Selective, High-Temperature O₂ Adsorption in Chemically Reduced, Redox-Active Iron-Pyrazolate Metal–Organic Frameworks

Abstract

Developing O₂-selective adsorbents that can produce high-purity oxygen from air remains a significant challenge. Here, we show that chemically reduced metal–organic framework materials of the type A_xFe₂(bdp)₃ (A = Na⁺, K⁺; bdp²⁻ = 1,4-benzenedipyrazolate; 0 < x ≤ 2), which feature coordinatively saturated iron centers, are capable of strong and selective adsorption of O₂ over N₂ at ambient (25 °C) or even elevated (200 °C) temperature. A combination of gas adsorption analysis, single-crystal X-ray diffraction, magnetic susceptibility measurements, and a range of spectroscopic methods, including ²³Na solid-state NMR, Mössbauer, and X-ray photoelectron spectroscopies, are employed as probes of O₂ uptake. Significantly, the results support a selective adsorption mechanism involving outer-sphere electron transfer from the framework to form superoxide species, which are subsequently stabilized by intercalated alkali metal cations that reside in the one-dimensional triangular pores of the structure. We further demonstrate similar O₂ uptake behavior to that of A_xFe₂(bdp)₃ in an expanded-pore framework analogue and thereby gain additional insight into the O₂ adsorption mechanism. The chemical reduction of a robust metal–organic framework to render it capable of binding O₂ through such an outer-sphere electron transfer mechanism represents a promising and underexplored strategy for the design of next-generation O₂ adsorbents.

Graphical Abstract

Introduction

The isolation of high-purity oxygen from air is vital for pre-combustion (i.e., carbonaceous fuel gasification) and post-combustion (i.e., oxy-fuel combustion) carbon capture technologies,¹ as well as for the steel, medical, chemical, food, glass, and waste-treatment industries.² Currently, purification methods such as cryogenic distillation are carried out on a large scale in industry, although these separations require enormous energy inputs.^2,3 On smaller scales, membranes can increase O₂ concentrations relative to air, but they are typically capable of only ~50% enrichment.³ Porous adsorbents stand as attractive alternatives for O₂ purification, given that they can operate with high energy efficiencies and therefore at lower cost than low-temperature methods. However, the cation-exchanged zeolites currently employed in adsorbent-based air separation are generally N₂-selective^4,5 and must be regenerated frequently, owing to the larger nitrogen fraction in air (78%) compared with oxygen (21%). Furthermore, the purity of oxygen derived from these zeolite adsorbents is also typically limited to ≤95%.^3,5

The development of an O₂-seleetive adsorbent, especially one capable of separating oxygen at ambient or elevated temperature, could both enhance efficiency for this important industrial separation and facilitate technologies that mediate CO₂ release into the atmosphere. Indeed, in pre-combustion carbon capture—where carbonaceous fuel is gasified using high-purity O₂, converted to syngas, and combusted to power a turbine—significant efficiency gains are achieved by feeding air to the separation unit directly from the turbine compressor.^3,6,7 In this integrated gasification combined cycle, air from the turbine compressor can exceed 300 °C due to compressive heating, and this heat must be rejected before entering the separation unit.³ Consequently, substantial energy savings could be achieved using an air separation unit capable of operating well above ambient temperature.

The design of O₂-selective adsorbents is particularly challenging given the similar physical properties of O₂ and N₂, such as kinetic diameter, polarizability, and quadrupole moment.⁸ Nitrogen is slightly greater in both polarizability and quadrupole moment—factors that are exploited by N₂-selective zeolites. Redox activity, however, is perhaps the most powerful characteristic of O₂ that distinguishes it from N₂. Indeed, biological systems leverage strategies based on redox-activity to reversibly bind dioxygen,⁹ and similar behavior has been engineered in synthetic complexes^10–12 and porous metal–organic frameworks through the use of coordinatively unsaturated, redox-active metal centers that provide open binding sites for O₂. Depending on the electronic properties of the metal centers and the coordination environments in these systems, O₂ can be reduced to either a superoxo (O₂⁻) or peroxo (O₂²⁻) species and can exhibit a variety of binding modes.

Given their high tunability, crystallinity, and chemical versatility,^8,13–22 metal–organic frameworks can provide appealing platforms for the design of O₂-seleetive adsorbents. Indeed, frameworks such as Cr₃(btc)₂ (btc³⁻ = 1,3,5-benzenetricarboxylate),²³ Cr-BTT (BTT³⁻ = 1,3,5-benzenetristetrazolate),²⁴ Fe₂(dobdc) (dobdc⁴⁻ = 2,5-dioxido-1,4-benzenedicarboxylate),²⁵ Co-BTTri (H₃BTTri = 1,3,5-tri(1H-1,2,3-triazol-5-yl)benzene),²⁶ Fe-BTTri,²⁷ PCN-224Mn^II,28 and Co₂(OH)₂(bbta) (H₂bbta = 1H,5H-benzo(1,2-d:4,5-d′)bistriazole)^29,30 have shown high selectivities and capacities for O₂. However, oxygen binding in many of these materials is either irreversible or very weak at ambient temperature, and the frameworks tend to suffer from poor thermal stability, capacity loss during cycling, or framework degradation under humid conditions. Given that only a small fraction of all reported metal–organic frameworks feature open metal sites and the tunable design of such frameworks remains a considerable challenge, it is crucial to explore alternate design strategies.

