All solvents and reagents except silica are commercially available without further purification. The silica was soaked overnight in a 10 wt% aqueous nitric acid solution, then washed repeatedly with deionized water.

The bisphosphine ligand PNP was synthesized by a slight improvement in the literature⁴³. Specifically, potassium tert-butoxide (7.0 g, 62.5 mmol) was added to 125 mL of anhydrous tetrahydrofuran (THF) under nitrogen atmosphere. Diphenylphosphine (7.0 mL, 40 mmol) was then added to obtain a deep red solution. The mixture was stirred for 5 min, then bis(2-chloroethyl)amine hydrochloride (3.575 g, 20 mmol) was added. The mixture was refluxed under nitrogen atmosphere at 80°C for 16 h. The mixture was poured into 200 mL of hexane and washed sequentially with 75 mL of 10% aqueous NaOH and saturated aqueous sodium chloride. The organic layer was separated, filtered and stirred vigorously with 200 mL of 2 mol/L aqueous HCl solution to give a thick white precipitate. Recrystallization from 75 mL of boiling acetonitrile gave white needles of bis[2-(diphenylphosphino)ethyl]ammonium hydrochloride.

Under nitrogen atmosphere, 2.0 g of pretreated SiO₂ was added to 30 mL of anhydrous toluene, and 3-chloropropyltriethoxysilane (CPTES) (4.0 mL, 16 mmol) was added slowly dropwise and the mixture was refluxed at 110°C for 24 h. The solid phase was washed twice with anhydrous toluene and twice with ethanol and distilled water to remove unreacted CPTES. The white solid was obtained by filtration and washed twice with anhydrous toluene and then twice each with ethanol and distilled water to remove unreacted CPTES. The sample was dried at 100°C for 24 h and denoted as Cl-SiO₂.

The mass fraction of phosphorus (P) in PNP-SiO₂ was determined via ICP-OES under varying PNP additions and reaction durations, allowing for the identification of the optimal reaction conditions (Fig. S1). Typically, 2.87 g (6 mmol) of PNP, 1.00 g of Cl-SiO₂ and 1.64 g (16 mmol) of triethylamine were added to 30 mL of anhydrous toluene under nitrogen atmosphere. The mixture was stirred and refluxed at 110°C for 36 h. Here, triethylamine was used as a deprotonator and also to trap HCl released during the reaction. After filtration, the solid was washed with dichloromethane and deionized water and then dried at 100°C overnight, denoted as PNP-SiO₂.

0.05 g of Rh(acac)(CO)₂ was dissolved in 25 mL of N,N-dimethylacetamide, and then 1.98 g of PNP-SiO₂ was added. The mixtures were stirred under nitrogen and a syngas atmosphere (CO/H₂ = 1) for 24 h at 30 ℃, respectively, then centrifuged and washed three times with N,N-dimethylacetamide. The solids were dried under vacuum at 30 ℃ for 24 h. The samples obtained were named Rh/PS-N₂ and Rh/PS-SG, respectively.

An autoclave reactor (Parr 4598) was applied to conduct the formaldehyde hydroformylation reaction. Typically, different amounts of the prepared catalysts (n_Rh = 0.020 mmol), FA (24 mmol, paraformaldehyde used as the precursor of FA), and 25 mL N,N-Dimethylacetamide were loaded. The reactor was then purged with nitrogen and syngas (CO / H₂ = 1), respectively, and charged with high-pressure syngas. The reactor was heated to 95°C with stirring and kept for 3 h. The reacted liquid was centrifuged and extracted by xylene and deionized water sequentially, followed by the addition of isopropanol as an internal standard. The aqueous phase was taken into Gas Chromatography/Mass Spectrometry (GC/MS, Agilent) with a VF-WAXms column (60m × 0.25mm) and a flame ionization detector (FID) for analysis. The conversion of HCHO (X_FA), selectivity of products (S_i) and the turnover frequency (TOF) were calculated according to the following formula:

wherein i represents the products, including GA and MeOH; n_{in, FA} refers to the initial molar amount of FA; n_Rh refers to the mole number of Rh atoms; D_Rh refers to the dispersion of Rh species calculated via CO pulse chemical adsorption; t is the reaction time.

The actual content of rhodium and phosphorus in the samples was determined by Inductively Coupled Plasma optical emission spectrometer (ICP-OES; Jobin Yvon, Ultima-2). X-ray powder diffraction (XRD) patterns were determined using a Rigaku Miniflex II powder diffractometer with the operating current of 15 mA and the tube voltage of 40 kV, the scanning range of 2θ = 5˚~70˚, and the scanning step of 3 ˚/min.

