1-s2.0-S0167572908000757-main

Surface Science Reports 63 (2008) 515582Contents

1. Introduction........................................................................................................................................................................................................................5162. Historical overview ............................................................................................................................................................................................................5163. Properties of TiO2 materials ..............................................................................................................................................................................................519

3.1. Crystal structures...................................................................................................................................................................................................5193.2. Electronic properties .............................................................................................................................................................................................5203.3. Surface structure studies.......................................................................................................................................................................................5233.4. Surface chemical studies: Interactions with water .............................................................................................................................................5233.5. Surface chemical studies: Interactions with dioxygen and other species .........................................................................................................5273.6. Bulk chemistryHydrogen....................................................................................................................................................................................5273.7. Electrochemical properties ...................................................................................................................................................................................5293.8. Photoelectrochemical properties..........................................................................................................................................................................534

4. Fundamentals of photocatalysis........................................................................................................................................................................................5344.1. Mechanisms of photocatalysis ..............................................................................................................................................................................534

4.1.1. Photoelectrochemical basis of photocatalysis ......................................................................................................................................5344.1.2. Time scales ..............................................................................................................................................................................................5384.1.3. Trapping of electrons and holes.............................................................................................................................................................541

Corresponding author. Tel.: +81 (0)44 819 2020; fax: +81 (0)44 819 2038.E-mail address: [email protected] (A. Fujishima).Contents lists available at ScienceDirect

Surface Science Reports

journal homepage: www.elsevier.com/locate/surfrep

TiO2 photocatalysis and related surface phenomenaAkira Fujishima a,, Xintong Zhang b, Donald A. Tryk ca Kanagawa Academy of Science and Technology, 3-2-1 Sakado, Takatsu-ku, Kawasaki-shi, Kanagawa 213-0012, Japanb Center for Advanced Optoelectronic Functional Materials Research, Northeast Normal University, 5268 Renmin Street, Changchun 130024, Chinac Fuel Cell Nanomaterials Center, University of Yamanashi, Takeda 4-3-11, Koufu, Yamanashi 400-8510, Japan

a r t i c l e i n f o

Article history:Accepted 1 October 2008editor: Y. Murata

Keywords:Titanium dioxideTitaniaTiO2Self-cleaning surfacesSuperhydrophilic effectAnion dopingWater splittingEnvironmental cleaning

a b s t r a c t

The field of photocatalysis can be traced back more than 80 years to early observations of the chalking oftitania-based paints and to studies of the darkening of metal oxides in contact with organic compoundsin sunlight. During the past 20 years, it has become an extremely well researched field due to practicalinterest in air andwater remediation, self-cleaning surfaces, and self-sterilizing surfaces. During the sameperiod, there has also been a strong effort to use photocatalysis for light-assisted production of hydrogen.The fundamental aspects of photocatalysis on the most studied photocatalyst, titania, are still beingactively researched and have recently become quite well understood. The mechanisms by which certaintypes of organic compounds are decomposed completely to carbon dioxide and water, for example,have been delineated. However, certain aspects, such as the photo-induced wetting phenomenon,remain controversial, with some groups maintaining that the effect is a simple one in which organiccontaminants are decomposed, while other groups maintain that there are additional effects in whichthe intrinsic surface properties are modified by light. During the past several years, powerful tools suchas surface spectroscopic techniques and scanning probe techniques performed on single crystals in ultra-high vacuum, and ultrafast pulsed laser spectroscopic techniques have been brought to bear on theseproblems, and new insights have become possible. Quantum chemical calculations have also providednew insights. Newmaterials have recently been developed based on titania, and the sensitivity to visiblelight has improved. The new information available is staggering, but we hope to offer an overview ofsome of the recent highlights, as well as to review some of the origins and indicate some possible newdirections.

2008 Elsevier B.V. All rights reserved.0167-5729/$ see front matter 2008 Elsevier B.V. All rights reserved.doi:10.1016/j.surfrep.2008.10.001

n6.10. Photochromism......................................................................................................................................................................................................5746.11. Microchemical systems .........................................................................................................................................................................................574

7. Summary ............................................................................................................................................................................................................................575Appendix. TiO2 film preparation methods ..................................................................................................................................................................576References...........................................................................................................................................................................................................................576

1. Introduction

Photocatalysis is generally thought of as the catalysis of aphotochemical reaction at a solid surface, usually a semiconductor[116]. This simple definition, while correct and useful, however,conceals the fact that theremust be at least two reactions occurringsimultaneously, the first involving oxidation, fromphotogeneratedholes, and the second involving reduction, from photogeneratedelectrons. Both processes must be balanced precisely in order forthe photocatalyst itself not to undergo change, which is, after all,one of the basic requirements for a catalyst.It will be seen in this review of the fundamentals and selected

applications of photocatalysis, principally on titanium dioxide,that there is a host of possible photochemical, chemical andelectrochemical reactions that can occur on the photocatalystsurface. The types of reactions occurring, their extent and theirrates depend upon a host of factors that are still in the processof being unraveled. Furthermore, there can indeed be changesthat occur, involving the surface and bulk structure and evendecomposition of the photocatalyst, a fact that appears to stretchthe definition of the term.This topic started its early history asmostly a nuisance involving

the chalking of titania-based paints [17,18] and then graduallytransformed into a highly useful approach to the remediation ofwater and air and then into an approach tomaintain surfaces cleanand sterile. Along theway, it has also transformed into an approachto photolytically split water into hydrogen and oxygen [1921]and also an approach to perform selective oxidation reactions inorganic chemistry [22].

have tried to put together an overview of some of the morefundamental aspects, which are in their own right extremelyscientifically interesting and which also need to be betterunderstood in order tomake significant progresswith applications.The review will be divided into several sections: 2. Historical

overview; 3. Properties of TiO2 materials; 4. Fundamentals ofphotocatalysis; 5. Fundamentals of the photo-induced hydrophiliceffect; 6. Brief review of applications; 7. Summary, and Appendix(film preparation methods).

2. Historical overview

We will give a brief overview of the early history ofphotocatalysis, which will be based just on papers that we havebeen able to access, which means that we will almost certainlybe ignoring some important papers. The earliest work that wehave been able to find is that of Renz, at the University of Lugano(Switzerland), who reported in 1921 [17] that titania is partiallyreduced during illumination with sunlight in the presence of anorganic compound such as glycerol, the oxide turning from whiteto a dark color, such as grey, blue or even black; he also foundsimilar phenomena with CeO2,Nb2O5 and Ta2O5. For TiO2, thereaction proposed was:

TiO2 + light Ti2O3 or TiO. (2.1)Baur and Perret, at the Swiss Federal Institute of Technology, werethe first to report, in 1924, the photocatalytic deposition of a silver516 A. Fujishima et al. / Surface Scie

4.1.4. Oxidizing species at the TiO2 surface ...........................4.1.5. Role of molecular oxygen..............................................4.1.6. Effect of crystal face.......................................................4.1.7. Remote photocatalysis ..................................................

4.2. Photocatalytic reactions ...............................................................4.2.1. Decomposition of gaseous pollutants ..........................4.2.2. Decomposition of aqueous pollutants..........................4.2.3. Decomposition of liquid and solid films ......................4.2.4. Photocatalytic sterilization ...........................................

4.3. Visible-light-induced photocatalysis...........................................4.3.1. Non-metal doping..........................................................4.3.2. Origin of visible light photoactivity..............................4.3.3. Activity and stability of N-doped TiO2 photocatalysts

5. Fundamentals of the photo-induced hydrophilic (PIH) effect ...............5.1. Overview .......................................................................................5.2. Mechanisms of the PIH effect.......................................................

5.2.1. Decomposition of organic films ....................................5.2.2. Reductive mechanism ...................................................5.2.3. Oxidative mechanism....................................................5.2.4. Combined redox mechanism ........................................5.2.5. Visible-light-induced PIH effect....................................

6. Brief review of applications......................................................................6.1. Self-cleaning surfaces ...................................................................6.2. Water purification ........................................................................6.3. Air purification ..............................................................................6.4. Self-sterilizing surfaces ................................................................6.5. Anti-fogging surfaces....................................................................6.6. Heat transfer and heat dissipation...............................................6.7. Anticorrosion applications ...........................................................6.8. Environmentally friendly surface treatment ..............................6.9. Photocatalytic lithography ...........................................................Clearly, with so many varied aspects, photocatalysis is nearlyimpossible to review comprehensively. In the present review, wece Reports 63 (2008) 515582

.........................................................................................................................542

.........................................................................................................................545

.........................................................................................................................547

.........................................................................................................................549

.........................................................................................................................551

.........................................................................................................................551

.........................................................................................................................552

.........................................................................................................................553

.........................................................................................................................553

.........................................................................................................................555

.........................................................................................................................555

.........................................................................................................................556

.........................................................................................................................558

.........................................................................................................................560

.........................................................................................................................560

.........................................................................................................................562

.........................................................................................................................562

.........................................................................................................................563

.........................................................................................................................563

.........................................................................................................................563

.........................................................................................................................564

.........................................................................................................................565

.........................................................................................................................565

.........................................................................................................................567

.........................................................................................................................568

.........................................................................................................................569

.........................................................................................................................570

.........................................................................................................................571

.........................................................................................................................571

.........................................................................................................................572

.........................................................................................................................572salt on zinc oxide to produce metallic silver [23]. Even at thisearly date, the authors suspected that both oxidation and reduction

nA. Fujishima et al. / Surface Scie

were occurring simultaneously. The reaction pathway proposedwas this:

ZnO+ h h+ + e (2.2)h+ + OH 1

4O2 + 12H2O (2.3)

e + Ag+ Ag0. (2.4)Three years later, Baur and Neuweiler proposed simultaneousoxidation and reduction to explain the production of hydrogenperoxide on zinc oxide [24].

