Continuous processing of sustainable mixed-matrix carbon hollow fiber precursors for ethylene/ethane separation membranes Authors: Guanran Zhao, Zhinan Fu, Zixin Chen, Xinkai Pan, Xiaoya Zhang, Linfeng Lei, Zhi Xu Published: 2026 ABSTRACT AbstractMembrane-based separations constitute energy-lean alternatives to olefin/paraffin purification, however, it remains challenging to translate sub-ångström discrimination into defect-free, large-area membranes via manufacturing-compatible routes. In this work, we realize a multiscale engineering design that strategically integrates engineered metal-organic frameworks (MOFs) with renewable, low-cost biomass-derived polymers via continuous-flow processing, followed by semi-batch carbonization into mixed-matrix carbon hollow fiber membranes. Controlled molecular entanglement imparts robust colloidal stability and interfacial cohesion, enabling thermal conversion of the monolithic precursor into contiguous inorganic-inorganic carbon nanophases. The intergrown, triphasic heterogeneous nanodomain enables fast ethylene transport and ultra-precise ethylene/ethane discrimination (C2H4/C2H6 selectivity of 75) under practical high-pressure conditions (20 bar) with long-term stability over 21 days. We also demonstrate a large-area membrane module (125 cm2) that retains attractive ethylene/ethane separation performance with an ideal single-gas C2H4/C2H6 selectivity of 51 and a mixed-gas selectivity of ~30 at 10 bar. A techno-economic analysis further confirms their cost competitiveness, highlighting module-relevant scalability and sustainability for petrochemical and broader gas separation applications. This continuous-flow hybrid strategy provides an applicable toolbox to translating sub-nanometer precision into robust, module-relevant molecular-sieving membranes.Subject terms: Chemical engineering, Chemical engineering, Nanoscale materials Introduction Energy demands from petrochemical separations are extremely intensive, and advancing their efficiency promises significant savings in both operating costs and environmental footprint1. Typically, purifying olefins from olefin/paraffin mixtures ranks among the most demanding industrial gas separations, currently achieved by cryogenic distillation that consumes 0.3% of global energy use2. Membrane-based technologies provide a promising candidate for gas separation, combining energy efficiency with scalability as an alternative or supplement to conventional distillation3–5. However, very few membrane materials could provide sufficient performance for separating molecules such as ethylene/ethane (C2H4/C2H6) with similar physicochemical properties under practical conditions, indeed, which remains a key barrier to the industrial implementation of membrane technologies6. Commercially available polymeric membranes cannot exhibit adequate selectivity, constrained by their tortuous and flexible polymer chain networks6–8. Alternatively, crystalline nanoporous membranes, such as metal-organic frameworks (MOFs) membranes, within long-range ordered lattice structures can deliver high permeability and selectivity, but their large-scale fabrication remains challenging, which are often susceptible to structural imperfections9,10. Solution-processable mixed-matrix membranes (MMMs) have therefore emerged as an attractive class of advanced membrane materials, integrating high-performance molecular sieves with the economical and mechanochemical robust polymeric fabrication11–14. By engineering low-resistance mass-transport pathways within the composite, MMMs can couple rapid permeation with molecular-level discrimination. However, the polymeric continuous phase of conventional MMMs renders the membranes prone to competitive sorption and pressure-induced plasticization under realistic operation, leading to unpredictable selectivity losses, especially exposed to high-pressure feeds of polarizable gases. Crosslinking polymer can suppress plasticization by restricting chain mobility, but it typically densifies the matrix and sacrifices gas permeability15. Thermal conversion offers an alternative to transform flexible polymer continuous phase into rigid carbon molecular sieve (CMS) networks, thereby enhancing separation efficiency and resistance to high-pressure swelling16–18. In addition, MOF-derived nanoporous carbons within membranes, featuring tunable porosity, high specific surface area, and abundant surface functionalities, may advance membrane performance, redefining the trade-off of permeability and selectivity19. However, the inorganic-inorganic interfaces (between polymer-derived and MOF-derived carbon) typically exhibit poor compatibility, making it difficult to eliminate interfacial mismatch and non-selective transport pathways20. Achieving uniform dispersion of extended filler networks within the continuous phase becomes more challenging because of the complex membrane fabrication processes. Such uncontrolled membrane-internal microstructure remains a fundamental bottleneck to scalable deployment. Herein, we present a straightforward, scalable strategy for the multiscale engineering design of free-standing mixed-matrix carbon hollow fiber membranes (Fig. 1). Enabled by controlled molecular entanglement, the continuous-flow integration of proof-of-concept engineered metal-organic frameworks (MOFs) with a renewable, low-cost biomass-derived polymer establishes robust colloidal stability and interfacial cohesion. Thermal conversion translates the monolithic structure into contiguous inorganic-inorganic carbon nanophases. The membrane features rigid, triphasic and chemically heterogeneous carbon nanodomains-locally shrinkage throats, an intermediate continuous phase, and more open gas reservoirs, enabling fast ethylene transport and atomic-level precise discrimination. Consequently, it delivers an unprecedented C2H4/C2H6 selectivity of 75, surpassing state-of-the-art membrane materials. Even under a high-pressure (20 bar) practical C2H4/C2H6 mixture feeding, the membranes exhibit plasticization-resistant, long-term stability performance over 21 days. More importantly, we demonstrate the fabrication of a large-area membrane module with 125 cm2, underscoring the module relevance and scale-up potential of this platform for practical ethylene/ethane purifications. A techno-economic analysis further confirms their cost competitiveness, supporting a viable pathway from sustainable precursors to industrial deployment. Schematic illustration of the continuous-flow route and multiscale nanostructure for mixed-matrix carbon hollow fiber membranes enabled efficient ethylene/ethane sieving. Defect-free mixed-matrix carbon hollow fiber membranes are fabricated through continuous-flow precursor fabrication followed by semi-batch carbonization, during