Atomically precise Au₈Pd₁(DPPF)₄²⁺ catalyst orchestrates the synthesis and utilization of hydrogen peroxide

Abstract

Atomically precise metal clusters that merge structural clarity, functional modularity, and tunable spatial programmability have gained momentum over the past decade. Here, we report the catalysis of an atomically precise Au₈Pd₁(DPPF)₄²⁺ (DPPF = 1,1’-bis(diphenylphosphino)ferrocene) cluster in the synthesis and utilization of hydrogen peroxide. The cluster shows a concerto-like synergy of metal core and surface ligand in the entire catalytic process, in which the metal core catalyzes the reaction of formic acid and oxygen to produce H₂O₂, and subsequently DPPF ligands work with the core to drive the conversion of H₂O₂ into hydroxyl radicals. Notably, with the computational fluid dynamics simulation assistance, the heterogeneous catalyst from the assembly of Au₈Pd₁(DPPF)₄²⁺ clusters into mesoporous Al₂O₃ monolith enables continuous generation and utilization of H₂O₂ performed in the fixed-bed reactor. This work offers an alternative route for the efficient upgrading of H₂O₂ synthesis and application.

Atomically precise metal clusters that combine structural clarity, functional modularity, and spatial programmability are becoming increasingly relevant. Here, the authors report the synthesis and applications of the cluster-based Au₈Pd₁(DPPF)₄²⁺/Al₂O₃ catalyst for hydrogen peroxide generation.

Introduction

The best heterogeneous catalyst requires atomically uniform sites for predictive structure-performance relationship^1–3, multi-functional components to streamline the complex catalytic process^4–6, and optimized macrostructure to enhance mass transfer efficiency in the diffusion-limited step^7–10. Conventionally, concurrent satisfaction of these requirements presents a paradox due to stochastic metal nucleation within traditional impregnation methods, forcing the trade-offs among above three goals^11–13. Atomically precise metal clusters protected by organic ligands, integrating structural clarity, functional modularity, and tunable spatial programmability, offer a solution to this paradox^14,15. Single crystal X-ray crystallography reveals their atomic-packing total structures (metal core plus ligand shell), in which their discrete metal cores with uniform active sites avoid the heterogeneity of traditional nanoparticles^16–18, alongside a ligand shell can serve as secondary sites^19–21. Furthermore, the self-assembly behavior of clusters can be precisely regulated through coordination chemistry, enabling programmable spatial organization from the isolated clusters to extended superstructures^22–24. These features render them exceptionally distinctive catalysts, completely different from typical heterogeneous catalysts.

Hydrogen peroxide (H₂O₂), as a versatile green oxidant in energy and environmental applications, faces challenges in its traditional anthraquinone oxidation production method, including high energy consumption, safety hazards, and difficulties in product transportation and storage^25–27. Meanwhile, the direct synthesis method using hydrogen and oxygen is limited in the practical applications due to the explosion risk of the gas mixture²⁸. Formic acid (HCOOH), a promising liquid hydrogen carrier, offers an alternative route to on-demand distributed H₂O₂ synthesis, thereby effectively avoiding the safety hazard and mitigating the decomposition of H₂O₂ during storage. Recent advances have demonstrated the feasibility of formic acid oxidation for H₂O₂ synthesis^29,30. Li et al. reported that the pair of single-atom Pd achieved the efficient conversion of HCOOH/HCOOK oxidation to H₂O₂ (Fig. 1a)³¹. Chen et al. realized the efficient synthesis of H₂O₂ from HCOOH/HCOOK oxidation using AuPt alloy catalyst and then coupled a single-atom Fe catalyst to utilize the generated H₂O₂ for the oxidation of coumarin (Fig. 1b)³². Both above systems were confined to batch operations. Integrating generation with utilization of H₂O₂ in the continuous-flow operations has emerged as an important challenge, and the design of catalysts needs to be systematically addressed. Specifically, the precise regulation of the active structures of catalysts is required to ensure highly efficient synthesis of H₂O₂^33–35, and meanwhile the utilization of H₂O₂ on the catalysts needs to be executed to avoid the diffusion loss^36,37. These catalysts might be ideally prepared into a monolith to resist continuous liquid flushing, and the optimized alignment of catalytic site location with the reactant distribution is also necessary to improve the overall reaction efficiency^38–40.

