Supplementary Materials for

Transcription

1 Supplementary Materials for The critical role of water at the gold-titania interface in catalytic CO oxidation Johnny Saavedra, Hieu A. Doan, Christopher J. Pursell, Lars C. Grabow,* Bert D. Chandler* *Corresponding author. (B.D.C.); (L.C.G.) This PDF file includes: Materials and Methods Supplementary Text Figs. S1 to S12 Scheme S1 Tables S1 to S8 Caption for movie S1 References Published 4 September 2014 on Science Express DOI: /science Other Supplementary Material for this manuscript includes the following: (available at Movie S1

2 Table of Contents: Supplementary Materials 1. List of Figures and Tables 2. Materials and Methods 2.1. Materials 2.2. Catalyst pretreatments 2.3. Catalysis experiments a. H 2 O reaction order b. O 2 kinetic dependence c. CO reaction order 2.4. Water adsorption equilibrium 2.5. Computational methods 2.6. Justification for Computational Model p. 4 p Kinetic Isotope Effect (KIE) 3.1. Isotope interchange p CO oxidation catalysis experiments with OH/H 2 O ads and OD/D 2 O ads exchanged samples 4. Catalyst water content and water adsorption equilibrium 4.1. Water content on the catalyst 4.2. Water adsorption equilibrium p Kinetics studies 5.1. H 2 O reaction order 5.2. Michaelis-Menten kinetic studies O 2 reaction dependence and reactivity 5.3. CO reaction order p. 20 2

3 6. Description of computationally investigated pathways 6.1. Water adsorption 6.2. O 2 binding a. Without H 2 O b. With H 2 O 6.3. O 2 activation 6.4. C-O bond formation 6.5. Migration of Ti-OH to Au 6.6. Carboxylate decomposition 6.7. Kinetic isotope effect calculation p Movie S1 p Example computational input files p References p. 39 3

4 1. List of Figures and Tables Figure S1. TEM data for Au/TiO 2. TEM micrograph (A) and Particle size distribution (B) of the catalysts used in this study. Mean calculated particle size is 2.9 ± 0.9 nm. p. 6 Table S1. Catalyst Pretreatments. p. 7 Figure S2. Ab-initio surface phase stability of TiO 2 as a function of P(H 2 0). The surface phases include: black stoichiometric s-tio 2 (110) with Ti cus and O br sites; blue fully hydroxylated h-tio 2 (110) with Ti cus -OH and HO br sites; green single water adsorption on h TiO 2 (110); red h TiO 2 surface with one Ti cus -OH and HO br pair at the Au/TiO 2 interface removed; orange h TiO 2 (110) with water adsorbed at the Au/TiO 2 interface. The top left panel spans pressures between 10-7 to 1 bar. The top right panel shows an enlarged view of the h-tio 2 (110) systems with weakly adsorbed water in the pressure range between 0.1 and 1000 Pa. Note, the partially reduced r-tio 2 (110) surface is too unstable to be shown in the left panel. Figure S3. FTIR spectra during isotopic interchange of Au/TiO 2 wafer at 20 C under flowing D 2 O / N 2 (100 ml/min, 700 Pa of D 2 O). Figure S4. Changes in FTIR spectra showing D 2 O treatment of an Au/TiO 2 (30 mg) wafer at 20 C. The top (blue) spectrum was collected after H 2 O saturation; the bottom (red) spectrum was collected after D 2 O interchange. Pretreatment details in Table S1. Figure S5. Kinetic isotope effect (KIE) experiment using a dried gas CO feed. Six separate experiments were performed for each plot; standard deviation error bars are shown. Figure S6. FTIR spectral features of a Au/TiO 2 wafer (30 mg) using the mild drying treatments shown in Table S1 (T<70 C, 50 ml/min N 2 ). Figure S7. TGA Results. (A) weight change during a TGA experiment applied to a 4 mg sample of the Au/TiO 2 catalyst (100 ml/min of N 2 ) and (B) heating ramp used. Figure S8. FTIR features of the catalyst after drying at 120 C (using procedure shown in Table S1g). Figure S9. FTIR integrated absorbance of δhoh 1623 bending water characteristic band (1623 cm -1 ) as a function of the treatment temperature applied to an Au/TiO 2 catalyst wafer (A), and as a function of the strongly adsorbed water on the catalyst (wt %) calculated using TGA. p. 10 p. 12 p. 13 p. 14 p. 15 p. 16 p. 17 p. 17 4

5 Figure S10. Changes in FTIR absorbance of water characteristic bands (A2, A3) when a wafer of Au/TiO 2 is stabilized with N 2 containing water vapor (at 20 C). A background spectrum was collected after pellet was dried at 120 C and used as reference before integration. Scheme S1. Reaction steps used in the development of Michaelis-Menten kinetic model. A * denotes an oxygen binding site in the proximity of Au sites. Figure S11. Catalytic rate (ν) vs P O2 partial pressure plot. Reaction order values (n) are extracted from log(ν) vs log (P O2 ) plot p. 19 p. 21 p. 23 Table S2. Oxygen reaction orders calculated for several H 2 O contents in the gas feed. p. 23 Figure S12. CO catalytic rate (ν) vs CO concentration (%) plot. Reaction order values (n) are extracted from log(ν) vs log (% CO) plot Table S3. Calculated binding energy (E b ) and optimized geometry of a water molecule stabilized at the interface of the Au/TiO 2 model catalyst. p. 24 p. 25 Table S4. Calculated O 2 binding energy (E b ) and optimized geometry on the Au/TiO 2 model catalyst. Adsorption in the absence of water (a and b), and in the presence of one water molecule at the interface (c and d) were considered. Table S5. Calculated O 2 activation pathways on the Au/TiO 2 model catalyst. Watermediated pathways are given in c e, and the CO-assisted pathway is shown in e. IS, TS, and FS refer to the initial state, transition state, and final state of the reaction, respectively. Table S6. Calculated C-O bond formation pathways on the Au/TiO 2 model catalyst in the presence of H 2 O. The Au-COOH species is shown as intermediate in the CO oxidation route to CO 2. IS, TS, and FS refer to the initial state, transition state, and final state of the reaction, respectively. Table S7. Interfacial reaction of Ti cus -OH with Au on Au/TiO 2 model catalyst during CO Oxidation. IS, TS, and FS refer to the initial state, transition state, and final state of the reaction, respectively. Table S8. Calculated carboxylate decomposition pathways on the Au/TiO2 model catalyst. IS, TS, and FS refer to the initial state, transition state, and final state of the reaction, respectively. p. 26 p. 28 p. 29 p. 30 p. 31 5