which the ZIF-8 nanofillers are synthesized by flow-based nanotechnology and cellulose/ZIF-8 hollow fiber precursors are continuously spun using a dry-jet/wet-quench method. Unlike conventional membranes with an isoporous sieving channel, this work establishes gradient heterophase carbon nanodomains, which consist of locally shrinkage throats (dt), intermediate continuous phase (di), and more open gas reservoirs (dr) that enable rapid ethylene transport with sub-ångström discrimination under practical mixed-gas operation. The upper-right inset presents a cross-sectional image of the membranes, scale bars: 1 μm and 50 μm, respectively. And the bottom-left inset shows a digital photograph of the membranes produced at scale, scale bar: 10 cm. Results Engineering of fillers and colloidal stability The ZIF-8 particle-polymer interface was engineered through tannic acid (TA) interfacial molecular self-assembly based on metal coordination between -OH and Zn2+, which was labeled as MPN-ZIF-8. Field-emission scanning electron microscopy (FESEM) revealed that the prepared nanoparticles preserved similar shapes and sizes to the initial ZIF-8 crystals (Supplementary Figs. 1, 2). And the hollow interior could be obviously observed from the high-resolution transmission electron microscopy (HRTEM) images (Fig. 2a). The TA molecules act as a proton donor to induce lattice etching, while its bulky structure blocks ZIF pores to limit further penetration and stabilize the remaining framework (Fig. 2b)21–23. X-ray diffraction (XRD) confirmed the characteristic peak of ZIF-8 after TA treatment (Supplementary Fig. 3). A Brunauer-Emmett-Teller (BET) specific surface area of 1581 m2 g−1 was obtained for ZIF-8. Conversely, MPN-ZIF-8 reduced to 387 m2 g−1, which was attributed to the elimination of the inner mesopore region (Supplementary Fig. 4a). According to the density functional theory (DFT) model, the N2-based pore size distribution of MPN-ZIF-8 represented a lower proportion of micropores (Supplementary Fig. 5a). Interestingly, using CO2 as the smaller molecular probe, the surface-attached metal-polyphenol network, with its intrinsic microporosity, contributed additional ultramicroporosity (< 7 Å) to the system (Supplementary Figs. 4b, 5b). A higher population of ultramicropores within ZIFs might promote sharper molecular sieving capacity in the derived carbon membranes. Preparation and characterization of MOF nanoparticles and mechanism of the enhanced colloidal stability. a HRTEM image of MPN-ZIF-8, scale bar: 200 nm. b Schematic illustration for the structural evolution process of the ZIF-8 to MPN-ZIF-8. c Zeta potential of ZIF-8 and MPN-ZIF-8 in deionized water (the inset image shows the contact angle). d DLS measurements of ZIF-8 and MPN-ZIF-8 in EmimAc solvents. e Photographs of ZIF-8 and (f) MPN-ZIF-8 with equal volume fraction dispersed in DMSO and EmimAc solvents for 24 h. g The adsorption amount comparison of ZIF-8 and MPN-ZIF-8 crystals to EmimAc molecules based on the QCM-D technique. h van der Waals and electrostatic interactions between ZIF-8/MPN-ZIF-8 and EmimAc. i The calculated RDF and coordination numbers in ZIF-8/EmimAc system and MPN-ZIF-8/EmimAc system for Zn-N. j 1H time-domain nuclear magnetic resonance spectra of pristine, MMMs and pMMMs precursor membranes. k DFT results of the interactions of cellulose-ZIF-8 and cellulose-MPN-ZIF-8 composite. l IGMH isosurface for the visualization of noncovalent interaction for cellulose and MPN-ZIF-8 complexes in a spatial region. Green and blue isosurfaces indicate van der Waals and hydrogen bonding interactions, respectively. The FTIR characteristic peaks of TA at 3300 cm−1 (-OH) and ZIF-8 at 420 cm−1 (Zn-N) confirmed the composite chemistry of MPN-ZIF-8 (Supplementary Fig. 6). X-ray photoelectron spectroscopy (XPS) further revealed a stronger oxygen atomic content in MPN-ZIF-8, attributable to TA-derived phenolic hydroxyl (Supplementary Fig. 7 and Table 1). Surface enrichment with polyphenolic hydroxyl raised interfacial energy, rendering the MOF surface from hydrophobic to hydrophilic while reversing the zeta potential from positive to negative (Fig. 2c, Supplementary Figs. 8, 9 and Table 2). Such a high density of interfacial uncoordinated phenolic hydroxyls and aromatic rings in the TA-stabilized dispersion of MPN-ZIF-8 within the high-viscosity spinning solution. Dynamic light scattering (DLS) confirmed ~ 200 nm MPN-ZIF-8 nanoparticles in EmimAc (viscosity of 139 mPa s) and DMSO (1.996 mPa s), matching TEM and SEM analyses (Fig. 2d and Supplementary Fig. 10). Conversely, pristine ZIF-8 displayed a broad size distribution beyond its ~ 200 nm original size in both solvents. With a high concentration dispersion (100 mg L−1) in both solvents, MPN-ZIF-8 remained uniformly suspended, while ZIF-8 was sedimented within 24 h (Fig. 2e, f). To further probe filler-solvent intermolecular forces, the adsorption behaviors of EmimAc molecules through testing the pristine ZIF-8 and MPN-ZIF-8 were quantitatively detected based on quartz crystal microbalance with dissipation (QCM-D) technique (Fig. 2g and Supplementary Fig. 11). MPN-ZIF-8 indicated an increased frequency decrease (|Δf | ) upon EmimAc exposure than pristine ZIF-8, implying the formation of a thicker, more solvent-rich adsorption adlayer24–26. In addition, a classical molecular dynamics (MD) simulation was performed to evaluate the essence of enhanced interaction relationships at the atomic level (Supplementary Figs. 12–14). At the MPN-ZIF-8 interface, van der Waals and electrostatic interactions were synergistically strengthened, driven by hydrogen bonding between phenolic-OH groups and acetate anions, cation π-π interactions between imidazolium rings and polyphenolic network, and the attraction of positively charged imidazolium cations to the negatively shifted MPN surface (Fig. 2h). The primary Zn-solvent shell is identified by the first peak in the radial distribution functions (RDF), and the coordination number is evaluated by integrating the RDF up to the first minimum following that peak27. Correspondingly, MPN-ZIF-8 exhibited a higher and sharper first-shell Zn-N peak together with an elevated coordination number relative to pristine ZIF-8 in EmimAc, evidencing a tighter first solvation shell of imidazolium cations adjacent to surface Zn nodes (Fig. 2i). Therefore, the strongly bound solvation shell suppressed particle-particle attractions under high-viscosity conditions and resulted in the improved nanoparticles dispersion. Membrane nanostructures and chemical characteristics ZIF-8 and MPN-ZIF-8 crystals were incorporated into a cellulose matrix to fabricate mixed-matrix hollow fiber membranes by a dry-jet/wet-quench spinning method (Supplementary Table 3 lists the spinning