Here, we discuss the application, advantages, and potential pitfalls of using transient UV/Vis (ultraviolet-visible) absorption spectroscopy to study photoelectrodes for water splitting. We revisit one of the most commonly studied water oxidation photoanodes (α-Fe2O3−x) to provide commentary and guidelines on experiment design and data analysis for transient absorption (TA) studies of photoelectrodes within a photoelectrochemical cell. We also assess the applicability of such in situ TA studies to understand photoelectrodes under operating conditions. A major limitation is that most, if not all, past in situ TA studies have been carried out using only pulsed light sources to generate carriers, with the electrode held in the dark at other times, which is shown to be a poor model for operating conditions. However, with a simple modification of existing TA experiments, a simple operando TA measurement is reported.

The use of solar energy to produce storable fuels offers a route to overcoming the intermittency of renewable resources. In particular, light-driven water splitting can be used to generate hydrogen and oxygen, providing a non-fossil-based route to hydrogen. Semiconductor water splitting photoelectrodes have received attention since the initial reports that metal oxides such as TiO2 are active for this reaction.1,2 Although significant advances have been made over the last ∼50 years, the efficiencies of scalable materials for the production of solar hydrogen are still below commonly stated targets.

The basic principles of operation of photoanodes for water oxidation have been reviewed extensively elsewhere and are discussed only briefly here.3–5 Following photon absorption, it is desired that photogenerated electrons are rapidly separated from holes and transported away from the semiconductor–liquid junction (SCLJ) and toward the external circuit, giving rise to a current flow. The presence of an electric field (depletion layer) within the semiconductor, close to the SCLJ, also facilitates hole transport to the surface where transfer to the electrolyte, i.e., the oxidation of water, can occur. However, in competition with these two positively contributing pathways exist a range of recombination (loss) pathways both at the semiconductor surface and in the bulk, Fig. 1(b). Whilst the theoretical limit of efficiency of the photoanode (and a photoelectrode in general) may be given by its energetic properties (e.g., positions of the top of the valence band and the bottom of the conduction band), the achieved efficiency depends on the balance of rates of the competing contributing and loss pathways.6,7 The kinetics of these competing processes are dependent on the chemical/physical properties of the electrode, the electric field strength in the electrode (and, hence, the applied bias), and the photogenerated carrier density (which depends upon the illumination intensity). Therefore, it is important that techniques that can measure the charge carrier dynamics under realistic operating conditions are developed to provide a way to target the development of modified materials with improved solar to fuel efficiencies. Ex situ studies can provide useful qualitative insights into photoelectrode behavior and into the role of electrode treatments/modifications. However, they will not provide quantitative data suitable for modeling the electrode under operating conditions. At worst, studies carried out under non-operating conditions can even lead to erroneous conclusions being drawn regarding the effect of a particular electrode modification on the underlying mechanism. In this perspective, we critically appraise how transient absorption (TA) spectroscopy has been applied to date to study the kinetics of photoelectrodes and highlight how few, if any, studies have been reported under true operating conditions.

FIG. 1.

(a) Transient absorption (TA) spectroscopy can be used in situ to probe charge carriers in photoelectrodes on time scales ranging from femtoseconds to seconds. (b) Photoanode internal efficiency is determined by the kinetic competition between positively contributing pathways (red, electron extraction and hole transfer across the SCLJ) and recombination pathways including bulk recombination giving rise to a current density (Jbr), recombination in the depletion region (Jdr), surface state (trap) mediated recombination (JSS), and electron transfer (Jet) and tunneling (Jt) to intermediates of the water oxidation reaction. (b) Adapted with permission from Lewis, Inorg. Chem. 44, 6900–6911 (2005). Copyright 2005 American Chemical Society.

FIG. 1.

(a) Transient absorption (TA) spectroscopy can be used in situ to probe charge carriers in photoelectrodes on time scales ranging from femtoseconds to seconds. (b) Photoanode internal efficiency is determined by the kinetic competition between positively contributing pathways (red, electron extraction and hole transfer across the SCLJ) and recombination pathways including bulk recombination giving rise to a current density (Jbr), recombination in the depletion region (Jdr), surface state (trap) mediated recombination (JSS), and electron transfer (Jet) and tunneling (Jt) to intermediates of the water oxidation reaction. (b) Adapted with permission from Lewis, Inorg. Chem. 44, 6900–6911 (2005). Copyright 2005 American Chemical Society.

Close modal

TA spectroscopy is a pump–probe time-resolved technique used to study dynamic processes in materials or chemical compounds. In a TA experiment, the change in concentration of species produced following excitation of the sample with a short pulse of light (typically from a laser) is monitored against time, Fig. 1(a). Here, we discuss experiments that use UV/Vis (ultraviolet-visible) absorption spectroscopy as a probe, where the change in optical density (often assumed to be equal to the change in absorbance) or reflectivity is measured as a function of time. TA spectroscopy using a UV/Vis probe allows for the study of processes with lifetimes as short as femtoseconds (fs), out to seconds, and longer if required. A very recent review provides an overview of the basic principles of TA spectroscopy, alternative probe spectroscopies, and their applications to semiconductor photocatalysts, and we do not revisit these points here. Instead, readers interested in the background to pump–probe spectroscopy and its application to photocatalysts are directed there.9 

TA spectroscopy is being increasingly applied to explore charge carriers within photoelectrodes on both ultrafast (here defined as fs–ns) and slow (μs–s) time scales. The earliest examination of semiconductor materials such as TiO2,10,11 WO3,12 and α-Fe2O313,14 identified characteristic absorption features which could be assigned to electrons and holes and also gave insights into the nature of processes such as fast trapping of charges; however, the semiconductors studied were not part of a photoelectrochemical (PEC) cell. To the best of our knowledge, the first study of a water splitting photoanode by TA within a PEC cell was carried out on nanocrystalline TiO2 in 2010 by some of us.15 In this work, it was shown that once a positive bias was applied, it was possible to retard electron–hole recombination enough to enable measurement of the kinetics of hole transfer (water oxidation) at the SCLJ. A wide number of photoelectrode materials under an anodic bias within a PEC cell have now been studied using TA spectroscopy16,17 including α-Fe2O3,18–20 TiO2,15 ZnO,21 WO3,22,23 and BiVO4.24–26 TA spectroscopy has also been used to explore the role of electrode treatments, including those designed to passivate/control the distribution of surface states (e.g., thermal annealing steps,8,22 acid treatments,27 and overlayers8,20), the roles of co-catalysts,28,29 and heterojunction dynamics.30–33 In the majority of these studies, the kinetics of the charge carriers was found to be strongly dependent upon the applied bias highlighting the benefits of studying photoelectrodes in situ.

