Technical Comments

Response to Comments on "Efficient Photochemical Water Splitting by a Chemically Modified n-TiO2"

Science  19 Sep 2003:
Vol. 301, Issue 5640, pp. 1673
DOI: 10.1126/science.1080271

We do not disagree with the contention in the comment by Fujishima (1) that we should have cited the work of Fujishima et al. (2) in which titanium oxide was made by flame oxidation, a common method used both in our work (3) and that of other authors. However, the Fujishima et al. work has nothing else in common with ours. Our report (3) focused on citing works that reported high efficiency for water splitting; the Fujishima et al. study (2) is not among them. Although Fujishima (1) now maintains that we should have discussed the differences between his earlier work and our own, it would have been impossible in a two-page report to include all the differences separating our work with his and (logically) with that of all other authors who have reported in the literature on making titanium oxide by flame oxidation.

In the limited space afforded by this response, we now offer some important differences between the work of Fujishima et al. (2) and our own. One major difference is the absence data on the intensity of the light Fujishima et al. used from their 500-W xenon lamp (2). The 500-W xenon lamp can easily generate more than 500 mW cm2 power density, and it is logical to assume that he used as much intensity as possible to have highest value of photocurrent density. We used 40 mW cm2 power density from our xenon lamp to gain our reported photocurrent density.

Fujishima et al. (2) did not report photocurrent density, but rather reported the current density under illumination. That includes the dark current density, which could have been quite high for the sample studied in (2). Consequently, the photocurrent density would have been considerably less. Assuming, however, that the reported illumination current density was photocurrent density—that is, taking the dark current density as zero—under illumination intensity of I0 = 500 mW cm2, and assuming that the electrode potential at open-circuit condition was Eaoc = –1.00 volt/SCE (saturated calomel electrode) under the same illumination condition, we can obtain photoconversion efficiency as a function of applied potential (Fig. 1).

Fig. 1.

Photoconversion efficiency as a function of applied potential, Eapp = Emeas/SCE–Eaoc/SCE, where Emeas is the potential of the photoelectrode with respect to saturated calomel electrode (SCE) at which the photocurrent density was measured and Eaoc = –1.0 volt/SCE is the potential of the photoelectrode with respect to SCE at open-circuit conditions under the same illumination intensity, I0 (mW cm2), used for photocurrent measurements. Calculation is shown for study of Fujishima et al. (2) and Khan et al. (3).

Note that it is quite reasonable that the best sample studied by Fujishima et al. (2) generated a maximum photoconversion efficiency of 0.62%, because the illumination current density of his best flame-oxidized sample was less than that generated by single-crystal titanium oxide [figure 1 in (2)]. It is well known that photoconversion efficiency of single-crystal titanium oxide is within 1%. The low efficiency of the sample studied by Fujishima et al. (2) may have stemmed from the fact that they made the sample at too high a temperature (1300°C) in air instead of oxygen and for 5 min oxidation time. We noticed that if the oxidation temperature was too high (>850°C) and if oxidation time was much more or much less than 13 min, sample quality and photoresponse declined tremendously.

We did not report or claim that our results were obtained under sunlight, but reported that they were for illumination by the xenon lamp, which reasonably mimics sunlight. Hence, we do not find it appropriate to guess what our result would have been under sunlight. It should be noted that we reported the band-gap lowering by using ultraviolet-to-visible (UV-Vis) spectral response.

Hägglund et al. (4), in a second comment, have calculated a theoretical 8.1% conversion efficiency (compared with our experimental results of 8.35% efficiency), assuming 100% quantum efficiency and 1985 data (5) on photon flux for the air mass (AM) 1.5 solar spectrum. However, more recent data for the photon flux for AM 1.5 (6, 7) imply 16.99% conversion efficiency, again assuming 100% quantum efficiency. Hence, our experimental result of 8.35% efficiency is only 49% of the theoretical maximum efficiency possible if all photons having energy 2.32 eV (535 nm) to 4.2 eV (295 nm) in AM 1.5 were used to produce hydrogen gas. The 51% efficiency loss relative to the theoretical maximum likely stems from the loss due to recombination of photogenerated carriers and losses in the interfacial barriers at the semiconductor (CM-n-TiO2)–solution interface, among other losses (8).

