Hydrophobic promoter in physical proximity stabilizes cobalt catalysts for durable Fischer-Tropsch synthesis

Abstract

Cobalt catalysts have been widely studied for Fischer-Tropsch synthesis, yet they gradually deactivate under water-rich reaction conditions. Current stabilization strategies rely on complex catalyst compositions or even use noble-metal promoters. Here, we demonstrate that cobalt deactivation can be effectively hindered by physically regulating water diffusion. By physical mixing of a classical Co/SiO₂ catalyst with hydrophobic polydivinylbenzene, a microenvironment is created that promotes directional water transport away from the cobalt. As a result, stable Fischer-Tropsch operation is achieved for 1200 hours, maintaining high activity with 88-90% selectivity toward C₅₊ hydrocarbons, minimal methane and CO₂ production. Mechanistic investigations suggest that the combination of hydrophobic polymer could promote directional water removal from the catalyst surface. These findings demonstrate a practical strategy for stabilizing highly water-sensitive catalysts.

Cobalt catalysts are widely used for Fischer–Tropsch synthesis but gradually deactivate under water-rich reaction conditions. Here, the authors demonstrate a practical strategy to stabilize these water-sensitive catalysts by regulating water diffusion.

Introduction

Fischer-Tropsch synthesis (FTS) provides a versatile route for converting syngas derived from coal, biomass, and natural gas into fuels and chemicals^1–7. Among the various catalytic systems, supported cobalt nanoparticles have been widely used in industrial FTS since the pioneering work of Fischer and Tropsch, owing to their high activity and selectivity toward long-chain hydrocarbons^{1,2,4,5,8–13}. Despite these advantages, cobalt catalysts remain fundamentally limited by deactivation during operation^1,11–16, which is closely associated with the accumulation of oxygen-containing species under FTS conditions. CO dissociation on cobalt generates adsorbed oxygen species (*O). If these oxygen species are not efficiently removed by hydrogen, they can accumulate on the cobalt surface and promote cobalt oxidation^14–18. Conversely, hydrogen reacts with these oxygen species to form water, thereby regenerating metallic cobalt sites^19,20. The generated water may subsequently reabsorb and dissociate on cobalt, further contributing to cobalt oxidation^18,21,22. In addition, elevated water concentrations shift the local reaction environment toward oxidation and inhibit the hydrogen-assisted removal of surface oxygen species^16,19,23. Consequently, the state of cobalt under FTS conditions is dynamically governed by the balance between reducing (e.g., H₂) and oxidizing (e.g., H₂O) capabilities of the reaction environment^14,16,19,23.

Considerable effort has been devoted to mitigating the deactivation by strengthening metal-support interactions using reducible oxides, such as Al₂O₃, TiO₂, and CeO₂^24,25. However, under hydrothermal FTS conditions, these approaches may lead to the formation of irreversibly inactive mixed oxides (e.g., CoAl₂O₄)^1,4,26–29. A variety of strategies have been developed to improve cobalt catalyst stability, including the incorporation of noble-metal (e.g., Ir¹², Pt^30,31, Rh³², Ru^4,31,33) or non-noble-metal promoters (e.g., Zr and Mn³⁴), which can facilitate oxygen removal and/or stabilize the hexagonal close-packed cobalt phase¹². While effective, this approach substantially increases catalyst cost and complexity. An alternative approach involves modifying the catalyst with hydrophobic organosilane layers, which can partially mitigate the detrimental effects of water^35–37. Chemical modification often introduces phyllosilicate species or alters surface chemistry, thereby reducing catalytic performance. Physical incorporation of hydrophobic modifiers has been shown to enhance the activity of carbide catalysts in syngas-to-olefin conversion³⁸. However, whether hydrophobic domains that are spatially separated from the catalyst surface can effectively influence the local water environment and even the chemical state of metallic cobalt catalysts remain unclear^39,40.

Here we demonstrate that physical regulation of water diffusion offers an effective and scalable route to stabilizing cobalt catalysts in FTS. By physically mixing a classical silica-supported cobalt catalyst (Co/SiO₂) with a low content of hydrophobic polydivinylbenzene (e.g., 5 wt%), cobalt deactivation is completely suppressed under FTS conditions. The resulting catalyst exhibits stable performance over 1200 h of continuous operation, maintaining high CO conversion and 88-90% selectivity toward C₅₊ hydrocarbons with low methane formation, whereas the unmodified Co/SiO₂ catalyst rapidly deactivates. Mechanistic investigations suggest that the coexistence of hydrophilic catalytic domains and hydrophobic polymer domains facilitates water transport away from the catalyst surface, thereby reducing water accumulation within the catalyst granules. This mitigates water-induced oxidation and structural evolution during operation, contributing to the exceptional catalyst stability observed under FTS conditions. These results demonstrate that regulating local water transport provides an effective strategy for improving the durability of cobalt FTS catalysts without catalyst redesign.

Results and discussion

Catalysis in FTS

The cobalt-based catalyst supported on silica was prepared by conventional deposition, yielding a Co/SiO₂ catalyst with a cobalt loading of 13.2 wt% (Supplementary Fig. 1). Under FTS conditions at 220 °C and a syngas pressure of 2.0 MPa, the catalyst (diluted with quartz sand) initially exhibited a CO conversion of 67.9% and a C₅₊ selectivity of 88.2% (Fig. 1a). However, significant deactivation was observed over the Co/SiO₂ catalyst: after 80, 201, and 592 h of reaction, CO conversion declined to 44.3%, 30.2%, and 7.4%, while C₅₊ selectivity also decreased to 87.2%, 83.6%, and 59.4%, respectively. These findings align with previous observations of cobalt catalyst deactivation during FTS^12,16. Such deactivation has been widely associated with multiple processes, including cobalt oxidation, structural evolution, and sintering under water-rich reaction environments¹⁷.

