Mapping surface-modified titania nanoparticles with implications for activity and facet control

The use of surface-directing species and surface additives to alter nanoparticle morphology and physicochemical properties of particular exposed facets has recently been attracting significant attention. However, challenges in their chemical analysis, sometimes at trace levels, and understanding their roles to elucidate surface structure–activity relationships in optical (solar cells) or (photo)catalytic performance and their removal are significant issues that remain to be solved. Here, we show a detailed analysis of TiO2 facets promoted with surface species (OH, O, SO4, F) with and without post-treatments by 31P adsorbate nuclear magnetic resonance, supported by a range of other characterization tools. We demonstrate that quantitative evaluations of the electronic and structural effects imposed by these surface additives and their removal mechanisms can be obtained, which may lead to the rational control of active TiO2 (001) and (101) facets for a range of applications.


Supplementary Note 1: 31 P MAS NMR analysis of TMP adsorbed on metal oxide.
Pioneered by Lunsford and co-workers, TMP was first adopted as a probe molecule to characterize the acidity of zeolite based on the observed 31 P chemical shift (δ 31 P) 3 . Thereafter, the technique has been widely utilized for acidity characterization of various solid acid catalysts 4 .
Supplementary Fig. 3 shows three scenarios of interactions between TMP and metal oxide: (a) with metal cation LA center; (b) with hydroxyl proton LA center (hydrogen bonding interaction); (c) on bridging hydroxyl proton (Brønsted acid, BA) site, the formation of TMPH + complex). The δ 31 P of adsorbed TMP spans over a wide range (-20~-58 ppm) when interacting with various metal cations on different solid acids (i.e. case (a)), whereas a TMPH + ionic complex formed when a TMP molecule adsorbs onto a bridging hydroxyl proton tends to give rise to a 31 P resonance in a much narrower range of -2 to -5 ppm (i.e. case (c)). Therefore, Brønsted (proton donor) and Lewis acid (electron acceptor) sites presented in a solid acid catalyst can be readily distinguished using 31 P ssNMR of adsorbed TMP.
On the other hand, TMP on an isolated hydroxyl proton surface usually gives a signal at higher field (~-61 ppm, i.e. case (b)).

Supplementary Note 2: EPR study of as-prepared PD, F-(101) and F-(001) samples with different treatments.
As shown by Wöll's group 5

Supplementary Note 3: Raman study of as-prepared F-(001), F-(101) and PD samples with different treatments.
According to previous literature [8][9][10] , the removal of surface fluorine can be monitored by Raman spectroscopy. It has been shown that the surface attached fluorine changes both "symmetry of Ti-O-Ti" and "coordination of surface Ti atom", resulting in the "shift of low-frequency E g " and "weakening of B 1g (cf. A 1g )" after fluorine removal. However, from our experiment result, only a marginal shift of lowfrequency E 1g is observed ( Supplementary Fig. 8a-c). Calcination treatment ( Supplementary Fig. 8d-f), as expected, results in B 1g > A 1g , while the intensity B 1g = A 1g case is observed herein on NaOH washed samples. These observations give hints the performances on the removal of surface F (calcination or NaOH wash) which showed the change in the coordination of surface Ti atom.

