1. Introduction

As a first-generation material, pristine has served well in the photoassisted (often dubiously referred to as photocatalytic) disposition of contaminants in aqueous and atmospheric ecosystems. The science that underlies heterogeneous photocatalysis has shown that the lowest energy level of the bottom of the conduction band (CB) of is a measure of the reduction potential of the photogenerated electrons, whereas the higher energy level of the valence band (VB) is a measure of the oxidation potential of photogenerated holes. pH-dependent flatband potentials, , of the CB and VB bands of this metal oxide determine the energy of electrons and holes at the interface. Accordingly, reductive and oxidative processes of couples with redox potentials more positive and more negative than the of CB and VB, respectively, can be driven by surface trapped electrons () and holes () that are poised to engage in various processes, the most important of which are photoreductions and photooxidations. An important issue regards the notion that once photogenerated, and tend to recombine somewhat efficiently and rapidly relative to an otherwise slow redox chemistry at the surface. An additional, no less important issue is that absorbs a relatively small fraction (ca. 3–5%) of the solar radiation reaching the Earth's surface.

Of the two important polymorphs of , anatase begins to absorb UV light around 387 nm (band gap energy, ), whereas the absorption onset of rutile occurs around 413 nm () increasing sharply to shorter wavelengths. Accordingly, in the late 1980s studies began to develop the next generation of titanium dioxides [1] that could absorb and make use of both UV (290–400 nm) and visible (400–700 nm) radiation to enhance process efficiencies. To achieve this feat required that the absorption edge of be shifted to longer wavelengths (>400 nm). One way to accomplish this necessitated photosensitizing either with suitable dyes that unfortunately led to their own destruction, or with suitable metal-ion dopants that sometimes act as recombination centers of and ; metal-ion dopants are often ineffective in aiding surface redox reactions, as appears to be the case when metal doping is achieved by wet impregnation [1]. However, metal-ion implantation methods have produced metal-doped specimens that enhance photoinduced surface processes even in the visible-light spectral region [2], where wet chemical methods failed.

First reports of anion-doped , began to appear in the early 1990s, although Sato [3] had earlier hinted at a N-doped . The 2001 study of Asahi and coworkers [4] on doping with various anions to prepare visible-light-active (VLA) N-doped s was the catalyst needed to produce second-generation materials that are photoactive over the UV and much of the visible-light region. Subsequent studies reported several other visible-light-active N-doped s, together with C-doped and S-doped . The reports by Asahi et al. [4, 5] has led to a lively debate on the causes that lead the absorption onset of to be red-shifted to the visible region. They proposed that N-doping of red-shifts the absorption edge of and increases photoactivity by a narrowing of the band gap. Carbon- and sulfur-doped displayed similar red-shifts accompanied by increased photoactivity. As we will see later, others have proposed that electronic transitions in N-doped systems activated by visible-light irradiation involve transitions from N 2p localized states to the CB of . Clearly, just as first-generation led to lively debates on the fundamental science that underpins -assisted photoredox surface processes following photo-activation, so are the second-generation VLA s generating enthusiastic discussions on the root cause that shifts the absorption onset to longer wavelengths. Three recent reviews have summarized some of the facets of first (undoped) and second generation (doped) titanias. [6–8].

This review article focuses briefly (a) on some preparative methods of visible-light-active N-doped materials and their XPS spectroscopic features, (b) on their visible absorption spectra that display the red-shift of the absorption edges, to terminate with (c) a brief visit into the lively debate concerning band gap narrowing.

2. Nitrogen-Doped TiO₂s– Syntheses and Characterization

Asahi et al. [4] initially set three requirements to achieve visible-light-activity for : (i) doping should produce states in the band gap of that absorb visible light; (ii) the CB minimum and the dopant states of doped should be as high as or higher than the level to ensure photoreductive activity; and (iii) the intragap states should overlap sufficiently with the band states of to transfer photoexcited carriers to reactive sites at the surface within their lifetime. Metal dopants were undesirable because they did not meet conditions (ii) and (iii) as they produce localized d states deep in the band gap of and tend to act more as recombination centers. Calculations of density of states (DOS) of substitutional doping with several nonmetals (C, N, F, P, or S) into O sites in anatase by the full-potential linearized augmented plane-wave (FLAPW) formalism in the framework of the local density approximation (LDA) led Asahi et al. [4, 5] to choose N since the 2p states apparently contribute to band gap narrowing through mixing with O 2p states in the valence band. On the other hand, Yates and coworkers [9] classified the methods of synthesizing N-doped s into (i) modification of existing by ion bombardment, (ii) modification of existing in powdered form, film, and single crystal, or else modify TiN by gas phase chemical impregnation, and (iii) grow TiO_2-xN_x (crystals) from liquid or gaseous precursors.

Early on, Sato [3] had noted that calcination of in the presence of (or aqueous ) led to photosensitization of when exposed to visible-light radiation. The powdered samples were deduced to be -doped with the impurity acting as the sensitizer. Noda et al. [10] reported a yellow-colored anatase powder obtained from aqueous hydrazine and solutions, and deduced that visible-light absorption was due to the presence of oxygen vacancies s.