The BET surface area and pore structure of samples were determined using an automated physicochemical adsorption tester (Micromeritics Tristar II 2460) with high purity nitrogen as the adsorbent, and adsorption and desorption were carried out − 196°C). The samples were vacuum treated at 150°C for 6 hours prior to testing to remove adsorbed water.

The surface chemical environment of the samples was analyzed by X-ray photoelectron spectroscopy (XPS) and measurements were performed using a Thermo Fisher Scientific ESCALAB 250Xi spectrometer equipped with a mono-chromatic Al-K radiation source (hν = 1486.6 eV) and the measured spectral data were corrected for the binding energy of C 1s (284.8 eV). Thermogravimetric-differential thermal analysis (TG-DTA) was tested using a Mettler Toledo thermogravimetric analyzer. The tests recorded the mass change of the samples from 30°C to 800°C at a rate of 10°C/min in an air atmosphere.

Transmission electron microscopy (TEM) and energy-dispersive X-ray spectroscopy (EDS) mapping images were obtained on a Tecnai F20.

The Fourier transform infrared (FT-IR) spectra of the samples were tested by an infrared spectrometer with a Bruker Vertex 70 with the measuring wave numbers in the range of 4000 to 400 cm^− 1 and a spectral resolution of 4 cm^− 1. In situ CO-diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) was measured on a Thermo Fisher Nicolet 6700 infrared spectrometer in the range of 900 to 4000 cm^− 1. ³¹P MAS nuclear magnetic resonance (NMR) spectra were performed on a Bruker 400 MHz AVANCE Ⅲ.

All geometry optimizations and frequency calculations were carried out using the ORCA 6.1 package^44,45. The PBE0 hybrid functional^46,47, which mixes the Perdew-Burke-Ernzerhof exchange-correlation functional with a fraction of Hartree-Fock exchange, was employed in combination with the def2-TZVP basis set⁴⁸ for all atoms. Grimme’s DFT-D3(BJ) empirical dispersion correction^49,50 was applied throughout all calculations. Where applicable, temperature (T = 368.15 K) and pressure (P = 12 MPa) corrections were included in the thermochemical analysis. Thus, the employed level of theory can be denoted as PBE0-D3(BJ)/def2-TZVP.

All optimized structures were confirmed as minima by the absence of imaginary frequencies, while each transition state exhibited one and only one imaginary frequency, as verified by vibrational frequency analysis.

The Gibbs free energies (G) were obtained by combining the electronic energies from single-point calculations with the corresponding thermochemical corrections derived from the vibrational frequency analyses. All three-dimensional molecular structures were visualized using VMD (Visual Molecular Dynamics)⁵¹ and Multiwfn 3.8(dev) ^52,53.

