DFT-GGA and DFT+U Simulations of Thin Water Layers on ReducedTiO2 AnataseAntonio Tilocca*,† and Annabella Selloni‡

†Department of Chemistry, University College London, London WC1H 0AJ, U.K.‡Department of Chemistry, Princeton University, Princeton, New Jersey 08544, United States

ABSTRACT: We investigated the effect of a reducing subsurface Ti interstitial defect onthe structure and reactivity of thin water layers adsorbed on the majority anatase(101)surface of titanium oxide using ab initio molecular dynamics simulations. We find thatstandard DFT-GGA and the DFT+U method predict similar energetics and dissociationbarrier for a single water molecule adsorbed on the reduced surface; moreover, the twoapproaches also lead to very similar structural features and reactivity for an adsorbed watermonolayer (ML) on the same surface. This allows us to model 1, 2, and 3 water layers onthe reduced surface through Car−Parrinello molecular dynamics simulations up to 20 pslong. Compared to the defect-free surface, the simulations highlight how the interstitialdefect alters the stability of surface adsorption sites, substantially enhancing the surfacereactivity and leading to a markedly different structure of the first water layers adsorbed onthe reduced surface.

■ INTRODUCTIONThe interaction of water with titania surfaces continues toattract considerable attention for both its fundamental scientificinterest and its important role in technological applications.1−5

After many years of intense experimental and theoreticalresearch, some aspects of this interaction, such as wateradsorption in the dilute limit, have been elucidated to a largeextent, but other features remain poorly understood. Amongthe latter, there are the properties of adsorbed thin water layersand how these are affected by the underlying surface structure.Knowledge of these properties is important for a betterunderstanding of the mechanisms of TiO2 photocatalysis andwould be very useful in other fields as well. For instance,immobilizing peptides, protein, and DNA strands on titaniumoxide surfaces is of key importance for biomolecularrecognition to develop biosensors and improve the perform-ances of Ti-based biomaterials.6 The interaction of peptideswith TiO2 is mediated by structured water adsorbed at thesurface: proteins, collagen, and DNA bases bind to a thin waterfilm in direct contact with the surface, rather than directly tothe surface itself.7,8 The properties of the first water layers inclose contact with the surface of titanium oxide and otherbiomaterials can thus control their biological response.9,10

Recently, there have been several computational studies ofwater on the majority (101) surface of anatase, the TiO2polymorph most relevant for photocatalytic applications. Thesestudies have shown that the stability of various adsorbates onanatase TiO2 is significantly influenced by coadsorbedwater.11−13 Ab-initio simulations have also shown that themain structural features of the first water layers in direct contactwith the surface14 are maintained at the interface with bulkliquid water.15 While these results highlight the interest ofinvestigating the properties of thin water layers, they generally

refer to the stoichiometric, defect-free surface, or the reducedsurface with a surface oxygen vacancy.17 On anatase (101),however, subsurface defects are usually present.18−20 In thispaper we thus use density functional theory (DFT) calculationsand first principles molecular dynamics (FPMD) simulations toexamine how the structure of up to three monolayers thickwater films is affected by a subsurface reducing defect.The structures of a water monolayer (ML), bilayer (BL), and

trilayer (TL) on defect-free anatase (101) have been studied byus previously.14 We also investigated the structures of a waterML and a BL on anatase (101) in the presence of a surfaceoxygen vacancy.17 Although informative, those studies weresomewhat limited by the use of a rather thin slab model and therelatively short duration of the simulations. Moreover, thesimulations on the reduced surface17 were based on standardDFT in the generalized gradient approximation (GGA), whichis now known to provide a rather unsatisfactory description ofthe electronic structure of intrinsic defects in TiO2 and othermetal oxide materials.21−23 In the present study, we overcomethese limitations by using thicker slab models and performingsignificantly longer FPMD simulations. To model the reducedsurface, we consider a subsurface Ti interstitial, a frequentintrinsic reducing defect in TiO2.

18,20,24,25 We assess theperformance of DFT-GGA by comparing the structure anddynamics of an adsorbed ML obtained from two parallelsimulations performed using DFT-GGA in one case and theDFT+U method in the other.26 This detailed comparisonshows that the ML structure and dynamics from DFT-GGA areactually quite similar to those obtained using the DFT+U

Received: February 18, 2012Revised: March 31, 2012Published: April 3, 2012

Article

pubs.acs.org/JPCC

© 2012 American Chemical Society 9114 dx.doi.org/10.1021/jp301624v | J. Phys. Chem. C 2012, 116, 9114−9121

method. On the basis of these results, we then use DFT-GGAfor extensive simulations of water ML, BL, and TL on thestoichiometric and reduced surfaces. The latter simulationsshow that the interstitial defect significantly perturbs theadsorbed water layers, affecting the adsorption strength and thereactivity of the surface sites nearest to it and altering thevertical and in-plane ordering of the water layers observed onthe defect-free surface.

