Carbon dioxide capture and storage is essential to reducing greenhouse gas emissions and meeting net-zero emissions targets1,2. A range of technologies are under development to meet the need for more energy efficient carbon capture. One promising strategy to improve on traditional aqueous amine technology is to use solid adsorbent materials for capture3,4,5. In particular, installation of reactive amine or hydroxide functional groups within a porous scaffold such as a metal-organic framework or a porous silica brings about selective reactivity with CO26,7,8,9,10, with the porous scaffold providing a large surface area for hosting the reactive groups while maintaining channels for CO2 transport. The increasingly complex adsorbent materials under consideration bring major challenges in the characterisation of new carbon capture chemistry, hindering the design of improved materials4. Existing characterisation tools for understanding CO2 capture modes include single-crystal diffraction11,12,13, powder diffraction14, infra-red spectroscopy6,10,15, X-ray absorption spectroscopy16, and NMR spectroscopy8,17,18,19,20, each of which has strengths and limitations in terms of the materials that can be studied and the information that can be obtained. Solid-state NMR spectroscopy is a promising tool for investigating CO2 binding modes in adsorbents as there is no requirement for long-range ordering and detailed information about the local structure and dynamics of the CO2 can be obtained. However, different CO2 adsorption products often give rise to very similar signals in the NMR spectrum and assigning these signals to specific CO2 binding modes remains very challenging8,17,18,19,20.

The most common experiment with the NMR approach is to dose the candidate adsorbent with 13CO2 gas and perform 13C magic angle spinning (MAS) NMR experiments. These experiments are relatively straightforward to perform, but often lead to ambiguous identification of the adsorption products. For amine-functionalised materials, the 13C chemical shifts give poor differentiation between closely related ammonium carbamate, carbamic acid, and ammonium bicarbonate adsorption products, with the signals from these species showing very similar 13C chemical shifts17,18. A similar problem arises for bicarbonate and carbonate products in hydroxide-based materials20. The prediction of NMR parameters with density-functional theory (DFT) calculations21 can improve confidence in the structural assignments, and more advanced multi-nuclear NMR experiments can give improved differentiation between adsorption products17,19,22. However, there remains a pressing need for the exploration of new NMR methods for understanding CO2 capture chemistry23.

A representative emerging class of CO2 adsorbents are amine-functionalised metal-organic frameworks. The framework M2(dobpdc) (dobpdc = 4,4′-dioxidobiphenyl-3,3′-dicarboxylate) (Fig. 1a) can straightforwardly be functionalised with amines to yield a family of (amine)–M2(dobpdc) adsorbents (Fig. 1b)14. These adsorbents have large capacities for selective and reversible CO2 uptake, and the adsorption thermodynamics can be tuned by varying the amine11,24,25,26,27, and the metal14,28,29. Importantly, these materials generally display steep adsorption isotherms making them promising for a range of energy efficient carbon capture applications24,25. Initial characterisation of CO2 adsorption modes in these materials has revealed a rich chemistry, with three CO2 adsorption products proposed to date: (i) ammonium carbamate chains (Fig. 1c), thought to be the dominant product in a range of variants11, (ii) carbamic acid pairs (Fig. 1d), identified in the Zn-based framework functionalised with the diamine dmpn (dmpn = 2,2-dimethyl-1,3-diaminopropane)17,24, and (iii) a mixed adsorption product (Fig. 1e) recently proposed for (dmpn)-Mg2(dobpdc)17. The adsorption thermodynamics of these three adsorption processes vary, motivating further characterisation to aid the design of metal-organic frameworks with the best CO2 capture performances.

Fig. 1: Structure and CO2 adsorption modes in (amine)–M2(dobpdc) metal-organic frameworks.
figure 1

a Structure of the metal organic framework M2(dobpdc). b Amine functionalisation yields (amine)-M2(dobpdc). Upon exposure to CO2 various adsorption products can form. The most dominant adsorption products are: c ammonium carbamate chains, d carbamic acid pairs and e mixed ammonium carbamate–carbamic acid.

Here we leverage the crystalline and tuneable family of (diamine)–M2(dobpdc) adsorbents to perform a systematic computational exploration of solid-state NMR parameters for different CO2 adsorption products. We show that 17O solid-state NMR spectroscopy is a powerful probe of CO2 capture chemistry, providing unambiguous identification of carbamic acid formation and a detailed picture of the hydrogen-bonding environments.