parameters). Microcrystalline cellulose (with a DP of 151.9 and Mw of 40,482 Da, Supplementary Fig. 15) was selected as the precursor matrix because: (i) renewable and low-cost biomass-derived polymer, making it a sustainable feedstock for scalable membrane manufacturing; (ii) rich in hydroxyl groups that can serve as abundant polar anchoring sites for interfacial interactions with MOF fillers; (iii) semi-crystalline with highly ordered short-range packing, which can be retained in part during thermal transformation into a rigid carbon phase with strong molecular-sieving capability; and (iv) readily processable in the DMSO/EmimAc solvent system, allowing the formation of a homogeneous spinning dope for continuous dry-jet/wet-quench hollow-fiber spinning. The corresponding membranes with identical filler mass loading amount were denoted as MMMs and pMMMs (Supplementary Fig. 16). The phenolic-OH groups of MPN layers were supposed to form stronger hydrogen bonds with cellulose, as evidenced by a red shift and broadening of the O-H stretching band from 3333 to 3320 cm−1 in pMMMs compared to MMMs (Supplementary Fig. 17). Moreover, the band at 1155 cm−1, assigned to the asymmetric aromatic C-O stretch, also showed a slight blue shift, attributable to interactions with hydrogen donors. These characteristics indicated that the MPN interphase introduced abundant hydrogen bonds and increased the heterogeneity of the H-bonding environment. Low field-nuclear magnetic resonance (LF-NMR) measurements of hydrogen spectra (H 1 s) further elucidated the organization of hydrogen-bond networks in precursor membranes28,29 (Fig. 2j). Relative to pristine membranes, pMMMs displayed an increased integrated peak area, indicating an increased H nucleus population attributable to stronger TA-cellulose interactions. In contrast, the incorporation of pristine ZIF-8 crystals decreased the peak area. Considering the pivotal role of hydrogen-bond topology packing of precursors in the separation performance of the resulting carbon membranes, we further examined the stacking state of cellulose molecular chains by wide-angle X-ray scattering (WAXS) (Supplementary Fig. 18). pMMMs exhibited an enhanced peak intensity of 1.43 Å−1 in the amorphous region by the 1D WAXS images, evidencing tighter interchain packing and enhanced short-range order30. In addition, the 2D WAXS patterns demonstrated more intense and sharper Debye-Scherrer rings, reflecting densification of the cellulose network, however, the MMMs showed the reverse trend. DFT calculations were performed to confirm the molecular interactions of cellulose with ZIF-8 and MPN-ZIF-8 (Supplementary Figs. 19, 20). The calculated interaction energies demonstrated that cellulose bound far more strongly to MPN-ZIF-8 (− 115.3 kcal mol−1) than pristine ZIF-8 (− 44.9 kcal mol−1) (Fig. 2k). Surface electrostatic potential (ESP) calculations revealed a more negative potential on MPN domains arising from phenolate O atoms and delocalized π electrons within the aromatic rings, which promoted directional interactions with cellulose -OH groups (Supplementary Fig. 21). Noncovalent interactions using the independent-gradient-model-based-on-Hirshfeld (IGMH) visualized contiguous attractive regions at the cellulose/MPN interface, evidencing cooperative reinforcement of hydrogen bonding and van der Waals interactions across the interphase (Fig. 2l and Supplementary Figs. 22, 23). These calculations aligned with the molecular characterizations above and validated the proposed interaction design in the cellulose-MPN-ZIF-8 unit. The MPN interphase acted as a supramolecular crosslinker, which promoted a synergistic enhancement in both bulk cohesion and interfacial adhesion through electrostatic anchoring and hydrogen-bond interlocking. The precursor membranes were converted to mixed-matrix carbon membranes (MMCMs) via a specific carbonization temperature (Supplementary Figs. 24, 25). The corresponding membranes were denoted as MMCMsTc and pMMCMsTc, where TC represented the carbonization temperature. Following thermal processing under an identical temperature-programmed protocol, both ZIF-8 and MPN-ZIF-8 maintain their framework31, verified by HRTEM images (Supplementary Fig. 26). As shown by cross-sectional FESEM (Fig. 3a, b), the poor interfacial compatibility between the evolving carbon matrix and pristine ZIF-8, together with weak particle adhesion, generated abundant interfacial defects and nonselective bypass pathways in MMCMs. However, the MPN-mediated interphase enabled pMMCMs to maintain a uniform-dispersed, defect-free morphology, consistent with the structural model proposed in Fig. 1. The HRTEM image further revealed the microstructure featuring a coherent triphasic nanodomain arrangement composed of ZIF-derived, MPN-derived, and cellulose-derived phases (Fig. 3c). Crystalline nanodomains with clear lattice-fringe contrast are observed and assigned to ZIF-derived regions, which mainly exist as ZnO-containing domains after pyrolysis. A thin carbonaceous boundary layer can be identified around the ZIF-derived crystalline domains and is assigned to the MPN-derived interphase, consistent with the localization of MPN between ZIF-8 and cellulose in the precursor membrane, where it serves as an interfacial transition layer after pyrolysis. This boundary layer shows weaker lattice contrast than the ZIF-derived domains but more locally aligned, turbostratic graphite-like lamellae than the distant cellulose-derived carbon matrix with disordered amorphous characters. Such enhanced short-range ordering indicates the formation of a more densified and structurally coherent interphase between the inorganic ZIF-derived domains and the surrounding cellulose-derived carbon matrix. The interfacial stitching is anticipated to reduce nonselective transport pathways and thereby enhance the molecular-sieving performance of the hybrid carbon membranes. XPS analysis was used to track the evolution of interfacial chemical bondings32. Deconvolution of the C 1 s/O 1 s spectra showed an enhanced ratio of sp2-hybridized carbon (related to graphitic hexagonal-ring) to sp3-hybridized carbon (related to edge/defect sites), together with an increased C-O-C (ether) intensity (Supplementary Fig. 27). These trends indicated that the stronger hydrogen-bond network in the precursors promoted dehydration/condensation during pyrolysis, converting labile -OH groups into ether linkages and driving aromatization into graphite-like nanodomains. Furthermore, the average interlayer spacing can be calculated from the WAXD patterns using Bragg’s equation33 (Supplementary Fig. 28). Relative to the pristine carbon membranes, MMCMs exhibited a larger d-spacing of 3.95 Å, due to additional ZIF-derived carbons within highly disordered microporosity. However, pMMCMs