To obtain useful information from such in situ studies, it is important that careful consideration is given to the experiment design and data analysis. Here, we revisit a hematite photoelectrode we have previously studied8 (α-Fe2O3−x), in which oxygen deficiencies have been deliberately incorporated into the sample to improve the electronic properties.34 Our aim is not to re-examine the role of oxygen defects in controlling the structure, but instead to use results from this sample to illustrate practical guidelines to the community for design of in situ studies. We also aim at demonstrating how with a few simple modifications of widely used TA configurations, it is possible to carry out true operando studies of photoelectrodes. To the best of our knowledge, all TA experiments of PEC cells have been carried out using the pulsed laser as the only (or at least dominant) excitation source of the photoelectrode, i.e., the electrode often in the dark except for short (fs–ns) periods of illumination. We define TA studies of photoanodes in a PEC carried out in this way as in situ. Although the electrode is within the electrolyte and under potentiostatic control, it is not under operating conditions due to the lack of CW illumination, and at any one time, the photogenerated carrier concentration will not be equivalent to that found during use. In catalysis research, operando studies are typically defined as those carried out with simultaneous measurement of catalytic activity (often through detection of products).35,36 In-line with these criteria here, operando TA studies are defined as those only in which the photoelectrode is examined whilst operating under both CW illumination and potentiostatic control with a simultaneous measurement of the photocurrent. The ability to use the electrical current as a facile measure of catalytic activity represents a significant advantage in the operando study of photoelectrodes. We propose that a move toward operando studies is important if accurate quantitative kinetic modeling of photogenerated charges is to be achieved.

TA spectra and selected kinetic traces recorded on both the ps–ns and μs–s time scales of α-Fe2O3−x collected in transmission mode are shown in Figs. 2 and 3, and these data will form the center of the discussion for the remainder of this paper. The experimental apparatus used for the μs–s TA experiments in this study has been described elsewhere,8 while commercially available apparatus was employed for the ps–ns experiments outlined in the supplementary material. Schematics alongside specific considerations for the instrument design relating to the study of photoelectrodes can be found in Figs. S1 and S2. The transmission UV/Vis spectrum of α-Fe2O3−x at the open circuit is shown in Fig. 4(a). SEM images and the PEC response in 1M NaOH are reported and reproduced here within supplementary material, Figs. S3 and S4. A common pump wavelength of 355 nm was chosen for all experiments. Experiments are carried out with the front face [α-Fe2O3−x, often described as electrolyte/electrode (EE) side] excitation only, but we note that an advantage of transmission TA experiments is that comparison of the TA response between the EE side and the SE (substrate/electrolyte) side is possible, which can provide insight into the kinetics of charge transport within the electrode and help us discriminate between the surface and bulk TA signals in a manner similar to that used previously in transient photovoltage, electrochemical impedance spectroscopy, and photocurrent studies.37,38 Typically, for an in situ TA study, a range of potential steps are recorded, chosen following consultation of the linear sweep voltammogram of the same cell under CW illumination (Fig. S4) to provide data points both before photocurrent onset and up to the potential where photocurrent plateaus. Here, we focus on one particular potential (+0.4 VAg/AgCl), but data recorded at other potentials are shown in Fig. S5, and we see a strong potential dependence on the kinetics observed.

FIG. 2.

TA spectra of Fe2O3−x held at 0.4 VAg/AgCl in 1M NaOH electrolyte [(a) and (c)] on the ps–ns time scale following 355 nm (31 µJ cm−2, 5 kHz) excitation and [(b) and (d)] on the μss time scale following 355 nm (200 µJ cm−2, 0.33 Hz) excitation. Experiments (a) and (b) are recorded in situ (no CW light source), and (c) and (d) are operando measurements (with an additional CW LED, 365 nm). Data in (a) and (c) are presented without any correction of negative time delays to show TA signals that are long-enough lived to persist between excitation events.

FIG. 2.

TA spectra of Fe2O3−x held at 0.4 VAg/AgCl in 1M NaOH electrolyte [(a) and (c)] on the ps–ns time scale following 355 nm (31 µJ cm−2, 5 kHz) excitation and [(b) and (d)] on the μss time scale following 355 nm (200 µJ cm−2, 0.33 Hz) excitation. Experiments (a) and (b) are recorded in situ (no CW light source), and (c) and (d) are operando measurements (with an additional CW LED, 365 nm). Data in (a) and (c) are presented without any correction of negative time delays to show TA signals that are long-enough lived to persist between excitation events.

Close modal
FIG. 3.

TA kinetic traces on the ps time scale (left) and μs–s time scale (right) at 575 nm (top) and 650 nm (bottom) following 355 nm excitation of α-Fe2O3−x held at 0.4 VAg/AgCl in 1M NaOH electrolyte. In situ experiments with no additional CW light are shown in black, and operando measurements using a CW LED are in blue. To facilitate comparison between the in situ and operando ps–ns traces, we have applied an offset to the operando traces to align the change in OD at negative times to that of the in situ measurement, denoted as “CW light on, offset.”

FIG. 3.

TA kinetic traces on the ps time scale (left) and μs–s time scale (right) at 575 nm (top) and 650 nm (bottom) following 355 nm excitation of α-Fe2O3−x held at 0.4 VAg/AgCl in 1M NaOH electrolyte. In situ experiments with no additional CW light are shown in black, and operando measurements using a CW LED are in blue. To facilitate comparison between the in situ and operando ps–ns traces, we have applied an offset to the operando traces to align the change in OD at negative times to that of the in situ measurement, denoted as “CW light on, offset.”

Close modal
FIG. 4.

UV/Vis spectra and difference spectra of α-Fe2O3−x in 1M NaOH. At the open circuit, room temperature (a). Change in optical density of α-Fe2O3−x held at 0.4 VAg/AgCl as a function of temperature, and data are shown as difference spectra vs 23 °C (b). Change in optical density of α-Fe2O3−x at 23 °C as the applied bias is changed with the electrode in the dark vs a potential of −0.3 VAg/AgCl (c). TA spectra of α-Fe2O3−x at an applied bias of 0.4 VAg/AgCl following 355 nm excitation (d). Spectra are at selected time scales and extracted from the data shown in Fig. 2 in the absence of a CW light source.

FIG. 4.

UV/Vis spectra and difference spectra of α-Fe2O3−x in 1M NaOH. At the open circuit, room temperature (a). Change in optical density of α-Fe2O3−x held at 0.4 VAg/AgCl as a function of temperature, and data are shown as difference spectra vs 23 °C (b). Change in optical density of α-Fe2O3−x at 23 °C as the applied bias is changed with the electrode in the dark vs a potential of −0.3 VAg/AgCl (c). TA spectra of α-Fe2O3−x at an applied bias of 0.4 VAg/AgCl following 355 nm excitation (d). Spectra are at selected time scales and extracted from the data shown in Fig. 2 in the absence of a CW light source.

Close modal

To assign the observed features in the in situ spectra to either photogenerated charges, surface intermediates, or experimental artifacts (see below), a steady-state spectroelectrochemical study is carried out for the electrode in the same cell using the same potential range studied in the TA experiment, Fig. 4(c). TA experiments provide the user only with a set of difference spectra. Without knowledge of the UV/Vis spectrum of the sample prior to excitation at the applied potential, it is not possible to fully understand the difference spectrum. At positive potentials, a strong narrow feature centered at 575 nm increases in intensity. In a number of past studies, this has been assigned to an electron trap state that lies close (∼200 meV) to the conduction band edge, which will be partially occupied at the open circuit. At positive applied biases, these states become oxidized enabling an optical transition to the localized trap states, Fig. 4(c).8,18,32 A schematic depicting the 575 nm transition can be found in Fig. S6. A detailed discussion of the transient behavior of the electron trap state in α-Fe2O3 was given by Barroso et al.18 As at +0.4 VAg/AgCl, the ground state spectroelectrochemistry shows that the electron trap states are partially depopulated, we can make an initial assignment, where the initial small increase in optical density (OD) at 575 nm (compared to the negative time delays) on the ps time scale is due to fast hole trapping at these trap states in the bulk of the electrode as previously described,18Figs. 2(a) and 3 left. At later times (10 µs–700 µs), the OD at 575 nm then decreases as photoelectrons become trapped in the vacant states within the depletion layer, Fig. 3 right. Finally, on the millisecond time scale, the trapped electron population decays and the bleach in the TA signal at 575 nm recovers, Fig. 3 right.

