Introduction

Photoelectrocatalysis has emerged as a prevalent technology with broad applications spanning energy conversion, environmental remediation, and organic synthesis1,2,3,4,5. High-performance photoelectrochemical (PEC) conversion typically requires sufficiently strong interactions (e.g., covalent bonding) between the photoelectrode surface and reactants to drive efficient charge transfer and activate inert chemical bonds within substrates, while avoiding excessively strong binding that may poison the catalyst (Fig. 1a)6,7,8,9,10. To achieve the balanced interaction, many efforts such as surface modification and defect engineering, etc., have been made on various photoelectrodes1,2,5,6,7,8,9,10,11, but this process is not only time-consuming but also requires elaborate design and precise operation. On the other hand, at the repulsive interfaces (e.g., cations on anodes), the direct covalent bonding cannot be achieved, challenging the interfacial charge transfer (Fig. 1b). These limitations underscore the need to move beyond covalent-interaction-dominated mechanisms, which currently restrict the further wide applications of photoelectrocatalysis. Compared to the covalent interaction, non-covalent weak interfaces between semiconductor surfaces and substrates are more favorable for achieving this due to their flexibility (Fig. 1c). Such interfaces exhibited structural adaptability, enhanced charge transport, and effective mitigation of Fermi-level pinning arising from disordered surface layers, which have been demonstrated in enzymatic systems and electronic materials12,13,14,15,16. Despite these merits, few PEC studies have investigated the non-covalent interactions between the catalyst surface and substrate molecules.

Fig. 1: Interfacial interactions between substrates and photoanodes.
figure 1

a Schematic of covalent bonding-dominant interface. Excessive covalent bonding of reactants on the catalyst surface facilitated the charge transfer, but this led to catalyst poisoning due to the tight bond connection. b Schematic of electrostatic repulsion interface. Long-range repulsion prevented reactant access, hindering the charge transfer process and reactant activation. c Schematic of non-covalent interface. Surface electric field-induced polarization of reactants via non-covalent interactions enabled efficient charge transfer while reversible adsorption mitigated surface poisoning.

Full size image

Direct cleavage of N−H bonds represents an ideal method to generate nitrogen-centered radicals (NCRs), which are essential intermediates in diverse transformations such as N–N coupling and C−N bond formation17,18,19,20,21. Nevertheless, the high bond dissociation energies (BDEs) of most N−H bonds (typically > 100 kcal mol−1) pose a challenge to break it17,19,22. Effective activation of N−H bonds generally necessitates a strong covalent interaction between the substrate and the catalyst. Unfortunately, such robust engagement introduces the potential risk of catalyst surface deactivation, complicating the design and application of efficient catalytic systems6,7,23,24. Consequently, current routes commonly involve the scission of the relatively weak N−O, N−S, N−N, and N−halogen (I, Br, Cl) bonds in various N-functionalized precursors21,25. Although recent advances in single-electron transfer (SET) strategies, particularly photocatalysis, have transformed N−H bonds into NCRs, these methods typically require exogenous oxidants/base, noble metal-based catalysts (such as Ru, Pd, and Ir), and expensive complex ligands20,21,26,27. Furthermore, inherent limitations in charge separation efficiency (or quantum yield) of photocatalysts also underscore the necessity for developing eco-friendly and inexpensive catalytic systems for directly generating NCRs.

Herein, we found that the non-covalent van der Waals interactions between BiVO4 photoanodes and a series of polarizable Cu-imine/amine/NH3 complexes enabled a notable charge transfer efficiency of 96%. This strategy facilitated direct generation of NCRs from native N−H bonds across various organic and inorganic nitrogen-containing substrates, circumventing the conventional requirements for pre-functionalization or stoichiometric reagents. Combined experimental and computational analysis revealed that Cu complexes in the Helmholtz layer strengthened the built-in electric field within the space charge region, which enhanced the photovoltage of bare BiVO4 from 0.18 to 0.53 V. The interfacial Cu ions simultaneously weaken N−H bonds through coordination effects. These interactions cooperatively promoted proton-coupled electron transfer and the subsequent chain propagation process, thereby regenerating Cu ions and releasing substantial NCRs. Towards the PEC conversion of model substrate benzophenone imine (BI) to benzophenone azine (BA), a key step in hydrazine synthesis, the Cu complex-BiVO4 system achieved a near-unity Faradaic efficiency (96.4%) and selectivity (>99%) for the bimolecular radical coupling. In the scale-up experiment, a photocurrent of 409 mA and a high yield rate of 6069 μmol h−1 were obtained in a solar-driven flow reactor.

Results and Discussion

Cu-assisted generation of NCRs on BiVO4

The BiVO4 layer was deposited on a fluorine-doped SnO2 (FTO) substrate through an electrodeposition-calcination process (more details in the Methods). Characterizations of the BiVO4 photoanodes were shown in Supplementary Fig. 1.

For the direct generation of NCRs from N−H precursors, linear sweep voltammetry (LSV) was first tested. Under AM 1.5 G illumination, BiVO4 photoanodes exhibited limited activity (2.3 mA cm⁻2 at 0.6 VAg/AgCl) for benzophenone imine (BI) oxidation, comparable to the low water oxidation performance (1.1 mA cm−2, Fig. 2a). Notably, the introduction of 10 ppm Cu ions in the electrolyte boosted the photocurrent by 257% to 5.9 mA cm⁻2, demonstrating the accelerated charge transfer kinetics. This enhancement was proven universal across various N-containing substrates, including imines, amines, and NH3. The photocurrent for benzylamine (PhCH2NH2) oxidation surged from 2.8 to 6.8 mA cm⁻2 upon adding 10 ppm of Cu ions (Fig. 2b), while the oxidation of alkylamine (such as EtNH2) and ammonia (NH3) exhibited 229% (2.8 to 6.4 mA cm⁻2) and 270% (2.3 to 6.2 mA cm⁻2) current density improvements, respectively (Fig. 2c, d). These measured photocurrents approaching the theoretical maximum (7.5 mA cm−2) underscored the highly improved charge transfer kinetics in the Cu complexes-BiVO4 system.