To improve the energy efficiency of air separation processes, an ideal adsorbent would be highly selective for O₂ and stable at ambient and even elevated temperature. In seeking underexplored O₂ adsorption routes, we considered the possibility of outer-sphere electron transfer from coordinatively saturated, redox-active metal centers (Figure 1). Here, post-synthetic chemical reduction of a stable framework is expected to generate a material capable of reducing O₂ that would also feature charge-balancing cations for stabilizing the reduced O₂ species. Beyond the choice of metal and ligand, these cations could also offer an additional functional handle for tuning adsorption properties. Such a strategy requires a material with redox-active centers, the potential for topotactic insertion of charge-balancing cations, chemical resistance to reactive O₂ⁿ⁻ species, and high thermal stability. We therefore turned to the framework Fe₂(bdp)₃ (bdp²⁻ = 1,4-benzenedipyrazolate),³¹ which is known to undergo chemical reduction with potassium naphthalenide to yield K_xFe₂(bdp)₃ (0 < x ≤ 2).³² This material and a related Fe^II-tetrazolate framework have been described as exhibiting reactivity in air,^32,33 but their O₂ adsorption properties were not investigated further. Herein, we show that the materials A_xFe₂(bdp)₃ (A = Na⁺, K⁺; 0 < x ≤ 2, Figure 1) are capable of selectively adsorbing O₂ over N₂ with high capacities at room temperature and as high as 200 °C and that they can be partially regenerated using heat and vacuum. Comprehensive characterization methods, including gas adsorption measurements, single-crystal X-ray diffraction, and numerous solid-state spectroscopies, provide evidence that O₂ is reduced to a superoxo species upon adsorption, ostensibly via an outer-sphere electron transfer mechanism. These results are the first illustration of the use of chemical reduction of a stable framework to generate new high-performance adsorbents capable of exceptionally selective O₂ capture.

Figure 1.

X-ray crystal structures of Fe₂(bdp)₃ (left), Na_0.5Fe₂(bdp)₃ (middle), and room-temperature O₂-dosed Na_1.2Fe₂(bdp)₃ (right). Orange, blue, gray, red, and purple spheres represent Fe, N, C, O, and Na atoms, respectively. Disordered atoms and H atoms are omitted for clarity.

Results and Discussion

Synthesis and Characterization.

The synthesis of Fe₂(bdp)₃ was performed following the previously reported procedure³¹ to afford a black microcrystalline solid. The structure of this material consists of one-dimensional μ²-pyrazolate-bridged chains of octahedrally coordinated iron(III) nodes, connected in three dimensions by bdp²⁻ linkers to yield a rigid framework with triangular channels (Figure 1). Importantly, the strong metal–pyrazolate bonds and structural rigidity of the framework should serve to prevent coordinative reorganization or material degradation upon O₂ adsorption.

Subsequent reduction of Fe₂(bdp)₃ with sodium or potassium naphthalenide in tetrahydrofuran yielded A_xFe₂(bdp)₃ (A = Na⁺, K⁺; 0 < x ≤ 2) in a topotactic manner, as previously described.³² Langmuir surface areas of 750–790 m²/g were reliably obtained for the half-reduced framework materials AFe₂(bdp)₃ following activation at 180 °C, whereas the fully reduced compound A₂Fe₂(bdp)₃ was found to be essentially non-porous to N₂. We therefore narrowed our initial focus to the half-reduced form of this framework due to its greater accessible porosity.³² Additionally, on the basis of its electrochemical behavior, AFe₂(bdp)₃ is less reducing than A₂Fe₂(bdp)₃,³² which should favor the formation of superoxo rather than peroxo species. We note that, because superoxide is a one-electron reduction product of O₂, a material that favors an alkali-stabilized superoxo species in its pores has twice the theoretical capacity of one that generates a peroxo species. Superoxo binding is also more likely to be reversible than peroxo binding, and indeed peroxo binding is often irreversible—as has been observed at elevated temperature for Fe₂(dobdc).²⁵ We further focused our initial studies on the potassiated congener, due to the relative stability of potassium superoxide compared to sodium superoxide.^34–36

O₂ Adsorption in Reduced Fe₂(bdp)₃.

We initially probed the interaction of reduced Fe₂(bdp)₃ with dioxygen by measuring the O₂ adsorption isotherm of K_1.09Fe₂(bdp)₃ at 298 K (Figure 2). Powder X-ray diffraction data collected at this temperature confirmed no loss of crystallinity or change in symmetry upon O₂ dosing (Figure S1). At low pressures, the material exhibits extraordinarily steep uptake of O₂ (Figure 2), achieving a loading of 0.44 mmol/g at just 1 mbar O₂. The O₂ uptake thereafter rises only gradually with increasing pressure, resulting in a loading of 0.51 mmol/g at 0.21 bar—close to the partial pressure of O₂ in air—and a maximum loading of 0.68 mmol/g at 1 bar. The steep O₂ uptake is suggestive of strong initial adsorption, whereas in contrast, K_1.09Fe₂(bdp)₃ exhibits little N₂ adsorption under these conditions, consistent with only weak physisorptive guest–framework interactions. The O₂ and N₂ adsorption isotherms were modeled using multi-site and single-site Langmuir-Freundlich equations, respectively (Figure S2; see details in Supporting Information), and the strong interaction of the framework with O₂ is exemplified by far higher Langmuir parameter values (Table S1). Given the stark differences in the O₂ and N₂ adsorption profiles, we would further expect near-perfect selectivity for O₂ over N₂ under the measured conditions. Many empirical and theoretical frameworks for describing adsorption, such as Ideal Adsorbed Solution Theory, poorly predict adsorption equilibria for mixtures containing adsorbates with substantially differing adsorption interactions (e.g., chemisorption compared with physisorption) and for adsorbents with heterogenous surfaces (e.g., cation-exchanged zeolites).^37,38 Nevertheless, a full discussion of multiple methods we used to quantify selectivity and strength of O₂ binding is available in the Supporting Information. These calculations show that O₂ can be efficiently separated from N₂ even at very low O₂ partial pressure. Furthermore, given the proper conditions for regeneration of this material after air separation, it is likely that O₂ with a purity greater than 99.9% could be generated.

Figure 2.

Adsorption isotherms for the uptake of O₂ and N₂ at 298 K (red and blue circles, respectively) and O₂ at 473 K (triangles) in K_1.09Fe₂(bdp)₃.

