Membrane synthesis and characterizationLeveraging the superior catalytic properties of aminocatalysts in the Knoevenagel condensation reaction, we selected 2,5-bis(2-(dimethylamino)ethoxy)terephthalohydrazide (Bth), known for its prominent tertiary amine group, as a crucial component in synthesizing a COF membrane. The combination of Bth with 1,3,5-triformylphloroglucinol (Tp) was specifically chosen because it effectively promotes the transformation of reversible imine bonds into irreversible β-ketoenamine structures, thereby enhancing the chemical stability of the COF. To optimize operational efficiency while maintaining the organic solvent resistance and water transport efficiency of the membrane, we grow the catalytically active COF layer onto a superhydrophilic polyethersulfone (PES) surface via liquid-solid-liquid interfacial polymerization technique. The procedure entailed positioning a Tp solution in ethyl acetate and mesitylene on one side of the membrane and a Bth solution in acetic acid on the opposite side, leading to the creation of the COF-TpBth/PES membrane (Supplementary Fig. 1). Post-reaction, the membrane underwent a color shift from white to dark yellow. A water contact angle test indicated a transition from superhydrophilic to hydrophobic properties, evident in the water droplet contact angle increasing from 0° to 128°. Conversely, the contact angle for organic compounds, such as benzaldehyde, decreased from 21° on the PES to 0° (Fig. 2a and Supplementary Fig. 2). This dual organophilic and hydrophobic nature is expected to be advantageous for catalytic processes that generate water as a by-product, by facilitating the concentration of organic reactants and promote the efficient removal of H2O.Fig. 2: Membrane characterization.a Depiction of the COF-TpBth/PES composite membrane structure and the wettability comparison between the PES substrate and the COF-TpBth/PES membrane. b SEM images of the COF-TpBth/PES membrane with an inset showing the cross-sectional view. c PXRD patterns, with an inset displaying the chemical structure of COF-TpBth.Scanning electron microscopy (SEM) analysis unveiled distinctive surface features on the COF-TpBth/PES membrane. In contrast to the original PES surface, which featured numerous micron-sized pores, the COF-TpBth/PES membrane exhibited a more compact and smoother surface (Fig. 2b and Supplementary Fig. 3). Cross-sectional SEM images provided additional insights into the structure, revealing a COF layer thickness of approximately 200 nm (Fig. 2b, inset). Dye exclusion tests indicated that the membrane possessed diffusion channels smaller than methyl blue (2.3 nm), affirming a defect-free structure in the COF-TpBth/PES membrane (Supplementary Fig. 4). To ascertain the successful creation of the β-ketoenamine COF structure in the COF-TpBth membrane, multiple analytical methods were employed. For characterization purposes, the PES substrate was dissolved in dimethyl sulfoxide (DMSO). Fourier-transform infrared spectroscopy (FT-IR) results were particularly informative, showcasing the disappearance of characteristic aldehyde group signals from Tp and –NH2 signals from Bth. Additionally, new signals corresponding to amide C = O and C = C vibrations emerged, strongly suggesting the formation of the β-ketoenamine COF (Supplementary Fig. 5)56. Solid-state 13C nuclear magnetic resonance (13C NMR) spectroscopy further substantiated this conclusion by displaying characteristic C = O signals at 190.0 ppm (Supplementary Fig. 6). Powder X-ray diffraction (PXRD) patterns demonstrated excellent crystallinity, with diffraction peaks aligning with the eclipse AA stacking structures (Fig. 2c). N2 sorption isotherms revealed a Brunauer–Emmett–Teller (BET) surface area of 459 m2 g−1 for COF-TpBth. Pore size distribution analysis indicated an average pore size of about 2.0 nm, consistent with the proposed structures (Supplementary Fig. 7). To assess the inherent basic strength of the COF-TpBth/PES membrane, we employed an in-situ probe molecule adsorption technique in conjunction with FT-IR spectroscopy (Supplementary Fig. 8). Deuterated chloroform (CDCl3), chosen for its mild acidity, was selected as the probe. Following the adsorption of CDCl3, FT-IR analysis revealed a broad absorption band between 2180 − 2290 cm−1, indicating the stretching vibrations (\({\nu }_{{CD}}\)) of the C − D bond in CDCl3. According to the equation: \(\log \Delta {\upsilon }_{{CD}}=0.0066{PA}-4.36\), where \(\Delta{\upsilon }_{{CD}}\) represents the shift in \({\nu }_{{CD}}\) (cm−1), and \({PA}\) is the proton affinity (kJ mol−1), revealing that COF-TpBth exhibits moderate basicity57.Catalytic efficiency evaluationTo assess the efficacy of COF-TpBth/PES as a membrane reactor, we selected the Knoevenagel condensation of benzaldehyde with malononitrile, resulting in the formation of benzalmalononitrite, as a representative model reaction. The catalytic process occurred within a specially designed dead-end membrane system (Fig. 3a). The experimental