Exploring robust electron transfer materials

As Pt is highly effective for surface HER and thus the best candidate for constructing the exterior metal layer^12,13, the first priority is to find excellent materials for the interface electron transfer layer. Herein, we adopted our previous approach of calculating the intrinsic electron transfer (IET) energy²⁸, defined by the energy change of moving an excess electron from TiO₂ bulk to supported metal particles in the absence of surface adsorbates, to compare the electron transferring ability among candidate metals. The initial state of a localized electron at a Ti site (Ti³⁺) in the bulk region of TiO₂ and the final state of delocalized electron densities on supported metal particles were confirmed by calculating the site-projected magnetic moments and Bader charges (see details in “IET energy calculation” in Methods)^15,28. It should be mentioned that, since there are many possible electron trapping sites in bulk TiO₂, we have calculated at least three different Ti sites for each metal/TiO₂ composite and the reported energies were the averaged value. Eventually, the IET results can be obtained by comparing the total energy along the electron transfer pathway, as summarized in Supplementary Table 1.

In Fig. 1, we assessed the electron transferring ability of 10 candidate metals (Supplementary Fig. 1) by calculating the IET energies. It is noteworthy that other transition metals (first ~ third periods) are generally too active with electron affinity < 1 eV³⁴ and can readily cause interface distortion (e.g., Hf and Ta) and/or oxide reduction (e.g., Ti, Mn, Co, Nb, Mo, and W) upon deposition (Supplementary Fig. 2)³⁵, and thus were not considered in the screening scope. It was found that noble metals in the VIII group (except Ru) are able to promote the directional electron transfer (namely, from bulk TiO₂ to the metal) with exothermic IET energies, but Ag and Au turn to be bad choices, consistent with experimental observations^5,11. In particular, the screening results (using the M₈/TiO₂ model) suggest that Rh, with strong IET energy of −0.27 eV, might be more efficient than Pt (−0.17 eV) in facilitating the directional electron transfer, but the main drawback is again its high price. Encouragingly, cheap metal Cu also shows an impressive IET energy of −0.15 eV close to Pt, serving as a promising electron transfer material. Moreover, the excellent electron transferring ability of Rh and Cu was further confirmed by using relatively larger M₁₉/TiO₂ models, obtained via ab initio molecular dynamics simulations (AIMD; Supplementary Fig. 3). Therefore, we theoretically proposed two bimetallic models, constructed by one layer of noble Rh or cheap Cu directly adhering to TiO₂ and a second layer of Pt capping on the top, to be competitive substitutes (either in benefits or costs) of the conventional Pt for photocatalytic H₂ production.

Fig. 1

Schematic of calculated IET energies in the presence of different metals on TiO₂.

Colors from red to blue indicate gradually enhancing IET energies (using M₈/TiO₂ or M₇/TiO₂ models) and the details are shown in Supplementary Table 1 and Supplementary Fig. 1.

Verifying the effectiveness of proposed catalysts

By utilizing geometry features of the stable Pt₁₃/TiO₂ composite (obtained via AIMD simulations; Supplementary Fig. 4) and replacing the first interface layer of six Pt atoms with Rh or Cu, we determined the bimetallic structure of Pt₇Rh₆/TiO₂(101) and Pt₇Cu₆/TiO₂(101) via unconstrained relaxations. No obvious structure changes were observed for the Pt₇Rh₆/TiO₂ (Fig. 2a) relative to that of Pt₁₃/TiO₂, indicating a lattice-matched alloying between Pt and Rh, whereas noticeable distortions for the Pt₇Cu₆/TiO₂ (Fig. 2c) can be attributed to the distinct atomic radius difference (~7%)³⁶. Further, IET energy calculations showed that the directional electron transfer (confirmed by charge density plots in Supplementary Fig. 5) indeed gets promoted for both the Pt₇Rh₆/TiO₂ and Pt₇Cu₆/TiO₂ composites, being −0.04 and −0.08 eV, respectively, as compared with the endothermic 0.05 eV for Pt₁₃/TiO₂^15,28. Such promotion is consistent with the downward shift of the Fermi level (E_f), from 0.45 eV below TiO₂ conduction band minimum (CBM) for Pt₁₃/TiO₂²⁸ to ~0.65 eV (see Fig. 2b, d), considering that a relatively lower energy level is beneficial to accept electrons. It is also necessary to address that the geometry of deposited metals has significant impacts on their physicochemical properties (Supplementary Fig. 6), signifying the importance of determining stable metal/oxide interfaces on reliable calculation results.