6 2. Materials and Methods 2.1 Materials The catalyst used in this study was a commercial AUROlite sample purchased from STREM Chemicals (nominal 1% Au/TiO 2 ). This catalyst was pretreated by the manufacturer to ensure that particles were of appropriate size (2.9 ± 0.9 nm, Fig. S1) to be active for CO oxidation. The catalyst was crushed and stored in a dark refrigerator. Powdered Silicon Carbide (400 mesh) was purchased from Aldrich. Gases (N 2, H 2, O 2, and 5% CO/He) were 5.0 grade supplied by Praxair and used with no additional purification. Deuterium oxide (99.9%) was purchased from Cambridge Isotope Laboratories. Water was purified to a resistivity of 18.6 Ω with a Barnstead Nanopure system. A C B Figure S1. TEM B data for Au/TiO 2. TEM micrograph D (A) and Particle size distribution (B) of the catalysts used in this study. Mean calculated particle size is 2.9 ± 0.9 nm. 2.2 Catalyst pretreatments Each experiment was performed with a fresh sample of catalyst. The catalyst was active with no further treatment; however, several pretreatments using N 2 were applied to the catalyst with three primary goals: 1) equilibrate the water adsorbed on the catalyst surface with a known, controlled H 2 O vapor pressure, 2) expose the catalyst to a feed containing D 2 O vapor in order to exchange the original OH/H 2 O ads species on the catalyst surface (to form OD/D 2 O ads ), and 3) remove surface water to dry the catalyst. Table S1 summarizes the pretreatments applied to the catalyst immediately prior to reactivity and IR studies. 2.3 Catalysis experiments CO Oxidation Catalysis. The CO oxidation reactor consisted of a home-built laboratory scale single pass plug-flow micro-reactor. The reaction zone consisted of 5 mg of finely ground fresh catalyst diluted in 750 mg of silicon carbide. Gas flows were controlled with 4 electronic 6

7 low pressure mass flow controllers (Porter Instruments). The composition of the feed and reactor effluent (CO and CO 2 ) was determined using a Siemens Ultramat 23 IR gas analyzer. Table S1. Catalyst Pretreatments. Procedure N 2 Flow (ml/min) Conditions a. H 2 O saturation min (N Pa H 2 O) + 30 min (N 2 ) at 20 C b. Isotope Interchange min (N Pa D 2 O) + 30 min (N 2 ) at 20 C c. Mild drying min (N Pa H 2 O) + 30 min (N 2 ) at 20 C d. Mild drying min (N Pa H 2 O) at 20 C + 30 min (N 2 ) at 70 C e. Mild drying min (N Pa H 2 O) at 20 C + 16 hr (N 2 ) at 20 C f. Mild drying min (N Pa H 2 O) at 20 C + 16 hr (N 2 ) at 70 C g. Complete Drying (loosely bound water) hr (N 2 ) at 120 C After loading into a glass U-tube, the diluted catalyst was stabilized for 4h with variable contents of moisture in the gas (0.1 to 700 Pa). CO oxidation activity was measured in a 60 min experiment immediately following the pretreatment. The feed (1% CO, 20% O 2, balance N 2, 180 ml/min; WHSV = 2.16 x 10 3 L/h/g cat ) was held constant and the reaction temperature was maintained at 20 ºC using a water bath. Gases and their analytical mixtures may contain water in ppm levels. Moisture in the gaseous feeds used in these experiments was controlled using a coil submerged into a dry ice trap (-78.5 C). In this way, water pressure can be decreased to an approximate value of 0.1 Pa, which corresponds to the equilibrium vapor pressure at the temperature of the dry ice. Water moisture in the gas was controlled by subsequently passing the gas through a water column submerged into a dry ice / isopropanol bath set at a constant, but adjustable temperature (-78.5 to 20 C). a. H 2 O reaction order. The H 2 O reaction order measurements were performed under CO oxidation catalysis conditions shown above. Prior to CO oxidation measurements, water adsorbed on the catalyst was equilibrated with N 2 (100 ml/min) containing variable amounts of water vapor for 4 hours. The water vapor content in the N 2 (1, 5, 20, 50, 125 Pa) was changed using a saturator set at different temperatures. b. O 2 kinetic dependence. The O 2 dependence measurements used in Michaelis-Menten kinetic analyses were performed using five different O 2 contents (10, 15, 18, 20, 24 % vol ). Prior to each kinetics experiment, the catalyst was equilibrated with H 2 O/N 2 for 4 hours with variable amounts of moisture in the gas (1, 5, 20, 50, 125 Pa). c. CO reaction order. CO kinetic dependence was performed using five CO contents (0.56, 0.8, 1.0, 1.2, and 1.4% vol ) at 0 and 60 Pa of water vapor. O 2 content was held constant at 20%. 7

8 2.4 Water adsorption equilibrium The amount of water adsorbed on the catalyst was determined using a TGA-50 Shimadzu Thermogravimetric Analyzer. Finely ground catalyst (4 mg) was placed in the aluminum sample holder. Using a 5 C/min ramp, the sample was heated from room temperature to 120 C, held at 120 C for 1 hr, and heated to 250 C. The sample weight was recorded continuously during the heating treatment. IR spectroscopy was used to monitor changes of the catalyst surface using the same thermal treatment used in the TGA experiments (5 C/min C, hold 1 hr, 5 C/min to 250 C). Finely ground catalyst (30 mg) was pressed into a 13 mm circular pellet using a stainless steel manual hydraulic press (5 tons of pressure for 1 min). The pellet was mounted into a heated flow cell ( C) and placed in the sample compartment of a Nicolet FTIR spectrometer where a thermocouple adjacent to the pellet monitored the temperature. 2.5 Computational methods Density functional theory (DFT) calculations using the DFT+U approach by Dudarev et al. (1) were performed using the Vienna ab-initio simulation package (VASP) (2, 3). Following the scheme proposed by Hu and Metiu (4), the effective U value (U eff = U J = 2 ev) that describes the on-site Coulomb interaction between Ti 3d orbitals was chosen such that it fits the reaction energy for the oxidation of Ti 2 O 3 to TiO 2. The core and valance electrons were represented by the projector augmented wave (PAW) method (5, 6) with a kinetic energy cut-off of 400 ev. Exchange and correlation were described by the Perdew-Wang (GGA-PW91) functional (7). A Gaussian smearing with a Fermi temperature of k b T = 0.1 ev was employed and the total energies were subsequently extrapolated to k b T = 0.0 ev (8). For geometry optimizations, forces were converged below 0.05 ev/å. The TiO 2 support was modeled as a four-layer rutile (110) slab with a (5x2) unit cell, where the top two layers were relaxed and the bottom two layers were fixed in their bulk position. The calculated lattice constants for TiO 2 are a = Å and c/a = A vacuum spacing of 20 Å along the normal direction (z) to the surface was used. The Au/TiO 2 interface was created in a two-step process in which we first determined the most stable position of a 10-atom nanocluster, initially connected to the TiO 2 support via its largest (111) facet. Numerous positions of the Au cluster with respect to the TiO 2 (110) support were tested to identify the most stable structure. We note that during the geometry optimizations significant reconstruction of the Au cluster occurred and the original crystalline structure of the Au cluster was not preserved during the relaxation. After placing the Au cluster onto the stoichiometric TiO 2 (110) surface, all accessible bridging oxygen atoms (O br ) were terminated with H and the Ti cus (coordinatively unsaturated sites) positions were populated with hydroxyl (OH) groups and the system was relaxed again. The Au cluster remains flexible and adjusts its shape quickly in the presence of hydroxyl groups. The effect of surface water at the Au/TiO 2 interface was tested in a zeroth-order approximation by placing one (or two) H 2 O molecule(s) at the interface. A dipole correction (9) to the electrostatic potential was included and the Brillouin zone was sampled using a 2 x 2 x 1 Monkhorst-Pack k-point mesh (10). All transition states were found using the climbing image NEB method (11) and confirmed via vibrational analysis. An example 8