procedure involved loading the feed chamber with a 0.5 mL ethanol solution containing 0.5 M benzaldehyde and 0.55 M malononitrile (ethanol was identified as the optimal solvent through solvent screening experiments, Supplementary Fig. 9). A constant vacuum suction was applied to facilitate the permeation of this mixture through the membrane, propelling the reactants to traverse the membrane and enhancing the interaction with the catalytic sites within the pore channels of the membrane reactor. The catalytic reaction initiated upon the contact of the reactants with the active sites of the COF reactor during the transmembrane process. Simultaneously, the filtrate was collected in a flask, and the yield of the membrane catalysis was determined using 1H NMR spectroscopy. The volumetric flow rate of the reaction mixture remained constant, as evidenced by a linear flow volume-time relationship (Supplementary Fig. 10). Furthermore, the membrane flux exhibited an almost linear increase as the vacuum suction increased up to 0.8 bar (Supplementary Fig. 11).Fig. 3: Setup for membrane reactions and evaluation of catalytic performance.a Schematic representation of the membrane catalysis experimental setup. b Comparative analysis of catalytic efficiency among the COF-TpBth/PES membrane, a shredded COF membrane, the corresponding COF powder, and a homogeneous catalyst (Bthba). The grey areas on the x-axis indicate the reaction temperatures corresponding to the catalytic data obtained above. The error bars represent the standard deviation of three independent measurements of the various membranes. c Arrhenius plots along with the corresponding apparent activation energies.The efficacy of the membrane reactor was benchmarked against three distinct catalytic systems: a shredded COF membrane, COF powder, and a homogeneous catalyst (Bthba) derived from Bth and benzaldehyde, and all the measurements were replicated with three independent experiments. Batch reactions for these catalytic systems, with the exception of the membrane reactor, were carried out using reaction volumes derived from the permeance of the membrane reactor, maintaining identical temperature conditions and stirring. The membrane reactor exhibited exceptional performance, achieving complete conversion of benzaldehyde under a vacuum suction of 0.05 bar, maintaining in a stable turnover frequency (TOF) of 60 ± 0.23 min–1 over a period of 8 min. This performance greatly surpassed that of the batch reactions, where the initial turnover frequencies for the shredded membrane, COF powder, and homogeneous catalyst were 18, 16, and 25 min–1, respectively, under similar conditions. Notably, the contact time between the solution and the membrane, calculated based on the volumes of the reaction solution and the COF active layer, was only 0.63 s to achieve 100% conversion (see details in the Supplementary Methods). In contrast, at 60 °C over an 8-minute period, the shredded COF membrane, the COF powder, and the Bthba catalyst reached conversion rates of 61.0%, 48.5%, and 71.0%, respectively (Fig. 3b).To further elucidate the experimental observations, we undertook additional experiments. We began by calculating the activation energies (\({E}_{a}\)) for the condensation of benzaldehyde with malononitrile. This involved analyzing Arrhenius plots, which correlate product yield with reaction temperature, as shown in Supplementary Fig. 12. The results revealed \({E}_{a}\) values of 15.1 kJ mol–1 for the membrane reactor, 91.5 kJ mol–1 for the COF powder, 51.5 kJ mol–1 for the shredded COF membrane, and 38.3 kJ mol–1 for the homogeneous Bthba catalyst (Fig. 3c). The relatively low \({E}_{a}\) for the membrane reactor suggests its superior efficiency in activating reactants, thereby enhancing the rate of the Knoevenagel condensation reaction. The disparity in catalytic performance among these systems, despite similar active sites, is primarily attributed to two factors. First, the synergy between the hydrophobic COF layer and the superhydrophilic PES substrate in the membrane reactor. This unique combination boosts reactant concentration while simultaneously repelling water (a by-product), continuously shifting the reaction equilibrium towards product formation, leading to higher single-pass conversion. This is also evidenced by the superior performance of the shredded COF membrane compared to the COF powder. Second, the pressure-driven transit of reactant molecules through the narrow COF pore channels, which likely aligns them in an orderly manner and enhances their interaction with active sites, thereby facilitating the reaction and reducing the activation energy58,59,60.Investigation of the influence of reactant concentration on the reactionSubsequently, experiments were carried out over a range of initial reactant concentrations from 0.1 to 0.9 M. In Fig. 4a, the rate of benzaldehyde conversion and TOF are plotted against the initial benzaldehyde concentration (the concentration of malononitrile is 1.1 times that of