The Role of Surface Texture on the Photocatalytic H 2 Production on TiO 2

: It has been often reported that an efﬁcient and green photocatalytic dissociation of water under irradiated semiconductors likely represents the most important goal for modern chemistry. Despite decades of intensive work on this topic, the efﬁciency of the water photolytic process under irradiated semiconductors is far from reaching signiﬁcant photocatalytic efﬁciency. The use of a sacriﬁcial agent as hole scavenger dramatically increases the hydrogen production rate and might represent the classic “kill two birds with one stone”: on the one hand, the production of hydrogen, then usable as energy carrier, on the other, the treatment of water for the abatement of pollutants used as sacriﬁcial agents. Among metal oxides, TiO 2 has a central role due to its versatility and inexpensiveness that allows an extended applicability in several scientiﬁc and technological ﬁelds. In this review we focus on the hydrogen production on irradiated TiO 2 and its fundamental and environmental implications.


Introduction
From the perspective of our ancestors (and of the billions of underprivileged people living in the developed and developing world), we are living an unprecedented golden age thanks to the large availability of fossil fuels that provide economical energy sources. This is translated into a large pool of available goods and services, which are becoming more and more available to an increasing share of the world population. The lifestyle of wealthy nations is simultaneously very appealing and dependent on energy. This pushes an increasing fraction of the population to adopt it, and to cope with its direct and indirect drawbacks, namely stress, obesity, pollution and finally, global warming.
In the 1970s, the Kippur and the Iran-Iraq wars created a sense of awareness about the dependence on fossil fuels and the consequences of their eventual depletion, and pushed the research to find alternatives. Almost 50 years later, 85% of the world's primary energy is still coming from fossil fuels [1], but we are presently facing a subtler threat related to global warming. To have at least a 50% chance of keeping warming below 2 • C throughout this century, two-thirds of the global fossil fuels reserves must remain untouched [2,3]. These "forbidden" reserves represent the so-called "unburnable carbon". This should further push researchers, but also policy makers, to find sustainable alternatives, because the transition should be completed significantly before the depletion of fossil fuels reserves.
Then again this poses another problem to citizens and policy makers when comparing different technologies. Usually such comparisons are based on performance and costs. Liquid fossil fuels are relatively cheap and have an unparalleled energy density, compared with hydrogen and batteries [4].
2 H 2 O (l) → 2 H 2(g) + O 2(g) ∆G 0 = 237 kJ mol −1 (1) 2 H 2 O (g) → 2 H 2(g) + O 2(g) ∆G 0 = 229 kJ mol −1 This means that the reaction is not spontaneous at room temperature, but requires a significant energy input, which can be chemical, electrochemical or electromagnetic ( Figure 1). To perform water photosplitting, the energy required will be provided in the form electromagnetic energy, i.e., light. Considering that two electrons are required to oxidize one water molecule to O 2 , the energy required is to be delivered by two photons, then the minimum photon energy will be 1.23 eV, i.e., ≈1010 nm: where N A is the Avogadro constant and e is the electron's elementary charge. But, because O 2 is the product of water oxidation, the whole process is actually a four-photon, four-electron process. Consequently, to promote water cleavage in an electrolyzer, the energy constraint will be overcome Catalysts 2019, 9, 32 3 of 28 thanks to an external bias, which must be larger than 1.23 V, because of the standard potentials of the couples: 2 H + (aq) + 2 e − → H 2(g) E 0 = 0.000 V 4 H + (aq) + O 2(g) + 4 e − → 2 H 2 O (g) E 0 = 1.229 V (4) Considering that ∆G 0 = nFE 0 (5) with n = 2 in the case of water, Equation (5) further demonstrates how the thermodynamic requirements of Equations (1)-(4) are linked together [7]. Due to its positive Gibbs energy variation, water photosplitting is a photosynthetic reaction. For this reason, it could be legitimately considered a form of artificial photosynthesis, and, in fact, the anodic semireaction involves the oxidation of water and the formation of molecular oxygen, while on the cathodic side there is the formation of a fuel, H 2 , finalizing the transformation of electromagnetic energy into chemical energy.
In the case of hydrogen production from photoreforming, a hydrogen-containing organic compound-such as methanol-produces hydrogen and carbon dioxide: CH 3 OH (l) + H 2 O (l) → 3 H 2(g) + CO 2(g) ∆G 0 = 8 kJ mol −1 (6) In this latter case the Gibbs free energy variation is still positive, but its value is almost two orders of magnitude less than in the case of Equation (1). The reason is the great stability of carbon dioxide, which has Gibbs' free energy of formation, ∆G 0 f , of −394 kJ mol −1 [8]. On the other hand, water also has a quite large ∆G 0 f (water formation reaction is the reverse of Equation (1)). Therefore, depending on the particular organic compound, its reforming could be either endergonic or exergonic, according to the number of water and carbon dioxide molecules involved in the reaction. For example, formic acid and glycerol are characterized by a H 2 O to CO 2 ratio of 0 and 1, respectively, and in both cases their reforming is a downhill reaction: HCOOH → H 2 + CO 2 ∆G 0 = −83 kJ mol −1 C 3 H 8 O 3 + 3 H 2 O → 7 H 2 + 3 CO 2 ∆G 0 = −300 kJ mol −1 (8) whereas in the case of ethanol and propan-2-ol, where the H 2 O to CO 2 ratio is larger than 1, the Gibbs free energy of their reforming is positive: Therefore, when the previous reforming reaction are promoted by light we can be in the presence of either photosynthetic or photocatalytic processes, depending on the ∆G. Nevertheless, we feel that the claim of artificial photosynthesis in the cases of reactions (6), (9) and (10) would be exaggerated, even though their positive ∆G would make such a claim technically correct. In the case of organic reforming the oxidation potentials are less positive, and the kinetics of reaction less complicated compared with water oxidation, even though also in these cases multielectron and multiproton transfers are involved ( Figure 1). Accordingly, also the minimum photon energy required will be lower. Focusing our attention to TiO2, its band gap is larger than 3 eV; therefore, its excitation requires UV photons, or in terms of energy 300 kJ mol −1 , which is larger than the ΔG 0 of all the reactions considered until now ( Figure  2a). Considering that reactions 1-10 involve all multiple electron transfers, the use of TiO2 in hydrogen photo production implies a significant waste of photon energy [9]. Returning to the energy diagram of Figure 1 we can argue that the energy loss is mainly due to the valence band, which is considerably more oxidizing compared with the reduction potential of the organic compounds considered until now.  [10][11][12][13] Monoelectronic couples are in red, bielectronic in black, the four-electron couple O2/H2O is in blue. (b) Simplified pictorial scheme of the methanol reforming on a irradiated Pt/TiO2 particle. (c) Photoinduced charge separation and charge distribution in Pt/TiO2 nanocomposites. EF, dark and EF, light represent the pseudo-Fermi levels in the dark and under irradiation, respectively. Adapted with permission from [14]. Copyright 1979 American Chemical Society.
Conversely, the conduction band potential is very favorable, because it allows water reduction with a reasonable overpotential, which ensures a sufficient drive to carry out the reaction, without important losses. To minimize energy losses of photocatalytic reforming on TiO2, hydrogen production should be coupled with the oxidation of recalcitrant compounds, when this reaction is kinetically limited or other strategies would not work, such as in the case of toxic compounds, aromatics, and antibiotics (scarcely removed in traditional wastewater treatment plants). Alternatively, the valorization of biomasses in remote regions could be another valuable and promising option, also because in that case H2/O2 separation could be ruled out [15].

History and Developments
Usually the first example of water photosplitting is attributed to Fujishima and Honda and their 1972 paper published on Nature [16]. In their contribution rutile TiO2 is used as a photoanode for the evolution of oxygen, while hydrogen evolution occurred at the platinum cathode. Their approach is therefore photoelectrochemical, because both irradiation and an external bias were required. There are only five references in their paper: the first is a 1910 paper [17], in which it was   Accordingly, also the minimum photon energy required will be lower. Focusing our attention to TiO 2 , its band gap is larger than 3 eV; therefore, its excitation requires UV photons, or in terms of energy 300 kJ mol −1 , which is larger than the ∆G 0 of all the reactions considered until now (Figure 2a). Considering that reactions 1-10 involve all multiple electron transfers, the use of TiO 2 in hydrogen photo production implies a significant waste of photon energy [9]. Returning to the energy diagram of Figure 1 we can argue that the energy loss is mainly due to the valence band, which is considerably more oxidizing compared with the reduction potential of the organic compounds considered until now.  Accordingly, also the minimum photon energy required will be lower. Focusing our attention to TiO2, its band gap is larger than 3 eV; therefore, its excitation requires UV photons, or in terms of energy 300 kJ mol −1 , which is larger than the ΔG 0 of all the reactions considered until now ( Figure  2a). Considering that reactions 1-10 involve all multiple electron transfers, the use of TiO2 in hydrogen photo production implies a significant waste of photon energy [9]. Returning to the energy diagram of Figure 1 we can argue that the energy loss is mainly due to the valence band, which is considerably more oxidizing compared with the reduction potential of the organic compounds considered until now.  [10][11][12][13] Monoelectronic couples are in red, bielectronic in black, the four-electron couple O2/H2O is in blue. (b) Simplified pictorial scheme of the methanol reforming on a irradiated Pt/TiO2 particle. (c) Photoinduced charge separation and charge distribution in Pt/TiO2 nanocomposites. EF, dark and EF, light represent the pseudo-Fermi levels in the dark and under irradiation, respectively. Adapted with permission from [14]. Copyright 1979 American Chemical Society.
Conversely, the conduction band potential is very favorable, because it allows water reduction with a reasonable overpotential, which ensures a sufficient drive to carry out the reaction, without important losses. To minimize energy losses of photocatalytic reforming on TiO2, hydrogen production should be coupled with the oxidation of recalcitrant compounds, when this reaction is kinetically limited or other strategies would not work, such as in the case of toxic compounds, aromatics, and antibiotics (scarcely removed in traditional wastewater treatment plants). Alternatively, the valorization of biomasses in remote regions could be another valuable and promising option, also because in that case H2/O2 separation could be ruled out [15].