Results and discussion

Computational discovery of 17O NMR as a probe of carbon capture

The broad tunability of the diamine–M2(dobpdc) family of materials and large number of available DFT-calculated adsorption structures presented an excellent opportunity to explore NMR parameters for differentiating CO2 adsorption products. To increase the diversity of structures explored, for select materials we explored not only the leading candidate adsorption product from previous studies11,17, but also the other adsorption products in Fig. 1 (see Supplementary Table 1 for a list of studied materials). While 13C NMR spectroscopy is the most commonly used tool for probing CO2 adsorption products, our DFT calculations show that poor differentiation of ammonium carbamate and carbamic acid binding modes is achieved by the 13C chemical shifts (Supplementary Fig. 1a, b, Supplementary Table 2). Better differentiation of products is achieved by considering the orientation dependence of the 13C chemical shift (i.e., the chemical shift anisotropy), consistent with recent work on amine-functionalised silicas19,23. However, in many cases carbamic acid and ammonium carbamate species still have similar 13C NMR parameters (Supplementary Fig. 1c) because both the protonation state (ammonium carbamate chain or carbamic acid) and other hydrogen bonding interactions have an impact. These findings support the idea that alternative NMR probes beyond 13C are needed to characterise CO2 adsorption modes with greater confidence.

We hypothesised that 17O NMR spectroscopy would be a powerful probe30 of CO2 adsorption modes as the two oxygen atoms per CO2 molecule can have significantly different local environments depending on the adsorption product formed. The investigated adsorption products have a total of four main types of oxygen environment (Fig. 2a). Ammonium carbamate chains have two oxygen environments with one of these bound to a metal ion, carbamic acid pairs have two oxygen environments corresponding to the carbonyl and hydroxyl oxygens, and the mixed adsorption mode features all four of these oxygen environments. Importantly, as 17O is a spin 5/2 nucleus, the resulting NMR spectra will be affected not only by the familiar chemical shift, (δiso), but also by the quadrupolar interaction, i.e., the interaction of the nuclear quadrupole moment with the surrounding electric field gradient, which is defined by the parameters CQ and ηQ31. CQ is the quadrupolar broadening defined as CQ = eQVzz/h and gives the magnitude of the interaction whilst ηQ measures the asymmetry of the interaction as ηQ = (VyyVxx/Vzz), where Vxx, Vyy and Vzz are the principal components of the electric field gradient tensor, e is the electronic charge, Q is the nuclear quadrupole moment, and h is the Planck constant. By performing a 17O MAS NMR experiment, δiso, CQ and ηQ values can be obtained, therefore providing more information than 13C NMR.

Fig. 2: Computation exploration of 17O NMR as a probe of adsorption modes.
figure 2

The calculated 17O NMR parameters for the different oxygen environments present in the main adsorption structures. a Figures illustrating the classifications, with red and blue corresponding to the oxygens found in carbamates and grey and yellow those of carbamic acids. b Plot of CQ vs δiso values for the various oxygen classes. c Plot of ηQ vs δiso values for the various oxygen classes. This figure shows the clear differentiation of adsorption products that is achieved with 17O NMR. Source data are provided as a Source data file.

The calculated 17O NMR parameters (Fig. 2, Supplementary Table 3) show that, broad clusters of data points are found for the various adsorption environments. Excitingly, the OH oxygens in carbamic acid species are differentiated from the other oxygen environments by a lower 17O chemical shift, a higher CQ and a lower ηQ (marked with a yellow box). This differentiation is unambiguous compared to that found by 13C NMR (Supplementary Fig. 1). For the non-protonated oxygen present in carbamic acids (grey), two distinct groupings are seen, corresponding to oxygen in the mixed and carbamic acid pair environments, with the former having a higher shift and CQ but a lower ηQ. Interestingly, the data also show that metal-bound carbamate oxygens (blue) are differentiated from free carbamate oxygens (red) by their generally lower chemical shifts. We note that some outliers are seen amongst the carbamate oxygens (blue and red), which correspond to ammonium carbamate chain structures with different hydrogen bonding arrangements. One of these is for (R,R)-dach–Mg2(dobpdc) (dach = trans-1,2-diaminocyclohexane) where additional hydrogen bonding is seen between pairs of carbamate chains32 and another outlier is for (i-2)-Mg2(dobpdc) (i-2 = n-isopropylethylenediamine) where the proposed structure has a hydrogen bond between an amine and the oxygen next to the metal11. Overall the DFT calculations show that 17O NMR should provide good differentiation between adsorption products, especially for carbamic acid species.