The framework uptake of 0.51 mmol/g (0.40 molecules of O₂ per formula unit, Figure S4) at 0.21 bar of O₂ corresponds to approximately 40% of the theoretical capacity (1.40 mmol/g) given a stoichiometry of K_1.09Fe₂(bdp)₃ and assuming one-electron reduction of O₂ to form superoxo species. This result implies that either a more reduced dioxygen species is being formed or there is a kinetic barrier to complete O₂ loading. Indeed, if the O₂ uptake at ambient temperature is kinetically limited, a substantial increase in available thermal energy could surmount the activation barrier and enable access to the full capacity of the material, assuming no change in the mechanism of adsorption. Examples of kinetic limitations could include hindered diffusion through the narrow triangular framework channels due to occlusion by reduced O₂ species, an activation barrier toward rearrangement of alkali cations upon introduction of O₂, sluggish movement of reduced O₂ species to preferred binding sites, or even a barrier to electron transfer.

Seeking to explain this partial loading, as well as to test the chemical stability of the material, we further measured O₂ adsorption at 473 K (200 °C). Significantly, K_1.09Fe₂(bdp)₃ retains its strong affinity for O₂ at this temperature and displays an enhanced adsorption capacity (Figure 2), achieving a loading of 0.98 mmol/g at 10 mbar. Subsequent dosing yielded loadings of 1.11 and 1.32 mmol/g at close to 0.21 and 1.0 bar, respectively. The uptake at 0.21 bar of O₂ corresponds to 0.86 molecules of O₂ per formula unit (Figure S4) or ~80% of the theoretical capacity, which critically rules out the formation of a peroxo species, at least at 473 K. Based on these data, we again expect the O₂/N₂ selectivity of K_1.09Fe₂(bdp)₃ at 473 K to be extraordinarily high. The apparent selectivity of this material for O₂ at ambient temperature and the substantial increase in capacity at such a high temperature ultimately indicate its great promise for O₂ separations.

Desorption isotherms collected for K_1.09Fe₂(bdp)₃ at 298 K show only the release of weakly bound O₂ (Figure S5), while the majority remains strongly adsorbed. A small amount of hysteresis is observed, likely due to strong but highly kinetically limited binding of O₂ during the shallow uptake region of the adsorption isotherm that is not removed upon desorption. However, upon heating to 453 K (180 °C) under vacuum after an adsorption and desorption isotherm cycle, the material can be partially regenerated. Over at least five adsorption/desorption cycles between 0 and 0.21 bar at 298 K followed by activation, the quantity of O₂ adsorbed during each adsorption isotherm appears to reach an asymptotic value of ~0.2 mmol/g (Figure S6). We note that the regeneration conditions were not optimized, and indeed we found that heating to 478 K (205 °C) under vacuum resulted in greater capacity recovery. Thus, higher regeneration temperatures and longer regeneration times may allow for more of the capacity of the material to be recovered. In fact, desorption at 473 K appears to be more reversible in the low-pressure region than at 298 K (Figure S7), and impressively the material can be cycled at least 10 times at these elevated temperatures, albeit with diminished capacities (Figure S8).

We also measured 298 and 473 K O₂ adsorption isotherms for Na_1.04Fe₂(bdp)₃ (Figure S9), which displayed almost identical behavior to the potassiated version. Thus, the stability of the reduced O₂ species within the framework pores appears not to be strongly dependent on the alkali metal cation, although complex cations could show different behavior.

The foregoing results imply that the adsorption of O₂ in A_xFe₂(bdp)₃ is either under kinetic control to some degree or that the mechanism of O₂ adsorption is altered at elevated temperature. We therefore sought a deeper understanding of (i) the lower O₂ uptake at ambient temperature—for example, whether O₂ diffusion is limited as described above and (ii) the nature of the adsorbed O₂ species.

Single-Crystal X-ray Structure Determinations.

Single-crystal X-ray diffraction was employed to determine the nature of the adsorbed O₂ species in A_xFe₂(bdp)₃ and the adsorption mechanism. Dark, acicular, X-ray quality single crystals of Fe₂(bdp)₃ were synthesized using a modification³⁹ of the original synthetic procedure (see Supporting Information).³² Following chemical reduction, activation, and room temperature O₂ dosing of the crystals (details available in the Supporting Information), we obtained structures of both Na- and K-reduced Fe₂(bdp)₃ (Figures 1 and 3). Virtually the same Fe–N bond lengths and ligand metrical parameters were found for reduced Fe₂(bdp)₃ and its O₂-dosed form; however, the pores of the latter structure are clearly occupied by O₂ species stabilized by intercalated cations. We note that the absence of observable O₂ species near the iron centers also helps refute the possibility that O₂ binding is simply occurring at defect sites. Attempts to collect structural data following O₂ dosing at 473 K revealed that the material was not sufficiently crystalline for a structure determination.

Figure 3.

Expanded sideview along one pore of K_0.74Fe₂(bdp)₃ dosed with 1 bar O₂ at 298 K. Orange, gray, blue, red, and green spheres represent Fe, C, N, O, and K atoms, respectively. Both crystallographically distinct K sites are shown. Disordered atoms created by symmetry and hydrogen atoms are omitted for clarity.

