Na+-gated water-conducting nanochannels for boosting CO2 conversion to liquid fuels

See allHide authors and affiliations

Science  07 Feb 2020:
Vol. 367, Issue 6478, pp. 667-671
DOI: 10.1126/science.aaz6053

Water-selective zeolite membranes

The yield of many gas-phase industrial reactions is limited by the formation of water as a by-product. Li et al. harnessed the water-sieving properties of NaA zeolite crystals by forming them into continuous defect-free membranes within tube reactors (see the Perspective by Carreon). These membranes can let water pass but reject gases such as hydrogen, carbon monoxide, and carbon dioxide (CO2). When these membranes were used in CO2 hydrogenation to form methanol with water as a by-product, substantial increases were observed in both the CO2 conversion and methanol yield.

Science, this issue p. 667; see also p. 624


Robust, gas-impeding water-conduction nanochannels that can sieve water from small gas molecules such as hydrogen (H2), particularly at high temperature and pressure, are desirable for boosting many important reactions severely restricted by water (the major by-product) both thermodynamically and kinetically. Identifying and constructing such nanochannels into large-area separation membranes without introducing extra defects is challenging. We found that sodium ion (Na+)–gated water-conduction nanochannels could be created by assembling NaA zeolite crystals into a continuous, defect-free separation membrane through a rationally designed method. Highly efficient in situ water removal through water-conduction nanochannels led to a substantial increase in carbon dioxide (CO2) conversion and methanol yield in CO2 hydrogenation for methanol production.

Protein-constructed water channels (1) or ion channels (2) exist in all living organisms and allow fast permeation of water or specific ions while rejecting other ionic species on the basis of size exclusion and electrostatic repulsion. These attributes have led scientists to attempt to create similar channels as a means of increasing the efficiency of industrial processes. Experimental strategies to obtain high water permeability while rejecting hydrated ions in artificial systems have included the incorporation of protein-based (3) or polymer-based artificial water channels (4, 5) into a polymer matrix or lipid layers, as well as packing of graphene-based laminates (69). Very promising results in terms of ion rejection for desalination have been successfully demonstrated (6, 7).

We explored whether nanochannels could be fabricated that reject small gas molecules approximately half the size of the hydrated ions (for example, Na+, 6.6 Å) at high temperatures and pressures for applications in catalysis. For example, by-product water strongly inhibits the kinetics and thermodynamics of CO2 hydrogenation to liquid fuels such as methanol (1012). Gas-rejecting water-conduction nanochannels could boost the reaction rate and shift the equilibrium toward product formation by removing water while retaining reactant gases and products (13). However, angstrom-scale nanochannels that can distinguish water molecules (kinetic diameter 2.6 Å) from gas molecules as small as H2 (kinetic diameter 2.9 Å) are very challenging to precisely engineer. Further, the nanochannel would need to be assembled without defects into a practical separation membrane that could operate at >200°C and >20 bar.

We report gas-impeding water conduction of NaA zeolitic nanochannels assembled into a centimeter-scale membrane with negligible defects through a rationally designed method (Fig. 1, A and B). Water conduction, which was confirmed by experimental gas dehydration at high temperatures and pressures, was likely the result of a gating effect of Na+ located in the 8-oxygen ring apertures that regulated their effective size, as supported by theoretical simulations (Fig. 2). The exceptional benefits of such a water-conduction membrane (WCM) were demonstrated for catalytic methanol synthesis from CO2 hydrogenation at elevated temperatures (200° to 250°C) and pressures (21 to 35 bar).

Fig. 1 Rationally designed preparation strategy and separation/permeation properties of water-conduction membrane (WCM).

