The anatase TiO₂ nanoparticle powder was prepared by metal organic chemical vapor synthesis (MOCVS) (Berger et al., 2005; Elser et al., 2006; Elser and Diwald, 2012). Specifically, titanium (IV)isopropoxide was decomposed at T = 800°C in a hot wall reactor system. For purification, the obtained powder samples were subjected to thermal treatment under high vacuum conditions (p < 10⁻⁵ mbar). First, the powder sample was heated to T = 600°C using a rate of r ≤ 5°C min⁻¹. Subsequent oxidation with O₂ at this temperature was applied to remove organic remnants from the precursor material and to guarantee the stoichiometric composition of the oxide (Berger et al., 2005; Elser et al., 2006; Elser and Diwald, 2012). Resulting powders are characterized by a very low degree of particle aggregation thus yielding an ensemble of virtually isolated nanoparticles (Elser et al., 2006). The powder (0.150 g) was ground in ultrapure water (Milipore, 18.2 MΩ cm, 1.30 ml) in the absence of any additives to avoid the adsorption of organic molecules on the high surface area material (Rettenmaier et al., 2019). The slurry was spread by doctor blading onto FTO-coated glass (Pilkington TEC 8, resistance 8 Ω/□) to obtain anatase TiO₂ nanoparticle electrodes (“pristine electrodes”). Subsequently, the dried layers were annealed at 100°C ≤ T ≤ 450°C for t = 1 h in air (“annealed electrodes”) yielding films with a thickness of ∼5 µm (Rettenmaier et al., 2019). As the spreading of the slurry leads to unavoidable variations in the amount of immobilized oxide material, some measurements were conducted using the same electrode for a series of measurements. In these cases, the electrode was thoroughly rinsed with ultrapure water after each measurement, dried in air and annealed repeatedly at temperatures between 100 and 450°C. Importantly, for both processing / characterization approaches the same trends were observed. As confirmed in a previous study (Rettenmaier et al., 2019) using X-ray diffraction, transmission electron microscopy (TEM) and nitrogen sorption, the crystal structure (anatase TiO₂), the specific surface area (∼90 m²/g) and the nanocrystal size are preserved upon annealing of the TiO₂ films at 100°C ≤ T ≤ 450°C. In-depth analysis (Rettenmaier et al., 2019) yields for both films a narrow particle size distribution with a median particle diameter of 14.5 ± 0.5 nm. Accordingly, observed variations in the charge separation efficiency can be attributed to annealing-induced interface formation between particles featuring invariant primary particle properties.

Cyclic voltammograms of pristine and annealed anatase TiO₂ nanoparticle electrodes were recorded with a scan rate v = 0.020 V s⁻¹ in the potential window −0.6 V ≤ E _Ag/AgCl ≤ + 0.6 V in N₂-purged 1 M MeOH/0.1 M HClO₄ aqueous solution. Perchloric acid (HClO₄, 70% w/w in water) and methanol (MeOH, ≥ 99.8%) were purchased from Sigma Aldrich and used without further purification. The electrochemically accumulated charge was determined from the cathodic (i.e., negative-going) scan. For photoelectrochemical measurements the electrodes were illuminated from the electrolyte side with polychromatic light (P = 500 mW cm⁻²) of a 300 W Xe discharge arc lamp (LOT QuantumDesign) equipped with a water filter.

Anatase TiO₂ nanoparticle films (A = 1.82 ± 0.05 cm², m = 1.5 ± 0.2 mg) were deposited onto microscopy glass slides and were annealed at T = 100°C and T = 450°C in air. The mass of the oxide was determined gravimetrically. For all experiments, a freshly prepared film was immersed into a 4.0 mM H₂O₂ aqueous solution (V = 100.0 ml, Sigma Aldrich, 30 wt%, containing inhibitor). TiO₂ films were photoexcited by monochromatic light (λ = 360 nm) with an irradiance of P = 10 mW cm⁻² corresponding to ≈3.3∙10¹⁶ photons per second. For this purpose a 1000 W Xe discharge arc lamp (LOT QuantumDesign) equipped with a water filter and a 360 nm band pass filter (FWHM = 40 nm, 38% transmission) was used. Both, photogenerated electrons and holes can react with H₂O₂ to form H₂O and O₂ (Hirakawa and Nosaka, 2002). To avoid thermal H₂O₂ decomposition due to sample heating upon UV exposure, the reaction batch was cooled by an ice bath. In addition, the aqueous H₂O₂ solution was continuously stirred to avoid concentration and temperature gradients. Blind tests were performed using blank glass substrates instead of TiO₂-covered substrates. The H₂O₂ concentration of the solution (V = 100 ml) before [C(H₂O₂) ∼ 4 mM, corresponding to a total of 2.4∙10²⁰ H₂O₂ molecules] and after the exposure to UV light was determined by KMnO₄ titration (Huckaba and Keyes, 1948; Nishimura et al., 2008). The titration consisted of