Fig. 2

Calculated structures, IET energies, and density of states for bimetallic metal/TiO₂ composites.

a, b for Pt₇Rh₆/TiO₂(101), c, d for Pt₇Cu₆/TiO₂(101), e, f for Pt₇Ir₆/TiO₂(101), g, h for Rh₇Pt₆/TiO₂(101), and i, j for Cu₇Pt₆/TiO₂(101), respectively. The DOS (density of states; using the hybrid HSE06 functional) for M₁₃ cluster and for TiO₂ surface are represented by blue and yellow-green curves, respectively. The valance band edge of TiO₂ is uniformly aligned to zero in DOS, and the vertical dot lines indicate the TiO₂ CBM and the E_f position, respectively. Calculated IET energies are also given in eV.

In addition, we also picked a less effective electron transfer metal of Ir for fair comparison, using similar approaches above. Consistently, the Pt₇Ir₆/TiO₂ composite (Fig. 2e) was found to be no better than Pt₁₃/TiO₂ with an endothermic IET energy of 0.09 eV, verifying the effectiveness of our screening approach. Furthermore, we investigated the electron transfer for models with inverted metal stacking sequence, namely the Rh₇Pt₆/TiO₂ and Cu₇Pt₆/TiO₂ composites (Fig. 2g, i). As expected, the electron transfer process turns to be retarded with endothermic IET energies and up-shifted E_f (Fig. 2g–j). Despite minor oscillations induced by the upper Rh or Cu layers, the electron transfer ability of the two inverted models generally resembles that of Pt₁₃/TiO₂. These results revealed evidently the electron transfer function of the first interface metal layer, and demonstrated that not only the right metal type (Rh or Cu) but also proper stacking sequence (Pt/Rh/TiO₂, Pt/Cu/TiO₂) are essential to achieve improved electron transfer efficiency.

Next, it is necessary to verify whether the catalytically efficient HER continues to proceed on the exterior Pt layer of the bimetallic composites. Considering the diverse ΔG_H results on different surface sites (Supplementary Table 2) and also the volcano-shaped ΔG_H ~ activity relationship^9,10, we uniformly selected two of the most active sites with ΔG_H approaches zero for computing the H–H coupling barrier (E_a^coup) and determining the HER activity of the alloy models. In Fig. 3a, b, one can see that both Pt₇Rh₆/TiO₂ and Pt₇Cu₆/TiO₂ give moderate hydrogen adsorption strength close to the volcano peak of ΔG_H = 0 eV^9,10. Moreover, we determined the coupling barrier E_a^coup to be 0.71 and 0.69 eV for Pt₇Rh₆/TiO₂ and Pt₇Cu₆/TiO₂ (Fig. 3d, e), respectively, being even slightly lower than the value of 0.75 eV on Pt₁₃/TiO₂²⁸. It is noteworthy that the coupling state on Pt₇Rh₆/TiO₂ shows quite late transition structure resembling the H₂ adsorption configuration, which was further confirmed by using the climbing image nudged elastic band method (CI-NEB; Supplementary Fig. 7)³⁷. Apparently, the HER activity on these alloy cocatalysts could be higher than extended Pt surfaces, owing to large E_a^coup of 0.88 and 1.07 eV on Pt(111) and Pt(100), respectively³⁸.

Fig. 3

Calculated structures and energies for H adsorption and H–H coupling on proposed metal/TiO₂ composites.

a–c H adsorption at two reactive sites (labeled as a and b) and d–f transition states of H–H coupling on the Pt₇Rh₆/TiO₂, Pt₇Cu₆/TiO₂, and Pt₆Cu₁₃/TiO₂ composites, respectively. Gibbs adsorption energies for the independent adsorption of H atom on each of the two reactive sites (H_a and H_b), the co-adsorption of two H atoms (H_ab), as well as the H–H coupling barrier E_a^coup, are given in eV.

Hence, by integrating good surface catalytic activity of Pt and robust electron transfer ability of Rh or Cu, the alloyed Pt/Rh or Pt/Cu cocatalyst breaks the intrinsic constraints between the two key photocatalytic factors for mono-type materials (e.g., Pt), and could in principle achieve superior performance than conventional Pt/TiO₂. It is worth noting that this strategy resembles the concept of using multiphase materials to break the scaling constraint between chemisorption energies of key intermediates (e.g., C and O in CO oxidation) in heterogenous catalysis³⁹ and also offers probabilities to approach the global activity maximum in photocatalysis. In particular, we underline the striking results of Pt/Cu/TiO₂ with enhanced photocatalytic performance but reduced catalyst cost as compared to pure Pt cocatalysts. This is further validated by using a trilayer Pt₆Cu₁₃/TiO₂ model with much lower Pt content, where suitable ΔG_H and E_a^coup are still well conserved (Fig. 3c, f). Furthermore, the Pt/Cu/TiO₂ composite was validated to be fairly stable via long-time AIMD simulations (>20 ps; Supplementary Fig. 8). Overall, TiO₂ supported Cu nanoparticles coated with a thin film of Pt skin are theoretically predicted to be efficient yet inexpensive catalysts for photocatalytic HER.

