About the Nitrogen Location in Nanocrystalline N-Doped TiO2

Jan 4, 2012 - Eventually, the starting electron populations on Ti, O, and N were always chosen so that the crystal cell was always electrically neutra...
2 downloads 6 Views 2MB Size
ARTICLE pubs.acs.org/JPCC

About the Nitrogen Location in Nanocrystalline N-Doped TiO2: Combined DFT and EXAFS Approach Michele Ceotto,*,† Leonardo Lo Presti,*,† Giuseppe Cappelletti,† Daniela Meroni,† Francesca Spadavecchia,† Roberto Zecca,‡ Matteo Leoni,‡ Paolo Scardi,‡ Claudia L. Bianchi,† and Silvia Ardizzone† † ‡

Dipartimento di Chimica Fisica ed Elettrochimica, Universita degli Studi di Milano, Via Golgi 19, 20133 Milano, Italy Dipartimento di Ingegneria dei Materiali e Tecnologie Industriali, Universita degli Studi di Trento, via Mesiano 77, 38123 Trento, Italy

bS Supporting Information ABSTRACT: N-doped titanium dioxide is one of the most promising materials for photocatalysis in the visible region. The exact location of nitrogen in the host lattice is still under debate. Here, we synthesized a series of N-doped titania nanoparticles. Average Ti nearest neighbors distances were obtained from EXAFS experiments and compared with DFT calculations at different levels of theory. The comparison shows that N substitutes oxygen at low levels of doping, whereas oxygen vacancy creation is observed at higher dopant concentrations. Overall, this article illustrates a general method for bulk characterization based on DFT and EXAFS approaches, which can be extended to several systems.

1. INTRODUCTION Titania (Titanium dioxide, TiO2) has been used for many years in several technological areas, embracing heterogeneous catalysis, photocatalysis, solar cells, gas sensors, waste remediation, and even biocompatible-materials. Such a widespread usage is granted by its peculiar electronic structure, where the photogenerated electrons at the bottom of the conduction band and the holes generated at the top of the valence band induce, respectively, photoreductions and photo-oxidations.1 Nevertheless, most of these applications have been hampered by titania wide and insulator-like band gap and, therefore, confined to the UV radiation, which is less useful. In order to extend titania photoactivity into the visible range, after the pioneering work of Asahi et al.,2 a lot of efforts have been devoted to the synthesis of variously N-doped titania.3 Nowadays, nitrogen incorporation can be achieved by both physical and wet-chemical synthetic methods (such as solgel synthesis, chemical treatments of the bare oxide, titanium nitride oxidation, sputtering with ArN2 gas, N-ion bombardments at high energies, molecular beam epitaxial growth, chemical vapor deposition, post-treatment by NH3 gas at elevated temperatures, spray pyrolysis, and other supercritical methods).1,3d Given such a jungle of synthetic routes, materials with largely different properties are produced. Therefore, direct comparisons and unifying photocatalytic conclusions are difficult to draw. Moreover, experimental characterization techniques failed as yet in determining the exact location of the nitrogen anion in the lattice, namely, interstitial or oxygen-substitutional. This limitation is a key point frequently recurring in literature.4 So far, the r 2012 American Chemical Society

anion has been generically labeled as Nb, where b stands for bulk.3a DFT calculations5 based on thermodynamic considerations under oxygen-rich conditions have suggested N atoms to be preferably interstitially located in the lattice. However, these calculations have never been backed by experimental results. Also, it is still under debate whether other chemical species like NOx or NHx are present in bulk and why, in some cases, the photocatalytic performance gets worse by increasing N content, even if visible light absorption increases. Overall, the experimental solid state picture of the N-doped titania remains scanty because detailed information about the crystal structure and the nature of the photoactive centers are still missing. A clear understanding of these issues is necessary to build up focused syntheses and to improve the photocatalytic performance of N-doped titania. It is essential to know whether Nb is primarily interstitial or substitutional because the very different behavior of N in these sites will accordingly affect the material photocatalytic properties. Figure 1 schematically represents the electronic structure of substitutional (left side) and interstitial (right side) N-doped titania.3a,6 In the first case, after suitable irradiation, both A and B compounds are oxidized, whereas in the second case, only A is oxidized. Electronhole recombination is more likely to occur in the interstitial case because of the apparent band gap narrowing induced by the N states, and any beneficial effect may vanish in this way. For these reasons, Received: October 11, 2011 Revised: November 23, 2011 Published: January 04, 2012 1764

dx.doi.org/10.1021/jp2097636 | J. Phys. Chem. C 2012, 116, 1764–1771

The Journal of Physical Chemistry C

Figure 1. Schematic photocatalytic oxidations with substitutional (left) and interstitial (right) N-doped titania. A and B are molecules with different electronic energy levels. The subscript ox stands for the oxidized form and red for the reduced one.

a rational comprehension of the synthetic route with the target of improved photocatalytic properties can only be achieved if the nature of the synthetic products is fully determined, in particular the location of the N atom within the lattice. Several experimental techniques have been employed to understand where nitrogen is located in titania. For example, X-ray photoelectron spectroscopy of the N1s electron is quite noisy due to the low dopant concentration and presents questionable attribution of poorly resolved peaks, which are the convolution of more than one signal component.7 Moreover, this technique is representative of the outer part (surface and subsurface electronic states) of the material, rather than the actual bulk one.8 However, electron paramagnetic resonance, which can detect very low concentrations of paramagnetic species, is not able to discriminate between interstitial and substitutional doping. The spin density on N indicates that both doping models may account for the observed species.5,8 X-ray absorption spectroscopy (XAS), and in particular extended X-ray absorption fine structure (EXAFS), is, however, a promising alternative for the localization of N, as it provides information on the local site symmetry and average bond lengths. It has been already employed on several nanostructured oxides.9 However, its potential in the field of doped titania has not been fully exploited yet. In this article, we studied the chemical environment surrounding the (heavy) Ti atoms for a series of N-doped samples by combining Ti K-edge EXAFS spectra with DFT calculations. Moreover, the N-doping effects on the bulk geometry have been computationally explored at different levels of theory and a joint experimental and theoretical answer is put forward.