History and Developments
Usually the first example of water photosplitting is attributed to Fujishima and Honda and their 1972 paper published on Nature [16]. In their contribution rutile TiO2 is used as a photoanode for the evolution of oxygen, while hydrogen evolution occurred at the platinum cathode. Their approach is therefore photoelectrochemical, because both irradiation and an external bias were required. There are only five references in their paper: the first is a 1910 paper [17] Figure 2. (a) Energy diagram showing the conduction and valence band potentials of TiO 2 anatase, C1 and water-related redox couples at pH 7 and 298 K [10][11][12][13] Monoelectronic couples are in red, bielectronic in black, the four-electron couple O 2 /H 2 O is in blue. (b) Simplified pictorial scheme of the methanol reforming on a irradiated Pt/TiO 2 particle. (c) Photoinduced charge separation and charge distribution in Pt/TiO 2 nanocomposites. E F, dark and E F, light represent the pseudo-Fermi levels in the dark and under irradiation, respectively. Adapted with permission from [14]. Copyright 1979 American Chemical Society.
Conversely, the conduction band potential is very favorable, because it allows water reduction with a reasonable overpotential, which ensures a sufficient drive to carry out the reaction, without important losses. To minimize energy losses of photocatalytic reforming on TiO 2 , hydrogen production should be coupled with the oxidation of recalcitrant compounds, when this reaction is kinetically limited or other strategies would not work, such as in the case of toxic compounds, aromatics, and antibiotics (scarcely removed in traditional wastewater treatment plants). Alternatively, the valorization of biomasses in remote regions could be another valuable and promising option, also because in that case H 2 /O 2 separation could be ruled out [15].

History and Developments
Usually the first example of water photosplitting is attributed to Fujishima and Honda and their 1972 paper published on Nature [16]. In their contribution rutile TiO 2 is used as a photoanode for the evolution of oxygen, while hydrogen evolution occurred at the platinum cathode. Their approach is therefore photoelectrochemical, because both irradiation and an external bias were required. There are only five references in their paper: the first is a 1910 paper [17], in which it was stated that water photolysis requires 190 nm or shorter wavelength light, while the other four are previous reports by the same authors to a more specialized and narrow audience [18][19][20][21]. The small bibliography and the large rate of self-citation were due to the seminal nature of their report, but also to the smaller number of people, with different habits towards scientific publication, working in the field compared to nowadays. These two factors are still determining the reception of that work by the scientific community. Due to its seminal character the article received the first citation only two years later, but then interest rose up in the end of the decade and does not appear to slow down (Figure 3a) [22].
Considered as a whole, the literature concerning photoreforming on TiO 2 is increasing almost exponentially (Figure 3b) as water photosplitting publications are. Nevertheless, the sole increase of publication rate as a function of time in not significant, because of the unprecedented increase in the volume of the scientific literature occurred in the last decades. Indeed, even smallpox related publications show an increasing trend, even though the last recorded case dates back to 1977 [23]. More significantly, the increasing trend of water photosplitting publications is more similar to that recorded for lithium ion batteries, whereas hydrogen storage is showing a less steep increase since 8-10 years.
With the buildup of interest, the number of reports, experiments, and observations increased, empowering the understanding of the processes and the mechanisms involved. Compared to the beginning of the 1970s, nowadays we have a much vaster literature dealing with hydrogen production, heterogeneous photocatalysis and artificial photosynthesis. But even more than 50 years ago the first evidence that species such as CN − , I − [20] Br − , Cl − , Ce 3+ , Mn 2+ [24,25] and certain organics [26] could be oxidized by irradiated semiconductor slurries such as ZnO, TiO 2 and CdS thanks to charge carrier transfer [27] were already recorded. Moreover, Grätzel and coworkers sketched and demonstrated different strategies to evolve hydrogen in the presence (and also in the absence [28]) of TiO 2 , even with visible light only [29,30]. The electron transfer from TiO 2 conduction band to N 2 was also reported [31]. In 1979, Bard published a comprehensive review [32] rationalizing the three possible approaches for the exploitation of electromagnetic energy with photoelectrochemical cells (PEC) based on semiconductors, namely photovoltaic, photosynthetic and photocatalytic. In photovoltaics, the goal is to obtain electricity from light, and the main development in the field came from dye-sensitized solar cells, which are regenerative photoelectrochemical cells capable of developing an electrical bias under irradiation and convert solar energy with 11.9% record efficiency [33].
Catalysts 2019, 9, x FOR PEER REVIEW 5 of 28 stated that water photolysis requires 190 nm or shorter wavelength light, while the other four are previous reports by the same authors to a more specialized and narrow audience [18][19][20][21]. The small bibliography and the large rate of self-citation were due to the seminal nature of their report, but also to the smaller number of people, with different habits towards scientific publication, working in the field compared to nowadays. These two factors are still determining the reception of that work by the scientific community. Due to its seminal character the article received the first citation only two years later, but then interest rose up in the end of the decade and does not appear to slow down ( Figure 3a) [22]. Considered as a whole, the literature concerning photoreforming on TiO2 is increasing almost exponentially (Figure 3b) as water photosplitting publications are. Nevertheless, the sole increase of publication rate as a function of time in not significant, because of the unprecedented increase in the volume of the scientific literature occurred in the last decades. Indeed, even smallpox related publications show an increasing trend, even though the last recorded case dates back to 1977 [23]. More significantly, the increasing trend of water photosplitting publications is more similar to that recorded for lithium ion batteries, whereas hydrogen storage is showing a less steep increase since 8-10 years.
With the buildup of interest, the number of reports, experiments, and observations increased, empowering the understanding of the processes and the mechanisms involved. Compared to the beginning of the 1970s, nowadays we have a much vaster literature dealing with hydrogen production, heterogeneous photocatalysis and artificial photosynthesis. But even more than 50 years ago the first evidence that species such as CN − , I − [20] Br − , Cl − , Ce 3+ , Mn 2+ [24,25] and certain organics [26] could be oxidized by irradiated semiconductor slurries such as ZnO, TiO2 and CdS thanks to charge carrier transfer [27] were already recorded. Moreover, Grätzel and coworkers sketched and demonstrated different strategies to evolve hydrogen in the presence (and also in the absence [28]) of TiO2, even with visible light only [29,30]. The electron transfer from TiO2 conduction band to N2 was also reported [31]. In 1979, Bard published a comprehensive review [32] rationalizing the three possible approaches for the exploitation of electromagnetic energy with photoelectrochemical cells (PEC) based on semiconductors, namely photovoltaic, photosynthetic and photocatalytic. In photovoltaics, the goal is to obtain electricity from light, and the main development in the field came from dye-sensitized solar cells, which are regenerative photoelectrochemical cells capable of developing an electrical bias under irradiation and convert solar energy with 11.9% record efficiency [33].  for photocatalysis do not convert energy, but employ light to overcome energetic barriers and promote thermodynamically favored reactions, such as CN − oxidation, or reactions (7) and (8). Apart from providing many examples dating back to the 1950s, Bard outlined the similarities between semiconductor photoelectrodes and irradiated semiconducting slurries, which can, in principle, promote the same reactions. An irradiated semiconductor particle behaves as a short-circuited and miniaturized photoelectrochemical cell operating without bias. It is then evident that with the semiconductor slurry the prevention of recombination is an extremely important task to ensure high efficiency [32].