In their 2001 seminal report, Asahi et al. [4] prepared crystalline TiO_2-xN_x films by sputtering a target in a gas mixture followed by annealing at 550 C in a atmosphere. The yellowish TiO_2-xN_x films absorbed light below 500 nm. X-ray photoelectron N 1s spectra (XPS) of the N-doped showed bands at 402, 400 and 396 eV; the undoped film showed no 396 eV band. The latter was assigned to atomic - in TiO_2-xN_x, whereas the 402-eV and 400-eV bands were attributed to molecularly chemisorbed dinitrogen - [11]. Powdered samples prepared with as the source of N followed by calcination at 550–600 C produced a TiO_2-xN_x systems that showed XPS peaks at 396 eV; these systems were photoactive toward the decomposition of methylene blue (optimal loading, at.% N). The sites for photoactivity under visible-light irradiation were those when N substitutionally replaced O, that is, sites associated with atomic - at 396 eV [4]. Lee et al. [12] fabricated N-doped anatase films by MOCVD using Ti(i-PrO)₄ and at 420 C; XPS Ti 2p spectra showed N was incorporated into the lattice to form Ti–N bonds. Hydrolysis of with in dry air at 400 C produced a visible-light-active anatase ( < 550 nm) [13]; however, XPS spectra showed only trace amounts of N, with visible-light response due to an oxygen-deficient stoichiometry.

Pale yellow, yellow, and dark green TiO_2-xN_x () powdered samples can be prepared by annealing anatase powder (ST-01) in a flow of at 550, 575, and 600 C, respectively [14]. XRD patterns indicated that the samples retain the anatase structure; no phase was present. The XPS peak at 396 eV confirmed substitutional doping of sites yielding O–Ti–NO bonds. Noticeable shifts of the absorption edge into the visible spectral region were evident for TiO_2-xN_x. The feature at > 550 nm was attributed to since decomposes into and at ca. 550 C, and reduces under these conditions. The band gap energy remained at 3.2 eV. Mineralization of isopropanol to with -light and visible-light radiations resulted in different quantum yields, suggesting that -doping forms a narrow 2p band above the valence band of . Note that band gap narrowing in TiO_2-xN_x, as inferred by Asahi et al. [4], would have required identical quantum yields. On irradiating with visible light, the quantum yields decreased with increase in x of the dopant because of the increase in oxygen vacancies, s, with increase of x in TiO_2-xN_x. In this case, s act as recombination centers for and . Under irradiation, the quantum yields also decreased with increase in x, suggesting that the doping sites also act as recombination centers.

Nanocrystalline porous -doped thin films, prepared by introducing N into anatase by means of DC magnetron sputtering in -containing plasma [15], displays new spectral features in the spectral range 410 < < 535 nm at low concentrations owing to excitation of to unoccupied states from local states located slightly above the edge. -doping had no effect on the conduction band edge. Band gap narrowing was deemed somewhat questionable by these authors [15]. Despite the intense recombination of charge carriers caused by -doping, the new band gap states created by -doping improved the visible-light photoresponse at the expense of some losses of the response.

Sakthivel and Kisch [16] prepared yellow -doped anatase with various loadings by hydrolysis of with a -containing base aqueous , or followed by calcination at 400 C. XPS spectra showed only a broad signal at (but no 396-eV peak) attributed to the hyponitrite () species that was confirmed by infrared spectral techniques. No changes in the valence band edge occurred on -doping, despite the red-shift of the absorption edge to . Contrary to the inference by Lindgren et al. [15], however, photoelectrochemical results [16] showed a slight change in the electrochemical potentials of the CB of (see Figure 1) for three of the specimens. According to the authors [16], -doping led only to a “modest band gap narrowing.”

Figure 1: Electrochemical potentials (versus NHE) of band edges of three -doped s: (a) , (b) TiO₂–N/1, (c) TiO₂–N/2, (d) TiO₂–N/3. Shaded areas denote surface states; the oxygen redox potential at pH 7 is also shown. Reproduced with permission from [16]. Copyright Wiley-VCH Verlag GmbH & Co. KGaA.

High-energy ball milling of P-25 (ca. 80% anatase and rutile) with various quantities of hexamethylenetetramine (HMT) at near-ambient temperatures yields yellowish N-doped rutile [17], which subsequent to calcination in air at 400 C gives an -doped product that displays absorption edges at and 550 nm and a good visible-light photoresponse toward oxidation of .

Ion-assisted electron-beam evaporation of rutile titania powder and yields crystalline anatase TiO_2-xN_x films with a considerable amount of substitutional atoms (1.8 at.%) and chemisorbed molecular [18]. XPS spectra showed peaks at 402 and 396 eV assigned to molecularly chemisorbed - and atomic -, respectively; 2p XPS spectra revealed and in the anatase film indicating that the majority of titanium in the TiO_2-xN_x film consisted of , thus confirming the results of XRD patterns. -doping caused no changes to the anatase structure as attested by Raman spectroscopy.

In a simplified synthesis, Gole and coworkers [19, 20] produced TiO_2-xN_x samples at room temperature using direct nitridation of anatase with alkylammonium salts. The samples could be tuned to respond to wavelengths up to . The method first yielded metal-oxide colloids by the controlled hydrolysis of - in aqueous/isopropanol media (pH 2; ), subsequent to which treatment in excess led to TiO_2-xN_x nanocolloids. XRD and HRTEM results demonstrated that the treated TiO_2-xN_x nanoparticles were predominantly anatase. Diffuse reflectance spectra (DRS) of the TiO_2-xN_x crystallites rose sharply at ; the corresponding DRS spectrum of nitrided TiO_2-xN_x from partially agglomerated nanoparticles rose at . XPS analysis with -ion sputtering revealed the presence of dopants at the surface and in the sub-layers of TiO_2-xN_x agglomerates ( content, 3.6–5.1 at.%). No evidence was found for conversion of the anatase structure into rutile on -doping for the initial nanocolloids and for the agglomerated gel solutions. Little if any XPS evidence of atomic - binding at 396 eV was found in any of the TiO_2-xN_x samples. Rather, the XPS results were consistent with nonstoichiometric surface-based –– bonding.