Supplementary Note 4: Na + ions left on TiO 2 surface after NaOH wash.
Detailed XPS scanning in the Na 1S region has been carried out over the samples with preferential exposure of (001) facet (i.e. (F-(001)) and (101) facet (i.e. PD, prepared without HF). No Na 1S signal at 1072 eV for both F-(001) and PD and their corresponding calcination samples (i.e. Cal-(001) and Cal-PD) is detected (Supplementary Fig. 10). Notice that the broad signals at 1067 eV and 1073 eV are the Ti LMM Auger signals. A very small trace of Na 1S signal can be marginally detected for both Na- (001) and Na-PD (green line) after the samples were pre-treated with 0.1M NaOH, followed by rinsing with DI water several times (> three times). This suggests majority of Na + ions had been removed without interfering to the measured chemical shift values of TMP by NMR. The Na + on surface can only be quantifiable by XPS for the sample treated with 0.5M NaOH (~6.29%, blue line) 11 .
In addition to (001) facet, similar result was also obtained for samples with preferential exposed (101) facet (i.e. PD, Cal-PD and Na-PD). As shown in Supplementary Fig. 16a, the photocatalytic activity is correlated to the overall concentration of Lewis acid (LA) sites: 712.1 μmol/g of PD > 596.9 μmol/g of Na-PD > 84.8 μmol/g of Cal-PD. We also carried out the photocatalytic testing on Degussa P25 for comparison. P25 with less than one fourth LA concentration (151.7 umol/g) to that of PD (712.1 umol/g) exhibit comparable photocatalytic activity. The large difference in LA concentration could be attributed to their surface area: PD (123.3 m 2 /g) > P25 (40.3 m 2 /g). However, the similar photocatalytic activity implies there is another factor overrides the total LA concentration in P25 case. It is noted that all TiO 2 samples compared in this study are single crystalline 100% anatase structure with different ratio of (001) and (101) surface. While P25 is a well-known polycrystalline TiO 2 nanoparticle containing more than 70% anatase with a minor amount of rutile and sometimes a small amount of amorphous phase. The ratio of crystalline composition (anatase to rutile) of P25 has been found changed from time to time even though they are from the same package 12 . Similar fluctuations of crystalline composition of P25 has also been reported before 13,14 . The intrinsic interfaces between those anatase and rutile domains have been demonstrated greatly improve charge separation efficiency because of the well-formed type-II band alignment at the anatase and rutile interface 15 . 31 P MAS NMR study of TMP-adsorbed Degussa P25 ( Supplementary Fig. 16b) shows a main signal of surface anatase Ti 5c (101) at -35 ppm as our PD sample, while the shoulder with irregular shape appearing at lower field can be attributed to Ti 5c from surface amorphous or rutile phase. Considering the factors from both inside (charge separation, poly/single crystallinity) and outside (Ti 5c from anatase/rutile/amorphous) particle, it is thus difficult to study the correlation between surface and catalytic result by a simple comparison of catalytic result between polycrystalline P25 and all other anatase samples in this study. However, this is a good example to illustrate the importance of factor isolation from particle side (both intrinsic and extrinsic) in an aim to correlate the corresponding catalytic activity. By carefully tuning those factors one at a time, we believe those different interpretations and frequently disagreements amongst researchers can be largely avoided.

Supplementary Note 6: The system setup for preparation of TMP-adsorbed samples.
About 150 mg of TiO 2 was placed in a home-made glass tube and activated at 150 o C for 2 h under vacuum (10 -1 Pa) to ensure maximum adsorption of TMP molecules. After cooling down to room temperature, the system connecting TMP tube and sample tube ( Supplementary Fig. 19) was isolated from the left part of vacuum system before the introduction of TMP molecules. 300 μmol/catalyst g (calculated by the pressure and volume of isolated system) of TMP was then introduced into this system.
Wait for ~10 min until the pressure of this isolated system reach a plateau, which means the equilibrium between TMP and catalyst surface has been achieved. The tap to TMP and sample tubes were then closed before the removal of extra TMP molecules by left vacuum system. These steps were repeated three times to ensure the fully adsorption of TMP on catalyst surface. The sample tube was then flame sealed for storage and transferred to Bruker 4 mm ZrO 2 rotor with a Kel-F endcap in a glove box under nitrogen atmosphere before NMR measurement.