■ COMPUTATIONAL PROCEDURE

All calculations were performed using density functional theory(DFT) within the plane wave-pseudopotential framework asimplemented in the Quantum-ESPRESSO (QE) package.27

Both the generalized gradient approximation (GGA) and theDFT+U method26 were employed, and in both cases the PBE28

exchange-correlation functional was adopted. The DFT-GGAcalculations were spin-restricted, whereas spin polarization wasincluded in the DFT+U calculations. For the latter, U = 3.5 eVwas used, as determined previously.16 Core−valence inter-actions were represented through ultrasoft pseudopotentials,29

with plane-wave basis set cutoffs of 25 and 200 Ry for thesmooth part of the wave functions and the augmented charge,respectively. Because of the relatively large supercell size, k-point sampling was restricted to the Γ point only. Theanatase(101) surface was represented as a periodically repeatedslab, using previously calculated bulk lattice parameters20 (a =3.77 Å and c = 9.54 Å). We used three-layers thick slab, with 1× 3 (10.26 × 11.31 Å2) surface supercells, corresponding to 36TiO2 units per simulation cell for the stoichiometric surface; forthe reduced surface, an additional Ti atom was included. Avacuum gap of 14.5 Å separated the periodic images of the slabalong the c direction.14 In all simulations, the Ti and O atoms inthe bottom layer were fixed at their bulk positions. Thiscomputational setup has been thoroughly validated in ourprevious studies of water interacting with TiO2 surfaces.

11,14,16

DFT-GGA-based Car−Parrinello molecular dynamics(CPMD)30 simulations were carried out with a fictitiouselectronic mass of 700 atomic units and a time step of 0.17 fs,using the deuterium mass for hydrogen atoms. The ionictemperature was controlled through a Nose thermostat in allsimulations; for the runs on the reduced surface, an additionalNose thermostat on the electronic degrees of freedom wasneeded to avoid drifts in the fictitious kinetic energy andpreserve adiabaticity.32 Spin-polarized DFT+U-based simula-tions were performed using Born−Oppenheimer moleculardynamics (BOMD)33 with a time step of 1 fs and a Berendsenthermostat to control the ionic temperature. Despite the longertime step, the spin-polarized DFT+U BOMD runs areobviously much more time-consuming34 than the spin-restricted DFT-GGA CPMD runs. The total electronicmagnetization was unconstrained in the DFT+U BOMDsimulations and converged to S = 4.0, which corresponds toa fully ferromagnetic arrangement of the four excess electronsassociated with the Ti interstitial. The nudged elastic band(NEB) method35 was used to determine the DFT+U energybarrier for water dissociation on the reduced surface. Ninereplicas were employed in the calculation; as in the BOMDruns, the unconstrained total magnetization converged to S =4.0 for all replicas.

■ ASSESSING THE PERFORMANCE OF DFT-GGA VSDFT+U

Single-Molecule Adsorption and Dynamics. Weconsidered two possible stable locations for the subsurface Tiinterstitial, previously labeled as T4 and T5.20 As shown inFigure 1, the Ti interstitial is closer to the surface in T4, which

is deemed to result in a stronger effect on the surface reactivity,compared to the deeper and more stable T5.31 After relaxingthe bare slabs, we modeled the adsorption of a single watermolecule on all the inequivalent 5-coordinated surface Ti atoms(Ti5c) on the T4 and T5 surfaces; these sites are labeled 1,2(3), 4(5), and 6 in Figure 1 for T4, whereas for T5 theinequivalent sites are those labeled 1(3), 2, 4, and 5(6). Allstructures were optimized at the DFT+U level and theiradsorption energies compared with those previously obtainedby standard DFT-GGA16,31 in Table 1. Even though the

Figure 1. Side (left) and top (right) views of the anatase(101) surfacewith T4 (top panels) and T5 (bottom panels) titanium interstitials.The Ti5c sites exposed on the two surfaces are numbered as in the text.Ti and O atoms are represented as silver and red spheres, respectively,with larger spheres highlighting the Ti interstitials.

Table 1. Adsorption Energy of a Single Water Adsorbed atDifferent Ti5c Sites in the Presence of T4 and T5Interstitialsa

DFT+U DFT-GGA

surface site Eads ΔE Eads16,31 ΔE

T4 1 0.57 0.82 0.67 0.432 (3) 1.39 (1.06) 0 (0.33) 1.10 (0.84) 0 (0.26)4 (5) 0.85 0.54 0.79 0.316 1.08 0.31 0.86 0.24

T5 1 (3) 0.73 0.39 0.71 0.342 1.12 (0.92) 0 (0.20) 1.05 04 0.86 0.26 0.73 0.325 (6) 0.89 0.23 0.74 0.31

aBoth absolute energies (Eads, eV) and energies relative to the moststable adsorption mode (ΔE) are reported. Bold (normal) valuesdenote dissociative (molecular) adsorption.