The diffraction data collected for crystals dosed with O₂ at 298 K were best modeled assuming two crystallographically unique Na⁺ or K⁺ ion sites. One position is very similar to the cation site near the phenyl groups of the ligand in the reduced, activated structures.³⁹ Indeed, for both O₂-dosed structures, the alkali metal ion⋯phenyl-centroid distances are 3.44(2) Å, compared with 3.429(4) Å in activated Na_0.5Fe₂(bdp)₃. The second alkali metal position—with longer alkali metal ion⋯phenyl-centroid distances of 3.87(2) and 3.88(2) Å for sodium and potassium, respectively—appears to stabilize the O₂ species, with a Na–O distance of 2.30(3) Å and a K–O distance of 2.31(3) Å. The distances between the two crystallographically distinct alkali sites are 1.21(3) Å (sodiated structure) and 1.24(3) Å (potassiated structure). The O₂ species and alkali metal ion sites reside near an inversion center at the middle of the triangular pore and are thus duplicated by symmetry. Importantly, the O–O distances of 1.29(6) and 1.34(6) Å in the sodiated and potassiated structures, respectively, are consistent with reported bond lengths for superoxo species.³⁶ The closest pyrazolate N⋯O distances are 5.81(2) and 5.79(2) Å for A = Na⁺ and K⁺, respectively, whereas the corresponding Fe⋯O distances are 7.16(2) and 7.14(2) Å. Assuming that outer-sphere electron transfer occurs at a point of closer contact, these relatively large distances between the iron–nitrogen coordination sphere and oxygen atoms imply that substantial rearrangement and movement of the reduced O₂ species must occur following electron transfer.

We note that both the crystal symmetry and stoichiometry necessitate partial occupancy of the alkali metal and O atom crystallographic sites. Additional factors, such as structural disorder, the proximity of adsorbed O₂ to a point of high symmetry in the center of the pore, large thermal motion, and the relatively low electron density of these species, make a precise determination of the O–O bond length and other distances difficult. However, these structures conclusively show the presence of adsorbed O₂ species and are consistent with an adsorption mechanism involving one-electron reduction of O₂.

Analysis of Extra-Framework Species.

Vibrational spectroscopy is often used to probe the nature of reduced O₂ species given that the bond order and therefore the vibrational frequency of the O–O bond are distinct for O₂⁻ and O₂⁻. However, despite numerous attempts, we were unable to assign any O–O signatures using Raman or in situ diffuse reflectance infrared spectroscopy (see the Supporting Information for details). We therefore utilized X-ray photoelectron spectroscopy (XPS) and solid-state NMR spectroscopy coupled with density functional theory (DFT) calculations to further elucidate the nature of the O₂ species adsorbed in AFe₂(bdp)₃ and the associated chemical environment of the alkali metal cations.

Owing to the strongly bound nature of the O₂ species in these materials, we were able to directly compare the O 1s signals in Fe₂(bdp)₃, K_1.06Fe₂(bdp)₃, and K_1.06Fe₂(bdp)₃ dosed with 1 bar O₂ at either 298 or 473 K, under the high vacuum of the XPS measurement chamber. As previously reported,³² Fe₂(bdp)₃ contains defects likely associated with ligand vacancies. For this material, we observe an O 1s signal that we accordingly assign to oxygen-containing defect species at a binding energy of 531.1 eV (Figure 4), consistent with hydroxides or oxygen-containing organics such as formate.^40–45 This peak is also observed at a similar binding energy in K_1.06Fe₂(bdp)₃, as well as in both O₂-dosed K_1.06Fe₂(bdp)₃ samples. Critically, for both of these O₂-dosed samples, a new O 1s peak is also present at a higher binding energy of ~534.1 eV, consistent with a superoxo, rather than more reduced peroxo or oxo species.^{40,43,45–48} The observation of this new peak for both samples suggests that (i) the nature of the reduced, adsorbed O₂ species is the same, regardless of dosing temperature, and (ii) the mechanism of O₂ adsorption is therefore likely also the same. As such, differences in O₂ adsorption capacity at 298 and 473 K most likely arise from kinetics effects. Furthermore, the relative area of this higher energy peak is greater for the sample dosed with O₂ at high temperature, as expected given the greater O₂ loading with increasing temperature. As further corroboration of this signal assignment, we measured the O 1s spectrum of potassium superoxide (KO₂) and observe a similar binding of energy of 533.8 eV (Figure S10).

Figure 4.

Oxygen 1s XPS spectra for (a) Fe₂(bdp)₃, (b) K_1.06Fe₂(bdp)₃, (c) K_1.06Fe₂(bdp)₃ dosed with 1 bar O₂ at 298 K, and (d) K_1.06Fe₂(bdp)₃ dosed with 1 bar O₂ at 473 K. Individual peak fits are shown in red and blue. Peak fit backgrounds and envelopes are shown with black lines.

We turned to solid-state magic-angle spinning (MAS) NMR spectroscopy as a more sensitive probe of local structural changes occurring upon O₂ dosing, and chose to study the Na analogue due to the greater ease of obtaining ²³Na data compared to ³⁹K NMR data. Typically, NMR spectroscopy of paramagnetic systems is challenging due to hyperfine interactions between unpaired electrons and the NMR-active nucleus. These interactions may be isotropic and through-bond (Fermi contact) and/or anisotropic and through-space (hyperfine dipolar coupling), often leading to large NMR shifts and highly broadened spectral features.⁴⁹ For example, the highest frequency ²³Na shift of Na₂FePO₄F is 450 ppm—well outside the typical diamagnetic range of −50 to 100 ppm. In this case, unpaired spin density is transferred along bond pathways from the Fe through O and onto Na.⁵⁰ On the other hand, in some paramagnetic systems, the Fermi contact shift may be unusually small due to competition between delocalization and polarization mechanisms. In NaO₂, where superoxide acts as the paramagnetic center, the room-temperature ²³Na shift is only – 30 ppm.⁵¹

The ²³Na MAS NMR spectra for activated Na_1.04Fe₂(bdp)₃ and for aliquots of the same sample after dosing with O₂ at 298 K and at 473 K, are shown in Figure 5a. Given the large signal width resulting from paramagnetic broadening, data collection required the use of variable-offset cumulative spectroscopy (VOCS), wherein spin echo sub-spectra are acquired at spaced frequency offsets and summed together.⁵² Two features are consistently observed in these spectra: a relatively sharp feature centered at −12 ppm with associated spinning sideband manifold and a very broad signal centered at ~50 ppm, which is somewhat obscured by the first signal. Baseline subtractions of the sub-spectra were performed to ensure that the intensity of the broad feature was accurate. Moreover, neither signal was observed in control experiments with an empty probe, confirming they arose from Na within the sample.