showed a decreased d-spacing of 3.79 Å, indicating tighter carbon plane packing mediated by the MPN interphase, which is further supported by WAXS analysis (Supplementary Figs. 29, 30). Chemistry and nanodomains evolution of membranes. a Cross-sectional FESEM images of MMCMs and (b) pMMCMs, scale bar: 1 μm; inset image presents a low-magnification cross-sectional morphology of the hollow fiber membrane, scale bar: 50 μm. c HRTEM image of pMMCMs, scale bar: 5 nm. Specifically, the green and red regions represent short-range ordered nanodomains and structurally disordered nanodomains within the carbon matrix, respectively. d Raman mapping images (5 × 5 µm2) of pristine carbon membranes, MMCMs and pMMCMs. Areas in blue and red correspond to higher (disorder) and lower ID/IG (local order) values, respectively. e AFM 3D topography projection of pMMCMs. f Nano-FTIR spectra of locations marked in the 3D topography projection image. g Hypothetical pore structure of pMMCMs with triphasic heterogeneous nanodomains. h Schematic illustration of conventional carbon membranes with an isoporous sieving channel and (i) pMMCMs with triphasic heterogeneous sieving channels. j Pore size distribution of pristine carbon membranes, MMCMs and pMMCMs based on non-local density functional theory using CO2 isotherms at 273 K. k C2H4 and (l) C2H6 adsorption isotherms of pristine carbon membranes, MMCMs and pMMCMs at 298 K. We further elucidated the microstructural properties of hybrid membranes using Raman mapping (5 × 5 µm2) to visualize spatial variations in the ID/IG ratio across each membrane, which the D band is associated with disordered or defect-rich carbon structures and the G band is related to graphite-like sp2 carbon domains34 (Fig. 3d). Unlike the conventional point Raman spectra, which only provide localized or averaged information from selected positions and may overlook spatial heterogeneity. MMCMs displayed a higher proportion of defective carbons compared to that of pristine carbon membranes. Notably, pMMCMs exhibit more continuous low-ID/IG regions (ordering MPN-derived molecular sieving nanodomains) that preferentially surround isolated high-ID/IG regions, which are generally associated with ZIF-derived defective carbon domains. Such MPN interphase reinforced more extensively interconnected ordered sp2 networks around the filler-derived domains35. A nano-FTIR spectroscopy technique was employed to access local chemical information of the pMMCMs at the nanodomain level36, which cannot be resolved by conventional macroscopic FTIR spectra (Supplementary Fig. 31). The AFM-correlated 3D topography reveals pronounced protrusions assigned to ZIF-derived domains, adjacent regions assigned to the MPN-derived interphase, and distant regions assigned to the cellulose-derived carbon matrix (Fig. 3e). At the ZIF-derived domains, IR vibrational features are nearly absent, consistent with their inorganic ZnO-containing character. Interestingly, IR spectra along particles (MPN-derived domains) exhibited an increased intensity and red shift of the C-O-C stretching band from 1120 cm−1 to 1109 cm−1 relative to regions far from the particles (cellulose-derived domains), indicated the more conversion of hydrogen bond into ether linkages within the MPN-derived boundary layer during pyrolysis (Fig. 3f). In addition, the band at 1520 cm−1, assigned to the aromatic C = C stretch, also exhibited an enhanced intensity of interphase layer consistent with locally densified sp2 domains. Therefore, the MPN layer mediated interfacial rich -OH into ether conversion, which chemically anchored the filler within the evolving carbon matrix, thereby minimized non-selective interfacial defects during pyrolysis. Such a triphasic, intergrown microstructure defined a hierarchical gas transport network that decoupled selectivity with permeability, as hypothesized in Fig. 3g: (i) narrower ultramicropore size constrictions at the MPN-derived interphase (“shrinkage throats”, dt) impose more precise size-sieving; (ii) intermediate pores in the cellulose-derived carbon matrix (di); and (iii) more open ultramicroporous/mesoporous reservoirs within ZIF-derived carbons and the inner hollow cavity (dr) reduced transportation resistance. Accordingly, Fig. 3h, i compares conventional carbon membranes with relatively isoporous sieving channels and pMMCMs with triphasic heterogeneous sieving channels, respectively. This cascade of pore sizes (dt < di < dr) is hypothesized to promote a stronger molecular sieving capability, and a lower permeation resistance and might offer a general route to break the permeability-selectivity limitations of single-phase carbon membranes. We further employed CO2 as a probe molecule to resolve the structure evolution in hybrid carbon membranes37 (Supplementary Fig. 32). Compared to pristine carbon membranes, the pMMCMs displayed an enhanced ultramicroporosity, whereas the MMCMs shifted toward larger pores, consistent with increased average interlayer spacings and less constrained sieving windows (Fig. 3j and Supplementary Fig. 33). Specifically, pMMCMs exhibited a higher CO2-accessible surface area arising from a more developed ultramicropore population, while MMCMs showed an elevated pore volume attributable to TA-induced ZIF etching (Supplementary Fig. 34 and Table 4). These results confirmed that the MPN interphase induced sp2-rich domains and shrinkage-throat constrictions at the interface. Pyrolysis temperature (TC) provided an additional handle to tune the carbon microstructure. Elevation of TC led to the tighter carbon chains packing, a narrowed distribution of ultramicropores and a higher pore surface area. This trend was further confirmed by C 1 s XPS deconvolution, which reveals an increase in the sp2/sp3 ratio, reaching 2.86 at 800 °C, indicating enhanced aromatization and more compact carbon nanodomains (Supplementary Fig. 35). Interestingly, CO2 uptake increased monotonically with TC (for instance, from 63.1 to 73.6 cm3 g−1 as the TC elevated from 700 to 800 °C), whereas C2H4 adsorption peaked at 750 °C and then declined, and C2H6 adsorption decreased markedly with increasing TC (Fig. 3k, l). At moderate TC, interlayer spacing constrictions produced sub-nanometer pores that still accommodate C2H4 while excluding larger paraffins. With further temperature increased, continued carbon framework densification suppressed accessible free volume and diffusion pathways. In summary, the sharpening of sub-Ångström size-exclusion “cut-off” effects well-developed carbon ultramicroporosity and maximized gas pairs selectivity38 (Supplementary Fig. 36). Furthermore, the as-prepared pMMCMs exhibited appreciable mechanical flexibility and could be bent to a minimum radius of approximately 2 cm (Supplementary Fig. 37). The