Fig. 2: Cu-assisted generation of NCRs on BiVO4 photoanodes.
figure 2

LSV curves (50 mV s−1) of BiVO4 photoanodes for the oxidation of 0.2 M (a) BI, b PhCH2NH2, c EtNH2 and d NH3 with or without 10 ppm of Cu ions. In situ EPR spectra of (e) DMPO-BI•, f DMPO-•NHCH2Ph, g DMPO-•NHEt, h DMPO-•NH2 with or without 10 ppm of Cu ions on BiVO4 photoanodes.

Full size image

In corresponding electron paramagnetic resonance (EPR) measurements with using 5,5-dimethyl-1-pyrroline N-oxide (DMPO) as the spin-trap reagent, only a faint signal of DMPO-NCRs adducts appeared during photoelectrolysis without Cu ions (Fig. 2e–h), consistent with the low photocurrents in LSV curves. This indicated that BiVO4 photoanodes had limited ability of directly cleaving the N−H bonds to produce NCRs. Upon addition of Cu ions, intensified EPR signals of various NCRs including iminyl (BI•), benzylamine-derived radicals (PhCH2NH•), aliphatic aminyl radicals (EtNH•), amino radicals (NH2•) and anilinyl radicals (PhNH•) were detected (Fig. 2e–h and Supplementary Fig. 2a–c). Cu alone (without photoelectrolysis) did not produce such intense EPR signals of NCRs (Supplementary Fig. 2d). The consistent enhancement observed in LSV and EPR measurements across various nitrogenous substrates demonstrated the universal effectiveness of Cu complexes-BiVO4 system in accelerating charge transfer process and producing NCRs from the cleavage of N−H bonds.

The model reaction: BI oxidation reaction

The oxidative homocoupling reaction of BI to benzophenone azine (BA) was selected as the model reaction (noted as BIOR) for detailed study. This reaction avoided the interference from α-H atoms and did not involve the transformation of NCRs into carbon-centered radicals, making it an ideal model reaction for elucidating the mechanisms of NCR formation (Supplementary Fig. 3). Moreover, BIOR was a critical step in the Hayashi or Bayer process for hydrazine synthesis28,29,30, and its execution via PEC methods offered a sustainable and environmentally friendly approach to producing hydrazine.

The detailed optimization procedure for the solvent, atmosphere conditions and the identification of cathodic process were showed in Supplementary Figs. 4–7. Under optimized conditions, the effect of Cu ions concentration was systematically tested. As shown in Fig. 3a, the presence of merely 1 ppm of Cu ions could enhance the photocurrent from 2.3 to 5.4 mA cm−2 at 0.6 VAgCl/Ag, accompanied by a decreased onset potential of 250 mV. The photocurrent exhibited a positive correlation with Cu ion concentration (Fig. 3b). The incident photon-to-current conversion efficiency (IPCE) further validated the enhanced performance. For water oxidation reaction (WOR) and BIOR, the maximum IPCEs were only 5% and 21%, indicating of the inherently low activity of BiVO4 toward these reactions. Subsequent introduction of Cu ions enhanced the IPCE to 78% and its integrated photocurrent density reached 5.9 mA cm−2, consistent with the measured photocurrent density (Supplementary Fig. 8). To further identify the product of BIOR, 30-min photoelectrolysis experiments under various Cu concentrations were carried out. In the absence of Cu ions, high performance liquid chromatography (HPLC) analysis revealed no detectable formation of the N−N coupling product BA (Supplementary Fig. 9), and the in-situ attenuated total reflectance Fourier-transform infrared (ATR FTIR) measurements identified that benzophenone (BP) derived from the hydrolysis of BI was the dominant product (Supplementary Figs. 10 and 11). Notably, the addition of 1 ppm Cu ions improved the yield rate of BA from 0 to 66.8 μmol cm−2 h−1, accompanied by a selectivity of 80.1% and a faradaic efficiency (FE) of 77.1% (Fig. 3c). As the Cu ions concentration increased to 5–15 ppm, the yield rate of BA plateaued at approximately 95 μmol cm−2 h−1, achieving near-unit selectivity (>99%) and a high FE of 96.4% for BA. Further increase in Cu ion concentration led to a slight decrease in the PEC performance due to the enhanced hydrolysis and the greater light harvesting losses from the yellow Cu-BI complexes in the solutions (Supplementary Fig. 11).

Fig. 3: The PEC performance of Cu-assisted BIOR on BiVO4 photoanodes.
figure 3

a LSV curves, b extracted photocurrent density and c product analysis obtained in electrolyte containing 0.2 M BI with variable concentration of Cu ions. d Time-dependent conversion profile of Cu-assisted BIOR at 0.6 VAgCl/Ag with 0.3 mmol BI and 10 ppm Cu ions. e The cyclic stability test at 0.6 VAgCl/Ag with or without 10 ppm Cu ions. f BA yield under various conditions.

Full size image

To evaluate the operational durability, we prolonged photoelectrolysis and performed the cyclic stability test. With increasing passed charge, the concentration of BI gradually decreased, while the generation of BA correspondingly increased (Fig. 3d and Supplementary Fig. 12). The selectivity of BA maintained within the range of 95-98% up to a passed charge of 25 C, accompanied by a high yield of 85.2%. The BiVO4 demonstrated high stability in the presence of 10 ppm Cu ions, maintaining a consistent photocurrent and high yield of 82.7 ± 2.7% over 8 cycles (Fig. 3e and Supplementary Fig. 13). In sharp contrast, in the absence of Cu ions, the BIOR process produced negligible amounts of N−N coupling products. Structural characterizations reflected that the BiVO4 photoanode tested in the presence of Cu ions maintained its pristine structure and morphology following the cyclic stability test. Conversely, the BiVO4 tested without Cu ions showed serious anodic photocorrosion (Supplementary Fig. 14).