Figure 5.

(a) ²³Na VOCS MAS NMR spectra of activated Na_1.04Fe₂(bdp)₃ (solid blue curve) and the same material dosed with 1 bar O₂ at 298 and at 473 K (red and dark red, respectively); spectra were acquired at 11.7 T at a MAS rate of 9 kHz. The ²³Na NMR spectra are deconvolved (vertically offset for comparison) into broad and sharp components, the latter including a fitted spinning sideband manifold. Difference plots are shown in Figure S11. (b) Quantitative relative intensities of the deconvolved spectral features as a function of dosing condition. Dashed lines are guides for the eye.

Spectral deconvolutions shown in Figures 5a and S11 strongly suggest that the sharp feature increases and/or the broad feature decreases in intensity with O₂ dosing. For paramagnetic systems, however, observed NMR spectral intensities are generally not quantitative due to rapid spin-lattice (T₁) and spin-spin (T₂) relaxation. For purposes of quantitation, we therefore measured both the ²³Na T₁ and T₂ relaxation times, as shown in Figures S12 and S13. The VOCS spectra are quantitative with respect to T₁ relaxation but not T₂ relaxation. In particular, across all samples, the broad feature has a very short T₂ time (on the order of 500 μs), such that between 35% and 55% of its intensity (depending on the specific sample) is lost prior to acquisition. Correcting the intensity of both deconvolved signals using the measured T₂ times gives quantitative relative intensities (Figure 5b). The broad feature contributes to 75% of the total spectral intensity for the undosed sample, and the relative intensity of the peak decreases to 61% and 45% in the samples dosed with O₂ at 298 and 473 K, respectively. Conversely, the relative intensity of the sharp component increases from 25% in the activated framework to 39% and 55% in the samples dosed with O₂ at 298 and 473 K, respectively.

These spectral trends imply that the broad feature corresponds to ²³Na sites in the reduced structure that are not interacting with oxygen. The sharp component therefore appears to correspond to a chemically distinct environment associated with incorporated oxygen. Though the chemical shift of the sharp component at −12 ppm is in the vicinity of the known room-temperature ²³Na shift of NaO₂ (−30 ppm),⁵¹ this feature is clearly observed even for the undosed sample and therefore cannot be assigned exclusively to superoxo-associated Na⁺. We hypothesize that this feature in the undosed spectrum corresponds to Na⁺ near oxygen-containing ligand-vacancy defects that are clearly observed in the XPS data (for reference, the ²³Na shift of sodium formate is ~0 ppm).⁵³ We also expect a smaller degree of spin density transfer from Fe to ligand vacancy sites, leading to a sharper ²³Na signal for associated Na⁺ relative to the signal arising from sodium ions in the rest of the framework.

In the O₂-dosed samples, then, the sharp feature likely comprises multiple sites, a conclusion supported by analysis of the spinning sideband intensities. In particular, were this feature due only to a single type of Na⁺ site in the activated framework that becomes more predominant with O₂ dosing, the intensity of its spinning sideband manifold should remain proportional to that of the centerband. Instead, the spinning sidebands comprise 65% of the total intensity of this feature in the undosed spectrum but only 48% and 45% of the feature in the dosed spectra at 298 and 473 K, respectively. Additional support comes from the T₁ measurements (Figure S12): the sharp component exhibits monoexponential relaxation for the undosed sample, suggesting a single site, whereas after O₂ dosing the relaxation behavior is bi- or multiexponential, indicating multiple distinct environments with differing T₁ times. We conclude that, with O₂ dosing, a second type of Na⁺ site that we assign to Na⁺ positioned near superoxo species begins to dominate the observed intensity of the sharp component centered at ∡12 ppm. However, paramagnetic broadening leading to a fwhm of ~50 ppm of the sharp component as well as the differing chemical nature of the sodium environment(s) in this system relative to either sodium formate or sodium superoxide mean that deconvolution or distinction by chemical shift alone are not possible, due to likely overlap between the signal observed from Na⁺ near defects and Na⁺ near superoxo species.

To corroborate the sign and magnitude of the assigned ²³Na NMR shifts, we performed DFT calculations on a small cluster model (Figure S14) generated from the refined single-crystal structure of the O₂-dosed sodiated framework (Figure 1). Many of the linkers were further converted to non-bridging phenyl pyrazolate units to minimize the system size, while still capturing the local sodium ion environments. The calculated isotropic hyperfine coupling constants for the ²³Na sites were found to be small and negative, and by using the experimental magnetic susceptibility data (see below), they could be further scaled^54,55 to obtain room-temperature Fermi contact shifts between −9 and −20 ppm. This shift range is in good agreement with the experimental ²³Na shift of the sharp component at −12 ppm, ascribed to Na⁺ sites near reduced O₂ species (and/or defects). These calculations suggest that, despite the highly paramagnetic nature of the framework, the ²³Na nuclei do not experience a significant Fermi contact shift, due to the relatively weak Na⁺⋯framework interactions. Moreover, the calculations confirm that the nearby O₂ species does not induce large ²³Na Fermi contact shifts. We note that experimental ²³Na NMR characterization of sodium superoxide similarly found the absence of a significant ²³Na Fermi contact shift.⁵¹

Finally, we performed static and MAS solid-state ¹⁷O NMR spectroscopy after dosing an activated sample of Na_1.04Fe₂(bdp)₃ with ¹⁷O-enriehed O₂ (see Supporting Information). We did not observe signal in these experiments even after long signal averaging times of ~12 h, suggesting the speciation of O₂ as paramagnetic superoxide rather than diamagnetic peroxide. We note that while ¹⁷O NMR spectra of alkali metal peroxides have been reported, this is not the case for the corresponding superoxides,⁵¹ as the unpaired spin localized on the NMR-active nucleus renders spectral acquisition extremely challenging. Taken together, the ²³Na and ¹⁷O solid-state NMR data support the hypothesis that, upon O₂ dosing, the Na⁺ ions move slightly further from the linkers to accommodate and associate with O₂, which is incorporated as a superoxide guest species.