engineered MPN interphase increased the Young’s modulus from 4.48 to 5.44 GPa, reflecting enhanced interfacial cohesion and the formation of a more rigid hybrid carbon framework (Supplementary Fig. 38 and Table 5). In contrast, the reduced mechanical robustness of MMCMs is attributed to interfacial defects arising from poor filler-matrix compatibility. Moreover, the Young’s modulus progressively increased with TC, consistent with framework densification and enhanced rigidity during pyrolysis. Membrane separation performance To probe the transport properties of our mixed-matrix carbon hollow fiber membranes, we investigated single gas permeation with an industrially important gas pair (C2H4/C2H6) at 25 °C and 2 bar (Supplementary Figs. 39, 40). As expected, introducing the MPN-ZIF interphase increased C2H4 permeability from 5.6 to 41.0 Barrer (ca. 7.3-fold) and C2H4/C2H6 selectivity from 7.8 to 11.5 (ca. 1.5-fold), relative to the pristine carbon membrane (Fig. 4a and Supplementary Fig. 41). By contrast, MMCMs only exhibited higher permeability but lower selectivity, consistent with the non-selective interfacial diffusion typical of conventional MMMs39,40, as evidenced in the cross-sectional SEM (Fig. 3a). In the ZIF-only case, enhanced free volume and interfacial mismatch broadened the pore size distribution and introduce fast bypass routes, which raised permeability at the expense of gas pairs selectivity. By contrast, the MPN interphase acted as a molecular solder that chemically bridged the carbon matrix and ZIF-derived domains, suppressing nonselective interfacial voids and confining gas transport to a narrower distribution of pore throats. Furthermore, the ZIF-derived carbon and additional hollow structure provided increased pore reservoirs, which promoted gas transport. Further increasing the TC of the pMMCMs sharpened the molecular sieving. At 800 °C, the pore size distribution shifted towards smaller ultramicropores, consistent with enhanced ethylene/ethane molecular sieving. Particularly, the pMMCMs800 demonstrated a C2H4/C2H6 selectivity of 75, one of the highest values reported in the literature (Fig. 4b). The clear permeability cut-off between smaller (i. e., H2, CO2) and larger gases (i. e., CH4, C2H6) revealed kinetic diameter dominated transport mechanism of molecular sieving41 (Supplementary Fig. 42). The higher permeabilities of H2 over He and C2H4 over N2, suggest additional contributions from molecular shape and penetrant-pore walls interactions. In addition, attractive O2/N2, CO2/CH4, H2/CH4 and H2/C2H6 separation selectivity were also achieved, making them promising for applications in other gas separations. Compared to the reported state-of-the-art polymer membranes, mixed-matrix membranes, MOF membranes, and CMS membranes, pMMCMs800 membranes in this work exhibited more attractive C2H4/C2H6 separation performance, establishing such mixed-matrix carbon hollow fiber membranes as a practical and competitive platform for petrochemical gas separation (Fig. 4c and Supplementary Table 6). Separation performance of mixed-matrix carbon hollow fiber membranes. a Gas permeability and ideal selectivity in membranes derived from pristine membrane, MMMs and pMMMs precursor at 25 °C and 2 bar. The modules (each made with 3 fibers) were tested for single-gas permeation. Error bars represent the standard deviations obtained from three membranes. b Effects of carbonization temperature of pMMCMs on the ethylene/ethane separation performance. The modules (each made with 3 fibers) were tested for single-gas permeation. Error bars represent the standard deviations obtained from three membranes. c Comparison of the single-gas ethylene/ethane separation performance of state-of-the-art polymer membranes, mixed-matrix membranes, MOF membranes and CMS membranes. The corresponding raw data are provided in Supplementary Tables 6 and 8. d Diffusivity and (e) solubility of ethylene/ethane for pristine CMS membranes, MMCMs and pMMCMs. f Diffusivity selectivity and solubility selectivity of ethylene/ethane for pristine carbon membranes, MMCMs and pMMCMs. g Long-term equimolar mixed-gas performance as a function of total feed pressure varying from 10 to 20 bar for the pMMCMs800. The modules (each made with 10 fibers; membrane area was at least 5 cm2) were tested for mixed-gas permeation. Gas transport in carbon membranes is conventionally described by the solution-diffusion, in which permeability decouples into diffusivity and solubility (Fig. 4d–f and Supplementary Table 7). With incorporating ZIF or MPN-ZIF crystals, the permeability increase was primarily driven by increases in DC2H4. However, the diffusion selectivity declined in MMCMs relative to an increase in pMMCMs. As the TC of pMMCMs rose, the drop in DC2H6 was much higher than for DC2H4, leading to a sharp increase in diffusion selectivity. Simultaneously, the solubility coefficient at higher TC was also strongly suppressed, causing the solubility selectivity to rise from ~ 1.1 at 700 °C to ~ 3.1 at 800 °C, which has also been reported by Koros42 and Pinnau17. We hypothesized that narrowing pore throats disfavored the bulkier C2H6, and that thermally induced loss of surface functional groups reduced adsorption affinity. For stringent and challenging C2 separations, a central barrier to membrane deployment lies in their lack of long-term stability under practical applications. However, such high-pressure mixed-gas evaluations are still scarcely reported, yet they more faithfully emulate the operating conditions that these membranes must withstand in practice. Accordingly, we evaluated C2H4/C2H6 separation for pMMCMs pyrolyzed at 800 °C using an equimolar feed at total pressures of 10, 15 and 20 bar up to 21 days (Fig. 4g, Supplementary Fig. 43 and Table 8). Even under harsh binary-gas conditions, pMMCMs800 still demonstrated promising separation performance, which far exceeded most reported membranes. At a total feed pressure of 10 bar, the prepared membrane performed mixed-gas C2H4/C2H6 selectivity of ~ 57. Further increasing the total pressure to 15 and 20 bar lowered the selectivity to the range of ~40-50, which was similar with the pressure-dependent trends widely reported for carbon-based membranes17,43. Under high-pressure olefin separation conditions, ethylene sorption can enhance polymer-chain mobility in polymer membranes, leading to a sharp increase in ethylene permeation-a phenomenon commonly referred to as plasticization. By contrast, the prepared pMMCMs800 showed no such pressure-induced increase in ethylene permeability, consistent with its rigid hierarchical pore structure rather than the flexible segment relaxation (Supplementary Fig. 44). The C2H4/C2H6 selectivity decline was governed primarily by a reduction in C2H4 