A series of control experiments were further performed. When H2O2 was introduced to scavenge holes, a decrease in BA yield was observed, dropping from 85.2% to 0.6% (Fig. 3f, column 2). It indicated that Cu-assisted BIOR required the involvement of holes from BiVO4 photoanodes. Without the assistance of external bias, the bulk carrier in photoanodes had insufficient lifetimes to participate in the catalytic reaction31, hence no reaction occurred under photocatalytic conditions (noted as PC, Fig. 3f, column 3). This also ruled out the possibility of Cu ions acting as catalysts to drive the BIOR under ambient conditions32,33. Tert-butyl alcohol (TBA), commonly recognized as the scavenger for •OH radicals and CuIII species34,35, had a negligible effect on the Cu-assisted BIOR upon addition. The contributions of •OH radicals and CuIII species were thus excluded (Fig. 3f, column 4). Subsequently, DMPO was introduced as a quencher to trap transient reactive species. While DMPO introduction exhibited minimal influence on the photocurrent (Supplementary Fig. 15a), the generation of BA was essentially suppressed (Fig. 3f, column 5). Meanwhile, the adduct formed by DMPO and iminyl radicals (BI•) was successfully detected via gas chromatography-mass spectrometry (GC-MS, Supplementary Fig. 15b). This observation confirmed the generation of BI• radicals in the Cu complex-BiVO4 system. The strong positive correlation between photocurrent and BI• formation rate in the in-situ EPR measurements and the semi-quantitative analysis also demonstrated that BI• radicals originated from the Cu-assisted BIOR (more details in Supplementary Fig. 16). There was negligible production of BA in the absence of Cu ions (Fig. 3f, column 6). These experiments demonstrated that the holes generated in BiVO4 photoanodes would oxidize BI substrates with the assistance of Cu ions, followed by the formation of BI• radicals.

The non-covalent interaction between BiVO4 and Cu-BI complexes

The rinse test and Cu distribution analysis revealed that Cu ions mainly operated within the solution or in close proximity to the electrode surface, without directly bonding to the photoanode (Supplementary Fig. 17a-d). We also intentionally deposited Cu onto the BiVO4 surface, but the Cu-modified BiVO4 photoanode failed to produce BA (yield = 2.3%) under the same reaction conditions, reconfirming that Cu ions functioned through solution-phase complexation rather than via catalyst surface modification (Supplementary Fig. 17e-l). To identify the chemical form of Cu ions, ultraviolet-visible (UV-vis) absorption spectroscopy and IR measurements were carried out. In a BI-free acetonitrile solution, Cu ions formed a [Cu(MeCN)4]2+ solvatocomplex, with two characteristic absorption peaks (Fig. 4a). The strong ligand-to-metal charge transfer (LMCT) absorption peak appeared at 330 nm, while a very weak d-d transition band peaked at 750 nm36. Consequently, the solution appeared almost colorless in the optical image (inset in Fig. 4a). When Cu ions and BI molecules coexisted, the LMCT absorption band shifted from the UV to the visible light region, increasing with higher Cu ions concentrations and resulting in a yellow solution. This result suggested that BI molecules had displaced MeCN ligands to form Cu-BI complexes. ATR FTIR spectroscopy was exploited to further investigate the coordination between Cu ions and BI molecules (Fig. 4b, c and Supplementary Fig. 18). As Cu ions concentration increased, the N−H stretching vibration peak at 3362 cm−1 gradually shifted to lower wavenumbers (Fig. 4b), indicating that the N−H bonds were progressively weakened through coordination with Cu ions. The addition of Cu had a minimal effect on the C−H vibrations of the aromatic ring (3083, 3068, and 3033 cm−1), confirming the selective activation of N−H bonds over inert C−H moieties. The red shift of the C=N stretching vibration (1604 cm−1) indicated an elongation of the C=N bond, directly validating BI-to-Cu coordination through the imine nitrogen (Fig. 4c). Additionally, the skeletal vibrations of the benzene ring (1570 cm−1) in the Cu-BI complexes would shift towards higher wavenumbers and be weaker with Cu ion concentrations. This phenomenon was likely attributed to the large steric hindrance of BI ligands and the square-planar-like structure of the four-coordinate Cu complexes, which enhanced the intrinsic rigidity of Cu complexes and restricted the vibrations of the benzene ring37. Furthermore, based on the fitted results of X-ray absorption spectra (XAS) and the work reported by Gagliardi et al.18, the Cu-BI complexes were identified as a four-coordinate geometry, namely CuBI4 complexes (Supplementary Fig. 19).

Fig. 4: The non-covalent interaction between BiVO4 and Cu complexes.
figure 4

a Uv-vis spectra of BI solutions in MeCN with variable concentration of Cu ions. Inset was the optical image of the solutions. b High wavenumber region and c low wavenumber region in ATR IR spectra of BI solutions in MeCN with variable concentration of Cu ions. d Calculated BDEN−H of BI molecules, Cu-BI complexes and corresponding adsorbed configurations on BiVO4 surface. e Radial distribution functions, g(r), between Cu atom and the N, O, Bi, V atoms. Inset was the representative configurations of Cu-BI complexes on the BiVO4 surface. Cu (orange), N (blue), H (white), C (sky blue), O (red), Bi (pink), V(Gray). f Isosurface plots (s = 0.3) of the RDG for the BiVO4/Cu-BI complexes interfaces. The Multiwfn analysis was based on the equilibrium configuration, without considering polarization effects that may arise from photovoltages. g Schematic illustration of the measured energy diagram at BiVO4/Cu-BI complexes interfaces. The diagram included the conduction band minimum (ECBM), valence band maximum (EVBM), electron quasi-Fermi level (Ee), hole quasi-Fermi level (Eh), and the redox potentials of active species in the Helmholtz layer (Eredox). h Comparison of photovoltage for WOR, BIOR and Cu-assisted BIOR with different Cu ions concentrations. The error bars represented the standard deviations across three parallel experiments.