The TA spectra recorded at +0.4 VAg/AgCl, a potential where light-driven water oxidation occurs, also show a broad absorption feature between 550 nm and 800 nm that persists for >0.1 s after excitation, Figs. 2 and 4(d). A broad bleach at wavelengths longer than 650 nm can be seen in the ground state (steady state) spectrum at this potential, Fig. 4(c), which could be rationalized by a decrease in free, or only shallow trapped, carriers, but this is contrary to the TA band, Fig. 4(d). In past studies, based on its very long lifetime, combined with the correlation of the amplitude of this signal on the millisecond time scale with the measured photocurrent, the TA response between 600 nm and 700 nm has been assigned to be due to photogenerated holes.39 Here, we also make the same assignment.

In addition to manual inspection of the TA data and comparison to the spectroelectrochemical response in the dark, global analysis approaches can also be a useful tool to initially identify spectra of individual charge carriers, and this is discussed further in Sec. VI. Once initial assignments are proposed, a series of additional experiments can then help us to confirm assignments. The use of a hole scavenger, a species that is more readily oxidized than water, such as H2O2, can help identify the presence of photogenerated holes and show if they are accessible to the scavenger. For example, in-line with other studies, here we find that the lifetime of the broad TA feature between 600 nm and 800 nm is greatly reduced in the presence of H2O2 helping confirm its assignment to surface accumulated photogenerated holes, Fig. S7.40 Transient photocurrent measurements are also of great benefit, and with α-Fe2O3−x, we have previously shown that the rate of the 575 nm bleach recovery, which we proposed, was due to electron detrapping correlated with the rate of charge extraction.8 

Additional experiments to test the hypothesized assignments to photogenerated species are important as it has been shown that with BiVO4,41 LaFeO3,42 and α-Fe2O3,43 the thermal effects arising from the laser-induced heating during a TA experiment can give rise to changes in the UV/Vis spectrum, which can be difficult to differentiate from excited state absorptions. In one study, on α-Fe2O3 at open circuit, it was found that the rate of lattice cooling was significantly slower than the rate of electronic relaxation with thermally induced signals dominating on the ns–μs time scales in a TA study.43 This has led some17 to raise valid questions about the assignment of spectral features in past TA studies. Therefore, we also recommend that the temperature dependence of the UV/Vis spectrum, ideally under applied bias, is reported, something which we are aware of in very few past TA studies.26 The temperature dependence of the α-Fe2O3−x photoelectrode at 0.4 VAg/AgCl in the dark is shown in Fig. 4(b) with data presented as difference spectra vs 23 °C. The thermal difference spectra observed in Fig. 4(b) are consistent with those previously reported by Hayes et al.43 for an α-Fe2O3 film produced from atomic layer deposition. The spectra can be split into three sections: features resulting in a reduction in absorption at wavelengths <550 nm, a positive feature centered at 575 nm, and a broad positive feature observed at wavelengths >600 nm. The features at wavelengths <550 nm have been assigned to d–d band shifts due to thermally induced structural distortions, whilst the thermally induced feature at 575 nm was proposed to originate from both contributions from d–d and the ligand to metal charge transfer (LMCT) transition as a result of density functional theory (DFT) calculations.43 From Fig. 4, it is clear that the 575 nm band is common in the spectroelectrochemical and thermal study, a conclusion also reached recently elsewhere.9 It is feasible that the features in the spectra arise from different species; however, given their very similar band shape, we believe this to be unlikely, Fig. S8. Polaron formation, a distortion of the lattice which can occur upon addition (or removal) of excess charge, can cause self-trapping, and photoinduced polaron formation is well known to occur with α-Fe2O3.44–46 It has been proposed that the common feature may be a signature of a Fe4+ hole polaron.9 However, experimental evidence against this assignment is that the TA kinetics of the 575 nm band correlates with the rate of electron extraction obtained through simultaneous transient photocurrent measurements, suggesting that the state is related to trapped electrons.8,18 Recent transient extreme ultraviolet,45 transient x-ray,44 and transient visible pump nIR probe46 measurements have demonstrated the ultrafast (<600 fs) formation of localized electron (Fe2+) polarons, and the lattice distortion following such ultrafast self-trapping may provide a way to reconcile the observed common spectral feature at 575 nm in Fig. 4. It is apparent that new measurements, such as extreme ultraviolet spectroscopy of photoelectrodes in situ on a wider-range of time scales (e.g., from fs-ms), will be required to unequivocally settle the debate.

Regardless of the uncertainty arising around the physical nature of the 575 nm trap state, it is clear that it is not possible to disentangle thermal and electronic contributions to the TA response of α-Fe2O3 by spectral deconvolution (in the wavelength/OD domain) alone. Instead, we suggest that the following precautions should be taken to enable evaluation of possible thermal relaxation contributions to the TA spectra. First, it is possible to calculate the maximum temperature change following photoexcitation during a TA experiment and use the variable-temperature UV/Vis data to estimate the maximum possible thermal contribution to the TA spectrum. Example calculations are given in the supplementary material (Fig. S9) for the fs and ns laser pulses used here to generate the data in Fig. 2, and these lead us to conclude that although thermal changes are a likely contributing factor, heating alone following laser excitation is not sufficient to induce the optical changes observed during our TA studies. Second, the sensitivity of the TA signal to applied bias should also be studied. A straightforward decrease in ΔOD at 575 nm as the applied potential is made positive would be compatible with decreased radiative recombination at these potentials, in-line with laser-induced heating being the dominant source of the TA response. However, here at potentials where water oxidation occurs, the optical density at 575 nm, which is increased immediately after excitation, decreases at early times (ps–μs), before increasing (ms) and then decreasing again. In contrast, at more negative potentials, we see a simple decay in the OD at 575 nm after excitation. The complex dependence of the TA amplitude with time at positive potentials appears incompatible with the signal being solely due to rapid laser-induced heating followed by lattice relaxation. This contrasts the findings of Hayes et al.,43 where thermal effects swamped the electronic contributions, likely due to the very high laser pulse energies employed (17 mJ cm−243 compared to ∼31 µJ cm−2 employed for the ultrafast experiments reported here). Therefore, we recommend working with the minimum pulse energies to minimize heating effects.

As in our example here, where the calculations indicate a contributing factor, decreasing the pump laser energy may itself be insufficient to remove thermal contributions to the TA spectrum, and a balance between the achievable signal to noise and concerns over thermal contributions must be reached. Therefore, in addition to checking the sensitivity of the TA signal to the applied bias, the effect of hole/electron scavengers on the TA response, such as that shown in Figs. S7 and S10, should also be studied as the thermal relaxation of the lattice should be insensitive to the presence of small concentrations of chemical scavengers such as H2O2 in the electrolyte. Here, we find that the TA response at 575 nm on the μs–ms at 0.4 VAg/AgCl in the presence and absence of H2O2 shows differences, again allowing us to be confident that the observed TA signal is dominated by photogenerated charges, Fig. S10.