Mechanistic Considerations
Since the earliest reports, it has been outlined that semiconductors such as ZnO or TiO 2 were able to oxidize organics and anions such as I − [20] and CN − , and reducing O 2 to form hydrogen peroxide or water [34]. The addition of cocatalysts such as Pt and other noble metals improved the rates and allowed the evolution of hydrogen in anaerobic conditions. Other cocatalyst, such as RuO 2 , allowed water oxidation to O 2 [35]. From these evidences, a first picture emerged regarding the mechanism of organics photoreforming on noble metal-modified semiconductors ( Figure 2b). The photogenerated electron-hole couples are separated; the holes are transferred to the organics [36,37], while the electrons are captured by the noble metal, where they can promote water reduction and hydrogen evolution thanks to the low activation energy displayed by most noble metals and especially Pt [38,39]. At this stage is worth clarifying that the mechanisms of reaction in the dark, or under visible irradiation, where plasmonic resonance may be activated, could be radically different, as recently reviewed by Panayotov and Morris [40].
This pictorial unpolished mechanism is actually in good agreement with the experimental evidences collected during the years. It has been shown that hole trapping is a fast process, requiring less than 50 fs to happen, and that hole transfer to surface hydroxyl requires ≈250 ns, while in the presence of methanol the hole transfer significantly accelerates, being completed within 1 ns [41,42].
Conversely electron trapping is considerably slower, requiring ≈160 fs. Electron lifetime is in the order of the microsecond in vacuum or in the presence of water vapor, but it is extended to more than one second in the presence of methanol vapors. This demonstrates how electron and hole dynamics could be radically different when specific scavengers are introduced, and that their recombination, predominantly non-geminate and showing second order kinetics, strongly depends on these factors [43].
When TiO 2 is decorated with noble metal particles, electron trapping becomes at least 10-15 times faster compared with pristine TiO 2 , whereas hole capture seems almost insensitive to noble metals [43]. This evidence is coherent with previous picture of electron capture by the metal where hydrogen evolution takes place (Figure 2b). A more refined treatment was carried out by Kamat and coworkers [14]: upon irradiation the Fermi level of the semiconductor is displaced towards more negative potential, because of the faster transfer of holes and to the accumulation of photoelectrons ( Figure 2c). The Fermi level shift in the semiconductor causes an identical shift in the noble metal particles. If this shift is such that water reduction is triggered, hydrogen evolution will occur. This picture is coherent with the noble metal particle forming an ohmic contact with the semiconductor, rather than a Schottky junction, which should repel semiconductor electrons, especially in the case of noble metals, which have large work functions.
Because hydrogen evolution occurs on the noble metal, and because of the analogy between irradiated semiconductor slurries and photoelectrochemical cells, the study of the mechanism can take advantage from the enormous amount of work done by electrochemists on the hydrogen evolution reaction (HER) on metal electrodes ( Figure 4).
HER is an archetypical reaction, in which two electrons and two protons are combined to form the simplest molecule H 2 . The reaction requires two steps; the first is the Volmer step, the reductive adsorption or electrochemical discharge [44]: In which a proton from solution is reduced to become an adsorbed hydrogen atom on the metal surface. Of course, the Volmer step needs empty metal sites M-to proceed. At this point a bifurcation occurs, because the reaction may advance through the Tafel pathway, or H-H recombination: Or through the Heyrovsky pathway, also known as reductive or electrochemical desorption [38]: HER is an archetypical reaction, in which two electrons and two protons are combined to form the simplest molecule H2. The reaction requires two steps; the first is the Volmer step, the reductive adsorption or electrochemical discharge [44]: In which a proton from solution is reduced to become an adsorbed hydrogen atom on the metal surface. Of course, the Volmer step needs empty metal sites M-□ to proceed. At this point a bifurcation occurs, because the reaction may advance through the Tafel pathway, or H-H recombination: Or through the Heyrovsky pathway, also known as reductive or electrochemical desorption [38]: The actual mechanism depends primarily on the metal. For example, on Pt, Pd and Rh HER proceeds mainly via the Tafel pathway. Interestingly, the same mechanistic features are found on Pt, Pd, and Rh decorated TiO2 slurries, further confirming that hydrogen evolution is not only occurring on the metal cocatalyst, but also that the metal nanoparticles behave very similarly to their macroscopic electrodes counterparts. Moreover, poisoning the cocatalyst or in its absence HER is hindered, and the mechanism modified [38,45]. Even though common wisdom would suggest that the Volmer step could be limiting on metals with lower H adsorption energy, while on metals forming strong M-H bonds the mechanism could shift towards the Heyrovsky pathway, the actual mechanism depends on the electronic structure of the metal catalyst [46,47]. Apart from the chosen metal element, other parameters such as particle size, defectivity and exposed surfaces can therefore determine the overall rate. Moreover, the selection of the Tafel or the Heyrovsky pathway depends also on H surface coverage, which is potential dependent. While the working potential could be modulated if HER were carried out on a macroscopic electrode, it becomes uncontrolled, and usually unknown, in the case of a semiconductor slurry. Nevertheless, the Fermi level of the system will respond to changes in the identity and concentration of the organic substrate and of the incident photon flux. When the slurry is concentrated, the optical path through the reactor is long, the slurry illumination will not be uniform due to scattering, potentially leading to changes in relative H2 reaction pathways along the optical path [48].
Hydrogen photoreforming is the result of the complex interplay among many elementary steps, such as light absorption, charge carrier recombination, hole transfer to substrate, which in turn can give current doubling or act as a recombination center, electron transfer from the semiconductor to the metal, proton reduction (electrochemical discharge) and hydrogen evolution on the metal, which occurs via catalytic or electrochemical desorption. In the case of readily oxidized organics giving current doubling, it is likely that the hydrogen evolution on the metal represents the ratedetermining step. Conversely, in the case of recalcitrant compounds, which can act as The actual mechanism depends primarily on the metal. For example, on Pt, Pd and Rh HER proceeds mainly via the Tafel pathway. Interestingly, the same mechanistic features are found on Pt, Pd, and Rh decorated TiO 2 slurries, further confirming that hydrogen evolution is not only occurring on the metal cocatalyst, but also that the metal nanoparticles behave very similarly to their macroscopic electrodes counterparts. Moreover, poisoning the cocatalyst or in its absence HER is hindered, and the mechanism modified [38,45]. Even though common wisdom would suggest that the Volmer step could be limiting on metals with lower H adsorption energy, while on metals forming strong M-H bonds the mechanism could shift towards the Heyrovsky pathway, the actual mechanism depends on the electronic structure of the metal catalyst [46,47]. Apart from the chosen metal element, other parameters such as particle size, defectivity and exposed surfaces can therefore determine the overall rate. Moreover, the selection of the Tafel or the Heyrovsky pathway depends also on H surface coverage, which is potential dependent. While the working potential could be modulated if HER were carried out on a macroscopic electrode, it becomes uncontrolled, and usually unknown, in the case of a semiconductor slurry. Nevertheless, the Fermi level of the system will respond to changes in the identity and concentration of the organic substrate and of the incident photon flux. When the slurry is concentrated, the optical path through the reactor is long, the slurry illumination will not be uniform due to scattering, potentially leading to changes in relative H 2 reaction pathways along the optical path [48].
Hydrogen photoreforming is the result of the complex interplay among many elementary steps, such as light absorption, charge carrier recombination, hole transfer to substrate, which in turn can give current doubling or act as a recombination center, electron transfer from the semiconductor to the metal, proton reduction (electrochemical discharge) and hydrogen evolution on the metal, which occurs via catalytic or electrochemical desorption. In the case of readily oxidized organics giving current doubling, it is likely that the hydrogen evolution on the metal represents the rate-determining step. Conversely, in the case of recalcitrant compounds, which can act as recombination centers, the rate determining step is likely to be on semiconductor side. Due to this variability, researchers devoted their efforts towards the improvement of both metal and semiconductor-and their coupling-as we will review in the next sections.

Shape Controlled TiO 2 Structures
While H 2 evolution occurs on the metal, charge separation and the oxidation semireaction efficiency are strongly related to TiO 2 morphology and surface texture.
Albeit a large amount of experimental and theoretical studies has been carried out on the characteristics of different surfaces of metal oxides [59][60][61][62][63][64][65][66][67], less attention has been paid on the dependence of the functional properties (reactivity, toxicity, osseointegration, etc.) on crystal surfaces exposed by metal oxides [68,69]. By tuning the ratio of different facets, the functional properties correspondingly change [70]. It is generally accepted that a surface with a high percentage of under-coordinated atoms has a greater reactivity [71][72][73][74]. A cooperative mechanism involving both favorable surface atomic structures and surface electronic structures has been recently proposed in order to explain the differences in reactivity [75]. Moreover, surfaces with higher reactivity usually have higher surface energy, so the possibility to obtain a pristine facet without the use of adsorbates was questioned [73], and upon removing the adsorbates the surface can reconstruct. However, reactive facets are highly desirable for improving and tuning functional properties. The shape evolution during the crystal growth is driven by the continue decrease of the total surface energy until the minimum surface energy point. Such process is conditioned by the presence of shape-controllers in the reaction environment.
The equilibrium shape of crystal structures can be predicted by using the Wulff construction [76], taking into account the surface energies of the crystal surfaces in the growth medium, if they are available. In the Wulff construction, surface energy minimization is used to optimize the composition of the crystal surface. The disadvantage is that only for clean crystal planes in vacuum are the surface energies available, and only in these conditions is the obtained equilibrium shape valid. The presence of liquid phases, electrolytes and adsorbates can change the equilibrium shape. The main TiO 2 polymorphs are anatase, rutile and brookite. All these three phases are constituted by TiO 6 octahedra, but in the case of anatase and rutile the crystal system is tetragonal, while for brookite is orthorhombic. The TiO 6 octahedra have different arrangements in the three polymorphs. Therefore, the surface terminations in various orientations are different, as well as the equilibrium morphologies of the crystals.
According to the Wulff construction, the equilibrium shape of the anatase is a lightly truncated bipyramid enclosed with eight trapezoidal {101} surfaces (>94%) and two top squares {001} surfaces, as shown in Figure  All the above predicted equilibrium shapes must be considered only in vacuum at 0 K [60]. By means of theoretical approaches, Barnard et al. found that the presence of water could influence the equilibrium shapes by changing the size and the aspect ratio of anatase and rutile [79]. Despite the high differences among vacuum conditions used in the calculations and hydrothermal/solvothermal conditions for crystal growth, the most stable facets predicted usually represent the largest fraction also in the real crystal surfaces.