benzaldehyde). For concentrations between 0.1 to 0.5 M, both plots exhibit linearity. However, at 0.5 M, the conversion rate peaks and then decreases with further concentration increases, while the TOF continues to rise as reactant concentration increases. This trend suggests a strong dependency of the reaction rate on reactant concentration. To quantitatively analyze the reaction order, the homogeneous Bthba catalytic system was studied. The time-dependent product yield curve in Fig. 4b shows a linear relationship between time (\(t\)) and the concentration expression \({{\mathrm{ln}}}\,[(b(a-x))/(a(b-x))]\). In this expression, \(x\) represents the product (benzalmalononitrite) concentration, \(t\) is the reaction time, and \(a\) and \(b\) are the initial concentrations of malononitrile and benzaldehyde, respectively. The correlation coefficient of this relationship is notably high at 0.982, suggesting that the reaction follows a pseudo-second-order kinetic model61,62. To further investigate the individual effects of benzaldehyde and malononitrile concentrations on the reaction, their concentrations were varied separately in a series of experiments. It was observed that the influence of these two reactants on the reaction kinetics was nearly identical, suggesting a comparable impact from both benzaldehyde and malononitrile (Supplementary Fig. 13). Both the COF powder and membranes used in batch reactions displayed similar rate constant terms for benzaldehyde and malononitrile as observed with the homogeneous catalyst Bthba (Supplementary Figs. 14 and 15). These findings underscore the significant role that reactant concentration plays in dictating the reaction rate; as concentrations decrease during the reaction, the reaction rate also tends to diminish.Fig. 4: Investigation of reaction order and transport kinetics across the membrane for various molecules.a Assessment the impact of reactant concentration on the performance of COF-TpBth/PES. The grey areas on the x-axis indicate the reactant concentrations and vacuum suction applied in relation to the catalytic data obtained above. b Time-dependent variation of product yield in Knoevenagel condensation catalyzed by Bthba at room temperature. The accompanying empirical equation is derived from linear fitting. c Evaluating the influence of water on reaction efficiency across different catalytic systems. d Snapshots illustrating the dynamic behavior of benzaldehyde, ethanol, and water within COF-TpBth nanochannels at specific time points (0, 0.37, and 1.28 ns). The diffusion rates follow the order: H2O > ethanol > benzaldehyde (red, O; white, H; cyan, C; grey, COF-TpBth). Some molecules in the pore channel are depicted as dots for clarity. e Statistical analysis of transmembrane molecule permeation across the COF membrane. (f) A magnified view highlighting a pore channel of the COF membrane reactor filled with a hydrogen-bonded water wire. Hydrogen-bonding interactions among the H2O molecules are observable. These interactions, combined with the hydrophobic environment created by surrounding benzaldehyde molecules, are expected to align the H2O molecules linearly along the pore channels.A comprehensive examination of the influence of reactant concentration on the membrane reactor, COF powder, and the homogeneous Bthba catalytic system revealed that the batch reaction system was more sensitive to decreases in reactant concentration. For example, a reduction in concentration from 0.5 M to 0.1 M resulted in a 35% decrease in conversion rate in the membrane reactor, falling from 100% to 65%. In contrast, the conversions of benzaldehyde in the homogeneous and COF powder systems dropped by 45% and 41% respectively, indicating that reactants can be effectively enriched during catalysis (Supplementary Fig. 16).Investigation of the influence of H2O on the reactionWe observed a significant decline in the growth rate once specific conversion thresholds were reached: 80% for COF powder and 90% for the homogeneous Bthba catalyst. Specifically, the conversion of benzaldehyde in the COF powder reached 78% after 80 min, then increased to 84% and 91% after 648 min and 1440 min, respectively (Supplementary Fig. 17). Further extending the reaction time to 48 h only minimally enhanced the yield to 92%. Consequently, the overall TOF values for the homogeneous catalyst and the COF powder were calculated to be 0.34 min−1 (at 100% conversion) and 0.16 min−1 (at 92% conversion), respectively. These values are 176 and 375 times smaller than that of the corresponding membrane reactor. This pattern led us to hypothesize that the accumulation of by-product H2O might be inhibiting the reaction. In the membrane-catalyzed process, produced H2O molecules are instantly separated from the phase, a process enhanced by the hydrophobic and superhydrophilic structure of the composite membrane. However, in batch reactions, all the generated H2O remains in the solution. As the reaction continues, the concentration of reactants decreases, while that of