are visible: coherent 2PPE from the valence band (VB) maximum, Auger features (see also Figures 3(a), S6), feature 1 and coherent 2PPE from the BGS. The clearer appearance of the coherent 2PPE BGS feature at higher photon energies (>3.65 eV) in FA-R110 compared with AA-R110 is in line with the UPS results in Figure 1(c). The origin of this feature can be identified from its photon energy dependence and has been reported previously. 12,13 The Auger feature has an identical distribution to that in spectra of AA-R110, making it an ideal point for normalization. There is no evidence of the feature 2 peak seen in the s-[001] spectra.  increases, there is an increase in contribution from coherent 2PPE from the valence band tail and the appearance of the coherent 2PPE feature from the BGS. No evidence of a feature 2 peak is observed at any photon energy.

S6 -Auger electrons in the 2PPE spectra of FA-and AA-R110
In the 2PPE spectra of AA-and FA-R110 (Figures 3(a), S5, S6) there is a broad signal centered at 5.2 eV (E-EF), that has increased intensity when light is polarized in the p-[001] orientation. The electron kinetic energy is unaffected by the photon energy and hence it is assigned to an Auger feature. The energy of the feature suggests a BGS origin, which would require an excitation energy of ∼6.0 eV from a recombination process, given the BGS is centered at ∼0.8 eV BE. A possible Auger process is therefore the ejection of BGS electrons following the multiphoton excitation and recombination of electrons from the VB to CB Ti 4+ d-orbital states (see Figure S6, right hand side). The initial excitation must be to a state between the energy of the Ti 3+ intermediate state (IS) and the vacuum level (Evac), labeled on the figure 'Auger range'. In this picture, if VB electrons are excited within the Auger range, they can relax to a high cross-section state ∼3.0 eV above EF. If VB electrons are excited above the Auger range (exceeding the workfunction) then they are ionized, and the Auger feature is not observed. This also coincides with the dominance of the coherent 2PPE VB feature. This can be seen in Figure S6 (left hand side) at ℎ > 4.00 eV (310 nm), which is an extension to Figure 3(a). Although the reduction in workfunction allows for the observation of the Auger feature, previous work suggests the carboxylate overlayers themselves are the crucial factor in its appearance. 2PPE spectra with monolayer water covered TiO2(110) does not contain an Auger feature despite the workfunction decreasing to 3.8 eV. 12,13 The possible reason for the difference between carboxylate terminated TiO2(110) and water covered TiO2(110) is that carboxylates facilitate the diffusion of photogenerated holes to the surface, where they are stabilized. [16][17][18] The Auger feature suggests a natural position for normalization. Because these electrons arise from the recombination of a distinct state in the CB, both the relative intensity and position should be independent of the photon energy. Normalizing at this location leads to a number of observations that agree with well-established trends in the 2PPE spectra of TiO2(110) such as the photon energy intensity dependence of feature 1 and the increase of the coherent VB feature at higher photon energies. We also note that at ℎ > 4.00 eV (310 nm), feature 1 is absent in the 2PPE spectra of AA-R110 and FA-R110 but present in 2PPE spectra of the clean surface. 10,13 The loss of feature 1 coincides with the dominant appearance of the coherent 2PPE VB feature, suggesting that multiphoton excited electrons in the Auger range may play a role in populating the Ti 3+ IS at ∼2.6 eV above EF at non-resonant 2PPE conditions.  TiO2(110) interface. As the photon energy increases, there is a large increase in contribution from coherent 2PPE from the valence band tail. When this coherent feature is particularly prominent, the Auger process is less pronounced, likely due to the decay process being less probable due to multiphoton photoemission. The inset shows the dashed box, expanded. Feature 1 follows the expected wavelength dependence. Right hand side -Schematic of processes leading to features in the 2PPE spectra of AA-R110.

S7 -Further 2PPE spectra of AA-R110 (s-[001])
This figure extends the AA-R110 dataset from Figure 3(d) in the main text. Feature 2 is not observed in the 2PPE spectra at any photon energy examined. Furthermore, as expected in this orientation, a weak feature 1 signal is observed. Figure S7 - 17 However, with a surface coverage of 0.5 ML formate, 1500 L of O2 is required for the 2PPE feature 1 signal to be reduced to the intensity before exposure to formic acid. Presumably this is due to formate blocking the O2 adsorption sites, which at the reduced/hydroxylated surface are known to be at Ovac and Ti5c atoms. 19 This is supported spectroscopically here by three pieces of evidence. First, upon exposure of the Hp-R110 surface to O2, there is a 0.4 eV increase of the workfunction after 1 L exposure. 17 Oxygen is known to be an electron accepting adsorbate, enhancing the surface dipole moment in the [110] direction and increasing the workfunction. In contrast, at the FA-R110 surface, there is a slow workfunction change that reaches a maximum increase of ∼0.15 eV after 1500 L. The slow rate of small change in workfunction suggests that O2 simply has very limited active adsorption sites at this termination. Secondly, no shift in the intermediate state energy is observed following O2 dosing on this surface (1500 L), suggesting that the 2PPE signal is still arising from surface localized BGS. 17 Thirdly, there is no change in the UPS spectrum after O2 exposure.