Control of the energy of the pump laser is essential, as in addition to causing sample heating, it has been shown that the rate of bulk electron–hole recombination is particularly sensitive to pump energy.47–49 In an attempt to provide kinetic data on photoelectrodes that are relevant to actual operating conditions, several studies have stated that laser pulse energies and repetition rates were set so that the photon flux from the pulsed light source is equivalent or less than that experienced under a CW solar simulator at 1 sun.38,50 Using this approach, a correlation between the yield of long-lived electrons and holes measured by TA spectroscopy and the measured photocurrent density under CW illumination has been reported for multiple electrodes.39,51 In one work using very low pulse energies (35 µJ cm−2, 0.33 Hz), we were even able to quantify charge separation efficiencies within a TiO2 electrode using TA spectroscopy with results that matched those achieved through more conventional CW incident photon to current efficiency measurements.38 However, whilst it is clear that such in situ TA experiments can provide insights into the mechanisms of operational PEC cells, it must be recognized that the very high peak laser powers achieved using short fs–ns pulses make a pulsed laser a poor model for solar illumination. Typically, for slow TA experiments (ns–s),18 a weak CW probe light is present, but this is usually monochromatic light of photon energy below the band gap. In ultrafast (fs–ns) measurements, a white light probe is generated but this too is pulsed. Therefore, the dominant (or in many cases sole) excitation source is the pulsed laser. Excessive geminate recombination may be controlled through decreasing the pulse energy, but the use of a very short pulsed light source alone is still likely to lead to kinetics being measured which are not representative of the electrode under operating conditions. CW illumination generates non-equilibrium charge carrier concentrations with the magnitude of quasi-electron and hole Fermi level splitting being dependent on the CW light intensity. Therefore, recombination dynamics, distribution of surface intermediates, surface state occupancies, and double-layer structures are expected to be dependent on steady-state illumination intensity.

Insight into the importance of carrying out experiments under operating conditions comes from photoinduced absorption (PIA) spectroscopic studies of photoelectrodes, particularly from the Durrant group.52–55 In a PIA measurement, the UV/Vis absorption spectrum of the photogenerated holes is measured to derive the surface hole concentration (hs+), and simultaneously, the photocurrent density (Jph) is recorded allowing for the derivation of the rate law for water oxidation, using Eq. (1), where α is the order of the water oxidation reaction with respect to surface hole density and kwo is the water oxidation rate constant,

Jph=kwo(hs+)α.
(1)

Consistently, in these studies, Durrant et al. observed that the rate of hole transfer into the electrolyte (the rate of water oxidation) is dependent upon surface hole density and that two distinct mechanisms can occur, again dependent upon the surface density of holes.53 For α-Fe2O3, a transition from first-order (α = 1) to third-order kinetics with respect to surface hole concentration occurs. In the third-order mechanism, two nearest neighbors are triply oxidized prior to O–O bond formation. This is important as the need to accumulate three holes at the reaction site means that, although this mechanism dominates under typical operating conditions (1 sun CW), during TA experiments, only a first-order mechanism that proceeds via a series of one-hole oxidation steps without the need to accumulate oxidized intermediates occurs.51 Therefore, TA spectroscopy has been previously blind to the presence of the second mechanism, and the lifetimes of holes during water oxidation measured by past in situ TA studies are significantly longer (typically 0.3 s–3 s for α-Fe2O3)11,12,18,24,51 than those under 1 sun and, therefore, should be treated with caution.

PIA is a powerful tool to measure water oxidation kinetics, but it provides limited information on the faster processes that may be occurring within the electrode (e.g., charge trapping and bulk recombination); therefore, to demonstrate the effect of CW illumination more widely on electron–hole kinetics, we have repeated the TA experiments of α-Fe2O3−x under CW illumination, Figs. 2 and 3. All experiments are reported at 0.4 VAg/AgCl as at this potential, a significant (∼0.35 mA cm−2, Figs. S4, S11, and S12) steady-state photocurrent is measured. Figure 3 shows the ultrafast TA kinetics (ps–ns) in the presence of the CW light source. The changes in OD at negative time delays in the ps–ns TA data are due to processes that occur on time scales slower than the laser repetition rate (10 kHz probe) being sensitive to the CW light which become evident from inspection of the μs–s TA spectrum. In some studies, ultrafast TA data are shown normalized with the change in OD being set to zero at negative time delays, and we highlight the need to avoid this. A limitation of many ultrafast TA measurements of photoelectrodes that use high repetition rate laser systems is that equilibrium is not reached between each laser shot. Observation of a non-zero change in optical density at negative time delays is an important indicator of the presence of long-lived species, and as is shown here, the negative time data can provide useful insights into the slow processes. On the ultrafast time scale, the kinetic traces recorded at 650 nm and 575 nm are near indistinguishable, showing the same initial decay rates. Several potential conclusions can be drawn here. First, it is possible that laser-induced heating dominates the TA spectrum on the ultrafast time scale and that the addition of a CW light source does not change the rate of thermal relaxation; however, as calculated above, the magnitude of the observed TA response is greater than that achievable through heating alone. Alternatively, it can be concluded that the TA signals on these time scales are dominated by charge carriers in the bulk and that as the electrode is under potentiostatic control, the bulk electronic properties are not significantly changed by the CW light. It is also possible that the high peak laser powers of the fs–ns pulses used lead to carrier injection levels so large that the CW light has a negligible effect on the excess carrier population at a given time, and this represents a common shortcoming of many fs experiments.

In contrast to the ultrafast study, we find that the lifetime of the TA signal at all wavelengths is significantly reduced under CW illumination in the slow (μs–s) operando TA experiment, Figs. 2 and 3. Fitting of these data to obtain kinetic parameters for individual processes is discussed in Sec. V. However, even without fitting, a simple comparison of the magnitude of the TA signals with and without CW illumination, we find that there is an ∼85% reduction in hole signal at 650 nm at very long times (0.1 s after excitation), in-line with the findings from the past PIA experiments described above, Fig. 3. The electron trap signal at 575 nm is also found to be sensitive to the CW light source on the μs–ms time scale with accelerated trapping and de-trapping from the surface trap states occurring under LED illumination, Fig. 3.8 The large changes in the kinetics of the charges on the slow time scales highlights the importance of carrying out operando studies. Perhaps, unsurprisingly, this study indicates that the photoelectrode surface is markedly different under pulsed and CW illumination.

Analysis of the kinetic data provides a way to extract kinetic parameters to construct models of the charge carrier kinetics and, in some cases, generate the spectra of transient species associated with a decay process. The simplest way to analyze the TA data is to fit kinetic data recorded at selected wavelengths independently. For example, here it might be considered appropriate to only fit the kinetics of the photogenerated hole (650 nm) and electron trap state (575 nm). In this case, each trace is fitted to a pre-determined combination of functions (e.g., combinations of exponential and power-law functions). A review that describes the fitting of photoelectrode TA data in this way has been recently published.9 While such an approach is straightforward to implement and has been most widely used for photoelectrode studies,29,51,56,57 we advise against considering the kinetics of only one or two wavelengths in isolation. Reducing the entire TA dataset to one or two kinetic traces runs the inevitable risk of arriving at an oversimplified kinetic model for the photoelectrode. Furthermore, the number and type of fitting functions are determined by the user, often by the assumption of a particular kinetic model occurring. Whilst a thorough study should assess the quality of the fit for multiple kinetic models, few works on photoelectrodes have presented these data, and we need to recognize the potential danger of confirmation bias leading an investigator to seek a fit to their preferred model.