Detection of different carbon capture products with 17O NMR

Motivated by the computational results, experimental 17O NMR spectra were acquired for a series of representative adsorbents. First, a control experiment for activated Mg2(dobpdc), i.e., with no amine functionalisation, revealed a relatively sharp resonance for physisorbed CO2 at 61.5 ppm (Fig. 3a)33.

Fig. 3: 17O MAS NMR spectra of representative 17O-enriched CO2-dosed adsorbents.
figure 3

a The 17O MAS NMR (23.5 T, ν0 17O  = 135.6 MHz, 20 kHz MAS, 32768 transients, experiment time ~27 min) of activated Mg2(dobpdc) i.e. diamine-free MOF. be The MAS 17O NMR spectra (20.0 T, ν0 17O  = 115.4 MHz, 14 kHz MAS, 32768 transients, experiment time ~27 min) of the four investigated diamine-functionalised frameworks. Black, red, blue lines show the experimental data, the fit, and the deconvoluted peaks, respectively. These spectra show strong support for the carbamate chain adsorption mechanism (ee-2, e-2 and i-2 analogues), for the mixed adsorption mechanism (dmpn analogue) and highlight the sensitivity of 17O NMR to different carbon capture modes.

(Ee-2)–Mg2(dobpdc)–CO2 (ee-2 = N,N-diethylethylenediamine) was then selected as the first amine functionalised sample, as previous characterisation has confidently assigned this material to form ammonium carbamate chains11,17. Excitingly, the 17O MAS NMR spectrum supports ammonium carbamate chain formation, with two oxygen environments of similar signal intensity observed (Fig. 3b). Deconvolution of these two resonances, aided by a multiple-quantum MAS (MQMAS) spectrum (Supplementary Fig. 2), gave δiso, CQ and ηQ values in good agreement with DFT-calculated parameters for ammonium carbamate chains (Table 1). The agreement for six NMR parameters gives much greater confidence in the structural assignment than previous NMR work17. It is important to note that the calculated DFT structures exclude the presence of water when considering the adsorption mechanism. Care has been taken to exclude water from the system experimentally (see ‘Methods’). The spectrum for (ee-2)-Mg2(dobpdc) presented two additional peaks, one at 70.5 ppm assigned to physisorbed CO2 and a smaller second peak at around –22 ppm. The identity of this minor peak is unknown, though it potentially arises from CO2 reacting with defects in the metal-organic framework, since this signal was also weakly observed in the spectrum of unfunctionalized Mg2(dobpdc). Finally, the 17O NMR spectrum could be acquired rapidly, with an acquisition time of ~27 min, and with an estimated cost of 17O enriched CO2 gas of £50 per sample, which could be reduced in future by optimising the gas dosing line used to prepare samples. Although care has been taken to ensure adequate delays during the acquisition of these experiments, the integrated signal intensities cannot be considered truly quantitative owing to the differences in T1 and T2 relaxation and the different nutation rates for 17O nuclei with different quadrupolar couplings.

Table 1 Overview of the experimental and DFT-calculated 17O NMR parameters for the various diamine-functionalised frameworks adsorbing CO2 to form carbamate chains

(Dmpn)-Mg2(dobpdc) (dmpn = 2,2-dimethyl-1,3-diaminopropane) is hypothesised to adsorb CO2 via a mixed adsorption mechanism (Fig. 1e)17 and is therefore an ideal candidate to assess whether 17O NMR spectroscopy can be used to determine more complex binding modes. The 17O MAS NMR spectrum (Fig. 3c, Supplementary Figs. 2, 3) shows a broad lineshape with multiple overlapping signals, which could be deconvoluted to reveal four oxygen environments consistent with CO2 binding via the mixed adsorption mechanism. To gain increased confidence in the extracted NMR parameters, we simultaneously fitted data from two magnetic field strengths (Fig. S4). The 12 measured NMR parameters are consistent with the DFT-calculated values for the mixed mechanism (Table 2), and provide important support for this recently hypothesised adsorption mode. Most notably, the experimental results show a clear carbamic acid OH resonance, which stands out with a lower chemical shift, larger CQ and a lower ηQ as predicted by the DFT calculations (see Supplementary Fig. 3 for spectrum at 23.5 T, where the acid resonance is more clearly resolved). Finally, we note that for dmpn–Mg2(dobpdc), two resonances are observed for physisorbed CO2 at 62.7 and 68.1 ppm, with the origin of these currently unclear. Summarising, the experiments on (dmpn)–Mg2(dobpdc) showcase the excellent ability of 17O NMR to determine complex adsorption modes and to distinguish ammonium carbamate and carbamic acid species.