In a later related detailed XPS study of a series of -based nanometer-sized photocatalysts that included nitrogen-doped nanoparticles, Chen and Burda [21] noted a broad 1s binding energy peak of the nitrogen-doped nanoparticles that was centered at ca. 401.3 eV (see Figure 2) and extended from 397.4 eV to 403.7 eV, a range greater than the typical binding energy of 397.2 eV in titanium nitride, . These findings were attributed to the formation of an O–Ti–N structure that was suggested as the chemical entity formed during the substitutional doping process and responsible for the significant increase in photocatalytic activity of synthesized nitrogen-doped nanoparticles. Moreover, the XPS observations reported by Chen and Burda [21] were not consistent with a - entity within the nanocolloid lattice [22]. The shift in the 1s binding energy for the TiO_2-xN_x nanocolloid to higher energies relative to was likely due to a more positive oxidation state of nitrogen in relation to the 1s binding energy in (408 eV), in (398.8 eV) and adsorbed (400–401.5 eV) [23]. Attribution of the 1s peak at 400 eV to - in -doped titanias has been questioned by Chen et al. [23] in that is not chemisorbed on metal oxides such as at ambient temperature, preferring instead to assign the 400 eV peak to species consistent with the attendant heat release on the creation of these sites in the lattice.

Figure 2: X-ray photoelectron spectra of various specimens; -Degussa is nitrided P25 and TiO_2-xN_x denotes TiO_2-xN_x nanoparticles. Reproduced with permission from [21]. Copyright (2004) American Chemical Society.

Gopinath [24] recently questioned the validity of the conclusions reached by Chen and Burda [21] as not being consistent with the reported XPS observations, suggesting (among others) rather that there may have been surface contamination of the analyzed nanocolloidal -doped material as a result of atmospheric degradation. Based on earlier studies of XPS spectra of and primary alkyl/aromatic amines (398–399 eV), anionic in (396–397 eV), adsorbed and on (401 and 405 eV, resp.) and Sato and coworkers’ [25] observation of the 1s core level at 400 eV for -doped prepared by a wet method and attributed to an impurity sensitization such as , Gopinath [24] argued that the 401.3-eV peak in the – nanocolloid reported by Chen and Burda [21] was more likely due to oxidized nitrogen as in N–O–Ti–O or O–N–Ti-O, and that the high binding energy of 401.3 eV was associated with some partial positive charge on as noted by György and coworkers [26] in the nitridation of titanium for the 400–401 eV peak attributed to an O–N–Ti structure. Also questioned by Gopinath [24] were the origins of the 1s core levels at binding energies of 530 and 532 eV ascribed by Chen and Burda [21] to O-Ti-O and to N-Ti-O structural units, respectively. Gopinath [24] insisted that the 532-eV peak was likely due to surface contamination by some carbon oxide, such as , and further claimed that the large level of -doping in nanocolloids (ca. 4–8%) was rather unlikely. In an appropriate response, Burda and Gole [27] pointed out that Chen and Burda [21] used the shorthand notation to mean that , Ti, and were bound to next neighboring atoms ( or ) in the lattice and not to mean an isolated triatomic molecular species with a nonoxidized nitrogen, and were thus rather surprised by Gopinath's misinterpretation since such notation was chemically obvious even to others [28]—also see below. In fact, as clearly stated by the groups of Gole and Burda [19, 20, 22, 23] the discussion dealt with a minimal structure fragment within a doped lattice (characterized by sites), and not some species where the nitrogen was not oxidized. As well, the apparent discrepancy noted by Gopinath [24] in the XPS data reported earlier by others (see refs 13–20 in [27]) did not take into account that the data were taken under vastly different conditions across distinct physical entities that included films and undoped nanoparticles, not to mention the anatase and rutile structures. Burda and Gole [27] further noted that Gopinath’s misinterpretation of their studies originated with the misconception that , , and are isolated from their anatase environment, when in fact the XPS results reported by Chen and Burda [21] clearly demonstrated that nitrogen was partially oxidized and that the neighboring Ti was reduced relative to Ti in pure . The discussion concerning the valencies of the nitrogen had indeed attributed the 401-eV peak to an site [22, 23]; however, the precise location of the site within the titania lattice remains a subject yet to clarify. Finally, that the doping levels are as high as 4 to 8% in the -doped colloids are not unusual based on nucleophilic substitution chemistry, as pointed out previously [23].

The above reports, misinterpretations and misunderstandings, together with other XPS results described below, clearly demonstrate that XPS results do not in themselves provide unequivocal understanding of the exact nature of -doped titania specimens, not to mention the nature of other anion- and cation-doped s.

Nakamura and coworkers [29] produced anatase TiO_2-xN_x materials by two methods: (a) in the dry method anatase ST-01 was heated to 550 C in a dry flow and (b) in the wet method the - precursor was hydrolyzed in aqueous at 0 C followed by calcination at 400 C. Nitrogen-doped nanocrystalline (yellow) powders have also been synthesized by a procedure developed by Ma and coworkers [30]. Commercial anatase ST-01 heated at 500 C under a dry gas flow in the presence of a small quantity of carbon also yields -doped yellow whose XRD patterns reveal the sample to have the anatase structure, even after annealing at 500 C. XPS spectra showed peaks at 396.2, 398.3, and 400.4-eV in the 1s region, with the first two peaks attributed to chemically bound -species and to O–Ti–N linkages within the crystalline lattice, respectively, whereas the 400.4-eV peak was ascribed to molecularly chemisorbed species.