Photocatalytic Hydrogen Production from Anatase TiO2
In the last decade, several works have been devoted to the influence that the TiO2 structure, particle size and morphology have on the charge carrier separation. The photocatalytic production of hydrogen can be performed using several TiO2-based catalysts, with different morphologies and different exposed facets. In this Section, we will describe some TiO2 anatase systems employed for H2 evolution; in particular, the attention will be focused on the effect of the crystal morphology on the HER.

Effect of Crystal Shape and Structure
The interaction of the TiO2 crystals with the light can be changed modifying their size and shape in order to maximize the light absorption and the amounts of photogenerated charge carriers, in accordance with the Mie theory [80,81].
In 2011, Yun and coworkers [82] investigated the role of the TiO2 nanoparticles shape in the photocatalytic hydrogen evolution. They compared nanospheres and nanorods using Pt as cocatalyst (1 wt %) and ethanol as sacrificial agent. They found that TiO2 nanorods have higher UV photocatalytic activity due to a decreased probability of ecb − /hvb + recombination. The authors, however, did not investigate the role of the different surfaces in their materials, probably responsible for the different electron hole recombination.
In 2011, Cho et al. [83], synthesized hierarchically branched TiO2 nanorods for more efficient PEC devices, due to their large contact area, excellent light absorption and the conductive path for the charge carrier collection. They obtained three different nanostructured films, made of TiO2 nanoparticles (NPs), nanorods (NRs) and branched nanorods (B-NRs). Under Xe lamp irradiation (880 W m −2 total output, UV matched to AM 1.5G) NRs and B-NRs had higher performances when used in PEC devices, due to their enhanced light absorption and charge carriers transport. In particular, the higher surface area and the more effective charges separation in branched TiO2 nanorods (B-NRs), resulted in an increased activity in comparison to simple nanorods (NRs). All the above predicted equilibrium shapes must be considered only in vacuum at 0 K [60]. By means of theoretical approaches, Barnard et al. found that the presence of water could influence the equilibrium shapes by changing the size and the aspect ratio of anatase and rutile [79]. Despite the high differences among vacuum conditions used in the calculations and hydrothermal/solvothermal conditions for crystal growth, the most stable facets predicted usually represent the largest fraction also in the real crystal surfaces.

Photocatalytic Hydrogen Production from Anatase TiO 2
In the last decade, several works have been devoted to the influence that the TiO 2 structure, particle size and morphology have on the charge carrier separation. The photocatalytic production of hydrogen can be performed using several TiO 2 -based catalysts, with different morphologies and different exposed facets. In this Section, we will describe some TiO 2 anatase systems employed for H 2 evolution; in particular, the attention will be focused on the effect of the crystal morphology on the HER.

Effect of Crystal Shape and Structure
The interaction of the TiO 2 crystals with the light can be changed modifying their size and shape in order to maximize the light absorption and the amounts of photogenerated charge carriers, in accordance with the Mie theory [80,81].
In 2011, Yun and coworkers [82] investigated the role of the TiO 2 nanoparticles shape in the photocatalytic hydrogen evolution. They compared nanospheres and nanorods using Pt as cocatalyst (1 wt %) and ethanol as sacrificial agent. They found that TiO 2 nanorods have higher UV photocatalytic activity due to a decreased probability of e cb − /h vb + recombination. The authors, however, did not investigate the role of the different surfaces in their materials, probably responsible for the different electron hole recombination. In 2011, Cho et al. [83], synthesized hierarchically branched TiO 2 nanorods for more efficient PEC devices, due to their large contact area, excellent light absorption and the conductive path for the charge carrier collection. They obtained three different nanostructured films, made of TiO 2 nanoparticles (NPs), nanorods (NRs) and branched nanorods (B-NRs). Under Xe lamp irradiation (880 W m −2 total output, UV matched to AM 1.5G) NRs and B-NRs had higher performances when used in PEC devices, due to their enhanced light absorption and charge carriers transport. In particular, the higher surface area and the more effective charges separation in branched TiO 2 nanorods (B-NRs), resulted in an increased activity in comparison to simple nanorods (NRs). Nanorods were studied also by Jiang et al. in 2013 [84]. They synthesized different immobilized anatase nanostructures from titanate nanowires by means of hydrothermal treatment. Changing the conditions of synthesis, two different structures were obtained: in neutral or acidic aqueous solution a chemical dehydration of titanate to anatase porous framework (TiO 2 -PF) and nanowires (TiO 2 -NW) occured, on the other hand mild alkaline conditions led to a dissolution-recrystallization mechanism, in this case nanorods were the final product with well-defined {010} and {101} surfaces (TiO 2 -NR). Under solar simulated spectrum (41 W m −2 in the UV), the authors found that the presence of the reactive {010} and {101} facets promoted the photocatalytic hydrogen evolution, overcoming the benefits of the intimate inter-particle contact of the hierarchical structures synthesized in neutral and acidic media.
Among different TiO 2 architectures, photonic crystal can also be exploited for studying the correlation between shape and photocatalytic activity. Photonic crystals are periodic optical nanostructure in which the refractive index varies periodically in the space. In this way they can affect the light propagation in the material [85]. If a given wavelength matches the lattice periodicity, a constructive interference of the reflected waves inside the material will be possible. At that wavelength, a complete reflection occurs and the light cannot propagate because of the photonic band gap. The group velocity of the light is strongly reduced at the edges of the photonic band gap (where it can propagate inside the crystal); therefore, the interaction with the material increases. This stronger interaction is called "slow light" or "slow photons" and can be exploited in several fields, including photocatalysis [86][87][88][89].
In 2015 Sordello and Minero [88] studied the hydrogen evolution rate from TiO 2 inverse opals (TIOs) and macroporous disordered (MP) structures that they synthesized using different polymeric sacrificial templates [88,90]. They used polymethylmethacrylate (PMMA) and polystyrene (PS) to obtain TiO 2 architectures with different pores size. The authors then compared the photocatalytic activity of the synthesized materials using Pt as cocatalyst and formate buffer (0.1 M) as hole scavenger.
To study the effect of the photonic crystal slow photons, the hydrogen photoproduction tests were carried at two different irradiation wavelengths: at 365 nm, where the inverse opals studied were able to take advantage of the slow light, and 254 nm, where the contribution of the slow light was negligible. They concluded that under 254-nm irradiation, MP TiO 2 materials presented similar photoactivity compared to inverse opals, given the same polymer template. Conversely, at 365 nm, the slow photons effect induced a larger H 2 evolution for the inverse opals. Thanks to this approach they were able to prove that the rate increase stems from improved light absorption when photonic band gap edges and incident light emission maximum are matched and slow photons can be exploited. Without slow photons, inverse opals behave as disordered macroporous materials, confirming that charge transfer dynamics are not modified. The activities were also influenced by the chemical characteristics of the sacrificial polymer that controlled the crystalline phase of the TiO 2 materials. While PMMA induced the formation of anatase only, conversely, with PS a mixture of anatase and rutile was obtained.
The photonic crystal properties of TiO 2 nanotube arrays were exploited by Chiarello and coworkers [91] to enhance the photocatalytic hydrogen production under a polychromatic UV irradiation. Thin films of TiO 2 nanotube arrays on a Ti support were synthesized by electrochemical anodization. Figure 6 shows the vertically self-assembled nanotube arrays. The authors were able to linearly increase the inner diameter of the nanotubes and their growth rate increased with increasing applied voltage. From theoretical calculations, they demonstrated that the position and the width of the photonic band gap could be affected by three morphological characteristics of the nanotubes: (1) inner diameter, (2) wall thickness and (3) distance between pores. The possibility to change these characteristics by simply modifying the synthesis conditions, allowed tuning the photonic band gap and fitting it within the TiO 2 absorption band, red shifting the photoactivity threshold. This led to a > 50% increase of the H 2 evolution rate under polychromatic irradiation.
Catalysts 2019, 9, x FOR PEER REVIEW 11 of 28 Figure 6. Dependence of the photonic band gap (Left) and H2 production rate (Right) from the nanotubes inner radius. Reprinted with permission from [91]. Copyright © 2016 American Chemical Society.