Table 2 The experimental and DFT calculated 17O parameters for dmpn-Mg2(dobpdc)

CO2-dosed (i-2)-Mg2(dobpdc), (i-2 = N-isopropylethylenediamine), and (e-2)-Mg2(dobpdc), (e-2 = N-ethylethylenediamine), were then investigated to further test the robustness of 17O NMR in identifying CO2 adsorption products. The 17O MAS NMR spectra of these compounds closely resemble that of (ee-2)-Mg2(dobpdc), consistent with ammonium carbamate chain formation (Fig. 3d, e). Interestingly, discrepancies arise when comparing the experimental and DFT-calculated NMR parameters for (e-2)-Mg2(dobpdc) and especially (i-2)-Mg2(dobpdc) (Table 1), suggesting that the DFT-proposed models are inaccurate. The models for (ee-2)–Mg2(dobpdc) and (i-2)–Mg2(dobpdc) (based on single crystal diffraction structures for the analogous zinc frameworks)11 show important differences in their hydrogen bonding arrangements, with a hydrogen bond formed from the ammonium to the “free” carbamate oxygen for the ee-2 variant (Supplementary Fig. 5a), and the metal bound carbamate oxygen for the i-2 variant (Supplementary Fig. 5b). In the DFT model for (e-2)–Mg2(dobpdc) additional hydrogen bonding interactions are also present. The findings that (i) the experimental 17O NMR parameters for (i-2)-Mg2(dobpdc) and (e-2)-Mg2(dobpdc) show poor agreement with the DFT values, and (ii) the spectra closely resemble that for (ee-2)-Mg2(dobpdc), challenge these alternative hydrogen bonding arrangements and suggests that all structures instead have hydrogen bonds solely between the ammonium proton and the “free” oxygen as in (ee-2)–Mg2(dobpdc) (Supplementary Fig. 5a). DFT parameters for an improved (e-2)-Mg2(dobpdc) model are shown in Supplementary Fig. 6 and Supplementary Table 4, with our findings further highlighting the excellent ability of 17O NMR experiments to differentiate subtly different CO2 adsorption products.

Observation of carbamic acid formation in (ii-2)–Mg2(dobpdc)

As a final test of 17O NMR as a probe of carbon capture mechanisms in MOFs, we examined the capture mode of (ii-2)-Mg2(dobpdc) for the first time (ii-2 = N,N-diisopropylethylenediamine). We initially assumed this material would form ammonium carbamate chains upon CO2 adsorption, as in the related material (ee-2)–Mg2(dobpdc). Excitingly, the 17O MAS NMR spectrum acquired at 23.5 T (Fig. 4a) instead reveals a clear carbamic acid resonance (δiso = 125.8 ppm, CQ = 7.99 MHz, ηQ = 0.43, see carbamic acid OH groups in Fig. 2). This peak was also seen in a spectrum on an independent sample at 20.0 T (Supplementary Fig. 7). Furthermore, the integral of the carbamic acid peak relative to the rest of the peaks is 1:3.34, likely indicating the presence of four oxygen environments which would be expected in a mixed ammonium carbamate–carbamic acid adsorption structure. The NMR spectrum seen is different from that of dmpn–Mg2(dobpdc) (Supplementary Fig. 3), hinting that this is a different mixed mechanism to that previously reported (Fig. 1e). The left-hand overlapped feature is harder to deconvolute, likely consisting of three oxygen environments at similar shifts.

Fig. 4: A mixed CO2 adsorption mode for (ii-2)–Mg2(dobpdc).
figure 4

a 17O NMR (23.5 T, ν0 17O = 135.6 MHz, 20 kHz MAS, 1048576 transients, experiment time ~15 h) spectrum of (ii-2)-Mg2(dobpdc) dosed with 17O-CO2. The red line shows the deconvolution of a single oxygen environment assigned to a carbamic acid OH group. b MAS 13C NMR (16.4 T, 15 kHz MAS) spectra acquired by cross polarization (contact time 1 ms) for ii-2–Mg2(dobpdc)–CO2 dosed at 950 mbar 13CO2. c 1H → 13C heteronuclear correlation (contact time 100 µs) spectrum, with key correlation peaks indicated by blue arrows on hypothesised Lewis structures. d Lewis structure of the DFT-calculated proposed structure. Key 1H–13C correlation groups are shown. These spectra show strong evidence for a previously unknown adsorption mechanism in amine-functionalised where carbamate chains and carbamic acids form on opposite chains.