Ion implantation of atomically clean (110) surfaces in single crystals with mixtures of and ions, followed by subsequent annealing under ultrahigh-vacuum conditions, led to incorporation of N into the lattice [31]. XPS spectra revealed only the N 1s feature at 396.6 eV attributed to substitutionally bound nitride nitrogen ( ions substituted by anions). Contrary to expectations, N-doped crystals containing only nitride ions exhibited a shift in the photothreshold energy of 0.2 eV to higher (shorter wavelengths) rather than lower (longer wavelengths) energy compared to undoped (110). N-doped (110) rutile single crystals previously treated in the presence of an gas mixture at ca 600 C exhibited photoactivity at the lower photon energy of 2.4 eV, that is, 0.6 eV below the band gap energy of rutile (3.0 eV) [32]. The active dopant state of the interstitial N responsible for this effect showed a N 1s binding energy at 399.6 eV attributed to a form of nitrogen likely bound to H. This is distinctively different from the substitutional nitride state, which displays a N 1s binding energy at 396.7 eV. Apparently a co-doping effect of N and H probably enhanced the visible-light photoactivity. Doped and undoped (110) samples also showed an impurity XPS feature at 399.6 eV, which on UV treatment in air and/or -ion sputtering led to extensive depletion of the signal indicating that traces of nitrogen may have contaminated the metal-oxide surface. Such inferences by the Yates group are in stark contrast to those of Asahi et al. [4] and those of others who claimed that nitridic nitrogens that substitute ions in the lattice are the necessary dopant species for photoactivity in the visible-light spectral region.

In a later study, Thompson and Yates [33] re-emphasized that the exclusive XPS N 1s signal at 396.7 eV attributed to substitutional - in ion-implanted N-doped does not account for the decrease in the photothreshold of (110), as observed for interstitially located – bound species. Rather, they pointed out that the 0.2-eV increase in photothreshold energy of -doped systems arose from deposition of charge in the low levels of the CB (the band-filling mechanism). Although clear XPS evidence exists for the incorporation of -substitutional in -doped , there is no strong and firm evidence of any appreciable photoactivity when these doped systems are irradiated with visible light according to Yates et al. [9], a point also raised by Frach et al. [34] who further noted no improvement in visible-light activity on -doping , and by Li and coworkers [35] who reported that the nature and level of visible-light activity depended on the nitriding compound employed.

With , ethyl acetate and as precursors and as the carrier gas, Yates and coworkers [9] used an atmospheric pressure thermal CVD coater to grow thin films of -doped on glass substrates. Only three grown samples displayed the XPS 1s peak at 396 eV of atomic -substituted . XPS 1s spectra showed no evidence of the 397-eV signal typically due to the ion (), but did reveal weak signals at 400 and 402 eV probably arising from molecularly chemisorbed , or from species located at interstitial sites (399.6 eV), or from , or (400 eV), or from an oxynitride (399.3 eV) of stoichiometry equivalent to . Some of the films displayed visible spectral absorption features, but so did nominally undoped films indicating that N incorporation cannot be assumed on the basis of red-shifts of the absorption edge. Even though the -doped specimens revealed the presence of - incorporation and absorption spectral features in the visible region, they were photo-inactive under visible light irradiation, while the UV photoactivity was reduced considerably compared to films grown in the absence of . Clearly, the presence of -nitrogen alone cannot be claimed to induce visible-light activity in N-doped films, a point also raised by Mrowetz and coworkers [36] who prepared two different yellow-colored -doped samples: sample A prepared by the method of Gole et al. [19, 20] and sample H prepared by the high-temperature nitridation of commercial anatase at 550 C in a gas flow. XPS spectra of sample A surface revealed intense peaks at 399.6 and 404.5 eV in the 1s region, whereas the peak at 396 eV in the XPS spectra of sample H powders was weak and diffuse, even after -ion sputtering. Despite these observations, the -doped materials failed to catalyze the oxidation of to , and – to under visible-light illumination.

The solvothermal process using a /HMT/alcohol (methanol or ethanol) mixed solution in an autoclave at 90 C and then at 190 C yields -doped nanoparticles consisting of pure anatase (pH 9, methanol), rutile (pH 9, ethanol) and brookite (pH 1, methanol) phases, which showed good visible-light absorption and visible-light activity at > 510 nm [37]. The two-step absorption in the reported diffuse reflectance spectra (DRS) became apparent only after calcination at 400 C. The first absorption edge is related to the band structure of nondoped , whereas the second absorption edge around 520–535 nm is due to the formation of a 2p band located above the 2p valence band in the TiO_2-xN_x specimen.

Electrochemical anodization of titanium in HF/ electrolyte, followed by calcination at different temperatures (range 300–600 C) in pure , yields self-organized -doped nanotubes [38]. The initial amorphous nanotubes were converted to anatase, with some rutile present depending on the heat-treatment temperature. Absorption spectra of the TiO_2-xN_x nanotubes displayed a sub-band gap energy of (that may be referred to as an extrinsic band gap ) and the intrinsic band gap of anatase (). The 1s XPS spectrum showed two clear peaks, one at 400 eV, ascribed to molecularly chemisorbed dinitrogen (-N₂ state), and the other at 396 eV, attributed to the atomic - state.

The RF-MS deposition method produces -substituted – photoactive thin films using various mixtures as the sputtering gas and a calcined plate as the source material [39]. The absorption edge of the films red-shifted to 550 nm, with the extent of the shift depending on the concentration of substituted within the lattice (range 2.0–16.5%). The specimen with 6.0% N exhibited the highest visible-light activity in the photooxidation of isopropanol (aqueous media; irradiation ≥ 450 nm), and the photooxidation of (irradiation ≤ 550 nm). XPS and XRD measurements showed significant substitution of lattice atoms of by atoms, which Kitano and coworkers [39] suggested as playing a crucial role in the band gap narrowing of the thin films (range 2.58–2.25 eV relative to 3.2 eV, depending on X) enabling the visible-light photoresponse. In samples with , the species formed in the – samples acted as recombination centers of and and thus led to a decreased photoactivity.