Alternatively, global fitting procedures such as global lifetime analysis (GLA) offer a way to assess all wavelengths simultaneously.58 In GLA, all kinetic traces are simultaneously fitted to a discreet, often small (2–5), number of exponential functions, with the pre-exponential factors allowed to vary across the wavelengths. Such an approach allows for the construction of decay associated spectra (DAS) for each lifetime component, which are a plot of the pre-exponential factors vs wavelength, providing a composite of the spectra of the species contributing to that particular kinetic process. A number of examples59,60 of the application of GLA exist being applied to photoelectrodes including a TA study of BiVO4 on the kinetics of charge trapping on the ultrafast time scale.25 In a similar approach to GLA, it is also possible to carry out a global analysis by spectral fitting, where TA data are fitted to a discreet number of component spectra with the amplitude allowed to vary for each across the time slices. This approach has been used in a study on oxygen deficient WO3 photoanodes, where singular value decomposition was used initially to extract spectral components prior to amplitude weighted fitting.23 Global fitting methods such as GLA can be used without a priori knowledge of the number of experimental decays or spectral components and, when combined with target modeling, can be used to test specific kinetic models.58 However, photoelectrodes present a challenge for GLA due to their complex kinetics and the often time-dependent spectra of semiconductor excited states, which continue to evolve over prolonged periods. The presence of multiple species (free carriers, trapped carriers, and surface intermediates), which react via a range of different pathways, is not easily captured using only a small discreet number of exponential functions, and solutions obtained can appear similar. A further complication is that common nanostructured photoelectrode materials contain high levels of defects and a wide distribution of trap and reaction sites. Therefore, single exponential functions are often a poor fit to the kinetics of a single pathway. In photoelectrode TA studies, where the kinetics at single wavelengths have been fitted, the heterogeneity of sites has been partially accounted for through the use of stretched exponential functions with the stretching parameter providing useful insights into the photoelectrode.1,15 We are not aware of the implementation of stretched exponential functions within the GLA procedure; however, this would represent an interesting route to aid the analysis of photoelectrode TA data.

Instead, here we demonstrate the use of lifetime density analysis (LDA) to study photoelectrode kinetics. LDA is a powerful technique for analyzing complex kinetic data, and it has been applied to a number of fields including TA studies on light-harvesting photosynthesis,61 photo-switches,62 and photoconductive hydrogels.63 Despite its potential advantages, we do not believe it has been applied to photoelectrodes previously. LDA is based on the principle that TA spectroscopic data (ΔA, which here we approximate to ΔOD) can be represented by the integral of a continuous distribution of single exponential functions as follows:

ΔAt,λ=0Φt,λetτδt.
(2)

To make Eq. (2) readily addressable, LDA approximates the integral to a semicontinuous sum of a large number (typically n > 100, here we use n = 200) of exponential functions, which are evenly distributed across the time scale. Typically, experimental data will also be convoluted (⊗) by an instrument response function [IRF(t, λ)] as follows:

ΔAt,λ=j=1nxj(τj,λ)etτjIRF(t,λ).
(3)

Equation (3) indicates that LDA and GLA are similar, different only in the value of n (LDA >100, GLA typically ∼2–5). However, a significant advantage of LDA is that it is effectively a model independent fitting process with no user input into the expected number of pathways. As a model-free fitting process, it is important to note that Eq. (3) is not based on the underlying kinetic equations governing the photoelectrode, but it is instead designed to provide an initial visualization of common lifetime components without assumption as the step toward construction of a kinetic model. Furthermore, the very large number of functions available makes it particularly amenable to complex and dispersive kinetics. LDA results are usually shown as lifetime density maps (LDMs), where the amplitude (xj) for each lifetime is plotted on the z-axis of a x–y surface of wavelength vs lifetime (τj), Fig. 5. A LDM provides a simple visualization of the number of kinetic processes occurring, indicated by peaks in the LDMs with positive and negative values indicating a lifetime component that leads to a decrease or increase in ΔA, respectively. The distribution of lifetimes associated with each peak can give important information; for example, it may potentially inform on the distribution of a particular trap state within a material or provide insights into carrier transport mechanisms. However, before drawing conclusions relating to lifetime distributions, it is advisable to consider the limitations and assumptions of the regularization procedure used within the LDA analysis, and a summary of these is reported elsewhere.64,65 Regularization is required as the large number of parameters makes overfitting a concern. Fitting of un-regularized spectra leads to fitting to the noise; therefore, procedures to penalize the use of large coefficients of fitting are required, effectively smoothing the data. Here, we present a LDA of the TA spectra of α-Fe2O3−x recorded on the μs–s time scale both in the presence (operando) and absence (in situ) of the CW light source, Fig. 5. Freely available software packages for LDA are now available,64,66 and here, we used OPTIMUS, which employs a Tikhonov regularization, and the appropriate regularization parameter (α) was chosen from the point of maximum curvature of the L-curve, available as Figs. S13 and S15. Full details of the fitting parameters, fits to all wavelengths, and reconstruction of the TA contour maps are available in supplementary material, Figs. S13–S16.

FIG. 5.

LDM generated from the TA data shown in Figs. 2(b) and 2(d) of α-Fe2O3−x held at 0.4 V following 355 nm excitation in the dark [in situ study, (a)] and under CW illumination [operando study, (b)]. The dotted lines indicate the five lifetime distributions identified for each study.

FIG. 5.

LDM generated from the TA data shown in Figs. 2(b) and 2(d) of α-Fe2O3−x held at 0.4 V following 355 nm excitation in the dark [in situ study, (a)] and under CW illumination [operando study, (b)]. The dotted lines indicate the five lifetime distributions identified for each study.

Close modal

Figure 5 shows at least five lifetime distributions, indicated by the dashed lines, within the TA data on the μs–s time scale demonstrating the complexity of the charge carrier dynamics on this time scale. It is also worth noting that a limitation of the Tikhonov regularization is that it can lead to broadening of kinetic distributions, making it hard to identify close lying distributions of lifetimes, so it is feasible that more lifetime distributions are present than the five identified.