2. EXPERIMENTAL AND THEORETICAL SETUP 2.1. Synthesis. Pure and doped titania samples were prepared by solgel synthesis, with titanium(IV) isopropoxide and triethylamine (for the doped samples) used as precursors and calcined at 400 C in oxygen stream, as reported in our previous work.7 The undoped sample is denoted as T, while the doped ones are labeled as TN_x, with x = 0.05, 0.10, 0.20, 0.40, and 0.50, according to the initial N/Ti molar ratio.

ARTICLE

2.2. EXAFS Analysis. To avoid thickness effects, sample tablets suitable for recording XAS spectra were prepared by carefully diluting the material in CaCO3. X-ray absorption curves (Figure S1 in the Supporting Information) were then collected around Ti K-edge at room temperature in transmission mode at the BM01B beamline of ESRF (European Synchrotron Radiation Facility, Grenoble, France). A Si (111)-monochromated beam was employed to probe six nanostructured samples (pure TiO2 plus the five N-doped specimens described above) in the 4.95.4 (pure TiO2) and 4.95.8 keV (N-doped TiO2) energy ranges. The Horae suite of programs,10 based on the IFEFFIT library,11 was used throughout for data processing and fitting. As our samples contained a variable non-negligible fraction of brookite (see the Supporting Information), we explicitly took into account backscattering paths belonging to both anatase and brookite crystal structures. In regard to the anatase phase, we independently refined three Δr parameters, corresponding, respectively, to the TiO first-shell equatorial, TiO first-shell axial, and Ti-outer shell distances. To maintain the total number of parameters below the upper limit set by the Nyquist theorem (Nfree = 2ΔkΔr/π + 1), we never introduced explicitly backscattering paths involving bulk N atoms. Rather, we preferred to indirectly infer the effect of dopant N atoms from the distortions of the average axial and equatorial TiO bond distances in the Ti first coordination shell (see section 3.1). More details on the final parameter statistics and agreement factors, together with the statistical assessment of the fitting model employed to interpret experimental data can be found in the Supporting Information. 2.3. Plane Wave Computational Setup. All calculations were spin-polarized and performed using the projector augmented wave (PAW) pseudopotentials to treat the valence-core interactions.12 The PerdewBurkeErnzerhof parametrization13 of the generalized gradient approximation14 was adopted for the exchange-correlation potential. The cutoff energy of the plane wave basis was 400 eV. Forces calculated through the HellmannFeyman theorem included the HarrisFoulkes corrections,15 and optimizations were performed using the conjugate-gradient scheme.16 Iterative relaxation of atomic positions was stopped when the change in total energy between successive steps was less than 0.001 eV and residual forces were below 0.01 eV/Å. As far as the DFT+U approach is concerned,17 this introduces an on-site correction in order to describe systems with localized d electrons. Here, the effective on-site Coulombic interactions U (U = U0  J) for Ti 3d were also used, where U0 and J represent the energy cost of adding an extra electron at a particular site and the screened exchange energy, respectively. A value of U = 5 eV was used, which has previously been shown to properly account for the electronic structure of the Ti 3d states.18 For the Ti54O108 supercell, reciprocal space sampling was restricted to the Γ-point, which is justified due to the rather large size of the used simulation supercells. For the Ti16O32 one, instead, a 5  5  5 MonkhorstPack19 k-point mesh was adopted. 2.4. Atom Centered Gaussian DFT (All-Electron) Computational Setup. Solid-state calculations based on the atomcentered linear combination of Gaussian-type functions (LCGTF) are widely employed in studies of inorganic compounds.20 As concerns the present case, we performed periodic optimizations of both pure and defective TiO2 anatase with the CRYSTAL06 code adopting triple-ζ basis sets previously optimized for calculations of inorganic solids21 and both spin-polarized B3-LYP20d,22 and PBE013,23 Hamiltonians. To speed up convergence, the frozen-core 1765

dx.doi.org/10.1021/jp2097636 |J. Phys. Chem. C 2012, 116, 1764–1771

The Journal of Physical Chemistry C

ARTICLE

Figure 2. Fitted magnitude of the forward EXAFS Fourier transform (left) and corresponding fitted k2-weighted Ti K-edge EXAFS spectra (right) of six TiO2 samples with various initial N/Ti molar ratios. Blue curves, experimental data; red curves, least-squares fitting; green curves, least-square estimate of the background.

StuttgartDresden ECP10MDF pseudopotential24 was applied to Ti atoms in all the Nint interstitial solid state optimizations. Eventually, the starting electron populations on Ti, O, and N were always chosen so that the crystal cell was always electrically neutral. A more complete discussion concerning the all-electron LCGTF method employed in this work, inclusive of relevant technical computational details, can be found in the Supporting Information.