Joung et al. [40] used the hydrolysis of – in anhydrous ethanol containing , followed by treatment of the resulting colloids in an stream at 400 C, 500 C and 600 C at various time intervals (5 to 60 min) to prepare visible-light-active -doped materials. The highest photoactivities were seen for samples prepared at 400 C and 600 C and calcination times of 5 and 10 min, whereas for samples prepared at 500 C the highest photoactivity was observed for a calcination time of 60 min. XRD patterns of the powders before and after -doping showed that -doping caused no changes to the anatase phase. The 1s XPS spectra of the TiO_2-xN_x samples displayed a peak at 399.95 eV that was tentatively assigned to adsorbed or to in Ti–O–N; however, no peak attributable to – bonding at 396 eV was observed. Band gap energies of -doped inferred from absorption spectra ranged from 2.92 to 3.04 eV, for samples prepared at 400 C (5 min) and at 500 C (60 min), respectively. For samples prepared at 400 C (5 and 10 min), the active species were described as being , , , , whereas for samples prepared at 500 C (60 min) the active species was identified as doped atomic ; the active species for samples prepared at 600 C (5 and 10 min) were inferred to be doped atomic along with , , , and species.

At this time, it is important to realize that visible-light photoactivity of -doped materials appears to be highly sensitive to the preparative routes, because although such materials may absorb visible light, they are nonetheless frequently inactive in photooxidations. To the extent that photogenerated charge carriers in and by themselves do not impart photoactvity and that charge carrier recombination must be muted to allow the carriers to reach the metal-oxide surface, In and coworkers [41] prepared a series of TiO_2-xN_x systems with nominal loadings from 0.2 to 1.0 wt.% involving the sequential reaction of with a small known excess of in toluene (step 1) under dry -free argon, followed by the stoichiometric reaction of the remaining with a standard solution of in dioxane (step 2). The resulting species were heat-treated in air at 400, 500, and 600 C. Tests of the TiO_2-xN_x specimens for visible-light photoactivity revealed (i) that calcination at 400 C yields a solid with pronounced absorption in the visible spectral region but yet no visible-light photoactivity, (ii) that 500 C calcination produces an effective (yellow) visible-light-active sample, and (iii) that the heat treatment at 600 C results in an inactive white material.

In two extensive reports, Belver and coworkers [42, 43] prepared and characterized a series of nanosized -doped -based materials by a reverse micelle microemulsion method using a - precursor and three sources, for example, 2-methoxyethylamine, N,N,N',N'-tetramethyl-ethylenediamine and 1,2-phenylenediamine, to produce Ti(IV) complexes in dry isopropanol under a atmosphere. Dropwise addition of the solution to the inverse microemulsion, that contained dispersed in n-heptane and Triton X-100 as the surfactant with hexanol as the cosurfactant, produced materials that were subsequently calcined at 200 C and then at 450 C. A XANES examination confirmed the anatase structure of TiO_2-xN_x. XANES spectra also showed a lack of correlation between the number of oxygen vacancies (s) and the content in the samples. Above a certain limit, the association of point defects, such as s and/or the presence of nonpoint extended defects, was detrimental to photoactivity. The distribution of defects and the nature of defects present in the -doped samples were examined in a joint XANES/EXAFS investigation, which revealed that the distribution of defects was not simply related to the oxygen vacancies s, since strong differences existed in the first cation-cation coordination shell that inferred the possible presence of nonpoint defects. The joint study also confirmed the point defects to be the s; no interstitial defects were seen and the O/Ti atom ratio was <2. Evidently, there exists an optimal O/Ti ratio for maximum photoactivity achieved when oxygen vacancies are located in the bulk lattice that act as electron traps subsequent to visible-light photoactivation of the doped specimens. No apparent effect due to -doping on the valence band edge was detected. Some localized states were, however, detected at the bottom of the conduction band with broad absorption around 500 nm. Results from DRIFT spectra indicated the presence of several anion-related impurities of a substitutional () and interstitial () nature. Although these species contributed to the absorption features, the authors [42, 43] found no clear correlation between any of these species and photoactivity. In fact, photoactivity best correlated with an optimal number of oxygen vacancies, above and below which a decrease of steady-state reaction rates occurred.

N-doped titania samples with high visible-light activity have been synthesized using a layered titania/isostearate nanocomposite from a sol-gel technique [44], with -doping achieved by treating the composite with aqueous followed by calcination either in an mixture or in pure at various temperatures (300, 350, 400, 450, and 500 C). The vivid yellow samples absorbed visible light in the 380–500 nm spectral region, and correlated with the extent of doped- content in the samples. However, the visible-light photoactivity did not correlate with content. The highest visible-light photoactivity was observed for the 400-C calcined sample. The quantity of content in the sample decreased on increasing the calcination temperature, which was particularly significant between 300 and 350 C, with the decrease being more important for the sample calcined in than for the sample calcined in pure .

Thus far we have witnessed that visible-light-active (VLA) systems doped with nitrogen possess, in most cases, good attributes toward the photooxidation of organic and inorganic (e.g., ) substrates. Of particular interest have been the materials doped with whose preparative methods have been varied but otherwise simple in a large number of cases. Most important, however, although all the -doped materials displayed absorption features and absorption edges red-shifted to the visible spectral region (at least to 550 nm), photoactivity of these systems under visible-light irradiation has not always correlated with these absorption features. In a recent study, Tachikawa and coworkers [45] addressed some of these issues and described mechanisms of the photoactivity of VLA specimens. Using solid-state NMR measurements combined with transient diffuse reflectance (TDR) spectroscopy, these workers provided direct evidence of the degradation of ethylene glycol with VLA-active TiO_2-xN_x under visible-light irradiation. It appears that photoassisted oxidations of organic compounds on the surface of TiO_2-xN_x proceed by surface intermediates generated from oxygen reduction (the superoxide radical anion, ) or otherwise water oxidation (the radical) and not by direct reaction with that may be trapped at the -induced mid-gap level (see Figure 3). Based on their experimental results, it is rather evident that both an appropriate lower-energy photo-threshold for visible-light absorption and high carrier mobilities are needed for advanced visible light-active -based photocatalysts.