Mesoporous silica materials are widely used as supports for various catalytic transformations. It is well known that the SiO₂ surface can be functionalized by dehydration condensation of Si-OH with silane coupling agents. The synthesis of the support PNP-SiO₂ through a grafting technique is shown in Scheme 1. First, we covalently grafted the PNP pincer ligand bis[2-diphenylphosphinoethyl]amine onto SiO₂ using a surface modification method to obtain PNP-SiO₂ samples. Subsequently, PNP-SiO₂ was impregnated in an organic solution of the Rh precursor under syngas and N₂ atmosphere, respectively. Fourier transform infrared (FT-IR) spectroscopy (Fig. 1) can be used to verify whether the support has been successfully functionalized. All the supports and catalysts showed absorption peaks at 1630 cm^− 1, 1105 cm^− 1, 805 cm^− 1, and 470 cm^− 1, which were assigned to δ(H-O-H) of H₂O in the silica framework^54–56, ν_as(Si-O-Si) and ν_s(Si-O-Si) and δ(Si-O-Si)⁵⁷. The absorption band of the SiO₂ sample located at 974 cm^− 1 was assigned to ν(Si-OH) on the silica surface, whereas on the grafted sample, the redshift to 958 cm^− 1 was due to the grafting with silane coupling agent, which functionalized the surface of SiO₂. In addition to the bands of silica, the grafted samples not only showed the stretching vibrational absorption peaks of methyl and methylene at 1442 cm^− 1 and 1404 cm^− 1, but also detected a C-H out-of-plane bending vibrational peak of monosubstituted benzene at 756 and 698 cm^− 1. The appearance of the above bands indicated that the silane coupling agent and PNP ligand were successfully introduced into SiO₂. In Rh/PS-SG, infrared spectroscopy revealed rhodium geminal dicarbonyl species (2064 and 1992 cm^− 1), consistent with CO adsorption onto Rh sites. In contrast, no carbonyl stretches were detected in Rh/PS-N₂, indicating complete CO desorption. Furthermore, the characteristic vibrational modes of acetylacetone which present in the precursor were absent in both Rh/PS-SG and Rh/PS-N₂. Instead, acetylacetone was detected in both post-loading filtrates (Fig. S2), indicating that acetylacetone anion captured a hydrogen proton and then detached from the immobilized catalysts into the liquid system. It is hypothesized that in the syngas, hydrogen cleaves and binds to the acetylacetone anion and Rh species, respectively, whereas in nitrogen, the Si-OH provides a hydrogen proton to the acetylacetone anion to form an acetylacetone molecule, simultaneously accompanied by the formation of Si-O-Rh (Scheme 1). The successful anchoring of Rh species was further confirmed by ³¹P MAS NMR spectroscopy (Fig. 2). The ³¹P MAS NMR spectra of PNP-SiO₂ showed peaks at 16.4 and 26.5 ppm corresponding to the PNP ligand immobilized on the SiO₂ surface. In contrast, the Rh/PS-SG and Rh/PS-N₂ immobilized PNP ligands show a decrease in the intensity of the main peak and the appearance of a new peak at -23.3 ppm, which was attributed to the coordination of the PNP pincer ligand to Rh via the P donor. The ³¹P chemical shift is a sensitive indicator of the coordination state of the PNP pincer ligand^17,58. This proved that Rh species are anchored to the support surface by coordination with PNP.

The X-ray diffractograms of the powders of the supports and catalysts (Fig. S3) show only one plate peak attributed to amorphous SiO₂ at 21.6°, and no characteristic diffraction peaks representative of Rh or Rh₂O₃ are observed in either Rh/PS-SG or Rh/PS-N₂ samples, implying highly dispersed Rh species. The textural properties of supports and catalysts were depicted in Fig. 3 and Table 1. The N₂ adsorption-desorption isotherms for all samples are classified as type IV with H4 type hysteresis loop⁵⁹. In addition, the specific surface area, pore volume and the average pore size of the grafted samples were reduced due to the modification of the organics blocking part of the pore structure of the silica. Thermogravimetric analysis of the supports (Fig. S4) revealed that the organic functional groups introduced by grafting started to oxidatively decompose at 230°C, which indicates good thermal stability at the catalyst evaluation temperature of 95°C for the formaldehyde hydroformylation reaction.

High-resolution transmission electron microscopy (HRTEM) images of the catalysts (Fig. 4) reveal that the Rh species are highly dispersed on the grafted supports, with no observable agglomeration. CO-probe Fourier transform infrared spectroscopy (CO-FTIR) of Rh/PS-SG and Rh/PS-N₂ (Fig. 5) exhibited symmetric and asymmetric stretching vibrational peaks (2071 and 1997 cm^− 1) characteristic of Rh(CO)₂ species⁶⁰ in Rh/PS-SG and Rh/PS-N₂. Notably, the absence of bridging CO adsorption bands (typically observed at 1800–1900 cm^− 1)¹, which would indicate Rh clusters or nanoparticles, provides additional evidence for the atomically dispersed nature of the Rh species.

XPS measurements were used to further investigate the valence and coordination environment of Rh species on the catalyst. After deconvolution, the P 2p spectra of Rh/PS-SG and Rh/PS-N₂ (Fig. 6) displayed two peaks at 132.5 and 133.3 eV. The former peak was attributed to the uncoordinated PNP, while the appearance of the high binding energy peak was due to the formation of coordination bond between Rh and P, which reduced the electron density around the P atom. The N 1s spectra of the samples (Fig. S5) before and after rhodium loading only showed a single peak of 399.8 eV, indicating that N atom is not involved in the coordination with Rh species. This further proves that the rhodium species is anchored to the carrier by coordination with the P atom in PNP. XPS analysis of Rh 3d (Fig. 7) showed two peaks at binding energies (BE) of 307.5 eV and 308.6 eV attributed to Rh⁰ and Rh⁺, respectively^19,61, indicating that the binding energies of Rh⁺ species decreased after coordination with PNP, which is also in accordance with the results of P 2p spectra. In addition, Table S1 reveals a higher proportion (45.9%) of Rh⁰ species in Rh/PS-SG due to the strong reducing nature of syngas. Recent literature has demonstrated that the synergistic interplay between Rh⁰ and Rh⁺ sites is critical for enhancing the hydroformylation of formaldehyde⁴⁰. Rh⁰ species serve as active sites for the adsorption and activation of H₂ and formaldehyde, whereas Rh⁺ sites facilitate the formation of the Rh⁺(CO)_x(PPh₃)_y active species. Consequently, the tailored distribution of Rh⁰ and Rh⁺ in Rh/PS-SG is anticipated to favor the enhancement of catalytic activity toward this reaction.