Fig. 1. Conceptual comparison of H₂O₂ synthesis and utilization systems from HCOOH/HCOOK oxidation.

Previous works: a single-atom Pd pair sites catalyzed H₂O₂ synthesis from HCOOH/HCOOK oxidation performed in batch reactors; b AuPt alloy nanoparticles generated H₂O₂ coupled with single-atom Fe catalysts for H₂O₂ utilization in batch reactors. This work: c working principle in this Au₈Pd₁(DPPF)₄²⁺ cluster for H₂O₂ production and utilization.

In this work, we report a distinctive performance of Au₈Pd₁(DPPF)₄²⁺ cluster on the H₂O₂ production and utilization, which is mainly due to the definite congruity between metal core and ligands. Figure 1c highlights the advancement of the catalyst system. The Au and Pd sites in the core enable the reaction of formic acid and oxygen to obtain H₂O₂, followed by the Fe and P sites in the ligand work with Au and Pd sites together to improve the conversion of H₂O₂ to hydroxyl radicals, in which the four sites in this cluster compose a whole melody to combine the two independent processes into one pot reaction. Based on this discovery, by controlling spatial distribution of clusters into the mesoporous alumina monolith, the continuous synthesis and application processes of H₂O₂ are finally accomplished in the fixed-bed reactor using prepared Au₈Pd₁(DPPF)₄²⁺/Al₂O₃ heterogeneous catalysts. This Au₈Pd₁(DPPF)₄²⁺ example demonstrates the power of atomically precise metal clusters for achieving remarkably catalytic performances on important chemical reactions.

Results

Structure and reactivity of Au₈Pd₁(DPPF)₄²⁺ cluster

The atom-packing structure of the Au₈Pd₁(DPPF)₄²⁺ cluster was shown in Fig. 2a, which featured a typical core-shell structure of metal core-organic ligand⁴¹. The octahedral like metal core comprised eight gold atoms surrounding a central palladium atom, with an average Pd-Au bond length of 2.62 Å, significantly shorter than that of bulk Pd-Au alloy, indicating strong metal-metal interactions within the cluster. The peripheral ligand shell consisted of four bidentate DPPF ligands, which was anchored by eight P-Au bonds, leaving the central Pd atom coordinatively unsaturated while positioning four redox-active ferrocene sites at a uniform 6.1 Å distance from the Pd center. We next evaluated the catalytic activity of Au₈Pd₁(DPPF)₄²⁺ cluster for the reaction of formic acid and oxygen in the batch reactor. Au₈(DPPF)₄²⁺ without Pd center⁴², Au₈Pd₁(PPh₃)₈²⁺ (PPh₃ = triphenylphosphine) without DPPF ligand (Supplementary Fig. 1)⁴¹, mononuclear complex Pd(DPPF)Cl₂, DPPF ligand and pure ferrocene (Fc) were selected as reference catalysts. Au₈Pd₁(DPPF)₄²⁺ exhibited much higher activity than Au₈(DPPF)₄²⁺, Au₈Pd₁(PPh₃)₈²⁺ and Pd(DPPF)Cl₂. In addition, pure DPPF ligand, pure Fc and the physical mixture of Au₈Pd₁(PPh₃)₈²⁺ and Fc had little activity for formic acid oxidation (Supplementary Fig. 2). Thus, the observed catalytic performance of the Au₈Pd₁(DPPF)₄²⁺ cluster was likely due to the synergy of Au, Pd and DPPF in the cluster.