(A) Schematics of different preparation routes. In route a, 50- to 200-nm seeds are fixed at high loading density onto and into the support through dehydration of surface hydroxyl groups as illustrated, for growth of WCM-a with defects largely suppressed, whereas in route b, these seeds are used directly for growth of membrane (M-b) with defects. In route c, the seeds are diluted by a factor of 2 or 10 relative to route a for growth of M-c-02 and M-c-10, respectively. In route d, larger seeds (300 to 400 nm and 400 to 700 nm) are fixed onto and into the support through dehydration for growth of M-d-300 and M-d-400, respectively. (B and C) Molecular transport pathway through WCM-a (B) and through membranes prepared by routes b to d (C). (D) Separation performance of membranes prepared by different routes for H2O/CO2/CO/H2/MeOH mixture with composition of 1.77 ± 0.14% / 23.52% / 0.98% / 73.50% / 0.23 ± 0.02% at 250°C and 21 bar. The minimum H2O/H2, H2O/CO, and H2O/MeOH selectivities of the WCM are estimated from the detection limits of GC. The fraction ratio at the top of each column represents the selectivity ratio from top to bottom. Error bars denote SD. (E) Permeability of pure gases and water through the WCM at 25° to 200°C and 2 to 8 bar (solid red symbols) and comparison with previous results (open symbols) (2344). The shaded area is the permeability zone of gases and water, based on reported results. Error bars denote SD.

Fig. 2 DFT simulation of the passage of molecules through NaA zeolitic channels.

(A) Top view of the structural framework of NaA zeolite with three Na+ sites in the dehydrated crystal structure. (B) The passage of molecules through the zeolitic channels of NaA. In state 1, the gas molecule is away from the 8-oxygen ring. In state 2, the gas molecule is adsorbed onto the ring. In state 3, the gas molecule is in the ring. In state 4, the gas molecule is in the α-cage of the NaA zeolite. Numbers are energy levels calculated by DFT for three molecules moving into the α-cage of NaA zeolite through the 8-oxygen ring.

We designed a rational and effective process (Fig. 1A, route a) to prepare WCMs. We dip-coated a ceramic hollow-fiber support (length 300 mm, inner diameter 0.75 mm, outer diameter 1.5 mm, pore size 400 nm; fig. S1) with 50- to 200-nm NaA crystals (fig. S2) and then dried the support at 80°C for at least 3 hours. Prior to membrane synthesis, the coated support was subjected to thermal annealing at 200°C overnight. During annealing, physically loaded nanocrystals were chemically bonded to the support surface and inside the pores through dehydration of surface hydroxyl groups of the zeolite and the support (14) to ensure stable and adequate nuclei for inducing membrane growth. After a single cycle of hydrothermal growth, membranes without visible surface defects grew on the support (fig. S3). The penetration of small crystals, also called seeds, inside the 400-nm support pores also induced membrane growth within the pores and resulted in an indistinguishable boundary between the membrane layer and the support. The membrane thickness was 3 to 4 μm, as indicated by energy-dispersive x-ray spectroscopy (EDX) (fig. S3).

We investigated the separation performance of our WCM for a H2O/CO2/CO/H2/MeOH gas mixture (MeOH, methanol) with composition of 1.77 ± 0.14% / 23.52% / 0.98% / 73.50% / 0.23 ± 0.02% in our homemade apparatus (fig. S4) at 250°C and 21 bar. The mixture was generated by bubbling a 75% H2 / 1% CO / 24% CO2 gas mixture through a liquid tank containing 1.5 wt % MeOH in H2O at 70°C. The selectivity (permeance ratio) of H2O/CO2 for this mixture (Fig. 1D and table S1) was ~551 ± 33. The minimum selectivities of H2O/H2, H2O/CO, and H2O/MeOH were 190, 170, and 80, respectively, as estimated from the detection limits of gas chromatography (GC), because no H2, CO, and MeOH were detected on the permeate side. Relative to previously reported separation results (fig. S5 and table S2), our membrane showed two to three orders of magnitude lower gas permeances but comparable water permeance, and thus considerably higher H2O/gas selectivity, suggesting effective gas-permeation blockage by the membrane layer (Fig. 1B).

At higher pressures (up to 38 bar), different temperatures (200° and 250°C), and various water concentrations (1 to 4.2 mol %), the WCM maintained excellent mixture separation performance (fig. S6). When a binary H2O/gas mixture was used as feed, the selectivity of H2O/CO2 was as high as 10,722 ± 222 at 250°C and 21 bar (table S1). The much higher CO2 partial pressure in the binary mixture did not lead to a proportional increase of its flux and thus drastically decreased CO2 permeance (1.39 × 10–11 mol m–2 s–1 Pa–1), whereas water permeance was almost the same (1.49 × 10–7 mol m–2 s–1 Pa–1). No obvious selectivity decline (fig. S7) was detected throughout 12-hour tests, suggesting high membrane stability.