Fig. 1. Catalytic data.

a Data showing the durability of Co/SiO₂ and Co/SiO₂-P_mix catalysts in FTS. Reaction conditions: 0.5 g of Co/SiO₂ catalyst, or a mixture of 0.5 g of Co/SiO₂ and 0.025 g of PDVB in a granule mixing manner, granule size at 40–60 mesh, 2.0 MPa of syngas (H₂/CO ratio at 2), 2400 mL g_cat⁻¹ h⁻¹, 220 °C. b Product distribution of the FTS over Co/SiO₂-P_mix catalyst. c Data showing the performance of Co/SiO₂ catalyst mixed with PDVB with different mixing manners. The reaction conditions are the same as those in Fig. 1a, except for changing the mixing manners of 0.5 g of Co/SiO₂ and 1.0 g of PDVB. More PDVB was used to facilitate experimental operation and comparison. d Data showing the deactivation rates of various catalysts with different mixing manners. The error bounds for CO conversion and product selectivity were ±2%.

The catalyst, denoted as Co/SiO₂-P_mix, was prepared by mechanically mixing Co/SiO₂ granules with hydrophobic polydivinylbenzene granules, both having sizes of 40–60 mesh (PDVB, water-droplet contact angle 144.9°, irregular morphology, surface area <3 m²/g; Supplementary Figs. 2 and 3). Under the FTS conditions, PDVB was chemically inert and stable and did not melt or flow, which might otherwise encapsulate or block the catalyst surface (Supplementary Figs. 4–7). The catalyst bed volume of Co/SiO₂-P_mix was kept consistent with that in the test over the Co/SiO₂ catalyst diluted with quartz sand. The Co/SiO₂-P_mix catalyst, having a weight ratio of PDVB to Co/SiO₂ at 1/20, demonstrated an initial CO conversion of 44.1% and a C₅₊ selectivity of about 90.0% (Fig. 1a). This performance remained stable over a 1200-h time-on-stream test with undetectable deactivation, suggesting the potential for an even longer lifetime. Throughout the reaction, methane and CO₂ selectivities were consistently maintained at about 5.7% and <1.0%, respectively (Fig. 1a, b). This result underscores the critical role of PDVB in enhancing the durability of the Co/SiO₂ catalyst. Generally supported cobalt catalysts exhibit limited durability and rapid deactivation in FTS (Supplementary Table 1).

The optimal PDVB contents within the catalyst system were determined by adjusting the PDVB-to-Co/SiO₂ weight ratio (Supplementary Table 2). All these tests were performed by mixing Co/SiO₂ granules with hydrophobic polydivinylbenzene granules at 40–60 mesh. A ratio of 1/20 was initially identified as effective in stabilizing performance. To further probe the minimum requirement, the PDVB content was reduced to a 1/50 ratio; this catalyst exhibited deactivation, declining from an initial conversion of 69.0% to 54.1% after 50 h (Supplementary Fig. 8). Nevertheless, this deactivation rate was slower than that observed for the bare Co/SiO₂ catalyst, indicating a beneficial but insufficient effect of PDVB. Increasing the PDVB content to ratios between 1/15 and 1/0.5 effectively suppressed deactivation, maintaining stable CO conversion with undetectable activity loss over time (Supplementary Fig. 9). These results confirm that sufficient PDVB content is critical for stabilizing catalytic performance. However, exceeding this threshold of 1/20 inversely reduced the initial CO conversions, although the C₅₊ selectivity remained unaffected.

Different manners for mixing the Co/SiO₂ catalyst with PDVB for FTS were studied (Fig. 1c, d and Supplementary Table 3). In a dual-bed manner, where Co/SiO₂ and PDVB granules were segregated into upper and lower beds with an intervening quartz sand layer, the catalytic performance was comparable to that of the unmodified Co/SiO₂. This setup proved ineffective, as an initial CO conversion of 75.9% was followed by a decline to 36.5% within 94 h (Supplementary Fig. 10). Alternatively, in the granule mixing manner, we varied the granule size to control the proximity between the two components (Supplementary Table 4). A mixture of 40–60 mesh granules sustained a stable CO conversion of 32.7% over the catalyst with PDVB-to-Co/SiO₂ weight ratio of 1/0.5, and 44.1% over the catalyst with PDVB-to-Co/SiO₂ weight ratio of 1/20. In contrast, mixtures with larger granules (10–20 mesh and 20–40 mesh) exhibited gradual deactivation. In the powder mixing manner with the closest proximity, where the Co/SiO₂ powder was mixed with PDVB powder and then made into 40–60 mesh granules but having adjusted PDVB-to-Co/SiO₂ weight ratio at 1/20 and 1/0.5 (Supplementary Fig. 11 and Fig. 1c, d), the CO conversion was 32.4% and 27.2%, with C₅₊ selectivity of 86.9% and 87.5%, respectively, which are lower than that over Co/SiO₂-P_mix catalyst with 40–60 mesh granule mixture. Importantly, these performances were also stable with undetectable deactivation in the continuous test (Supplementary Fig. 12). A control experiment, mixing the Co/SiO₂ powder with inert quartz powder, resulted in even more severe deactivation (CO conversion of 18.5% after 49 h of reaction, Supplementary Fig. 13) than the bare Co/SiO₂ catalyst.