The Journal of Physical Chemistry C Article

dx.doi.org/10.1021/jp301624v | J. Phys. Chem. C 2012, 116, 9114−91219115

absolute values are different, the relative trends of DFT+U andDFT-GGA adsorption energies are exactly the same: theincreasing order of stability calculated by the two approaches is1 < 4 < 6 < 2 both for T4 and T5. Both DFT+U and DFT-GGA identify the dissociative state at site 2 (close to theinterstitial) as the most stable adsorption mode, whereasadsorption is weaker and molecular on the other Ti5c sites. OnT4, dissociation at site 2 is more favorable than molecularadsorption by 0.33 eV (0.26 eV with DFT-GGA): the initialand final states are shown in Figure 2. The main effect of the+U repulsion is to enhance the adsorption energy differencebetween the different states.

Previous NEB/DFT-GGA calculations found a low energybarrier (0.24 eV) for water dissociation at site 2 on surfaceT4.16 With DFT+U, NEB calculations give an even lowerbarrier, 0.14 eV. The strong tendency of a water molecule todissociate at site 2 on the T4 surface was confirmed by a 110 KDFT+U BOMD trajectory: the adsorbed molecule dissociatedafter 0.7 ps and the resulting pair of OH groups (right panel ofFigure 2) remained stable for the rest of the trajectory. Thedynamic behavior obtained with the DFT+U approach is inclose agreement with that observed in DFT-GGA CPMDsimulations.16 In summary, we find that DFT-GGA reproduceswell both the energetics and the kinetics of single-wateradsorption in the presence of Ti interstitials given by the DFT+U calculations.Monolayer. Reactivity. A DFT-GGA CPMD trajectory at

160 K was started after placing a water molecule ∼2.5 Å aboveeach of the six Ti5c exposed on the T4 surface. Figure 3illustrates the time evolution of the water monolayer. The watermolecules in the left row, above the interstitial, show asignificantly richer dynamic behavior compared to themolecules in the adjacent row. The water molecule initiallyadsorbed on site 1 (W1 hereafter) leaves this site and rapidlymoves closer to the region above the interstitial, between thewater molecule on site 2 (W2 hereafter) and the adjacent rowof 2-fold coordinated bridging oxygens (BOs), accepting ahydrogen bond (Hb) from W2 and donating Hbs to two BOs.W1 reaches this stable configuration after 1 ps and maintains itfor the rest of the trajectory. In this configuration, the moleculeassists the proton transfer from W2 to one of the BOs: thisprocess starts after 8 ps, when W2 dissociates and donates aproton to W1, which rapidly transfers another proton to a BO.Whereas W2 remains dissociated for the rest of the trajectory,frequent proton exchanges occur between W1 and the twoBOs, essentially resulting in two different configurations, shownat 12 and 28 ps in Figure 3. The latter configuration is observedmore frequently.

Starting from the same initial configuration used for theDFT-GGA CPMD trajectory, a DFT+U BOMD simulation at160 K showed a dynamical evolution very similar to thatobserved in the DFT-GGA run: W1 desorbs from its initial siteand moves to the same stable configuration where it assists W2dissociation and coordinates its proton transfer to a BO (seeFigure 4). These processes occur however on a slightly shortertime scale compared to the DFT-GGA runs. As in the DFT-GGA run, Figure 4 shows that also with DFT+U the W1molecule leaves the initial site and reaches its stableconfiguration in ∼1 ps; at this point, however, it immediatelygives rise to the proton transfer events leading to W2dissociation and protonation of a BO, whereas the reactionoccurred after an additional ∼7 ps in the DFT-GGA run. Thedriving force for the observed process likely is the interstitial-induced destabilization of site 1 (see Table 1). This couldindeed explain the water desorption from that site andmigration to a more favorable configuration, which in turnenables dissociation of an adjacent water molecule. Thisinterpretation is supported by the fact that a CPMD run forsurface T5 also led to desorption from site 1, which representsthe site with the lowest single-water adsorption energy for thesurface with the T5 interstitial as well.We determined the relative stabilities of the intermediate

structures observed during the dynamical evolution of the watermonolayer by calculating the adsorption energies of thegeometry-optimized structures in Figure 5 (see Table 2). Theinitial flat ML (Figure 5a) corresponds to the most stablestructure on the defect-free surface.14 Migration of W1 is drivenby the higher stability of structure (b), whose energy is actuallyvery close to that of the corresponding dissociated structure(c). The similar stability of molecular and dissociated structuresleads to the frequent proton transfers observed during thedynamics, corresponding to breaking and reassociation of O−Hbonds. The trend calculated at the DFT-GGA level is the same,even thoughas in the case of single water adsorptiontheabsolute adsorption energies are lower.

Vertical Ordering. Figure 6 shows the vertical distance fromthe surface of each water oxygen during the DFT-GGA and

Figure 2. Molecular (left) and dissociative (right) adsorption states ofa single water molecule on site 2 on surface T4. Color codes as inFigure 1, with water oxygen highlighted in yellow and hydrogen white.

Figure 3. Selected snapshots extracted from the DFT-GGA CPMDsimulation of a water monolayer on the T4 surface. The oxygens of thethree water molecules on the row atop the interstitial are highlightedin yellow (W1), green (W2), and pink.