Investigating this adsorption mechanism further, a 13C MAS NMR spectrum (Fig. 4b) showed two chemisorbed CO2 resonances at 163.8 and 160.1 ppm, assignable to ammonium carbamate and carbamic acid, respectively, and consistent with the observations from 17O NMR. The 13C peaks had relative intensities of 1:0.9 (quantitative NMR) further supporting a mixed adsorption mechanism consisting of two different CO2 environments. Support for the 13C peak assignments is provided by a 2D 1H–13C heteronuclear correlation experiment with a short contact time (Fig. 4c), which reveals 1H–13C correlations for hydrogens nearby the carbons of the chemisorbed CO2. Most importantly, the 13C resonance at 160.1 ppm shows a strong correlation with a 1H resonance at 10.3 ppm, assigned to a carbamic acid COOH group17. Strong N–H correlations are observed for both resonances supporting reaction with CO2 at the primary amine in both cases. The 163.8 ppm 13C resonance also shows a weak correlation with an ammonium group at 14.6 ppm, with the 1H chemical shift of this species comparable to those observed previously for tertiary ammonium groups in ammonium carbamate chains for related materials17.

Finally, an adsorption structure was proposed to explain the above results (Fig. 4d). The proposed structure features a mixture of ammonium carbamate chains and carbamic acid chains, with CO2 insertion occurring between the metal-amine bond in both cases. The relative signal ratios from 17O and 13C NMR suggests that these two chain variants are present in similar proportions and hence have similar free energies. To check the proposed model, NMR parameters were obtained from DFT (Table 3) and showed a good agreement for the OH peak corresponding to carbamic acid, however, further work is needed to confidently assign the environments of the other resulting three peaks. Importantly, ii-2 is only the second diamine variant (after dmpn) that has been shown to form carbamic acid. In common with dmpn, ii-2 possesses bulky alkyl groups, suggesting that steric bulk is an important handle for tuning adsorption chemistry in amine-functionalised metal-organic frameworks. This may suggest that a range of diamines which contain bulky alkyl groups can access this new adsorption mechanism and it will be important to understand the kinetics of this mechanism and how it relates to the application of MOF carbon capture systems.

Table 3 The experimental and DFT calculated 17O parameters of (ii-2)-Mg2(dobpdc)

Applying 17O NMR to study CO2 capture in amine-grafted silicas

Overall, the work shown here demonstrates that 17O NMR spectroscopy is a powerful tool for understanding CO2 adsorption mechanisms in MOFs. To highlight the versatility of this approach we further applied this technique to amine-grafted silica materials, another important adsorbent class for CO2 separations19,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54. The CO2 binding modes in amine-grafted silicas remain widely debated and an array of adsorption products have been proposed, including, ammonium carbamate34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50, carbamic acid19,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52, ammonium bicarbonate34,44,45,46,47,48,49,50,51,53,54, and surface-bonded silyl carbamates36,37,48,49,50.

Here, two amine-grafted silicas, propylamine-SBA15 (Pr-Si) and triamine-SBA15 (Tri-Si) (Supplementary Table 6) were synthesised and investigated by 17O MAS NMR spectroscopy at 23.4 T (Fig. 5). The 17O MAS NMR spectra show broad signals at 177.5 ppm and 28.7 ppm for Tri-Si, and 175.5 ppm and 33.3 ppm for Pr-Si, with no signal corresponding to physisorbed CO2 observed in these samples. The left hand signals at ~175 ppm are consistent with ammonium carbamate oxygens, see Fig. 2 and additional DFT calculations for amine-grafted silicas in Supplementary Table 7 and Supplementary Fig. 11, suggesting that ammonium carbamate is the major adsorption product for the two materials. The right hand resonances at ~30 ppm in the two spectra are harder to assign, however, it is clear that these signals arise from an interaction between the CO2 and the amine-grafted silica material given the lack of any 17O resonances in the control experiment without CO2 (Fig. 5c). We considered the previously proposed carbamic acid, ammonium bicarbonate and silylpropylcarbamate species as candidate products (Supplementary Fig. 11, Supplementary Table 7), but DFT NMR calculations gave chemical shifts that differed significantly from the experimental values, suggesting that none of these species are present in any significant quantity (Supplementary Table 7). This new resonance is therefore suggestive either of a new adsorption mode in amine-grafted silica materials, or of isotopic enrichment of oxygen atoms in the silica backbone55,56. Overall our experiments on amine-grafted silicas highlight the broad applicability of 17O NMR measurements as a probe of CO2 capture modes, and we envisage that these techniques will lead to the discovery of new CO2 chemistry in a wide range of materials.

