A short overview of the state of the art and perspectives on the main basic factors hindering the development of photocatalytic treatment of water

Water treatment by all advanced oxidation processes (AOPs) generates intermediate products resulting from the degradation of the initial organic pollutants. Therefore, the toxicity of water treated by AOPs may not be decreased and may even be increased if complete mineralization cannot be achieved because of time constraints and cost. Moreover, ions – particularly HCO₃⁻/CO₃²⁻ – present in most waters compete with organic pollutants to react with •OH radicals generated by the AOPs. This competition can markedly diminish the efficacy of the AOPs concerning the degradation and mineralization of the targeted organic pollutants.

Among the AOPs, photocatalysis based on the photo-excitation of a semiconductor (Pichat 2013) suffers also from other limitations. The main one is that the electron-hole pairs formed upon photo-excitation are prone to recombine. The higher the recombination rate, the higher the loss of energy. Then, the adsorption of the organic pollutants on the semiconductor – regarded as an asset with respect to the AOPs taking place in homogeneous phase – can be limited by the partition of these pollutants between the semiconductor surface and the water bulk. On the other hand, the use of solar irradiation (Malato et al. 2013) to excite the semiconductor is often put forward as a potential advantage of photocatalysis because of electricity saving, to the detriment, however, of the water treatment regularity. In particular, small units have been proposed to purify agricultural waters on-site (Pichat et al. 2004) and to supply drinkable water at isolated locations. Nevertheless, this use may not be sufficiently efficient because the semiconductors that are stable when irradiated in water have a low sensitivity in the visible region of the solar spectrum.

Extensive research has been carried out over the years to cope with these three basic obstacles which restrict the applications of heterogeneous photocatalysis for water purification.

First, the principal means that have been explored to decrease the recombination rate of the photoproduced charges have included:

• modifying the structure, morphology and crystallinity of the semiconductor;

• doping the semiconductors with cations;

• depositing tiny metal particles on the semiconductor surface.

Second, the effects of the addition of another adsorbent have been studied in view of achieving better contact between the semiconductor and the organic pollutants, and of simultaneously decreasing the toxicity level of the treated water.

Third, regarding a better fitting of the semiconductor absorption with the solar spectrum, the following methods have been explored:

• the synthesis of a variety of semiconductors with smaller band gaps, keeping in mind the necessity that the conduction band position must enable the scavenging of photoproduced electrons by the oxygen dissolved in water;

• the doping of semiconductors with cations or anions.

The main results of these investigations are summarized in this article.

A photon, with energy at least equal to the band gap and absorbed by a semiconductor, promotes an electron from the filled valence band to the vacant conduction band, which leaves behind a ‘hole’ (i.e., an electron vacancy) in the valence band (Figure 1). Unfortunately, this activation process is readily reversible, that is, the out-of-equilibrium electron-hole pairs can recombine through radiative and nonradiative processes, either directly (i.e., band-to-band recombination) or most often indirectly (viz., via bulk or surface defects). Charge recombination is an inherent and extremely important drawback of heterogeneous photocatalysis. It corresponds to a significant loss of the energy supplied.

Indirect charge recombination occurs via energy levels located within the band gap. First, these levels can correspond to impurities, that is, cations other than Ti⁴⁺. These cations come from the preparation of TiO₂, principally because they are present in the ore (presently, FeTiO₃ is the main one). Purification is costly and cannot be total. Hence, this pathway of charge recombination can be minimized but not completely avoided.