Fig. 2. Fabrication and characterization of Au₈Pd₁(DPPF)₄²⁺ supported on Al₂O₃ catalysts.

a Crystal structure of Au₈Pd₁(DPPF)₄²⁺ cluster. b Illustration of the preparation process of supported catalysts via the electrostatic interaction. c Photographs showing the time-dependent process of cluster immobilization on the Al₂O₃ support and d corresponding UV–Vis absorption spectra of the solution. e TEM images and elemental mapping of the cluster layer.

The Au₈Pd₁(DPPF)₄²⁺ cluster exhibited structural stability under reaction conditions. Ultraviolet-visible (UV–Vis) absorption spectra of fresh and spent clusters showed no significant difference in the characteristic absorption bands of the two cases (Supplementary Fig. 3a). Electrospray ionization mass spectrometry (ESI-MS) data of fresh and spent clusters further confirmed that the formula of spent cluster was identical to that of fresh one with a dominant peak at m/z = 1949 (Supplementary Fig. 3b). Fourier transform infrared spectra (FT-IR) of the clusters before and after reaction showed that no obvious change was observed (Supplementary Fig. 3c), including the aromatic C-H stretching vibration at 3054 cm⁻¹, the bending vibration of the benzene ring at 1436 cm⁻¹, and the P-C bond stretching vibration at 1098 cm⁻¹⁴³. Meanwhile, no new band appeared in the region of phosphine ligand oxidation from 1100 to 1220 cm⁻¹⁴⁴, excluding the possibility of significant oxidation of the phosphine ligand.

Heterogeneous catalysts of Au₈Pd₁(DPPF)₄²⁺ supported on Al₂O₃ and catalytic performances

The heterogeneous catalyst of Au₈Pd₁(DPPF)₄²⁺ clusters supported on Al₂O₃ was next prepared via the electrostatic immobilization strategy (Fig. 2b). Utilizing the cationic nature of Au₈Pd₁(DPPF)₄²⁺, it generated electrostatic attraction with the hydroxyl groups on the γ-Al₂O₃ microsphere⁴⁵. During the immobilization, the CH₂Cl₂ solution of Au₈Pd₁(DPPF)₄²⁺ faded from deep red to colorless within 90 min (Fig. 2c), and UV-Vis spectra showed that the intensities of the characteristic bands decreased by >99%, confirming the efficient immobilization (Fig. 2d). Transmission electron microscope (TEM) images, corresponding element mapping, and X-ray diffraction (XRD) analysis verified that the clusters were highly dispersed on the alumina support (Fig. 2e and Supplementary Fig. 4). In addition, the loading of clusters did not alter the mesoporous framework of Al₂O₃ support (Supplementary Fig. 5).

Computational fluid dynamics (CFD) simulations were conducted on the formic acid-O₂ co-feeding system to help optimize clusters loaded into Al₂O₃. Simulation results showed that the concentration of formic acid decreased in the bulk phase of the Al₂O₃ microsphere due to the convective diffusion, while the oxygen molecules enriched in the bulk phase, and finally the two curves formed a concentration intersection zone 0.26 mm away from the surface edge of Al₂O₃ (Fig. 3a, b and Supplementary Fig. 6). Total mass-transfer resistance increased along the surface-to-bulk pathway, exhibiting zero slope at 0.17 mm from the surface (Fig. 3c), which indicated that the diffusion resistance escalated with deeper penetration, thereby predicting an optimal cluster embedding depth of ~0.2 mm into Al₂O₃.

Fig. 3. Optimization of the cluster distribution on Al₂O₃ and catalytic performances.

Mass transfer processes revealed by CFD simulation: a simulated radial distributions of formic acid in alumina spheres; b radial concentration distribution curves of formic acid and oxygen; c radial mass transfer resistance in alumina spheres. d Schematic diagram of supported catalysts sectioning to obtain the radial cluster distribution. e-h Sectional images of supported catalysts and corresponding pixel brightness distributions. From left to right samples were labeled AA-0, AA-1, AA-2, and AA-3 prepared with the addition of 0%, 1%, 2%, and 3% methanol, respectively. Performances of the above catalysts: i catalytic activities of AA-0, AA-1, AA-2 and AA-3, and error bars represent standard deviation (SD) from three independent experiments; j catalytic stability of AA-1; k formaldehyde degradation with AA-1 catalytic system.

