In this work, building on the structural framework of PIM-1, we strategically modify its chemical architecture to enable precise alcohol/hydrocarbon separation. Spirocyclic moieties embedded in the polymer backbone critically inhibit chain packing, generating uniform micropores29,34,35. Prior work has shown that judicious insertion of flexible linkers between these spirocycles enhances chain alignment while preserving microporosity21,22. Leveraging these insights, we synthesize a spirocyclic poly(vinylene ether ketone) (SPVEK) via click chemistry36,37, fabricating thin-film composite (TFC) membranes on porous polyimide substrates (Fig. 1b). Rigorous characterization validates microporous architecture of the SPVEK membrane, organic solvent stability, and alcohol/hydrocarbon permselectivity. Complementary molecular dynamics (MD) simulations reveal the dual mechanism for azeotrope separation: alcohols exhibit preferential surface adsorption and internal diffusion, while hydrocarbons face steric hindrance effect within sub-nanometer channels (Fig. 1c). This work establishes a materials design blueprint for OSRO membranes, achieving azeotrope separation while operating at ambient temperatures -- a transformative advance toward sustainable chemical separations.
We engineered a pressure-driven membrane for OSRO using an established TFC architecture (Fig. 1b). The selective layer comprises spirocyclic poly(vinylene ether ketone) (SPVEK), synthesized via a click reaction based the spirocyclic monomer 3,3,3',3'-tetramethyl-1,1'-spirobisindane-6,6'-diol (SBI) and the aromatic monomer 1,4-dipropioloylbenzene (DPPB) (Fig. 1b, and Supplementary Fig. 1-6). Gel permeation chromatography (GPC) confirmed a weight-average molecular weight (Mw) of 20 kDa and a polydispersity index (PDI) of 2.4 (Supplementary Fig. 7). SPVEK exhibits good solubility in volatile solvents such as tetrahydrofuran and chloroform (Supplementary Table 1, Supplementary Fig. 8), enabling facile fabrication of uniform films. Differential scanning calorimetry (DSC) revealed no detectable glass transition below the decomposition temperature of the SPVEK (Supplementary Fig. 9-10), attributable to its rigid architecture -- comprising benzene rings and spirocyclic motifs -- that restricts chain mobility. This behavior mirrors that of PIM-1, which shares analogous spirocyclic units, further validating the structural design's role in enhancing thermal stability.
To engineer SPVEK with tailored microporous topology, we introduced vinylene ether ketone (VEK) units to enhance backbone flexibility while maintaining microporosity. MD simulations revealed conformational flexibility of the VEK segment, with kernel density estimation (KDE) mapping the dynamic distributions of interatomic distances (d) and angles (θ) between interatomic connection line and benzene ring (Fig. 2a, b, and Supplementary Fig. 11). Comparative analysis demonstrated broader conformational space of VEK (Δd ≈ 0.7 Å, Δθ ≈ 40°) relative to the rigid dioxane units in PIM-1 (Δd ≈ 0.16 Å, Δθ ≈ 10°) (Fig. 2a, b and Supplementary Fig. 11), confirming enhanced chain mobility. This structural distinction manifests in diminished N (77 K) and CO (273 K) adsorption capacities compared to PIM-1 (Fig. 2c, and Supplementary Fig. 12). Crucially, Negligible N uptake (0.2 cmg STP) for SPVEK paired with measurable CO adsorption (4.6 cmg STP) signifies a dense structure with enhanced microporosity, which consists with smaller kinetic diameter for CO (3.3 Å vs. 3.64 Å for N).
To elucidate microporous architecture of SPVEK, we performed MD simulations using the Gromacs package, constructing atomic-scale models of the dry polymer (Supplementary Fig. 13-15, Supplementary Table 2-4). Triplicate models per system ensured statistical reliability, with thermal equilibration achieved through 21-step annealing protocols. While SPVEK and PIM-1 exhibited comparable densities (~1.0 g cm), their pore topology diverged markedly (Fig. 2d, e, and Supplementary Fig. 15-16). Monte Carlo pore analysis revealed a maximum interconnected pore diameter (D) of 2.06 Å for SPVEK versus 3.60 Å for PIM-1 (Supplementary Table 4) -- directly explaining impermeability for SPVEK to N (3.64 Å kinetic diameter). This aligns with experimental N adsorption isotherms (Supplementary Fig. 12), validating the computational models. Crucially, micropores (1-6 Å) in SPVEK demonstrated narrow distribution versus wide distribution (2-11 Å) in PIM-1 (Fig. 2f).
Swelling studies confirmed structural resilience for SPVEK: exposure to ethanol/cyclohexane induced minimal mass changes (swelling degree < 1.5-10% vs. > 43-57% for PIM-1; Supplementary Table 5), while simulations revealed preserved micropores (1-6.5 Å post-swelling; Supplementary Fig. 17). This solvent stability establishes SPVEK as suited for maintaining size-sieving microporosity during azeotrope separations.