Time resolved microwave conductivity (Kunst & Beck 1986; Warman & de Haas 1991) is an excellent method to compare k_r for various samples. For example, the conductivity – which is related to k_r – increased substantially with the elementary particle size of anatase samples prepared in a flame reactor (Warman et al. 1991), even though other factors, such as the particle morphology, could also intervene. This result was expected because the bigger particles, obtained at a higher temperature, should have a lower density of structural defects. A decrease in the number of defects corresponds to an increase in crystallinity. A higher crystallinity is most often obtained through sintering. Indeed, the conductivity of Millennium Chemicals/Cristal samples was higher, the higher the sintering temperature (Pichat et al. 2010; Alaoui et al. 2012). However, sintering decreases the surface area available for reactants. Indeed, a tradeoff between k_r and the surface area has been shown to exist (Enriquez & Pichat 2006; Agrios & Pichat 2006; Enriquez et al. 2007). The surface area is the dominant factor for pollutants requiring close contact with TiO₂, that is, those for which direct reaction with h⁺ (Figure 1) prevails. The degradation rate of pollutants for which the •OH radical-mediated mechanism is prominent depends more on k_r. Accordingly, much lower decreases in removal rates were observed for these latter pollutants when TiO₂ became partially inaccessible because of the use of a SiO₂ binder (Enriquez et al. 2004). These factors were also shown to affect differently various photocatalytic reactions in the aqueous phase or the gas phase (Kominami et al. 2002). In practice, the consequences are that (i) one test pollutant does not suffice to quantitatively compare photocatalytic activities, (ii) selection of a photocatalyst requires case-by-case trials, and (iii) the use of photocatalytic reactors in series with different kinds of TiO₂ might be envisaged.

Moreover, a high k_r inhibits the increase in the efficacy of the water photocatalytic purification that could be effected by an increase in irradiance. For low irradiance, a linear relationship has indeed been observed between irradiance and the photocatalytic removal rate of a pollutant in water. Unfortunately, for high irradiance, the dependence becomes half order, so that the benefit due to irradiance increase becomes lower (D'Oliveira et al. 1990; Al-Sayyed et al. 1991). The explanation is that the additional h⁺ generated by a higher irradiance are predominantly consumed by recombination if the concentration of pollutants is low, which is the case in water treatment.

Long ago, deposition of metal particles (mainly from groups IB and VII) on TiO₂ was explored as a means of diminishing the charge recombination rate (Pichat 1987; Highfield & Pichat 1989; Linsebigler et al. 1995). It was assumed that electrons could be attracted to the metal provided that the metal work function has an adequate value. As expected, the electron migration was found to depend not only on the metal, but also on its amount, particle size and dispersion on the TiO₂ surface. Obviously, factors that favor charge recombination in or on TiO₂ somewhat counter the electron migration to the metal. The electrons collected by the metal are thought to be transferred to O₂ (Gerischer 1980). However, at a very low wt % of metal, the O₂ amount adsorbed on the metal can be low compared with the O₂ amount adsorbed on TiO₂. This may attenuate the relative significance of the effect arising from O₂ adsorption on the deposited metal. Because of the transfer of electrons to the metal, the scavenging of holes by pollutants (if the corresponding redox potential allows this to happen), OH⁻ anions, H₂O (note that this occurrence has been questioned (Salvador 2007; Montoya et al. 2011)) and surface O²⁻ anions would be increased. Unfortunately, charge recombination can also occur at the metal particles if they are numerous and/or well dispersed on TiO₂, even though it is hindered by a Schottky barrier between the metal and TiO_2. Accordingly, an optimum amount of deposited metal has always been observed (most often, around 0.5 wt %). In spite of this low optimum amount, the resulting cost can be still too high for photocatalytic water treatment. Indeed, costly metals, such as Pt and Ir, are much more effective than metals that are less expensive, such as Ni, because their work function makes the electron transfer from the metal to TiO₂ easier.

On the whole, the expected favorable effect on metal deposits upon removal of water pollutants is difficult to tailor and anyhow limited. The adsorption of pollutants on the metal could also affect the photocatalytic degradation rate and pathways, but to an extent that is likely not significant for water treatment.

Introducing selected cations in the TiO₂ lattice has also been proposed to create energy levels in the band gap capable of decreasing the charge recombination rate by trapping one type of charge carrier (Bin-Daar et al. 1983). The effect obviously depends on the cation, its amount and its location (substitutional vs. interstitial; bulk vs. surface; homogeneous vs. heterogeneous distribution). Co-doping has also been investigated with the expectation that each cation would trap opposite charges. Negative and positive photocatalytic effects, as well as pollutant dependency, have been reported even for the same cation. That is not surprising because, as aforementioned, many parameters intervene (e.g., Choi et al. 1994; Fuerte et al. 2001; Fujishima et al. 2008; Zhang et al. 2009; Bak et al. 2010; Henderson 2011). Obviously, the detrapping rate of trapped charges should affect the doping effect on the photocatalytic activity. Depending, inter alia, on its amount and energy level, the dopant can, unfortunately, act as a recombination center of charges (e.g., Pichat 1987; Highfield & Pichat 1989). To the author's knowledge, trials have not been undertaken for water purification beyond the laboratory scale because of these mixed results.