permeability, presumably arising from competitive sorption under mixed-gas operation. Membrane scalable fabrication and economic feasibility To enable scalable membrane fabrication, the continuous and controllable scale-up of MOF synthesis is essential. However, conventional batch methods are often time-consuming and challenging to scale, and these syntheses typically provide limited control over the resulting MOFs structure and often cause poor dispersity, and relatively low productivity, and batch-to-batch inconsistency. In this work, we demonstrate a proof-of-principle investigation of the efficient, controllable and continuous manufacture of ZIF-8 nanoparticles using a flow-based nanotechnology. The schematic diagram of the continuous-flow system is presented for synthesis, along with photographs of the four-stream micromixer (Fig. 5a and Supplementary Fig. 45). The microchannel structure of the micromixer facilitated vortex-intensified mixing of reactants containing zinc nitrate and ligands within milliseconds (~10 ms) and minimized molecular diffusion distances. Benefiting from the enhanced mass transfer efficiency in microscale channels, the reaction kinetics were remarkably accelerated, which greatly shortened the nucleation and crystal growth period of ZIF-8 crystals, thereby allowing for the prepared ZIF-8 with highly uniform morphology (Supplementary Fig. 46). The output obtained in continuous-flow synthesis was 36 g h−1 in terms of the total weight of ZIF-8 powder under the high flow rates, representing the high productivity (Fig. 5b). Furthermore, a major highlight of this flow-based technique is the ability to effectively downsize the ZIF-8 nanoparticles through flow condition control (Fig. 5c and Supplementary Table 9), as it promotes the formation and dispersion of small-sized nuclei in a short time. In contrast to conventional batch synthesis, the flow-based platform allowed precise control of fluid volumes in microchannels, thereby achieving accurate regulation of mixing time and efficiency, and ultimately realizing the controllable growth of ZIF-8 nanoparticles. Followed by a simple TA modification, MPN-ZIF-8 can be produced in a batch. Building on this continuous MOF supply, we further translated the above protocol into a module-relevant manufacturing workflow for hollow fiber pMMMs by employing renewable cellulose as a sustainable, low-cost feedstock (Fig. 5b). After MPN engineering, the continuously produced ZIF-8 nanoparticles were effectively incorporated into the cellulose spinning dope without compromising its processability. In the current laboratory-scale spinning line, the pMMMs production rate was approximately 450 m h−1 for a single spinning line. Through a continuous dry-jet/wet-quench spinning line, we validated the scalable manufacturing of at least 2000 mixed-matrix carbon hollow-fiber membranes through subsequent semi-batch carbonization (Fig. 5d, e and Supplementary Fig. 47). We then assembled a 125 cm2 membrane module (100 fibers of membrane) (Fig. 5f and Supplementary Fig. 48). Importantly, the resulting module preserved attractive separation performance. It showed an ideal single-gas selectivity of 51 and a mixed-gas C2H4/C2H6 selectivity of ~ 30 at 10 bar (Fig. 5g), confirming no macroscopic defects during fabrication and the uniformity and reproducibility of the resulting membrane materials. Compared with other laboratory-scale membrane samples (≤ 1 cm2), the module-scale membrane still delivers separation performance that surpasses that of most state-of-the-art membranes reported to date. Large area membrane fabrication and techno-economic analysis. a Digital photograph of a continuous-flow synthesis system. Specifically, stream 1 contains the zinc nitrate solution, streams 2 and 3 contain the 2-methylimidazole solution, and stream 4 is methanol. These streams are combined in the four-stream micromixer to enable rapid and continuous ZIF-8 nucleation and growth. b Digital photograph of the prepared MPN-ZIF-8, cellulose, and spinning dope. c HRTEM images of ZIF-8 crystals with different size, scale bar: 100 nm. The ZIF-8 particle sizes were tuned by controlling the flow conditions during the flow-based synthesis. d Digital photograph of a scalable dry-jet/wet-quench hollow fiber spinning line. e Digital photograph of pMMMs and pMMCMs. f Digital photograph of a large-area membrane module of 125 cm2. g Single-gas separation performance and equimolar mixed-gas of 10 bar separation performance of 125 cm2 membrane module, compared to state-of-the-art polymer membranes, mixed-matrix membranes and CMS membranes. The corresponding raw data are provided in Supplementary Tables 6 and 8. h Comparison of production costs of our pMMCMs and other reported carbon membranes derived from petrochemical precursors, such as PIM−1 and polyimides (Matrimid® 5218). A techno-economic analysis (TEA) indicates that pMMCMs can be manufactured at a lower unit cost than membranes derived from conventional petrochemical precursors, by 58.4% relative to PIM-1 and by 53.5% relative to representative polyimides (for example, Matrimid® 5218). (Fig. 5h and Supplementary Tables 10–12). Enabled by the availability of biomass-derived feedstocks, the low-cost cellulose contributes only ~ 3% of the raw-material cost (Supplementary Fig. 49). The main cost factors instead originate from the ionic-liquid solvent and methanol (~ 40.5%), importantly, both of which are recyclable, further supporting the low-cost potential of this manufacturing route. This breakthrough highlights the transformative potential of the pMMCMs approach for membrane fabrication, delivering enhanced separation efficiency and providing a critical step toward overcoming the persistent barriers to industrial deployment of membrane-based gas separations. Discussion The proposed approach of continuous-flow processing makes it possible to fabricate defect-free, large-area hybrid hollow fiber membranes for highly selective and stable discrimination of olefin/paraffin under real process conditions. Rational matching of engineered MOFs with polymeric precursors enables the construction of rigid, multilevel sub-nanometer transport channels with tailored gas-transport properties. More broadly, this work establishes a molecular-to-module pathway for energy-relevant separations, accelerating the deployment of high-performance membrane materials in industrially demanded separations. Future work will be directed toward advancing manufacturing protocols that yield ultrathin selective layers and large-area membrane modules. Methods Materials Microcrystalline cellulose (MCC) was purchased from Sigma-Aldrich (Avicel PH-101). Dimethyl sulfoxide (DMSO, 99%), ethanol (99.9%) and methanol (> 99.7%) were purchased from Greagent. 