The distribution of clusters onto the Al₂O₃ was experimentally optimized by regulating solvent polarity. We introduced methanol into the CH₂Cl₂ solvent of clusters to change the adsorption preference of clusters onto the support⁴⁶, which promoted the migration of clusters from the surface to the interior of the support. As the volume fraction of methanol increased, a series of the Au₈Pd₁(DPPF)₄²⁺ supported on Al₂O₃ catalysts named as AA-0, AA-1, AA-2, and AA-3 were obtained with the addition of 0%, 1%, 2%, and 3% methanol, respectively. To quantify the regulatory effect on the cluster distribution, the semi-quantitative optical microscopy analysis was adopted, based on the linear relationship between the characteristic deep red absorbance of clusters and concentration. The Al₂O₃ microsphere loaded with clusters were sectioned along the center (Fig. 3d), and the relative density of clusters was characterized, and distribution parameters were determined by the image analysis (Fig. 3e-h). Among them, the 95% cluster localization depth (D_95%, defined as the radial thickness of the region where 95% of clusters accumulated) was used as the main evaluation index. As the volume fraction of methanol increased from 0% to 3%, the radial range of the cluster-enriched region expanded and D_95% increased from 0.14 mm (AA-0) to 0.18 mm (AA-1), 0.28 mm (AA-2), and 0.42 mm (AA-3), respectively. This indicated that enhanced solvent polarity could lead to an increase in cluster embedding depth with the increase of methanol ratio (Supplementary Fig. 7-10), verifying the intrinsic spatial programmability of cluster-based catalyst design.

We next explored catalytic performances of these heterogeneous catalysts of Au₈Pd₁(DPPF)₄²⁺ supported on Al₂O₃ for the formic acid oxidation performed in the fixed-bed reactor. As expected, AA-1 catalyst showed an activity toward the target product of 737 μmol g_cat.^-1 h^-1, which was higher than another catalysts such AA-0, AA-2 and AA-3 (Fig. 3i). The AA-1 catalyst retained >95% of its initial activity after continuous operation for 36 h, indicating its long-term stability (Fig. 3j). Moreover, the heterogeneous catalysts were robust under continuous-flow conditions. FT-IR spectra of the Au₈Pd₁(DPPF)₄²⁺/Al₂O₃ catalysts before and after reaction showed no obvious change (Supplementary Fig. 11), implying that no obvious ligand degradation and oxidation occurred on the supported clusters under reaction conditions. Moreover, from the TEM images of the catalyst after reaction (Supplementary Fig. 12), the clusters exhibited no aggregation during the reaction. Additionally, this catalytic system realized the coupling of H₂O₂ synthesis and formaldehyde degradation (Fig. 3k).