2h+ + CH2O+ 2OH CO+ 2H2O (2.5)2e + 2H+ + O2 H2O2. (2.6)It was not until many years later that this was absolutelyconfirmed, however. In 1932, Renz reported the photocatalyticreduction of silver nitrate to metallic silver and gold chloride tometallic gold on anumber of illuminated oxides, including TiO2 andNb2O5, [25] and discussed the results in terms of the Baur redoxmechanism.It has been recognized for quite a long time that titania-based

exterior paints tend to undergo chalking in strong sunlight.This means that a non-adherent, white powdery substance tendsto form on the surface, similar to the chalk on a blackboard.This effect was recognized to result from the actual removal ofpart of the organic component of the paint, leaving the titaniaitself exposed. With this background, Goodeve and Kitchener, atUniversity College, London, carried out an excellent study on thephotocatalytic decomposition of a dye on titania powder in air in1938, including absorption spectra and determination of quantumyields (Fig. 2.1) [26]. These authors proposed that titania acts as acatalyst to accelerate the photochemical oxidation and also studieda number of other oxides and speculated on the precisemechanism[27]. In 1949, Jacobsen, at the National Lead Company (USA), alsoattempted to explain the paint chalking phenomenon in terms ofa redox mechanism. He found a correlation between the tendencyof a number of different titania powders to undergo photo-inducedreduction in the presence of organic compounds to their chalkingtendency [18]. The photo-induced reduction was measured as aloss of reflectivity, due to the discoloration of the powder uponreduction, presumably to various oxygen-deficient forms, all theway to Ti2O3. The author proposed a cyclic redox process in whichthe titania was reduced while the organic paint components wereoxidized, followed by re-oxidation of the titania by oxygen fromthe air. The changes experienced by the titania were recognized tobe completely reversible, while those experienced by the organicpaint were recognized to be irreversible, leading to the formationof water-soluble organic acids and CO2. Even though Jacobsen wasapparently unaware of the work of Baur on the redox mechanism,he referred to the 1921 paper of Renz on the photo-reduction ofmetal oxides and proposed the same basic mechanism that hadbeen proposed by Baur; thus, a foundation was laid for later workon the redox mechanism.During the 1950s, the development of photocatalysis shifted to

zinc oxide. In 1953, two studies appeared in which the puzzlingphenomenon of hydrogen peroxide production on zinc oxideilluminated with UV light was studied [28,29], followed by a seriesof follow-up studies in ensuing years [3034]. In these studies, theoverall reactions and mechanisms were completely clarified, andit became apparent that an organic compound was oxidized whileatmospheric oxygen was reduced. Even in the earliest study, anoverall reaction with phenol to produce catechol was proposed,

and the involvement of radical species such as the hydroxyl radical(OH) was also speculated upon [28]. Thus, the original proposalce Reports 63 (2008) 515582 517

Fig. 2.1. Original data of Goodeve and Kitchener showing the photocatalyticdecomposition of a dye (chlorazol sky blue) adsorbed on anatase powder underUV illumination at 365 nm [26]. 1938, Royal Society of Chemistry.

of Baur and Neuweiler was finally confirmed, with the overallreaction:

RHOH+ H2O+ O2 H2O2 + R(OH)2. (2.7)Markham, first at the Catholic University of America and laterat St. Josephs College (USA), continued to study photocatalyticreactions on ZnO, and her papers constitute an impressive, yetunderappreciated, body of work [28,30,31,3539]. This workculminated in a highly intriguing study in which Markham andco-worker Upreti constructed and studied a number of differenttypes of photo-assisted fuel cells, using illuminated ZnO as thephoto-anode with formamide or several alcohols as the organicsubstrates [39]. At the dark cathode (platinum), several differentredox mediators were examined, with atmospheric oxygenultimately being the electron acceptor. The authorsmay have beendiscouraged by the inevitable problem of ZnO photocorrosion,which prevented this system from reaching practical application.It was not until years later that the same basic ideas were re-examined with TiO2. Unfortunately, Markham and Laidler, in theirinitial study in 1953, examined TiO2 but subsequently abandonedit, since it did not produce measurable amounts of hydrogenperoxide [28].It is also interesting to note that Stephens et al. (Wayne

State University), in their study in 1955 of hydrogen peroxideproduction on a large assortment of illuminated semiconductors,but, unfortunately, not TiO2, remarked that zinc oxide and theother catalytic solids should not be abandoned as devices forcapturing solar energy in a form capable of transfer to somechemical system [32]. These authors found that CdS was themostactive photocatalyst, exceeding ZnO in activity.In a study reported in 1956 in Nature, Hindson and Kelly

(Defense Standards Laboratory) reported on the effects of variousrot-inhibitors on tent fabrics for use in Australia. They examinedthe effects of fabric strength after one year of exposure to sunlight.They stated: The effect of anatase is startling. Fabrics containing3% of this pigment lost 90% in strength.In 1958, Kennedy et al., at the University of Edinburgh, studied

the photo-adsorption of O2 on TiO2 in order to try to more fullyunderstand the photocatalytic process [40]. They concluded thatelectrons were transferred to O2 as a result of photoexcitation, andthe resulting reduced form of O2 adsorbed on the TiO2 surface.These authors found a correlation between the ability of the TiO2

sample to photocatalytically decompose chlorazol sky blue (thesame dye used earlier by Goodeve and Kitchener) and the ability

n518 A. Fujishima et al. / Surface Scie

to photo-adsorb O2. This phenomenon is certainly important forphotocatalysis and will be commented upon later.During this period, researchers in Russia were active. Photo-

adsorption of O2 on illuminated ZnO was studied by Terenin andSolonitzin at the University of Leningrad (now University of St.Petersburg) [41]. In a very interesting early work, Filimonov, atthe same institution, compared the photocatalytic oxidation ofisopropanol to acetone on ZnO and TiO2 [42] and concluded thatthemechanism on TiO2 involved an overall reduction of O2 to H2O,while the reduction of O2 on ZnO onlywent as far as H2O2. On TiO2,the surface reactions were proposed to be:

TiO2 + (CH3)2CHOH (CH3)2CO+ TiO+ H2O (2.8)TiO+ 1

2O2 TiO2. (2.9)

Thus, this mechanism is a more detailed version of the Baur cyclicmechanism. It involves the removal of a surface lattice oxygenatom,whichwould be a kind of reduction process. Thismechanismwill be discussed later also, in Section 3.3.In Japan, at the Kyoto Institute of Technology, an early study

(1964) by Kato and Mashio also found that various types oftitania powders had different photocatalytic activities, specificallyto oxidize hydrocarbons and alcohols, simultaneously producinghydrogen peroxide [43]. Interestingly, these authors found thatanatase powders were more active than the rutile ones.In further work at the University of Edinburgh, McLintock and

Ritchie, using gas-phase adsorption measurements, studied thephotocatalytic oxidation of ethylene and propylene at TiO2 [44].This study is one of the first that we have found that shows thatit is possible to oxidize organic compounds completely to CO2 andH2O:

C2H4 + 3O2 2CO2 + 2H2O. (2.10)The mechanism was proposed to involve the production ofsuperoxide from oxygen:

O2 + e O2 . (2.11)Markham et al. had already proposed this reaction to take placeon illuminated ZnO [31]. Work on similar photo-reactions hascontinued into more recent years [45].In an important study for the relationship between photoelec-

trochemistry and photocatalysis, whichwewill come back to later,in Section 3.2, Lohmann, at the Cyanamid European Research Insti-tute, in 1966 published a highly detailed study of the photoelectro-chemical (PEC) behavior of ZnO, both in the presence and absenceof redox couples, including ferro/ferricyanide and methylene blue[46]. He clearly showed that the overall current at the ZnO elec-trode under illumination is the sum of anodic and cathodic cur-rents, the anodic current being a combination of the dissolution ofthe ZnO itself and the oxidation of any redox species present. Thecathodic process was the reduction of O2 to H2O2. This same ap-proach had been introduced in 1938 by Wagner and Traud, at theTechnical University of Darmstadt, to help explain the corrosion ofmetals, coupled with either hydrogen evolution or oxygen reduc-tion [47,48].Another PEC study that wewill mention in this overview is that

of Morrison and Freund, of the Stanford Research Institute, whoalso studied ZnO [49]. These authors also demonstrated in detailthe various situations that arise in the presence and absence ofredox couples. They also showed that oxidation products of someorganic compounds are different in the case of the PEC electrodepoised at the open circuit potential, i.e., with both oxidation andreduction currents balanced, compared to the case of a purely

electrochemical oxidation. This difference was proposed to be dueto the presence of cathodically generated superoxide. This is one ofce Reports 63 (2008) 515582