Fig. 5: Studying CO2 capture by amine-functionalised silicas.
figure 5

17O (23.5 T, ν0 17O  = 135.6 MHz, 20 kHz MAS, 65536 transients, experiment time ~55 min) NMR spectrum of a Tri-Si and b Pr-Si dosed with C217O, c undosed Pr-Si. These spectra show signals consistent with ammonium carbamate oxygens and show a lack of signal intensity expected for carbamic acids. Furthermore, 17O NMR spectroscopy reveals a currently unassigned signal at δiso = 30 ppm, suggesting a previously undetermined adsorption mechanism.

In conclusion, this work shows that 17O NMR is an excellent probe of different CO2 adsorption products in amine functionalised adsorbents. In particular, 17O NMR can differentiate between ammonium carbamate chains and carbamic acids in a wide range of materials. Our measurements provide new support for ammonium carbamate chain formation in a series of (amine)–Mg2(dobpdc) variants, and also provide strong evidence for a recently proposed mixed ammonium carbamate–carbamic acid mechanism for the material (dmpn)–Mg2(dobpdc). We reveal carbamic acid formation in a previously poorly studied adsorbent, (ii-2)-Mg2(dobpdc), highlighting the prevalence of carbamic acid in frameworks with bulky amine groups. Finally, initial measurements on amine-grafted silica materials showcase the excellent versatility of the technique, and support the formation of ammonium carbamates in these materials, while also suggesting a new adsorption mode may be in operation. It is worth noting that care was taken to ensure no water was present during the preparation of samples. It is known that the presence of water would impact the CO2 adsorption mechanism and 17O NMR spectroscopy would be a powerful tool to explore additional mechanisms further. In the future 17O NMR spectroscopy will be extended to a range of carbon capture technologies and will ultimately enable the design of improved materials that can help tackle the climate crisis.



All of the chemicals used in this project were purchased from commercial suppliers and were used without further purification. The ligand 4,4′-dihydroxy-[1,1′-biphenyl]-3,3′-dicarboxylic acid (H4dobpdc) was purchased from Hangzhou Trylead Chemical Technology. 17O-enriched CO2 gas was purchased from ICON/Berry & Associates, Inc, with ~20 at.% 17O.

Mg2(dobpdc) synthesis

Mg2(dobpdc) was synthesised according to a previously reported procedure11. Mg(NO3)2·6H2O (11.5 g, 45.0 mmol, 1.24 equiv), H4dobpdc (9.90 g, 36.0 mmol, 1.00 equiv), N,N-dimethylformamide (DMF) (90 mL), and methanol (110 mL) were mixed together in a 350 mL glass heavy wall pressure vessel (Chemglass, CG-1880-42). The reaction mixture was sonicated for 15 min until all of the solids had dissolved, and was then sparged with N2 for 1 h. The reaction vessel was sealed and heated at 120 °C with stirring for 21 h. This resulted in the precipitation of a white solid from solution. The solid was collected via vacuum filtration and quickly returned to the reaction vessel along with fresh DMF (250 mL). The reaction vessel was then heated to 60 °C for 3 h with stirring. Following this, the solid was again collected via vacuum filtration and returned to the reaction vessel with fresh DMF (250 mL) and again heated to 60 °C for 3 h with stirring. This washing process with DMF was repeated a total of three times, after which the solid was washed three more times in methanol (250 mL) at 60 °C to yield the desired product, Mg2(dobpdc). A small portion of the product (ca 0.1 g) was collected via filtration and activated for characterisation by powder diffraction (Supplementary Fig. 10) by heating to 60 °C in N2 for 15 h. The remaining Mg2(dobpdc) was stored in methanol.