In the case of AOPs in liquid water, the active species have to encounter the trace pollutants dispersed in the bulk phase. Consequently, adsorption is often presented as an advantage of photocatalysis because the pollutants are attracted to TiO₂ where the active species, including •OH radicals, are generated. However, a closer look leads to moderating the importance of the effect of this apparent asset. The water molecules in the layers near TiO₂ are bond together more strongly than in the liquid bulk. Indeed, for the 101 plane of anatase, calculations have shown that (i) the adsorption enthalpy for the first layer is about 30% higher than for the second layer (Opalka et al. 2004) and (ii) the mobility of the water molecules at 160 K is substantially decreased for both the first and second layers in comparison to the third layer (Figure 2; Tilocca & Selloni 2004). Accordingly, it is understandable that these adsorbed water layers could hinder the approach of organic molecules to TiO₂.

For moving, non-polar pollutants have to change the orientations of the water molecules that surround them (Israelachvili 1991). This phenomenon becomes less probable in the adsorbed phase because of the aforementioned restricted mobility of the water molecules. A molecule of a pollutant – such as phenol, pyridine and their derivatives – which can form a hydrogen-bond with one water molecule is also surrounded by other water molecules. This creates a kind of ‘cavity’ whose size depends on the pollutant (Malaspina et al. 2002). For moving, this ensemble must break hydrogen-bonds, thus disorganizing the water network, which is more difficult near TiO₂.

Indeed, extremely low adsorbed amounts on TiO₂ have been measured for many organic compounds in aqueous suspensions (Doherty et al. 1995; Agrios & Pichat 2006), which is in agreement with this reasoning. On the other hand, the photocatalytic removal rates of poorly adsorbed pollutants are often not very different from those of pollutants adsorbed to a higher extent. That observation has led to suggestions that •OH radicals can move into the near-surface water layers (Turchi & Ollis 1990; Peterson et al. 1991; Cunningham et al. 1994; Enriquez et al. 2004, 2007). The size of a •OH radical is close to that of a water molecule. Therefore, the reorientation and restructuration of the water network should be very small when a •OH radical replaces a H₂O molecule (Figure 3) (Vassilev et al. 2004; Belair et al. 2005; Du et al. 2006). The mobility of •OH radicals could understandably be much higher than those of pollutants. Also, the •OH radicals are supposed to be linked to the surface by weak hydrogen-bonds because they stem from either oxidation of adsorbed H₂O molecules or reduction of H₂O₂ molecules formed in situ. Accordingly, their detachment from the surface can be envisaged. However, their mobility should be restricted within the more rigid network of adsorbed water molecules with respect to that in bulk water. It is then foreseeable that their properties could somewhat differ from those of the •OH radicals formed in bulk water by homogeneous AOPs.

Adding activated carbon (AC) to TiO₂ appears as a good means to overcome the problem of the approach of organic pollutants to TiO₂ across the well-organized layers of adsorbed water. Following a pioneering article (Uchida et al. 1993), many papers have been published on the TiO₂-AC combination (e.g., see the references in Li Puma et al. (2008)). Unlike TiO₂, AC is hydrophobic and organophilic. The adsorption of organic pollutants by AC tends to be non-specific. Because of that, AC is commonly employed in water treatment. Regarding AC combination with TiO₂, the reasoning is that the trace organic pollutants in liquid water will be attracted to AC, and then could be transferred to TiO₂ owing to the interfaces between the two solids (e.g., Paz 2010). Alternatively, a transfer of active species from TiO₂ to the pollutants adsorbed on AC has been postulated, in line with the preceding discussion about the mobility of •OH radicals. Indeed, remote degradation of a solid like soot has been reported (Lee & Choi 2002; Lee et al. 2004; Chin et al. 2009). Diffusion of pollutants over distances up to 20 μm has also been deduced from experiments using well-defined structures consisting of alternate microstripes or islands of TiO₂ on Si (Haick & Paz 2001).