The Journal of Physical Chemistry C Article

dx.doi.org/10.1021/jp301624v | J. Phys. Chem. C 2012, 116, 9114−91219116

DFT+U MD trajectories of the water monolayer. In both cases,there is a noticeable difference between the height of the watermolecules in the row atop the interstitial and those in theadjacent row, which are closer to the surface. An additional gapis present between the W1 and W2−W3 molecules atop theinterstitial: W1, not being coordinated to a Ti5c, is farther awayfrom the surface than W2 and W3. These features are commonto both DFT-GGA and DFT+U trajectories, indicating that theperturbation induced by the interstitial on the ML verticalordering is qualitatively reproduced already at the DFT-GGAlevel. However, the W4−W6 molecules are slightly (0.15 Å)closer and the W5−W6 slightly (0.1 Å) farther from the surfacein the DFT+U simulation compared to the vertical distances inthe DFT-GGA CPMD run. This results in a more pronouncedvertical split between W4−W6 and W2−W3 molecules, asshown by the vertical distance (Δz) distributions in the leftpanels of Figure 6. This difference could partially reflect the factthat the proton transfer events take place earlier in the DFT+Uruns, whereas, as mentioned above, no dissociation occurs inthe DFT-GGA runs within the time frame covered in Figure 6.

Water−Water and Water−Surface Interactions. The radialdistribution functions for pair interactions involving watermolecules in the ML are displayed in Figure 7. We can see thatDFT-GGA and DFT+U yield quite similar local structures forthe water ML. The characteristic peaks for the water−surfaceinteraction are shown in the bottom panel of the figure. Themain Ti−Ow peak (where Ow is a water oxygen) is split in twodue the contributions from the Ti−OH bond (around 1.9 Å)formed upon W2 dissociation, much shorter than the Ti−OH2distance of ∼2.3 Å. The Ow−Os (where Os is an oxygenbelonging to the TiO2 slab) main peak at 2.8 Å arises fromhydrogen bonds formed between W2 and the two adjacentbridging oxygens; this peak has a short-distance shoulderaround 2.5 Å, which reflects a shorter Ti−OsH···Ow hydrogenbond formed upon W1 dissociation.The water−water radial distribution functions are displayed

in the top panel of Figure 7. Whereas the Ow−Ow peak at 3.8 Åreflects water molecules not hydrogen-bonded and in registrywith the lattice of Ti5c sites, as in the unperturbed ML on thedefect-free surface (see ref 14 and also the discussion below),new O−H and O−O peaks at 1.5 and 2.5−2.7 Å reveal

Figure 4. Time evolution of characteristic interatomic distances duringthe DFT+U and DFT-GGA MD runs of water ML on the T4 surface.The top panels (a) illustrate the migration process of the W1 moleculeleaving its initial site and settling into the final configuration; panels(b) highlight the proton transfers to the bridging oxygens; panels (c)describe the dissociation of molecule W2; and panels (d) describe theproton transfers involving W1. The vertical dashed line identifies thestart of the dissociation processes in each case.

Figure 5. DFT+U-optimized structures of a water monolayer on theT4 surface. Color codes as in Figure 1, with water oxygen highlightedin yellow and hydrogen white.

Table 2. Adsorption Energy (eV/H2O) of a WaterMonolayer on T4 and T5 Surfaces (Structures Are Labeledas in Figure 5)

Eads

surface structure DFT+U DFT-GGA

T4 (a) 0.80 0.77(b) 0.88 0.81(c) 0.91 0.82

T5 (a) 0.78 0.73(b) 0.82 0.76(c) 0.81 0.76

Figure 6. Distribution (left panels) and trajectory (right) of water-surface vertical distances. Δz = z − z, where z is the average zcoordinate of the six Ti5c. Top panels: DFT-GGA; bottom panels:DFT+U.

The Journal of Physical Chemistry C Article

dx.doi.org/10.1021/jp301624v | J. Phys. Chem. C 2012, 116, 9114−91219117

additional hydrogen bonds within the ML involving almostexclusively W1 and W2. These features characterize both theDFT+U and DFT-GGA description of the ML−surface andintra-ML interactions. The most significant difference is theOw−Ow H-bond distance, which is 0.2 Å shorter in the DFT-GGA model.

■ INFLUENCE OF THE TI INTERSTITIAL ON THESTRUCTURE OF 1−3 WATER LAYERS

From the results in the previous section, it is reasonable toconclude that the DFT+U and DFT-GGA descriptions ofstructural and dynamical features of a water monolayer on thereduced surface are sufficiently close to justify the use of thestandard DFT-GGA approach to investigate systems whichwould be too cumbersome to simulate with spin-polarized DFT+U-based BOMD. Therefore, we now use DFT-GGA CPMDto study the effects of the interstitial defect on the structure ofup to three layers of adsorbed water. For this purpose, we useas a reference the corresponding structures on the defect-freesurfaces, obtained with the same computational settings.Thin Water Layers on the Stoichiometric Surface.