9 input INCAR file and the POSCAR file with the atomic positions of the supported Au 10 /TiO 2 (110) system for use with VASP is given in Section S Justification for Computational Model Our active site model was created by combining model elements and features that have previously been used, but never concurrently. While the model is one of the most comprehensive models in the literature, it clearly remains a simplified approximation of the real catalyst system. However, it serves the purpose of supporting our interpretation of the experimental data and proposing a plausible reaction mechanism. The 10-atom Au cluster has a diameter of ca. 0.7 nm and is smaller than the average nanoparticle size of our Au/TiO 2 catalyst samples. Yet, supported atom clusters are a reasonable computational approximation that balances accuracy and computational cost (12-14). In some cases, constrained clusters have been used to study the effect of the coordination number, but surface relaxation effects are then neglected (15, 16). Larger Au clusters have been approximated by using an infinite rod model (17, 18). While the infinite rod remains in a well-defined crystalline shape upon the adsorption of reactants, the lattice constant mismatch between TiO 2 and Au causes a small amount of artificial strain. At the expense of being smaller than experimentally observed Au clusters, the supported Au cluster in our model is free of any artificial strain and surface relaxation is possible. The role of hydroxyl groups on the TiO 2 support has been the subject of extensive scientific debate (19, 20). Fairly similar to our model, Ganesh et al. used a 5-atom Au cluster on TiO 2 (110) and varied the number of hydroxyl groups at the interface between 0 and 5 (21). The authors reported that interfacial hydroxyl groups stabilize the Au cluster and lower the reaction barrier for the reaction Au-CO + O 2 -TiO 2 at the interface. Instead of varying the number of hydroxyl groups, we used ab-initio constrained thermodynamics and calculated the surface free energy for a stoichiometric TiO 2 (110) surface (s-tio 2 ), a partially hydroxylated surface containing only HO br groups (r-tio 2 ), and a fully hydroxylated surface with Ti-OH and HO br groups (h-tio 2 ). The surface energy was estimated as: where represents the number of atoms of specie i in the surface model, is the total energy of the surface and is the bulk energy per TiO 2 unit. The chemical potentials and are obtained from the known O 2 and H 2 O pressures, under the assumption that the reaction ½ O 2 + H 2 H 2 O is equilibrated. With P(O 2 ) = 0.2 bar even at the lowest partial pressure of H 2 O (P(H 2 O) = 0.1 Pa) the fully hydroxylated h-tio 2 surface is significantly more stable than s-tio 2 or r-tio 2 (Fig. S2, left panel). This agrees with the exothermic dissociative adsorption of water on the stoichiometric 9

10 TiO 2 (110) surface: Ti cus + O br + H 2 O Ti cus -OH + HO br ( E = ev/h 2 O). Furthermore, adding water molecules to the Au/TiO 2 interface (orange line in Fig. S2) lowers the free surface energy even more, which corresponds to the computational model we have chosen for our study. The most sophisticated approach to address the possible surface termination of TiO 2 in the presence of H 2 O is first-principles molecular dynamics, which suggests that water can dissociatively adsorb and form the bridging HO br and terminal Ti cus -OH groups we have included in our model (22). The same group further found that dissociated and molecular water can coexist, and that multilayer water adsorption is feasible. The formation of the second water layer may even commence before the first layer is complete (23, 24). It was also noted in this study that proton transfer between water molecules occurs frequently during the MD simulations. Figure S2. Ab-initio surface phase stability of TiO 2 as a function of P(H 2 O). The surface phases include: black stoichiometric s-tio 2 (110) with Ti cus and O br sites; blue fully hydroxylated h- TiO 2 (110) with Ti cus -OH and HO br sites; green single water adsorption on h TiO 2 (110); red h TiO 2 surface with one Ti cus -OH and HO br pair at the Au/TiO 2 interface removed; orange h TiO 2 (110) with water adsorbed at the Au/TiO 2 interface. The top left panel spans pressures between 10-7 to 1 bar. The top right panel shows an enlarged view of the h-tio 2 (110) systems with weakly adsorbed water in the pressure range between 0.1 and 1000 Pa. Note, the partially reduced r-tio 2 (110) surface is too unstable to be shown in the left panel. The formation of multiple water layers on TiO 2 is consistent with our measured surface IR data and the calculated exothermic binding energy of ev for H 2 O on h-tio 2 (110) (green line in Fig. S2). Yet, we decided to approximate the presence of surface water by adding only one (or two) individual water molecules instead. This is the same approach Ganesh et al. used to study the effect of hydroxyls (21) and is motivated by two reasons: (i) the computational cost 10

11 increases quickly when multiple water layers are included, and (ii) the structure of the water layer is highly speculative and not well defined. It is obvious that a discrepancy between the actual water concentration on the surface and the model approximation exists, but it turns out to be advantageous in this case. Under reaction conditions, particularly in our case where the feed water content is carefully controlled, the adsorbed water layer can be approximated as being in thermodynamic equilibrium with the gas phase and the chemical potential of gas phase water and adsorbed (dissociated) water must be identical. The hydroxyl groups at the Au/TiO 2 interface of our model are significantly less stable than hydroxyl groups away from the interface, which results in reversible, non-activated ads- /desorption with a desorption energy of only 0.14 ev. Using constrained ab-initio thermodynamics and equating the chemical potentials for the reaction Au/Ti cus -OH + Au/HO br Au/Ti cus + Au/O br + H 2 O gas, where Au indicates the vicinity to the Au cluster, we find that this reaction is in chemical equilibrium for a partial water pressure of P(H 2 O) = 394 Pa (intersection of the orange and blue line in Fig. S2 right panel). This pressure is within the range of water pressures used in our experiments (0.1 Pa < P(H 2 O) < 700 Pa). Hence, the OH/H species at the Au/TiO 2 interface can be described as being quasi-equilibrated with the remaining water in the system and represent the weakly adsorbed water necessary to enhance the reaction rate. Furthermore, protons in the presence of water are highly mobile on oxide surfaces (25, 26) and we may safely assume that all available protons are quasi-equilibrated. Hence, the proton source does not need to be further specified and is irrelevant for our conclusions. Due to the limited model size, the protons in the DFT calculation originate from the HO br group, but in reality, the protons would be readily available within the weakly adsorbed surface water layers. 11