• adding 10 ml H₂SO₄ (Merck, for analysis EMSURE, 25%) to 25 ml of analyte (H₂O₂ solution),

From the photon energy (E _ph = 3.44 eV, corresponding to λ = 360 nm) and the light irradiance (P = 10 mW cm⁻² = 6.242∙10¹⁶ eVs⁻¹ cm⁻²) the incident photon flux (Φ _q) was calculated. The apparent quantum yield (Φ) of H₂O₂ decomposition on TiO₂ films annealed at 100 and 450°C was calculated from the incident photon flux (Φ _q) and the number of H₂O₂ molecules decomposed upon UV exposure ( N H 2 O 2 ) according to Φ = N H 2 O 2 Φ q ⋅ t ⋅ 100 %

Prior to charge separation experiments, all electrodes were characterized (in the absence of UV light) by cyclic voltammetry. The results resemble observations from our previous study (Rettenmaier et al., 2019). As they set the base for charge separation experiments, they will be shortly discussed in the following.

The cyclic voltammograms (CVs) of a pristine film and of films annealed at different temperatures feature—in the absence of electron acceptors in the electrolyte—capacitive currents at potentials −0.6 V < E _Ag/AgCl < 0.2 V (Figure 1A) in line with previous studies (Fabregat-Santiago et al., 2003; Bisquert et al., 2008; Berger et al., 2012a; Bertoluzzi et al., 2013). These currents are associated with the population and depopulation of band gap states at solid/electrolyte and solid/solid interfaces in the semiconductor film (Bisquert et al., 2008; Berger et al., 2012a). A high reversibility of electron accumulation in the mesoporous semiconductor film gives rise to a symmetrical shape of the CV as the values of cathodic and anodic currents are comparable at each electrode potential (Bertoluzzi et al., 2013). While the CVs of films annealed at T ≥ 100°C virtually overlap at E _Ag/AgCl ≤ −0.3 V and feature a symmetrical shape, significantly lower current densities as well as a distorted shape are observed in the CV of a pristine film.

All films are characterized by comparable specific surface areas and primary particle sizes [d = 14.5 nm, see Ref. (Rettenmaier et al., 2019)]. Importantly, CVs of all annealed films yield comparable current densities (Figure 1A). It has been shown that for mesoporous anatase TiO₂ electrodes the capacitive current densities as well as the amount of accumulated charge (as extracted from the CVs) scales with the electrochemically active surface area (Berger et al., 2012a). Obviously, annealing at a temperature as low as 100°C imparts to the electrodes sufficient conductivity to electrochemically address—at least at sufficiently negative potentials—the whole nanoparticle network. Correspondingly, the accumulated charge (as determined from the cathodic scan) in TiO₂ electrodes annealed at 100°C ≤ T ≤ 450°C is virtually constant (Q = 2.2 ± 0.1 Cg⁻¹, Figure 2A).

In the pristine TiO₂ electrode, significantly less charge is accumulated (Q = 0.5 Cg⁻¹). This points (at comparable primary particle sizes) to a lower electrochemically active surface area and thus to limited interparticle charge transfer. Consequently, only a minor part of the films’ internal surface area contributes to the capacitive currents. Clearly, particle/particle contacts resulting from the sole drying of immobilized colloidal particles at room temperature do not impart sufficient conductivity to pristine TiO₂ electrodes. The high resistance of the resulting pristine film gives rise 1) to the distorted shape of the CV and 2) to low current densities (Figure 1A). An annealing-induced modification of particle/particle contacts at temperatures as low as 100°C, however, imparts to the films sufficient conductivity to address the entire TiO₂ film by electrochemical means (Rettenmaier et al., 2019).