3. RESULTS 3.1. EXAFS. EXAFS spectroscopy is a unique tool to probe the local, average structure around the absorbing atom. 25 This technique is sensible to short-range distortions offering complementary information with respect to X-ray diffractometry that retrieves the long-range structure. Therefore,

EXAFS appears to be particularly suitable to highlight indirect effects of disordered nitrogen atom dopants on the local structure around Ti atoms. Actually, XAS-based techniques were employed, both recently and in the past, for the analysis of titania-based materials, both doped26 and undoped.27 In particular, Belver et al.26d analyzed several N-containing TiO2 crystalline samples by using EXAFS among other methods. They observed a shrinking of the average TiO bond distances with respect to the reference material and this was associated to the increment of the 5-coordinate titanium fraction, i.e., with an increase of the concentration of the oxygen vacancies. No information was given about the possible location of the dopant atoms; they could not distinguish between axial and equatorial distances of the Ti centered octahedron. In this article, we present a more detailed representation of the 1766

dx.doi.org/10.1021/jp2097636 |J. Phys. Chem. C 2012, 116, 1764–1771

The Journal of Physical Chemistry C

ARTICLE

Figure 3. (a)Average axial (blue triangles) and equatorial (red circles) TiO distances within the first coordination shell of titanium as a function of the N-doping extent. The error bars correspond to 1 estimated standard deviation; (b) Crystallographic tetragonal unit cell of the TiO2 anatase structure, with the distorted octahedral coordination of oxygen atoms (red spheres) around the Ti ions (gray spheres) highlighted.

Ti neighboring environment, including N location at low dopant level. Figure 2 summarizes, both in the real and k spaces, the overall matching among data and least-squares model functions. However, Figure 3a shows the final estimates for the average axial and equatorial first shell TiO distances in the anatase structure (scheme in Figure 3b). The numerical entries corresponding with data reported in Figure 3a can be found in Table S1 within the Supporting Information. Even if the estimated standard deviations are quite high with respect to the observed range of variation, especially for ÆTiOæaxial distances, a clear trend appears. In particular, (i) ÆTiOæequatorial remains basically constant throughout the overall doping range, whereas (ii) ÆTiOæaxial undergoes a ∼0.1 Å reduction when the doping nitrogen exceeds the nominal 0.1 N/Ti molar ratio. Then, it remains constant up to the maximum dopant concentration. In other words, it is possible to distinguish between a low N/Ti doping regime from a high one, by inspecting the average distances of the Ti first coordination shell. The first doping region goes from pure TiO2 to 0.10 N/Ti molar ratio: here, the ligand geometry around Ti is invariant with respect to the pure sample. The other relevant region appears when 0.4 e N/Ti nominal e0.5, and it is characterized by a significant reduction of the average TiO axial distances with respect to the low doping regime. In details, the average value of ÆdTiOæaxial among 0 and 0.1 N/Ti nominal molar ratio (first three columns of Table S1, Supporting Information) is as large as 2.003(13) Å, to be compared with the axial distance evaluated by averaging the entries among 0.4 and 0.5 N/Ti doping concentrations (last two columns in Table S1, Supporting Information), 1.919(3) Å. The entry in Table S1 of the Supporting Information corresponding to the 0.2 N/Ti ratio appears to be somewhat intermediate between these two limit situations, with ÆdTiOæaxial = 1.97(3) Å. It should be stressed that the values reported in Figure 3a and Table S1 (Supporting Information) are not directly comparable with crystallographic results, since EXAFS, as said before, is sensible to the local environment around titanium, and it is not able to provide information on the long-range structure. In this context, we use this technique to estimate the relative, average degree of distortion with respect to the undoped nanostructured material. As we take into account the presence of brookite in the fitting procedure of the experimental signals, such local distortions of the anatase structure should be indirectly related to the

Figure 4. Nearest titanium neighbors: (a) and (b) for substitutional N doping, (c) and (d) for interstitial ones.

amount of lattice nitrogen, and most importantly, to the preferential doping type (substitutional or interstitial) of N atoms for each specific doping concentration. 3.2. Titania DFT Modeling. In order to have a fair comparison between EXAFS results and DFT calculations, first principles periodic boundary supercell calculations were arranged in a way to reproduce the whole N/Ti molar range of the experimental samples. A realistic supercell arrangement for DFT calculations ranged from 48 up to 162 atoms. In a symbiotic EXAFS and DFT approach, we had to switch from the usual point of view based on a primitive or crystallographic cell to a less common one, where the fundamental unit is represented by Ti (the scatterer) and its nearest neighbors (oxygen and nitrogen). In other words, from a long-range to a local point of view. In anatase, coordinated nonmetals make up a Ti-centered octahedron (Figure 3b). Possible substitutional and interstitial N-doping of the octahedron are represented in Figure 4. In order to put theory on a solid footing, we performed DFT calculations with PAW as well as LCGTF bases. The calculated distances between the Ti and neighbors in stoichiometric anatase are reported in the second column of Table 1. The axial oxygen atoms are at a 2.00 Å distance, while all the equatorial ones are at 1.94 Å, and the agreement between the different DFT approaches is within three digits. Then, we simulated how the shape of the octahedron changes under both substitutional and interstitial doping. The geometries are reported in Figure 4. Even if panels a and b, as well as panels c and d, geometries look different for a single octahedron, they are potentially the same in a supercell periodic framework. Actually, the substitution in panel a is equivalent to that in panel b (and panel c to panel d) when the neighboring Ti atom is taken as the octahedron center. Figure 4 shows that substitutional doping preserves the octahedral symmetry and the bond distances of the original anatase phase. The distances are reported in Table 1 (3rd column) for the equatorial oxygen atom substitution and (4th column) for the axial one, at three different levels of DFT calculations. In all cases, the Ti neighbor-atom distances are almost unchanged: equatorial ones are on average equal to 1.94 Å, and the axial ones are on average equal to 2.03 Å. The picture changes in the case of interstitial doping. On panel c of Figure 4, the interstitial doping 1767