12 N 2 + D 2 O Time 3. Kinetic Isotope Effect (KIE) Kinetic isotope effect experiments were performed in order to better understand how H 2 O participates in CO oxidation over Au/TiO 2 catalysts. Surface water and hydroxyl groups were first exchanged with D 2 O by contacting the fresh catalyst with a D 2 O/N 2 feed at room temperature (20 C). Differences in the measured CO oxidation activity before and after interchange were then measured to evaluate the magnitude of the isotope effect. Absorbance 3.1 Isotope exchange H/D exchange was accomplished by passing an N 2 feed (100 ml/min) through a water saturator thermostatted at 5 C to produce a 700 Pa D 2 O/N 2 feed. This feed was then contacted with a catalyst pellet (30 mg) and placed in an in-situ IR spectroscopy cell. The exchange process was followed with IR spectroscopy to monitor the degree of exchange (Fig. S3). OH Stretching δhoh OD Stretching δdod 0.5 initial Isotopic Interchange at 20 C 1 min δhoh δdod N min Wavenumbers (cm -1 ) Figure S3. FTIR spectra during isotopic interchange of Au/TiO 2 wafer at 20 C under flowing D 2 O / N 2 (100 ml/min, 700 Pa of D 2 O). The top (blue) spectrum in Fig. S3 shows a fresh Au/TiO 2 catalyst. The δhoh bending vibrations centered at 1623 cm -1 are exclusively due to adsorbed water. The broad high energy band ( cm -1 ) is due to hydrogen-bonded OH stretching arising from interactions between adsorbed water molecules and possibly surface hydroxyl groups. The small shoulder on this band (~3670 cm -1 ) is due to non hydrogen-bonded surface Ti OH groups. Small peaks in 12

13 Absorbance the cm -1 region are attributed to small amounts of surface carbonates, which are common for these materials (27). Upon contact with the 700 Pa D 2 O/ N 2 feed, the bands associated with adsorbed H 2 O and surface Ti-OH attached to the support quickly disappear and are replaced by the corresponding OD stretching ( cm -1 ), δdod bending (1200 cm -1 ), and δhod bending (1430 cm -1 ) bands (purple spectra)(28). This appears to be an extremely facile process, as the only observed delay is the time it takes for the wet gas to travel through the gas lines. This exchange is largely complete 10 minutes after the initial contact with D 2 O. After 30 minutes of interchange, all the adsorbed H 2 O was removed from the sample (monitored by disappearance of δhoh band at 1623 cm -1 ). A small amount of HOD likely accounts for the remaining broad O-H stretching (symmetrical band centered at 3400 cm -1 ) observed as δhod bending is also observed (28). The D 2 O was removed from the gas feed and the sample was flushed for additional 30 minutes with N 2 (green spectra in Fig. S3) until no further changes occurred to the spectrum. After this N 2 flush, no δhoh modes are observed and only a trace of O-H stretching modes are observed, indicating that, at least to the detection limit of infrared spectroscopy, all of the adsorbed H 2 O had been removed. 0.2 H Au/TiO 2 (20 C) δhoh δdod D Wavenumbers (cm -1 ) 1500 Figure S4. Changes in FTIR spectra showing D 2 O treatment of an Au/TiO 2 (30 mg) wafer at 20 C. The top (blue) spectrum was collected after H 2 O saturation; the bottom (red) spectrum was collected after D 2 O interchange. Pretreatment details in Table S1. The net interchange process is summarized in Fig. S4, which shows only two spectra: 1) a spectrum of the sample after it had been saturated with 700 Pa of H 2 O for 30 min and then flushed with N 2 for 30 min to remove the vapor phase water (top, blue), and 2) a spectrum of the same pellet after interchange with 700 Pa of D 2 O for 30 min followed by flushing with N 2 for 30 min to remove the vapor phase D 2 O (bottom, red). The red spectrum shows that these conditions allow for essentially complete exchange of all OH(D) groups on the surface (at least to the detection limit of IR spectroscopy) as there are no observable OH stretches remaining. Additionally, the area under the OH spectrum and the area under the OD spectrum are within 10% of each other, further indicating complete H/D exchange. 13

14 3.2 CO oxidation catalysis experiments with OH/H 2 O ads and OD/D 2 O ads exchanged samples Kinetic Isotope Effects (KIE) were examined with fully (in-situ) exchanged catalysts having the spectral features shown in Figs. S3 and S4. Six separate experiments were performed with H 2 O saturated samples (Table S1a); reaction kinetic results are shown in Fig. 1B in the main text ( H). Another six separate fresh samples were interchanged with D 2 O (Table S1b) before CO oxidation catalysis; the results are shown in Fig. 1B ( D). The resulting calculated KIE value is 1.8±0.1. A similar set of experiments was performed by rigorously drying the gas mixture with a dry ice / isopropanol trap (Fig. S5). The rates with the rigorously dried feed are roughly 30% lower, but the calculated KIE value is similar (1.8 ± 0.3). The considerable deactivation and larger measurement errors compared to the KIE experiment performed with the untreated (wet) gas is due to changing amounts of H 2 O/D 2 O on the catalyst during the experiment as the surface is dried by the feed gas. 0.3 Au/TiO 2 (k H/ k D ) 60 =1.8 ± 0.3 Catalytic rate ν (mol CO /mol Au /s) H D Pa H 2 O in gas TOS (min) Figure S5. Kinetic isotope effect (KIE) experiment using a dried gas CO feed. Six separate experiments were performed for each plot; standard deviation error bars are shown. 14

15 Absorbance 4. Catalyst water content and water adsorption equilibrium The amount of water adsorbed on the catalyst surface is an important issue that will be correlated with catalysis results and FTIR measurements in the following sections. Fig. 3 in the main text correlates the catalytic activity with the amount of water adsorbed on the catalyst. The water content can be expressed as weight percentage of the total mass of the catalyst, or molecules per square nanometer (nm 2 ), using the catalyst surface area. Water adsorbed on the catalyst was studied using two primary tools: 1) measuring changes in sample weight during TGA experiments and 2) monitoring changes in the characteristic FTIR water bands (OH-stretching at cm -1 and δhoh bending at 1623 cm -1 ) using a sample pellet. 4.1 Water content on the catalyst FTIR spectra of a catalyst pellet (Fig. S6) show that the dominant surface species are Ti-OH groups and adsorbed water on the TiO 2. The non hydrogen-bonded hydroxyl groups bound to the TiO 2 surface occur at ~3670 cm -1 consistent with previous studies (29, 30). There are multiple peaks in this region due to OH groups in a variety of coordination geometries. This includes both anatase and rutile TiO 2 in the catalyst, along with terminal and bridging hydroxyls (31, 32). Additionally, the peak positions for these OH groups are known to shift with the addition of adsorbates that participate in hydrogen bonding (adsorbed water in this case) (33-35). However, while these peaks shift with the drying treatments applied to the catalyst (Fig. S6), their overall area does not change appreciably and there is no correlation between the peak shifts and catalytic activity. -OH OH Stretching δhoh No treatment 20 C, 30 min 70 C, 30 min 20 C, 16 hr 70 C, 16 hr Wavenumbers (cm -1 ) Figure S6. FTIR spectral features of a Au/TiO 2 wafer (30 mg) using the mild drying treatments shown in Table S1 (T < 70 C, 50 ml/min N 2 ). 15

