Metal single-site catalyst design for electrocatalytic production of hydrogen peroxide at industrial-relevant currents

Direct hydrogen peroxide (H2O2) electrosynthesis via the two-electron oxygen reduction reaction is a sustainable alternative to the traditional energy-intensive anthraquinone technology. However, high-performance and scalable electrocatalysts with industrial-relevant production rates remain to be challenging, partially due to insufficient atomic level understanding in catalyst design. Here we utilize theoretical approaches to identify transition-metal single-site catalysts for two-electron oxygen reduction using the *OOH binding energy as a descriptor. The theoretical predictions are then used as guidance to synthesize the desired cobalt single-site catalyst with a O-modified Co-(pyrrolic N)4 configuration that can achieve industrial-relevant current densities up to 300 mA cm−2 with 96–100% Faradaic efficiencies for H2O2 production at a record rate of 11,527 mmol h−1 gcat−1. Here, we show the feasibility and versatility of metal single-site catalyst design using various commercial carbon and cobalt phthalocyanine as starting materials and the high applicability for H2O2 electrosynthesis in acidic, neutral and alkaline electrolytes.


Theoretical calculations
DFT calculations are conducted to screen for highly active and selective two-electron ORR electrocatalyst among TMPc including MnPc, FePc, CoPc, NiPc, CuPc and ZnPc ( Fig. 1a and Supplementary Figs. 1, 2). Figure 1b shows the volcano plot describing the two-electron ORR activity as a function of the Gibbs free energy for *OOH adsorption (ΔG *OOH ). MnPc and FePc, situated on the left side of volcano plot, show strong *OOH binding to favorably dissociate the O-O bond adsorbed on active metal sites, resulting in selective H 2 O formation over H 2 O 2 14,15 . NiPc, CuPc and ZnPc, located at the right side, present weak *OOH binding and thus high selectivity to H 2 O 2 formation; however, the ORR activities are proposed to be inferior due to the ratelimiting protonation of O 2 toward *OOH. In contrast, CoPc shows a ΔG *OOH value of 4.15 eV that is closest to the activity-volcano peak (~4.22 eV) 14 , demonstrating the highest activity and selectivity toward the two-electron ORR among all TMPc assessed here. Free energy diagrams for the two-electron ORR in Fig. 1c display the largely uphill barrier, up to 0.66-0.80 eV, to form the *OOH intermediate on NiPc, CuPc and ZnPc at 0.7 V, confirming the poor ORR activities. In contrast, the energy barrier of 0.40-0.48 eV for *OOH dissociation on both MnPc and FePc, originated from the excessively downhill step to form the *OOH, would drastically reduce the selectivity to H 2 O 2 . In line with the activity-volcano relationship, CoPc shows the lowest energy barrier of 0.07 eV for the entire reaction process of oxygen reduction to H 2 O 2 formation, although there is still a small energy gap to the ideal reaction process where the free energy change is flat with zero energy barrier 11,15 . It is worth noting that the ΔG *OOH values of MnPc, FePc, CoPc, NiPc and CuPc show the correlation with the d-band center (Supplementary Figs. 3,4), where the lower d-band center of metal atoms contributes to more positive ΔG *OOH and thus weaker binding ability 22,23 . Furthermore, the adjustment of adjacent atomic environment of the metal sites can also effectively optimize the *OOH binding by modifying the electronic structures [17][18][19] . Compared to CoPc, the too strong *OOH binding away from the top of the volcano plot on the typical CoN 4 structure suggests the possibility of tuning ORR selectivity by coordinating the Co center. The O-modified graphitic carbon is employed to anchor the CoPc molecule for modifying the local configuration of the Co center. It's worth noting that O modified graphitic carbon have poor ORR activity due to weak *OOH binding, while the Co site of CoPc has good *OOH binding and is highly ORR active that can be ascribed to the high electrophilicity to *OOH (Supplementary Figs. [5][6][7]. Interestingly, it is found that different types of O atoms on the carbon plane ( Supplementary Fig. 8) result in two opposite effects that strengthen or weaken the *OOH binding on CoPc.