All the aforementioned results confirm the effect of PDVB in inhibiting cobalt deactivation during FTS, where proximity between PDVB and Co/SiO₂ is necessary to achieve this goal, as in powder or small-size granule mixtures. Notably, the PDVB and quartz sand were non-porous, with negligible porosity, as confirmed by N₂ sorption measurements (Supplementary Figs. 2 and 14), thereby excluding the effects of porosity and particle packing on the catalytic performance. Regarding the influence of heat transfer to catalysis, we performed control experiments using silicon carbide (SiC) granules as a diluent for the Co/SiO₂ catalyst. SiC exhibits negligible porosity (Supplementary Fig. 15), similar to quartz sand and PDVB, thereby minimizing differences arising from pore-related transport effects. In terms of surface properties, SiC is hydrophilic, which is similar to quartz sand but distinctly different from hydrophobic PDVB (Supplementary Figs. 3 and 16). Importantly, SiC possesses a significantly higher thermal conductivity (>50 W m⁻¹ K⁻¹) than both quartz sand (<2 W m⁻¹ K⁻¹) and PDVB (<0.3 W m⁻¹ K⁻¹)⁴¹. Therefore, employing SiC as a reference enables decoupling heat dissipation from other factors. As shown in Supplementary Fig. 17, the Co/SiO₂ catalyst diluted with SiC exhibits a continuous decline in conversion over time, similar to the behavior observed with quartz sand. We even conducted control experiments at a slightly elevated reaction temperature (225 °C). As shown in Supplementary Fig. 18 and Supplementary Table 5, the Co/SiO₂ catalyst diluted with either SiC or SiO₂ still exhibits obvious deactivation, whereas the Co/SiO₂-PDVB system maintains stable performance. Furthermore, we physically mixed Co/SiO₂ with other hydrophobic materials, including hydrophobic graphite, methyl-modified silica (SiO₂-Me), and polytetrafluoroethylene (PTFE) (Supplementary Fig. 19). In all cases, we observed a similar enhancement in catalytic stability compared to the unmodified Co/SiO₂ catalyst (Fig. 1a, Supplementary Fig. 20, and Supplementary Table 6). These results indicate that the improved performance is not specific to PDVB, but instead arises from the introduction of hydrophobic domains into the catalyst bed.

Oxidation state of cobalt during catalysis

Transmission electron microscopy (TEM) images of the spent Co/SiO₂ showed cobalt nanoparticles with an average size of 28.5 nm, which is obviously larger than 13.1 nm of the as-reduced catalyst before reaction (Supplementary Figs. 21 and 22). This phenomenon is generally observed on supported cobalt nanoparticle catalysts, due to the redox cycle during CO cleavage and water production, which leads to dynamic changes in cobalt^15,17,42. Further high-resolution transmission electron microscopy (HRTEM) characterization revealed a distinct oxide layer surrounding the cobalt metal particles in the Co/SiO₂ catalyst (Fig. 2a, b). The corresponding inverse fast Fourier transform (IFFT) images showed that the oxide shell has a characteristic lattice spacing of the CoO(111) plane (Fig. 2e, green region), which is consistent with the general phenomena observed in cobalt catalysts in FTS¹⁷. The internal nanoparticle region has a characteristic lattice spacing of the metallic Co(111) plane (red region), which is unaffected by surface oxidation. In contrast, the surface CoO_x layers were minimized in the HRTEM images of the spent Co/SiO₂-P_mix catalyst (Fig. 2c, d). The corresponding IFFT images show some oxide regions on the cobalt surface (Fig. 2f, green region), but they are much less than those on the spent Co/SiO₂ without PDVB. Multiple regions on each sample were examined to ensure a representative assessment of the cobalt state, where similar results were obtained (Supplementary Figs. 23 and 24). This indicates that the PDVB-containing catalyst exhibits a lower extent of cobalt oxidation than Co/SiO₂. Furthermore, we investigated the evolution of cobalt oxidation states by H₂-TPR analysis of the spent catalysts, using hydrogen consumption as an indicator of reducible cobalt oxide species (Supplementary Fig. 25). The spent Co/SiO₂-P_mix catalyst shows a hydrogen consumption comparable to that of the fresh Co/SiO₂ catalyst, while the spent Co/SiO₂ catalyst exhibits a significantly higher hydrogen consumption. These results are consistent with the HRTEM observations and further support that PDVB could suppress cobalt oxidation under reaction conditions. During the FTS, the Co nanoparticles on the Co/SiO₂-P_mix catalyst have an average size of 20.7 nm (Supplementary Fig. 26)^43,44. Previous studies have shown that cobalt activity becomes much less sensitive to particle size once the cobalt crystallites exceed approximately 6–8 nm^44,45. The observed increase in particle size may contribute to some loss of activity. Indeed, a slight decrease in CO conversion was observed during the initial stage of the reaction (Supplementary Fig. 27). Although the spent Co/SiO₂ and Co/SiO₂-P_mix catalysts exhibited different average cobalt particle sizes (28.5 and 20.7 nm, respectively), such a difference in cobalt particle size alone cannot account for their substantial performance difference^45,46, where the spent Co/SiO₂ and Co/SiO₂-P_mix catalysts showed CO conversions of 7.4% (after 592 h) and 44.1% (after 1200 h), respectively.

Fig. 2. Characterization of spent cobalt catalysts.