Vertical Ordering. DFT-GGA CPMD simulations of one, two,and three water layers on the defect-free surface were carriedout at 160 K for 18 ps. The vertical distances in Figure 8 showthat the ML and BL quickly settle into stable configurations,characterized by well-defined distances from the surface foreach molecule and corresponding sharp peaks in the distancedistributions at Δz = 2 and 3 Å. The left and middle panel inFigure 9 shows that water molecules in the first layer (Δz = 2Å) are coordinated to Ti5c, whereas the second layer (Δz = 3Å) is formed by strongly oriented molecules, all arranged withone hydrogen pointing upward and the other toward thesurface and forming a strong H-bond with a BO. The verticalordering of the first two layers in contact with the surface ismaintained also in the TL: the position of the first two peaks inthe z distribution is the same already observed for the ML andBL. However the presence of the additional water molecules

introduces some disorder in the first layer, with a moleculebecoming slightly higher than the others. Moreover, only threemolecules are now in the second layer, H-bonded to BOs, whilea third layer is present at Δz = 4.5 Å, formed by molecules H-bonded to the first two water layers (Figure 9, right panel). Thecorresponding peak in the height distribution has a tailextending up to 6 Å, suggesting a higher mobility and a moredisordered arrangement within the top layer.

Intra- and Interlayer Structure. H2O−H2O Interactions.The radial distribution functions (rdfs) in Figure 10 allow us tofurther characterize the structure of the thin layers on thedefect-free surface. The top panel shows that Ti5c-coordinatedwater molecules in the ML do not significantly interact witheach other, due to the large Ti5c−Ti5c distance.14 Clearsignatures of a H-bonded system (intermolecular Ow−Hw andOw−Ow peaks at 1.7 and 2.7 Å, respectively) emerge for the BL,and are maintained for the TL, with a generally moredisordered arrangement evident from the broader peaks.

H2O−Surface Interactions. The persistent Ti−Ow peakaround 2.3 Å indicates that water molecules remain strongly

Figure 7. Water−water (top) and water−surface (bottom) radialdistribution functions for a ML on surface T4, calculated from theDFT-GGA and DFT+U MD simulations. The DFT+U curves areshifted vertically for clarity.

Figure 8. Distribution (left panels) and trajectory (right) of water−surface vertical distances for mono-, bi-, and trilayer adsorbed on thedefect-free surface. Δz = z − z, where z is the average z coordinate ofthe six Ti5c.

Figure 9. Structure of a water mono-, bi-, and trilayer on the defect-free surface. In each case, the last configurations of the correspondingMD runs at 160 K is shown. H2O molecules bonded to Ti5c and Os arehighlighted in yellow and blue, respectively, whereas the remainingmolecules are green.

The Journal of Physical Chemistry C Article

dx.doi.org/10.1021/jp301624v | J. Phys. Chem. C 2012, 116, 9114−91219118

coordinated to Ti5c at the different coverages, and the strengthof this interaction increases at coverages higher than one ML.Water molecules in the ML are anchored to the Ti5c sites andcan only form weak Hbs with the surface BO (bottom panel,dashed red and cyan curves). On the other hand, the BL andTL also include molecules that are not coordinated to Ti5c andare thus able to form stronger Hbs with the surface. Inparticular, water−surface Hb interactions are strongest in theBL due to the contribution from strongly oriented moleculespointing a hydrogen down (Figure 9, middle panel), whichleads to sharp Hw−Os and Ow−Os peaks at 1.5 and 2.5 Å. Eventhough the weight of molecules in this specific configuration islower in the TL, for which the sharp peaks noted aboveessentially disappear, water−surface interactions are stillsignificant for the TL.Thin Water Layers on the Reduced Surface. Vertical

Ordering. DFT-GGA simulations of 1−3 water layers on theT4 surface were carried out for 20 ps at 160 K. The finalconfigurations are shown in Figure 11. As already pointed out,the presence of the interstitial affects substantially the MLstructure compared to the defect-free surface. The orderedarrangement of Ti5c-coordinated molecules observed on thestoichiometric surface changes into a partially dissociated layerwith an empty Ti5c site, due to the migration of the W1 water

to an intermediate position between the Ti5c and BO rows(Figure 11, left panel). The molecules directly above theinterstitial tend to stay farther away from the surface, leading toa secondary peak in the z-distribution (Figure 12, bottompanels).

The Ti interstitial has an even more marked influence on theBL structure, whose vertical arrangement changes from theordered double layer with sharp peaks seen on the defect-freesurface to a multipeak distribution reaching up to 8 Å from theT4 surface. Even though the peaks at 2 and 3 Å characteristic ofthe BL on the defect-free surface are still present, the structureof these layers is significantly perturbed, with a dissociatedmolecule in the first layer and only half of the molecules left inthe second layer, H-bonded to BOs (Figure 11, middle panel).The remaining molecules produce an additional peak at z = 4.2Å and further extend up to 8 Å away. For the TL, the structureof the first two layers in contact with the surface (Figure 11,right panel) is similar to that just described for the BL, with adissociated water in the first layer and a depleted second layerof molecules H-bonded to the surface. Above these two layersin direct contact with the surface sites, the remaining moleculesin the BL and TL are significantly spread for about 3 Åperpendicularly to the surface; this upper region is uniformlyfilled only at TL coverage, whereas low-density gaps are evidentfor the BL (Figures 11 and 12).