The O dopant at the defective carbon site helps the Co site (denoted as CoPc-OCNT) achieve a slightly increased ΔG *OOH that is closest to the activity-volcano peak (Fig. 1b). The corresponding free energy diagram exhibits a negligible uphill energy barrier (Fig. 1c), suggesting excellent two-electron ORR activity of CoPc-OCNT. However, axial O coordination at the carbon basal plane to the Co center (denoted as CoPc/ OCNT) causes a stronger *OOH binding with a reduced ΔG *OOH of 3.77 eV and increases the energy barrier (0.45 eV) for H 2 O 2 formation, adversely reducing the two-electron selectivity. The effect of axial O coordination is also confirmed on FePc or CuPc, which causes stronger *OOH binding and thus lower H 2 O 2 selectivity on Fe site while moderately improved ORR activity on Cu site ( Supplementary Fig. 9). We further examine the H 2 O 2 selectivity of catalysts by inspecting the entire ORR process containing three primitive steps of *+O 2 →*OOH→*H 2 O 2 →*+H 2 O 2 as well as the O-O bond dissociation of the adsorbed *OOH or *H 2 O 2 ( Supplementary Fig. 10). The CoPc-OCNT catalyst shows a small energy barrier during the entire two-electron ORR process, indicating a low overpotential for producing H 2  close to the peak of the volcano ( Supplementary Fig. 12). Thus, we infer that Co sites of these Co SSC coordinated with pyrrolic N are intrinsically active to selectively catalyze two-electron ORR.
To uncover the origin for ORR selectivity tuning on Co SSC, the electronic properties of the Co atom in CoPc/OCNT and CoPc-OCNT are analyzed by studying the density of states and Bader charge. Projected density of states (PDOS) of CoPc/OCNT and CoPc-OCNT in Supplementary Fig. 13 reveal that the different degree of orbital hybridization before and after *OOH adsorption determine the properties of *OOH adsorption. We calculate the projected crystal orbital Hamilton populations (pCOHP) between the Co atom and the O atom of *OOH to explain the bond strength (Fig. 1d). The pCOHP of CoPc/ OCNT shows that all antibonding σ* orbitals are located at unoccupied states above the Fermi level, which contributes to reinforced *OOH adsorption. In contrast, pCOHP of CoPc-OCNT reveals that antibonding electrons of the σ* orbital with highly occupied states below the Fermi level reduce the strength of the Co-O bond, which well accounts for weaker *OOH adsorption. Moreover, the absence of the backdonation effect due to antibonding π* with unoccupied state above the Fermi level is also responsible for the weak  16,17), which is responsible for the weaker *OOH binding.

Metal single-site catalyst synthesis and characterization
Guided by the theoretical predictions, TMPc-OCNT (TM=Fe, Co or Cu) catalysts are synthesized by combining TMPc with the OCNT support through a sufficient mixing process (see Methods). This mild synthetic method at room temperature is adopted to avoid generating the strong bonding between the TM site in TMPc and the O atom in OCNT and to prevent the decomposition of TMPc molecules in typical hightemperature processing. The OCNT material is prepared by oxidatively treating the original CNT (ori-CNT) to create oxygen-containing functional groups. As illustrated in Supplementary Fig. 18, no visible morphology changes are observed on OCNT after oxidation. X-ray photoelectron spectroscopy (XPS) measurements show that OCNT has a significantly increased O content (9.1%) compared to ori-CNT (2.5%), and these introduced oxygen functional groups exist mainly in the form of C-O-C (O at defective site) as well as minor portion of C=O (O on carbon plane) 18,26 , as also revealed