Voltammetry is a powerful tool for studying the density of electrochemically active states in mesoporous semiconductor electrodes (Bisquert et al., 2008; Berger et al., 2012a). An in-depth analysis of the CVs of annealed electrodes allows to extract additional information on the energetic distribution of band gap states. The cathodic branches of the CVs feature in addition to an exponentially increasing current density at E _Ag/AgCl < −0.35 V (Figure 1A) a capacitive peak of low intensity at −0.35 V < E _Ag/AgCl < 0.1 V (Figure 1B). Whereas the exponential increase in current density has been assigned to the population of shallow traps located at the semiconductor/electrolyte interface (Bisquert et al., 2008), the capacitive peaks were attributed to the population of deep trap states located at grain boundaries (Lana-Villarreal et al., 2010; Jankulovska et al., 2012). Upon increasing the annealing temperature of the films, these peaks are shifted towards higher potentials (from E _Ag/AgCl = −0.20 V for a film annealed at 150°C to E _Ag/AgCl = 0.05 V for a film annealed at 450°C; Figure 1B). This shift can be associated with a modification of the energy of deep traps. Thermal annealing obviously induces changes in the atomic arrangement at grain boundaries and leads to an energetic stabilization of associated deep traps, which shift deeper into the band gap upon particle consolidation. From our experimental results we can not deduce conclusively, whether thermal annealing furthermore leads to a minor decrease of the intensity of capacitive peaks associated with deep traps. Due to the overlap of different capacitive contributions (resulting from traps at the solid/electrolyte interface and at solid/solid interfaces, respectively) we were not able to resolve any significant changes.

Photocurrents can be detected for semiconductor nanoparticle electrodes when the applied potential (Fermi level, E _F, of the conducting substrate) is more positive than the potential corresponding to the conduction band edge (E _CB) of the semiconductor (Berger et al., 2012a). For anatase TiO₂ nanoparticle electrodes in 0.1 M HClO₄, the position of the conduction band edge was estimated from photocurrent onset measurements in the absence of surface recombination and trapping at E _CB = −0.7 V vs. Ag/AgCl (3 M KCl) RE (Berger et al., 2012b). However, the photocurrent onset may significantly shift to more positive potentials due to electron-hole recombination, trapping of holes at surface defects and/or poor hole transfer to acceptors in solution (Bai et al., 2015). The detected photocurrent is thus a direct measure of the charge carrier separation efficiency in semiconductor electrodes.

The pristine and annealed anatase TiO₂ nanoparticle electrodes were used as photoanodes. After each annealing step, CVs were recorded upon exposure of the film to UV light (Figure 1C). Methanol, a frequenly employed hole scavenger in photoelectrochemical studies (Jiang et al., 2001; Monllor-Satoca et al., 2011; Mesa et al., 2017), was employed to minimize surface recombination. At negative potentials (E _Ag/AgCl < −0.5 V), the current response of the electrodes resembles the one measured in the absence of UV light and results from capacitive dark currents (compare with Figure 1A) and the absence of significant photocurrents (Figure 1C). When scanning the electrode potential from the photocurrent onset potential towards more positive values, a linear increase of the photocurrent is observed for all annealing temperatures. At sufficiently positive potentials, a saturated photocurrent is reached (Figure 1C). Importantly, the saturated photocurrent (determined at E _Ag/AgCl = 0.6 V) increases almost linearly with annealing temperature in the range 100°C ≤ T ≤ 300°C and levels off at T = 400 and 450°C (Figures 1C, 2B).

Annealing temperature affects—in addition to the saturated photocurrent—also the onset of photocurrent saturation, which is observed in the potential range −0.25 V ≤ E _Ag/AgCl ≤ 0.10 V (Figure 1C). This potential shifts towards more positive values upon an increase of the annealing temperature from 100 to 450°C. The maximum photocurrent is attained at E _Ag/AgCl ≥ −0.25 V, E _Ag/AgCl ≥ 0.06 V and E _Ag/AgCl ≥ 0.10 V for a film annealed at 100, 300, or 450°C, respectively. This shift resembles the shift of the energy of deep trap states located at the grain boundaries (Figure 1B).

The separation efficiency of photogenerated charge carriers (and thus the measured photocurrent) is the result of the kinetic competition between different processes including transport, transfer and recombination of charge carriers, which depend on electrode potential and thus on trap population. To rationalize the impact of deep traps and their population state on charge separation efficiency, we summarized the experimental findings from this and from our previous study (Rettenmaier et al., 2019) in Scheme 1.

For all electrodes investigated, we observe that the maximum obtainable photocurrent (Scheme 1A) is reached once deep trap states are mostly depopulated (Scheme 1C) as inferred from voltammetric measurements in the dark (Scheme 1D). This points to the strong impact of the population state of deep traps (Scheme 1C) on charge carrier separation. A possible explanation for this dependency is the involvement of deep traps in charge carrier recombination. Apparently, populated deep traps act as more efficient recombination centers than empty ones (Scheme 1, bottom). The increased recombination rate possibly results from the action of local electrostatic fields resulting from an uncompensated negative charge thus attracting photogenerated holes (Scheme 1, bottom). Indeed one may envisage—especially at high trap population—an incomplete charge compensation by counter ions from solution (such as e.g., H⁺) (Jiménez et al., 2016) at the buried interface of the grain boundaries. Consequently, photocurrents saturate at potentials, at which deep traps are depopulated. In such a case, the branching ratio between hole transfer to water or methanol and electron-hole recombination at populated grain boundary sites is increased and the photocurrent reaches its maximum value. Due to the energetic stabilization of deep traps upon thermal processing, this situation is reached at more positive potentials for annealed electrodes (Scheme 1A).