Intra- and Interlayer Structure. Water−Water Interac-tions. The radial distribution functions for water on the T4surface (Figure 13) further detail the substantial influence ofthe interstitial defect on the structure of adsorbed water. Atvariance with the defect-free surface, signatures of hydrogenbonds between water molecules are observed already in theML. In this case, Hbs always involve OH groups produced by

Figure 10. Radial distribution functions for water−water (top) andwater−surface (bottom) interactions for mono-, bi-, and trilayers onthe defect-free surface.

Figure 11. Structure of a water mono-, bi-, and trilayer on the T4interstitial surface. In each case, the last configurations of thecorresponding CPMD runs at 160 K is shown. Color codes as inFigure 9.

Figure 12. Distribution (left panels) and trajectory (right) of water−surface vertical distances for mono-, bi-, and trilayer adsorbed on thesurface with T4 interstitial. Δz = z − z, where z is the average zcoordinate of the six Ti5c.

The Journal of Physical Chemistry C Article

dx.doi.org/10.1021/jp301624v | J. Phys. Chem. C 2012, 116, 9114−91219119

water dissociation and are stronger than Hbs betweenundissociated water molecules, which become dominant inthe bi- and trilayer: Ow−Ow and Ow−Hw peaks lie at 2.5 and1.45 Å in the ML, compared to values of 2.7 and 1.7 Å in theBL and TL. The water−water hydrogen bond patterns in theBL and TL are very similar, with distances approaching those ofa liquid water environment.Water−Surface Interactions. The water−surface rdfs are

shown in the bottom panel of Figure 13. The formation of aTi−OH bond results in the short-distance peak in the Ti−Owrdf, whose weight is obviously higher in the ML case. The Ti−Ow rdf remains structured for the ML and increasingly lessgoing to BL and TL. At variance with the defect-free surface,hydrogen bonds between water and the reduced surface arealready present at ML coverage (compare the Hw−Os rdfsbetween 1.5 and 2 Å), mainly involving the W1 molecule. Themore prominent water−surface Hbs peaks in the BL than in TLcase are somewhat a remnant of the highly ordered pattern forthe BL on the defect-free surface.

■ DISCUSSION AND FINAL REMARKSRecent scanning tunneling microscopy (STM) and temper-ature-programmed desorption (TPD) studies of water onreduced anatase (101) surfaces have provided evidence of ahigher water desorption temperature on highly reduced surfacesas well as a clear preference for water adsorption close tosubsurface defects.16 These findings were supported by DFTcalculations, which showed higher adsorption energies at sitesabove subsurface oxygen vacancies and Ti interstitials. Low-coverage water dissociation turned out to be more favorable onTi interstitials than on O vacancies, indicating a higherreactivity of interstitial defects.16

Motivated by these results, in this paper we have focused onTi interstitials, and in particular we have studied the influenceof a subsurface Ti interstitial on the structure and dynamics ofthin water layers on the anatase (101) surface. By performing adetailed comparison of DFT-GGA CPMD and DFT+U

BOMD simulations for an adsorbed monolayer on the defectedsurface, we have shown that DFT-GGA can adequately describethe structure and dynamics of thin water films on reducedTiO2, even though this approach provides an unsatisfactorydescription of the electronic structure of localized defect statesin the material.21−23 In this context, a particularly importantfinding is that DFT-GGA and DFT+U predict the same orderof stability for single-water adsorption on the surface with asubsurface interstitial defect, analogously to what previouslyobserved for the surface with a subsurface O vacancy.16

The improved description of the localized electronic statesassociated with the interstitial defect, obtained with the +Ucorrection, does not affect significantly the main structuralfeatures of the adsorbed water layers. The only small differencesconcern the time scale of water dissociation and the Ow−Owdistance between two H-bonded water molecules in a ML. Theshorter time for the onset of dissociation in the GGA+Uapproach is consistent with the smaller computed waterdissociation barrier. Moreover, taking into account the well-known GGA underestimation of the O−O distance in liquidwater,38 the longer O−O distance between two H-bondedadsorbed water molecules obtained with the +U correctionsuggests that DFT+U can indirectly improve the description ofH-bonds in adsorbed water, similarly to the effect of hybridfunctionals for liquid water.39 Experimental techniques such asinfrared reflection absorption spectroscopy (IRRAS)40,41 couldprobe the H-bond structure of the adsorbed layers andcomplement our findings.On the stoichiometric anatase(101) surface, the main factors