Effect of Exposed Surfaces
The strong dependence of the anatase nanoparticle photocatalytic activity on the typology of exposed surfaces, depend on its different reaction mechanisms occurring on different facets, due to their characteristic physicochemical properties [70]. In 2015 D'Arienzo et al. [92] studied the relationship between the different anatase surface and the H2 generation yield in methanol photosteam reforming under UV irradiation. They synthesized three different anatase crystals with shape nearly rectangular (RC), rhombic (R) and nanobar (NB) (Figure 7). They exposed {101}, {001} and, only for the RC, the {010} surfaces. They carried out Electronic Paramagnetic Resonance (EPR) experiments, correlating the amount of Ti 3+ (electron traps) with the H2 evolution rate. They observed that the higher is the Ti 3+ EPR signal, the higher is the H2 evolution rate. As we will report later, the {101} surface for anatase seems to be a reducing surface. Then, high exposition of this kind of facet should favor the formation of Ti 3+ centers under illumination and, therefore, better H2 production. From their results, the RC materials were the most active in methanol photoreforming and showed the largest density of Ti 3+ centers, however these materials had also the lowest {101} surface area, exposing preferentially the {010} surface. This points out that the H2 production cannot be related only to the extension of the reductive {101} surface, but also to presence of the {010} facet seems to play a role. The authors suggested the presence of a "surface hetero-junction" between the

Effect of Exposed Surfaces
The strong dependence of the anatase nanoparticle photocatalytic activity on the typology of exposed surfaces, depend on its different reaction mechanisms occurring on different facets, due to their characteristic physicochemical properties [70]. In 2015 D'Arienzo et al. [92] studied the relationship between the different anatase surface and the H 2 generation yield in methanol photo-steam reforming under UV irradiation. They synthesized three different anatase crystals with shape nearly rectangular (RC), rhombic (R) and nanobar (NB) (Figure 7). They exposed {101}, {001} and, only for the RC, the {010} surfaces. They carried out Electronic Paramagnetic Resonance (EPR) experiments, correlating the amount of Ti 3+ (electron traps) with the H 2 evolution rate. They observed that the higher is the Ti 3+ EPR signal, the higher is the H 2 evolution rate. As we will report later, the {101} surface for anatase seems to be a reducing surface. Then, high exposition of this kind of facet should favor the formation of Ti 3+ centers under illumination and, therefore, better H 2 production.
Catalysts 2019, 9, x FOR PEER REVIEW 11 of 28 Figure 6. Dependence of the photonic band gap (Left) and H2 production rate (Right) from the nanotubes inner radius. Reprinted with permission from [91]. Copyright © 2016 American Chemical Society.

Effect of Exposed Surfaces
The strong dependence of the anatase nanoparticle photocatalytic activity on the typology of exposed surfaces, depend on its different reaction mechanisms occurring on different facets, due to their characteristic physicochemical properties [70]. In 2015 D'Arienzo et al. [92] studied the relationship between the different anatase surface and the H2 generation yield in methanol photosteam reforming under UV irradiation. They synthesized three different anatase crystals with shape nearly rectangular (RC), rhombic (R) and nanobar (NB) (Figure 7). They exposed {101}, {001} and, only for the RC, the {010} surfaces. They carried out Electronic Paramagnetic Resonance (EPR) experiments, correlating the amount of Ti 3+ (electron traps) with the H2 evolution rate. They observed that the higher is the Ti 3+ EPR signal, the higher is the H2 evolution rate. As we will report later, the {101} surface for anatase seems to be a reducing surface. Then, high exposition of this kind of facet should favor the formation of Ti 3+ centers under illumination and, therefore, better H2 production. From their results, the RC materials were the most active in methanol photoreforming and showed the largest density of Ti 3+ centers, however these materials had also the lowest {101} surface area, exposing preferentially the {010} surface. This points out that the H2 production cannot be related only to the extension of the reductive {101} surface, but also to presence of the {010} facet seems to play a role. The authors suggested the presence of a "surface hetero-junction" between the From their results, the RC materials were the most active in methanol photoreforming and showed the largest density of Ti 3+ centers, however these materials had also the lowest {101} surface area, exposing preferentially the {010} surface. This points out that the H 2 production cannot be related only to the extension of the reductive {101} surface, but also to presence of the {010} facet seems to play a role. The authors suggested the presence of a "surface hetero-junction" between the {101}, {010} and {001} facets which drives the electrons photogenerated on the {001} not just toward the {101}, but also to the {010}, favoring the charge separation. The idea of "surface hetero-junction" is one of the most employed concepts for explaining the synergistic effect of the different surfaces of TiO 2 nanocrystals [93,94]. Figure 8 reports the charge transfer process using energy level diagram of the "surface hetero-junction" between facets {101} and {001} of TiO 2 anatase.  [93,94]. Figure 8 reports the charge transfer process using energy level diagram of the "surface hetero-junction" between facets {101} and {001} of TiO2 anatase. The importance of the exposed surfaces is highlighted also in the recent paper of Liu, Page and coworkers [95]. They synthesized anatase NPs with different relative amounts of {101} and {001} surfaces. They obtained that the catalyst with an intermediate amount of the {101} and {001} surfaces, normalizing for the SSA, showed the best UV photocatalytic activity in the H2 production from a methanol solution (Figure 9). They concluded that a synergistic effect may exist between {001} and {101} facets. In 2012 Gordon et al. [49] used TiF4 as precursor in the synthesis of anatase nanocrystals to produce HF in situ for controlling the crystal shape. Through the change of the ratio of TiF4 precursor in combination with TiCl4 and the type of surfactant (oleylamine or octadecanol), a tunability of the titania shape based on bipyramidal geometry was demonstrated ( Figure 10). The percentage of {101} and {001} facets was therefore well controlled and depended on the amount of HF released during the reaction, resulting in samples where F was still present on the surface. Then, treatment with NaOH exchanged the fluorides at the surface with OH groups. The materials were then tested for H2 production in the presence of methanol as sacrificial agent and photodeposited Pt The importance of the exposed surfaces is highlighted also in the recent paper of Liu, Page and coworkers [95]. They synthesized anatase NPs with different relative amounts of {101} and {001} surfaces. They obtained that the catalyst with an intermediate amount of the {101} and {001} surfaces, normalizing for the SSA, showed the best UV photocatalytic activity in the H 2 production from a methanol solution (Figure 9). They concluded that a synergistic effect may exist between {001} and {101} facets.  [93,94]. Figure 8 reports the charge transfer process using energy level diagram of the "surface hetero-junction" between facets {101} and {001} of TiO2 anatase. The importance of the exposed surfaces is highlighted also in the recent paper of Liu, Page and coworkers [95]. They synthesized anatase NPs with different relative amounts of {101} and {001} surfaces. They obtained that the catalyst with an intermediate amount of the {101} and {001} surfaces, normalizing for the SSA, showed the best UV photocatalytic activity in the H2 production from a methanol solution (Figure 9). They concluded that a synergistic effect may exist between {001} and {101} facets. In 2012 Gordon et al. [49] used TiF4 as precursor in the synthesis of anatase nanocrystals to produce HF in situ for controlling the crystal shape. Through the change of the ratio of TiF4 precursor in combination with TiCl4 and the type of surfactant (oleylamine or octadecanol), a tunability of the titania shape based on bipyramidal geometry was demonstrated ( Figure 10). The percentage of {101} and {001} facets was therefore well controlled and depended on the amount of HF released during the reaction, resulting in samples where F was still present on the surface. Then, treatment with NaOH exchanged the fluorides at the surface with OH groups. The materials were then tested for H2 production in the presence of methanol as sacrificial agent and photodeposited Pt In 2012 Gordon et al. [49] used TiF 4 as precursor in the synthesis of anatase nanocrystals to produce HF in situ for controlling the crystal shape. Through the change of the ratio of TiF 4 precursor in combination with TiCl 4 and the type of surfactant (oleylamine or octadecanol), a tunability of the titania shape based on bipyramidal geometry was demonstrated ( Figure 10). The percentage of {101} and {001} facets was therefore well controlled and depended on the amount of HF released during the reaction, resulting in samples where F was still present on the surface. Then, treatment with NaOH exchanged the fluorides at the surface with OH groups. The materials were then tested for H 2 production in the presence of methanol as sacrificial agent and photodeposited Pt as cocatalyst under simulated solar light. The results showed that the samples with higher content of {101} facets were better than {001} for this particular reaction, both with and without fluorine on the surface. This clearly demonstrates the importance of the presence of the {101} to have efficient reductive processes.  The photocatalytic production of hydrogen from methanol steam reforming was also investigated over a series of fluorinated Pt/TiO2 samples (F for O nominal molar substitution ranging from 5 to 15%) synthesized by flame spray pyrolysis by Chiarello et al. in a work published in 2014 [96]. A substitution lower than 10% led to an increase of surface hydroxylation which favored an indirect hydroxyl mediated mechanism, with the consequent increase of hydrogen production rate. Conversely, fluorination > 10% introduced an excess of bulk structural defects that can act as electron-hole recombination centers. Moreover, an excess of fluorination increases the Finduced surface electronegativity, inhibiting CB electron transfer. Both these effects led to lower hydrogen production rate under UV irradiation.
The effect of the nanoparticles shape and the surface fluorination were investigated by Yu and coworkers in 2010 [97], investigating the hydrogen production over Pt/TiO2 nanosheets that largely exposed {001} facets. They prepared anatase nanosheets by acidic hydrolysis of Ti(IV) butoxide using hydrofluoric acid as capping agent. A part of the obtained nanoparticles was then cleaned with diluted 0.1 M NaOH solution in order to remove the surface fluorides. To investigate the influence of the shape, the same synthesis procedure was repeated but replacing the hydrofluoric acid with the same volume of distilled water. Pt nanoparticles were photodeposited from a 0.26 mM H2PtCl6 solution in order to obtain a final weight ratio of Pt to Ti of 2%. The fluorinated TiO2 material presented higher photocatalytic activity than the other two materials, due to the synergistic effect of fluorination (reduction of the charge carriers recombination rate) and exposed {001} facets.  The photocatalytic production of hydrogen from methanol steam reforming was also investigated over a series of fluorinated Pt/TiO 2 samples (F for O nominal molar substitution ranging from 5 to 15%) synthesized by flame spray pyrolysis by Chiarello et al. in a work published in 2014 [96]. A substitution lower than 10% led to an increase of surface hydroxylation which favored an indirect hydroxyl mediated mechanism, with the consequent increase of hydrogen production rate. Conversely, fluorination > 10% introduced an excess of bulk structural defects that can act as electron-hole recombination centers. Moreover, an excess of fluorination increases the F-induced surface electronegativity, inhibiting CB electron transfer. Both these effects led to lower hydrogen production rate under UV irradiation.
The effect of the nanoparticles shape and the surface fluorination were investigated by Yu and coworkers in 2010 [97], investigating the hydrogen production over Pt/TiO 2 nanosheets that largely exposed {001} facets. They prepared anatase nanosheets by acidic hydrolysis of Ti(IV) butoxide using hydrofluoric acid as capping agent. A part of the obtained nanoparticles was then cleaned with diluted 0.1 M NaOH solution in order to remove the surface fluorides. To investigate the influence of the shape, the same synthesis procedure was repeated but replacing the hydrofluoric acid with the same volume of distilled water. Pt nanoparticles were photodeposited from a 0.26 mM H 2 PtCl 6 solution in order to obtain a final weight ratio of Pt to Ti of 2%. The fluorinated TiO 2 material presented higher photocatalytic activity than the other two materials, due to the synergistic effect of fluorination (reduction of the charge carriers recombination rate) and exposed {001} facets. The higher reactivity of the {001} facet is in contrast compared with most of the works published in the last few years. Ohtani and coworkers [98], in 2009, described a synthetic procedure to obtain decahedral anatase nanoparticles through a gas-phase reaction using TiCl 4 as precursor and with a rapid heating and quenching. The obtained material showed ≈40% of {001} surface, with a BET specific surface area of 9.4 m 2 g −1 . The activity of the material was evaluated on methanol photoreforming using UV light and Pt (2 wt %) as cocatalyst, deposited on the semiconductor from two different precursors: H 2 PtCl 6 or [Pt(NH 3 ) 4 ]Cl 2 . The authors compared the synthesized nanoparticles and the commercial TiO 2 P25 material. Due to the low density of crystalline defects, the decahedral nanoparticles were more reactive than TiO 2 P25 in the hydrogen evolution from methanol solution.