by Fourier transform infrared (FT-IR) spectra ( Supplementary Fig. 19). Raman spectra reveal an obvious increasement of carbon defects in OCNT due to more O dopant. The analysis of N 2 adsorption-desorption isotherms shows a two-fold increase in the specific surface area of OCNT compared to ori-CNT ( Supplementary Fig. 20), which is favorable to provide more opportunities to anchor TMPc molecules. The atomic contents of Fe, Co and Cu are determined by XPS to be 0.57, 0.30 and 0.79 at% on the surface of FePc-OCNT, CoPc-OCNT and CuPc-OCNT, respectively, ( Supplementary Fig. 21). The X-ray diffraction (XRD) patterns show that all TMPc-OCNT catalysts contain the characteristic diffraction peaks that are well-matched with reference TMPc molecules (Supplementary Fig. 22), suggesting that the molecular structures of TMPc are retained after being anchored on OCNT. The XRD patterns for CoPc-OCNT (Fig. 2a) show that the characteristic peaks located at 7.0°and 9.2°are markedly increased in intensity when the CoPc content is increased from 0.23 to 0.45 at% (Supplementary Tables 2, 3)     results are also observed on CuPc-OCNT and FePc-OCNT (Supplementary Fig. 23), indicating the successful CuPc and FePc loading. Furthermore, we examine the chemical interactions between Co of CoPc and O of the OCNT by high-resolution XPS spectra, which reveals a slight peak shift in the Co peak of CoPc-OCNT to higher binding energy comparing to CoPc, which is companied by a reverse shift of the O peak ( Fig. 2c). Similar shifts of spectral peaks are also observed on FePc-OCNT and CuPc-OCNT but not on Pc-OCNT (Supplementary Fig. 24), indicating more positive valence for the metal sites, likely due to charge transfer from the metal centers to surrounding coordinated atoms [29][30][31][32] . Additionally, the O 1s high-resolution spectrum reveals that the C-O-C group is dominant in CoPc-OCNT compared to C=O. To identify the structural configuration of metal sites in CoPc-OCNT, X-ray absorption near edge spectroscopy (XANES) and extended X-ray absorption fine structure (EXAFS) of the Co K-edge are analyzed. As illustrated in Fig. 2d, the near-edge absorption intensity of the Co K-edge for CoPc-OCNT is between that of Co foil and Co 2 O 3 , suggesting a positively charged Co atom 32,33 . The detection of the pre-edge peak at 7715 eV (a signature of the square-planar Co-N 4 structure in CoPc) indicates that the configuration of Co-N 4 remains intact on OCNT [33][34][35] . Referring to the Fourier transformed EXAFS (FT-EXAFS) for Co foil ( Fig. 2e and Supplementary Fig. 25), the Co-Co scattering path (2.2 Å) is absent in CoPc-OCNT, indicating that Co exists as isolated single sites. In comparison with CoPc, the EXAFS fitting for CoPc-OCNT shows a main peak at 1.5 Å and the minor peaks at 2.5 and 3.0 Å that can be assigned to the first scattering path of Co-N and the second paths of Co-C-N and Co-N-C, respectively (Fig. 2f). The quantitative fitting of the EXAFS spectra reveals the first coordination shell of four Co-N bonds at the distance of 1.9 Å and the second shell of Co-N-C and Co-C-N at the distances of 3.3 and 3.0 Å (Supplementary Fig. 26 and Supplementary Table 5). Overall, the results above reveal the CoPc-like Co-(pyrrolic N) 4 structure with O atom modification in CoPc-OCNT.

Electrocatalytic two-electron ORR performance
The two-electron ORR activity and selectivity of TMPc-OCNT catalysts are investigated by rotating ring-disk electrode (RRDE) using a standard three-electrode system in 0.01 M KOH (pH 11.9), 0.  the simulation of kinetics involving water layers using advanced dynamic models 24,25 .