Fig. 2.2. Photocurrent vs. potential for illuminated rutile single crystal. SCE refersto the saturated calomel electrode, which is 0.241 V vs. the standard hydrogenelectrode (SHE) (taken from Fujishima et al. [214]). 1971, Chemical Society of Japan.

the key points in understanding photocatalysis, and wewill returnto it later.During the late 1960s, one of the present authors, at the

University of Tokyo, began to study the photoelectrochemistry oftitania and found that oxygen gas was evolved at potentials verymuch shifted from the thermodynamic expectation, for example,with an onset of ca. 0.25 V vs. the standard hydrogen electrode(SHE), compared to the standard potential of +0.95 V in pH 4.7aqueous buffer (Fig. 2.2) [50,51]. At first, there was skepticism ofthis result, but then it slowly became accepted. One reason thatthis result was difficult to understand is that the photoexcitationprocess converts the photon energy to chemical energy with littleloss, and thus the photogenerated hole has a very high reactivity, sothat it can react directly with either water or quite robust organicand inorganic compounds. Subsequently, a number of studieswerecarried out in which the photoelectrochemical oxidation processon TiO2 was examined for the competitive oxidation of waterto O2 with the oxidation of a variety of inorganic and organicsubstrates [52,53]. Both types of reactions, of course, involve theuse of light energy to get over an energy barrier, either an overalluphill process, as in the case of O2 evolution, or an overall downhillprocess, as in the case of organic oxidations.With the report of the ability to simultaneously generate hydro-

gen gas in 1972 (see Fig. 2.3) [19], the PEC field started to receivemuch wider attention, due to its implications for solar energy con-version [54,55]. From this point, also, photoelectrochemistry be-came closely associatedwith photocatalysis.We shall return to thistopic later, in Section 3.2, and more carefully describe the detailedrelationships.In this overview, we briefly mention some of the early work

of Bard and co-workers at the University of Texas. Frank and Bardwere the first ones to propose that illuminated TiO2 could be usedfor the purification of water via the photocatalytic decompositionof pollutants [56,57]. They suggested that cyanide and sulfite couldbe photocatalytically oxidized to cyanate and sulfate, respectively.In one of these studies, they found that photocatalytic oxidationscould also occur at other illuminated semiconductors, such asZnO, CdS, Fe2O3 and WO3. The most active semiconductor wasfound to be ZnO [57]. These authors expanded this study to along list of inorganic and organic species [58] and speculated thatphotocatalysis could be a useful approach to both environmentalcleanup and photo-assisted organic synthesis. The Bard group

also suggested that each small illuminated semiconductor particlecould be considered as a PEC cell, with both photo-assisted


Fig. 2.3. Photoelectrochemical cell used in the photolysis of water [19]. 1972, Nature Publishing Group.

oxidation and dark reduction reactions taking place [59]. The Bardgroup also proposed photocatalysis as a way to remove toxicmetals from wastewater [60].For a period of several years, the photocatalysis area continued

to expand as a technology for both the selective oxidation oforganic compounds [22] and the unselective oxidation of organiccompounds for purposes of water purification [1,2,6164] and,to some extent, also air purification [6569]. There have alsobeen reviews and listings of references of work on both airand water purification [4,7072]. For these technologies, it istypically necessary to use powerful ultraviolet (UV) light sources.For passive purification, without special light sources, it becameapparent in the early 1990s that the amount of light present ineither natural sunlight or artificial light was insufficient to processlarge amounts of organic compounds. Therefore, attention wasturned to applications in which a relatively small number of UVphotons could be used to carry out reactions at the TiO2 surface,for example, to decompose thin organic films on solid surfaces orto kill bacteria on surfaces [5,6,7376]. Thus, the focus turned fromwater purification to passive, self-cleaning, self-sterilizing solidsurfaces, which, with sometimes only slight modification, couldalso be used to purify air. For these types of applications, it wasnecessary to develop ways to coat various materials with TiO2films. Such applications included the self-cleaning glass cover forhighway tunnel lamps, as well as a number of others, which havebeen reviewed previously and which will also be reviewed brieflylater in this article.The large number of applications has also generated a renewed

scientific interest in photocatalysis, and indeed on photo-assistedreactions on semiconducting metal oxides in general. One of theways that we have tracked this activity is by looking at thenumber of citations of the 1972 Nature paper on water photolysis(Fig. 2.4(a)). This number of yearly citations has been climbingsteadily over the past ten years or so and of course is correlated

with the number of publications appearing on photocatalysis(Fig. 2.4(b)).ce Reports 63 (2008) 515582 519

Fig. 2.4. (a) Citations per year of the 1972 Nature paper: Electrochemicalphotolysis of water at a semiconductor electrode [19]; (b) Numbers of researcharticles appearing on photocatalysis per year: search results in the period of19722007with the Webof Science (a) by the keyword photocataly (blue bars)and (b) the keywords TiO2 AND photocataly* (green bars). (For interpretationof the references to colour in this figure legend, the reader is referred to the webversion of this article.)

3. Properties of TiO2 materials

3.1. Crystal structures

As often described, there are three main types of TiO2structures: rutile, anatase and brookite. The size dependence ofthe stability of various TiO2 phases has recently been reported[77,78]. Rutile is the most stable phase for particles above 35 nmin size [77]. Anatase is the most stable phase for nanoparticlesbelow 11 nm. Brookite has been found to be the most stablefor nanoparticles in the 1135 nm range, although the Grtzelgroup finds that anatase is the only phase obtained for theirnanocrystalline samples [79,80]. These have different activitiesfor photocatalytic reactions, as summarized later, but the precisereasons for differing activities have not been elucidated in detail.Since most practical work has been carried out with either rutileor anatase, we will focus more attention on these.Rutile has three main crystal faces, two that are quite low

in energy and are thus considered to be important for practicalpolycrystalline or powder materials [81]. These are: (110) and(100) (Fig. 3.1a, b). The most thermally stable is (110), andtherefore it has been the most studied. It has rows of bridgingoxygens (connected to just two Ti atoms). The corresponding Tiatoms are 6-coordinate. In contrast, there are rows of 5-coordinateTi atoms running parallel to the rows of bridging oxygens andalternating with these. As discussed later, the exposed Ti atomsare low in electron density (Lewis acid sites). The (100) (Fig. 3.1b)

surface also has alternating rows of bridging oxygens and 5-coordinate Ti atoms, but these exist in a different geometric

nFig. 3.1. Schematic representations of selected low-index faces of rutile: (a) (110); (b) (100); and (C) (001).

relationship with each other. The (001) face (Fig. 3.1c) is thermallyless stable, restructuring above 475 C [81]. There are double rowsof bridging oxygens alternating with single rows of exposed Tiatoms, which are of the equatorial type rather than the axial type.Anatase has two low energy surfaces, (101) and (001)

(Fig. 3.2a, b), which are common for natural crystals [80,82].The (101) surface, which is the most prevalent face for anatasenanocrystals [79], is corrugated, also with alternating rows of5-coordinate Ti atoms and bridging oxygen, which are at theedges of the corrugations. The (001) (Fig. 3.2b) surface is ratherflat but can undergo a (1 4) reconstruction [82,83]. The (100)surface is less common on typical nanocrystals but is observedon rod-like anatase grown hydrothermally under basic conditions(Fig. 3.2c) [80]. This surface has double rows of 5-coordinate Tiatoms alternating with double rows of bridging oxygens. It canundergo a (1 2) reconstruction [84].Recently, the brookite phase, which is rarer and more difficult

to prepare, has also been studied as a photocatalyst (see later). Theorder of stability of the crystal faces is (010) < (110) < (100)(Fig. 3.3) [85].Recently also, the discovery of high-pressure phases of TiO2

was made [86]. These are expected to have smaller band-gaps but similar chemical characteristics [87]. Their existencewas theoretically predicted and then experimentally proven;specifically, a form of TiO2 with the cotunnite structure wasprepared at high temperature and pressure and then quenched inliquid nitrogen. It is the hardest known oxide.

excess titanium, such as the Magneli phases, TnO2n1, where ncan range from 4 up to about 12 and the titanium oxide layeredcompounds, in which there can be as much as several percentexcess oxygen. The oxygen-deficient Magneli phases, which alsoexist for V, Nb, Mo, Re and W, have been known for many years[8891]. In these compounds, oxygen vacancies are ordered andlead to the slippage of crystallographic planes with respect toeach other; this leads to formation of planes in which, instead ofcorner or edge-shared TiO6 octahedra, there are now face-sharedoctahedra. Fig. 3.4 shows a schematic diagram of this situation. Thecorresponding Ti atoms are then unusually close and can interactelectronically [92]. It has been found recently that laser ablation ofa TiO2 rutile target can produce Magneli-phase nanoparticles [93].There are also quite a number of layered titanate compounds

in which there is an apparent excess of oxygen. For example, thelayered protonic titanate HxTi2 x/4x/4O4 H2O has been preparedand exfoliated into single sheets, termed titania nanosheets.Fig. 3.5 shows (a) a diagram of the layered structure and (b) TEMand AFM images of single sheets.