Full size image

Benefiting from the formation of Cu complexes, the bond dissociation energies of N−H bonds (BDEN−H) of BI molecules decreased from 1.61 eV to 1.32 eV, as obtained in the density functional theory (DFT) calculations (Fig. 4d and Supplementary Fig. 20). Taking the influence of BiVO4 surfaces into consideration, the BDEN−H was further reduced to 1.19 eV, whereas that of pure BI molecules remained nearly unchanged. This indicated a synergistic interaction between the BiVO4 surface and Cu-BI complexes, proving a facile pathway to cleavage N−H bonds. To validate theoretical calculations, the H atoms in C=NH moiety of BI molecules and the reactive H atoms of H2O molecules were substituted with D atoms to examine H/D kinetic isotope effect (KIE, Supplementary Fig. 21). The Cu-free system showed pronounced KIE values (1.4–2.0), indicating of the kinetically sluggish N−H bond dissociation as the rate-determining step (RDS). Notably, the Cu complex-BiVO4 system exhibited near-unity KIE, which suggested that the cleavage of N−H bonds was highly accelerated and no longer involved in the RDS.

To dissect the structural information of the Cu complex and its interaction with the BiVO4 surface, the ab initio molecular dynamics (AIMD) simulations were conducted with an explicit solvent model (acetonitrile). The radial distribution function (RDF) centered on Cu atoms was plotted based on the simulation trajectories, as shown in Fig. 4e. The first coordination shell of Cu ions consisted of N atoms, with a Cu−N bond length of approximately 2 Å and a coordination number of 4, consistent with the fitted results of Cu-BI complexes in the extended X-ray absorption fine structure (EXAFS) data (Supplementary Fig. 19). In contrast, the distances between Cu and the surface O, Bi, and V atoms of BiVO4 exceeded 4 Å, exceeding the bonding threshold of covalent bonds (dmax(Cu−Bi/O/V) = 2.24 Å) but falling within the range of Helmholtz layer (about 3–6 Å in thickness38,39). Consequently, Cu complexes mainly resided in the Helmholtz layer rather than bonded onto the BiVO4 surface, consistent with the experimental results (Supplementary Fig. 17). To further decipher the interaction type between this Cu complex and the BiVO4 surface, a noncovalent interaction (NCI) analysis was employed and visualized based on the Multiwfn software40,41,42. This analysis method relies on the principle of identifying regions characterized by low electron density with a reduced density gradient (RDG)−sign(λ2)ρ function, enabling the distinction of various noncovalent interactions such as strong attractive (SA, sign(λ2)ρ < 0 a.u., dipole-dipole or hydrogen bonding), strong repulsive (SR, sign(λ2)ρ > 0 a.u., nonbonding) or van der Waals (vdW, sign(λ2)ρ close to 0 a.u.) interactions40. As shown in Fig. 4f, the RDG among various groups was visualized with the colored isosurfaces, while the color bar reflected the value of sign(λ2)ρ for various non-covalent interactions. It exhibited green isosurfaces (i.e., sign(λ2)ρ close to 0 a.u.) between the Cu-BI complexes and BiVO4 surface, demonstrating their non-covalent combination with a vdW interaction. The Cu-BI complex was highly polarizable (Supplementary Figs. 22 and 23), and thus the surface-charge-induced dipole contributes to the efficient vdW interaction.

With the presence of the non-covalent interaction between BiVO4 surface and Cu complexes, a high charge transfer efficiency of 96% was observed in both intensity-modulated photocurrent spectroscopy, scavenger experiments and transient photocurrent measurements (Supplementary Fig. 24). This efficiency was competitive with conventional PEC systems relying on covalent interactions4,9,43. To unravel the intrinsic mechanisms of non-covalent interactions on interfacial charge transfer kinetics, open-circuit potential (OCP) measurements combined with adsorption energy analysis and Mott-Schottky analysis were carried out (more details in Supplementary Fig. 25 and 26)44,45. As shown in Fig. 4g, it was shown that equilibrium energy diagram, particularly the electrochemical potentials of the solution-phase redox couple (Eredox), exhibited minimal changes after the introduction of BI. This implied that BI molecules did not penetrate the Helmholtz layer, or the Eredox of BIOR was close to that of WOR. Upon adding 10 ppm of Cu ions, a notable shift in Eredox was observed. It indicated that Cu-BI complexes predominantly functioned in the Helmholtz layer near the photoanode surface and altered the solution-phase redox couple from H2O or BI molecules to Cu-BI complexes (Supplementary Fig. 26f). To further support this transition in redox species, the adsorption energies of free BI molecules and Cu-BI complexes were calculated (Supplementary Fig. 25). Compared with free BI molecules (0.07 eV), the more negative adsorption energy of Cu-BI complexes (−1.50 eV) indicated their preferential enrichment near the BiVO4 surface, consistent with the shift in the Eredox of solution-phase redox couples. Benefiting from the band edge pinning of BiVO4 photoanodes (more details in Supplementary Fig. 26), changes in the Eredox induced by Cu ions would trigger a compensatory band bending reconfiguration to maintain equal Fermi levels on both sides of the interface45,46. As a result, a much higher photovoltage of 0.53 V was obtained upon adding 10 ppm of Cu ions, compared to 0.18 V measured in the BIOR without Cu ions (Fig. 4h). Electrochemical impedance spectroscopy (EIS) studies further confirmed the outer-sphere charge transfer pathway, supporting the non-covalent interactions between Cu complexes and BiVO4 (Supplementary Fig. 27). It is noted that on anodes, the Helmholtz layer should be mostly formed by anions (e.g., electrolyte anion, BF4−), but in this work the possibility of anion-mediated modulation of solution-phase redox potentials was excluded (Supplementary Fig. 28). Instead, the presence of Cu-BI complexes induced a notable change of the solution-phase Fermi level. Consequently, the direct interaction between Cu-BI complexes and BiVO4 surfaces was considered as the critical factor in this study.