The SPVEK membrane was fabricated through spin-coating of a chloroform polymer solution onto a crosslinked polyimide porous support, forming a TFC architecture with a defect-free selective layer (Fig. 1b). Cross-sectional scanning electron microscope image revealed a distinct interface between the dense SPVEK selective layer (∼170 nm thick) and the underlying crosslinked polyimide support (Fig. 1b, Supplementary Fig. 18). To evaluate separation performance, we focused on the ethanol/cyclohexane azeotrope -- a system of broad industrial relevance in bioethanol dehydration and chemical production processes where cyclohexane serves as an entrainer. Prior to testing, membranes underwent preconditioning through continuous mixture permeation for 48 h at 8 MPa transmembrane pressure to stabilize swelling-induced structural rearrangements. The preconditioning process and separation performance tests were conducted in a cross-flow filtration system (Supplementary Fig. 19) operated at 25 °C.
The SPVEK TFC membrane demonstrates exceptional ethanol/cyclohexane separation performance with concentration-dependent selectivity (Fig. 3a, b). Ethanol consistently dominates permeate streams (cyclohexane content < 3.5 wt.%) across feed ethanol concentrations spanning 10-90 wt.%. Remarkably, decreasing ethanol feed fraction from 90 wt.% to 10 wt.% enhances separation factors from 14 to 330, albeit with a concomitant flux reduction from 1.38 to 0.17 kg m h (Fig. 3b). This inverse correlation between selectivity and flux arises from ethanol-induced plasticization: swelling degree measurements reveal SPVEK undergoes 10% mass expansion in ethanol versus merely 1.5% in cyclohexane (Supplementary Table 5). At elevated ethanol concentrations, this preferential swelling enlarges micropores, compromising size-sieving precision. In stark contrast, PIM-1 membranes exhibit negligible selectivity (< 1.5) for this mixture, aligning with their larger molecular weight cutoff (500-800 Da) optimized for organic solvent nanofiltration rather than molecular differentiation.
The permeate ethanol fraction consistently surpasses feed concentrations, revealing reverse osmosis characteristics through generation of counteracting osmotic pressure. Quantitative analysis of feed/permeate concentration gradients enabled calculation of apparent osmotic pressures (2.4-11.5 MPa range) that remarkably exceeded applied transmembrane pressures (8 MPa, Fig. 3c, Supplementary Fig. 20, 21). This abnormal osmotic pressure phenomenon may be related to the preferential adsorption of ethanol -- adsorption experiments and molecular dynamics indicate that ethanol accumulates at the membrane-solvent interface and within the membrane (Supplementary Fig. 22, and Supplementary Table 6). We hypothesize that ethanol exhibits preferential adsorption at the feed solution interface and recalculate the activity and osmotic pressure using molecular simulation concentration data (at the feed solution-membrane interface). The resulting corrective osmotic pressure (Fig. 3c) appears more reasonable. This also reminds us that when evaluating osmotic pressure and the OSRO membrane process, attention should be paid to the influence of the membrane material's specific adsorption of feed solution components. Additionally, the observed inverse relationship between ethanol feed concentration (10-90 wt%) and osmotic pressure (11.5-2.4 MPa) directly explains flux enhancement at higher ethanol fractions through increased effective driving force.
The separation performance of the SPVEK membrane exhibits strong pressure dependence, revealing a critical pressure threshold for optimal operation (Fig. 3d, and Supplementary Fig. 21). Increasing transmembrane pressure from 4 to 8 MPa drives separation factor surges from 10 to 84 - a 740% enhancement demonstrating pressure-tunable molecular discrimination. This pressure-activation behavior aligns with established OSRO mechanisms where elevated hydraulic forces overcome solution-diffusion limitations. Flux analysis reveals a transport mechanism: ethanol permeance escalates from 0.16 to 1.37 kg m⁻² h⁻¹ with pressure, while cyclohexane flux remains suppressed (< 0.012 kg m⁻² h⁻¹) across all tested conditions (Supplementary Fig. 23). This striking disparity is likely attributed to ethanol dual advantageous characteristics: (1) enhanced intermolecular interaction capability with SPVEK, and (2) a substantially smaller kinetic dimensions (Supplementary Fig. 24), which facilitates selective microporous transport.