1-Ethyl-3-methylimidazolium acetate (EmimAc, 98%) was procured from Lanzhou Green Chemistry. Zinc nitrate hexahydrate (Zn(NO3)2·6H2O, > 99%) was purchased from Shanghai Lingfeng Chemical Reagent. 2-methylimidazole (2-MI, 99%) was purchased from Adamas. Tannic acid (TA, 99%) was purchased from Aladdin. The high-purity gas (Ar, 99.9995%) used as the protective gas during the carbonization process was provided by Air Liquide, Shanghai. Preparation of ZIF-8 crystals ZIF-8 nanoparticles were synthesized by dissolving 10.71 g of Zn(NO3)2·6H2O in 360 mL of a mixed solvent consisting of deionized water and methanol (1:1, v/v), followed by the addition of a solution of 9.85 g of 2-methylimidazole (2-MI) in 120 mL of methanol. The mixture was sonicated at room temperature for 30 min, and the resulting ZIF-8 nanocrystals were collected by centrifugation at 21912.8 g, washed with ethanol several times, and finally vacuum-dried at 40 °C overnight. Preparation of MPN-ZIF-8 crystals The as-prepared ZIF-8 nanoparticles (3.0 g) were first dispersed in 1.57 L of methanol. This suspension was then added to 0.525 L of an aqueous tannic acid solution (5 g L−1) under stirring. After stirring at room temperature for 10 minutes, the product was collected by centrifugation at 21912.8 × g, washed with ethanol several times, and finally vacuum-dried at 40 °C overnight. Preparation of ZIF-8 crystals by a continuous-flow nanotechnology ZIF-8 nanoparticles were synthesized using a continuous-flow technique with a four-stream micromixer. The four-stream micromixer is equipped with feeding channels of 1.1 mm in height and 1.5 mm in width, which converge into a vortex-shaped confined chamber with a diameter of 6.0 mm to intensify fluid mixing. Briefly, an aqueous solution of Zn(NO3)2·6H2O (0.2 M) was prepared as stream 1. Two separate ethanolic solutions of 2-MI (1.0 M) were used as streams 2 and 3, while pure methanol served as stream 4. All four streams were delivered by two digitally controlled syringe pumps (Harvard Apparatus, PHD 2000). Specifically, streams 1 and 4 were injected via one pump connected to inlets 1 and 4, respectively, while streams 2 and 3 were injected via another pump connected to inlets 2 and 3, respectively. The resulting suspension containing ZIF-8 nanoparticles was continuously collected from the outlet of the micromixer. The precipitate product was then obtained by centrifugation at 21912.8 g, washed several times with ethanol, and vacuum-dried at 40 °C overnight. The flow rates of streams in the micromixer were continuously adjusted to obtain ZIF-8 nanoparticles with different particle sizes. Preparation of mixed-matrix cellulose hollow fiber membranes A cellulose dope was prepared using DMSO and EmimAc as solvents (12/66/22 wt% for cellulose/DMSO/EmimAc). ZIF-8/MPN-ZIF-8, with 10% of the total polymer precursor mass, was then incorporated to prepare mixed-matrix membranes. ZIF-8 or MPN-ZIF-8 was first dispersed in DMSO and ultrasonicated to obtain a homogeneous suspension. Then cellulose was dispersed in DMSO to clarity, after which EmimAc was added. Lastly, the mixture was roll-mixed at 60 °C and 60 rpm for 12 h to ensure complete homogenization. Mixed-matrix cellulose hollow fibers were fabricated by dry-jet/wet-quench spinning through a coaxial (double-layer) spinneret (ID/OD = 0.5/0.8 mm) with deionized water as the bore fluid. During the spinning process, the fluid was maintained at 25 °C. The dope was delivered by a metering pump (6 rpm) and extruded across a 4 cm air gap, the bore fluid flow was 1.5 mL min−1. Coagulation occurred in a water bath (temperature was 25 °C), after which fibers were guided into a washing tank and collected on a take-up reel (line speed 7.5 m min−1; gel line 7.8 m min−1; winding reel 3.8 rpm). The as-spun membranes were then soaked in deionized water for three days to remove residual solvent. Preparation of carbon hollow fiber membranes Precursor hollow fibers were carbonized in a tube furnace under a continuous flow of ultrahigh-purity argon (Ar, 99.999%) at 70 mL min−1. Before introducing Ar, the furnace was evacuated to ~3 mbar for 2 h, and then underwent a specific carbonization procedure. After carbonization, the furnace was allowed to cool under Ar to <50 °C. Fresh carbon hollow fibers were then sealed immediately into stainless-steel housings using an epoxy (Araldite 2012) and terminated with tee connectors (X-tec, Shanghai Xitai Fluid Technology Co., Ltd.). Prior to module fabrication, hollow fibers with no visible defects were selected and cut to the desired length. It should be noted that, in the small-scale module tests, three hollow fibers were assembled in a 1/4-inch tube for single-gas permeation measurements, while ten hollow fibers were assembled in a 1/4-inch tube for mixed-gas separation measurements. In the large-area membrane module, 100 hollow fibers were packed into a 1/2-inch tube, which was connected to the gas permeation testing apparatus through a 1/2-inch-to-1/4-inch adapter. The detailed mixed-gas module potting configuration is schematically illustrated in the Supplementary Fig. 47. Epoxy resin was carefully introduced at both ends of the module to seal the shell side and mechanically fix the hollow fibers in place. During potting, we carefully ensure complete infiltration of the epoxy around the outside of the fiber bundle while avoiding penetration of epoxy into the fiber bores. The epoxy was cured at room temperature for 24 h. After curing, the excess fiber and epoxy at both ends of the module were carefully trimmed to expose open fiber bores and to ensure a clean cross-section for gas flow through the feed side. Before gas permeation measurements, each module was connected with appropriate 1/4 inch fittings and tested for leakage by pressurizing the feed side with N2 to 40 bar and monitoring the pressure stability. Modules that showed no pressure leakage were used for gas permeation tests. Gas separation performance After sealing by epoxy for 2 hours, single-gas permeation of H2, He, CO2, N2, CH4, C2H4 and C2H6 through the carbon hollow fiber membranes was measured using the constant-volume method. Three modules (each made with 3 fibers) were tested for single-gas permeation at each condition. Prior to testing, modules were vacuumed for at least 12 h, and the feed side was then pressurized to 2 bar at 25 °C. Gas permeability, P (Barrer), was calculated from Eq. (1), and ideal selectivity from Eq. (2): Here, V (cm3) is the calibrated permeate-side volume, l (cm) is the membrane thickness, A (cm²) is the effective membrane area, T (K) is the measurement temperature, Δp = pF − p is the trans-membrane pressure difference (bar) with pF and p represents the feed and permeate pressures, respectively, and dpdtsteady the steady-state rate of permeate-side pressure rise (cmHg s−1), which is at least one order of magnitude higher than the leak rate dpleakdt of the system. Mixed-gas performance was assessed with a 50/50 mol.