Based on the analytical results of FT-IR, solid-state ³¹P NMR and XPS, we found that in Rh/PS-SG the Rh atoms were coordinated to P and carbonyl groups, while in RhN₂, the Rh atoms were only coordinated to the P atoms of PNP. Furthermore, the additional band at 2104 cm^− 1 in the CO-FTIR spectrum (Fig. 5) of the Rh/PS-N₂ catalyst is assigned to Rh(O)(CO) species⁶², which is consistent with different coordination environments of Rh in the two catalysts. The specific structure of the catalyst was then inferred in combination with the effective atomic number (EAN) rule, and a reasonable Rh coordination model was proposed (Scheme 2). In formaldehyde hydroformylation, the active species of the reaction is usually considered to be the HRh(CO)₃L complex (where L represents the phosphine ligand)^63,64, which is also consistent with the structure of Rh/PS-SG in Scheme 2(a).

We assessed the catalytic performance of Rh/PS-SG and Rh/PS-N₂ for formaldehyde hydroformylation and the homogeneous catalyst Rh(CO)₂(acac) served as comparators. As shown in Fig. 8, the FA conversion and GA selectivity of Rh/PS-SG were 12.5% and 88.3%, respectively, which were remarkably higher than those of the homogeneous catalyst Rh(CO)₂(acac) of 3.4% and 66.5%. Despite the similar metal loading (ICP-OES, Table S1), the FA conversion (4.7%) and GA selectivity (74.8%) were considerably lower for Rh/PS-N₂. The enhanced activity of the Rh/PS-SG was attributed to the fact that Rh was coordinated with PNP pincer ligand in the syngas to produce the HRh(CO)₂L species, which is structurally similar to the homogeneous Rh-P complex catalysts, and therefore can replace soluble triphenylphosphine in catalyzing formaldehyde hydroformylation.

Additionally, color differences of the fresh mixture extracted immediately after reactions are observed in Fig. 8, indicating that more losses of Rh occurred in the catalytic system of Rh/PS-N₂ than in the catalytic system of Rh/PS-SG. This viewpoint has been further confirmed by the ICP-OES results (Table S1) that the Rh loading of the Rh/PS-N₂ catalyst decreased from 0.31% to 0.16% after the reaction, whereas that of the Rh/PS-SG catalyst decreased only marginally from 0.39% to 0.36%. Structurally, the Rh center within Rh/PS-N₂ has not enough coordination space to bind CO and H₂ due to steric hindrance of four PNP ligands. Consequently, the formation of active species requires the disruption of the parent framework, which in turn leads to significant leaching of Rh. Meanwhile, massive loss of rhodium species resulted in a decrease in activity. Based on experimental phenomena and ICP-OES results, we speculated that the structure of Rh/PS-N₂ is unstable in the formaldehyde hydroformylation.

To further probe catalyst practicality, the recyclability of the Rh/PS-SG catalyst was evaluated by consecutive cycle tests (Fig. S6). Over multiple cycles, the selectivity for GA remained at ~ 90%, whereas the FA conversion declined, which was attributed to the agglomeration of Rh species during the reaction (Fig. S7). Notably, all heterogeneous catalysts reported to date require the addition of PPh₃ in reactions. In comparison with these catalysts (Table S3), the Rh/PS-SG catalyst synthesized in this work exhibits a relatively high performance in GA production without the addition of soluble phosphine ligands, with a turnover frequency (TOF_GA) of 99.6 h^− 1.

In this study, density functional theory (DFT) calculations were performed to compare the geometric structures, electrostatic potential distributions, and frontier molecular orbitals of two rhodium-based complexes (Rh/PS-SG and Rh/PS-N₂), aiming to elucidate the intrinsic mechanisms underlying their differences in catalytic activity. The results show that Rh/PS-SG exhibits good catalytic performance, whereas Rh/PS-N₂ undergoes immediate dissociation upon CO coordination and fails to proceed through any further catalytic steps.