3.2. Electronic properties

It was reported in 1942 by Earle that rutile and anatase TiO2in the form of powders are n-type semiconductors and thatthe conductivity decreases with increasing O2 partial pressure attemperatures above 600 C [94]. The effect of O2 was explained on520 A. Fujishima et al. / Surface ScieThere are actually quite a variety of different structures forcompounds with compositions close to TiO2, including those withce Reports 63 (2008) 515582the basis of an equilibrium involving thermal release of O2 from thelattice. We recognize today that this leads to the creation of Ti3+

nFig. 3.2. Schematic representations of selected low-index faces of anatase: (a) (101); (b) (100); and (C) (001).

sites, which are responsible for the electronic conductivity. Theactivation energy for the electronic conductivity was found to be1.75 eV for unsintered rutile powder and 1.7 eV for sintered rutilepowder. No evidence for ionic conduction was found. Cronemeyerand Gilleo reported in 1951 that rutile single crystals exhibita band-gap energy of 3.05 eV [95]. Absorption spectra werereported for both normal and slightly reduced crystals. For thelatter, the blue color was based on a very broad absorption thatpeaked at 1.8 m. In the following year, Cronemeyer publisheda very extensive study of the electronic properties of singlecrystal rutile in which the preliminary findings were substantiated[96]. Detailed photoconductivity measurements were made. Darkconductivity and photoconductivity measurements were alsomade on a slightly reduced sample (reduction in H2 at 600 C).Interestingly, therewas found to be amarked hysteresis in the darkconductivity when the sample was raised from room temperatureto 250 C and then cooled back to room temperature. After coolingwith a high applied electric field, the blue color was found to be

possible that the migration involved interstitial hydrogen. This isan ambiguity that has persisted for many years.Strong reduction of various types of samples was examined,

at various temperatures between 300 and 1150 C. The strongreduction turns the samples blueblack. The activation energyfor electronic conduction had already been reported to be0.07 eV at room temperature, to produce a conductivity of ca.1 1 cm1. The conductivities were found to increase withincreasing reduction time. A ceramic sheet sample heated inhydrogen at 800 C was found to experience a weight loss of0.1%, corresponding to a release of oxygen that would provide3 1020 electrons cm3. Hall effectmeasurements showed a closeagreement between the numbers of carriers and those calculatedon the basis of the weight loss, indicating that all of the electronswere electrically active.Breckenridge and Hosler also published extensive work on the

electrical properties of rutile [97]. The effective electron mass wasfound to be anomalously large, 30100 times greater than that ofA. Fujishima et al. / Surface Scieconcentrated at the negative electrode; the author ascribed this tomovement of oxygen vacancies, but this is not confirmed. It is alsoce Reports 63 (2008) 515582 521the free electron. These authors presented convincing argumentsthat the source of electronic conductivity in rutile is Ti3+, which


Fig. 3.3. Schematic representation of the brookite structure (taken from Beltranet al. [85]). 2006, American Chemical Society.

Fig. 3.4. Schematic representation of the crystallographic shear process to formMagneli phases from rutile (taken from Marezio et al. [632]). 2000, Elsevier Science.

results from the loss of oxygen, which produces oxygen vacancies

Ov. It was proposed that these vacancies (valence +2) can have0, 1 or 2 electrons associated with them, with distinct energies,ce Reports 63 (2008) 515582

Fig. 3.5. Layered lepidocrocite-like protonic titanate: (a) schematic representation;(b) AFM image (taken from Shibata et al. [633] and Sasaki [634], respectively). 2007, Royal Society of Chemistry; 2007, Ceramic Society of Japan.

the neutral (fully reduced) vacancy being the lowest, followedby the singly reduced and finally the unreduced vacancy. Theobserved optical absorption edge was proposed to correspond tothe transition between the valence band (based on O2) to theneutral oxygen vacancy, which was considered to be a narrowimpurity band,with ahigh effective electronmass. The band at 3.67eV above the valence band was proposed to be due to Ti4+. Theobserved temperature and oxygen partial pressure dependenceswere fully explained by the equilibrium:

O2 = 12O2 + O2+v + 2e. (3.1)

The electronic properties of rutile and anatase thin films werestudied by Tang et al. [98]. There were large differences in theelectronic conductivities of the two types of films after reductionby heating in vacuum at either 400 or 450 C. The anatase filmsbecame essentially metallic, with no change in conductivity withtemperature. The rutile films, in contrast, retained measurableactivation energies, 0.076 eV for 400 C and 0.06 eV for 450 C. Thedifference in behavior was considered to be due to the followingproperties for rutile: the average static dielectric constant of ca.100, the effective electronmass of 20m0, and the donor state radiusof ca. 2.6 . Since the latter is similar to the distance between Ti4+sites, there is little overlap between donorwave functions. Anatasehas the following properties: static dielectric coefficient of ca. 30,and reduced effective mass of ca. 1 m0, based on an estimateddonor state radius of ca. 15 . Based on optical absorption spectra,

the band-gap energieswere estimated to be 3.0 eV for rutile and 3.2eV for anatase. Forro et al. reported on the electronic properties of


high purity anatase single crystals and found an activation energyfor electronic conduction of 0.004 eV [99].Recently, Hendry et al. pointed out the problem that the exact

nature of electron transport had not been solved, given the verywide range of values of Hall mobilities (0.0110 cm2 V1 s1)and polaron (electron + accompanying lattice distortion) effectivemasses (8me190me, where me is the free electron mass) [100].Part of the problem might involve the presence or absenceof dopants, which were found to decrease the mobility. Theseworkers, using THz spectroscopy with undoped rutile singlecrystals, based on Feynmans analysis [101], found intermediatesize polarons and a mobility of 1 cm2 V1 s1. Thus, theyconcluded that, in TiO2 films such as those used in dye-sensitizednanocrystalline solar cells, the main limiting factor might beinterparticle contacts. However, this assumes a zero dopant level,which is unrealistic. Clearly, even now, further work needs to bedone to clarify this basic issue.Other recent studies have also focused on the electronic

properties from the standpoint of the dye-sensitized solar cellapplication (see, e.g., work of Aduda et al. [102]). For example, theeffect of the morphology of porous titania films on the electrondrift mobility was studied. Other studies have been focused onthe gas-sensing properties, e.g., for hydrogen gas [103105]. Innanostructured form, titania has very high sensitivity for H2,increasing greatly in electronic conductivity in the presence oflow concentrations, for example, an increase of three ordersof magnitude upon introduction of 1000 ppm H2 [103]. Themechanism proposed for the increased conductivity was thoughtto involve the adsorption of hydrogen on the titania surface,rather than incorporation into the bulk. The platinum electrodesthat were used to contact the surface in this study were alsopossibly involved, acting to dissociate the H2 molecule, so that Hatoms could be produced and adsorbed more easily. The subject ofhydrogen interactions with titania has been well studied and willbe treated in more detail in Section 3.5.

3.3. Surface structure studies

It is quite difficult to separate work that has been carried outon the surface structure of titania from work that has been carriedout on surface science in general and also on surface chemistry. Thelatter two subjects will be taken up in the two sections that follow.In this section, we will briefly treat the stoichiometric rutile andanatase surfaces. A detailed treatment has been given as part ofDiebolds extensive review on the surface science [106].Diebold shows how rutile can be cleaved to produce the

commonly shown (110) surface, which is the most stable rutilesurface (see Fig. 3.6). Ramamoorthy et al. carried out theoreticalcalculations on the rutile structure and found the (110) surfaceto be the most stable, based on the fact that it has the leastdangling bonds [81]. The structure shown in Fig. 3.1a is that ofthe unrelaxed bulk and is rather flat, but these authors predictedthat this structure should pucker slightly upon relaxation, with thefive-fold-coordinated Ti atoms depressed by 0.32 a.u. (0.169 ),and the bridging oxygens also depressed, by 0.15 a.u. (0.079 ).Vogtenhuber et al., using similar calculations, found that the five-fold Ti atoms were depressed by 0.180 , the bridging O atomsby 0.156, the planar O atoms by 0.115 and the six-fold Ti atomsby 0.049 [107]. An experimental study that made use of surfaceX-ray diffraction found that the five-coordinate Ti was depressedby 0.16 , while the six-coordinate Ti was pushed out by 0.12 ,and the bridging O was depressed by 0.27 . A total of seventheoretical studies were compared with the experimental surfaceX-ray diffraction results in the review of Diebold [106]. Most of

these studies have agreed on the depression of the five-coordinateTi, on the pushing out of the six-coordinate Ti and the depressionce Reports 63 (2008) 515582 523