Supplementary Note 7: 31 P MAS NMR experiments.
Solid state magic angle spinning (MAS) NMR experiments were carried out using a Bruker Avance III 400WB spectrometer at room temperature. To remove the effect of proton spins on 31 P spectra, a strong radio frequency field (B) is usually applied in a pulsed at the resonance frequency of the non- A simple one-pulse sequence as shown in Supplementary Fig. 20a  peak intensity as 100, the relative 31 P peak areas of their mixture B and C were found to match very well with the numbers of 31 P nuclei in each mixture.
Although cross polarization (CP) technique has been widely employed in solid-state NMR to enhance the signal of nuclei with low gyromagnetic ratio or long T 1 relaxation. For the case of TMP, the abundant nucleus is 1 H and the observed nucleus is 31 P. If the abundant 1 H is excited, and its energy is transferred to the observed 31 P by using a CP on both channels ( Supplementary Fig. 21). The 31 P signal intensity can thus be enhanced by exploiting the polarization of the nearby proton nuclei. Since this process involves transfer from 1 H to 31 P in the solid state, the number and distance of proton nearby could significantly vary 31 P signal intensity. However, as the surface probe molecule for solid metal oxide, the number and distance of proton around 31 P (TMP) vary with its interactions with different surface features. As shown in Supplementary Fig. 3, the TMP molecule can bind to metal cation (a), isolated (b) and bridging (c) hydroxyl proton. Both the bottom cases (especially for the bottom right case with chemical bond formation between 1 H and 31 P) can give stronger 31 P intensity as an additional proton in a close proximity (cf. upper case). As CP could lead to variation in signal intensities with multiple 31 P environments, we thus adopted HPDEC rather than CP in this paper for the quantification of various surface features.
As we know, the T 1 for adsorbed TMP should be shorter than pure TMP as a result of the additional interactions between adsorbed TMP and solid adsorbent. Under the same acquisition parameters, if a delay time is sufficiently long enough for pure TMP sample, it will be enough for bound TMP on adsorbents and can be employed for the 31 P MAS NMR experiments in this paper. To shorten the delay time and obtain a better signal-to-noise ratio in a given time, we have used 30 o pulse with a pulse width of 1.2 s in the 31 P MAS NMR experiments for both the pure TMP and also the adsorbed TMP in this paper. First, we introduced a fixed quantity of TMP into a home-made glass tube, which fitted into a 4 mm Bruker zirconia rotor, with the help of liquid nitrogen in a vacuum line. Then, we chose 12, 15 and 20 s as the delay times while keeping other parameters unchanged. 31 P MAS NMR spectra were recorded accordingly and can be seen in Supplementary Fig. 22. The parameters shown in the right side of the picture were the acquisition and processing parameters for those spectra when the delay time of 20 s was chosen as an example. We defined the peak area in 31 P MAS NMR spectrum obtained at a delay time of 12 s as 100, the peak area in the other two spectra obtained at a delay time of 15 s and 20 s, was found to be 99 and 100, respectively. So, a delay time of 15 s was sufficiently long enough for pure TMP in the present acquisition conditions, and was therefore chosen for the 31 P MAS NMR experiments in this paper.

Supplementary Note 8: Computational details.
In DFT calculations, we employed projector-augmented waves (PAW) 22 slab systems shown in Supplementary Fig. 23. Initially, the primitive unit cell of TiO 2 was constructed to consist of tetragonal anatase TiO 2 structure containing eight O atoms with four Ti atoms; the system was then allowed to reach its lowest energy configuration by a relaxation procedure. The k-point grid determined by the Monkhorst-Pack method was 7 × 7 × 3 for bulk calculations in this study. The calculated lattice parameters of TiO 2 were 3.776 × 3.776 × 9.486 Å, which was in good agreement with the experimental value (3.785 × 3.785 × 9.514 Å) 25 .
For the modeling of a-TiO 2 (001), we adopted a slab containing six Ti-O units. The surface was constructed as a slab within the three dimensional periodic boundary conditions. This model was separated from their images in the z direction perpendicular to the surface by a 14 Å vacuum layer (the x and y directions being parallel to the surface). The bottom three layers were kept fixed to the bulk coordinates; full atomic relaxations were allowed for the top six layers. For these calculations, a 3 × 3 × 1 k-Point mesh was used in the 4 × 4 super cell. A suitable dimension of supercell (11.328 × 11.328 × 26.255 Å 3 ) was found to perform the adsorption of trimethylphosphine (TMP) on a-TiO 2 (001). The atoms in the cell were allowed to relax until the forces on unconstrained atoms were less than 0.02 eV/Å. The adsorption energy in TMP-a-TiO 2 (001) system, E ad , is defined as the sum of interactions