dx.doi.org/10.1021/jp2097636 |J. Phys. Chem. C 2012, 116, 1764–1771

The Journal of Physical Chemistry C

ARTICLE

Table 1. Ti Nearest Neighbors DFT Distances N@O equatorial

N@O axial

Nint equatorial

Nint axial

bond

undoped

(panel a in Figure 4)

(panel b in Figure 4)

(panel c in Figure 4)

(panel d in Figure 4)

TiO1

2.00/2.00/1.98a

2.01/2.04/2.02

2.03/2.03/2.00

2.00/2.04/2.02

1.89/1.85/1.85

TiO2

2.00/2.00/1.98

2.01/2.01/1.99

1.97/1.96/1.94

2.37/2.38/2.35

TiO3

1.94/1.95/1.94

TiO4

1.94/1.95/1.94

TiO5 TiO6 TiN NO a

1.93/1.94/1.93

1.90/1.91/1.90

1.94/1.96/1.95

1.94/1.94/1.92

1.93/1.94/1.93

2.12/2.13/2.11

1.94/1.96/1.94

1.94/1.95/1.94

1.93/1.94/1.93

1.94/1.95/1.94

1.98/2.04/2.02

1.94/1.95/1.94

1.94/1.95/1.94

1.93/1.94/1.93

1.95/1.95/1.94

1.99/1.99/1.98

1.94/1.95/1.94

1.96/1.98/1.96

2.08/2.10/2.09

2.08/2.06/2.04 1.34/1.37/1.36

2.33/2.48/2.42 1.34/1.37/1.36

First PAW calculations with PBE; second and third for LCGTF with B3-LYP and PBE0 Hamiltonians, respectively.

Figure 5. Nearest Titanium neighbors in the presence of an oxygen vacancy: (a) and (b) respectively for axial and equatorial substitutional N doping; (c) and (d) respectively for the equatorial and axial interstitial N doping. Geometries (e) and (f) respectively for undoped axial and equatorial vacancy.

occurred at the equatorial region. The octahedral symmetry is broken, and the distances, except one, are increased. The exception is due to the O3 atom, which is trans-equatorial with respect to O4 and quite distant from the interstitial nitrogen. Therefore, it is less affected by changes of the bonding network around titanium. In the case of interstitial doping (geometries reported in panel d of Figure 4), axial distances are very much elongated, while equatorial ones are left almost unchanged. This geometry could resemble the undoped one. However, one should not forget that geometry reported in panel d of Figure 4 is the same as that in panel c, if taken from another point of view. In other words, both geometries are present in the case of interstitial N-doping. The octahedron is broken for the geometry in panel d of Figure 4 as well. Moreover, Table 1 reports the comparison between different DFT functionals. Despite the well-known differences among the LCGTF and PAW approaches, it is important to note that the LCGTF perfectly reproduces the PAW-DFT and the experimental findings on the coordination geometry around Ti (Table 1). 3.3. Defected Titania DFT Modeling. For a more realistic modeling, one needs to consider the effect of the presence of vacancies in the local geometrical arrangement. It is then our

purpose to see how the original octahedral geometry changes by placing a vacancy at one of the oxygen atoms both with or without the presence of N-dopant. The arrangement where both O-vacancy and dopant are located in the same octahedron is quite rare, because the actual dopant bulk concentration has been estimated to be very small. Nevertheless, we cannot exclude such an eventuality, as in nanostructured materials the surface-to-bulk atom ratio is significantly higher with respect to conventional bulk compounds. Moreover, the eventuality that some kind of defect clustering26d could take place within the lattice cannot be a priori neglected. Figure 5 reports the PAW-DFT equilibrium geometries for the concomitant occurrence of N-doping and O-vacancy around the same Ti ion, as well as for only the O-vacancy. In geometries a and b, the N-dopant atom is located, respectively, at equatorial and axial substitutional locations, while the oxygen vacancy is, respectively, axial and equatorial. In geometry a, the original octahedral shape is preserved, while in b, the TiN axis is tilted with respect to the original octahedral axis in such a way to compensate for the equatorial oxygen vacancy. Geometries c and d represent the interstitial kind of doping: In c, the dopant was originally located at the equator and the vacancy at the axes, and in d, viceversa. From Figure 5c, one can see how, during the 1768

dx.doi.org/10.1021/jp2097636 |J. Phys. Chem. C 2012, 116, 1764–1771

The Journal of Physical Chemistry C

ARTICLE

Table 2. Ti Nearest Neighbors PAW-DFT Distances for Oxygen Defected Octahedron undoped axial

undoped equatorial

N@O equatorial

N@O axial

Nint equatorial

Nint axial

vacancy

vacancy

(panel a in Figure 5)

(panel b in Figure 5)

(panel c in Figure 5)

(panel d in Figure 5)

bond TiO1

Xa

1.92

X

TiO2

1.86

1.96

1.88

TiO3

1.93

X

TiO4

1.92

1.81

1.90

TiO5

1.90

1.92

TiO6

1.90

1.91

TiN NO a

2.03

2.03

1.88

X

X

1.93

1.93

1.93

1.93

1.93

1.93

1.94

1.81

1.81

1.94

1.94

X

1.81

1.94

1.90

1.90

1.90 1.46

2.24 1.46

X indicates the presence of an oxygen vacancy.