16 The water adsorbed on the TiO 2 gives rise to the following infrared bands: hydrogenbonding stretching vibrations (ν 1 and ν 3 ) centered at 3400 cm -1 and HOH bending vibration at 1623 cm -1. These assignments and peak positions are consistent with previous infrared spectroscopic studies of TiO 2 (36-38). Treatments applied to the catalyst induce changes in the characteristic water species present on the surface of the catalyst. To determine the amount of water adsorbed on the catalyst, TGA experiments were performed using a 4 mg sample of the Au/TiO 2 catalyst. Fig. S7A shows the weight decrease as a function of temperature ramp used (Fig. S7B) during a typical TGA experiment. The heating ramp (5 C/min) was the same ramp used in pretreating samples for catalysis and FTIR experiments. Fig S7 shows that there is an initial rapid desorption of loosely bound water in the C temperature range, followed by a regime of slower desorption associated with more strongly bound water. Approximately 2/3 of the total adsorbed water is weakly bound to the TiO 2. Weight differences allow calculating the total water content on the catalyst (1.92 wt%) as well as the weakly (1.32 %) and strongly (0.61 wt%) bound water. B A Figure S7. TGA Results. (A) weight change during a TGA experiment applied to a 4 mg sample of the Au/TiO 2 catalyst (100 ml/min of N 2 ) and (B) heating ramp used. The TGA experiments (Fig. S7) were performed with a 60 minute hold at 120 C; during this time, very little water desorbed. This indicates that the pretreatment temperature is the key feature to determining the amount of strongly adsorbed water. Fig. S8 shows the FTIR spectrum of a catalyst dried at 120 C (Table S1g); the remaining adsorbed water can be quantified with the A2 band (OH stretching of water, red peak in Fig. S8) and A3 band (H-O-H bending, green peak in Fig. S8). 16

17 IR Integrated Absorbance δhoh A1 : cm -1 A1 A2 : cm -1 A3 : cm -1 Absorbance A2 Area = (0.61 wt %) 0.1 T=120 C A Wavenumbers (cm -1 ) Figure S8. FTIR features of the catalyst after drying at 120 C (using procedure shown in Table S1g) IR Integrated Absorbance δhoh R² = A R² = Temperature of the cell (ºC) water on catalyst (wt. %) B Figure S9. FTIR integrated absorbance of δhoh 1623 bending water characteristic band (1623 cm -1 ) as a function of the treatment temperature applied to an Au/TiO 2 catalyst wafer (A), and as a function of the strongly adsorbed water on the catalyst (wt %) calculated using TGA results. Treatments applied to the catalyst induce changes in IR spectra associated with the adsorbed water and Ti-OH groups. Fig. S7 shows that the primary change to the catalyst after mild thermal treatments is loss of adsorbed water, i.e. no support dehydroxylation was observed at temperatures below 120 C. Complete water desorption and dehydroxylation can only be 17

18 achieved using severe thermal treatments (Temperature >300 C). As an example, Fig. S9A plots the decrease in the IR integrated area of the δhoh 1623 bending band as a function of the temperature applied to the catalyst. Concomitant with the decrease in the integrated δhoh 1623 bending band at high temperatures ( Cº), the loss of strongly adsorbed water can be correlated to the decrease in weight of the sample during the TGA experiment, establishing a clear linear correlation between the adsorbed water (% wt) with the intensity of the IR δhoh 1623 band. This correlation is shown in Fig. S9B. A similar linear correlation between integrated IR absorbance and adsorbed water on TiO2 films has beed previously reported (36). The linear correlation between IR integrated area (Fig S9B) and the water content on the catalyst after drying at 120ºC, A2 and A3 peak areas can be correlated with the catalyst water content (wt%). In this case, the FTIR integrated areas (317.9 (A2) and 20.3(A3) integrated units) correspond to a water content of wt%. 4.2 Water adsorption equilibrium After drying (120 C, Table S1g), weakly adsorbed water can be restored to the catalyst surface by contact with a N 2 flow containing controlled amounts of moisture. This adsorption process was studied using FTIR spectroscopy after the dried pellet at 20 C was collected as the background. As Fig. S10 shows, the water adsorbed on the catalyst increases with the gas phase water content, following a typical adsorption isotherm. The A2 and A3 IR band areas (as shown in Fig. S10) can therefore be used to monitor the amount of water adsorbed on the catalyst surface and correlated with the TGA experiments above. Notably, the A2 peak assigned to hydrogen bound O-H stretching vibrations ( cm -1 ) tracks with the A3 δhoh bending vibration of adsorbed water (centered at 1623 cm -1 ). Since the A3 peak can only come from water (not hydroxyls), the A2 peak also correlates with changes in the amount of water on the support. The concentration of surface hydroxyls is therefore described only by the A1 band. Further, this adsorption isotherm can be readily fit to a logarithmic equation, allowing us to determine the peak area at any water pressure: 18

19 1000 IR area = ln(p H2O ) R 2 =0.98 IR Integrated Absorbance A cm -1 A3* cm Pa 5 Pa Water vapor pressure in the gas (Pa) Figure S10. Changes in FTIR absorbance of water characteristic bands (A2, A3) when a wafer of Au/TiO 2 is stabilized with N 2 containing water vapor (at 20 C). A background spectrum was collected after pellet was dried at 120 C and used as reference before integration. IR ln( ) 113.7, R 2 =0.98, P 2 H2O in Pascal absorbance P H O The IR area -wt% relation determined with Fig. S9 and the TGA data can be combined with Eq. 1 (IR area -P H2O relationship, Fig. S10) to estimate the amount of water adsorbed on the catalyst at any water pressure in the range studied. This wt% water adsorbed information is used in Fig. 3 (B and D) in the main text. The amount of weakly adsorbed water can also be expressed in terms of a surface concentration by normalizing to the surface area of the catalyst (45 m 2 /g, Fig. 3 B and 3D, secondary x-axis) 19

20 5. Kinetics Studies Kinetic studies were performed in order to examine the influence of various species on the reactivity of the Au/TiO 2 catalyst. Kinetics measurements were studied by varying the composition of the reactive gas at a constant flow. Conditions were maintained to be similar to a basic reforming mixture composition, nominally 1% CO, 20% O 2, balanced with N H 2 O reaction order Prior to catalytic experiments, the weakly bound water was removed using the drying procedure in Table S1g. The surface was then equilibrated with water by passing a H 2 O (or D 2 O) saturated N 2 flow (100 ml/min) over the catalyst for 4 hours. The kinetics for each water vapor pressure were measured with a separate set of experiments on a fresh catalyst maintaining constant O 2 and CO concentrations (20 and 1%, respectively). Fig. 3B shows water dependence in CO oxidation activity of the Au/TiO 2 catalyst with respect to the water content used in the experiment. The dots in Fig. 3B ( ) represents the CO oxidation activity (ν) as a function of the water present in the gas (P H2O ). The triangles ( ) plots the CO oxidation activity when water content is expressed as the equilibrium weight% of water adsorbed on the catalyst. The reaction order is then calculated with the slopes of the trend lines in Fig. 3B. 5.2 Michaelis-Menten kinetic studies O 2 reaction dependence and reactivity A key question in evaluating or comparing catalysts is the underlying causes of activity differences. Two primary factors in determining activity are the total number of active sites on a catalyst and the intrinsic activity of the active sites; distinguishing between these two factors allows one to better understand changes in reactivity. This can be accomplished with a Michaelis-Menten (M-M) treatment that we have previously developed and employed (39-41). In this treatment, activity data as a function of oxygen pressure can be evaluated with double reciprocal plots (Fig. 3B) to provide a means of extracting quantitative parameters that describe O 2 reactivity on Au catalysts. A full derivation of this kinetic treatment has been previously reported (39). Briefly, a simple characterization mechanism (Scheme 1), is used to describe the reaction. We note that this is an intentionally simplified characterization mechanism that may not capture all of the molecular complexity of every elementary step in the reaction mechanism. The characterization mechanism is intentionally vague regarding the nature of the O 2 binding site the purpose is to evaluate changes to the number of active sites rather than to presuppose the structural characteristics of the O 2 binding site. 20