Generalized correlations

Finally, given the crucial role of electron transfer in photocatalysis, one may ask what physical properties might be related with the electron transfer efficiency across the interface. Here we calculated the effective potential, which represents the carrier interaction with other electrons in the system and the external electrostatic field^40,41, of relevant metal/TiO₂ composites using the HSE06 functional (Supplementary Table 1). In particular, the tunneling barrier height (Φ_TB), defined as the potential difference between the peak at the interface and TiO₂ slab (Fig. 4a), indicates the minimum energy cost of directional electronic injection via quantum tunneling^40,41. Intriguingly, the IET energies was found to correlate well with the Φ_TB, which although is typically mentioned in van der Waals junctions (e.g., metal/MoS₂ interfaces)⁴², among a wide range of metal/TiO₂ composites (Fig. 4b) holding short interface distances (Supplementary Table 1). Furthermore, despite the significant impact of particle size on IET energies as previously discovered²⁸, the linear relationship between the IET energy and the Φ_TB was found irrelevant to either the metal type (indicated by different colors) or particle size (different shapes in Fig. 4b). We notice that the number of sampling points might still be limited, but the good correlation coefficient in a wide sampling range encourages to regard such correlation as a general trend (at least for supported metals on TiO₂), which is beyond the considered metal/TiO₂ models in this work. As far as we know, this is the first piece of report on the electron transfer descriptor for photocatalytic reactions on metal/oxide catalysts. Considering the significance of electron transfer efficiency in photocatalysis, this finding could possibly promote future photocatalysts design via coordinating conventional catalytic descriptors (e.g., ΔG_H for HER) and the electron transfer properties Φ_TB.

Fig. 4

Calculated effective potential curve and the generalized linear correlation.

a Schematic of the effective potential (black curve) and Φ_TB (indicated by blue arrows) of Pt₁₃/TiO₂ using the hybrid HSE06 functional, as well as b the linear correlation between IET energies and Φ_TB on a wide range of metal/TiO₂ composites. While Pt, Rh, and Cu are highlighted in blue, olive, and orange color, the M₅/TiO₂, M₈/TiO₂, M₁₃/TiO₂, and M₁₉/TiO₂ are illustrated by solid square, dot, triangle, and pentagon in plot b, respectively. Symbols filled with mixed colors represent alloyed metal particles, where the stacking pattern from surface to the interface is consistent with the top-down coloring sequence.

DFT calculations

All density functional theory (DFT) calculations were performed using the Vienna Ab-initio Simulation Package (VASP) with the spin-polarization being considered. The DFT functional was utilized at the Perdew–Burke–Ernzerhof level. The project-augmented wave method was used to represent the core-valence electron interaction. The valence electronic states were expanded in plane wave basis sets with energy cutoff at 450 eV. The transition states were searched using a constrained optimization scheme and were verified when (i) all forces on atoms vanish and (ii) the total energy is a maximum along the reaction coordination but a minimum with respect to the rest of the degrees of freedom. The force threshold for the transition state search and structural optimization was 0.05 eV/Å. The dipole correction was applied throughout the calculations to take the polarization effect into account.

For the TiO₂ system, we have demonstrated previously^43,44 that the DFT + U method can yield similar structures and energies as those from the hybrid HSE06 functional. Here we mainly applied the DFT + U method in computing the thermodynamic properties (e.g., adsorption energy and energy barriers), where the on-site Coulomb correction was set on Ti 3d orbitals with an effective U value of 4.2 eV as suggested in literature works^45,46. To produce electronic structure properties more accurately (e.g., band gap, band edge position, and effective potential), we further performed hybrid HSE06 calculations utilizing the DFT + U geometry, in which the electronic minimization algorithm was set as ALGO = All with a very soft augmentation charge (PRECFOCK = Fast). The lattice parameters and band gap of bulk anatase TiO₂ determined via the HSE06 functional (a = b = 3.766, c = 9.448 Å; E_g = 3.31 eV) agree well with the experimental values (a = b = 3.776, c = 9.486 Å; E_g = ~3.2 eV).