Concerto-like catalysis of Au₈Pd₁(DPPF)₄²⁺

To reveal the reaction mechanism of this catalytic system, we focused on the activation of formic acid on the cluster. In-situ attenuated total reflection infrared spectroscopy (ATR-IR) was employed to track surface species evolution during formic acid oxidation (Fig. 4a). Two sets of characteristic absorption bands were observed, where one was the characteristic bands at 1714 cm^-1 and 1210 cm^-1 assigned to the adsorbed HCOOH, the other characteristic bands at 1575 cm^-1 and 1386 cm^-1 were assigned to the *HCOO⁴⁷. These signals showed that formic acid underwent rapid dehydrogenation on the cluster. Furthermore, electron paramagnetic resonance (EPR) technology combined with 5,5-dimethyl-1-pyrroline-N-oxide (DMPO) trapping agent was carried out to track the radicals formed in different systems (Fig. 4b). In the system of Au₈Pd₁(DPPF)₄²⁺ and H₂O₂, a clear four-line signal of the DMPO-OH adduct was detected, while no obvious radical signal was observed for Au₈Pd₁(PPh₃)₈²⁺ under identical case. A similar phenomenon was observed in the HCOOH + O₂ system, as DMPO-OH adduct was detected for Au₈Pd₁(DPPF)₄²⁺ but not for Au₈Pd₁(PPh₃)₈²⁺. In addition, pure DPPF ligand alone also produced the DMPO-OH signal with H₂O₂, while no radical was generated in the HCOOH + O₂ system with pure ligand, confirming that pure DPPF ligand could not catalyze the reaction of HCOOH and O₂ to form H₂O₂. Cyclic voltammetry (CV) tests further revealed that pure DPPF exhibited a clear anodic oxidation peak at 1.35 V and a corresponding cathodic reduction peak at 1.21 V, with a half-wave potential E_1/2 = 1.28 V (Fig. 4c), verifying its reversible redox activity. The Au₈Pd₁(DPPF)₄²⁺ retained a pair of redox peaks at 1.26 V and 1.12 V (E_1/2 = 1.19 V), indicating that the coordination interaction did not destroy the electron transfer ability of the DPPF ligand. Notably, the CV curve of Au₈Pd₁(PPh₃)₈²⁺ did not exhibit any relevant redox peaks (Supplementary Fig. 13), further highlighting the significance of the DPPF ligand in the formation of hydroxyl radicals.

Fig. 4. Mechanistic investigation of formic acid oxidation and hydroxyl radical generation.

a In-situ ATR-IR spectra obtained from the formic acid oxidation over Au₈Pd₁(DPPF)₄²⁺. b EPR spectra obtained from different reaction systems using DMPO as the spin-trapping agent. c CV curves of Au₈Pd₁(DPPF)₄²⁺ cluster and DPPF ligand. d Reaction mechanism proposed by DFT calculations and corresponding reaction energy curves of Au₈Pd₁(DPPF)₄²⁺ and ligand-free Au₈Pd₁.

To further probe the cooperativity of active sites in Au₈Pd₁(DPPF)₄²⁺, the formic acid oxidation and the hydroxyl radical generation were computed by density functional theory (DFT) calculations (Supplementary Data). As shown in Fig. 4d, a concerto-like catalytic mechanism was proposed. First, HCOOH attached to the cluster (state I, -0.90 eV) and underwent the first dehydrogenation producing *HCOO + *H (state III, -1.08 eV). The removal of the first hydrogen atom proceeded via a transition state TS1 involving the transfer of the H atom from the O–H moiety to the Au site (state II, -0.49 eV), surmounting an energy barrier of 0.41 eV. In contrast, the ligand-free Au₈Pd₁ cluster required a substantially higher barrier of 0.63 eV for this step. Subsequently, the C-bound hydrogen atom migrated from the *HCOO + *H (state III, -1.08 eV) to generate the *COO + *2H intermediate (state V, -0.95 eV) through the H-transfer transition state TS2 (state IV, -0.70 eV), with a remarkably low energy barrier of 0.38 eV. By comparison, the corresponding step on the bare Au₈Pd₁ encountered a much higher barrier of 0.85 eV, resulting in a significantly slower reaction rate than that observed for Au₈Pd₁(DPPF)₄²⁺ (Fig. 4d and Supplementary Fig. 14). Following this, O₂ competitively adsorbed onto the cluster to afford the *O₂ + *2H intermediate (state VI, -1.14 eV), which was then successively hydrogenated to *OOH + *H (state VII, -1.43 eV) and ultimately to *H₂O₂ (state VIII, -1.74 eV).