Fig. 3.6. Schematic diagram of the cleavage of rutile along the (110) plane (takenfrom Diebold [106]). 2003, Elsevier Science.

of the bridging O. However, a more recent experimental study thatused LEED found that the bridging O is pushed out by 0.12 , incontrast to the earlier experimental study [108]. New theoreticalcalculationsweremostly in agreementwith the LEED study, exceptthat the bridging O position was almost unchanged from theunrelaxed structure [109].The titania surface may undergo significant structural changes

when it is exposed to water. Onishi and co-workers have shownthat single crystal rutile surfaces that have been prepared bystandard methods used for UHV can actually erode and roughenwhen exposed to an aqueous electrolyte [110]. Subsequently,Nakato and co-workers reported on amethod by which atomicallyflat single crystal surfaces could be prepared that were stable inaqueous electrolyte [111]; thismethod involved etching in 20%HF,followed by air-annealing at 600 C.The X-ray crystal truncation rod (CTR) technique has been

used by Zhang et al. to examine the rutile (110) surface in thepresence of pure water and of 1 molal RbOH aqueous solution[112]. Interestingly, the five-coordinate Ti, which had been foundto be significantly depressed in UHV, was found to be depressedto a much smaller degree in pure water (0.051 ) and only slightlydepressed in 1 m Rb+. This is because the terminal position, whichis empty in UHV, is occupied by a water molecule in aqueoussolution and by a hydroxide ion in the alkaline Rb+ solution. Thesix-coordinate Ti, which had been found to be pushed out in UHV,was found to be depressed to a small degree (0.002 in water and0.019 in Rb+ solution). The bridging O, which had been previouslybeen observed to be depressed in the X-ray study and pushed outin the LEED study, was found to be pushed out, by 0.004 in waterand 0.010 in Rb+ solution. All of the displacements were smallerthan those found in vacuum, which the authors propose to be duethe fact that either water molecules or Rb+ ions occupy positionsthat would be occupied in the bulk lattice. Further conclusionsfrom this study will be discussed in Section 3.4, in which we treatinteractions of titania with water.Other predictions from the Ramamoorthy work were in regard

to the relative stabilities of the other rutile single crystal surfaces.The order of stability was found to be (110) > (100) >(011) > (001). This calculation is strictly only valid for 0 K andis for vacuum. Based on the results of the X-ray CTR study foraqueous solution, this ordering might be modified slightly, sincethe stabilities are based in part on the presence of dangling bonds,which would of course not be present any longer in the presenceof water.

3.4. Surface chemical studies: Interactions with water

The interactions of titania surfaces with water have been stud-

ied extensively and have been reviewed [106,113]. The earlierwork involved conventional surface science methods. The studies


Fig. 3.7. High-resolution electron energy loss spectra of rutile (110) at variousdosages of water, starting from the clean surface at bottom (taken from Hendersonet al. [114]). 1996, Elsevier Science.

reported during the past five years on this subject have involvedboth theoretical and experimental studies, the latter including alarge number of scanning tunnelingmicroscopy studies. The inter-actions with water are important to understand, because water,either liquid or vapor, is almost always present in photocatalyticreactions. These interactions are especially important for the laterdiscussion of the photo-induced hydrophilic effect.Much of the work that has appeared over the past decade has

been targeted at the question of whether water is adsorbedmolec-ularly or dissociatively. Going back to one of the pioneering workson this subject, Henderson reported a high-resolution electron en-ergy loss spectroscopy (HREELS)-temperature-programmed des-orption (TPD) study that concluded that the adsorption of wateron rutile (110) is molecular on the stoichiometric surface and dis-sociative on the reduced surface, which is conventionally producedby heat treatment, presumably forming oxygen defects [114]. Theprogression of HREELS spectra as a function of water coverage isshown in Fig. 3.7. In the background spectrum at the bottom, thereis a very small peak at 3690 cm1. This is due to the OH stretchfor OH groups that are not hydrogen-bonded, often called iso-lated OH groups. This vibrational frequency is close to that forOH groups that stick out from the surface of liquid water, withoutbeing hydrogen-bonded to any neighbors, as observed with sum-frequency generation (SFG) spectroscopy [115]. At higher cover-ages, the peaks that appear are shifted to lower wavenumbers,indicating hydrogen bonding. There is no longer any evidence ofthe high wavenumber peak, except at the highest coverage. Thereis also a peak that appears at 1605 cm1, which is due to theHOHbending mode of liquid water. Thus, it is certain that there are wa-ter molecules adsorbed. However, it is not certain whether thereare also dissociated water molecules present that are hydrogenbonded to neighboring water molecules or to bridging oxygens.

There is not much doubt that water dissociates at oxygen

vacancies that are produced by heating in vacuum [113]. Thece Reports 63 (2008) 515582

Fig. 3.8. Schematic diagram of a mixed molecular water-dissociated watermonolayer on the rutile (110) surface (taken from Lindan [127]). 2003, American Institute of Physics.

main question is whether or not this is true for the non-reduced,stoichiometric surface. Theoretical studies have been divided intothose that predict molecular adsorption [116122], those thatpredict dissociative adsorption [123129], and those that also findstability for mixed molecular-dissociative adsorption [122,125127]. This is a particularly difficult problem, since the energydifferences are rather small. One of the studies that has predicteddissociative adsorption also predicts a mixed layer of molecularand dissociated water at higher coverages [127]. Fig. 3.8 is aschematic diagram taken from this paper that shows how themixed monolayer is arranged; the two structures both includehydrogen bonding between a water molecule adsorbed at a five-coordinate Ti site and an OH group adsorbed at an adjacent 5-coordinate Ti site. There is also a weak interaction of the watermolecule with the bridging oxygen. For reference, the diagramsfor purely dissociative and purelymolecular adsorption are shown.Zhang and Lindan have also calculated a theoretical vibrationalspectrum, which we show along with one of the HREELS spectra

from Hendersons work (Fig. 3.9). Even though the simulatedspectrum is significantly shifted upward in wavenumber, the two


Fig. 3.9. (a) A HREELS spectrum for water adsorbed on rutile (111), taken fromHenderson [114]; (b), (c) and (d) show simulated vibrational spectra for the varioustypes of submonolayers and monolayers studied by Zhang and Lindan [127]. 2003, American Institute of Physics.

peaks have an appearance that is similar to the experimentalspectrum. The strong peak at highwavenumber is due to an almostcompletely non-hydrogen-bonded OH, presumably the terminalhydroxyl group, and theweaker, lowerwavenumber peak is due tothe OH group of the water molecule that is bonded to the hydroxylgroup.This type of double-peak structure is rather commonly

observed in experimental infrared spectra for both rutile andanatase powders (Fig. 3.10). The quite sharp peak or peaks athigh wavenumber are due to isolated OH groups, and the broader,lower wavenumber peak or peaks are due to hydrogen-bonded OHgroups. In all of the spectra shown, there is some fine structure.Although not certain, this could be due to the existence of differentcrystal faces, with slightly different geometries for adsorption.Thus, it appears likely that, at least on powders, with coverages

on the order of a monolayer, there could be mixed monolayers.Certainly, there is molecular water, and there must also behydroxyl groups, with the OH group pointing up, normal to thesurface, so that there is little opportunity for hydrogen bonding.However, for powders, there are, of course, a variety of crystalfaces exposed, and distinct situations might be found on each. Thisis expected from the work of Henderson in a comparison of the(110) and (100) surfaces [130]. The latter was found to supportdissociative, while the former was found to support molecularadsorption.Direct evidence for molecular adsorption on stoichiometric

rutile (110) can also be found in STM work that has been targetedat interactions of water with oxygen vacancies. This general topicwill be discussed next.

The background of the work on the interaction of water with

oxygen vacancies on rutile (110) has been given by Hendersonce Reports 63 (2008) 515582 525

Fig. 3.10. A series of three sets of infrared spectra for (a) anatase [635] and (b) [636],(c) [637]) rutile powders acquired at various temperatures and water coverages. In(a), the water coverage decreases with spectrum number, and in (b) and (c), withspectrum letter. In (b) and (c), the original spectrawere obtained in the transmissionmode; all spectra have also been replotted with increasing wavenumber. In (b), themain peaks are listed. 1988, Royal Society of Chemistry; 1987, American Chemical Society; 1971, RoyalSociety of Chemistry.