HRTEM images of the spent (a, b) Co/SiO₂ and (c, d) Co/SiO₂-P_mix catalysts. e, f HRTEM images characterizing the spent Co/SiO₂ and Co/SiO₂-P_mix catalysts, and the corresponding IFFT images with colored images showing the distribution of CoO (green) and Co (red) in real space. g Normalized Co K-edge XANES spectra of spent catalysts after THF extraction and vacuum drying. h H₂-TPR profiles of Co/SiO₂ and Co/SiO₂-P_mix catalysts after a pre-treatment in ambient-pressure syngas for 100 h. i XRD patterns of as-synthesized Co/SiO₂, spent Co/SiO₂, and spent Co/SiO₂-P_mix catalysts, which show the hcp-to-fcc ratio through refinement via a full Rietveld analysis.

Figure 2g shows the X-ray absorption near-edge structure (XANES) spectra of the spent Co/SiO₂ and Co/SiO₂-P_mix catalysts, showing the curve falling between that of the CoO and Co foil. The spectrum of the spent Co/SiO₂-P_mix catalyst is close to the cobalt foil, while the spectrum of the spent Co/SiO₂ catalyst is located nearer to CoO. These data also confirm the maximized metallic state of cobalt on the spent Co/SiO₂-P_mix catalyst, while the cobalt on the Co/SiO₂ catalyst was more oxidized. To gain further insight into the effect of PDVB on cobalt oxidation by syngas, the cobalt catalysts were pretreated with atmospheric-pressure syngas for 100 h, then switched to hydrogen to collect the temperature-programmed reduction (TPR) signals⁴⁷. Although this condition differs from the high-pressure syngas in FTS reactions, it allows a direct comparison of the cobalt oxidation with and without PDVB. The syngas pretreated Co/SiO₂ showed a water signal (m/z of 18) at 248 °C assigned to the reduction of active lattice oxygen⁴⁸, while it is extremely weak on the syngas pretreated Co/SiO₂-P_mix (Fig. 2h). This result confirms efficient hindrance of the oxidation of metallic cobalt upon the addition of PDVB, in good agreement with TEM and XANES results.

Under FTS conditions, the redox cycle of cobalt can induce a phase transition from close-packed (hcp) to face-centered cubic (fcc), which in turn contributes to a decline in catalyst activity^16,49–52. We find that suppressing the detrimental effects of water markedly stabilizes the hcp cobalt phase. XRD analysis of the as-synthesized Co/SiO₂ catalyst shows an initial hcp-to-fcc ratio of 1.15, which decreases to 0.40 after 592 h of FTS reaction, reflecting pronounced hcp-to-fcc transformation (Fig. 2i and Supplementary Fig. 28). In contrast, the spent Co/SiO₂-P_mix catalyst retains an hcp-to-fcc ratio of 0.97 after reaction for 1200 h, indicating effective preservation of the hcp cobalt phase by mixing the hydrophobic promoter (Supplementary Fig. 29).

Rapid water escape

Previous studies employed the ratio of the partial pressures of water and hydrogen (p_H2O/p_H2) as an important indicator of the oxidation tendency of cobalt catalysts under FTS conditions^53,54. Under FTS with CO conversion of about 44%, the Co/SiO₂ catalyst gradually deactivated at a p_H2O/p_H2 ratio of about 0.40, whereas physical mixing with PDVB substantially mitigated this deactivation (Supplementary Figs. 30 and 31). To further probe the effect of water, additional water was introduced into the syngas feed to increase the p_H2O/p_H2 ratio to 0.52. Under these more water-rich conditions, the Co/SiO₂-P_mix catalyst exhibited remarkable water tolerance, maintaining a stable CO conversion of 34.7% (Fig. 3a). In contrast, the reference Co/SiO₂ catalyst under the same conditions underwent pronounced deactivation, with the CO conversion decreasing from 52.8% to 23.8% after water addition.

Fig. 3. Study on PDVB effect.

a Data showing the performances of Co/SiO₂ and Co/SiO₂-P_mix catalysts with artificially induced additional water in the syngas feed. Reaction conditions: catalyst granule at 40–60 mesh, 2.0 MPa of syngas (H₂/CO ratio at 2), 2400 mL g_cat⁻¹ h⁻¹, 220 °C. The water feed rate was ~78 mg g_cat⁻¹ h⁻¹. b H₂-TPR profiles of CoO_x/SiO₂ and a mixture of CoO_x/SiO₂ and PDVB. The humid hydrogen (H₂/H₂O/Ar of 9.7/3.1/87.2) was used. c Temperature-programmed water desorption profiles of Co/SiO₂ and Co/SiO₂-P_mix samples. d Normalized water responses after switching the feed gas from syngas to argon in the TKA test over Co/SiO₂ and Co/SiO₂-P_mix catalysts at 220 °C. e Water adsorption profiles of Co/SiO₂ and Co/SiO₂-P_mix catalysts. f, g In situ XANES spectra at Co K-edge showing the change of Co/SiO₂ and Co/SiO₂-P_mix catalysts in syngas containing water (H₂/CO/H₂O/Ar of 62/31/3.1/3.9) at 220 °C.

Cobalt oxidation arises not only from reaction with water but also from the accumulation of surface ^*O atoms after CO cleavage, if they are not promptly removed by hydrogen^39,55. To explore the role of PDVB in this process, H₂-TPR was performed under humid conditions to examine the reduction behavior of cobalt oxide on silica (CoO_x/SiO₂). The pristine CoO_x/SiO₂ catalyst exhibited hydrogen consumption starting at 319 °C, with a principal reduction peak at 379 °C (Fig. 3b and Supplementary Fig. 32). Conversely, the mixed sample containing PDVB demonstrated a markedly lower onset temperature of 283 °C and a main peak at 340 °C. This shift toward lower temperatures indicates that PDVB promotes the reductive removal of reactive oxygen species from cobalt oxide.