Table 3. Averaged Ti Nearest Neighbors (Including N) from PAW-DFT Distances in Tables 1 and 2 distances (Å)

undoped

N@O

Nint

undoped and O-vacancy

N@O and O-vacancy

Nint and O-vacancy

ÆTiO(N)æequatorial

1.94

1.94

1.98

1.90

1.91

1.90

ÆTiO(N)æaxial

2.00

2.03

2.11

1.91

1.94

2.05

DFT optimization procedure, the NO moiety (see Figure 4c) migrated to compensate the O-vacancy by placing the oxygen next to the axial vacancy. Similarly, in Figure 5d, the original axial NO moiety (see Figure 4d) has arranged in a way to place the oxygen next to the equatorial vacancy. In geometries e and f, no dopant is present, and the vacancy is located, respectively, in axial and equatorial location. For a better comparison, the distances between the central Ti atoms and the nearest neighbor atoms are reported for the undoped (Figure 5ef) and doped (Figure 5ad) cases in Table 2. For the undoped defected octahedron (Figure 5ef and second and third columns of Table 2), the axial distances are reduced and are quite similar to the equatorial ones, which are left unchanged. In these cases, one would observe a single set of distances, located at the equatorial anatase value. On the fourth column, the distances of the geometry in Figure 5a for a substitutional doping at the equator are reported: The equatorial distances are left unchanged, while the axial one is greatly reduced. On the fifth column in Table 2, one can see that the axial distance is left invariant and that the average equatorial one is slightly shorter because the dopant is substitutionally located at the equator and an O-vacancy at the axis. Thus, in the case shown in Figure 5b, the two sets of distances of the original anatase are almost unchanged. When the interstitial doping occurs, one can observe shorter equatorial and unchanged axial distances when the vacancy is placed on the axes, as reported in the sixth column of Table 2. Finally, in the last column, the distances of Figure 5d are almost identical with the ones of the last column in Table 1: the equatorial distances are unchanged, while the axial one are elongated in the case of TiN distance and shortened for TiO. Now, one may wonder if by placing both N doping and oxygen vacancy at the equators, the picture would have changed. This is indeed a rare configuration. Specifically, if both the N substitutional doping and oxygen vacancy had been equatorial, this would correspond to originally have two nearest neighbor vacancies. Instead, if the interstitial doping and oxygen vacancy would have been equatorial, then the nitrogen would migrate into the vacancy (as can be seen from DFT calculations), and the

arrangement would be equivalent to a substitutional equatorial doping.

4. DISCUSSION 4.1. EXAFS vs DFT Results. When trying to compare the EXAFS results with the DFT calculations, we should remember that EXAFS signals contain an average picture of the specimen, accounting for both doped and undoped regions. Thus, for a more realistic modeling, we report in Table 3 the average equatorial versus axial distances values calculated from the PAW-PBE results in Tables 1 and 2 since EXAFS results in Figure 3 clearly allow us to make this distinction. For example, under the column labeled N@O, the average of all PBE equatorial Ti-nearest neighbor atoms taken from the third and fourth columns of Table 1 is reported. This would be the EXAFS result if all Ti centers experienced a nearest-neighbor substitutional doping. Clearly, this is never the case, and the EXAFS signal would presumably be an average (with different weights) between all possible scenarios reported in Table 3. The point is to see how the weight of each scenario changes with N content. Table 3, third column, reinstates that the original octahedral distances are invariant when substitutional doping is predominant. Instead, if the interstitial doping prevailed, one should observe elongated axial distances (see fourth column). The creation of O-vacancies strongly affects the original octahedral geometry since distances become almost degenerate, as reported in the fifth column. Then, if we take into consideration that doping happens to be within the same octahedron where an O-vacancy has been created, then the equatorial and axial distances are about the same when the doping is substitutional (sixth column). Instead, last column, axial distances are elongated and equatorial ones shortened when it is the interstitial to share the same Ti center as the O-vacancy. As said above, a realistic picture includes all the different kind of Ti centers presented in Table 3. However, column six and seven are clearly less frequent because they refer to a doping scenario where the dopant N atom and the O vacancy are both located near the same Ti atom. Although such a situation cannot be excluded a priori 1769