O₂ Adsorption in More-Reduced Fe₂(bdp)₃ and in an Expanded-Pore Analogue.

Measurable porosity is still observed for O₂-dosed K_1.02Fe₂(bdp)₃ (Figure S15), suggesting that restricted O₂ diffusion—potentially due to pore occlusion by reduced O₂ species—does not completely explain the apparent kinetically limited O₂ adsorption at ambient temperature. To further investigate the possible restriction of O₂ diffusion, we prepared a more reduced form of the framework material, K_1.88Fe₂(bdp)₃, as well as an expanded-pore analogue Fe₂(bpeb)₃ (bpeb²⁻ = 1,4-bis(pyrazolide-4-ylethynyl)benzene)⁵⁶ (see below). In the more reduced material, the pores should be even more occluded and O₂ diffusion more restricted. Indeed, as noted above, this material is essentially nonporous to N₂ at 77 K, with a Langmuir surface area of only ~70 m²/g (Figure S16). As such, if O₂ diffusion represents the primary kinetic barrier to O₂ adsorption, it would be expected that the fully reduced material should show far lower O₂ uptake.

At 298 K, K_1.88Fe₂(bdp)₃ exhibits steep initial O₂ uptake to a loading of 0.68 mmol/g at ~6 mbar that then tapers off to yield a loading of 0.87 mmol/g at 1 bar (Figure 6). Significantly, for all pressures measured, the O₂ capacities are higher than those in the half-reduced material. As observed for K_1.09Fe₂(bdp)₃, the O₂ loading at 0.21 bar is far lower than the theoretical capacity of 2.46 mmol/g. However, since this material adsorbs an appreciable quantity of O₂, it is unlikely that restricted diffusion is the primary reason for the apparent kinetic control of O₂ adsorption. When dosed with O₂ at 473 K, K_1.88Fe₂(bdp)₃ again exhibits steep uptake with even greater capacities of 1.23, 1.40, and 1.64 mmol/g at 10 mbar, 0.21 bar, and 1 bar, respectively, suggesting the fully reduced framework operates under similar kinetic limitations as the half-reduced form. Despite the overall improvement in capacity at elevated temperature, the 1.40 mmol/g uptake (1.13 molecules of O₂ per formula unit, Figure S17) in K_1.88Fe₂(bdp)₃ near atmospheric oxygen partial pressure is only ~60% of its theoretical capacity, whereas K_1.09Fe₂(bdp)₃ achieves approximately 80% under similar conditions. This result indicates that increased kinetic limitations may occur with increasing reduction above K_1.09Fe₂(bdp)₃—such as more restricted movement of cations or sluggish rearrangement of reduced O₂ species. Additionally, it is possible that different redox behavior for the fully reduced material³² could lead to partial formation of more reduced O₂ⁿ⁻ products.

Figure 6.

Adsorption isotherms for the uptake of O₂ at 298 K for K_1.88Fe₂(bdp)₃, K_0.82Fe₂(bpeb)₃, and K_2.07Fe₂(bpeb)₃ (red, light purple, and dark purple filled circles, respectively). Adsorption of O₂ at elevated temperature is shown with triangles. K_1.88Fe₂(bdp)₃ was measured at 473 K whereas the expanded pore analogues were measured at 453 K.

The framework Fe₂(bpeb)₃⁵⁶ features larger interchain separations relative to Fe₂(bdp)₃, (18.2 versus 13.2 Å, respectively; Figure 7), which give rise to larger pores that should reduce or preclude restricted O₂ diffusion. Additionally, this framework displays thermal stability above 350 °C in air, although it is less stable in the presence of water relative to Fe₂(bdp)₃.⁵⁶ We prepared H₂bpeb according to reported procedures^56,57 and synthesized Fe₂(bpeb)₃ in an analogous manner to Fe₂(bdp)₃. Notably, we determined a Langmuir surface area of 2270 m²/g for this expanded material (Figure S18), far higher than the previously reported value of 1600 m²/g.⁵⁶ Treatment of Fe₂(bpeb)₃ with potassium naphthalenide aimed at half and full reduction yielded K_0.82Fe₂(bpeb)₃ and K_2.07Fe₂(bpeb)₃, respectively. Powder X-ray diffraction data confirmed topotactic reduction of Fe₂(bpeb)₃ (Figure S19) as well as adsorption of O₂ without significant loss in crystallinity or changes in symmetry (Figure S20).

Figure 7.

(Left) Solid-state structure of Fe₂(bpeb)₃.⁵⁶ Orange, blue, and gray spheres represent Fe, N, and C atoms, respectively. H atoms are omitted for clarity. (Right) The organic linker H₂bpeb.

Interestingly, the 298-K O₂ adsorption behavior of K_0.82Fe₂(bpeb)₃ is very similar to that of K_1.09Fe₂(bdp)₃, despite its substantially higher Langmuir surface area (1700 m²/g compared to ~750 m²/g, respectively; Figure S21) and much larger pore diameter. After an initial steep uptake until ~1 mbar (reaching a loading of 0.26 mmol/g; Figure 6), the capacity increases gradually with loading to 0.33 mmol/g (0.30 molecules O₂ per formula unit) close to 0.21 bar of O₂ and 0.53 mmol/g at 1 bar. These loadings correspond to 37% and 59% of the theoretical capacity, respectively. The similarities in O₂ adsorption behavior between K_0.82Fe₂(bpeb)₃ and K_1.09Fe₂(bdp)₃ continue at higher temperature, where at 453 K, the expanded-pore material exhibits steep uptake and greater capacities of 1.10, 1.29, and 1.43 mmol/g at 12 mbar, 0.21 bar, and 1 bar, respectively. Note that 453 K represents the activation temperature of the reduced framework and was chosen for this measurement as it should produce very similar behavior to data obtained at 473 K for A_xFe₂(bdp)₃ without increasing the risk of framework degradation for the expanded-pore analogue.