The aim of this perspective is to provide guidelines and to critically appraise current experiment and data analysis methodologies used in the TA spectroscopy of photoelectrodes. A future study will focus on the assignment of each lifetime distribution and the development of a complete kinetic model. Here, we only briefly outline a few features to demonstrate the potential of LDA for operando studies of photoelectrodes. First, the LDMs also allow for the identification of new spectral features and kinetic pathways not previously included in past kinetic models. In the absence of CW illumination, the lifetime distribution centered around 640 µs, which has a strong contribution to the TA data between 550 nm and 625 nm, can be assigned to surface trapping of electrons, Figs. 5(a), 6(a), and 6(b). The recovery of the bleach signal across this region (indicated by the negative amplitude lifetime clusters between 1 ms and 100 ms) is markedly different at 575 nm and 600 nm, which we propose is due to the presence of a second trap state with a TA spectrum with maxima around 600 nm, Fig. 5. Further experiments are required to confirm the nature of this TA band, but its identification through the use of LDA highlights the benefits of the application of this procedure to TA studies of photoelectrodes. The LDA analysis also allows us to test the validity of our claim to be carrying out a true operando study through comparison of the measured lifetime of the surface hole in the TA studies to the rate of hole transfer measured by PIA. The lifetime of holes taking part in water oxidation can be assumed to be the longest-lived feature around 650 nm in the LDM. In the absence of a CW light, the LDM shows a distribution centered at 0.40 s, with a significant tail of lifetimes to beyond 2 s in-line with previous TA studies,39,51Fig. 6. The peak at 0.4 s indicates that initially, the third-order water oxidation may occur; however, as the electrode is in the dark after the laser excitation, the surface hole density rapidly decreases and a switch to the slower first-order mechanism occurs, giving rise to the very long lifetimes (>1 s), Figs. 5(a), 7(c), and 7(d).53 In contrast, under CW illumination, the hole lifetime is significantly decreased, and a single distribution centered at 0.14 s (7.1 s−1) is present, indicating that under CW illumination, only the third-order mechanism occurs. The measured rate of hole transfer in our operando TA study is in excellent agreement with the PIA measured hole turnover frequency for α-Fe2O3 at ∼1 sun condition (∼4 s−1),53 confirming the validity of our claim to be carrying out operando TA spectroscopy.

FIG. 6.

LDA fits (red, green lines) to the experimental TA data (blue, black) at 575 nm (a) and 650 nm (c) for a α-Fe2O3−x photoelectrode held at 0.4 V in 1M NaOH following 355 nm excitation. The lifetime distribution at these specific wavelengths is shown in (b) and (d). It is notable that a similar pattern in lifetime distributions exists, but for that under CW illumination (black lines), the lifetimes are significantly decreased.

FIG. 6.

LDA fits (red, green lines) to the experimental TA data (blue, black) at 575 nm (a) and 650 nm (c) for a α-Fe2O3−x photoelectrode held at 0.4 V in 1M NaOH following 355 nm excitation. The lifetime distribution at these specific wavelengths is shown in (b) and (d). It is notable that a similar pattern in lifetime distributions exists, but for that under CW illumination (black lines), the lifetimes are significantly decreased.

Close modal

The development of new more efficient photoelectrodes for water splitting is a pressing goal that requires input from a broad range of expertise. In presenting a revised study of α-Fe2O3−x, we have re-examined many standard practices used during TA studies of photoelectrodes. This has led to the generation of a series of best practice recommendations designed to ensure that results generated are of greatest benefit to the wider solar fuel community, and these are summarized in Table I. Of these, we highlight the need to clearly indicate to the community the conditions, and, hence, the limitations, of a particular study, and definitions are proposed based on the common terminology used within the catalysis community. TA studies of photoelectrodes in the dark (in situ) have great value, providing relevant data on bulk carrier kinetics and charge separation yields. However, recent PIA experiments and the work presented here show that experiments carried out in the dark with only short-pulsed excitation lead to unrealistic surface and double-layer conditions. Therefore, we look forward to more future studies being carried out under true operando conditions, which when coupled with appropriate data analysis protocols will enable the accurate, quantitative kinetic modeling of photoelectrodes.

TABLE I.

Guidelines for TA studies of photoelectrodes.

RecommendationDetails
Adoption of a standard terminology to explain the experimental conditions In situ: electrode at an applied potential within a PEC cell 
 Operando: At an applied potential within a PEC cell and under CW light with simultaneous measurement of photocurrent or chemical products 
Reporting of steady-state spectroelectrochemical response of electrodes as standard Essential for both in situ and operando studies to enable analysis of TA spectra. It is proposed that reviewers should question the absence of these data 
Assessment of thermal contributions to the TA spectra Variable-temperature steady-state UV/Vis allows for calculation of the maximum contribution to the TA spectra 
 Where thermal contributions are shown to be a potentially significant contributor analysis of the dependence of the TA spectra on applied potential, and CW light and chemical scavengers can provide evidence for the presence of photogenerated species 
Operation at lowest practical laser pulse energies Helps minimize heating and provide a closer model to solar illumination, minimizing excess bulk electron–hole recombination 
Use global analysis procedures, ideally model independent, to analyze/fit kinetic data LDA is found to be particularly suitable for photoelectrodes due to its ability to handle complex dispersive kinetics 
RecommendationDetails
Adoption of a standard terminology to explain the experimental conditions In situ: electrode at an applied potential within a PEC cell 
 Operando: At an applied potential within a PEC cell and under CW light with simultaneous measurement of photocurrent or chemical products 
Reporting of steady-state spectroelectrochemical response of electrodes as standard Essential for both in situ and operando studies to enable analysis of TA spectra. It is proposed that reviewers should question the absence of these data 
Assessment of thermal contributions to the TA spectra Variable-temperature steady-state UV/Vis allows for calculation of the maximum contribution to the TA spectra 
 Where thermal contributions are shown to be a potentially significant contributor analysis of the dependence of the TA spectra on applied potential, and CW light and chemical scavengers can provide evidence for the presence of photogenerated species 
Operation at lowest practical laser pulse energies Helps minimize heating and provide a closer model to solar illumination, minimizing excess bulk electron–hole recombination 
Use global analysis procedures, ideally model independent, to analyze/fit kinetic data LDA is found to be particularly suitable for photoelectrodes due to its ability to handle complex dispersive kinetics 

See the supplementary material for a description of the experimental methods and apparatus, electrode characterization data, calculations of the thermal contributions to the TA spectra, and supporting figures for LDA.

The authors thank Professor Yat Li (UCSC) for the α-Fe2O3−x sample and the Imaging Centre at Liverpool (ICaL) for access to STEM. Financial support for this work was from the INTERREG Atlantic Area Programme (Grant Reference No. EAPA_190_2016) and the EPSRC (Grant No. EP/P034497/1). The EPSRC is also acknowledged for equipment funding (Grant No. EP/S017623/1).

The data that support the findings of this study are available from the corresponding author upon reasonable request.