The key difference lay in the activation of H₂O₂. On Au₈Pd₁(DPPF)₄²⁺, the Fe site bound to ferrocene took part in breaking the O-O bond evenly, generating two ·OH radicals via transition state TS3 (state IX, -1.32 eV) with an energy barrier of merely 0.42 eV. One ·OH attached to the Pd center, while the second was temporarily stabilized by a Fe-O-H···P interaction before preferring to bind to Fe. The calculated dissociation free energy of ·OH on Pd and Fe were 1.09 and 0.83 eV (Supplementary Fig. 15), respectively, enabling rapid desorption of ·OH into the solution phase, where it subsequently catalyzed formaldehyde oxidation. Although the OH moiety on Fe-OH may partially participate in the direct oxidation of formaldehyde, this route did not constitute the dominant pathway. The free energy barrier for Fe-OH catalyzed formaldehyde oxidation was 0.96 eV, which exceeded the corresponding barrier for ·OH desorption into the solution phase (Supplementary Fig. 16). In contrast, decomposing H₂O₂ on bare Au₈Pd₁ needed to overcome a very high energy barrier (1.52 eV) (state IX relative to state VIII). As an alternative reaction pathway, hydrogen peroxide generation may proceed through pre-adsorption of O₂ on the catalyst surface, followed by hydrogen abstraction from liquid-phase HCOOH to afford H₂O₂. DFT calculations revealed that the free energy barriers for H transfer from the C-H and O-H bonds were 1.60 eV and 1.93 eV, respectively, and both are substantially higher than the barrier for the initial adsorption and activation of HCOOH (Supplementary Fig. 17). This computational evidence showed that the DPPF ligands might not only speed up the dehydrogenation of formic acid but also reduce the activation energy needed for H₂O₂ to turn into hydroxyl radicals.

Cluster-mediated O₂ enrichment microenvironment proposed by MD simulations

The physical accessibility of reactants to catalytic centers was further investigated by molecular dynamics (MD) simulations (Fig. 5a). The results showed that the Au₈Pd₁(DPPF)₄²⁺ cluster created a specific microinterfacial environment, in which the hydrophobic ligand shell drove O₂ to the cluster surface and concentrated poorly soluble oxidant within the reaction zone (defined as a 10 Å diameter from the Pd center). Analysis of radial distribution functions (RDFs) demonstrated that O₂ dominated the first solvation shell. Importantly, the Pd-O_O2 distance was substantially shorter than the Pd-C_HCOOH distance (Fig. 5b). Kinetically, O₂ had much higher local diffusion coefficient (0.11 Ų ps^-1) than formic acid (0.05 Ų ps^-1), which allowed O₂ to be rapidly supplied at the Pd center (Fig. 5c). As a result, the residence time and collision frequency of O₂ at the cluster surface were significantly higher than those of formic acid (Fig. 5d).

Fig. 5. The O₂ enrichment microenvironment within the Au₈Pd₁(DPPF)₄²⁺ system.

a Molecular dynamics simulation snapshots of reactant distribution and diffusion on Au₈Pd₁(DPPF)₄²⁺/Al₂O₃. Color code: Pd (cyan), Au (gold), P (orange), Fe (turquoise), Al (pink), O (red), C (gray), H (white). b Radial distribution functions (RDFs) of Pd-O_O2 and Pd-C_HCOOH. c Mean square displacement of O₂ and HCOOH within 10 Å of the Pd center. d Residence times and collision frequencies of reactants with the cluster surface.

Discussion

Atomically precise Au₈Pd₁(DPPF)₄²⁺ cluster with well-defined modular structure provides mechanistic insights into concerto-like catalysis between the Au₈Pd₁ core and DPPF ligand, where the two components collaborate to couple the generation and utilization of H₂O₂. Through optimizing gradient cluster distribution on Al₂O₃ monolith, heterogeneous Au₈Pd₁(DPPF)₄²⁺/Al₂O₃ catalysts achieve the concurrent enhancement of catalytic activity and stability under continuous-flow conditions. This study demonstrates a feasible strategy for realizing controlled gradient spatial arrangements of clusters into supports at the macroscopic scale, while maintaining the cluster structure at the nanoscale. It offers the rational design of high-performance catalysts, advancing the development of precise and efficient catalytic systems for distributed chemical manufacturing in the continuous-flow production lines.