The regeneration of Cu ions on BiVO4 surfaces

As shown in Fig. 5a, the yield rate of BA exhibited a zero-order kinetics with respect to Cu ions near the optimized Cu ions concentration of 10 ppm. Considering the low concentration of Cu ions (at ppm levels), we speculated that the zero-order kinetics originated from the rapid regeneration of Cu ions near the BiVO4 surface, thereby enabling sustainable catalytic cycling with merely trace amounts of Cu ions (inset in Fig. 5a). And the relatively positive adsorption energy of BI• radicals (+0.95 eV) compared to that of Cu-BI complexes (−1.50 eV) also suggested that the generated BI• radicals were more likely to desorb into the solution, rather than being retained at the photoanode surface (Supplementary Fig. 25).

Fig. 5: The regeneration of Cu ions on BiVO4 surface.
figure 5

a The rate law analysis of Cu ions. Inset described the proposed pathway for the rapid regeneration of Cu ions. b Schematics of polarization-cathodic CV scans. Cathodic CV of BiVO4 photoanodes measured in BI solutions with or without 10 ppm Cu ions after 40 s of polarization at 0.6 VAgCl/Ag. c Comparison of Jpolar and Jpc measured in variable BI and Cu ions concentrations. d Energy profile of Cu ions regeneration on BiVO4 via ligand exchange and chain propagation pathways. e The configurations of the initial state (IS), transition state (TS), and final state (FS). Cu (orange), N (blue), reactive H (pink), other H (white), C (sky blue).

Full size image

Polarization-cathodic cyclic voltammetry (CV) scans were carried out to further support the hypothesis. Under PEC conditions, positive polarization facilitated the generation and accumulation of oxidized species on the BiVO4 surface. Subsequent dark cathodic scans revealed the reduction behavior of these metastable species (Fig. 5b). A well-defined reduction peak emerged at −0.4 VAgCl/Ag after the BIOR. Introduction of 10 ppm Cu increased the cathodic peak current from −0.7 to −3.4 mA cm⁻2, indicating of Cu-mediated enhancement of these redox-active species. This enhancement aligned with the promoted generation of BI• radicals with the assistance of Cu ions (Fig. 2). Therefore, the observed reduction peak may correspond to the reduction of BI• radicals (Supplementary Figs. 29 and 30). The strong correlation between anodic polarization current (Jpolar) and cathodic peak current (Jpc) confirmed the direct proportionality between the generation of oxidized species and their subsequent reduction (Fig. 5c). Notably, Jpc exhibited pronounced concentration dependence on the BI but unaffected by Cu ions concentrations above 10 ppm. Complementary experiments using DMPO as scavenger for free radicals caused 47% peak suppression (Supplementary Fig. 31). These results collectively identified the reducible species as free BI•, and reconfirmed the critical role of Cu ions in facilitating NCRs generation. Although the potential reduction of Cu ions was initially considered, the TBA scavenger experiments (Fig. 3f) and quantitative analysis based on Randles-Sevcik equation (0.02 ~ 0.09 mA cm−2, Supplementary Note 1) effectively ruled out this possibility.

The observed BI-dependent yet Cu-independent Jpc also revealed that the concentration of free BI• radicals was not limited by the low concentration of Cu ions but governed by BI concentrations, demonstrating the existence of a substrate-driven Cu ions regeneration process. This dependency of BI• yield rate on the concentration of BI concentrations was also observed in the in-situ EPR measurements and semi-quantitative analysis (Supplementary Fig. 16). Therefore, the substrate-driven Cu ions regeneration involved a proton-coupled electron transfer (PCET) step to generate metastable Cu-BI• intermediates from Cu-BI complexes, followed by subsequent reaction with additional BI molecules that simultaneously liberated free BI• radicals and regenerated the Cu-BI complexes (inset in Fig. 5a). This chain propagation process circumvented the limitation imposed by low Cu ions concentration, rationalizing the observed zero-order kinetics in Cu ions (Fig. 5a). The activation order among all four BI ligands of the Cu–BI complexes was also considered. The BDEN−H of the four ligands showed minimal difference. Given kinetic factors such as mass transport and steric hindrance, the dissociation of N–H bonds was expected to be more favorable on the ligands oriented toward the solution phase (Supplementary Fig. 32). Besides chain propagation pathway, an alternative ligand exchange pathway was considered, where additional BI molecules displaced BI• ligands in Cu-BI• intermediates to release free radicals and regenerate Cu ions (Supplementary Fig. 33). Static DFT calculations revealed that the chain propagation process exhibited a low activation free energy of 0.34 eV (Fig. 5d, e), while the ligand exchange involved considerable configurational reorganization, resulting in a much higher energy barrier of 1.11 eV. Therefore, the chain propagation became the predominant pathway to regenerate Cu ions, thereby continuously facilitating the BIOR process.

Overall pathways for Cu-assisted BIOR on BiVO4 surface

Finally, the N−N coupling pathway was investigated through CV scans at varying scan rates. As illustrated in Fig. 6a, the oxidation peak observed in the forward scan was attributed to the generation of free BI• radicals via the above chain propagation pathway, while the backward peak current corresponded to the reduction of these BI• radicals. Since the voltammetric current magnitude directly correlated with the concentration of redox-active species (i.e., BI• radicals), this feature enabled quantitative analysis of BI• radical concentrations. It was found that the ratio of cathodic and anodic peak currents (Jpc/Jpa) was less than unity and decreased with the scan rate (Fig. 6b). This observation indicated the occurrence of a chemical reaction consuming free BI• radicals during the elapsed time between the anodic and cathodic peaks, consequently reducing the backward peak current. By modulating scan rates, the reaction time (t) for radical consumption was controlled, and reaction kinetics of the chemical step could be extracted through the peak ratio evaluation (derivation detailed in Supplementary Note 2)47,48. A linear correlation observed in the second-order kinetic plot (R2 = 0.995) demonstrated a bimolecular radical coupling mechanism between two free BI• radicals, ultimately yielding the N−N coupling product, BA (Fig. 6c). And its slope provided an estimate of the BI• radical lifetime on the order of tens to hundreds of seconds, which was further demonstrated by the relaxation-EPR detection process (Supplementary Fig. 34). Therefore, the chain propagation pathway sustained radical generation while bimolecular coupling served as the product-forming termination step.