High-density small seeds (50 to 200 nm) that had been bonded to the support by annealing were critical for high membrane quality, as evidenced by the results of a series of comparative experiments (Fig. 1D and table S1). Water/gas selectivities near the as-reported values (fig. S5 and table S2) were obtained after the elimination of the annealing step (Fig. 1A, route b), highlighting the importance of annealing treatment. A marked decrease of water/gas selectivities occurred when the 50- to 200-nm seed suspensions were diluted by factors of 2 and 10 (Fig. 1A, route c), indicating that the initial high-density loading of seeds was essential. When 300- to 400-nm or 400- to 700-nm seeds were applied, the accumulated selectivities were a factor of 5 to 8 lower than with the 50- to 200-nm seeds (Fig. 1A, route d).

The decrease of water/gas selectivity was mainly the result of the increase of gas permeances (by factors of 60 to 300) because the water permeance did not exhibit large variation (less than a factor of 3). Relatively high gas permeances can be attributed to the presence of nonselective defects that allow water and gases to pass at comparable rates (Fig. 1C). High-density, stabilized seeds on the surface resulted in the growth of a continuous membrane layer on the support surface, whereas the penetrated small seeds in the support pores facilitated the growth of the crystals in the pores and subsequent seamless merging. These two intergrown layers ensure the high quality of the WCM with negligible defects and thus high water/gas selectivities.

To further understand the mechanism of gas blockage by the WCM, we performed single-gas permeation measurements for gas molecules with different sizes after membrane degassing at 200°C for at least 3 days to exclude the influence of adsorbed water in the zeolitic nanochannels. Even without adsorbed water in NaA nanochannels, gas permeances in single-gas permeation (fig. S8 and table S3) were in the same range (10–10 to 10–9 mol m–2 s–1 Pa–1) as those in the mixture separation (Fig. 1D and table S1); this finding suggested that adsorbed water in mixture separation had negligible influence on gas permeation and that NaA zeolite nanochannels effectively blocked gas permeation.

We then conducted a thorough comparison of pure gas and water permeabilities with results on NaA zeolite membranes reported within the past 20 years (Fig. 1E). To have a fair comparison, we used gas permeability, calculated as the product of gas permeance and membrane thickness. The pure gas permeabilities of the WCM were two to three orders of magnitude lower than the reported results and completely dropped out of their respective permeability zones. However, the water permeability of the WCM was two to three orders of magnitude higher than that of gases and was well within the water permeability zone, comparable to the reported results. These findings, combined with mixture separation results, indicate that the WCM has very high quality and negligible defects. Moreover, the WCM also exhibited good thermal stability in dry gas up to 400°C (table S4) and good water stability up to 300°C (fig. S9).

The accepted channel diameter of the NaA structure, dominated by the 8-oxygen ring (Fig. 2A), is 4.2 Å, as calculated from the interatomic distance of two opposing oxygen atoms across the ring (15). However, Na+, by neutralizing the negatively charged NaA framework (Fm3¯c space group with a = 24.555 Å, unit cell composition of Na96Si96Al96O384) and positioning inside zeolite nanocavities with three locations (Fig. 2A), partially blocks the nanochannels and decreases the effective aperture size (16, 17). We further speculate, on the basis of our permeation results, that molecules entering these nanochannels can be influenced strongly by Na+. The passage of small, polar water molecules to the zeolitic nanochannel could be facilitated by Na+, whereas the passage of larger and less polar molecules, such as H2 and CO2, would be hindered, leading to much faster transport of water molecules through the zeolitic nanochannel.

To better understand the potential gating effect of Na+, we performed density functional theory (DFT) simulations to mimic the process of molecules passing the 8-oxygen ring aperture (Fig. 2B). We consider the passage of a molecule through the 8-oxygen ring as a three-step process with four defined states of a molecule (Fig. 2B): (i) approaching the surface of the 8-oxygen ring; (ii) entering the ring, referred to as admission of a molecule to the ring; and (iii) moving out of the ring and entering the α-cage of the NaA zeolite. Our result suggests that the most stable Na+ location in NaA zeolite is 1.27 Å away from the center, which is consistent with previous reports (17, 18); in the presence of H2O, H2, and CO2, respectively, stable Na+ positions are ~1.33 Å away from the center (fig. S10). At its stable positions, in contrast to the case without Na+ in the 8-oxygen ring, the introduction of Na+ decreases the admission energy barrier for H2O but increases the barrier for CO2 and H2 (fig. S11 and table S5), supporting our hypothesis. Specifically, in Fig. 2B, the DFT calculation shows a substantial energy decrease (1034 meV) when a polar H2O molecule adsorbs onto the aperture. The corresponding decreases were 300 meV for the less polar CO2 molecule and 98 meV for nonpolar H2, indicating a much more favorable interaction between Na+ and H2O than with CO2 and H2.