In addition to recombination, changes in the population state of deep traps are also associated with differences in conductivity (Scheme 1B) (Rettenmaier et al., 2019; Zhang et al., 2014). The electronic conductivity of the films increases exponentially in a potential range where deep traps get electrochemically populated and reaches an annealing temperature-independent value only at very negative potentials i.e., at potentials at which all deep traps are filled (Schemes 1B–D) (Rettenmaier et al., 2019). Such negative potentials corresponding to very high levels of electron accumulation are, however, not relevant for photoelectrodes assuring an efficient extraction of photogenerated electrons and generating thus measurable photocurrents (Schemes 1A,C).

The onset potential of the conductivity increase shifts positive upon increasing annealing temperature (Scheme 1B) in line with the displacement of the pair of capacitive peaks associated with deep trap states (Scheme 1D). This results—at a given electrode potential and under conditions, where deep traps are only partially populated—in a higher electronic conductivity for annealed electrodes. Consequently, photogenerated electrons are transported faster to the conductive substrate, which may explain the increase of the saturated photocurrent upon particle consolidation (Scheme 1A).

For mesoporous nanoparticle electrodes there exists typically a wide potential range where the conductivity of the semiconductor film is high enough to allow for a homogeneous charging of the electrode and thus for a homogeneous displacement of the semiconductor Fermi level with respect to the conduction band as long as the band edges are pinned (Bisquert et al., 2008; Berger et al., 2012a). Consequently, a homogeneous equilibrium occupancy of band-gap states corresponding to the externally controlled substrate potential can be established. However, upon UV exposure and at sufficiently positive potentials (i.e., potentials more positive than the onset potential for charge accumulation in the dark), a gradient of the carrier concentration (and thus the Fermi level, E _F) is established within the semiconductor film (Scheme 2) (Berger et al., 2012a). In direct vicinity of the conductive substrate (FTO), the Fermi level of the TiO₂ nanoparticles possibly equilibrates with the potential applied to the substrate. However, the Fermi level of the TiO₂ nanoparticles, which are exposed to UV light, is shifted towards more negative potentials as the photogenerated holes are at least partly transferred to solution species whereas the electrons accumulate in the semiconductor. Consequently, a gradient of electron concentration arises along the TiO₂ film thickness. At electrode potentials, where electrochemically addressable deep traps are virtually depopulated in the dark, they may be at least partially populated upon UV excitation (Scheme 2) (Zhang et al., 2014). The gradient in the occupancy of deep traps—which will vary according to the annealing temperature-dependent trap energy—will give rise to local variations of the electron conductivity and of the recombination rate. The overall charge separation efficiency (as tracked by the photocurrent) represents, therefore, an integral electrode property resulting from the kinetic competition of generation, transport, recombination, and interfacial transfer of charge carriers.

The beneficial effect of particle consolidation upon thermal annealing is also reflected in the photoactivity of anatase TiO₂ nanoparticles immobilized on glass substrates. The apparent quantum yield for H₂O₂ decomposition (at λ = 360 nm) accounts for Φ = 5.1 ± 0.6% and Φ = 8.3 ± 0.9% for films annealed at 100 and 450°C, respectively (Figure 3). In the absence of the TiO₂ layer, but under otherwise identical experimental conditions, the quantum yield accounts for 0.21 ± 0.04%. After 6 h of UV exposure a H₂O₂ conversion of 14 and 22% was observed for films annealed at 100 and 450°C, respectively. These results are fully in line with previous observations on the impact of aggregation and the associated formation of extended photocatalyst structures on charge separation (Wang et al., 2006; Lakshminarasimhan et al., 2008; Bian et al., 2012; Park et al., 2013; Wang et al., 2018). The promotion of charge carrier separation was attributed to the creation of additional pathways involving interparticle charge transfer and thus requiring sufficient conductivity. The beneficial effect of thermal annealing as observed in the present study can therefore be attributed to the increase of electronic conductivity upon particle consolidation and preservation of the active surface area (Rettenmaier et al., 2019). As highlighted for nanoparticle electrodes, the difference in film conductivity is especially pronounced under experimental conditions where deep traps at grain boundaries are not fully occupied (Rettenmaier et al., 2019). This will be the case in photocatalytic reactions under photostationary conditions, where an electron acceptor (such as H₂O₂) (Hirakawa and Nosaka, 2002; Friedmann et al., 2010) is typically added to the solution.