Fig. 6: Overall pathways for Cu-assisted BIOR on BiVO4 surface.
figure 6

a CV curves of BiVO4 photoanodes measured in BI solutions containing 10 ppm Cu ions at variable scan rates. b Scan rate dependence of cathodic-to-anodic peak current ratio (Jpc/Jpa). c Second-order kinetic analysis of BI• radicals derived from peak current ratio evolution. d Illustration of the overall mechanism for BIOR in the Cu complex-BiVO4 system.

Full size image

According to the above experiments and calculations, the overall mechanism of Cu-assisted BIOR on BiVO4 photoanodes was schematically depicted in Fig. 6d. Cu ions firstly coordinated with BI molecules to form Cu-BI complexes, inducing N−H weakening of BI ligands. The non-covalent interactions between Cu complexes and BiVO4 synergistically enhanced the built-in photovoltage and established an efficient outer-sphere charge transfer pathway, thereby accelerating hole transfer from BiVO4 to Cu-BI complexes. This process was coupled by the dehydrogenation of the Cu-BI complexes, generating metastable Cu-BI• intermediates. Subsequently, Cu-BI• intermediates underwent chain propagation pathway with additional BI molecules, simultaneously releasing free BI• radicals and regenerating the Cu-BI complexes. The rapid catalytic cycle rationalized the observed zero-order dependence on Cu ion concentration. Ultimately, bimolecular coupling of two free BI• radicals produced the final BA product.

Scale-up experiments of PEC Cu-assisted BIOR

Based on the synergistic interplay between Cu complexes and BiVO4 photoanodes, we successfully scaled up the N − N coupling reaction by using a PEC flow reactor49. The size of the BiVO4 photoanode was enlarged by 35-fold to approximately 70 cm2 (Supplementary Fig. 35a), paired with a Pt-coated Ti plate (Pt@Ti) as the cathode in a two-electrode system (Fig. 7a). This scaled system delivered a photocurrent of 409 mA under a low cell voltage of 0.8 V, demonstrating its scalability (Fig. 7b). Subsequently, a commercial Si solar cell (0.78 V, Supplementary Fig. 35b) was integrated to drive the Cu-assisted BIOR. The process maintained a stable photocurrent of 315 mA, achieving a high BA yield rate of 6069 μmol h−1 with FE exceeding 90% (Fig. 7c). This yield rate was comparable with conventional thermal, photochemical and electrochemical N−N coupling methods (typically <100 μmol h−1) (Supplementary Table 1). After 2.5 hours of photoelectrolysis, the reaction mixture underwent purification with n-pentane, yielding 4.88 g of pale-yellow solid (inset in Fig. 7d). 1H and 13C nuclear magnetic resonance (NMR) analysis confirmed the solid as target product BA, achieving the gram-scale synthesis (Fig. 7d, e). The high photocurrent, FE, and compatibility with solar energy input underscored the practical viability of this approach.

Fig. 7: Scale-up experiments of PEC Cu-assisted BIOR.
figure 7

a Photograph of the experimental setup in the gram-scale synthesis and the illustration of the two-electrode PEC flow reactor. b LSV curves of the scaled BiVO4 photoanodes measured in 0.2 M BI with 10 ppm Cu ions in a two-electrode cell. c The average photocurrent, yield rate and FE(BA) as a function of reaction time. d 1H and e 13C NMR spectra (400 MHz, CDCl3) of collected solid products. The insets were the photograph of separated product and its structure, respectively.

Full size image

Finally, the collected BA solid, serving as the hydrazine precursor, was dissolved in MeCN solutions containing 10 vol% HCl to facilitate hydrolysis (more details in Supplementary Fig. 36). After 1 h of hydrolysis, nearly all of BA substrates were successfully converted into hydrazine and BP products with a yield exceeding 99%, demonstrating an efficient pathway for hydrazine synthesis.

Discussion

We established that non-covalent interactions dominated the fast interfacial dynamics between BiVO4 photoanodes and N-coordinated Cu complexes. Mechanistic studies using BIOR as a model reaction revealed that the charge transfer process was highly enhanced via the non-covalent interactions, which synergized with the coordination effect to promote the NCRs generation from directly cleaving inert N−H bonds in various imines/amines/NH3. The practicality of this strategy was validated in a PEC flow reactor, achieving a photocurrent of 409 mA and the gram-scale synthesis for the N−N coupling products. This work offered valuable insights into the role of non-covalent interactions in the PEC system and provided an environmental-friendly and inexpensive PEC method for the direct generation of NCRs from N−H precursors.

Methods

Materials and reagents

The fluorine-doped tin oxide substrates (FTO, 2.2 mm) were manufactured by Nippon Plate Glass Co., Ltd. Potassium iodide (>99%), p-Benzoquinone (99%), Vanadyl acetylacetonate (99%), Benzophenone azine (99%), and Tert-butyl alcohol (TBA, 99.8%) were purchased from Acros. Tetraethylammonium hydroxide (25% in H2O), Trifluoroacetic acid (99%), and Carbon tetrachloride (99.5%) were purchased from Aladdin. Petroleum ether (60-90 °C) and Acetonitrile (99.9%) were purchased from Concord. 5,5-Dimethyl-1-pyrroline N-oxide (DMPO) was purchased from Concord Dojindo Laboratories. Tetraethylammonium nitrate (TEANO3, 99%), Tetraethylammonium perchlorate (TEAClO4, 98%), Tetraethylammonium chloride (TEACl, 98%), Hydrochloric acid (37% in H2O), Ethylamine (68.0-72.0% in H2O), Benzylamine (99%), Benzophenone imine (95%), Benzophenone (99%), Tetraethylammonium tetrafluoroborate (TEABF4, 99%), Copper nitratehydrate (>99%), Sodium perchlorate (99%), Ammonia solution (28.0-30.0%), Sodium hydroxide (99%), Sodium sulfite (98%), and Deuterium oxide (99.9 atom% D) were purchased from Innochem. The deionized water (D.I. water) was obtained by a purification equipment (Millipore, Milli-RO Plus). All chemicals were used as received without further purification.