During FTS, water forms on the cobalt and then desorbs into the reaction atmosphere^15,56. We evaluate the effect of hydrophobic PDVB on water removal through water desorption experiments on water-saturated Co/SiO₂ (Fig. 3c and Supplementary Fig. 33). While both Co/SiO₂ and Co/SiO₂-P_mix catalysts behaved similarly at low temperatures (<60 °C), the full water desorption was achieved for Co/SiO₂-P_mix at 133 °C, at which temperature desorption on Co/SiO₂ was only 80%, requiring 160 °C for full desorption.

The kinetics of water escape were quantified via transient kinetic analysis (TKA)^57,58. Following catalyst pretreatment with syngas at 220 °C, which generated a detectable water signal confirming catalytic activity, the feed was switched to argon to collect water-escape signals. The subsequent decay of the water signal was monitored to determine the parameter τ, representing the water residence time. As shown in Fig. 3d, the signal decayed more rapidly for the Co/SiO₂-P_mix catalyst, yielding residence times of about 80 and 175 s for Co/SiO₂-P_mix and Co/SiO₂, respectively. The spatial proximity of PDVB to Co/SiO₂ was critical, with more intimate mixing within the granule mixture accelerating water diffusion (Supplementary Figs. 34 and 35). PDVB-promoted water desorption was directly visualized in a model dehydration experiment with CuSO₄·5H₂O, which undergoes a distinct color change from blue to white upon removal of crystallization water. When a small amount of PDVB (3.3 wt%) was physically mixed with CuSO₄·5H₂O, the dehydration‑induced color change was significantly accelerated under nitrogen purging at 100 °C (Supplementary Fig. 36a). In parallel, it was examined using a CoCl₂ color‑change experiment, in which CoCl₂ turns from blue to pink upon hydration by water vapor. As shown in Supplementary Fig. 36b, when CoCl₂ was placed at the same position, replacing quartz sand with PDVB resulted in a faster color change, indicating decreased average water accumulation in the presence of PDVB. Based on these results, the slightly lower initial activity of Co/SiO₂-P_mix may originate from the reduced local water concentration around the catalyst, because moderate amounts of water have previously been reported to promote FTS on silica-supported cobalt catalysts⁵⁹. Therefore, while enhanced water removal improves long-term catalyst stability, it may simultaneously reduce the beneficial contribution of water during the initial stage of the reaction.

Beyond facilitating rapid water removal from the catalyst, the PDVB also effectively suppresses the re-adsorption of ambient water onto the catalyst. This is suggested by markedly slower water-adsorption kinetics for the Co/SiO₂-P_mix compared to the Co/SiO₂ (Fig. 3e and Supplementary Fig. 37). In situ water-adsorption Fourier transform infrared (FTIR) spectra were collected by introducing water vapor to the pre-dehydrated catalysts (Supplementary Fig. 38). Characteristic bands at 1615 and 3500–3000 cm⁻¹ were enhanced on the Co/SiO₂ sample, corresponding to adsorbed water molecules and increased O-H stretching vibrations, respectively^60,61. In contrast, these spectral features were negligible on the Co/SiO₂-P_mix catalyst under equivalent conditions. X-ray photoelectron spectroscopy (XPS) studies further supported these observations. We performed a steam treatment on the Co/SiO₂ and Co/SiO₂-P_mix catalysts in an ambient-pressure gas (9.7H₂/3.1H₂O/87.2Ar) at 220 °C for 10 min. As evidenced by the Co 2p XPS spectra (Supplementary Fig. 39a), steam treatment led to the disappearance of the Co⁰ signal for the Co/SiO₂ catalyst, while XRD still clearly showed reflections characteristic of metallic Co and CoO_x (Supplementary Fig. 39b). In contrast, the Co⁰ signal remained detectable in the water-treated Co/SiO₂-P_mix catalyst, suggesting that the presence of PDVB minimizes cobalt oxidation.

To further probe changes in cobalt, the catalysts were exposed to a syngas-water vapor mixture, simulating the oxidation/reduction characteristics of the reaction atmosphere in the in situ XAS (Fig. 3f, g). Linear combination fitting (LCF) of the Co K-edge XANES spectra (Supplementary Figs. 40 and 41) was applied to quantify the fractions of metallic cobalt and cobalt oxides. The analysis indicated that the freshly reduced Co/SiO₂ catalyst contained about 71% metallic cobalt, which transiently decreased to about 33% after 12 min upon exposure to humid syngas. This initial change reflects the dynamic evolution of cobalt species during catalyst startup rather than a thermodynamic oxidation boundary. Upon extended testing, partial re-reduction by syngas was observed, restoring the metallic cobalt fraction to about 56% after 60 min. In contrast, the Co/SiO₂-P_mix showed a high fraction of metallic cobalt at about 71% with negligible change during the equivalent test.

Mechanism study

Previous studies have established that chemical grafting hydrophobic groups onto solid surfaces facilitates water desorption, which can be attributed to direct contact between the hydrophobic groups and the active sites^5,62. However, for physically mixed Co/SiO₂ and PDVB, a spatial distance exists between the hydrophilic catalyst and the hydrophobic polymer. It remains unclear how a hydrophobic surface, spatially separated from the hydrophilic Co/SiO₂, can optimize water diffusion across the hydrophilic domain.