In situ attenuated total reflectance surface-enhanced infrared absorption spectroscopy (ATR-SEIRAS) testes are performed to detect the key adsorbed OOH* on CoPc-OCNT during the electrolytic H 2 O 2 synthesis ( Supplementary Fig. 27). Figure 4 shows that a weak absorption band at about 1264 cm −1 appears when applying a potential of 0.8 V vs RHE, and this band is gradually enhanced by decreasing the potential. These absorption band on CoPc-OCNT can be assigned to O-O stretching vibration of *OOH, which are slightly shifted to higher wavenumber compared with the values in previous studies likely due to different adsorption sites 17,37,38 . Additionally, the bands at 835 cm −1 that increase with negatively shifted potential can be reasonably assigned to the M-O stretching mode of *OOH [39][40][41] . Moreover, the bands assigned to adsorbed hydroperoxide (typically at 1386 cm −1 ) are not detected 37 , because H 2 O − product rather than H 2 O 2 is produced at pH>11.6. Overall, the detection of the potential-dependent adsorbed hydroperoxy bands supports the *OOH mediated two-electron ORR pathway on the CoPc-OCNT catalyst.
To evaluate the ability for electrocatalytically producing H 2 O 2 , the CoPc-OCNT catalyst is deposited on hydrophobic gas-diffusion electrode for enhancing O 2 supply. We employ a two-electrode system assembled in a customized flow-cell reactor, as illustrated in Supplementary Fig. 28. As shown in Fig. 5a Supplementary Fig. 29), only 0.17% Co is loaded on the CNT-H support using the similar preparation method, which is lower than that on CoPc-OCNT (0.30%) and CoPc-oriCNT (0.48%, prepared using original CNT). Consequently, the CoPc-CNT-H exhibits poor H 2 O 2 electrosynthesis performance (65% FE at 100-170 mA cm −2 ) than CoPc-OCNT and CoPc-oriCNT (96-98% and 67-83% FE at 100-200 mA cm −2 ) ( Supplementary Fig. 30). Although a higher Co loading, CoPc-oriCNT deliveries more inferior Faradaic efficiency for producing H 2 O 2 especially at larger currents compared to the CoPc-OCNT. It is inferred that there are moderately agglomerated CoPc molecules in the CoPc-oriCNT due to less strong interaction between Co sites and insufficient O-containing groups on oriCNT ( Supplementary Fig. 29c), which demonstrates the crucial contribution of sufficient O modification to catalytic Co centers to the two-electron ORR selectivity. Figure 5c shows that when increasing the current on the CoPc-OCNT cathode, the concentration of directly outflowed H 2 O 2 almost linearly increases from 366 to 1084 mM (3.7 wt%) that can meet the concentration requirement for disinfection or wastewater treatment applications 42 (3 wt% is generally sufficient). The CoPc-OCNT cathode can delivery ORR current density up to 300 mA cm −2 with remarkable FE of 96-100% that is higher than that achieved by majority of electrocatalysts currently reported and is comparable to the best values reported by Wang et al. 36,43 (Supplementary Table 7 and Supplementary Fig. 31). To better understand the intrinsic activity, the turnover frequency (TOF) per metal site for H 2 O 2 production is calculated. The CoPc-OCNT catalyst presents current-dependent TOF values that reach up to 649-1921 min −1 at 100-300 mA cm −2 , demonstrating superior activity of the Co sites to Cu or Fe sites (Supplementary Figs. 32,33). Significantly, the heterogeneous H 2 O 2 electrosynthesis method shows great potential in maximizing the intrinsic activity of catalysts compared to homogeneous ORR that relies on Co-based molecular catalysts in organic system (limited TOF even at large overpotentials) 44 . The results from the current study show continuous H 2 O 2 production in high-concentration, while which remains a great challenge for homogeneous molecular catalysis process that also suffers from H 2 O 2 separation from the catalyst. Additionally, the CoPc-OCNT delivers a remarkable H 2 O 2 production rate of 3892-11,527 mmol h −1 g cat −1 under the current densities of 100-300 mA cm −2 ( Supplementary Fig. 34), higher than the values reported to the best of our knowledge ( Fig. 5d and Supplementary Table 8). As shown in Fig. 5e, the continuous 30 h electrosynthesis of H 2 O 2 at 200 mA cm −2 stably produces H 2 O 2 at a concentration of 677-755 mM at a remarkable FE of 91-100%, demonstrating the electrochemical stability of the CoPc-OCNT electrode. In contrast, the activity and stability of CuPc-OCNT declines rapidly with increasing current, which might be ascribed to less robust bonding of CuPc on OCNT ( Supplementary Fig. 32).