dx.doi.org/10.1021/jp2097636 |J. Phys. Chem. C 2012, 116, 1764–1771

The Journal of Physical Chemistry C even if the O vacancies are few and randomly distributed, it is clear that it should become really significant only when the concentration of vacancies is very high. Therefore, columns six and seven of Table 3 should be considered as limiting scenarios, corresponding, respectively, to (i) predominant substitutional N and (ii) predominant interstitial N together with the presence, in both cases, of high vacancy concentration. Therefore, the averaging over all possible scenarios confirms that the original anatase octahedral distances are unchanged only when substitutional doping prevails. Instead, the starting geometry would not be preserved if interstitial doping became dominant or O-vacancy concentration is significantly increased with respect to the original undoped sample. EXAFS averaged distance results reported in Figure 3 and Table S1 (Supporting Information) show that there are much less Ti centers that experience a nearest-neighbor O-vacancy than a stoichiometric arrangement in the undoped sample, as reasonable. The original anatase octahedral geometry is preserved for doping below or equal to ca. 0.1 N/Ti initial molar ratio. By comparing the EXAFS results with all possible DFT scenarios in Table 3, one can safely deduce that N-doping is either superficial or bulk substitutional up to ca. 0.1 N/Ti initial molar ratio. Instead, the scenario at higher dopant concentration is more complex. The decrement of axial distance observed by EXAFS can be explained by DFT calculations by assuming an indirect N-doping effect, which increases the number of O-vacancies. According to DFT calculations, only the generation of O-vacancies can induce such an average axial distance shrinking (see the undoped and O-vacancy column of Table 3). The same calculations allow for a concomitant N-doping increase: this can be either substitutional or interstitial since both the N@O and Nint scenarios (see the third and fourth columns of Table 3) are possible when O-vacancies predominate. At this level, we cannot distinguish if the N-content is either substitutional or interstitial. As far as the contemporary occurrence of a nitrogen atom and an oxygen vacancy near the same metal center (last two columns of Table 3), we consider these possibilities quite remote (see the discussion above). In conclusion, in the light of theoretical simulations, EXAFS results point toward a picture where N-dopant atoms are present together with higher concentration of oxygen vacancies, as doping increases. An independent set of experiments on the same samples were reported by some of us elsewhere.6a There, a different experimental approach was used, and no information about the location of the N-dopant at any concentration was obtained. Specifically, in that article, the quasi-Fermi energy levels have been measured under irradiation by photovoltage. This electrochemical analysis showed that the quasi-Fermi levels are invariant under N-doping and concluded that the amount of N-dopant, that acts as an electron acceptor, is always less than the O-vacancy concentration, as the present EXAFS results show. The confirmation of the presence of O-vacancies can be supported also by XPS analyses in the Ti 2p region. Figure S2 (Supporting Information) shows the presence of Ti3+ as a shoulder of the Ti4+ main component for the case of 0.5 N/Ti initial molar ratio; the presence of Ti3+ species in the oxide external layers may suggest the occurrence O-vacancies even in the bulk of that sample. 4.2. Interstitial vs Substitutional Doping and Oxygen Vacancies Formation. As far as the N location is concerned, some authors claim that there are “some evidences for a preference for interstitial sites”3a and that interstitial doping may be first to occur under oxygen rich conditions, based on theoretical

ARTICLE

thermodynamic considerations.1a,5 We reproduced these results by calculating the N-doped titania formation energy starting from stoichiometric titania. The gas phase chemical potential for nitrogen and oxygen were also calculated using DFT. At both levels of theory, i.e., PAW and LCTGF, the interstitial dopant location is favored under rich oxygen conditions. This is indeed the present case since nanoparticles were calcined under O2(g) stream. These thermodynamics considerations have been used as an explanation why interstitial doping is favored with respect to the substitutional one.5 However, one should not rush to conclusions. We think that during our synthetic procedure, one cannot consider the reactant as perfect or partially defected crystals. Actually, inclusion of nitrogen is concomitant with TiO2 synthesis, which already starts during the gel formation at room temperature. Subsequent calcination at T = 400 C is not able to reinstate a full thermodynamic equilibrium to a perfect crystal and to validate the above energetic considerations. Moreover, these calculations take the classical crystal geometries (i.e., the bottom of the optimization well) bulk titania crystal as the thermodynamic reference, while this is obviously not the case for real synthesized systems. In our opinion, the reasons why the substitutional doping should be favored, even under oxygen rich conditions,1a,3a may lay on the kinetic barrier involved in breaking the octahedral shape. In other words, interstitial doping pays a higher kinetic price than the substitutional one, and it may occur before the interstitial one, as contemplated by the EXAFS data up to ca 0.1 N/Ti nominal molar ratio. A set of DFT calculations on vacancies, interstitial, and substitutional nitrogen migration has shown that kinetic barriers change drastically depending on the bulk setup.28 As far as the oxygen vacancy creation is concerned, these are favored under N doping likely because of the electron transfer from an F center (or Ti3+ state) to a midgap N electronic state.6a,29,30

5. CONCLUSIONS In this article, a joint EXAFS and DFT investigation is presented in order to determine the nature of the N-dopant inclusion in the bulk titania. First, the EXAFS structural models have given a reasonable estimate of the axial and equatorial nearest Ti neighbor distances in the undoped sample. These fittings have showed that for a nominal molar ratio lower than ca. 0.1 N/Ti, the average coordination geometry of Ti is left unchanged upon doping. Then, the average TiO axial distances are gradually but significantly reduced and the equatorial ones left invariant, as the doping is increased up to 0.5 N/Ti nominal molar ratio. DFT calculations, both at the level of PAW and LGCTF bases, have been employed to describe the different scenarios that the Ti centers may experience. This synergistic approach allows us to conclude that (i) at low dopant concentrations (0 < nominal molar ratio N/Ti < 0.1) substitutional doping is preferred; and (ii) at higher dopant concentrations, a significant increase in the number of oxygen vacancies occurs together with substitutional and/or interstitial doping. At least, this is the case for the synthetic route adopted here. Interestingly, these findings agree well with the DFT electronic structure results according to the O-vacancy extra electrons that are partially accommodated by empty N electronic orbitals.6a Purely thermodynamic explanations, however, are biased by the choice of an ideal bulk crystal at its classical geometry as a reference and certainly do not take into account the kinetic aspects of the defect 1770

dx.doi.org/10.1021/jp2097636 |J. Phys. Chem. C 2012, 116, 1764–1771

The Journal of Physical Chemistry C formation. Therefore, great care should be taken in transferring tout court results derived from thermodynamic theoretical models to real systems. In conclusion, our joint EXAFS and DFT investigation has unequivocally identified the Nb locations at low dopant concentrations and clearly proved the generation of oxygen vacancies by N doping. Last but not least, the procedure presented here can be used for the characterization of many doped materials.