by-product H2O increases, making the reaction increasingly difficult. To validate this presumption, we evaluated the effect of H2O on catalytic efficiency. The results showed that introducing 10 vol% H2O into the reaction mixture decreased the efficiency of the membrane reactor by roughly 9%, while the efficiencies of the COF powder and homogeneous catalyst systems dipped by 52% and 42% respectively. This finding reinforces the assumption that by-product water exerts a negative influence on the reaction efficiency (Fig. 4c).Molecular dynamics (MD) simulationTo assess the effectiveness of the COF membrane reactor in concentrating reactants and excluding water at a molecular level, MD simulations were performed (Supplementary Fig. 18). These simulations analyzed the diffusion dynamics of benzaldehyde, ethanol, and water through the membrane. The findings showed a consistent decrease in the transmembrane transport rates for these molecules, with water having the highest rate, followed by ethanol, and benzaldehyde the slowest (Fig. 4d and Supplementary Movie 1). This outcome was further corroborated by statistical analysis of molecule permeation across the COF membrane (Fig. 4e). Specifically, MD simulation snapshots recorded the penetration times of the first water, ethanol, and benzaldehyde molecules through the membrane as 0.09 ns, 1.28 ns, and 1.52 ns, respectively (Supplementary Fig. 19). A detailed analysis was conducted on the transmembrane behavior of these molecules through the channels of COF-TpBth membranes. It was observed that benzaldehyde molecules oscillated within the pore channels, while water molecules quickly passed through. In the transport process of water molecules, an intermittent water wire formed, which, surrounded by hydrophobic benzaldehyde molecules, arranged the water molecules into single-file, one-dimensional configurations, enhancing their movement along the pore channel (Fig. 4f)63,64. The oscillations of the side chains on the COF pore walls contributed to the transient nature of the water wire. The rapid diffusion of water molecules led to their effective exclusion from the reactor. Meanwhile, the faster diffusion of ethanol, serving as the solvent, compared to the reactant benzaldehyde, resulted in a higher concentration of the reactant within the membrane reactor. This differential behavior significantly enhanced the efficiency of the COF membrane reactor by maintaining high reactant concentrations and rapidly excluding water, thus potentially reducing its inhibitory impact on the reaction process. Furthermore, through detailed MD simulations analyzing the transmembrane transport behavior of benzaldehyde molecules, we indeed observed that upon entering the confined pore channels, the benzaldehyde molecules form intermolecular hydrogen bonds, which orient them within the channels. Additionally, these molecules also engage in hydrogen bonding with the COF backbone (Supplementary Fig. 20).The general advantage of transforming COF powder into a membrane reactorTo highlight the unique capability of promoting reactions through the transformation of COF powder into a membrane reactor, we developed an alternative membrane reactor by condensing Tp with terephthalohydrazide (Tph, Supplementary Fig. 21). This fabricated reactor (COF-TpTph/PES) revealed weakly basic sites of amide nitrogen as demonstrated by CDCl3 adsorption experiments (Supplementary Fig. 22). Under operation conditions of 0.5 M reactant concentration and 0.05 bar pressure, the reactor accomplished a 23% yield in benzaldehyde conversion. Notably, in direct comparison, the corresponding COF powder and homogeneous catalytic systems achieved conversion rates of only 4% and 7% respectively. Even more striking, when the reactant concentration was decreased to 0.1 M and the vacuum suction increased to 0.1 bar, the membrane reactor was still able to maintain an 8% benzaldehyde conversion rate, while no conversion was discernible in either the COF powder or homogeneous systems (Fig. 5a). These results underscore the ability of the COF membrane to reduce the activation energy of the reaction, likely due to the unique packing of reactants within the confined nanochannels of the membrane reactor, a process augmented by vacuum suction.Fig. 5: Comparative catalytic performance across diverse catalytic systems.a Comparative analysis of catalytic efficiency among the COF-TpTph/PES membrane, the corresponding COF powder, and a homogeneous catalyst (Tphba) derived from Tph and benzaldehyde. b Examination of the influence of vacuum suction magnitude on the catalytic efficiency of the COF-TpBth/PES membrane reactor. c Comparison between the COF-TpBth/PES membrane reactor and previously reported catalytic systems (see details in Supplementary Table 1).Long-term efficiency evaluationTo further evaluate the efficiency of the COF-TpBth/PES membrane reactor, a 48-hour continuous operation test was performed. Under optimized conditions consisting of a 0.5 M concentration of benzaldehyde and