Figure 3: Cartoons illustrating possible photoassisted processes of a substrate adsorbed on the surfaces of pure, -, -, and -doped nanoparticles. Reproduced with permission from [45]. Copyright (2007) American Chemical Society.

3. DFT Computations of Band Gap Energies in -Doped TiO₂

Different preparative methods and strategies of -doping as described above can lead to anion-doped metal-oxide materials with entirely different properties—these are the next generation photocatalysts. The key question that keeps recurring in the literature is the chemical nature and the location of the species that lead(s) the absorption edge of to be red-shifted and consequently to the visible-light activity of doped . Species such as , , and have been proposed, not to mention , and species that have been confirmed experimentally. Another key question regards the electronic structure(s) of the (anion)-doped materials and their fate when subjected to UV- and/or visible-light irradiation. Although, these questions have been addressed in several interesting computational studies, a consensual acceptance of the results has yet to be reached. Significant advances can be made in clarifying key questions by a combination of experimental and computational studies within the same laboratory or among collaborating laboratories.

Densities of states in anatase for substitutional doping of oxygen in the lattice by C, N, F, P, and S dopants were first reported by Asahi and coworkers [4, 46] using the full-potential linearized augmented plane-wave (FLAPW) formalism in the framework of the local density approximation (LDA). The calculations were carried out without geometry optimization for the five anion-dopings because the resulting atomic forces were apparently too large to obtain reasonable positions in the unit cell (eight units per cell). Three types of doping for were considered in the computations: (a) substitutional doping (), (b) interstitial doping (), and (c) both types of doping () in the anatase architecture. Optimization of the positions in the cell inferred molecularly bonding states ( and ) for cases (b) and (c) with bond lengths (improved by the generalized gradient approximation GGA) in fair agreement with accepted values: 1.20 A versus 1.15 A (–) and 1.16 A versus 1.10 A (). According to the authors [4], substitutional doping of was the most effective because its 2p states contribute to band gap narrowing by mixing with O 2p states of the valence band. Calculated imaginary parts of the dielectric functions of TiO_2-xN_x showed a shift of the absorption edge to lower energy by doping, with dominant transitions from rather than from as in . However, the calculated band gap energies were considerably underestimated relative to the experimental value (e.g., = 2.0 eV versus 3.2 eV for anatase) attributed, in part, to the well-known shortcomings of the LDA approach. The underestimated band gap was corrected using a scissors operator (a sort of fudge factor) that displaces the empty and occupied bands relative to each other by a rigid shift of 1.14 eV to bring the minimum band gap in line with experiment for the band gap of anatase (corrected = 3.14 eV). Accordingly, the band gaps of -doped TiO_2-xN_x systems were also adjusted by the factor 1.14 eV [4] on the assumption that the underestimated energy of the band gap in the LDA approach is not affected by -doping because long-range screening properties in TiO_2-xN_x were likely similar to those in .

The picture as to the exact cause of the red-shift of the absorption edge of in various -doped ( anatase ) powders becomes somewhat confused with the report from Yates group [30] that the absorption edge of a -doped rutile single crystal shifts to higher energy by 0.20 eV. Spin-polarized density functional theory (DFT) calculations within the GGA approximation by Di Valentin and coworkers [47] showed that whereas in anatase the localized 2p states, located just above the 2p states of the valence band, red-shift the absorption edge to lower energy, in rutile the tendency to red-shift the absorption edge is offset by a concomitant contraction of the 2p valence band resulting in an overall increase in the optical transition energy by ca. 0.08 eV. Compared to anatase, rutile has a wider (W) 2p band due to both its higher density and its different structure (see Figure 4). In this work [47], -doping was modeled by replacing 1, 2, or 3 oxygen atoms in a 96-atom anatase supercell and 1 or 2 oxygen atoms in a 72-atom rutile supercell giving a stoichiometry comparable to that used in experiments for TiO_2-xN_x for anatase and for rutile. Note that Asahi et al. [4, 46] used a higher level of -doping, which yielded a stoichiometry of . Inclusion of more atoms in the same supercell yielded more accurate results than using smaller supercells. Nonetheless, calculated band gaps [47] were still underestimated at 2.19 eV and 1.81 eV (at the position) versus the experimental 3.2 eV and 3.0 eV, respectively, for pure undoped anatase and rutile , again because of the shortcomings of the DFT method. Analysis of the electronic energy levels (see Figure 4) shows that -doping causes no shift of the position of both top and bottom of the 2p VB band and of the CB band relative to pure undoped anatase , in significant contrast with the conclusions of Asahi and coworkers [4, 46] with respect to the undoped material. Structural variations in rutile subsequent to substitution of one atom with in the 72-atom supercell appear significant in rutile relative to anatase in which the variations were inconsequential. In any case, the impurity states can act as deep electron traps in TiO_2-xN_x systems (see Figure 4). Di Valentin et al. [47] also considered the contribution of oxygen vacancies (s), estimated experimentally at 0.75 to 1.18 eV below the conduction band (DFT calculations placed them at 0.3 eV below ), to the overall visible-light photoactivity of -doped systems when s trap electrons to produce -type color centers. The simultaneous presence of dopants and s can also lead to charge transfer states (reaction 1) that can further contribute to the visible-light photoactivity,

Figure 4: Schematic representation of the energy band structure of pure and -doped anatase and rutile (energies not to scale). Note the modest shift of 0.03 eV in of the conduction band to higher energies and the contraction energy in the -doped rutile . represents the energy of the dopant levels above the valence band. Higher levels of doping, for example, three atoms per supercell, cause a small shift of to higher energies for which is overcompensated by the presence of -derived states just above VB, so that the excitation energy from these states to the conduction band is reduced by <0.1 eV compared to pure anatase ( denotes the width of the valence bands). Adapted from results reported in [47].