Diamine-functionalised Mg2(dobpdc) synthesis

Diamine-functionalised Mg2(dobpdc) materials were synthesised according to a procedure previously reported in literature11. Methanol-solvated Mg2(dobpdc) was filtered and washed with toluene (50 mL). The filtered MOF (ca 0.1–0.4 g) was then added to a toluene (4 mL) and diamine (1 mL) solution and left to soak for at least 12 h. The solid was then collected via vacuum filtration and washed with toluene (50 mL). e-2, dmpn, ee-2, i-2 and ii-2, functionalised Mg2(dobpdc) materials were activated by heating in an aluminium bead bath under N2 to 125 °C for 1 h, 150 °C for 1 h, 125 °C for 1 h, 130 °C for 1 h, and for 130 °C for 1.5 h, respectively. This activation step removes solvent as well as excess diamine, and ensures the samples are dry prior NMR sample preparation (see below). A portion (10–20 mg) was taken for powder X-ray diffraction analysis (Supplementary Fig. 10). To determine sample stoichiometries by 1H NMR (Supplementary Fig. 9, Supplementary Table 5), ~5 mg of the activated amine-functionalised MOFs was digested in a mixture of dimethyl sulfoxide (DMSO-d6) (1 mL) and two Pasteur pipette drops of deuterated hydrochloric acid (DCl) (35 wt% in D2O, ≥ 99 at.% D).

Preparation of amine-grafted SBA15

12 g of Pluronic P123 triblock copolymer, 90 g of distilled water, and 360 g of 2 M HCl aqueous solution were mixed in a Teflon-lined container. The mixture was stirred at 35 °C for about 2 h, until complete dissolution of P123. Then, 25.5 g of tetraethyl orthosilicate was added to this solution under vigorous stirring. Stirring was stopped after 5 min, and the mixture was kept under static conditions at 35 °C for 20 h, followed by 48 h at 100 °C in an autoclave. The solid product was collected by filtration, washed with distilled water, dried at ambient condition, and calcined at 550 °C in flowing air for 6 h.

Amine grafting was carried out as described elsewhere57. The SBA-15 support was dried at 120 °C for 4 h to remove residual moisture. Then, 1.0 g of the dried support was transferred to a multineck flask, to which 30 mL of toluene was added. The mixture was stirred at ambient temperature for 2 h, and 0.3 mL of water was added dropwise. The mixing continued for an additional 2.5 h. The temperature was then raised to 110 °C, followed by addition of 1 mL of propylamine silane or 3-[2-(2-aminoethylamino)ethylamino]propyl trimethoxysilane for propylamine-SBA15 (referred to as Pr-Si) and triamine-SBA15 (referred to as Tri-Si), respectively. The mixture was kept under reflux overnight. The grafted materials were filtered and washed with toluene followed by pentane, then dried at room temperature overnight, and archived in sealed vials. Activation before gas dosing for NMR studies was carried out by heating at 120 °C under flowing nitrogen for at least 1 h.

C17O2 dosing of amine functionalised adsorbents

The activated amine functionalised adsorbents were packed into 4 mm or 3.2 mm NMR rotors inside a nitrogen-filled glovebag, thereby excluding water as far as possible. Each sample was then evacuated for a minimum of 10 min in a home-built gas manifold, as described previously17. 17O-enriched CO2 gas (20 at.% 17O) was then used to dose the samples with gas at room temperature, before sealing the rotors inside the gas manifold with a mechanical plunger. (ee-2)-Mg2(dobpdc) was dosed for 0.5 h with a final gas pressure of 896 mbar. The (dmpn)-Mg2(dobpdc) sample for measurements at 20.0 T was dosed for 15 h with a final gas pressure of 1253 mbar, and the second independent (dmpn)-Mg2(dobpdc) sample for measurements at 23.5 T was dosed with 17O-enriched CO2 for 15 h with a final gas pressure of 1113 mbar. (e-2)-Mg2(dobpdc) was dosed for 1 h, and the final gas pressure was 448 mbar. (i-2)-Mg2(dobpdc) was dosed for 0.5 h, and the final gas pressure was 1116 mbar. The (ii-2)-Mg2(dobpdc) sample for measurements at 20.0 T was dosed for 0.75 h with a final gas pressure of 1015 mbar, and the second independent (ii-2)-Mg2(dobpdc) sample for measurements at 23.5 T was dosed with 17O-enriched CO2 for 0.5 h with a final gas pressure of 1039 mbar. For activated Mg2(dobpdc) (i.e., with no amines), activation was first carried out by heating in flowing nitrogen gas at 180 °C for 15 h. This sample was then packed in an NMR rotor (as above) and dosed with gas for 0.5 h, and the final gas pressure was 1075 mbar. For silica grafted amines, samples were dosed inside 3.2 mm NMR rotors for 0.5 h with final gas pressures for Pr-Si and Tri-Si of 1083 mBar and 993 mBar, respectively.