a pressure of 0.05 bar, the membrane reactor consistently achieved a conversion rate exceeding 95% for benzaldehyde, demonstrating its stable and efficacious performance over the prolonged period. Contrastingly, the homogeneous catalytic system and the COF-TpBth powder, under identical conditions, achieved lower conversion rates of 85% and 49% respectively. Moreover, to achieve a benzaldehyde conversion rate analogous to that of the membrane reactor, the homogeneous Bthba catalytic system and COF-TpBth powder necessitated substantially longer durations of 98 h and an extensive 210 h, respectively (Supplementary Fig. 23). Importantly, the integrity and the crystallinity of the membrane were maintained after catalysis (Supplementary Figs. 24 and 25).Subsequent investigations focused on the impact of pressure on the catalytic efficiency of the membrane reactor. A key finding was the observed enhancement in the TOF of benzaldehyde with increasing pressure (Fig. 5b). This improvement in performance is likely due to the increased pressure, which enlarges the diffusion rates of the reactants, solvent, and water, thereby concentrating the reactants more effectively. As a result, a significant increase in TOF was observed, rising from 60 to 185.2 min−1 as the vacuum suction intensified from 0.05 to 0.8 bar, establishing a benchmark in the field of Knoevenagel condensation (Fig. 5c and Supplementary Tables 1–3).Exploration of reaction scopeTo evaluate the adaptability of our membrane reactor across diverse substrates, we broadened our investigation to encompass a variety of substituted benzaldehydes undergoing Knoevenagel condensation with malononitrile. This exploration allowed us to further accentuate the advantages of the membrane reactor by comparing its performance with a homogeneous catalytic system. To optimize the reaction rate, we maintained a constant pressure of 0.7 bar and utilized aldehyde and malononitrile at concentrations of 0.5 M and 0.55 M, respectively. Our findings revealed that the membrane reactor efficiently catalyzes the Knoevenagel condensation of various aldehydes. We observed that electron-withdrawing groups on the benzene ring, such as in 2-fluorobenzaldehyde, afforded a greater efficiency, achieving a TOF of up to 183 min–1. Conversely, electron-donating groups tended to reduce the reaction rate, particularly when positioned at the para position of the benzene ring, as in 4-methoxybenzaldehyde, though the TOF remained impressively high. Significantly, the membrane reactor consistently outperformed the homogeneous system, with TOF values ranging from 1.45 to 2.86 times higher. It is important to note that these TOF values were determined within the initial 8 minutes. As the reactant concentration decreased and water molecules were generated, the reaction rate of the batch reaction decreased significantly. Consequently, there is expected to be a more pronounced difference in TOF values between the batch reaction and the membrane catalysis when the reactions achieve full conversion. This disparity inversely correlated with the reaction efficiency: lower efficiencies in the homogeneous system corresponded to larger improvements in the membrane reactor. This suggests that the membrane reactor can effectively lower the apparent activation energy, facilitating the reaction process. Moreover, the membrane reactor also exhibited excellent catalytic performance in the Knoevenagel condensation of heterocyclic aldehyde. In this case, the aldehyde TOF achieved was 159 min–1, which is 2.12 times higher than that of the corresponding homogeneous system (Fig. 6a). This underscores the broad applicability and efficiency of membrane reactor in various organic transformations.Fig. 6: Exploration of reaction scope.a Comparative analysis between the COF-TpBth/PES membrane reactor and the homogeneous Bthba catalyst. b Catalytic efficiency in the condensation of benzaldehyde and ethyl cyanoacetate across different catalytic systems. c Catalytic efficiency in the ring-opening reaction between styrene oxide and ethanol across different catalytic systems.To further demonstrate the efficacy of the COF-TpBth/PES membrane reactor, a comparative analysis was conducted on its performance in the condensation of benzaldehyde with ethyl cyanoacetate. The COF-TpBth/PES membrane reactor achieved a TOF of 7.0 min–1, whereas the homogeneous catalyst Bthba and the COF powder catalyst did not produce any detectable products (Fig. 6b). Additionally, the membrane reactor proved effective in the ring-opening reaction between styrene oxide and ethanol, delivering a TOF of 11.5 min–1. Under identical conditions, Bthta achieved only a TOF of 2.4 min–1, and the COF powder catalyst yielded no detectable product (Fig. 6c). These results not only highlight the advantages of the membrane reactor in lowering the activation energy and enhancing the overall efficiency of the reaction but also suggest that the COF membrane could be applicable to a broader range of base-catalyzed reactions.