’ ASSOCIATED CONTENT

bS

Supporting Information. Normalized Ti K-edge absorption curves in the relevant edge and postedge regions for six TiO2 samples with various N/Ti nominal molar ratios; XANES and XPS combined analysis; details of the EXAFS data refinement; atom centred Gaussian DFT (all-electron) computational setup; atom centred Gaussians DFT (pseudopotential) computational setup; TiO2 polymorph content. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: +390250314258. Fax: +390250314258. E-mail: michele. [email protected] (M.C.); [email protected] (L.L.P.).

’ ACKNOWLEDGMENT M.C. acknowledges the Universita degli Studi di Milano for starting grant “5 per mille”. The European Synchrotron Radiation Facility is acknowledged for the provision of beamtime on ID31 and BM01B. Dr A. Fitch and V. van Beek are acknowledged for their experimental support. We also wish to thank one anonymous reviewer for thoughtful comments on the EXAFS results. ’ REFERENCES (1) (a) Emeline, A. V.; Kuznetsov, V. N.; Rybchuk, V. K.; Serpone, N. Int. J. Photoenergy 2008, 2008, 2583941–25839419. (b) Roy, P.; Berger, S.; Schmuki, P. Angew. Chem., Int. Ed. 2011, 50, 2904–2939. (c) Ardizzone, S.; Cappelletti, G.; Meroni, D.; Spadavecchia, F. Chem. Commun. 2011, 47, 640–2642. (d) Meroni, D.; Ardizzone, S.; Cappelletti, G.; Oliva, C.; Ceotto, M.; Peolman, D.; Peolman, H. Catal. Today 2011, 161, 169–174. (2) Asahi, R.; Morikawa, T.; Ohwaki, T.; Aoki, K.; Taga, Y. Science 2001, 293, 269–271. (3) (a) Di Valentin, C.; Finazzi, E.; Pacchioni, G.; Selloni, A.; Livraghi, S.; Paganini, M. C.; Giamello, E. Chem. Phys. 2007, 339, 44–56. (b) Graciani, J.; Nambu, A.; Evans, J.; Rodriguez, J. A.; Sanz, J. F. J. Am. Chem. Soc. 2008, 130, 12056–12063. (c) Zhang, H.; Chen, G.; Bahnemann, D. W. J. Mater. Chem. 2009, 19, 5089–5121. (d) Zhang, J.; Wu, Y.; Xing, M.; Leghari, S. A. K.; Sajjad, S. Energy Environ. Sci. 2010, 3, 715–726. (4) Othani, B. J. Photochem. Photobiol. C 2010, 11, 157–178. (5) Di Valentin, C.; Pacchioni, G.; Selloni, A.; Livraghi, S.; Giamello, E. J. Phys. Chem. B 2005, 109, 11414–11419. (6) (a) Spadavecchia, F.; Cappelletti, G.; Ardizzone, S.; Ceotto, M.; Falciola, L. J. Phys. Chem. C 2011, 115, 6381–6391. (b) Meroni, D.; Ardizzone, S.; Cappelletti, C.; Oliva, C.; Ceotto, M.; Poelman, D.; Poelman, H. Catal. Today 2011, 161, 169–174. (7) Spadavecchia, F.; Cappelletti, G.; Ardizzone, S.; Bianchi, C. L.; Cappelli, S.; Oliva, C.; Scardi, P.; Leoni, M.; Fermo, P. Appl. Catal., B 2010, 96, 314–322. (8) Oropeza, F. E.; Harmer, J.; Egdell, R. G.; Palgrave, R. G. Phys. Chem. Chem. Phys. 2010, 12, 960–969.