In a later study that combined experiments (EPR, XPS) and DFT calculations performed using the plane-wave-pseudopotential approach, together with the Perdew-Burke-Ernzerhof (PBE) exchange correlation functional and ultrasoft pseudo-potentials, Di Valentin and coworkers [48] characterized the paramagnetic species present in -doped anatase powders obtained by sol-gel synthesis, and unraveled some of the mechanistic details of the visible-light activity of -doped as to whether photoactivity is due to or species or simply to substitutional -doping. The TiO_2-xN_x sample was obtained by hydrolysis of - in isopropanol media in the presence of aqueous as the source, followed by calcinations of the -doped specimen at ca. 500 C for 2 hrs. XPS 1s spectra showed only a peak at ca. 400 eV attributed to interstitial (without precluding others); no 396 eV peak was seen that might have originated from substitutional -doping. EPR spectra indicated the presence of two different paramagnetic species that were attributed to substitutional and interstitial species. These results led them to consider two structurally different locations for the dopant in their DFT calculations: substitutional and interstitial atoms in the anatase matrix. In the substitutional model, the that replaces in the lattice was taken to be bonded to three Ti atoms in the 96-atom supercell (a) so that the paramagnetic species is formally , whereas in the interstitial model is added to the 96-atom supercell bonded to one or more and thus is in a positive oxidation state as in , and/or (b)(see Figure 5). Figure 6 illustrates the DFT band structure of the -doped and reports the calculated albeit still underestimated band gap energy of titania. The two bonding states in Figure 6(b) for lie deep below the 2p band but the two still occupied states lie above (0.73 eV) the 2p band and lie at higher energy than the 2p states of substitutional (0.14 eV). A more interesting consequence of the picture of Figure 6 is that the two electrons left in the formation of an oxygen vacancy, which typically would form two color centers, may also be trapped by substitutional and interstitial yielding the azide species ( denoted as ) and the hyponitrite species (). Another significant result deriving from the DFT calculations of Di Valentin and coworkers [48] is that -doping leads to a substantively reduced energy of formation of s (4.3 eV to 0.6 eV for anatase) with important consequences in the generation of -type and color centers. Experimentally, which of the two types of dopants predominates in the -doped will depend on the experimental conditions, for example, nitrogen and oxygen concentrations, and calcination temperatures. What Figure 6 also implies is that moving from substitutional to interstitial is in fact an oxidative step, which according to DFT estimates is ca. 0.8 eV exothermic. Thus, there is a cost for the reverse, that is, interstitial -doping is preferred when TiO_2-xN_x systems are prepared in excess nitrogen and oxygen, whereas high-temperature calcination of -doped systems, commonly done in most experiments, both substitutional N and formation of oxygen vacancies s are likely the preferred occurrences.

Figure 5

Figure 6: Electronic band structure for (a) substitutional and (b) interstitial -doped anatase as given by PBE calculations at a low-symmetry -point. In the former, the site contains the paramagnetic species making the site electrically neutral (replaced ), whereas in the latter the site is occupied by the radical . The estimated band gap energy is also indicated. Reproduced with permission from [48]. Copyright (2005) American Chemical Society.

The question on the blue-shift of the absorption edge of -doped rutile single crystals contrasting the red-shifts in -doped powders was also taken up in a DFT study by Yang and coworkers [49] using the plane-wave method. Results confirmed those of Di Valentin et al. [47] that some 2p states lay above the 2p valence band when substitutes in the lattice and when is located at interstitial positions. However, no band gap narrowing was predicted by the calculations of Yang et al. [49]. When substitutes Ti atoms in the rutile lattice, a band gap narrowing in the rutile crystal is apparently possible because the dopant introduces some energy states (the 2p states) into the bottom of the conduction band [49]. The authors rationalized this inference by the fact that removal of electrons from the supercell on replacing one Ti with a atom leads to a reduction of the Coulomb repulsion and thus to a shift of the energy band edges, that is, the band gap energy is reduced by ca. 0.25 eV relative to the undoped rutile supercell whose estimated band gap energy was calculated to be 1.88 eV. This is reminescent of the assertion by Asahi et al. [4, 46] that -doping causes band gap narrowing because the 2p states mix with the 2p states in the valence band, thereby widening the valence band and shrinking the band gap.