We investigate the H 2 O 2 electrosynthesis performance of CoPc-OCNT catalyst in less alkaline electrolyte (0.01 M KOH), which shows 97% H 2 O 2 FE at 10 mA cm −2 , but the FE drops to 27% at 20 mA cm −2 due to large overpotential ( Supplementary Fig. 35a). The cell voltage can be significantly reduced by increasing electrolyte concentration or shortening the electrode distance to reduce solution resistance (Supplementary Figs. 35,36). Results show that the cell voltage can be markedly reduced even in 0.01 M KOH, and the current density for H 2 O 2 production is obviously lifted (Supplementary Fig. 35). When using 1 M KOH+0.5 M H 2 SO 4 or 1 M KOH as catholyte and anolyte, the H 2 O 2 FE are 83% and 75% at 1100 mA, and the partial H 2 O 2 currents can reach 913 and 823 mA, respectively (Fig. 6a). The electrosynthesis process consumes an electric energy of 0.20-0.34 kWh for producing per kg 3 wt% H 2 O 2 in the small reactor ( Supplementary Fig. 37, calculations are described in Method) that is reduced to 0.12-0.28 kWh by the upgrading of the electrolyzer configuration, showing highefficiency electrical energy conversion into valuable chemicals. Furthermore, the feasibility studies of scalable catalytic material preparation are explored, based on the simplicity of the CoPc-OCNT catalyst synthesis compared to the catalysts typically prepared by high-temperature carbonization. Five types of highly conductive, commercially available carbon materials are adopted as the supports for CoPc after introducing O-containing groups via the similar oxidization treatment to that of OCNT ( Supplementary Figs. 38, 39). CoPc-OVX (Cabot VXC72) and CoPc-OAB (acetylene black) catalysts show 2000 1800 1600 1400 1200 1000 800  higher FE of 85-98% for H 2 O 2 production at 100-170 mA cm −2 compared to CoPc-OBP (black pearl 2000), CoPc-OECP (Ketjenblack ECP-600JD) and CoPc-OYP (Kurary YP-80F) (Supplementary Fig. 40). CoPc-OVX and CoPc-OAB can produce about 330 and 550 mM H 2 O 2 at 100 and 150 mA cm −2 that are higher than that on CoPc-OBP, CoPc-OECP and CoPc-OYP. The performance of H 2 O 2 electrosynthesis is closely related to the Co loading, and the best activity is achieved at a moderate loading while insufficient or excessive Co loading would degrade the activity and selectivity due to limited catalytic sites or CoPc aggregation. It's worth noting that 0.2-0.6 at% CoPc loadings on OCNT or commercial carbon materials approach the theoretical upper limit value of 0.6 at% that is estimated based on the densest monodisperse packing of CoPc on OCNT (described in Supplementary Fig. 41), where the space of adjacent Co sites is far enough to prevent the CoPc from aggregating and thus favor the two-electron ORR over the fourelectron process. The results not only prove the wide versatility of the design of metal single-site catalysts, but also demonstrate the feasibility for scalable catalyst manufacture that is not limited by specific starting carbon materials. Furthermore, it is promising to prepare the high-density O-modified Co-(pyrrolic N) 4 sites to improve the activity for H 2 O 2 production by using the CoPc and commercial carbon materials as raw materials in further studies 24,25 .