The admission energy barriers ∆E are 255 meV for CO2, 38 meV for H2, and 3 meV for H2O. Whereas H2O encounters a negligible energy barrier when entering the aperture, apparently resulting from the favorable interaction between polar H2O and Na+ and its small size, H2 and CO2 must overcome higher energy barriers, especially for larger CO2, to enter the aperture. When further entering the α-cage, H2O shows the largest energy drop (200 meV), followed by H2 (68 meV); for CO2, energy input (229 meV) is needed to enter the cage. The favorable interaction between H2O and the α-cage also facilitates its fast passage of the 8-oxygen ring. Overall, both the admission into the 8-oxygen ring and subsequent entry into the α-cage energetically favor the H2O molecule.

We applied a WCM to a reaction for the production of liquid fuels (Fig. 3A). We loaded copper-zinc-alumina (CZA) catalysts (fig. S12), which have been commercialized for methanol production, on the outer surface of a WCM to catalyze CO2 hydrogenation for methanol production at 21 to 35 bar and 200° to 250°C and gas hourly space velocity (GHSV) between 5100 and 10,500 ml gcat–1 hour–1. The CZA catalyst was reduced in pure H2 in situ at 250°C and atmospheric pressure for 10 hours with a ramping rate of 2°C/min in a homemade apparatus [membrane reactor (MR); fig. S13]. For comparison, the performance of the same catalysts without incorporation of the WCM [traditional reactor (TR)] was also evaluated with the same catalyst packed in a nonpermeable dense tube with the same dimensions as the WCM. Among all results obtained (Fig. 3B and figs. S14 and S15), CO2 conversion (up to 61.4%) exceeded the equilibrium conversion (19) after incorporation of the WCM and was 2.6 to 3.0 times that obtained without the WCM.

Fig. 3 Schematics of the membrane reactor (MR) concept and catalytic performance of the MR.

(A) Schematics of WCM-incorporated dehydration MR for high-purity methanol direct synthesis from renewable resources. (B) Catalytic CO2 conversion (points) and methanol yield (columns) obtained in the traditional reactor (TR; orange) and in the MR (purple) as a function of temperature at 35 bar and feed (CO2/H2 = 1/3) gas hourly space velocity (GHSV) of 5100 ml gcat–1 hour–1. Membrane length is 45 mm. Error bars denote SD. (C) Catalytic performance of methanol synthesis from CO2 hydrogenation over 100 hours in the MR at 220°C and 35 bar. The feed (CO2/H2 = 1/3) GHSV is 4200 ml gcat–1 hour–1; membrane length is 45 mm. (D) Demonstration of high-purity methanol direct synthesis in the MR at different feed GHSVs at 220°C and 35 bar with three hollow-fiber WCMs 210 mm in length.

Because NaA zeolite has no catalytic activity for CO2 hydrogenation (20) and only CO2 and H2 were detected with only the WCM in the reactor at 35 bar and 250°C, effective in situ water removal by the WCM was the only reason for the drastically increased CO2 conversion; moreover, it was the only reason why the thermodynamic equilibrium limitation on CO2 hydrogenation was overcome. The adsorption of water on the catalyst was greatly inhibited by the low water concentration (less than 2 mol %) inside the MR, which should have enhanced the kinetics of the reaction and catalyst activity. Indeed, the methanol yield (25.1% to 38.9%) in the MR was 2.8 to 3.5 times that in the TR, with no obvious difference in product selectivity (figs. S16 and S17). Correspondingly, the space-time yield (STY) of methanol was highly promoted, for example, from 339 mg gcat–1 hour–1 in the TR to 809 mg gcat–1 hour–1 in the MR with GHSV of 10,500 ml gcat–1 hour–1 at 250°C and 35 bar (fig. S18), the highest value ever reported under similar conditions.