The internal architecture of catalyst granules (40–60 mesh) comprising PDVB and Co/SiO₂ before and after reaction was examined using X-ray computed tomography (CT) with a monochromatic beam (Fig. 4a). This technique allows non-destructive visualization of the granule internal structure⁶³. Owing to their differing densities, PDVB and silica exhibit distinct contrast in the CT images, permitting clear identification of each component. Figure 4b, c reveal a random distribution of PDVB and Co/SiO₂ within the granule (Supplementary Movie 1), separated by interstitial channels with an average diameter of 4.0 ± 1.2 μm. A three-dimensional reconstruction of the pore network unveiled a highly interconnected channel system across the granule (Fig. 4d). After reaction, the internal structure of the spent catalyst remains essentially unchanged compared to the fresh sample (Supplementary Fig. 42). The channel walls consist of a combination of hydrophilic and hydrophobic surfaces. The hydrophilic surface, which has a higher surface free energy, creates regions with elevated local water vapor pressure (high-concentration region with higher chemical potential). Conversely, the hydrophobic surface, with its lower free energy, sustains areas of reduced water vapor pressure (low-concentration region with relatively lower chemical potential) (Fig. 4e). This disparity forms a spatial vapor pressure gradient, accelerating the diffusion of water from high chemical potential regions toward low chemical potential regions according to Fick’s first law⁶⁴.

Fig. 4. Water diffusion in a catalyst granule.

a Schematic diagram of X-ray CT experiments. b Schematic of the imaging plane at the middle-cross section in one Co/SiO₂-P_mix granule with powder mixture of Co/SiO₂ and PDVB. c Image of the two-dimensional slice showing the distribution of Co/SiO₂ (white) and PDVB (black). d Three-dimensional rendering of Co/SiO₂-P_mix granule. Co/SiO₂ (yellow) and PDVB (blue). e Scheme showing the water sorption equilibrium on the hydrophilic catalyst surface, water concentration gradient in the free region, and rapid water escape close to the hydrophobic surface. f Enhanced water transport model showing the water concentration map without and with a hydrophobic promoter. The pink dashed rectangle highlights the hydrophobic channel region (in the case of a hydrophobic promoter). The black circles indicate the concentration fronts, with relative concentrations of 67.8% (C₁) and 35.7% (C₂) of the injection concentration (C₀). The models were obtained using OpenPNM.

Moreover, the adsorption-desorption equilibrium of water on hydrophilic surfaces is influenced by local water vapor pressure, according to Langmuir equation⁶⁵; therefore, facilitating water escape from high-concentration region promotes a shift in this equilibrium on the hydrophilic surface (Fig. 4e). These observations suggest that hydrophilic-hydrophobic contrasts within the catalyst granules facilitate water transport and reduce water accumulation around cobalt particles, which is in good agreement with our experimental observations. Concurrently, water diffusion on hydrophobic surfaces is superfast compared with that in the general system, because of the extremely low effective viscosity of these water molecules^66,67. This feature significantly enhances overall water-diffusion efficiency in the catalyst granules within the continuous-flow reactor. Indeed, similar phenomena are commonly observed in nature, where hydrophilic-hydrophobic surface contrasts facilitate molecular transport and promote the removal of water from hydrophobic regions^68,69.

We developed a directional water diffusion model through a widely used pore network modeling package (OpenPNM)⁷⁰ to study the influence of hydrophobic granules on the water diffusion in the reactor. As shown in Fig. 4f, the water molecules were injected from the left side, diffused through the model in the parallel channels, and escaped from the right side. In some of the channels, part of the walls was made hydrophobic (pink dashed rectangle), simulating the presence of hydrophobic catalyst granules. This model can visually represent the apparent water diffusion efficiency with hydrophobic regions in the reactor. In the model without hydrophobic regions, water diffusion is equivalent across all channels, resulting in a water concentration gradient from left to right. With the presence of hydrophobic regions, the water concentration in both the hydrophobic region and the surrounding areas is significantly reduced. To facilitate comparison of concentration distributions, two concentration fronts (C₁/C₀ at 67.8% and C₂/C₀ at 35.7%) are indicated by black circles. Upon inducing hydrophobic channel walls, both concentration fronts evolve from straight to curved profiles and exhibit a pronounced leftward shift. This phenomenon is attributed to accelerated water transport in the hydrophobic region, thereby reducing water concentration.

In summary, this work resolves a long-standing challenge in cobalt-based Fischer-Tropsch synthesis by demonstrating that cobalt catalyst deactivation can be effectively suppressed through engineering control of water transport. The PDVB-mixed cobalt catalyst exhibits fully stable performance over 1200 h of continuous operation, maintaining 88–90% selectivity toward C₅₊ hydrocarbons. These results establish that even highly water-sensitive metallic cobalt catalysts can be durably stabilized by regulating local water diffusion, rather than by modifying catalyst composition.

Compared with conventional stabilization strategies that rely on noble metals, this approach employs a low-cost, practically available hydrophobic material and is readily compatible with existing catalyst formulations, offering clear advantages for potentially practical use. In addition, the results suggest that hydrophilic-hydrophobic contrasts within catalyst granules facilitate water transport and reduce water accumulation around cobalt particles, thereby mitigating water-induced deactivation. More broadly, this study highlights local water transport as an important and tunable factor in stabilizing catalysts operating in water-rich reaction environments.

Methods

Materials

Co(NO₃)₂⋅6H₂O (99%), divinylbenzene, azobis isobutyronitrile, and fumed nanosilica were obtained from Shanghai Aladdin Co., Ltd. The gases were supplied by Hangzhou Jingong Special Gas Co., Ltd.