Methods

Materials

1,1’-bis(diphenylphosphino)ferrocene (DPPF, ≥99%), ferrocene (Fc, ≥99%), formic acid ( ≥ 99%), o-phenylenediamine (OPD, ≥99%), horseradish peroxidase (HRP, ≥150 U/mg enzyme powder), formaldehyde (37 wt% in H₂O), acetylacetone ( ≥ 99%), ammonium acetate ( ≥ 99%), and acetic acid ( ≥ 99%) were all purchased from Aladdin Reagent Co., Ltd.

Synthesis of clusters

Au₈Pd₁(DPPF)₄²⁺: Au₈Pd₁(PPh₃)₈²⁺ (50 mg) and DPPF (150 mg) were dissolved in a small amount of dichloroethane (1 mL). Tetrahydrofuran (20 mL) was added, and the mixture was transferred to a 50 mL PTFE-lined stainless steel autoclave reactor. The vessel was sealed, heated at 100 °C for 24 hours, yielding Au₈Pd₁(DPPF)₄²⁺ as crystalline form. The successful preparation of Au₈Pd₁(DPPF)₄²⁺ was confirmed via UV-Vis spectra and ESI-MS. The clusters Au₈Pd₁(PPh₃)₈²⁺ and Au₈(DPPF)₄²⁺ were prepared according to reported procedures^41,42.

Preparation of heterogeneous catalysts

Au₈Pd₁(DPPF)₄²⁺ clusters (10 mg) were dissolved in CH₂Cl₂ (10 mL). Al₂O₃ microspheres (1 g) were added, and the suspension was stirred for 90 min at room temperature. Subsequently, the solid product was separated and dried at room temperature, and the obtained catalyst was labeled as AA-0.

To tune the spatial distribution of clusters within Al₂O₃ microspheres, methanol was added to the CH₂Cl₂ solution to modulate solvent polarity and electrostatic interaction strength. Au₈Pd₁(DPPF)₄²⁺ clusters (10 mg) were dissolved in CH₂Cl₂ (10 mL). 100, 200, and 300 μL of methanol, corresponding to 1%, 2%, and 3% v/v relative to the 10 mL solvent volume, were added to the cluster solution, respectively. 1 g of Al₂O₃ microspheres was added, and the mixture was stirred for 90 min. After the solid was separated and dried, the obtained catalysts were labeled as AA-1, AA-2, and AA-3 in sequence.

Catalytic evaluation

Synthesis of H₂O₂ in the batch system: high purity air was bubbled through deionized water (900 μL) at 20 mL min^-1 for 5 min to achieve O₂ saturation. Catalysts (1 mg) were dispersed in DMF (1 mL), and 20 μL of this solution was added to the O₂ saturated water. Formic acid (40 μL, 500 mM) was then injected. After the reaction for 2 min, 20 μL of horseradish peroxidase (HRP, 0.1 mg mL^-1) and 20 μL of o-phenylenediamine (OPD, 25 mM, aqueous solution) were sequentially added to perform the color reaction. The absorbance was determined at a wavelength of 450 nm using a UV-Vis spectrophotometer.

Synthesis of H₂O₂ in the fixed-bed reactor: 1 g of supported catalysts was packed into a quartz fixed-bed reactor (outer diameter: 8 mm), 500 mM formic acid aqueous solution was pumped at a flow rate of 200 μL min^-1, and high purity air was simultaneously introduced at a flow rate of 20 mL min^-1. After the reaction stabilized, 40 μL of liquid sample was collected at the outlet and diluted to 920 μL with deionized water, and the subsequent detection method was the same as above. The data were presented as mean ± standard deviation.