Moreover, because of the in situ water removal (by 95%) from methanol, very-high-purity methanol was directly obtained by simply condensing liquid products after the reactor—for example, 95.9 wt % in the MR versus 54.3 wt % in the TR at 230°C and 35 bar (figs. S19 and S20). The purity can be further increased by optimizing the MR (see detailed analysis in supplementary materials). This process could be expected to save a considerable amount of energy in purification processes such as distillation. The selectivities of H2O over other components during the reaction (fig. S21) were consistent with the mixture separation results (fig. S6).

We performed stability tests at the highest methanol yield (TR, 14.0%; MR, 39.8%) under conditions of 220°C and 35 bar. We obtained a minor decrease (4%) of CO2 conversion (57.2% to 54.8%) and methanol yield (39.8% to 38.1%) in the MR, as well as very stable high-purity methanol (~95 wt %) production for more than 100 hours (Fig. 3C); by contrast, in the TR, in which catalysts were exposed to large amounts of product water, we obtained a 10% decrease of CO2 conversion (22.7% to 20.4%) and a 20% decrease of methanol yield (14.0% to 11.2%) (fig. S22). These results suggest excellent stability of our WCM under reaction conditions and effective protection of catalysts from water poisoning and catalyst sintering. Moreover, no obvious difference between SEM images of the WCM (fig. S23) and Brunauer-Emmett-Teller (BET) surface areas of catalysts (table S6) before and after reaction supported the above observation.

We tested our WCM for potential use at larger scale by assembling three 210-mm-long membranes (Fig. 3D, lower left inset) and loading ~2.8 g of catalysts in one reactor. We ran bench-scale methanol synthesis by drawing vacuum on the permeate side (Fig. 3D, upper inset) to generate driving force for water permeation at 250°C and 35 bar (movie S1). CO2 conversion of 41.0% to 57.6% and methanol yield of 21.9% to 30.3% were achieved with the long MR comparable to those for the short MR; the methanol production rate of the long MR approached 50 g/day with an average purity of 94.6 wt % (Fig. 3D, lower right inset, and fig. S24).

We compared our catalytic reaction results with those previously reported (Fig. 4 and table S7) in terms of CO2 conversion and STY of methanol. With only CZA catalysts, the TR showed only moderate CO2 conversion (23.0%) and methanol STY (339 mg gcat–1 hour–1). After the incorporation of the WCM, both CO2 conversion and methanol STY were greatly boosted (61.4% and 809 mg gcat–1 hour–1, respectively). Our results obtained in the MR represent the highest values reported to date under similar reaction conditions and are even higher than those obtained at much higher temperatures and pressures. It may be possible to use WCMs for other reactions that are kinetically or thermodynamically restricted by water—such as CO2 hydrogenation to dimethyl ether (21) and Fischer-Tropsch synthesis (22)—by loading corresponding catalysts onto the outer surface of the WCM.

Fig. 4 Comparisons of catalytic results in the MR using WCMs in this study with TR results from literature in terms of CO2 conversion and methanol space-time yield.

Data points 1 and 12 are from (45); 2, (46); 3, (47); 4, (48); 5, (49); 6, (50); 7, (51); 8, (52); 9, (53); 10 and 11, (54); 13, (55); 14, (56); 15, (57); 16 to 19, (58); 20, (59). Orange stars, TR in this study; purple stars, MR in this study.

Supplementary Materials

Materials and Methods

Figs. S1 to S24

Tables S1 to S7

Movie S1

References (6075)

References and Notes

Acknowledgments: Funding: Supported by Advanced Research Projects Agency–Energy (ARPA-E), U.S. Department of Energy, award DE-AR0000806; Rensselaer Polytechnic Institute (RPI) startup funds; and National Natural Science Foundation of China grant NSFC-21625604 (J.W.). Author contributions: M.Y. and H.L. conceived the project; H.L. prepared and evaluated the membranes and membrane reactors; S.R. and X.L. supplied and characterized the catalysts; Q.D. conducted the BET test; S.Z. and S.L. drew and discussed some figures; F.Z. performed the XRD; C.Q. and J.W. performed the DFT calculation; M.Y. directed all aspects of the project; and M.Y. and H.L. wrote the manuscript. Competing interests: A U.S. provisional patent (application number 62/795,082), with M.Y. and H.L. as co-inventors, has been filed by RPI. Data and materials availability: All data are available in the main text or the supplementary materials.

Stay Connected to Science

Navigate This Article