The permselectivity of the SPVEK membrane was systematically evaluated using industrially relevant azeotropic mixtures: ethanol/heptane, ethanol/benzene, methanol/toluene, and alcohol/ester mixtures (Fig. 3e). The SPVEK membrane demonstrated a separation factor (α) of 74 for ethanol/heptane. However, performance declined for ethanol/benzene (α ≈ 2) and methanol/toluene (α ≈ 6), attributed to membrane swelling in aromatic solvents (swelling degree > 180%, Supplementary Table 5). This result demonstrates the importance of anti-solvent plasticization in molecular sieving for membrane material design. Notably, the membrane exhibited a certain degree of selectivity for alcohol/ester mixtures (Fig. 3e, mixtures 4-7), highlighting potential for esterification process intensification. Long-term stability testing under industrial-relevant conditions (8 MPa, 30 days) revealed < 5% flux decline and stable permselectivity (Fig. 3f, and Supplementary Fig. 25), demonstrating exceptional operational durability. Remarkably, we found that during the OSRO process (driven solely by pressure at ambient temperature), the SPVEK membrane exhibited high separation performance (α = 330, Flux=0.17 kg m h) comparable to most energy-intensive pervaporation processes (operated under heated and vacuum conditions, Supplementary Fig. 26). This establishes a benchmark for membrane-based azeotrope separation.
To elucidate the molecular mechanism of azeotrope permselectivity, we conducted non-equilibrium molecular dynamics (NEMD) simulations using a fully atomistic SPVEK membrane model (Fig. 4a, and Supplementary Fig. 13, 27-28; Supplementary Table 2, 7-8). The simulation system comprised a 4 nm-thick SPVEK membrane flanked by azeotropic mixtures, with graphene pistons generating a 100 MPa pressure gradient to overcome rare-event kinetics at experimental operating pressures (4-8 MPa)-a validated acceleration strategy for reverse osmosis simulations.
Two-dimensional density distribution plot of the SPVEK membrane (Fig. 4b, and Supplementary Fig. 29) revealed uniform polymer backbone distribution with localized variations, indicative of solvent-induced swelling during diffusion. To elucidate the permselectivity mechanism, we analyzed solvent density distribution along the transport direction (z axis, Fig. 4c, d). Ethanol exhibited preferential adsorption at the feed-side membrane surface (near z = 22.5 nm), with concentrations significantly exceeding the feed, while the concentration of cyclohexane is lower than the feed (Fig. 4d). Permeate-side analysis revealed that the ethanol/cyclohexane ratio at the interface is similar to that on the permeate side, with no significant selective adsorption. This is attributed to ethanol being the main component (> 96.5 wt.%) in the permeate. Coordination number analysis (Supplementary Fig. 31) revealed reduced yet non-isolated solvent clustering (< 5 Å) within membrane channels, suggesting cluster-mediated transport through confined pathways.
Analysis of molecular dynamics simulations revealed distinct transport behaviors through mean square displacement (MSD) calculations (Fig. 4e). The ethanol diffusion coefficient (2.20 × 10⁻ cm s⁻) significantly surpasses that of cyclohexane (0.73 × 10⁻ cm s⁻), confirming preferential ethanol permeation (Supplementary Table 9). Applying the solution-diffusion model, we quantified the selectivity components: diffusion selectivity (58.8%) and sorption selectivity (41.2%), which combine to produce the overall membrane selectivity. Molecular interaction analysis reveals that ethanol engages with SPVEK chains via hydrogen bonding (-33.26 kJ mol), whereas cyclohexane exhibits exclusively van der Waals interactions (-17.53 kJ mol, Fig. 4f, g; Supplementary Fig. 32-33). Stronger hydrogen bonding interactions lead to greater adsorption of ethanol on the membrane surface. Meanwhile, due to its larger kinetic dimensions, cyclohexane exhibits a three times lower diffusion coefficient within the membrane (Supplementary Fig. 24). The preferential adsorption of ethanol and the lower diffusion rate of cyclohexane within the membrane collectively determine high selectivity for ethanol/cyclohexane azeotrope separation, demonstrating the dual selectivity mechanism of SPVEK membranes.
In membrane separation processes, separation efficiency is critically governed by membrane selectivity. Conventional membranes typically exhibit performance degradation with varying feed compositions, significantly limiting their operational flexibility. Remarkably, our SPVEK membrane demonstrates an inverse correlation between selectivity and ethanol concentration in the feed (Fig. 3b). This counterintuitive behavior, where selectivity increases as ethanol concentration decreases, suggests a molecular discrimination mechanism fundamentally different from conventional pervaporation membranes. To evaluate practical applicability, we engineered a two-stage cascade separation process (Fig. 5a) using an azeotropic ethanol/cyclohexane mixture (1:1 w/w). The first-stage permeate achieved > 95 wt.% ethanol purity, while the retentate concentrated to > 95 wt.% cyclohexane (Fig. 5b). Subsequent second-stage processing yielded ethanol purity exceeding 98 wt.% (Fig. 5c), demonstrating exceptional multi-stage scalability. Energy analysis reveals the transformative potential of our OSRO process (Fig. 5d). Compared to conventional distillation and pervaporation, our membrane system reduces energy consumption by 2-3 orders of magnitude. This breakthrough efficiency stems from the synergetic combination of microporous molecular sieving and pressure-driven operation, bypassing the phase-change energy penalties inherent to thermal processes.