The electrosynthesis of H 2 O 2 is also explored in different pH for satisfying specific on-site applications related with environmental remediation that requires for acidic H 2 O 2 , such as Fenton reaction only effectively works pH~3. When using 0.1 M K 2 SO 4 catholyte and 0.5 M    47 . The drastic increase in local pH on electrode-electrolyte interfaces due to the depletion of H + , especially when applying a current density over 100 mA cm −2 , should also contribute to the enhanced H 2 O 2 current density 48 . The experimentally observed pH increasements of effluent catholyte during electrosynthesis of H 2 O 2 also confirm an increase in the pH (Fig. 6b). We highlight that such electrocatalytic reaction interfaces including metal cations and pH gradients are difficult to be accurately modelled by the DFT approach in this study based on purely thermodynamic analysis. Advanced computational methods, such as ab initio molecular dynamics simulations, are the potential tools to better describe and simulate the electrocatalytic properties of catalytic materials under real experimental conditions by considering the effects of solvent and ions and the kinetic information after a preliminary screening on the twoelectron ORR catalysts 24,25,47,49 . We built a flow-cell system combining electrosynthesis of H 2 O 2 and the Fenton reaction for wastewater treatment, which shows almost 100% removal of the biodegradable pollutants with a treatment capacity of 700 mL h −1 , wherein 40-52% total organic carbons (TOC) can be converted into CO 2 (Fig. 6c). This flow-cell Fenton system also shows good performance for real coking wastewater treatment with almost 90% TOC removal. The 12 h acidic H 2 O 2 electrosynthesis at 500 mA for refractory organic pollutant degradation and wastewater treatment by cooperating with the Fenton reaction demonstrates the potential in actual small-scale applications.

Discussion
In summary, the ORR volcano plot with ΔG *OOH as a key descriptor is established by DFT calculations to screen for highly selective twoelectron electrocatalyst among TMPc catalysts. The ΔG *OOH values of TMPc are associated with the d-band center of TM, which accounts for the ΔG *OOH on CoPc being much closer to the peak of the volcano. The O dopant at the defective carbon site is demonstrated to modify the local electronic structure of the Co center in CoPc and helps achieve the ΔG *OOH closest to the peak of the volcano. The prepared CoPc-OCNT catalyst with the O-modified Co-(pyrrolic N) 4  densities up to 300 mA cm −2 with 96-100% FE are achieved for continuous H 2 O 2 production at a record rate of 11,527 mmol h −1 g cat −1 in a flow-cell electrolyzer. The high H 2 O 2 production capacity at industrialrelevant currents is meaningful to help reduce infrastructure cost for promising scalable applications of H 2 O 2 electrosynthesis. Furthermore, this study demonstrates the versatility of the metal single-site catalyst design using various commercial carbons as starting materials and high applicability for H 2 O 2 electrosynthesis, presenting a promising potential in future large-scale H 2 O 2 manufacture and in-situ wastewater treatment and disinfection.

Computational details
Density functional theory calculations were conducted using the Vienna Ab-initio Simulation Package (VASP 5.4.4 version) 50 . The electron-ion interactions and electron exchange-correlation interactions are described by the projector augmented wave (PAW) potential and Perdew-Burke-Ernzerhof (PBE) functional with the generalized gradient approximation (GGA) method 51,52 . The structural models of TMPc, OCNT and TMPc-OCNT were constructed in a cell with lattice parameters of 17 Å × 17 Å × 20 Å (Supplementary Figs. 1, 2 and 5). The kpoint meshes of (1 × 1 × 1) was set. The spin polarization effect was considered for all calculations expect for OCNT. The cutoff energy of plane wave was set at 420 eV. All structural configurations were optimized until the forces on each atom were converged to less than 1 × 10 −2 eV Å −1 . Total free energy changes between two steps were less than 1 × 10 −5 eV atom −1 in electronic relaxation. The two-electron ORR pathway is described as the following steps: * + O 2 ðgÞ + H + + e À ! *OOH ð1Þ where * and *OOH