Structural characterization

X-ray diffraction (XRD) patterns were collected on an X-ray diffraction instrument (Rigaku, D/max 2500), scanning 2θ angles from 15o to 60o at a rate of 5o min−1. The surface morphology of samples was examined by scanning electron microscope (SEM, Hitachi, SU8010) and transmission electron microscope (TEM, Thermo Fisher Scientific, FEI Talos 200X). Ultraviolet-visible diffuse reflectance spectrum (UV-vis DRS) was acquired on a Hitachi U-3900 spectrophotometer, equipped with an integrating sphere. Chemical states and elemental compositions were analyzed via X-ray photoelectron spectroscopy (XPS, Thermo Scientific, ESCALab 250Xi) using 200 W monochromatic Al Kα radiation. All XPS spectra were energy-calibrated using the adventitious carbon C 1 s peak at 284.8 eV as a reference.

Photoanodes preparation

BiVO4 photoanodes were fabricated through an electrodeposition-calcination method50. Initially, 0.4 M KI solution was prepared by dissolving 3.32 g of KI in 50 mL of D.I. water, followed by adjusting the pH to 1.7 with HNO3. Subsequently, 0.97 g of Bi(NO3)3•5H2O was added to the solution, forming a solution containing 0.04 M Bi(NO3)3 and 0.4 M KI. This mixture was then combined with 20 mL of ethanol containing 0.23 M p-benzoquinone to form the deposition electrolyte. Electrodeposition was performed in a standard three-electrode cell. The fluorine-doped tin oxide (FTO) substrates acted as the working electrode (WE), while Pt wire and saturated AgCl/Ag electrode were used as the counter electrode (CE) and the reference electrode (RE), respectively. Optimal parameters of −0.1 VAgCl/Ag applied potential and −0.26 C cm−2 deposition charge yielded uniform red-brown BiOI films on the FTO substrate. Next, 0.2 mL cm−2 of 0.2 M VO(acac)2 in DMSO solution was drop-casted onto the BiOI electrode, followed by the thermal treatment at 450 °C for 2 h (2 °C min−1 ramp rate) to form BiVO4. Post-calcination samples were etched in 1.0 M NaOH to remove surface V2O5 residues. To further improve the bulk charge-separation efficiency, the BiVO4 photoanodes were photo-treated in 1.0 M potassium borate (KBi) buffer solution (pH = 9.3) containing 0.2 M Na2SO3 under the AM 1.5 G illumination for 13 min51. The final BiVO4 photoanodes were rinsed with DI water and dried with N2 flow.

The Cu-modified BiVO4 photoanodes were fabricated via an optimized electrodeposition approach22,52. The deposition solution consisted of NaClO4 as the supporting electrolyte and 5 mM CuSO4 as the Cu source. Electrodeposition was conducted at −1.0 VAgCl/Ag for 20 s using a bare BiVO4 as working electrode, with Pt wire counter electrode and AgCl/Ag reference electrode.

PEC measurements

PEC experiments were carried out in a three-electrode undivided cell controlled by an electrochemical workstation (CHI760E for photoelectrolysis experiments, Autolab PGSTAT302N for PEC characterizations). This system employed the synthesized photoanode (effective area = 2 cm2), Pt wire, and saturated AgCl/Ag electrode as the WE, the CE, and the RE, respectively. iR-compensation was not performed in all PEC tests. Data without error bars were derived from single measurements, and error bars represented the standard deviations across three parallel experiments. The electrolyte consisted of 0.3 M tetraethylammonium tetrafluoroborate (TEABF4) as the supporting electrolyte, various amines/imines as substrates, and Cu(NO3)2 as the Cu source in a mixed solvent. The solvent optimization details were discussed in Supplementary Fig. 4. A 300 W Xenon lamp (CEAULIGHT, CEL-HXF300-T3) equipped with an AM 1.5 G filter provided simulated sunlight, utilizing back illumination for BiVO4. Linear sweep voltammetry (LSV) curves were conducted at a scan rate of 50 mV s−1 under AM 1.5 G irradiation. In a typical photoelectrolysis, the electrolyte was firstly purged with O2 flow for 5 min. Chronoamperometric (I-t) measurements were conducted at a constant applied potential such as 0.6 VAgCl/Ag for 30 min. All reactions were conducted at room temperature and atmospheric pressure.

Reaction products were analyzed using high-performance liquid chromatography (HPLC, Shimadzu, LC-20ADXR) and gas chromatography-mass spectrometry (GC-MS, Agilent, 7890B-5977A). The HPLC instrument was equipped with a 5 μm C18 column (250 × 4.6 mm) and a diode array detector (DAD). Detection wavelength was set as 275 nm, and mobile phase was optimized as 10% aqueous trifluoroacetic acid (1.0%): 90% CH3CN. For GC-MS analysis, a HP-5MS capillary column (30 m× 0.25 mm) was employed with the following temperature program: injector at 300 °C, initial column temperature 80 °C (2 min), ramped to 150 °C at 10 °C min−1, then to 300 °C at 20 °C min−1, followed by a 10-min isothermal hold.

A 470-nm LED (Metrohm, LDC470) was employed for PEC characterizations. Electrochemical impedance spectroscopy (EIS) was performed at varying applied potentials combined with a sinusoidal voltage of 10 mV amplitude. The frequency of AC voltage ranged from 10 kHz to 0.1 Hz. Intensity modulated photocurrent spectroscopy (IMPS) utilized a DC light intensity of 10 mW cm−2 with a 10% modulation, where the frequency was varied from 10 kHz to 0.1 Hz. Mott-Schottky analysis was conducted under both dark and illuminated conditions via acquiring EIS data over a series of DC potentials. For each DC potential, the imaginary impedance components at 500, 1000 and 1200 Hz were extracted to calculate the reciprocal of the square capacitance, thereby obtaining Mott-Schottky plots.