Obviously, the area of the TiO₂-AC interfaces must be maximized to increase the efficacy. To achieve that, many syntheses have been explored, from simple mixing in the dry state or in aqueous phase to chemical vapor deposition (Li Puma et al. 2008). The ratio TiO₂/AC has been varied; in the extreme cases, the resulting solid could be regarded as TiO₂ deposited on AC or the opposite. Not surprisingly, the morphology of each solid plays a role. Optimizing the efficacy also means (i) preserving irradiation of TiO₂ and (ii) avoiding blockage of AC pores by TiO₂ in order to not diminish the adsorbing surface available to pollutants.

Because of the variety of parameters, the numerous papers and patents have reported recipes rather than clear guidance. Once the improvement only due to additional adsorption of pollutants was correctly taken into account, several studies have reported significant increases in photocatalytic efficacy for waste water samples and distilled water containing one pollutant or a mixture of some pollutants. These results have led to the appearance of commercialized TiO₂ + AC materials, though not available on a large industrial scale. Added assets of these materials are: (i) the well-known use of AC to purify water, (ii) an AC price compatible with water treatment, (iii) the increased duration of AC usage without regeneration because of in situ photocatalytic cleaning, (iv) the easy recovery of the material by filtering if granular AC is utilized, and (v) the enhanced adsorption of intermediate products of photocatalytic degradation which can be toxic and whose complete mineralization can be too costly.

TiO₂ is not excited beyond about 400 nm and thus does not take advantage of by far the main part of the solar spectrum received on Earth. Consequently, to make solar purification of water more efficient, numerous attempts have been undertaken to replace TiO₂ by semiconductors capable of being photo-activated in the visible spectral region. Most semiconducting sulfides are excluded because of their instability in water under solar irradiation. Multiple trials at displacing the absorption of TiO₂ towards higher wavelengths have also been reported.

For purification of water exposed to ambient air or aerated, the requirement for any semiconductor is that the energy of the conduction band is sufficiently high to permit the reduction of O₂, which is the main acceptor of the photogenerated electrons (Figure 4). This requirement has sometimes been forgotten. For instance, it rules out the use of the following, chemically stable, simple semiconducting oxides, even if their band gap – narrower than that of TiO₂ – would seem of interest for solar applications: WO₃, Fe₂O₃, Bi₂O₃. This is also the case for semiconductor oxides comprising two metal elements, such as BiVO₄ and Bi₂WO₆ (Castillo et al. 2010; Saison et al. 2011). It should also be taken into account that if rarer elements were included in the semiconductor composition it will likely not be cost-effective for water treatment. Indeed, numerous attempts in this domain have concerned water splitting to produce H₂ as an energy source, in which case higher-cost photocatalysts can be tolerated. Finally, toxic elements must be excluded for water treatment, which eliminates the use of visible-light sensitive oxides such as Pb_xNb_yO_z (Li et al. 2007) for fear that Pb cations might be released. It can then be concluded that the quest for substituting TiO₂ with other semiconducting oxides has not been successful until now.

Introducing foreign cations or anions has been a method to add energy states within the band gap of TiO₂, thereby allowing absorption of visible light near the UV spectral region. Depending on numerous factors, these ions can either form a band (Figure 4) or be discrete. In the latter case, the mobility of charges located in these energy states is low, so that these charges are prone to recombine before reaching the surface to participate in photocatalytic reactions. Also, native point defects resulting from charge imbalance due to the introduction of aliovalent ions may behave as recombination centers of photoproduced charges. These adverse effects with respect to photocatalytic activity are difficult to minimize in order to obtain samples that are not only notably active under visible irradiation, but whose activity under UV irradiation is not significantly diminished.

In practice, for cation doping no sure guidance regarding the nature, amount and preparation has emerged in spite of numerous investigations, although ion-implantation has produced valuable results (Anpo & Takeuchi 2003). Most importantly, the overall improvement in photocatalytic activity under solar light has not been sufficiently attractive to induce trials beyond the laboratory scale.