which control multilayer water adsorption are the balancebetween water−surface and water−water interactions (thelatter being weaker14) and the surface specific geometry whichdoes not favor interactions between water moleculescoordinated to the surface sites. This leads to substantialvertical and in-plane ordering of the first two layers in directcontact with the surface, which is maintained even at highercoverages, although surface-induced ordering is weaker abovethe second layer. On the reduced (101) surface, on the otherhand, the structure of water overlayers is affected also by thedefect-induced stress, which destabilizes H2O−Ti5c bonds andbreaks the ordered arrangements formed on the stoichiometricsurface. We previously observed a similar effect caused by asurface oxygen vacancy,17 which also leads to a partiallydissociated and highly disrupted water ML, with H2O−Ti5ctransformed into H2O−Os, as well as water−water and water−surface Hbs not present on the clean surface.Intermolecular water interactions at higher coverages have

previously been found to favor water dissociation in thepresence of both surface17 and subsurface16 oxygen vacancies.This effect is less noticeable for an interstitial due to the higherreactivity of the defect which strongly favors dissociationalready at low coverage. The mechanism of O vacancy- and Tiinterstitial-induced dissociation may also differ: whereas asurface oxygen vacancy can directly participate to thedissociation mechanism at low and high coverages (the drivingforce being the filling of the vacancy by an OH group),17,36 asubsurface interstitial affects the process indirectly, through thedestabilization and consequent mobilization of water moleculeswhich enables water dissociation by a concerted mechanism. Inany case, the present simulations confirm that defects, eithersurface or more likely subsurface,18 are required to enable waterdissociation on anatase(101) at any coverage.

Figure 13. Radial distribution functions for water−water (top) andwater−surface (bottom) interactions for mono-, bi-, and trilayers onthe T4 surface.

The Journal of Physical Chemistry C Article

dx.doi.org/10.1021/jp301624v | J. Phys. Chem. C 2012, 116, 9114−91219120

The structure of thin water layers on the defect-free surfacethat we have found in this study is very close to that previouslyobtained with less demanding simulations involving thinnerslabs and shorter MD runs.14,17 However, among the differentBL configurations observed in our previous simulations,14 thepresent results strongly favor the highly ordered arrangementwhere each molecule in the second layer is oriented with ahydrogen atom pointing upward. This arrangement denotesunfavorable lateral interactions between molecules H-bondedto a surface BO (Os). Alternative BL configurations can only beattained by removing a molecule from either a Ti5c site or an Ossite,14 but we have not observed these alternative and moredisordered BL configurations in the present simulations,suggesting that their energetic cost is too high when a morerealistic surface model is adopted. Therefore, the most stablestructures of a ML and a BL on the defect-free surface containwater molecules in registry with all the Ti5c and Os exposed onthe surface; only when a third water layer is adsorbed, a lessordered arrangement appears also in the first two layers, whichdo not longer match all the surface sites.The Ti interstitial strongly perturbs the ordered adsorption

structure present on the stoichiometric surface. Watermolecules in the row atop the defect are much more perturbedthan those in the adjacent row; moreover, adsorption just abovethe interstitial is strengthened whereas adsorption at theadjacent sites is weakened, ultimately leading to W1 desorptionand W2 dissociation. The disrupted structure of the first twocontact layers at BL and TL coverage represents a less favorablelandscape for the attachment of additional molecules, whichcould at least in part explain the significant vertical spread onthe reduced compared to the defect-free surface (compareFigures 8 and 12). These effects could also reflect the excess(negative) charge introduced on the surface by the interstitial.37

In particular, the higher spread and also the larger distance fromthe surface of molecules directly above the interstitial suggestthat reducing defects provide some hydrophobic character tothe surface. Additional simulations of bulk liquid water on thereduced surface would be needed to assess the long-rangeconsequences of this effect: the results of this work show thatthis task can be effectively accomplished through standardDFT-GGA CPMD simulations.15,42

■ AUTHOR INFORMATIONCorresponding Author*E-mail: a.tilocca@ucl.ac.uk.

NotesThe authors declare no competing financial interest.

■ ACKNOWLEDGMENTSA.T. acknowledges financial support from the U.K.’s RoyalSociety (University Research Fellowship). A.S. thanks thesupport of DoE-BES, Chemical Sciences, Geosciences andBiosciences Division, Contract DE-FG02-12ER16286. We usedresources of the TIGRESS high performance computer centerat Princeton University, which is jointly supported by thePrinceton Institute for Computational Science and Engineeringand the Princeton University Office of Information Technol-ogy.

■ REFERENCES(1) Henderson, M. A. Surf. Sci. Rep. 2002, 46, 1.(2) Henderson, M. A. Surf. Sci. Rep. 2011, 66, 185.