1.
P. J.
Boddy
,
J. Electrochem. Soc.
115
,
199
203
(
1968
).
2.
A.
Fujishima
and
K.
Honda
,
Nature
238
,
37
38
(
1972
).
3.
M. G.
Walter
,
E. L.
Warren
,
J. R.
McKone
,
S. W.
Boettcher
,
Q.
Mi
,
E. A.
Santori
, and
N. S.
Lewis
,
Chem. Rev.
110
,
6446
6473
(
2010
).
4.
Y.
He
,
T.
Hamann
, and
D.
Wang
,
Chem. Soc. Rev.
48
,
2182
2215
(
2019
).
5.
C.
Ros
,
T.
Andreu
, and
J. R.
Morante
,
J. Mater. Chem. A
8
,
10625
10669
(
2020
).
6.
H.
Gerischer
,
J. Electroanal. Chem.
150
,
553
569
(
1983
).
7.
N. S.
Lewis
,
Inorg. Chem.
44
,
6900
6911
(
2005
).
8.
M.
Forster
,
R. J.
Potter
,
Y.
Ling
,
Y.
Yang
,
D. R.
Klug
,
Y.
Li
, and
A. J.
Cowan
,
Chem. Sci.
6
,
4009
4016
(
2015
).
9.
T. J.
Miao
and
J.
Tang
,
J. Chem. Phys.
152
,
194201
(
2020
).
10.
D.
Bahnemann
,
A.
Henglein
,
J.
Lilie
, and
L.
Spanhel
,
J. Phys. Chem.
88
,
709
711
(
1984
).
11.
J.
Tang
,
J. R.
Durrant
, and
D. R.
Klug
,
J. Am. Chem. Soc.
130
,
13885
13891
(
2008
).
12.
F. M.
Pesci
,
A. J.
Cowan
,
B. D.
Alexander
,
J. R.
Durrant
, and
D. R.
Klug
,
J. Phys. Chem. Lett.
2
,
1900
1903
(
2011
).
13.
A. G.
Joly
,
J. R.
Williams
,
S. A.
Chambers
,
G.
Xiong
,
W. P.
Hess
, and
D. M.
Laman
,
J. Appl. Phys.
99
,
053521
(
2006
).
14.
G.
Xiong
,
A. G.
Joly
,
G. P.
Holtom
,
C.
Wang
,
D. E.
McCready
,
K. M.
Beck
, and
W. P.
Hess
,
J. Phys. Chem. B
110
,
16937
16940
(
2006
).
15.
A. J.
Cowan
,
J.
Tang
,
W.
Leng
,
J. R.
Durrant
, and
D. R.
Klug
,
J. Phys. Chem. C
114
,
4208
4214
(
2010
).
16.
D. V.
Esposito
,
J. B.
Baxter
,
J.
John
,
N. S.
Lewis
,
T. P.
Moffat
,
T.
Ogitsu
,
G. D.
O’Neil
,
T. A.
Pham
,
A. A.
Talin
,
J. M.
Velazquez
, and
B. C.
Wood
,
Energy Environ. Sci.
8
,
2863
2885
(
2015
).
17.
K. E.
Knowles
,
M. D.
Koch
, and
J. L.
Shelton
,
J. Mater. Chem. C
6
,
11853
11867
(
2018
).
18.
M.
Barroso
,
S. R.
Pendlebury
,
A. J.
Cowan
, and
J. R.
Durrant
,
Chem. Sci.
4
,
2724
2734
(
2013
).
19.
Z.
Huang
,
Y.
Lin
,
X.
Xiang
,
W.
Rodríguez-Córdoba
,
K. J.
McDonald
,
K. S.
Hagen
,
K.-S.
Choi
,
B. S.
Brunschwig
,
D. G.
Musaev
,
C. L.
Hill
,
D.
Wang
, and
T.
Lian
,
Energy Environ. Sci.
5
,
8923
(
2012
).
20.
T.-P.
Ruoko
,
A.
Hiltunen
,
T.
Iivonen
,
R.
Ulkuniemi
,
K.
Lahtonen
,
H.
Ali-Löytty
,
K.
Mizohata
,
M.
Valden
,
M.
Leskelä
, and
N. V.
Tkachenko
,
J. Mater. Chem. A
7
,
3206
3215
(
2019
).
21.
R. L.
Willis
,
C.
Olson
,
B.
O’Regan
,
T.
Lutz
,
J.
Nelson
, and
J. R.
Durrant
,
J. Phys. Chem. B
106
,
7605
7613
(
2002
).
22.
S.
Corby
,
L.
Francàs
,
A.
Kafizas
, and
J. R.
Durrant
,
Chem. Sci.
11
,
2907
2914
(
2020
).
23.
S.
Corby
,
E.
Pastor
,
Y.
Dong
,
X.
Zheng
,
L.
Francàs
,
M.
Sachs
,
S.
Selim
,
A.
Kafizas
,
A. A.
Bakulin
, and
J. R.
Durrant
,
J. Phys. Chem. Lett.
10
,
5395
5401
(
2019
).
24.
Y.
Ma
,
S. R.
Pendlebury
,
A.
Reynal
,
F.
Le Formal
, and
J. R.
Durrant
,
Chem. Sci.
5
,
2964
(
2014
).
25.
J.
Ravensbergen
,
F. F.
Abdi
,
J. H.
van Santen
,
R. N.
Frese
,
B.
Dam
,
R.
van de Krol
, and
J. T. M.
Kennis
,
J. Phys. Chem. C
118
,
27793
27800
(
2014
).
26.
S.
Selim
,
E.
Pastor
,
M.
García-Tecedor
,
M. R.
Morris
,
L.
Francàs
,
M.
Sachs
,
B.
Moss
,
S.
Corby
,
C. A.
Mesa
,
S.
Gimenez
,
A.
Kafizas
,
A. A.
Bakulin
, and
J. R.
Durrant
,
J. Am. Chem. Soc.
141
,
18791
18798
(
2019
).
27.
Y.
Yang
,
M.
Forster
,
Y.
Ling
,
G.
Wang
,
T.
Zhai
,
Y.
Tong
,
A. J.
Cowan
, and
Y.
Li
,
Angew. Chem., Int. Ed.
55
,
3403
3407
(
2016
).
28.
L.
Jing
,
J.
Zhou
,
J. R.
Durrant
,
J.
Tang
,
D.
Liu
, and
H.
Fu
,
Energy Environ. Sci.
5
,
6552
(
2012
).
29.
M.
Barroso
,
C. A.
Mesa
,
S. R.
Pendlebury
,
A. J.
Cowan
,
T.
Hisatomi
,
K.
Sivula
,
M.
Gratzel
,
D. R.
Klug
, and
J. R.
Durrant
,
Proc. Natl. Acad. Sci. U. S. A.
109
,
15640
15645
(
2012
).
30.
I.
Grigioni
,
K. G.
Stamplecoskie
,
E.
Selli
, and
P. V.
Kamat
,
J. Phys. Chem. C
119
,
20792
20800
(
2015
).
31.
A.
Yamakata
,
C. S. K.
Ranasinghe
,
N.
Hayashi
,
K.
Kato
, and
J. J. M.
Vequizo
,
ACS Appl. Energy Mater.
3
,
1207
1214
(
2020
).
32.
T.-P.
Ruoko
,
K.
Kaunisto
,
M.
Bärtsch
,
J.
Pohjola
,
A.
Hiltunen
,
M.
Niederberger
,
N. V.
Tkachenko
, and
H.
Lemmetyinen
,
J. Phys. Chem. Lett.
6
,
2859
2864
(
2015
).
33.
D.
Barreca
,
G.
Carraro
,
A.
Gasparotto
,
C.
Maccato
,
M. E. A.
Warwick
,
K.
Kaunisto
,
C.
Sada
,
S.
Turner
,
Y.
Gönüllü
,
T.-P.
Ruoko
,
L.
Borgese
,
E.
Bontempi
,
G.
Van Tendeloo
,
H.
Lemmetyinen
, and
S.
Mathur
,
Adv. Mater. Interfaces
2
,
1500313
(
2015
).
34.
Y.
Ling
,
G.
Wang
,
J.
Reddy