Effect of the Cocatalyst
Numerous are the cocatalysts coupled with TiO 2 proposed to maximize the rate of electron transfer to H + and consequently increase the overall efficiency of both water photosplitting and photoreforming. The most efficient cocatalysts are noble metals such as Pt, Au and Pd [99][100][101][102][103][104][105][106][107][108], but also materials based on Ag, Rh, Ni, Mn, Ru, Cr, and Co were proposed [109][110][111][112][113][114][115]. Furthermore, different carbon-based materials (e.g., graphene, graphene oxide, single-wall and multi-wall carbon nanotubes, carbon nanocones, graphitic-C 3 N 4 . . . ) [101,104,[116][117][118][119][120][121] have been recently reported as active materials to (i) delocalize the electron in the conduction band thanks to their proper work functions (significantly lower than the CB of the titania), (ii) decrease the overpotentials for the proton reduction. An overall agreement on the role of the carbonaceous phase during the photocatalytic process under irradiated hybrid systems has not been reached yet. One reason is that different materials, or the same materials synthesized with different procedures, can have significantly different electronic structures. The partitioning of the photogenerated charges between the semiconductor and the carbon-based material is strongly affected by their electronic properties (especially in terms of energetic positioning of the bands) and, consequently, this point is not easy to be rationalized comparing the photocatalytic behavior of materials only seemingly similar. An agreement on the role of the carbonaceous phase (graphene, grapheme oxide, reduced-GO . . . ) has been reached for the photocatalytic degradation of dyes on hybrid systems. In this case a dye-sensitized degradation mechanism seems to be active: an electron promoted in the LUMO of the dye-as a consequence of the absorption of a visible photon-is rapidly delocalized toward the carbonaceous phase-where the dye is usually strongly adsorbed-and here it reacts with the dissolved oxygen to generate the radical O 2 •-/HO 2 • . Then the oxidized dye can autonomously evolve forming stable or transients byproducts [122,123]. Pastrana-Martínez et al. investigated the photocatalyzed degradation of the colorless diphenhydramine and the organic dye methyl orange and demonstrated that under visible irradiation the self-oxidation of the visible-absorbing species is the dominant mechanism [124]. Under UV irradiation the transfer of the charged carriers between the two solid phases and between the hybrid catalyst and the reactants in solutions (e.g., H 2 O to be reduced for the production of H 2 or a sacrificial organic compound in the case of carbon photoreforming) has not been totally clarified. It was suggested by Minella et al. that on UV-irradiated TiO 2 /rGO hybrids, it is possible both the electron transfer from exited states of rGO onto the titania conduction band and the hole transfer from the TiO 2 to the carbonaceous phase where the sacrificial substrate is adsorbed and oxidized, if and only if the HOMO of the substrate has a less positive redox potential (higher energy) than the empty state of excited rGO [123]. Similarly, Hu and Tang suggested that during the photocatalytic transformation of phenol on visible irradiated hybrids composed of rGO and titanate nanotubes (TNTs), the photogenerated electrons in the graphene can be injected into the TNTs conduction bands although the presence of a Schottky barrier. In this case the degradation of phenol under visible irradiation was related to the indirect formation of highly reactive hydroxyl radical mediated by the formation of O 2 •-/HO 2 • as also suggested by Qianqian et al. [125]. The kinetics of transfer of electrons from high-energy rGO states toward the TiO 2 are characterized by kinetics in the 0.1-0.2 ps range. The ab-initio calculations carried out by Long et al. [126] suggested that the electron transfer process can occur from photoexcited states of graphene sheets without defects onto the titania surface with a non-adiabatic mechanism. On the contrary, the EIS evidences reported by Wang et al. investigating the TiO 2 -rGO/electrolyte interface suggested that electrons in the TiO 2 conduction bands can be transferred onto rGO sheets, but also that the holes in the valence bands can move toward the rGO phase promoting an effective degradation of adsorbed molecules [127]. The main conclusion from the analysis of the scientific literature on this topic (see as an example the recent reviews [128,129]) is that it is quite difficult to generalize because the numerous modifications of the synthesis and the intrinsic differences between materials only seemingly similar can affect abruptly their photocatalytic performance, the electronic structures of the hybrids and-ultimately-the dynamics of the photogenerated charge carries.
Recently, numerous Cu-containing species (Cu [105,130,131], Cu 2 O [132], CuO x [133], CuO [134], Cu(OH) 2 [135] and Cu 2 (OH) 2 CO 3 [136,137]) showed relatively high photocatalytic rate, even if a complete agreement on the operative electron transfer mechanism has not been reached yet. The complex redox speciation of the copper-based materials seems to force the ecb of TiO 2 to be delocalized on the cocatalyst and here react to produce hydrogen. The role of Pt and Au in the photoreforming of methanol under irradiated Pt-and Au-supported titania was investigated by Naldoni and coworkers [100] by in-situ electron spin resonance (ESR) spectroscopy. Both the nature of the cocatalyst and its loading strongly affected the H 2 evolution rate and the nature and relative concentration of the byproducts of the methanol oxidation. The best performance was reported for Pt-TiO 2 and this was attributed to the better ability of Pt to act as electron sink with compared with Au. This was directly demonstrated evaluating the concentration of the Ti +3 paramagnetic sites under irradiation that followed the order: pristine TiO 2 > Au-TiO 2 > Pt-TiO 2 . The electrons photopromoted were better delocalized on the metal phase-especially with platinum-giving a lower concentration of Ti +3 species. Interestingly, in the adopted experimental conditions the oxidation of methanol was only partial and significant amounts of formaldehyde were reported (see Figure 11).
Catalysts 2019, 9, x FOR PEER REVIEW 15 of 28 carried out by Long et al. [126] suggested that the electron transfer process can occur from photoexcited states of graphene sheets without defects onto the titania surface with a non-adiabatic mechanism. On the contrary, the EIS evidences reported by Wang et al. investigating the TiO2-rGO/electrolyte interface suggested that electrons in the TiO2 conduction bands can be transferred onto rGO sheets, but also that the holes in the valence bands can move toward the rGO phase promoting an effective degradation of adsorbed molecules [127]. The main conclusion from the analysis of the scientific literature on this topic (see as an example the recent reviews [128,129]) is that it is quite difficult to generalize because the numerous modifications of the synthesis and the intrinsic differences between materials only seemingly similar can affect abruptly their photocatalytic performance, the electronic structures of the hybrids and-ultimately-the dynamics of the photogenerated charge carries.
Recently, numerous Cu-containing species (Cu [105,130,131], Cu2O [132], CuOx [133], CuO [134], Cu(OH)2 [135] and Cu2(OH)2CO3 [136,137]) showed relatively high photocatalytic rate, even if a complete agreement on the operative electron transfer mechanism has not been reached yet. The complex redox speciation of the copper-based materials seems to force the ecb of TiO2 to be delocalized on the cocatalyst and here react to produce hydrogen. The role of Pt and Au in the photoreforming of methanol under irradiated Pt-and Au-supported titania was investigated by Naldoni and coworkers [100] by in-situ electron spin resonance (ESR) spectroscopy. Both the nature of the cocatalyst and its loading strongly affected the H2 evolution rate and the nature and relative concentration of the byproducts of the methanol oxidation. The best performance was reported for Pt-TiO2 and this was attributed to the better ability of Pt to act as electron sink with compared with Au. This was directly demonstrated evaluating the concentration of the Ti +3 paramagnetic sites under irradiation that followed the order: pristine TiO2 > Au-TiO2 > Pt-TiO2. The electrons photopromoted were better delocalized on the metal phase-especially with platinum-giving a lower concentration of Ti +3 species. Interestingly, in the adopted experimental conditions the oxidation of methanol was only partial and significant amounts of formaldehyde were reported (see Figure 11). Figure 11. Hydrogen, formaldehyde, carbon monoxide and carbon dioxide production rates on 4 different TiO2-based photocatalysts. TiO2-A and TiO2-Rf are pristine nanometric TiO2, while Au/TiO2-A and Pt/TiO2-A are as TiO2-A but decorated with nanometric particles of Au and Pt, respectively. Reprinted with permission from [100]. Copyright Elsevier, 2013.
In water photosplitting the back reaction giving water from the evolved products (H2 and O2) has to be limited to maximize the quantum yield of the process. As an example, Au, NiO and RuO2 are suitable cocatalysts also because the back reaction hardly proceeds because the reductive and oxidative sites on these catalytic systems are spatially separated. Interestingly, a suitable cocatalyst for the H2 evolution from water on irradiated semiconductor-not only TiO2-is NiO, which is Figure 11. Hydrogen, formaldehyde, carbon monoxide and carbon dioxide production rates on 4 different TiO 2 -based photocatalysts. TiO 2 -A and TiO 2 -Rf are pristine nanometric TiO 2 , while Au/TiO 2 -A and Pt/TiO 2 -A are as TiO 2 -A but decorated with nanometric particles of Au and Pt, respectively. Reprinted with permission from [100]. Copyright Elsevier, 2013.
In water photosplitting the back reaction giving water from the evolved products (H 2 and O 2 ) has to be limited to maximize the quantum yield of the process. As an example, Au, NiO and RuO 2 are suitable cocatalysts also because the back reaction hardly proceeds because the reductive and oxidative sites on these catalytic systems are spatially separated. Interestingly, a suitable cocatalyst for the H 2 evolution from water on irradiated semiconductor-not only TiO 2 -is NiO, which is usually formed by H 2 reduction of islands of nickel oxide on the semiconductor photocatalyst surface and subsequent partially oxidation with O 2 to form biphasic Ni/NiO layer structure. This structure is usually denoted as NiO x and is able to favor the charge carrier separation promoting the evolution of H 2 on the surface of the cocatalyst (in aqueous suspension mainly formed by Ni(OH) 2 ) and avoiding the back reaction between H 2 and O 2 being the oxygen evolution sites spatially separated from the cathodic sites. For further information on the electrochemical behavior of NiO, please refer to [138].
The chemical nature of the cocatalysts is as important as their structural features and consequently the same material structured in different nanometric forms can have significantly different efficiency. An in-depth analysis of the most investigated and the most promising cocatalysts is far from the scope of this review. To deepen this topic, please refer to [139,140].