(3) Fujishima, A.; Zhang, X. T.; Tryk, D. A. Surf. Sci. Rep. 2008, 63,515.(4) Diebold, U. Surf. Sci. Rep. 2003, 48, 53.(5) Sun, C. H.; Liu, L.; Selloni, A.; Lu, G.; Smith, S. C. J. Mater.Chem. 2010, 20, 10319.(6) Cohavi, O.; Corni, S.; De Rienzo, F.; Di Felice, R.; Gottschalk, K.E.; Hoefling, M.; Kokh, D.; Molinari, E.; Schreiber, G.; Vaskevich, A.;Wade, R. C. J. Mol. Recognit. 2010, 23, 259.(7) Monti, S. J. Phys. Chem. C 2007, 111, 16962.(8) Monti, S.; Walsh, T. R. J. Phys. Chem. C 2011, 105, 24238.(9) Skelton, A. A.; Liang, T.; Walsh, T. R. ACS Appl. Mater. Interfaces2009, 1, 1482.(10) Tilocca, A.; Cormack, A. N. ACS Appl. Mater. Interfaces 2009, 1,1324.(11) He, Y.; Tilocca, A.; Dulub, O.; Selloni, A.; Diebold, U. Nat.Mater. 2009, 8, 585.(12) Miller, K. L.; Musgrave, C. B.; Falconer, J. L.; Medlin, J. W. J.Phys. Chem. C 2011, 115, 2738.(13) Sanchez, V. M.; de la LLave, E.; Scherlis, D. A. Langmuir 2011,27, 2411.(14) Tilocca, A.; Selloni, A. Langmuir 2004, 20, 8379.(15) Sumita, M.; Hu, C.; Tateyama, Y. J. Phys. Chem. C 2010, 114,18529.(16) Aschauer, U.; He, Y.; Cheng, H.; Li, S.; Diebold, U.; Selloni, A. J.Phys. Chem. C 2010, 114, 1278.(17) Tilocca, A.; Selloni, A. J. Phys. Chem. B 2004, 8, 4743.(18) He, Y.; Dulub, O.; Cheng, H.; Selloni, A.; Diebold, U. Phys. Rev.Lett. 2009, 102, 106015.(19) Cheng, H.; Selloni, A. Phys. Rev. B 2009, 79, 092101.(20) Cheng, H.; Selloni, A. J. Chem. Phys. 2009, 131, 054703.(21) Di Valentin, C.; Pacchioni, G.; Selloni, A. Phys. Rev. Lett. 2006,97, 166803.(22) Finazzi, E.; Di Valentin, C.; Pacchioni, G.; Selloni, A. J. Chem.Phys. 2008, 129, 154113.(23) Ganduglia-Pirovano, V. M.; Hofmann, A.; Sauer, J. Surf. Sci. Rep.2007, 62, 219.(24) Wendt, S.; Sprunger, P. T.; Lira, E.; Madsen, G. K.; Li, Z.;Hansen, J.; Matthiesen, J.; Blekinge-Rasmussen, A.; Laegsgaard, E.;Hammer, B.; Besenbacher, F. Science 2008, 320, 1755.(25) Henderson, M. A. Surf. Sci. 1999, 419, 174.(26) Anisimov, V. I.; Zaanen, J.; Andersen, O. K. Phys. Rev. B 1991,44, 943.(27) Giannozzi, P.; et al. J. Phys.: Condens. Matter 2009, 21, 395502.(28) Perdew, J. P.; Burke, K.; Ernzerhof, M. Phys. Rev. Lett. 1996, 77,3865.(29) Vanderbilt, D. Phys. Rev. B 1990, 41, 7892.(30) Car, R.; Parrinello, M. Phys. Rev. Lett. 1985, 55, 2471.(31) Aschauer, U.; Selloni, A. Proc. SPIE 2010, 7758, 77580B.(32) Blochl, P. E.; Parrinello, M. Phys. Rev. B 1992, 45, 9413.(33) Marx, D.; Hutter, J. In Modern Methods and Algorithms ofQuantum Chemistry; Grotendorst, J., Ed.; NIC Series; FZ Julich:Germany, 2000; Vol. 1.(34) Using 32 cores connected by Infiniband on the Della Linuxcluster (http://www.princeton.edu/researchcomputing/), a trajectoryof 1 ps required 300−400 wall clock hours using spin-polarized DFT+U BOMD and 16 h with spin-restricted DFT-GGA CPMD.(35) Henkelman, G.; Uberuaga, B. P.; Jonsson, H. J. Chem. Phys.2000, 113, 9901.(36) Tilocca, A.; Selloni, A. J. Chem. Phys. 2003, 119, 7445.(37) Finazzi, E.; Di Valentin, G.; Pacchioni, G. J. Phys. Chem. C 2009,113, 3382.(38) Sit, P.H.-L.; Marzari, N. J. Chem. Phys. 2005, 122, 204510.(39) Zhang, C.; Donadio, D.; Gygi, F.; Galli, G. J. Chem. TheoryComput. 2011, 7, 1443.(40) Xu, M.; Noei, H.; Buchholz, M.; Muhler, M.; Woll, C.; Wang, Y.Catal. Today 2012, 182, 12.(41) Kimmel, G. A.; Baer, M. A.; Petrik, N. G.; VandeVondele, J. A.;Rousseau, R.; Mundy, C. J. J. Phys. Chem. Lett. 2012, 3, 778.(42) Cheng, H.; Selloni, A. Langmuir 2010, 26, 11518.

The Journal of Physical Chemistry C Article

dx.doi.org/10.1021/jp301624v | J. Phys. Chem. C 2012, 116, 9114−91219121