Many anions have been used in attempts to render TiO₂ active when irradiated by visible radiations. After a long-ignored pioneering work (Sato 1986), an article (Morikawa et al. 2001) about N ‘doped’ TiO₂ initiated many investigations on TiO₂ ‘doped’ not only by N, but also by S, B, P, halides and even C (e.g., Ohno et al. 2003; Sakthivel & Kisch 2003; Kitano et al. 2006; Sun et al. 2010; Dozzi et al. 2011). Combinations of elements were also studied (e.g., Liu et al. 2008; Di Valentin et al. 2008; Palaez et al. 2009; Xu et al. 2010, 2011). The objective was, in principle, to replace O by these elements and thereby to create energy states above the valence band and/or moving the valence band top toward higher energy (Figure 4). Extremely diverse synthesis methods have been used and have resulted in samples having very different photocatalytic activity, even when the same foreign element was introduced.

Regarding N, for which more papers have been published, there are uncertainties about the chemical state and location of the N atoms depending on the preparation method, the amount and source of N atoms, and other factors (Serpone 2006; Henderson 2011). Moreover, the interpretation of the results issuing from the characterization methods is sometimes debated. In addition to the existence of substitutional and interstitial N (in the form of N⁻² and N⁻³ anions), the formation of Ti–O–N bonds and the presence of embedded N₂ and NO have been reported. It has also been suggested that inorganic and organic N-containing surface species could also act as sensitizers beyond 400 nm (e.g., Cheng et al. 2012). New energy levels within the band gap (Figure 4) could also be associated not directly with N atoms but with oxygen vacancies arising from the preparation (Serpone 2006; Henderson 2011). Similar uncertainties have been reported for the other non-metals, although the picture may be less complex for halides and still more complex for C atoms.

For photocatalytic removal of pollutants, the key factor is the charge trapping and detrapping rates at these new energy levels. Cases have been reported where oxidation of pollutants supposed to proceed with valence band holes did not occur with holes located within the band gap (that is, upon visible-light excitation; Figure 4) because these rates were too low (e.g., Su et al. 2008; Tojo et al. 2008). Accordingly, the visible-light photocatalytic activity was proposed to take place through •OH radicals formed from the reduction of O₂ via superoxide and hydrogen peroxide, as occurs with UV radiations. Anyhow, the low rates regarding holes formed at levels within the band gap are unfavorable for the overall photocatalytic rate under visible irradiation compared with UV irradiation. Consequently, it reduces the interest of making TiO₂ photocatalytically active beyond 400 nm.

It seems that these results have restrained the development of commercially available TiO₂ modified with non-metals, even though these modifications might be cost-effective because some of the chemicals that could be utilized as N or C sources have reasonable prices. Although research aiming at modifying TiO₂ to better take advantage of solar irradiation is still very active, TiO₂ KronoClean (kronosww.com) seems to be one of the rare products presently on the market; it has been used as a reference for testing laboratory-prepared samples (e.g., Palaez et al. 2009).

This overview has presented some of the fundamental means for improving the efficacy of photocatalytic water treatment by (i) decreasing the recombination rate of the photoproduced charges, (ii) increasing the adsorption of trace organic pollutants on the photocatalytic material, and (iii) extending the spectral range of photocatalytic activity to better use solar irradiation when it is appropriate to use sunlight. Numerous papers have been published in these three domains. Various solutions have been proposed. However, most of them concern other applications of photocatalysis (Agrios & Pichat 2005; Fujishima et al. 2008; Zhang et al. 2009) because water treatment requires photocatalysts comprising innocuous and low-cost chemical elements. Consequently, no commercialized material has emerged to decisively overcome the recombination of charges and/or to substantially increase the efficacy under solar irradiation. Possibly, combinations of TiO₂ with AC might be susceptible of offering valuable materials to increase the photocatalytic removal rates of trace organic pollutants in water because AC is organophilic. But, to the best of our knowledge, such materials are not yet commercially available on a large industrial scale.

In the author's view, studies to better understand the photocatalytic events at the molecular level, including interactions between TiO₂, H₂O and probe organics, are still needed in the hope of perhaps overcoming the obstacles presented in this short overview. These studies imply the utilization of photocatalysts that are better defined with well-characterized surfaces (Henderson 2011). They should involve composites with AC and possibly other non-semiconducting adsorbents. The use of probe molecules capable of revealing the mechanisms and pathways (Pichat 2003; Jenks 2013), and that of in situ studies such as Fourier transform infrared spectroscopy (FTIR) (Pichat 2014), preferentially time-resolved, should be more developed.