[113]. One of the interesting aspects is that Ti3+ sites by themselvesdo not have special reactivity; for example, such sites on rutile(100) and on Ti2O3 are not reactive. Only on the (110) surface arethey reactive.The background of the STM work has also been discussed

in the thorough reviews of Henderson [113] and Diebold [106].For example, the latter discusses the problems of distinguishingbetween oxygen vacancies and hydroxyl groups that have beenproduced as a result of a water molecule reacting with anoxygen vacancy. A number of authors concluded that themedium-brightness spots that they observed in STM between bright rowswere due to oxygen vacancies. Diebold et al. pointed out that thereappeared to be two types of defects that were observable on rutile(110), which they termed A and B [131]. Between the rowsof 5-coordinate Ti atoms, which appear bright due to their highelectron density, there are darker rows that are due to the bridgingoxygens. The A type were observed to be significantly brighterthan the B type and were proposed to be oxygen vacancies. TheA defects were removed by scanning the tip at a voltage of +3 V,while the B type remained. The A-type defects were also foundto be quite mobile. Suzuki et al. subsequently reported similarimages and also found that the brighter spotswere removablewitha scan at+3 V [132]. These authors found that the spots were alsoremovable by electron-stimulated desorption. They were also ableto produce additional spots by dosing with atomic hydrogen. Thus,they proposed that the bright spots were due to hydroxyl groups

formed by hydrogen adsorption on bridging oxygens. Brookes et al.also carried out STMmeasurements on rutile (110) and found that,

nvon neighboring bridging oxygens [133]. A second water moleculecan then further catalyze the splitting of the hydroxyl pair inan energetic reaction that can result in one of the protonsjumping several rows away. The initial water-dissociation processis shown in Fig. 3.12. Both the dissociation and splitting processesare also available as movies [134,135]. This paper also correctsthe assignments that had been given in work focused on theinteraction of oxygen (O2)with oxygen defects [136].A paper by Bikondoa et al. appeared in early 2006 [137],

apparently written without the knowledge of the one by Wendtet al., which appeared in 2005 [129]; this paper also clearly showedthe whole situation regarding previously published assignmentsby the various groups. These authors can be credited with havingrecognized, from the beginning, the fact that the bright spotsthat were being observed by various groups were in fact due tohydroxyl groups, either single or double, that had been formed viawater reaction at oxygen vacancies.Additional work has recently appeared on the STM observation

of the reaction of water molecules with oxygen vacancies. Zhanget al. reported that the two bridging OH groups that are produced

appears to be consistent with the report of Wendt et al., in whichit was shown that proton can jump several rows away [133].The implication is that the electrons associated with the originalvacancy are rather localized. All the theoretical results have notbeen in agreement with this picture.The work just described on the interactions of water with

vacancies on rutile (110) has a more general implication, inaddition to the obvious one. Some of the studies that have beencarried out over the years that have discussed reactions of oxygenvacancies may have been actually dealing with bridging hydroxylgroups. Henderson had already pointed out this effect in 1996[114].Wenote also thework ofMezhenny et al., whichwas focused on

the question of whether or not UV light produces oxygen vacancieson the rutile (110) surface [139]. This work showed little effectof ordinary intensity levels of UV light, i.e., similar to those thatare present in sunlight, in producing oxygen vacancies on thesurface. It is likely that this work might have also suffered fromunrecognized background levels of water, since they report STMimages that are characterized by the brighter spots that have been526 A. Fujishima et al. / Surface Scie

Fig. 3.11. STM images of rutile (110) showing (a) oxygen vacancies and ( 2005, Else

when they purposefully dosed the surface with water, it adsorbedwithout dissociation at 150 K and then dissociated at 290 K.Significantly, no terminal features were observed, i.e., there wasno evidence for oxygens or hydroxyls adsorbed on 5-coordinate Tisites,whichwouldhave resulted if thewater haddissociated on thestoichiometric surface. Features were only observed on the dark,bridging oxygen rows. Thus, the authors concluded that the waterhad dissociated at oxygen vacancies.Schaub et al. carried out further work and found that there

were two types of A defects, a smaller type and a larger, brighterone [117]. The latter was assigned to an oxygen vacancy and theformer to a hydroxyl group, based on theoretical calculations.These authors continued to support the idea of water beingdissociatively adsorbed only at oxygen vacancies, in line withtheir theoretical calculations. In further work from this group,however, the assignmentsweremodified [129]. In thiswork, it wasrecognized that the surfaces that had been examined earlier weremostly hydroxylated. Special care was taken to achieve extremelylow levels of background water, and thus it became clear that thedarker featureswere the actual oxygen vacancies and themedium-bright features were individual hydroxyl groups that had beenformed via water dissociation (Fig. 3.11).Subsequent work showed even more clearly how a water

molecule moves along a row of 5-coordinate Ti sites and thenreacts with a vacancy, first producing a pair of hydroxyl groupsfrom the reaction are actually not identical [138]. The one that isproduced at the site of the original oxygen vacancy is not observedce Reports 63 (2008) 515582

b) bridging hydroxyl groups (taken fromWendt et al. [129]).ier Science.

Fig. 3.12. STM images of rutile (110) showing the dissociation of a water moleculeat an oxygen vacancy (taken from [133]). 2006, American Physical Society.

to move, whereas the one that is produced at an adjacent bridgingoxygen due to the addition of a proton, is quite mobile. This resultassigned by later workers to bridging hydroxyls. Nevertheless, ifoxygen vacancies had indeed been produced, there would have


been an increase in the number of such hydroxyls, whereasno increase was observed. This result is in contrast to earlierresults obtained with second harmonic generation and X-rayphotoelectron spectroscopy that concluded that large numbers ofdefects were in fact produced and were reactive toward dioxygen[140]. These issues will be discussed at greater length in the nextsection and in Section 4, which deals with the photo-inducedhydrophilic effect.Work on the interaction of water with titania surfaces has also

been carried outwith other techniques. For example, anatase pow-der was examined with gas-chromatographymass spectrometry(GCMS) and quantum chemical calculations. Experimental evi-dencewas found forwater dissociation, whichwas consistent withthe theoretical calculations; the latter showed that the adsorptionis molecular on the anatase (101) surface and dissociative on the(001) surface.The previously mentioned X-ray CTR study of Zhang et al.

provides rather clear evidence for molecular adsorption for purewater on the stoichiometric rutile (110) surface and dissociativeadsorption in alkaline aqueous solution (pH 12) [112].

3.5. Surface chemical studies: Interactions with dioxygen and otherspecies

It was realized in 1998 that dioxygen does not only react withoxygen vacancies (or possibly, as discussed above, bridging hy-droxyls) to produce a near-stoichiometric surface at temperaturesabove 600 K, but also, at temperatures below 600 K, it can leave be-hind an oxygen atom adsorbed at a 5-coordinate Ti site [141]. Thisrealization led to doubts concerning previously published workthat had found dissociative water adsorption at rutile (110). It alsoled to a reassessment of what had been an accepted procedure forthe preparation of high quality, clean surfaces. The scheme thatEpling et al. proposed to explain the interactions of O2 with oxygenvacancies and subsequent reactionwithwater is shown in Fig. 3.13.In the same paper, these authors also found evidence to support asimilar end product that resultedwhenwaterwas present initially,so that bridging hydroxyls had already been formed.Henderson et al. reported later that O2 can adsorb at a reduced,

i.e., vacancy-containing, rutile (110) surface without dissociationat temperatures below 150 K [142]. One of the more interestingaspects was the observation that O2 can adsorb, probably as O2 ,at a ratio of up to three molecules per oxygen vacancy, whichnecessarily means that it does not have to interact directly withthe vacancy but can reside on an adjacent cation site. Anotherpaper from the same group appeared more recently exploring thereaction of O2 with bridging hydroxyl groups in more detail [143].These authors conclude that the role played byO2 in photocatalysisinvolves specifically this reaction. They also found, in agreementwith their earlier work, that a second monolayer of water blocksthe access of O2 to the bridgingOHgroups, effectively impeding theelectron transfer. On the basis of these results and other studies inwhich superoxidewas generated both on thermally reduced titaniaand on UV-illuminated titania, the authors proposed that bridginghydroxyls are a key intermediate in the photocatalytic process.We agree with this proposal and also propose (see later) that suchbridging hydroxyls can be generated electrochemically.The effect of gas-phase O2 on nanocrystalline titania films has

been studied in terms of the gas-sensing application [144]. Itwas found that the film conductivity decreased in the presenceof O2. This effect could also be related to the effect discussedabove but is more likely to involve the scavenging of bulk trappedelectrons, as discussed in the next section. The authors alsoobserved photo-induced adsorption of O2, a phenomenon that had

been described by Kennedy et al. in 1958, as mentioned in thehistorical overview [40].ce Reports 63 (2008) 515582 527

Fig. 3.13. Schematic diagram of an O2 molecule reacting at an oxygen vacancy onrutile (110) and dissociating, with further reaction with a water molecule (takenfrom Epling et al. [141]). 1998, Elsevier Science.

We include here a brief mention of surface photochemicalreactions involving O2. The work of Thompson and Yates hasemployed the photodesorption of O2 as a means of monitoringthe arrival of photogenerated holes to the surface of a rutile singlecrystal with exposed (110) face [145]. This process is essentiallythe reverse of the photo-adsorption process just alluded to, whichrequires a trapped electron, creating a partially or fully reducedO2, i.e., O2 . The presence of methanol as a hole trapping agentsignificantly decreased the photodesorption. In furtherwork by thesame authors, they proposed a fractal rate law to fit the observedkinetics of the reaction of trapped electrons with trapped holes[146]. It was assumed that the electrons were associated withoxygen vacancies, but this picture may be in some doubt, based onthe ability of trace water to convert these to bridging hydroxyls.