IR measurements

Attenuated total reflectance Fourier-transform infrared (ATR-FTIR) measurements were performed on the same home-made PEC cell8,53. The spectra were recorded by a FT-IR spectrometer (Thermo Fisher, Nicolet iS50) equipped with a liquid nitrogen-cooled MCT detector, while the polarization conditions of WE were controlled by an electrochemical workstation (CHI760E). A 45° angle ZnSe crystal (80 mm × 10 mm × 4 mm) was used as the external reflection element. To study the coordination of Cu ions with BI substracts, BI acetonitrile solutions containing different concentrations of Cu ions were spread over the ZnSe crystal to acquire the signals of Cu-BI complexes. For the Operando ATR-FTIR measurements, the BiVO4 photoanode was pressed onto the ZnSe prism with its conductive surface facing down. Ti wire was used as the working electrode connection. Each spectrum was acquired by averaging 64 scans at a resolution of 4 cm−1.

KIE measurements

For H/D kinetic isotope effect (KIE) experiments, the benzophenone imine (NH=CPh2, noted as BI) and H2O were replaced by deuterated BI (ND=CPh2) and D2O. The steady-state current was obtained through the chronoamperometry at each potential.

EPR measurements

Electron paramagnetic resonance (EPR) measurements were conducted by using 5,5-dimethyl-1-pyrroline N-oxide (DMPO) as the spin-trap reagent, adopting a sampling method reported by Wand and co-workers54. Specifically, during the photoelectrolysis, 50 μL DMPO was added onto the photoanodes through a pipette. Close interfacial contact between the pipette tip and photoanode was maintained to facilitate instantaneous trapping of transient radical species. If radicals were generated during the photoelectrolysis, they would in-situ react with the DMPO to form corresponding adducts. Subsequently, the final solution was collected in a capillary tube for the EPR measurements. The optical image of the cell used in EPR measurements is shown in Supplementary Fig. 16.

Computational details

The spin-polarized DFT calculations were carried out using the Perdew-Burke-Ernzerhof (PBE) functional and DZVP basis sets in the CP2K package55,56, and the dispersion correction was applied in all calculations with the Grimme D3 method57. Only Γ point was used in all calculations here due to the large size of the supercell. An orbital transformation (OT) procedure was used for the wave function optimization. The cutoff of 500 Ry and GTH pseudopotential were used for static calculations or ab-initio molecular dynamics (AIMD) simulations. The energy convergence for the self-consistent field calculation was set to 10−6 Hartree (10−5 Hartree for AIMD simulations), and the largest forces on atoms were set to less than 4.5 × 10−4 Hartree/Bohr. The (110) facet of monoclinic scheelite BiVO4 was constructed for the catalytic calculations due to its selective hole-accumulation nature based on the optimized bulk structure cell58. The BiVO4 (110) slab model possesses a 4 × 2 × 1 supercell structure (20.6454 × 25.4525 × 43.0662), containing 512 atoms, where 3-atomic layers in the middle were fixed in all calculations. The solvation effect of the electrolyte was considered with an explicit solvation model by adding 120 CH3CN molecules to the BiVO4 (110) slab model. The N-coordinated Cu complex with four benzophenone imine (BI) molecules has been introduced to the semiconductor/electrolyte interface, whose structure is relaxed to the equilibrium state via the AIMD simulation over 5 ps in a canonical ensemble (NVT) with a Nose-Hoover thermostat at 300 K. For static calculations, i.e., the bond dissociation energies of N−H bonds, non-covalent interaction (NCI) analysis and reaction path, a slightly smaller-size BiVO4 (110) slab model (20.6454 × 25.4525 × 33.6051) with Cu-BI4 and 50 CH3CN molecules was employed after additional 5 ps equilibrium of AIMD simulation (NVT, 300 K). Three to five random snapshots from the last 2 ps of the AIMD simulations were optimized to produce the structural input with the lowest total energy for static calculations. The bond dissociation energies were determined by using the computational hydrogen electrode (CHE) method59. The transition state (TS) was located with the climbing image nudged elastic band (CI-NEB) method integrated into CP2K. For the frequency calculations, we considered five atoms (H, N, and C within the free BI molecules, N and Cu atoms within the complex) related to the reaction intermediates. Then the free energy correction was conducted using Shermo2.4 software developed by Lu et al.60. The NCI analysis based on promolecular density was carried out via Multiwfn software. The computational data generated in this study have been deposited in the Supplementary data 1.

Scale-up of PEC N−N couplings

The scale-up PEC reaction was implemented in a reported PEC flow reactor employing a two-electrode configuration49. A BiVO4 photoanode paired with a Pt-coated Ti cathode (Pt/Ti) constituted the electrode assembly. The 500 mL electrolyte solution consisted of 0.3 M TEABF4, 0.2 M BI, and 10 ppm Cu ions in a mixture solvent comprising 5% H2O, 20% CCl4, and 75% MeCN. Two white-light LEDs were utilized the light source, while a commercial solar panel (0.76 V output, Supplementary Fig. 35) provided supplementary power. A gear pump (Leadfluid, CT3001F) maintained continuous electrolyte flow in the reaction chamber at a rate of 500 mL min−1, facilitating efficient flow-type photoelectrolysis. Following the reaction, the mixture underwent dichloromethane extraction followed by solvent removal via rotary evaporation. The resultant crude product was purified through sequential washing with petroleum ether, yielding a pale-yellow solid. The structure of the final product was confirmed through 1H and 13C nuclear magnetic resonance (NMR) spectroscopy. The hydrolysis of BA for the preparation of hydrazine (N2H4) was discussed in Supplementary Fig. 36.