Synthesis of Co/SiO₂

1.0 g of fumed nanosilica was dispersed in 5 mL aqueous cobalt (Ⅱ) nitrate solution (0.68 mol/L), then the liquor was dried at 120 °C under milling to evaporate the water. The obtained solid powder was calcined in air at 400 °C for 4 h (temperature ramping rate of 2 °C/min). The obtained sample was noted as CoO_x/SiO₂. After reducing in flowing hydrogen at 400 °C for 6 h (temperature ramping rate of 1 °C/min) and passivating in 1% O₂/N₂ at 50 °C for 30 min, the sample of Co/SiO₂ was finally obtained. The cobalt loading was 13.2 wt% by an inductively coupled plasma optical emission spectrometer (ICP-OES) analysis.

Synthesis of PDVB

Five hundred milligrams of azobis isobutyronitrile was added to 10 g of divinylbenzene under stirring at room temperature for 1.0 h, then the liquor was transferred to an autoclave and thermally treated at 100 °C for 24 h. The obtained solid was washed with ethanol and dried at 100 °C overnight to obtain non-porous polydivinylbenzene (PDVB).

Catalyst characterization

Powder X-ray diffraction (XRD) profiles were collected on a Rigaku D/MAX 2550 and Bruker D8 Advance diffractometer with Cu Kα radiation (λ = 1.5418 Å). Nitrogen sorption isotherms were measured using a High-Performance Microporous Analyzer (UtraSorb 4802 system). The specific surface area (S_BET) was calculated using the Brunauer-Emmett-Teller (BET) method. Transmission electron microscopy (TEM) images were recorded on a JEM-2100F electron microscope (JEOL, Japan) operated at 200 kV. High-resolution transmission electron microscopy (HRTEM) images were obtained using an FEI Titan G2 80-200 ChemiSTEM microscope equipped with a spherical aberration corrector, operated at 200 kV. The probe convergence angle was set to about 21 mrad, and the annular detection angle of BF images was about 0–20 mrad. The low-order aberrations were adjusted to an acceptable level, e.g., C_s < 0.5 μm, A1 < 2 nm, A2 < 20 nm, B2 < 20 nm. Scanning electron microscopy (SEM) characterizations were performed with Hitachi SU-8010 electron microscopes. Metal loadings of the catalysts were determined by an inductively coupled plasma optical emission spectrometer (ICP-OES, PerkinElmer Avio^TM 200). Fourier Transform infrared spectroscopy (FTIR) characterization was performed with the Thermo Nicolet NEXUS 470 FT-IR system. Water droplet contact angles were measured with a video-based contact angle measuring device (OCA 20, Dataphysics Co. Ltd). The water vapor adsorption and desorption measurement was performed by a gravimetric vapor adsorption analyzer (3H-2000PW, BeiShiDe Instrument Co. Ltd., China). Thermogravimetric curves (TG) were performed on a NETZSCH STA 449 F5 Jupiter® simultaneous TG-DSC in flowing air with a heating rate of 10 °C/min. The X-ray photoelectron spectra of the catalysts were recorded on a Kratos Amicus spectrometer using Al Kα radiation (1486.6 eV). The measurement was conducted on the ThermoFisher ESCALAB Qxi spectrometer under a high vacuum at ambient temperature. To obtain the XPS spectra, the pressure of the analysis chamber was maintained at 1 × 10⁻⁶ Pa.

Temperature-programmed reduction with hydrogen (H₂-TPR) was performed using a fixed-bed glass reactor connected to a mass spectrometer (SRD200M, TILON GRP TECHNOLOGY LIMITED). 0.05 g of CoO_x/SiO₂ was localized in the middle of the reactor tube (40–60 mesh) and pre-treated under a pure Ar flow at 400 °C for 30 min (flow rate at 30 mL/min). Then, the temperature cooled down to 50 °C. The hydrogen and water signals (m/z at 2 and 18) were collected in flowing 10% H₂/Ar (flow rate 10 mL/min, humidified with water at room temperature) by raising the temperature from 50 to 700 °C at a ramping rate of 10 °C/min.

Water-adsorption Fourier transform infrared spectra (FTIR) were collected using an FTIR spectrometer with an in situ cell (Thermo, Nicolet 6700). In a typical run, 16 mg of catalyst was pre-treated under 10% H₂/Ar at 300 °C for 30 min. The system was cooled down in flowing Ar, and the background spectra were recorded at different temperatures. Then, the syngas was introduced to the catalyst at 220 °C for 1.0 h with a flow rate of 20 mL/min under atmospheric pressure, and the spectra were recorded every 3 min. Water vapor was introduced into the system by controlling the feed gas flowing through water at room temperature.

In situ X-ray absorption fine structure (XAFS) spectra at Co K-edge were obtained on the 1W1B beamline of Beijing Synchrotron Radiation Facility (BSRF) operated at 2.5 GeV and 250 mA. The measurement was performed at 300 °C in a 10% H₂/Ar flow or at 220 °C in a syngas flow (H₂/CO = 2). The Linear Combination Fitting (LCF) range was from −20 to 30 eV, based on E0, to calculate the component fraction content. Ex situ XAFS spectra at Co K-edge (E₀ = 7709 eV) were collected at the TableXAFS-500. The XAFS data were recorded in transmission mode using a standard Lytle ion chamber. The energy was calibrated according to the absorption edge of pure Co foil. The data were extracted, and the profiles were fitted according to the Athena and Artemis codes. X-ray Computed Tomography (CT) experiments were performed at station BL13HB of Shanghai Synchrotron Radiation Facility (SSRF) using a monochromatic X-ray beam (energy = 15 KeV). Throughout, the catalysts were supported in the pipette tip. For characterizing the spent catalysts after FTS, the wax was removed, followed by vacuum drying, then the samples were handled in a nitrogen-filled glovebox and loaded into the XANES sample holder. The sample was sealed with a Kapton film during transfer and measurement, thereby minimizing exposure to air.