NMR spectroscopy

17O MAS and MQMAS experiments were performed using Bruker spectrometers equipped with a 20.0 T wide-bore and 23.5 T standard bore magnets, corresponding to a 1H Larmor frequencies, ν0, of 850 MHz and 1 GHz. For experiments at 20.0 T, a Bruker Avance III spectrometer was used, alongside a Bruker 4 mm low-ɣ HX double resonance probe, and experiments were performed with an MAS frequency, νR, of 14 kHz. 17O MAS spectra were acquired using a spin-echo pulse sequence with radiofrequency field strength, ν1, of ~50 kHz and a recycle delay of 0.05 s. NMR parameters were optimised experimentally and therefore the spin-echo experiments cannot be considered quantitative. MQMAS experiments were acquired using a z-filter pulse sequence58,59,60,61 with triple-quantum excitation/conversion pulses with ν1 ≈ 50 kHz and a central-transition selective π/2 pulse at ν1 ≈ 11 kHz. All MQMAS spectra are shown after shearing using the convention described in ref. 58. For experiments at 23.5 T, a Bruker Avance NEO spectrometer equipped with a Bruker 3.2 mm HX double resonance probe was used with a MAS rate of 20 kHz. 17O MAS spectra were acquired using a spin-echo pulse sequence, with ν1 = 25 kHz, and with a recycle delay of 0.05 s. Chemical shifts are given in ppm, and are referenced relative to liquid H2O at 0 ppm.

DFT calculations

The candidate structures were first geometry optimised using CASTEP21,62,63,64,65,66,67,68,69,70,71,72,73. This was done with (i) plane-wave basis set with an 80 Ry (1088 eV) cut-off energy, (ii) the on-the-fly generated ultrasoft pseudopotential (C17), (iii) a 1 × 1 × 3 k-point grid, (iv) the Perdew-Burke-Ernzerhol (PBE) functional with a G06 Van de Waals correction.

The NMR parameters were calculated using the same parameters and this gave values of δiso, anisotropy, asymmetry, CQ and ηQ which converged within 0.1 ppm, 0.25 ppm, 0.001, 0.0 MHz and 0.0, respectively, for the investigated oxygen and carbon nuclei at the selected k-point grid and cutoff energy.

For 13C and 17O NMR, the principal components of the chemical shielding tensor (σ11, σ22 and σ33 where σ33 ≥ σ22 ≥ σ11) were obtained directly from the CASTEP calculations, in terms of σxx, σyy and σzz where |σzzσiso| ≥ | σxxσiso| ≥ |σyyσiso|. The principal components of the chemical shielding tensor were converted to chemical shift principal components using δ = –(σcalcσref) where the reference values for 13C and 17O were 171.2 and 249.8 ppm, respectively. These values were obtained from CASTEP calculations on cocaine (13C)74 and the amino acids tyrosine and valine (17O)75, and correlation of the calculated values with the experimental values with a linear fit with a fixed gradient of –1.

Gaussian calculations

The cluster models were created using Avogadro76 based off the models given in ref. 23. Dangling silicon bonds at the surface edges were terminated by OH species50. All calculations were performed using the Gaussian 09 software77. Geometry optimisations and frequency calculations were performed on the model structures prior to the calculation of NMR parameters. Note: no imaginary values were observed in the frequency calculations, and as such the structures were determined to be at the true minima. All calculations were carried out at the CAM-B3LYP/pcS-2 level of theory.

Thermogravimetric analysis

CO2 uptake measurements were carried out using a thermogravimetric analyser (Q500, TA Instruments). Samples (10–20 mg) were pre-treated in flowing nitrogen at 120 °C for 2 h, to remove residual impurities. After cooling the samples to 25 °C, the purge gas was switched to 15% CO2 in N2. After 1 h, the temperature was raised to 50, then 75 °C. The CO2 uptake at each temperature was calculated based on the corresponding weight gain. The samples were then heated to 700 °C under N2 gas, before switching to air for 30 min to remove any residual carbon deposit. Amine content was determined based on the weight loss during decomposition.