Formaldehyde degradation in the fixed-bed reactor: under the same reactor and catalyst packing conditions, a mixed aqueous solution containing formic acid (500 mM) and formaldehyde (333 mM) was introduced at a flow rate of 200 μL min^-1, and high purity air was simultaneously introduced at flow rate of 20 mL min^-1. After the reaction, 5 mL of the liquid product was taken, 500 μL of color developing agent (aqueous solution containing 500 g L^-1 ammonium acetate, 6% (v/v) acetic acid, and 0.5% (v/v) acetylacetone) was added, and the mixture was heated in a 60 °C water bath for 10 min. The absorbance was determined at a wavelength of 414 nm using a UV-Vis spectrophotometer, the residual formaldehyde concentration was calculated through the formaldehyde standard curve.

Characterizations

Powder X-ray diffraction (XRD) patterns were recorded on a Bruker D8 ADVANCE diffractometer using Cu Kα radiation operating at 40 kV and 40 mA.

Transmission electron microscopy (TEM) was performed on a JEOL JEM-2100F instrument operated at 200 kV.

Attenuated total reflectance Fourier transform infrared (ATR-IR) spectra were acquired on a Nicolet iS50 spectrometer equipped with an ATR attachment. In-situ ATR-IR measurements were conducted to probe formic acid activation pathways. High purity air was bubbled through deionized water (600 μL) at 20 mL min^-1 for 5 min to achieve O₂ saturation. Au₈Pd₁(DPPF)₄²⁺ clusters (1 mg) were dispersed in DMF (1 mL), and 200 μL of this solution was added to the O₂ saturated water. Background spectra were collected after cluster addition. Formic acid (200 μL, 500 mM) was then injected, and time-resolved spectra were recorded.

Ultraviolet-visible (UV-Vis) spectra were characterized using a Shimadzu UV-2600 spectrophotometer.

Cyclic voltammetry (CV) measurements were conducted on a CHI660E electrochemical workstation. The measurements were performed using a three-electrode system in a dichloromethane (DCM) solution containing 0.1 M tetrabutylammonium hexafluorophosphate as the electrolyte. The clusters to be tested were prepared at a concentration of 10 μM. The working electrode was a glassy carbon electrode, and a platinum sheet electrode served as the counter electrode, an Ag/AgCl electrode saturated with 3 M KCl was employed as reference electrode, with a scan rate of 50 mV s^-1.

Electron Paramagnetic Resonance (EPR) spectra were recorded at room temperature using a Bruker EMX PLUS spectrometer. High purity air was bubbled through deionized water (940 μL) at 20 mL min^-1 for 5 min to achieve O₂ saturation. Catalysts (1 mg) were dispersed in DMF (1 mL), and 20 μL of this solution was added to the O₂ saturated water. Formic acid (40 μL, 500 mM) was then injected. The resulting solution was mixed with 100 μL of 5,5-dimethyl-1-pyrroline N-oxide (DMPO), and 30 μL of the mixture was transferred into a quartz standard sample tube. The bottom of the tube was sealed with wax, and the sample tube was then inserted into a dedicated EPR test tube for measurement.

Electrospray ionization mass spectra (ESI-MS) were obtained on a Thermo Fisher Scientific Q Exactive combined Quadrupole-Orbitrap Mass Spectrometer.

Computational details

Computational details related to DFT calculation, molecular dynamics (MD) simulation and computational fluid dynamics (CFD) simulation are provided in Supplementary Information.

Author contributions

Y.Z. conceived the project. X.C. performed the experiments. J.Y. performed the theoretical calculations. S.H. performed the TEM characterizations. All authors wrote the manuscript.

Peer review

Peer review information

Nature Communications thanks Hsing-Yin Chen, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

Y.Z. discloses support for the research of this work from National Natural Science Foundation of China [grant number 22125202, 92461312 and U24A20487]. X.C. discloses support for the research of this work from National Natural Science Foundation of China [grant number 22302091]. J.Y. and S.H. declare no relevant funding.

Data availability

The data that support the findings of this study are presented in the manuscript and supplementary information file. The source data are provided in the Source Data file. Source Data file has been deposited in the figshare database under accession code 10.6084/m9.figshare.32149171. All data are available from the corresponding author upon request. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74604-w.