21 Au + CO fast Au-CO k 1 A* + O 2 A*-O 2 k -1 k 2 A*-O 2 + Au-CO A*-O + Au + CO 2 fast A*-O + Au-CO A* + Au + CO 2 Scheme S1. Reaction steps used in the development of the Michaelis-Menten kinetic model. A * denotes an oxygen binding site in the proximity of Au sites. This intentional vagueness is particularly important for CO oxidation over Au catalysts, where there has been such substantial debate regarding O 2 binding and activation. While many studies indicate that O 2 binding and activation occurs on the Au nanoparticles, many others suggest that oxygen vacancies on the titania support are the key O 2 activation sites. These two suggested active site models are considerably different in their structural requirements, so we did not want to presuppose which one might more closely resemble the true active site. Hence, the only structural requirement in this M-M treatment is that oxygen binding occurs in close proximity to Au CO sites. Further, the characterization mechanism did not include surface water, as we do not want to assume a precise role for water. In the development of this method, adsorption of CO and activation of O 2 are considered the key kinetic steps; desorption of produced CO 2 and regeneration of the active site are considered to be fast and after the rate determining steps and are therefore kinetically unobservable. Applying a typical kinetic derivation employing the steady-state approximation yields the following expression: where rxn is the measured reaction rate and We note that a similar set of equations can be derived using a Langmuir-Hinshelwood mechanism, although this requires additional assumptions (39). In these equations, CO designates the coverage of the CO binding sites, which are a subset of the total Au surface sites and [A*] T is the total number of active sites. The total number of active sites is also assumed to involve a subset of the total number of surface Au sites. K R and ν max are descriptive kinetic 21

22 parameters comparable to those employed in enzyme kinetics (42). Analogous to the Michaelis- Menten constant, K R is a measure of the reactivity or instability of adsorbed O 2 (cf. A*-O 2 ). Similarly, ν max depends both on the intrinsic reaction barrier and the number of active sites. This kinetic treatment has been previously applied to model kinetic data for CO oxidation over several Au (39) and bimetallic NiAu catalysts (43), including a series of intentionally poisoned Au catalysts (40) and a catalyst with extended carbonate buildup (27). One of the primary utilities of the M-M kinetic treatment is that it provides separate measures of the intrinsic activity of the active sites (K R ) and the total number of active sites (related to both K R and ν max ). By applying this kinetic model to the CO oxidation kinetics while varying the feed H 2 O content, this study can help determine if the amount of water on the catalyst affects the number of active sites and/or the inherent reactivity of the active sites. Fig. 3D shows that K R ( ) is relatively unaffected by the amount of water adsorbed on the catalyst, while υ max ( ) values change significantly with the amount of water adsorbed on the catalyst. This indicates that the intrinsic activity of the active sites is relatively unchanged and that the primary change is to the number of active sites. The literature shows a wide variation in kinetic measurements for O 2 dependence in the CO oxidation reaction. The most representative reference is the tabular summary published by Gates et al. (44). In this review, O 2 reaction orders in the literature vary in a broad range (0.05 to 0.4) based on studies with remarkable differences in the experimental conditions. As an example, the space velocity used in those studies vary in the range ml/min g cat, which is, roughly, >40 times slower than experimental conditions in this study (36000 ml/min g cat ). Very few studies maintain low conversions, however, so direct comparisons between studies are difficult. Other important variation in experimental conditions is the oxygen content in the gaseous streams, which varies from stoichiometric (O 2 /CO = 2/1) to excess (O 2 /CO = 20/1), more consistent with reformate streams and used in this experimental design. The O 2 dependence data, shown in Fig. S12 and Table S2, are within the range of orders reviewed in ref (44). The presence of water in the reacting mixture has little effect on O 2 reaction orders and the average value (n=0.18) is significantly lower than the reaction orders for water (n=0.33 and 1.33, based on gas phase and weakly adsorbed water, respectively). These results stress the importance of water and, most importantly, weakly adsorbed water in the reaction kinetics. These implications are discussed in detail in the main text of this paper and were used as the starting point for the design of the model Au/TiO 2 catalyst and the development of DFT calculations. 22

23 Catal. rate υ (mol CO /mol Au /s) Pa, n=0.21± Pa, n= 0.12 ± Pa, n=0.12 ± Pa, n=0.23 ± Pa H 2 O, n=0.2 ± P O2 Figure S11. CO catalytic rate (ν) vs P O2 partial pressure plot. Reaction order values (n) are extracted from log(ν) vs log (P O2 ) plot. Table S2. Oxygen reaction orders calculated for several H 2 O contents in the gas feed. Feed water content (Pa) O 2 reaction order Calculated Error in O 2 reaction order We note that there is no clear trend in the O 2 reaction orders, and that their apparent differences are largely due to an artifact of the logarithm function when the slope of the reaction order plot is close to zero. 23

24 5.3 CO reaction order The CO reaction order was studied using five CO concentrations (0.56, 0.8, 1.0, 1.2, and 1.4% Section 2.3.c), at two different feed moisture contents (0.1 and 60 Pa) maintaining constant O 2 concentration (20%). Results are shown in Fig. S13. There is little to no change in reaction rate upon changing the CO concentration, resulting in reaction order values close to zero (n=0.05 and for 60 ( ) and 0.1 Pa ( ) of H 2 O in the feed, respectively). These results are in agreement with the bulk of the literature, which shows virtually no kinetic dependence on the CO content of the feed (44). This further supports the conclusion that CO is readily available for the reaction even at low concentrations in the gas, and that the general kinetics of the reaction is not affected by the CO coverage on the Au clusters Pa H 2 O n=0.05 Catalytic rate ν (mol CO /mol Au /s) Pa H 2 O n= ml/min 20% O 2 in gas CO in gas (%) Figure S12. CO catalytic rate (ν) vs CO concentration (%) plot. Reaction order values (n) are extracted from log(ν) vs log (% CO) plot. 24