The effective potential was calculated by including (i) the Hartree potential caused by the mean-field electrostatic interaction, (ii) the exchange-correlation potential due to the quantum mechanical nature of the electrons, and (iii) other external electrostatic interactions in the system. Technically, this can be done by invoking the LVTOT parameter in the input file using the VASP program (version 5.4 or later).

The adsorption energy of hydrogen atom (E_ad^H) is defined as the energy difference after and before the adsorption with respect to the gas phase H₂ molecule according to E_ad^H = E(total) − E(surface) − 1/2E(H₂), where E(surface), E(H₂), and E(total) are the energies for the clean surface, H₂ molecule in the gas phase, and hydrogen atom adsorbed on the surface, respectively. It is noteworthy that ΔG_H is obtained by adopting the entropy correction proposed by Norskov et al.^12,13 according to ΔG_H = E_ad^H + 0.24. The more negative the E_ad^H (or ΔG_H) is, the more strongly the H atom binds on surface.

Model construction for metal/TiO₂ composites

To properly accommodate the metal cluster, we utilized a relative large supercell of anatase (101) surface, that is a p(2 × 3) three-Ti-layer slab (72 Ti and 36 O atoms) with a vacuum thickness of 16 Å. We have checked the electron transfer energetics by enlarging the slab thickness to four-Ti-layer models, which produce similar results as that from three-layer calculations (Tables S1 and S2 in ref. ¹⁵). It is noteworthy that all the metal/TiO₂ composites containing non-Pt metals were constructed by utilizing geometry features of the optimal Pt₁₃/TiO₂ or Pt₈/TiO₂, or Pt₇/TiO₂ composites, determined via extensive AIMD simulations as shown in Supplementary Fig. 4, and replacing Pt with other metals by needs. Both the conjugate gradient and Quasi-Newton optimization methods were used to determine the stable structures.

Regarding the optimal structure of Rh₁₉/TiO₂, Cu₁₉/TiO₂ and bimetallic Pt₇Cu₆/TiO₂ composites, AIMD simulation was performed to investigate their structural evolutions. The simulation was carried out in the canonical NVT ensemble employing Nosé–Hoover thermostats. The temperature was set at 450 K that is taken from the temperature of hydrothermal treatment commonly used in experiments^21,47 and the time step was 1 fs. More than 20 ps AIMD simulations were performed (Supplementary Figs. 3 and 8) and from the equilibrated trajectory we selected structural snapshots in every 2 ps interval and fully optimized them until all forces diminish. The optimal composite structure can be determined as the lowest-energy configuration.

IET energy calculation

The extra photoelectron in the system was simulated by adding an excess electron into the supercell as common practice^8,46,48. We have also previously validated the approach by comparing the electron transfer results with those obtained by introducing an additional H atom on the opposite layer of TiO₂ slab in charge-neutral systems (Table S1 in ref. ²⁸.). The localization of the electron on a particular Ti site of TiO₂ can be initially configured and followed by DFT + U electronic structure optimizations. Initial magnetic moments on each atom are usually necessary in the input settings, although they will be optimized during the calculation. None of any constraints on the distribution/population of spin-polarized charges were imposed during the optimization.

To calculate the IET energy, the initial state is set as a localized electron at a Ti site in the bulk region of TiO₂. The as-formed Ti³⁺ cation shows distinctive features relative to the regular lattice Ti⁴⁺ and one can distinguish them easily as follows. (i) Geometry structures: the electron localization (generation of Ti³⁺ cation) is accompanied with the distinct elongations of Ti-O bonds by the outward movement of surrounding lattice O, consistent with literation reports^8,48. (ii) Electronic structure analysis: the position of electron localization can be further confirmed by calculating the site-projected magnetic moments (~0.8 µ_e) and Bader charges (negatively charged with ~0.4 | e| relative to lattice Ti⁴⁺ in bulk TiO₂), as visualized by the spin density plots in Supplementary Fig. 5. Whereas for the final state, the electron delocalization on metal particles can also be confirmed by (i) the absence of Ti³⁺ cation in TiO₂ but (ii) the increased electron quantities of ~0.4 |e| on metals (relative to the charge-neutral metal/TiO₂ model) via Bader charge analysis. It should also be noted that, our research focused on the post-photoexcitation electron migration process, which can be approximately regarded as a ground-state event, and therefore no excited states were simulated in calculating the IET energies.