denote an unoccupied active site and adsorbed *OOH intermediate, respectively. Gibbs free energy (ΔG) for each step at the given potential U was calculated by the following equation: 14,53 where ΔG represents the free energy that is equal to the calculated ΔE after being corrected by zero-point energies (ΔZPE) and entropic contributions (ΔU 0→T -T×ΔS) of adsorbates at 298.15 K. ΔE is the electronic energy difference between reactants and products for each step. These correction items can be conveniently calculated using a VASPKIT code 54  During the two-electron ORR process, the limiting step is determined by both *OOH formation (Eq. (1)) and *OOH removal (Eq. (2)) from the catalytic sites. The theoretical overpotential is demonstrated to be a function of the *OOH binding energy, thus the limiting potential can be expressed as: 11 U L1 = À ΔG *OOH + 4:92 ð4Þ

Catalyst preparation
The OCNT material was prepared by chemically creating oxygen functional groups on the surface of original CNT, which was fulfilled by the oxidization treatment using concentrated nitric acid. In a typical operation, 3 g CNT power was homogeneously dispersed in 68 wt% HNO 3 (150 mL) under continuous magnetic stirring. The oxidization treatment was conducted at 140°C for 12 h, followed by water washing and drying. It is important to note that the evaporation of concentrated HNO 3

Catalyst characterization
The morphology of catalysts was observed by SEM (Hitachi S-4800) and TEM (Thermo-Tecnai G2 F30 S-Twin). The oxygen functional groups on OCNT and chemical constitutes of TMPc-OCNT were identified by FT-IR (RRUKER, VERTEX 70) and XPS (Thermos K-Alpha+ instrument with Al Kα X-ray excitation source). The phase structure was measured by XRD (Smartlab 9 kW, Nippon Neoku Electric Co. Ltd., Japan) and Raman (laser confocal microscopy Raman with laser excitation at 532 nm). The surface area and pore characteristics were assessed using N 2 adsorption-desorption measurements (Quantachrome, Autosorb-IQ-C The electrodes were prepared by loading as-synthesized TMPc-OCNT catalysts on GDE. In a typical process, 10 mg of catalyst was dispersed in the mixture that consists of 0.8 mL H 2 O, 0.2 mL isopropanol and 50 μL of Nafion (5 wt%) to form catalyst ink. 50 μL of catalyst ink was coated on GDE by hand-painting, followed by drying treatment at 40 o C. The catalyst loading is about 0.5 mg cm −2 . Pt foil was used as the counter electrode to construct a two-electrode system for electrochemical H 2 O 2 production. The proton exchange membrane (Nafion 117) was used to separate the cathode and anode. The H 2 O 2 electrosynthesis was carried out at the constant currents using two types of flow-cell electrolyzers shown in Supplementary Figs. 28 and 36. When employing the small reactor with a working area of 1 cm −2 , KOH electrolyte (0.01, 0.1, 0.2, 0.5 or 1.0 M) was used as the catholyte with onepass flow rate of 5 mL h −1 to bring out the generated H 2 O 2 , and the anolyte of 0.5 M H 2 SO 4 was cycled at a rate of 33 mL h −1 . Pure O 2 was supplied to the cathode side faced to the gas chamber at a flow rate of 20 mL min −1 that was controlled by a mass flow controller (Sevenstar D07, China). The constant currents were provided by a workstation (Chenhua CHI760E, Shanghai, China). When employing the bigger reactor with a working area of 4 cm −2 , the catholyte was one-time outflowed at a rate of 80-100 mL h −1 , while the anolyte was cycled at the same flow rate.
It should be noted that the consumed volume (V ðCe 4 + Þ, mL) of Ce 4+ solution (the initial concentration C 0 ðCe 4 + Þ is 1 mM) for detecting H 2 O 2 was at the range of 5-40 mL, which was far larger than V ðH 2 O 2 Þ of the H 2 O 2 solution (10 μL). Otherwise, this simplified equation above needs to be corrected.

Electricity consumption calculation
The electricity consumption for per kg H 2 O 2 (3 wt%) production is calculated according to the equation of E represents electricity consumption (kWh kg −1 3 wt% H 2 O 2 ), I and U represent applied current and cell potential (V, A), respectively, and C, M(H 2 O 2 ) and v represent H 2 O 2 concentration (mol L −1 ), H 2 O 2 molecular weight (34 g mol −1 ) and flow rate, respectively.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.