25 6. Description of computationally investigated pathways In order to develop a better molecular understanding of the experimental observations and probe potential mechanisms in which water might be involved in the reaction mechanism, we undertook a computational study using density functional theory (DFT) simulations (computational details are included in Section 2.5 of this SM). Our Au/TiO 2 model consisted of a 10-atom gold nano-cluster (Au 10 ) residing on a (5x2) unit cell with 4 layers of the TiO 2 (110) support. The hydroxylated environment was modeled by populating the bridging oxygen (O br ) and coordinatively unsaturated titanium sites (Ti cus ) of the TiO 2 (110) surface with H and OH, respectively. The effect of small amounts of surface water was approximated by adding one (or two) water molecule(s) at the Au/TiO 2 interface. 6.2 Water adsorption Table S3 shows the interaction of a water molecule at the metal-support interface of the Au/TiO 2 model catalyst. We calculated that water adsorbs to an Au atom at the interface with a binding energy of E b = ev (Table S3a). This is significantly stronger than H 2 O adsorption on an unsupported Au 12 cluster (E b = ev) (45) or on the fully hydroxylated h TiO 2 (110) surface (E b =-0.36 ev, Table S3c). Yet, water adsorption on a HO br group near the Au/TiO 2 interface is even more exothermic (E b = ev, Table S3b). The strong binding energy at the Au/TiO 2 interface is a result of the formation of a stable surface hydronium ion (H 3 O + ), which effectively anchors and stabilizes the water molecule. Location E b (ev) a. On Au 10 cluster b. At Au/TiO 2 interface on HO br site c. On HO br site of h TiO 2 (110) support Table S3. Calculated binding energy (E b ) and optimized geometry of a water molecule stabilized near and far from the interface of the Au/TiO 2 model catalyst. 25

26 6.2 O 2 binding a. Without H 2 O. O 2 binding studies and the different interactions with the Au/TiO 2 model are summarized in Table S4. O 2 adsorption on top of the Au 10 cluster (Table S4a), and at the Au/TiO 2 interface (Table S4b) were examined in the absence of H 2 O. For the former, the calculated binding energy E b = ev is much stronger than previously obtained on an unsupported Au 12 cluster (E b = ev) (15, 45). We speculate that the large binding energy is an artifact of the highly under-coordinated Au atoms on top of the cluster. Based on the very large barrier for O 2 dissociation from this initial state (E a = 2.32 ev, Table S5a) and the stronger O 2 binding at the water-free Au/TiO 2 interface (Table S4b), we did not consider this adsorption geometry further. O 2 adsorption at the Au/TiO 2 interface leads to the spontaneous (non-activated) formation of H 2 O 2 adsorbed on the Au cluster (Table S4b). We attribute the strongly exothermic nature of this adsorption step (E b = ev) to the transfer of two protons from surface OH groups. The energy required for breaking the HO-OH bond at the metal-support interface (Table S5b) is 1.22 ev lower than O-O bond dissociation at the top of the cluster. The strong binding of O 2 as a peroxide and the low activation barrier point to the key role that proton transfer plays in catalyzing O 2 binding and dissociation. Location E b (ev) a. On Au 10 cluster (without H 2 O) b. At Au/TiO 2 interface (without H 2 O) c. On Au 10 cluster (with H 2 O) d. At Au/TiO 2 interface (with H 2 O) Table S4. Calculated O 2 binding energy (E b ) and optimized geometry on the Au/TiO 2 model catalyst. Adsorption in the absence of water (a and b), and in the presence of one water molecule at the interface (c and d) were considered. 26

27 b. With H 2 O. In the presence of the surface hydronium generated by water adsorption, we found that O 2 adsorptions on the metal cluster and at the metal-support interface were equally favorable with a binding energy of approximately -1.4 ev (Table S4c,d). In both cases, the adsorbed O 2 molecule was stabilized by accepting a proton from either the surface hydronium or the surface hydroxyl. In particular, the formation of OOH on the metal cluster involves O 2 adsorption followed by a non-activated (barrierless) proton transfer from the hydronium molecule. The exact location of the adsorption site strictly at the interface or elsewhere on the cluster appears to be far less important than the access to protons. This also suggests that Au sites away from the Au/TiO 2 interface can be utilized for O 2 adsorption, provided that they have access to protons from co-adsorbed water. This general conclusion is in good agreement with the Michaelis-Menten analysis (Section 5.2 in this SM) showing that the number of active sites increases in the presence of surface water. 6.3 O 2 activation Based on the O 2 adsorption results, several O 2 activation pathways were explored; the results are summarized in Table S5. Oxygen activation can occur in the absence of water either on the Au cluster (Table S5a) or at the Au/TiO 2 interface (Table S5b). While the adsorption of O 2 on top of the Au cluster (Table S4a) is strong (E b = ev), its dissociation is highly activated (E a = 2.32 ev) and this O 2 activation pathway may be excluded. This is in good agreement with the large reported O 2 dissociation barrier (E a = 1.62 ev) on unsupported Au 12 clusters (45). The adsorbed H 2 O 2 at the Au/TiO 2 interface has a moderate HO-OH dissociation barrier (E a = 1.10 ev, Table S5b). This suggests that, in the absence of water and CO, it may be possible to activate O 2 at the metal-support interface, leading to Au-OH species. The presence of H 2 O lowers the activation barriers for O-OH dissociation at the Au/TiO 2 interface (E a = 0.71 ev, Table S5c) and on Au (E a = 0.49 ev, Table S5d). In both pathways the final state is stabilized by a hydrogen-bonding network involving the co-adsorbed H 2 O molecule and support OH groups. These relatively low energy barriers suggest that the presence of surface water at the metal-support interface opens up new reaction channels that provide for more facile activation of the O 2 molecule. We note again, that the presence of H 2 O renders a larger fraction of the Au sites accessible for O 2 activation through water-assisted pathways (Table S5d). This further indicates that the primary role of the metal support interface it to provide access to protons, rather than to generate Au sites with special reactivity. The presence of CO has been shown to lower the O 2 activation barrier on Au (13, 15, 17, 46, 47) and we investigated the CO-assisted O 2 activation in the presence of H 2 O on the TiO 2 supported Au 10 cluster. Indeed, the barrier for the formation of a carboxyl intermediate (COOH) is only E a = 0.10 ev and the process is 2.23 ev exothermic (Table S5e). The O 2 activation pathway in the presence of H 2 O and CO is clearly favored over all other considered pathways (Table S5a-d), while simultaneously forming the necessary C-O bond. The extremely low activation barrier indicates a quasi-spontaneous O 2 activation step, which is a substantial departure from the predominant view that O 2 activation is rate limiting in CO oxidation catalysis over Au. 27

28 Pathway Energies IS TS FS a. On Au 10 cluster (without H 2 O) E a = 2.32 ev ΔE =-2.00 ev b. At Au/TiO 2 interface (without H 2 O) E a = 1.10 ev ΔE =-0.42 ev c. At Au/TiO 2 interface (with H 2 O) E a = 0.71 ev ΔE =-2.00 ev d. On Au 10 cluster (with H 2 O) E a = 0.49 ev ΔE =-1.75 ev e. CO-assisted (with H 2 O) E a = 0.10 ev ΔE =-2.23 ev Table S5. Calculated O 2 activation pathways on the Au/TiO 2 model catalyst. Water-mediated pathways are given in c e, and the CO-assisted pathway is shown in e. IS, TS, and FS refer to the initial state, transition state, and final state of the reaction, respectively. In sections of this supplementary material we have shown that low energy pathways for O-O bond activation require a proton transfer. The availability of protons is increased in the presence of surface water and the DFT results agree qualitatively with the strong experimental rate dependence on surface water. However, the proton transfer to O 2 was in all cases non-activated and can therefore not explain the measured primary KIE, which implicates O-H(D) bond formation/scission as a kinetically important step. In the subsequent sections we now focus only on pathways that involve co-adsorbed H 2 O and determine the kinetically relevant step, which can explain the observed primary KIE. 6.4 C-O bond formation As discussed above, CO-assisted O 2 activation in the presence of H 2 O leads to the formation of an Au-COOH intermediate (Table S6a, identical to Table S5e). This concerted process of COOH formation and O-OH dissociation yields the lowest activation energy barrier, but we also need to consider C-O bond formation with Au-O leading to CO 2 and Au-OH species 28

