It was confirmed by X-ray diffraction that the TiO₂ samples used in this work had single-phase rutile or anatase phase (Figure S1). Specific surface areas (S_BET) of these samples are listed in Table 1. Using the TiO₂ samples as support materials, Co₃O₄ nanoparticles were loaded, and photocatalytic activities of Co₃O₄/TiO₂ for water oxidation under visible light (λ > 500 nm) were investigated. It should be noted that loading Co₃O₄ onto these TiO₂ supports did not change the original XRD patterns, and that no peak derived from any cobalt species could be identified, as confirmed by our previous work using R1.²⁰ In this work, this was reconfirmed using A2 that had the largest surface area (Figure S2). It indicates that solid-state reaction between Co₃O₄ and TiO₂ was negligible under the present preparation conditions.

Photocatalytic activity was first compared using rutile and anatase TiO₂ particles (here R2 and A1, see Table 1) that have identical S_BET (14–17 m² g⁻¹) and morphology (see Figure S3). As shown in Figure 1, TEM observation indicated that most of the deposited Co₃O₄ in both samples was in the form of nanoparticles a few nm in size, but some of them aggregated to form larger secondary particles (10–20 nm). Figure 2 shows that this modification resulted in the generation of a new absorption band in visible light region in both cases. Our previous study indicated that the 400–600 nm band results from electronic interaction between Co₃O₄ and TiO₂, while a peak at around 700 nm could be assigned to bulky Co₃O₄.²⁰

Figure 1. TEM images of 1.0 wt% Co₃O₄-loaded R2 and A1.

Figure 2. UV-VIS-NIR diffuse reflectance spectra of R2 and A1 with and without loading of 1.0 wt% Co₃O₄.

While no O₂ was evolved without Co₃O₄ due to the lack of visible light absorption, the Co₃O₄-loaded TiO₂ samples produced O₂ upon visible light (λ > 500 nm). Note that Co₃O₄-loading onto SiO₂, which is an insulator, did not yield O₂ evolution. As listed in Table 2 (Entries 1–6), activities were improved in both cases with increasing the loading amount of Co₃O₄ to 1.0 wt%, then decreasing with further loading. This trend is consistent with our previous study, which showed that there is a trade-off between activity and Co₃O₄ loading in the visible-light water oxidation using rutile TiO₂ powder (R1).²⁰ More concretely, the appearance of an optimal loading amount indicates that the activity is governed by visible light absorption capability of surface water oxidation catalysis. As shown in Figure S4, more pronounced visible light absorption was provided by increasing the loading amount of Co₃O₄, which should also increase the density of active sites for water oxidation. Excess loading, on the other hand, decreases the active site density, leading to the drop in activity. Here, no significant difference in activity was observed between R2 and A1 samples at the same loading amount of Co₃O₄. These results suggest that the crystal phase of TiO₂ support has little influence on the water oxidation activity of Co₃O₄/TiO₂, most likely because it does not largely affect morphology of the deposited Co₃O₄ (Figure 1) and the resulting visible light absorption at least unless S_BET and morphology of TiO₂ are similar to each other (Figure 2).

Table 2. Photocatalytic activities of Co3O4-modified rutile and anatase TiO2 with different specific surface areas (SBET) for water oxidation under visible light (λ > 500 nm)a

Using various TiO₂ support materials that have different S_BET values, how S_BET of TiO₂ support influences the optimal loading amount of Co₃O₄ as well as photocatalytic activity was then investigated in detail. Table 2 (Entries 7–19) displays the results. In all cases examined, there were certain optimal loading amounts of Co₃O₄, depending on the TiO₂ support employed. For example, the optimal loading amount was 2.0 wt% for R1 (Entry 8), while 3.0 wt% for A2 (Entry 12). Results of photocatalytic reactions also indicated that the optimal loading amount tended to become lower when a lower surface area TiO₂ was used.

Among TiO₂ supports examined, A2 having the largest surface area (225 m² g⁻¹) was found to give the highest activity, which corresponded to an apparent quantum yield of ca. 0.15% at 500 nm.^19,20 As shown in Figure 3, the best-performing material, 3.0 wt% Co₃O₄/A2, contained highly dispersed Co₃O₄ nanoparticles of ~1 nm in size without noticeable aggregation.

Figure 3. TEM images of Co₃O₄-loaded R1 and A2. Some of the deposited Co₃O₄ nanoparticles are indicated by arrows. The data for 2.0 wt% Co₃O₄/R1 is reproduced from ref. 20 with permission from American Chemical Society.