3.6. Bulk chemistryHydrogen

In this section, we briefly review the literature on thebulk chemistry of titania. This subject is mostly limited to theincorporation of elemental hydrogen. It also can include theincorporation of lithium or sodium, but this is beyond the presentscope. The subject of hydrogen incorporation can also includeelectrochemically-induced processes; these will be treated in thenext section. The characteristics of hydrogen as a bulk impurityin titania are central to the understanding of its electrical,electrochemical and photoelectrochemical behavior, which is, inturn central to the understanding of the photocatalytic behavior.An early paper on the optical and infrared absorption spectraof rutile single crystals by Soffer showed evidence for theincorporation of H, already present in the as-received crystal,

nmaxima: 3276 and 3317 cm1 for OH and 2435 and 2463 cm1for OD. That paper had proposed that the peak splitting was dueto slightly differing OTi distances. Johnson et al., however, deniedthis possibility due to the symmetry of the structure. H and Ddoping was carried out by heating in an atmosphere of H2O orD2O, plus O2 at 850 C, or in some cases, H2 or D2 below 550 C.Conduction-band electrons produced a broad absorption band at1.5 m. This was remarked to not be due to conventional freeelectron behavior, which should produce no peak. These authorscarried out a detailed analysis of the various possible bindingsites for H or D within the crystal. The same group reported theuse of the IR absorption band in the precise determination of Hand D concentrations in rutile [150]. DeFord and Johnson studiedthe H/rutile system in detail from the viewpoint of theoreticalsemiconductor and thermodynamic properties [151]. Later, theymade measurements of H and D diffusion using the isotopeexchange technique in order to avoid internal electric fields [152].The diffusion was carried out simply by heating the samples in theappropriate atmosphere (see above) for various times and thenmeasuring the H or D concentrations via the IR absorption. The

high dielectric constant oxide gate materials [157,158]. They findthat the H0 energy level lies above the conduction band in ZnO,TiO2 and SrTiO3, consistent with the fact that H is a shallowdonor in all the three. Work of Park et al. also confirmed theseresults for rutile [159]. Koudriachova et al. have also carried out arecent ab initio quantum chemical calculation on H incorporationin rutile [160]. These authors sought to recheck the binding site,due to the difficulty already mentioned, i.e., the OHO bonddistances were not consistent with established rules relating themto vibrational frequencies. The new calculation found a distortionin the cage surrounding the H atom such that the distancesbecame highly consistent with the correlation. They found thatthe H atoms are most favorably located at ordered positions, asshown in Fig. 3.15a. The lattice expands linearly with increasingH incorporation (Fig. 3.15b).It is also appropriate to mention here theoretical calculations

that were carried out on the surface adsorption of hydrogen asH2 [161]. In that work, it was found that up to one monolayer isadsorbed, with all of the bridging oxygens becoming hydroxylatedand the underlying Ti4+ ions being reduced to Ti3+. Wemight note528 A. Fujishima et al. / Surface Scie

Fig. 3.14. (a) H binding site within the rutile structure and (b) proposed diffusion pat 1979, American Physical Society.

and D, which was introduced by heating at 900 C in D2; thesewere evidenced by the appearance of IR bands at 3277 (main)and 3322 cm1 due to an OH stretch and at 2442 cm1 for thecorresponding OD stretch [147]. The author remarked that thebands are unusually narrow for a solid-state OH stretch and alsodiscussed the possibility of a hydrogen-bonding-related shift.Prior to the work of Hill in 1968 [148], a number of different

studies had suggested that non-stoichiometry in rutile TiO2was associated with increased electrical conductivity, but themechanismwas not clear. It had also been suggested that hydrogenacts as a dopant in rutile. In Hills work, heating rutile crystalsin hydrogen below 600 C led to increases in the bulk hydrogenconcentration, measured by IR. Heating above 650 C in vacuumled to decreases in hydrogen concentration, coupled with oxygenloss to produce water. This led to increased conductivity, probablydue to the formation of faults involving Magneli phases. For theheated crystals, the electrical behavior was modeled involvinga series network of two parallel RC circuits, with one beingassociated with an exhaustion layer, i.e. either a depletion layeror layer that is very low in carriers, as discussed in the next section.Johnson et al. reported further detailed work on the optical and

infrared spectra for H and D incorporated in rutile single crystals[149]. This paper referred to the earlier work of von Hippel et al.that provided presumably more accurate values for the absorptiondiffusion coefficients for H varied from ca. 3 108 cm2 s1 at350 C to ca. 1.7 106 cm2 s1 at 700 C along the c-axis,ce Reports 63 (2008) 515582

h within the solid structure, along a c-channel (taken from Bates and Perkins [154]).

compared with 8 108 cm2 s1 at 698 C along the a-axis. Thediffusion coefficient estimated for room temperature was 1.8 1013 cm2 s1, which was in reasonable agreement with the valuevery roughly estimated by Chester and Bradhurst, in the range10111013 cm2 s1, based on electrochemical insertion.Bates and Perkins measured the infrared frequencies for H, D

and T in rutile TiO2 and carried out a detailed structural analysis,comparing the results with theory for anharmonic oscillators andalso with that of hydrogen bonding [153]. The agreement with thelatter was poor. Bates et al. later published a much more detailedstudy, including a review of the literature up to 1979 [154]. Theycarried out a detailed analysis of themechanism of diffusion of H inrutile. The binding site within the lattice for the proton is shown inFig. 3.14a, and the proposed path for diffusion in Fig. 3.14b. Thebinding site was later confirmed by Klauer and Whlecke usingpolarized Raman [155]. The understanding of this system achievedin this work is excellent. It was concluded that the wavenumbershift of the IR absorption was not due to hydrogen bonding, whichis consistent with the observed sharpness of the band. Instead, itwas proposed to be due to the electrostatic environment withinthe lattice (however, see below). Further work was also publishedon tritium diffusion [156].Peacock and Robertson have carried out quantum chemical

calculations for H in a variety of oxides that are considered asthat this type of surface could in principle be produced thermallyby removing half of the bridging oxygens, followed by exposure to


Fig. 3.15. (a) Illustration of a high stability ordered arrangement of interstitialhydrogen atoms in the rutile structure, with stoichiometry H1/4TiO2 (taken fromKoudriachova et al. [160]). (b) Plot of lattice volume for various numbers ofinterstitial H (open circles) and Li atoms (filled circles) (taken from Koudriachovaet al. [160]). 1994, American Physical Society.

water. It could also be produced in principle electrochemically (seeSection 3.7).It has been found that titania nanotubes respond to the

presence of hydrogen in the gas phase, as already discussed inthe section on electronic properties. In that work, it was notconsidered that hydrogen could actually be absorbed. However,work of Lim et al. showed that for nanotubes that were preparedhydrothermally, there was a reversible uptake of ca. 2% [162]. Theincorporation led to an increase in the IR absorption (3427 cm1),which is significantly higher than that for single crystal rutile. Only75% of the uptake was reversible at room temperature with theremaining 25% requiring temperatures up to 130 C to desorb.A relatively detailed study has been carried out on small

rutile crystals by the use of optical and IR absorption and Ramanscattering [163]. The effect of neutron irradiation was also studied.The incorporation of H in minerals is of interest to geologists,because it affects the macroscopic properties and can be a way

by which water is incorporated in minerals that normally do notabsorb water.ce Reports 63 (2008) 515582 529

Panayotov and Yates recently reported on experiments inwhich they reduced titania pellet samples (Degussa P25) witha source of H atoms [164]. They were able to observe thebroad, featureless background visible-IR spectrum expected forconduction-band electrons. In addition, there were increases inelectronic conductivity. However, there was no discernable IRabsorption at ca. 3280 cm1 for the internally bound OH stretchalready discussed above. The activation energy for the diffusion ofH into the solid was estimated to be 0.09 eV.