Transient kinetic analysis (TKA) was performed on a tubular reactor connected to an online mass spectrometer. Typically, the catalysts were reduced in a hydrogen flow (50 mL/min) at 300  °C for 0.5 h. Then the samples were treated at 220 °C in syngas flow (20 mL/min) for reaction until the water signal (m/z of 18) is constant. Ar flow (20 mL/min) was then introduced to replace syngas by quickly switching through a 4-way valve. Signals characterizing the water desorption from the catalyst granules were collected.

Catalytic tests in FTS

The feed gas conversion was performed in a fixed-bed flow reactor with an inner diameter of 8 mm. Typically, 0.5 g of Co/SiO₂ catalyst (40–60 mesh) and 0.025 g of PDVB (40–60 mesh) were mixed and localized in the middle of the reactor unless otherwise stated. Additional quartz sand was placed in the reactor to maintain the catalyst bed. The reactor was heated to 220 °C, and the feed gas (H₂/CO/Ar molar ratio at 64/32/4) was introduced into the reactor under a pressure of 2 MPa. The gaseous products from the reactor were analyzed using two online gas chromatographs (Fu Li-9790). One is equipped with a TDX-01 column (3 m × 3 mm) and a thermal conductivity detector (TCD), and the other is equipped with a capillary PLOT-Al₂O₃ column (50 m × 0.53 mm × 0.25 μm) and a flame ionization detector (FID). Argon was used as an internal standard for calculating CO conversion and product selectivity (CO₂ and CH₄) in TCD-GC. The liquid-phase products collected by the cold and hot traps were analyzed using an offline Shimadzu GC-14C gas chromatograph equipped with an HP-5 column (60 m × 0.32 mm × 0.25 μm) and FID. Product selectivity was calculated based on the number of carbon atoms, and the total carbon balance was typically greater than 95%. To study the effect of water on catalysts, water vapor was introduced into the system by switching the valve to flow through the bubble bottle.

After the FTS reaction, the reactor was cooled to room temperature under an inert gas flow. The spent catalyst granules were collected and transferred into a vacuum bag. The wax products were removed by washing the spent catalyst with THF at room temperature for 30 min. The washing procedure was followed by vacuum drying at room temperature for 3 h.

Computational Methods

Pore network modelling (PNM) was conducted to elucidate the water transport behavior on the mixed PDVB and Co/SiO₂ catalyst granule. The corresponding pore-network information is derived from CT image analysis. The simulation was performed using an open-source PNM package (OpenPNM)¹. The pore network was configured with the following parameters: pore spacing of 200 nm, throat length of 100 nm, pore diameter of 100 nm, throat diameter of 25 nm, and a total of 50 × 50 pores. For reaction-diffusion settings, water enters the network at the left boundary via a reaction with a rate of R_i = 10⁻⁴C_i $\cdot {\pi \mathrm{D}}_{\mathrm{i}}^2$, while the right boundary maintains a concentration of 0.1 mol/m³. The diffusion coefficient of water in the pore network is 2.2 e⁻⁷m²/s. No reactions occur in other pores, and throats are diffusion-only. Sixteen interparticle pores, with a pore diameter of 5 μm, a throat length of 20 μm, and a throat diameter of 25 μm, are randomly distributed. The diffusion coefficient of water in the interparticle pores is 4.7 e⁻⁷m²/s. The diffusion coefficient was set based on previous studies². Mass transfer follows Fick’s law: ${\text{q}}_{\text{i}}=\frac{{\text{D}}_{\text{i}}{\text{A}}_{\text{i}}\mathrm{\Delta }{\text{C}}_{\text{i}}}{{\text{L}}_{\text{i}}}$, ensuring equal inflow and outflow rates under the same concentration difference.

Author contributions

W.F. performed the catalyst preparation, characterization, and catalytic tests. Y.H., C.Z., and Z.L. performed the diffusion simulation and wrote the corresponding part. S.C. performed the XANES and EXAFS characterization and analyzed the data. Y.C. and Y.J. performed the HRTEM characterization and wrote the corresponding part. Z.W. and Y.Z. performed the TKA characterization and analyzed the data. H.J.L. participated in the characterization, catalysis, and discussion. Y.D., G.-H.D., and L.L. processed the X-ray CT characterization. L.B. and P.M. participated in catalysis to reconfirm the PDVB effect and provided helpful discussions. S.Y., W.Y., and P.H. provided helpful discussions on the catalyst structure and reaction process. J.D., S.Z., and W.T. provided helpful discussion. L.W. designed the study, analyzed the data, and wrote the paper. All the authors discussed the results and improved the manuscript.

Peer review

Peer review information

Nature Communications thanks Amol P. Amrute and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the National Key Research and Development Program of China (2022YFA1503502), the National Natural Science Foundation of China (22525204, 22521204, 22401275), the National Postdoctoral Program for Innovative Talents (BX20240303), the China Postdoctoral Science Foundation (2025M770945), the Key R&D Program of Zhejiang (2024SSYS0067) and the Science and Technology Project of Zhejiang Provincial Energy Group Co., Ltd. (ZNKJ-2023-118).

Data availability

All data that led us to understand the results presented here are available in the paper or from the corresponding author upon request. Source data are provided with this paper.

Competing interests

The authors declare that there is no competing interest.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76571-8.