Effect of the Sacrificial Agent
The increment of the H 2 photoproduction rate can be obtained choosing properly the hole scavenger. Despite of the huge works on the use of methanol as a reductant, its implementation cannot at this stage be considered a green technology, because no green process for the production of methanol is really operative and its production is mainly carried out through the catalytic conversion of synthesis gas to methanol in fixed bed reactor [141]. However, methanol is an ideal substrate for the investigation of the process and this justifies the efforts carried out to clarify the operative mechanism. Particularly, the oxidation of gaseous methanol to CO 2 involves successive oxidative steps comprising both · OH mediated oxidation and direct hole transfer to the adsorbed methanol with the production of formaldehyde, formic acid and CO 2 [45].
Bahruji and coworkers investigated the photocatalytic reforming of a variety of oxygenates compounds under UV irradiated Pd/TiO 2 providing insight into the effect of molecular structure on the reaction kinetics [108]. The rate of H 2 production with the different hole scavengers is reported in Figure 12. From these experimental evidences some general rules were proposed to achieve high H 2 production rate from the reforming of alcohols over Pd -TiO 2 catalysts: the alcohol must have a hydrogen in the adjacent position (α) with respect to the oxygen atom (few H 2 production was observed in the presence of acetic acid or acetone); the primary alcohols release CO 2 for decarbonylation and the related alkyl groups yields the corresponding alkanes (with the clear exception of methanol) and the carbon in the methylene groups finally undergoes complete oxidation to yield CO 2 . Bowker et al. [142] underlined the essential role that the hydroxyl groups in the organic reductant have to achieve effective photoreforming, being chemicals such as alkanes and alkenes inactive. usually formed by H2 reduction of islands of nickel oxide on the semiconductor photocatalyst surface and subsequent partially oxidation with O2 to form biphasic Ni/NiO layer structure. This structure is usually denoted as NiOx and is able to favor the charge carrier separation promoting the evolution of H2 on the surface of the cocatalyst (in aqueous suspension mainly formed by Ni(OH)2) and avoiding the back reaction between H2 and O2 being the oxygen evolution sites spatially separated from the cathodic sites. For further information on the electrochemical behavior of NiO, please refer to [138]. The chemical nature of the cocatalysts is as important as their structural features and consequently the same material structured in different nanometric forms can have significantly different efficiency. An in-depth analysis of the most investigated and the most promising cocatalysts is far from the scope of this review. To deepen this topic, please refer to [139,140].