Even better results are theoretically possible using the irradiance data for the AM 1.5 standard from the U.S. National Renewable Energy Laboratory (NREL), Denver, Colorado (9). Using the NREL data (Fig. 2), we found that the total power input from 295 nm (4.2 eV) to 535 nm (2.32 eV) was 21.95 mW cm2 when we integrated the data using the trapezoidal method, which implies a theoretical maximum photoconversion efficiency of 21.95% for total power of 100 mW cm2 if all input power were used to produce hydrogen. Hence, our experimental efficiency of 8.35% for hydrogen production is only 38% of the maximum possible efficiency of 21.95% for a 2.32 eV band-gap material using the NREL data. An earlier analysis by Gerischer (10), meanwhile, showed a maximum theoretical solar efficiency of 16.53% for a band-gap of 2.32 eV, roughly double our experimental 8.35% conversion efficiency. After taking into account the recombination loss inside the semiconductor, Gerischer (10) obtained maximum theoretical efficiency of 12.65% for the band-gap of 2.32 eV, which is 1.5 times higher than our experimental results. All these results clearly demonstrate that our experimental result of 8.35% is obtainable from a 2.32 eV band-gap material using AM 1.5 sunlight.

Fig. 2.

NREL irradiance standard for AM 1.5 (direct and global 37° tilt). Integration of the data from a wavelength of 295 nm to a wavelength of 535 nm implies a theoretical maximum photoconversion efficiency of 21.95% for total power of 100 mW cm2 if all input power were used to produce hydrogen.

Hägglund et al. (4) also compared our UV-Vis spectral absorption data (in arbitrary units) with old AM 1.5 data (in units of photon flux). We view this as an inappropriate comparison, however, because the spectral distribution of electric bulbs used in the UV-Vis spectrometer will be different from that of the xenon lamp that we used in our water-splitting experiments. Hägglund et al. (4) also mentioned that the current generation efficiency of a 150-W xenon arc lamp is at least six times higher than that of standard AM 1.5 sunlight, and suggested that additional infrared (IR) light filtering argues for an even higher spectral mismatch in our experiments. The filter, however, was used to provide a better match with AM 1.5 sunlight. We used a total intensity of only 40 mW cm2 power from the 150-W xenon arc lamp; the trend of the spectral response of the output from that light source at that intensity, through the filter, generally follows the trend of the spectral response of NREL standard AM 1.5 sunlight (Fig. 3).

Fig. 3.

Spectral irradiance for AM 1.5 standard (NREL data) at 100 mW cm2, xenon lamp at 40 mW cm2, and xenon lamp normalized to 100 mW cm2 (i.e., 2.5 × 40 mW cm2).

When these data are integrated, the power density becomes 7.86 mW cm2 up to 535 nm for a total output of 40 mW cm2 from the IR-filtered 150-W xenon arc lamp, compared with a power density of 21.95 mW cm2 up to 535 nm for a total output of 100 mW cm2 from AM 1.5 sunlight. The power output that we obtained for hydrogen production was 3.34 mW cm2, for a photocurrent density, jp, of 3.591 mA cm2 at light intensity of 40 mW cm-2 from a 150-W xenon arc lamp at an applied potential, Eapp, of 0.3 V. Hence, the power density of 3.34 mW cm2 [= jp × (1.23–Eapp) for hydrogen production] is only 42.5% of the total power input of 7.86 mW cm2 that our sample of band-gap 2.32 eV (535nm) can absorb from an incident light intensity of 40 mW cm2 (Fig. 3).

In any event, the data in Fig. 3 clearly indicate that incident light at an intensity of 40 mW cm2 from a 150-W xenon arc lamp can never generate six times more current than a AM 1.5 sunlight. Fig. 3 also includes a normalized plot at 2.5 times the 40 mW cm2 light intensity (equal to the total light intensity of 100 mW cm2 for AM 1.5 sunlight). It is clear from this plot that our sample could generate more photocurrent density and higher efficiency if it were illuminated by AM 1.5 sunlight: Although the normalized xenon arc lamp gains light in the UV region, it loses 1.8 times as much light in the visible range up to 535 nm (Fig. 3).