ARTICLE

(9) Fernandez-Garcıa, M.; Martınez-Arias, A.; Hanson, J. C.; Rodriguez, J. A. Chem. Rev. 2004, 104, 4063–4104. (10) Ravel, B.; Newville, M. J. Synchrotron Radiat. 2005, 12, 537–541. (11) Newville, M. J. Synchrotron Radiat. 2001, 8, 322–324. (12) (a) Kresse, G.; Hafner, J. J. Phys. Rev. B 1993, 47, 558–561. (b) Kresse, G.; Furthm€uller, J. Phys. Rev. B 1996, 54, 11169–11186. (13) Perdew, J. P.; Burk, K.; Ernzerhof, M. Phys. Rev. Lett. 1996, 77, 3865–3868. (14) Perdew, J. P.; Wang, Y. Phys. Rev. B 1992, 45, 13244–13249. (15) (a) Harris, J. Phys. Rev. B 1985, 31, 1770–1779. (b) Foulkes, W. M. C.; Haydock, T. Phys. Rev. B 1989, 39, 12520–12536. (16) (a) Teter, M. P.; Payne, M. C.; Allan, D. C. Phys. Rev. B 1989, 40, 12255–12263. (b) Bylander, D. M.; Kleinman, L.; Lee, S. Phys Rev. B 1990, 42, 1394–1403. (17) Dudarev, S. L.; Botton, C. A.; Savarsov, S. Y.; Hunphreys, C. J.; Sutton, A. P. Phys. Rev. B 1998, 57, 1505–1509. (18) (a) Finazzi, E.; Di Valentin, C.; Pacchioni, G.; Selloni, A. J. Chem. Phys. 2008, 129, 154113/1–154113/9. (b) Mattioli, G.; Filippone, F.; Alippi, P.; Amore Bonapasta, A. Phys. Rev. B 2008, 78, 241201/1–241201/ 4. (c) Morgan, B. J.; Watson, G. W. Surf. Sci. 2007, 601, 5034–5041. (d) Cheng, H.; Selloni, A. J. Chem. Phys. 2009, 131, 0547031–1/054703110. (e) Stausholm-Moeller, J.; Kristoffersen, H. H.; Hinnemann, B.; Madsen, G. K. H.; Hammer, B. J. Chem. Phys. 2010, 133, 144708/1–144708/8. (19) Monkhorst, H. J.; Pack, J. D. Phys. Rev. B 1976, 13, 5188–5192. (20) (a) Meyer, A.; Pascale, F.; Zicovich-Wilson, C. M.; Dovesi, R. Int. J. Quantum Chem. 2010, 110, 338–351. (b) Meyer, A.; Catti, M.; Dovesi, R. J. Phys.: Condens. Matter 2010, 22, 146008/1–1460008/8. (c) Lo Presti, L.; Destro, R. J. Chem. Phys. 2008, 128, 044710/1–044710/9. (d) Dovesi, R.; Saunders, V. R.; Roetti, C.; Orlando, R.; ZicovichWilson, C. M.; Pascale, F.; Civalleri, B.; Doll, K.; Harrison, N. M.; Bush, I. J.; D’Arco, P.; Llunell, M. CRYSTAL06 User’s Manual; University of Torino: Torino, Italy, 2006. (21) (a) Cora, F. Mol. Phys. 2005, 103, 2483–2496. (b) Pandey, R.; Jaffe, J. E.; Harrison, N. M. J. Phys. Chem. Solids 1994, 55, 1357–1361.(c) Towler, M. CRYSTAL Resources Page. http://www.tcm.phy.cam.ac.uk/ ∼mdt26/crystal.html. (22) (a) Becke, A. D. J. Chem. Phys. 1993, 98, 5648–5652. (b) Volsko, S. H.; Wilk, L.; Nusair, M. Can. J. Phys. 1980, 58, 1200–1211. (23) Adamo, C.; Barone, V. J. Chem. Phys. 1999, 110, 6158–6170. (24) (a) Dolg, M.; Wedig, U.; Stoll, H.; Preuss, H. J. Chem. Phys. € nderung, L. Pseudopotentials, ECPs. http:// 1987, 86, 866–872.(b) A www.theochem.uni-stuttgart.de/pseudopotentials/index.en.html. (25) Bunker, G. Introduction to XAFS. A Practical Guide to X-ray Absorption Fine Structure Spectroscopy; Cambridge University Press: Cambridge, UK, 2010. (26) (a) Bard, A.; Faulkner, L. R. Electrochemical Methods. Fundamental and Applications, 2nd ed.; John Wiley & Sons, Inc.: New York, 2001; (b) Ward, M. D.; White, J. R.; Bard, A. J. J. Am. Chem. Soc. 1983, 105, 27–31. (c) Mitoraj, D.; Kisch, H. J. Phys. Chem. C 2009, 113, 20890–20895. (d) Belver, C.; Bellod, R.; Stewart, S. J.; Requejo, F. G.; Fernandez-García, M. Appl. Catal., B 2006, 65, 309–314. (e) Rodrıguez-Torres, C. E.; Cabrera, A. F.; Errico, L. A.; Adan, C.; Requejo, F. G.; Weissmann, M.; Stewart, S. J. J. Phys.: Condens. Matter 2008, 20, 135210–135218. (f) Angelome, P. C.; Andrini, L.; Fuertes, M. C.; Requejo, F. G.; Soler-Illia, G. J. A. A. C. R. Chim. 2010, 13, 256–269. (27) (a) Bandara, J.; Pradeep, U. W. Thin Solid Films 2008, 517, 952–956. (b) Luca, V.; Djajanti, S.; Howe, R. F. J. Phys. Chem. B 1998, 102, 10650–10657. (c) Chen, L. X.; Rajh, T.; Wang, Z.; Thurnauer, M. C. J. Phys. Chem. B 1997, 101, 10688–10697. (d) Angelome, P. C.; Andrini, L.; Calvo, M. E.; Requejo, F. G.; Bilmes, S. A.; Soler-Illia, G. J. A. A. J. Phys. Chem C 2007, 111, 10886–10893. (28) Tsetseris, L. Phys. Rev. B 2010, 81, 165205/1165205/7. (29) (a) Yang, K.; Dai, Y.; Huang, B.; Feng, Y. P. Phys. Rev. B 2010, 81, 033202/1–033202/4. (b) Kuznetsov, V. N.; Serpone, N. J. Phys. Chem. C 2009, 113, 15110–15123. (30) Serpone, N. J. Phys. Chem. B 2006, 110, 24287–24293.

1771

dx.doi.org/10.1021/jp2097636 |J. Phys. Chem. C 2012, 116, 1764–1771