% C2H4/C2H6 feed in a flow configuration at 10, 15, 20 bar and 25 °C. The feed flow was fixed at 20 mL min−1. Argon was used as a counter-current sweep on the bore side at 5 mL min−1; the mixed gas was introduced on the shell side. The modules (each made with 10 fibers; membrane area was at least 5 cm2) were tested for mixed-gas permeation. In addition, the large-area membrane modules of 125 cm2 with 100 fibers were tested for gas permeation. Permeate flow was tracked with a bubble flow meter, and composition was quantified by gas chromatography (GC-2014). The separation factor was obtained from the permeate and feed compositions via: where yi and xi denote the mole fractions of component i in the permeate and feed, respectively. Gas permeability (P, Barrer) is the product of diffusivity (D) and sorption coefficient (S). Therefore, gas permeance (P/l, GPU) can be described by Eq. (4): Where l (µm) is the thickness of the selectivity layer in carbon hollow fiber membranes. Membrane characterization FESEM images were obtained using a field emission scanning electron microscopy of which the model is SU 8600 at the operating voltage of 1.50 kV. HRTEM images were obtained using a FEI Talos F200X G2 transmission electron microscope operated at a 200 kV accelerating voltage in high-resolution bright-field TEM imaging mode. Specifically, the samples were ultrasonically dispersed in ethanol for 30 min, and the resulting dispersion was deposited onto a copper TEM grid before imaging. XRD analysis was carried out by Rigaku SmartLab SE instrument operated at 45 kV and 200 mA with 2θ ranging from 5° to 90° at a scan speed of 2°/min (Cu-Kα radiation, λ = 0.154 nm). Raman analysis was conducted using a Horiba LabRAM HR Evolution with a 514 nm laser. Raman mapping was performed at a position located 5*5 μm2 from the reference point. CO2 adsorption was measured by MICROMERITICS TRISTAR Ⅱ PLUS automated gas sorption analyzer at 273 K, which conducted via surface area and pore size distribution analyzer of which the model is GCMC CO2 carbon slit. C2H4 and C2H6 adsorption was measured by MICROMERITICS TRISTAR Ⅱ PLUS automated gas sorption analyzer at 298 K. The samples were preliminarily outgassed for 12 h at 120 °C via MICROMERITICS VacPrep 061. WAXS analysis were conducted using a X - ray wide - Angle scattering instrument of which the model is Xenocs Xeuss 2.0. The q vector ranged from 3 to 28 nm−1. XPS spectra were obtained by Thermo Scientific K-Alpha operated at 150 W and 200 eV with monochromatic Al-Kα radiation. FTIR analysis were conducted by Thermo Fisher Scientific Nicolet iS20, and the wavelength range was selected from 500 to 4000 cm−1 at a scan rate of 4 cm−1. LF-NMR analysis was measured by MesoMR23-060H-I, which was applied to test the degrees of freedom for the H-nucleus operating at a proton resonance frequency of 42.58 MHz. DLS and Zeta potential were measured by Malvern Zetasizer Nano ZS90. For zeta-potential measurements, the particles were dispersed in deionized water at a concentration of 10 mg L−1, and the pH of the dispersion was 7.0. For DLS measurements, the particles were dispersed in DMSO and EmimAc at a concentration of 100 mg L−1. All dispersions were freshly prepared and thoroughly homogenized before measurement. Nano-FTIR was conducted using Bruker NanoIR3 by tapping-mode atomic-force-microscopy infrared (AFM-IR) probe. QCM-D was performed by Biolin Q-Sense Analyzer to measure the sorption ability of pristine ZIF-8 and MPN-ZIF-8 samples toward EmimAc molecules. MD Simulation methods Classical molecular dynamics simulations were employed to investigate the solvation structure of EmimAc with ZIF-8/MPN-ZIF-8 at the atomic level. The components and amount of the two simulated systems are listed in Table 1. The initial configurations of solvent systems were constructed through the software of PACKMOL44, all the molecules were randomly inserted in a cubic simulation box. The force field parameters for ZIF-8 were obtained from this ref. 45 and the OPLS-AA force field46 was employed to describe the IL and TA. The molecular force field is consisted of nonbonded and bonded interactions. The nonbonded interaction contains van der Waals (vdW) and electrostatic interactions. The molecular force field is consisted of nonbonded and bonded interactions. The nonbonded interaction contains van der Waals (vdW) and electrostatic interaction, which is described by the Eq. (5) and Eq. (6), respectively. For different kinds of atoms, the Lorentz-Berthelot mix rules were adopted for vdW interactions, which is following the Eq. (7). The cutoff distance of vdW and electronic interactions was set to 1.2 nm, and the particle mesh Ewald (PME) method was employed to calculate long-range electrostatic interactions. For this simulation, an energy minimization was firstly employed to relax the simulation box. Then, an isothermal-isobaric (NPT) ensemble with a 1.0 fs time step is employed to optimize the simulation box, where the temperature is set to 298 K, and the pressure is set to 1.0 atm. The temperature and pressure are kept via the Nose-Hoover thermostat and Parrinello-Rahman barostat, respectively. The NPT optimization time was set to 10.0 ns, which is enough long to obtain a stable box size. Following the NPT simulation, a canonical (NVT) ensemble with 50.0 ns was performed to furtherly optimize the simulation box, the time step is set to 2.0 fs. In all the MD simulations, the motion of atoms was described by classical Newton’s equation, which was solved using the velocity-Verlet algorithm. All simulations were performed using the GROMACS 2025.1 package47. DFT Simulation methods All calculations were carried out with the Gaussian 16 A.03 software48. The B3LYP functional was adopted for all calculations in combination with Grimme’s D3(BJ)49,50 dispersion correction. For geometry optimization and frequency calculations, the 6-311 G** 51,52 basis set was used for all atoms. The binding energy (ΔE) was calculated at B3LYP-D3(BJ)/ def2-TZVP53,54 basis set. ΔE was defined as the difference between the complex and the sum of energies of monomers, which can be obtained by the following formulas ΔE = E(A + B)-E(A)-E(B). All wave function analyses were finished via the Multiwfn 3.8(dev)55,56 code. The electrostatic potential (ESP) of van der Waals surface projection was obtained by quantitative analysis of the electrostatic potential surface57. The isosurface maps were rendered by means of the VMD58 visualization program based on the files exported by Multiwfn. Supplementary information Supplementary Information Transparent Peer Review file Source data Source data