3.7. Electrochemical properties

The intrinsic electrochemistry of TiO2 has been studiedcontinuously for a long period, since the first report of Boddy in1968 on the oxygen evolution reaction [165]. This work made useof single crystal rutile electrodes. Interestingly, this work containsa figure in which the photocurrent vs. potential behavior is given,one year prior to the reports of one of the present authors [50,166].This will be discussed further, in the next section.Many aspects of the behavior of TiO2 can be explained on the

basis of a semiconductor model, as already discussed. One of theways of characterizing a semiconductor is to measure its flat-bandpotential electrochemically, for example, with capacitance. In ouroriginal work reported in 1969, we described such measurementsand concluded that the flat-band potential EFB for rutile (001) ispH-dependent, with a relationship similar to the following [167]:

TiO(O)+ H+ + e = TiO(OH) (3.2)with the resulting Nernst relationship:

EFB = 0.00+ FRT ln[H+] = 0.0591pH at 25 C. (3.3)

This result has essentially been confirmed by subsequent workers,with the value at pH 0 being +0.01 0.05 V vs. SHE for rutile(001) [169,171,175]. The pH dependence is typical of the behaviorof most oxide semiconductors and has generally been consideredto be due to a surface acidbase equilibrium for these oxides. Thevalue for anatase is more negative:0.20 V vs. SHE [168].There has been a certain amount of discussion devoted to the

question of exactly how the capacitance measurements should beconducted and how the results should be interpreted, even forsingle crystals. This discussion is interesting, because it displaysthe convergence of an ideal, simple theoretical model with a real,non-ideal complicated material. The model, already discussed,involves an ideal semiconductor with a space charge region whosethickness is dependent upon the potential difference between theFermi level, to which we can assign an electrochemical equivalentEFL and the conduction-band-edge energy, again given on theelectrochemical scale, ECB. The energy difference between ECB andEFL deep within the bulk of the material is dependent upon thecarrier concentration. The space charge capacitance CSC based onthis simple model is given by

1C2SC= 2(E EFB)

0eND(3.4)

where ND is the bulk concentration of donors, with the conse-quence that a plot (MottSchottky) of C2 vs. potential yields thecarrier concentration from the slope and the flat-band potentialfrom the intercept with the potential axis. Many workers havefound that these plots are either non-linear or that the interceptgives a result that is not reasonable, for example, more negativethan that given in Eq. (3.3). Such results have been explained invarious ways, including (1) non-uniform depth profile of carriers

and (2) deep trap levels. There appears to be some consensus onthe merits of the first explanation.


Fig. 3.16. MottSchottky plots for a rutile (001) surface in pH 4.7 buffer: (a) rawdata for a more lightly doped sample (circles), with a solid line showing the fitto Eq. (3.5) (including both space charge CSC and passive layer CPL contributions);(b) calculated intrinsic CSC behavior, after removing the effect of CPL; (c) raw datafor a more heavily doped sample (squares), with a solid line showing the fit toEq. (3.5); (d) calculated intrinsic CSC behavior, after removing the effect of CPL . SCErefers to the saturated calomel electrode,which is 0.241V vs. the standard hydrogenelectrode (SHE) (based on [167]).

A model that can explain the appearance of a steep slope nearthe intercept and a shallower slope at more positive potentialsis that of a thin layer near the surface that has a lower carrierconcentration. This model is reasonable, because, as has beenreported often, the effect can arise as a result of oxidizing etchingtreatments, which could remove electrons from such a surfacelayer, especially if the treatment is conducted for a relativelyshort time. In this case, the overall capacitance behaves as ifthere is a smaller, potential-independent capacitance due to thepassive layer CPL, in series with the potential-dependent spacecharge capacitance. The slope of the upper portion of the curveis somewhat increased from that which is characteristic of thecarrier concentration in the bulk of the material, and the interceptis shifted to the negative. The overall behavior is described by

1C2TOT= 2(E EFB)

0eND+(2(E EFB)0eND

)1/2 ( 1CPL

)+(1C2PL

)(3.5)

which is derived by substituting

1CTOT= 1CSC+ 1CPL

(3.6)

into Eq. (3.4). In our original paper, we reported curves with thistype of double slope (see Fig. 3.16) and made a mathematicalcorrection of this type in order to estimate the true flat-bandpotential [167]. Our explanation at that time invoked a Helmholtzlayer capacitance, which was later correctly disputed by Dutoitet al. [169]. Based on our nitric acid etching procedure usedin that work, however, it is quite reasonable to invoke thepassive layer model just described. DeGryse et al. also found faultwith the involvement of the Helmholtz layer and proposed aninhomogeneous carrier concentration [170]. Tomkiewicz proposeda model based on surface states with energies in the middle ofthe band-gap [171]; this model was criticized by Ullman, whoproposed a surface passive layer [172].It should be noted that this same model has been discussed by

Schoonman et al. [173], but the equation given in that paper wasdifferent, in that the cross-term was not included, leading to theerroneous result that the curve is simply raised but has the sameslope. If theMottSchottky behavior does not fit this simplemodel,it is still possible that it can be fitted with a more complicatedmodel, in which the slope at any given point in the curve can

be related to the carrier concentration at a certain depth; thus acomplete concentration profile can be estimated [174].ce Reports 63 (2008) 515582

Finklea has given an excellent summary of the practicalmethods of obtaining linear MottSchottky plots with correctintercepts [175]. The opposite situation can also arise if a surfacelayer exists with a higher concentration of carriers compared tothe bulk. This leads to a shallow slope near the flat-band potential,followed by a steeper slope at more positive potentials.In certain cases, if the carrier concentration is sufficiently

high, the space charge capacitance can approach that of theelectrical double layer, and the Helmholtz capacitance must alsobe considered.It can be argued in general that the behavior of TiO2 is simply

not ideal, and, in each individual case, a full impedance treatmentis necessary to extract the space charge capacitance or even toevaluate whether or not it exists. Similar situations also exist foroxide films; the situation is particularly complicated for oxide filmsgrown on metals, for example, iron and zirconium [176].For certain types of TiO2 electrodes, it may be doubtful that the

behavior can be strictly described with a semiconductor model.For example, the MottSchottky plots may exhibit a significantfrequency dispersion, either with or without non-linearities. Thiscan signal the fact that a different type of model might bemore appropriate. In most cases, it is advisable to examine thefull impedance spectrum over a range of potentials in orderto determine whether or not the behavior does in fact followthat expected for a semiconductor. One specific type of behaviorthat has been observed is that of an electrochromic film, inwhich all of the charge injected into the film is associated withincreased absorption of visible light (see Refs. [177181] and laterdiscussion).The electrochemical impedance spectral (EIS) behavior can

be quite powerful in elucidating the behavior of electroactivematerials, i.e., those that can undergo electron transfer reactionswithin pores and even within the solid material itself. Nogamiexamined the EIS behavior of several single crystal rutile samplesand obtained sets of straight lines, with varying slopes in Bodeamplitude (log |Z | vs. log frequency) plots [182]. He explainedthis behavior on the basis of a disordered layer on the surfaceof the single crystal. The Bode plots show straight lines for thelogarithmof themagnitude of the impedance (absolute value of thecomplex impedance) vs. the logarithm of the frequency. At morepositive potentials, the slopes of the lines are close to 0.7, while,at more negative potentials, the slopes are close to 0.5. For a purecapacitance, the slope should be 1.0, while, for a pure resistance,it should be 0.0. For intermediate cases, there are both resistanceand capacitance distributed through the material. The simplestelectrical case is that of the uniform semi-infinite transmissionline, in which there is a ladder of constant resistances andcapacitances (Ref. [183] and references therein). This type ofbehavior has been shown to be followed by a porous mediumwith cylindrical pores. A rate-limiting diffusion process inwhich anelectroactive species diffuses through a uniform medium behaveselectrically in the sameway, and thus there is some ambiguity as towhich process is actually operating. In either case, the slope shouldbe 0.5, as observed by Nogami for more negative potentials.For the general case, in which the slope of the Bode plot can

have an arbitrary value between 0.5 and 1, the same type ofRC ladder is also valid, and the physical picture is also similarto that for the uniform network, but with pores that are non-cylindrical. For example, Wang and Bates have shown that horn-shaped pores can give rise to Bode slopes in this range [184]. Thus,the results of Nogami could be explained by a situation in whichhorn-shaped pores of larger diameter (lower roughness factor)exist at higher potentials; at intermediate potentials, the diametersbecome smaller (higher roughness factor); finally, at the most

negative potentials, the pores behave as if they are cylindrical andlonger than the penetration length of even the lowest frequencies

nslopes are often referred to as constant phase elements, meaningsimply that the phase angle, which is 0 for a pure resistance, 90for a pure capacitance and 45 for the uniform transmission line, isconstant but is not one of these standard ones. In any case, it seemsclear thatNogamis conclusion, i.e., that the frequencydispersion oftheMottSchottky plots is due to the presence of a disordered layeron the single crystal surface, might be valid. Frequency dispersionis a result of the mixing of resistive (i.e., electronically conducting)character with capacitive (i.e., ionically conducting) character.Furthermore, his proposal that the behavior involves surface

states might also be valid. Wang also has shown that constant-phase angles between 90 and 45 can be obtained if theresistances and capacitances in the ladder are non-uniform,e.g., with the resistance increasing and the capacitance decreasingin a logarithmic manner [183]. Even a 45 angle can result witha non-uniform situation. A non-linear gradient of resistance andcapacitance could well result if there is a similar gradient ofconcentration of Ti3+ sites within the film. Cahan and Chen havediscussed such gradients in the passive oxide film on iron. Theyhave also pointed out that these gradients can naturally give riseto linear log current vs. potential behavior (linear Tafel slope; seebelow) that mimics that for rate-limiting electron transfer [185

1-s2.0-S0167572908000757-main

Documents

surface chemical studies

surface structure studies

electronic properties

surface science reports

properties of tio2 materials

fundamentals of photocatalysis

mechanisms of photocatalysis

bulk chemistryhydrogen