Hägglund et al. (4) are also concerned with differences between the recent study (3) and our earlier similar work (11) in which we used the same temperature (850 °C) and same heating time (13 min) to generate a material with a higher band-gap (2.82 eV) and a lower eficiency, despite a (trivial) increase in the intensity of the light used [50 mW cm2 in (11), compared with 40 mW cm2 in (3)]. The answer to this dilemma, however, is quite simple. In earlier work we heated a Ti metal sheet with a surface area of 1.0 cm2 at 850 °C for 13 min with a small flame area of 0.25 cm2. Over the years, we have observed that if the sample surface area is larger than the flame area, the chemical modification by carbon is not optimum. We also tried to increase the heating time corresponding to surface area increase of the sample, but we found that this led to overheating that tended to spoil the sample quality and conductivity. The best solution we found was the use of a sample area equal to the flame area. In (3), we used a sample area of 0.25 cm2 and heated it for 13 min at 850 °C using an equal flame size of 0.25 cm2. This procedure generated optimum modification by carbon to create the 2.32 eV band-gap TiO2 material.

Finally, although we appreciate the interest of Lackner (12) in our study (3), the analysis of our results expressed in his comment is incorrect, because he appears to be confusing a three-electrode electrochemical measurement under potential control with measurements in a two electrode system.

We obtained a maximum photoconversion efficiency of 8.35% at a measured potential of –0.7 volt/SCE in 5 M KOH solution at a pH of 14.7. Lackner mentioned that at measured potential Emeas = –0.7 volt/SCE in an electrolyzer, one could produce equal amounts of hydrogen without additional light energy. To verify that statement, we carried out a three-electrode electrolyzer experiment using two platinum electrodes (working and counter) and a reference electrode in 5 M KOH of pH 14.7 at the measured potential of –0.7 volt/SCE. We did not observe any hydrogen or oxygen gas evolution and also did not observe any current at this measured potential in the three-electrode system. That is because, at a pH of 14.7, water splitting cannot occur at an applied electrode potential of –0.7 volt/SCE; the water-splitting potential at pH 14.7 is +0.36 V [= 1.23 V–(0.0592 V × 14.7)]. We observed substantial hydrogen and oxygen gas evolution, however, at the same potential (–0.7 volt/SCE) and in the same solution when CM-TiO2 was illuminated by light. Thus, Lackner's statement lacks experimental or theoretical validation and is rather misleading. It follows that his other comments, which are based on his misunderstanding of the three-electrode electrochemical cell, are also incorrect.

Lackner's misunderstanding may have arisen because he thought that our applied potential was between a working electrode (photoanode) and counter electrode (cathode) in a two-electrode system from a power supply having two connections, one for cathode and one for anode. It should be noted that we did not use a power supply as a source of our applied potential. We used a potentiostat to apply potential to the working electrode with respect to a reference electrode in a three-electrode electrochemical cell.

Lackner argues that our efficiency would have been negative if we had used only measured potential, Emeas, in the efficiency equation [equation 1 in (3)] and had not subtracted the electrode potential under illumination of light at open-circuit conditions, Eaoc (measured with respect to a reference electrode). To verify this, we used Emeas = –0.7 volt/SCE at pH = 14.7, which converts to 0.41 volt/SHE (standard hydrogen electrode) at pH = 0.0. Hence, using E0rev = 1.23 volt/SHE at pH = 0.0 (where E0rev is the standard-state reversible potential for the water-splitting reaction) and Emeas = 0.41 volt/SHE, we obtained a 7.4% photoconversion efficiency—close to our reported result, notwithstanding Lackner's (12) assertion to the contrary. Indeed, we point out that photoconversion efficiency could not be negative as long as photocurrent is much higher than the dark current (or as long as dark current is zero) under any applied potential condition. Only the light photons generate the photocurrents, not the applied potential, because at the same applied potential the dark current is zero or negligible [figure 3 in (3)].

Lackner states that “[i]n an extreme, hypothetical case, light might only act as an on/off switch that either opens or closes an electric circuit that is completely driven by the external power supply.” This is an incorrect statement, because semiconductor n-TiO2 is not an insulator in the absence of light. It is well known that the semiconductor–electrolyte junction generates photovoltage under an illumination that helps to photosplit water with minimal amount of externally applied potential (0.3 V instead of 1.23 V).

Finally, we note that the hydrogen overvoltage is negligible (∼10 mV) at a platinum electrode (the hydrogen-producing counter electrode in our experiments) when the current density is in less than around 10 mA cm2. Also, Emeas = –0.7 volt/SCE, discussed above, equates to an applied potential of 0.3 V relative to Eaoc, the electrode potential measured under open-circuit condition. Eaoc is not the what is commonly called the open-circuit voltage (cell voltage), but is rather a working electrode potential measured with respect to a reference electrode at open-circuit conditions under illumination.


Related Content

Navigate This Article