,
C.
Wang
,
J. Z.
Zhang
, and
Y.
Li
,
Angew. Chem., Int. Ed. Engl.
51
,
4074
4079
(
2012
).
35.
B. M.
Weckhuysen
,
Phys. Chem. Chem. Phys.
5
,
4351
(
2003
).
36.
M. A.
Bañares
,
Catal. Today
100
,
71
77
(
2005
).
37.
P. S.
Bassi
,
L.
Xianglin
,
Y.
Fang
,
J. S. C.
Loo
,
J.
Barber
, and
L. H.
Wong
,
Phys. Chem. Chem. Phys.
18
,
30370
30378
(
2016
).
38.
A. J.
Cowan
,
W.
Leng
,
P. R. F.
Barnes
,
D. R.
Klug
, and
J. R.
Durrant
,
Phys. Chem. Chem. Phys.
15
,
8772
8778
(
2013
).
39.
S. R.
Pendlebury
,
A. J.
Cowan
,
M.
Barroso
,
K.
Sivula
,
J.
Ye
,
M.
Grätzel
,
D. R.
Klug
,
J.
Tang
, and
J. R.
Durrant
,
Energy Environ. Sci.
5
,
6304
6312
(
2012
).
40.
H.
Dotan
,
K.
Sivula
,
M.
Grätzel
,
A.
Rothschild
, and
S. C.
Warren
,
Energy Environ. Sci.
4
,
958
964
(
2011
).
41.
J. K.
Cooper
,
S. E.
Reyes-Lillo
,
L. H.
Hess
,
C.-M.
Jiang
,
J. B.
Neaton
, and
I. D.
Sharp
,
J. Phys. Chem. C
122
,
20642
20652
(
2018
).
42.
S. Y.
Smolin
,
A. K.
Choquette
,
J.
Wang
,
S. J.
May
, and
J. B.
Baxter
,
J. Phys. Chem. C
122
,
115
123
(
2018
).
43.
D.
Hayes
,
R. G.
Hadt
,
J. D.
Emery
,
A. A.
Cordones
,
A. B. F.
Martinson
,
M. L.
Shelby
,
K. A.
Fransted
,
P. D.
Dahlberg
,
J.
Hong
,
X.
Zhang
,
Q.
Kong
,
R. W.
Schoenlein
, and
L. X.
Chen
,
Energy Environ. Sci.
9
,
3754
3769
(
2016
).
44.
J.
Husek
,
A.
Cirri
,
S.
Biswas
, and
L. R.
Baker
,
Chem. Sci.
8
,
8170
8178
(
2017
).
45.
L. M.
Carneiro
,
S. K.
Cushing
,
C.
Liu
,
Y.
Su
,
P.
Yang
,
A. P.
Alivisatos
, and
S. R.
Leone
,
Nat. Mater.
16
,
819
825
(
2017
).
46.
E.
Pastor
,
J.-S.
Park
,
L.
Steier
,
S.
Kim
,
M.
Grätzel
,
J. R.
Durrant
,
A.
Walsh
, and
A. A.
Bakulin
,
Nat. Commun.
10
,
3962
(
2019
).
47.
J.
Schneider
and
D.
Bahnemann
,
J. Phys. Chem. C
122
,
13979
13985
(
2018
).
48.
N.
Serpone
,
D.
Lawless
,
R.
Khairutdinov
, and
E.
Pelizzetti
,
J. Phys. Chem.
99
,
16655
16661
(
1995
).
49.
D. W.
Bahnemann
,
M.
Hilgendorff
, and
R.
Memming
,
J. Phys. Chem. B
101
,
4265
4275
(
1997
).
50.
J.
Tang
,
A. J.
Cowan
,
J. R.
Durrant
, and
D. R.
Klug
,
J. Phys. Chem. C
115
,
3143
3150
(
2011
).
51.
A. J.
Cowan
,
C. J.
Barnett
,
S. R.
Pendlebury
,
M.
Barroso
,
K.
Sivula
,
M.
Grätzel
,
J. R.
Durrant
, and
D. R.
Klug
,
J. Am. Chem. Soc.
133
,
10134
10140
(
2011
).
52.
F.
Le Formal
,
E.
Pastor
,
S. D.
Tilley
,
C. A.
Mesa
,
S. R.
Pendlebury
,
M.
Grätzel
, and
J. R.
Durrant
,
J. Am. Chem. Soc.
137
,
6629
6637
(
2015
).
53.
C. A.
Mesa
,
L.
Francàs
,
K. R.
Yang
,
P.
Garrido-Barros
,
E.
Pastor
,
Y.
Ma
,
A.
Kafizas
,
T. E.
Rosser
,
M. T.
Mayer
,
E.
Reisner
,
M.
Grätzel
,
V. S.
Batista
, and
J. R.
Durrant
,
Nat. Chem.
12
,
82
89
(
2020
).
54.
Y.
Ma
,
C. A.
Mesa
,
E.
Pastor
,
A.
Kafizas
,
L.
Francàs
,
F.
Le Formal
,
S. R.
Pendlebury
, and
J. R.
Durrant
,
ACS Energy Lett.
1
,
618
623
(
2016
).
55.
A.
Kafizas
,
Y.
Ma
,
E.
Pastor
,
S. R.
Pendlebury
,
C.
Mesa
,
L.
Francàs
,
F.
Le Formal
,
N.
Noor
,
M.
Ling
,
C.
Sotelo-Vazquez
,
C. J.
Carmalt
,
I. P.
Parkin
, and
J. R.
Durrant
,
ACS Catal.
7
,
4896
4903
(
2017
).
56.
A.
Kaouk
,
T.-P.
Ruoko
,
M.
Pyeon
,
Y.
Gönüllü
,
K.
Kaunisto
,
H.
Lemmetyinen
, and
S.
Mathur
,
J. Phys. Chem. C
120
,
28345
28353
(
2016
).
57.
J. K.
Cooper
,
Y.
Ling
,
C.
Longo
,
Y.
Li
, and
J. Z.
Zhang
,
J. Phys. Chem. C
116
,
17360
17368
(
2012
).
58.
I. H. M.
van Stokkum
,
D. S.
Larsen
, and
R.
van Grondelle
,
Biochim. Biophys. Acta
1657
,
82
104
(
2004
).
59.
R. E.
Rex
,
Y.
Yang
,
F. J.
Knorr
,
J. Z.
Zhang
,
Y.
Li
, and
J. L.
McHale
,
J. Phys. Chem. C
120
,
3530
3541
(
2016
).
60.
I.
Grigioni
,
L.
Ganzer
,
F. V. A.
Camargo
,
B.
Bozzini
,
G.
Cerullo
, and
E.
Selli
,
ACS Energy Lett.
4
,
2213
2219
(
2019
).
61.
R.
Croce
,
M. G.
Müller
,
R.
Bassi
, and
A. R.
Holzwarth
,
Biophys. J.
80
,
901
915
(
2001
).
62.
C.
Slavov
,
C.
Yang
,
L.
Schweighauser
,
C.
Boumrifak
,
A.
Dreuw
,
H. A.
Wegner
, and
J.
Wachtveitl
,
Phys. Chem. Chem. Phys.
18
,
14795
14804
(
2016
).
63.
C. L.
Smith
,
L. L. E.
Mears
,
B. J.
Greeves
,
E. R.
Draper
,
J.
Doutch
,
D. J.
Adams
, and
A. J.
Cowan
,
Phys. Chem. Chem. Phys.
21
,
26466
26476
(
2019
).
64.
G. F.
Dorlhiac
,
C.
Fare
, and
J. J.
van Thor
,
PLOS Comput. Biol.
13
,
e1005528
(
2017
).
65.
V. A.
Voelz
and
V. S.
Pande
,
Proteins Struct. Funct. Bioinf.
80
,
342
351
(
2012
).
66.
See https://optimusfit.org/ for information on lifetime density analysis and free download of software with user documentation.

Supplementary Material