Similar behavior is also observed for the fully reduced compound K_2.07Fe₂(bpeb)₃, which exhibits a Langmuir surface area of 600 m²/g (Figure S22). For this material, the K:Fe ratio is slightly greater than 1, which can likely be ascribed to a combination of metals analysis measurement error and defect site reduction. Sharp O₂ adsorption occurs in this framework at 298 K until ~5 mbar, corresponding to a loading of 0.52 mmol/g, and loadings of 0.55 and 0.65 mmol/g are achieved close to 0.21 bar and at 1 bar, respectively (Figure 6). At 453 K, 1.9 mmol/g O₂ is adsorbed at 50 mbar and 2.3 mmol/g at 1 bar. Notably, though the 0.55 mmol/g O₂ uptake (0.53 molecules per formula unit) in K_2.07Fe₂(bpeb)₃ at 298 K and ~0.21 bar corresponds to only 26% of its theoretical capacity, at 453 K and ~0.21 bar, the material is capable of adsorbing 95% of its theoretical capacity (1.96 equivalents of O₂ per formula unit). These results for the expanded-pore system indicate that there still appears to be some form of kinetic control over the adsorption of O₂ that is almost certainly not associated with restriction of O₂ diffusion. Indeed, if diffusion was the solely limiting factor for O₂ uptake, we would expect the quantity of adsorbed gas to be independent of temperature for Fe₂(bpeb)₃, wherein the large pores should dramatically reduce or eliminate any restriction of diffusion.

Collectively, the reduced forms of Fe₂(bdp)₃ or Fe₂(bpeb)₃ show exceptional selectivity for O₂ at 298 K that only increases at high temperature. The highest O₂ uptake we obtained in the Fe₂(bdp)₃ system at ~0.21 bar O₂ was 1.40 mmol/g at 473 K for K_1.88Fe₂(bdp)₃, compared with the theoretical capacity for K₂Fe₂(bdp)₃ of 2.46 mmol/g. The maximum O₂ capacity at ~0.21 bar O₂ that we measured in the expanded-pore analogue was 2.04 mmol/g at 453 K for K_2.07Fe₂(bpeb)₃, compared with a theoretical capacity for K₂Fe₂(bpeb)₃ of 2.09 mmol/g. To contextualize these results relative to the state-of-the-art, we are not aware of any commercial O₂-selective adsorbents used for air separation. Though the previously mentioned N₂-selective cation-exchanged zeolites have been commercialized, their O₂ output purity is limited, and they must either be regenerated more frequently or used in larger quantities, since N₂ is more abundant than O₂ in air. Several other reported O₂-selective adsorbents exhibit high capacities, but only operate well at temperatures below 298 K, since O₂ binding is either irreversible or too weak at room temperature. For instance, Fe₂(dobdc)²⁵ exhibits a largely reversible uptake of 5.33 mmol/g O₂ under 0.21 bar O₂ at 226 K, whereas O₂ becomes permanently bound at room temperature. On the other hand, though Co₂(OH)₂(bbta)^29,30 and Co-BTTri²⁶ adsorb ~4.5 and ~3.3 mmol/g O₂, respectively, at 195 K and 0.21 bar O₂, they do not show appreciable O₂/N₂ selectivity at 298 K. Some Cr^II-based frameworks such as Cr₃(btc)₂²³ and Cr-BTT²⁴ show strong selectivity for O₂ at room temperature, but degrade or lose capacity with cycling and are unlikely to be able to operate at high temperature. To the best of our knowledge, this is the first report of framework materials that maintain both stability and high selectivity for O₂ at temperatures as high as 473 K, while allowing for at least partial regeneration and cycling. Importantly, we have not optimized the reduced forms of Fe₂(bdp)₃ or Fe₂(bpeb)₃ for O₂ purification. Further modification of these systems may lead to improved reversibility and applying a similar chemical reduction approach to other frameworks could yield enhanced capacities. Therefore, this demonstration of an alternative strategy and mechanism for O₂ binding points to the value of these materials’ continued study and their potential promise for air separation at ambient or elevated temperature.

Electronic and Magnetic Properties of KFe₂(bdp)₃ and O₂-dosed KFe₂(bdp)₃.

To further examine O₂ adsorption behavior in chemically reduced Fe₂(bdp)₃, we turned to a combination of ⁵⁷Fe Mössbauer spectroscopy and magnetic susceptibility measurements. Mössbauer analysis has been used previously to confirm the increasing presence of high-spin Fe^II centers as well as a high degree of electron delocalization with increased chemical reduction of Fe₂(bdp)₃.³² The 5-K Mössbauer spectrum for a sample of K_1.06Fe₂(bdp)₃ dosed with O₂ at 473 K (i.e., the most oxidized, O₂-rich sample) reveals two distinct Fe environments (Figure 8). The primary spectral feature has an isomer shift of 0.129(2) mm/s that matches the shift for Fe₂(bdp)₃ (Fe^III, <δ> = 0.129(1) mm/s) and is distinct from the shift for K_1.1Fe₂(bdp)₃ (Fe^III, <δ> = 0.214(5) mm/s).³² This result indicates that the introduction of O₂ causes electron transfer from the reduced framework, resulting in re-oxidation of the iron centers back to low-spin Fe^III. The isomer shift of the second feature is 0.42(2) mm/s, which is considerably lower than the value reported for Fe^II in K_1.1Fe₂(bdp)₃, and is indicative of a new electronic environment. The exact nature of this species remains unclear, but the signal could correspond to remnant low-spin Fe^II. Another assignment consistent with this second feature is high-spin Fe^III.⁵⁸ In either case, interactions between pore-dwelling superoxo species and the framework may also contribute to the observed signal parameters. The full assignment of this signal and complete understanding of the complex electronic and magnetic structure in either reduced Fe₂(bdp)₃ or its O₂-dosed congeners is beyond the scope of this work, but a more detailed discussion of the Mössbauer measurements can be found in the Supporting Information. The Mössbauer spectrum for K_1.06Fe₂(bdp)₃ dosed with O₂ at ambient temperature is very similar (Figure S25), displaying two distinct features with similar isomer shifts and quadrupole splittings as those for the sample dosed at 473 K. Both datasets indicate that K_1.06Fe₂(bdp)₃ is substantially re-oxidized in the presence of O₂ and features a new iron electronic environment.