29 resulting in Au-COOH. Au-O and Au-OH are obtained from O 2 dissociation in the absence of CO (Table S5b-d). Au-O is also produced in the CO-assisted O 2 dissociation pathway (Table S5e). The initial state for the reaction of CO with Au-OH (Table S6b) and Au-O (Table S6c) was chosen to be the final state of the OOH dissociation step at the Au/TiO 2 interface in the presence of water (Table S5d). A CO molecule was then adsorbed to the Au cluster. The formation of the COOH intermediate through the reaction of CO with Au-OH has a moderate activation barrier of 0.40 ev and is essentially thermoneutral (Table S6b). CO 2 can be formed from the same initial state by the reaction of Au-CO with Au-O, which requires a higher barrier (E a = 0.65 ev, Table S6c). Overall, Table S6 shows that the concerted CO-assisted O 2 activation step (Table S6a) is clearly the preferred route for C-O bond formation. To close the catalytic cycle an additional C- O bond formation step according to Table S6b or c is required. Reactant for C-O bond formation a. Au-OOH (identical to Table S5e) Energies IS TS FS E a = 0.10 ev ΔE =-2.23 ev b. Au-OH E a = 0.40 ev ΔE = 0.05 ev c. Au-O E a = 0.65 ev ΔE =-1.03 ev Table S6. Calculated C-O bond formation pathways on the Au/TiO 2 model catalyst in the presence of H 2 O. The Au-COOH species is shown as intermediate in the CO oxidation route to CO 2. IS, TS, and FS refer to the initial state, transition state, and final state of the reaction, respectively. 6.5 Migration of Ti-OH to Au We examined the possibility of oxygen exchange between the TiO 2 support and the Au 10 cluster by studying two pathways for oxygen exchange, shown in Table S7. The migration of an OH group from a Ti cus site to the Au 10 cluster is thermodynamically very unfavorable (ΔE = 1.57 ev, Table S7a), because OH binds significantly weaker to gold than to titanium. We also studied the interfacial reaction between Au-CO and Ti cus -OH to form carboxyl. While the activation barrier E a = 0.72 ev is moderate, it is larger than all other C-O bond formation barriers discussed 29

30 in Section S6.4. Additionally, this pathway is thermodynamically uphill (ΔE = 0.60 ev), whereas the CO assisted Au-OOH dissociation reaction has a substantial thermodynamic driving force. This, along with the experimental evidence described above and in the main text suggests that CO oxidation on Au/TiO 2 does not involve any lattice oxygen atoms from the TiO 2 support, in agreement with isotopic labeling studies by the Davis and Iglesia groups (48, 49). Ti-OH reaction Energies IS TS FS a. OH migration to Au E a = 1.63 ev ΔE = 1.57 ev b. Ti cus -OH reaction with Au-CO E a = 0.72 ev ΔE = 0.60 ev Table S7. Interfacial reaction of Ti cus -OH with Au on Au/TiO 2 model catalyst during CO oxidation. IS, TS, and FS refer to the initial state, transition state, and final state of the reaction, respectively. 6.6 Carboxylate decomposition Carboxylate decomposition processes to CO 2 are shown in Table S8. The energy required for proton removal from carboxyl depends on whether co-adsorbed O and/or H 2 O are in close proximity. When neither is present and the proton is left on the Au cluster, the process is highly activated with E a = 1.36 ev (Table S8a). The presence of neighboring Au-O (e.g. final state in Table S6a) provides another possible proton acceptor, and Au-COOH decomposition becomes a mildly exothermic process with no calculated activation barrier (ΔE =-0.27 ev, E a = 0.00 ev, Table S8b). The resulting Au-OH subsequently requires reaction with adsorbed CO (ΔE = 0.05 ev, E a = 0.40 ev, Table S6b) forming another Au-COOH species. To close the catalytic cycle, Au-COOH must eventually decompose by proton transfer back to water (Table S8c and d). The proton transfer from Au-COOH to adsorbed water at the interface is an endothermic process with a small additional activation barrier (ΔE = 0.61 ev, E a = 0.76 ev, Table S8c). Our primary computational model included only a single co-adsorbed water molecule to examine proton transfer at the Au/TiO 2 interface. However, our IR and kinetics experiments show that the catalyst is most active when more than a monolayer of weakly bound water is adsorbed on the support. While the computational cost did not allow us to model a full surface water layer, we added a second co-adsorbed H 2 O molecule and investigated the effect on proton transfer to surface water (Table S8d). Notably, we observe that the inclusion of an additional water molecule lowered the activation barrier for Au-COOH decomposition to the thermodynamic limit, i.e. the endothermicity of the step ( E = 0.70 ev). 30

31 Proton acceptor Energies IS TS FS a. Transfer to Au (no Au-O, or H 2 O nearby) E a = 1.36 ev ΔE =-0.34 ev b. To Au-O E a = 0.00 ev ΔE =-0.27 ev c. To H 2 O E a = 0.76 ev ΔE = 0.61 ev d. To H 2 O with second H 2 O present E a = 0.70 ev ΔE = 0.70 ev Table S8. Calculated carboxylate decomposition pathways on the Au/TiO 2 model catalyst. IS, TS, and FS refer to the initial state, transition state, and final state of the reaction, respectively. Further, we (and others) observed an experimental KIE that shifts towards an equilibrium isotope effect at high surface water contents (49). The computational results are, again, in excellent agreement with the experimental observations as no kinetic barrier is predicted for Au- COOH decomposition when multiple water molecules are present near the reactive species on the Au cluster. A KIE for this step is also calculated and described in section S6.7 below. Overall, the moderate barriers for the proton transfer to H 2 O during Au-COOH decomposition (Table S8c and d) renders this step the best candidate for the rate-determining step, which is consistent with the experimental primary KIE when there is relatively little water on the surface. 6.7 Kinetic isotope effect calculation Further evidence for step c in Table S8 being the rate-limiting step during CO oxidation on Au/TiO 2 is the calculated KIE of 2.55, which is in agreement with the experimentally obtained value of 1.8. The KIE was calculated assuming that all H atoms, not only the transferred proton, are replaced with D atoms. A movie showing the pathway and the corresponding vibrational mode is available as Movie S1. For KIE calculations, the zero point energy (ZPE) and entropy contributions were estimated in the harmonic oscillator approximation, where all degrees of freedom of adsorbed species (including frustrated rotations and translations) were treated 31