The highly dispersed state of Co₃O₄ nanoparticles on A2 was also confirmed by X-ray absorption fine-structure spectroscopy (XAFS) measurements. As discussed in our previous study using the R1 sample, the loaded cobalt species on R1 are Co₃O₄ nanoclusters, when the loading amount is 2.0 wt%.²⁰ As shown in Figure 4, the EXAFS oscillation and the non-phase shift corrected FT-EXAFS spectra (k range: 3–12 Å⁻¹) of the A2 samples (modified with 3.0 and 5.0 wt% Co₃O₄) are similar to those of the 2.0 wt% Co₃O₄/R2 sample, indicating that the loaded cobalt species in the A2 samples are identical to Co₃O₄ nanoclusters. The lack of shell peaks at longer distances clearly indicates the nanosized nature of the loaded Co₃O₄ in these samples. In the FT of EXAFS spectra, the peak at approximately 2.5 Å, which can be assigned to Co(oct)–Co(oct) shell,²⁹ respectively, is stronger in the 5.0 wt% Co₃O₄/A2 sample than that in the 3.0 wt% analogue. This indicates the generation of larger Co₃O₄ nanoparticles at higher loadings. It is also noted that the possibility of the formation of CoTiO₃ phase in Co₃O₄/TiO₂ samples is excluded judging from the absorption edge of XANES spectra (Figure S5).

Figure 4. (A) Co–K edge EXAFS spectra of Co₃O₄-loaded TiO₂ samples, along with bulk Co₃O₄ for comparison. (B) FT of the k³-weighted EXAFS spectra of the same samples.

A TiO₂ support that is able to accommodate larger amounts of Co₃O₄ without aggregation, here A2, appears to be important for the construction of an efficient water oxidation system using Co₃O₄ nanoparticles and TiO₂. In this regard, a high surface area support would be useful. In the case of R1 that has lower S_BET than A2, increasing the loading amount of Co₃O₄ from 2.0 to 3.0 wt% resulted in aggregation of Co₃O₄, leading to lower activity.²⁰

Interestingly, however, R1 and A2, which exhibited almost the same activity with 2.0 wt% Co₃O₄ (Table 2, Entries 8–10), had different light-absorption profiles, as shown in Figure 5. Obviously, visible light absorption is more pronounced in 2.0 wt% Co₃O₄/R1 than 2.0 wt% Co₃O₄/A2 (even with 3.0 wt%). Therefore, the visible light absorption capability might be less important than the dispersion of Co₃O₄ in the Co₃O₄/TiO₂ system.

Figure 5. UV-VIS-NIR diffuse reflectance spectra of R1 and A2 with and without loading Co₃O₄.

It is known that surface structures of a semiconductor (e.g., exposed crystal facets) can affect photocatalytic activity, as individual surfaces may possess different reactivity, as reported by Ohno et al.³⁰ As shown in Figure 6, R3 and R4 have different morphologies despite their identical S_BET (1–2 m² g⁻¹). R3 consisted of aggregated primary particles of 100–400 nm, while R4 was well-crystallized particles 0.5–1 µm in size having exposed crystal facets. After modification of R3 and R4 with Co₃O₄, water oxidation reactions were conducted in the same manner. The results (as listed in Table 2, Entries 14–19) indicated that R4 worked better than R3, in particular at higher Co₃O₄ loadings (≥1.0 wt%).

Figure 6. SEM images of R3 and R4.

TEM images of R3 and R4 modified with 1.0 wt% Co₃O₄ are shown in Figure 7. In both cases, Co₃O₄ nanoparticles were highly dispersed in both cases. However, the average size of Co₃O₄ on R4 (2.4 ± 0.1 nm) was smaller than that on R3 (3.5 ± 0.3 nm), as indicated by the Co₃O₄ particle size distribution (Figure S6). Even though S_BET of these TiO₂ samples are small (1–2 m² g⁻¹), it was possible to realize highly dispersed Co₃O₄ nanoparticles on these supports. It thus appears that distribution of Co₃O₄ is not simply governed by S_BET of TiO₂ support, but may differ with respect to the surface structure (e.g., crystal facets) of support.

Figure 7. TEM images of 1.0 wt% Co₃O₄-loaded R3 and R4.

UV-VIS-NIR diffuse reflectance spectra of these samples were also compared. As shown in Figure 8, visible light absorption was more pronounced in the case of R3 than in R4, even though the same amount of Co₃O₄ (1.0 wt%) was loaded in both cases. It suggests that surface structure of TiO₂ can affect the light absorption properties of Co₃O₄/TiO₂, most likely due to different electronic interaction between Co₃O₄ and the TiO₂ surface. Despite the inferior visible light absorption of Co₃O₄/R4, R4 was a better-performing support than R3 (Table 2). Better light absorption of a photocatalyst should in principle contribute to an increase in the number of available photons to photocatalytic reactions. However, this was not the case. The present result strongly suggests that smaller particle size of Co₃O₄, which appears to be beneficial for water oxidation catalysis, on R4 is more important for enhancement of visible-light water oxidation activity, rather than the visible light absorption capability. According to a report by Li et al. on a plasmonic Au/TiO₂ photocatalyst for water oxidation, it is claimed that the interface between Au and TiO₂ is identified as the reaction site for water photooxidation.¹⁷ Because of the similarity in the reaction mechanism between Au/TiO₂ and Co₃O₄/TiO₂, active sites for water oxidation on the Co₃O₄/TiO₂ photocatalyst may be the interface between Co₃O₄ and TiO₂. However, identifying the real active site for water oxidation reaction in Co₃O₄/TiO₂ still remains a challenge.

Figure 8. UV-VIS-NIR diffuse reflectance spectra of R3 and R4 with and without loading of 1.0 wt% Co₃O₄.