Figure 8.

The 5-K Mössbauer spectrum for a sample of K_1.06Fe₂(bdp)₃ dosed with 1 bar O₂ at 473 K. The red and blue doublet fits correspond to low-spin Fe^III and a previously unobserved Fe species, respectively.

Given the paramagnetic nature of both the framework and the observed superoxo species, we further investigated the electronic structure of activated and O₂-dosed samples of K_1.06Fe₂(bdp)₃ using magnetic susceptibility measurements (Figure 9). At 300 K, the value of the molar magnetic susceptibility times temperature (χ_MT) for the activated sample is 3.19 emu·K/mol, while a Curie-Weiss fit of the inverse susceptibility versus temperature (Figure S28) over the range 85 to 300 K yielded values of C = 3.51 emu·K/mol and θ_CW = −33 K. Notably, this value of C is close to the value of 3.53 emu·K/mol expected if the added electrons result in the conversion of low-spin Fe^III sites to high-spin Fe^II sites. However, this result differs from our previously reported Mössbauer spectroscopy results, which show a small fraction of high-spin Fe^II in the half-reduced framework.³² As such, the large χ_MT value of the sample may indicate the presence of clusters of low-spin Fe^III centers strongly coupled by conduction electrons, although further measurements are needed to understand fully the magnetic structure of this material.

Figure 9.

Variable-temperature molar magnetic susceptibility times temperature (χ_MT) versus T obtained at 7 T for Fe₂(bdp)₃ (black), K_1.06Fe₂(bdp)₃ (blue), and K_1.06Fe₂(bdp)₃ dosed with 1 bar of O₂ at 298 K (red), and 473 K (dark red).

Both the ambient-temperature and high-temperature O₂-dosed samples exhibit lower χ_MT values of 1.74 and 2.03 emu·K/mol, respectively, compared to the activated sample (Figure 9), consistent with the removal of conduction electrons upon electron transfer to adsorbed O₂. Importantly, the larger χ_MT value of the high-temperature dosed sample is consistent with an increased concentration of S = 1/2 superoxo species within the pores of the framework. Indeed, the room-temperature χ_MT value of this sample is 0.29 emu·K/mol larger than the room temperature moment of Fe₂(bdp)₃, reasonably close to the expected increase of 0.32 emu·K/mol, assuming one S = 1/2 spin per adsorbed O₂ and no change in the moment of the host framework. Unfortunately, quantitative analysis of the magnetic susceptibility of these samples is complicated by contributions from temperature-independent paramagnetism (Figure S29), even under an applied field of 7 T, as was previously observed for Fe₂(bdp)₃. However, the data qualitatively agree with the formation of an S = 1/2 superoxo species upon adsorption of O₂.

O₂ Adsorption Mechanism.

The foregoing results from gas adsorption, X-ray diffraction, spectroscopic, and magnetic susceptibility experiments provide a consistent picture for the mechanism of O₂ adsorption in these chemically reduced Fe^III-pyrazolate frameworks. Upon introduction of O₂ even at very low pressures, the strongly reducing framework drives what is ostensibly outer-sphere electron transfer, thereby reducing O₂ to a superoxide (O₂⁻) guest species, followed by movement of this reduced species to a more favorable binding position stabilized by the alkali metal cations residing within the pores. The theoretical O₂ uptake expected for this mechanism is not realized at 298 K due to a kinetic limitation. This kinetic limitation is almost certainly not due to the restriction of O₂ diffusion resulting from pore occlusion by reduced O₂ species. Instead, it is likely a result of a large reorganization energy associated with rearrangement of the alkali cations from their preferred positions prior to O₂ dosing and/or the movement and ordering of superoxo species after O₂ reduction. This explanation seems especially plausible when considering the crystal structures of the O₂-loaded frameworks. The alkali metal cation sites that stabilize the O₂ species are over 1.2 Å away from the other alkali metal sites that interact with the ligand phenyl rings and the reduced O₂ species are positioned relatively far away from the iron centers. These distances imply large degrees of rearrangement after reduction of O₂ and suggest that back-transfer of electrons from the reduced O₂ species to the framework may also be subject to such a barrier, requiring significant thermal energy for framework regeneration. One possible approach to enhance the cycling ability of the material would thus be to employ cations that could enhance O₂ reduction by positioning reduced O₂ species much closer to the framework. These templating cations might also facilitate the reversal of this reduction.

Conclusion

Molecular complexes and materials such as metal–organic frameworks that reversibly bind dioxygen traditionally do so at coordinatively unsaturated, redox-active metal sites, which transfer an electron to O₂ by an inner-sphere mechanism.^{9–12,23–26,28} Here, we have presented an alternative strategy for the selective capture of O₂, via outer-sphere electron transfer to O₂ from a robust, chemically reduced framework with coordinatively saturated, redox-active metal sites. Through a suite of characterization techniques, we have shown that the O₂ species adsorbed in A_xFe₂(bdp)₃ (A = Na⁺, K⁺) are superoxide moieties stabilized by sodium or potassium cations. The deeper understanding gained here of a relatively unexplored mechanism of O₂ reduction and binding is of fundamental interest, yet these results can also inform the design of new O₂ selective adsorbents for numerous industries and important pre- and post-combustion carbon capture technologies that require high-purity oxygen.