Effect of the Sacrificial Agent
The increment of the H2 photoproduction rate can be obtained choosing properly the hole scavenger. Despite of the huge works on the use of methanol as a reductant, its implementation cannot at this stage be considered a green technology, because no green process for the production of methanol is really operative and its production is mainly carried out through the catalytic conversion of synthesis gas to methanol in fixed bed reactor [141]. However, methanol is an ideal substrate for the investigation of the process and this justifies the efforts carried out to clarify the operative mechanism. Particularly, the oxidation of gaseous methanol to CO2 involves successive oxidative steps comprising both  OH mediated oxidation and direct hole transfer to the adsorbed methanol with the production of formaldehyde, formic acid and CO2 [45].
Bahruji and coworkers investigated the photocatalytic reforming of a variety of oxygenates compounds under UV irradiated Pd/TiO2 providing insight into the effect of molecular structure on the reaction kinetics [108]. The rate of H2 production with the different hole scavengers is reported in Figure 12. From these experimental evidences some general rules were proposed to achieve high H2 production rate from the reforming of alcohols over Pd -TiO2 catalysts: the alcohol must have a hydrogen in the adjacent position (α) with respect to the oxygen atom (few H2 production was observed in the presence of acetic acid or acetone); the primary alcohols release CO2 for decarbonylation and the related alkyl groups yields the corresponding alkanes (with the clear exception of methanol) and the carbon in the methylene groups finally undergoes complete oxidation to yield CO2. Bowker et al. [142] underlined the essential role that the hydroxyl groups in the organic reductant have to achieve effective photoreforming, being chemicals such as alkanes and alkenes inactive.  The mechanism of reaction of organic substrates in photocatalytic condition has been in-depth investigated since from the seminal works on photocatalysis [143][144][145][146][147][148]. The fundamentals of semiconductor photocatalysis are now well understood and the presence of a complex network of oxidative paths involving both direct and mediated electron transfer at different reactive sites well accepted. Moreover, catalysts with different surface properties can promote transformation in which the relative weight of the different paths is different, with the consequent production of different byproducts [149,150]. Furthermore, the photocatalytic reactivity is dramatically influenced by both the intrinsic (crystallographic phase, surface defects, band potentials, doping, etc.) and extrinsic (pH, composition of the solution, presence complexing ions, etc.) properties of the photocatalyst in a manner often difficult to be rationalized. Even if rationalized with some difficulties, also the optical properties (absorption vs scattering) of the photocatalyst affects dramatically the ability of a semiconductor to harvest the radiation and consequently to promote the photocatalytic process (e.g., the photocatalytic reforming of organic compounds) [48,151].
As underlined above, photoreforming can be really considered a green technology only if the organic compounds used as sacrificial agents are wastes or they are produced through CO 2 neutral processes. In this light the best candidates and the most studied substrates are: (1) Ethanol, which is produced in a large number of fermentation processes of second-generation biomasses, as well as by the transformation of cellulose and lignocellulose [152]. The mechanism of reaction for the ethanol photoreforming was investigated on Pt/TiO 2 catalyst. The mechanism proposed followed the scheme described above where protons are reduced by electrons cumulating into Pt particles, while the chemisorbed or physisorbed ethanol is stepwise oxidized by holes [153]. The stepwise oxidation of the substrate gives predominantly acetaldehyde, acetic acid and CO 2 , but also compounds with larger number of carbon atoms, such as acetone and 2-butenal. 2-butenal (crotonaldehyde) can be formed by the aldol condensation of two acetaldehyde molecules or through the formation of the · CH 3 CHO radicals. In a similar way a possible path for the production of acetone is the reaction of acetaldehyde (or acetic acid) with · CH 3 radicals [154]. The accumulation of byproducts can give a partial poisoning of the catalyst with its progressive deactivation [155].
(2) Glycerol, which is the main byproduct of the biodiesel industry, is alternatively produced through the sorbitol hydrogenolysis or the fermentation of glucose [156,157]. Considering that (i) the total reforming of one molecule of glycerol gives 7 molecules of H 2 and (ii) the raw glycerol obtained as byproduct of the biodiesel industry is a complex mixture-containing methanol, water, inorganic salts, free fatty acids, triglycerides, and methyl esters-that requires expensive treatments to be purified, the energetic valorization of glycerol is a very promising path. The mechanism of transformation of glycerol on bare TiO 2 was investigated by Minero et al. [140] Glyceraldehyde (GAD), dihydroxyacetone (DHA), formaldehyde (FORM) and glycolaldehyde (GLY) were identified as the main byproducts obtained through two different reaction paths. GAD and DHA are the products formed from the OH radical mediated oxidation, while FORM and GLY are produced through a direct electron transfer (hole transfer from TiO 2 to the chemisorbed glycerol). Interestingly, the ratio between the produced byproducts is strongly affected by the surface properties of the catalyst. The direct hole transfer is favored on catalysts with more defective surfaces able to strongly adsorb the substrate on the more defective (and more reactive) sites. (3) Glucose, and other simple or complex sugar, which are mainly obtained from the degradation of cellulose materials [158,159]. The photoreforming of sugars, starches and cellulose was demonstrated for the first time on the ternary photocatalyst RuO 2 /TiO 2 /Pt [155]. As reported above, these substrates are ideal candidates to promote an efficient photoreforming because of the presence in the sugar molecules of numerous hydroxyl groups with hydrogen atom in α position. The more complex the carbohydrates, the lower is the efficiency. Fu and coworkers investigated systematically the glucose photoreforming under irradiated TiO 2 decorated with different metal cocatalysts in the absence of oxygen, and observed the following reaction rate: Pd > Pt > Au ≈ Rh > Ag ≈ Ru. The presence of O 2 -as expected-and the decrement of the pH inhibited the hydrogen production [159]. Recently, Kennedy et al. compared the hydrogen production with C2-C5 polyols, cyclic alcohols and mono and di-saccharides under irradiated Pd-TiO 2 . For the compounds in the first class the hydrogen evolution rate is directly related to the number of OH groups and the availability of hydrogen atoms in α position to the hydroxyl. This rule is not followed by the cyclic alcohols and especially sugar (glucose, fructose and sucrose). For the reforming of cyclic alcohols it was also observed the dehydrogenation and decarbonylation of α CC bond [107].
Furthermore, other organic compounds were investigated such as diols, superior alcohols and polysaccharides.
Caravaca and coworkers focused on the photoreforming of cellulose on a TiO 2 catalyst loaded with both Pt and Ni with UV irradiation. In the presence of both cocatalysts the photoreforming occurred with significant hydrogen production. The first step of the process was the (photo)hydrolysis of cellulose into glucose, which was then easily oxidized thanks to the highly hydroxilated structure. These results demonstrated the possible production of hydrogen directly from the main components of the biomass (the cellulose) without any intermediate steps [114]. To give more insights into the features of the photocatalytic reforming process of the above cited compounds, please refer to the excellent review by Cargnello et al. [160] and Rossetti [161].
An alternative approach is to consider the waste of the industrial or civil Waste Water Treatment Plants (WWTPs) as potential source of carbon to be photocatalytically reformed. In this line, Speltini et al. investigated the photocatalytic production of hydrogen in the presence of swine sewage as sacrificial agent observing significant evolution of hydrogen on Pt-TiO 2 upon UV-A and solar light [162]. Moreover, Liu et al. described the contemporary degradation of waste activated sludge and the simultaneous production of hydrogen in a fluidized tubular photocatalytic reactor [163].

Environmental Implications
An in-depth evaluation of the environmental impacts of a new technology-especially in the energetic field-is always mandatory to avoid that its employment causes severe perturbations of the fragile environmental equilibria. In this light, the lessons from the past (e.g., the depletion of stratospheric ozone as a consequence of the release of chlorofluorocarbons or the dramatic decrement of biodiversity as a consequence of the overuse of pesticide in vast area of the Earth) should drive the decisions related to the implementation of new technologies in our society. This has a general validity, but it is especially true for the technologies related to a massive market such as that of the energy.
The first environmental consideration about the processes for the production of hydrogen through photochemically driven and TiO 2 -based technologies described and commented above, should start from the abundance of the elements on the Earth's crust. Titanium is an abundant element (with an average concentration in the continental crust equals to 4010 ppm [164]). Its availability does not represent an obstacle for a massive employment of titanium dioxide-based technologies. However, we have previously underlined the essential role of cocatalysts to significantly increase the rate of H 2 production, low on pristine TiO 2 . In this light, as the abundance of the most investigated cocatalysts, namely Au, Pt and Pd, is very scarce (0.25, 0.4 and 0.4 ppb, respectively [164]) and a vast application of a technology based on these rare elements could not be envisaged, also because these new applications would be in competition with their more traditional industrial uses. The employment of cocatalysts based on copper seems more plausible thanks to its average higher abundance (25 ppm [164]). Clearly, the use of carbon-based cocatalyst or hybrid organic/inorganic structures does not have any hindrance from the point of view of the elemental abundance. In this case, the attention must be paid on the technologies employed for the production of innovative, complex and nano-engineered organic structures. A careful evaluation of the environmental impact of the production of these materials (e.g., graphene-like materials, CNTs, fullerenes . . . ) in terms of both energy request and environmental impact of their production process must be carried out. In this light, the evaluation of the overall process based on Life Cycle Assessment (LCA) is mandatory to really evaluate environmental merits and demerits of a technology. As an example, the actual processes for the production of significant amount of graphene-like structures are mainly based on a top-down approach starting from graphite. The strong oxidation of the original graphite is carried out through modifications of the original Hummer's method [165,166] that allows the production of huge amount of graphene oxide, but through a process heavy from the environmental point of view (e.g., use of concentrated mineral acids and oxidants).
Furthermore, the carbon photoreforming through the use of sacrificial agents (see Section 3.4) cannot be considered an a priori green technology because the nature of the sacrificial agents and the process used to produce them can modify enormously the overall green metrics of the process. The production of plants with pure energetic aims must be totally avoided not to compete with the food production. Furthermore, the sacrificial hole scavengers should be either an organic waste of primary industrial processes (e.g., the glycerol that is a byproduct of the biodiesel industry) or compounds produced from wastes of low-energy demanding processes and/or from other production chains. In all the cases, the overall life cycle for the hydrogen production technology investigated must be compared with that of the electrochemical production of hydrogen through the electrolysis of water that has yield almost quantitative (the electro-splitting of water on platinized anodes reach η ≥ 90%). This process, if based on the use of renewable energy sources, could have a dramatically low environmental impact.
Even though hydrogen production from irradiated semiconductors will probably have a marginal importance in the short-to medium-term, other technologies could be the drivers to the eventual transition toward a hydrogen-based economy. This scenario must be carefully evaluated taking into account not only the physical limits of our Planet, but also the complex network of reactions at the basis of the terrestrial equilibria. These could be perturbed by a significant hydrogen leakage during its production, transportation and use [167]. In this respect, the main environmental concerns are the possible impacts on the stratospheric ozone and its role as indirect greenhouse gas [168,169].

Conclusions
From the prior discussion, we can summarize the progresses of the research on photocatalytic H 2 production on TiO 2 into these three major areas:
Structuration. The crystal shape and size can be modulated in order to maximize the light absorption and, therefore, the charge carrier density under irradiation. The same effect can be achieved with photonic crystal structures, which can increase the performance of TiO 2 , as discussed in the Section 3.1. b.
Exposed Facets. Besides light absorption, morphology can also decrease the recombination of the charge carriers, accelerating their interfacial transfer. This was highlighted by several works, demonstrating the different behavior of the exposed facets for TiO 2 anatase and rutile. In the case of hydrogen evolution on anatase, the surface {101} has a prominent role due to his reductive nature. In the presence of a cocatalyst as Pt, the higher is the amount of the {101} surface the higher is the rate of H 2 photoproduction. Conversely, on pristine TiO 2 surface, a synergistic effect results between the different surfaces of anatase crystals, indicating the formation of a "surface heterojunction" that increases the efficiency of nanoparticles exposing considerable amount of oxidative surface.

2.
Cocatalyst. The role of the cocatalyst is crucial for improving the TiO 2 activity in the H 2 photoproduction. The common characteristics of a good cocatalyst in HER are the ability to drain photoelectrons from the conduction band and then decrease the required overpotential for reducing the proton. Noble metals are still in a predominant position, and especially Pt shows