Substitutional -doping can be stabilized by the presence of oxygen vacancies under oxygen-poor experimental conditions, whereas under oxygen-rich conditions interstitial species () become favored. The -doped sample prepared by the sol-gel process of stoichiometry near , examined earlier by Di Valentin and coworkers [48], was re-examined more closely by Livraghi et al. [50] in a series of experiments and DFT calculations aimed at determining the fate of the doped specimen when irradiated at different wavelengths in the presence of adsorbates. The experiments asserted that species were responsible for the absorption of visible-light radiation, and consequently for the visible-light activity, as well as for the photoinduced electron transfer from the solid to surface electron scavengers (adsorbates) such as molecular . The UV-visible diffuse reflectance spectrum of the sample (see Figure 7) is nearly identical to the many reported DRS spectra of -doped specimens prepared in a variety of ways. However, as we will see below, this spectral behavior is identical to the spectral behavior of so many other doped samples that have been doped with different types of dopants (e.g., transition metal ions, C, S, and others) and synthesized by different methods. Previous EPR work (Di Valentin et al. [48]) had identified two distinct -related paramagnetic species in -doped , one of which was the molecular radical [51] located in closed pores within the crystals and thus had no influence on the electronic structure of the solid. No evidence of -type paramagnetic species was found as had been reported by Yates’ group [31, 32]. Whatever the nature of the paramagnetic species, it was stable to washing and to calcination in air up to ca. 500 C. This was taken to mean that the nitrogen radical species, identified as in Figure 8 interact strongly with the lattice. DFT calculations carried out on the 96-atom supercell involved two interstitial nitrogens () or two substitutional nitrogen () paramagnetic species plus an oxygen vacancy () located far away from these -centers to avoid direct defect/impurity interactions. Note that removal of an atom from the lattice leaves behind two electrons to form the neutral center ( in the Kroger-Vink notation), or they may be trapped by neighboring species to give two color centers, which Henderson et al. [52] positioned at 0.8 eV below the bottom of the conduction band. Other studies indicated otherwise, although there are electron traps around this energy. color centers certainly do exist as demonstrated by EPR measurements [53, 54]. One of the consequences of the high number of s under oxygen-poor conditions in -doped is the partial quenching of paramagnetic species, which are transformed [48] into through reduction by color centers (see Figure 8). The energetically favored reduction of species may be the cause for the small energy cost in the formation of s in -doped (see above). The EPR peaks attributed to centers disappeared on reduction of the sample (reaction (2)) whether by annealing in vacuo or by other means to then reappear on re-oxidation. Thus, the -doped specimen (at least the one prepared by the sol-gel method) contained paramagnetic related species in the bulk lattice () and a number of diamagnetic species (), the presence of which depended on the oxygen content in the metal-oxide sample. The EPR signal due to increased on irradiating the doped sample at 437 nm in . A new EPR line appeared that was attributed to radical anions (reactions (3) and (4)). These anions are apparently stabilized on two different surface species, which Livraghi et al. [50] claimed could be typical of -doped . Figure 9 illustrates the process embodied in the formation of the superoxide radical anions.

Figure 7: UV-visible diffuse reflectance spectra of undoped and -doped . Reproduced with permission from [50]. Copyright (2006) American Chemical Society.

Figure 8: Electronic band structure changes from interactions between ( or ) and color centers. Reproduced with permission from [50]. Copyright (2006) American Chemical Society.

Figure 9: Sketch of the proposed mechanism for the processes induced by vis-light irradiation of the -doped sample in atmosphere. Reproduced with permission from [50]. Copyright (2006) American Chemical Society.

So far all the DFT-based calculations have failed to calculate experimentally commensurate band gap energies for undoped anatase and rutile , and consequently for all anion-doped systems unless, as some have done, one resorts to the scissor operator. In a comprehensive theoretical investigation of substitutional anion doping in , Wang and Lewis [55] explored the electronic properties of -, -, and -doped materials using an ab initio tight-binding method (FIREBALL) based on density functional theory and a nonlocal pseudopotential scheme. The method uses confined atomic-like orbitals as the basis set, which led to a calculated direct band gap from to of 3.05 eV for rutile that accords with the experimental band gap of 3.06 eV [56]. The local density approximation (LDA) approach generally underestimates the experimental band gap for insulators and semiconductors; ab initio plane-wave calculations for also place the band gap at around 2.0 eV [57]. In their theoretical treatment, Wang and Lewis [55] compensated for the underestimation by the LDA approach. For anatase, the direct band gap from to of 3.26 eV accords with the experimentally observed 3.20 eV [58]—see Table 1. The upper valence bands are composed mainly of 2p states, whereas the lower conduction bands consist primarily of unoccupied Ti 3d states.

Table 1: Comparison of calculated band gaps of undoped rutile and anatase with experimental and calculated values reported by others ( is the direct band gap). From Wang and Lewis [55].

In the computations of Wang and Lewis [55], two atoms were replaced by one because of the odd number of electrons in the atom to yield an effective -doping level of 0.52% in the low concentration case. The CB minimum remained unchanged. However, new states were introduced by -doping just above the valence band edge of bulk , as well as states that penetrated into the upper valence band of the bulk states. No significant energy shift (<0.05 eV) in the VB edge was seen in -doped rutile at the high content of 5.2%, contrary to the significant shift observed for the low -doping level of 0.52% which gave a narrowed band gap of 2.55 eV. Density-of states (DOS) calculations thus inferred that low doping levels greatly improve the visible-light photoactivity. No significant overlap occurred between the 2p states and the 2p states for the 0.52% -doping level. Moreover, at the low concentration the valence band edge was more localized compared to the high -doping level in -doped rutile . Unlike -doped rutile , however, identical shifts of ca. 0.44 eV were obtained in -doped anatase at both high and low doping levels resulting in a band gap of ca. 2.82 eV (440 nm). DOS calculations indicated significant overlap between the N 2p states and 2p states for the 5.2% N-doping. By contrast, at the low N-doping level of 0.52% N, the states introduced by were distinct, highly localized on the single dopant state, and there was no significant overlap. A result of this is that the high -doping level in -doped anatase would lead to greater visible-light photoactivity.

The brief discussion above points to a lack of consensus on whether or not there is band gap narrowing in doped materials based on DFT calculations. Some of the theoretical studies have deduced from these calculations that there is a rigid shift of the valence band edge to higher energies, thus narrowing the intrinsic band gap of as a consequence of doping. The discrepancies cannot be attributed to a simple semantic problem. Experimentally, anion-doping and cation-doping of do red-shift the absorption edge of in the doped samples, thus yielding potentially visible-light photoactive materials that might prove useful in several important applications of surface processes occurring on the